prompt
stringclasses 1
value | text
stringlengths 39
387k
|
|---|---|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The filoviruses (family Filoviridae ) from the genera Ebolavirus and Marburgvirus are etiologic agents of sporadic viral hemorrhagic fever outbreaks in humans with high mortality rates. An unprecedented outbreak of Ebola virus (EBOV; species Zaire ebolavirus ) disease that began in Guinea during December 2013 [ 1 ] subsequently spread into neighboring West African countries of Sierra Leone and Liberia, prompting the World Health Organization (WHO) to declare the epidemic a public health emergency of international concern ( http://www.who.int/mediacentre/news/statements/2014/ebola-20140808/en/ ). Phylogenetic analysis of viral isolates from this epidemic suggests a single transmission event introduced the virus, named the EBOV Makona variant [ 2 ], from an undetermined natural reservoir into humans in Guinea, followed by transmission between humans to spread the virus throughout Guinea and into Sierra Leone and Liberia [ 3 ]. Implementation of containment measures such as patient isolation and improved burial practices eventually controlled the epidemic, which resulted in 28,616 reported cases with a mortality rate of approximately 40% ( http://www.who.int/csr/disease/ebola/en/ ).
The severity of this epidemic and principle transmission from human to human underscored the need for efficacious vaccines (and therapeutics) against EBOV, accelerating the placement of candidate EBOV vaccines into clinical safety trials [ 4 – 6 ]. This need for safe and efficacious vaccines was again evident with the onset of the 10 th and largest outbreak in the Democratic Republic of the Congo (DRC) from 2018–2020. The 11 th outbreak of EVD continues in the Western DRC.
The characteristics of filovirus infection, where infected patients are contagious only after manifestation of symptoms, allows one to use a ring vaccination strategy for disease containment. Ring vaccination strategy relies on the combination of contact tracing for case identification and a rapid effective vaccine for use in contacts and contacts of contacts of infected patients. The application of this strategy led to the approval of rVSV-ZEBOV (ERVEBO ® ), a single dose vaccine, using the safety and efficacy data from the clinical trial during the 2014 outbreak in West Africa by the Food and Drug Administration in December 2019 ( https://www.fda.gov/news-events/press-announcements/first-fda-approved-vaccine-prevention-ebola-virus-disease-marking-critical-milestone-public-health ). The effectiveness of ERVEBO in a ring vaccination response provides an important countermeasure for public health but does not address all unresolved questions in filovirus vaccine utilization including duration of protection, alternate dosing regimens, and the effectiveness of filovirus vaccines based on other viral platforms or alternative strategies. The development of multiple countermeasures against a disease necessitates the use of a common assay based on a surrogate of protection which can be used to compare the elicited immune response between vaccines and provide valuable information as to the effectiveness and durability of protection. Ideally, this assay is not only informative but simple, reproducible, species independent, and transferrable between labs. For example, during the development of countermeasures against anthrax, a lethal toxin neutralization assay was developed and used by many laboratories [ 7 ].
The development of vaccine candidates for Ebola virus disease prophylaxis [ 8 ] continues today, including deployment of a heterologous prime boost vaccine with European Commission Market Authorization during the last outbreak. However, the demonstration of efficacy for new filovirus vaccines will be complicated in the absence of a large outbreak and may require evaluation under the FDA Animal Rule or via non-inferiority trials against ERVEBO. Regulatory evaluation using these approaches is only possible with a correlate of protection and a well-developed assay that can measure the response in well-characterized animal challenge models as well as in human clinical trials. The species-neutral ELISA is ideal for bridging data between humans and animal models. Also, since the assay likely will be utilized in multiple experiments at many sites, it is important to demonstrate that the assay is reproducible among different laboratories.
In order to facilitate the development of additional vaccine countermeasures and to address such questions as the durability of immunity, the FANG has supported the development of a human anti-EBOV GP IgG ELISA. This study describes the FANG efforts to determine if the performance of the human anti-EBOV GP IgG ELISA [ 9 ] is acceptable for sample evaluation across five participating laboratories. Each laboratory used an anti-EBOV GP IgG ELISA to measure levels of binding in human serum samples from a FANG designed human proficiency panel. The panel consisted of ten human serum samples created by the differential dilution of human serum lot number BMIZAIRE105 (pool of serum with an approximate anti-GP IgG concentration of 1,000 ELISA units/mL) with control human serum (BMI529) without antibody activity. The concentration of the proficiency panel samples ranged from 0.00 ELISA units/mL to approximately 800 ELISA units/mL.
Each participating laboratory received sufficient volume of the proficiency panel for initial testing plus repeats and used their own anti-EBOV GP IgG ELISA established assay. The assay was validated at some laboratories and qualified at others [ 9 ]. Data from the participating laboratories were compared by statistical analysis. Both intra-laboratory and inter-laboratory analyses were performed to evaluate repeatability, intermediate precision, dilutional linearity, and accuracy. This paper summarizes both the intra- and inter-laboratory analysis of the results generated in the five separate laboratories. Results from the laboratories are de-identified in the analysis and reported as Laboratory A through E. The repeatability estimate for Laboratory B was greater than the acceptance criteria as established in laboratories that validated the anti-EBOV GP IgG ELISA with human serum, and, as a result, the proficiency panel assay runs were repeated. Results from both the original and repeated runs were included in the analysis and labeled as being from Laboratory B1 and B2, respectively.
Assay method
A common assay method [ 9 ] was tech-transferred to the participating laboratories, but there were minor variations in equipment/materials/procedures between laboratories. The analysis of the proficiency panel in the ELISA was performed similarly at Labs A, B1, and B2. All three used two separate operators on separate days. Samples were analyzed using a starting dilution of 1:62.5 and followed the plate layout as illustrated in Table 1 . These plate layouts represent 15 plates with specific proficiency panel samples on each plate. All 15 plates were run twice for a total of 30 plates for each of Labs A, B1, and B2.
10.1371/journal.pone.0238196.t001
Table 1 Plate layout used at Laboratories A, B1, and B2.
Sample ID
Plate Number
1
2
3
4
5
6
7
8
BMI-ZPP-11
X (3)
X
X
X
BMI-ZPP-12
X
X
X
X
X (3)
BMI-ZPP-13
X (3)
X
X
X
X
X
BMI-ZPP-14
X
X
X
X
X
X
X (3)
BMI-ZPP-15
X
X (3)
X
X
BMI-ZPP-16
X
X
X
X
X
X
BMI-ZPP-17
X
X (3)
X
X
X
BMI-ZPP-18
X
X
X
X
X
BMI-ZPP-19
X
X (3)
X
X
BMI-ZPP-20
X
X
X
X
Sample ID
Plate Number
9
10
11
12
13
14
15
BMI-ZPP-11
X
X
X
X
X
X
BMI-ZPP-12
X
X
X
X
X
BMI-ZPP-13
X
X
X
X
BMI-ZPP-14
X
X
X
BMI-ZPP-15
X
X
X
X
X
X
BMI-ZPP-16
X (3)
X
X
X
BMI-ZPP-17
X
X
X
X
X
BMI-ZPP-18
X (3)
X
X
X
X
BMI-ZPP-19
X
X
X
X
X
X
BMI-ZPP-20
X
X (3)
X
X
X
X
An “X” indicates that sample was analyzed on the indicated plate.
An “X (3)” (shaded) indicates that sample was analyzed on the indicated plate three times.
The analysis of the proficiency panel in the ELISA was performed at Lab C by two separate operators over three days and at Lab D by two separate operators over five days. Samples were analyzed using a starting dilution of 1:50 and followed the plate layout as illustrated in Table 2 . These plate layouts represent 12 plates with specific proficiency panel samples on each plate. All 12 plates were run at least twice for a total of 24 plates for each of Labs C and D.
10.1371/journal.pone.0238196.t002
Table 2 Plate layout used at Laboratories C and D.
Sample ID
Plate Number
1
2
3
4
5
6
BMI-ZPP-11
X (3)
X
X
X
BMI-ZPP-12
X
X
X
X
X
X
BMI-ZPP-13
X (3)
X
X
X
BMI-ZPP-14
X
X
X
X
X
X
BMI-ZPP-15
X
X (3)
X
X
BMI-ZPP-16
X
X
X
X
X
X
BMI-ZPP-17
X
X
X (3)
X
BMI-ZPP-18
X
X
X
X
X
X
BMI-ZPP-19
X
X
X (3)
X
BMI-ZPP-20
X
X
X
X
X
X
Sample ID
Plate Number
7
8
9
10
11
12
BMI-ZPP-11
X
X
X
X
X
X
BMI-ZPP-12
X (3)
X
X
X
BMI-ZPP-13
X
X
X
X
X
X
BMI-ZPP-14
X (3)
X
X
X
BMI-ZPP-15
X
X
X
X
X
X
BMI-ZPP-16
X
X (3)
X
X
BMI-ZPP-17
X
X
X
X
X
X
BMI-ZPP-18
X
X
X (3)
X
BMI-ZPP-19
X
X
X
X
X
X
BMI-ZPP-20
X
X
X (3)
X
An “X” indicates that sample was analyzed on the indicated plate.
An “X (3)” (shaded) indicates that sample was analyzed on the indicated plate three times.
The analysis of the proficiency panel in the ELISA was performed at Lab E by two separate operators over four days. Samples were analyzed using a starting dilution of 1:50 and followed the plate layout as illustrated in Table 3 . This plate layout represents six plates with specific proficiency panel samples on each plate. The six plates were each run four times for a total of 24 plates. For all laboratories, some samples were analyzed three times on the same plate [denoted with “X (3)” in the plate layouts]. These contributed to assay repeatability.
10.1371/journal.pone.0238196.t003
Table 3 Plate layout used at Laboratory E.
Sample ID
Plate Number
1
2
3
4
5
6
BMI-ZPP-11
X
X
X
X
X
X
BMI-ZPP-12
X (3)
X
X
X
BMI-ZPP-13
X
X
X
X
X
X
BMI-ZPP-14
X (3)
X
X
X
BMI-ZPP-15
X
X
X
X
X
X
BMI-ZPP-16
X
X (3)
X
X
BMI-ZPP-17
X
X
X
X
X
X
BMI-ZPP-18
X
X
X (3)
X
BMI-ZPP-19
X
X
X
X
X
X
BMI-ZPP-20
X
X
X (3)
X
An “X” indicates that sample was analyzed on the indicated plate.
An “X (3)” (shaded) indicates that sample was analyzed on the indicated plate three times.
Samples on a given plate were excluded from analysis if the within-assay CV of at least three dilution-adjusted concentrations determined for that sample was greater than 20%. Samples were also excluded if the plate including that sample failed to meet system suitability criteria. Some samples and plates that failed to meet the sample suitability criteria or system suitability criteria were repeated on later days. The ELISA concentrations of each qualification test sample by laboratory are provided in the supplemental information ( S1 – S6 Tables).
This study, and specifically the use of human serum samples, was approved in writing by the Battelle Institutional Review Board in April of 2015 (approval number HSRE 0223–100062052). Human serum samples were collected from subjects by the sponsor (Crucell Holland) via written consent according to their IRB-approved protocol. These samples were not specifically collected for this interlaboratory study but rather for a different study. Battelle nor any authors were affiliated with this initial study. The sponsor subsequently provided Battelle volumes of these samples for the purposes of conducting the study described in this manuscript. Throughout its analysis of human biological materials and reporting, Battelle had no access to volunteer subjects’ identifiers nor any access to any code-key that would allow Battelle researchers to attribute any results of analysis to the original volunteer human research subjects.
Statistical methods
Inter-laboratory analysis was performed using the combined results across all laboratories. A mixed-effects analysis of variance (ANOVA) model was fitted to the base-10 log-transformed concentrations to evaluate both inter-laboratory precision (i.e., between lab precision) and intra-laboratory precision (i.e., within-laboratory precision). The model included a fixed effect for test sample and random effects for laboratory, test date nested within laboratory, and plate nested within day. Here, test operator was excluded as a random effect because this variable was indistinguishable from test day in most laboratories. Because of this confounding of effects, any variability attributable to test day may also be due to the different test operators.
Results were screened for outliers within each laboratory separately. Deleted studentized residuals were computed for each observation. If the absolute value of the deleted studentized residual was greater than four, then the observation was considered a statistical outlier and removed from the inter-laboratory analysis.
Variability associated with the random effects as well as intermediate precision, repeatability, and total assay variability were estimated separately for each lab using model-based percent coefficient of variation (CV). The percent CV for each source of variance was calculated using Tan’s [ 10 ] relative standard deviation as
100 × e ln ( 10 ) 2 × σ 2 − 1
where σ 2 is the model-estimated variance for the specific variance source. The percent CV associated with the residual variance served as an estimate for the assay repeatability. The percent CV associated with the test day and plate effects served as an estimate for the intermediate precision of the assay. Total assay variability was estimated using all variance components from the model (both inter- and intra-run variability).
The model intercept was obtained for each test sample from the mixed effects ANOVA model to serve as test sample consensus values across the laboratories. Agreement among laboratories was evaluated by comparing individual assay results from each laboratory to the consensus values. Boxplots were produced for each test sample to show the distribution of concentrations by laboratory in relation to the corresponding consensus value. The ratio of individual test results to consensus values was calculated by test sample to evaluate the level of agreement among laboratories based on two one-sided tests (TOST) of equivalence.
To assess dilutional linearity, a random coefficients linear regression model was fitted to the log-transformed observed concentrations versus the log-transformed target concentrations. The model included both a random intercept and slope effect for each laboratory, along with random effects for laboratory, test day nested within laboratory, and plate nested within laboratory. The random slope coefficients were modeled as laboratory-specific differences from the overall slope. The overall slope was used to assess the dilutional linearity based on a test of equivalence (TOST) and random slope coefficients were used to evaluate the level of agreement among the laboratories.
Results
Across all six laboratory runs, there were some false positive observations for Sample 18, a sample with a known negative concentration. All reportable values from Sample 18 were excluded from the statistical models. Table 4 lists five outliers that were removed from their respective intra-laboratory analyses that were also removed from this inter-laboratory analysis. One outlier each were removed from Laboratories B1 and B2. Three outliers were removed from Laboratory C. In the final analysis, Lab A contributed 204 reportable values, Lab B1 had 179 reportable values, Lab B2 had 214 reportable values, Lab C had 268 reportable values, Lab D had 216 reportable values, and Lab E had 218 reportable values.
10.1371/journal.pone.0238196.t004
Table 4 Statistical outliers identified during analysis of intra-laboratory data.
Laboratory
Test Sample
Observed Concentration (ELISA Units/mL)
Target Concentration (ELISA Units/mL)
Studentized Residual
B1
BMI-ZPP-17
4.28
200
-9.48
C
BMI-ZPP-13
896.47
300
5.34
C
BMI-ZPP-16
236.02
500
-4.74
B2
BMI-ZPP-19
51.20
100
-4.39
C
BMI-ZPP-14
1845.88
700
4.33
These observations were deleted from both intra- and inter-laboratory analyses.
Table 5 presents ANOVA variance estimates and %CV for each source of variability, intermediate precision, and total assay variability by laboratory. For Laboratory A, the %CV for test date and plate nested within test date were 0.0 and 9.8, respectively. For Laboratory B1, the %CV for test date and plate nested within test date were 10.8 and 15.3, respectively. For Laboratory B2, the %CV for test date and plate nested within test date were 4.5 and 8.9, respectively. For Laboratory C, the %CV for test date and plate nested within test date were 9.8 and 8.5, respectively. For Laboratory D, the %CV for test date and plate nested within test date were 18.9 and 10.5, respectively. Finally, for Laboratory E, the %CVs for test date and plate nested within test date were 7.3 and 5.0, respectively. Laboratory E had the lowest %CV for intermediate precision (8.9) while Laboratory A had the lowest %CV for repeatability (7.2) and total assay variability (12.2). Laboratory B1 had the highest repeatability and total assay variability (23.7%CV and 30.6%CV, respectively) while Laboratory D had the highest %CV for intermediate precision (21.7).
10.1371/journal.pone.0238196.t005
Table 5 Summary of variance components obtained from mixed ANOVA model fit to data from all laboratories (results shown by laboratory).
Laboratory A
Source of Variability
Variance
%CV
Test Date
0.0000
0.0
Plate Nested in Test Date
0.0018
9.8
Intermediate Precision 1
0.0018
9.8
Residual (Repeatability)
0.0010
7.2
Total Assay Variability 2
0.0028
12.2
Laboratory B1
Source of Variability
Variance
%CV
Test Date
0.0022
10.8
Plate Nested in Test Date
0.0044
15.3
Intermediate Precision 1
0.0065
18.8
Residual (Repeatability)
0.0103
23.7
Total Assay Variability 2
0.0169
30.6
Laboratory B2
Source of Variability
Variance
%CV
Test Date
0.0004
4.5
Plate Nested in Test Date
0.0015
8.9
Intermediate Precision 1
0.0019
9.9
Residual (Repeatability)
0.0033
13.3
Total Assay Variability 2
0.0052
16.7
Laboratory C
Source of Variability
Variance
%CV
Test Date
0.0018
9.8
Plate Nested in Test Date
0.0014
8.5
Intermediate Precision 1
0.0031
13.0
Residual (Repeatability)
0.0027
11.9
Total Assay Variability 2
0.0058
17.7
Laboratory D
Source of Variability
Variance
%CV
Test Date
0.0066
18.9
Plate Nested in Test Date
0.0021
10.5
Intermediate Precision 1
0.0087
21.7
Residual (Repeatability)
0.0023
11.2
Total Assay Variability 2
0.0110
24.6
Laboratory E
Source of Variability
Variance
%CV
Test Date
0.0010
7.3
Plate Nested in Test Date
0.0005
5.0
Intermediate Precision 1
0.0015
8.9
Residual (Repeatability)
0.0015
9.0
Total Assay Variability 2
0.0030
12.7
1 . Comprised of test date and plate nested within test date sources of variability.
2 . Comprised of repeatability and intermediate precision.
Table 6 shows the consensus values (geometric means) along with 95% confidence intervals for each test sample generated from the mixed model ANOVA fitted to the data. Boxplots by sample of the reportable values from each laboratory, with each plot including a horizontal line for the consensus value estimate for the given sample, are provided in the supplemental information ( S1 – S9 Figs).
10.1371/journal.pone.0238196.t006
Table 6 Consensus values by test sample generated from intercept of mixed ANOVA model fit to data from all laboratories.
Sample ID
Target Concentration
Consensus Value
95% CI Consensus Value
BMI-ZPP-11
600
695.93
(677.31, 715.06)
BMI-ZPP-12
400
475.20
(462.47, 488.27)
BMI-ZPP-13
300
325.47
(316.72, 334.46)
BMI-ZPP-14
700
844.08
(821.24, 867.55)
BMI-ZPP-15
800
871.34
(847.91, 895.41)
BMI-ZPP-16
500
561.81
(546.71, 577.33)
BMI-ZPP-17
200
226.25
(220.18, 232.49)
BMI-ZPP-19
100
110.63
(107.66, 113.68)
BMI-ZPP-20
50
70.81
(68.89, 72.79)
Table 7 shows the ratio of the mean concentration for each of the six individual laboratory runs to the consensus value for a given sample along with a 90% confidence interval for the ratio. Agreement among laboratories implies that these ratios should be close to one, indicating that the average concentrations are about the same as the consensus values. The ratios range from 0.95 to 1.08 for Laboratory A; from 0.96 to 1.19 for Laboratory B1; from 0.83 to 1.12 for Laboratory B2; from 0.96 to 1.16 for Laboratory C; from 0.71 to 0.97 for Laboratory D; and from 0.90 to 1.06 for Laboratory E. Fig 1 shows a graph of the mean ratio and 90% confidence interval for each test sample by laboratory.
10.1371/journal.pone.0238196.g001
Fig 1
Graph of ratio of laboratory mean concentration to consensus value with 90% confidence intervals for each test sample by laboratory.
Dotted lines show equivalence region (0.80 to 1.25) and perfect agreement with consensus value (1.00). All means and confidence bounds are entirely within equivalence region for Laboratories A, B2, C, and E.
10.1371/journal.pone.0238196.t007
Table 7 Ratio of laboratory mean concentration to overall consensus value with 90% confidence intervals for each test sample.
Sample ID
Laboratory A
Laboratory B1
Laboratory B2
Ratio
90% Confidence Interval
Ratio
90% Confidence Interval
Ratio
90% Confidence Interval
BMI-ZPP-11
1.08
(1.04, 1.13)
1.09
(0.97, 1.23)
0.83
(0.81, 0.85)
BMI-ZPP-12
1.02
(0.98, 1.08)
0.97
(0.87, 1.09)
1.12
(1.08, 1.16)
BMI-ZPP-13
1.02
(0.98, 1.06)
1.15
(1.02, 1.30) *
0.93
(0.89, 0.97)
BMI-ZPP-14
0.99
(0.96, 1.02)
0.96
(0.87, 1.05)
1.09
(1.04, 1.13)
BMI-ZPP-15
1.01
(0.97, 1.06)
1.08
(0.98, 1.18)
0.88
(0.84, 0.91)
BMI-ZPP-16
1.00
(0.96, 1.04)
1.07
(1.00, 1.14)
1.05
(1.00, 1.09)
BMI-ZPP-17
1.06
(1.01, 1.11)
1.01
(0.88, 1.15)
1.12
(1.06, 1.18)
BMI-ZPP-19
0.95
(0.91, 0.98)
0.98
(0.87, 1.09)
0.83
(0.79, 0.87) *
BMI-ZPP-20
0.98
(0.95, 1.02)
1.19
(1.02, 1.39) *
0.92
(0.88, 0.97)
Sample ID
Laboratory C
Laboratory D
Laboratory E
Ratio
90% Confidence Interval
Ratio
90% Confidence Interval
Ratio
90% Confidence Interval
BMI-ZPP-11
1.10
(1.05, 1.15)
0.85
(0.81, 0.88)
0.90
(0.87, 0.94)
BMI-ZPP-12
1.08
(1.03, 1.12)
0.74
(0.72, 0.77) *
0.96
(0.92, 0.99)
BMI-ZPP-13
1.09
(1.04, 1.14)
0.79
(0.75, 0.83) *
0.95
(0.92, 0.99)
BMI-ZPP-14
1.00
(0.96, 1.05)
0.75
(0.71, 0.80) *
1.06
(1.02, 1.10)
BMI-ZPP-15
1.10
(1.04, 1.15)
0.90
(0.84, 0.97)
0.98
(0.95, 1.01)
BMI-ZPP-16
1.06
(1.02, 1.11)
0.76
(0.73, 0.79) *
1.01
(0.97, 1.06)
BMI-ZPP-17
0.96
(0.93, 1.00)
0.71
(0.69, 0.74) *
1.02
(0.99, 1.05)
BMI-ZPP-19
1.16
(1.10, 1.22)
0.97
(0.94, 1.01)
0.92
(0.89, 0.96)
BMI-ZPP-20
1.10
(1.02, 1.19)
0.77
(0.73, 0.82) *
1.02
(0.98, 1.06)
* 90% confidence interval is outside the acceptance bounds of (0.80, 1.25). Therefore, the concentrations for this test sample are not equivalent to those of other laboratories.
An equivalence test was conducted to determine if the mean test sample concentrations for each laboratory were equivalent to the corresponding test sample consensus value. An equivalence interval of 0.80 to 1.25 (representing a difference of 20% on the log scale) for the ratio of laboratory mean concentration to consensus concentration was used. The mean laboratory concentration for a given test sample is said to be equivalent to the consensus value for that sample if the 90% confidence interval for the ratio of these two values falls completely within the interval (0.80, 1.25).
Following this equivalence criteria: two intervals from Laboratory B1 (corresponding to BMI-ZPP-13 and BMI-ZPP-20) had an upper bound greater than the upper acceptance limit of 1.25 (1.30 and 1.39); one interval from Laboratory B2 (corresponding to BMI-ZPP-19) had a lower bound less than the lower acceptance limit of 0.80 (0.79); and six intervals from Laboratory D (corresponding to BMI-ZPP-12, BMI-ZPP-13, BMI-ZPP-14, BMI-ZPP-16, BMI-ZPP-17, and BMI-ZPP-20) had a lower bound less than the lower acceptance limit of 0.80. Furthermore, three of the six intervals are entirely below the lower acceptance bound of 0.80. These findings indicate that mean concentrations observed at Laboratory D are not equivalent to the other laboratories for six of the nine test samples.
Table 8 presents the estimated slope across the five laboratories and the corresponding 90% confidence interval obtained from the random regression model fit to assess the relationship between log 10 (observed concentration) and log 10 (target concentration). The overall slope was estimated to be 0.95 with a 90% confidence interval of (0.93, 0.97). An equivalence test was conducted to determine if the overall slope was equivalent to 1.00 (perfect dilutional linearity). An equivalence interval of 0.80 to 1.25 for the overall slope was used. Because the 90% confidence interval for the overall slope was completely within the interval (0.80, 1.25), the concentrations were found to be dilutionally linear across the laboratories. The slope estimates specific to each laboratory ranged from 0.94 to 0.96 ( Table 8 ) and were consistent with the overall slope.
10.1371/journal.pone.0238196.t008
Table 8 Estimated slope and lower and upper 90% confidence interval bounds by laboratory from random coefficients regression model fit to all data.
Laboratory
Slope Estimate
90% Confidence Interval #
Overall (All Labs)
0.95
(0.93, 0.97)
A
0.96
(0.90, 1.02)
B1
0.94
(0.88, 1.01)
B2
0.96
(0.90, 1.02)
C
0.96
(0.90, 1.02)
D
0.95
(0.89, 1.00)
E
0.95
(0.90, 1.01)
# 90% confidence interval is within the acceptance bounds of (0.80, 1.25). Therefore, the concentrations were dilutionally linear across the laboratories.
Discussion
The value of an assay as a regulatory tool is dependent on its accuracy, consistency, simplicity, and reproducibility. An assay that is relevant, is species independent, and replicable among laboratories is a powerful tool for product development. The data from a number of clinical trials utilizing ERVEBO strongly suggest that the anti-EBOV GP IgG ELISA provides data that correlate with product efficacy against Ebola infection. The development of new vaccines, or the evaluation of durability or alternative dosing regimens will be based on interpretation of data using the human anti-EBOV GP IgG ELISA. Our ability to use, or trust the data generated from non-clinical studies in different laboratories and clinical trials carried out with sera evaluated at different sites will require an understanding regarding the consistency and reproducibility of the assay among laboratories. In particular, assays using material from animal studies may be performed in laboratories different from those where the assay was performed to evaluate clinical trials. If the assay performance is not consistent among species and across laboratories, then data interpretation will not be possible. This interlaboratory study provided a direct head-to-head comparison of the ELISA performance in five different laboratories. The results from this study confirm the assay can be a universal tool for Ebola virus vaccine evaluation since results were similar when using the assay at multiple labs. However, the small differences in assay performance reinforce that for regulatory purposes, it is still ideal to rely on only one test site where the assay is fully validated.
Intermediate precision for the six laboratory runs ranged from 8.9 to 21.7%CV and repeatability ranged from 7.2 to 23.7%CV. The total assay variability %CVs range from 12.2 to 30.6. As a point of reference, laboratories that validated the anti-EBOV GP IgG ELISA have used the following precision acceptance criteria: (1) The intermediate precision of the assay for samples within the analytic range of the assay must be no larger than 25% CV; and (2) the repeatability of the assay for samples within the analytic range of the assay must be no larger than 20% CV. The repeatability estimate for Laboratory B1 was greater than the upper acceptance bound as established in laboratories that validated the anti-EBOV GP IgG ELISA with human serum. However, a repeat of the proficiency panel run at this laboratory following additional training of laboratory staff resulted in a repeatability estimate less than the upper acceptance bound; thus, illustrating the importance of rigorous training of laboratory staff and the strict adherence to assay procedures to ensure consistent results between runs.
Similarly, laboratories that validated the anti-EBOV GP IgG ELISA have used the following dilutional linearity (relative accuracy) acceptance criteria: the 90% confidence interval for the slope from the random regression model fit to data between the limits of quantitation and relating log 10 (concentration) to log 10 (spike level) will be entirely within (-1.20, -0.80). The interlaboratory study models dilutional linearity as log 10 (observed concentration) to log 10 (target concentration) resulting in a positive relationship between the two variables. Therefore, to conclude that dilutional linearity is acceptable in relation to the validation in human serum, the 90% confidence interval for the slope should be positive and fall entirely between 0.80 and 1.20. The overall slope was 0.95 and has a 90% confidence interval estimate of (0.93, 0.97); thus, the dilutional linearity is within the acceptance criteria as established in the assay validation with human serum.
Agreement among laboratories implies that the ratios of the mean concentration for the five individual labs to the overall laboratory consensus value for a given sample should be close to one. The ratios range from 0.95 to 1.08 for Laboratory A; from 0.96 to 1.19 for Laboratory B1; from 0.83 to 1.12 for Laboratory B2; from 0.96 to 1.16 for Laboratory C; from 0.71 to 0.97 for Laboratory D; and from 0.90 to 1.06 for Laboratory E. Equivalence test results showed that the 90% confidence interval for the ratio were within the equivalence bounds of 0.80 to 1.25 for each laboratory except for samples BMI-ZPP-13 and BMI-ZPP-20 in Laboratory B1, BMI-ZPP-19 in Laboratory B2, and six samples in Laboratory D.
The assessment of between-laboratory performance revealed lower observed concentrations at Lab D and greater variability in assay results at Lab B1 relative to the other laboratories. The lower observed concentrations at Lab D illustrate the importance of monitoring assay performance and harmonizing across laboratories. Given the inherent differences from subject-to-subject in clinical trials and animal-to-animal in non-clinical studies, these differences observed at Lab D relative to the other laboratories are not likely to affect interpretation of study results. The variability in assay results at Lab B1 was mitigated by additional laboratory staff training.
The evaluation of the proficiency panel at these laboratories provides a limited assessment of assay precision (intermediate precision, repeatability, and total assay variability), dilutional linearity, and accuracy. This limited evaluation suggests that the within-laboratory performance of anti-EBOV GP IgG ELISA as implemented at the five laboratories is performing consistently with the intended use of the assay based on the acceptance criteria used by laboratories that have validated the assay.
Supporting information
S1 Fig
Observed concentration (ELISA Units/mL) by laboratory for sample BMI-ZPP-11.
(Consensus Concentration = 695.93). Center line in the box depicts the median concentration while the height of the box represents the 25 th and 75th percentile of the concentration distribution. Vertical lines extending above and below the box represent the maximum and minimum concentration values for the laboratory. Open circles show the observed concentrations.
(TIF)
S2 Fig
Observed concentration (ELISA Units/mL) by laboratory for sample BMI-ZPP-12.
(Consensus Concentration = 475.20). Center line in the box depicts the median concentration while the height of the box represents the 25 th and 75th percentile of the concentration distribution. Vertical lines extending above and below the box represent the maximum and minimum concentration values for the laboratory. Open circles show the observed concentrations.
(TIF)
S3 Fig
Observed concentration (ELISA Units/mL) by laboratory for sample BMI-ZPP-13.
(Consensus Concentration = 325.47). Center line in the box depicts the median concentration while the height of the box represents the 25 th and 75th percentile of the concentration distribution. Vertical lines extending above and below the box represent the maximum and minimum concentration values for the laboratory. Open circles show the observed concentrations.
(TIF)
S4 Fig
Observed concentration (ELISA Units/mL) by laboratory for sample BMI-ZPP-14.
(Consensus Concentration = 844.08). Center line in the box depicts the median concentration while the height of the box represents the 25 th and 75th percentile of the concentration distribution. Vertical lines extending above and below the box represent the maximum and minimum concentration values for the laboratory. Open circles show the observed concentrations.
(TIF)
S5 Fig
Observed concentration (ELISA Units/mL) by laboratory for sample BMI-ZPP-15.
(Consensus Concentration = 871.34). Center line in the box depicts the median concentration while the height of the box represents the 25 th and 75th percentile of the concentration distribution. Vertical lines extending above and below the box represent the maximum and minimum concentration values for the laboratory. Open circles show the observed concentrations.
(TIF)
S6 Fig
Observed concentration (ELISA Units/mL) by laboratory for sample BMI-ZPP-16.
(Consensus Concentration = 561.81). Center line in the box depicts the median concentration while the height of the box represents the 25 th and 75th percentile of the concentration distribution. Vertical lines extending above and below the box represent the maximum and minimum concentration values for the laboratory. Open circles show the observed concentrations.
(TIF)
S7 Fig
Observed concentration (ELISA Units/mL) by laboratory for sample BMI-ZPP-17.
(Consensus Concentration = 226.25). Center line in the box depicts the median concentration while the height of the box represents the 25 th and 75th percentile of the concentration distribution. Vertical lines extending above and below the box represent the maximum and minimum concentration values for the laboratory. Open circles show the observed concentrations.
(TIF)
S8 Fig
Observed concentration (ELISA Units/mL) by laboratory for sample BMI-ZPP-19.
(Consensus Concentration = 110.63). Center line in the box depicts the median concentration while the height of the box represents the 25 th and 75th percentile of the concentration distribution. Vertical lines extending above and below the box represent the maximum and minimum concentration values for the laboratory. Open circles show the observed concentrations.
(TIF)
S9 Fig
Observed concentration (ELISA Units/mL) by laboratory for sample BMI-ZPP-20.
(Consensus Concentration = 70.81). Center line in the box depicts the median concentration while the height of the box represents the 25 th and 75th percentile of the concentration distribution. Vertical lines extending above and below the box represent the maximum and minimum concentration values for the laboratory. Open circles show the observed concentration.
(TIF)
S1 Table
ELISA concentration of each test sample—Laboratory A.
(XLSX)
S2 Table
ELISA concentration of each test sample—Laboratory B1.
(XLSX)
S3 Table
ELISA concentration of each test sample—Laboratory B2.
(XLSX)
S4 Table
ELISA concentration of each test sample—Laboratory C.
(XLSX)
S5 Table
ELISA concentration of each test sample—Laboratory D.
(XLSX)
S6 Table
ELISA concentration of each test sample—Laboratory E.
(XLSX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
There is a clear link between the combined activity of neurons and specific neural computations [1] , [2] . A common observation from population recordings is that the correlation between the activities of pairs of neurons can be modulated – for instance, by the spatiotemporal structure of stimuli [3] , [4] , the perceptual state of the subject [5] , [6] , or the spatial focus of attention [7] – [9] . Theoretical work has focused on the cellular and circuit mechanisms that both determine and modulate correlation [10] – [20] . However, the general applicability of these theories is unclear [21] , and how neural populations modulate the correlation between their spiking activity remains an open question.
One complication is that spike train correlations reflect common activity that may be measured at different timescales, ranging from a few (synchrony) to hundreds of milliseconds (co-variation of firing rates). For example, pairs of neurons in visual cortex [22] , [23] , olfactory bulb [24] , and attention responsive cortical areas [7] – [9] show increases in spike time synchrony which accompany simultaneous decreases of rate co-variation. To indicate the complex temporal aspects of this modulation, we label a differential change in correlation over distinct timescales correlation shaping [19] , [24] . In this study, we use a combination of in vivo recordings and computational modeling of electrosensory neurons to study how the spatial structure of a stimulus shapes the correlation of primary sensory neurons.
Weakly electric fish detect perturbations of their self-generated electric field through an array of electroreceptor neurons scattered on their skin surface which synapse onto pyramidal neurons within the electrosensory lateral line lobe (ELL) [25] . Relevant stimuli can be broadly categorized as either local , stimulating only a small fraction of the skin, or global , projecting to a broad area of the animal's body. Local inputs are a reasonable approximation to the spatial scale of prey inputs, while global inputs mimic communication calls from conspecifics [26] . We recorded simultaneously from pairs of ELL pyramidal neurons and found that global inputs increased spike train correlations at short timescales while simultaneously decreasing correlations at long timescales, when compared to the spike train correlation induced by local inputs. While there is a general understanding about how local and global stimuli control single neuron responses [26] – [30] , the cellular and circuit mechanisms that allow the spatial extent of stimuli to shape correlated population activity in the electrosensory system are a new area of study.
Based on the well-characterized anatomy and physiology of electrosensory circuits [25] , we developed a spiking network model of ELL pyramidal neurons that captured the experimentally observed correlation shaping. Diffuse inhibitory feedback was activated preferentially by global stimuli and provided a decorrelating signal that reduced correlations at long timescales. Further, global stimuli recruited feedforward circuitry that increased correlations at short timescales which were immune to feedback decorrelation. For sufficiently weak stimuli, we use a linear response framework [28] , [31] to show how correlation shaping is consistent with a shaping of the single neuron stimulus-response gain function. We tested our model predictions experimentally by selectively blocking feedback input, causing spike train correlations at long timescales to increase, rather than decrease. This directly demonstrates how inhibition can be a source of decorrelation to pyramidal neurons, rather than a source of synchrony as described in many previous studies [10] , [11] , [32] – [35] . Finally, we used our understanding of the population's response properties to study how feedback selectively attenuates responses to distractor stimuli, improving the system's ability to represent relevant signals. In total, our results reveal novel principles by which feedforward and feedback neural circuits are differentially activated by stimuli to shape population spike train correlations.
Methods
Ethics Statement
Animals were obtained from local importers and were acclimated to the laboratory as per published guidelines [36] . All experimental procedures were approved by the McGill University Animal Care Committee and have been described in detail elsewhere [37] .
Electrophysiology
Briefly, dual extracellular recordings from the lateral and centrolateral ELL segments of Apteronotus leptorhynchus were made using metal-filled micropipettes [37] . Pyramidal cells within these segments can be distinguished from cells within the centromedial segment based on recording depth, the medio-lateral and rostro-caudal positions of the recording electrode with respect to surface landmarks such as the “T0” vein and its afferent veins [38] , and their responses to sensory input as previously described [39] . Superficial pyramidal cells were identified based on their low ( ) whereas deep cells were identified based on their high ( ) mean firing rates in the absence of EOD modulations [26] , [30] , [40] . All data was sampled at 10 kHz.
Random amplitude modulations of the animal's electric organ discharge (EOD) consisting of white noise low-pass filtered with a cutoff of 120 Hz (8th order Butterworth filter) were presented either globally via two electrodes positioned on either side of the animal or through a dipole located close to the skin surface [37] . The stimulus lasted and consisted of 6 concatenated segments of the same frozen noise epoch that lasted 20 s [37] .
Pharmacological blockade of the indirect feedback from EGp was performed by either applying the non-NMDA glutamate receptor antagonist CNQX within the ELL molecular layer [30] or by applying a 2% lidocaine solution to the praeminential-cerebellar tract (PECB) as done previously [41] . Since both manipulations gave rise to similar effects, the data was pooled.
Data Analysis
Spike train cross-covariance functions
The recorded signals from a pair of neurons in response to the stimulus were thresholded in order to obtain the spike times , where is the number of spikes from neuron ( ). The spike train from neuron is then given by: (1) Here is the discrete approximation of the Dirac delta function with if and is zero otherwise; throughout so that at most one spike was contained in any time window. We note that this is equivalent to discretizing time in bins of width ms and setting the content of bin to when there is a spike time such that and to otherwise, as was done previously [30] .
The firing rate for neuron is then estimated as: (2) where is the duration of a recording (typically 120 s). The spike train covariance at time lag between neurons and is defined as: (3) where the number of time bins in the discrete spike train is . We refer to as the auto-covariance, while for is called the cross-covariance.
Spike count correlations
We also considered the correlations between the spike counts of pairs of neurons. The spike count from neuron is simply defined as the number of spikes occurring in the time window . It is a random integer given by: (4) For a given window size , we computed a sequence of spike counts from neuron as , using overlapping windows to increase the number of estimates. We have that , where denotes the mean value of the sequence . We can also obtain second order statistics from including the spike count variance and co-variance, which are defined by: (5) (6)
From these one can define the correlation coefficient between the spike counts and over a time window : (7) We use to denote the average value of across all pairs and similarly for other pairwise statistics. For small , the correlation coefficient measures the degree of synchrony between the two trains, while, for large , measures the co-variation in the firing rates of a pair of neurons [12] , [13] .
The variance and covariance functions of the spike count and spike train are related by: (8) These equations are the well known relations between second order spike count and spike train statistics [42] , with resulting from the convolution of the windowing function that converts spike trains to spike counts.
Within-trial vs. across-trial covariance functions and correlation coefficients
We note that both the spike train covariance function and correlation coefficient are within-trial measures of co-variability, since they incorporate both signal induced as well as trial-to-trial variable (i.e noise) aspects of common input fluctuations. Since we presented the same (i.e frozen) realization of the signal six times in succession, we were able to compute the spike train covariance and spike count correlation that were due purely to the common signal by computing joint statistics from neuron pairs recorded in different trials (i.e. across-trial). Specifically, denote the spike train of neuron in response to the realization of the stimulus ( ) by . The across-trial spike train covariance between neurons and is then given by: (9) In Eq. (9) , . Eq. (9) measures the joint spike statistics from neuron pairs when the spike trains were not recorded simultaneously but were stimulated with the same signal. This is because the summation runs over all possibly non-repeating combinations ( ) of the responses of each neuron to the six presentations of the frozen stimulus.
Similarly, one can define the spike count sequence for neuron during stimulus realization as . The across-trial spike count correlation coefficient between neurons and is then given by: (10) where Cov with the sequence of spike counts from the realization of the stimulus.
Linear Response Approximation
We use linear response theory in order to derive an expression for the correlation coefficient in terms of the stimulus gain, as done in past studies [12] – [14] , [19] , [28] , [31] , [43] , [44] . We consider the Fourier transform of the spike train covariance function as the length of the trial becomes large and assuming the processes are stationary: (11) Throughout, we will refer to with as the cross spectrum and as the power spectrum. To relate spike count statistics to spike train statistics, we use the Wiener-Khinchin theorem to rewrite Eq. (8) (assuming is large): (12) (13) with . Note that approaches a -function centered at 0 as and a constant function on as . Therefore, for large , only the zero-frequency components of the spectra contribute to the integral, while for small , all frequencies contribute. A similar relation holds between and .
For a fixed stimulus , we assume that [13] , [28] , [31] , [43] : (14) where is the Fourier transform of the mean-subtracted spike train given a particular realization of , is the Fourier transform of the stimulus, and denotes an expectation over repeated presentations of the stimulus. is the single neuron stimulus-response gain of the neuron (which we refer to as the stimulus gain for brevity). It relates the amplitude of the response to that of a signal at a particular frequency. For both experimental data and numerical simulations, we compute as: (15) where is the cross spectrum between and and is the power spectrum of the signal.
Assuming that the spike trains are conditionally independent given the stimulus, we can write , where denotes an expectation over the random stimulus. Substituting Eq. (15) into Eq. (14) , (16) Finally, combining Eqs. (13) and (16) yields the following approximation: (17) Eq. (17) relates the joint spike count variability to the stimulus gain , and has been derived in several past studies [13] , [19] . We can then approximate the predicted across-trial correlation as: (18)
Modeling
ELL anatomy
The neuroanatomy and physiology of the electrosensory system have been extensively characterized [25] . Pyramidal neurons in the ELL are subdivided according to several criteria. Roughly half of all pyramidal neurons have a basilar dendritic tree (BP neurons) and receive direct electrosensory afferent input. The other half lack a basal dendrite (nBP neurons) and receive afferent input only indirectly via interneurons [45] . Both BP and nBP neurons have an apical dendritic arbor; however, the extent of the arbor is variable across neurons. Pyramidal neurons with small apical dendritic trees are called deep neurons and do not receive much feedback input [30] , [45] , [46] . In contrast, pyramidal neurons with large apical dendritic trees are called superficial neurons and receive large amounts of feedback [30] , [45] , [46] . It has been recently shown [45] that the spatial projection of electroreceptor input to individual pyramidal neurons establishes a putative column, composed of BP and nBP deep and superficial pyramidal neurons.
The afferent and efferent projections between the ELL and higher brain structures further distinguish ELL pyramidal neurons. Indeed, only deep pyramidal neurons project to the praeminentialis dorsalis (Pd) [46] , a second order isthmic structure that directly projects to the posterior eminentia granularis (EGp), which in turn projects back to the ELL along the dorsal molecular layer via parallel fibers [25] that make synaptic contact onto the large apical dendritic trees of superficial pyramidal neurons. Thus, the deep ELL EGp superficial ELL feedback pathway can be characterized as open-loop [46] . Electrophysiological studies suggests that EGp granule cells show temporal locking to electrosensory input [46] , [47] and that the indirect feedback input onto ELL pyramidal neurons is in the form of a negative image of the stimulus that is activated by spatially diffuse but not by spatial localized stimuli [30] , [46] .
ELL model description
Our model of the deep pyramidal neuron to superficial ELL feedback via the nP and EGp contained three distinct neural populations: a deep (Dp) ELL population that projected to a population of granule cells in the EGp, which in turn provided feedback to a population of ELL superficial (Sf) neurons. All cells were modeled with leaky integrate-and-fire (LIF) dynamics [48] . Numerical values of model parameters can be found in Table 1 , and a detailed model summary [49] can be found in Table S1 . The membrane potential obeyed linear subthreshold dynamics supplemented with a spike-reset rule so that implied that , and was marked as a spike time. The deep population consisted of neurons, and the membrane potential of the deep neuron obeyed: (19) The first two terms of the right hand side of Eq. (19) model a static rest state and an intrinsic leak process, respectively. The process models Gaussian stimulus locked electroceptor activity, while models stimulus independent activity afferent to neuron in population ( ). As in the experiments, we set , but the temporal structure of the processes was white with , , and for or . The electroreceptor input contrast was set by and the correlation of the stimulus locked component by .
10.1371/journal.pcbi.1002667.t001 Table 1
Parameter values used in numerical simulations.
Parameter
Description
Value
Number of deep neurons
800
Number of EGp neurons
200
Number of superficial neurons
2
Deep membrane time constant
10 ms
EGp membrane time constant
10 ms
Superficial membrane time constant
15 ms
Deep bias
−56 mV
EGp bias
−60 mV
Superficial bias
−56 mV
Threshold voltage
−55 mV
Reset voltage
−65 mV
Noise strength
1 mV
Deep to EGp synaptic strength
mV
EGp to Superficial synaptic strength
mV
EGp to Superficial synaptic time constant
5 ms
Local input correlation
0.1
Global input correlation
0.2
The EGp population consisted of neurons, and the membrane potential of the EGp granule cell followed: (20) Here is the spike train from the deep neuron, and is the strength of excitation from the Deep ELL EGp. The time constant was chosen as 10 ms, based on recent measurements of input resistance for these cells of approximately 2 G [47] and data from cerebellar granule cells indicating typical capacitance values of 3–5 pF [50] – [52] .
Finally, since we are only interested in the pairwise correlation between superficial neurons and because the feedback is open-loop, it is only necessary to consider a pair of superficial pyramidal neurons. As such, we set . The superficial pyramidal cell's membrane dynamics are given by: (21) Here where is the Heaviside function. The operation denotes convolution. The inhibitory coupling from EGp to the ELL was set by .
During local stimulation, a fraction of deep neurons received coherent, stimulus-locked electroreceptor input ( ), while all other deep neurons received uncorrelated input modeling spontaneous afferent activity. During global stimulation, all deep neurons ( ) received stimulus-locked input ( ). The increased value of reflects the fact that global stimuli will spatially saturate the receptive field center and will thus more effectively drive the afferent population [29] , [53] .
In our model, a pair of neurons in a given layer could receive correlated input from the previous layer in two ways. First, a neuron in the previous layer could project to both downstream neurons and thus correlate their input. Second, neurons in the previous layer could become locked to the stimulus and their pooled activity could correlate the downstream neurons, even if their projections did not overlap anatomically. In the linear model, we assumed that the first source of common input is negligible relative to common input from stimulus locked, pooled activity, as is often the case in feedforward networks [54] . Consequently, correlations between model neurons were due only to external signals that synchronously recruited electroreceptors. Therefore, for the model.
To evaluate for our model using the linear response approximation, we computed the superficial neuron stimulus gain . For numerical simulations, we estimated using Eq. (15) . However, following past work [28] , [31] , we derived a theoretical approach to compute . For global stimulation and assuming that both the input correlations and the effective coupling and are sufficiently small, we compute the feedback filter from the Deep ELL EGp Superficial ELL using the serial computation (22) where is the Fourier transform of the exponential synaptic kernel . This result follows simply from the linear convolution of Deep ELL activity to EGp and then from EGp activity to superficial ELL through . Here we have introduced , the single neuron cellular response function (which we refer to as the cellular response for brevity) that measures a neuron's response to an applied current, independent of network feedback. can be computed using standard techniques from statistical mechanics (see Text S1 ).
We note that can be calculated for mixed excitatory and inhibitory feedback to superficial neurons. In this case, the value of should be interpreted as the effective input strength from both excitatory and inhibitory populations. For example, if the fraction of excitatory synapses from EGp to superficial neurons is given by and the synaptic strength of excitation and inhibition are and , respectively, then we have . Previous studies have established that the stimulus-locked EGp feedback is net inhibitory [46] , and we therefore model the pathway as purely inhibitory for simplicity.
With , we calculate the stimulus gain of a superficial ELL neuron as given in Eq. (25) . Further, these techniques also permit a calculation for the power spectrum . With theoretical expressions for and , and assuming the signal is Gaussian white noise with unit variance, we use Eqs. (17) and (18) to obtain a theoretical prediction for the spike count correlation between the two superficial ELL neuron spike trains: (23) Here we have used the homogeneity of the spike trains to set and for all superficial neurons.
Results
Correlation Shaping with Global and Local Stimuli
We examined the response of ELL pyramidal neurons to time-varying electrosensory input. Broadband electrosensory stimuli (Gaussian, 0–120 Hz) were applied to awake, behaving weakly electric fish ( Apteronotus leptorhynchus ; see Methods). Throughout the study, we delivered stimuli in one of two spatial arrangements: a local or global configuration [26] , [27] , [29] . In the local configuration, stimuli were spatially compact, delivered through a small dipole (tip spacing of 2 mm), and excited only a small region of the skin surface ( Figure 1A , left, blue). Local inputs mimic prey stimuli which drive only a spatially localized portion of the receptive field of an ELL pyramidal neuron [55] . In the global configuration, stimuli were spatially broad, delivered through a pair of electrodes located on each side of the animal, and affected the entire surface of the animal ( Figure 1A , left, orange). Global inputs mimic stimuli caused by conspecifics which drive nearly the entire surface of one side of the animal, stimulating both the classical and non-classical receptive field of a target pyramidal neuron [29] , [56] . During both local and global stimulation, simultaneous extracellular recordings of ELL pyramidal neuron pairs were collected ( Figure 1A , right). There was an intentional selection bias for superficial basilar pyramidal (BP) neurons [25] , since these neurons are known to receive feedback projections that shape their responses to sensory input [30] , [45] , [46] . Superficial neuron firing rates in the local and global configurations were similar ( and respectively).
10.1371/journal.pcbi.1002667.g001
Figure 1
The spatial extent of electrosensory stimuli shapes the temporal correlation between the spike times from pairs of ELL pyramidal neurons.
A , Stimulus protocol for local and global stimulation. Left: Gaussian distributed electric field stimuli with broadband spectral content (uniform over 0–120 Hz) were applied in a spatially compact (local) or diffuse (global) manner. Right: Paired extracellular recordings of ELL pyramidal neurons were made during stimulation. B1 , Spike train cross-covariance function in the local and global stimulus configuration for pairs of simultaneously recorded superficial BP neurons (within-trial correlation). Correlation function is normalized by firing rate. B2 , Same as B1 except computed between spike trains recorded during distinct trials. C1 , Within-trial spike count correlation as a function of window length ( ) in the local and global stimulus configuration. C2 , Across-trial spike count correlation as a function of window length in the local and global stimulus configuration. D1 , Ratio of global and local within-trial spike count correlations shown in panel C1. D2 , Ratio of across-trial global and local spike count correlations shown in panel C2. The data set consists of n = 10 pairs of neurons, and all curves are population average quantities. In all panels, shaded regions denote standard error.
We used the simultaneous unit recordings to estimate the spike train cross-covariance function (see Methods Eq. 3 ) for neuron pairs in both the local and global stimulus configurations. Global stimulation set a narrow peak of the cross-covariance function with a high maximum at zero lag, while it was broad with a lower peak value for local stimulation ( Figure 1B1 ), consistent with previous reports [37] .
To quantify this shift in covariance at different timescales, we computed the correlation coeffcient between the spike counts of neuron pairs' outputs [22] , [42] . This provided a normalized measure of the similarity between the two spike trains as observed over windows over a particular length (see Methods Eq. 7 ). At small window sizes ( ), spike count correlation was larger during global stimulation than during local. For large window sizes ( ), this relationship was reversed ( Figure 1C1 ). Correlation is generally a rising function of window size [57] , since for small few spikes will occur in the same window. However, even small values of correlation (e.g. in magnitude) have substantial influence on the propagation of neural information [54] , [58] and neural coding [59] . To provide a relative measure of the shift in correlation between the two states, we considered the ratio of global correlation to local correlation. This was a decreasing function of window size which was substantially greater than 1 for small window sizes and lower than 1 for large window sizes ( Figure 1D1 ).
We performed statistical tests to confirm that the trends observed were significant. Nonparametric tests confirmed that the distributions for the local and global conditions were different ( , evaluated at , , two-sample Kolmogorov-Smirnov test). The trends with timescale were also significant ( , compared with , for local and for global stimulation, two-sample Kolmogorov-Smirnov tests). The means of the distributions were also different ( , evaluated at , , paired t-test). In summary, the spatial extent of the electrosensory signal shaped the timescales over which spike train pairs were correlated.
Shifts in Single-Neuron Response Gain Predict Correlation Shaping
In general, correlated neural activity can be decomposed into stimulus induced and non-stimulus induced components [21] , [60] . Stimulus induced correlations reflect the two neurons locking to a dynamic stimulus, while the non-stimulus induced correlations reflect the neurons sharing a portion of their trial-variable noise, presumably from a common pre-synaptic source. To uncover the cellular and circuit mechanisms underlying correlation shaping, we first determined whether the changes in correlation observed were present across trials and therefore related to how neurons responded to the repeated stimulus. Using spike trains across different trials of identical stimulus presentations, we computed the across-trial spike train cross-covariance functions and spike count correlations ( Figure 1B2,C2 ; see Methods Eqs. 9 , 10 ). The magnitude of these across-trial correlations was less than that of the within-trial correlations, indicating the presence of some trial-variable noise (compare Figure 1C1 and 1C2 ). Nevertheless, the differential shaping of correlations at short and long timescales was still present in the across-trial spike count correlation ( Figure 1C2,D2 ). This suggests that the way stimulus processing shifts between local and global conditions is related to the mechanisms responsible for correlation shaping.
To investigate this relationship, we considered the stimulus gain , which measures a neuron's response to an external electrosensory stimulus at frequency ( Figure 2A , see Methods Eq. 15 ). We computed the gain in the two stimulus conditions and found that during local stimulation, the gain function was low-pass, while during global stimulation, it was high-pass ( Figure 2B ), consistent with previous studies [29] , [30] . We then asked if the observed changes in correlation could be related to this shift in frequency selectivity.
10.1371/journal.pcbi.1002667.g002
Figure 2
Shifts in stimulus gain predict spike train correlation shaping.
A , Schematic illustration of stimulus gain. The gain is described as the ratio of the change in the output firing rate of a neuron that is evoked by an input sine wave stimulus of amplitude . B , Gain for neuron pairs during local and global stimulation. The signal was assumed to have unit amplitude. C , Across-trial spike count covariance (solid) and the prediction from a linear response theory (dashed, see Methods Eq. 17 ), in both global and local stimulus conditions. The data set consists of n = 10 pairs of neurons, and all curves are population averages. In all panels, shaded regions denote standard error.
Motivated by past studies [12] , [13] we assumed that the cross-spectrum between the spike trains was proportional to the product of their stimulus gain functions (see Methods Eq. 16 ). This amounts to assuming that the common stimulus is the only source of correlation in the neural responses. This theory predicts that the correlation for large window sizes is determined by stimulus gain at low frequencies. In contrast, correlation for small windows involves gain at high frequencies. The shift in from low frequency transfer for local inputs to high frequency transfer for global inputs therefore implies global stimulus correlation will be enhanced for small and attenuated for large , with the inverse true for local stimulation. We verified this hypothesis, obtaining a prediction of the spike count correlation in the two states that matched the experimental data (see Methods Eq. 18 ; Figure 2C , solid versus dashed curves). Thus, the shift in the frequency-selectivity of superficial neurons' stimulus gain between the local and global conditions indeed predicted the changes in correlation.
Modeling ELL Pyramidal Cell Responses
To understand mechanisms behind the shift in neuronal responses under the local and global stimulus conditions, we constructed a simplified population model of ELL pyramidal neurons based on known anatomical and functional data as well as our experimental results ( Figure 3A ; for a detailed discussion of the anatomy, see Methods). This model captured two generic circuit features that modulated population responses: feedforward sensory input and feedback inhibition. All pyramidal neurons received feedforward electrosensory input via electroreceptors, but were divided into two classes based on their feedback afferents: deep neurons did not receive feedback from higher regions, but superficial neurons did receive inhibitory feedback. This feedback arrived from the posterior eminentia granularis (EGp), which was in turn innervated by the deep neurons. In total, this structure formed an open-loop inhibitory feedback pathway, from deep neurons to EGp neurons to superficial neurons. Motivated by past studies, ELL pyramidal neurons were modeled as simple leaky integrate-and-fire units [27] , [28] , [46] . Consistent with experimental data [30] , superficial firing rates in the model were lower than deep firing rates (12 Hz and 36 Hz, respectively) in both local and global stimulation conditions.
10.1371/journal.pcbi.1002667.g003
Figure 3
Open loop feedback inhibition in electrosensory neural circuitry.
A , Detailed schematic of peripheral neural circuitry in the electrosensory system. Basilar (BP) and non-basilar (nBP) pyramidal neurons in the electrosensory lateral line lobe (ELL) have their somata located in the Pyramidal cell layer (PCL). Deep pyramidal neurons (green) have small apical dendritic arbors, projecting only to the Ventral Molecular Layer (VML). In contrast, superficial pyramidal neurons (red) have large apical dendritic arbors, projecting to the Dorsal Molecular Layer (DML). Pyramidal neurons receive direct and/or indirect input from feedforward electroreceptor afferent input to the Deep Fiber Layer (DFL). Deep pyramidal neurons excite neurons in the praminentialis dorsalis (Pd), which in turn excite granule cells in the posterior eminentia granularis (EGp). The EGp projects parallel fiber feedback along the DML exclusively targeting ELL superficial pyramidal neurons. In total the deep ELL EGp superficial ELL pathway is an open loop feedback structure. Pyramidal neuron graphics were from example neurolucida traced neurons [46] . B , Stimulus correlation for pairs of experimentally recorded deep pyramidal neurons (n = 45 pairs; 10 neurons were used) that were driven by the stimulus in local and global (bottom). Little correlation shaping is present. For comparison purposes we show the stimulus correlation for pairs of superficial neurons (top, Figure 1C2 ). C , Simplified model of the ELL-EGp circuit. Individual neurons in the deep ELL, EGp, and superficial ELL were modeled with leaky integrate-and-fire neuron dynamics (example realizations on right). Electroreceptor input was modeled as white noise, with 5% of deep pyramidal neurons receiving a stimulus-locked component in local and 100% in global. We studied the spike responses the pair of superficial pyramidal neurons (labeled 1 and 2) that receive both afferent and EGp feedback inputs.
Previous studies have shown that EGp feedback modulates both the static [41] and dynamic [30] gain of single neuron responses. However, how it controls the ELL population response, and in particular correlations between pyramidal neurons, is unknown. To determine whether feedback is responsible for stimulus-dependent correlations, we recorded from deep pyramidal neurons receiving a frozen stimulus and computed stimulus correlations between the pairs of spike trains. Consistent with the lack of feedback projections to this subpopulation, these neurons did not show substantial shaping of correlations between the local and global conditions ( Figure 3B , bottom), in contrast with superficial pyramidal neurons ( Figure 3B , top). The small decrease in correlation for large time windows observed during global stimulation for deep neurons ( Figure 3B , bottom) is consistent with these neurons receiving little feedback input [40] .
Recruitment of Feedback in the Model During Local and Global Stimulation
We used our model to examine the stimulus dependence of EGp feedback. In our model, electrosensory stimulation caused the firing of deep pyramidal neurons to become stimulus-locked. When the stimulus was local, only a small fraction of this population was stimulus-locked, so that the average correlation across the deep population was low ( across the population, Figure 4B1 ). The weak stimulus correlation across the deep population failed to recruit coherent activity in the EGp granule cell population, resulting in a near tonic inhibitory feedback to the ELL ( Figure 4C1 ). In contrast, when the stimulus was global, the entire deep population was correlated by the stimulus ( , Figure 4B2 ). This led to a dynamic, stimulus locked EGp feedback to the superficial neuron pair ( Figure 4C2 ). Thus, our model captured a link between the temporal locking of EGp feedback and the spatial extent of the external stimulus, which has been suggested in past experiments [46] , [47] .
10.1371/journal.pcbi.1002667.g004
Figure 4
Model EGp feedback is stimulus locked for global, but not local, stimulation.
A Low-pass (0–60 Hz) filtered version of the electrosensory stimulus. Filtering was done as a visual aid in relating the stimulus to the feedback in (C2). B1 , Raster plot of the deep neuron population during local stimulation. The signal weakly correlated only a small fraction of the population. B2 , Same as (b1), but during global stimulation. The spatially broad stimulus correlated the entire deep population. C1 , EGp feedback current during local stimulation, showing little stimulus locking. C2 , EGp feedback was stimulus-modulated by the global signal, due to recruitment of the deep population by the stimulus. The inhibitory feedback is a negative image of the stimulus (A2).
Having characterized the EGp feedback, we next determined how it shaped the responses of superficial neuron pairs. The total input to a model superficial pyramidal neuron, from both feedforward and feedback sources, is: (24) Here is the strength of the afferent activity to an ELL pyramidal neuron and and are Gaussian white noise processes modeling stimulus locked and unlocked (noise) afferent inputs, respectively. The parameter is the fraction of receptor afferents that are stimulus-locked, which determines the correlation between the electroreceptor input to neuron pairs. The function is the parallel fiber feedback kernel and involves compound processing of the stimulus by the population of deep ELL neurons, the EGp granule cells, and finally the inhibitory feedback pathway from the EGp to the ELL (see Methods Eq. 21 ). Assuming weak stimulus correlations (small ) and weak EGp feedback, we use linear response theory [28] , [31] , to obtain an expression for the stimulus gain of a superficial pyramidal neuron (see Methods): (25) Here is the Fourier transform of the feedback kernel (see Eq. 22 in Methods), and is the cellular response of a superficial neuron, which measures its response to a fluctuating current applied directly to the neuron (see Eq. 8 in Text S1 ). In contrast to the stimulus gain, the cellular response does not depend on network feedback. The parameter is the spatial extent of the stimulus ( ), with modeling the lack of stimulus-coherent EGp feedback for local stimulation, and the full recruitment of EGp feedback for global stimulation ( Figure 4 ). With this model of how shifts between local and global stimulus configurations, we next build a theory for the correlation shaping within the superficial ELL pyramidal neuron population.
Correlation Shaping in the ELL-EGp Network Model
We used our ELL-EGp network model to relate the spatial extent of an electrosensory stimulus and the timescale of the pairwise correlation between spike trains from superficial BP neurons. During local stimulation, pairs of nearby superficial neurons received correlated electroreceptor input ( Figure 3C ). The degree of correlation between the afferent input to the superficial pair was . The EGp feedback did not exhibit a substantial stimulus-locked component ( ) during local stimulation, and hence did not contribute to common fluctuations ( Figure 4C1 ). Thus, the stimulus gain in the local condition, denoted , reduced to: (26) Our theoretical (see Methods) quantitatively matched estimates from simulations of the ELL-EGp network of leaky integrate-and-fire neurons ( Figure 5a , blue curve and blue dots) and qualitatively matched the low-pass nature of obtained from experiments ( Figure 2B , blue). The calculation demonstrates that the gain to local stimuli of superficial pyramidal neurons is primarily determined by the cellular response , suggesting that feedback network dynamics can be ignored.
10.1371/journal.pcbi.1002667.g005
Figure 5
Model ELL-EGp network captures correlation shaping between local and global stimulation.
A , Stimulus gain of superficial BP neurons in the model (compare to Figure 2B ). Our analytical theory (solid) matches the simulation results from the ELL-EGp network (dots). B , Correlation between superficial BP neuron pairs during local and global stimulation of the model (compare to Figure 1C ). Since our theory predicts a linear relationship between output correlation and input correlation, the output is shown in units of input correlation in the local state , which was 0.1 in simulations. C , Idealized schematic illustrating the effect of feedback on shared fluctuations. Left: local inputs fail to recruit EGp feedback via deep population (see Figure 4 ), so common input arises purely through feedforward stimulus drive. Center: Low frequency global input recruits a negative image of the stimulus, which cancels the common input to the pair of superficial pyramidal neurons. Right: The cancellation signal is weak for high frequency global inputs due to the low-pass nature of the feedback. Hence, the common fluctuations are not cancelled.
The lack of network activity for local stimulation ( ), was contrasted with the recruitment of EGp feedback for global stimulation ( ). During global stimulation, we also assumed that the receptive fields of neurons were fully saturated, rather than being partially driven due to the limited extent of the stimulus, as suggested by experimental estimates [53] . We therefore increased the correlation of electroreceptor afferents in the global state, so that . Combining these two model assumptions, we expressed the gain in the global configuration, , as: (27) If – that is, if the negative feedback were a perfect replica of the feedforward signal – the stimulus gain would be zero, indicating complete stimulus cancellation by the feedback pathway. However, since the negative feedback was low-pass due to neuronal integration and synaptic filtering along the feedback pathway, only the low frequency components of the gain were strongly attenuated. Consequently, for sufficiently low frequencies ( Figure 5A , compare orange and blue curves for ). However, for high frequencies ( Figure 5A , compare orange and blue curves for ), because of the increase in receptive field saturation ( ). Our theoretical matched simulations of the ELL-EGp network ( Figure 5A , orange curve and orange dots). Thus, the combination of feedback recruitment and feedforward saturation during global stimulation captured the experimentally determined shift in stimulus gain known to occur between local and global stimulation ( Figure 2B and see [29] , [30] ).
Next, we examined how this gain shift controlled correlations across the population of superficial pyramidal neurons. Using the linear response theory we used to predict signal correlations in the experimental data ( Figure 2 , see Methods Eq. 23 ), we calculated theoretically the correlations between model pyramidal neurons. Global stimulation simultaneously increased short correlation and decreased long correlation compared to local stimulation ( Figure 5B ). These findings matched the experimental results (compare Figures 1C and 5B ) and are the primary theoretical result of this study.
Our model provides clear intuition for how the combination of receptive field saturation and the recruitment of EGp feedback during global stimulation shapes the correlation of ELL pyramidal neuron activity ( Figure 5C ). During local stimulation, EGp feedback was not recruited and the feedback did not cancel the feedforward signal from the electroreceptors ( Figure 5C , left). This case is contrasted with global stimulation, in which a broad stimulus-induced synchronization of all of the deep ELL neurons recruited a stimulus-locked EGp feedback. This feedback was low-pass, and therefore canceled the low frequency components of the signal ( Figure 5C , middle), but not the high frequency components ( Figure 5C , right). Thus, correlations due to global stimulation were canceled only for sufficiently long timescales ( Figure 5B , ). Furthermore, the saturation of the receptive field input ( ) enhanced the correlation for small ( Figure 5B , ). In total, feedforward and feedback circuitry shaped depending on the spatial profile of the electrosensory signal.
Our ELL-EGp network model distills correlation shaping into two hypotheses that link the spatial properties of an electrosensory stimulus and the timescale of pairwise correlation between the spike responses of ELL superficial pyramidal neurons:
Receptive field saturation for spatially broad signals increases the short timescale correlation between the spike trains from superficial pyramidal neurons.
Recruitment of EGp feedback by spatially broad signals decreases the long timescale correlation between the spike trains from superficial pyramidal neurons.
To study these two components of correlation shaping in isolation from one another, we used a combination of analysis on a subclass of ELL pyramidal neurons and pharmacological blockade of EGp feedback.
Correlation Shaping of nBP Neuron Responses
We first tested how short timescale correlation was affected by receptive field saturation (Hypothesis 1). The ELL has two classes of pyramidal neuron: non-basilar pyramidal (nBP) and basilar pyramidal (BP) neurons, distinguished by the extent of their basilar dendritic arbor ( Figure 3A ). While BP neurons respond to positive deflections of the electric field, nBP neurons are oppositely tuned, due to their afferent inputs arriving solely via an inhibitory interneuron population [25] . This difference in the feedforward afferent architecture to nBP neurons compared to BP neurons produces nBP neuron classical receptive fields that are smaller than those of BP neurons [26] . Despite the difference in feedforward afferent input for BP and nBP neurons, both superficial BP and nBP neurons receive near equivalent feedback from EGp parallel fibers ( Figure 3A ). Thus, a comparison between BP and nBP neurons is sensitive to a difference in feedforward afferent drive, and not to EGp feedback. We hypothesized that global inputs would not drive nBP neurons as strongly as BP neurons because of their smaller classical receptive fields. Hence, short timescale correlation during global stimulation for nBP neurons should be less than for BP neurons.
We first calculated the stimulus gain for nBP neurons. The difference in gain between local and global stimuli for nBP neurons was different than that for BP neurons ( Figure 6A1 ; [30] ). In particular, while nBP and BP neurons both exhibited a reduction in low frequency gain during global stimulation, nBP neurons exhibited little enhancement of high frequency response. Our model network replicated this difference ( Figure 6A2 ) when we assumed that the nBP neurons integrate stimuli over smaller regions of space, such that local inputs saturate the receptive field ( ), in contrast to the BP neuron case ( ). The lack of high frequency shaping of gain for nBP neurons across local and global configurations predicts that the small correlations do not substantially increase in the global state, while EGp feedback still attenuates low frequency gain and hence large correlations. Measurements of for nBP neurons in both the ELL-EGp model ( Figure 6B2 ), as well as nBP neurons recorded in vivo ( Figure 6B1 ) supported this prediction. Thus, the known differences between the receptive field sizes of nBP and BP neurons, provide evidence for the link between the spatial extent of electrosensory stimuli and short timescale correlation observed for superficial BP neurons.
10.1371/journal.pcbi.1002667.g006
Figure 6
Saturation of the receptive field for both local and global stimuli makes short timescale response insensitive to the spatial extent of electrosensory stimuli.
A1 , Experimental stimulus gain for nBP neurons (n = 14) in local and global stimulus configurations. The gain for BP neurons in the global configuration is shown for comparison (see Figure 2B ). A2 , Stimulus gain for model nBP neurons ( ) in local and global configurations, and the model BP neurons ( ) in global for comparison. B1 , Recorded spike count correlation over windows of length for pairs of nBP neurons. As with BP neuron pairs, firing rates in the local and global states were similar ( and , respectively). B2 , Spike count correlation for pairs of model nBP superficial neurons in the ELL-EGp network. For the model results (A2,B2) our analytical theory (solid) matches the simulation results from the ELL-EGp network (dots). Values are shown in units of input correlation in the local state .
Feedback Inhibition Cancels Long Timescale Correlations
We next tested how long timescale correlation is affected by recruitment of EGp feedback by global stimuli (Hypothesis 2). In our model, the EGp feedback was responsible for the decrease in low frequency stimulus gain and long timescale correlation in the global state. To experimentally confirm that this pathway was responsible for these effects, we pharmacologically blocked feedback from EGp to superficial neuron pairs (see Methods). We first tested whether attenuation of low frequency components of the stimulus gain was removed by the block. In experiments with global stimulation, we found that firing rates during the block were decreased significantly from the control condition (block: ; control: , , paired t-test). We remark that while the net impact of EGp feedback may be excitatory, the signal locked components of EGp feedback are thought to be inhibitory [46] , consistent with our model. To correct for the change in firing rates across control and block conditions, we normalized the gain by firing rate to show the relative modulation of firing rate by the stimulus. The normalized gain increased at low frequencies, yet remained unchanged at high frequencies ( Figure 7B1 , compare orange and gray curves), consistent with model predictions ( Figure 7B2 ). This effect was removed after a washout of the drug ( Figure 7B1 , compare orange and light orange curves).
10.1371/journal.pcbi.1002667.g007
Figure 7
EGp feedback reduces correlations on long timescales when stimuli are global.
A Schematic indicating block of feedback with CNQX in the ELL circuit. B1 , Stimulus gain for individually recorded superficial BP neurons in control, block, and recovered conditions. Gain is normalized to output firing rate in the data. B2 , Stimulus gain for model superficial neurons for global stimuli when feedback was intact or absent. C1 , Left: Spike count correlation at and 200 ms for paired recordings of superficial BP neurons. Right: Spike count correlation as a function of for individual recordings with a frozen stimulus in control, block, and recovered conditions. The standard error bars overlap for both the pre-drug and recovery curves, while they do not overlap with those for the block. Differences between control and recovered conditions could be due to incomplete drug washout or the preparation being in different states before and after the application of the drug. C2 , Spike count correlation as a function of for model neuron pairs when feedback was intact or absent. Values are shown in units of input correlation in the local state .
The spike count correlations for simultaneously recorded superficial neurons in the global state with and without pharmacological block of feedback verified its role in shaping long timescale correlations. Specifically, the spike count correlations for showed a significant increase during the block ( , paired t-test), while correlations for were similar ( Figure 7C1 ; left). Due to the difficulty in obtaining paired recordings under pharmacological blockade, we further verified our theory with units recorded individually with frozen noise in the global state with and without pharmacological block of EGp feedback ( Figure 7C1 ; right). Correlations at long timescales were increased during the block compared to control ( Figure 7C1 ; left, compare orange and gray curves) and recovered to control values after drug washout ( Figure 7C1 ; left, compare orange and light orange curves), consistent with our model ( Figure 7C2 ). Thus, despite EGp feedback being a source of common synaptic input to a pair of superficial ELL pyramidal neurons, removing it during global stimulation increased the spike correlation between the neuron pair. In total, these data supported our second hypothesis: stimuli with large spatial extent recruit inhibitory feedback that cancels the input correlation expected from feedforward afferent projections.
Correlation Shaping and Population Coding of Natural Electrosensory Scenes
We have presented a general mechanism for how spike train correlations from pairs of ELL pyramidal neurons are shaped by the spatial extent of an electrosensory signal. We explored the mechanism with simple noise signals categorized into either spatially local or global inputs. However, natural electrosensory scenes are complex, with a broad range of spatial and temporal scales. In this section, we speculate on how correlation shaping influences the population representation of natural electrosensory scenes.
Sensory systems must produce high fidelity representations of biologically relevant signals, while ensuring that distractor inputs do not contaminate the neural code. The ELL pyramidal neuron population is responsible for coding two distinct electrosensory inputs. First, electric fish routinely perform prey detection, tracking, and capture, during which prey organisms produce electric images with low frequency components ( ) that stimulate a limited portion of the animal's electroreceptive field [55] . Second, electric fish use their electric organ to communicate with conspecifics, using signals that contain primarily high frequency components ( ) and drive a large region of the skin [56] , [61] . However, these two signals often coexist with distractor inputs that the electrosensory system must ignore. Natural distractors arise from the superposition of background electric fields from many neighboring fish [62] , or self generated signals from body and tail positioning [47] . These inputs consist of mostly low to mid range frequencies ( ) and drive a broad sensory area. A critical sensory computation in the ELL is the pyramidal neuron population faithfully locking to prey and communication signals, with minimal locking to distractor electrosensory inputs. The linear response analysis of the ELL-EGp network suggests that EGp feedback to the ELL plays an important role in this computation.
Using our linear theory, we calculated the response of a population of superficial BP neurons to mixed signal and distractor input, with and without EGp feedback. The signal was either a local 4 Hz sine wave ( Figure 8A1–D1 ), or a 50 Hz global sine wave ( Figure 8A2–D2 ). In both cases, the distractor input was 0–10 Hz broadband noise. The population response was modulated by the signal and the distractor, with relative strengths determined by the corresponding gain ( Figure 8D ). To test how EGp feedback affects the coding of relevant signals, we computed the signal to noise ratio (SNR) of this population response, defined as the ratio of the signal power integrated over all frequencies to the distractor power integrated over all frequencies. For both the 4 Hz local and 50 Hz global signals, the SNR was greater with feedback than without feedback ( Figure 8B,C . SNR decreased from 2.3 to 0.70 for the 4 Hz local signal and from 2.8 to 0.70 for the 50 Hz global signal when feedback was removed). This is because EGp feedback was recruited by distractor input, attenuating any distractor induced correlation (low gain for distractor inputs in Figure 8D1,D2 ). In contrast, prey inputs lacked sufficient spatial power to recruit EGp feedback, meaning an EGp cancellation signal was not passed and ELL population stimulus gain was high ( Figure 8D1 ). Communication calls have large spatial power, yet their high frequency power cannot be transmitted by the low pass parallel fiber pathway, again meaning ELL population stimulus gain was high ( Figure 8D2 ). The ELL-EGp network was therefore capable of removing spurious correlations due to distractors while still coding for relevant signals.
10.1371/journal.pcbi.1002667.g008
Figure 8
EGp feedback cancels the ELL population response to global distractor inputs but not prey or communication signals.
A1 , Schematic of response to a prey signal, which occupies a limited spatial extent and contains power at low frequencies. A2 , Schematic of response to a communication call from a conspecific, which is a global, high frequency signal. B1 , Average population firing rate for ELL neurons responding to a local, 4 Hz signal (red) and the same signal with 0–10 Hz distractor noise (black). The SNR was 2.3. B2 , Same as B1, but with a global, 50 Hz signal. The SNR was 2.8. C1 , Same as B1, but without EGp feedback. The SNR was reduced to 0.70. C2 , Same as B2, but without EGp feedback. The SNR was reduced to 0.70. D1 , ELL pyramidal neuron stimulus gain for local inputs (which do not recruit feedback) and global inputs with and without feedback. The frequency of the signal is marked. Note that because the distractor is a global 0–10 Hz signal, its transfer will be enhanced by the removal of feedback, reducing SNR (compare gray and orange curves). D2 , Same as D1 but with a global, 50 Hz signal. Since the signal is high frequency, its stimulus gain is not substantially affected by feedback.
Discussion
Temporal shaping of correlated spiking activity has been observed in a variety of systems [7] , [9] , [19] , [22] – [24] . We have shown that the spatial extent of an electrosensory signal controls the timescale of correlation between the spiking outputs of principal neurons in the ELL of weakly electric fish. Specifically, an increase in the spatial extent of a signal increased pairwise spike time synchronization, while simultaneously decorrelating long timescale rate co-variations. Using a combination of computational modeling and targeted physiological analysis, we identified that correlation shaping in the ELL is mediated both by an increase in the strength of feedforward afferent drive and the recruitment of a feedback pathway for spatially broad signals. Electric fish offer a neuroethologically inspired functional context for correlation shaping, where it promotes an accurate population representation of relevant signals, even in the presence of distractor inputs. The generic circuit features that support correlation shaping and its use in feature selective population temporal codes suggest that the basic principles exposed here may be at play in other neural systems.
Correlation Shaping with Neural Architecture in the Electrosensory System
There has been extensive investigation of the gain shifts of single ELL pyramidal neurons between local and global stimulus configurations [26] – [30] , [46] . These studies have shown that both feedforward and feedback mechanisms mediated these shifts. Indeed, pharmacological manipulations of descending feedback to the ELL provided strong evidence for its role in controlling gain shifts of single unit response at low frequencies [27] , [29] , [30] , [46] . However, previous studies have shown that local stimuli only excited a fraction of the receptive field center [26] , [29] and that spatial saturation of the receptive field center mediated the gain shifts of single unit response at high frequencies only by recruiting a greater fraction of feedforward afferent input [29] , [30] . This importance of feedback activity prompted network models of the ELL and higher brain regions, and these models captured the sensitivity of single unit dynamics to the spatiotemporal structure of electrosensory stimuli [27] , . However, the models relied on heretofore untested assumptions about the population spike train statistics of ELL pyramidal neurons. In parallel to these single-unit studies, other work presented simultaneous recordings from pairs of ELL pyramidal neurons showing significant stimulus evoked correlation in spike activity [63] , and that the spike train correlation is sensitive to a stimulus' spatiotemporal structure [37] . However, these studies did not attempt to relate the dependence of pairwise statistics on stimulus structure to the extensive ELL single neuron experimental gain and network modeling literature. Our study merges the two avenues of research and shows that pairwise correlation shaping is related to gain shifts, as our linear response treatment of the ELL-EGp network model predicts. Thus, our analysis directly tests the proposed feedback mechanisms for single neuron response shifts [30] .
Previous studies of the ELL have focused on the generation of oscillations due to feedback from area nP to pyramidal neurons (the direct feedback pathway) [27] . Theoretical studies have demonstrated that such oscillations arise from a combination of spatially correlated noise and delayed inhibitory feedback [28] , [31] . Unlike neurons receiving closed-loop inhibitory feedback from nP, the superficial pyramidal neurons modeled in our study lack input from this direct pathway, and hence do not exhibit oscillations. Superficial neurons were excluded from the analysis in [27] and [28] , so that the results of our study concern a cell class that is distinct from these previous studies. This distinction emphasizes the qualitative differences in the dynamics induced by open- and closed-loop feedback pathways.
We used well-characterized anatomical data and pharmacological manipulation to study the network architecture that codes for time-varying electrosensory stimuli. This is in contrast to techniques such as the generalized linear model [64] that statistically determine the spike response and network filters that generate a response to a sensory signal with fixed statistics. Our approach allowed us to study the response of the system in distinct stimulus conditions, with varying input statistics. Further, network coupling suggested clear architectural predictions for the mechanisms behind correlation shaping (hypotheses 1 and 2). These predictions were validated with a combination of the known heterogeneity of ELL feedfoward architecture ( Figure 3 ), and a pharmacological blockade of feedback activity ( Figure 7 ). Organisms exist in environments with ever-changing sensory statistics yet must code these environments, often with a single neural population. Our study shows how neural architecture can help shift the response dynamics of neural populations as signals change to better meet this computational need.
Our results also highlight how architectural differences may lead to differential population activity in different layers. Recently, it has been shown that synchronization between neurons in visual cortex is layer-dependent [65] . Furthermore, the cognitive demands of a task may control the recruitment of feedback and influence spike train correlations [66] . Our results demonstrate that both layer-specific recruitment of feedback and connectivity profiles influence correlated population activity.
Finally, theoretical communities have recently made some progress in understanding how network architecture combines with cellular dynamics to determine the correlation between pairs of cells [44] , [67] , [68] . However, the work is general, and a clear neural motivation to base a concrete example upon is lacking. Our study demonstrates that the electrosensory system offers a prototypical system where cellular dynamics, a clear feedforward/feedback architecture, and a single stimulus feature (spatial extent) interact to shape the temporal structure of pairwise spike train correlation.
Decorrelating with Inhibition
The role of inhibition in neural circuits is a complex topic of study. Inhibition is linked to rhythmic, temporal locking between pairs of pyramidal neurons [32] . On fast timescales, inhibition is often thought to synchronize the activity of pairs of pyramidal neurons in both recurrent [10] , [27] , [33] – [35] and feedforward architectures [11] , [32] . However, on longer timescales, inhibition mediates competitive dynamics between populations of pyramidal neurons, and as such may be a source of anti-correlated activity [24] . Recently, studies of densely coupled cortical networks with balanced excitation and inhibition [17] , [18] and feedforward inhibitory cortical circuits [20] , [69] have provided new insights into the role of inhibitory dynamics. In these studies, fluctuations in correlated excitation to a pair of pyramidal neurons are cancelled by correlated inhibitory dynamics, yielding a roughly asynchronous cortical state. This cancellation of correlation is similar to the one explored in our study responsible for the reduction of correlation for global stimuli. However, our study was motivated by a primarily feedforward sensory architecture in which an external signal can drive correlated activity.
The strengths of the electrosensory preparation allowed us to extend the correlation cancellation mechanism along two important directions. First, the ease in controlling the spatiotemporal properties of external stimuli allowed an analysis of the limitations of correlation cancellation. The diffuse ELL EGp feedforward path restricts correlation cancellation to signals with broad spatial scale, while the slow filtering by the parallel fiber pathway can only cancel correlations of low frequency stimuli. Second, the well segregated parallel fibers that mediate EGp feedback to the ELL permitted a pharmacological blockade of inhibition, directly providing evidence for correlation cancellation. The parallel fibers are a source of common input to pyramidal neurons, and a naive analysis would predict that their removal would thus decrease pyramidal neuron spike train correlation. Nevertheless, the blockade of parallel fiber inputs increased the spike train correlation, suggesting that the common inhibition interacts with the common feedforward afferent input in a destructive, rather than cooperative, manner.
Studies of neural codes often investigate the distinction between signal evoked, across-trial correlations and additional ‘noise’ induced, within-trial correlations [60] . Across-trial correlations are attributable to a dynamic locking of the spike train pairs to the common signal. Within-trial correlations measure the trial-to-trial co-variability of a pair of spike trains and may be increased relative to across-trial correlations due to common synaptic input to the neuron pair. These common fluctuations are often deleterious to cortical population codes [60] , acting as a source of variability that cannot be removed through population averaging. The majority of our study presented simultaneously recorded spike train data which contains across-trial correlation as well as additional within-trial correlation. However, the shaping of correlation by the spatial profile of a stimulus was explained from knowledge of only of the across-trial correlation ( Figures 1 and 2 ), and thus our ELL-EGp network model ignored other sources of correlation entirely. Our analysis did study the effects of irrelevant distractor inputs which can act as a source of noise, though originating from external signals rather than internal circuit mechanisms. We found that low frequency distractors that drive a substantial portion of the network recruit a cancellation signal. We therefore predict that within-trial correlations may be cancelled by a similar mechanism if they drive a large number of neurons synchronously. This may be the case when within-trial correlations are driven by the local field potential, which is often low frequency and widespread to populations of neurons [70] , [71] .
In summary, we have identified the combination of feedforward and feedback architecture that allows the spatial extent of a stimulus to shape the temporal correlations between the spike trains of pairs of electrosensory principal cells. Furthermore, correlation shaping allows populations of neurons to respond to stimuli that match a specific spatiotemporal profile and ignore distractor inputs. The generic architectural features of our network and the fact that sensory systems must filter irrelevant signals suggest that our findings may generalize to other systems.
Supporting Information
Table S1
Model summary.
(PDF)
Text S1
Computation of cellular response function and power spectrum.
(PDF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Foreseeably, past generations of patients have used their physicians as the key source of health related information, however, there is evidence that people are increasingly turning to the Internet to supplement their information needs [ 1 ]. For example, a Swedish study found that just over three-quarters (76.2%) of people diagnosed with cancer accessed the Internet for cancer-related information and more than one quarter used social media relating to their health [ 2 ]. Patients commonly report using webpages, blogs, interactive forums and social media to obtain information to help make informed decisions, find practical information or answers to health related questions, stay in touch with others, and share experiences [ 2 , 3 ].
Social media has become ubiquitous to our lives where we share, connect and communicate our experiences with friends, family, organisations and people otherwise unknown to us. Worldwide, approximately 2.5 billion people use social media and almost two-thirds of American adults use social networking sites: an almost ten-fold increase over the past decade [ 4 ]. The portability of these websites via mobile applications has no-doubt accelerated their uptake and allows for the capture of life’s most ephemeral events. The differences in user demographics that are seen between platforms (such as age, ethnicity, gender, education or income), lend themselves to being targeted for various health campaigns, health promotions or health research seeking to reach different audiences [ 5 , 6 ]. Social media data collection foreseeably provides large-scale and easily accessible data for patient reported information, particularly when compared with traditional patient-focused data collection methods [ 7 ].
One of the most popular social media platforms, Instagram, has almost one billion active monthly users [ 8 ]. Instagram is differentiated from other social media platforms by user-posts’ being dominated by a photo. Most users choose to add accompanying text to their photos as well as tags or labels, termed ‘hashtags’ denoted as # label , (e.g. #cancer). The accompanying hashtags provide a method of grouping photos to create virtual social communities of similarly themed content or purpose and allows users to easily connect and share content.
Acute myeloid leukaemia (AML) is a relatively rare and aggressive blood cancer that can occur at any age [ 9 , 10 ]. The standard treatment is immediate intensive chemotherapy, requiring lengthy hospital stays [ 11 ]. Additionally, research shows most patients have a reduced quality of life and persistent side effects or symptoms even after the completion of therapy or in remission [ 12 – 14 ].
AML makes up less than 1% of all cancer diagnoses per year, making research challenging to accrue participants, particularly in young adulthood where incidence is at its lowest [ 9 , 10 ]. However due to the popularity of Instagram, particularly in early adulthood [ 15 ] and the search functionality of hashtags, the Instagram platform presents an opportunity for proposing unique research questions, particularly those focused on rare-diseases (as with AML), or research with participants that are traditionally difficult to access. Despite the popularity of Instagram, and the unique participant group it can reach, little health research has been undertaking using this platform [ 1 ].
Given large numbers of people with cancer are accessing online health-related messages and the relative absence of Instagram research, this exploratory study will be the first to characterise AML-related content on Instagram; specifically who is posting AML-related content and what types of content are being posted. Characterising AML-related content on social media could be useful for targeting people most likely to benefit from health messages, interventions, or support. Using Instagram for this type of extant research has the potential to provide unique insights into the lived experience, as well as observing individuals providing or receiving support through virtual communities and the sharing of health-related information. Additionally we detail a method potentially of interest to other researchers.
Materials and methods
The methods outlined in this paper adhered to Instagram’s terms of service at the time of the research and all the content analysed in this study was publicly available on Instagram (available at www.instagram.com/instagram ). However the data is not owned by the authors and they do not have permission for reproduction of the data used in the analysis. Ethical approval for the study was granted from Monash University, Human and Research Ethics Committee (Project ID 18540) and included a waiver of consent (no individual consent was necessary from Instagram users). The ethics approval prohibits the publication of data that may inadvertently identify any individual.
Data collection
Instagram is primarily a mobile application but has a desktop website with limited functionality. Only the website accessible version of Instagram was used in this study, to ensure complete separation between the researchers and their private accounts. This also ensured that only publicly available posts were being accessed (no Instagram account or login required). One hundred posts were chosen as a convenient number for a time-consuming manual exploratory method. The posts were found by using Instagram’s search bar at the top of the webpage using hashtags that had been previously scoped as being used by people with AML: #acutemyeloidleukemia, #acutemyeloidleukaemia and #amlsurvivor. Each hashtag was searched for separately.
Posts were excluded from the study if they were videos (we were unable to extract these using our context extraction method), had non-English accompanying text or the subject matter was focused on children’s cancer. Children’s cancer was able to be determined by accompanying hashtag, such as #childhoodcancersucks or though examining the post photo and/or the accompanying text.
This study was undertaken after Instagram removed the automatic application program interface (API), which allowed for automation in downloads and much of the accompanying meta-data. Therefore, we detail a manual method of data extraction that may be of use to other researchers. This manual method allowed for retrospective capture for all eligible posts made over seven consecutive days in February 2019; eight consecutive days in April 2019 and twelve consecutive days in May 2019, to obtain a consecutive sample of 100 posts during the collection periods. Posts in chronological order (as opposed to most popular) can be found by scrolling past the initial “top posts” to the “most recent”. It is was these most recent posts that were accessed taking note of the date of the post to ensure it complied with our sampling time-frame. This sampling method was used to avoid awareness campaigns or trending content, (which may generate atypical Instagram posts and traffic), cultural and ethnic influences between users’ geographical location and for researcher convenience. One hundred posts was deemed to be sufficient given practicality of methods employed and the rarity of AML for an initial exploratory study.
As a user can modify or delete the content or their Instagram account, a screenshot was taken of the post and the user profile, thereby creating a ‘post-record’, which became the main unit of analysis. The post-record was made using Microsoft Word. We analysed the content of a post to include both the photo and the accompanying text and hashtags but excluded subsequent comments (and hashtags) that were made by either the ‘post-owner’ or other users.
For each post, basic data points were gathered about the user and the post: age and gender of the user (self-reported in the user profile), and country of origin data by using the location specified as part of the post or contained in the user profile, as well as post-specific information (description of the photo/s, accompanying text and hashtags and the number of likes and comments etc.). We also captured the username, but as duplicates became apparent, we adopted our identification system for each post to be able to distinguish between different posts from the same users.
Whilst the Instagram posts are publically available the data cannot be reproduced to comply with the Instagram terms of service, comply with the ethical approval of this study and to protect the privacy of the individuals posting on Instagram.
Data analysis
We used an adapted mixed-method social network model to frame our analysis [ 16 ]. The model describes sourcing data (Instagram), constructing the data (organising and preparing for analysis) and analysing the data (using network analysis or linear modelling). The framework was appropriate as the study was exploratory and observational and employed a content analysis, however it was modified as we did not employ the network analysis or linear modelling. Fig 1 demonstrates our approach.
10.1371/journal.pone.0250641.g001
Fig 1
Method process, adapted from a mixed methods social network analysis framework [ 16 ].
The content analysis is ideal for exploratory research, as this method seeks to unobtrusively explore the explicit description of the communication and the trends, patterns and frequency of this communication found within data [ 17 , 18 ]. No a priori coding was developed owing to an absence of literature relating to the content of AML on Instagram. An inductive approach was employed to identify frequently occurring content categories and themes [ 18 ].
In brief the process included open coding, creating higher headings and then categories. After reviewing the post records multiple times, open codes were developed in a consultative and iterative process of reviewing the first ten post-records, at which time an open coding scheme was generated, that we thought could be applied to the whole data set [ 19 ]. A further ten posts were classified according to our coding scheme and codes were refined as necessary. The first initial ten posts were re-coded as per this scheme. This process was repeated twice, until we had a open coding scheme (after coding 40 posts) that could be applied to the entire data set. Higher order headings were then able to be developed from the open codes using researcher interpretation as to which codes belong in each higher order heading and then into categories ( Fig 2 ) [ 20 ]. The process was undertaken by two reviewers and any discordance in coding between reviewers was discussed to reach consensus [ 18 ].
10.1371/journal.pone.0250641.g002
Fig 2
Process of generating categories.
After the process was finished and the researchers were reflecting on the findings, we went back and coded for a single theme: ‘hope and/or gratitude’, as the researchers felt that even though this was outside the content analysis it was an interesting finding relevant to the research. This theme was based on the researchers’ interpretation of the image and accompanying text.
Most data were expressed as both means with standard deviations, medians with interquartile ranges (IQRs), as well as frequency and range because the content and distribution varied considerably. Microsoft Excel was used for these descriptive statistics.
Results
During the search period, almost all posts were found using #acutemyeloidleukemia (94%). During the window of analysis, 51 unique users posted content and 16 of these posted more than once resulting in the analysis of 100 posts, consisting of 138 photos—one post can contain up to 10 photos.
Age and gender were mostly unavailable. Only two profiles stated age but we have chosen to conceal this for re-identification protection. Gender was rarely specified in the user profile, and we deemed it unreliable to discern gender either, from self-description (e.g. mom, wife etc.), appearance or socially gender-normative names, so this has not been reported. As shown in Table 1 , we were mostly unable to determine the country of the post origin for most users (34/51).
10.1371/journal.pone.0250641.t001
Table 1 Country of post origin of the post or user account (n = 51).
Country
Frequency n (%)
United States
11 (22)
Canada
1 (2)
United Kingdom
3 (6)
Hong Kong
1 (2)
Malaysia
1 (2)
Unknown
34 (66)
We identified three user categories from the data: patient personal stories, personal support networks and professional organisations ( Table 2 ). The most frequent users were patients themselves (66% of the posts), followed by personal support networks that we interpreted as family and friends (24% of the posts) and lastly professional organisations (10% of the posts).
10.1371/journal.pone.0250641.t002
Table 2 Frequency of posts and photos in each user category.
User categories
Number of posts (n = 100) n (%)
Number of photos (n = 138) n (%)
Patient personal stories
66 (66)
99 (72)
Personal support networks
24 (24)
26 (19)
Professional organisations
10 (10)
13 (9)
As shown in Table 3 , the most frequent content posted in the analysis was patients communicating their health update (31% of the whole sample). The majority of posts made by personal support networks was a health update on behalf of a patient (50% of the personal support networks user category). Professional organisations only accounted for 10% of the total sample and the majority of the content was either patient information provision (40% of the posts) or raising disease awareness (50% of the posts).
10.1371/journal.pone.0250641.t003
Table 3 The content classification frequency by user category and content classification.
User category
Content classification
Frequency of posts for each user category n (%)
Frequency of content classification for whole sample (n = 100) %
Patient personal stories (n = 66)
Personal health
31 (47)
31
Reflection
24 (36)
24
Self-care
11 (17)
11
Personal support networks (n = 24)
Someone else’s health
12 (50)
12
Remembrance
4 (17)
4
Raising awareness
8 (33)
8
Professional organisations (n = 10)
Information provision for patients
4 (40)
4
Raising awareness
5 (50)
5
Other
1 (10)
1
The 10 organisational posts comprised of seven users and thirteen photos. Five of the seven users had an unknown country of origin, while one was based in the United States and the other in the United Kingdom as discerned from their profile or dot-org websites.
One-quarter of all posts detailed symptoms that were being experienced by patients and 19/25 posts containing symptoms came from patients with the remaining posts being made by personal support networks. Please note to the protect privacy of individuals (for example via reverse identification), the quotes chosen below have been altered to encompass the overall sentiment of the quote [ 21 ].
“#selfie #nofilter long term chemotherapy effects have mostly subsided. Still can’t shake that #red eye…” (picture of a person smiling into the camera). Patient personal story .
“…Hubby had platelets to fix his bleeding gums…” (picture of a person sitting upright in bed, surrounded by medical equipment) Personal support networks .
Likes and comments were used as a proxy measure for engagement ( Table 4 ). Overall there was between three and 394 likes and between zero and 54 comments. There was little engagement with organisational posts as measured by ‘likes’ and comments. There were between eleven and 41 likes on the posts and five posts had no comments.
10.1371/journal.pone.0250641.t004
Table 4 Engagement with posts by user category as measured by likes and comments.
User category
Likes
Comments
Mean (SD)
Median (IQR)
Range
Mean (SD)
Median (IQR)
Range
Patient personal stories
67.41 (68.61)
36.5 (51.5)
251
7.47 (10.16)
4 (8.75)
54
Personal support networks
79.71 (41)
91.87 (52.75)
391
5.58 (7.9)
3 (4.75)
31
Professional organisations
28.9 (8.64)
31 (12.75)
30
1 (1)
1 (1)
11
All posts
66.51 (73.59)
35 (50)
391
6.43 (9.35)
3(6)
54
Additionally, throughout the analysis, we noticed a prominent theme of hope often accompanied by gratitude, in the posts, either implicitly but commonly through the use of the accompanying text or hashtags (e.g. #gratitude or #grateful or #thankyou or #hopeful). Almost half (49%) of all the posts demonstrated this theme hope and/or gratitude. Thirty-four of these were made by the user category of patient personal stories, eleven by personal support networks and four by professional organisations.
“…Each day has something good in it, even on the toughest of days…” (Image of a motivational meme) Patient personal story .
Discussion
While much of the social media cancer communication research has focused on Facebook and Twitter, very few studies have focused on Instagram, particularly with a focus on such an emotionally and physically burdensome cancer like AML. Instagram differs from Facebook and Twitter by incorporating visual cancer communication and to our knowledge this is the first study to describe the content of Instagram communication concerning AML, thereby addressing this research gap. The novel method we have outlined is most useful for other investigators looking to utilise social media in the their research and our findings should be considered in the context of the limitations of our methods.
Our exploratory descriptive research showed in our sample, that people with AML communicating personal health updates, was the most common content being posted about AML on Instagram. Personal story sharing related to AML was also prominent by the personal support networks user category of people with AML. This finding was congruent with other Instagram disease-related research [ 3 , 22 , 23 ].
Why people tell such personal stories through Instagram may be explained by social media use being linked with patient empowerment through improved self-management and enhanced psychological and subjective well-being [ 1 , 3 ]. These benefits may be obtained through real or perceived social connectedness of users of Instagram where they feel a sense of intimacy through sharing or social support, through community [ 24 – 26 ]. By posting intimate stories, users may also provide and receive social and emotional support through these virtual online communities [ 23 ]. This is further supported by the high prevalence of hope and/or gratitude in our data, where Steffen et al found in a study of advanced lung cancer patients, that hope may be important in providing support to social and role functioning, irrespective of physical symptom severity [ 27 ]. In sentiment analysis, Cho and colleagues also found hope was the most commonly expressed emotion in their melanoma study [ 23 ]. Whether hope is a common finding on social media contained to people with a malignant disease remains unknown.
In contrast to a Facebook content analysis including breast, prostate and other reproductive cancers, Instagram users concerned with AML do not appear to be information seeking, which may be due to the inherent functionalities of the platform [ 28 ]. This means that health professionals, researchers and professional organisations should endeavour to tailor their communication respectively to the most appropriate platform. However, if users are predominately seeking or providing support through personal storytelling, Instagram presents an opportunity for health providers and other organisations tasked with awareness-raising or support and wellbeing. Furthermore, it is likely patients and their friends and family are highly motivated to sustain the engagement with cancer communication initiated by reputable professional organisations [ 28 , 29 ]. It is worth noting, we were unable to identify any health providers (individually or part of a health facility) posting during our data extraction period. The content of what patients communicate via social media outside the immediate doctor-patient consult provides an unique viewpoint unhindered by bustling waiting rooms or the interpretation of clinicians, to contextualise patient experience and decision making [ 7 ]. In our study, only about 10% of posts were organisational suggesting that Instagram may represent an untapped resource for cancer support communities and awareness campaigns. This suggestion possibly holds relevance for all cancer types. Furthermore, public awareness is particularly relevant for malignant haematological diseases where up to 70% of patients need to seek bone marrow transplant donors outside of their family and only 7% of the American population are registered bone marrow donors [ 30 ]. Increasing public awareness through emotional appeal and capitalising on hope as a concept, may increase the number of registered donors to ensure sufficient diversity in the donor pool to meet the patient demand for bone marrow transplant [ 31 , 32 ].
Social media research can complement other research methods: Crawford et al used YouTube to complement a literature review about the patient experience of haematological malignancies and found that YouTube provided supplementary information that highlighted the multifactorial experiences of patients that may not have been otherwise apparent through traditional research methods [ 7 ].
Certainly some individual healthcare professionals can and are using social media. A recent Italian study of neurologists showed that 56% of the sample used social media to have direct contact with patients and most of these health professionals were in favour of this communication method [ 33 ]. Instagram may provide an opportunity for clinican-led content that is trustworthy and appeals to patients, yet clinician-led social media posts are lacking, yet [ 34 ]. Moorhead et al. [ 35 ] suggests that both health professionals and their patients may need training to maximise the use of social media in their healthcare interaction. However, as yet it remains unknown how effective social media can be in its’ perceived role in healthcare [ 35 ] and how this applies to inherently passive platforms such as Instagram where interactivity between user and viewer is limited.
Given the popularity of Instagram and the potential reach of posts, further research is warranted to understand the implications of online visual communication and how this information can be harnessed to improve health communication, patient experience and the experience of healthcare and balancing this with minimising the perpetuation of misinformation to vulnerable individuals.
Our study is not without limitations: critics rightfully observe that Instagram is often curated and may not reflect real life—experiences are complex and Instagram is a snapshot in time. Additionally, our sample may not reflect the breadth of posts due to our sample size, which was limited by the practicality of employing a manual method and resourcing. The manual method we employed and limitations in the search function also meant the study was unable to capture videos and Instagram stories (which are only available for 24 hours from posting). The sample used in this study had many users posting multiple times, potentially meaning our results may be less diverse and biased towards fewer individual experiences of AML. The retrospective nature of this study only allowed for the capture of data about age, gender and location that the user chose to share and it is therefore unknown whether there are dominating age groups, gender or country of origin in our analysis.
The strengths of this study are that we have demonstrated a unique and innovative way to potentially reach and/or observe hard to reach populations or people suffering rare conditions. Additionally, photos are a unique and expressive medium not conventionally used in cancer support services so other researchers with appropriate research question could also choose to employ an interactive image-based study design.
Conclusion
This exploratory study, presents a novel method whereby we have characterised AML-related Instagram content that contributes to the understanding of how social media fits into the lives of people affected by AML. Our results suggest that social media may have a role to play particularly for social connectedness and support and that there is a potential role to play for health professionals and health organisations.
Further research should focus on exploring the feasibility and effectiveness of targeted awareness campaigns, as well as deploying support networks or health interventions to aid people by providing or seeking support.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The c-MYC proto-oncogene encodes a transcription factor that plays a central role in cell proliferation, differentiation, apoptosis, metabolism, and survival [ 1 , 2 ]. It can promote tumorigenesis in a variety of human malignancies [ 3 , 4 ]. c-MYC alteration occurs through various mechanisms, including chromosomal translocation, gene amplification, and perturbation of upstream signaling pathways [ 5 , 6 ]. Gene copy-number (GCN) gain or amplification is the most common c-MYC alteration in solid tumors [ 7 ].
Nevertheless, few studies have examined the clinicopathological implications of c-MYC status in colorectal cancer (CRC). Previous reports have shown that c-MYC GCN gain in CRC is found in approximately 10% of patients [ 8 ]. A recent study reported that several significant amplifications were focused on chromosome 8, including the 8q24 region which contains c-MYC , and suggested that c-MYC was a new marker for aggressive disease in CRC [ 9 ]. However, more recently, Christopher et al . reported data obtained by immunohistochemistry (IHC), indicating that c-MYC protein overexpression was significantly associated with improved prognosis in CRC patients [ 10 ]. Consequently, the prognostic value of c-MYC alterations in CRC is controversial.
Recently, the range of options for systemic chemotherapy has expanded and targeted therapy has been used in advanced CRC patients, increasing patient survival [ 11 ]. However, some CRC patients respond poorly to targeted therapy despite showing positive results in targeted therapy-specific mutation studies [ 12 ]. Tumor heterogeneity is a potential cause for failure of targeted therapy and several studies have reported that CRC possess a heterogenic genotype including KRAS , p53 , and BRAF [ 13 – 15 ]. Therefore, genetic variation between the primary tumor and corresponding metastatic sites needs to be clarified to improve the management of CRC patients with metastatic disease.
The heterogeneity of c-MYC and its prognostic implications have not been systematically studied in primary CRC patients. The aim of this study was to evaluate c-MYC gene status and its clinical significance in CRC. We also analyzed the heterogeneity of c-MYC in the primary tumor and distant metastasis.
Materials and Methods
Patients and samples
A total of 519 CRC patients treated with radical surgery at Seoul National University Bundang Hospital were enrolled in this retrospective study. First, to evaluate the clinicopathologic significance of c-MYC gene status, 367 consecutive CRC patients treated between January 2005 and December 2006 were enrolled (cohort 1). Second, to investigate the discordance between the primary and metastatic tumors, 152 advanced CRC patients with synchronous or metachronous metastasis who had undergone surgical resection for primary CRC between May 2003 and December 2009, were enrolled (cohort 2). All the cases were reviewed by two pathologists (K. S. L. and H. S. L.). The clinicopathological characteristics were obtained from the patients’ medical records and pathology reports. Follow-up information including patient outcome and the interval between the date of surgical resection and death was collected. Data from patients lost to follow-up or those who had died from causes other than CRC were censored.
Ethical statement
All samples were obtained from surgically resected tumors examined pathologically at the Department of Pathology, Seoul National University Bundang Hospital. All samples and medical record data were anonymized before use in this study and the participants did not provide written informed consent. The use of medical record data and tissue samples for this study was approved by the Institutional Review Board of Seoul National University Bundang Hospital (reference: B-1210/174-301).
Tissue array method
Surgically resected primary CRC specimens were formalin-fixed and paraffin-embedded (FFPE). For each case, representative areas of the donor blocks were obtained and rearranged into new recipient blocks (Superbiochips Laboratories, Seoul, South Korea) [ 16 ]. A single core was 2 mm in diameter and those containing > 20% tumor cells were considered valid cores.
Dual-color silver in situ hybridization
The c-MYC gene was visualized by using a blue-staining system (ultraView silver in situ hybridization [SISH] dinitrophenol [DNP] detection kit and c-MYC DNP probe, Ventana Medical Systems, Tucson, AZ, USA). The centromere of chromosome 8 (CEP8) was visualized by using a red-staining system (ultraView red ISH digoxigenin [DIG] detection kit and chromosome 8 DIG probe, Ventana Medical Systems). Positive signals were visualized at 60 × magnification and counted in 50 non-overlapping tumor cell nuclei for each case ( Fig 1 ) [ 17 ]. Small and large clusters were scored as 6 and 12 signals, respectively.
10.1371/journal.pone.0139727.g001
Fig 1
Representative figures of c-MYC status detected by dual-color silver in situ hybridization (A and B) in colorectal cancer patients.
(A) c-MYC gene copy number gain (60 × magnification); (B) c-MYC gene disomy (60 × magnification).
Immunohistochemistry
IHC analysis of c-MYC was carried out using a commercially available rabbit anti-c-MYC antibody (clone Y69, catalog ab32072, Abcam, Burlingame, CA, USA). The staining procedures were carried out using the ultraView Universal DAB kit (Ventana Medical Systems) and an automated stainer (BenchMark®XT, Ventana Medical Systems), according to the manufacturer’s instructions. Nuclear immunostaining of c-MYC was negative in normal mucosa. For statistical analysis, c-MYC nuclear staining of any intensity in greater than 10% of neoplastic cells was scored as positive ( S2 Fig ) [ 10 ].
Microsatellite instability
Microsatellite instability (MSI) was assessed in CRC cases with available tissue. MSI results were generated by comparing the allelic profiles of 5 microsatellite markers (BAT-26, BAT-25, D5S346, D17S250, and S2S123) in the tumor and corresponding normal samples. Polymerase chain reaction (PCR) products from the FFPE tissues were analyzed using an automated DNA sequencer (ABI 3731 Genetic Analyser, Applied Bio systems, Foster City, CA, USA) according to the protocol described previously [ 18 ].
KRAS mutation analysis
KRAS mutation detection was achieved by melting curve analysis using the cobas 4800 System (Roche, Branchburg, NJ, USA) with automated result interpretation software. This is a TaqMelt-based real-time PCR assay designed to detect the presence of 21 KRAS mutations in codons 12, 13, and 61. The workflow and testing process have been described previously [ 19 ].
Statistical analyses
The association between the clinicopathological features and c-MYC status was analyzed using the chi-square or Fisher’s exact test, as appropriate. The correlation between the detection methods was examined using the Pearson correlation coefficient. The patients’ survival was analyzed by using the Kaplan-Meier method and the log-rank test was used to determine if there were any significant differences between the survival curves. Univariate and multivariate regression analysis were performed by using Cox’s proportional hazards model to determine the hazard ratio and 95% confidence intervals for each factor. A P- value < 0.05 was accepted as statistically significant. All statistical analyses were performed using the SPSS statistics 21 software (IBM, Armonk, NY, USA).
Results
c-MYC gene status and clinical implications for consecutive primary CRC patients
In consecutive primary CRC cases (cohort 1), the median c-MYC :CEP8 ratio was 1.29 (range, 0.58–5.17). c-MYC gene amplification, defined by a c-MYC :CEP8 ratio ≥ 2.0 and similar to that established for HER2 [ 20 ], was detected in 31 (8.4%) of 367 patients. The mean c-MYC GCN was 2.88 (range, 1.22–13.12). In the present study, we defined the GCN gain as ≥ 4.0 c-MYC copies/nucleus [ 21 ], and this was detected in 63 (17.2%) of 367 CRC patients. All c-MYC amplification was included in c-MYC GCN gain. A c-MYC GCN gain ≥ 4 had the lowest P -value ( P = 0.015) and thus, was observed to be the most predictive cut-off point for patient prognosis ( Fig 2 ); ≥ 5.0 c-MYC copies/nucleus also influenced patient prognosis ( P = 0.026). There was no significant association between patient prognosis and either c-MYC amplification ( P = 0.149) or > 2, ≥ 3, and ≥ 6 c-MYC copies/nucleus ( P = 0.752, P = 0.175, and P = 0.122, respectively).
10.1371/journal.pone.0139727.g002
Fig 2
Kaplan-Meier survival curves illustrating the prognostic effect of c-MYC status in colorectal cancer (cohort 1).
(A) c-MYC gene copy number (GCN) gain; (B) c-MYC GCN gain in the stage II-III subgroup; (C) c-MYC amplification.
Table 1 shows the relationships between c-MYC status and the clinicopathological parameters in consecutive primary CRCs (cohort 1). Amplification of c-MYC correlated with early-stage disease ( P = 0.039). c-MYC GCN gain was frequently observed in sigmoid colon and rectum tumors ( P = 0.034), small tumors ( P = 0.041), and those classified as microsatellite stable or MSI-low ( P = 0.029).
10.1371/journal.pone.0139727.t001
Table 1 The association between clinicopathological parameters and c-MYC status in 367 CRC patients (cohort1).
Total
c-Myc
P -Value
c-Myc
P -Value
c-Myc IHC
P -Value
4 > GCN
4 ≦ GCN
Non-amplification
Amplification
Negative
Positive
Age
0.983
0.383
0.537
mean
64.2
64.2
64.2
64.1
66.0
64.6
63.9
Sex
0.740
0.619
0.431
male
205
171 (83.4%)
34 (16.6%)
189 (92.2%)
16 (7.8%)
89 (43.4%)
116 (56.6%)
female
162
133 (82.1%)
29 (17.9%)
147 (90.7%)
15 (9.3%)
77 (47.5%)
85 (52.5%)
Location
0.034
0.437
< 0.001
Rectum/sigmoid
237
189 (79.7%)
48 (20.3%)
121 (93.1%)
9 (6.9%)
90 (38.0%)
147 (62.0%)
others
130
115 (88.5%)
15 (11.5%)
215 (90.7%)
22 (9.3%)
76 (58.5%)
54 (41.5%)
pT stage
0.692
0.571
< 0.001
0–2
58
47 (81.0%)
11 (19.0%)
52 (89.7%)
6 (10.3%)
14 (24.1%)
44 (75.9%)
3–4
309
257 (83.2%)
52 (16.8%)
284 (91.9%)
25 (8.1%)
152 (49.2%)
157 (50.8%)
Differentiation
0.139
0.055
0.007
LG
331
271 (81.9%)
60 (18.1%)
300 (90.6%)
31 (9.4%)
142 (42.9%)
189 (57.1%)
HG
36
33 (91.7%)
3 (8.3%)
36 (100.0%)
0 (0.0%)
24 (66.7%)
12 (33.3%)
LN metastasis
0.609
0.070
0.058
absent
168
141 (83.9%)
27 (16.1%)
149 (88.7%)
19 (11.3%)
67 (39.9%)
101 (60.1%)
present
199
163 (81.9%)
36 (18.1%)
187 (94.0%)
12 (6.0%)
99 (49.7%)
100 (50.3%)
Lymphatic invasion
0.152
0.896
0.073
absent
158
136 (86.1%)
22 (13.9%)
145 (91.8%)
13 (8.2%)
63 (39.9%)
95 (60.1%)
present
209
168 (80.4%)
41 (19.6%)
191 (91.4%)
18 (8.6%)
103 (49.3%)
106 (50.7%)
Perineural invasion
0.631
0.530
0.025
absent
154
212 (83.5%)
42 (16.5%)
231 (90.9%)
23 (9.1%)
49 (58.7%)
105 (41.3%)
present
113
92 (81.4%)
21 (18.6%)
105 (92.9%)
8 (7.1%)
61 (54.0%)
52 (46.0%)
Venous invasion
0.776
0.999
0.814
absent
296
246 (83.1%)
50 (16.9%)
271 (91.6%)
25 (8.4%)
133 (44.9%)
163 (55.1%)
present
71
58 (81.7%)
13 (18.3%)
65 (91.5%)
6 (8.5%)
33 (46.5%)
38 (53.5%)
Tumor border
0.524
0.327
0.544
expanding
60
48 (80.0%)
12 (20.0%)
53 (88.3%)
7 (11.7%)
25 (41.7%)
35 (58.3%)
infiltrative
307
256 (83.4%)
51 (16.6%)
283 (92.2%)
24 (7.8%)
141 (45.9)
166 (54.1%)
Size (cm)
0.041
0.061
< 0.001
mean
5.3
5.4
4.7
5.3
4.5
5.8
4.8
Distant metastasis
0.123
0.544
0.252
absent
299
252 (84.3%)
47 (15.7%)
275 (92.0%)
24 (8.0%)
131 (43.8%)
168 (56.2%)
present
68
52 (76.5%)
16 (23.5%)
61 (89.7%)
7 (10.3%)
35 (51.5%)
33 (48.5%)
pTNM stage
0.822
0.039
0.050
I, II
162
135 (83.0%)
27 (17.0%)
140 (88.1%)
19 (11.9%)
64 (39.5%)
98 (60.5%)
III, IV
205
169 (82.4%)
36 (17.6%)
193 (94.1%)
12 (5.9%)
102 (49.8%)
103 (50.2%)
MSI status
0.029
0.256
0.490
MSS/MSI-L
323
264 (81.7%)
59 (18.3%)
294 (91.0%)
29 (9.0%)
141 (38.4%)
182 (49.6%)
MSI-H
32
31 (96.9%)
1 (3.1%)
31 (96.9%)
1 (3.1%)
16 (1.6%)
16 (4.4%)
Abbreviations: CRC, colorectal cancer; T, tumor; LG, low grade; HG, high grade; LN, lymph node; MSS, microsatellite stable; MSI-L, microsatellite instability-low; MSI-H, microsatellite instability-high; GCN, gene copy number; IHC, immunohistochemistry
P -values are calculated by using χ 2 -test or Fisher’s exact test
Prognostic significance of c-MYC gene status in CRC patients
All CRC patients were successfully followed up for inclusion in the survival analysis ( Fig 2 ). In cohort 1, the mean follow-up period was 55 months (range, 1–85 months) and 101 (27.5%) patients died during the follow-up period. Kaplan-Meier analysis showed that c-MYC GCN gain was significantly associated with poor survival in CRC patients ( P = 0.015), but c-MYC amplification was not ( P = 0.149). In the stage II-III subgroup, c-MYC -GCN gain also predicted poor prognosis ( P = 0.034). Multivariate analysis of c-MYC status is summarized in Table 2 , and showed that c-MYC -GCN gain independently predicted poor prognosis in the consecutive cohort ( P < 0.001) and in the subgroup of patients with stage II-III CRC ( P = 0.040).
10.1371/journal.pone.0139727.t002
Table 2 Multivariate Cox proportional hazard models for the predictors of overall survival (cohort 1).
Univariate survival analysis
Multivariate survival analysis
Factors
HR (95% CI)
P value
HR (95% CI)
P value
c-MYC GCN SISH (4≦ vs. 4>)
1.756 (1.117–2.763)
0.015
2.350 (1.453–3.801)
<0.001
Age
1.026 (1.008–1.045)
0.005
1.025 (1.007–1.043)
0.006
Size
1.244 (1.059–1.244)
0.001
1.099 (0.995–1.214)
NS (0.062)
Histologic grade (high vs. low)
3.143 (1.904–5.188)
<0.001
2.844 (1.625–4.977)
<0.001
Stage (3/4 vs. 1/2)
6.151 (3.494–10.829)
<0.001
3.069 (1.603–5.878)
0.001
Lymphatic invasion
3.661 (2.242–5.980)
<0.001
1.251 (0.709–2.205)
NS (0.439)
Perineural invasion
3.942 (2.648–5.870)
<0.001
2.325 (1.487–3.636)
<0.001
Venous invasion
3.985 (2.671–5.946)
<0.001
2.304 (1.490–3.676)
<0.001
c-MYC GCN SISH (4≦ vs. 4>) in a subgroup of stage II/III
2.057 (1.039–4.073)
0.038
2.058 (1.032–4.105)
0.040
Age
1.037 (1.009–1.067)
0.010
1.036 (1.007–1.066)
0.014
Stage (3 vs. 2)
2.955 (1.493–5.850)
0.002
1.705 (0.802–3.623)
NS (0.165)
Lymphatic invasion
2.882 (1.456–5.703)
0.002
1.846 (0.887–3.845)
NS (0.101)
Perineural invasion
3.536 (1.952–6.405)
0.001
2.921 (1.558–5.476)
<0.001
Abbreviations: SISH, silver in-situ hybridization; GCN, gene copy number; HR, hazard ratio
P -values are calculated by using χ 2 -test or Fisher’s exact test
Correlation between the c-MYC GCN gain and protein overexpression
Overexpression of c-MYC protein was detected in 201 (54.8%) of 367 CRC patients (cohort 1) and was associated with early pT stage ( P < 0.001), low grade of histologic differentiation ( P = 0.007), absence of perineural invasion ( P = 0.025) and smaller size ( P < 0.001) ( Table 1 ). Overexpression of c-MYC protein was associated with GCN gain (ρ, 0.211; P < 0.001), which was categorized as weakly correlation according to Dancey and Reidy’s categorization (2004) [ 22 ]. Furthermore, only 46 (22.9%) of 201 patients with c-MYC overexpression showed a GCN gain.
c-MYC status and heterogeneity according to tumor location in advanced CRC patients
To evaluate the regional heterogeneity of c-MYC status, we examined tissue from 3 sites including the primary cancer, distant metastasis, and lymph-node metastasis for each patient with advanced CRC (cohort 2). In the primary tumors of cohort 2, the median c-MYC :CEP8 ratio was 1.14 (range, 0.57–2.97). c-MYC gene amplification was detected in 8 (5.3%) of 152 patients. The mean c-MYC GCN was 2.97 (range, 1.40–9.94). c-MYC GCN gain was detected in 48 (31.6%) of 152 CRC patients. In addition, c-MYC GCN gain was found in 33 (21.7%) patients with distant metastatic tumors. Lymph-node metastasis was observed in 79 of 152 advanced CRC patients and c-MYC GCN gain was observed in 18 (22.8%) of these cases. The heterogeneity of c-MYC GCN gain according to tumor location is shown in Table 3 . Of 152 cases, discordance between c-MYC GCN gain in the primary tumor and distant metastasis was noted in 39 (25.7%) cases. Discordance between c-MYC GCN gain in the primary tumor and lymph-node metastasis was detected in 24/79 (30.4%) cases. Thus, regional heterogeneity of c-MYC GCN gain was quite common in advanced CRC. c-MYC GCN heterogeneity was not correlated with clinicopathological factors and prognosis ( P > 0.05; data not shown).
10.1371/journal.pone.0139727.t003
Table 3 Heterogeneity of c-MYC GCN gain with respect to tumor location in advanced CRC (cohort 2).
c-MYC GCN gain (%)
Primary
negative
positive
total
Distant metastasis
negative
92 (60.5)
27 (17.8)
152 (100)
positive
12 (7.9)
21 (13.8)
LN metastasis
negative
44 (55.7)
17 (21.5)
79 (100)
positive
7 (8.9)
11 (13.9)
Abbreviations: GCN, gene copy number; LN: lymph node
P -values are calculated by using χ 2 -test or Fisher’s exact test
There was no statistically significant correlation between the clinicopathological factors and c-MYC GCN gain in primary, distant metastatic, and lymph-node metastatic tumors from cohort 2 CRC patients ( P > 0.05; data not shown). The mean follow-up time was 42 months (range, 1–105 months) and 67 patients (44.1%) died from cancer during the follow-up period. Kaplan-Meier analysis showed that patients with c-MYC GCN gain in the primary tumor had a poor outcome than those without, but this result was not statistically significant ( P = 0.499). However, ≥ 3.0 c-MYC copies/nucleus in the primary tumor was significantly associated with a poor prognosis ( P = 0.044; S1 Fig ). There was no significant correlation between the patients’ prognosis and c-MYC GCN gain in distant or lymph-node metastases ( P = 0.981 and P = 0.417, respectively; data not shown).
KRAS mutations in advanced CRC
The cobas KRAS test was performed on 152 primary tumors from advanced CRC cases (cohort 2). KRAS gene mutations were observed in 84 (55.3%) cases and were associated with tumors located in the right colon ( P = 0.019), but were not correlated with other clinicopathological factors ( P > 0.05; data not shown). Additionally, there was no statistical difference between the survival of CRC patients with mutated or wild-type KRAS ( P = 0.688; data not shown). Of 68 cases with wild-type KRAS , c-MYC amplification was noted in 4 (5.9%) and a c-MYC GCN gain in 28 (41.2%). Of 84 cases with mutated KRAS , 4 showed c-MYC amplification (4.8%) and 20 (23.8%) revealed a c-MYC GCN gain. c-MYC GCN alterations occurred in patients with both wild-type and mutated KRAS. Therefore, c-MYC GCN alterations and KRAS mutations were not mutually exclusive.
Discussion
Although there have been several reports on c-MYC status in human cancers, there are no established criteria for GCN gain. Cancers with a c-MYC GCN gain are often associated with a poor prognosis. A previous study of mucinous gastric carcinoma showed that c-MYC amplification, defined as a c-MYC :CEP8 ratio > 2.0, was strongly correlated with the advanced stages of cancer [ 23 ]. Another report found an association between c-MYC amplification (> 4 copies/cell in a minimum of 10% of tumor cells) and the advanced stages of ovarian cancer [ 21 ]. In a study of prostate cancer, the c-MYC GCN gain included the criterion of a c-MYC/ CEP8 ratio > 1.5, and a poor prognosis was observed for patients in this category [ 24 ]. In recent research on adenocarcinoma of the lung, patients with > 2 c-MYC copies/nucleus were classified as having an increased c-MYC GCN, which was found to be an independent poor prognostic factor [ 25 ]. In CRC patients, it was reported that c-MYC amplification, defined as a c-MYC/ CEP8 ratio > 2, was frequently detected by using fluorescent in situ hybridization (9.0–14.2%), but was unrelated to clinical outcome and pathological data [ 26 ]. Therefore, we have applied diverse criteria to determine c-MYC amplification or GCN gain in this study, and have defined the c-MYC GCN gain as ≥ 4 copies/nucleus, because it had the lowest P -value for disease prognosis ( Fig 2 ). In cohort 1, the large consecutive cohort, CRC patients with a c-MYC GCN gain had a poor survival than those without ( P = 0.015). Furthermore, in multivariate analysis, c-MYC GCN gain was a significant CRC prognostic factor, both in the consecutive cohort and for those with stage II-III disease. The predictive value of the c-MYC GCN was found to be independent of known prognostic factors. The c-MYC GCN gain criteria used in the present study, together with the SISH method, may be useful in assessing CRC patients because it clearly identified patients expected to have poor survival, regardless of the c-MYC :CEP8 ratio.
In cohort 2, we showed that there was c-MYC GCN regional heterogeneity between the primary site and its related metastases. A c-MYC GCN gain (c-MYC GCN ≥ 4.0) in the primary cancer was not significantly associated with poor survival ( P = 0.499; S1 Fig ), which might be because all of cohort 2 consisted of advanced CRC patients with synchronous and metachronous metastasis and cohort 2 was largely comprised of stage IV CRC (98 cases; 64.5%). They received a variety of personalized treatment respectively and these might reflect the statistical insignificance. Interestingly, we applied slightly non-restrictive criteria of GCN gain ( c-MYC GCN ≥ 3.0) and its prognosis was changed to statistically significant ( P = 0.044; S1 Fig ). In a broad sense, c- MYC GCN gain of primary cancer tends to correlated with poor survival in advanced CRC. On the other hand, c-MYC status in distant and lymph-node metastatic lesion was not related to patient prognosis although we tried every possible GCN criteria. Even though, c-MYC heterogeneity was observed frequently in advanced CRC, a c-MYC GCN gain in the primary cancer was often associated with poor survival. Consequently, the c-MYC GCN in the non-metastatic lesion should be used when evaluating prognosis.
In a previous study, overexpression of c-MYC mRNA in CRC was found to be associated with a better prognosis [ 27 ], but this result was contradicted by another study [ 28 ]. Christopher et al . recently demonstrated that c-MYC overexpression determined by IHC alone, was significantly associated with a better survival in CRC patients when assessed by univariate analysis, but not by multivariate analysis [ 10 ]. Interestingly, we found conflicting results in a previous c-MYC overexpression study; presumably, because c-MYC expression is controlled by a complex regulatory pathway involving multiple interactions with other molecules, rather than just simple GCN gain [ 29 ]. Furthermore, we found a weak correlation between c-MYC protein overexpression and GCN gain in CRC patients. c-MYC GCN gain was not observed in most c-MYC protein overexpression cases. Unlike the c-MYC GCN gain, overexpression of c-MYC protein was correlated with less aggressive features ( Table 1 ). These results suggest that c-MYC GCN gain is probably only partly responsible for protein overexpression. As overexpression of c-MYC is not the same as a c-MYC GCN gain, further research is needed to explain the difference of c-MYC overexpression and GCN gain in CRC tumorigenesis.
Mutations in KRAS are evident in 30–40% of colorectal tumors [ 30 – 32 ]. Indeed, previous studies reported that a KRAS mutation was associated with resistance to anti-epidermal growth factor receptor (EGFR) monoclonal therapy and a poor survival [ 33 – 35 ]. In our study, KRAS mutations were present in 55.3% of advanced CRC patients (cohort 2) and were not associated with prognosis. It may be because we investigated KRAS mutation status in advanced CRC patients. Phipps et al. also reported that KRAS -mutation status was not associated with poor disease specific survival in cases who presented with distant-stage CRC [ 33 ]. c-MYC amplification was observed in 5.9% of wild-type KRAS and 4.8% of mutated KRAS CRCs. A c-MYC GCN gain was observed in 41.2% of wild-type KRAS and 23.8% of mutated KRAS CRCs. It is noteworthy that these 2 genetic events were not mutually exclusive. Further studies are required to investigate the possibility of using c-MYC genetic alterations as therapeutic targets in advanced CRC patients with primary and secondary resistance to anti-EGFR therapies.
In summary, we comprehensively analyzed the c-MYC gene status of CRC patients by using SISH. c-MYC GCN gain and amplification were observed in 17.2% and 8.4% of consecutive CRC patients, respectively. The c-MYC GCN gain was an independent poor prognostic factor, both in the consecutive cohort and in the subgroup of patients with stage II-III disease. These findings show that c-MYC status can be used to predict the prognosis of CRC patients, and may inform future studies on the pathogenesis and mechanisms involved in the progression of CRC.
Supporting Information
S1 Fig
Kaplan-Meier survival curves illustrating the prognostic effect of c-MYC status in primary lesions of colorectal cancer (cohort 2).
(A) c-MYC gene copy number (GCN) ≥ 3.0; (B) c-MYC GCN ≥ 4.0.
(TIF)
S2 Fig
Representative figures of c-MYC overexpression by immunohistochemistry (A and B) in colorectal cancer patients.
(A) c-MYC overexpression (40 × magnification); (B) No c-MYC expression (40 × magnification);
(TIF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Hantaan (HTNV) and Seoul viruses (SEOV), etiologic agents of hemorrhagic fever with renal syndrome (HFRS) in the Old World, pose serious health threats to military and civilian personnel residing, working, or conducting routine military operations in rodent-infested environments. HTNV, transmitted through the inhalation of dusts containing HTNV-infected Apodemus agrarius excreta, poses the greatest threat due to its extended incubation period (up to 50 days), mean duration of illness from onset of symptoms to complete recovery, and overall morbidity and mortality rate of 9.46% reported for US military personnel in the presence of quality medical management [ 1 – 4 ]. The morbidity of SEOV ranges from mild to moderate and mortality rate <1%, is similarly transmitted by inhalation of dusts containing SEOV-infected Rattus norvegicus excreta, and while much less severe than HTNV infections, poses a threat for personnel occupying or conducting operations in rat infested urban environments [ 5 ].
In 2000, the 18 th Medical Command launched a comprehensive rodent-borne disease surveillance aimed primarily to assess HFRS risks based on activities of military personnel, habitat characteristics, relative rodent population densities, and HTNV IgG antibody-positive (Ab+) at United States (US) and Republic of Korean (ROK) operated military training sites [ 4 , 6 – 10 ]. Rodent-borne disease surveillance at training sites near the demilitarized zone (DMZ) showed that A . agrarius populations, based on trap rates, remained relatively stable and that high gravid rates were observed in August-September, which correlated with increased numbers of HFRS cases from late September-December [ 4 , 7 , 9 , 10 ]. The proposed consolidation of US military forces to Camp (Cp) Humphreys resulted in the purchase of adjoining farmlands (rice paddies and associated irrigation and drainage systems). By 2008, the area immediately adjacent to the preexisting Cp Humphreys boundary was under development (land graded and paddies filled) to make way for the construction of roads, drainage systems, housing, and other structures that support military operations and other related activities. From 1986–2016, when detailed epidemiological data were available, there were no reported HFRS cases attributed to exposure at Cp Humphreys and nearby local training areas. However, due to landscape modifications and construction activities that may alter HFRS risks, a comprehensive rodent-borne disease surveillance program was conducted at the Cp Humphreys and the new expansion site. HFRS risk factors were estimated based on associated human activities that considered environmental, biological, and hantavirus prevalence associated with rodents and soricomorphs for both the established cantonment area of Cp Humphreys and the fallow farmlands that were being graded/filled that may alter HFRS risks. This report focuses on epidemiological data that identifies environmental factors, seasonal rodent population densities, biological factors, and HTNV serological and molecular prevalence related to HTNV transmission risks for Cp Humphreys and the expansion site, Pyeongtaek. The whole genomic sequences of HTNV newly obtained for Cp Humphreys and the expansion site, Pyeongtaek, were phylogenetically characterized demonstrating a greater diversity of rodent-borne hantaviruses.
These data are useful for developing local HFRS disease threat awareness, analysis, and risk reduction strategies for civilians and military personnel in southern Gyeonggi province, ROK.
Materials and methods
Ethics statement
All trapping of small mammals was approved by US Forces Korea (USFK) in accordance with USFK Regulation 40–1 “Prevention, Surveillance, and Treatment of Hemorrhagic Fever with Renal Syndrome” at US Military Installations and US and ROK Operated Military Training Sites [ 11 ]. Standard procedures were followed for the collection and transportation of specimens to minimize hazards from potentially infected animals as previously described and all personnel processing animals at the Korea University laboratory were vaccinated using a Korean-approved Hantavirus vaccine (Hantavax ® ) [ 12 , 13 ]. Small mammals were euthanized by cardiac puncture under isoflurane anesthesia in strict accordance with the Korea University Institutional Animal Care and Use Committee (KUIACUC, #2010–212) protocol approved for this study. All efforts were made to minimize suffering.
Site description
Cp Humphreys, located near Pyeongtaek in southern Gyeonggi Province, was designated as a US military hub incorporating US military installations near the DMZ and elsewhere ( Fig 1 ) . The areas surveyed included: (1) Cp Humphreys (hereafter referred to as “Cp Humphreys”) and (2) the new expansion site (referred to hereafter as “expansion site”) for the construction of structures for military operations and housing and development of roads, drainage systems, and outdoor recreational areas.
10.1371/journal.pone.0176514.g001
Fig 1
Location of Camp Humphreys, Gyeonggi province, Republic of Korea where rodent-borne disease surveillance was conducted from 2008–2010 and 2012.
Cp Humphreys is bounded by a small town (Anjeong-ri, Paengseong-eup), farmland, and the expansion site. It consists of an airstrip, asphalt roads, and structures for military operations and housing, streams, man-made water impoundments, and drainage ditches. Most areas bordering structures and roadside ditches were well maintained, while unmanaged grasses/herbaceous vegetation and shrubs border ponds, streams, and some drainage ditches, providing limited space, cover, and food for small mammals.
The expansion site under development is bounded by Cp Humphreys and the Anseongcheon River. During 2008, areas immediately adjacent to Cp Humphreys were under development for the construction of structures for military operations and housing, roads, drainage systems, and recreation areas. A large area not under development consisted of low lying hills of grasses/herbaceous vegetation, small groves of trees, major irrigation/drainage systems to reduce flooding, rice paddies, and limited dry-land farmland that lay fallow with unmanaged grasses/herbaceous vegetation. The major irrigation/drainage systems were bordered by moderate/tall grasses along the banks extending to roadways that harbored relatively high small mammal populations. Grading and filling of the unmanaged fallow lands continued and by 2010 much of the expansion site had been graded and cleared of grasses/herbaceous vegetation in preparation for land fill and construction. Habitats of small mammals (e.g., rodents and soricomorphs) and predators (e.g., weasels, raccoon dogs, and feral cats) were destroyed as lands were sequentially graded and filled, leaving these areas devoid of vegetation except for limited areas of grasses/herbaceous vegetation bordering ditches, major drainage systems for flood control, and temporary roads that provided habitat for small mammals.
Small mammal trapping
Trapping was conducted monthly from January 2008-December 2009, quarterly during 2010, and semiannually in 2012. Areas surveyed included: limited tall grasses/herbaceous vegetation that bordered streams, retaining ponds, fence lines, park perimeters, and open fields at Cp Humphreys and expansive tall grasses/herbaceous vegetation bordering dirt and paved roads, streams, ditches, water impoundments, flood control drainage systems, spillways, and the Anseongcheon River at the expansion site, which did not interfere with military or construction activities. Collapsible live-capture Sherman® traps (7.7x9x23 cm; H.B. Sherman, Tallahassee, FL) were set in grasses/herbaceous vegetation (providing shade) at 4–5 m intervals (25–50 traps/trap line) in the late afternoon over a 2–3 day period and picked up early the following morning as previously described [ 4 , 6 – 10 ]. Traps positive for small mammals were sequentially numbered, placed in a secure container, and transported to Korea University where they were euthanized, identified to species using morphological techniques, sexed, weighed and then, tissues (lung, liver, kidney, and spleen) collected and stored at -80°C until used [ 14 ].
Serological and molecular screening for hantaviruses
Small mammal sera were diluted 1:16 in phosphate buffered saline and examined for IgG antibodies against HTNV, SEOV, Prospect Hill virus, and Imjin virus (MJNV) by indirect immunofluorescent antibody test (IFAT) [ 15 – 18 ]. Lung tissues of hantavirus Ab+ rodents and soricomorphs were used for the identification of the hantavirus gene by RT-PCR that amplified a portion of the G C glycoprotein-encoding M segment. Total RNA, extracted from lung tissues of the seropositive animals using the RNA-Bee Kit (TEL-TEST Inc., Friendswood, TX), was reverse-transcribed using Superscript® II RNase H-reverse transcriptase kit (Invitrogen, Carlsbad, CA) according to the manufacturer’s instructions. Nested-PCR using the primers (outer primer set, 5’-TGGGCTGCAAGTGC-3’ , 5’-ACATGCTGTACAGCCTGTGCC-3’ ; inner primer set, 5’-TGGGCTGCAAGTGCATCAGAG-3’ , 5’-ATGGATTACAACCCCAGCTCG-3’ ) was performed to recover a 373-nucleotide (nt) region of the hantavirus M segment [ 19 – 21 ]. Amplified products were size-fractionated by electrophoresis on 1.5% agarose gels containing ethidium bromide (0.5 mg/mL). The PCR products were cloned using the pST Blue-1 vector (Novagen, Dormstadt, Germany) and plasmid DNA purified using the QIAprepSpinMiniprep kit (QIAGEN Inc., Chatsworth, CA). DNA sequencing was performed in both directions from at least three clones of each PCR product using the Big-Dye® Terminator v3.1 cycle sequencing kit (Applied Biosystems, Foster City, CA) on an automated sequencer (Model 3730, Applied Biosystems).
Real-time quantitative PCR (RT-qPCR)
cDNA was synthesized using a High Capacity RNA-to-cDNA kit (Applied Biosystems) by adding 1 μg of total RNA from lung tissues of HTNV-positive A . agrarius . RT-qPCR was conducted using SYBR Green PCR Master mix (Applied Biosystems) on a StepOne Real-Time PCR System (Applied Biosystems). The primer sequences included a forward primer; 5’-TTATTGTGCTCTTCATGGTTGC-3’ and a reverse primer; 5’-CATCCCCTAAGTGGAAGTTGTC-3’ for HTNV S segment [ 22 ]. The cycling program was a cycle of 95°C for 10 min, followed by 40 cycles of 15 s at 95°C and 1 min at 60°C.
Whole genome sequencing of HTNV
Total RNA was isolated from lung tissues of HTNV-positive A . agrarius using a Hybrid R Kit (GeneAll Biotechnology, Seoul, ROK). cDNA was synthesized using High Capacity RNA-to-cDNA kit (Applied Biosystems) with random hexamer or 5′-TAGTAGTAGACTCC-3′ . Using Ex Taq DNA polymerase (TaKaRa BIO Inc., Shiga, Japan), first and second RT-PCR were performed at 95°C for 4 min, followed by 6 cycles of denaturation at 94°C for 30 sec, annealing at 37°C for 30 sec, elongation at 72°C for 1 min 30 sec, and then 32 cycles of denaturation at 94°C for 30 sec, annealing at 42°C for 30 sec, and elongation at 72°C for 1 min 30 sec, and 72°C for 5 min (ProFlex PCR System, Life Technology, CA, USA). To complete whole genome sequencing of HTNV, rapid amplification of cDNA ends (RACE) for 3’ and 5’ termini were performed using 3’ and 5’-Full RACE Core Sets (Takara, Shiga, Japan) according to manufacturer specifications.
Genetic and phylogenetic analyses
Alignments of whole genome sequences of HTNV L, M, and S segments were facilitated using the Clustal W method (Lasergene program version 5, DNASTAR Inc. Madison, WI). The phylogenetic tree was generated by neighbor joining (NJ) and maximum likelihood (ML) methods (Molecular Evolutionary Genetics Analysis, 6.0). Genetic distances were computed, and topologies evaluated by bootstrap analysis of 1,000 iterations [ 23 ].
Results
Capture rates for species
A total of 2,364 (capture rate = 20.92%) small mammals over 11,300 trap nights, belonging to the Orders Rodentia [Families Muridae (3 species) and Cricetidae (3 species)] and Soricomorpha [Family Soricidae (1 species)], were captured monthly from 2008–2009, quarterly during 2010, and biannually during 2012 at Cp Humphreys (11.91%) and the expansion site (28.88%) ( Table 1 ) .
10.1371/journal.pone.0176514.t001
Table 1 Number, capture rate, and percent of small mammals captured, by species, at both Camp Humphreys and the expansion site (under development), monthly 2008–2009, quarterly 2010, and biannually 2012.
Location a (# Traps) b
Species
Apodemus agrarius
Microtus fortis
Micromys minutus
Mus musculus
Myodes regulus
Rattus norvegicus
Crocidura lasiura
Tscherskia triton
Total
Camp Humphreys (5,300)
Number Captured
510
34
12
25
1
0
49
0
631
Capture Rate
9.62
0.64
0.23
0.47
0.02
0.00
0.92
0.00
11.91
Percent Captured
80.82
5.39
1.90
3.96
0.16
0.00
7.77
0.00
Expansion Site (6,000)
Number Captured
1,302
220
29
42
1
3
135
1
1,733
Capture Rate
21.70
3.67
0.48
0.70
0.02
0.05
2.25
0.02
28.88
Percent Captured
75.13
12.69
1.67
2.42
0.06
0.17
7.79
0.06
Total (11,300)
Number Captured
1,812
254
41
67
2
3
184
1
2,364
Capture Rate
16.04
2.25
0.36
0.59
0.02
0.03
1.63
0.01
20.92
Percent Captured
76.65
10.74
1.73
2.83
0.08
0.13
7.78
0.04
a Preexisting Camp Humphreys installation and the projected expansion site under construction (grading and land fill of rice paddies).
b Number of traps set at Camp Humphreys and the expansion site.
Overall, A . agrarius (striped field mouse) accounted for 76.65% (1,812) of all small mammals captured, followed by Microtus fortis (reed vole) (254, 10.74%), Crocidura lasiura (Ussuri white-toothed shrew) (184, 7.78%), Mus musculus (house mouse) (67, 2.83%), Micromys minutus (harvest mouse) (41, 1.73%), Rattus norvegicus (brown rat) (3, 0.13%), Myodes regulus (Royal or Korean red-backed vole) (2, 0.08%), and Tscherskia triton (greater long-tailed hamster) (1, 0.04%). The overall capture rate of A . agrarius for unmanaged tall grass habitats at the expansion site were higher (21.70%) than for limited tall grass habitats at Cp Humphreys (9.62%) ( S1 Fig ) . A . agrarius accounted for 80.82% and 75.13% of all small mammals at Cp Humphreys and the expansion site, respectively. The lower proportion of A . agrarius captured at the expansion site was due to large numbers of canals and flooded fallow rice paddies that are primary habitats for M . fortis .
Capture rates for years and months
Annual capture rates for A . agrarius for both Cp Humphreys and the expansion site for years 2008–2009 were more than 4-fold and 3-fold higher, respectively, than for years 2010 and 2012, when a large portion of unmanaged vegetation was cleared at the expansion site ( Table 2 , S2 Fig ) .
10.1371/journal.pone.0176514.t002
Table 2 Annual capture rates for small mammals, by species, at Camp Humphreys and the expansion site, captured monthly during 2008–2009, quarterly 2010, and biannually 2012.
Collection Site
Year (No. Traps)
Apodemus agrarius
Microtus fortis
Micromys minutus
Mus musculus
Myodes regulus
Rattus norvegicus
Crocidura lasiura
Tscherskia triton
Total
Camp Humphrey a
2008 (1,600)
14.31
0.75
0.44
0.00
0.00
0.00
1.81
0.00
17.31
2009 (2,400)
9.17
0.75
0.21
0.58
0.00
0.00
0.58
0.00
11.29
2010 (1,000)
4.90
0.40
0.00
0.00
0.00
0.00
0.60
0.00
5.90
2012 (300)
2.00
0.00
0.00
1.83
0.17
0.00
0.00
0.00
4.00
Expansion Site b
2008 (2,050)
29.66
0.93
0.73
0.24
0.05
0.00
1.41
0.00
33.02
2009 (2,475)
25.01
7.84
0.36
1.29
0.00
0.12
4.28
0.00
38.91
2010 (1,175)
4.68
5.61
0.50
0.26
0.00
0.00
0.00
0.00
5.87
2012 (300)
6.67
0.33
0.00
0.67
0.00
0.00
0.00
0.33
8.00
Overall Mean
2008 (3,650)
22.93
0.85
0.60
0.14
0.03
0.00
1.59
0.00
26.14
2009 (4,875)
17.21
4.35
0.29
0.94
0.00
0.06
2.46
0.00
25.31
2010 (2,175)
4.78
0.46
0.23
0.14
0.00
0.00
0.28
0.00
5.89
2012 (600)
5.33
0.17
0.00
2.17
0.17
0.00
0.00
0.17
8.00
a Preexisting Camp Humphreys installation where uncut tall grasses/ herbaceous vegetation and shrubs predominated along ditches, streams, and water impoundment areas.
b Expansion site under development (clearing vegetation, grading and land fill) and construction of structures for housing and military operations, roads, drainage systems, and recreational areas.
Monthly capture rates for A . agrarius varied for both Cp Humphreys (range 3.50–17.00%) and the expansion site (range 9.63–55.79%) ( Fig 2 ) .
10.1371/journal.pone.0176514.g002
Fig 2
Monthly capture rates of Apodemus agrarius collected at Camp Humphreys and the new expansion site from 2008–2010 and 2012.
Capture rates for sexes
Apodemus agrarius capture rates for males (46.69%, range 33.33–55.56%) and females (53.31%, range 44.44–66.67%), were not significantly different ( p = 0.062) ( Table 3 ) . Gravid females were only observed from April-November, with the highest gravid rates observed during June (40.00%), August (41.67%), and September (45.16%) and ranged from 9.43–16.21 for the other months ( Table 3 ).
10.1371/journal.pone.0176514.t003
Table 3 The total number and percentage of Apodemus agrarius males and females captured and number females (%) gravid at Camp Humphreys and the new expansion site, monthly from 2008–2009, quarterly 2010, and biannually 2012.
Month
No. Males
(%)
No. Females
(%)
No. Gravid
(%)
January
41
40.59
60
59.40
0
0
February
98
50.00
98
50.00
0
0
March
74
46.83
84
53.16
0
0
April
51
40.80
74
59.20
12
16.21
May
61
42.96
81
57.04
13
16.05
June
48
51.61
45
48.39
18
40.00
July
164
54.85
135
45.15
21
15.56
August
60
55.56
48
44.44
20
41.67
September
77
45.29
93
54.71
42
45.16
October
53
33.33
106
66.67
10
9.43
November
17
50.00
17
50.00
2
11.76
December
102
44.93
125
55.07
0
0
TOTAL
846
46.69
966
53.31
138
14.29
Capture rates for weights
Few A . agrarius weighed ≤10 g (0.61%) or >40 g (2.37%), with most weighing 10–20 g (39.96%), 20–30 g (37.97%) and 30–40 g (19.09%) ( Table 4 ) .
10.1371/journal.pone.0176514.t004
Table 4 The total number and percentage of male and female A . agrarius captured, by weight category, mean weights for each category, and differences of mean male weights—Mean female weights for each weight category during 2008–2010 and 2012.
Categories
≤10 g
10–20 g
20–30 g
30–40 g
>40 g
Total
Number (%) a ♂
6 (54.54)
244 (33.70)
315 (45.78)
242 (69.94)
38 (88.37)
845 (46.63)
Number (%) a ♀
5 (45.45)
480 (66.30)
373 (54.21)
104 (30.06)
5 (11.63)
967 (53.37)
Total (♂,♀) b
11 (0.61)
724 (39.96)
688 (37.97)
346 (19.09)
43 (2.37)
1,812
Mean Weight c ♂
9.3
17.0
24.7
34.2
43.7
25.5
Mean Weight ♀
8.8
16.3
25.3
33.5
42.9
23.2
Difference (♂-♀)
+0.5
+0.7
-0.6
+0.7
+0.8
+2.3
a Percent males or females = number males or females captured for each weight category/total number of males or females collected for each weight category, respectively.
b Percent = total number of males and females captured by weight category/total number captured.
c Mean weights = total weight of all specimens, by weight category/total number of specimens for each category.
Males accounted for 69.94% and 88.37% of those weighing 30–40 g and >40 g (old rodents), while females accounted for 66.30% and 54.21% of those weighing 10–20 g and 20–30 g (young rodents), respectively. On the average, male A . agrarius weighed 0.4–0.8 g more than females for each weight category, except for those weighing 20–30 g where females was heavier by 0.6 g more than the males.
Seasonal variations in population weights coincided with reproductive activity, e.g., increased proportions of lower weight individuals following high gravid rates ( Table 5 ) .
10.1371/journal.pone.0176514.t005
Table 5 Quarterly total number and percentage (%) of A . agrarius captured, by weight category, at Camp Humphreys and the expansion site from 2008–2010 and 2012.
Trapping Season
Weight Class (g)
Total (%)
≤ 10
10–20
20–30
30–40
>40
Winter (Jan-Mar)
0
274 (60.21)
159 (34.95)
21 (4.61)
1 (0.22)
455 (25.11)
Spring (Apr-Jun)
4 (1.11)
92 (25.56)
202 (56.11)
57 (15.83)
5 (1.38)
360 (19.86)
Summer (Jul-Sep)
5 (0.87)
86 (14.90)
224 (38.82)
230 (39.86)
32 (5.55)
577 (31.84)
Fall (Oct-Dec)
2 (0.48)
272 (64.76)
103 (24.52)
38 (9.05)
5 (1.19)
420 (23.18)
TOTAL (%)
11 (0.61)
724 (39.96)
688 (37.97)
346 (19.09)
43 (2.37)
1,812
The proportion of A . agrarius weighing ≤20 g declined from a high of 82.18% in January to a low of 10.37% by July before increasing to a high of 71.37% by December ( Fig 3 ) . In contrast, the proportion of A . agrarius weighing >30 g increased from a low of 1.98% in January to a high of 51.76% by September as a result of a maturing populations, gravid females, and abundant food supply, but rapidly declined to 1.76% by December that followed high numbers of gravid females during August (41.67%) and September (45.16%).
10.1371/journal.pone.0176514.g003
Fig 3
Overall percentage of Apodemus agrarius captured monthly, by weight, at Camp Humphreys and the new expansion site from 2008–2010 and 2012.
Serological prevalence of HTNV at Cp Humphreys and the expansion site, Pyeongtaek
Among 1,812 A . agrarius captured at Cp Humphreys and the expansion site, IFAT showed that 39 (2.15%) rodents were positive for anti-HTNV IgG. HTNV Ab+ rates among A . agrarius were significantly higher for Cp Humphreys (2.41%, range 0.0–7.89%) than for the expansion site (2.11%, range 0.0–3.51%) (χ 2 = 17.279, df = 1, P<0.001) ( Fig 4 ) . However, based on trap rates (Cp Humphreys, 9.62; expansion site, 21.70), the number of HTNV Ab+ A . agrarius captured/100 traps was 2-fold greater at the expansion site (0.46) compared to Cp Humphreys (0.23). The quarterly seasonal proportions of HTNV Ab+ A . agrarius for all weight categories at Cp Humphreys and the expansion site varied from 0.95% (Oct-Dec) to 2.86% (Jan-Mar) ( Table 6 ) .
10.1371/journal.pone.0176514.g004
Fig 4
Overall percentage of Apodemus agrarius captured monthly at Camp Humphreys and the new expansion site that were antibody-positive for Hantaan virus.
10.1371/journal.pone.0176514.t006
Table 6 Quarterly number (%) of A . agrarius (total number HTNV Ab+/total number captured, by weight category, at Camp Humphreys and the expansion site, Pyeongtaek, Gyeonggi Province, 2008–2010 and 2012.
Trapping Season
Weight Class (g)
Total (%)
≤ 10
10–20
20–30
30–40
>40
Winter (Jan-Mar)
0
8/274 (2.92)
5/159 (3.14)
0/21 (0.0)
0/1 (0.0)
13/455 (2.86)
Spring (Apr-Jun)
0/4 (0.0)
0/92 (0.0)
5/202 (2.48)
1/57 (1.75)
0/5 (0.0)
6/360 (1.67)
Summer (Jul-Sep)
0/5 (0.0)
2/86 (2.33)
4/224 (1.79)
6/230 (2.61)
4/32 (12.50)
16/577 (2.77)
Fall (Oct-Dec)
0/2 (0.0)
1/272 (0.37)
2/103 (1.94)
1/38 (2.63)
0/5 (0.0)
4/420 (0.95)
TOTAL (%)
0/11 (0.0)
11/724 (1.52)
16/688 (2.33)
8/346 (2.31)
4/43 (9.30)
39/1,812 (2.15)
Overall, monthly HTNV Ab+ rates for males (2.25%) and females (2.07%) were similar, but varied monthly from 0.0–4.08% for males and 0.0–5.95% for females ( Fig 5 ) . The proportion of A . agrarius males and females that were serologically positive for HTNV for weight categories 10–20 g (1.52%), 20–30 g (2.33%), and 30–40 g (2.31%) were similar ( Fig 6 ) . However, the serological positivity (9.30%) of HTNV Ab for A . agrarius weighting > 40 g (oldest rodents) was only observed in males and was significantly higher than the other weighted groups (One-way ANOVA test, p<0.0001).
10.1371/journal.pone.0176514.g005
Fig 5
Overall monthly percentage of male and female Apodemus agrarius captured at Camp Humphreys and the new expansion site that were antibody-positive for Hantaan virus, 2008–2010 and 2012.
10.1371/journal.pone.0176514.g006
Fig 6
Overall percent of Apodemus agrarius , by sex, captured at Camp Humphreys and the new expansion site that were HTNV Ab+ for each weight category (lines) and percent HTNV Ab+ within each weight category (bars).
Molecular prevalence of HTNV at Cp Humphreys and the expansion site, Pyeongtaek
To identify HTNV RNA in the seropositive A . agrarius , RT-PCR was performed by targeting the G C glycoprotein-encoding M segment (373 bps). A total of 11 (28.21%) A . agrarius harbored HTNV RNA in 39 HTNV Ab+ rodents. The molecular prevalence of HTNV for males and females was 26.32% (5/19) and 30.00% (6/20), respectively ( Table 7 ) .
10.1371/journal.pone.0176514.t007
Table 7 Serological and molecular prevalence of HTNV from A . agrarius captured, by sex, weight, season category, at Camp Humphreys and the expansion site from 2008–2010 and 2012.
Categories
Seropositive rate for anti-HTNV IgG antibody (%)
HTNV RNA positivity (%)
Sex (n = 1,812)
Males
19/845 (2.25)
5/19 (26.32)
Females
20/967 (2.07)
6/20 (30.00)
Weight (n = 1,812)
≤10 g
0/11
0
10–20 g
11/724 (1.52)
2/11 (18.18)
20–30 g
16/688 (2.33)
6/16 (37.50)
30–40 g
9/346 (2.60)
2/9 (22.22)
>40 g
3/43 (6.98)
1/3 (33.33)
Season (n = 1,812)
Winter (Jan-Mar)
7/455 (1.54)
4/7 (57.14)
Spring (Apr-Jun)
20/360 (5.56)
3/20 (15.00)
Summer (Jul-Sep)
8/577 (1.39)
2/8 (25.00)
Fall (Oct-Dec)
4/420 (0.95)
2/4 (50.00)
The proportion of HTNV-positive A . agrarius for weight categories 10–20 g (18.18%), 20–30 g (37.50%), 30–40 g (22.22%), and 40 g (33.33%) was similar. Seasonal positivity of HTNV in A . agrarius showed 57.1% (Jan-Mar), 15% (Apr-Jun), 25% (Jul-Sep), and 50% (Oct-Dec).
The partial sequences of HTNV M segment (328nt length) were trimmed and used for analysis. All HTNV strain sequences were submitted to GenBank (Accession numbers; KX119152-119162). The partial M segment sequences (coordinates 1,994 to 2,321) of 11 HTNV strains from Cp Humphreys and the expansion site were phylogenetically compared to HTNV strains previously identified in military training sites, northern Gyeonggi province ( Fig 7 ) . The nucleotide and amino acid homologies of the 11 HTNV strains from Cp Humphreys and the expansion site varied between 0–3.1% and 0–2.8%, respectively.
10.1371/journal.pone.0176514.g007
Fig 7
Phylogenetic analysis of Hantaan virus strains identified in Cp Humphreys and the new expansion site, Pyeongtaek, based on a 328-nt region of the G C glycoprotein-encoding M segment.
The phylogenetic tree was generated by Neighbor-joining (NJ) method. Branch lengths are proportional to the number of nucleotide substitutions, while vertical distances are for clarity. The numbers at each node are bootstrap probabilities (expressed as percentages), as determined for 1000 iterations (GenBank accession numbers; KX119152-119162).
Quantitative RT-PCR, whole-genome sequencing, and phylogenetic analyses
To obtain whole genome sequences of HTNV in serological and molecular positive A . agrarius from Cp Humphreys and the extension region, Pyeongtaek, HTNV RNA copies were quantified in the lung tissue by RT-qPCR. The threshold of Cycle (Ct) value are shown in the Fig 8 . Aa09-198 demonstrated the lowest Ct value (highest viral loads), followed by Aa08-1111 and Aa09-189. The whole genome sequences of the three HTNV strains were recovered by conventional RT-PCR and RACE PCR for both 3’ and 5’ end sequences. The whole genome sequences of the HTNV strains deposited in GenBank (Accession numbers; KY594712- KY594720).
10.1371/journal.pone.0176514.g008
Fig 8
Determination of threshold cycle (Ct) values of Hantaan virus RNA in HTNV-positive Apodemus agrarius collected at Camp Humphreys and the new expansion site.
HTNV RNA was examined for the HTNV S segment in anti-HTNV IgG seropositive and HTNV RNA positive rodents (n = 11). The vertical axis represents the Ct value of HTNV S segment RNA.
The genetic diversity and phylogenetic relationship of HTNV in Cp Humphreys and the extension region were determined in comparison to strains obtained from lung tissue of seropositive rodents previously captured at a variety of HFRS-endemic areas, e.g. Twin Bridge Training Area (TBTA) North, TBTA South, and Dagmar North in Paju, Yeoncheon, and Pocheon ( Fig 9 ) . The L segment of the HTNV formed an independent outgroup of all of HTNV in Gyeonggi province. The phylogenetic analysis of the M segment showed a well-supported genetic lineage with HTNV 76–118. The S segment formed a geographic-specific group within HTNV strains, including HTNV 76–118, in Gyeonggi province.
10.1371/journal.pone.0176514.g009
Fig 9
Phylogenetic analyses of the whole genome sequences of Hantaan virus L, M, and S segments identified at Camp Humphreys and the new expansion site, Pyeongtaek, Gyeonggi province.
The phylogenetic tree was generated by Maximum-likelihood (ML) method. The phylogeny of the L segment (a), M segment (b), and S segment (c) is described. Branch lengths are proportional to the number of nucleotide substitutions, while vertical distances are for clarity. The numbers at each node are bootstrap probabilities (expressed as percentages), as determined for 1000 iterations. The phylogenetic positions of HTNV strains are shown in relationship to representative Aa10-434 (L segment, KT934970; M segment, KT935004; S segment, KT935038), Aa10-518 (L segment, KT934971; M segment, KT935005; S segment, KT935039), Aa14-204 (L segment, KT934977; M segment, KT935011; S segment, KT935045), Aa14-207 (L segment, KT934978; M segment, KT935012; S segment, KT935046), Aa05-331 (L segment, KT934962; M segment, KT934996; S segment, KT935030), Aa05-771 (L segment, KT934963; M segment, KT934997; S segment, KT935031), Aa09-410 (L segment, KU207177; M segment, KU207185; S segment, KU207193), Aa09-948 (L segment, KT934966; M segment, KT935000; S segment, KT935034), Aa14-172 (L segment, KT934974; M segment, KT935008; S segment, KT935042), Aa14-188 (L segment, KT934975; M segment, KT935009; S segment, KT935043), HTNV 76–118 (L segment, X55901; M segment, M14627; S segment, M14626) and HV004 (L segment, JQ083393; M segment, JQ083394; S segment, JQ083395).
Discussion
Cp Humphreys was designated a major US military hub with an estimated final US military and civilian population of >20,000 personnel. To accommodate for the increased population and expansion for military operations and outdoor recreational areas, adjacent lands that consisted mostly of low-lying rice paddies were purchased. The resulting environmental modifications of purchased properties included sequential grading of terrain and filling of low-lying fallow rice paddies for the construction of roads, ditches, major drainage system/recreational areas, military housing, schools, hospital and medical clinics, and other structures designed for military operations
Apodemus agrarius is associated with unmanaged lands characterized by abundant grasses/herbaceous vegetation in rural areas, including military training sites [ 24 – 27 ]. Similar to this survey and other annual and multi-year surveys, A . agrarius was the most commonly collected small mammal at US and ROK operated military training sites and installation field environments [ 4 , 6 – 10 ]. Compared to the expansion site of unmanaged grasses during 2008–2009, rodent populations were much higher than for Cp Humphreys where habitat was often limited to narrow strips of unmanaged vegetation along drainage systems/holding ponds. While capture rates of A . agrarius associated with the expansion site were high and movement of large trucks that created dusts on dirt, gravel, and hardened roads, the transmission risks of HTNV were reduced by very low HTNV Ab+ rates. Although no cases were reported among US military and civilian personnel, there may have been cases among local contractors and truck drivers that we were not aware of since these cases were not reported through the military medical system.
The Korea Centers for Control and Prevention (KCDC) [ 28 ] reports approximately 400–500 cases of HFRS annually, which are mainly caused by HTNV and SEOV. HTNV is the most common causative agent of HFRS in rural areas of the ROK and is characterized by severe medical manifestations and high mortality rate (9.46%) among US military personnel in Korea with good quality medical care from 1986–2014 [ 2 , 4 – 15 ]. In Korea, human infections of HTNV among military members are usually associated with high populations of A . agrarius in field environments or mice-infested vacant buildings in combination with “dust-creating” activities (e.g., back-blast from artillery, convoy operations, and track and wheeled vehicle maneuvers/operations in field environments), while SEOV infections are usually associated with urban environments activities (e.g., dry sweeping or vacuuming rodent infested buildings) where R . norvegicus predominates [ 3 , 5 ]. While HFRS caused by HTNV infections poses a serious health threat in Korea, it is classified by the US National Medical Intelligence Center (NCMI) as a rare disease, frequently occurring is small clusters. The most recent cluster among US military personnel deployed to the ROK was observed in 2005, when three US soldiers acquired HFRS at TBTA associated with exposure of contaminated dusts in wheeled vehicle cabs (cavalry unit) [ 3 ]. During the same year, another HFRS case was acquired at Firing Point 60, Yeoncheon, associated with the back-blast of artillery. More recently (Nov., 2014) a single case was reported for a US soldier conducting convoy and driver’s training at Dagmar North when HTNV seropositive rates in A . agrarius were 19.3% (considered a HFRS high-risk area), 27–30 days post-exposure that preceded infection [ 29 ]. The epidemiology of these cases was only accomplished through comparative analysis of the HTNV RNA from HFRS patients and associated rodents where the soldiers had trained, as the HTNV varies geographically [ 3 ]. Since 1986, only one case of SEOV has been reported from a US Airman that was vacuuming a rat-infested building and who had a relatively mild case of HFRS [ 5 ]. R . norvegicus , the primary reservoir for SEOV, is routinely captured by the Department of Public Works near housing and other facilities at Cp Humphreys. These resources would provide risk analyses for SEOV risks among US populations residing or working in buildings infested with rats.
Similar to rodent-borne disease surveillance conducted at training sites near the DMZ and other US military installations, A . agrarius was the most frequently collected small mammal [ 4 , 6 – 10 ]. Low to moderate A . agrarius capture rates were reported for limited tall grass habitats at Cp Humphreys and were similar to capture rates observed at other installations, e.g., Osan, Gunsan, and Gwangju Air Bases (unpublished data). During 2008–2009, high capture rates were observed for expansive tall grasses/herbaceous vegetation habitats at the expansion site and were similar to capture rates observed for expansive tall grass habitats at US and ROK operated training sites near the DMZ [ 4 , 6 – 10 ]. However, trap rates were significantly lower during 2010 and 2012 following grading and removal of much of the vegetation from the landscape that provided food and harborage for small mammals. Additionally, in part, the decline may have been due to over predation as the predator populations (e.g., raccoon dogs, feral cats, weasels, and predatory birds) were pushed into space-limited habitats surrounded by farming activities and urban environments. Over time, predator populations will likely stabilize based on available food sources and small mammal populations may rebound to near previous levels for undisturbed areas.
HFRS risks are associated with a combination of factors, including: environmental, reservoir host bionomics, and types of human exposure. Overall, HTNV Ab+ rates for Cp Humphreys/expansion site and Osan Air Base (50 km south of Seoul and 20 km north of Pyeongtaek), are very low, usually ≤6%, when compared to US and ROK operated training areas near the DMZ where seasonal HTNV Ab+ rates varied up to 60% during monthly surveys and overall annual rates varied from 15% to 25% [ 4 , 7 , 9 , 10 ]. Limited surveys at Gunsan and Gwangju Air Bases, near the southern tip of the Korean Peninsula, were indicative of low small mammal populations, as well as none of the A . agrarius were HTNV Ab+ (TA Klein, personal communication). The reason for high HTNV Ab+ rates near the DMZ that decrease over distance to the tip of the peninsula is not understood, but may be related to reproductive behaviors. For training areas near the DMZ, there were observed low reproductive periods during the winter (0–0.3%), followed by a late spring increase in reproduction (4.2–24.6%) (Apr-May), low reproduction during the summer (0–1.3%), and very high reproduction in the late summer/early fall (27.3–70.0%) (Aug-Sep). A large influx of HTNV naïve mice observed at training areas near the DMZ during the fall/early winter periods when temperatures become cooler and habitat is shrinking as vegetation dies likely results in increased territorial disputes, wounding, and higher rates of HTNV transmission [ 30 ]. At Cp Humphreys and the expansion site, peak numbers of gravid females were observed earlier (June, 40.0%) and similarly in August and September (42.5 and 25.2%, respectively), while moderate numbers of gravid females were observed during the early spring (April/May, 16.0–16.2%), July (15.6%), and early winter (October/November, 9.4–11.8%) [ 4 , 6 – 10 ]. In the southern area, young naïve A . agrarius broods throughout the summer may reduce territorial disputes in the fall due to relatively sufficient habitat and food. This proposed decreased movement and competition of naïve young rodents at Cp Humphreys and the expansion site may impact negatively on rodent-to-rodent HTNV transmission and result in lower HTNV Ab+ rates than those observed at the military training areas located near the DMZ, northern Gyeonggi province [ 31 ]. Additionally, the greatly reduced numbers of A . agrarius during 2010 likely reduced the potential for acute infections and corresponding viral shedding during the late fall/early winter when the majority of HFRS cases are reported. Although gravid females were observed throughout the early spring and summer, similar to training areas near the DMZ, the overall age (based on weight) of the population increased through September before rapidly declining as a result of the influx of young naïve rodents during the late fall reproductive cycle. By January, much of the population (based on weight) was replaced by young mice born during the late fall, indicating that the life span of A . agrarius live is approximately one year [ 24 ].
The overall HTNV seropositive rates of Cp Humphreys were higher than observed for the extensive tall grass habitats for undisturbed fallow rice paddies of the expansion area. However, the numbers of HTNV seropositive mice/100 traps were nearly 2-fold greater for the expansion site compared to Cp Humphreys, thereby increasing HFRS risks associated with less disturbed and unmanaged lands. The movement of potentially contaminated soil and vegetation and soil covered concrete roads created the potential for contaminated dusts and HTNV infections, especially for truck drivers and construction site monitors and workers. HTNV risks, while present at Cp Humphreys, are very low as a result of hard surface roads and recreation sites with short-cut grasses in the center, greatly reducing HTNV reservoir host habitats.
In this study, 39 (2.15%) of 1,812 A . agrarius were HTNV seropositive. The partial genome sequence of HTNV M segment was identified from 11 (28.21%) rodent lung tissues of HTNV Ab+ samples. RT-qPCR results showed varied viral loads in both sero- and molecular positive samples. The whole genome sequences of HTNV tripartite RNA were obtained from Aa08-1111, Aa09-189, and Aa09-198 that contained higher number of HTNV RNA copies. The termini of 3’ and 5’ sequences were determined by RACE PCR. Both end sequences of HTNV L, M, and S segments contained a mismatch at 9 th and the noncanonical U-G pair at 10 th nucleotides, suggesting the incomplete complementarity as previously described [ 32 ]. The total length of HTNV L segment for Cp Humphreys and the expansion site, Pyeongtaek, was three nucleotides shorter (6,530nt) than that of HTNV 76–118, demonstrating the deletion of 5’-AUC-3’ at the 5’ end of the L segment. U at the 12 th nucleotide on the M segment was defined compared to that on the HTNV 76–118 M segment.
The phylogenetic analyses of HTNV strains from Cp Humphreys and the extension site demonstrated a greater diversity of the rodent-borne hantavirus; the L segment showed distinct outgroup from entire HTNV strains, previously described in Gyeonggi province. The M segment formed a genetic cluster with HTNV 76–118, while the S segment was a geographic lineage within HTNV strains in Gyeonggi province. The natural reassortment and recombination of HTNV tripartite RNA genomes were observed near DMZ areas, northern Gyeonggi province [ 29 , 33 ]. Thus, the phylogenetic position and characterization of HTNV in Pyeongtaek will be clarified when additional genomic sequences of HTNV are acquired in southern areas of Korean peninsula.
A total of 10 (5.43%) C . lasiura were positive for MJNV, which was identified from shrews distributed in ROK and China, and the sera do not cross react with other rodent-borne hantaviruses [ 17 ]. Recently, there was a report that African shrew-borne hantaviruses were likely to infect humans [ 34 ]. Whether MJNV in C . lasiura poses a human health threat remains to be investigated. A total of 2/254 (0.79%) M . fortis and 1/41 (2.44%) M . minutus were serologically positive for hantaviruses, which was likely the result of interspecies transmission of HTNV since tissues were negative for hantaviruses by RT-PCR.
In summary, the characterization of US military installations undergoing expansion, in combination with small mammal surveillance, provides epidemiological information for the relative abundance of reservoir populations, hantavirus Ab+ rates, and other bionomic and environmental factors that are necessary to identify potential HTNV transmission risks. These transmission risks combined with human activities and exposure, which can be applied for disease risk analyses, are essential to the process of developing strategies for disease prevention. Comprehensive and long term rodent-borne disease surveillance should be the goal of US military preventive medicine to not only identify changes in HFRS disease risks due to modification of feral lands, but subsequently to better understand HFRS disease risks to soldiers, civilians, and family members residing and/or working on the installation. The whole genome sequences of HTNV at Cp Humphreys and the extension site show a greater diversity of rodent-borne hantaviruses in the ROK. Taken together, these data provide the robust impact to increase our knowledge of military activities, environmental conditions, and the genetic diversity of HTNV that can be applied to strategies to improve land management, disease risk mitigation, and the understanding of hantavirology.
Supporting information
S1 Fig
Overall capture rates (%) for rodents and soricomorphs captured at Camp Humphreys and the new expansion site from 2008–2010 and 2012.
(TIF)
S2 Fig
Annual capture rates and overall mean capture rates for years 2008–2010 and 2012 (bars) for rodents and soricomorphs captured at both Camp Humphreys and the new expansion site.
Numbers are the overall mean capture rates, by species, for all years (2008–2010 and 2012).
(TIF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Cutaneous leishmaniasis (CL) is an important neglected tropical skin disease (skin NTD) of public health importance and is the commonest form of leishmaniasis, characterized by skin lesions which may result in ulcers, scars, disability and stigma [ 1 , 2 ]. Globally, it is estimated that between 0.7 to 1.3 million new cases of CL are reported annually [ 3 ].
A localized outbreak of skin ulcers suspected to be cases of CL was first reported in Ghana from the Ho municipality of the Volta Region in 1999 based on the identification of Leishmania amastigotes in some skin lesion biopsies [ 4 ]. Subsequent studies have identified L . major , uncharacterized Leishmania species, and recently, members of the Leishmania enriettii complex from suspected CL cases in the Ho municipality, suggesting a complex epidemiology of CL in the region [ 5 – 7 ].
Although the Oti region has been part of the Volta region until the year 2019, no previous confirmation of CL cases had been made there. This study was therefore initiated following reports of skin ulcers which were suggestive of CL in some communities of the Oti region, after leishmanin skin test (LST) had been conducted to establish Leishmania infection (reported elsewhere).
Materials and methods
Ethics statement
Ethical approval to conduct this study was obtained from the ethics review committee of the Ghana Health Service (GHS-ERC006/08/18). Written informed consent was obtained from all study participants. For participants under 18 years, written consent was obtained from a parent or guardian.
Study design
This study was based on a cross-sectional study design approach. The study was conducted from October to December 2018 in three communities of the Oti region of Ghana having at least three suspected cases of active CL (ACL). A suspected ACL lesion was defined clinically as any open ulcer with diameter bigger than 5mm. Prevalence of CL among study participants with skin ulcers was investigated. Demographic and epidemiological data were obtained by a structured interviewer administered questionnaire.
Study area
This study was conducted in the following three communities of the Oti region of Ghana: Ashiabre, Keri, and Sibi Hilltop. Ashiabre is in the Tutukpene sub-district of the Nkwanta South municipality while Keri is in the Keri sub-district of the municipality. Sibi Hilltop is in the Sibi sub-district of the Nkwanta North district of the region.
The population of Nkwanta South municipality is estimated to be 117,878 with males constituting 49.6% of the population. Covering a land area of approximately 2733 km 2 , the Nkwanta South municipality is located between latitudes 7 o 30’ and 8 o 45’ North and longitude 0 o 10’ and 0 o 45’East [ 8 ].
The population of the Nkwanta North district is estimated to be 64,553 with males constituting 50.2% of the population. The district is located between Latitude 7°30’N and 8°45’N and Longitude 0°10’W and 045’E. It shares boundaries with Nkwanta South municipality to the south, Nanumba South to the north, Republic of Togo to the east, and Kpandai District to the west [ 9 ].
Inclusion criteria
Eligible study participants were residents in the study community for ≥ 12 months, aged between 2 to 65 years (inclusive).
Sample size considerations
For active case detection, assuming a current CL prevalence (P) of 22.1% [ 4 , 10 ], Z 2 = (1.96) 2 for 95% confidence interval D 2 = maximum 0.05, a minimum sample size (N) of 265 individuals was required for screening for active case detection using the formula:
N = ((Z) 2 P/D 2 ) *(1-P)
Selection of households for study inclusion
Using a sorted list of households, 200 households (with an average of 5–7 persons per household) were selected for study inclusion in each study community using a systematic sampling approach. For this study, a household was defined as a person or a group of persons, who live together in the same house or compound and share the same house-keeping arrangements. The head of each household was defined as a male or female member of the household recognised as such by the other household members. The head of a particular household is generally the person with economic and social responsibility for the household. As a result, household relationships were defined with reference the household head [ 11 ]. The community household list was obtained for each study community based on a household census. The number of households per study community determined by household census was 945, 795, and 1184 in Ashiabre, Keri, and Sibi Hilltop respectively.
A sampling interval I was determined, where I = N/n with N being the sum of individual households in the study community while n was the number of households to be selected. The I was rounded to 2 decimal places.
Using Microsoft excel, the RANDBETWEEN command was used to generate a random decimal integer R between 0 and 1 rounded up to two decimal points. The sequence of households that were selected in each study community were R*I, R*I + I, R*I +2*I, R*I +3*I,….R*I + (n -1 )*I, each rounded up to the next highest whole number [ 12 ]. With the assistance of community-based volunteers, the selected households were identified after which all members of the selected households aged 2 to 65 years were invited to participate in the study, using a door-to-door invitation approach. Because the invitation to participate in the study was extended to households, a household was not included in the study if the household head declined to allow his or her household to participate in the study. However, the agreement of the household head did not make it compulsory for every household member of age 2 to 65 years to participate in the study. Each household member was given the opportunity to go through the informed consent process to decide whether they wish to participate or not.
Sampling of suspected active cutaneous leishmaniasis (ACL) lesions (ulcers)
Each study participant was asked to disclose the occurrence of any skin ulcer(s) on their body. Interviewers also examined the exposed parts of participants body such as legs, arms, neck, and face to identify any suspected active CL (ACL) lesion(s). The location, size, and duration of each suspected ACL was documented. For each suspected ACL lesion, a non-invasive diagnostic sampling technique using sequential tape strips with a diameter of 22 mm (D-Squame, CuDerm Corporation, Texas, USA) was used to obtain samples for subsequent DNA isolation [ 13 ].
For the non-invasive skin sampling, one tape disc was placed on each suspected skin lesion after which even pressure was applied to the disc on the lesion using a plunger which was gently held on the disc and pressed for approximately 20 seconds. The tape disc was then detached and transferred into a sterile 1.5ml Eppendorf vial and stored at 4°C for transportation to the laboratory for further analysis ( Fig 1 ). Participants received standard wound care after sample collection.
10.1371/journal.pntd.0009416.g001
Fig 1
Non-invasive sampling of skin lesions.
DNA isolation from tape strip disc and PCR amplification of Leishmania species
DNA extraction was performed using SpeedTools Tissue DNA Extraction Kit (Biotools, Inc).
A nested polymerase chain reaction (Ln-PCR) approach was used to amplify DNA of Leishmania species from the human skin lesions following an adaptation of the protocol by Cruz et al., 2002 [ 14 ], with the target being the small subunit ribosomal ribonucleic acid (SSU rRNA) gene. Positive control used was Leishmania infantum (JPC strain) with distilled water as negative control.
Data management
Data was managed using Microsoft Access software version 2013 and analyzed using STATA software version 14. Association between nominal variables was assessed using Pearson’s chi square test of association and Fishers exact test. All statistical tests were performed at a 95% confidence level.
Results
Of 600 households (200 in each study community) invited to participate in this study, a total of 587 households comprising 189 (32.2%), 200 (34.1%), and 198 (33.7%) from Ashiabre, Keri and Sibi Hilltop respectively, were included in this study. The study households had a total of 3718 members out of which 3,440 (92.5%) consisting of 1,194, 941, 1305 from Ashiabre, Keri, and Sibi Hilltop respectively were enrolled in the study.
The average household size was 6.3 with a range of 1 to 18 household members. Ashiabre and Sibi Hilltop had an average household size of 7 while Keri had an average household size of 5.
Out of 3440 persons physically examined for ulcers, a total of 595 skin ulcers were observed on 426 (12.4%) ( Table 1 ). Of the 426 persons, 314 (73.7%) were within the age group 5–15 years while those under five constituted 13.6%. The number of skin ulcers observed on the participants ranged from 1 to 7 with those having one ulcer (47.1%) and two ulcers (27.6%) being the majority. Although skin ulcers were observed on various parts of the participants’ body, majority occurred on the lower legs (71.3%) and feet (17.1%). In Ashiabre, Keri, and Sibi Hilltop, 65.2%, 70.1%, and 74.3% of persons with skin ulcers had the ulcer on their lower legs respectively ( Table 1 ).
10.1371/journal.pntd.0009416.t001
Table 1 Individuals with skin ulcers, ulcers sampled and result of Leishmania PCR test.
Characteristic
Category
Ashiabre
Keri
Sibi Hilltop
Total
n
%
n
%
n
%
n
%
P value
Age of individuals with skin ulcers
<5 years
12
21.4
22
11.7
24
13.2
58
13.6
0.141
5–15 years
35
62.5
145
77.1
134
73.6
314
73.7
16–45 years
9
16.1
19
10.1
18
9.9
46
10.8
>45 years
0
0
2
1.1
6
3.3
8
1.9
Total
56
100
188
100
182
100
426
100
Sex of individuals with skin ulcers
Male
36
64.3
109
58
110
60.4
255
59.9
0.684
Female
20
35.7
79
42
72
39.6
171
40.1
Total
56
100
188
100
182
100
426
100
Number of Skin ulcers tested
1
31
44.9
116
41.3
133
54.3
280
47.1
<0.001
2
28
40.6
84
29.9
52
21.2
164
27.6
3
8
11.6
46
16.4
37
15.1
91
15.3
4
0
0
30
10.7
11
4.5
41
6.9
5
2
2.9
5
1.8
0
0
7
1.2
6
0
0
0
0
5
2
5
0.8
7
0
0
0
0
7
2.9
7
1.2
Total
69
100
281
100
245
100
595
100
Skin ulcer locations
Face/Head
3
4.3
5
1.8
6
2.4
14
2.4
0.029
Upper arm
0
0
2
0.7
0
0
2
0.3
Lower arm
1
1.4
13
4.6
11
4.5
25
4.2
Palm/Back of palm
0
0
2
0.7
3
1.2
5
0.8
Chest
0
0
1
0.4
0
0
1
0.2
Back (upper part below neck))
0
0
0
0
2
0.8
2
0.3
Stomach
2
2.9
0
0
0
0
2
0.3
Buttocks
1
1.4
0
0
2
0.8
3
0.5
Thighs
1
1.4
8
2.8
6
2.4
15
2.5
Lower legs(crus/cnemis)
45
65.2
197
70.1
182
74.3
424
71.3
Feet
16
23.2
53
18.9
33
13.5
102
17.1
Total
69
100
281
100
245
100
595
100
Leishmania pcr result
Negative
55
79.7
219
77.9
171
69.8
445
74.8
0.061
Positive
14
20.3
62
22.1
74
30.2
150
25.2
Total
69
100
281
100
245
100
595
100
PCR test of the 595 ulcer samples indicated that 150 (25.2%) of them were Leishmania positive. In the study communities, 14 (20.3%), 62 (22.1%), and 74 (30.2%) of skin ulcers tested from Ashiabre, Keri, and Sibi Hilltop respectively were positive for Leishmania ( Table 1 ).
Of the 595 ulcer samples tested, 365 (61.3%) were obtained from males while 90 (60.0%) of the 150 Leishmania positive samples were also obtained from males. Also, 437 (73.4%) of the ulcer samples tested as well as 112 (74.7%) of the Leishmania positive ulcer samples were obtained from people within the age group 5–15 years ( Table 2 ).
10.1371/journal.pntd.0009416.t002
Table 2 Skin ulcers tested for Leishmania parasite using PCR by age and sex.
Sex
Age
Number of skin
Leishmania positive ulcers
ulcers tested
n (%)
Males
< 5 years
48
8 (16.7)
5–15 years
276
70 (25.4)
16–45 years
36
10 (27.8)
>45 years
5
2 (40.0)
Subtotal
365
90 (24.7)
Females
< 5 years
45
13 (28.9)
5–15 years
161
42 (26.1)
16–45 years
19
4 (21.1)
>45 years
5
1 (20.0)
Subtotal
230
60 (26.1)
Total
< 5 years
93
21 (22.6)
5–15 years
437
112 (25.6)
16–45 years
55
14 (25.5)
>45 years
10
3 (30.0)
Total
595
150 (25.2)
The 150 Leishmania positive ulcer samples were obtained from 136 study participants of which 123 (90.4%) had single Leishmania positive skin ulcer, 12 (8.8%) had two Leishmania positive skin ulcers and 1 person had three Leishmania positive skin ulcers ( Table 3 ). Majority of individuals with Leishmania positive ulcers were within the age group of 5–15 years (73.5%) followed by children under five (14.0%) and persons aged 16–45 years (10.3%). Across the study sites and among males and females respectively, majority of persons with Leishmania positive skin ulcer(s) were within the age group 5–15 years ( Table 3 ).
10.1371/journal.pntd.0009416.t003
Table 3 Distribution of individuals with Leishmania positive skin ulcers by age, sex, and community of residence.
Characteristic
Category
Ashiabre
Keri
Sibi Hilltop
Total
Male (%)
Female (%)
Male
Female
Male
Female
Male
Female
Total
Individuals with one Leishmania positive skin ulcer
<5 years
1 (25.0)
1 (12.5)
3 (9.4)
5 (27.8)
4 (9.8)
3 (15.0)
8 (10.4)
9 (19.6)
17 (13.8)
5–15 years
3 (75.0)
6 (75.0)
23 (71.9)
11 (61.1)
31 (75.6)
15 (75.0)
57 (74.0)
32 (69.6)
89 (72.4)
16–45 years
0
1 (12.5)
6 (18.8)
1 (5.6)
4 (9.8)
2 (10.0)
10 (13.0)
4 (8.7)
14 (11.4)
>45 years
0
0
0 (0)
1 (5.6)
2 (4.9)
0
2 (2.6)
1 (2.2)
3 (2.4)
Sub total
4 (100)
8 (100)
32 (100)
18 (100)
41 (100)
20 (100)
77 (100)
46 (100)
123 (100)
Individuals with two Leishmania positive skin ulcers
<5 years
0
0
0
1 (33.3)
0
1 (25.0)
0
2 (28.6)
2 (16.7)
5–15 years
1 (100)
0
3 (100)
2 (66.7)
1 (100)
3 (75.0)
5 (100)
5 (71.4)
10 (83.3)
16–45 years
0
0
0
0
0
0
0
0
0
>45 years
0
0
0
0
0
0
0
0
0
Sub total
1 (100)
0
3 (100)
3 (100)
1 (100)
4 (100)
5 (100)
7 (100)
12 (100)
Individuals with three Leishmania positive skin ulcers
<5 years
0
0
0
0
0
0
0
0
0
5–15 years
0
0
0
0
1 (100.0)
0
1 (100.0)
0
1 (100.0)
16–45 years
0
0
0
0
0
0
0
0
0
>45 years
0
0
0
0
0
0
0
0
0
Sub total
0
0
0
0
1 (100)
0
1 (100)
0
1 (100)
Individuals with Leishmania positive skin ulcer(s)
<5 years
1 (20.0)
1 (12.5)
3 (8.6)
6 (28.6)
4 (9.3)
4 (16.7)
8 (9.6)
11 (20.8)
19 (14.0)
5–15 years
4 (80.0)
6 (75.0)
26 (74.3)
13 (61.9)
33 (76.7)
18 (75.0)
63 (75.9)
37 (69.8)
100 (73.5)
16–45 years
0
1 (12.5)
6 (17.1)
1 (4.8)
4 (9.3)
2 (8.3)
10 (12.0)
4 (7.5)
14 (10.3)
>45 years
0
0
0 (0)
1 (4.8)
2 (4.7)
0
2 (2.4)
1 (1.9)
3 (2.2)
Sub total
5 (100.0)
8 (100.0)
35 (100)
21 (100)
43 (100)
24 (100)
83 (100)
53 (100)
136 (100)
The overall prevalence of cutaneous leishmaniasis ( Leishmania infection observed among those with skin ulcers) was 31.9% (136/426) with prevalence of 23.2% (13/56), 29.8% (56/188), and 36.8% (67/182) observed in Ashiabre, Keri and Sibi Hilltop respectively.
The average size of the skin ulcers observed was 10.2mm by 10.3mm with 573 (96.3%) of them reported to have started in the year 2018. Among the ulcers which started in the year 2018, 17 (3.0%) started between January to July 2018 while 13 (2.3%), 70 (12.2%), 346 (60.4%), 127 (22.2%) of them started in August, September, October and November of the year 2018 respectively. Examples of Leishmania positive skin ulcers observed is captured as Fig 2 .
10.1371/journal.pntd.0009416.g002
Fig 2
Examples of skin ulcers which tested positive for Leishmania parasite.
A. Location: Left lower leg; dimension:10.1mm by 5.9mm. B. Location: Left lower arm; dimension:17.0mm by 15.1mm. C. Location: Left lower arm; dimension:17.6mm by 11.0mm. D. Location: Left lower leg; dimension:14.4mm by 5.2mm.
Of the 426 individuals with skin ulcers, 419 (98.4%) indicated that they applied some form of treatment. Majority of them (67.5%) used herbs while 35.3%, and 14.2% of them used hot stone and hot water respectively as treatment of their skin ulcers ( Table 4 ).
10.1371/journal.pntd.0009416.t004
Table 4 Summary of ulcer treatment methods reported by study participants.
Treatment
Ashiabre
Keri
Sibi Hilltop
Total
method
No.
%
No.
%
No.
%
No.
%
Herbs
21
40.4
117
62.6
145
80.6
283
67.5
Hot stone
3
5.8
72
38.5
73
40.6
148
35.3
Dermacot
7
13.5
31
16.6
3
1.7
41
9.8
Penicillin
7
13.5
14
7.5
8
4.4
29
8.1
Amoxycillin
5
9.6
10
5.3
2
1.1
17
4.7
Hotwater
5
9.6
19
10.2
27
15
51
14.2
Other treatment
5
9.6
4
2.1
6
3.3
15
4.2
Total
52
100
187
100
180
100
419
100
Discussion
Cutaneous leishmaniasis among study participants
The control of CL requires an understanding of the disease epidemiology [ 15 ]. This study confirmed cutaneous leishmaniasis in the study communities by detecting Leishmania infection in 150 (25.2%) out of 595 ulcer biopsies tested by PCR. The overall prevalence of cutaneous leishmaniasis among persons with skin ulcers was 31.9% (136/426) with prevalence of 23.2% (13/56), 29.8% (56/188), and 36.8% (67/182) observed in Ashiabre, Keri and Sibi Hilltop respectively. In Mali, a systematic review reported a prevalence of 40.3% for cutaneous leishmaniasis among suspected CL cases[ 10 ].
Majority of the persons with CL in this study (73.5%) were in the age group of 5–15 years, with males in this age group constituting majority of those infected among persons with skin ulcers. A study in Mali which screened study participants with skin lesions for CL using PCR, confirmed Leishmania infection in samples from 8 persons who were all under 18 years [ 16 ].
A review of literature on CL suggests that although Leishmania infection and subsequent leishmaniasis disease generally tends to be influenced by factors associated with the host, the parasite, as well as the disease vectors, the prevalence of CL usually increases with age till about 15 years [ 17 ]. It is assumed that the prevalence of CL levels of at about 15 years because persons exposed early on in life to Leishmania infection may have acquired some level of immunity to the infection by then [ 17 ]. Observation of the highest prevalence of CL in 5–15 years age group in this study suggest a need to prioritize this group in future CL control planning in the study area.
Treatment of persons with cutaneous leishmaniasis
An important aspect of disease control is treatment of affected people. The data on treatment of skin lesions by study participants indicate that majority of them use herbs (67.5%) followed by those who use hot stone (33.5%) and hot water (14.2%) respectively.
In the case of cutaneous leishmaniasis, the first choice of treatment is pentavalent antimonials with its attendant cost and possible adverse effects [ 18 – 22 ]. However, the evidence for what can be described as optimal treatment for CL has been described as patchy and generally weak. There is therefore a need for the development of improved guidelines for management of CL in addition to the conduct of more robust studies to improve the existing body of evidence for treatment of CL [ 18 , 23 – 25 ].
Furthermore, although efforts are ongoing to develop a vaccine against leishmaniasis, there is currently no vaccine licensed for use against leishmaniasis [ 26 , 27 ]. Given the gaps in the treatment of leishmaniasis and ongoing global efforts to develop vaccines, there is a need to develop measures in the local Ghanaian context, to protect people who are affected by leishmaniasis while research continues to provide data on critical aspects of the disease such as the vectors and reservoirs.
Need for investigation of skin ulcers which were negative for Leishmania infection
Given that not all skin ulcers observed in the study communities were infected with Leishmania parasites, there is a need for continuous diagnoses of skin ulcers observed in the study communities in order to identify the ulcers infected by Leishmania parasite for the appropriate treatment to be applied [ 28 – 30 ].
Some studies have reported occurrence of other skin ulcers such as buruli ulcer, and yaws in Ghana [ 31 – 34 ]. A pilot study aimed at using azithromycin as treatment for yaws in some communities of the West Akim district of Ghana for instance, used sero-positivity based on a point of care dual treponemal and non-treponemal test as the primary outcome in addition to presentation with clinically active yaws like lesions (as secondary outcome) to select yaws cases [ 35 ].
As a result, future studies aimed at screening a larger sample of persons in the study area for yaws and other skin ulcer causing diseases such as buruli ulcer, incorporating more sophisticated laboratory diagnostic approaches may help to better characterize the causes of skin ulcers in the study area.
Conclusions
Out of 426 individuals observed with various numbers of skin ulcers in the study communities, 136 (31.9%) individuals had various numbers of confirmed Leishmania positive skin ulcers. The observation of skin ulcers which tested negative to Leishmania infection suggests a need to test for additional causes of skin ulcers such as Treponema pallidum pertenue and Mycobacterium ulcerans in the study area.
Limitations of the study
Molecular characterization of the ulcer samples for agents of other skin ulcer causing diseases reported in Ghana such as yaws, and buruli ulcer would have enriched the data.
Inclusion of a household in the study depended on the consent of the household head. This may have led to the exclusion of a few households, given that 587 households were included out of 600 households invited.
Supporting information
S1 STROBE checklist
Checklist according to The Strengthening the Reporting of Observational studies in Epidemiology (STROBE) guidelines.
(DOCX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Cyclodextrins are water-soluble cyclic oligosaccharides with hydrophilic outer surface and hydrophobic inner cavity. Their chemical structure enables them to form inclusion complexes with lipophilic molecules in aqueous solutions leading to the increment of aqueous solubility of guest molecules. The complex formation ability of cyclodextrins is utilized mainly in pharmaceutical industry for the formulation of water insoluble or poorly soluble drugs of Class II and Class IV of the Biopharmaceutics Classification System (BCS). Solubility- and absorption-enhancing effects of cyclodextrins lead to higher bioavailability of intestinal formulations, and complex formation can increase the stability of active substances [1] [2] . Several cyclodextrin derivatives were synthesized to improve the complexation efficacy and decrease toxicity. Lipophilic cyclodextrins such as methylated cyclodextrins (e.g. randomly methylated β-cyclodextrin) and hydrophilic cyclodextrins like hydroxypropyl derivatives (e.g. 2-hydroxypropyl-β-cyclodextrin) are distinguished, even if their solubility in water is high [3] . Besides the pharmaceutical applications, β-cyclodextrins are also used in cell biology research for the removal of cholesterol from cell membrane [4] and to study the role of cholesterol on cellular functions. In the case of β-cyclodextrins a relationship could be identified among the substituents of the cyclodextrin ring, cholesterol solubilization, hemolytic activity and cytotoxicity [5] . Membrane cholesterol extraction can induce several cellular effects. The activity of membrane transporters, such as P-glycoprotein is sensitive to the presence of cholesterol [6] , [7] , [8] . The disruption of cholesterol rich membrane rafts alters the integrity of tight junctions and barrier functions of cell layers [9] , [10] . These effects can also increase the permeability and absorption of drug molecules from the intestine. On the other hand membrane cholesterol depletion with high cyclodextrin concentration inhibits endocytotic processes [11] , [12] and increases exocytosis [13] .
The chemical structure, number of hydrogen donors and acceptors, relatively high molecular weight (>1000 Da) and the hydrophilicity of cyclodextrins predict that these molecules are not able to permeate biological membranes and have poor absorption [14] ; only lipophilic cyclodextrins are considered to be absorbed from the gastrointestinal tract to some extent [3] . In general, only the free form of drug, which dissociates from the cyclodextrin complex, is thought to be absorbed. According to this mechanism cyclodextrin delivers the drug to the surface of cell membrane, the drug molecule penetrates into the lipophilic membrane, but after delivery the cyclodextrin remains extracellular [3] . Interestingly in vivo studies showed that relatively high amount of hydroxypropyl-β-cyclodextrin and dimethyl-β-cyclodextrin were absorbed via rectum of rats and excreted into the urine, suggesting that not only the free form of drugs, but also cyclodextrin complexes may be absorbable through the rectal mucosa [15] .
Although cyclodextrins most likely cannot permeate the cell membrane by diffusion, recent findings revealed that they are able to enter cells. Methyl-β-cyclodextrin-dextran conjugates and hydroxypropyl-β-cyclodextrin were found to enter cells by endocytosis, as they reduced intracellular cholesterol accumulation in Niemann-Pick type C mutant cells acting at the level of endocytotic organelles inside the cells [16] . Intracellular accumulation of the fluorescent mono-4-(N-6-deoxy-6-amino-β-cyclodextrin)-7-nitrobenzofuran (NBD-β-CD) was also detected in HepG2 and SK-MEL-24 cells, and endocytosis as a possible mechanism for the transmembrane passage of NBD-β-CD was suggested [17] . Macropinocytosis of amphiphilic cationic cyclodextrin transfection complexes were also observed in Caco-2 intestinal epithelial cells [12] , and clathrin-dependent endocytosis of a fluorescent methyl-β-cyclodextrin by HeLa cells was demonstrated [18] .
These results raise the possibility that cyclodextrin molecules not only increase the solubility of poorly soluble drugs and act as permeation enhancers in the intestine, but are able to enter intestinal cells by the endocytotic pathway. This mechanism, the intracellular route and fate of cyclodextrins have not been investigated on intestinal epithelial cells yet, although transcytosis is known in the case of intestinal epithelial Caco-2 cells [19] . There is also limited information about the permeability of cyclodextrins on Caco-2 monolayers.
In the present study our aim was to examine the interaction of the fluorescently labeled randomly methylated β-cyclodextrin (FITC-RAMEB) with Caco-2 colon cell layer and examine the cellular uptake of cyclodextrins on intestinal epithelial cells.
Materials and Methods
Randomly-methylated β-cyclodextrin (RAMEB) was purchased from Wacker Chemie (Munich, Germany). 6-monodeoxy-6-mono[(5/6)-fluoresceinylthioureido]-RAMEB (FITC-RAMEB) (DS = 1 for FITC, DS ∼ 12 for methyl) was the product of CycloLab Ltd(Budapest, Hungary). FITC-RAMEB was prepared by reacting methylated amino-cyclodextrin with fluorescein-5(6)-isothiocyanate as described elsewhere [18] . By this reaction FITC was covalently coupled to RAMEB. CellMask Deep Red plasma membrane stain and CellLight ® Early Endosomes-RFP *BacMam 2.0* was from Invitrogen (Budapest, Hungary). All other reagents were purchased from Sigma-Aldrich (Budapest, Hungary).
Caco-2 Cell Culture
Caco-2 cell line originates from the European Collection of Cell Cultures (ECACC UK). Caco-2 cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% heat-inactivated foetal bovine serum, 1% non essential amino acid and 1% penicillin-streptomycin solution at 37°C in an incubator containing 5% CO 2 . The passage number of the cells was between 25 and 40.
For permeability experiments and release studies, Caco-2 cells were seeded at density of 200,000 cells/well on Transwell® (Corning Costar, USA) polycarbonate filters (pore size 0.4 µm, surface area 1.12 cm 2 ). Culture medium was replaced with fresh medium every two or three days in the inserts. Monolayers were used for the experiments between 20 and 35 days after seeding. The formation of functional epithelial layers was monitored by the development of transepithelial electrical resistance (TEER) and measured with a Millicell–ERS voltohmmeter (Millipore, USA). In permeability experiments TEER values were also measured at the beginning and at the end of sampling to check monolayer integrity and follow the effects of cyclodextrin treatments.
Transepithelial RAMEB Permeability Measurements
In permeability measurements two different cyclodextrin solutions were used for the treatments: 0.05 mM FITC-RAMEB (FR) solution and 5.0 mM RAMEB solution containing 0.05 mM FITC-RAMEB (FRR). The solvent was Hank’s Balanced Salt Solution (HBSS). Caco-2 monolayers were washed twice and pre-incubated with HBSS for 20 minutes at 37°C and then incubated apically with FR or FRR solutions for 2 hours at 37°C. Samples were collected from the basolateral side at 60 and 120 minutes and the volume was replenished with HBSS. The monolayers were washed five times with HBSS and cells were lysed with 1% Triton X-100 (TX-100) (Roche Diagnostics GmbH (Mannheim, Germany). The permeated amount and the FITC-RAMEB content of the cell lysates were determined by FLUOstar Optima microplate reader (BMG LABTECH, Offenburg, Germany) at 492 nm excitation and 520 nm emission wavelength.
FITC-RAMEB permeation rates across the monolayers were determined from the concentration values. With the formula below the apparent permeability coefficients were calculated:
P app : apparent permeability coefficient (cm/s)
dQ/dt: permeability rate of substances (mol/s)
C 0 : initial concentration of the substances in the apical chamber (mol/ml)
A: surface area of the membrane (cm 2 ).
FITC-RAMEB Uptake Studies of Caco-2 Monolayers
In this experiment Caco-2 cells were seeded in black 96 well plates at the density of 10 4 cells/well. After 7 days the cells were washed twice with HBSS and incubated with FR or FRR solutions at 37°C for 5-, 10-, 30-, 60- or 120 minutes. After the treatment cells were washed four times with ice cold HBSS, kept on ice and fixed with 3% paraformaldehyde solution (37°C, 15 min). Fluorescence intensities of the samples were measured with FLUOstar Optima microplate reader. After the measurement 4′,6-diamidino-2-phenylindole dihydrochloride (DAPI), at 300 nM final concentration was added to each well and incubated for further 15 minutes. DAPI was measured at 355 nm excitation and 485 nm emission wavelengths. This dye was used to normalize FITC-RAMEB for DAPI fluorescence intensities.
FITC-RAMEB Release Studies on Caco-2 Monolayers
The first part of this experiment was the same like permeability measurements, except that the apical chamber contained 0.5 mM FITC-RAMEB solution. At the end of the 120 minutes long incubation, inserts were washed five times with ice cold HBSS and divided into two groups. The monolayers of the control group were lysed with 1% Triton X-100 solution and FITC-RAMEB contents were determined with FLUOstar Optima microplate reader. The other group of the inserts was incubated in HBSS at 37°C for another 120 minutes. During the second incubation samples were collected from apical and basolateral chambers at 10-, 30-, 60- and 120 minutes and the released amount of FITC-RAMEB was measured. After incubation these monolayers were also washed twice with HBSS, lysed and the FITC-RAMEB content of the cell layers was determined. The rate of the release was expressed as the percentage of FITC-RAMEB content of the monolayers of control group.
Confocal Microscopy
For microscopic investigations 80,000 cells/well were seeded on round glass cover-slips in 12-well plates. 24 hours later cells were treated with CellLight ® Early Endosomes-RFP *BacMam 2.0* at density of 30 particles per cell and incubated for further 48 hours in cell culture medium. Then samples were washed twice with HBSS and treated with FR or FRR solutions for 30 minutes at 37°C. To completely remove FITC-RAMEB, cells were washed eight times with ice cold HBSS and samples were stained with 1 µg/ml solution of CellMask Deep Red plasma membrane stain for 5 minutes at 37°C. After washing cells twice with HBSS and fixing them with 3% paraformaldehyde solution, cell nuclei were stained with DAPI (300 nM). In some experiments Caco-2 monolayers were treated applying the same protocol for plasma membrane and cell nucleus staining but in these samples Early Endosomes-RFP was not used. In experiments performed in Transwell®, the insert membranes were excised and placed on slides. Confocal microscopy measurements and analyses were carried out by a Zeiss LSM 510 META (Jena, Germany) confocal microscope. To eliminate spectral cross talk samples were illuminated with three different excitations subsequently using multi-track mode (UV lines: 351.1 nm and 363.8 nm of an Ar-ion laser, these two lines were used simultaneously; blue line: 488 nm of another Ar-ion laser; and red line: 633 nm of a He-Ne laser). Emissions above 420 nm, above 505 nm and above 650 nm were detected subsequently in three channels with the META detector, respectively. For confocal imaging pinhole size was set to 1 Airy unit.
Flow Cytometry
Flow cytometric experiments were used to verify endocytosis of FITC-RAMEB by Caco-2 cells. For these experiments cells were trypsinized, washed twice with HBSS and resuspended at 1×10 6 cells/ml concentration. Cells were incubated with FITC-RAMEB, Lucifer Yellow (LY) or calcein AM solutions in different concentrations for 30 minutes at 37°C or at 0°C. Dyes were used in the following concentration ranges: FITC-RAMEB from 0 to 500 µM, calcein AM from 0 to 1 µM, and LY from 0 to 960 µM. At the end of the treatments cells were washed three times with ice cold HBSS and kept on ice until measurements. Propidium-iodide was added to the cells at the concentration of 2 µg/ml to recognize dead cells. In uptake inhibition experiments cells were pre-incubated with 10 µM rottlerin for 45 minutes before adding FITC-RAMEB or LY. Cells were analyzed by five-laser BD FACSaria II flow cytometer (BD Biosciences, San Jose, CA). In the case of FITC-RAMEB and calcein AM staining, cells were illuminated with 488 nm laser line, while for LY staining with the more optimal 445 nm. In all previous cases fluorescence emission was detected via 502 nm long pass dichroic mirror and 530/30 nm band pass filter. Single cell events were recognized using both the area and width of the forward-scattered light and side-scattered light signals. Viable cells were gated in according to their low intensity propidium iodine fluorescence excited at 561 nm and detected via 590 nm long pass filter.
Statistical Analysis
For statistical analysis SigmaStat softver (version 3.1; SPSS Inc.) was used. Data are presented as means ± SD. Comparison of two groups was performed by unpaired or paired t-test, while comparison of more than two groups was performed using ANOVA. Differences were considered significant at p<0.05.
Results
Transepithelial FITC-RAMEB Permeability in Caco-2 Cell Monolayers
In order to investigate the permeability of the fluorescent derivative of RAMEB through the intestinal epithelial barrier we applied Caco-2 monolayers. Two cyclodextrin solutions, 0.05 mM FITC-RAMEB (FR) and 5.0 mM RAMEB solution containing 0.05 mM FITC-RAMEB (FRR) were used. The permeability of FITC-RAMEB was determined in both cases and the results were expressed in apparent permeability values (P app ). The apparent permeability of FITC-RAMEB was very low both in FR and FRR treatments, 3.35±1.29×10 −8 and 4.23±1.46×10 −8 cm/s, respectively. There was no significant difference between these two average permeability values (n = 9 for FR and n = 6 for FRR treatments, p>0.05), indicating that 5 mM RAMEB co-treatment had no effect on the permeability of FITC-RAMEB and the integrity of the monolayer.
The integrity of monolayers was tested by measuring transepithelial electrical resistance (TEER). The TEER values did not decrease significantly after the cyclodextrin treatments (p>0.05) ( Fig. 1 ).
10.1371/journal.pone.0084856.g001 Figure 1
Transepithelial electric resistance (TEER) of Caco-2 monolayers before and after 120 minutes permeability experiments.
Cell layers were treated with 0.05-RAMEB (FR) alone or in the presence of 5 mM RAMEB (FRR). Untreated monolayers were kept in HBSS. Values are expressed as means ± SD, n = 9 for FR, n = 6 for FRR treatment and n = 6 for untreated samples. There were no significant differences between TEER values before and after the treatments (p>0.05) and among the groups (p>0.05).
At the end of the permeability measurements cell layers were washed thoroughly with HBSS and lysed with 1% TX-100. The fluorescence of cell lysates were measured with FLUOstar Optima microplate reader. The fluorescence of FR and FRR treated samples were significantly higher than the untreated monolayers (p<0.001), indicating, that Caco-2 cell layers accumulated fluorescently-labeled RAMEB ( Fig. 2 ). FRR, containing 5 mM RAMEB did not change the accumulation of FITC-RAMEB in the cell layers (p>0.05).
10.1371/journal.pone.0084856.g002 Figure 2
Accumulation of FITC-RAMEB in Caco-2 monolayers after 120 minutes permeability experiments.
Caco-2 monolayers were treated with 0.05 mM FITC-RAMEB (FR) alone or in combination with 5 mM RAMEB (FRR) for 120 minutes. Monolayers were washed and the fluorescence intensity of the accumulated FITC-RAMEB was determined with FLUOstar Optima microplate reader. Presented values are means ± SD, n = 7 for FR, n = 4 for FRR treatment and n = 5 for untreated samples. FR and FRR treatments increased significantly the fluorescence of monolayers compared to the untreated control (p<0.001).
FITC-RAMEB Accumulation in Caco-2 Monolayers
The time dependence of FITC-RAMEB accumulation was measured in 96-well plates with a microplate reader. Caco-2 cells were treated with FR and FRR solutions for 5-, 10-, 30-, 60- or 120 minutes. No difference could be seen between FR and FRR treatment up to 120 minutes. The accumulated amount of FITC-RAMEB increased during the 120 minutes of the experiment. The rate of uptake was fast during the first 5–10 minutes and slower in the remaining period ( Fig. 3 ).
10.1371/journal.pone.0084856.g003 Figure 3
Kinetics of FITC-RAMEB uptake in Caco-2 monolayers.
Cell monolayers were treated with 0.05-RAMEB (FR) alone or in combination with 5 mM RAMEB (FRR) and in different time points the incubation was stopped. After washing cells were fixed with 3% paraformaldehyde solution and the accumulated FITC-RAMEB was determined by microplate reader. Cell nuclei were labeled with DAPI and fluorescence intensities of FITC-RAMEB were normalized for DAPI fluorescence intensities. Values are expressed as means ± SD, n = 3 for FR and FRR treatments.
Release of Accumulated FITC-RAMEB from Caco-2 Monolayers
In FITC-RAMEB release studies Caco-2 monolayers were treated with 0.5 mM FITC-RAMEB solution for 120 minutes, washed five times with ice-cold HBSS and the fluorescence of control group of monolayers was determined and considered as 100%. The second group of monolayers was kept for further 120 minutes in fresh HBSS at 37°C and the released amount of FITC-RAMEB in apical and basal chambers was determined as a function of time. FITC-RAMEB appeared rapidly in both chambers. About 85% of initial fluorescence was released into the apical chamber within the first hour, while only about 7.4% of the accumulated FITC-RAMEB was released into the basal chamber from monolayers after 2 hours of incubation ( Fig. 4 ). At the same time the fluorescence of the monolayers decreased drastically, from 100±8.8% to 8.9±1.5%.
10.1371/journal.pone.0084856.g004 Figure 4
Release of FITC-RAMEB from Caco-2 monolayers.
Treatment with 0.5-RAMEB was carried out on Transwell® inserts for 120 minutes, then the monolayers were washed and FITC-RAMEB release was followed in the apical and basolateral chambers during the next 120 minutes. FITC fluorescence intensities were determined in the control group of the samples after the first 120 minutes and considered as 100% accumulation. The rate of release in the second group was compared to this value. Values are means ± SD, n = 4.
FITC-RAMEB Internalization in Undifferentiated and Differentiated Caco-2 Cells and Colocalization with Rab5a Early Endosome Marker
The accumulation of FITC-RAMEB in Caco-2 cells and monolayers was visualized by confocal laser scanning microscopy. In undifferentiated Caco-2 cells, FITC-RAMEB could be detected on CLSM images as small bright particles, located in the cytoplasm ( Fig. 5 ). In the mid-sections the cyclodextrin-loaded granules were found under the cell membrane and near the cell nuclei.
10.1371/journal.pone.0084856.g005 Figure 5
Confocal images of undifferentiated Caco-2 cells.
Cells were treated with the solution of 0.05-RAMEB and 5 mM RAMEB (FRR). FITC-RAMEB (green) is localized in small vesicles (white arrows) under the CellMask labeled cell membrane (red) or in larger vesicles near the DAPI stained cell nucleus (light blue). Aggregated particles of FITC-RAMEB can be also seen outside the cell membrane (A). Nine consecutive confocal sections of a cluster of cells were recorded (B). Each section is one and half micrometer thick. FITC-RAMEB (green) is located in cytoplasmic granules. The granular bright particles are observed inside the cell membrane and outside of the cell nuclei.
To examine the FITC-RAMEB uptake of differentiated Caco-2 cells, monolayers grown on Transwell® inserts were used. Fig. 6 shows that FITC-RAMEB is able to enter the Caco-2 monolayer and it is localized in granules within the cytoplasm. There was no difference between cellular uptakes of FITC-RAMEB after FR or FRR treatments on the confocal images.
10.1371/journal.pone.0084856.g006 Figure 6
Cyclodextrin enters differentiated Caco-2 cells of a high resistance Caco-2 cell layer.
A confluent layer was treated with 0.05-RAMEB and imaged by confocal microscope in twelve two-micrometer thick sections, of which six are demonstrated in the panels on the left side (A–F). On the right, one middle section of the image shows the top view of the cell layer (H) at the level indicated by blue lines in side sections. Upper (G) and right (I) side images are appropriate sections from perpendicular directions at green and at red lines. Crosshair (green and red lines at the long white arrow) set to an intense FITC-RAMEB (green) granule (indicated by arrows), which is located at the nuclear (light blue DAPI stain) level of cells. Several smaller FITC-RAMEB green granules can be seen below the cell membrane marked by CellMask (dark blue).
These observations suggested that cyclodextrin molecules enter the cells by endocytosis. To confirm this hypothesis, we investigated the colocalization of FITC-RAMEB with the small GTPase Rab5, which is a key determinant of early endosomes [20] . Caco-2 cells were transiently transfected with a plasmid coding for a red fluorescent protein tagged Rab5a GTPase fusion protein (RFP-Rab5a) that was strongly expressed in the cell membrane. As Fig. 7 shows, FITC-RAMEB colocalizes with RFP-Rab5a. Colocalization is marked by yellow pixels. Pearson’s correlation coefficients were calculated and they were between 0.55 and 0.78 after 30 minutes incubation, indicating that the entry of RAMEB into the cytoplasm and the formation of early endosomes are associated. High degree of colocalization could be observed after 2 minutes of incubation, and after 30 minutes colocalization could be still detected indicating, that endocytosis functioned continously (see also Figure S1 and S2 .).
10.1371/journal.pone.0084856.g007 Figure 7
FITC-RAMEB colocalized with Rab5 proteins.
RFP-Rab5a transfected Caco-2 cells were treated with 0.05 mM FITC-RAMEB (green) for 30 minutes. Colocalization is indicated by yellow pixels in the confocal microscopic images on the sections marked by the 18-micrometer and 21-micrometer label. Figure shows nine subsequent, three-micrometer thick confocal sections giving altogether a twenty-four-micrometer cross layer of a Caco-2 cell. CellMask (dark blue) was used for labeling membrane at the cell surface and nucleus was stained by DAPI (light blue). Rab5 proteins (red) are visualized by transient transfection of a plasmid coding Red Fluorescent Protein (RFP) tagged Rab5.
Internalization of FITC-RAMEB and its Inhibition by the Fluid Phase Endocytosis Inhibitor Rottlerin
Internalization of FITC-RAMEB was also investigated by flow cytometry. Caco-2 cell suspensions were treated by FITC-RAMEB, the macropinocytosis marker Lucifer Yellow and the lipophilic membrane permeability marker calcein-AM, both at 37°C and 0°C. Major differences could be seen between the uptake of hydrophilic and lipophilic molecules ( Fig. 8 ). Intracellular accumulation of calcein was dependent on dye concentration, but was independent of temperature. At the same time, both FITC-RAMEB and Lucifer Yellow uptake increased as a function of the dye concentration, but it was inhibited at 0°C.
10.1371/journal.pone.0084856.g008 Figure 8
Cellular uptake of calcein-AM (A), FITC-RAMEB (B) and Lucifer Yellow (C) as a function of ligand concentration.
Cells were treated at 37°C and 0°C and the cellular fluorescence was determined by flow cytometry, after excluding dead cells with propidium iodide.(Graphs show results of a representative experiment).
Rottlerin, a macropinocytosis inhibitor decreased significantly both FITC-RAMEB and Lucifer Yellow internalization in Caco-2 cells. Although the inhibition was not complete, the extent of inhibition of FITC-RAMEB uptake was similar to that of Lucifer Yellow ( Fig. 9 ).
10.1371/journal.pone.0084856.g009 Figure 9
Effect of 10 µM rottlerin on the cellular uptake of FITC-RAMEB and Lucifer Yellow.
Caco-2 cells were pre-incubated for 45 minutes with rottlerin and the internalization of the fluorescent molecules was detected by flow cytometer. (n = 3, p<0.01).
Discussion
In the present study we investigated the permeability and cellular uptake of the fluorescent methyl-β-cyclodextrin in intestinal Caco-2 cells. The available data regarding the absorption and oral bioavailability of cyclodextrins is very limited and there are no Caco-2 permeability values in the literature. In early studies of intestinal absorption of 14 C-labelled β-cyclodextrin in rats only 5% of the administered activity could be detected in the blood. It was concluded that β-cyclodextrin was not absorbed from the stomach and the small intestine, and the low absorption was explained with the amylase action: only the open-chain dextrins and the glucose formed from cyclodextrins were absorbed [21] . According to recent publications the oral bioavailability of HPBCD is less, than 0.03% and it is approximately 0.3% for β-cyclodextrin [22] , while RAMEB has an oral bioavailability of about 12% in rats, [23] .
Our permeability results with FITC-RAMEB are in accordance with the low intestinal absorption of cyclodextrins, as the P app values were 3.35±1.29×10 −8 and 4.23±1.46×10 −8 cm/s for FR and FRR treatments, respectively. These data are also in agreement with the permeability results of natural cyclodextrins (α-, β-, and γ-cyclodextrin) on pulmonary Calu-3 cell layers, which were in the same order of magnitude [24] .
Methylated cyclodextrins are also used for membrane cholesterol depletion in 5–10 mM concentration [4] , [25] and can enhance the penetration of drugs [10] . 0.05 mM cyclodextrin presumably does not affect membrane cholesterol significantly, since it is 1/100 of the usually applied concentration. At 5 mM RAMEB concentration we did not observe cytotoxicity on Caco-2 cells previously [5] , therefore we investigated the effect of 5 mM RAMEB on FITC-RAMEB permeability and the TEER of the monolayers. No significant difference could be observed on permeability and resistance values of FR and FRR treatments (p>0.05), and 5 mM RAMEB had no effect on the permeability of the monolayer.
Examining the fluorescence of the cells at the end of the permeability experiments we found that a significant amount of FITC-RAMEB accumulated in the cell layers. These cyclodextrins could not be removed by extensive washing. We investigated the time-dependence of FITC-RAMEB accumulation in the monolayers and as Fig. 3 shows Caco-2 cells successively accumulated both FR and FRR up to 120 min of the experiment. To reveal the fate of the accumulated FITC-RAMEB we loaded the cells with 0.5 mM FITC-RAMEB solution, using 10 time higher concentration than in permeability studies. This resulted in 10 time higher accumulation, but the permeability of FITC-RAMEB did not increase (2.28±0.34×10 −8 cm/s). After 120 minutes, the release of the accumulated cyclodextrins was followed both in apical and basolateral directions. Interestingly FITC-RAMEB appeared in both the apical and basal chambers, but the majority of the accumulated cyclodextrin was released to the apical direction. Only 7.4% of the accumulated FITC-RAMEB reached the basal chamber.
These results indicated that although the intestinal Caco-2 monolayer is an almost impermeable barrier for the cyclodextrin molecules, the cells are able to take up cyclodextrins from solutions with a mechanism different from simple diffusion. Studies on Calu-3 monolayers suggested that cyclodextrins traverse these monolayers by paracellular route, although transcytosis could not be excluded [24] . Recent publications revealed that certain cell types are able to internalize cyclodextrins by endocytosis [16] , [17] , [18] ; therefore we investigated the intracellular localization of FITC-RAMEB by confocal microscopy. Fig. 5 and 6 show that FITC-RAMEB is able to enter into the cytoplasm of both undifferentiated and differentiated Caco-2 cells. Since the labeled cyclodextrin was localized in vesicles in the cytoplasm, the possibility of endocytosis was investigated hereafter. In the cell membrane RFP-Rab5a fusion protein and FITC-RAMEB showed colocalization. Rab5 is a key organizer of early endosomes, but it cannot be detected in late endosomes [20] . In our confocal microscopy images Rab5 and FITC-RAMEB did not exhibit colocalization in vesicles in deeper layers. The colocalization of RFP-Rab5a and FITC-RAMEB suggests that endosome formation is involved in the initiation of cyclodextrin internalization.
Endocytosis has two major routes, phagocytosis and pinocytosis or fluid-phase uptake. Fluid-phase endocytosis, which requires the cargo molecules to be dissolved, can be subdivided into macropinocytosis, clathrin-mediated, caveolin-mediated and clathrin- and caveolin-independent endocytosis [26] . The widely used marker of macropinocytosis is Lucifer Yellow [27] , [28] , [29] . Flow cytometry analyses revealed that in Caco-2 cells Lucifer Yellow was internalized in a concentration dependent manner and its uptake could be inhibited at 0°C [28] , [29] . FITC-RAMEB showed similar cellular uptake: at 37°C accumulated in the cells as a function of concentration, while at 0°C FITC-RAMEB uptake was diminished. On the other hand lipophilic calcein AM showed the same cellular accumulation at 0°C and 37°C, as it rapidly permeated the lipid membrane [30] , and the intracellular accumulation was not inhibited by cooling. The macropinocytosis inhibitor rottlerin [27] had similar inhibitory effect on FITC-RAMEB and LY accumulation. These results indicate, that in Caco-2 cells macropinocytosis is involved in the entry of FITC-RAMEB. It also explains why the majority of the accumulated FITC-RAMEB was released to the apical direction. It was demonstrated in human epidermoid A431 cells, that macropinosomes recycle their content to the cell surface [31] . It seems that the same mechanism could be observed in Caco-2 cells, as the mechanism of internalization was macropinocytosis, the majority of accumulated cyclodextrin was guided to the apical cell surface. Nevertheless, the total recycling of the internalized cyclodextrin molecules took at least one hour, which means that this process prolongs the contact between cylodextrins or cyclodextrin-drug complexes and the membrane of macropinosomes. It is important to note, that other endocytotic mechanisms should be also taken into consideration. Previous studies implicated fluid-phase endocytosis and clathrin-dependent endocytosis [16] , [17] , [18] for the mechanism of cyclodextrin internalization. Nevertheless, phagocytosis could be also a possibility for cyclodextrin internalization in concentrated cyclodextrin solutions. It is reported that at high concentrations, natural β-cyclodextrin [32] and the fluorescent tetraamino rhodaminyl hydroxypropyl-β-cyclodextrin [33] form large, nano-sized aggregates in water. However, the substitution of OH groups with methyl groups on the cyclodextrin ring inhibits the aggregation of RAMEB and at 12 mM no aggregation was observed [32] . In this study 0.05 mM FITC-RAMEB was applied alone or in combination with 5 mM RAMEB, which is 40–240 times lower cyclodextrin concentration than what was found to form aggregates above, thus phagocytosis can be excluded from among the possible mechanisms of cyclodextrin uptake.
In summary, our results on Caco-2 cells are in accordance with earlier findings, the cellular internalization of water soluble FITC-RAMEB is governed by fluid phase endocytosis in intestinal Caco-2 cells. It is hard to predict the quantitative importance of this mechanism. Even if permeability data are suitable to value the extent of absorption of cyclodextrins, it is difficult to quantify the amount of continuously internalized and released cyclodextrins with this setup of the model. The intestinal absorptive surface area relative to the volume of the gut is much bigger than the surface area of Caco-2 monolayers and on the other hand the peristaltic movement should be also considered. Thus the extent of internalization can be much higher in vivo, even if cyclodextrins are released back to the lumen of the gut and as the process can be continuous along the small intestine its efficiency can be much higher.
Conclusions
Cyclodextrins are used to increase solubility, bioavailability and stability of poorly water-soluble drugs. Our results demonstrate for the first time that randomly methylated-β-cyclodextrins can enter into intestinal epithelial cells by endocytosis. This process can contribute to the enhancement of the intestinal delivery and bioavailability of drugs by cyclodextrins in several ways. It can help to overcome the intestinal membrane barrier, the endosome formation increases the contact surface area between the cyclodextrin-drug complexes and the cell membrane and prolongs the retention time of cyclodextrins in the epithelial cells. Since this study has demonstrated the role of macropinocytosis in the uptake of methylated-β-cyclodextrin in intestinal cells, this mechanism merits further investigations in connection with drug absorption mediated by cyclodextrins.
Supporting Information
Figure S1
Colocalization of RFP-Rab5a with FITC-RAMEB, cell nucleus and cell membrane. RAMEB shows high degree of colocalization with Rab5a immediately below the cell surface membrane of two connected cells at 2 minutes incubation (R = 0.93). Caco2 cells attached to surface of coverslip were transiently transfected by Rab5a (red) tagged by red fluorescent protein (RFP), treated by Fitc-RAMEB (green) for 2 minutes, and fixed. Before imaging surface membrane and nuclei were labeled by CellMask (dark blue) and DAPI (cyan), respectively, for 5 minutes. Panel A shows colocalization of Fitc-RAMEB and RFP-Rab5a; panel B indicates colocalization of DAPI and RFP-Rab5a as negative control (R = −0.39) and panel C specifies colocalization of surface membrane (CellMask) and RFP-Rab5a as positive control (R = 0.95) at a confocal image section crossing RAMEB granules (white areas in panel A right side, section thickness is 1.5 micrometer). Right panels show two channel images of signals tested for colocalization (bar is 10 micrometer), while corresponding left panels show two parameter histograms of signals of the two channels. White areas in right side images were chosen by setting channel signals above thresholds indicated by red signs on scales of corresponding left side two parameter histograms. R indicates Pearson correlation coefficients calculated in images at location of the white areas. In two parameter histograms the highest colocalization between tested channels would be indicated by a 45° diagonal line corresponding to R = 1 (in left panels of A and C R is close to this value), while a 135° diagonal line would indicate a negative correlation (left panel of B).
(TIF)
Figure S2
Colocalization of FITC-RAMEB with RFP-Rab5a in the function of the time. Colocalization of RAMEB and Rab5a was monitored in time during the endocytosis process. The highest average colocalization (0.76±0.01) was measured at 2 minutes after initiation of the endocytosis at 37°C. In later time points, at 5, 10, 20 and 30 minutes R was dropped to a lower but still significant value (R = 0.5–0.6). R, Pearson correlation coefficient was measured in region of interests (ROI) set to those locations where RAMEB granules were observed in confocal sections (one section was 1.5 micrometer thick). Pattern of the intracellular localization of colocalized molecules also changed in time. At 2 minutes colocalization was either dispersed in the surface membrane of cell or in the cytoplasm close to surface membrane. At later time points RAMEB granules moved closer to cell nuclei with lower, but still significant R for Rab5a colocalization (means±SD).
(TIF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
To colonize the host, bacterial pathogens and commensals are limited to metabolites present in tissues to fulfill nutritional requirements. While strategies employed by bacteria to acquire nutrient transition metals such as iron have been an intense area of research, studies defining mechanisms of nutrient sulfur procurement from host environments are garnering heightened interest [ 1 – 6 ]. Sulfur is essential due to its capacity to fluctuate between redox states and therefore catalyze numerous cellular reactions [ 7 , 8 ]. Ultimately, cells require sulfur to synthesize cysteine (Cys) as it is the fulcrum of sulfur metabolism by serving as an intermediate for methionine (Met) and sulfur-containing cofactors such as Fe-S clusters [ 8 – 11 ]. In host cells and some bacterial species, Cys is also required to generate the low molecular weight thiol glutathione (GSH) [ 12 ].
GSH concentrations range between 0.5 and 10 mM in mammalian tissues, making it a relatively abundant source of nutrient sulfur for invading pathogens [ 3 , 12 , 13 ]. In addition to Cys, GSH consists of glutamate and glycine. A unique γ-peptide bond links the glutamate γ-carboxyl to the Cys amine. To maintain Cys reservoirs, organisms rely on GSH catabolism via the γ-glutamyl cycle [ 3 , 12 ]. Liberation of Cys from GSH is a two-step process that requires γ-glutamyl transpeptidase (Ggt), a specialized protease conserved in all domains of life due to its role in the γ-glutamyl cycle [ 12 , 14 , 15 ]. Ggt is localized within the Gram-negative periplasm or the periphery of eukaryotic cells where it degrades endogenous GSH [ 16 – 18 ], but it has also been shown to fulfill the nutritional sulfur requirement of Francisella tularensis [ 3 , 19 ].
Staphylococcus aureus is the leading cause of superficial and invasive bacterial diseases in the United States and Europe [ 20 , 21 ]. Strategies S . aureus employs to obtain nutrient sulfur during pathogenesis are largely unknown. Previous work demonstrated that reduced cysteine (Cys), oxidized cysteine or cystine (CSSC), sodium sulfide, thiosulfate, or GSH stimulated in vitro proliferation of S . aureus [ 22 ]. Further investigation revealed that the TcyP and TcyABC transporters support Cys and CSSC utilization; however, mechanisms of GSH acquisition have not been defined [ 22 , 23 ]. S . aureus does not synthesize GSH but encodes a putative Ggt. This fact supports the hypothesis that staphylococcal Ggt catabolizes exogenous host GSH, liberating Cys as a means to satisfy the nutrient sulfur requirement [ 24 ]. Here we demonstrate that S . aureus utilizes oxidized GSH (GSSG) as a nutrient sulfur source and isolate mutants that display significantly decreased proliferation in a medium supplemented with GSSG or GSH as the sole source of nutrient sulfur. These mutants harbor transposon (Tn) insertions within a five-gene locus, SAUSA300_0200 – 0204 , that encodes a predicted ATP-binding-cassette (ABC) transporter ( SAUSA300_0200–0203 ) and putative Ggt ( SAUSA300_0204 ). Based on the mutant proliferation defects observed in GSH- or GSSG-supplemented media, we name this transporter the G lutathione i mport s ystem (GisABCD). We determine that S . aureus Ggt is localized within the cytoplasm and that the recombinant enzyme cleaves both GSH and GSSG. A search for GisABCD-Ggt across Firmicutes revealed that only a select clade within the Staphylococcus genus, one that excludes S . epidermidis , encodes homologues of the system. Consistent with this finding, S . aureus outcompetes S . epidermidis in GSSG- or GSH-supplemented conditions in a GisABCD-Ggt-dependent manner. Therefore, this newly described nutrient sulfur acquisition system provides a competitive advantage for S . aureus over other staphylococci associated with the human microbiota.
Results
S. aureus proliferates in medium supplemented with GSSG as the sole source of nutrient sulfur
A previous study qualitatively reported that S . aureus proliferates on a chemically defined agar medium supplemented with GSH as the sole sulfur source, indicating that the abundant host metabolite is a viable source of nutrient sulfur [ 22 ]. However, S . aureus likely encounters both reduced and oxidized GSH (GSSG) as the pathogen induces a potent oxidative burst during infection [ 25 ]. Therefore, we hypothesized that S . aureus also utilizes GSSG as a source of nutrient sulfur. To quantitatively assess GSH and GSSG utilization as sulfur sources by S . aureus , a chemically defined medium, referred to as PN, was employed [ 26 ]. PN contains sulfate (MgSO 4 ) and methionine (Met), but S . aureus is not capable of assimilating sulfate or utilizing Met to generate Cys due to an apparent lack of a methionine S -methyltransferase homologue [ 9 ]; thus, oxidized cysteine or cystine (CSSC) is typically added as the source of nutrient sulfur [ 22 , 27 ]. In keeping with this, a USA300 LAC strain of S . aureus (JE2) exhibits substantially decreased proliferation in PN that lacks CSSC ( Fig 1A ). Notably, replacing CSSC with either 50 μM GSH or 25 μM GSSG stimulates robust S . aureus proliferation ( Fig 1A ). To determine whether utilization of GSSG is conserved throughout the species, we examined proliferation of clinical isolates in GSSG-supplemented medium. Growth of methicillin-susceptible and methicillin-resistant clinical isolates was quantified in PN supplemented with GSSG as the sole sulfur source. Compared to PN lacking a viable sulfur source, GSSG supplementation stimulates proliferation ( Figs 1B and S1A ). GSSG supplementation also promotes growth of other S . aureus strains ( Figs 1C and S1B ). S . aureus utilization of GSSG expands the number of sulfur-containing metabolites present in host tissues that are capable of supporting its proliferation.
10.1371/journal.pgen.1010834.g001
Fig 1
Supplementation of GSSG as the sole source of nutrient sulfur supports proliferation of S . aureus .
A. S . aureus JE2 was cultured in PN medium lacking a viable sulfur source (-S) or supplemented with CSSC (25 μM), GSSG (25 μM), or GSH (50 μM). B. JE2 and either MSSA or MRSA clinical isolates were grown in PN supplemented without a viable sulfur source (-S) or 25 μM GSSG for 19 hrs. C. S . aureus strains were cultured in PN medium supplemented without sulfur (-S) or 25 μM GSSG and cultured for 25 h. Bars depict the mean terminal optical density at 600 nm (OD 600 ), and circles represent individual replicate terminal OD 600 . The mean of at least three independent trials and error bars representing ± 1 standard error of the mean are presented.
The SAUSA300_0200–0204 locus supports S . aureus utilization of GSH and GSSG as sulfur sources
To determine genetic factors required for S . aureus utilization of GSSG as a sulfur source, we screened the Nebraska Transposon Mutant Library for mutants that exhibit decreased proliferation in PN medium supplemented with 25 μM GSSG as the sole source of sulfur [ 28 ]. Five GSSG proliferation-impaired mutants were identified in the screen, each harboring an independent Tn insertion in one of five genes present in the SAUSA300_0200-ggt locus ( Fig 2A ). The ggt gene ( SAUSA300_0204 ) encodes an annotated γ-glutamyl transpeptidase, an enzyme that catabolizes GSH in other organisms [ 12 , 14 , 15 ]. Notably, SAUSA300_0200–0203 encodes a putative nickel-peptide ABC transporter. Backcrossing these Tn-inactivated genes into an otherwise wild type (WT), JE2 strain significantly decreased proliferation in medium supplemented with 25 μM GSSG ( Fig 2B ). The Tn mutants also displayed decreased proliferation in PN supplemented with 50 μM GSH as the sole sulfur source ( Fig 2C ). However, the mutant strains demonstrated WT-like growth in medium supplemented with 25 μM CSSC or in a rich medium, indicating the proliferation defect is specific to GSH and GSSG ( Fig 2C inset ). To address complications associated with auto-oxidation of GSH to GSSG in aerobic environments, we tested anaerobic proliferation in media supplemented with GSH or GSSG. Mutant strains harboring a Tn in SAUSA300_0201 or an in-frame deletion of all five genes were used in this assessment. In these conditions, the mutant strains proliferate to WT levels in PN supplemented with CSSC but display diminished growth upon GSH or GSSG supplementation ( S2 Fig ). This finding confirms that SAUSA300_0200 – ggt supports S . aureus utilization of both GSH and GSSG as distinct sources of nutrient sulfur. Complementation experiments tested whether proliferation of a ggt mutant cultured in PN medium supplemented with GSH or GSSG could be restored by providing WT or a C-terminal His-tagged ggt encoded on a plasmid. ggt mutant strains harboring either plasmid display WT-like growth, confirming that decreased proliferation in GSSG- or GSH-supplemented medium is due to genetic inactivation of ggt ( S3 Fig ).
10.1371/journal.pgen.1010834.g002
Fig 2
SAUSA300_0200- ggt supports S . aureus utilization of GSSG and GSH as sources of nutrient sulfur.
A. Orientation of SAUSA300_0200- ggt encoded within the S . aureus genome. B and C. Strains cultured in PN supplemented with 25 μM GSSG (B), 50 μM GSH (C), or 25 μM CSSC (C-inset). The mean of at least three independent trials and error bars representing ± 1 standard error of the mean are presented.
Given the decreased proliferation exhibited by the mutants in GSH- or GSSG-supplemented PN media and the fact that SAUSA300_0200 – SAUSA300_0204 encodes a putative ABC-transporter with a predicted γ-glutamyl transpeptidase (Ggt, SAUSA300_0204), we propose to rename SAUSA300_0200 – SAUSA300_0203 the g lutathione i mport s ystem ( gisABCD ) ( Fig 2A ). Domain architecture analysis of the GisABCD-Ggt system reveals that GisA contains ATP-binding cassette domains ( S4 Fig ). Consistent with the prediction, recombinant GisA purified from Escherichia coli exhibits ATP hydrolysis activity ( S5 Fig ). GisB and GisC contain nine predicted transmembrane regions, consistent with their membrane-bound annotation, while GisD contains a signal peptide with a putative lipid attachment site ( S4 Fig ). Finally, our domain architecture analysis predicts that ggt encodes a γ-glutamyl transpeptidase domain spanning most of the protein ( S4 Fig ).
We next sought to determine whether GisABCD promotes S . aureus GSSG acquisition. To test this, liquid chromatography-tandem mass spectrometry (LC-MS/MS) was used to quantify GSSG in WT, a mutant strain harboring an inframe deletion of gisABCD-ggt (Δ gis ), and the previously described ggt transposon mutant. Cells were cultured to mid-exponential phase using 25 μM CSSC as the sulfur source, washed, and then incubated in the absence or presence of 25 μM GSSG for 5 min. In keeping with the fact that S . aureus does not synthesize GSH and therefore is incapable of using it as its low molecular weight thiol, we were unable to detect GSSG in the absence of supplementation across the three strains ( Fig 3 ) [ 29 ]. Importantly, GSSG levels in supplemented WT were significantly increased compared to the supplemented Δ gis mutant ( Fig 3 ). There was no significant difference between GSSG levels in the transporter encoding ggt mutant compared to WT upon exposure ( Fig 3 ). Taken together with the proliferation phenotypes of the gis mutants and the bioinformatic evidence, these data support the conclusion that GisABCD functions as an importer.
10.1371/journal.pgen.1010834.g003
Fig 3
GisABCD promotes acquisition of GSSG.
Levels of GSSG in WT, Δ gis , or ggt mutant cells after a 5 min exposure measured by liquid chromatography-tandem mass spectrometry (LC-MS/MS). Samples were normalized by OD 600 (ODU). The mean and standard deviation of three biological replicates are presented. * P -value < 0.05 as determined by two-way ANOVA with Dunnett’s multiple comparisons test.
Ggt hydrolyzes GSH and GSSG γ-peptide bonds and localizes to the staphylococcal cytoplasm
A hallmark of γ-glutamyl transpeptidases is their capacity to cleave the GSH γ–peptide bond, liberating glutamate. To validate the domain prediction, we sought to determine whether S . aureus Ggt cleaves the γ–peptide bond linking glutamate to Cys in GSH and GSSG [ 30 ]. C-terminal His-tagged recombinant Ggt (rGgt) was expressed and purified from E . coli ( S6 Fig ). In other species, Ggt is translated as an inactive polypeptide that is auto-catalytically cleaved to generate approximate 40 kDa and 35 kDa subunits [ 30 , 31 ]. In keeping with this, a tripartite banding pattern consisting of full-length pro-Ggt (75 kDa) and smaller, mature enzyme subunits (40 kDa and 35 kDa) are observed ( S6 Fig ). Mature Ggt cleaves GSH by attacking the glutamyl residue, transferring it to the enzyme. Ultimately, water hydrolyzes the γ-peptide bond, liberating glutamate [ 32 ]. Therefore, to quantify S . aureus Ggt γ-glutamyl transpeptidase activity, rGgt was incubated with increasing concentrations of GSH or GSSG and glutamate release was measured via mass spectrometry. Glutamate was detected in reactions containing rGgt incubated in the presence of either GSSG or GSH ( Fig 4A and 4B ). Importantly, glutamate was not detected in reactions lacking substrate or rGgt, indicating glutamate release resulted from enzymatic activity. K m values of Ggt for GSSG and GSH were determined to be 38.6 μM and 58.5 μM, respectively. These values are similar to previously reported Ggt homologues expressed in other organisms (e.g. the K m of E . coli Ggt for GSH is 29 μM [ 33 ]). S . aureus rGgt V max for GSSG and GSH are 1.1 μmoles min -1 and 1.0 μmoles min -1 , respectively. These data support in silico predictions and provide a molecular explanation for the ggt mutant proliferation defect in medium supplemented with GSH or GSSG as sources of nutrient sulfur— ggt mutant cells fail to initiate Cys liberation due to an inability to hydrolyze the GSH or GSSG γ-peptide bond. Next, we sought to determine whether hydrolysis of GSH and GSSG occurs intracellularly or extracellularly by probing Ggt localization in S . aureus subcellular fractions.
10.1371/journal.pgen.1010834.g004
Fig 4
S . aureus Ggt liberates glutamate from GSH and GSSG in the cytoplasm.
A and B. rGgt was incubated with indicated concentrations of GSSG (A) or GSH (B). Mean glutamate release per minute was measured using four independent rGgt protein preparations. Glutamate released per minute was calculated and data were fit with the Michaelis Menten equation using GraphPad Prism. Error bars represent ± standard error of the mean. C. Supernatant or whole cell lysate (WCL) fractions of mid-exponential S . aureus ggt ::Tn harboring pOS1 P lgt :: ggt (His -) or pOS1 P lgt :: ggt -His (His +) probed with an anti-His tag antibody (αHis) or anti-Hla antibody (αHla). His-tagged recombinant Ggt (rGgt) was used as a positive control. D. Cell wall and protoplast lysate fractions generated from strains cultured to mid-exponential phase expressing His-tagged (+) or untagged (-) ggt were probed with αHis antibodies or anti-protein A antibodies (αSpa). E. Cell wall or intact protoplasts derived from the indicated cells were incubated in the presence (Prot K+) or absence (Prot K -) of proteinase K (Prot K). To monitor fractionation and Prot K activity, samples were probed with αSpa. F. Bioinformatic predictions and experimental evidence support the presented model of S . aureus import and catabolism of exogenous GSH and GSSG. The predicted substrate-binding protein, GisD, binds GSH or GSSG in the extracellular milieu, which are transported into the cytoplasm by the transmembrane permease complex, GisBC. GisA hydrolysis of ATP provides energy needed for import. Finally, GSH and GSSG are cleaved in the cytoplasm by Ggt, generating glutamate and cysteinyl-glycine or cysteinyl-glycine disulfide, depending on the substrate. The model illustration was created using BioRender.
Bacillus spp. secrete Ggt [ 19 ]; however, structural and cellular localization predictions of S . aureus Ggt did not detect canonical secretion signal sequences within the primary sequence ( S4 Fig ). For example, SignalP predictions detected considerably low likelihoods of signal peptide, twin-arginine translocation (TAT) signal peptide, or lipoprotein signal peptide sequences (0.007, 0.003, 0.008, respectively). TatP 1.0 also failed to predict a signal peptide [ 34 , 35 ]. Ggt localization varies across organisms, but in this case, localization has implications for the substrate of GisABCD. Extracellular Ggt supports a model whereby GisABCD imports Ggt products, whereas intracellular Ggt suggests GisABCD imports GSH and GSSG intact. Evidence for the latter is supported by the presence of GSSG in GSSG-exposed WT but not Δ gis mutant cells ( Fig 3 ). We used the previously described His-tagged Ggt expression vector (Ggt-His) that functionally complements the ggt ::Tn mutant ( S3 Fig ) to determine subcellular localization of the enzyme. S . aureus cells expressing native or Ggt-His were cultured in PN supplemented with 25 μM GSSG, collected at mid-exponential phase, and fractionated into supernatant and whole cell lysate (WCL). An αHis tag antibody (αHis) was used to monitor Ggt-His within each fraction. rGgt served as a size comparison control. Bands corresponding to Ggt-His were not detected in the supernatant fractions; however, a band at ~35kDa was observed in both the Ggt-His WCL and rGgt samples. This band is specific to Ggt-His as it was not observed in WCL generated from cells expressing Ggt lacking the His-tag and corresponds to the rGgt subunit containing the His-tag. Presence of Ggt-His signal within the WCL fraction supports the conclusion that Ggt is cell-associated ( Fig 4C ). To further resolve Ggt localization, cells were fractionated into peptidoglycan cell wall and protoplast fractions. Protoplasts were lysed, generating protoplast lysate that contained the cytoplasm and membrane. A His-dependent ~35 kDa signal was increasingly apparent in the protoplast lysate fraction compared to the cell wall fraction when equivalent levels of total protein are assessed ( Fig 4D ). Protein A (Spa), a protein covalently linked to the peptidoglycan cell wall, was used as a fractionation control. To rule out peripheral, external association of Ggt with the outer leaflet of the plasma membrane, intact protoplasts were isolated and sensitivity to proteinase K (Prot K) was monitored. Spa was used to control for fractionation and Prot K activity. As expected, Spa predominantly localizes to the cell wall and was sensitive to Prot K ( Fig 4E ). On the other hand, a His-dependent band corresponding to Ggt was present in the protoplast fraction in samples treated with or without Prot K ( Fig 4E ). This finding indicates that the membrane shields Ggt from Prot K, supporting the conclusion that the protein resides within the cytoplasm. Taken together, the lack of a secretion signal sequence, cell lysate association, and Prot K protection support a model whereby GSH and GSSG are imported intact and catabolized in cytoplasm, liberating Cys to satisfy the nutrient sulfur requirement ( Fig 4F ). Cytoplasmic localization of Ggt has been reported in only one other bacterial pathogen, Neisseria meningitidis [ 18 ].
GisABCD-Ggt is not required for systemic infection of S . aureus
Given its critical role in nutrient sulfur acquisition in vitro and the abundance of GSH in host tissues, we next tested the potential role of GisABCD-Ggt in S . aureus pathogenesis using a systemic murine model of infection. Contrary to our expectations, we found that mice infected with a gisB ::Tn mutant contained equivalent bacterial burdens compared to WT-infected animals ( S7 Fig ). This result indicates that GisABCD-Ggt is dispensable for S . aureus pathogenesis during systemic infection. There are at least two possibilities for the WT-like virulence exhibited by the gisB ::Tn mutant. First, acquisition of other sulfur sources, such as Cys or CSSC, is sufficient to fulfill nutrient sulfur acquisition in the absence of GSH or GSSG scavenging. In keeping with this, a previous study showed that the TcyP and TcyABC cysteine transporters support S . aureus fitness during heart and liver colonization [ 4 ]. The second possible explanation is that S . aureus encodes multiple GSH and GSSG transporters. In fact, while GisABCD-Ggt supports proliferation of S . aureus in micromolar concentrations of GSSG or GSH, increasing GSH concentrations to levels present in host tissues restores Δ gis mutant proliferation in PN medium ( S8 Fig ). However, amplifying GSSG concentrations did not stimulate Δ gis mutant proliferation. These results support the conclusion that GisABCD-Ggt is an absolute requirement for GSSG utilization but that another, potentially low-affinity, GSH transporter is also active in this pathogen.
GisABCD-Ggt is conserved in select Firmicutes
To define a function for GisABCD-Ggt beyond systemic host colonization, we traced the conservation and evolution of the system throughout Firmicutes using a molecular evolution and phylogenetic approach [ 36 ]. Due to the ubiquity of ABC-transporters across bacterial genera, we first focused on potential Ggt homologues. We found that many genera encode Ggt homologues, including Bacillus , Gracilibacillus , Lysinibacillus , and Brevibacterium ; however, subsequent identification of potential GisABCD homologues was limited to a small subset ( S9 Fig ). Overall distribution of GisABCD-Ggt homologues across Firmicutes revealed six distinct clusters. Cluster 3 is the least populated, containing species encoding only Ggt homologues (e.g., Clostridium tetanomorphum ). Firmicutes in clusters 2 and 4 encode Ggt and either a GisB or a GisC homologue, respectively (e.g., Gracibacillus thailandensis ). Genomes in Cluster 5 contain GisB, GisC, and Ggt homologues (e.g., Bacillus licheniformis ; S9 Fig ). Cluster 6 includes a few bacilli species that encode nearly the complete S . aureus GisABCD-Ggt system (e.g., B . subtilis ), but Cluster 1 stands apart as these genomes encode the full GisABCD-Ggt system similar to the S . aureus query sequences (>80% similarity; e.g., Staphylococcus simiae ). Notably, this cluster is restricted to members of the Staphylococcus genus ( S9 Fig ).
Homologues of a complete GisABCD-Ggt system were exclusively identified in the S . aureus -related cluster complex (Cluster 6)—which includes S . argenteus , S . schweitzeri , and S . simiae ( Fig 5A ) [ 37 , 38 ]. Conservation rapidly diverged with increasing 16S rRNA phylogenetic distance as the next closest related species, S . epidermidis , lacks apparent GisB and GisD homologues ( Fig 5B ). Furthermore, the GisA, GisC, and Ggt homologues exhibit exceedingly low percent similarities ( Fig 5A ). This finding suggests that S . epidermidis is incapable of utilizing low concentrations of GSH or GSSG as sources of nutrient sulfur.
10.1371/journal.pgen.1010834.g005
Fig 5
GisABCD is conserved in exclusive staphylococci and promotes competition in GSH- or GSSG-supplemented media.
A. Heatmap depicting percent similarity of GisA, GisB, GisC, GisD, and Ggt proteins across select staphylococcal species ( S . aureus USA300_FPR3757 GisA, GisB, GisC, GisD, and Ggt amino acid sequences were used as the starting point for the homology search). Conservation cross Firmicutes is depicted in S9 Fig . B. Phylogenic relationship between Staphylococcus species based on 16S rRNA sequences. C. In vitro competition experiments between S . epidermidis strain RP62a and WT (closed circles) or Δ gisABCD-ggt (Δ gis ) S . aureus (open squares). The mean and standard deviation are presented. *** indicates P -value = 0.0003, **** indicates P -value <0.0001 as determined by one-way ANOVA with a Sidak multiple test correction.
GisABCD-Ggt promotes interspecies Staphylococcus competition in a GSH-specific manner
To quantify S . epidermidis organic sulfur source-dependent proliferation, we first needed to define whether this species is capable of utilizing Met, a component of PN medium, as a source of nutrient sulfur. Additionally, S . epidermidis encodes predicted sulfate assimilation enzymes; thus, sulfate (MgSO 4 ) was removed from PN medium [ 39 , 40 ]. Supplementation of sulfate depleted PN with Met as the sole source of nutrient sulfur stimulates proliferation of S . epidermidis , but not S . aureus ( S10A and S10B Fig ). Therefore, Met was also removed and the resulting medium, PN mod , was used to assess GSH- and GSSG-dependent proliferation of S . epidermidis . Compared to S . aureus , S . epidermidis growth exhibited considerable lag phases in media supplementation with either 25 μM GSSG or 50 μM GSH ( S10C and S10E Fig ). Increasing GSH concentrations to 750 μM resulted in comparable proliferation between S . epidermidis and S . aureus ( S10F Fig ). Increasing GSSG concentrations from 25 μM to 375 μM slightly decreased the S . epidermidis lag phase ( S10D Fig ). While GSH and GSSG eventually promote S . epidermidis growth, clearly this species struggles to proliferate when GSSG and GSH are added as sulfur sources compared to S . aureus unless a physiologically relevant concentration of GSH is supplied. These results are consistent with the conclusion that GisABCD-Ggt facilitates efficient S . aureus GSSG and GSH utilization and suggest that low affinity GSH acquisition might be conserved between the two species.
Next, the capacity of GisABCD-Ggt to provide a competitive advantage to S . aureus over S . epidermidis was determined. To evaluate this, competitive indices between S . aureus and S . epidermidis were quantified after a 24 h co-culture in PN mod supplemented with different sulfur sources. As expected, S . aureus outcompeted both a S . epidermidis clinical isolate and the laboratory RP62a strain in PN mod supplemented with 25 μM GSSG or 50 μM GSH ( Figs 5C and S10G ). Conversely, S . epidermidis strains outcompeted S . aureus in medium with 50 μM Met. S . aureus exhibited a competitive advantage over S . epidermidis in 750 μM GSH, despite equivalent S . epidermidis monoculture proliferation ( Figs 5C and S10G ). Both S . epidermidis strains outcompeted S . aureus Δ gis in 25 μM GSSG and 50 μM GSH, underscoring the importance of GisABCD-Ggt in promoting S . aureus competition over S . epidermidis in environments containing GSH or GSSG. Equivalent quantities of S . epidermidis and S . aureus Δ gis were recovered in medium supplemented with 750 μM GSH ( Figs 5C and S10G ). These findings support the conclusion that Gis-independent GSH acquisition is conserved between S . aureus and S . epidermidis , while GisABCD-Ggt promotes S . aureus competition over S . epidermidis .
Discussion
This study increases our knowledge of S . aureus nutrient sulfur acquisition strategies by expanding the host-derived metabolites capable of satisfying the sulfur requirement to include GSSG and identifying proteins that play a pivotal role in the use of GSSG and GSH. The proteins include an ABC transporter, which we named GisABCD, and the γ-glutamyl transpeptidase, Ggt. Lithgow et al . previously showed that chemically defined agar medium supplemented with GSH, CSSC, Cys, sulfide, or thiosulfate stimulated proliferation of S . aureus while sulfate and methionine failed to promote appreciable growth [ 22 ]. Previous investigations into the sulfur acquisition pathways used by S . aureus revealed strategic employment of redundancy. In fact, S . aureus encodes two transporters, TcyP and TcyABC, to acquire Cys and CSSC while the current study highlights multiple mechanisms of GSH acquisition and catabolism [ 4 ]. This is exemplified by the finding that Δ gis mutant proliferation defects in medium supplemented with 50 μM GSH can be suppressed by increasing GSH quantities to physiologically relevant concentrations. A multifaceted strategy to acquire GSH makes implicit sense given the relative abundance of GSH in host cells and because the host invests considerable energy maintaining GSH in the reduced form [ 12 , 41 ]. S . aureus likely encounters an oxidized environment as a consequence of the host innate immune response; therefore, targeting GSSG increases the number of sulfur sources available within the overall host sulfur metabolite reservoir, potentially extending the dynamic range of tissue environments and conditions conducive to staphylococcal growth. Whether the abundance GSH and GSSG change relative to each other as a result of staphylococcal infection is a focus of ongoing investigations, though a recent report showed that GSH is the fourth most abundant metabolite increased in bronchoalveolar lavage as a consequence of S . aureus infection [ 42 ].
While this is the first study to identify a transporter involved in S . aureus GSH and GSSG utilization, GSH transporters have been established in other bacteria. The first bacterial GSH transporter to be discovered was the E . coli ABC transporter GsiABCD (formerly yliABCD ) [ 43 ]. Though GisABCD and GsiABCD are both members of the ABC transporter family, there are several noteworthy distinctions. First, gsiACBD is entirely operonic, while gisA is presumed to be divergently transcribed from gisBCD-ggt . Second, while gisBCD is operonic with ggt , GsiABCD lacks ggt , but encodes an ORF of unknown function called iaaA [ 43 , 44 ]. In addition, the amino acid sequences of the GisD and GsiB substrate binding lipoproteins indicate different mechanisms of GSH and GSSG recognition [ 45 ]. Together, these observations support a distinct nomenclature. Other mechanisms of GSH import have been reported in Gram-positive pathogens. For example, in Streptococcus mutans the substrate binding lipoprotein, GshT, promotes GSH import [ 46 , 47 ]. Subsequent studies revealed that GshT works in concert with TcyBC to acquire GSH and satisfy the sulfur requirement [ 48 ]. GshT homologues are encoded in other streptococci including S . pyogenes and S . pneumoniae . In S . pneumoniae , a Δ gshT mutant was shown to be more sensitive to superoxide, copper, cadmium, zinc, and innate-derived hypothiocyanous acid [ 13 , 49 , 50 ]. The Δ gshT mutant exhibited decreased burdens compared to WT in the nasal cavity and blood [ 13 ]. S . pyogenes Δ gshT mutants were more sensitive to H 2 O 2 and challenge with human neutrophils compared to WT [ 51 ]. Whether S . aureus scavenges GSH as a mechanism to protect against reactive oxygen species, in addition to satisfying nutritional requirements, is a focus of ongoing study.
GSH acquisition has also been explored in Gram-negative pathogens. For instance, Francisella tularensis is capable of replicating within macrophages and independent transposon mutagenesis screens identified Ggt as an important intracellular growth factor [ 3 , 5 ]. Additionally, genes harboring Tn insertions in the Cys-Gly transporter, DptA, and an alternative GSH peptidase, ChaC, also exhibited decreased intracellular proliferation within murine macrophages [ 5 ]. Together, these data support a model whereby GSH is catabolized in the F . tularensis periplasm by either ChaC or Ggt, generating Cys-Gly which is transported into the cytoplasm via DptA, fulfilling the sulfur requirement. The finding that Ggt and ChaC both participate in GSH catabolism reveal that multiple GSH degradation pathways can be active in a bacterial cell. Evidence of redundant mechanisms of GSH catabolism are also presented here as S . aureus proliferation in medium supplied with 750 uM GSH is independent of Ggt. These results suggest S . aureus encodes another peptidase capable of cleaving the GSH γ-peptide bond. Identification of this hypothetical peptidase is a focus of active investigation.
Ggt is conserved throughout all three kingdoms and is typically expressed extra-cytoplasmically. In humans, GGT is peripherally associated with the outer leaflet of the plasma membrane where it plays important roles in sulfur, Cys and redox homeostasis [ 52 ]. In the Gram-positive bacterial species B . subtilis and B . licheniformis , Ggt is secreted [ 17 , 19 ]. In Gram-negative bacteria, like E . coli , Proteus mirabilis , and Helicobacter pylori , Ggt is secreted to the periplasm where it partakes in GSH catabolism [ 16 , 53 – 55 ]. A notable exception is N . meningitidis in which Ggt resides in the cytoplasm [ 18 ]. S . aureus produces the low molecular weight thiol bacillithiol in lieu of GSH suggesting that cytoplasmic Ggt expression primarily functions in nutrient sulfur acquisition rather than GSH or Cys homeostasis [ 56 ]. However, N . meningitidis synthesizes GSH despite expressing Ggt within its cytoplasm [ 57 ]. These observations demonstrate that the relationship between Ggt function and localization is increasingly complex and will need to be resolved by additional analyses of Ggt localization in other bacterial species. Nonetheless, Ggt is an established virulence factor for F . tularensis , N . meningitidis , H . pylori , Acinetobacter baumanii , Campylobacter jejuni , and Bacillus anthracis [ 3 , 5 , 55 , 58 – 61 ]. Therefore, understanding whether the enzyme functions in nutrient sulfur acquisition, Cys and redox homeostasis, or both could provide a framework for designing toxic analogues that synergize with the host oxidative burst. While a role for Ggt in staphylococcal systemic infection can be ruled out, the enzyme may prove to be important in other models of pathogenesis. Alternatively, identification of the secondary GSH peptidase will help determine whether GSH catabolism is important for S . aureus blood stream infections.
Tracing conservation of GisABCD-Ggt across Firmicutes revealed that the system is encoded by an exclusive clade of Staphylococcus species which includes S . argenteus , S . schweitzeri , and S . simiae , but excludes S . epidermidis [ 38 ]. S . epidermidis proliferates poorly in a GSSG-supplemented medium and GisABCD-Ggt promotes S . aureus competition over S . epidermidis in GSH- and GSSG-limiting environments. These results further underscore the importance of GisABCD-Ggt to acquisition of GSSG and GSH as sources of nutrient sulfur. Both S . aureus and S . epidermidis are common residents of the human skin microflora and S . argenteus causes skin and soft tissue infections and sepsis in humans [ 62 – 64 ]. Therefore, the skin represents a dynamic host niche that could be further explored to determine whether GisABCD-Ggt provides a competitive advantage to S . aureus and S . argenteus over S . epidermidis . S . epidermidis , S . aureus , and S . schweitzeri also colonize nasal passages of humans or closely related primates [ 37 , 65 ]. However, the lack of GisABCD-Ggt conservation in S . epidermidis suggests that the system is not essential for nasal colonization. Interestingly, de novo Met synthesis is a critical pathway for S . aureus nasal colonization with Cys presumably serving as the precursor [ 66 ]. Previous reports indicate negative associations between S . aureus and S . epidermidis in the nasal cavity, but accounts of co-colonized individuals have also been established [ 67 , 68 ]. The presence of sulfate in nasal secretions seems to favor S . epidermidis [ 66 ], and yet roughly 30% of the population is colonized by S . aureus [ 69 ]. Given that the nares represent a potential environment where nutrient sulfur competition between the species could promote niche expansion of the winner, it is tempting to speculate that the distinct nutrient sulfur sources targeted by S . epidermidis and S . aureus allow them to gain a foothold, compete, and or coexist within the complex nasal environment.
Materials and methods
Ethics statement
This study was conducted in meticulous accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The approved protocol, PROTO201800068, was reviewed by the Animal Care and Use Committee at Michigan State University.
Bacterial strains used in this study
The WT S . aureus strain used in these studies was JE2, a laboratory derivative from the community-acquired, methicillin-resistant USA300 LAC [ 28 ]. SAUSA300_0200 -ggt mutant strains were generated via transduction of the transposon-inactivated gene from the Nebraska Transposon Mutant Library strain into JE2 using previously described techniques [ 28 , 70 ]. Bacterial strains used in this study are presented in S1 Table .
A strain harboring an in-frame deletion of gisABCD - ggt (Δ gis ) was constructed using a previously described allelic exchange methodology for S . aureus [ 71 ]. One kb upstream of SAUSA300_0200 and one kb downstream of ggt were amplified using primers listed in S2 Table and cloned into pKOR1-mcs. pKOR1-mcs was confirmed to have correct 1kb homology sequences by Sanger sequencing. The deletion strain was screened for hemolysis on blood agar plates and displayed WT-like hemolysis.
Isolation of clinical isolates
Four clinical isolates of S . aureus were obtained from de-identified specimens at a regional hospital clinical laboratory. Three abscess isolates were confirmed to be methicillin-resistant (strains 1055–1057) and the other was a methicillin-susceptible bone isolate (strain 1059). Identification and minimum inhibitory concentration assays were performed following Clinical and Laboratory Standards Institute approved methods. After initial isolation, subcultures were grown on tryptic soy agar (TSA, Remel) overnight.
Sulfur source-dependent proliferation analysis
Chemically defined (PN) medium was prepared as previously described [ 4 , 27 ]. PN medium was supplemented with 5 mg mL -1 glucose for this work. Prior to inoculation in PN S . aureus was cultured in tryptic soy broth (TSB, BD Bacto) overnight, washed with phosphate-buffered saline (PBS), and resuspended in PN medium to an OD 600 equal to 1. Round bottom 96-well plates containing PN supplemented with 5 mg mL -1 glucose and the indicated sulfur sources were inoculated with S . aureus strains at an initial inoculum of OD 600 0.01. Growth analysis was carried out in a Biotek Epoch 2 plate reader set to 37° C with continuous shaking for the indicated time. PN was modified to test sulfur source-dependent proliferation of S . epidermidis and S . aureus by replacing MgSO 4 with MgCl 2 and omitting Met, resulting in PN mod . S . aureus and S . epidermidis growth curves were performed as described above in PN mod supplemented with the indicated sulfur sources for 25 h. Sulfur sources were purchased from Millipore Sigma and GSH solutions were freshly prepared prior to each trial to limit oxidation. Alternatively, to ensure CSSC, GSH, and GSSG were maintained in their respective reduced or oxidized forms, stock solutions were prepared by weighing the appropriate amount of the chemical aerobically and immediately transferring it to an anaerobic chamber (Coy) with a 95%:5% nitrogen:hydrogen atmosphere. Sulfur sources were then resuspended in either anaerobically acclimated water (GSH and GSSG) or anaerobically acclimated 1 N HCl (CSSC). Anaerobic proliferation was monitored within the Coy chamber using a Biotek Epoch 2 plate reader.
Isolation of GSSG-proliferation impaired transposon mutants
Mutant strains that comprise the Nebraska transposon mutant library (NTML) were transferred from flat bottom 96-well plates archived at -80° C into flat bottom 96-well plates containing TSB supplemented with 10 μg mL -1 erythromycin using a 96-pronged replica plater (Millipore Sigma). Mutant cells were cultured overnight at 37° C and subcultured using a 1:150 dilution into a fresh round bottom 96-well plate containing 150 μL PN media supplemented with 25 μM GSSG instead of 25 μM CSSC as the primary sulfur source. Proliferation of each mutant within the 96-well plate was assessed by monitoring OD 600 over 22 h using a H1 Biotek plate reader set to 37° C with continuous shaking. This process was repeated for all twenty 96-well plates that comprise the NTML. To account for plate-to-plate variation, proliferation was averaged across all 96 transposon (Tn) mutant strains within a given 96-well plate. Tn mutants that exhibited decreased proliferation equal to at least two standard deviations below the 96-well plate average across mid-exponential and stationary phase were isolated by streak plating onto TSA supplemented with 10 μg mL -1 erythromycin. Isolated Tn mutants were validated in a secondary screen that compared proliferation in PN media supplemented with 25 μM GSSG to WT JE2 in technical and biological triplicate using round bottom 96-well plates and monitoring OD 600 over 22 h using a H1 Biotek plate reader set to 37° C with continuous shaking. Growth in TSB was used as a positive proliferation control. The following Tn mutants displayed significantly reduced growth compared to WT JE2 in 25 μM GSSG supplemented PN but not TSB: NE392 ( SAUSA300_0200 ::Tn), NE541 ( SAUSA300_0201 ::Tn), NE457 ( SAUSA300_0202 ::Tn), NE215 ( SAUSA300_0203 ::Tn), NE254 ( ggt ::Tn) ( S1 Table ). These mutants were selected for two additional steps of identification and validation. First, the location of the bursa aurealis Tn insertion for each mutant strain was verified using a previously described inverse PCR method followed by Sanger sequencing [ 28 ]. Second, the Tn mutations were backcrossed into the WT JE2 parental strain using a previously described phage transduction method [ 28 , 70 ] and the resulting backcrossed mutants were analyzed for proliferation as mentioned above in PN supplemented with 25 μM GSSG or the indicated sulfur source ( S1 Table ).
Liquid Chromatography-Tandem Mass spectrometry quantification of S . aureus -associated GSSG
WT JE2, Δ gis and ggt ::Tn strains were cultured in a 1 L Erlenmeyer flask containing 250 mL PN supplemented with 25 μM CSSC (4:1 flask-to-volume ratio). Cultures were incubated at 37° C with 225 rpm to an OD 600 of 0.4–0.5. At this time the three cultures (WT JE2, Δ gis and ggt ::Tn) were split into two 50 mL conical tubes each containing 50 mL aliquots of culture, the cells washed once in PBS, and resuspended in 50 mL of prewarmed PN media supplemented with either 0 μM or 25 μM GSSG. The six 50 mL conical tubes were incubated at 37° C without shaking for 5 min. Cells were centrifuged, washed twice with 10 mL PBS, and resuspended in 500 μL HPLC grade methanol containing [ 2 H 10 ] GSSG (Toronto Research Chemicals, Canada) as the internal standard. Cells were lysed via bead beating using 0.1 μm zirconia beads (BioSpec, Bartlesville, OK) in a Bullet Blender Storm 24 (Next Advanced, Troy, NY) at maximum speed for 2 min. The lysate was collected and centrifuged at maximum speed for 10 min to remove insoluble debris. Metabolite-containing supernatants for each of the six samples were dried using a Savant DNA 120 SpeedVac concentrator (Thermo Scientific) and were stored overnight at -20° C. Samples were reconstituted 500 μL of 0.1% formic acid in preparation for mass spectrometry.
Metabolites from a 10 μL aliquot from each sample were separated in reversed-phase mode using an Acquity HSS T3 column (1.8 μm 100 X 2.1 mm 2 ; Waters, Milfor, MA). Solvent A was 0.1% formic acid in water and solvent B was methanol. The mobile phase gradient was as follows: 0 min−100% A, 1 min − 100% A, 5 min − 60% A, 7 min − 1% A, 8 min − 1% A; 8.01 min − 100% A, and 10 min − 100% A with a flow rate of 0.3 mL/min. Mass spectrometry detection was performed using a Xevo G2-XS quadrupole time-of-flight (QTof) by positive ion electrospray ionization. GSSG detection was based on the retention time of 2.96 min to 3.00 min and mass accuracy using MassLynx Version 4.2 (Waters). GSSG calibration curves were generated with a six-point curve of serially diluted unlabeled GSSG standards with the corresponding concentration of [ 2 H 10 ] GSSG.
Expression and purification of Ggt and GisA
Open reading frames corresponding to the ggt or gisA genes were PCR amplified from the S . aureus JE2 genome with primers listed in S2 Table . Subsequently, the genes were cloned into the pET28b expression vector using NEB HiFi Gibson assembly kit (NEB, New England, MA) after the plasmid has been linearized with NcoI-HF and XhoI-HF. The assembly mixture was transformed into E . coli , cells were recovered in lysogeny broth (LB), and plated onto LB agar (Fisher) containing 50 μg mL -1 kanamycin and 5 mg mL -1 glucose. Plasmids were confirmed using Sanger sequencing and transformed into an E . coli NEB strain 3016 slyD mutant [ 72 ]. Transformed E . coli were cultured in LB with 50 μg mL -1 kanamycin overnight at 37° C with shaking at 225 rpm, sub-cultured 1:50 into 500 mL LB with 50 μg mL -1 kanamycin in a 2 L flask and grown to an OD 600 of 0.4–0.7. Ggt or GisA protein expression was induced by addition of 200 μM isopropyl-1-thio-β-D-galactopyranoside (IPTG) and the culture was separated into five 500 mL flasks containing 100 mL of culture and incubated for 4 h at 27° C and 225 rpm shaking. After induction, cells were centrifuged at 10,000 x g for 10 min at 4° C and washed with PBS. Resulting GisA and Ggt induction pellets were resuspended in 40 mL of buffer containing 50 mM tris, 200 mM KCl, 20 mM imidazole at pH 8, or 40-mL buffer containing 50 mM tris, 500 mM NaCl, 20 mM imidazole at pH 8, respectively. Cells were lysed via five consecutive cycles through a fluidizer set to 20,000 psi. Lysates were then centrifuged at 15,000 x g for 15 min to remove intact cells and the resulting supernatant was retained. To purify the target proteins, Ni-NTA chromatography was used. Purification was performed by incubating the cleared lysate with 1 mL Ni-NTA resin (Qiagen, Hilden, Germany) on a rotating platform at 4° C for 2 h. Protein was eluted with 50 mM tris 400 mM imidazole. Buffers used to purify GisA contained 200 mM KCl while buffers used to purify Ggt contained 500 mM NaCl. Each buffer contained 1x protease inhibitor cocktail (Millipore-Sigma). The GisA elutant was dialyzed using 10 mM tris, 200 mM KCl at pH 7.5 as the dialysis buffer for 18 h. The Ggt elutant was dialyzed using 10 mM tris, 150 mM at pH 7.0 as the dialysis buffer for 18 h. Both elutions were concentrated using a 10 kDa molecular weight cutoff protein concentrators. Purification was confirmed via electrophoresis using 12% SDS-PAGE gels. Protein concentrations were determined with the bicinchoninic acid (BCA) protein kit (Pierce ThermoFisher).
Quantitation of Ggt enzyme kinetics
γ-glutamyl transpeptidase reactions contained 5 μg recombinant Ggt, reaction buffer (10 mM tris with 150 mM NaCl), and the indicated concentrations of GSH and GSSG dissolved in reaction buffer. Reactions proceeded for 30 min at 37° C after which samples were incubated at 80° C for 5 min to stop the reaction. Samples were dried using a roto-vac speed vacuum and stored at -80° C until they were hydrated via resuspension in water, derivatized with carboxybenzyl (CBZ), and applied to a Waters Xevo TQ-S triple quadrupole mass spectrometer as previously described [ 73 ]. Peak processing was performed by MAVEN, and the signal was normalized to a 13 C-glutamine internal standard [ 74 ]. An external glutamate standard curve was generated using the same chromatographic conditions, and the signal was normalized to a 13 C-glutamine internal standard. A fit equation to the standard curve was employed to quantify glutamate within the samples. Glutamate released per min was calculated and data were fit to the Michaelis-Menten equation using GraphPad Prism. Data represent the average of glutamate quantified from four independent protein purifications.
Fractionation and western blot analysis of His-tagged S . aureus Ggt
The His-tagged ggt ORF was amplified from pET28b:: ggt and cloned into pOS1 P lgt digested with NdeI and HindIII using Gibson assembly to generate Ggt-His (His +). The ggt ORF was amplified from JE2 genomic DNA and pOS1 P lgt digested with NdeI and HindIII using Gibson assembly to generate the untagged Ggt (His -). Plasmids were confirmed by Sanger sequencing and transformed from E . coli DH5α into S . aureus RN4220 via electroporation. Plasmids were purified from RN4220 and transformed into JE2 ggt ::Tn. An empty vector control strain was generated by transforming JE2 and ggt ::Tn with pOS1 P lgt . To assess Ggt localization, S . aureus ggt ::Tn pOS1 P lgt :: ggt (His -) and ggt ::Tn pOS1 P lgt :: ggt -His (His +) cultures were prepared as previously described in the proliferation analysis section by sub-culturing into three, 250 mL flasks each containing 100 mL PN supplemented 25 μM GSSG and 10 μg mL -1 chloramphenicol at a starting OD 600 equal to 0.1. Cells were cultured for 4 h at 37° C and 225 rpm shaking to mid-exponential phase. At this time cells were collected via centrifugation, the supernatant recovered, and the cell pellet washed with PBS. A total of 50 mL of supernatant was precipitated with trichloracetic acid (TCA, final percent of 10% v/v), incubated overnight at 4° C, pelleted, and the resulting pellet washed twice with 95% ethanol. Whole cell lysates (WCL) were prepared by resuspending the washed cell pellet in membrane buffer (50 mM Tris-HCl pH 7.0, 10 mM MgCl 2 , 60 mM KCl) and transferring the suspension to a 2 mL bead beating tube containing 500 μL volume of 0.1 mm zirconia/silica beads (BioSpec Bartlesville, OK). Cells were bead beaten using a Mini-beadbeater 16 (BioSpec) thrice for 1 min and centrifuged 4,000 x g for 10 min to remove cells that were not lysed. Cell wall and lysed protoplast fractions were generated from whole cells cultured as described above. Staphylococci were pelleted via centrifugation, resuspended in TSM (100 mM Tris-HCl pH 7.0, 500 mM sucrose, 10 mM MgCl 2 ), and incubated with 100 μg lysostaphin for 1 h at 37° C. Protoplasts were recovered by centrifugation at 13,000 x g for 15 min. The supernatant containing the cell wall fraction was collected and the protoplasts were washed twice with TSM prior to resuspension in membrane buffer and lysed via bead beating as previously described. Intact protoplasts were removed via centrifugation at 4,000 x g for 10 min, generating the protoplast lysate. Protein concentrations of the whole cell lysate (WCL), cell wall, and protoplast lysate were determined using the BCA method (Pierce ThermoFisher). The cell wall fraction was further concentrated using the TCA precipitation method previously described for the supernatant. WCL and lysed protoplast fractions were mixed 1:1 by volume with 2x Laemmli buffer, boiled for 10 min, and an equivalent of 40 μg was loaded onto a 12% SDS-PAGE gel. Concentrated cell wall pellet was resuspended in 1x Laemmli buffer, boiled for 10 min, and 40 μg was loaded onto a 12% SDS-PAGE gel. PAGE was performed using Tris-glycine running buffer and samples were transferred at 65 volts for 1 h to a PVDF membrane (GVS North America) at 4° C. Membranes were incubated overnight in phosphate buffered saline tween-20 (PBST) with 3% bovine serum albumin (BSA) at 4° C with agitation. An αHis mouse antibody (Millipore) was used as the primary antibody at a 1:4,000 dilution in PBST supplemented with 5% BSA and incubated for 1 h with shaking. The membrane was washed thrice with PBST. An α-mouse IgG conjugated to horseradish peroxidase (HRP) was used as the secondary antibody at a dilution of 1:4,000 (Sigma-Aldrich). To assess supernatant and WCL fraction a rabbit anti-α-hemolysin (αHla) primary antibody (Sigma-Aldrich) was used at a 1:8,000 dilution. To control for proper cell wall fractionation a mouse anti-protein A (αSpa) primary antibody (Sigma-Aldrich) was used at 1:6,000 dilution. Membranes were washed thrice in PBST after incubation with either a 1:10,000 diluted horseradish peroxidase (HRP)- linked goat anti-rabbit IgG (Sigma) or 1:5,000 diluted HRP-linked goat anti-mouse IgG (Millipore) secondary antibodies. Membranes were developed using the ECL Prime kit (Cytiva, Marlborough, MA) and imaged using Amersham Imager 600 (GE Healthcare, Amersham, Buckinghamshire, UK). Fractionation was repeated with three independent sets of cultures and one set of immunoblots is presented.
Proteinase K protection of S . aureus Ggt
S . aureus ggt ::Tn pOS1 P lgt :: ggt and ggt ::Tn pOS1 P lgt :: ggt -His were cultured as described in the western blot analysis section. Cells were collected via centrifugation after normalizing OD 600 . The resulting pellet was washed with PBS, resuspended in TSM, and incubated with 100 μg lysostaphin for 1 h at 37° C prior to treatment with 10 μg mL -1 proteinase K (Thermo-Scientific) for 30 min at 37° C. At this time 5 mM phenylmethylsulfonyl fluoride was added to halt Proteinase K activity. Protoplasts were isolated from the cell wall fraction via centrifugation at 13,000 x g at 4° C for 15 min and washed twice with TSM. The cell wall fraction protein concentration was determined using BCA and was TCA precipitated as previously described. Protoplasts were lysed, and protein concentrated using the Wizard Genomic DNA Purification Kit (Promega) nuclei lysis buffer and protein precipitation solution, respectively. Cell wall and protoplast protein pellets were reconstituted in Laemmli buffer and immunoblotted as described above.
Identification of GisABCD-Ggt homologues across bacteria
The USA300_FPR3757 (assembly GCF_000013465.1) Ggt protein sequence (ABD22038.1) was used as the query protein for homology searches with MolEvolvR using DELTA-BLAST and the NCBI RefSeq database [ 36 , 75 – 77 ]. Data were filtered to include only Firmicutes that encoded Ggt homologues containing a glutamyl transpeptidase domain. Only genomes harboring Ggt homologues were used to query USA300_FPR3757 GisABCD. Percent similarities to the S . aureus GisABCD-Ggt protein amino acid sequences were used to generate a heatmap. The heatmap and hierarchical clustering of similar protein profiles were generated using the R package, pheatmap.
16S rRNA phylogenic analysis of Staphylococcal species
16S ribosomal RNA DNA sequences were retrieved from NCBI for the following strains: S . aureus USA300 FPR3757, S . epidermidis RP62s, S . simiae NCTC 13838, S . schweitzeri NCTC13712, and S . argenteus 58113. Parsimonious reconstruction was conducted using kSNP4 using default parameters and selecting Macrococcus caseolyticus FDAARGOS_868 as the outgroup [ 78 , 79 ].
Quantitation of S . aureus and S . epidermidis competition
S . aureus , S . aureus Δgis , and S . epidermidis were cultured overnight in TSB, pelleted, washed in PBS, and normalized to the same OD 600 in PN mod . Strains were mixed in a 1:1 ratio (v/v) and inoculated into 5 mL PN mod . PN mod was supplemented with 25 μM GSSG, 50 μM GSH, 750 μM GSH, or 50 μM Met. Dilution plating of the freshly mixed co-culture was plated onto mannitol salt agar (MSA) to quantify initial counts of each organism. Cultures were incubated for 24 h at 37° C with 225 rpm shaking after which the cultures were dilution plated onto MSA and allowed to grow for 48 h at 35° C. S . aureus ferments mannitol and appears yellow on MSA, while S . epidermidis does not and maintains a pink color; consequently, yellow and pink colored colonies were enumerated to assess quantities of each organism. Competitive indices (CI) were calculated by dividing the S . aureus to S . epidermidis output ratio by the S . aureus to S . epidermidis input ratio. A CI greater than one indicates more S . aureus than S . epidermidis while a CI less than one signifies greater quantities of S . epidermidis compared to S . aureus .
Supporting information
S1 Text
Supporting Materials and Methods.
(DOCX)
S1 Table
Staphylococcus strains used in this study.
(DOCX)
S2 Table
Primers used in this study.
(DOCX)
S1 Fig
GSSG supplementation as the sole source of nutrient sulfur stimulates proliferation of S . aureus .
(DOCX)
S2 Fig
GisABCD-Ggt promotes anaerobic proliferation in PN medium supplemented with GSSG or GSH.
(DOCX)
S3 Fig
Ectopic expression of native or His-tagged Ggt complements ggt mutant proliferation in medium supplemented with reduced or oxidized GSH.
(DOCX)
S4 Fig
Domain architectures and secondary structure predictions for the S . aureus GisABCD-Ggt system.
(DOCX)
S5 Fig
GisA encodes ATPase domain signatures and demonstrates ATP hydrolysis activity.
(DOCX)
S6 Fig
Heterologous expression and purification of S . aureus Ggt from Escherichia coli .
(DOCX)
S7 Fig
Virulence of a gisB ::Tn mutant strain mimics wild type.
(DOCX)
S8 Fig
S . aureus acquires GSH independent of GisABCD-Ggt in physiologically relevant concentrations of GSH.
(DOCX)
S9 Fig
Conservation of Ggt and GisABCD across Firmicutes.
(DOCX)
S10 Fig
S . epidermidis and S . aureus nutrient sulfur source utilization is distinct and promotes interspecies competition.
(DOCX)
S1 Data Table
Raw data that support the graphs in the figures.
(XLSX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Bacterial pathogens that are resistant to conventional antibiotics are becoming more difficult to treat. Natural products are a valuable and fundamental source for new drug discovery. Among the potential candidates, antimicrobial peptides (AMPs) have recently drawn increasing interest and have been proposed to be a promising alternative to conventional antibiotics [1] , [2] . AMPs exist in various species, including animals, plants, fungi and bacteria, and act in the innate defense of the organisms [3] . Investigating the antimicrobial mechanisms of the AMPs is important for their application.
Among the various AMPs, the fungi-derived cationic peptaibols are an important group that constitute a large family of approximately 850 members and could be a potential source for new antimicrobial drugs [4] . Peptaibols are characterized as linear peptides of 5–20 residues that contain a C-terminal amino alcohol, an acylated N-terminus, and a high proportion of non-standard amino acid residues, including α-amino isobutyric acid (Aib), isovaleric acid (Iva) and imino acid hydroxyproline (Hyp).
Peptaibols have been isolated from at least 23 fungal genera, and the genus Trichoderma and related genera are the most abundant sources of peptaibols [5] , [6] . While most conventional antibiotics do not act against fungi, and fungi-killing drugs are not active against bacteria [7] , peptaibols and other AMPs have excellent activity against both bacterial and fungal pathogens [7] . Moreover, earlier research has shown that peptaibols can also function as suppressors of tumor cells by inducing apoptosis and autophagy in hepatocellular carcinoma cells while causing no obvious harm to normal liver cells [8] .
Alterations in the morphological and nanomechanical properties of bacteria induced by AMPs are directly related to the mechanisms of antimicrobial action of the peptaibol agents. Therefore, atomic force microscopy (AFM), which is advantageous for investigating the ultrastructural and nanomechanical properties of bacteria [9] , [10] , [11] , is useful in researching the mechanisms of AMPs against microorganisms. However, currently published AFM studies on the antimicrobial effects of AMPs have mostly focused on Gram-negative bacteria, such as E. coli and Pseudomonas aeruginosa [12] , [13] , [14] , [15] , [16] , [17] , [18] , [19] , [20] , and studies using AFM to observe the morphological and nanomechanical properties of Gram-positive bacteria treated with AMPs are rare.
Here, we report the antimicrobial effects of trichokonin VI, a peptaibol produced by Trichoderma pseudokoningii, on the Gram-positive bacterium B. subtilis . Changes to the morphological and nanomechanical properties were monitored and probed by AFM in combination with other experimental methods. The mechanism of the action of trichokonin VI on bacterial membranes is also discussed.
Materials and Methods
Preparation of Trichokonin VI
Trichokonin VI was prepared from T. pseudokoningii SMF2 using solid-state fermentation following previously described methods [21] . The purity of the prepared Trichokonin VI was confirmed by HPLC (data not shown). The purified trichokonin VI (10 mg) was first dissolved in methanol (0.1 ml) and diluted with Milli-Q water (4.9 ml) to a final concentration of 2 mg/ml as a stock solution. The stock solution of trichokonin VI was stored at 4°C.
Preparation of Bacterial Cells
B. subtilis from a single colony was grown at 37°C in Mueller Hinton Broth (MHB; Hangzhou Microbial Reagent Co., Ltd.; China) overnight to a final concentration of 10 8 ∼10 9 CFU/ml. The bacterial culture was diluted with fresh MHB to a concentration of 10 6 CFU/ml for the susceptibility test and time-killing test.
Susceptibility Test
The minimal inhibitory concentration (MIC) was determined for trichokonin VI according to a modified microtiter-broth dilution method [22] . Trichokonin VI stock solution was diluted with Milli-Q water to a concentration of 200, 100, 50, 25, 12.5 6.25, 3.12, 1.56, 0.78 and 0.39 µM. Each dilution (100 µl) was transferred to a microtiter plate well, and 100 µl of bacterial suspension (at a concentration of 10 6 CFU/ml) was added to each cell. The final concentration of trichokonin VI was 100, 50, 25, 12.5 6.25, 3.12, 1.56, 0.78, 0.39 and 0.2 µM per well. The well that contained only growth medium was set as sterility control. The well that contained only bacterial suspension was set as growth control. Three rows were used for replicates on each microtiter plate for every concentration, and three microtiter plates were used for parallel experiment. The plates were covered with a plastic lid to avoid contamination and incubated at 37°C for 16–20 h without shaking. The MIC was defined as the lowest concentration that inhibited the visible growth of bacteria compared with the control sample. The experiment was repeated three times.
Time-kill Curves
The stock solution of trichokonin VI was diluted with Milli-Q water to a concentration of 1, 2 and 4 × MIC (corresponding to 25 µM, 50 µM and 100 µM), and 1 ml of the trichokonin VI solution was mixed with 1 ml of bacterial suspension in a test tube, which resulted in final trichokonin VI concentrations of 0.5, 1 and 2 × MIC (corresponding to 12.5 µM, 25 µM and 50 µM). The mixture of 1 ml bacterial suspension and 1 ml Milli-Q water was used as a control. The bacteria were incubated at 37°C with agitation. Samples were collected at predetermined time points (0, 2, 4, 6, 8, 10, 12 and 20 h). The samples were serially diluted with Milli-Q water and spread on Luria-Bertani (LB) broth agar plates, with three replicates used for every dilution of each time point. The number of viable colonies was counted after the plates were incubated at 37°C for 18–24 h.
AFM Imaging
The B. subtilis cells were incubated at 37°C with trichokonin VI at 0.5, 1 and 2 × MIC (corresponding to 12.5 µM, 25 µM and 50 µM). Control samples were not treated with trichokonin VI. Sample preparation process was the same with that described in Time-kill curves section. Samples were collected at 0.5, 1, 2, and 5 h time points and were centrifuged at 7,000 g for 10 min. The B. subtilis cells were suspended in Milli-Q water. A drop (2.5 µl) of B. subtilis was spread onto freshly cleaved mica and air-dried at room temperature before imaging. AFM images were obtained using a Multimode Nanoscope V (Bruker AXS; German) in tapping mode, and a probe (NSC11, MikroMasch) with a cantilever length of 90 µm was used.
Surface Roughness Analysis
The AFM data of the B. subtilis cells treated with trichokonin VI at the MIC (25 µM) for different time periods were flattened and used to calculate surface roughness. The surface roughness of a selected area was calculated using the NanoScope Analysis AFM software. Root mean square average roughness ( Rq ) was calculated with Equations 1 . (1) Z i is the height value at the i th point, and N is the number of points within the selected area. For each sample, roughness was measured near the center area on a bacterial cell with fixed sizes of 400 × 400 nm 2 . At least 5 different areas from different bacterial cells were measured.
Force-curve Measurements
B. subtilis cells treated with trichokonin VI at the MIC for different time periods were selected for the force-curve measurements. The spring constant k c of the probe cantilever (NSC11, MikroMasch) with a length of 200 µm was determined in air using the thermal noise method and Nanoscope software. Samples were first imaged in tapping mode to identify bacterial cells. At least 10 force curves per sample were collected in contact mode near the center of the selected bacterial cell. The spring constant of the cantilever used in the force experiments was approximately 2.0 N/m, which is smaller than that used to measure the mechanical properties of bacteria in air in some other works [23] , [24] . The spring constant of the cantilever was larger than the spring constant used in liquid [25] , [26] , [27] , due to the higher rigidity of the cells in air.
Mechanical Property Analysis
From the approaching branch of the force-distance curves, the stiffness of the bacteria could be determined. When force was applied on a softer sample, the slope of the linear portion of the approaching branch in the contact region was lower. The bacteria and cantilever could be modeled as two connected springs, and the spring constant of the bacteria could be determined by the slope ( s ) of the linear portion of the force curve [28] . The spring constant of the bacteria can be calculated according to Equation 2 . (2)
In Equation 3 , k b is the spring constant of the bacteria, and k c is the spring constant of the cantilever.
The jump-off contact point of the retract branch from the force-distance curve reflects the tip-sample adhesion interaction. This adhesive force may be due to biomacromolecules and thin water films on the cellular surface. The adhesion force between the probe tip and bacterial surface could be calculated according to Equation 3 . (3)
In Equation 4, F is the adhesion force and d is the deflection of the cantilever that resulted from the adhesive interaction between the tip and bacterial surface.
As the area in the retract branch from the force-distance curve below the zero force line represents the work performed by the adhesion force, the adhesion energy, which is also an indicator of bacterial surface properties, was subsequently calculated.
Leakage of Cellular UV-absorbing Materials
To analyze the damage of the bacterial cytoplasmic membrane caused by trichokonin VI, leakage of UV-absorbing cellular substances following sample treatment was monitored. Bacterial cell suspensions were mixed with trichokonin VI to a final concentration of 0.5, 1, 2 and 4 × MIC (corresponding to 12.5 µM, 25 µM, 50 µM and 100 µM), and the controls were cells without trichokonin VI treatment. Samples were collected at different time points. The samples were centrifuged at 7,000 g for 10 min to remove the bacterial cells. The supernatants were diluted, and their absorbance at 210 nm, 260 nm and 280 nm was recorded. The absorbance of trichokonin VI at concentrations of 0.5, 1, 2 and 4 × MIC (corresponding to 12.5 µM, 25 µM, 50 µM and 100 µM) were measured to exclude its absorption. Absorbance was measured at room temperature using a UV/VIS-550 spectrophotometer (Jasco; Japan).
Membrane Permeabilization Test
Trichokonin VI was added to B. subtilis cultures to a final concentration of 0.5, 1 and 2 × MIC (corresponding to 12.5 µM, 25 µM and 50 µM). Bacterial samples without treatment were used as a control. After treatment for 60 min, samples were centrifuged at 7,000 g for 10 min and washed twice in phosphate buffer. The bacterial cells were suspended in phosphate buffer to a concentration of 10 5 –10 6 CFU/ml and incubated with SYTOX Green (to a final concentration of 2 µM) for 10 min. Fluorescence of SYTOX Green was examined by flow cytometry (FACSCalibur, Becton-Dickinson).
Results
Antimicrobial Susceptibility Tests
The antimicrobial activity of trichokonin VI against B. subtilis was determined using the broth-dilution method. Gram-positive B. subtilis showed susceptibility to trichokonin VI in the test. In the wells on the microtiter plates with the concentration of trichokonin VI at 12.5 µM or lower, the deposits or turbidity which indicated the growth of bacteria was visible. However, in the wells with the concentration of trichokonin VI at 25 µM or higher, the liquids in the wells were clean, and neither turbidity nor deposits could be visualized. By carefully examining all parallel experiments for the antimicrobial susceptibility tests, the MIC value of trichokonin VI to B. subtilis was determined to be 25±0 µM, which is smaller than those of the AMPs such as LEAP-2 [29] and Cn-AMP1 [30] , similar to that of Magainin2 [29] and higher than those of LL-37 [29] and MSI-594 [31] . The error in the MIC could not be determined as it was less than the dilution factor used.
Time-kill Curves
The time-kill curves of trichokonin VI against B. subtilis are shown in Fig. 1 . Trichokonin VI exhibited a concentration-dependent antimicrobial activity in the bacterial viability test. Trichokonin VI at a concentration of 0.5 × MIC reduced cell growth in the first few hours when compared with the control; however, the growth curves were similar after 10 h of incubation. Trichokonin VI at the MIC inhibited the growth of B. subtilis ; however, the bacteria number increased slightly after treatment for 20 h. At a higher concentration (2 × MIC), trichokonin VI led to a progressive decrease in number of bacterial colony forming units, and no living bacteria were detected after 20 h.
10.1371/journal.pone.0045818.g001
Figure 1
Time-kill curves of trichokonin VI against B. subtilis .
The concentrations of trichokonin VI used were 0 µM, 12.5 µM, 25 µM and 50 µM, which corresponded to 0, 0.5, 1 and 2 × MIC, respectively.
AFM Images of B. subtilis
Images of the B. subtilis cells freshly collected from the bacterial culture were acquired ( Fig. 2 ). The surfaces of cells are reasonably smooth with a bacillary shape. No visible pores or ruptures could be observed in all examined cells. Cross sections of the bacterial cells were acquired. The measured length, width and height, shown in Fig. 2A , were 4.4, 1.3, and 0.44 µm, respectively, which are comparable to the reported dimensions [24] . However, the bacteria were observed to have a variation in length, and this variation may be a result of detecting living bacteria at different growth stages. Trace amount of methanol in trichokonin VI solution had no obvious effect on the morphology of the B. subtilis cells ( Fig. S1 ). Suspending bacteria in deionized water would bring about the hypo-osmotic shock to the bacterial cells. However, the sample preparation step in this experiment was short and no obvious random damage to the bacteria cells of B. subtilis was observed ( Fig. S2 ).
10.1371/journal.pone.0045818.g002
Figure 2
Morphology and section analysis of B. subtilis .
A and B are 3-dimensional AFM height images of B. subtilis cells, and B shows a dividing cell. C and F show the 2-dimensional height data from A and B, respectively, indicating the cross section position. D and E are the cross sections of the image indicated in C; and G and H are the cross sections of the image indicated in F. Scale bar is 2 µm.
AFM Images of B. subtilis Cells Treated with Trichokonin VI
The antimicrobial effect of trichokonin VI on B. subtilis at different treatment times was monitored. For each sample, randomly selected cells were examined and analyzed. B. subtilis was treated with trichokonin VI at the MIC for a period of 0.5, 1, 2 and 5 h, and the treated cells were collected and imaged. After treatment, the bacterial cells retained their rod-like form ( Fig. 3A–D ). However, minor corrugations to the bacterial surface could be distinguished after a 0.5-h incubation. The corrugation was more evident as the incubation time was increased. Treatment for 2 h or longer induced greater disruption in the cell morphology and collapse of the cell wall was observed. After incubation, the height of the B. subtilis cells declined from approximately 500 nm to less than 400 nm ( Fig. S3 ), which suggests a dramatic decrease in cell volume after trichokonin VI treatment.
10.1371/journal.pone.0045818.g003
Figure 3
The 3-dimensional height images of B. subtilis treated with trichokonin VI.
B. subtilis were treated with trichokonin VI at the MIC (25 µM, top row), 0.5 × MIC (12.5 µM, middle row) and 2 × MIC (50 µM, bottom row) for 0.5 h (first column), 1 h (second column), 2 h (third column) and 5 h (fourth column). Scale bar is 2 µm.
Changes to the surface characteristics of B. subtilis as a result of trichokonin VI treatment are obvious. Treatment at 0.5 × MIC resulted in minor surface perturbations after 2 h, and cells with more obvious changes could be detected after 5 h ( Fig. 3E–H and Fig. S4 ). Upon treatment at 2 × MIC, a more pronounced collapse of the cell wall could be detected ( Fig. 3I–L ), and collapse of the apical end of the cell was evident. Surface characteristics changed after the first 0.5 h of incubation. After 5 h of incubation, the height of the treated bacterial cells declined to less than 300 nm ( Fig. S5 ), suggesting a greater reduction in cell volume. The formation of granules on the B. subtilis cells after treatment with trichokonin VI was observed, and the granules are most likely formed from the condensation of cells because of the leakage of cellular materials.
Bacterial cells diluted only with growth broth and incubated for different time periods were set as control. These bacterial cells were also imaged ( Fig. S6 ). The morphologies of the bacteria were comparable to those freshly collected from bacteria culture. No obvious morphological collapse was observed. Thus, the results indicated that the alteration in bacterial morphological properties was induced by trichokonin VI.
Surface Roughness Analysis
B. subtilis cells treated with trichokonin VI showed alterations in surface characteristics compared with untreated cells. Images were analyzed by calculating the roughness of the bacterial cell surface. Surface roughness increased as the treatment time increased ( Fig. 4 ). The surface roughness, Rq , of the untreated cells over an area of 400 × 400 nm was calculated to be 9.9±0.9 nm. After incubation with trichokonin VI for 0.5 h, Rq slightly increased (10.2±1.4 nm) compared to the untreated cells. After 1 h of treatment, Rq increased to11.5±4.1 nm and it increased to 15.6±4.0 nm after 5 h of incubation. The results showed that the antimicrobial effects of trichokonin VI on B. subtilis are time dependent.
10.1371/journal.pone.0045818.g004
Figure 4
Surface roughness of B. subtilis treated with trichokonin VI at the MIC for different periods of time.
Rq was calculated in fixed-square size with a side length of 400 nm.
Mechanical Properties of the B. subtilis Cells
For force-distance measurements, the force curves were collected near the center of the bacterial cells. The force measurements were performed on the B. subtilis cells treated with trichokonin VI at the MIC for different incubation times. The spring constants of the B. subtilis cells were first calculated. For the untreated cells, the spring constant was 12.1±4.1 N/m ( Table 1 ). The measured value was larger than the values obtained from bacteria in liquid [27] . The influence of trichokonin VI to the nanomechanical properties of B. subtilis incubated with trichokonin VI for 0.5, 1 and 2 h was analyzed. It was observed from the force-distance curves that the slopes of the curves on the bacteria are all less steep than on the slopes on mica, and an increased incubation time led to shallower slopes ( Fig. 5 ). Statistical analysis revealed that the spring constant of the cells decreased with an increase in the incubation time. The spring constants of the B. subtilis cells treated with trichokonin VI are summarized in Table 1 . After a 0.5-h treatment, the spring constant decreased to 9.8±3.8 N/m and further dropped to 7.3±4.1 N/m after a 1-h incubation. When treated with trichokonin VI for 2 h, the spring constant of the B. subtilis cells decreased to 6.2±3.1 N/m, or approximately half the spring constant of untreated cells.
10.1371/journal.pone.0045818.g005
Figure 5
Representative approaching branch of force-distance curves on mica and B. subtilis .
Curves were collected on mica and B. subtilis incubated with trichokonin VI at the MIC (25 µM) for 0 min (A), 30 min (B), 60 min (C), and 120 min (D).
10.1371/journal.pone.0045818.t001 Table 1
Nanomechanical properties of B. subtilis incubated with trichokonin VI at MIC for different time periods.
Control
30 min
60 min
120 min
spring constant (N/m)
12.1±4.1
9.8±3.8
7.3±4.1
6.2±3.1
adhesive force (nN)
26.4±3.3
23.2±4.4
25.9±8.5
44.5±11.9
adhesive energy (×10 −18 J)
406±61
339±127
427±183
753±332
The retract branch of the force curves between the AFM tip and B. subtilis cells treated with trichokonin VI for different time periods showed that the adhesion force had only minor changes compared with the untreated cells in the first 1 h. However, the force dramatically increased after 2 h of incubation with trichokonin VI ( Fig. S7 and Table 1 ), which suggests marked changes to the bacterial surface properties. Work performed by adhesion forces, such as adhesion energy, was also calculated ( Table 1 ), and these forces exhibited similar trends as the adhesion force upon trichokonin VI treatment.
Leakage of Cellular UV-absorbing Materials
Leakage of cellular UV-absorbing substances is an indicator of changes in membrane permeability. Thus, we monitored the UV-absorption of the supernatant after B. subtilis cells were treated with trichokonin VI. Low doses of trichokonin VI had little effect on the leakage of cellular UV-absorbing substances ( Fig. 6 ). An increase in absorption at 260 and 280 nm could only be observed when B. subtilis cells were treated with high concentrations of trichokonin VI for a relatively long time. However, the absorption at 210 nm increased after incubation for only 20 min at high trichokonin VI concentrations. For treatment with low trichokonin VI concentrations, variations in the 210-nm absorption values after long incubation times were observed.
10.1371/journal.pone.0045818.g006
Figure 6
Release of UV-absorbing substances from B. subtilis treated with trichokonin VI at different concentrations.
B. subtilis were treated with trichokonin VI at 0.5 × MIC (12.5 µM), MIC (25 µM), 2 × MIC (50 µM) and 4 × MIC (100 µM). Absorbance was measured at 280 nm (A), 260 nm (B) and 210 nm (C).
Membrane Permeabilization Test
SYTOX Green is a high affinity nucleic acid dye that becomes more fluorescent when bound to DNA [32] , [33] . SYTOX Green does not cross the membranes of live cells but can enter into cells with a compromised membrane [32] , [33] . When SYTOX Green was applied to the healthy bacterial cells which were untreated with trichokonin VI, a weak but detectable fluorescence emission could be noticed ( Fig. 7A ). This weak fluorescence resulted from the binding of dye to bacterial surfaces, which is consistent with previous report [34] . Fluorescence intensity from B. subtilis cells treated with trichokonin VI at low concentrations appeared similar to the untreated cells. However, when the bacterial cells were treated with trichokonin VI at MIC or higher, an enhancement of SYTOX Green fluorescence could be observed ( Fig. 7B ), suggesting that the membranes of B. subtilis cells have been disrupted by the membrane-active compound trichokonin VI at these concentrations.
10.1371/journal.pone.0045818.g007
Figure 7
Membrane permeabilization effects of trichokonin VI on B. subtilis cells.
The B. subtilis cells were stained with SYTOX Green and analyzed with flow cytometry. The bacteria were untreated (A), treated with trichokonin VI at a concentration of 0.5 × MIC (B), MIC (C) or 2 × MIC (D) for 1 h.
Discussion
Trichokonin VI was determined to be effective against Gram-positive B. subtilis with a MIC of 25 µM. The MIC of trichokonin VI against B. subtilis was comparable to some AMPs, such as Magainin2 [29] , and higher than other AMPs, such as LL-37 [29] and MSI-594 [31] . Here, we combined AFM and other methods to investigate the antimicrobial effects of trichokonin VI on B. subtilis and determined that the peptaibol induced changes to the morphological and mechanical properties of the bacterium as well as permeabilized the bacterial membrane. Our results suggest a concentration- and time-dependent mode of action in modifying the morphological and nanomechanical properties of B. subtilis cells.
Microscopic methods are important tools for studying the interactions between AMPs and microorganisms. AFM is among the most powerful of the microscopic tools. Bacterial ultrastructures can be investigated with a high resolution, and their mechanical properties can be monitored by AFM. Height and surface roughness of the cells can be analyzed from the AFM images. Cellular morphological and mechanical properties are directly related to the mode of action of the antimicrobial agents; therefore, these advantages, in combination with other tools, may help us better understand the antimicrobial mechanisms of AMPs. Efforts have been made to evaluate the morphological changes of the bacteria treated with AMPs. Most of this work has focused on Gram-negative bacteria, which has an outer membrane that surrounds the peptidoglycan layer. However, there has been little work on the application of AFM to monitor the antimicrobial effects of AMPs on Gram-positive bacteria.
Imaging with a 3-dimensional scale using AFM allows for analysis of the surface roughness of bacterial cells. Roughness analysis of the cell surface can help us to understand and quantify the cellular alteration process, especially the detection of subtle changes caused by AMPs after treatment. It is widely accepted that AMPs interact with membranes, and the cell membrane of Gram-positive bacteria is surrounded by a thick layer of peptidoglycan. Therefore, changes in the roughness properties might be due to changes beneath the peptidoglycan layer. Along with the increased treatment time, surface roughness of B. subtilis cells gradually increased. Alteration to the surface roughness was a time-dependent process, which is in agreement with the morphological observations.
The results show that the mechanical strength of B. subtilis was altered under treatment with trichokonin VI. The mechanical properties of bacteria usually originate from the cell wall and turgor pressure of the cell [35] . Maintaining the bacterial structure necessarily involves the mechanical strength of the cell, which needs the cell wall to constrain the cellular contents under turgor pressure and defines the cell shape. Turgor pressure is generated from the cellular contents in the cytoplasm, which push against the cytoplasmic membrane and peptidoglycan wall. The permeabilization experiments indicated that the membranes are the target of trichokonin VI. With an increase in incubation time, it was observed that the spring constant of the B. subtilis cells gradually decreased. Thus, alterations in the mechanical properties are most likely due to the leakage of the intracellular materials, which led to changes in the turgor pressure, and resulted in a reduction in cellular elasticity as well as a collapse of the cell structure. From the retract part of the force-distance curve, adhesion forces between the AFM tip and bacterial surface can be measured. The adhesion force may originate from the interactions between the probe tip and molecules on the bacterial cell wall, including peptidoglycans, leaked biomacromolecules and water layers. A change in the adhesion force is an indicator of a change in the cell surface characteristics.
Research on morphological alterations has clearly exhibited a time- and concentration-dependent process for the antimicrobial effects of trichokonin VI on B. subtilis . Structural collapse usually results from leakage of intracellular materials or starvation. Structural shriveling of the bacterial cells could be observed in nutrient-free buffer [13] . In our experiments, the collapse was not a result of starvation ( Fig. S6 ). Alves et al. discovered that the AMPs induced a collapse in the septional region of the E. coli envelopes [13] . However, in the case of B. subtilis , it is interesting that the polar region of the B. subtilis cells appears to be sensitive to attack and is dependent of the trichokonin VI concentrations used in the experiments.
At 0.5 × MIC, the effects of trichokonin VI on B. subtilis cells are less clear. Minor morphological alterations could be observed after 2 h of treatment in the AFM images. After longer treatment times, the cell surface became rougher and section analysis showed a minor reduction in height, which indicated a loss of cellular volume. The minor morphological alterations in the B. subtilis cells suggest that trichokonin VI only has a slight effect on the bacteria at this concentration. The growth of B. subtilis could not be completely inhibited, which was determined using the time-killing assay.
At higher trichokonin VI concentrations (MIC), the antimicrobial effects on B. subtilis cells are much more obvious in the AFM experiments. After the first 0.5 h, only a slight collapse could be detected. After 1 h of treatment, granules were detected. A reduction in cell volume suggests that leakage of the intracellular material was occurring, and the changes in the mechanical properties suggest a decrease in bacterial turgor pressure, which was also a potential result from the leakage of the intracellular materials. Small molecules were allowed to cross the cell membrane, which indicates that the membranes have been permeabilized. Permeabilization of the membrane led to a reduction in cell volume, which caused alterations in morphology, surface roughness, and resulted in cell death.
When the concentration of trichokonin VI was increased to 2 × MIC or higher, more profound changes were detected. The AFM images show that the collapse of the cell structure was more obvious and led to the appearance of more granules on cellular surfaces. Considering the profound reduction in cell height, which was determined using section analysis, the granules are most likely are result of the collapse of the cell structure. Detection of leakage of cellular materials showed that trichokonin VI severely permeabilized the membrane at high concentrations, which caused a much quicker reduction in cellular volume and more profound changes in cell morphology.
In our leakage and membrane permeabilization experiments, the results indicated that the membrane did not allow small organic molecules to cross at low concentrations. At higher concentrations, the membranes were permeabilized and small molecules were allowed to cross the membrane. Trichokonin VI has a helical structure, which is a requirement for channel forming peptides, and the helical structure of trichokonin VI was determined using circular dichroism [36] . Furthermore, trichokonin VI has a length of 20 amino acid residues. Works on other helical peptaibols indicated that the long chain is also a requirement for the peptaibols to be able to penetrate the lipid bilayers and form channels [37] , [38] .Thus, based on our results and previous studies, it is possible that the target of trichokonin VI is the membrane of bacterial cells, and leakage of intracellular materials induced by trichokonin VI appears to be the reason for the changes in the morphological and nanomechanical properties probed by AFM.
In summary, we have demonstrated that trichokonin VI, a peptaibol isolated from T. pseudokoningii , is effective against the Gram-positive bacterium B. subtilis . Morphological and mechanical studies observed a concentration- and time-dependent effect against the B. subtilis cells. Nanoindentation experiments revealed a progressive decrease in the stiffness of the cells. Furthermore, we monitored the membrane permeabilization effect and suggest that leakage of intracellular materials is a potential mechanism of action of trichokonin VI on B. subtilis .
Supporting Information
Figure S1
Effect of methanol on B. subtilis cells. When preparing trichokonin VI stock solutions, 10 mg of trichokonin VI was dissolved in methanol (0.1 ml), and the solution was diluted with Milli-Q water (4.9 ml) to a concentration of 2 mg/ml as a stock solution. When trichokonin VI stock solutions were diluted to the MIC, the solution contained methanol at a concentration of 0.05% (v/v). Thus, it is necessary to examine whether trace amounts of methanol had any effect on B. subtilis . We incubated B. subtilis cells (10 6 CFU/ml) with methanol at a concentration of 0.05% (A) and 0.2% (B) (v/v), and these concentrations correspond to methanol content at the MIC and 4 × MIC. Samples were collected after 5 h of treatment. Cells were centrifuged at 7,000 g for 10 min and suspended in Milli-Q water. The cells were imaged by AFM, and the representative results are shown. As shown in the figures, B. subtilis cells retained their smooth surfaces and rod shape, and the sizes of the treated cells are comparable to that of the untreated cells ( Fig. 2 ), which suggests that methanol at the concentrations used in our experiments had no visible influence on the B. subtilis cells.
(TIF)
Figure S2
Effect of water on B. subtilis cells during sample preparation. Suspending the bacteria in deionized water would bring about the hypo-osmotic shock to the bacteria cells. To check the osmotic effect of incubation with water on the morphology of B. subtilis , we incubated B. subtilis in Milli-Q water for 0.5 h (A) and 1 h (B). These bacterial cells appeared intact with no visible holes, granules, or breakages in the cell envelop, and the morphologies of the bacteria are distinctly different from those treated by antimicrobial peptide. Thus we consider that this sample preparation step did not leave to random damage to the bacteria cells.
(TIF)
Figure S3
Section analysis of B. subtilis treated with trichokonin VI at the MIC for different incubation times. B. subtilis were treated with trichokonin VI at the MIC (25 µM) for 0.5 h (A, B), 1 h (C, D), 2 h (E, F) and 5 h (G, H). A and B are cross sections of the image in Fig. 3A ; C and D are cross sections of the image in Fig. 3B ; E and F are cross sections of the image in Fig. 3C ; and G and H are cross sections of the image in Fig. 3D . A, C, E and G are section profiles along the short axis of the bacterial cells. B, D, F and H are section profiles along the long axis of the bacterial cells.
(TIF)
Figure S4
Section analysis of B. subtilis treated with trichokonin VI at 0.5 × MIC for different times. B. subtilis were treated with trichokonin VI at 0.5 × MIC (12.5 µM) for 0.5 h (A, B), 1 h (C, D), 2 h (E, F) and 5 h (G, H). A and B are cross sections of the image in Fig. 3E ; C and D are cross sections of the image in Fig. 3F ; E and F are cross sections of the image in Fig. 3G ; G and H are cross sections of the image in Fig. 3H . A, C, E and G are section profiles along the short axis of the bacterial cells. B, D, F and H are section profiles along the long axis of the bacterial cells.
(TIF)
Figure S5
Section analysis of B. subtilis treated with trichokonin VI at 2 × MIC for different times. B. subtilis were treated with trichokonin VI at 2 × MIC (50 µM) for 0.5 h (A, B), 1 h (C, D), 2 h (E, F) and 5 h (G, H). A and B are cross sections of the image in Fig. 3I ; C and D are cross sections of the image in Fig. 3J ; E and F are cross sections of the image in Fig. 3K ; and G and H are cross sections of the image in Fig. 3L . A, C, E and G are section profiles along the short axis of the bacterial cells. B, D, F and H are section profiles along the long axis of the bacterial cells.
(TIF)
Figure S6
Representative images of the control B. subtilis cells which were not treated with trichokonin VI. The B. subtilis cells which were not treated with trichokonin VI were set as control samples to those which were treated with trichokonin VI. The morphologies of the control bacterial cells incubated for different time periods were monitored. The morphological properties of the bacterial cells incubated for 2 h (A) or 5 h (B) was comparable to that freshly collected from broth medium ( Fig. 2 ). Scale bar, 2 µm.
(TIF)
Figure S7
Representative retract branches of force-distance curves on mica and B. subtilis. Curves were collected on mica and B. subtilis incubated with trichokonin VI at the MIC (25 µM) for 0 h (A), 0.5 h (B), 1 h (C) and 2 h (D). The jump-off contact point represents the adhesion force between the probe tip and bacterial surface.
(TIF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Colorectal cancer (CRC) is one of the most common malignancies in the Western societies. Long-term survival of CRC-diagnosed patients is correlated with disease stage at diagnosis. In early stages as well as in selected patients with advanced disease, surgery is the main modality of treatment [1] . At least 40% of patients with CRC will develop either synchronous or metachronous distant metastases, most of them will succumb to their disease and die [2] . Some of characteristics of the malignant phenotype of CRC are correlated with overexpression and hyper-activation of receptor tyrosine kinases such as epidermal growth factor receptor (EGFR), which make these receptors attractive targets for cancer treatment [3] . In most CRC patients, the progression from normal colonic mucosa to cancer involves a defined cascade of molecular changes, that spreads over years [4] . Endoscopic polypectomy was shown to reduce CRC-related mortality [5] . This procedure requires fibro-optic colonoscopy visualization of the CRC tissue followed by histological evaluation. In addition the CRC tissues are often evaluated by RT-PCR [6] , immunohistochemistry [7] and in situ hybridization [8] techniques, which showed a much higher degree of discordance between primaries and related CRC metastases [9] .
EGFR is frequently overexpressed in a variety of solid tumor, of the brain, breast, lung, ovary and pancreas, and is associated with increased metastatic potential and poor prognosis of CRC [10] . Biological agents that inhibit EGFR have demonstrated clinical activity as single agents or in combination with chemotherapy, the most promising of these agents being cetuximab and panitumumab. Unfortunately, these antibodies are clinically effective in only a minority of patients with CRC [11] . The clinical success of these monoclonal antibody therapies is uniformly limited by the development of acquired resistance to EGFR blockade [12] . One mechanism to resistance was recently elucidated: cetuximab resistant cells contain an EGFR mutation in the extracellular domain (S492R) that impairs cetuximab, but not epidermal growth factor (EGF) binding [12] . Therefore, since the response to therapy require the EGFR target to be present, the development of BOI methods for quantitative detection of EGFR protein levels in CRC primary and secondary tumor tissues is necessary, in order to guide the treatment of individual selected for EGFR targeted antibody treatment and in particular those who relapse while on EGFR targeting therapies [13] . The advent of EGFR-targeted antibodies, cetuximab and panitumumab has paved the way to individualized medicine of mCRC. Current data suggests that the evaluation of KRAS and bRAF mutation and PI3K/PTEN alteration could be useful for selecting patients who are unlikely to respond to anti-EGFR-targeted antibodies. It was found that responsive CRC tumors carry wild type KRAS/bRAF and tend to have a modest, increase copy number of the EGFR gene, which is translated into a modest increase in EGFR level. Therefore, the EGFR gene copy number detection and quantitative evaluation of EGFR protein level will likely improve tailoring of cetuximab and panitumumab therapies for mCRC patients [12] . However, the technical difficulties of the immunohistochemistry technique, which is used to assess the expression of the EGFR in fixed tissues, may have limited the detection of small EGFR protein level increase so far [11] . Therefore, novel sensitive BOI methods of EGFR protein level in mCRC are needed. EGFR scintigraphy, represents such a method which is based on the binding, internalization, and retention of the radiolabeled EGFR-targeted agents in intracellular compartments and has been demonstrated with radiolabeled EGF and with radiolabeled monoclonal antibody directed against EGFR [14] . However, the disadvantage of these methods is the use of radioactive materials.
Near infrared (NIR) optical BOI offers unique advantages for diagnostics of mCRC: it offers high sensitivity, it can be used with different NIR tags and it can provide dynamic, real time in vitro and in vivo images by non radioactive means [15] . NIR light (700–1000 nm wavelength) can penetrate into tissue, and offers a potentially safe, noninvasive method of characterizing tumors [16] . In most applications, NIR BOI is used for assisting targeted fluorescent contrast agents that not only provide enhanced contrast, but also, more importantly, reveal specific molecular events associated with CRC tumor initiation and progression [17] . Recent studies have established the use of Affibody-mediated targeting of NIR excitable fluorescent contrast agents for the detection of malignant cells and tumors [18] .
Therefore, the aims of the present study were to develop and characterize novel in vitro CRC models that resembled CRC heterogeneity and to assess whether it is possible to quantify the level of EGFR in ex vivo fresh CRC tissue samples, orthotopic tumor in mice and newly developed cell lines models by using EGF conjugated with IRDye 800CW (EGF-NIR) probe. We found optimal conditions for BOI of EGFR using EGF-NIR probe in these models, applicable for endoscopic and Odyssey Infrared Imager analyses. Furthermore, by using image processing analysis and western blotting we confirmed that the intensity of EGF-NIR signal to background ratio reflects EGFR protein level in the in vitro CRC models and in situ human CRC tissues investigated.
10.1371/journal.pone.0048803.g001
Figure 1
Synthesis, purification, spectrum, electrophoresis properties and signaling of EGF-NIR.
(A) Reaction scheme for the synthesis of EGF-NIR conjugate, the first amino acid asparagine at the amino terminal is indicated as Asn1. (B) Separation of synthesis reaction mixture on gel permeation chromatography and of EGF-NIR sample from gel permeation on (C) anion exchange chromatography; EGF-NIR-full line (800 nm); gradient of NaCl-broken line; unconjugated EGF-dotted line. (D) HPLC separation of EGF-NIR purified from anion exchanger chromatography. Full line represents absorbance at 226 nm and dotted line indicates the gradient. Insert-12% SDS-PAGE analysis of 10 µg of EGF-NIR scanned with Odyssey and unmodified EGF stained with coomassie blue. (E) NIR spectrum of EGF-NIR [excitation (gray line) and emission (black line)]; Insert-IRDye 800CW NHS ester; (F) EGF-NIR induced Erk phosphorylation.
10.1371/journal.pone.0048803.g002
Figure 2
The specificity and selectivity of EGF-NIR probe binding measured by IC-NIR imaging.
HT-29 cells were incubated for 15 minutes at 37°C (A) and 4°C (B) with 7 nM EGF-NIR in the presence or absence of 100 nM EGF. Competition experiments with 500 nM of cetuximab, TGF-α or NRG1 were also conducted. In control experiments the cultures were incubated with 7 nM NIR-Dye to evaluate nonspecific labeling of the cells. The NIR intensity at 800 nm was estimated under identical conditions for all cultures and the mean ± SD (n = 9) is presented. Upper inserts: NIR scans; lower inserts: phase-contrast photomicrographs of the cultures,* p<0.05 vs. NIR-Dye; ** p<0.05 vs. EGF-NIR.
10.1371/journal.pone.0048803.g003
Figure 3
siRNA-induced knock down of EGFR evaluated by IC-NIR imaging.
HT-29 cells were transfected for 2 days with 5 nM anti-EGFR Silencer select siRNA or scrambled RNA or left untreated (control). Evaluation of EGFR expression was performed by In Cell NIR imaging using 7 nM EGF-NIR (black bars) and western blotting (gray bars). The values are mean ± SD (n = 3). * p<0.05 vs. scrambled or control.
Materials and Methods
Cell Culture
Human colon carcinoma cell lines HT-29, SW620, COLO205, A431 human epithelial squamous carcinoma cells and rat small intestine epithelial cell clone IEC 6 were purchased from American Type Culture Collection (ATCC, Manassas, VA) and adjusted for growth in Dulbecco’s Modified Eagle’s Medium (DMEM) containing 10% fetal bovine serum, 2 mM L-glutamine and 10000 U/ml penicillin and 100 µg/ml streptomycin. The cells were grown at 37°C, 6% CO 2 in a humidified incubator. All experiments were carried out under GLP conditions using a clean room according to ISO7 requirements (10,000 particles/m 3 ).
Preparation, Purification and Characterization of EGF-NIR
EGF-NIR was synthesized according to LI-COR Biosciences (Lincoln, NE, USA) instructions using IRDye 800CW NHS ester (2-(3-{5-[7-(5-amino-1-carboxy-pentylcarbamoyl)-heptanoylamino]-1-carboxy-pentyl}ureido)-pentanedioic acid) for conjugation to human recombinant EGF (Peprotech, Asia, Rehovot, Israel). Briefly, EGF (32 nmole) was incubated with 5 equivalents of IRDye 800CW NHS ester in 1 M K2HPO4, pH 9.0, for 2 hours in darkness at 20°C, with stirring. Following conjugation, the coupling mixture of free reagents and EGF-NIR was applied to Hiprep 26/10 (Fine Sephadex G-25, particles size 90 µm) desalting column (GE Healthcare, Life sciences, Buckinghamshire, UK) of a volume of 55 ml (Vo = 15 ml). This column was equilibrated and eluted at 10 ml/min with distilled water using FPLC AKTA P900 instrument (GE-Healthcare Life Sciences, Buckinghamshire, UK). This was followed by collecting the excluded peak and adjusting it to pH 8.0 by addition of 1 ml of 0.02 M Tris HCl buffer (pH 8.0). The solution was applied for anion exchange chromatography on Hitrap DEAE FF (DEAE Sepharose High Flow, GE Healthcare, Life sciences, Buckinghamshire, UK) equilibrated with 0.02 M Tris HCl buffer (pH 8.0) at a flow rate of 1 ml/min and EGF-NIR was eluted from the column using a gradient of 1–100% NaCl in the equilibration buffer. The purified EGF-NIR was dialyzed in 3000 cut off dialysis bags (Thomas Scientific, Swedesboro, NJ, USA) for 14 hours in dark at 4°C against distilled water. The distilled solution was lyophilized, and 0.1 mg of dry samples of EGF-NIR dissolved in 1% triflouroacetic acid (TFA) were finally separated by HPLC using a Sperisorb DDS2 column (LKB instruments, Gaithersburg MD) using two linear gradients: the first gradient from 10–35%, followed by a second gradient of 35–85% acetonitrile in 1% TFA. The purification was performed at a flow rate of 4.7 ml/min (100 bar pressure). The full run continued for 45 min and the EGF-NIR peak was estimated by the optical absorbance at 226 and 700 nm. Molar concentrations of dyes and EGF were calculated using molar extinction coefficients of 270,000 M −1 cm −1 for IRDye 800CW at 780 nm, and 18,000 M −1 cm −1 for EGF. Absorbance at 280 nm was used to calculate EGF protein concentration based on its molar extinction coefficient. Dual wavelength absorbance was used to determine dye: protein ratio. EGF-NIR emission was measured in PBS using a Fluoro-Max 4 spectrofluorimeter (JY Horiba, Edison, NJ, USA) with a Xenon arc lamp as excitation source of 774 nm in 1 cm cuvette, and at a scanning rate of 80 nm/sec. For validation, EGF-NIR from LI-COR Biosciences (Lincoln, NE, USA) was also used. EGF-NIR was submitted for analysis on 12% SDS-PAGE by comparison with native, unlabeled EGF. Samples of 10 µg proteins were separated and visualized by commassie blue staining and the NIR emission was measured by positioning and scanning the gel in the Odyssey® Infrared Imager (LI-COR Biosciences, Lincoln, NE, USA).
Western Blotting of EGFR and Erk Phosphorylation
The levels of EGFR in cell lines and CRC tissue and Erk phosphorylation, were estimated upon extraction with cell lysis buffer (Cell Signaling Technology, Inc. Danvers, MA, USA). The ability of EGF-NIR to stimulate ERK phosphorylation in A431 cells was compared to that of unlabeled EGF. 2×10 6 cells at 90% confluence in a 6 well plate were serum starved for 2 hours. Starvation media was replaced with regular medium containing 7 nM EGF-NIR. Cells were incubated for 15 min at room temperature and harvested. 50 µg protein lysates were separated by 10% polyacrylamide SDS-PAGE and transferred on ice to nitrocellulose membranes (90 V for 1.5 hours; Whatman, Dassel, Germany). Non-specific binding was blocked by incubation of the membranes for 2 hours at room temperature (RT) with 5% non-fat powdered milk (Bio-Rad, Hercules, CA, USA) in Tris buffered saline containing 0.1% Tween-20. Immunodetection was performed with monoclonal anti-EGFR antibody (Cell Signaling Technology, Inc. Danvers, MA, USA) or primary antibodies (1∶1,000) against phospho-or pan-Erk1/2 (Cell Signaling Technology, Inc. Danvers, MA, USA), followed by horseradish peroxidase (Jackson ImmunoResearch, West Grove, PA, USA) and developed with ECL (Pierce, Rockford, IL, USA).
10.1371/journal.pone.0048803.g004
Figure 4
Saturation, kinetics and sensitivity of EGF-NIR binding using IC-NIR imaging of CRC cultures.
(A) Left: a scheme of CRC polyp with high spots of transformed CRC cells (focal area) and heterogeneous area including both normal and transformed CRC cells; Middle: generation of a heterogeneous mixture at different ratios (%) between HT-29 and SW 620; 0 – no cells; +/++ -presence of different cell concentrations; Right: focal plating (in a ring) of HT-29 and A431 monolayer surrounded by SW620 monolayer; insert-the level of EGFR (170 kD) and non-mature EGFR (150 kD) in the cells; (B) The relationship between the NIR intensity of 15 min binding with 7 nM EGF-NIR (mean ± SD, n = 9) and the percentage of SW620 in the cell mixture with either HT-29 (closed circles) or HCT116 (open circles) * p<0.05 vs. 100% SW620; (C) The relationship between SBR (mean ± SD, n = 9) and EGF-NIR concentration; A431 (open circles); HT-29 (closed circles); binding was performed for 15 minutes. Insert: NIR scans; * p<0.05 vs. 0.01 nM; (D) The kinetics of 7 nM EGF-NIR binding (mean ± SD, n = 9) to focal cultures of A431 (open circles) or HT-29 (closed circles); Insert: NIR scans; * p<0.05 vs. 0 min.
Preparation of in vitro CRC Models and in Cell NIR Imaging (IC-NIR)
Homogenous monolayer of an individual cell line
CRC cells were plated at a density of 150,000 cells/well in 12 wells tissue culture plates (Nunc, Rochester, NY, USA) two days before the experiments generating homogenous cell monolayer. Thereafter, the culture medium was replaced with fresh medium containing 7 nM EGF-NIR, for 15 min at 4°C and 37°C, to measure total binding. At the end of the experiment the cultures were washed three times with 1 ml PBS and cell associated NIR intensity was estimated. To evaluate the non specific binding, sister cultures were incubated with the same concentration of EGF-NIR, at the same conditions, in the presence of excess of 100 nM EGF. Specific binding by EGF-NIR imaging is defined as the difference between the NIR intensity of total binding and NIR intensity of non specific binding. Competition experiments with 500 nM of either cetuximab, transforming growth factor α (TGF-α) or neuregulin 1 (NRG1) were performed by concomitant incubation of the competitor with EGF-NIR. The results are presented as the mean ± SD of at least three independent experiments (n = 9). The NIR imaging was estimated using Odyssey Infrared Imager at the following conditions range: resolution: 170–340 microns; pixel area: 0.03 mm 2 (approximately 15–20 cells); quality: medium-low; focus offset: 1–3; channels: 800 nm; intensity: 1–3.
Heterogeneous focal monolayer of CRC cells
To mimic high spots of transformed CRC cells in the polyp [19] and to enable direct measurements of signal to background ratio in the same experiment, a focal cell culture approach was used. Different colorectal cancer cell lines (with different levels of EGFR) or 15×10 3 A431 (high levels of EGFR) were plated to confluence inside a 4 mm inner diameter cloning ring (Sigma-Aldrich, St Louis, MO, USA) placed at the center of a well. Cells were left to adhere for 2 hours in an incubator. Thereafter, 15×10 3 SW620 cells (lacking EGFR) were plated in the cell-free area surrounding the cloning ring and left to adhere for 2 hours at the same conditions. Two types of such experiments were performed: a. one focus. b. two foci. At the end of cell adherence step, the cloning ring was removed and the cells were washed with culture medium. Two days after generating the model, the cultures were subjected to binding and imaging experiments as previously described. Signal/Background ratio (SBR) values indicate the ratio of NIR fluorescent signal of the central circle (focal area of CRC or A431 cells) to NIR fluorescent signal of outside area (SW620).
10.1371/journal.pone.0048803.g005
Figure 5
IC-NIR BOI of EGFR levels in CRC clones, in relation to the level of expression of CEA.
Different CRC cells were focally plated on a background of IEC6 enterocyte monolayer. The cultures were incubated for 15 min at 37°C with 7 nM EGF-NIR in the presence (nonspecific binding – white bars) or absence (total binding – grey bars) of 100 nM unmodified EGF. The signal (CRC cell line)/ background (IEC6) ratio was estimated at identical conditions for all cultures and is presented as the mean ± SD (n = 9). Significance: * p<0.05 compared to IEC6 values; ** p<0.05 compared to total binding of the respective group p< 0.05 compared to A431 and & p<0.05 compared to COLO 205; Lower inserts: NIR scans; Upper inserts: left-EGFR protein expression by western blotting; arrow indicate the position of mature EGFR 170 kD protein and non-glycosylated 150 kD protein; right-mRNA expression of CEA and β-actin in cell cultures.
Heterogeneous suspensions of CRC cells
To mimic heterogeneous distribution of transformed CRC cells in the polyp [20] and to enable evaluation of sensitivity of detection of the minimal amount of EGFR expressing cells among control normal enterocytes, heterogeneously mixed cultures of HT-29 and SW620 cells were prepared. Suspension cultures of HT-29 were mixed with suspension cultures of SW620 to generate different percentage of the individual cell culture in the same volume. SW620 cells could be distinguished from HT-29 cells by their elongated morphology. At conditions designated as “0%” the suspension contains 0% HT29 cells and 100% SW620 cells, and at “100%”, the suspension contains 100% HT-29 cells and 0% SW620 cells. In the other cases the percentages indicate the ratio between percentage HT-29 cells and percentage of SW620 cells. The binding experimental conditions and measurements of NIR fluorescent intensity (arbitrary units/mm 2 ) of the different heterogeneous cultures was performed as above.
10.1371/journal.pone.0048803.g006
Figure 6
Whole body in vivo and isolated tissue ex vivo EGF-NIR BOI of mice with HT-29 orthotopic tumors.
(A) Photograph of an orthotopic tumor and EGFR protein expression in the tumors; (B) Time course of EGF-NIR accumulation in tissues of tumor-bearing mice. The mice were injected i.v. with 1 nmol of EGF-NIR in untreated mice (upper row, n = 6) or mice pre-injected with 1 µg/ml cetuximab (lower row, n = 4); high resolution BOI of a mouse injected with EGF-NIR and circles indicate ROI measurements (C) Time course of tissue accumulation of EGF-NIR at 48 hours from mice presented in B; Signal intensity at 800 nm were normalized to background fluorescence using an arbitrary tumor circle (10–20 ROIs/mouse) compared to an identical area on the flank (adjacent muscle); * p<0.05 compared to EGF-NIR 4 hours; ** p<0.05 compared to mice injected with EGF-NIR; (D) EGF-NIR signal/background ratio in isolated tissue from the tumor-bearing mice 48 hours after injection. * p<0.05 compared to muscle, ** p<0.05 compared to liver; Insert: Upper-photographs of tissues in the dish; Middle-NIR images; Lower-spectral intensity maps; Intensity scale-red-brown (5) high expression; blue (3) very low expression.
RT-PCR and EGFR siRNA Silencing
Total RNA was isolated and genomic DNA was degraded from the RNA preparations, using the SV total RNA isolation system (Qiagen GmbH, Hilden, Germany). 1 µg of total RNA was reverse transcribed (Promega, Madison, WI, USA), according to the manufacturer’s instructions. PCR was performed in a final volume of 50 µl containing 5 µg cDNA, 50 pmol of each upstream sense and downstream sense primers of CEA or EGFR [21] , [22] , and 25 µl of GoTaq® Green Master Mix (Promega, Madison, WI). PCR experiments were conducted for 35 cycles. To generate various cDNA fragments, a Mastercycler gradient (Eppendorf, Hamburg, Germany) was programmed as follows: denaturation at 95°C for 1 min, annealing at 61°C and elongation at 72°C for 1 min. To knock down EGFR the standard amine transfection agent protocol of Ambion (Applied Biosystem, Austin, TX, USA) was followed. Briefly, 5 nM of 21 mer anti-EGFR Silencer select small interference RNA (siRNA) and scrambled RNA were reverse transfected into A431 cell cultures using siPORTNeoFX transfection agent according to manufacturer protocol. The cells at a density of 80,000 cells/ml were applied on 12 well plates and 2 days after transfections were analyzed. Knock down of EGFR mRNA was confirmed by Western blotting. The following carcino embryonic antigen (CEA) and EGFR primers, prepared by SyntezzaBioScience Ltd., Jerusalem, Israel, were used:
Human CEA CAM5: Sense: 5′-CGCATACAGTGGTCGAGAGA-3′ ; Antisense: 5′-ATTGCTGGAAAGTCCCATTG-3′
Rat CEA1: Sense: 5′-CTACAGGCTGAGGGATGCTC-3′ Antisense: 5′-GGTCCCGTCACAGTTACGTT-3′
Human EGFR: Sense: 5′-CGAGGGCAAATACAGCTT-3′ Antisense: AAATTCACCAATACCTATT-3′
Human EGFR siRNA: Sense: 5′-CCAUAAAUGCUACGAAUAUtt-3′ Antisense: 5′-AUAUUCGUAGCAUUUAUGGag-3′
Scrambled siRNA: Sense: 5′-UAACGACGCGACGACGUAATT-3′ Antisense: 5′-UUACGUCGUCGCGUCGUUATT-3′
Preparation and NIR Imaging of CRC Orthotopic Tumors in Mice
This study using Male Balb/c nude (Harlan, Israel) mice was approved, performed and supervised by the guidelines of The Chaim Sheba Medical Center Animal Care and Use Committee. HT-29 cells were trypsinized, washed and resuspended at concentration of 1×10 7 cells/ml in PBS. For tumor implantation, the mice were anesthetized by intra-peritoneal injection of a mixture of ketamine (100 mg/kg) and xylazine (20 mg/kg). Trans-anal injection of 1×10 6 HT-29 cells was performed under microscope magnification (X40) using a 27 g needle. The injection was directed submucosally into the distal, posterior rectum, approximately 2–3 mm beyond the anal canal and into the rectal mucosa [23] . Mice were monitored two times weekly for tumor initiation and progression. Tumors reached ∼ 0.75 cm in size at 3–4 weeks. In vivo imaging of EGF-NIR fluorescence in mice was performed with a LI-COR Biosciences small-animal imager Odyssey MousePOD ® . To visualize the tumors, 1 nmol of EGF-NIR in 100 µl saline was injected via the tail vein into tumor-positive mice in the presence (n = 4) or absence (n = 6) of cetuximab (1 µg/ml) and evaluated for systemic clearance by NIR imaging at intervals of 1–8 hours over a period of three days, after which time >95% of the signal had cleared. The mice were imaged up to two days post injection and euthanized. Statistical analysis of the images for each mouse was normalized using the same intensity scales, under the conditions previously described. SBR was calculated as follow: mean NIR intensity of the tumor divided by mean NIR intensity of the background of the adjacent muscle. Regions of interest (ROI) with identical areas were used for both tumor and background. The standard deviation of mean backgrounds was calculated using 10–20 ROIs. Due to tumor size differences between the animals receiving EGF-NIR, tumor signal divided by the background signal of similar size ROI corrected for area (pixels), provided the tumor two dimensional (2D) total labeling. Tumors, skeletal muscle and liver tissue of euthanized mice were dissected and their urine samples were collected. The tissues were weighted and introduced into plastic tubes dishes. The dishes were scanned on Odyssey Infrared Imager. NIR intensity of ROI of the tumor was compared to adjacent muscle to generate SBR. Isolated tissue analyses were performed by scanning at 800 nm channel, for the tissue accumulated EGF-NIR fluorescence signal and the SBR was calculated. The NIR imaging was estimated at the following conditions: resolution: 170–340 microns; pixel area: 0.03 mm 2 ; quality: medium-low; focus offset: 1–3; channels: 800 nm; intensity: 1–3.
Patients and CRC Tissue Specimen Collection
18 patients over the age of 18 years with histologically confirmed primary adenocarcinoma of the colon were offered participation in the study. 5 patients who received prior radiation or chemotherapy were ineligible for the study. The study protocol was approved by the Institutional Review Board (IRB, Helsinki Committee) of Hadassah-Hebrew University Medical Center. All samples were obtained from consenting study subjects undergoing surgical tumor resection who signed a written informed consent. All specimens underwent routine macroscopic and microscopic analysis by a board certified pathologist according to the College of American Pathologists (CAP) guidelines of histopathology reporting ( www.cap.org ). Tissues identified by the study pathologist as colonic adenocarcinoma and adjacent tissues [24] were then used for IC-NIR imaging.
10.1371/journal.pone.0048803.g007
Figure 7
BOI of EGF-NIR binding in human CRC tissues.
36 Slices of CRC tissues and 19 slices of adjacent colon tissue (n = 10–15 ROI in each slice) were submitted for ex vivo binding assay for 45 min at 37°C with 70 nM EGF-NIR in the presence (non specific) or absence (total binding) of 1 µM unlabeled EGF. The NIR intensity was estimated at identical conditions for all slices (n = 12). Significance: * p<0.01 compared to respective group in adjacent colon EGFR-, ** p<0.05 compared to respective group in CRC tissue EGFR-; Insert: typical western blotting for EGFR of the slices investigated.
10.1371/journal.pone.0048803.g008
Figure 8
Spectral intensity maps of BOI images of specific EGF-NIR binding in human CRC tissues.
Images of five typical slices, specifically labeled with EGF-NIR, as described in legend for Fig. 7. The 800 nm Odyssey Infrared Imager acquired images were processed using applied spectral imaging software, Spectral View. Intensity scale-red-brown (5) high expression of EGF-NIR binding; green (4) intermediate expression; blue (3) very low expression.
NIR BOI of CRC Tissues
Fresh tumor tissue or adjacent colon tissues were divided in horizontal slices, 230 µm thick, which were prepared using a vibratome VT1000S (Leica, Nussloch, Germany) and incubated in ice-cold DMEM binding solution. The slices were transferred to 24 well tissue culture plates filled with DMEM saturated with 95% O 2 and 5% CO 2 similar to conditions enabling rectal organ cultures [25] . The plates were maintained on ice for 45 min duration in DMEM and the binding experiment was performed by addition of 70 nM EGF-NIR in the presence (non specific binding) or absence (total binding) of 1 µM of unmodified EGF. From each tissue, triplicate slices were incubated with 70 nM NIR-Dye (IRDye800CW) to evaluate the non specific binding of the dye. The binding experiment was terminated by washing the tissue three times with cold PBS. The wet slices were transferred to new 24-well plates in 1 ml/well PBS and scanned for NIR imaging intensity using Odyssey Infrared Imager, under the conditions described above. Serially, over a two years period, 43 CRC and 23 adjacent colon tissue samples were evaluated. ROIs with identical areas were used for slices from the different experimental groups. The means and standard deviations of NIR intensity (arbitrary fluorescent units/mm 2 area) were calculated using 10–15 ROIs in each individual slice. Each slice submitted for the EGF-NIR binding experiment was also evaluated after the imaging for EGFR expression by western blotting. The data achieved was categorized according to EGFR positive CRC tissues, EGFR negative CRC tissues and adjacent colon tissue which in a majority were EGFR negative. The lack of EGFR was proved by Western blotting. 85% of the slices were included in the statistical analyses according to the following pharmacological criteria: a. total binding; b. specific binding were higher than the value of non specific absorption of NIR-Dye. The slices images were processed by high resolution imaging [26] using applied spectral imaging software, Spectral View TM (ASI, Migdal Ha’Emek, Israel).
Statistical Analysis
All results are presented as the mean ± SD of large series of independent experiments using different batches of EGF, cells, CRC and mice tumor tissues. All data were evaluated using the InStat statistics program (GraphPad, La Jolla, CA, USA). Statistically significant differences between experimental groups were determined by analysis of variance (ANOVA) with Bonferroni post-hoc test and considered significant when p<0.05.
Results
Preparation, Characterization and Validation of EGF-NIR Probe
The reaction scheme for synthesis of EGF-NIR is shown in Figure 1A: EGF molecule has one free amino group available for conjugation which is located at the amino-terminal of the protein and can be conjugated with IRDye800CW [27] , [28] . After reaction with these reagents the EGF-NIR was purified on size exclusion ( Fig. 1B ) followed by anion exchange chromatography ( Fig. 1C ). Final purification of the EGF conjugate was achieved by high pressure liquid chromatography ( Fig. 1D ) resulting with a single peak found very homogenous upon SDS-PAGE electrophoresis and NIR visualization ( Fig. 1D -insert). EGF absorbance and excitation spectra clearly indicates the presence of IRDye800CW ( Fig. 1E ) in the conjugate. Compared to the unmodified EGF, EGF-NIR stimulated EGFR resulting with increased phosphorylation of Erk ( Fig. 1F ). These findings indicate that the modified EGF preserved bioactivity, as evident from the ability to stimulate EGFR-induced Erk signaling pathway. Validation of EGF-NIR binding properties on CRC cell line was performed using HT-29 cultures expressing relative high level of EGFR. Comparison of NIR intensity signals of homogeneous monolayer cultures of HT-29 incubated with EGF-NIR in the presence or absence of different ligands at 4°C ( Fig. 2B ) and 37°C ( Fig. 2A ) conditions, reflects EGF-NIR cell surface binding and ligand-induced receptor internalization, respectively. Furthermore, the results clearly indicate specific and selective binding of EGF-NIR probe, similar to unmodified EGF, as concluded from lack of competition with NRG1 and partial competition with EGF, TGF-α and cetuximab.To demonstrate a direct relationship between EGFR level and binding of the EGF-NIR, we used siRNA anti-EGFR to knockdown the receptor. The efficiency of siRNA was validated using RT-PCR and indicated a knock down of ∼ 50% of EGFR mRNA. EGFR protein level, as evident from the western blotting and EGF-NIR binding, was reduced by 65% and 35% (p < 0.05), respectively, as compared to untreated controls or cultures transfected with scrambled RNA ( Fig. 3 ), as previously documented with other cells [22] . Therefore, we conclude that the decrease NIR intensity (decreased specific binding of EGF-NIR) reflects the reduced expression of EGFR protein level.
Development of Novel CRC in vitro Models for BOI
To mimic high spots of transformed CRC cells in the tumor, a focal cell culture approach ( Fig. 4A -focal plating) was developed using HT-29 cells expressing high level of EGFR and SW620 lacking EGFR ( Fig. 4A -insert) and for comparative purposes A431 overexpressing EGFR. In another approach to mimic clinical presentation of diffused tumor cells, a suspension of both types of above cells, at different ratios was prepared and plated ( Fig. 4A -heterogeneous plating). Heterogeneous mixed cultures of HT-29 or HCT116 and SW620, at different ratios were incubated with 7 nM EGF-NIR for 15 min ( Fig. 4B ). It is evident that the presence of 15–30% HT-29 or HCT116 cells in a heterogeneous mixture with SW620 cells, represented the threshold of detection of the minimal amount of EGFR expressing cells among EGFR negative cells, providing a significant NIR intensity signal ( Fig. 4B ). Using the focal culture model, saturation ( Fig. 4C ) and time-course ( Fig. 4D ) binding experiments were preformed using IC-NIR imaging of the CRC HT-29 cell line (EGFR++) compared to A431. As found for A431, 7 nM EGF-NIR induced maximal SBR, however the maximal binding was optimal between 1 and 5 min and thereafter decreased due to increased nonspecific binding to SW620. The lower SBR in CRC HT29 compared to A431 in the different experiments is in direct correlation with the lower level of EGFR in the cells. Based on these experiments, the optimal conditions for IC-NIR imaging with the focal and heterogeneous cultures models using either A431 or HT-29 cells were setup on 7 nM EGF-NIR and 15 min of incubation since at higher concentrations of EGF-NIR or time of incubation the SBR decreased due to the increased background (increased nonspecific binding of EGF-NIR).
Since EGFR-targeting therapies are currently in use for the treatment of metastatic CRC [29] the above in vitro CRC models may be considered for BOI. To further evaluate the relationship between the EGFR level and SBR using EGF-NIR we took advantage of a panel of human CRC cell lines with different expression levels of EGFR and CEA ( Fig. 5 , top). Using the focal model, we performed IC-NIR BOI with these cell lines which also exhibit highly variable growth and metastatic capacities [30] . For comparison purposes COLO 205 (EGFR++) and A431 (EGFR+++) or HT 29 cells (EGFR++) were plated in different rings surrounded by small intestine epithelial IEC6 cells (EGFR -). The binding experiment was performed in the absence and presence of EGF to measure total and non specific binding, respectively ( Fig. 5 ). It was found that the SBR of EGF-NIR binding to COLO 205 is significantly higher than that of HT 29, and lower than that of A431, in accordance to EGFR 170 kDa protein isoform expression level ( Fig. 5 -insert). Although the SBR of total binding of CRC clones was between 3–6 (about 3 fold lower than A431), the non specific values of SBR between 1–2 allowed calculations of specific binding (NIR intensity) values of 3–4, which are in the sensitivity range of NIR detection systems [31] , [32] .
BOI of Tumors Positive for EGFR in Orthotopic Mice Model
To confirm the suitability of EGF-NIR for in vivo BOI, HT-29 CRC orthotopic tumors in nude mice were generated ( Fig. 6A ). Western blotting analysis of dissected tumors confirmed the expression of EGFR in the tumors ( Fig. 6A ). Figs. 6B presents typical NIR fluorescence images of mice bearing EGFR positive tumors 4, 24 and 48 hours after i.v. injection of 1 nmol EGF-NIR, an optimal dose which affords the highest SBR, clearance and imaging results [28] . The control groups were i.v. injected first with 1 µg/ml cetuximab to block EGFR in orthotopic tumor and tissues [28] and after 5 hours injected with 1 nmol of EGF-NIR. In the first four hours the animal injected with the imaging agent show a very strong whole body fluorescence signal. After 24 hours, about 80% of the fluorescence signal cleared, and after 48 hours the signal intensity fell back to background level, in contrast to EGF-NIR labeling in tissues expressing EGFR: tumor, bladder and liver ( Fig. 6B right image, taken at high resolution). Quantitative 2D surface measurements of the SBRs 4, 24 and 48 hours after injection, at each pixel (mm 2 ), on ROI taken from the tumor and liver region, compared to an identical area on the flank (adjacent muscle) region are presented in Fig. 6C . It is evident that already after 24 hours EGF-NIR is specifically (competitive with cetuximab) and significantly accumulating in the tumor, providing an SBR value around 4.2 ± 0.6. This value is further increased at 48 hours. In the liver, a tissue highly abundant in EGFR [16] , a significant accumulation or slowly clearance was observed after 48 hours and EGF-NIR binding was fully competitive with i.v. injected cetuximab. We assume that the fast accumulation at 24 hours of EGF-NIR reflects the very high level of EGFR in the tumors responsible for its uptake. Macroscopic 3D NIR estimations of harvested organs and urine were performed 48 hours after injection ( Fig. 6D ). The urine was strongly fluorescent, since EGF is known to be excreted unmodified into the urine [33] , providing the highest SBR NIR fluorescence compared to that of the muscle. The liver/muscle SBR NIR fluorescence strongly exceeded compared to tumor ( Fig. 6D ) since the NIR intensity of the isolated tissue is higher than its value upon scanning the whole animal. Tumor/ muscle SBR NIR fluorescence was 11.3 ± 1.7 while the SBR 2D measurements of the whole animals was 6.6 ± 1.1. NIR imaging of isolated organs, in close proximity with the laser instrument, is more sensitive than the imaging of the organs in the animals, due to the lack of absorbance and scattering of the NIR fluorescence by the animal tissues [28] .
Specificity and Heterogeneity of CRC Tissue Upon ex vivo NIR Imaging
Figs 7 and 8 present ex vivo NIR imaging of human CRC tissues performed with EGF-NIR and processed by high resolution imaging to emphasize the distribution of EGFR. The binding experiments were conducted with fresh slices from CRC tissues identified by western blotting as either EGFR positive or EGFR negative ( Fig. 7 -insert). We also assessed slices from adjacent, colonic tissue close to the tumor, identified as EGFR negative in the same way previously performed for detection of CRC associated transcript-1 biomarker in malignant and pre-malignant CRC human tissues [24] . The binding was performed for 45 min to allow optimal diffusion of EGF-NIR into the slice. Significant total and specific binding of EGFNIR imaging agent was detected only in the CRC tissues positive for EGFR ( Fig. 7 ). The ratios between specific binding of EGF-NIR to “CRC tissue EGFR+/adjacent colon EGFR-” or “CRC tissue EGFR+/CRC tissue EGFR-” were 46 and 16 (p < 0.05), respectively. These high values are easily detected by NIR instruments scanners. Fig. 8 presents 5 slices randomly chosen from the different tissues and their NIR imaging was processed by high resolution analyses. Spectral intensity maps of the slices indicate distinct focal areas of high level of expression of EGFR in CRC tissues EGFR+, with very high specific NIR labeling, few high spots of EGFR in adjacent colon tissue EGFR-with practically no specific NIR labeling, and very low level of EGFR in CRC tissue EGFR-with extremely low specific NIR labeling. These findings indicate a direct relationship between EGFR expression and EGF-NIR BOI and further indicate heterogeneity of CRC tissues in EGFR expression analyzed by spectral imaging software.
Discussion
A new platform of NIR reagents based on IRDye 800CW was developed and used for preparation of IRDye 800CW conjugated EGF [27] , [28] , [34] which was accepted by NIH database as a molecular imaging and contrast agent for optical visualization of prostate carcinomas in vitro and in mice. We took advantage of this progress to prepare an EGF-NIR bio-imaging agent according to this procedure, and evaluated its pharmacological properties for recognition of EGFR in novel CRC models in vitro , resembling tumor heterogeneity, orthotopic CRC tumors in mice and CRC tissue slices ex vivo , thus translating results from CRC cells to human tissue specimens. EGF-NIR was homogeneous and pharmacologically active, as evident from binding experiments using CRC models. It specifically and selectively binds EGFR but not NRG 1 and its binding at 37°C was significantly higher than at 4°C, indicating that it is internalized at 37°C as expected from native growth factor. The binding of the probe measured by SBR directly reflected the level of EGFR, as evident from binding experiments with cells expressing lower EGFR levels, due to siRNA knockdown of EGFR and experiments with CRC cultures expressing different levels of EGFR. The EGF-NIR, rapidly saturated the EGFR in a focal CRC set-up, and was able to detect with high sensitivity 15–30% of EGFR expressing cells in a heterogeneous mixture of carcinoma/enterocyte cells in vitro . Consistent with the in vitro findings, specific targeting of EGF-NIR to EGFR was demonstrated by analyzing the NIR images of mice bearing EGFR positive CRC orthotopic tumors. Selective accumulation of EGF-NIR as evident from in vivo competition with cetuximab, was clearly seen after 24 and 48 hours in the tumor and EGFR positive organs such as the liver. The measurements of EGF-NIR in the tumors normalized to skeletal muscle and compared to the liver, indicated a faster accumulation of the agent in the tumor, compared to the liver, supporting the possibility of a specific EGFR- mediated process. These findings are reminiscent of accumulation studies using EGF- conjugated to Cy5.5 fluorophore in mice with breast cancer xenografts [16] and experiments of confocal endomicroscopy targeting EGFR with fluorescently-labeled antibodies [35] . We found relatively high SBRs for EGF-NIR binding to the whole animal and isolated tumors, supporting the notion that EGF-NIR is a suitable imaging agent for CRC tumor visualization using NIR endoscopy. This conclusion is also in line with studies in which dual labeling of a peptide with IRDye800CW and In 111 , generated satisfactory SBRs of 1.6 for NIR and 1.7 for nuclear imaging providing high quality images of the tumors in mice with human melanoma xenografts [36] . As documented by Goetz et al. [35] , which measured a factor of 10 fold signal distinction of neoplastic from non-neoplastic CRC tissue using confocal laser endomicroscopy, we would like to stress that SBR value of 10 for isolated organs, scanned with the Odyssey Infrared Imager, represent a much higher value than the sensitivity threshold required by a NIR endoscope. SBR is an important parameter influencing the sensitivity of detection. Since EGFR is overexpressed in CRC tissue, it enables upon binding EGF-NIR, a relatively high SBR, facilitating detection with NIR scanners.
Following satisfactory results obtained in mice, we further characterized the applicability of EGF-NIR for ex vivo NIR imaging of human CRC tissues. Mouse and human EGF show more than 70% homology [37] , therefore the EGF-NIR probe efficiently cross reacts with both mouse and human EGFR, enabling experiments with both mouse and human CRC tissues. We proved its specific binding to CRC tissue-EGFR positive, but not CRC tissue-EGFR negative and adjacent colon tissue-EGFR negative ( Fig. 7 and 8 ), as previously demonstrated using FITC-labeled anti-EGFR antibody imaging of CRC with confocal endomicroscopy [35] . The ratios between specific binding of EGF-NIR to “CRC tissue-EGFR positive/CRC tissue-EGFR negative” or “CRC tissue-EGFR positive/adjacent colon tissue-EGFR negative” were 46 and 16 (p < 0.05), respectively. These high values prove that it is possible to generate a strong NIR fluorescence signal to detect fresh, unfixed CRC tissue-EGFR positive, using NIR scanners, as previously documented with an anti CEA antibody labeled with indocyanin green (ICG) NIR-Dye [31] . The heterogeneity of EGFR distribution among different slices and in the same slice of CRC tissue as observed in the present study may explain previous reports on the lack of correlation between clinical response to EGFR protein expression on immunohistochemical analyses of patients with refractory metastatic CRC [38] , [39] . Since in the clinic, during the tissue processing, CRC EGFR may lose affinity upon handling and fixation, and antibodies used for diagnosis recognize epitopes different then EGF binding domain, we would like to propose that EGFNIR imaging of CRC fresh tissue slices may complement the above method [40] . This possibility needs to be further addressed using a larger cohort of human CRC tissues, in parallel with immunohistochemical analyses.
On one hand, EGF-NIR is significantly smaller then antibodies, it has higher tissue permeability, and its clearance as unbound agent can be achieved with simple washing steps, compared to antibodies labeled with NIR probes. Also its use can be more accurate in measuring EGFR expression than cetuximab in CRC tissue with variable affinity of EGFR for cetuximab [41] . On another hand, the risk in using such an agent is its ability to activate EGFR mediated downstream signaling such as RAS-Erk required for tumor proliferation. This problem can be solved by future designing of biomimetic peptides with reduced proliferative activity compared with native EGF or using EGFR targeting nanoparticles. Present findings provide EGF-NIR as an enabling platform technology for possible implementation of the binding protocol to visualize ex vivo EGFR in human CRC tissue, complementary to the gold standard technique of RT-PCR for EGFR mRNA quantification [6] , in situ hybridization [8] and EGFR immunohistochemistry [7] .
Fluorescence optical imaging has become a valuable tool as an adjunct to fiber-optic endoscopy due to its low cost and its ability to track multiple probes in a real time manner. Compared with the visible spectrum, the NIR dyes overcome the endogenous autofluorescence, maximize tissue penetration and are suitable for non-invasive whole body and organs imaging in small animals [17] . Over the last decade, NIR endoscopes were developed for high resolution imaging in small animals [42] and in identification of hyperplastic and adenomatous polyps in the colon [43] . The instruments currently available and/or under development, for clinical human imaging, such as Zeiss Pentero, with the microendoscope consisting of 20-gauge fiber optic catheter and dichroic beam splitters that simultaneously display visible light and 700 nm and 800 nm NIR fluorescent are used for intra-operative vascular flow and vascular surgery [44] . This instrument follow up the endogenous NIR autofluorescence of hemoglobin in the blood or of the FDA approved fluorescent ICG dye, either as a free molecule injected i.v. or conjugated to a relevant antibody such as anti-CEA and anti-mucin antibodies in local application [45] . Therefore, it is anticipated that present results, achieved with Odyssey Infrared Imager, will be reproduced with fiber-optic NIR endoscopes.
In conclusion, our study underscores the potential benefits of NIR optical imaging for BOI of EGFR in CRC tissues as evident from cell lines models in vitro , orthotopic tumors in mice at low dose (1 nmol/mouse) and in situ human tissues. The results suggest that BOI of EGFR using EGF-NIR probe may provide a sensitive, highly selective, non invasive tool for the detection and characterization of EGFR expressing CRC tissues without the need of radioactive imaging. This technology may complement immunohistochemical assessment of EGFR protein expression in CRC tissue which might contribute to future standardization methods measuring EGFR protein level, to improve the identification of patients who will benefit from anti-EGFR monoclonal antibody therapy.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Malaria remains a significant global public health burden that is responsible for an estimated 229 million infections worldwide and 409,000 deaths annually, with the vast majority of malaria cases and deaths occurring in Sub-Saharan Africa [ 1 ]. The knowledge of malaria epidemiology and implementation of control measures in any endemic setting are imperative for the reduction of transmission and eventual transition to elimination efforts [ 2 ]. Utilizing rapid diagnostic tests (RDTs), detection of Plasmodium antigens directly from human blood provides an effective measure of active malaria infection [ 3 ]. In 2019, 348 million RDTs were sold by global manufacturers with the most commonly used RDTs detecting the presence of Plasmodium falciparum histidine rich protein 2 (HRP2), though tests are available which also detect Plasmodium aldolase and lactate dehydrogenase (LDH) [ 1 ]. These RDTs are evaluated at a sensitivity of detection of 200 parasites/μL, although actual field results can be influenced by a number of test, operator, and parasite factors [ 4 ]. A positive HRP2-based RDT result could indicate an active P . falciparum infection (clinical or subpatent) or a recently-cleared P . falciparum infection with HRP2 antigen still in systemic circulation. Due to the length of time for post-treatment clearance of HRP2, HRP2-based RDTs can be positive for weeks after resolution of infection [ 5 , 6 ]. Clearance of aldolase and LDH is substantially quicker and their presence is more indicative of active infection [ 7 ]. A negative result for any type of RDT can indicate a true negative, or false negative due to a low-density Plasmodium infection or low production of the antigen target (or non-production of HRP2 due to a gene deletion, [ 8 ]). RDTs provide a qualitative, point-of-care measurement for specific Plasmodium antigens (presence or absence of the antigen), and quantitative measurement of these malaria antigens can occur in the laboratory setting through different immunoassay platforms [ 9 – 11 ].
Further molecular testing, such as real-time polymerase chain reaction can be used as a sensitive method to detect the presence/absence of active infections for all malaria species and inform estimates of malaria prevalence [ 12 ]. However, testing is more laborious and costly as multiple steps and assays are needed for confirmation of Plasmodium DNA. Initial screening of samples using antigen detection can serve as a more economical and high-throughput method to screen samples to predict parasite presence/absence status using PCR. Additionally, beyond the simple presence or absence of a Plasmodium antigen in a sample, using lab assays for detection of multiple antigens can provide quantitative estimates for each target, as well as generate an antigen profile (interpretation for +/- to multiple targets) for a specimen [ 9 ].
Recent advances in statistics and computing power have seen the increase of use of sophisticated machine learning approaches for classification in the context of complex datasets [ 13 , 14 ], including random forest machine learning approaches to predict protection to malaria based on antibody profiles [ 15 ]. In this study, we evaluated the use of machine learning approaches using continuous concentration of antigen data to predict PCR presence/absence classification. Specifically, we evaluated the use of conditional inference trees using antigen concentration and log concentration to predict the presence/absence of infection and classification of five parasite density levels using dried blood spot samples from high and low transmission areas in Angola and Haiti. Predictive models can provide greater nuance to epidemiological estimates and inform the selection of samples as a screening method for further downstream molecular testing.
Materials and methods
Samples and ethics statement
Dried blood spot (DBS) samples used in this study were previously collected for a therapeutic efficacy study (TES) in Angola in 2015 (n = 193) [ 16 ], an Angolan health facility (HF) survey in 2016 (n = 208) [ 17 ], and a bednet study in Haiti enrolling persons seeking care in health facilities from 2012–2013 (n = 425) [ 18 ]. The TES samples were from symptomatic children seeking care at health facilities with microscopy confirmed P . falciparum infection. The 2016 health facility survey samples were from a representative sample of febrile and afebrile outpatients of all ages in Angola. The Angolan TES activity was classified as non-research by human subjects research boards at CDC (#2014-233b) and the Angolan Ministry of Health. Blood sample collection during the Angolan health facility survey was approved by the Angolan Ministry of Health and further laboratory investigation approved by the Office of the Associate Director for Science in the Center for Global Health at the CDC (#2018–034). The Haiti bednet study enrolled febrile patients presenting to health facilities and capillary blood was collected blood for an RDT and also spotted onto filter paper. The Haiti study protocol was approved by the National Bioethics Committee of Haiti and the Institutional Review Board at the CDC. For all studies, written consent was obtained from all participants, and assent obtained from minors upon consent from minor parent or guardian.
Plasmodium falciparum detection and parasite density calculation for different studies
Molecular detection of P . falciparum infection and parasite density estimation were determined using real-time PCR and/or microscopy. Parasitemia for samples collected during the 2015 TES in Angola was determined using traditional microscopy [ 16 ]. Parasitemia for samples collected during the 2016 health facility survey in Angola was determined using sensitive quantitative PCR (sen-qPCR) using methods previously described [ 9 ], with an analytical sensitivity of 0.02 parasites/μL [ 19 ]. Parasitemia for samples collected in Haiti was determined using PET-PCR using methods previously described [ 18 ].
PCR assays and multiplex antigen detection
For samples with PCR results, total DNA was extracted from blood specimens by column purification with the Qiagen DNA easy kit according to manufacturer’s protocol (Qiagen, Valencia, CA), and purified DNA subjected to either PET-PCR or sen-qPCR as denoted above. To translate from real-time PCR signal to estimated parasite density, appropriate standard curves were prepared specific for each assay as described previously [ 20 ].
Concentrations of HRP2, pan- Plasmodium lactate dehydrogenase (pLDH), and pan- Plasmodium aldolase (pAldolase) were calculated for each sample using the multiplexed antigen bead-based assay and extrapolation from assay signal to antigen concentration performed using methods previously described [ 9 ]. For all laboratory data collected for analyses, it was assumed there was no sample contamination.
Data analysis and malaria infection status classification
Principal components analysis (PCA) was performed using antigen concentration and log concentration for PCR presence/absence and infection level parasitemia (parasites/μL) based on five categories (none- 0, lowest—> 0–20, low—> 20–200, mid- > 200–2,000, high- > 2,000) using the prcomp function in R (R Foundation for Statistical Computing). Categories were selected on a log10 scale with the 200 p/μL as the benchmark, being the minimum parasite density RDT product testing employs [ 4 ]. As the Angola (microscopy dataset) were nearly all high density infections, Infection level categories for that dataset were the following: lowest = > 0–5,000; low = > 5,000–10,000; mid = >10,000–15,000; high = > 15,000. Conditional inference trees were constructed using the ctree function in the party package in R. Conditional inference trees were selected as a non-parametric regression analysis, which uses unbiased recursive partitioning on continuous, multivariate data to identify the most informative features (e.g., antigenic concentration) and quantitative thresholds for prediction. The decision trees were trained using leave-one-out cross validation using antigen concentration and log concentration as features to classify PCR presence/absence and infection level status as described above. Accuracy, sensitivity, and specificity of conditional inference trees was calculated using the following: accuracy = (true positive (tp) + true negative (tn)) / (tp + false positive (fp) + tn + false negative (fn)); sensitivity = tp / (tp + fn); specificity = tn / (fp + tn). All figures were created using the ggplot2 and cowplot libraries, unless otherwise stated. All R code for data analysis and figure generation is available at https://github.com/SESchmedes/plasmodium_falciparum_infection_status .
Results
The P . falciparum prevalence of the study population was 28% from the 2016 Angola survey [ 17 ], and 4.0% from the Haiti study [ 18 ]. The mean age of participants for the 2015 Angola TES was 2.8 years with a median 3 years and range 7 months– 12 years. The mean age of participants for the 2016 Angola HF survey was 21 years with a median 15 years and range 1 month– 90 years. The mean age of participants for the 2012–2013 Haiti HF study was 17.4 years with a median 11 years and range 0–99 years.
HRP2, pLDH, and pAldolase concentrations were generated from a total of 826 dried blood spot samples collected in P . falciparum high-transmission (Angola) or low-transmission (Haiti) areas. Persons from the Angola TES had parasitemia determined by microscopy with a range of 2,175 to 184,464 parasites/μL (mean 49,230 parasites/μL); parasite densities for the Angolan health facility samples ranged from 0 to 43,290 parasites/μL (mean 618 parasites/μL); parasite densities for Haitian samples ranged from 0 to 18,463 parasites/μL (mean 908 parasites/μL) ( Fig 1 ).
10.1371/journal.pone.0275096.g001
Fig 1
Parasitemia distribution for each countries’ dataset.
For each study, parasite density depicted as parasites/μL blood. Middle bar = median. Upper box hinge = 75 th percentile. Lower box hinge = 25 th percentile. Upper whisker = largest value no further than 1.5 * IQR (inter-quartile range or distance from first and third quartiles) from the hinge. Lower whisker = smallest value at most 1.5 * IQR from the hinge.
Conditional inference trees were trained using leave-one-out cross validation with HRP2, pAldolase, and pLDH concentrations (and log-transformed concentrations) for classification of PCR presence/absence status ( Fig 2 ) and infection level ( Fig 3 , S1 Fig ). Both HRP2 and pAldolase informed the models for PCR presence or absence for the full Angolan and Haitian datasets, but pLDH concentrations did not. Using the non-transformed antigen concentrations, the Angolan sen-qPCR data from the health facility survey only predicted one node (bifurcation point) at a pAldolase concentration greater than 325.3 pg/mL to predict P . falciparum presence or absence ( Fig 2A ), and the Haitian PET-PCR provided two nodes with the first at a HRP2 concentration of 183 pg/mL and the second at pAldolase at 274 pg/mL ( Fig 2C ). When log-transforming the antigen data, additional prediction nodes were generated with the Angolan sen-qPCR predicting two nodes with the first of pAldolase at 325.1 pg/mL and second of HRP2 concentration at 595.7 pg/mL ( Fig 2B ). The log-transformed Haitian antigen data also provided two nodes for infection presence/absence, the first at HRP2 concentration of 182.8 pg/mL, and the second at a higher HRP2 concentration of 779.8 pg/mL ( Fig 2D ). PCR infection status was predicted with accuracies ranging from 73–96%, while infection level was predicted with accuracies ranging from 59–66% ( Table 1 ).
10.1371/journal.pone.0275096.g002
Fig 2
Conditional inference trees using HRP2, pLDH, and pAldolase antigen concentration for classification of Plasmodium falciparum presence or absence as determined by PCR assay.
A) Angola (sen-qPCR). B) Angola (sen-qPCR), log scale. C) Haiti (PET-PCR). D) Haiti (PET), log scale. Y-axes at base of trees indicate probability of correct classification on a scale of 0.0 to 1.0.
10.1371/journal.pone.0275096.g003
Fig 3
Conditional inference trees using HRP2, pLDH, and pAldolase antigen concentration and log concentration for malaria infection level classification.
Infection level categories: None = 0 parasites/μL; Very low = > 0–20; Low = > 20–200; Mid = > 200–2,000; High = > 2,000. A) Angola (sen-qPCR). B) Angola (sen-qPCR), log. C) Haiti (PET-PCR). D) Haiti (PET), log. Y-axes at base of trees indicate probability of correct classification on a scale of 0.0 to 1.0.
10.1371/journal.pone.0275096.t001
Table 1 Percent accuracies for malaria infection status prediction.
Country/Dataset
Attribute
Presence/Absence
Infection Level % accurate
% accurate (Se, Sp) *
Concentration
NA
70
Angola (microscopy)
Log concentration
NA
72
Angola (sen-qPCR)
Concentration
73 (94%, 67%)
66
Log concentration
75 (73%, 78%)
66
Haiti (PET-PCR)
Concentration
96 (97%, 95%)
59
Log concentration
96 (97%, 95%)
66
* Se: sensitivity; Sp: specificity; Accuracy, Se, and Sp are based on correct classification of malaria parasite presence/absence by utilizing PCR result as gold standard
P . falciparum infection data was further modeled by conditional inference trees after sub-dividing into five categories based on levels of estimated parasite densities. As the Angola TES only enrolled participants based on a microscopically-confirmed parasite density above 2,000 p/μL, those data were not able to be evaluated using the same categorization scheme as the Angola (sen-qPCR) and Haiti (PET-PCR) datasets; therefore, higher concentration infection levels were used for resolution of level of infection ( S1 Fig ). In assessing the non-transformed antigen data, the Angolan sen-qPCR generated four nodes for infection level, with the first three from pAldolase and the fourth node from HRP2 ( Fig 3A ). The Haitian dataset provided further resolution with a primary node at 311 pg/mL of pAldolase, and further downstream nodes based on HRP2 or pAldolase concentrations ( Fig 3C ). As was the case for infection presence/absence, log-transformed antigen data provided more nodes for level of infection. For Angolan sen-qPCR data, log-transformed antigen data provided five nodes with the first three based on pAldolase concentration and the final two on HRP2 concentration ( Fig 3B ). Notably, log-transforming the Haitian antigen data now provided the first node at HRP2 (concentration of 182.8 pg/mL) with downstream nodes involving both HRP2 and pAldolase ( Fig 3D ). Percent accuracy for predicting P . falciparum infection level ranged between 59 and 72%, all which were lower than the accuracies of predicting simple presence/absence ( Table 1 ).
For both presence/absence and infection level analyses with PCR data, pLDH did not provide significant decision nodes in the full dataset. However, if pAldolase data was removed, leaving only HRP2 and pLDH, then pLDH did provide nodes. An example is shown for the Angolan sen-qPCR dataset with pAldolase removed where pLDH provides the first nodes for both the non-transformed and transformed antigen concentrations ( S2 Fig ). The corresponding classification accuracy was approximately the same for both the non-transformed (75%) and log-transformed (75%) antigen data when compared to the full dataset with all three antigens included ( Table 1 ). For all three studies, the pLDH and pAldolase concentrations were shown to correspond with each other ( S3 Fig ).
When performing principal component analysis (PCA) on the datasets, scatterplots displaying the concordance between the first and second principal components (PC) showed a degree of clustering based on infection presence/absence as well as level of infection. For data from PCR assays, no discernable clustering was observed with non-transformed antigen data for infection presence/absence ( S4A and S4C Fig ), but when antigen data were log-transformed, the Haiti PET-PCR infection presence was strongly connected to lower values of PC1 (which explained 81.1% of variance) ( S4D Fig ). When assessing data by level of infection, the non-transformed antigen data again did not show defined visual clustering ( Fig 4A and 4C ). Higher PC1 values were strongly connected to higher parasite densities for the log-transformed antigen data from Angola ( Fig 4B ), but the inverse was true for the Haiti dataset with lower PC1 values connected with the ‘mid’ and ‘high’ infection levels. Infection level categories by microscopy did not show any visual clustering by scatterplots of PC1 and PC2 ( S5 Fig ).
10.1371/journal.pone.0275096.g004
Fig 4
Principal components analysis of HRP2, pLDH, and pAldolase concentrations and log concentrations for infection level using qPCR.
Infection level categories: None = 0 parasites/μL; Very low = > 0–20; Low = > 20–200; Mid = > 200–2,000; High = > 2,000. A) Angola (sen-qPCR). B) Angola (sen-qPCR), log scale. C) Haiti (PET-PCR). D) Haiti (PET-PCR), log scale.
Discussion
Our results suggest that machine learning algorithms can be trained using quantitative malaria antigen data to reliably predict P . falciparum presence/absence and as well as different levels of peripheral parasite densities. Antigen detection is utilized globally for diagnosis of malaria by RDTs, but these are designed to detect clinically-relevant parasite densities, and only provide a binary result [ 3 , 4 ]. Additionally, standard RDT use would also require a point-of-contact action (i.e. administering anti-malarial drugs) upon a positive result, and the multiplex antigen data and analyses presented here would be more utilized for epidemiological purposes. By being able to categorize sample sets into levels of estimated parasite densities based on multiplex antigen data alone, an additional benefit could arise by being able to select samples within these higher levels for greater success with DNA-based assays. The ability to use quantitative antigen concentrations to train machine learning algorithms to predict peripheral parasite densities represents a novel step forward for these efforts.
Collection of quantitative multiplex antigen data presents many advantages to other laboratory assays for the estimation of malaria status from a patient sample. These immunoassays are formatted to a 96-well format, the per-sample cost is approximately an order of magnitude less than nucleic-acid based assays, and hands-on time in the laboratory is short due to the simplicity of the antigen detection assays [ 9 , 10 ]. For the datasets available for this study, quantitative data for three Plasmodium antigens was available: pan- Plasmodium LDH and aldolase antigens (pLDH and pAldolase, respectively), and P . falciparum- specific HRP2. As all three datasets were specifically capturing P . falciparum infections by microscopy or PCR assays, this panel of three antigens was appropriate for this investigation. However, non-falciparum infections have been reported in both Angola [ 9 , 21 ] and Haiti [ 22 ], so the possibility also exists that non-detected mixed infections with P . vivax , P . malariae , or P . ovale (only in Angola) could have skewed the pan- Plasmodium antigen concentrations beyond what would be expected for a P . falciparum- only infection. However, as a proportion of the total malaria burden on these populations, non-falciparum infections are rare in these two countries, so presence of mixed infections would likely not have influenced the models. With quantitative data potentially available for other malaria antigen targets not utilized here, these similar machine learning approaches could be expanded to be even more robust in predicting P . falciparum infection or be modeled against infection with another of the human malarias.
In assessing the model input of the pan- Plasmodium antigens for the complete datasets, pLDH was only informative for the Angola TES study (infection detected by microscopy), whereas pAldolase was informative for nearly every other decision tree with both non-transformed and log-transformed antigen concentrations. It may not be surprising that one of these pan- antigens would “out-compete” the other as high collinearity was observed in absolute concentrations between these two targets in the same sample ( S3 Fig ) [ 9 ]. As an example, when the pAldolase data was removed from the Angolan health facility dataset, the pLDH replaced pAldolase as the first node on the decision tree. As both are metabolic enzymes in Plasmodia spp ., though concentrations of these two antigens would be expected to be largely concordant, they both may provide unique information for different strains of P . falciparum parasites which may express slightly different isoforms of these two antigens [ 23 , 24 ]. By far, the most informative input for the models creating the most nodes was the HRP2 antigen, which is abundantly expressed during blood-stage P . falciparum infection [ 25 ]. This was true in training models for both P . falciparum presence/absence as well as level of parasite density during infection. The added advantage of being able to measure signals for multiple antigens at a time is consistent with previous reports showing that HRP2/LDH ratios are predictive of determining active from recently cleared infection [ 26 ].
When compared to models for parasite presence/absence, modeling for discrete infection levels had noticeably lower accuracy. This was not surprising as blood-stage malaria infection is characteristic of billions of parasites and high amounts of antigen being produced, so an identified infection typically has very high amounts of these antigens in the host without precise gradations. Additionally, the “very low” (1–20 parasites/μL) and “low” (20–200 parasites/μL) categories of P . falciparum infection are both under the parasite density levels evaluated by the World Health Organization RDT evaluation program for product qualification [ 4 ]. The highest accuracy, up to 96%, was observed for prediction of PCR presence/absence in Haiti, and this could be explained by the low-transmission setting in this country; individuals would have been less likely to have had a recent infection with antigen concentrations creating “noise” that makes it more difficult to distinguish from active infection.
A limitation to this study was that datasets from each of the three studies provided different sample sizes in terms of number of persons infected with P . falciparum and utilized different enrollment criteria and samples from persons with different exposure histories. Additionally, the only sample type utilized in these surveys was DBS, and during drying and storage, potential degradation of antigen or DNA may have occurred to understate the quantity of these biomarkers. Different PCR assays were used for Haiti and Angola, and while both estimated parasite densities from quantity of DNA in the samples, comparison of classification trees between these two sample sets should consider the differences in PCR assays. A more recently detected phenomenon of P . falciparum strains with deletions or alterations of the pfhrp2 gene has been seen in numerous countries but was not evaluated in this study [ 8 ]. However, these deletions have not been reported in Haiti [ 27 ], and only reported at very low levels in Angola [ 9 ], so these potential deletions likely did not affect our analyses. High-transmission areas (like Angola) might not perform as well using this model compared to low-transmission areas (like Haiti) due to lingering HRP2 antigen in circulation [ 5 , 7 ], which could negatively impact specificity estimates.
Future studies on larger datasets should address optimal statistical tests and machine learning models for infection status prediction, as well as employ methods to correct for dataset imbalance. Conditional inference trees were selected for use in this study to perform a nonparametric regression analysis as a method for unbiased recursive partitioning to easily identify the most informative antigen for the model, in addition to predictive quantitative thresholds of antigenic concentrations. As such, it could not be stated that this approach utilized here would produce optimal accuracy, and further investigation of other statistical approaches, such as k-nearest neighbor regression, linear discrimination analysis, random forest, gradient boosting, or finite mixture models, should be conducted on future antigenic concentration datasets collected addressing the limitations stated above.
This study provides a pilot methodology and the results can be used to design and conduct additional studies. Specifically, future validation studies should have datasets with: substantial numbers of negative and positive samples with a wide range of parasite densities; molecular detection/parasite density measured using a variety of quantitative PCR techniques; and samples from different geographical areas (including pre-elimination, low-transmission and high-transmission countries/regions). Further investigation of machine learning approaches could provide greater resolution for determination of infection status from quantitative antigen data to support malaria surveillance activities and epidemiologic studies.
Supporting information
S1 Fig
Conditional inference trees using HRP2, pLDH, and pAldolase antigen concentration for malaria infection level classification by microscopy.
Infection level categories: Lowest = > 0–5,000; Low = > 5,000–10,000; Mid = >10,000–15,000; High = > 15,000. A) Angola (microscopy). B) Angola (microscopy), log scale. Y-axes at for all plots indicate probability of correct classification on a scale of 0.0 to 1.0.
(TIF)
S2 Fig
Conditional inference trees using only HRP2 and pLDH data with pAldolase removed for classification of Plasmodium falciparum presence or absence as determined by PCR assay.
Data shown for Angola (sen-qPCR) classification with antigen concentrations on non-transformed scale on left and log-transformed on right. Y-axes at base of trees indicate probability of correct classification on a scale of 0.0 to 1.0.
(TIF)
S3 Fig
Comparison of concentrations of pLDH and pAldolase antigens from persons enrolled among the three surveys.
Antigen concentration data shown for the Angola 2015 TES (A), Angola 2016 health facility (B), and Haiti bednet (C) studies with concentration of pAldolase on x-axis and pLDH on y-axis for each.
(TIF)
S4 Fig
Principal components analysis of HRP2, pLDH, and pAldolase concentrations and log concentrations for qPCR presence/absence.
A) Angola (sen-qPCR). B) Angola (sen-qPCR), log scale. C) Haiti (PET-PCR). D) Haiti (PET), log scale.
(TIF)
S5 Fig
Principal components analysis of HRP2, pLDH, and pAldolase concentrations for infection level using microscopy.
Infection level categories: Lowest = > 0–5,000; Low = > 5,000–10,000; Mid = >10,000–15,000; High = > 15,000. A) Angola (microscopy). B) Angola (microscopy), log scale.
(TIF)
S1 File
Inclusivity in global research.
(DOCX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The metabolic syndrome (MS) is a composition of cardiovascular risk factors–namely adiposity, high blood pressure, impaired glucose and fat metabolism–and represents a major burden to both the individual patient and society [ 1 ]. Organ-specific complications comprise type 2 diabetes, coronary artery disease (CAD), obstructive sleep apnea or fatty liver disease (FLD), among others [ 1 , 2 ]. Of these, cardiovascular disease (CVD) is the leading cause of death in Europe, therefore prevention and early treatment are crucial [ 3 ].
Another important manifestation of the MS is fatty liver disease (FLD), i.e. non-alcoholic fatty liver disease (NAFLD) [ 1 , 4 , 5 ] and accordingly metabolic associated fatty liver disease (MAFLD) [ 6 ]. The currently introduced MAFLD definition puts emphasis on metabolic disorders resulting in FLD, regardless of alcohol consumption [ 6 ].
Complications of FLD comprise development of steatohepatitis, which can progress to fibrosis and cirrhosis [ 7 ]. Moreover, patients with FLD have an increased risk to develop hepatocellular carcinoma, even in cases without advanced fibrosis or cirrhosis [ 8 – 10 ]. Therefore, patients with advanced FLD show a higher liver specific morbidity and mortality [ 11 ]. In addition, NAFLD patients also show a higher overall mortality due to extrahepatic reasons, e.g. CAD [ 11 , 12 ].
Several studies suggest an association between severity of FLD and presence of relevant CAD, which may be induced by similar risk factors [ 12 – 14 ]. A recent meta-analysis revealed a prevalence of 44.6% (95% confidence interval (CI), 36.0%–53.6%) of CAD among FLD-patients [ 15 ]. Interestingly, the prevalence of CAD increased to 60.5% when invasive angiography was used (95% CI, 43.8%-75.1%). There is strong evidence that NAFLD is an independent risk factor for cardiovascular events [ 16 ]. Furthermore, some studies suggest an association between the severity of NAFLD and CAD risk [ 12 , 15 – 17 ].
In clinical practice, prevention, early detection, and treatment of FLD and CAD are important but specific preventive measures are not always in awareness or available [ 18 – 20 ]. Moreover, risky alcohol consumption is frequent among patients with CAD and there is evidence for an increased cardiovascular risk among heavy drinkers [ 21 ]. Such patients are neglected by risk stratification algorithms of traditional NAFLD guidelines [ 20 ]. Therefore, the MAFLD definition better reflects real life scenarios than the traditional NAFLD definition and is thus increasingly applied [ 22 ].
A major problem for studying the relationship between FLD and CAD is the need for a reliable liver tissue characterization because grade of steatosis, inflammatory activity and grade of fibrosis provide important information on disease severity and risk of progression. The traditional gold standard liver biopsy is, however, associated with bleeding risk, especially in CAD patients due to antiplatelet or anticoagulation drugs and, therefore, restricted to carefully selected patients [ 23 ]. Recent developments of non-invasive ultrasound-based methods now allow a thorough FLD screening in CAD patients at a larger scale [ 24 ]. Elastography [ 25 ] combined with attenuation quantification [ 26 ] and risk score calculation [ 27 ] provide reliable and fast tools for estimation of fibrosis severity and steatosis grade as well as risk assessment of steatohepatitis.
In this pilot study, we non-invasively investigated the presence and especially the severity of FLD, patients at MAFLD risk, in a cohort with suspicion of advanced CAD.
Patients and methods
Study design and patient selection
This prospective, cross-sectional, single-center, diagnostic study was designed to evaluate the interplay between FLD and CAD risk.
The study protocol was approved by the ethical committee of the University Leipzig (registration number 148-15-09032015) and registered in the Clinical Trials Register of the U.S. National Library of Medicine (ClinicalTrials.gov, NCT02779946). All patients provided written informed consent.
Between October 2015 and February 2017 adult patients (≥18 years) with indication for coronary angiography (CA) were recruited at the Division of Cardiology of Leipzig University Medical Center. We considered patients either scheduled for routine angiography or undergoing emergency angiography due to (suspected) myocardial infarction (MI). In the latter case, patients were only considered after successful cardiological therapy with stabilized heart function prior to discharge from hospital.
Patients who have had known CA before were excluded from the analysis. Further exclusion criteria comprised age < 18 years, pregnancy or lactation, active malignant disease within 12 months before inclusion, patients with a liver transplant or altered liver anatomy, e.g. after resection of right liver lobe, cholestasis on ultrasound imaging, severe congestive heart failure (EF <30%, NYHA III or IV, diastolic dysfunction III or IV) and pulmonary hypertension (WHO III or IV) [ 28 , 29 ]. In addition, patients with history of chronic viral hepatitis were not considered.
Study examination
The study examinations were performed immediately prior to or within 21 days after the cardiological intervention but did not delay any cardiological diagnostics. Ultrasound and liver risk assessment were performed by a blinded examiner (JB and SB) to the results of CA to avoid possible investigator bias.
All patients underwent a thorough clinical characterization including case history, anthropometric data, lifestyle, alcohol and nicotine consumption and medication. Alcohol consumption was recorded as average alcohol intake per week (grams). From this, the daily alcohol consumption was derived. For sub-analysis, patients were stratified according to risky drinking behavior. Average alcohol intake of >20g/day in men and >10g/day in women was considered relevant according to the German NAFLD guideline [ 20 ]. For the laboratory data, a recent blood sample in an overnight fasting state with a maximum interval of 3 weeks was required.
Additionally, clinically established risk scores (NAFLD-fibrosis score (NFS), FIB-4-index (FIB4) as well as the newly developed FibroScan-AST score (FAST), see below ) were calculated. Prior to CA–or in case of emergency CA after clinical stabilization within 21 days–, a thorough abdominal ultrasound examination including liver stiffness measurement (LSM) with vibration controlled transient elastography (VCTE) combined with measurement of controlled attenuation parameter (CAP) were performed by an experienced examiner using Siemens Acuson S2000 ultrasound device (linear and curved array probe). A fasting period for at least four hours prior to examination was required.
MAFLD was defined by the presence of elevated CAP (see below) and at least one of either overweight/obesity (BMI >25kg/m 2 ) or diabetes mellitus type 2 or the presents of at least two metabolic risk abnormalities [ 30 ].
Vibration controlled transient elastography
VCTE (kPa) of the liver including quantification of steatosis with CAP (dB/m) was performed by a trained examiner as previously described [ 31 ]. In brief, ten valid measurements were performed in supine position in a right intercostal space according to the manufacturer’s recommendation [ 25 ]. LSM were performed using the M- and XL-probe of the Fibroscan® (FibroScan; Echosens, Paris, France) according to the skin-liver-capsule distance (SLCD) [ 32 ]. The SLCD was measured by a conventional linear ultrasound transducer. Patients with SLCD exceeding 25mm were examined with the XL-probe. Focal lesions affecting the measurement area of VCTE, hepatic cholestasis, presence of ascites as well as liver congestion due to right heart failure were ruled out by conventional ultrasound (see above) [ 25 ].
An interquartile range (IQR) of less than 30% in patients with LSM >7.1 kPa was required [ 33 ]. According to the German NAFLD guideline, LSM values of ≥8 kPa defined an increased risk for significant fibrosis [ 20 , 24 ]. CAP values ≥302 dB/m indicated presence of steatoses (any grade), a value ≥ 331 dB/m indicated advanced steatosis (S2/3) [ 34 ].
Laboratory values and clinical risk scores
We recorded full blood count, alanine and aspartate aminotransferases (ALT and AST), gamma-glutamyl-transferase (GGT), alkaline phosphatase (AP), albumin, creatinine, calculated glomerular filtration rate (GFR), lipid profile (cholesterol, high- and low-density lipoprotein (LDL and HDL) triglycerides) and glycohemoglobin (HbA1c). The upper limit of normal of AST at time of the study was 0.6 μkat/l in women and 0.85 μkat/l in men, and of ALT 0.58 μkat/l in women and 0.85 μkat/l in men, respectively.
Clinically well-established fibrosis risk scores (NFS; FIB4) were calculated [ 35 , 36 ]. For NFS, sensitive/specific cut-offs were −1.455 (age 36–65) and 0.12 (age ≥ 65)/0.676 (age ≥ 36). For FIB4, sensitive/specific cut-offs were 1.3 (age < 65) and 2.0 (age ≥ 65)/2.67 (all ages), respectively.
FAST is a recently developed elastography based tool for assessing the risk of active fibrotic (≥F2) steatohepatitis (NAS≥4) [ 27 ]. We calculated FAST according to the published formula using CAP, LSM and AST. We applied sensitive (0.35) and specific (0.67) cutoff values for the detection and exclusion of active fibrotic steatohepatitis [ 27 ].
Cardiological diagnostics and coronary angiography
Results of routine echocardiography were used to rule out right heart stress and severe insufficiency. CA was performed by an experienced interventionalist according to current recommendations [ 37 ] using the radial or femoral artery as vascular access. Severity of CAD was evaluated according to the grade of stenosis (wall irregularities, moderate stenosis of ≥ 50% lumen reduction and severe stenosis of ≥ 75% lumen reduction, respectively) [ 37 ].
The indication for revascularization was based on the angiography findings as well as the patient’s clinical presentation in accordance with the valid guideline at the time of intervention [ 38 , 39 ].
For the main data analysis, patients were divided into two groups based on the need for revascularization: CAD requiring revascularization (percutaneous transluminal coronary angioplasty (PTCA), stenting or cardiac bypass) and no intervention needed.
Statistics
Statistical analyses were performed using SPSS Version 27. Values are displayed as median and interquartile range (IQR) by X (X, Y) and absolute numbers and percentage by X (Y%). Chi-square test without continuity correction was used to evaluate association between two categorical variables. Mann-Whitney U-test was used to compare medians between two non-normally distributed groups. Logistic regression was used to identify independent risk factors for coronary intervention using the software R. For logistic regression, a p-value was based on the Wald statistic and the 95% CI determined with a profiling method.
A p-value of <0.05 was considered significant.
Results
A total of 216 patients referred for CA were screened. After exclusion of 96 patients due to various reasons, 120 patients were available for the final analysis ( Fig 1 ). Of these 120 patients, n = 19 (15.8%) underwent emergency CA due to (suspected) myocardial infarction. Indication for routine CA (n = 101) was stable angina pectoris.
10.1371/journal.pone.0286882.g001
Fig 1
Summary of patient recruitment.
NYHA New York Heart Association, EF Ejection fraction; LSM Liver stiffness measurement.
Baseline characteristics of the total study cohort and the MAFLD sub-cohort are shown in Table 1 , stratified for the presence or absence of MAFLD. Using only CAP for steatosis detection 49 patients (41%) were diagnosed with MAFLD as defined by the expert consensus statement [ 37 ]. A total of 50 patients (42%) of the total cohort required intervention including nine patients being referred for coronary bypass surgery. The remaining 70 patients had no indication for coronary intervention. In the MALFD subgroup the proportion of patients requiring intervention was comparable with the non MAFLD subgroup: 45% (22 patients) vs. 39% (28 patients; p = 0.55).
10.1371/journal.pone.0286882.t001
Table 1 Baseline characteristics of the total cohort stratified for the presence or absence of MAFLD.
Variables
Total cohort
non-MAFLD group
MAFLD group
p-value
n = 120
n = 71 (59%)
n = 49 (41%)
Age, years
65.2
(58.1;73.6)
64.0
(55.7;73.2)
66.4
(58.9;74.5)
0.47
Male gender
80
(66.7%)
42
(59.2%)
38
(77.6%)
0.036
BMI, kg/m 2
27.6
(25.0;30.4)
25.7
(24.0;28.1)
30.5
(28.4;34.6)
<0.001
BMI >25 kg/m 2
90
(75.0%)
43
(60.6%)
47
(95.9%)
<0.001
Relevant alcohol consumption
50
(41.7%)
30
(42.3%)
20
(40.8%)
0.88
High blood pressure
86
(71.7%)
49
(69.0%)
37
(75.5%)
0.44
Diabetes mellitus
29
(24.2%)
11
1(5.5%)
19
(38.8%
0.004
Waist-hip ratio
0.97
(0.91;1.02)
0.95
(0.89;1.00)
1.00
(0.96;1.03)
0.001
CAD
50
(41.7%)
28
(39.4%)
22
(44.9%)
0.55
hs-CRP, mg/l
2.79
(1.33;7.23)
2.11
(0.82;6.43)
3.38
(1.85;7.27)
0.18
Trigylcerides, mmol/l
1.34
(0.98;1.78)
1.11
(0.91;1.54)
1.70
(1.29;2.84)
<0.001
HDL cholesterol, mmol/l
1.38
(1.11;1.71)
1.48
(1.20;1.83)
1.24
(1.02;1.39)
<0.001
LDL cholesterol, mmol/l
3.30
(3.68;4.30)
3.30
(2.65;4.06)
3.36
(2.84;4.49)
0.21
AST, μkat/l
0.45
(0.39;0.57)
0.46
(0.39;0.58)
0.45
(0.40;0.57)
0.76
AST % of ULN
60.0
(50.6;76.7)
63.3
(50.6;86:0)
58.8
(50.6;68.3)
0.24
AST > ULN (n)
12
(10%)
9
(12.7%)
3
(6.1%)
0.24
ALT, μkat/l
0.41
(0.33;0.57)
0.40
(0.31;0.57)
0.44
(0.35;0.62)
0.21
ALT % of ULN
57.3
(45.6;75.9)
56.9
(43.5;75.0)
58.8
(48.2;77.6)
0.83
ALT > ULN (n)
11
(9,2%)
6
(8,5%)
6
(12,2%)
0.50
GGT, μkat/l
0.50
(0.35;0.73)
0.49
(0.32;0.71)
0.57
(0.41;0.73)
0.19
AP, μkat/l
1.14
(0.97;1.37)
1.14
(0.99;1.34)
1.14
(0.95;1.41)
0.90
INR
1.0
(1.0;1.1)
1.0
(1.0;1.1)
1.0
(0.9;1.1)
0.97
PTT, sec
29.5
(28.0;31.5)
29.1
(27.8;30.6)
29.8
(28.6;32.6)
0.044
Albumin, g/l
45.1
(43.4;46.7)
45.1
(43.2;47.2)
44.9
(43.6;46.5)
0.62
Platelets, 10 9 /l
231
(201;284)
232
(200;289)
229
(205;277)
0.91
Bilirubin, μmol/l
8.5
(6.5;11.5)
8.4
(6.4;11.5)
8.6
(6.9;11.9)
0.96
HbA1c, %
5.56
(5.36;5.98)
5.54
(5.31;5.74)
5.61
(5.46;6.39)
0.011
Values given in median (IQR) and absolute numbers (%)
MAFLD, metabolic associated fatty liver disease; BMI, Body Mass index; CAD, coronary artery disease; hs-CRP, high-sensitive C-reactive protein; HDL cholesterol, high-density lipoprotein cholesterol; AST, aspartate aminotransferase; ALT, alanine transaminase; ULN, upper limit of normal; gGT, gamma-glutamyltransferase; AP, alcalic phosphatase; INR, international normalized ratio of prothrombin time; PTT, partial thromboplastin time; HbA1c, glycated haemoglobin
Assessment of fatty liver disease by elastography
LSM including CAP could not be applied in four patients due to cysts of the right liver lobe (n = 1) or insufficient fasting status (n = 3). Cases with invalid (less than 10 single measurements; n = 1) or unreliable (n = 7) LSM had high BMI (median of 34 kg/m 2 vs. 28 kg/m 2 ; p = 0.002) or live-capsule distance (median of 27 mm vs. 21 mm; p = 0.002). Thus, LSM was valid and reliable in 120 (94% of eligible) patients ( Fig 1 ).
Table 2 shows LSM and CAP results stratified by the results of coronary diagnostics. Median LSM values were within the normal range but differed significantly between the sub-groups (intervention vs. no intervention: 4.9 kPa vs 4.2 kPa; p = 0.018). Suspicion of significant fibrosis in the overall cohort was found rarely in only 6 cases (5%), also equally distributed between the groups. Results in the MAFLD subgroup were similar ( S1 Table ).
10.1371/journal.pone.0286882.t002
Table 2 Severity of liver disease.
Variables
Total cohort
Coronary Intervention
Yes
No
n = 120
n = 50
n = 70
p-value
Steatosis risk
CAP (dB/m)
289
(245;327)
290
(263;340)
289
(244;318)
0.43
S0
CAP ≤ 302 dB/m
70
(58.3%)
28
(56%)
42
(60%)
0.66
S1
302 < CAP < 331 dB/m
22
(18.3%)
8
(16%)
14
(20%)
0.58
S2/3
CAP ≥ 331 dB/m
28
(23.3%)
14
(28%)
14
(20%)
0.31
Fibrosis risk
Liver stiffness (kPa)
4.5
(3.6;5.6)
4.9
(4.2;5.7)
4.2
(3.5;5.5)
0.018
Low risk
< 8 kPa
114
(95%)
47
(94%)
67
(95.7%)
0.67
Increased risk
≥ 8 kPa
6
(5%)
3
(6%)
3
(4.3%)
NFS a (n = 117)
-1.4
(-2.13;-0.21)
-1.15
(-1.81;-0.20)
-1.62
(-2.41;-0.49)
0.15
≥ Sens. Cut-off
35
(29.9%)
15
(31.3%)
20
(29.0%)
0.79
≥ Spec. Cut-off
9
(7.7%)
5
(10.4%)
4
(5.8%)
0.36
FIB4 b (n = 118)
1.5
(1.12;1.94)
1.64
(1.26;2.16)
1.47
(1.03;1.84)
0.064
≥ Sens. Cut-off
53
(44.9%)
25
(51.0%)
28
(40.6%)
0.26
≥ Spec. Cut-off
10
(8.5%)
7
(14.3%)
3
(4.3%)
0.056
Indicators of steatohepatitis
AST % of ULN
60
(50.6;76.7)
60
(50.6;89.6)
60
(51.7;69.9)
0.35
AST > ULN (n)
12
(10%)
9
(18%)
3
(4.3%)
0.014
ALT % of ULN
57.3
(45.6;75.9)
64.8
(46.2;82.1)
55.3
(45.7;70.7)
0.20
ALT > ULN (n)
11
(9.2%)
10
(20%)
1
(1.4%)
<0.001
FAST c
0.15
(0.10;0.25)
0.22
(0.13;0.34)
0.12
(0.09;0.21)
<0.001
≥ Sens. Cut-off
16
(13.3%)
12
(24%)
4
(5.7%)
0.004
≥ Spec. Cut-off
3
(2.5%)
3
(6%)
0
(0%)
Values given in median (IQR) and absolute numbers (%)
CAP, Controlled Attenuation Parameter; NFS, NAFLD-Fibrosis Score; FIB4, FIB4-index; AST, aspartate aminotransferase; ALT, alanine aminotransferase; ULN, upper limit of normal; FAST, Fibrosis-AST-score
a sensitive/specific cut-offs were −1.455 (age 36–65) and 0.12 (age ≥ 65)/0.676 (age ≥ 36)
b sensitive/specific cut-offs were 1.3 (age < 65) and 2.0 (age ≥ 65)/2.67 (all ages)
c sensitive/specific cut-offs were 0.35/0.67
Steatosis detected by CAP could be seen in 50 patients (41.7%) of the total cohort with 28 patients (23.3%) with signs of moderate or severe steatosis (≥S2). CAP values did not differ significantly between the groups (see Table 2 and S1 Table ).
Laboratory based risk indices
Calculation of the NFS revealed a relatively high proportion of patients above the sensitive cut-off ( Table 2 ). No difference between the groups was seen (31%, n = 15 in the interventional group vs. 29%, n = 20; p = 0.79). Only 5 patients (10%) of the interventional group and 4 patients (6%) of the non-intervention group (p = 0.36) were identified as high-risk according to the specific cut-off of NFS. Similar numbers were seen in the MALFD group ( S1 Table ).
FIB4 did neither show a significant difference between the groups using the sensitive (51%, n = 25 vs. 41%, n = 28; p = 0.26) nor the specific cut-off (14%, n = 7 vs. 4%, n = 3, p = 0.056).
Unexpectedly, there was a poor agreement in risk assessment between LSM, FIB4 and NFS: none of the high-risk FIB4 patients and only one of the high-risk NFS patients had increased LSM values. The small number of patients with significant elevated LSM ≥8kPa had significantly higher BMI, waist-to-hip ratio and CAP resulting in an increased prevalence of MAFLD and elevated FAST score ( S2 Table ).
Risk of fibrotic steatohepatitis
The non-invasive evaluation of fibrotic steatohepatitis using the FAST score revealed significantly higher (median 0.22 vs. 0.12; p < 0.001) values in patients with indication for coronary intervention ( Table 2 and S1 Table ). This difference persisted after exclusion of n = 11 patients with myocardial-infarction-to-LSM-interval below one week (n = 109; median 0.22 vs. 0.12; p = 0.012). In addition, we performed a multivariate logistic regression analysis of the whole patient sample taking known anthropometric and behavioral parameters including alcohol consumption into account ( Table 3 ). In the MAFLD subgroup, the values were similar although the limited patient number did not allow a multivariate logistic regression.
10.1371/journal.pone.0286882.t003
Table 3 Variables associated with coronary intervention using multivariate logistic regression analysis.
Estimate (OR)
95% CI
p-value
Sex (female vs male)
0.25
0.09 to 0.66
0.007
Age (per 10 years)
1.66
1.05 to 2.76
0.038
BMI (per kg/m 2 )
0.90
0.81 to 1.00
0.054
Smoking (yes vs no)
1.25
0.49 to 3.29
0.64
Risky alcohol consumption (yes vs no) a
0.80
0.32 to 1.96
0.63
FAST score (per log odds ratio)
2.28
1.40 to 3.96
0.001
OR, Odds ratio; CI, confidence interval; BMI, Body-Mass index; FAST, Fibrosis-AST score
a >20g/day in men and >10g/day in women
Discussion
FLD and CAD are both associated with the metabolic syndrome, but their pathophysiological interplay is not fully clarified [ 40 ]. Prevalence of FLD is increasing, especially since the MAFLD incorporated many patients with coexisting other liver disease [ 41 ]. FLD is associated with increased liver related morbidity and mortality in patients with advanced liver disease, whereas cardiovascular events are a leading cause of mortality in FLD patients without relevant fibrosis [ 42 ]. Therefore, patients with FLD should be carefully assessed for co-existing CAD [ 18 , 20 ]. However, there is less evidence available if patients at high risk for CAD require intensified FLD screening.
Our cohort represents the typical spectrum of patients at risk of coronary artery disease with an expectable rate of coronary intervention [ 43 , 44 ]. 41% of patients fulfilled the MAFLD definition, which is slightly higher compared to the estimated NAFLD prevalence in the general adult population in Germany [ 20 ]. However, we only observed a low frequency (5%) of cases at risk of advanced liver disease defined by elevated VCTE values or serum-based fibrosis markers in the total cohort, which converts to a prevalence of advanced liver disease of approximately 15% in the MALFD sub-group. This is in line with a recent observational study from Germany that revealed an overall prevalence of relevant fibrosis of 19% among NAFLD patients at secondary referral institutions [ 45 ]. Interestingly, the traditional risk factors diabetes, obesity and arterial hypertension were neither strongly associated with advanced liver disease nor with the need for cardiovascular revascularization therapy. This might result from the relatively small study population and may reflect the small effect size in preselected cases. Only male gender had a significant association with CAD risk which is in line with reported data and due to the practice of patient selection at the time of recruitment, in which women show a higher rate of false positive ergometry and myocardial perfusion scintigraphy [ 44 , 46 ].
Our findings demonstrate, that FLD prevalence in CAD patients shows no relevant deviation from age-adjusted values from the normal population. Hence, a close interplay between the drivers of CAD and FLD progression remains questionably, and thus, intensified screening measures for advanced FLD in patients undergoing CA beyond the actual recommendations [ 18 , 20 ] seems not indicated for routine medical care. In addition, the poor correlation between different non-invasive markers of advanced fibrosis in our cohort underlines that confounding factors, e.g., medication or intervention interfering with platelet counts, must be taken into account when assessing FLD risk. However, patients at risk for advanced liver disease (e.g. with metabolic syndrome) should undergo screening as recommended by the current guideline independent of the presence CAD [ 20 , 47 ].
As a secondary finding, we observed associations of higher liver stiffness within the limits of normal in patients with relevant CAD compared to those without need of revascularization therapy. Even after exclusion of patients with LSM ≥8 kPa, the difference remained significant (4.8 kPa (IQR 4.1;5.5) vs. 4.1 (IQR 3.3;5.2) kPa; p = 0.018). Elevated liver stiffness at lower levels does not necessarily reflect tissue fibrosis but is also an indicator of tissue vascularization and inflammation [ 25 , 48 ]. These observations were incorporated in the development of the FAST score, which reflects the risk of fibrotic steatohepatitis with good accuracy [ 27 ] and has meanwhile gained attention as potential guidance in the referral pathways of patients with FLD [ 31 , 49 ]. Higher FAST score values were the strongest predictor of relevant CAD in our cohort. This phenomenon has not been reported before and points to inflammatory-driven links between CAD and FLD progression. Low grade systemic inflammation has been described in many metabolic diseases including Type-2-Diabetes-mellitus, obesity and NAFLD [ 40 ] and is also relevant in cardiovascular pathologies [ 50 ]. Therefore, future studies should focus on the potential role of FAST score as a stratification method not only in the field of liver diseases but also in settings where cardiac risk is the dominant question. Ideally, this could be accompanied by analysis of liver tissue, visceral fat and endothelial function to get further insights in the mutual pathophysiological mechanisms of these disease.
Our study has several limitations. The monocentric design restricted the case numbers and may have led to selection bias according to local healthcare pathways. However, our cohort is comparable to other studies comparing CAD risk and liver disease [ 51 , 52 ]. In this pilot study, we chose a dichotomous endpoint (presence of CAD requiring intervention). Further correlation with the severity of CAD would have required pressure analysis for estimation of coronary flow reserve and microvascular resistance index [ 19 ], which was not always available in our patients. Moreover, we could only use non-invasive indicators of FLD severity instead of liver histology. This reference standard is, however, associated with relevant bleeding risk in cohorts with the need of anticoagulation therapy, which is very common among cardiovascular patients. Although there is no general recommended cut-off for CAP, we used the cut-offs proposed by Eddowes et al. [ 34 ], one of the largest prospective biopsy-controlled studies providing data in CAP accuracy for steatosis quantification. Established non-invasive markers are therefore an accepted substitute for liver biopsy [ 20 , 48 ] and may help to avoid selection bias.
Conclusion
In conclusion, our data show that the prevalence of advanced FLD is low in CAD patients requiring invasive procedures. Noninvasive estimates of fibrosis and steatosis severity were not associated with the need for coronary intervention. Elevated FAST score values underline the pathophysiological importance of inflammatory activity in both FLD and CAD.
Supporting information
S1 Table
Severity of liver disease on the subgroup of patients with MAFLD.
(DOCX)
S2 Table
Characterization of patients at high and low risk of relevant liver fibrosis.
(DOCX)
S1 Dataset
(XLSX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The factor for inversion stimulation (FIS) is a global regulator of gene expression and chromosome compaction in Gamma-proteobacteria. In Escherichia coli and Salmonella enterica , transcription of the fis gene is very high during rapid growth, consequently FIS is one of the most abundant DNA-binding proteins during exponential growth (50,000 - 100,000 monomers/cell) [ 1 - 3 ]. Transcription of fis declines dramatically as cells enter stationary phase, and the FIS protein concentration drops to undetectable levels [ 1 - 3 ]. Control of fis expression is best understood in E. coli , where the global regulatory proteins CRP, IHF, and FIS modulate only the degree of induction, but none of them is absolutely required for fis expression in laboratory culture [ 4 - 6 ]. In their absence, fis continues to be induced by nutritional upshift and repressed during stationary phase [ 4 - 6 ]. A major determinant of growth phase-dependent fis expression is the GC-rich sequence between the −10 RNA polymerase binding site and the transcription start site, which creates a protein-independent barrier to transcription [ 7 , 8 ]. When DNA becomes highly supercoiled during rapid growth, the topological stress exerted on the DNA double strand facilitates melting of the GC-rich discriminator by RNA polymerase. Thus, DNA supercoiling directly controls fis expression by removing a repressive barrier [ 9 ]. The discriminator is so-called because it allows RNA polymerase to discriminate between promoters that are subject to the stringent response (like the fis promoter) and those that are not [ 9 ]. Stringently-regulated promoters respond negatively to the alarmone guanosine tetraphosphate (ppGpp) and to the protein DksA, both of which act via RNA polymerase [ 10 , 11 ].
FIS is an important transcription factor in natural environments where bacteria are starved for nutrients. For example, S. enterica serovar Typhimurium ( S . Typhimurium) requires FIS for pathogenicity gene expression during the initial stages of tissue invasion and later inside the macrophage vacuole [ 12 , 13 ]. Transcriptome analysis has revealed that S. enterica serovar Typhi fis is induced inside macrophage vacuoles despite nutrient-poor conditions and slow growth [ 14 ]. The expression of fis in intracellular environments may be explained in part by the recent finding that S . Typhimurium fis expression is elevated in stationary phase if oxygen availability is reduced, a condition referred to as “sustained expression” because the fis promoter remains active in stationary phase [ 13 ]. Sustained fis expression results in highly elevated FIS protein levels, which contribute directly to increased invasion of epithelial cells in in vitro assays [ 13 ]. It is superficially similar to the elevated expression seen with genetically altered derivatives of the E. coli fis promoter where the initiation nucleotide +1C is changed to either A or G: in these mutants fis transcription is elevated in stationary phase [ 15 ]. The activity of the native fis promoter shadows the fluctuation in the CTP pool of the cell. Our observations of sustained fis expression involve the native promoter and the goal of the study presented here was to identify the regulatory mechanisms that link oxygen availability to fis expression.
The redox sensors ArcAB and FNR regulate transitions from aerobic to anaerobic growth, making these proteins prime candidates for regulating the sustained expression of fis in low oxygen environments. In the ArcAB two-component system, ArcA is a site-specific DNA binding protein that is activated by phosphorylation when ArcB senses a drop in redox potential [ 16 ]. FNR is a site-specific DNA binding protein that becomes active as a transcriptional regulator when its Fe-S cluster is reduced in the absence of oxygen [ 17 ]. ArcA and FNR regulate dozens and hundreds of genes, respectively, many of which are required for metabolism in anaerobic conditions, but neither transcription factor has been implicated in controlling fis expression.
In bacteria, DNA supercoiling changes in response to environmental conditions such as nutrient and oxygen availability [ 10 , 18 ]. Because DNA shape influences gene promoter activity, DNA supercoiling is a mechanism for orchestrating global cellular transcription in response to changing conditions [ 19 ]. During rapid growth, E. coli maintains high levels of DNA supercoiling and this stimulates transcription of fis along with other growth phase-dependent promoters, such as rRNA promoters [ 9 , 19 , 20 ]. Several studies have found that a minimal promoter region containing only the σ 70 (RpoD) binding site is sufficient for the induction of the fis promoter during rapid growth [ 6 , 8 ]; this suggests that DNA supercoiling alone can activate the fis promoter in the absence of protein transcription factors, at least in laboratory conditions.
Genetic analysis revealed that neither the stringent-response-associated alarmone ppGpp nor the DksA protein were essential for the sustained expression of fis in stationary phase [ 21 ]. We examined the relative contributions made by other protein transcription factors and DNA supercoiling to the expression of S . Typhimurium fis in the non-aerated growth conditions that up-regulate the fis promoter. This led to the discovery of a second transcription start site in the S . Typhimurium fis promoter, which becomes more active as transcription from the primary transcription start site decreases in the absence of oxygen. Together the results suggest that the full-length fis promoter region activates fis expression in response to DNA supercoiling in low-aerated conditions. This topological switch may work in concert with multiple transcription start sites and the alternate sigma factor RpoS to integrate environmental and physiological signals at the S . Typhimurium fis promoter.
Materials and Methods
Bacterial strains and reporter gene constructs
Salmonella enterica serovar Typhimurium strain SL1344 and mutant derivatives were used for all experiments. Detailed descriptions of strains used in this study are provided in Table S1 . Deletion mutants were constructed by the method of Datsenko and Wanner [ 22 ] using the recombineering primers listed in Table S2 . Mutations were transduced into a fresh SL1344 background by bacteriophage P22 generalized transduction [ 23 ] and were confirmed by PCR and DNA sequencing. The pCP20 plasmid, which encodes the FLP recombinase, was used to remove the kanamycin resistance cassette from the dusB-fis-gfp TCD :: kan R construct in the S . Typhimurium chromosome as described in [ 22 ]. After curing cells of pCP20, loss of antibiotic resistance was confirmed by screening colonies for kanamycin sensitivity using LB-plates containing kanamycin at 50 μg/ml. Construction of SL1344 Δ crp was confirmed phenotypically by sugar fermentations tests using peptone broth containing a sugar or sugar alcohol at 0.5% final concentration and bromothymol blue as an indicator. As expected SL1344 Δ crp was unable to ferment mannitol, sorbitol or maltose but retained its ability to ferment glucose.
Promoter truncates were generated by inverse PCR using the primers listed in Table S2 and pZepfisS as template [ 13 ]. PCR amplicons were phosphorylated by T4 polynucleotide kinase and then circularised by ligation with T4 DNA ligase, followed by transformation into chemically-competent E. coli DH5α. PCR primers targeting the fis promoter region included a restriction enzyme site to facilitate confirmation of correct ligation products.
Culture growth
Well-aerated conditions were achieved by using 10 ml of LB (1% tryptone, 0.5% yeast extract, 0.5% NaCl) in a 250 ml glass conical flask, whereas non-aerated conditions used the same culture volume in a glass tube with an interior diameter 14 mm, as in [ 13 ]. All containers were loosely capped to allow air flow, and shaken at 200 RPM at 37 °C. Cultures were started by diluting cells 1/100 from well-aerated overnight cultures and grown for 22 hours to ensure that they had reached steady-state stationary phase before sampling. Anaerobic shock was administered by transferring 15 ml of well-aerated exponentially-growing cell culture into a 15 ml centrifuge tube, capping tightly, and incubating without agitation for 30 min at 37 °C. When required, antibiotics were used at the following final concentrations: carbenicillin 100 µg/ml, chloramphenicol 20 µg/ml, and kanamycin 50 µg/ml.
Flow cytometry and quantitative PCR
GFP levels in cells expressing fis-gfp TCD and P fis - gfp + were quantified using flow cytometry as follows: 2-30 µl of culture was fixed in 700 µl of freshly prepared phosphate buffered saline containing 2% formaldehyde. Fixed samples were stored overnight in the dark at 4 °C. The median fluorescence of 20,000 cells/sample was measured on a Dako CyAn ADP flow cytometer (PMT voltage 800-875 V). To quantify transcript abundance, RNA was isolated using TRIzol as described earlier [ 24 ]. Quantitative PCR was conducted as in [ 25 ] using primers listed in Table S2 .
DNA supercoiling analysis
Plasmids were isolated from cultures using the HiYield Plasmid Mini Kit (RBC Bioscience). All electrophoresis was conducted in 27-cm-long 1% agarose gels with 2x Tris Borate EDTA (TBE) as gel and running buffer. Approximately 1 µg of plasmid DNA (8–15 µl) was loaded on a gel using 4 µl of loading buffer (80% glycerol, 0.5 mg/ml bromophenol blue). The gel and buffer contained 2.5 µg/ml chloroquine, and electrophoresis was performed at 3 V/cm for 16 hours. To remove the chloroquine after electrophoresis, gels were washed by gentle rocking in large volumes of tap water for at least 3 hours; the wash water was replaced every 20–30 minutes. After washing, gels were stained by gentle rocking in water containing ethidium bromide (1 µg/ml) for at least 1 hour, then washed briefly in water, and plasmid topoisomers were visualized with UV light. Analyses of topoisomer distributions were conducted as in [ 26 ].
WebSIDD [ 27 ] was used to predict how DNA supercoiling affects stability of the S . Typhimurium and E. coli fis promoter regions. WebSIDD uses a default 5 kbp window that slides by 500 bp. Thus the stability of each base pair, G(x), is the average of 10 calculations, where the influence of proximal bases is weighted according to how close the base pair is to the center of the sliding window.
5′ RACE and western blot analysis
5′ RACE was conducted as in [ 28 ] using the primer Pfis.expTSS.RT.Rev to reverse transcribe dusB-fis transcripts. PCR products were cloned into the linearized vector pJET (Fermentas) and transformed into E. coli strain XL-1 and at least 5 clones were DNA sequenced. Western blot analysis was conducted as in [ 25 ] using the E. coli σ S (RpoS) monoclonal antibody (Neoclone) diluted 1:5,000.
Results
Comparison of S . Typhimurium and E. coli fis promoter regions
The regulation of fis expression has been studied primarily in E. coli . As a first step to identify the regulatory elements that control S . Typhimurium fis expression, the S . Typhimurium fis region was aligned with the same region in E. coli . The bicistronic dusB - fis operon encodes FIS in both species. Because dusB has attracted very little research attention, the upstream promoter region (P fis ) bears the fis name. The dusB and fis open reading frames have 92% and 98% nucleotide identity, respectively, between E. coli and S . Typhimurium, indicating that very little change has occurred in these genes since divergence of the Escherichia and Salmonella genera. As has been noted previously, the region containing the primary transcription start site is identical between E. coli and S . Typhimurium [ 20 ], but there is significant sequence divergence upstream of position −49 ( Figure 1 ). In E. coli there is an intergenic space of 296 bp between the dusB-fis transcription start site and the stop codon of the upstream gene, prmA . In S . Typhimurium, the prmA stop codon is 628 bp upstream, suggesting that >300 bp of non-coding DNA was inserted upstream of the ancestral promoter region. Alignment of DNA sequence from other Enterobacteriaceae species confirms that the ancestral promoter region was likely around 300 bp (not shown), making Salmonella unusual in the Enterobacteriaceae family.
10.1371/journal.pone.0084382.g001 Figure 1
Gapless sequence alignment of the E. coli and S . Typhimurium fis promoter regions. CRP and FIS (grey boxes) and IHF (underline) binding sites identified in E. coli are shown [ 5 , 6 , 40 ]. Promoter elements, transcription start sites and dusB translation initiation codons are boxed. The angled arrows show the locations of P fis -1 (both species) and P fis -2 ( S . Typhimurium). The locations of promoter truncations in S . Typhimurium are indicated by black triangles.
To test whether the full-length intergenic region contributes to regulation, the gfp TCD reporter gene was fused to the S . Typhimurium dusB-fis operon in the native chromosomal location. Because FIS is autoregulatory, we avoided creating a fis null mutant by inserting gfp TCD downstream of the fis open reading frame, but upstream of a predicted transcriptional terminator. The kanamycin resistance marker linked to gfp TCD was subsequently excised to prevent the kan promoter from influencing fis expression. This chromosome-based reporter differs from the study by Ó Cróinín and Dorman that first identified sustained fis expression [ 13 ]. Ó Cróinín and Dorman used a plasmid-based reporter gene fusion containing the fis promoter region up to -298 bp, thus omitting the 330 bp of upstream non-coding sequence.
In stationary phase cultures the dusB-fis-gfp TCD construct demonstrated elevated expression in non-aerated conditions compared to aerated conditions ( Figure 2A ). This modest two-fold increase in expression corresponds to a dramatic rise in FIS protein level, which causes S . Typhimurium to become more invasive in a fis -dependent manner [ 13 ]. Overall, the dusB-fis-gfp TCD chromosomal fusion exhibited the classic expression pattern of strong induction in exponential growth in both well-aerated and non-aerated conditions ( Figure 2C ), followed by a rapid decrease in expression as growth slowed. Two-fold higher fluorescence was sustained from 3 to 22 hours of growth in non-aerated compared to aerated conditions ( Figure 2A & C ). The two-fold difference we observed with the dusB-fis-gfp TCD construct was less than the increase seen previously in the plasmid-based reporter system [ 13 ], and quantitative PCR confirmed that our chromosomal construct is an accurate reporter of native fis gene expression ( Figure S1 ). The reduced intensity of sustained expression in the chromosome gfp fusion compared to the plasmid-borne reporter indicates that the plasmid system is a reliable but exaggerated reporter of sustained fis expression in non-aerated conditions.
10.1371/journal.pone.0084382.g002 Figure 2
Effects of transcription factor mutations on fis expression in aerated and non-aerated conditions. A) Column graph comparing dusB-fis :: gfp TCD expression at 22 hours. All data are expressed relative to wild type at 22 hours in well-aerated conditions (dashed line). B ) Average and standard deviation of four replicate culture densities after 22 hours. C-F) Time courses of dusB-fis::gfp TCD expression in S . Typhimurium wild type and Δ fnr (C), Δ arcA (D), Δ rpoS (E), and Δ arcA /Δ rpoS (F) mutants. The blue arrow in D indicates the increased sustained expression observed in the Δ arcA mutant after 22 hours in non-aerated conditions. A , C - F ) Mean and standard deviation of GFP fluorescence from three or four biological replicates are plotted in arbitrary units. All data are expressed relative to wild type levels at time point 0 in well-aerated conditions. The same wild type data are presented in panels C-F. G ) Growth dynamics of strains in aerated and non-aerated conditions. Smoothed curves were generated by GraphPad Prism 5.0d from the average of four or more replicate growth curves.
ArcA is a repressor of S . Typhimurium fis
The redox sensors ArcAB and FNR are transcriptional activators in low oxygen conditions, making them prime candidates for activating sustained fis expression. Deletion of fnr had no detectable effect on fis expression or on cell growth ( Figure 2B, C & G ). Surprisingly, deletion of arcA caused an increase in fis expression in non-aerated stationary phase ( Figure 2A & D ), implicating arcA as a repressor of sustained fis expression. We found that although the Δ arcA mutant had a slightly prolonged lag phase in fresh medium, its growth rate and final cell densities were very similar to wild type cells in both aerated and non-aerated conditions ( Figure 2B & G ). Thus, the effect of the Δ arcA mutation on fis expression is not due to altered growth in non-aerated conditions.
In E. coli ArcA is a known repressor of rpoS transcription [ 29 ], and RpoS has been proposed to repress fis expression in S . Typhimurium [ 13 ], thus an S . Typhimurium Δ arcA mutant would be predicted to have reduced fis expression due to up-regulation of rpoS . This led us to test whether ArcA may act through rpoS in S . Typhimurium, perhaps by up-regulating rpoS expression as opposed to repressing it as in E. coli . First though we were surprised to discover that the chromosomal dusB-fis-gfp TCD fusion was not up-regulated in the Δ rpoS mutant ( Figure 2E ), which contrasts with the previously observed strong activating effect of the Δ rpoS mutation on the plasmid-based P fis :: gfp fusion [ 13 ]. Quantitative PCR confirmed that the dusB-fis-gfp TCD fusion is a reliable reporter of fis expression in non-aerated conditions ( Figure S1 ). The Δ rpoS mutant showed an extended lag phase, but unlike the Δ arcA mutant it did not achieve wild type densities in aerated culture after 22 hours; instead it grew slightly better than wild type in non-aerated conditions ( Figure 2B & G ). Neither of these minor growth phenotypes had a detectable effect on fis expression.
To further address whether ArcA may act indirectly on fis through RpoS function, we constructed a Δ arcA /Δ rpoS double mutant. In aerated and non-aerated growth conditions, the Δ arcA /Δ rpoS double mutant showed consistently reduced fis expression during exponential growth ( Figure 2F ), unlike either single mutant. Finding an exacerbating effect of combining the two mutations suggests that ArcA acts separately from RpoS to influence fis expression in exponential growth conditions. During stationary phase in non-aerated conditions, the Δ rpoS mutation nullified the effect of the Δ arcA mutation of enhancing sustained fis expression ( Figure 2A & F ), implicating RpoS as a mediator of ArcA’s effect on fis expression.
The minimal fis promoter is deficient in sustained expression Because neither FNR nor ArcA was required for sustained fis expression in non-aerated conditions, we created a series of fis promoter (P fis ) truncates to identify promoter regions required for sustained expression. Three truncates were constructed in a plasmid-based gfp reporter system. The amplification of fis expression profiles by the plasmid system provided two experimental advantages: first, the ability to create truncated promoters, and second, improved sensitivity to detect subtle changes in gene expression. Unfortunately, we were unable to clone the full intergenic region from S . Typhimurium, suggesting that an unidentified element upstream of position −300 is toxic to cells when present in multicopy. The construct P fis (−298) contained a promoter region identical in length to the E. coli intergenic region ( Figure 3A ), which is the same S . Typhimurium promoter region used previously to study sustained expression [ 13 ]. The medium length construct P fis (−198) removed sequence that in E. coli contributes to regulation ( Figure 1 ). The shortest construct, P fis (−49), preserved only the region of identity between S . Typhimurium and E. coli .
10.1371/journal.pone.0084382.g003 Figure 3
Defining functional regions and transcription factor input at the fis promoter. A) Schematic of the E. coli and S . Typhimurium intergenic regions. The locations of truncations in S . Typhimurium are indicated. B) Time courses of expression of plasmid-borne P fis :: gfp promoter truncates in aerated and non-aerated S . Typhimurium cultures. C and D) Expression of promoter truncates in S . Typhimurium transcription factor mutant backgrounds at 22 hours. In B-D, mean and standard deviation of GFP fluorescence from three or more biological replicates are plotted in arbitrary units that are reported relative to the chromosomal dusB-fis::gfp TCD fluorescence output presented in Figure 2 .
All three constructs demonstrated the classic peak in expression during exponential growth. However, P fis (−198) and P fis (−49) showed reduced overall expression levels at all stages of growth in both aerated and non-aerated conditions ( Figure 3B ), indicating that the full 300-bp upstream region contributes to activation. Furthermore, the minimal RpoD-driven promoter, P fis (−49), was sufficient for strong induction during exponential growth in S . Typhimurium, as reported in E. coli [ 6 , 8 ]. In non-aerated conditions, sustained fis expression was lower in P fis (−198) and was even further reduced in P fis (−49) ( Figure 3B ), indicating that one or more regulatory elements upstream contribute to sustained expression.
Transcription factor control of S . Typhimurium fis expression
To determine which transcription factors contribute to sustained fis expression and to identify regions of the fis promoter required for transcription factor function, P fis truncates were tested in several mutant backgrounds. In E. coli , FIS represses its own promoter by binding the FIS-I and FIS-II binding sites that flank the transcription start site [ 4 , 6 ], and these sites are perfectly conserved between the E. coli and S . Typhimurium fis promoters ( Figure 1 ). Up-regulation of the S . Typhimurium fis promoter in the Δ fis mutant confirmed that FIS represses its own promoter, and this repression occurs in both aerated and non-aerated conditions ( Figure 3C & D ). The FIS-I and FIS-II sites are intact in the shortest construct, P fis (−49), allowing FIS to repress each of the three promoter truncates.
All three lengths of P fis showed elevated expression in the Δ arcA mutant in non-aerated conditions ( Figure 3D ). Elevated expression regardless of promoter length supports a model in which ArcA has an indirect influence on fis during stationary phase, consistent with the absence of a predicted ArcA site in the S . Typhimurium fis promoter. fis expression was slightly elevated in aerated conditions ( Figure 3C ), which was not detected using the chromosomal system — again indicating that the plasmid-borne system enhances or exaggerates fis promoter activity.
The Δ rpoS mutation did not have a detectable effect on the P fis (−298) and P fis (−198) truncates, as observed above with the dusB-fis-gfp TCD chromosomal fusion ( Figure 3C & D ). However, P fis (−49) revealed that RpoS can exert a mild repressive effect in the absence of upstream fis promoter DNA, both in aerated and non-aerated conditions.
CRP is a global-acting transcription factor that directly regulates FIS expression in E. coli [ 4 ]. We found that deletion of crp caused a slight reduction in fis expression from all three promoter truncates in both aerated and non-aerated conditions ( Figure 3C & D ). The absence of a DNA site matching the CRP binding site consensus ( Figure 1 ) suggest that CRP plays an indirect regulatory role at the fis promoter in S . Typhimurium.
The S . Typhimurium fis regulatory region has a second, anaerobically-induced promoter
The reduction in fis expression caused by progressive truncation of the fis promoter could be explained by removal of one or more upstream transcription start sites (TSS). Thus, we used recently published RNA-seq data [ 24 ] to search for evidence of transcripts originating upstream of the primary TSS, which revealed a putative TSS at position −216 relative to the primary TSS ( Figure 1 ). 5′-RACE confirmed the existence of the second TSS ( Figure 4A ), which we named P fis -2 to distinguish it from the gene proximal TSS, called P fis -1. P fis -2 was not included in Kröger et al. [ 24 ] because it did not pass the conservative threshold used in that study.
10.1371/journal.pone.0084382.g004 Figure 4
Characterisation of S . Typhimurium Pfis-2. A) RNA-seq and 5′-RACE identification of the P fis -2 transcription start site. The number of sequencing reads in RNA-seq analysis of an early stationary phase S . Typhimurium culture is aligned with DNA sequencing results from 5′-RACE analysis of the same RNA sample (RNA-seq data available in [ 24 ]). B) Quantitative PCR measurement of transcripts originating from P fis -1 and P fis -2 during exponential growth. C) Quantitative PCR measurement of rpoS expression. D) Quantitative PCR measurement of transcripts originating from P fis -1 and P fis -2 during anaerobic shock. E) Quantitative PCR measurement of rpoS expression. In B-E , the mean and standard deviation from 3 to 5 biological replicates are plotted.
To determine the activity of P fis -2 in laboratory culture, quantitative PCR was used to distinguish between transcripts originating from P fis -1 and P fis -2. During exponential growth, less than one percent of dusB-fis transcripts originated at P fis -2 ( Figure 4B ). Transcription from P fis -2 was up-regulated two-fold in non-aerated stationary phase cultures, but this accounted for only 7±1% of total fis transcripts, indicating that P fis -2 is not responsible for sustained fis expression. Transcription was moderately reduced at both P fis -1 and P fis -2 in the Δ arcA mutant ( Figure 4B ), suggesting that ArcA causes mild activation of fis expression during exponential growth. Deletion of rpoS revealed RpoS to be a weak repressor of P fis -1 and a stronger repressor of P fis -2 during exponential growth ( Figure 4B ).
To further test whether ArcA acts indirectly through RpoS by repressing rpoS transcription, we measured rpoS transcript abundance. No significant difference was detected in rpoS transcript levels between wild type and Δ arcA mutant cells in exponential growth ( Figure 4C ). RpoS is controlled primarily by post-transcriptional mechanisms; unfortunately attempts to quantify RpoS protein levels in the wild type and Δ arcA mutant in exponential growth were unsuccessful because RpoS protein levels were below the detection limit of western blot assays.
Two-fold up-regulation of P fis -2 in non-aerated growth conditions led us to test the effects of more severe anaerobic shock, a condition that, like non-aerated growth, may imitate conditions encountered by S . Typhimurium in the intestine and in intracellular environments. Exponentially growing cells were transferred to sealed tubes to induce anaerobic shock, and this caused a significant up-regulation of P fis -2 concomitant with 18-fold repression of P fis -1 ( Figure 4D ). In these conditions, deletion of arcA caused a decrease in P fis -1 and P fis -2 activity, consistent with the decrease in exponential phase fis expression seen in Figure 2D . To once again test whether the ArcA effect might arise through RpoS activity, rpoS expression was measured in anaerobic shock. ArcA was found to exert a repressive effect on rpoS as revealed by higher rpoS transcript levels in the Δ arcA mutant ( Figure 4E ). Although higher levels of RpoS protein in the Δ arcA mutant could explain the repression of P fis -2 observed in Figure 4C , the Δ rpoS mutant showed wild type levels of P fis -2 expression during anaerobic shock ( Figure 4D ). The absence of an explicit RpoS effect in aerated, non-aerated, and anaerobic conditions ( Figures 2E , 4C & D ), suggests a model in which RpoS represses fis expression only when RpoS protein levels are unusually high, as in a Δ arcA mutant.
DNA supercoiling control of S . Typhimurium fis expression
DNA supercoiling is a key driver of fis expression [ 9 ], raising the intriguing question of whether DNA supercoiling and the fis promoter respond to small incremental changes in oxygen availability. Conversely, there may be specific oxygen concentrations at which regulatory control undergoes a transition that facilitates sustained fis expression in non-aerated conditions. To assess the relationship between oxygen availability, DNA supercoiling, and fis promoter activity, aeration was adjusted by growing cells in tubes with increasing culture volumes. DNA supercoiling and fis expression were measured in parallel cultures containing either pUC18 or the P fis (-298) reporter plasmid, respectively. The small, high copy number plasmid pUC18 (2,686 bp) was ideal for quantifying supercoiling levels, whereas the larger (6,424 bp) and low copy number reporter plasmid pZec-Pfis was less reliable for resolution of topoisomers; nevertheless, we did find that pZec-Pfis had the same topological responses as pUC, as was previously confirmed for pZec [ 26 ]. In the present study, wild type cells carrying pUC18 were assayed in parallel with strains having either the plasmid-borne P fis (-298) reporter or the chromosomal dusB-fis-gfp TCD reporter.
A continuous increase in DNA supercoiling was observed as aeration decreased due to increased culture volume ( Figure 5A ). In these same conditions P fis expression demonstrated a gradual increase with decreased aeration. Increased expression was observed with both the plasmid-borne and chromosomal reporters, but as expected was more pronounced in the plasmid system. P fis expression was the same in 6 ml and the 10 ml “non-aerated” condition used in the experiments above (compare Figures 3D and 5A ); thus cells appear to be severely oxygen limited in volumes above 5 ml in standard culture tubes. These results suggest that changes in DNA supercoiling during non-aerated growth may be involved in sustained fis expression.
10.1371/journal.pone.0084382.g005 Figure 5
DNA supercoiling control of fis expression. A) Median and interquartile ranges of DNA supercoiling from four biological replicates, plotted as in Figure 4E . Fluorescence data from P fis (-298) (green circles) and dusB-fis::gfp TCD (blue diamonds) for each biological replicate is plotted on the right y-axis; units as in Figure 2 . For both the DNA supercoiling and gene expression measurements, wild type cells were grown to stationary phase cells in the indicated volume of culture medium. The dashed lines are nonlinear curves fit to the expression data, with goodness-of-fit R 2 >0.85 in both cases. The degree of DNA supercoiling (σ) was determined by measuring the migration of topoisomers relative to fully relaxed DNA in chloroquine gels; each topoisomer represents a change of 1 in the linking number. B) DNA supercoiling states in S . Typhimurium wildtype and Δ arcA mutant cells at 22 hours in non-aerated conditions. Medians (black bar) and interquartile ranges of pUC18 topoisomer distributions in stationary phase in well-aerated cultures. For each strain, the average interquartile range from four biological replicates is plotted. C) SIDD profile of the fis promoter region. The energy required for DNA strand separation at a base pair, G(x), is a function of adjacent and distant DNA sequence [ 30 ], and G(x) values below 10 indicate positions prone to SIDD. G(x) values for linear DNA were calculated by WebSIDD [ 27 ] using 3,500 bp of chromosomal DNA sequence on either side of the dusB start codon (7,000 bp total); only the 550 bp region containing P fis -1 and P fis -2 is shown.
The correlation between DNA supercoiling and sustained fis expression in non-aerated cultures prompted us to test whether increased fis expression in the ∆ arcA mutant might also correlate with an increase in DNA supercoiling in this genetic background. DNA supercoiling was not elevated in the Δ arcA mutant ( Figure 5B ).
DNA supercoiling reduces the amount of energy required for transcription initiation by exerting torsional stress on the DNA double helix, which weakens base pairing and facilitates melting; this is referred to as stress-induced duplex destabilization (SIDD) [ 30 ]. Because DNA supercoiling stimulates fis expression we predicted that the fis promoter region would have a SIDD profile indicative of significant stress-induced denaturation, which may be focused near P fis -1 and P fis -2. Figure 5C plots the predicted stability of the fis promoter at the supercoiling level observed both in exponential growth and in non-aerated stationary phase (superhelix density around -0.05). This analysis suggests that the S . Typhimurium fis promoter region is particularly prone to destabilization and melting when DNA is highly supercoiled. The region immediately upstream of P fis -1 (position −10 to −160) is predicted to be extremely destabilized, with a second smaller region of destabilization surrounding P fis -2 (position −190 to −260). This very strong SIDD profile raises the possibility that DNA supercoiling alone is able to activate transcription. Further, the observed SIDD profile helps explain why removal of the most destabilized region makes the P fis (−49) truncate deficient in sustained expression.
Discussion
FIS is a global regulator of gene expression, but its abundance in the cell fluctuates dramatically depending on growth phase. During periods of rapid growth in laboratory conditions the fis promoter is highly expressed, which accounts for the high levels of FIS protein during exponential growth. Activation of the fis promoter during rapid growth relies on highly supercoiled DNA and RpoD [ 9 ]. It is possible then that the fis promoter is inactive in stationary phase because DNA is in a more relaxed state and because effective concentrations of RpoD are reduced due to competition with RpoS [ 31 ]. A reduction in oxygen availability during stationary phase causes DNA to become highly supercoiled in both S . Typhimurium and E. coli [ 26 , 32 , 33 ]. Thus a simple model predicts that fis will be expressed in oxygen-limited conditions, and our results suggest that DNA supercoiling may be an important contributor to sustained fis expression. Further, our findings fit very well with the recent observation that E. coli fis expression is much more dependent on global cellular physiology than on the direct activity of transcription factors [ 34 ]; we suspect that the regulatory influence of “cellular physiology” posited by the authors is due largely to changes in DNA supercoiling, which they did not test.
The observed reduction in fis expression with progressive truncation of the promoter region suggests that the entire promoter region functions as a topological switch, and that the switch loses potency when shortened. DNA supercoiling appears to sit atop the regulatory hierarchy because promoter truncation had the same effect of reducing transcriptional output in each of the fis , arcA , rpoS , and crp regulatory mutant strains. A topological switch mechanism is further supported by the prediction of a long destabilised region upstream of the GC-rich discriminator. During rapid growth, the GC-rich discriminator is readily melted and transcription proceeds at a high rate even with a very short promoter region, as in the P fis (−49) truncate. However, our data suggest that during the lower energy state of stationary phase, discriminator melting requires a long destabilised region to focus stress-induced melting at the fis promoter. The fis promoter is particularly striking for its large dynamic range, from very strong to silent. This makes it an ideal model for understanding how DNA supercoiling can be a dominant force in transcriptional control.
How ArcA and RpoS influence fis expression remains enigmatic. Both proteins are global regulators, and the pleiotropic effects of mutating global regulators can make it difficult to distinguish direct from indirect mechanisms of gene regulation. The simplest explanation is that ArcA indirectly influences fis expression through its activity as a repressor of rpoS expression. The expression of fis and rpoS is negatively correlated, the former being elevated in exponential growth and the latter in stationary phase, therefore it is unsurprising that FIS and RpoS are antagonistic at some gene promoters. For example, both proteins influence DNA supercoiling; FIS represses gyrB [ 35 ], whereas RpoS transcribes gyrB [ 36 ]. The interplay between these two regulators is particularly intriguing in conditions that promote fis expression during stationary phase, when RpoS is most active.
RpoS functions as a non-traditional repressor by (1) competing with RpoD for access to core RNA polymerase and (2) by competing with RpoD for DNA-binding sites because RpoD and RpoS bind very similar DNA sequences. RpoD depends on higher levels of DNA supercoiling than RpoS to initiate transcription [ 37 ]. For this reason, promoters can be differentially regulated by RpoD and RpoS through a mechanism in which relaxation of DNA supercoiling causes a transition from RpoD binding to RpoS binding. The dps promoter may present a useful model for understanding fis regulation by RpoS. Dps is a nucleoid-associated protein with an expression profile that is the opposite of FIS; it is absent during exponential phase growth but becomes highly abundant in stationary phase [ 3 ]. During exponential growth, RpoD and FIS bind together and remain locked at the dps promoter, thus preventing RpoS from accessing the −35 and −10 elements [ 38 ]. When FIS levels decrease as growth slows, RpoS gains access to the dps promoter and transcription is up-regulated. It may be that during stationary phase RpoS is able to gain access to and initiate transcription at the otherwise RpoD-driven fis promoter. Thus, up-regulation of rpoS in the Δ arcA mutant causes an elevation in sustained fis expression because of elevated RpoS activity at the fis promoter.
In E. coli , at least five transcription start sites have been detected in the fis promoter region [ 5 ], with the highly conserved start site (P fis -1 in Figure 1 ) being chiefly responsible for fis expression in laboratory conditions [ 39 ]. We have identified a novel transcription start site, P fis -2, in S . Typhimurium, bringing to two the number of characterised start sites in the S . Typhimurium dusB-fis operon. P fis -2 is up-regulated in anaerobic shock and even overtakes P fis -1 as the primary source of transcript in the absence of ArcA or RpoS. Anaerobic shock is particularly relevant to the study of S . Typhimurium pathogenesis, thus we are currently investigating whether P fis -2 plays a significant role in infection systems.
Supporting Information
Figure S1
Quantitative PCR measurement of fis transcript in mutants. Total (P fis -1 plus P fis -2) fis transcript abundance, expressed relative to wild type at 22 hours in well-aerated conditions. (TIFF)
Table S1
Bacterial strains and plasmids used in this study.
(DOCX)
Table S2
Oligonucleotide primers used in this study.
(DOCX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Over the last decade, online social network sites (SNSs) have grown dramatically. For instance, in April 2018, Facebook (FB), one of the most well-known SNSs, had 2.20 billion active monthly users ( https://zephoria.com/top-15-valuable-facebook-statistics/ ). The success of SNSs can be viewed as being closely tied to the inherent human need to belong to communities and to affiliate with peers to engage in a role within the community and obtain the social acceptance and gratification that comes with its fulfillment. Their success also relates to the need for self-presentation, which is part of the process of impression management [ 1 ].
FB use (i.e., time spent on FB) seems to be associated with offline social capital, as suggested by some studies. For instance, a cross-sectional study highlighted a positive link between FB use and offline connections, i.e. relationships in real life [ 2 ]. Similarly, a longitudinal study showed that regular FB users tend to benefit from a rich offline social network [ 3 ]. Perceived social isolation seems, however, to increase with high involvement in social media use (including FB), as found in a recent study on representative US individuals aged 19–32 years [ 4 ]. Yet, because this was a cross-sectional study, causality cannot be established. On the one hand, social isolation can lead to increased involvement in SNS as an emotional coping strategy or as a strategy to search for social support; on the other hand, overuse of SNS per se can induce social isolation (e.g., a sense of exclusion from online social groups, self-negative comparison with the idealized life of online peers, decreased available time for offline social relationships). Furthermore, a Dutch study conducted on 10- to 19-year-old adolescents [ 5 ] showed that receiving positive feedback on SNS profiles increased self-esteem and well-being, whereas receiving negative feedback on SNS profiles had the opposite effect. Another study [ 6 ] highlighted that self-management of SNSs positively influenced self-esteem. More precisely, these authors found that participants who were instructed to update and view their own profile during an experiment reported a higher level of self-esteem.
Despite ongoing controversies regarding the status of excessive SNS use as a potential addictive disorder [ 7 ], a growing number of research articles have reported that SNS users may suffer from excessive and potentially impairing use [ 8 ]. For instance, overinvolvement in SNSs has been linked to addiction-like symptoms [ 9 ] and a variety of negative psychological correlates or consequences [ 10 – 13 ]. Some studies have also tried to disentangle the psychological factors involved in problematic use of SNS, with, on the one hand, a focus on user preferences and motivations and, on the other, a focus on the normal and pathological personality traits that could act as risk factors. In the current paper, we decided to use the label “problematic SNS use” to define an involvement linked to negative (personal, social, professional/academic) consequences, in order to avoid a priori considering it as a genuine addictive disorder and allows for potential alternative conceptualizations [ 14 ]. For example, some studies showed that excessive SNS use can be viewed as a maladaptive coping [ 15 ], and assimilating it to a disorder might result in (over)pathologization [ 16 ].
SNSs, and FB in particular, can be used in different ways and can fulfill various motives [ 3 ], including, but not limited to, entertainment, social sharing, information seeking, relationship maintenance (e.g., interacting with an existing offline social network), and emotional coping (e.g., facing boredom, loneliness, or negative affect; [ 17 ]. In particular, FB usage preferences (and underlying motives) have been related to its healthy versus dysfunctional use [ 8 , 18 – 20 ]. Directed communication, for instance (e.g., one-on-one exchanges), was found to have more impact on social capital than does broadcasting or passive consumption of social news [ 3 ]. Moreover, Baek, Bae, & Jang [ 21 ] emphasized the opposite effects of social (i.e., bidirectional, such as chatting or messaging) versus para-social (i.e., unidirectional, such as checking other’s status or profile) online activities: Social activities were negatively correlated with a perceived feeling of loneliness, whereas para-social activities were positively correlated with it. In another study, well-being was found to increase when time spent on FB was used to maintain relationships, but to decrease when it was used to create new relationships [ 18 ].
In recent years, a growing number of studies have explored the normal and pathological personality traits likely to promote problematic use of SNSs, in particular FB [for a review, see 8 ]. The initial studies were conducted by using the five-factor model (FFM) of personality [ 22 ] and on the whole showed that problematic SNS use is associated with higher neuroticism and extraversion and with lower conscientiousness[ 9 , 23 – 24 ]. Symptoms of SNS addiction were also found to be positively related to lower levels of self-esteem [ 25 , 26 ] and insecure attachment [ 27 ]. Problematic and addictive-like use of SNSs has also been linked to pathological personality traits, namely, proneness to narcissistic personality [ 28 ], obsessive-compulsive symptoms [ 10 ], and borderline personality traits [ 29 ].
Strikingly, existing data on the role of impulsive personality traits in problematic use of SNSs remain scarce, although a few reports have documented heightened impulsivity in individuals displaying such use [ 30 ]. This is particularly surprising, given the tremendous number of studies documenting the associations between impulsivity and a wide range of addictive behaviors [ 31 ], including general or specific types of Internet-related problematic behaviors such as problematic video gaming [ 32 – 34 ]. Despite recent models of Internet-related disorders underscoring the pivotal role of impulsivity traits in the disorder [ 35 ]), the available data regarding problematic use of SNSs are to date limited.
An important gap in the literature is the exploration of the links between impulsivity and problematic SNS use built upon a theoretically driven model of impulsivity that assumes its multidimensional nature. In this regard, the UPPS-P (Urgency-Premeditation-Perseverance-Sensation seeking-Positive urgency) model of impulsivity [ 36 , 37 ] measures five impulsivity facets: (1) negative urgency, defined as the tendency to act rashly while faced with intense negative emotional contexts; (2) premeditation, defined as the tendency to take into account the consequences of an act before engaging in that act; (3) perseverance, defined as the ability to remain focused on a task that may be boring and/or difficult; (4) sensation seeking, considered as the tendency to enjoy and pursue activities that are exciting and openness to trying new experiences; and (5) positive urgency, defined as the tendency to act rashly while faced with intense positive emotional contexts. In the last decade, this model has proven to be a relevant theoretical framework to elucidate the associations between specific impulsivity traits and various forms of psychiatric disorders and problematic behaviors and thus became a dominant model in the field of psychopathology [ 38 ] and neuropsychology [ 39 ]. However, to date, the UPPS model has not been used to explore the links between specific impulsivity facets and problematic use of SNSs.
Accordingly, the aims of the current study were twofold. The first objective was to explore the heterogeneity of FB usage preferences in a community sample and to determine which type of FB activity puts people more at risk for developing problematic usage. The second objective was to test whether specific impulsivity facets predict problematic use of SNSs by capitalizing on their multidimensional nature. Our study is of an exploratory nature and thus formal operational hypotheses were not formulated. Yet, it can be expected that problematic FB use will be associated to elevated levels of positive and negative urgency, as previous research identified ICT-mediated problematic behaviors (e.g., dysfunctional use of online sexual activities, sexting) to be predicted by these specific impulsivity components [ 40 , 41 ]. If the current study reproduces these findings, it will suggest that FB problematic use constitutes an unregulated form of behavior displayed to regulate emotional states in the short term, despite the potential delayed negative consequences.
Methods
Participants and procedure
The study consisted of a survey that was accessible online and circulated via FB and other social networks and research networks, as well as via email by using snowballing techniques among the researchers’ contacts. The survey was disseminated by using an online platform (Qualtrics, Provo, UT). All items used in the online survey can be obtained from https://osf.io/xt4zv/ . The survey items were administered in fixed order (FB actual-use items, FB problematic use items, a self-esteem item, and impulsivity items; see below). The survey was available between October 1 and November 30, 2015. Anonymity of the participants was guaranteed (no personal data or Internet Protocol [IP] address was collected). Inclusion criteria were being French-speaking, aged 18 years or older, and being a FB user. The sample included 857 participants, of whom 676 fully completed the questionnaire. The mean age of the completers was 35.8 years ( SD = 12.0, range: 18.3–80.3 years), and 466 (68.9%) were women. The median age of women (30.8 years) was lower than that of men (38.8 years; W = 60770, p < 0.001). The ethical committee of the Psychological Sciences Research Institute of the Catholic University of Louvain, Louvain-la-Neuve, approved the study.
Measurements
The study included several questions generated to measure actual FB use, including frequency of use (separately for weekday and weekend), types of activities performed on FB (see Table 1 ), preferred activities performed on FB (participants ranked the activities from preferred to less preferred; see Table 2 ), device(s) used to access FB (personal computer, computer at work, smartphone, tablet), preferred device used to access FB, use of Messenger (do participants use FB alone, Messenger alone, or both applications when using FB via smartphones or tablets; this variable was coded as Messenger use: yes/no), and usage of notifications (yes/no).The frequency of use was not taken into account for further analyses, as the reliability of self-reported time spent in SNS. The different activities measured resulted from a consensus of the research team. All activities identified by the research team were considered for the study. The frequency item was finally not used in our analyses, as recent evidences demonstrated that self-reported frequency of FB is not a reliable variable [ 42 , 43 ].
10.1371/journal.pone.0201971.t001
Table 1 FB usage preferences and gender differences.
Whole sample ( n = 676)
Men ( n = 210)
Women ( n = 466)
χ 2
p -Value
Reading the news feed n (%)
389 (57.5%)
137 (65.2)
252 (54.1)
6.9
>0.01
Viewing friends’ pictures n (%)
363(53.7)
94 (44.8)
269 (57.7)
9.3
0.1
Commenting n (%)
353 (52.2)
114 (54.3)
239 (51.3)
0.4
0.5
Reading friends’ timelines n (%)
346 (51.2)
98 (46.7)
248 (53.2)
2.3
0.1
Contributing to a group n (%)
235 (34.8)
70 (33.3)
165 (35.4)
0.2
0.7
Updating status n (%)
229 (33.9)
95 (45.2)
134 (28.8)
16.8
>0.001
Gaming n (%)
66 (9.8)
12 (5.7)
54 (11.6)
5.0
0.025
Sharing stuff from the Internet n (%)
44 (6.5)
14 (6.7)
30 (6.4)
0.0
1
Using Messenger app n (%)
294 (43.5%)
90 (42.9%)
204 (43.8%)
0.02
0.89
FB main access
24.08
>0.001
Personal computer n (%)
241 (35.7%)
99 (47.1%)
142 (30.5%)
Professional computer n (%)
46 (6.8%)
19 (9.0%)
27 (5.8%)
Smartphone n (%)
351 (51.9%)
81 (38.6%)
270 (57.9%)
Tablet n (%)
38 (5.6%)
11 (5.2%)
27 (5.8%)
Notifications n (%)
302 (44.7%)
75 (35.7%)
227 (48.7%)
9.38
>0.01
10.1371/journal.pone.0201971.t002
Table 2 Favorite FB activities and related gender differences.
Whole sample 1
Men
Women
Wilcoxon
p -Value
Reading the news feed ( n = 389) mean ( SD )
1.9 (1.2)
1.8 (1.1)
1.9 (1.2)
17432
0.9
Reading friends’ timelines ( n = 346) mean ( SD )
2.2 (1.3)
2.4 (1.3)
2.2 (1.3)
13256
0.2
Updating status ( n = 229) mean ( SD )
2.6 (1.4)
2.2 (1.2)
2.8 (1.4)
4694
0.0005
Commenting ( n = 353) mean ( SD )
2.7 (1.1)
2.6 (1.0)
2.8 (1.1)
12422
0.2
Contributing to a group ( n = 235) mean ( SD )
2.7 (1.5)
2.8 (1.5)
2.6 (1.5)
6370
0.2
Sharing stuff from the Internet ( n = 44) mean ( SD )
2.8 (1.7)
3.4 (1.9)
2.5 (1.6)
272
0.1
Viewing friends’ pictures ( n = 363) mean ( SD )
2.9 (1.4)
3.1 (1.4)
2.8 (1.3)
14123
0.08
Gaming ( n = 66) mean ( SD )
2.9 (1.7)
1.9 (1.0)
3.1 (1.8)
198
0.032
1 These scores are ranks, meaning that lower scores indicate higher preferences.
Problematic use of FB was measured with a revised version of the Internet Addiction Test (IAT-R). The IAT-R is a modified version of the IAT-20 Young questionnaire on Internet addiction. It has been modified for statistical reasons in order to take into account current Internet use and the fact some items of the original version are outdated [ 44 , 45 ], and to include a question on craving, which was not measured in the original version of the IAT. Indeed, previous research emphasized the relevance of using a craving item when assessing problematic and addictive-like online activities [ 46 ]. The IAT-R comprises 18 items scored on a 5-point Likert scale ranging from 1 to 5. Cronbach’s alpha for the entire scale, computed on the polychoric correlation because of the categorical nature of the item responses, is 0.92.
Impulsivity traits were measured with the short version of the UPPS-P Impulsive Behavior Scale [ 47 ]. The UPPS-P is a self-reported questionnaire that assesses five distinct impulsive traits: negative urgency (e.g., “When I am upset I often act without thinking”; Cronbach’s alpha for the current sample: 0.82), positive urgency (e.g., “When I am really excited, I tend not to think of the consequences of my actions”; Cronbach’s alpha for the current sample: 0.75), lack of perseverance (e.g., “I finish what I start”; Cronbach’s alpha for the current sample: 0.86), lack of premeditation (e.g., “Before making up my mind, I consider all the advantages and disadvantages”; Cronbach’s alpha for the current sample: 0.82), and sensation seeking (e.g., “I sometimes like doing things that are a bit frightening”; Cronbach’s alpha for the current sample: 0.84). Previous research conducted with the UPPS-P was characterized as having a solid factorial structure, adequate test-retest validity, and high internal reliability [ 47 ].
Data analyses
Descriptive statistics about FB use and preferences, along with gender comparisons, are presented in Tables 1 and 2 . Chi-square tests were computed for FB use, whereas the Wilcoxon test was used for age and rank variables, since the distributions were skewed.
Regression analyses were computed to identify predictors of problematic FB use. In all regression models computed, the dependent variable was the total score on the IAT-R and the independent variables entered were sex, age, FB activities, and impulsivity traits (UPPS-P). A hierarchical regression strategy was performed. In a first step, only sex, age and FB activities were entered in the model. In a second step, impulsivity subscales (UPPS-P) were entered. Residual analysis showed problems such as non-normality and outliers; therefore, the logarithm of the IAT score was used instead and this transformation provided satisfactory residual analysis. From the variance inflation factors (VIF), no sign of multicollinearity was found.
All statistical analyses were done with R 3.3.0 [ 48 ]). All study data are available from https://osf.io/xt4zv/ .
Results
Fig 1 shows the distribution of the IAT scores. The theoretical scores range is between 18 and 90 while the actual scores range is between 18 and 74. 25% of the participants had a score lower that 28, the median score was 33 and 75% of the subjects had a score lower than 40.
10.1371/journal.pone.0201971.g001
Fig 1
Distribution of the IAT scores.
Theoretical score range: 18–90; Actual score range: 18–74.
The first part of Table 1 shows the types of activities performed on FB. The most prevalent FB-related activity appeared to be reading the news feed, followed next by viewing friends’ pictures. Gender differences appeared, however, among the types of activities performed on FB. Indeed, reading the news feed was the most often reported activity by men, whereas it was the second-most often reported activity for women, for whom viewing friends’ pictures was the most prevalent activity. Notably, gaming was the seventh most prevalent reason to use FB, which was the penultimate one (the last for men). Statistically significant differences occurred between genders, the first one being reading the news feed, which men reported doing more often than women did (65.2% vs. 54.1%). In addition, twice as many men update their status compared to women (45.2% vs. 28.8%), whereas gaming had the reverse pattern (11.6% of women reported gaming on FB vs 5.7% of men). The second part of Table 1 shows the type of FB access, with 43.5% of respondents declaring that they use the Messenger app and 44.7% that they use notifications. Women reported using notifications (48.7%) more often than men did (35.7%). Most of the subjects (87.6%) reported using FB mainly on their personal computer or their smartphone, but there was a gender difference for main access, as men preferred to use their personal computer (47.1%) and women preferred to use their smartphone (57.9%).
Table 2 reports ranked preferred activities performed on FB. Although reading the news feed had the smallest rank for both men and women, men notably ranked gaming as the second activity, whereas women ranked reading friends’ timelines as second. Moreover, the mean rank for gaming was statistically lower for men (1.9) than for women (3.1), whereas updating their status was higher for women (2.8) than for men (2.2).
Table 3 shows the results of the hierarchical regression analysis computed to predict problematic use of FB (logarithm of the total score on the IAT-R). In a first step, only gender, age, FB activities, and use were entered in the model. Age was negatively associated with problematic FB use, whereas both updating one’s status and gaming were positively associated with it. Interestingly, the use of notifications was also positively associated with problematic use of FB. In a second step, impulsivity subscales were added to the model. The increase in the R -squared value between Model 1 and Model 2 was statistically significant. All the previous statistically significant associations from the first model remained significant in the second model, except for notifications. Notifications were nevertheless barely not significant, which may be due to a decrease in power or impulsivity that could have had a mediating effect because negative urgency, positive urgency, and lack of premeditation were positively associated with problematic FB use. The R -squared value of Model 2 was more than twice that of Model 1 and was statistically significant, suggesting that adding impulsivity facets dramatically increased the quality of the model.
10.1371/journal.pone.0201971.t003
Table 3 Predictors of FB problematic use: Hierarchical regressions.
Model 1
Model 2
Estimate
SE 1
t
p -Value 2
Estimate
SE
t
p -Value 2
Women vs. Men
-0.039
0.024
-1.66
0.10
-0.023
0.023
-1.01
0.31
Age
-0.003
0.001
-3.29
>0.01
-0.003
0.001
-2.82
>0.01
FB activities (yes vs. no):
Updating status
0.090
0.024
3.76
>0.001
0.090
0.022
4.03
>0.001
Reading the news feed
0.035
0.022
1.58
0.11
0.018
0.021
0.87
0.39
Reading friends’ timelines
0.012
0.022
0.57
0.57
0.002
0.020
0.09
0.92
Commenting
-0.019
0.024
-0.81
0.42
-0.030
0.022
-1.37
0.17
Viewing friends’ pictures
0.014
0.022
0.63
0.53
0.009
0.021
0.44
0.66
Gaming
0.096
0.036
2.68
>0.01
0.079
0.034
2.34
0.02
Contributing to a group
0.040
0.023
1.77
0.08
0.034
0.021
1.64
0.10
Sharing stuff from the Internet
-0.013
0.042
-0.30
0.77
-0.025
0.040
-0.63
0.53
Messenger app (yes vs. no)
0.014
0.023
0.61
0.54
0.010
0.022
0.48
0.63
FB main access (vs. personal computer):
Professional computer
-0.037
0.043
-0.85
0.40
-0.052
0.041
-1.27
0.20
Smartphone
-0.039
0.024
-1.59
0.11
-0.044
0.023
-1.93
0.05
Tablet
-0.070
0.048
-1.48
0.14
-0.072
0.044
-1.63
0.10
Notifications (yes vs. no)
0.049
0.023
2.15
0.03
0.039
0.021
1.85
0.06
Impulsivity traits :
Negative urgency
0.009
0.005
2.03
0.04
Positive urgency
0.022
0.005
4.15
>0.001
Lack of premeditation
-0.006
0.005
-1.07
0.29
Lack of perseverance
0.032
0.005
6.53
>0.001
Sensation seeking
0.001
0.004
0.37
0.72
R 2
0.075
0.206
Delta R 2
0.130
>0.001
1 Standard error of the regression coefficient.
2 p-Values smaller than 0.05 are in bold.
Discussion
In the current study, we first aimed to highlight how FB use is multidetermined and related to a range of specific activities, and we second sought to identify which types of usage preferences and impulsivity profiles predict problematic involvement in FB. On the whole, we identified that various motives and related activities “fuel” FB use and that gender differences exist in relation to FB usage preferences. Specific usage preferences (i.e., updating status, gaming via FB, and using notifications) and impulsive personality traits (i.e., positive urgency, negative urgency, lack of perseverance) are related to problematic use of FB. Similar to what has been found in previous studies, younger age was also associated with increased excessive use of FB [ 10 ].
The current study first provides, in a sample of French-speaking individuals, insight into FB usage preferences (see Tables 1 and 2 for related results) and their potential predictive value in explaining symptoms of problematic use. Among the various results related to FB usage preferences, three specific findings are of much importance and deserve further elaboration. The first is that the tendency to update FB status is an important predictor of problematic use of FB. A possible explanation for this finding is that constantly updating one’s status probably interferes with daily activities and thus engenders negative consequences. For example, proneness to frequently update one’s status likely negatively impacts on professional or academic productivity through the frequent switching of one’s focus of attention from the activity in question to FB. Proneness to frequently update one’s status is also likely to have negative social consequences. This will particularly be the case when smartphone users update their status in social situations, resulting in “phubbing” behaviors, i.e., the act of snubbing someone in a social setting by using one’s phone instead of interacting [ 49 ]. A second important finding pertains to the fact that gaming through FB is an important predictor of problematic usage. In this regard, it is possible to assert that gaming is probably a highly addictive SNS-mediated activity. Such a result supports the view that labels such as “FB addiction” or “SNS addiction” are misleading and that a focus on the actual activities performed on SNSs is crucial when making inferences regarding potential dysfunctional usage. Such an argument is in line with recent criticisms made about other umbrella constructs such as “Internet addiction” [ 50 ] or “mobile phone addiction” [ 51 ]. In relation to FB-mediated gaming, our study also highlighted that there are a higher proportion of female gamers, which may seem at first sight surprising. Nevertheless, a possible explanation is the nature of video games that are generally played via FB, which mainly belong to the category of “casual games” (prototypical examples: Candy Crush Saga, Farmville). These video games, which are generally easy to handle and can be played for even short sessions of a few minutes are more popular among females [ 52 ], which could explain the gender difference observed in our study. We highlighted that using notifications is a predictor of problematic FB use. Even if this result becomes marginally significant ( p = .06) when impulsivity-related variables are included in the regression model, it suggests that switching off FB-related notifications can constitute a protective factor against problematic use.
The current research report also showed that specific impulsivity traits predict problematic involvement in FB. First, our results emphasized that problematic use of FB is predicted by heightened positive and negative urgency. Accordingly, disordered FB use is linked to emotion-laden impulsive behaviors and can thus be conceptualized as a dysfunctional strategy to cope with aversive emotions (for individuals with high negative urgency) or a way to promote or maintain pleasant emotions (for individuals with high positive urgency), despite the potential resulting negative consequences (e.g., conflicts, interference with daily life routines, reduced sleep). In other words, individuals with a high (positive and/or negative) urgency trait are at increased risk of developing FB usage that constitutes a maladaptive emotion regulation strategy associated with tangible negative consequences. Such an explanation is in line with the large body of evidence suggesting that a wide range of problematic behaviors, such as alcohol abuse [ 53 ] and compulsive buying [ 54 ], can be viewed as urgency-related behaviors displayed to regulate (suppress and/or exacerbate) emotional states in the short term despite their delayed negative consequences [ 36 , 55 , 56 ]. Of importance, heightened urgency has also been related to reduced inhibitory control and poor decision making [ 57 , 58 ], implying that the deregulated use of FB demonstrated by individuals with a high urgency trait could be sustained by conjoint deficits in emotion regulation and executive control skills. Yet, the few preliminary data available from the literature tend to emphasize persevered frontal lobe functioning in individuals displaying excessive FB use [ 59 , 60 ], implying that further research is needed to better understand the psychological and neurological processes at play. Second, the lack of a perseverance facet of impulsivity was also found to predict addictive use of FB. This impulsivity component, defined as difficulty in remaining focused on cognitively demanding and/or boring tasks [ 37 ], has been related to attentional difficulties [ 61 ] and to an increased occurrence of distractions or irrelevant thoughts that may interfere with ongoing tasks or project completion [ 62 , 63 ]. Lack of perseverance, among other things, was found to predict procrastination behaviors [ 64 ] and elevated frequency of mobile phone use [ 65 ]. Accordingly, we postulate that a lack of perseverance increases actual and potentially exaggerated use of FB because of attentional fluctuation, mind wandering, or irrelevant thoughts (e.g., someone checking a friend’s page following an intrusive thought about this friend; someone checking FB following difficulties in concentrating on an ongoing task). Such an assumption is in line with previous results that linked problematic use of SNSs with symptoms of attention-deficit hyperactivity disorder [ 10 ]. Notably, because of the correlational design of the study, we cannot exclude a reverse explanation, i.e., problematic use of FB in itself interferes with other competing tasks and thus promotes lack of perseverance.
Some limitations of the study warrant consideration. First, the sample is self-selected and thus not necessarily representative of the population under study. For example, according to Sprout Social ( https://sproutsocial.com/insights/new-social-media-demographics/ ), 44% of FB users are females (at the worldwide level), implying that they are overrepresented in our sample. Although not problematic in relation to our findings regarding the factors involved in problematic FB use, this nonetheless calls for extra caution when considering the usage patterns and preferences as potentially reflecting those of the FB French-speaking community. Second, the questions related to FB use and preferences were generated in the framework of the current study, implying that we cannot exclude the possibility that additional reasons or motives to use FB would have been identified with other methodologies (e.g., focus groups, open-ended questions, usage time trackers). Third, the analyses regarding gaming were drawn from a rather small sample (12 men and 54 women), therefore the results should be taken with caution. Finally, the current study relied on self-reports, which makes the data collected prone to be influenced by lack of insight and social desirability biases [ 66 ].
Despite these limitations, the current study offers several original results that could fuel further research. Indeed, given the heterogeneous nature of SNS-related activities, future studies should systematically measure the activities performed, rather than considering the SNS user population as a homogenous group. As an illustration, our results suggest that individuals gaming on FB are more prone to display symptoms of problematic use, which could imply that some video games playable via FB are addictive, but not necessarily that FB is addictive per se. At a broader level, the links observed between various impulsivity traits and problematic FB use are in accordance with those presented in recent theoretical models [ 35 , 67 ], which posit that impulsivity and related mechanisms (e.g., impairment in inhibitory control and decision making) play a pivotal role in the onset and maintenance of specific Internet use disorders. At a more applied level, the finding that using notifications predicts addictive-like use of FB has straightforward implications in terms of prevention of problematic SNSs use. Furthermore, it is reasonable to suppose that a similar approach can be generalized to other types of notifications (e.g., most smartphone apps send notifications) and thus to a wide range of potentially problematic online applications.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Paleogenomics, the study of genomes from extinct organisms, is an emerging scientific field that has been fuelled by recently developed technologies in high-throughput DNA sequencing [1] , [2] . In the first of such approaches to be undertaken [3] , about 27,000 base pairs (bp) of cave bear ( Ursus spelaeus ) genomic sequences were obtained with cloning vectors from 42,000 and 44,000 years-old cave bear samples, respectively. Using sequencing-by-synthesis (SBS) technology, as applied through the Roche/454 Life Sciences GS20 and FLX pyrosequencing platforms [4] , 13 million bp of the woolly mammoth ( Mammuthus primigenius ) genome were generated from a 28,000-year-old permafrost mammoth bone [5] , complete mitochondrial genomes from mammoth and thylacine hairs [6] , [7] and, finally, about 80% of the nuclear genome from ∼20,000-year-old mammoth hairs [8] . The same approach has been applied to other ancient bones, including Neanderthal samples that provided around 1 million bp of its genome [9] , as well as other Pleistocene mammals from Denisova cave in Siberia [10] . However, the efficiency of these metagenomic analyses is notably variable: while in the mammoth bone it was possible to identify from 45.4% (mainly in bone samples) to 90.45% (in hair samples) of the sequences as endogenous, this fraction was significantly reduced in cave bear (between 1.1 and 5.8%), Neanderthal (6%, although a significant fraction of contamination was posteriorly estimated to be present in this particular extract [11] ), ancient horse (0.7%), ancient wolf (1.8%) and cattle (1.1%) [3] , [5] , [9] , [10] . In addition, due to the low genomic coverage, the degradation of the template DNA, and the innate error rate of the sequencing platforms [12] , the paleogenomic data contains a significant number of sequencing errors, resulting in an excess of C to T substitutions due to cytosine deaminations as compared to the corresponding reference genome [3] , [5] , [9] , [13] . Thus, it is likely that in the future, specific loci in regions with low shotgun coverage would need to be verified by targeted approaches, such as the polymerase chain reaction (PCR). Several studies in mammoths and Neanderthals have already focussed on the specific retrieval of nuclear genes and the problems of distinguishing endogenous variants from DNA damage [14] – [16] .
Paleogenomic data can be useful for understanding the rate and nature of some evolutionary processes, because it allows us to investigate the genetic basis of adaptive traits in extinct species [14] . At present, however, it is not clear what the limits of these new technical approaches are, in terms of efficiency (ratio of endogenous versus exogenous DNA retrieved), age of the sample, geographic location and/or thermal history. For instance, some mammal species, including goats, cervids, elephants and hippos, have gone extinct in the last few thousands years in the Mediterranean islands [17 and references therein] , a temperate area which is clearly not favourable to DNA preservation. The possibility of having access to the genome of these species is therefore of great interest for exploring unique insular evolutionary patterns.
In previous studies [18] – [20] , we have retrieved by PCR mitochondrial genes (Cytochrome b , 12S rRNA) and a multi-copy nuclear gene (28S rRNA) from one of these species, Myotragus balearicus , an extinct goat from the Western Balearic Islands (Western Mediterranean). Myotragus is an extremely modified caprine [21] that evolved in insularity conditions since the end of the Messinian crisis (5.35 million years ago) in the islands of Mallorca and Menorca [22] , [23] . It became extinct between 3,700 and 2,040 years BC, probably after the arrival of modern humans to the Balearic Islands [24] , that took place between 2,350 and 2,150 years BC [25] . The unclear taxonomic position of this caprine is related to its amazing morphological peculiarities, which include extreme size reduction (250–500 mm shoulder height), a single, ever-growing rodent-like lower incisor, shortened distal limb bones, frontal eyes, and reduced brain size [26] – [31] . Although ancient mitochondrial DNA (mtDNA) data have provided statistical support for a Myotragus clade with Ovis [20] , the general phylogeny of the caprine group is not yet fully established [20] . Here, we demonstrate that it is possible to undertake shotgun sequencing approaches from ancient bones from the thermally unfavourable Mediterranean area. Additionally, the paleogenomic data obtained from Myotragus support phylogenetic relationships previously generated with mtDNA sequences.
Results
A total of 96,357 singleton GS-FLX sequence reads were obtained and analysed by means of database searches. No significant identity was found for 98.49% of the sequences, a figure higher than that found in the ancient wolf, horse and cattle (86.8% on average), Neanderthal (79%) and mammoth shotgun (5.53%, 18.4% and 24.92%, depending on the study) [5] , [8] , [9] , [10] . A fraction of those sequences could be endogenous, but remain unidentified due to the incompleteness of the cow and specially of the sheep genome. Alternatively, the high fraction of sequences without any match may reflect a lack of environmental DNA studies in the Mediterranean area. The remaining 1.51% of the sequences were taxonomically classified by the highest identity found in the database.
Only 0.27% of the sequences, comprising 15,832 nucleotides, gave the best hit to the cow genome, with an average percentage identity of 94.95%. This figure seems to be in agreement with divergence times of about 12–14.3 million years for the cattle-caprine lineages, as suggested from genetic and morphological data [32] . In addition, 0.35% of the sequences gave the best hit to the human genome, with almost 100% of identity, indicative of exogenous contamination. The most represented taxonomic group, however, was bacteria (0.69%), followed in decreasing order by invertebrates (0.12%), plants (0.05%), fungi (0.02%) environmental sequences (0.02%) and others (0.02%). The average length of these sequences was 59.97 nucleotides, and they ranged from 30 bp (determined by the length cut-off in the analysis) to 245 bp (limited by the GS-FLX technology) ( Figure 1 ). The length was similar to those of putatively endogenous sequences found in the Neanderthal and the cave bear metagenomic library (52 and 69 nucleotides, respectively) [3] , [33] . The presence of two sequences deriving from the Bos Y chromosome indicated that the Myotragus specimen analysed was likely male.
10.1371/journal.pone.0005670.g001 Figure 1
Size distribution, plotted in 10 bp bins, of Neandertal [10] , cave bear [3] and Myotragus sequences obtained from metagenomic analyses.
The average hit size in each case is indicated by a dotted line.
The human contaminant sequences were significantly longer on average than the Myotragus ones (81.57 and 59.97 respectively, P <0.0001), suggesting that they were more recent and therefore, less degraded. The longest (>200 bp) Myotragus sequences did not have higher identities to the cow sequences than the shorter ones (85.6% versus 94.95%, respectively), indicating that they are unlikely to derive from recent cow-mediated DNA contamination. Furthermore, no remains of cow were found inside the Cova Estreta cave.
To additionally confirm that the bovid-like DNA fragments were endogenous, we designed five primer pairs from the shotgun sequences that matched Myotragus specific substitutions in their 3′ ends and represented unambiguous (those that did produce only one match to the Bos genome) BLAST hits. These nuclear fragments, varying between 80 and 112 nucleotides in length, were co-amplified with a previously known 113 bp fragment of the 12S mtDNA gene [20] . In the PCR, we used, to overcome inhbitors present in ancient extracts, rat serum albumin (RSA) [34] instead of the usual bovine serum albumin (BSA) to avoid possible cow contamination in the BSA. One nuclear fragment, along with the mtDNA gene, showed an amplification product and was subsequently cloned and sequenced. The nuclear sequence was identical to that obtained in the shotgun sequencing except for two nucleotide changes that could be related to DNA damage, both in the shotgun and in the PCR-generated sequence (Figure S1).
The plotting of the Myotragus sequences along the Bos chromosomes showed an excess of sequences in chromosomes 3, 16 and 23, although they were not statistically significant after applying a Bonferroni correction ( Figure 2 ). This pattern could correspond to chromosomal duplications unique to the Myotragus lineage or shared by all the Caprinae species, although more sequences and the completion of the Ovis genome are needed to explore in the future this possibility. Most of the identified Myotragus sequences correspond to unannotated genomic regions of the cow, with only 3.42% of the sequences and 3.90% of the nucleotides belonging to coding regions (Figure S2). The predicted Myotragus genes are listed in Table S1.
10.1371/journal.pone.0005670.g002 Figure 2
The proportion of the Bos genome contained on each chromosome (blue bars) is shown with the proportion of Myotragus sequences (red line) aligning to each Bos chromosome with exactly 1 hit with e-value<1e-3 BLAST.
The observed distribution is not significantly different from the expected one when we compare all the chromosomes together ( P = 0.081) or when we tested each one independently and correct for multiple testing.
To explore the phylogenetic signal of the Myotragus sequences, we further searched for orthologous sequences in three Bovidae species ( Bos taurus , Ovis aries and Capra hircus ) and one Cervidae species ( Muntiacus ) in GenBank. However, we noticed a greater genomic coverage of the Bos genome that generated an excess of matches due to the presence of multiple paralogs. These sequences might remain undetected in the other genomes due to their more limited coverage. Therefore, we created a sub-dataset of 80 sequences (accounting for a total of 1,987 nucleotides after removing gaps and missing data) that included only those sequences that did not produce multiple matches in none of the genomes. With these sequences, we generated a maximum-likelihood phylogenetic tree that showed the topology previously established from mtDNA data for these species [20] ( Figure 3 ), in which Myotragus grouped first with Ovis . However, the bootstrap support for this tree was low (64%). The same topology was found with Bayesian trees with a probability of 0.97 for the Myotragus - Ovis group. The overall congruence of this partial genomic phylogeny and the mtDNA tree further supports the authenticity of the Myotragus sequences.
10.1371/journal.pone.0005670.g003 Figure 3
Maximum-likelihood phylogenetic tree of Myotragus balearicus and other artiodactyls.
The tree was rooted in the cervid Muntiacus reevesi . Numbers along the branches indicate bootstrap support of the maximum-likelihood analyses (first number) and Bayesian support of an independent Bayesian analysis (second number). The scale bar represents 0.01 substitutions/site.
The large branch found for Myotragus in the phylogenetic tree ( Figure 3 ) could be attributed to sequence changes due to DNA damage or to an accelerated evolution of the Myotragus genome. To test these possibilities, we characterised the nucleotide changes exclusively present in the aligned Myotragus sequences (and different to those from Ovis and Bos ) and found a statiscally significant ( P <0.05) bias towards higher C to T/G to A ratios, as compared to the T to C/A to G ( Figure 4 , Table S1). This feature has been previously described as damage-derived lesions due to cytosine deaminations [35] . However, the removal of these substitutions from the alignment only barely shortened the Myotragus branch in subsequently generated trees. Specifically, the Myotragus branch was 3.1 times longer than the Ovis branch in the original alignment and it was still 2.8 times longer than Ovis after removing putatively damaged positions. Thus, an important contribution of accelerated evolution in the Myotragus genome cannot be discarded. However, a similar acceleration in Capra indicates that this phenomenon is not specific to the Balearic lineage.
10.1371/journal.pone.0005670.g004 Figure 4
Frequency distribution of 113 Myotragus -specific substitutions observed in 3,602 bp of aligned Ovis , Myotragus and Bos genomic sequences.
Complementary substitutions (such as C to T and G to A) are considered equivalent events. Fisher's exact test was used to calculate the excess of Myotragus -specific C to T and G to A transitions. The total number of each substitution is in parentheses.
Discussion
Molecular studies, mainly based on mtDNA data, have failed until now to fully resolve the caprine phylogeny, probably due to the explosive radiation of this group [20] . The phylogenetic analysis of the present paleogenomic data supports the previous caprine relationships established from mtDNA, but also indicates the potential of this approach for testing evolutionary hypothesis and establishing robust phylogenies.
Despite being excavated in a region with a mean annual temperature of 14°C and below 40 degrees North latitude, we have been able to successfully retrieve nuclear genome sequences from a ∼6,000 years old Myotragus balearicus bone. The extremely low efficiency of the paleogenomic retrieval is striking, as is the fact that the level of human contaminant sequences is higher than that of the endogenous ones (0.34 vs 0.27). In contrast, the ratio of endogenous to contaminant human sequences among the colder preserved Denisova mammalian samples was 49∶1 [10] , and the human sequences accounted for less than 0.015% [10] . In a mammoth sample from the Artic Circle [5] , this ratio was 32∶1, and the human contaminants up to 1.4% of the total sequences. The Myotragus sample was retrieved with no special precautions against contamination. However, the histological structure of the bone also correlates with contamination levels [36] , and the cortical tissue in Myotragus limb bones is thinner than in other, larger extinct mammals, such as mammoths and Neanderthals. Thus, it is not clear if this figure can be taken as an estimate of potential human contamination in ancient bone specimens stored in museums.
The efficiency ratio of retrieval of endogenous Myotragus sequences is the lowest among those observed in some other bone-based metagenomic studies, obtained from samples at higher latitudes: 47.4°N [3] , 74°N [5] , 45.5°N [8] and 51.23°N [10] . Despite the low efficiency values, the mean fragment length and the range value of the Myotragus sequences are similar to those found in samples with higher efficiencies. In addition, the frequency of damage-derived lesions in the Myotragus sequences is 4.2 times lower than those found in Neanderthal sequences [33] . Even so, some inconsistencies have been found between the shotgun and the PCR-based sequence, indicating the need for targeted approaches in genomic regions with low coverage. A previous study has estimated that a 12-fold coverage would be needed to have an error rate of 1 in 10,000 nucleotides [37] , something extremely expensive to achieve in highly degraded ancient samples.
These somehow contradictory results between low retrieval efficiency and low DNA damage can be due to a combination of factors. On one hand, the temperate climatic conditions of the Mediterranean islands are highly unfavourable to paleogenomic preservation, although the cave where the bones were found has maintained a rather constant temperature inside. On the other hand, the Myotragus sample used is much more recent than those from wolf, horse, cattle, cave bear, Neanderthal and mammoth, all of them dated between 20,000 and 69,000 years ago [3] , [5] , [8] , [9] , [10] . However, the estimated thermal age [38] for this bone at the excavation is 26,206 years at 10°C (David Harker, personal communication). This age is older than that estimated for the 100,000 years-old Scladina Neanderthal [38] , which is in agreement with the low efficiency of DNA retrieval found in the present study.
Our findings imply that we are working at the very limits of the current paleogenomic approaches, but still they are more efficient than PCR-based strategies, which are problematic for genomic studies on similarly preserved samples. In fact, under these unfavourable environmental conditions, only paleogenomic approaches can provide the amount of sequence data generated here. In the future, with greater genomic coverage, paleogenomic approaches could provide further data to study other aspects of this Balearic endemism, such as evidences for selective sweeps in the Myotragus genome related to its particular adaptations. Also, our results suggest that livestock domestication events that took place in the Fertile Crescent could be approachable from paleogenomics.
Materials and Methods
A left Myotragus radius bone (IMEDEA 43619) from Cova Estreta (Pollença, Mallorca) was chosen for analysis because of its excellent macroscopic preservation. Previous analyzed bones were excavated in a different site, Cova des Gorgs (Escorca, Mallorca) [20] . Cova Estreta is a deep and narrow cave discovered in 1996 that acted as a natural trap for Myotragus [31] . Radiocarbon dates from bones obtained from the same stratigraphical unit [UtC-5175, 6,357±44 BP (5469–5225 calBC)] and [UtC-5171, 5,720±60 (4716–4449 calBC)] allow us to establish a narrow chronological age for the studied material of ∼6,000 years.
A sample of 3 g of cortical tissue was powdered, digested with proteinase K and extracted with phenol-chlorophorm, following a protocol described elsewhere [5] . Previous metagenomic studies have described an overwhelming fraction of environmental DNA found in ancient bones. Following a previously published procedure, the bone powder was incubated with bleach for five minutes, prior to extraction [39] . It was assumed that this could remove part of the pervasive exogenous DNA and thus increase the efficiency of the endogenous DNA retrieval. The fact that so much contamination is still seen afterwards is intriguing. Further studies could help clarify the efficiency of the bleach treatment prior to GS-FLX 454 sequencing.
One hundred microliters of extract were subjected to the GS-FLX 454 sequencing platform. The nebulization and Ampur purification steps were omited for the library building process, following, except for this, the manufacturer's guidelines (Roche Diagnostics). The amount of DNA in the libraries was estimated by Quantitative PCR (qPCR) [40] and found to be too low for successful sequencing. Therefore, libraries were amplified with the emulsion primers prior to the emulsion PCR (ePCR) to increase the amount of DNA. This procedure generated redundant sequences that were posteriorly identified and eliminated. Subsequently half of a full sequencing run was performed on the commercial Cogenics Genome Express FLX platform (Grenoble, France). To confirm the authenticity and accuracy of the GS-FLX generated data, a small number of mtDNA and nuclear DNA sequences were targeted using conventional PCR protocols, following a two-steps protocol [41] and 50 degrees of annealing temperature. Amplification products were cloned using the TOPO TA cloning kit (Invitrogen), and sequenced using an ABI3730 capillary sequencer (Applied Biosystems).
Obtained sequences were identified with BLAST searches [42] (using the megaBLAST program with an e-value threshold of 0.001) using the cow and human genomes, the environmental sample sequences database in the GenBank env , and the general nucleotide sequences nt . Sequences of other bovids ( Bos taurus , Ovis aries and Capra hircus ) as well as one cervid ( Muntiacus reevesi ) were aligned to those of Myotragus with Multialin [43] . Discrepancies in homopolymeric tracts were not considered, as 454 technology is known to have problems dealing with these regions [33] . Best match to target sequences in the blast didn't include the edge nucleotides, since these are known to accumulate postmortem damages associated to the breakage DNA process.
Phylogenetic trees were constructed by maximum likelihood with the Phyml program, version 2.4.4 [44] . A general time reversible (GTR) model with four rate categories and a proportion of invariable sites was used, with parameters estimated from the data. A bootstrap analysis with 100 replicas was also performed. In addition, a Bayesian tree was calculated with MrBayes 3.1 [45] using a GTR model with invariable sites and rate heterogeneity. Two runs of four chains of 5,000,000 trees were generated, sampling every 100 trees, with burning completed by the 20,000th tree.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Expanding the footprint of humans has modified ~ 70% of Earth’s land surface [ 1 , 2 ], causing not only the loss of carnivore habitat but also affecting the animal movement and global species recovery efforts [ 3 ]. Because of landscape modification, an increasing number of large carnivore species have been forced to inhabit with humans in modified landscapes (hereafter shared landscape) [ 4 ], which likely have escalated the human-carnivore interactions [ 5 ]. The extent to which humans co-adapt with the carnivore in the shared landscape is the key to the success of coexistence [ 6 ]. However, carnivores pose several real and perceived threats to humans (i.e., economic loss and deaths) living in their vicinity [ 7 ]. These threats directly lead to retaliatory killings and local extinction of large carnivores due to intolerant behaviours by humans, threatening carnivore conservation efforts [ 8 ]. However, the behavioural adaptation of large carnivores (i.e., reduced home range in human-dominated areas and increased nocturnal activity) [ 9 ], their socio-ecological importance [ 10 , 11 ], and human tolerance are key mechanisms facilitating coexistence in shared landscapes [ 12 – 14 ].
Despite the several national and international efforts to protect carnivores globally, the populations continue to decline in response to increasing human populations and political instability [ 15 ]. World current population projections indicate further expansion of human footprint globally when ~2.0 billion people are expected to add to the current human population by 2050 [ 16 ]. For the long-term persistence and strengthening of the conservation efforts of carnivores in the Anthropocene, an understanding of (i) how carnivores adapt and use the matrix (remnant of natural vegetation) in shared landscapes [ 17 , 18 ], and (ii) how carnivores render benefits to humans’ well-being, are the key determinants [ 19 ]. For example, spotted hyenas ( Crocuta crocuta ) in northern Ethiopia are valued and tolerated for their service of removal of carcasses of livestock and reducing the risk of disease [ 20 ]. In the Himalayan region, because of the cultural and religious beliefs, the Tibetan Buddhist monasteries protect snow leopards ( Panthera uncia ) and their habitats, although snow leopard heavily depredates on their livestock [ 21 ]. After the eradication and recent recolonisation of large carnivores in certain regions of Europe and the USA, changes in human behaviour and tolerance towards carnivores have been observed [ 7 , 15 ].
India is home to 23% of global carnivore species, wherein the carnivores share space with a population of 1.3 billion people in multiuse landscapes (e.g., forest, agroforests, scrublands, barren lands, grasslands) [ 22 ]. Among these, ~4% forest is protected, while ~19% are unprotected forest cover of the whole country’s land area [ 23 ], which has been used by carnivores for different purposes, including foraging, dispersal, and reproduction [ 24 ]. Carnivore generally adopted behavioural mechanism can lead to coexistence in shared landscapes via (i) spatial avoidance of human-dominated areas [ 5 ] (ii) overlap in the same space with humans but temporally avoid humans, e.g., nocturnality [ 25 ]. However, shared landscapes are recently recognised as potential habitats for many wildlife species of conservation interest [ 15 , 26 ]. However, for the long-term persistence of carnivores in shared landscapes, it is essential to identify the ecological and anthropogenic factors that facilitate human-carnivore coexistence [ 27 ].
In this paper, we focused on striped hyena ( Hyaena hyaena) , a species listed as Near Threatened by the International Union for Conservation of Nature (IUCN) Red List [ 28 ], as a model species to understand the coexistence pattern with humans. The striped hyena is a large-bodied, asocial, and solitary carnivore [ 29 ]. It exhibits nocturnal activity [ 29 ] and is widely distributed in India’s arid and semiarid landscapes [ 30 ]. It is a ‘facultative scavenger’ adapted to coexist with humans and mostly scavenges on domestic and wild ungulate carcasses [ 30 ]. We studied habitat use and interactions between the striped hyena and humans in the shared landscape (dominated by agro-pastoral activities) of Sawai Mansingh Wildlife Sanctuary (SMS WLS), Rajasthan, India. Based on our prior knowledge of the ecology of the species and previous research in the landscape [ 31 ], we used data derived from motion sensors-based surveys and satellite remote sensing images and household questionnaires to understand how ecological and anthropogenic factors facilitate their persistence in the shared landscape. We aimed to understand the major drivers of coexistence between the striped hyena and human in the shared landscape in the following hypotheses; (i) being a scavenger; hyena does not attack livestock and humans; hence humans tolerate striped hyenas as they clean the organic waste generated by humans and reduce disease risk [ 32 ]. Hence, we predict a higher density of striped hyenas in the shared landscape, (ii) behavioral adjustments (i.e., nocturnality) of striped hyena tend to minimise human interactions. Hyenas in the landscape use the same space as humans but temporally avoid humans and (iii) spatial partitioning of spatial avoidance of humans via spatial use of the landscape by which hyenas reduce interaction with humans. In this study, we direct a comprehensive view towards the ecological attributes of striped hyenas, which are helpful to managers and conservationists to accurately determine parameters influencing striped hyena’s presence for optimising investment in the management of resources.
Material and methods
Ethics statement
This study was conducted after getting permission from Rajasthan Forest Department (letter no- F 19(11) permission/cwlw/2017/1678). We followed all guidelines for animal care and scientific research ethics.
Study area
We conducted this study inside and buffer area of Sawai Mansingh Wildlife Sanctuary (SMS WLS), Rajasthan, India ( Fig 1 ). The SMS WLS is a part of the tiger conservation and management unit of the Ranthambhore Tiger Reserve [ 33 ]. The total area of SMS WLS is 127.6 km 2 , while another adjacent forested area is Qualji area 7.58 km 2 and another forested area 132.96 km 2 [ 34 ]. The entire Ranthambhore landscape forms a transition zone between the true desert and seasonally wet peninsular India [ 35 ]. The area falls in the 4B semiarid zone and Gujarat-Rajwara biotic province [ 36 ]. The region’s average annual rainfall is 800 mm, of which 500 mm falls in the monsoon season. The temperatures can be ≤ 2°C in January and ≥ 47°C in May [ 31 ]. The landscape is undulating and dominated by humans; there are 75 villages within a 5 km buffer of SMS WLS with more than 104261 people inhabiting in and around [ 37 ]. The rolling landscape mosaic is interspersed with forest, scrublands, grasslands, riverine areas, and agricultural lands [ 31 ]. The residents are mostly engaged in agriculture, livestock farming, cutting grass, grazing livestock, lopping trees, and mining (illegally) to supplement household incomes. All the villages are primarily dependent on agriculture for their livelihood, and their economy is supplemented by animal husbandry. They have numerous cows, buffaloes, and goats but very few herds of sheep and camel. The villagers tend to graze their animals in the fallow agricultural lands and the village commons during the lean periods of the year, viz., January to June. However, the villagers enter the peripheral forest area to graze their animals throughout the year. The area is dominated by northern tropical, dry, deciduous, and thorny forest [ 38 ]. The forests are mainly of edaphic climax and belong to the subgroup 5B- Northern Tropical Dry Deciduous forests and subgroup 6B -DS1-Zizyphus scrub [ 38 ]. The degradation stages are DS1-Dry deciduous scrub and SS4 -Dry Grasslands [ 38 ]. The vegetation was representative of a typical dry deciduous dhok forest ( Anogeissus pendula ). Apart from dhok, the species commonly found are kadaya ( Sterculia urens ), salai ( Boswellia serrata ), raunj ( Acacia leucophloea ), amaltas ( Cassia fistula ), Palash ( Butea monosperma ), tendu ( Diospyros melanoxylon ), gurjan ( Lannea coromandelica ), and Jamun ( Syzygium cumini ). Apart from humans’ landscape is shared with the striped hyena, tiger ( Panthera tigris ), leopard ( Panthera pardus ), sloth bear ( Melursus ursinus ), jackal ( Canis aureus ), fox ( Vulpes bengalensis ), and the wild ungulates, including sambar ( Rusa unicolor ), chital ( Axis axis ), nilgai ( Boselaphus tragocamelus ), chinkara ( Gazella gazelle ) and wild boar ( Sus scrofa ).
10.1371/journal.pone.0266832.g001
Fig 1
Location map of the study area and camera trap location in Sawai Mansingh Wildlife Sanctuary Rajasthan from November 2019 to March 2021.
Hyena detection data
We obtained striped hyena detection data via a camera-trap survey in different land use categories in the landscape. The landscape was characterised into five different land-use classes (forest habitat, scrubland, agriculture area, riverine habitat, and water) based on recent satellite imagery (S1 File in S1 Text , S1 Table in S1 Text ). We conducted a reconnaissance survey initially to define the extent of the hyena distribution in different land-use classes. Then, we overlaid the grid cell network of 1 × 1 km 2 over the distribution area of hyena in the area. Forty-three sampling sites for the winters of both years ( Table 1 ) were selected randomly from the grid of 52 cells, with forest = 14, scrubland = 06, agriculture = 11, riverine habitat = 12 located landscapes ( Fig 1 ). We used ten camera trap devices for sampling belonging to Cuddeback C1 type; WI, USA digital camera, (20.0 megapixel) enabled with white flash. Camera traps were enabled to take three subsequent photos bursts every time the sensor was triggered. We set the sensitivity of the camera minimum value and set traps 6-7m apart from the animal trail, so each camera had sufficient time to detect the animal and take full-frame pictures.
10.1371/journal.pone.0266832.t001
Table 1 Summary of camera trapping effort at the study site of Sawai Mansingh Wildlife Sanctuary, Rajasthan during 2019–2020 and 2020–2021.
Sampling duration
Sampled area (Km 2 )
No. of trapping station
Sampling days
Trap Night
Photo Captures
Identified hyena
Density(D)±SE/100 km 2
g0±SE
Sigma(σ)±SE (in Km)
Total
Left
Right
November 2019- March 2020
38.76 Km 2
28
34
952
125
64
61
14
12±0.03
0.04±0.01
1.46±0.18
November 2020-February 2021
36.33 km 2
15
34
510
143
73
70
14
11±0.03
0.05±0.01
1.55±0.23
The single-camera traps were installed per sampling site on either side of the road. To capture hyena’s natural behaviour, no lures or baits were used. According to Coordinated Universal Time (UTC), camera time was set as the time standard approaches and time and date stamps imprinted to each image when the camera was triggered. The sampling was conducted during winter seasons only from Nov 4 2019, to Mar 18, 2020, and Nov 21 2020, to Mar 7 2021. Camera trap stations were spaced 1 km apart from nearby traps (average trap distance = 924m). The camera traps were installed 45–60 cm above ground and operational for 34 days per sampling site. Both the animals and humans share the same space; hence, to avoid cameras having been stolen or damaged by locals, 21 camera traps (close to human habitation) were deployed at evening 18:00 and removed in the morning 07:00 daily. Despite these four-camera trap units being stolen, the data were not considered in the analysis. Instead, we collected the variables within a 200-m radius around each camera trap site. This was considered the area over which localised conditions may influence species detectability. The minimum convex polygon for season 1 and season 2 covered an area of 38.8 km 2 and 36.3 km 2 .
Density estimation
As the number of photo captures of striped hyenas was higher in left flanks, we considered it to estimate the density. Individual striped hyenas were identified from photographs obtained using the camera traps by visually examining the markings on the pelage of the hind limbs, forelimbs, and forequarters ( Fig 2 ) [ 31 ]. Photographs in which hyenas were individually identified were assigned unique identification numbers, and the specific trap location, sampling period, date, and time of capture were recorded. We constructed a capture history of striped hyenas in SECR data format for analysis for each sampling session that considered a continuous 34-days sampling occasion. Using the camera trapping data, we followed the spatial explicit capture-recapture (SECR) approach to obtain maximum likelihood density estimates for striped hyenas [ 39 ]. The likelihood SECR models were implemented package SECR V. 4.4.8 in the R and DENSITY 5.0.3 [ 40 , 41 ] ( www.Otago.ac.NZ/density ). The detection probability of each individual was modelled using the spatial detection function [ 42 ] and was explained by two parameters (one-night detection probability at the centre of an individual’s home range, [g o ] and a function of the scale of animal movements [σ]; [ 42 ]. We used a half-normal detection function because it seemed appropriate for mark-recapture data from large carnivores. We evaluated the log-likelihood function by integrating the Poisson distribution of the home range centres by adding a buffer of 10,000 m around the trapping grids (this distance was chosen to ensure that no individual outside of the buffered regions had any probability of being photographed by the camera trap during the survey [ 43 ].
10.1371/journal.pone.0266832.g002
Fig 2
Photo-capture of striped hyena in Sawai Mansingh Wildlife Sanctuary, Rajasthan.
Influence of human activity on striped hyena nocturnality
We examined the diel activity pattern of hyenas and humans in different land use categories (i.e., forest, agriculture, riverine, and scrubland) using the date and time imprint on camera trap images. We considered only independent captures taken from different stations or at least 30 min apart from the same station or depicted unambiguously different individuals in the same station [ 44 , 45 ]. The association between the hyena and humans’ activity was compared using the Chi-Square test in a different land-use category using Cramer’s V strength of association [ 46 ]. The effect size was calculated in the R package ‘effect size’ [ 47 ]. The value of association ranges from 0 (no association) to 1 (perfect association). The activity pattern and overlapping coefficient (Δ) was calculated using a non-parametric kernel density estimation method [ 48 ] for hyenas and humans in different land-use categories using the ‘overlap’ package [ 49 ] in program R v.4.0.2 [ 50 ]. The value of overlapping coefficient (Δ) ranges from 0 [no overlap] to 1 [complete overlap] [ 48 ]. Recommended by [ 49 ], Dhat 4 (Δ4) coefficient of overlap was used for a larger sample size that was more than 75 (>75), and Dhat 1 (Δ1) was used for a smaller sample size that was less than 75 (<75). We used both Δ1 and Δ4 depending on the sample size. We obtained a 95% confidence interval for activity overlap using 10000 bootstrapping iterations. After that, we calculated the activity pattern of a striped hyena at each terrain category using a non-parametric kernel density estimation method [ 48 ] in the ’overlap’ package [ 49 ] using program R v.4.0.2 [ 50 ].
Spatial use of the landscape by hyena
To assess the effects of landscape and anthropological variables on the detection of hyenas at camera trap stations, we used a Generalised Linear Model (GLM) because it fits the count data well [ 51 ]. We used seven environmental variables at every camera trap within the grid, i.e., distance from the human settlement (m), aspects (degrees), slope (degrees), scrubland (%), forest cover (%), and riverine habitat (%) and distance from water (m) (S1 File in S1 Text ). A 200m buffer was laid using the “buffer” feature of “Proximity” under “Analysis tools” of ArcMap v.10.2.2 (Esri 2014) to calculate the habitat parameters at each sampling point. We also calculated the Relative abundance Indices (RAI) for humans per site as covariates. RAI was calculated as RAI = E/TNx100, where E is the number of events (photo-captures), and TN is the total number of trap nights [ 52 ]. We used the RAI of a hyena (i.e., detection rate) per camera trap station as a response variable, while landscape and anthropogenic variables as a predictor variable. The multicollinearity among variables was examined using IBM SPSS Statistics (ver. 21.0; SPSS Inc., Chicago, IL., USA, variables with VIF (variance inflation factor) <3 were included in the analysis [ 53 ]. VIF ranged from 1.3 to 2.8 for all variables (S2 Table in S1 Text ). Therefore, all variables were retained in the modelling. We used Poisson distribution with log link function [ 54 ]. The list of all possible models was created to examine the relationship between our response variable and predicted variables. We considered the final Model with ΔAIC < 2 using the ‘dredge’ function of package ‘MuMIn’ in program R [ 55 ]. Model selection was based on Akaike’s information criterion and Akaike weights [ 56 ]. We averaged the parameter coefficients of all models with a cumulative Akaike weight > 0.9 [ 57 , 58 ]. All analysis was performed in program R v.4.0.2 [ 50 ]—data analysis codes used in program R and data given in S1 and S2 Appendices .
People’s perception of hyenas
We interviewed 200 random people selected from our study area. Respondents were questioned about their occupation, types of crops, conflicting species, hyena’s role in the ecosystem, conflict with the hyena, and attitude towards striped hyena (S2 File in S1 Text ). We also asked about livestock mortality rate, mortality reasons, and the process of livestock dumping after death. Furthermore, the livestock number was collected from livestock census data collected by the government for each village in the year 2019–2020 (S3 Table in S1 Text ).
Results
Striped hyena density
A total camera trapping effort of 1462 days captured 28 unique striped hyena individuals spanning two years ( Table 1 ). The number of individual striped hyena captures did not differ between years, although the highest number was captured during the winter season of 2020. The RAI of hyenas and humans was calculated at 18 captures/100 trap days and 139/100 trap days consequently (S4 Table in S1 Text ). The RAI of striped hyena was recorded higher in the forest followed by scrubland, riverine, agricultural land, while the human activity was recorded higher in the agricultural area followed by riverine, forest and scrubland ( Fig 3 ). The presence of striped hyenas was least associated with humans in the landscape (r = 0.40, p = <0.05). We recorded the highest striped hyena density for both years in the landscape. The hyena density for each season was estimated at 12±0.03 individuals/100 km 2 and 11±0.03 individuals/100 km 2 , respectively. However, the animal movements from the centre of the home range (σ) for were 1.46 (SE = 0.18) km and 1.55 (SE = 0.23) km. Consequently.
10.1371/journal.pone.0266832.g003
Fig 3
Relative abundance indices (RAI) of hyena in different land-use categories in human dominated landscape of Sawai Mansingh Wildlife Sanctuary, Rajasthan.
Influence of human activity on striped hyena nocturnality
Striped hyenas were crepuscular and nocturnal, showing bimodal peak activity ( Fig 4A ). The activity of hyenas was reduced during the daytime when human activities were at their peak ( Fig 4A ). In all land-use patterns, hyena activity was crepuscular and nocturnal with a bimodal peak of activity. However, there was no sign of activity during the daytime when human activity was high ( Fig 4B–4E ). The overall activity overlap between humans and hyenas [Δ4 = 0.30, CI = 0.29–0.37, Fig 4A ]. The highest activity overlap was observed in the agricultural area [Δ1 = 0.39, CI = 0.28–0.49, Fig 4B ], followed by riverine area [Δ1 = 0.25, CI = 0.28–0.44, Fig 4C ], scrubland [Δ1 = 0.23, CI = 0.006–0.26, Fig 4D ], and forest [Δ4 = 0.19, CI = 0.16–0.27, Fig 4E ]. We recorded no major difference in the activity of striped hyenas in both rugged and flat terrain ( Fig 4F and 4G ). The overall temporal association between human and striped hyenas was calculated at 0.46 (CI = 0.38–0.53, p = <0.05). We observed the highest association between striped hyena and human in agricultural area (0.92 [CI = 0.75–0.99], p = <0.05) followed by riverine area (0.75 [0.63–0.86], p = <0.05), forest (0.66 [0.58–0.73], p = <0.05) and scrubland (0.63 [042–0.82], p = <0.05).
10.1371/journal.pone.0266832.g004
Fig 4
The kernel density function of activity and overlap plot of striped hyena and human in different land-use categories (a) along with human (b) agricultural area (c) riverine area (d) scrubland (e) forest (f) rugged terrain (g) flat terrain, in the human-dominated landscape of Sawai Mansingh Wildlife Sanctuary, Rajasthan.
Spatial use of the landscape by hyena
In our analysis, we included all predictor variables: the top three models’ performance with less than an ΔAIC < 2, explained the site used by hyena and selected the top model that had Akaike weights 0.22 ( Table 2 ). The coefficient of predictor variables was generated using model-averaged. The predictor variable, including distance from human settlement, scrubland area, distance from the water, was the best predictor for striped hyena on-site use ( Table 3 ). The distance from the human settlement (β = 0.28, p = <0.05) variable was positively associated with the detection of hyena predicted the detection rate of striped hyena increased with decreasing distance from the human settlement ( Fig 5A ). While the scrubland (β = -0.01, p = <0.05) and water availability (β = -0.16, p = <0.05) were negatively associated with the detection of hyena ( Table 3 ) ( Fig 5B and 5C ). The striped hyena capture rate decreased with increasing distance to water ( Fig 5C ). However, the aspect (β = -0.00, p = >0.05), slope (β = -0.005, p = >0.05), riverine (β = 0.001, p = >0.05) and forest (β = -0.00, p = >0.05) had no significant effect on detection of hyena ( Fig 5D–5G ). Human presence had no significant effect on detection of hyena (β = -0.00, p = >0.05) ( Fig 5H ).
10.1371/journal.pone.0266832.g005
Fig 5
The spatial relationship between the striped hyena and environmental variables (a) hyena vs distance to human settlements or village (km), (b) hyena vs scrubland, (c) hyena vs distance to water (km), (d) hyena vs aspect, (e) hyena vs slope (f) hyena vs riverine area, (g) hyena vs forest cover (h) hyena vs human presence, in the human-dominated landscape of Sawai Mansingh wildlife sanctuary, Rajasthan.
10.1371/journal.pone.0266832.t002
Table 2 Result of Generalised Linear Model used to evaluate the environmental and anthropogenic variables on-site use by striped hyena in the human-dominated landscape of Sawai Mansingh wildlife sanctuary, Rajasthan. Top candidate models predicting the habitat selection of striped hyaena in the landscape in Sawai Mansingh Wildlife Sanctuary, Rajasthan, India; aspect; area of riverine (arrive); area of scrubland (arscrb); slope; distance from the human settlement (hsdist); distance from water (watdist); area of forest cover (arfcm); human RAI (human).
Covariates
Degree of freedom
log link
AICc
ΔAIC
Weightage of Model
aspect+ arrive + arscrb + hsdist + watdist
6
-478.45
971.23
0.00
0.22
aspect+ arrive + arscrb +hsdist+ slope+ watdist
7
-477.78
972.75
1.53
0.10
aspect+ arscrb +hsdist+ watdist
5
-480.63
972.87
1.65
0.10
10.1371/journal.pone.0266832.t003
Table 3 GLM model average coefficient (β) with standard error values (SE) of the variables to explain the site use by striped hyena in the sampling area. * = statistically significant at P ≤0.05.
Covariates
Coefficient (β)
SE (coefficient)
P-value
Intercept
2.894
0.146
0.00***
Aspect
-0.0005
0.00
0.26
Area of riverine habitat
0.0001
0.001
0.28
Area of scrubland
-0.017
0.003
0.00***
Distance from human settlement
0.280
0.073
0.0001***
Distance from the water body
-0.161
0.037
0.00***
Slope
-0.006
0.012
0.62
Forest cover
-0.0001
0.001
0.89
Human presence
-0.00
0.00
0.93
Local’s perception and livestock density
The people reported no conflict with striped hyenas in and around SMS WLS. Most people (63%) had a positive attitude, while 25.5% had a negative and 11.5% had a neutral attitude toward hyena. A total of 78% of people considered hyena’s role in the ecosystem as cleaning off carrion, while 17.5% people considered predation of livestock and 04% people were not aware of any significant role played by a hyena. In the study area, every month, an average of 40–50 livestock died due to diseases (26%), natural death (36%) or predation by carnivores (13.5%), and carcasses were dumped ~1–2 km distance away from their villages into dumping ground. As per the animal husbandry records, there were 21,272 livestock heads in 22 villages within 75 km 2 . The estimated livestock density in the study area was 283.62 animals/km 2 .
Discussion
Striped hyena density
Our results suggest that low-risk conflict species and the scavenging nature of hyena allow them to coexist with humans in the landscape [ 30 ]. The density estimates of hyena in our study area were higher than in the adjoining protected area Ranthambhore National Park (5.49 individuals/100 km 2 ) [ 31 ], but lower than previous studies in similar habitat in Sariska Tiger Reserve (STR) (15.1 individuals/100 km 2 ) [ 59 ]. However, the hyena densities in the outside protected area in India range from 3.67 to 5.03 individuals/100 km 2 [ 30 , 60 ]; whereas estimates for protected areas (i.e., Rajaji National Park and Gir National Park) were 3.91, 6.50 individuals/100 km 2 , respectively in India [ 61 , 62 ]. Several ecological factors may be driving carnivore densities, with prey abundance among the important factors [ 63 , 64 ]. STR had the highest wild (107 animals/ km 2 ) and domestic (222 animals/km 2 ) prey densities [ 65 , 66 ], which may be the reason for the high density of hyenas. While in Israel density of striped hyenas was higher away from the agriculture and human settlement [ 67 ]. In our study area, there are 75 villages located within the five km of boundary of SMS WLS, and the livelihood of locals is primarily dependent on agriculture, and their economy is supplemented by animal husbandry. In our study area, livestock density was estimated to be 283.62 animals/km 2 , and it was observed that 40–50 livestock/month died due to various causes (i.e., starvation, inadequate veterinary care, depredation by tiger or leopard, disease, poor sanitation; personal observation), which were not buried or consumed by locals’ dues to their religious views [ 68 ]. The availability of livestock carcasses likely provides subsidised scavenged food sources for striped hyenas in the landscape and lower competition from other predators. Being a specialised scavenger, it was implicit that the striped hyena’s distribution and density might be related to livestock abundance (i.e., carcasses) [ 30 ]. However, studies suggested human impact caused low density of striped hyena in East Africa [ 69 ]. While in South Africa, the high densities of brown hyenas were observed near cattle farms compared to neighbouring protected areas [ 70 ]. Hence the shared landscape provides scavenging opportunities for hyenas, suggesting the higher densities.
Influence of human activity on striped hyena nocturnality
Temporal avoidance of humans may ease human-hyena coexistence in the shared landscape. Hyenas exhibited nocturnal behaviour in this study. The nocturnal activity of hyenas has been interpreted as a response to human activity, having high human activity during the daytime. We observed that the hyena activity was the crepuscular onset of activity to occur around sunset around 1800 hrs. in comparison, another set of activities occurred around early in the morning at 0600 hrs. In Israel, the striped hyena’s activity was affected by high human activities near agricultural areas [ 71 , 72 ].
Similarly, our result also revealed the high activity overlapped and high temporal association with humans and hyenas in the agriculture area. In our study area, winter crops (i.e., mustard, chilli, wheat, maze, etc.) were guarded by peoples for protection from wild ungulates (i.e., nilgai and wild boar), and crops provide the cover to hyena for movement in the landscape; hence the overlaps may be expected in agriculture area. While the villages near riverine and scrubland habitats provide abundant food for hyenas, it was common for peoples to discard livestock carcasses and leave a substantive, easily attainable food for hyenas to exploit. People used these habitats for grazing livestock during the daytime, and hyenas utilised areas when people were least likely to be active, i.e., 1800–1900 hrs; hence the activity of hyena was increased during the crepuscular period to benefits of accessing resource subsidies. Likewise, many carnivores (i.e., lion, tiger, wolf, spotted hyenas, brown hyenas) adapt in human-dominated areas to Spatio-temporal avoidance of humans via shifting the activity timing to a preference for nocturnality [ 9 , 25 , 73 , 74 ]. The hyena activities were minimally overlapped during 0600 hours to 1200 hours with humans in forest habitat. Due to the forest department restrictions, the human activities are minimal in the forested area; however, the locals illegally graze the cattle and collect firewood for cooking purpose, while hyena uses the forest area for daytime resting and denning purposes [ 31 ]. Hyenas are vulnerable to being killed by people and predation by feral dogs. Therefore, they generally avoid human interactions, especially during daylight hours [ 30 , 75 ].
Spatial use of the landscape by hyena
Our results are consistent with the general perception that in India, the hyena is recognised as low-risk species compared to other large carnivores and can coexist with humans in a shared landscape [ 22 ]. Our results are similar to other studies on habitat selection of hyenas from India [ 30 , 31 , 59 ], Nepal [ 76 ], and Africa [ 20 ], which found the distance from the human settlement as the significant indicator for hyena detection due to foraging opportunities [ 30 , 31 , 77 ]. Hyenas feed easily on domestic waste in slaughterhouses, garbage dumps, livestock carcasses, and poultry farms [ 30 , 78 , 79 ]. In our study area, due to animal husbandry practices of peoples, generates abundant anthropogenic food (i.e., livestock carcasses). Hyenas likely get most food from scavenging and readily use anthropogenic food in the landscape; hence the detection rate of hyenas is predicted to be high near human settlements. It is often reported that large carnivores (i.e., tiger, leopard, lion) predate on livestock and attack humans, which often leads to conflict [ 80 ]. Previous studies reported that the attitude of positive or neutral towards the species is the key factor of coexistence between humans and wildlife [ 81 ]. In our study area, people’s attitude towards hyenas was positive, and locals reported no conflict and active livestock predation as they considered the hyena a scavenger. Previous studies reported forest cover and scrubland as important factors for hyena detection [ 31 , 76 ]. We found that these habitat variables were negatively associated with hyena site use. Scrublands are treated as ‘wastelands’, and people diverted them for commercial use and converted them into agriculture [ 82 ]. Hyena prefers open or thorn habitats in arid and semiarid environments [ 32 , 76 , 83 ], and the presence of the hyena may have been associated with such habitats because hyenas used scrubland for movement. Habitat use studies indicate the distance to the water body and riverine habitat found important factors for site use by hyenas [ 76 ]. Our results generally agree with this pattern in the area both humans and hyenas temporally used the same water resources. The rugged and gullied terrain of riverine habitat provides a suitable denning site to hyenas and easy access to the water body from the nearby Chambal river. The riverine habitats are barren areas which are alluvium deposited by the Chambal river itself [ 84 ], have less interest to villagers but are used by different faunal species for denning. A previous study suggested the rugged terrain in the landscape provides disturbance-free denning refugia (i.e., not used by livestock, humans, or guard dogs) of hyenas may provide optimal conditions for breeding and raising pups [ 30 , 31 ]. Our study demonstrates that the scrubland, agricultural lands, and riverine habitats may serve as supplementary habitats for hyenas. The coexistence of hyenas and humans in the shared landscape is supported by mutual benefits, where hyenas get benefits from food and humans’ benefit from waste removal. Thus, sharing landscape in a human-dominated landscape without negatively impacting each other is a possible key factor of human-wildlife coexistence [ 81 ].
Supporting information
S1 Appendix
Data analysis codes of R used in the paper.
(DOCX)
S2 Appendix
Data used for analysis.
(XLSX)
S1 Text
(DOCX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Pancreatic islets in mouse embryos are developed between embryonic day (E) 13.5 and 15.5 [1] , [2] . After birth, the number of islet cells is determined by the balance of cell renewal and cell loss [3] . The dynamic change of islet cells number is essential in maintaining euglycemia and thus it is important to understand how islet cells balance in the adult pancreas is achieved, specifically the mechanisms involved in stimulating pancreatic islet cells growth and preventing pancreatic islet cells from apoptosis. The growth rate of pancreatic islet cells is normally low but changes in response to different stimuli. The cell death or apoptosis is also an important factor in maintaining the appropriate number of islet cells. Increasing reactive oxygen species (ROS) concentration is one of the major causes to induce apoptosis of cells. It is continuously derived from glucose metabolism and cannot be effectively eliminated by endogenous antioxidant enzymes. Pancreatic islet cells undergo apoptosis in either physiological or pathological conditions. There is also an evidence that pre-existing β-cells are the major source of new β-cells during adult life span and after pancreatectomy in mice [4] . However, the exact mechanisms involved in the regulation of these processes are not yet clarified. In particular, the factors involved in these important physiological or pathological conditions are not fully identified.
Insulin gene enhancer binding protein-1 (ISL1), which belongs to LIM homeobox gene family, was first discovered and cloned in 1990 [5] . ISL1 is mainly expressed in adult islet endocrine cells (α, β, γ, ε) as well in the central nervous system [6] , [7] . As a key transcription factor, the functions of ISL1 involve cell fate specification and embryonic development. In pancreas its function is twofold: 1) control the four endocrine islet cell lineages development and 2) control of dorsal pancreas mesenchyme development. Complete loss of dorsal pancreatic mesenchyme and endocrine islet cells was found in ISL1 knock-out mice embryos. It has also been shown that ISL1 could regulate the expression of several islet specific genes, such as proglucagon/glucagon (α cells), somatostatin (γ cells), amylin (δ cells) and insulin (β cells), although it is not the master regulator for these genes [8] , [9] , [10] , [11] . However, it is not clear whether ISL1 plays more important roles rather than the regulation of these endocrine hormones secretion in postnatal pancreatic islets. Recent studies demonstrate that ISL1 is required for proliferation, migration and survival of cardiac progenitor cells [12] . It also promotes proliferation and repairing of injured motor neurons [13] , [14] . Overexpressing ISL1 in endothelial cells and mesenchymal stem cells can promote blood vessel formation [15] . ISL1 is also involved in the establishment of pancreatic endocrine cells during the secondary transition (E13.5–E15.5) and controls the proliferation and survival of endocrine cells during embryonic islet developmental stage [16] . However, the roles of ISL1 in adult islets are yet not clear.
Based on reports that ISL1 can promote some types of cell proliferation as mentioned above, we designed this study in order to investigate whether ISL1 plays roles in maintaining the balance of islet cell renewal and cell loss. We show that ISL1 can promote adult pancreatic islet cells proliferation and attenuate cell apoptosis against oxidative stress. The mechanism involving ISL1 in promoting adult pancreatic islet cells proliferation includes the direct activation the cell autonomous factors c-Myc and CyclinD1. Our findings advance our understanding of the roles of ISL1 in adult pancreas and provide insights into the regulation of adult pancreatic cell proliferation.
Results
ISL1 was highly expressed in adult pancreatic islets
The important role of ISL1 in pancreatic islet development has been well established. However, its role in adult pancreatic islets remains unclear. The high level of ISL1 expression in adult pancreatic islets indicates that ISL1 must play important tissue specific roles. We previously reported that ISL1 enhances the transcriptional activation of the insulin gene in vitro [17] . In order to explore the biological functions of ISL1 in adult pancreatic islets, we detected the expression of ISL1 in different diabetes animal models. To our surprise, the increasing expression of ISL1, compared to normal control mice, was detected in all diabetes models, which is not parallel to the expression of insulin ( Fig. 1A ). The results indicate that ISL1 may play other roles in diabetic mice other than regulating insulin.
10.1371/journal.pone.0022387.g001
Figure 1
ISL1 was highly expressed in adult pancreatic islets and could reduce apoptosis in HIT-T15 cells.
(A) Relative mRNA expression level of insulin in pancreatic islets from STZ mice (n = 6), Akita mice (n = 5) and db/db mice (n = 6) were measured by real-time RT-PCR. C57BL/6 mice (n = 6) were used as controls to STZ mice and Akita mice, db/w mice (n = 6) were as controls to db/db mice. Each bar represents mean ± SD (**p<0.01, *p<0.05, vs. the controls). (B) Level of ROS production was measured by flow cytometry analysis in HIT-T15 cells treated with glucose oxidase (GO) at various concentrations (0–100 mU/mL) for 4 h. (C) Level of apoptosis rate was measured by flow cytometry analysis in HIT-T15 treated with GO at various concentrations (0–100 mU/mL) for 4 h. (D) Relative level of ISL1 mRNA expression in HIT-T15 cells treated with different GO concentrations was examined by real-time RT-PCR. (E) Level of apoptosis rate was measured by flow cytometry analysis in stable ISL1 knockdown HIT-T15 cells treated with or without 5 mU/mL GO. Each bar represents mean ± SD from three samples (**p<0.01, vs. the control).
ISL1 reduced apoptosis in pancreatic HIT-T15 cells
It has been reported that ISL-1 controls the proliferation and survival of endocrine cells in postnatal pancreas [16] . Considering that pancreatic islet cells are challenged by continuous oxidative stress derived from the glucose metabolism throughout life, we suspected that ISL1 plays a role in anti-oxidative stress mechanism. We used glucose oxidase (GO) to generate H 2 O 2 from glucose, mimicking oxidative stress in HIT-T15 (a pancreatic β-cell line) cells. The flow cytometry (FCM) results showed that GO strongly promoted the production of ROS ( Fig. 1B and Fig. S1 ) and increased the level of apoptosis ( Fig. 1C and Fig. S2 ), in a dose-dependent manner (0–100 mU/mL). Real-time RT-PCR results showed that GO also dynamically changed the level of ISL1 expression, with a peak at 5 mU/mL of GO ( Fig. 1D ).
To further confirm the anti-apoptosis role of ISL1, ISL1-specific siRNA (ISL1-siRNA) was designed and transferred into HIT-T15 cells. The expression of ISL1 at both mRNA and protein levels was significantly reduced by ISL1-siRNA compared to non-silence siRNA (data not shown). The FCM result showed that knockdown of ISL1 could increase HIT-T15 cells apoptosis three folds regardless GO stimulation ( Fig. 1E ). These results implied that ISL1 could protect cells against apoptosis under physiological or oxidative stress conditions.
ISL1 promoted pancreatic islet cells proliferation
To further define whether ISL1 plays a role in adult islets, islet mass were isolated from adult Sprague-Dawley (SD) rats and infected with ISL1 overexpressing lentivirus or ISL1-siRNA lentivirus (multiplicity of infection, MOI = 10). The infection efficiency reached approximately 53% and 66%, respectively ( Fig. 2A ). Real-time PCR and Western blotting results showed that the expression level of ISL1 was ameliorated ten folds in ISL1 overexpressed islet cells ( Fig. 2B, 2C ) and was attenuated to 30% ( Fig. 2D, 2E ) in ISL1 knockdown islet cells, which provided a model for further study.
10.1371/journal.pone.0022387.g002
Figure 2
The expression of ISL1 was altered in adult islet mass by ISL1 overexpression or knockdown.
(A) Infection efficiency (as indicated by the percentage of GFP positive cells in gated cells) of ISL1 overexpression lentivirus or ISL1-siRNA lentivirus were detected by flow cytometry analysis after 72 h infection. Real-time RT-PCR (B, D) and Western blotting (C, E) results showed the expression level of ISL1 in ISL1 overexpressed (B and C) islet cells and in ISL1 knockdown (D and E) islet cells. Data represent 3 independent experiments, each performed in triplicate. Each bar represents mean ± SD (**p<0.01, vs. the control). Lentivirus without any insert was used as a control.
To examine the impact of ISL1 on the proliferation of adult pancreatic cells, the cell cycle profile was analyzed using propidium iodide staining and flow cytometry. Compared with the control (infected with control lentivirus), ISL1 overexpression was associated with a decreased cell population in G 0 /G 1 phases (from 76.27±1.17% to 66.72±1.62%) and an increased cell population in the G 2 /M and S phases ( Fig. 3A ). Conversely, adult islet mass exposed to ISL1-siRNA lentivirus exhibited an increase in the proportion of cells in G 1 phase (from 76.76±0.67% to 82.74±0.92%) and a decrease in the proportion of cells in G 2 /M and S phases ( Fig. 3B ). These data indicate that ISL1 plays a role in promoting adult pancreatic cells proliferation.
10.1371/journal.pone.0022387.g003
Figure 3
ISL1 promoted proliferation of adult pancreatic cells.
Adult pancreatic islet cells infected with ISL1 lentivirus (A) or ISL1-siRNA lentivirus (B) were subjected to cell cycle analysis by flow cytometry. The data represent 3 independent experiments. The representative cytometric results from these experiments are shown. EdU incorporation assay was analyzed by confocal microscopy (scale bar, 50 µm; scale bar in magnified field, 10 µm) in adult islets infected with ISL1 lentivirus (C) or ISL1-siRNA lentivirus (D). (E) The EdU incorporation rate was expressed as the ratio of EdU positive cells to total Hoechst33342 positive cells. Each bar represents mean ± SD from 3 samples (**p<0.01, vs. the control).
Subsequently, we employed the EdU incorporation assay, a more sensitive and specific method [18] , [19] , to further define the function of ISL1 in promoting cell proliferation. The number of EdU positive cells was increased by 2.5 folds in ISL1 overexpressing adult islets cells relative to control cells ( Fig. 3C, 3E ). More importantly, the number of EdU positive cells in ISL1-siRNA lentivirus-infected cells was reduced by 40% relative to that of the cells infected with non-silencer siRNA lentivirus ( Fig. 3D, 3E ). These results indicate that the knockdown of ISL1 could inhibit the proliferation of adult islets in vivo , while the overexpression of ISL1 promotes adult pancreatic islet cells proliferation.
We also constructed a stable ISL1 overexpressing HIT-T15 cell line with the pcDNA3.1-ISL1 expression plasmid and a stable ISL1 knockdown HIT-T15 cell line with ISL1-siRNA. RT-PCR and Western blotting results showed that both overexpression ( Fig. 4A ) and knockdown ( Fig. 4B ) were established successfully in cell lines. Then, cell proliferation was determined by CCK-8 analysis. As shown in Fig. 4C , stable transfection of pcDNA3.1-ISL1 expression plasmid promoted the proliferation of HIT-T15 cells three folds relative to that of with pcDNA3.1 (control) after 72 h culture. As expected, the knockdown of ISL1 inhibited cells growth by 20% compared to inhibition with non-silencer siRNA after 48 h culture ( Fig. 4D ). EdU assay also showed more EdU positive cells in ISL1 overexpressing cells ( Fig. 4G ) and less EdU positive cells in ISL1 knockdown ( Fig. 4H ) cells, indicating that ISL1 promoted cell proliferation. These results are well in agreement with cell-cycle analysis that showed a decrease (from 75.94±1.45% to 63.78±1.76%) of the percentage of cells in G 0 /G 1 phase and an increase (from 24.05±1.45% to 36.21±1.76% ) in the G 2 /M and S phases in ISL1 stable HIT-T15 cells ( Fig. 4E ). In contrast, ISL1 knockdown increased cell population in G 0 /G 1 phase from 76.54±0.28% to 86.27±0.56% and decreased cell population in the G 2 /M and S phases from 23.45±0.28% to 13.73±0.56% ( Fig. 4F ). Colony formation assays revealed that ISL1 overexpressing cells resulted in a significant increase in colony number compared with the control cells; while ISL1 knockdown resulted in a significant decrease in colony number ( Fig. 4I ). These results further confirm that ISL1 promotes pancreatic islet cells proliferation.
10.1371/journal.pone.0022387.g004
Figure 4
ISL1 promoted proliferation of HIT-T15 cells.
The expression of ISL1 in stable overexpression (A) or knockdown (B) HIT-T15 cell lines was examined by RT-PCR and Western blotting. The cell proliferation was determined by CCK-8 analysis. A total of 1×10 3 cells (either stably overexpressing (C) ISL1 or stably knockdown (D) ISL1) per well were seeded in 96-well plate and measured for their proliferation after 12 h, 24 h, 48 h, and 72 h. The data represent 3 independent experiments, each performed in triplicate. Each bar represents mean ± SD (**p<0.01, *p<0.05, vs. the control). The cell cycle profile was analyzed by flow cytometry in ISL1 (E) and ISL1-siRNA (F) stable HIT-T15 cell lines. EdU incorporation was detected by confocal microscopy (scale bar, 50 µm) in ISL1 (G) and ISL1-siRNA (H) stable cells. (I) Stable cells were maintained in G418 or puromycin-containing medium for 21 days before staining with crystal violet and counting for colony numbers. Each bar represents mean ± SD from 3 samples (**p<0.01, *p<0.05, vs. the control).
ISL1 stimulated islet cells proliferation through the regulation of cell cycle regulators
Several cell cycle regulators, including CyclinD1, c-Myc, and CDK4 have been shown to control pancreatic islet cell proliferation in vivo [20] , [21] . To identify the mechanism of ISL1-stimulated pancreatic islet cell proliferation, we tested whether the expression of these established cell cycle regulators was controlled by ISL1. HIT-T15 cells stably expressing ISL1 were subjected to real-time RT-PCR analysis for the expression of c-Myc, CyclinD1, CyclinA and p53. The results indicated that the overexpression of ISL1 did not significantly affect the expression levels of CyclinA and p53, but led to a three-fold increase in CyclinD1 expression and approximately two folds increase in c-Myc expression, relative to the control ( Fig. 5A, 5B ). Consistently, the knockdown of ISL1 did not alter the expression level of CyclinA, but was associated with the decrease in CyclinD1 expression level by 60% and in c-Myc expression level by 50% as compared with that in the cells transfected with non-silencer siRNA ( Fig. 5C, 5D ). Interestingly, the knockdown of ISL1 significantly enhanced the expression of p53 ( Fig. 5C ). To investigate the biological significance of ISL1-promoted cell proliferation, we then assessed the trans-activation activity of ISL1 on c-Myc or CyclinD1 promoters. In these experiments, HIT-T15 cells with ISL1 overexpression were transfected with a luciferase reporter construct: firefly luciferase gene inserted with c-Myc or CyclinD1 promoter. The results showed that the activation of c-Myc or CyclinD1 promoter exhibited a dose dependent manner within 1 µg of ISL1 with a constant amount of 0.2 µg c-Myc ( Fig. 5E ) or 1 µg of ISL1 with a constant amount of 0.2 µg CyclinD1 ( Fig. 5F ). These results indicate that ISL1 is able to activate the c-Myc and CyclinD1 promoters efficiently. Taken together, ISL1 promotes pancreatic islet cells proliferation possibly through the activation of the c-Myc and CyclinD1 promoters and thus increasing the expression of c-Myc and CyclinD1.
10.1371/journal.pone.0022387.g005
Figure 5
ISL1 promoted the expressions of c-Myc and CyclinD1.
The mRNA level (A, C) and protein level (B, D) of c-Myc and CyclinD1, Cyclin A, p53 were analyzed by real-time RT-PCR and Western blotting in ISL1 stable overexpression HIT-T15 cells (A and B) and ISL1 knockdown HIT-T15 cells (C and D). The transcriptional activity of ISL1 was analyzed by luciferase reporter assay. ISL1 activated the promoter of c-Myc (E) or CyclinD1 (F) in a dose-dependent manner. The data represent 3 independent experiments, each performed in triplicate. Each bar represents mean ± SD (**p<0.01, *p<0.05, vs. the control).
ISL1 activated CyclinD1 or c-Myc transcription by binding to an evolutionarily conserved site
We have shown that ISL1 could act as a transcriptional activator of c-Myc or CyclinD1. It is unknown whether ISL1 could directly control c-Myc or CyclinD1 transcription. Bioinformatic analysis with MatInspector software revealed a conserved ISL1 binding sequence (TAAT) 645 bp upstream of the ATG translation start site on the c-Myc promoter ( Fig. 6A ) and 684 bp upstream of the ATG translation start site on the CyclinD1 promoter ( Fig. 6B ). A region covering the ISL1 binding site located between −645 and −650 bp in mouse c-Myc promoter ( Fig. 6A Left) or at −684 to −689 bp in mouse CyclinD1 promoter ( Fig. 6B Left) was amplified using PCR and was used as a probe for subsequent electrophoretic mobility shift assays (EMSA). The results showed that a specific complex was formed with c-Myc ( Fig. 6C , lane 3) or CyclinD1 ( Fig. 6D , lane 3) probe. The complex specificity was confirmed by performing incubation with a 100-fold molar excess of unlabeled oligonucleotide prior to the addition of the labeled probes ( Fig. 6C , lanes 4–5 and Fig. 6D , lanes 4–5). The mutation of ISL1 binding site on c-Myc or CyclinD1 probe failed to compete with the complex formation ( Fig. 6C , lanes 6–7 and Fig. 6D , lanes 6–7). The addition of normal rabbit IgG, mouse IgG and the unrelated antibody did not affect the protein-DNA complex formation ( Fig. 6C , lane 8 and Fig. 6D , lanes 8–10), while the complex bands were dramatically attenuated in the presence of anti-ISL1 antibody, further proving the specificity of the protein-DNA complex ( Fig. 6C , lane 9 and Fig. 6D , lane 11).
10.1371/journal.pone.0022387.g006
Figure 6
ISL1 band on the c-Myc or Cyclin D1 promoter directly.
Consensus binding site (TAAT, box) for ISL1 on the c-Myc (A) or CyclinD1 (B) promoter was analyzed by Matinspector software. (C and D) Nuclear extracts were subjected to EMSA for the ISL1 proteins binding ability to the 32 P-labeled oligonucleotides containing the consensus sequence of the c-Myc or CyclinD1 promoter. (C) Lane 1, free probe. Lane 2, nuclear extracts from HIT-T15 cell without transfection. Lanes 3–9, nuclear extracts from HIT-T15 cells transfected with ISL1 expression construct. Lane 3 shows the direct binding of ISL1. Lanes 4 and 5 show wild-type unlabeled oligonucleotide competition. Lanes 6 and 7 show mutant unlabeled probe competition. Lane 8, the mouse normal IgG was used as a negative control. Lane 9, 1 µg anti-ISL1 antibody (H00003670-M05, Abnova) was added. (D) Lanes 1–7, same as C, Lane 8, rabbit normal IgG was used as a negative control. Lane 9, 1 µg unrelated rabbit anti-GATA4 was added. Lane 10, mouse normal IgG was used as a negative control. Lane 11, 1 µg mouse anti-ISL1 polyclonal antibody was added. (E and F): ISL1 recruited on the c-Myc or Cyclin D1 promoter was analyzed by ChIP assay. Soluble chromatin was prepared from HIT-T15 cells stably transfected with 2 µg of pcDNA3.1-ISL1 plasmid followed by immunoprecipitation with antibodies against ISL1. The DNA extractions were amplified using the primers that cover the ISL1 binding sites on the c-Myc or Cyclin D1 promoter by PCR (E) or real-time PCR (F) with normal IgG as a control. The data represent 3 independent experiments, each performed in triplicate. Each bar represents mean ± SD (**p<0.01, *p<0.05, vs. the control).
Promoter chromatin immunoprecipitation (ChIP) assay was performed in HIT-T15 cells to determine if ISL1 could occupy the c-Myc promoter or CyclinD1 promoter region in vivo . Results show that the ISL1 antibody specifically immunoprecipitated with the DNA fragments containing the c-Myc promoter or CyclinD1 promoter in ISL1 overexpressing cells. As shown in Fig. 6E , the overexpression of ISL1 significantly accentuated the binding of ISL1 on the c-Myc or CyclinD1 promoter compared to IgG, suggesting that ISL1 could bind on the c-Myc or CyclinD1 promoter in vivo . As shown in Fig. 6F , ISL1 was recruited to the Cyclin D1 promoter four folds as compared with lgG, whereas the binding level on the c-Myc promoter was less than 1.5 folds relative to IgG. Collectively, these in vitro and endogenous binding data indicate that ISL1 is a direct regulator of c-Myc and CyclinD1 transcription in pancreatic β cells.
Discussion
ISL1 is a LIM-homeodomain transcription factor and plays important roles in the early development of heart, motor neurons and pancreas at embryonic stage. The expression of ISL1 is down-regulated in heart cells after birth but remains at a high level in normal adult islet cells, indicating that ISL1 may have some functions in terminally differentiated islets. Previous studies provided evidences that ISL1 is required for proliferation of human cardiac progenitor cells and motor neurons in injured zebrafish [13] , [14] . Furthermore, Du A et al. [16] demonstrated the requirement of ISL1 in the maturation, proliferation, and survival of hormone-producing islet cells after the secondary transition and in postnatal lives, indicating a crucial role of ISL1 in regulating endocrine cell growth and survival in young animals [16] . However, whether ISL1 can prevent islet cells from apoptosis or promote mature islet cell proliferation are still questions to be answered.
Our results show that ISL1 knockdown in HIT-T15 cells could increase the level of apoptosis even in normal cell culture condition. However, the level of apoptosis was dramatically increased in response to the glucose oxidase stimulation, suggesting that ISL1 may play roles in anti-apoptosis under the oxidative stress. Moreover, our study demonstrated that ISL1 could promote mature pancreatic islet cell proliferation and identified several downstream targets for ISL1 in the islets. It was reported that ISL1 could directly bind to MafA in the early stage of pancreas development [16] , but its direct downstream targets in adult pancreatic islets were uncharacterized. As the adult islet cell proliferation is a complex process and requires a dynamic change to meet the requirement of pancreatic function, the regulation of the islet cell proliferation must result from the orchestration of many transcriptional factors and growth factors. CyclinD1, CyclinD2, CDK4, and c-Myc, amongst others, have been reported to control the islet cell proliferation during pancreas development [20] , [21] . We investigated whether the expression of these established cell cycle regulators are controlled by ISL1. As a transcription factor, ISL1 exerts its function by binding to a consensus sequence TAAT on the promoter of target genes. Our promoter chromatin immunoprecipitation (ChIP) assay results revealed that ISL1 could regulate the transcription of c-Myc and CyclinD1 by directly binding on the c-Myc or CyclinD1 promoter in vivo . Furthermore, the luciferase assays demonstrated that the overexpression of ISL1 was positively correlated with the promoter activity of c-Myc or CyclinD1 promoter, while knockdown of ISL1 inhibited their activity.
Wild-type p53 acts as a potent regulator of the cell cycle, especially coordinating transcriptional responses to pathological stress [22] . In response to DNA damage, p53 accumulates in cell nuclei causing cell arrest at the G 1 phase and inducing apoptosis [23] , [24] . We found that the overexpression of ISL1 did not affect the expression level of p53, but the knockdown of ISL1 significantly enhanced the expression of p53 and increased the apoptosis. Whether p53 is involved in the anti-apoptosis role of ISL1 in the islet cells requires further investigation. It should be mentioned that we have observed that ISL1 overexpression also slightly increased the level of apoptosis of HIT-T15 cells (data not shown). This phenomenon is possibly due to the high level of endogenous ISL1 expression in HIT-T15 cells and the overexpression of ISL1 may disrupt the orchestration of ISL1 regulation. We have noticed that EdU incorporation rate in ISL1 expressing adult islet cells is about 2.5 folds than that in the control ( Fig. 3C, 3E ). However, the differences are not so obvious in FCM analysis. EdU incorporation and FCM analysis are two different methods reflecting cell proliferation. EdU incorporation detected the newly synthesized random DNA regardless of cell cycle phase transition. The data are expressed as folds of positive islet cells relative to negative cells. While FCM detected the cell proliferation, represented by S/G 2 M phases transition in one cell cycle. Additionally, as ISL1 target genes, we also noted CyclinD1 was regulated more significantly than c-Myc by ISL1. This implies that CyclinD1 might be a more potent downstream factor to mediate the ISL1 proliferation effects in pancreatic islet cells.
In conclusion, ISL1 can promote mature pancreatic islet cells proliferation and attenuate cell apoptosis against oxidative stress. c-Myc and CyclinD1 are identified as novel downstream targets of ISL1 and are involved in ISL1 regulation on the proliferation of adult islet cells. Our findings extend the knowledge about the crucial role of ISL1 in maintaining mature islet cells homeostasis. Our results also provide insights into the new regulation relationship between ISL1 and other growth factors.
Materials and Methods
Ethics Statement
The animal experiments were performed in accordance with the ethical principles and guidelines for scientific experiments on animals of the Swiss Academy of Medical Sciences (1995). All protocols were approved by the Animal Care and Use Committee of Peking University (LA 2010-066). The animals in the postprandial state were anesthetized with 5 mg/100 g body weight of sodium pentobarbital and pancreatic tissue was removed.
Plasmid constructs and recombinant lentivirus
The plasmid construct pcDNA3.1-ISL1 was previously described [25] and was used as stable transfection. The pLL3.7-ISL1-siRNA plasmid was commercially constructed by the GeneChem Company (Shanghai, China) and was used for stable transfection. The luciferase reporter constructs c-Myc-luc and CyclinD1-luc were generous gifts from Prof. Yongfeng Shang (Department of Biochemistry and Molecular Biology, Peking University Health Science Center).
The sequences of ISL1-siRNA are: sense, GAGACAUGGUGGUUUAtt ; antisense: UUUCUCCUUGCACCUCtt . The sequences of non-silencer siRNA are: sense: UUCUCCGAACGUGUCACGUtt ; antisense: ACGUGACACGUUCGGAGAAtt . Recombinant lentiviruses with ISL1 cDNA or ISL1-siRNA were obtained from the Genechem Company (Shanghai, China). Lentivirus infection was carried out following the manufacturer's instruction with multiplicity of infection as 10 and rat adult islet cells were used as target cells. The infection efficiency was detected by flow cytometry analysis.
Cell cultures, stable cell lines and adult islets isolation
Monolayer cultures of hamster pancreatic islet β cell line HIT-T15 (ATCC number: CRL-1777) were maintained in RPMI 1640 (GIBCO BRL) complete medium, which contained 2 mM/L glutamine, 100 U/ml penicillin, 100 U/ml streptomycin, and 10% fetal bovine serum. The medium was changed every 1–2 days.
To establish stable cell lines, 2×10 5 HIT-T15 cells were plated into a 60 mm culture dish. When approximately 50% confluence was reached, HIT-T15 cells were transfected with 2 µg pcDNA3.1-ISL1 plasmid, or a control pcDNA3.1 plasmid, respectively, with Lipofectamine2000. G418 selection (1000 µg/ml) was performed and single colonies were picked up at about 21 days. The cells transfected with pLL3.7-ISL1-siRNA or pLL3.7-non-silencer were selected using puromycin resistant screening (1 µg/ml). Identification of stable cell lines were performed using RT-PCR and Western blotting for quantifying the expression levels of ISL1.
Pancreatic islets were isolated from 100–120 g male Sprague-Dawley rats as previously described [26] . Briefly, donor pancreases were perfused in situ with collagenase V (3 mg/ml, Sigma-Aldrich, St. Louis, MO, USA), and the pancreatic tissues were harvested after the perfusion and were further incubated at 37°C with gentle vortex (375 rpm) for 30 min. Islets were released from the pancreas and were handpicked. The isolated islets were washed with Hank's solution twice at 4°C, counted and cultured in RPMI complete medium.
Cell proliferation assays and clone formation assay
The pancreatic cell proliferation was measured by performing WST-8 assay and 5-ethynyl-20-deoxyuridine (EdU) incorporation assay, using a CCK-8 cell proliferation kit (Dojindo Laboratories, Kumamoto, Japan) and EdU assay kit (Ribobio, Guangzhou China), respectively, according to the manufacturers' instructions. For CCK-8 assay, cells were seeded into a 96-well plate at 1×10 3 cells per well with 100 µl complete medium and cultured at 37°C, 10 µl CCK-8 solution was added to each well after 12 h, 24 h, 48 h, and 72 h, respectively. Plates were incubated at 37°C for 2 h, and then the absorbance at 450 nm was measured with Microplate Reader (Bio-Rad, La Jolla, CA, USA). All experiments were done in triplicate and three independent repeating experiments were performed.
For EdU incorporation assay, cells were cultured in triplicate in 96-well plates at a density of 1×10 3 HIT-T15 cells or 50 rat islets mass per well for 48 h at 37°C, and then 50 µM of EdU was added to each well and cells were cultured for additional 4 h at 37°C. The cells were fixed with 4% formaldehyde for 15 min at room temperature and treated with 0.5% Triton X-100 for 20 min at room temperature for permeabilization. After washing with PBS three times, 100 µl of 1× Apollo® reaction cocktail was added to each well and the cells were incubated for 30 min at room temperature. Then the cells were stained with 100 µl of Hoechst33342 for 30 min and visualized under a fluorescent microscope (Olympus Corporation, Tokyo, Japan). The EdU positive cells (red cells) were counted using Image-Pro Plus (IPP) 6.0 software (Media Cybernetics, Bethesda, MD, USA). The EdU incorporation rate was expressed as the ratio of EdU positive cells to total Hoechst33342 positive cells (blue cells). All experiments were done in triplicate and three independent repeating experiments were performed.
A total of 1×10 5 ISL1 or ISL1-siRNA stable transfection cells were plated onto a 60-mm culture dish. Cells were maintained in culture medium supplemented with 1 mg/ml G418 or 1 µg/ml puromycin for 21 days and stained with crystal violet for colony counting. The colony diameter greater than 4.5 mm was counted using IPP 6.0 software. All experiments were done in triplicate and three independent repeating experiments were performed.
Flow cytometry analysis for cell cycle, apoptosis and ROS generation
Cell cycle analysis was performed by flow cytometry. Briefly, cultured cells were trypsinized into single cell suspensions and fixed with 70% ethanol for 30 min on ice. RNA was degraded by incubation with 20 mg/ml RNase (Sigma-Aldrich, St. Louis, MO, USA) for 1 h at 37°C. DNA was labeled with 20 mg/ml propidium iodide (PI, Sigma-Aldrich) and DNA content was assessed by FACS Calibur flow cytometry (Becton Dickinson, Franklin Lakes, NJ, USA) equipped with the ModiFit LT v2.0 software. For apoptosis analysis, cultured cells were harvested by trypsinization and washed with PBS. 1×10 6 cells from each sample were processed for Annexin V FITC/PI apoptosis detection (Becton Dickinson) according to the manufacturer's instructions.
Reactive oxygen species (ROS) were detected using reactive oxygen species assay kit C1300 (Applygen, Beijing, China). Cells were incubated with variable concentrations of glucose oxidase (0 to 100 mU/mL) in fresh serum-free RPMI1640 for 4 h, before the absorbance at 530 nm was assessed by FACS Calibur flow cytometry. Data are presented as the percentage of ROS positive cells amongst total cells. All experiments were done in triplicate and three independent repeating experiments were performed.
RT-PCR and Real-time RT-PCR
Cells were seeded into a 6-well plate at 1×10 5 cells per well and harvested after 48 h culture. Total RNA extraction was performed using Trizol Reagent (Invitrogen, Carlsbad, CA, USA) based on the manufacturer's instructions. Amplifications were performed in the ABI 7300 Real-Time RT-PCR System (Carlsbad, CA, USA) with different primers ( Table 1 ). All annealing temperatures were 60°C. Transcription levels were normalized to 18S rRNA levels. Each value presents the average of at least 3 independent experiments.
10.1371/journal.pone.0022387.t001 Table 1
Primers used in RT-PCR and real-time RT-PCR.
Primers
Primer sequences (5′–3′)
Product size (bp)
18SRNA
F: GTAACCCGTTGAACCCCATT
151
R: CCATCCAATCGGTAGTAGCG
ISL1
F: CTGCTTTTCAGCAACTGGTCA
123
R: TAGGACTGGCTACCATGCTGT
c-Myc
F: GCCACGTCTCCACACATCAG
141
R: TCTTGGCAGCAGGATAGTCCTT
CyclinD1
F: GCGTACCCTGACACCCCTCTC
183
R: CTCCTCTTCGCCTGATCC
CyclinA
F: GCCTTCACCATTCATGTGGAT
118
R: TTGCTGCGGGTAAAGAGACAG
p53
F: TTCCACCTGGGCTTCCTG
144
R: GGATAGGTCGGCGGTTCAT
Insulin
F: AGGACCCACAAGTGGAACAACT
140
R: CAACGCCAAGGTCTGAAGGT
Western blotting
Western blotting was performed as previously described [27] . Cells were seeded into a 6-well plate at 1×10 5 cells per well and were harvested after 48 h culture. Total protein was prepared and subjected to 12% SDS polyacrylamide gel electrophoresis and subsequently transferred onto nitrocellulose membranes. Antibodies, including ISL1 (H00003670-M05, Abnova, Taipei, China), CyclinD1 (ab61758, Abcam, Hong Kong, China), c-Myc (sc-764, Santa Cruz, CA, U.S.A.) and horseradish peroxidase-conjugated secondary antibody from Santa Cruz were used.
Luciferase assays
HIT-T15 cells were seeded at a density of 10×10 4 per well in a 24-well plate, constructs of c-Myc-luc, CyclinD1-luc, or ISL1 were transfected using Lipofectamine™2000 (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's instructions when cells reached 60% confluence. The total amount of DNA was kept constant using pcDNA3.1/β-gal plasmid. Luciferase activity was measured and normalized to Renilla luciferase activity. All experiments were done in triplicates and three independent repeating experiments were performed.
Electrophoretic mobility-shift assay (EMSA)
Nuclear extracts were prepared from HIT-T15 cells. Oligonucleotides used in EMSA assay were labeled with [γ- 32 P] ATP as described previously [27] . The sequences of the sense strand of these oligonucleotides are as follows (ISL1 binding motif is under-lined and mutant bases are in lower cases):
Wild-type c-Myc probe: 5′-ATTAGAGTCGGCTTT TAAT TAGTTAACACACAC-3′ .
Mutant c-Myc probe: 5′-ATTAGAGTCGGCTTTgccgTAGTTAACACACAC-3′ .
Wild-type CyclinD1 probe: 5′-AATTTAATTTCTTTTT TAAT TAAAAAAAATGAGT C-3′ . Mutant CyclinD1 probe: 5′-AATTTAATTTCTTTTTgccgTAAAAAAAATGA GTC- 3′ . The binding reactions were performed at 4°C (20 µl final volume) with 10 µg of nuclear protein and 0.25 ng (20 kcpm) of radiolabeled, double-stranded oligonucleotide in a binding buffer. Antibody analyses were performed by pre-incubation of nuclear extract protein with specific ISL1 antibody (Abnova) for 20 min at 4°C prior to addition of the radiolabeled probe. Protein-DNA complexes were resolved on a 6% native polyacrylamide gel and visualized by autoradiography.
ChIP assay
HIT-T15 cells were stably transfected with 2 µg of pcDNA3.1-ISL1 plasmid and cells were harvested after 48 h culture. ChIP experiments were performed according to the method described previously [17] . After crosslink reversal, precipitated DNA was analyzed by PCR for fragments of the c-Myc and CyclinD1 promoters with different primers. The annealing temperatures were 60°C. The input DNA and immunoprecipitated DNA were calculated by real-time PCR using SYBR® Green Real-time PCR Master Mix (TOYOBO, Japan). The data obtained were normalized to the corresponding DNA precipitated by IgG. c-Myc: forward (F), 5′-AATGCACAGCGTAGTATTCAGGA-3′ ; reverse (R), 5′ GGA GTG AAT TGC CAA CCC AGA 3′ (270 bp fragment). CyclinD1: forward (F), 5′- AGCTTCGGTGTCTGGTTC- 3′ ; reverse (R), 5′- ATTCCAGCAACGCTCAA GATG- 3′ (283 bp fragment).
Type 1 and type 2 diabetes animal models
STZ-induced diabetic mice (male C57BL/6 mice at the age of 8 to 10 weeks, obtained from Department of Laboratory Animal Science, Peking University, China), were established as described previously [17] . The mice with stable hyperglycemia (blood glucose level >20 mmol/L) were used as the type 1 diabetes animal model. Akita, db/db and db/w mice were provided by Prof. Youfei Guan (Peking University, China). The Akita mice at 12 weeks of age were used as an autosomal dominant mutation Mody animal model and showed hyperglycemia with notable pancreatic β-cell dysfunction. Db/db mice at 14–16 weeks of age were used as the type 2 diabetes animal model, which were homozygous for the db gene and, thus, exhibited an obese, diabetic phenotype. Db/w mice that were heterozygous for the db gene exhibited a nondiabetic, normal phenotype were used as controls to db/db mice. The type 1 and type 2 diabetes mice in the postprandial state were anesthetized with 5 mg/100 g body weight of sodium pentobarbital and pancreatic tissue was removed. RNA was prepared from these tissues and used for real-time RT-PCR assay.
Statistical analysis
The data are expressed as mean ± standard deviation (S.D.). Comparisons between groups were analyzed using Student's t -test or ANOVA, and the Student-Newman-Kleuss method was used to estimate the level of significance. Differences were considered to be statistically significant at p <0.05.
Supporting Information
Figure S1
Glucose oxidase stimulated ROS production. Level of ROS production was measured by flow cytometry analysis. HIT-T15 cells were treated with glucose oxidase (GO) at various concentrations (0–100 mU/mL) for 4 h. ROS production exhibited a dose dependent manner with GO concentration.
(TIF)
Figure S2
Glucose oxidase increased the number of apoptotic cells. Level of apoptosis rate was measured by flow cytometry analysis. HIT-T15 were treated with GO at various concentrations (0–100 mU/mL) for 4 h. the numbers of apoptotic cells increased in a dose dependent manner with GO concentration.
(TIF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Despite significant advances in anesthetic and surgical techniques, postoperative atrial fibrillation (POAF) remains the most common complication after cardiac surgery [1] – [3] . The incidence of POAF varies from 11% to 40%, depending on the definition and the method of monitoring [1] – [3] . Although this arrhythmia is usually benign and self-limiting, it may result in hemodynamic instability, a longer hospital stay, and increased health care costs [1] – [3] . Given the clinical consequences attributable to POAF, its prevention is of great importance. To date, many pharmacologic approaches have been attempted to prevent POAF, for example, β-blockers, amiodarone, and magnesium [4] . Most reviews reflect a growing consensus in favor of the prophylactic administration of β-blockers for cardiac surgery patients [5] . In addition, updated American College of Cardiology/American Heart Association (ACC/AHA) 2006 guidelines recommend β-blockers for the prevention of POAF [6] .
Despite the extensive studies, the exact pathophysiology of POAF is for the moment far from being fully elucidated [1] – [3] . A growing body of evidence suggests that markers of inflammation and oxidative injury are elevated in atrial fibrillation patients [7] – [10] . Carvedilol, a non-selective β-adrenergic blocking agent approved for use in heart failure cases, has a number of ancillary activities including anti-inflammatory and antioxidant properties [11] , [12] . Moreover, unlike other beta-blockers, carvedilol antagonizes the rapid-depolarizing sodium channel, the human ether-a-go-go-related gene potassium channel, and the L-type calcium channel [11] , [12] , which suggests a pharmacologic profile similar to amiodarone, a proven anti-arrhythmic agent for the prevention of POAF [13] . Theoretically, this should reduce the incidence of arrhythmia, including POAF. All these properties of carvedilol have generated interest in its use as a prophylactic agent for POAF. Recently, several relevant studies regarding prophylactic carvedilol in preventing POAF have been published [14] – [19] . However, the role of carvedilol in preventing POAF remains unknown. We therefore undertook a meta-analysis of published studies to the efficacy of carvedilol in preventing POAF for adult patients undergoing cardiac surgery.
Methods
Literature search and inclusion criteria
Two investigators searched PubMed database for relevant articles published up to September 2013. The initial search terms were carvedilol and atrial fibrillation. No language restriction was imposed. In addition, the reference lists of identified studies were manually checked to include other potentially eligible trials. This process was performed iteratively until no additional articles could be identified.
The following inclusive selection criteria were applied: (i) study design: comparative trial; (ii) study population: adult patients undergoing cardiac surgery; (iii) intervention: carvedilol (no matter what regimen applied); (iv) comparison intervention: control (placebo or other beta-blockers) and (v) outcome measure: the incidence of POAF.
Data extraction and outcome measures
Two investigators independently extracted the following data from each trial: first author, publication year, number of patients (carvedilol/control), patient characteristic, regimen of intervention (carvedilol/control), definition and monitoring of POAF, study design, the incidence of POAF, and length of hospital stay (LOS). Extracted data were entered into a standardized Excel file. The primary outcome was the incidence of POAF. Secondary outcome included LOS.
Statistical analysis
Differences were expressed as relative risks (RRs) with 95% confidence intervals (CIs) for dichotomous outcomes, and weighted mean differences (WMDs) with 95% CIs for continuous outcomes. Heterogeneity across studies was tested by using the I 2 statistic, which was a quantitative measure of inconsistency across studies. Studies with an I 2 statistic of 25% to 50% were considered to have low heterogeneity, those with an I 2 statistic of 50% to 75% were considered to have moderate heterogeneity, and those with an I 2 statistic of >75% were considered to have a high degree of heterogeneity [20] . An I 2 value greater than 50% indicates significant heterogeneity [21] . A fixed-effects model was used (I 2 ≤50%), and a random-effects model was used in the case of significant heterogeneity (I 2 >50%). We further conducted subgroup analyses according to type of control, surgery type, and study design. We also investigated the influence of a single study on the overall risk estimate by omitting one study in each turn. We did not assess publication bias [22] , because the pooled estimate included fewer than ten trials. A p value <0.05 was considered statistically significant. All statistical analyses were performed using Stata version 11.0 (Stata Corporation, College Station, Texas, USA).
Results
Study identification and selection
The initial search yielded 87 relevant publications of which 79 were excluded for various reasons (review, letter, case report, or irrelevant to the current analysis) based on the titles and abstracts. The remaining eight were then retrieved for full text review, two of them were also excluded because one was focused in patients undergoing coronary bypass graft with heart failure and one was currently ongoing [23] , [24] . Thus, six studies were included in the final analysis [14] – [19] . The flowchart of studies included in meta-analysis was shown in Figure 1 .
10.1371/journal.pone.0094005.g001 Figure 1
Selection process for clinical trials.
Study characteristics
The basic characteristics of studies included in the meta-analysis are shown in Table 1 . These studies were published between 2003 and 2010. The sample size of these studies ranged from 53 to 207 (total 765). Four studies in this meta-analysis enrolled patients undergoing coronary artery bypass grafting (CABG) only [15] – [18] . The remaining two included patients undergoing CABG and/or valve surgery [14] , [19] . Carvedilol was administered orally by different regimens and formulations. Timing of initiation for carvedilol prophylaxis was 3–10 days before the surgery in the preoperative prophylaxis studies [15] , [16] , [19] and within 24 hours of surgery in the postoperative group [17] , [18] . Definition of POAF in terms of duration varied among the studies. All the patients were monitored using electrocardiography.
10.1371/journal.pone.0094005.t001 Table 1
Characteristics of studies included in the meta-analysis.
Study (Reference)
Sample size (Carvedilol/Control)
Patient characteristic
Mean age (year)/Male (%)
Regimen of intervention
POAF
LOS (days)
Study design
Carvedilol
Control
Carvedilol
Control
Carvedilol
Control
Merritt 2003 [14]
115(26/89)
Adult patients undergoing CABG and/or VS
60.3/NA
NA
Metoprolol/atenolol
2/26
28/89
5.9±1.9
6.9±4.5
Non-RCT
Haghjoo 2007 [15]
120(60/60)
Adult patients undergoing CABG
61/52.5
6.25 mg twice daily, oral, starting from 10 days before surgery, then increasing until to the maximum
Metoprolol 25 mg twice daily, oral, starting from 10 days before surgery, then increasing until to the maximum
9/60
20/60
NA
NA
RCT
Acikel 2008 [16]
110(55/55)
Adult patients undergoing CABG
60/71.8
12.5 mg twice daily, starting on 3 days prior to surgery, lasting to the morning of surgery, then titrating according to hemodynamic responses after CABG
Metoprolol 50 mg twice daily, starting on 3 days prior to surgery, lasting to the morning of surgery, then titrating according to hemodynamic responses after CABG
9/55
20/55
NA
NA
RCT
Tsuboi 2008 [17]
160(80/80)
Adult patients undergoing CABG
66.5/70.6
5 or 10 mg/day, oral, starting on postoperative days 1 or 2, then increasing until to the maximum
Placebo
12/80
27/80
17.0±6.2
22.0±12.3
Non-RCT
Yoshioka 2009 [18]
53(31/22)
Adult patients undergoing CABG
67/68
2.5 mg/day, oral, starting on postoperative days 1 or 2
Placebo
4/31
7/22
NA
NA
Non-RCT
Ozaydin 2013 [19]
207(104/103)
Adult patients undergoing CABG and/or VS
63/72.5
6.25 mg twice daily, starting from 7 days before surgery, if not tolerated, a 3.125 mg twice daily dose was given
Metoprolol 50 mg once daily dose, starting from 7 days before surgery, if not tolerated, a 25 mg twice daily dose was given
25/104
37/103
NA
NA
RCT
CABG, coronary artery bypass grafting; LOS, length of hospital stay; NA, no data available; POAF, postoperative atrial fibrillation; RCT, randomized controlled trial; VS, valve surgery.
Primary outcome: POAF
The definition and monitoring of POAF in each trial are summarized in Table 2 . Overall, six studies including 765 patients were included in this analysis (356 in the carvedilol group and 409 in the control group). Meta-analysis of six studies using a fixed-effects model suggested that carvedilol significantly reduced the incidence of POAF in patients undergoing cardiac surgery compared with control (RR 0.49, 95% CI 0.37 to 0.64, p<0.001; Figure 2 ). There was no heterogeneity among the studies (I 2 = 0%, heterogeneity p = 0.645; Figure 2 ).
10.1371/journal.pone.0094005.g002 Figure 2
Effect of carvedilol versus control on the incidence of postoperative atrial fibrillation.
10.1371/journal.pone.0094005.t002 Table 2
Definition and monitoring of POAF.
Study (Reference)
Definition of POAF
Monitoring of POAF
Merritt 2003 [14]
NA
NA
Haghjoo 2007 [15]
Absent P wave before the QRS complex together with irregular ventricular rhythm on the rhythm strips, lasting longer than 5 minute.
ECG and 12-lead ECG were need to confirm
Acikel 2008 [16]
An irregular rhythm with no prominent P waves lasting 30 s or more
Automated arrhythmia detectors in cardiac ICU, and simultaneous telemetric display of ECG in the ward
Tsuboi 2008 [17]
Absent consistent P waves before each QRS complex and an irregular ventricular rate and as episodes of atrial fibrillation that persisted for over 10 min.
12-lead ECG
Yoshioka 2009 [18]
Lasted more than 5 minutes or required intervention for angina or hemodynamic compromise, or any episode that required intervention for angina or hemodynamic compromise.
Monitoring system on a rhythm strip or 12-lead ECG
Ozaydin 2013 [19]
An irregular rhythm with the absence of discrete P-waves lasting 5 min during hospitalization
Continuous ECG monitoring and all-day Holter
ECG, electrocardiogram; NA, no data available; POAF, postoperative atrial fibrillation; ICU, intensive care unit.
Then we further conducted subgroup analyses based on type of control (metoprolol vs. placebo), surgery type (CABG and/or valve surgery vs. CABG only), and study design (randomized trials vs. nonrandomized trials). Table 3 shows the results of subgroup analyses for POAF. The results suggested that carvedilol appeared to be superior to metoprolol for the prevention of POAF (RR 0.51, 95% CI 0.37 to 0.70, p<0.001; Figure 3 ). No evidence of heterogeneity was observed in subgroup analysis. Influence analysis suggested exclusion of any single study did not materially alter the overall combined RR, with a range from 0.41 (0.29 to 0.59) to 0.52 (0.39 to 0.68), which adds robustness to our results.
10.1371/journal.pone.0094005.g003 Figure 3
Effect of carvedilol versus metoprolol on the incidence of postoperative atrial fibrillation.
10.1371/journal.pone.0094005.t003 Table 3
Results of subgroup analyses for POAF.
Subgroup analysis
n (N)
Carvedilol
Control
OR (95% CI)
p value
I 2 (%)
Heterogeneity p
Study design
RCTs [15] , [16] , [19]
3 (437)
43/219
77/218
0.56 (0.40–0.77)
<0.001
0
0.489
Non-RCTs [14] , [17] , [18]
3 (328)
18/137
62/191
0.38 (0.23–0.64)
<0.001
0
0.723
Surgery type
CABG and/or valve surgery [14] , [19]
2 (322)
27/130
65/192
0.56 (0.37–0.85)
0.007
51.7
0.15
CABG only [15] – [18]
4 (443)
34/226
74/217
0.44 (0.31–0.64)
<0.001
0
0.999
Type of comparison
Metoprolol [14] – [16] , [19]
4 (552)
45/245
105/307
0.51 (0.37–0.70)
<0.001
0
0.408
Placebo [17] , [18]
2 (213)
16/111
34/102
0.44 (0.26–0.74)
0.002
0
0.886
CABG, coronary artery bypass grafting; RCT, randomized controlled trial; n, number of patients; N, number of trials.
Secondary outcome: LOS
Two trials reported the effect of carvedilol on LOS and provided available data (expressed as mean ± standard deviation) with a total of 275 patients. The combined analysis using a random-effects model showed that carvedilol did not significantly reduce LOS (WMD −2.75, 95% CI −6.64 to 1.14, p = 0.17), with a high degree of heterogeneity between the trials (I 2 = 82.9%, heterogeneity p = 0.016).
Publication bias
Publication bias was not assessed because of the limited number (below 10) of studies included in the analysis.
Discussion
Meta-analysis of all six included studies using a fixed-effects model illustrates that carvedilol may effectively reduce the incidence of POAF in adult patients undergoing cardiac surgery.
The mechanisms that carvedilol reduces the incidence of POAF are not entirely known. However, there is now an increasing body of evidences that oxidative stress [25] , and inflammation [26] , [27] , and increased sympathetic activation [28] are involved in the pathogenesis of POAF. Carvedilol is a β blocker with antioxidant and anti-inflammatory properties [11] , [12] , and reduces sympathetic activity [29] . From a pathophysiological point of view, it is plausible that the abovementioned properties of carvedilol might result in the favorable effect on the prevention of POAF.
Recently, Khan et al carried out a meta-analysis of randomized controlled trials and confirmed the efficacy of prophylactic beta-blockers against POAF [30] . Both the Khan meta-analysis and our meta-analysis showed that carvedilol appeared to be more effective than metoprolol for the prevention of POAF. Compared with metoprolol, carvedilol has been shown to increase the levels of antioxidant enzymes (superoxide dismutase and glutathione peroxidase). Moreover, carvedilol may have direct antiarrhythmic profile through electrophysiological traits, since it blocks multiple cationic channels (Na + , K + , and Ca 2+ ) [11] , [12] . These properties of carvedilol, which are not equally shared by metoprolol, may partly explained superior efficacy of carvedilol in preventing POAF. In addition, numerous trials indicate that carvedilol is better than conventional β1-selective β blockers on reducing sympathetic activation, a risk factor for atrial fibrillation [28] , [29] .
In this meta-analysis carvedilol did not significantly reduce the LOS. The total incidence of POAF is 26.1% (200 of 765), less than one-third of patients develop POAF and still fewer develop prolonged atrial fibrillation, so the effect of carvedilol on LOS in patients prone to atrial fibrillation would have to be very large to be able to detect an effect of LOS in the total population. In addition, a relatively small number of samples (only two studies) provided available data on LOS, additional studies or data are warranted.
One problem with the use of carvedilol to prevent POAF is that the majority of patients does not develop POAF after cardiac surgery but would still be exposed to possible side effects. In this meta-analysis, two trials reported carvedilol was well tolerated and side effects attributable to carvedilol were detected. And one trial reported complication rates were similar between carvedilol and control groups, including postoperative myocardial infarction and renal dysfunction.
Several potential limitations of this meta-analysis merit consideration. First, our study included only six studies and some of them have a modest sample size. Overestimation of the treatment effect is more likely in smaller studies compared with larger samples. Second, our analysis is based on six clinical studies, and half of them were non-randomized controlled trials. The targeted population, adopted carvedilol protocols, type of control, and study design differed among the included studies. These factors may result in the heterogeneity and have potential impact on our results. Furthermore, these studies lack homogeneity in both the method of postoperative monitoring and in their definition of POAF. This may lead to potential underestimation and/or overestimation of the true incidence of POAF. Finally, it was possible that some missing and unpublished data may lead to bias in effect size.
In conclusion, despite its various limitations, our study is clinically valuable because it revealed that carvedilol leads to lower incidence of POAF than control and appears to be superior to metoprolol as the current study clearly delineated. Carvedilol may effectively reduce the incidence of POAF in patients undergoing cardiac surgery. On the basis of this encouraging finding, we believe that research on the field is promising and should be continued. At least the ongoing COMPACT [24] , which is a prospective, multi-center, randomized, open-label, active-controlled trial, will answer the question of whether or not carvedilol is more superior to metoprolol in preventing POAF in patients undergoing CABG.
Supporting Information
Checklist S1
PRISMA Checklist.
(DOC)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Dengue viruses belong to the Flavivirus genus of the Flaviviridae family and include four antigenically different serotypes of dengue virus [1] . Dengue virus is a growing threat to public health, not only in terms of geographical distribution but also with respect to infection cases. Dengue occurs in as many as 128 countries throughout tropical and subtropical areas [2] . Vaccination has been proposed as a cost-effective strategy to combat the threat of infectious disease. Unfortunately, an approved dengue vaccine is not presently available, despite tremendous efforts in previous decades. Several vaccine candidates are proceeding in clinical trials [3] . The most advanced candidate is Sanofi Pasteur's recombinant live, attenuated tetravalent dengue-yellow fever chimeric virus vaccine. These vaccine candidates are based on the backbone of 17D yellow fever vaccine strain, each expressing the pre-membrane and envelope genes of one of the four dengue virus serotypes [4] . Recently, the results of a phase 2b trial of this tetravalent dengue vaccine in Thai schoolchildren of 4–11 years of age were reported [5] . The overall efficacy of the vaccine was 30.2%. One or more doses of the vaccine reduced the incidence of dengue-3 and dengue-4 febrile diseases by 80–90%, with a smaller reduction in diseases caused by dengue-1. However, there was no efficacy against dengue-2. Thus, there is an urgent need to complement the deficiency of the recombinant live, attenuated tetravalent dengue-yellow fever chimeric virus vaccine.
In most cases, dengue viral infection causes dengue fever, which is a self-limiting illness. However, infection with dengue virus can also develop into severe dengue hemorrhagic fever (DHF) or dengue shock syndrome (DSS) [6] , [7] . The mechanisms of DHF and DSS are still not fully understood. The pathogenesis of DHF and DSS may be due to antibody-dependent enhancement (ADE). ADE is mediated by nonneutralizing antibodies or subneutralizing antibody concentrations bound to the dengue virion, which enhances viral entrance into target cells via the Fc receptor (FcR) [8] . ADE is also mediated by dual-specific antibodies, which bind to both dengue virus particles and target cells lacking FcR expression [9] . In addition to ADE, dengue viral proteins induced antibodies cross-react with plasminogen, endothelial cells, and platelets have been proposed to play an important role in the pathogenesis of DHF and DSS [10] – [12] . The complex pathogenesis of DHF and DSS represents a barrier that complicates dengue vaccine development.
Dengue envelope protein is the major structural protein that mediates dengue virus infection. The envelope protein domain III (ED III) is responsible for viral attachment by binding to the cellular receptor [13] , [14] . ED III has been proposed as a suitable target for dengue vaccine development [15] . The immunogenicities of purified recombinant envelope protein or ED III have been evaluated in mice and nonhuman primates [16] – [19] . However, these purified recombinant proteins are poorly immunogenic. Adjuvants are often required in vaccine formulations to augment the immune response to antigens. However, aluminum-containing adjuvants, which are the most widely used in human vaccines, may not be suitable for use in dengue subunit vaccines to induce robust immune responses.
Antigens and immunostimulators are two major components of modern subunit vaccines. We [20] and others [21] – [23] have demonstrated that both bacterial-derived lipoproteins and synthetic lipopeptides can activate antigen-presenting cells via the Toll-like receptor signaling pathway and augment humoral and cellular responses. Based on these findings, we have developed technology to express recombinant lipoprotein in high yields for the development of subunit vaccines with high immunogenicity [24] . In the present study, we prepared recombinant lipidated dengue-2 ED III (LD2ED III) and evaluated its immunogenicity. We demonstrated that exogenous adjuvant is not required for the induction of a robust immune response to LD2ED III. These results provide important information for future clinical studies of ED III-based subunit vaccines.
Materials and Methods
Ethics statement
Animal studies were conducted in strict accordance with the recommendations of Taiwan's Animal Protection Act. The protocols were approved by the Animal Committee of the National Health Research Institutes (Protocol No: NHRI-IACUC-098014) and performed according to their guidelines.
Cloning and expression of recombinant proteins
The amino acid sequence of D2ED III was described previously [25] . Briefly, the sequence of D2ED III was obtained by aligning 13 amino acid sequences from different isolates of type II dengue (accession numbers P07564, P14339, P30026, P27914, Q9WDA6, P12823, P14338, P14337, P14340, P18356, P29984, P29990, and P29991). Based on the amino acid sequence of D2ED III, the DNA sequence of D2ED III using Escherichia coli codon usage was determined and fully synthesized by a biotechnology company (Purigo Biotechnology Co., Taipei, Taiwan). The synthesized DNA was then amplified by PCR. To generate an expression plasmid for D2ED III production, the following primers were used: forward primer, 5′-GGAATTC CATATG aaaggcatgagctatgC-3′ (NdeI site, underlined); reverse primer, 5′-CCG CTCGAG gctgctgcctt-3′ (XhoI site, underlined). The PCR product was then cloned into the NdeI and XhoI sites of the expression vector pET-22b(+) (Novagen, Madison, WI) to produce the plasmid pD2DE III. As a result, the C-terminus of the recombinant protein contained a hexahistidine tag (His-tag). To express protein, E. coli BL21 (DE3) (Invitrogen, Carlsbad, CA) was transformed with pD2DE III. The transformed cells were cultured at 37°C overnight, and protein expression was induced by adding 1 mM isopropylthiogalactoside (IPTG), followed by incubation for 3 hours at 37°C. To clone and express LD2ED III, the D1 domain and the lipid signal peptide of the lipoprotein Ag473 [24] were cloned into the NdeI and BamHI sites of the expression vector pET-22b(+) (Novagen, Madison, WI) to obtain the plasmid pLipo. The D2ED III gene was cloned into the BamHI and XhoI sites of the pLipo plasmid to produce the plasmid pLD2DE III. As a result, the C-terminus of the recombinant protein contained a His-tag. E. coli C43(DE3) (Lucigen, Middleton, WI) was transformed with pLD2DE III to express lipidated protein. The transformed cells were cultured at 37°C overnight. One ml of the overnight culture was scaled up to 600 ml in a 2 l-shake flask and incubated at 37°C for 4 h before induction. Protein expression was induced (OD 600 = 0.8) by adding 1 mM IPTG, followed by incubation at 20°C for 20 h.
Production of D2ED III and LD2ED III
D2ED III was purified by disrupting the harvested cells in a French press (Constant Systems, Daventry, UK) at 27 Kpsi in homogenization buffer [20 mM Tris (pH 8.0), 50 mM sucrose, 500 mM NaCl and 10% glycerol]. The cell lysate was clarified by centrifugation (80,000× g for 40 min). Most of the D2ED III was present in inclusion bodies. D2ED III was then solubilized with extraction buffer [50 mM NaH 2 PO 4 /5 mM EDTA/200 mM NaCl/0.5 M urea/1% Triton X-100 (pH 6.0)]. The extracted fraction was loaded onto immobilized metal affinity chromatography (IMAC) columns (QIAgen, Hilden, Germany) to purify D2ED III. The eluent from the IMAC column was further purified by passage through an anion exchange column (DEAE Sepharose fast flow; GE) after dialysis against DEAE buffer [50 mM NaH 2 PO 4 /1 M urea (pH 5.8)]. An E membrane (Pall Co., USA) was used to remove endotoxin. The endotoxin levels of the purified D2ED III were determined by the Limulus amebocyte lysate (LAL) assay (Associates of Cape Cod, Inc., Cape Cod, MA), and the resulting endotoxin levels were less than 0.06 EU/mg. After dialysis against 10 mM sodium acetate/3 mg/ml sucrose/4 mM glycine, the D2ED III was lyophilized and stored at −20°C. The fractions from each step were analyzed by SDS-PAGE and immunoblotted with anti-His-tag antibodies.
The disruption and purification steps in the production of LD2ED III were similar to those used for D2ED III. However, LD2ED III was not subjected to anion exchange chromatography. After IMAC purification of LD2ED III, Endotoxin Removing Gel (Pierce, Rockford, IL, USA) was used to remove lipopolysaccharide (LPS). The LPS levels of the purified LD2ED III were determined by LAL assay, and the resulting LPS levels were less than 0.06 EU/mg. After the LD2ED III was dialyzed against 10 mM sodium acetate/3 mg/ml sucrose, the LD2ED III was lyophilized and stored at −20°C. The fractions from each step were analyzed by SDS-PAGE and immunoblotted with anti-His-tag antibodies.
Identification of the lipid moiety in LD2ED III
LD2ED III was digested with trypsin (Sigma, St. Louis, MO). After digestion, the reaction mixture was further purified with a ZipTip (Millipore, Massachusetts). A 1-µl aliquot of the ZipTip-polished tryptic fragments was mixed with 1 ml of a saturated solution of α-cyano-4-hydroxycinnamic acid in acetonitrile/0.1% trifluoroacetic acid (1∶3 vol∶vol). One microliter of the mixture was placed on the target plate of a MALDI micro MX mass spectrometer (Waters, Manchester, UK) for analysis.
Virus
Dengue-1/Hawaii, dengue-2/PL046, dengue-3/H-087, and dengue-4/H241 were used for this study. The virus was laboratory-adapted virus and kindly provided by Yi-Ling Lin of the Institute of Biomedical Sciences, Academia Sinica, Taiwan [26] , [27] . The virus was propagated in C6/36 cells, and viral titers were determined by focus-forming assays with BHK-21 cells. Briefly, a monolayer of BHK-21 cells in 24-well plates was inoculated with supernatant obtained from C6/36 cultured medium infected with dengue virus. Supernatants were diluted by 10-fold serial dilution (starting at 1∶10). Viral adsorption was allowed to proceed for 3 h at 37°C. An overlay medium containing 2% fetal bovine serum and 0.8% methylcellulose in DMEM was added at the conclusion of adsorption. The infected monolayer was incubated at 37°C. After 72 h of infection, the overlay medium was removed from the wells, and the BHK cells were washed with cold PBS. The cells were fixed for 15 min in 3.7% formaldehyde/PBS. After washing with PBS, the cells were permeabilized with 0.1% Nonidet P-40/PBS for 15 min and blocked with 3% BSA/PBS for 30 min. Infected cells were detected with a monoclonal anti-dengue antibody (American Type Culture Collection, No. HB-114). After washing with PBS, antibody-labeled cells were detected with a secondary antibody conjugated to HRP. The labeling was visualized with 3,3′,5,5′-tetramethylbenzidine (TMB). The focus-forming units (FFUs) were counted, and the viral titers were determined by times dilution factor.
Mouse experiments
Five BALB/c mice (6–8 weeks of age) were immunized subcutaneously with recombinant D2ED III or LD2ED III. The lyophilized D2ED III and LD2ED III were reconstituted with PBS. Each mouse was received 10 µg/0.2 mL per dose. Mice were given 2 immunizations at a 2-week interval with the same regimen. This immunization protocol was used throughout the present study. Mice were inoculated intraperitoneally with live dengue-2 virus (1×10 7 FFUs) on the same schedule for comparison with the D2ED III and LD2ED III vaccine candidates. Blood was collected from each mouse at different time points as indicated. Sera were prepared and stored at −20°C until use.
Measurement of antibody titer
The levels of anti-D2ED III IgG in the serum samples were determined by titrating the samples. Sera were diluted by 3-fold serial dilution (starting at 1∶33). Briefly, purified D2ED III was coated onto 96-well plates. Bound IgG was detected with HRP-conjugated goat anti-mouse IgG Fc. After the addition of TMB, the absorbance was measured with an ELISA reader at 450 nm. ELISA end-point titers were defined as the serum dilution that produced an OD value of 0.5. The serum dilution was obtained from the titration curve by interpolation, unless the OD value was less than 0.5 at the starting dilution (1∶33).
Measurement of antibody avidity
Antibody avidity was determined on the basis of D2ED III-specific IgG dissociation induced by the chaotropic agent ammonium thiocyanate. Briefly, purified D2ED III was coated onto 96-well plates. After blocking with 1% bovine serum albumin (BSA)/PBS, serum at a dilution of either 1∶100 or 1∶300 was incubated at room temperature for 1 h. The plates were washed and incubated with 0–3.5 M ammonium thiocyanate in 0.5 M increments at room temperature for 15 min. The bound IgG was detected with HRP-conjugated goat anti-mouse IgG. After the addition of TMB, the absorbance at 450 nm was measured with an ELISA reader. The avidity index was calculated as the concentration of ammonium thiocyanate that resulted in a 50% decrease in the initial absorbance [28] , [29] .
Focus reduction neutralization tests (FRNT)
Sera were diluted via 2-fold serial dilutions (starting at 1∶8), and the sera were heat-inactivated prior to testing. A monolayer of BHK-21 cells in 24-well plates was inoculated with dengue-2 virus that had been pre-mixed at 4°C overnight with preimmunization or postimmunization sera to a final volume of 0.5 ml. The virus titer prior to pre-mixing was approximately 20–40 FFU per well. The FFUs were obtained as described previously, and the neutralizing antibody titer FRNT 50 was calculated as the reciprocal of the highest dilution that produced a 50% reduction in FFU compared with control samples containing the virus alone. For calculation purposes, the neutralizing antibody titer was designated as 4 when the neutralizing antibody titer was less than 8.
Antibody-dependent enhancement tests
Antibody-mediated enhancement of dengue virus infectivity was determined by flow cytometry in K562 cells. Sera were diluted via 4-fold serial dilutions (starting at 1∶8), and the sera were heat-inactivated prior to testing. Serially diluted sera and virus were mixed and incubated to form immune complexes for 1 h at 37°C. K562 cells were mixed with immune complexes (MOI = 0.1) and then incubated for 1.5 h at 37°C. After washing, the cells were resuspended in fresh medium and incubated for 3 days at 37°C. Infections with and without virus were performed in parallel as controls. Cells were stained for intracellular with monoclonal anti-dengue antibodies (American Type Culture Collection, No. HB-114 for dengue-1, dengue-2, and dengue-4; HB-49 for dengue-3). Antibody-labeled cells were detected with a secondary antibody conjugated to FITC. The data were acquired with CellQuest Pro software on a BD FACSCalibur flow cytometer and were analyzed with FACS 3 software. The fold enhancement was defined as the percentage of infected cells in the presence of sera divided by the percentage of infected cells in the absence of sera.
Statistical analyses
Statistical analyses were performed with the ANOVA Bonferroni post test using GraphPad Prism version 5.02 (GraphPad Software, Inc.). Differences with a p value of less than 0.05 were considered statistically significant.
Results
Preparation of recombinant antigens containing dengue-2 envelope protein domain III
The D2ED III gene was cloned into the expression vector pET-22b(+) to produce the plasmids, pD2ED III and pLD2ED III. They were used for the production of recombinant antigens, D2ED III and lipidated D2ED III (LD2ED III), respectively. Both antigens contained an additional hexahistidine sequence (His-tag) at their C-termini and were expressed under the control of the T7 promoter ( Figure 1A ).
10.1371/journal.pntd.0002432.g001 Figure 1
Production and identification of recombinant D2ED III and LD2ED III.
(A) The amino acid sequence of D2ED III is a consensus sequence of dengue virus type II (reference [25] ). The DNA sequence encoding D2ED III was optimized for E. coli codon usage and fully synthesized by a biotechnology company (see Materials and Methods ). The PCR product was cloned into a pET22b-based vector to generate the expression plasmid pD2ED III for the production of D2ED III. To clone LD2ED III, pD2ED III was cloned into the pET22b-based vector with a lipid signal peptide in front of the D2ED III gene to generate pLD2ED III for LD2ED III production. Both recombinant proteins contained an additional hexahistidine sequence (His-tag) at their C-termini and were expressed under the control of the T7 promoter. (B) The purification of the D2ED III protein was monitored by 15% reducing SDS-PAGE followed by Coomassie Blue staining and immunoblotting with anti-His-tag antibodies (lanes 1 to 8). D2ED III was expressed in E. coli BL21 (DE3). Lane 1, D2ED III expression after IPTG induction; lane 2, protein expression in the absence of IPTG induction; lane 3, extraction of D2ED III from inclusion body; lane 4, purified D2ED III. Lanes 5–8 show immunoblotting to monitor the D2ED III induction and purification processes, and the samples in these lanes are the same as those in lanes 1–4, respectively. The arrows indicate the electrophoretic positions of D2ED III in the gels or blots. LD2ED III purification was monitored by 15% reducing SDS-PAGE followed by Coomassie Blue staining and immunoblotting with anti-His-tag antibodies (lanes 9 to 16). LD2ED III was expressed in E. coli C43 (DE3). Lane 9, LD2ED III expression after IPTG induction; lane 10, protein expression in the absence of IPTG induction; lane 11, soluble fraction of LD2ED III; lane 12, purified LD2ED III. Lanes 13–16 show immunoblotting to monitor the LD2ED III induction and purification processes, and the samples in these lanes are the same as those in lanes 9–12, respectively. The arrows indicate the electrophoretic positions of LD2ED III in the gels or blots. (C) N-terminal LD2ED III fragments were obtained and identified after digestion of LD2ED III with trypsin. The digested sample was analyzed with a MALDI-TOF mass spectrometer. The MALDI-TOF MS spectra included three major peaks with m/z values of 1452, 1466, and 1480.
The purification of D2ED III and LD2ED III were monitored and analyzed by SDS-PAGE and immunoblotting ( Figure 1B ). After removing LPS, the residual LPS in D2ED III and LD2ED III were less than 0.06 EU/mg. The yields of D2ED III and LD2ED III were 40 mg/l and 8 mg/l, respectively. The immunogenicity and efficacy of endotoxin-free D2ED III and LD2ED III were comparatively analyzed in animal models.
We then measured the exact mass of trypsin-digested N-terminal fragments of LD2ED III. Three major peaks with m/z values of 1452, 1466, and 1480 were identified ( Figure 1C ). These peaks have been previously identified as a lipidation signature in other lipidated proteins [24] . We confirmed that the peaks of LD2ED III were associated with lipidated cysteine residues and verified that LD2ED III contained an N-acetyl-S-diacyl-glyceryl-cysteine at its N-terminus [30] .
Lipidated dengue-2 envelope protein domain III immunization induces durable neutralizing antibody responses
The immunogenicity of the purified D2ED III and LD2ED III was tested in mice. Groups of BALB/c mice were immunized with D2ED III or LD2ED III (10 µg per dose) two times with a two-week interval between immunizations. Animals infected with live dengue-2 virus (1×10 7 FFU) on the same schedule served as controls. Serum samples were collected from the immunized mice at different time points, as indicated in Figure 2 . Antibody responses against D2ED III were only detected in D2ED III- or dengue-2 virus-immunized mice after two vaccinations, 4 weeks post-priming ( Figure 2A ). By contrast, the recombinant LD2ED III was highly immunogenic and induced stronger antibody responses than D2ED III (p<0.05 by the ANOVA Bonferroni post test). Mice immunized with LD2ED III quickly elicited antibody titers 2 weeks post-priming. Antibody titers were further elevated following a booster immunization and were maintained for at least 20 weeks after the initial priming ( Figure 2A ).
10.1371/journal.pntd.0002432.g002 Figure 2
Humoral immune responses induced by vaccine candidates.
Groups of BALB/c mice (n = 5) were immunized subcutaneously with 10 µg D2ED III or LD2ED III two times at a two-week interval. Live dengue-2 virus (1×10 7 FFU) was injected intraperitoneally on the same schedule. Sera were collected from mice at the indicated time points after the first immunization. (A) IgG antibodies against D2ED III were evaluated by ELISA. Preimmune sera (week 0) were collected and used to determine basal levels for comparison. (B) Antibody avidity profiles were examined by ELISA. One of two representative experiments is shown. The results are expressed as the mean ± standard deviation. Statistical significance was determined by the ANOVA Bonferroni post test. ** p <0.01; *** p <0.001.
The antibody avidity profiles of serum samples collected from different groups at weeks 2 and 6 were examined. The avidity index was calculated as the concentration of ammonium thiocyanate that resulted in a 50% decrease in the initial absorbance [28] , [29] . As shown in Figure 2B , the avidity index of mice immunized with LD2ED III was 0.41±0.06 at 2 weeks after priming. However, mice immunized a single time with D2ED III or dengue-2 virus did not generate significant antibody responses for avidity analysis. After booster immunization, the avidity indexes of mice immunized with D2ED III and dengue-2 virus were 0.50±0.11 and 0.49±0.10 at week 6, respectively. Remarkably, the avidity index of mice immunized with LD2ED III increased to 0.82±0.15 by week 6 and was significantly higher than the avidity index of mice immunized with D2ED III or dengue-2 virus.
Next, we evaluated the neutralizing capacity of the antibodies induced by vaccination. As shown in Figure 3 , mice immunized with D2ED III could not stimulate significant neutralizing antibody responses (FRNT 50 = 5) at 2 weeks after priming. Even after booster immunization, the neutralizing antibody titers were 9 and 16 at 4 and 20 weeks after priming, respectively. Dengue-2 virus-infected mice generated low neutralizing antibody titers (FRNT 50 = 11) at 2 weeks after primary infection. After secondary infection, neutralizing antibody titers increased and reached 37 and 223 at 4 and 20 weeks after primary infection, respectively. Notably, mice immunized with LD2ED III generated significant neutralizing antibody responses (FRNT 50 = 84) at 2 weeks after priming. After booster immunization, the neutralizing antibody titers were further elevated to 588 and 1176 at 4 and 20 weeks after priming, respectively. These results suggest that mice immunized with LD2ED III without exogenous adjuvant elicit quick and durable neutralizing antibody responses.
10.1371/journal.pntd.0002432.g003 Figure 3
Neutralizing antibody titers in mice immunized with vaccine candidates.
Groups of BALB/c mice (n = 5) were immunized subcutaneously with 10 µg D2ED III or LD2ED III two times at a two-week interval. Live dengue-2 virus (1×10 7 FFU) was injected intraperitoneally on the same schedule. Sera were collected from mice at the indicated time points after the first immunization. The dengue-2 virus neutralizing capacity was determined by FRNT. The neutralizing antibody titer was calculated as the reciprocal of the highest dilution that resulted in a 50% reduction in FFU compared to control samples containing the virus alone. One of two representative experiments is shown. Statistical significance was determined by the ANOVA Bonferroni post test. * p <0.05; ** p <0.01; *** p <0.001.
Reduction of the potential risk of antibody-dependent enhancement of infection
Antibody-dependent enhancement of infection is a significant concern in the development of vaccines against dengue virus. Therefore, we measured the capacities of vaccine candidates to mediate antibody-dependent enhancement of infection. We employed K562 cells, which have been widely used for the measurement of dengue virus antibody-dependent enhancement of infection. The results are shown in Figure 4 . Serum samples obtained from mice immunized with dengue-2 virus possessed tremendous antibody-dependent enhancement capacities for heterotypic viral infection in K562 cells. The peak fold enhancement values were 442.5±389.1, 28.0±19.1, and 93.1±86.4 for dengue-1, dengue-3, and dengue-4, respectively, at the dilution 1/8–1/128. By contrast, antibodies generated from D2ED III-immunized mice did not promote antibody-dependent enhancement of heterotypic viral infection in K562 cells. Serum samples obtained from mice immunized with LD2ED III displayed minor antibody-dependent enhancement of heterotypic viral infection in K562 cells at the lowest dilution tested (1/8). The fold enhancement values were 10.1±7.1, 4.0±2.2, and 16.5±16.7 for dengue-1, dengue-3, and dengue-4, respectively, which were notably lower than the values for the serum samples obtained from mice immunized with dengue-2 virus. Compared with heterotypic viral infection in K562 cells, only low antibody-dependent enhancement capacities were observed for homotypic viral infection in K562 cells. The peak fold enhancement values were 3.1±0.7, 3.4±0.8, and 2.5±0.5 for serum samples obtained from mice immunized with dengue-2 virus, D2ED III, and LD2ED III, respectively. These results suggest that antibodies elicited by LD2ED III have less capacity for antibody-dependent enhancement than antibodies elicited by dengue-2 viral infection.
10.1371/journal.pntd.0002432.g004 Figure 4
Capacity for antibody-dependent enhancement in mice immunized with vaccine candidates.
Groups of BALB/c mice (n = 5) were immunized subcutaneously with 10 µg D2ED III or LD2ED III two times at a two-week interval. Live dengue-2 virus (1×10 7 FFU) was injected intraperitoneally on the same schedule. Sera were collected from mice at 6 weeks after the first immunization. K562 cells were infected (MOI = 0.1) with (A) dengue-1, (B) dengue-3, (C) dengue-4, or (D) dengue-2 viruses in the presence or absence of serial 4-fold dilutions of pooled serum samples. The fold enhancement values were calculated as the percentage of infected cells in the presence of pooled serum samples divided by the percentage of infected cells in the absence of pooled serum samples. The results are expressed as the mean ± standard deviation from triplicate. Statistical significance was determined by the ANOVA Bonferroni post test at the same dilution. * p <0.05 compared to D2ED III. # p <0.05 compared to LD2ED III.
Discussion
The development of novel subunit vaccines relies on a limited number of individual components, namely antigens of the specific pathogen under study. Importantly, the selected antigens must elicit protective immunity against the pathogen. To augment the rational design of subunit vaccines, we expressed LD2ED III as a dengue vaccine candidate using an E. coli -based expression system ( Figure 1 ). D2ED III of LD2ED III served as the antigenic component, and the lipid moiety of LD2ED III provided a danger signal that activated the immune system to induce an appropriate adaptive immune response. In the present study, we demonstrated that LD2ED III alone, without exogenous adjuvant, elicited higher D2ED III-specific antibody responses than D2ED III or dengue-2 virus ( Figure 2A ). In addition, the avidity ( Figure 2B ) and neutralizing capacity ( Figure 3 ) of the antibodies induced by LD2ED III were higher than those elicited by D2ED III or dengue-2 virus. The above properties would be beneficial to a host during dengue virus infection and suggest that LD2ED III could be a potential dengue vaccine candidate.
The role of antibodies in controlling dengue virus infection is complex. Antibodies are thought to mediate both neutralization and enhancement of dengue virus infection [31] , [32] . Antibody-dependent enhancement is the leading theory to explain the higher risk of DHF associated with heterologous serotype viral infections [8] . A reduction in the enhancement capacity of antibodies induced by vaccine candidates should increase the safety of dengue vaccines. Anti-ED III antibody titers in D2ED III-immune sera were comparable with those in dengue-2 virus immune sera ( Figure 2A ). Surprisingly, there was a remarkable capacity for ADE in dengue-2 virus-immune sera but little ADE capacity in D2ED III-immune sera ( Figure 4 ). Anti-ED III antibody titers in LD2ED III-immune sera were also significantly higher than those in dengue-2 virus-immune sera, and the ADE mediated by LD2ED III-immune sera in K562 cells was lower than that mediated by live virus-stimulated antibodies ( Figure 4 ). D2ED III and LD2ED III only induced anti-ED III antibody responses. However, dengue-2 virus induced antibodies against ED III and other viral antigens. These results suggest that anti-dengue virion antibodies other than ED III are the major antibodies that mediate ADE.
Although the exact mechanism of ADE is still not fully understood, it is believed that antibodies against envelope proteins either neutralize or enhance the viral infection, depending on the concentration and affinity of the antibodies [33] , [34] . As shown in Figure 4D , enhancement of dengue-2 with sera from mice vaccinated using infectious dengue-2 was observed. The peak fold enhancement was 3.1±0.7 at the dilution 1/512 - 1/2048. Neutralization was observed with antisera at the dilutions 1/8 - 1/32. Notably, a heterotypic enhancing response occurs at a wide range of serum concentrations ( Figure 4A–C ). These are general phenomena that a homotypic enhancing response are usually restricted to higher serum dilutions due to high neutralizing capacities, while a heterotypic enhancing response occurs at a wide range of serum dilutions because of little heterotypic neutralizing capacities. Similar profiles were observed in sera obtained from LD2ED III immunized mice for homotypic enhancement, the peak fold enhancement was 2.5±0.5 at the dilution 1/2048 - 1/8192 and neutralization was observed with antisera at the dilutions 1/8 - 1/128 ( Figure 4D ). Most importantly, sera obtained from LD2ED III immunized mice have less enhancement capacities for heterotypic virus than sera from mice vaccinated using infectious dengue-2 ( Figure 4A–C ). In contrast, D2ED III immunized mice did not elicit significant neutralizing antibodies. The enhancement infection was observed at the sera dilutions 1/8 - 1/32 ( Figure 4D ). These results suggest that LD2ED III is a good vaccine candidate with low risk of antibody-dependent enhancement. Recently, Mady et al. demonstrated that the delivery of dengue virus to the cell surface at a location other than Fc receptors by a bispecific antibody can also increase viral infectivity [35] . Furthermore, Huang et al. observed that anti-prM antibodies were cross-reactive with heat shock protein 60, which enhanced dengue virion binding and infection of cells lacking Fc receptors [9] . LD2ED III induced high-affinity antibody responses ( Figure 2B ) while providing only the ED III antigen. These properties could partly explain the notably lower ADE capacity of LD2ED III compared to dengue-2 virus.
Some conserved motifs located in the dengue envelope protein domain II and non-structural protein-1 have been shown to induce autoantibodies. Cross-reaction of dengue viral protein-induced antibodies with host antigens can trigger cell damage or induce harmful effects, which may facilitate DHF/DSS development [9] – [12] . Indeed, such autoantibodies were detected in DHF/DSS patients [10] , [36] . All dengue viral antigens are absent from the LD2ED III candidate with the exception of ED III, which may not induce cross-reactive antibodies. Taken together, these results suggest that LD2ED III is a safe vaccine candidate in terms of its reduced ADE capacity and autoantibody induction of LD2ED III.
The results of the current study suggest that the use of lipidated ED III from the four serotypes of dengue virus may have potential for the development of tetravalent dengue vaccines. Alternatively, the strategy of priming with live attenuated dengue vaccines followed by boosting with a lipidated ED III vaccine candidate may enhance ED III-specific immune responses to elicit safe and effective immunity against dengue virus infection. In conclusion, LD2ED III is an effective dengue vaccine candidate for inducing long-lasting neutralizing antibody responses with a low risk of detrimental effects. Future work should examine the suitability of this candidate for clinical use.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Since identification of the human immunodeficiency virus (HIV-1) as the causative agent for acquired immunodeficiency syndrome (AIDS), more than 70 million people have been infected, and an estimated 36 million people live with HIV-1 today [ 1 ]. Basic science advances in the understanding of HIV-1 have occurred at an unprecedented pace, allowing the development of numerous antiretroviral (ARV) agents, and advances in clinical science have determined optimal ways of using these drugs to reduce the morbidity and mortality associated with HIV-1 infection. Notwithstanding these impressive successes in the management of HIV-1, individuals who receive effective ARV therapy nonetheless have excess mortality compared to HIV-1 negative populations, due to the effects of inflammation and accelerated aging, manifested as increased risk of cardiovascular, metabolic, and malignant diseases [ 2 ]. Thus, a cure for HIV-1 infection is needed [ 3 ].
To date, only 1 case, known as the “Berlin patient,” has been cured of HIV-1 [ 4 ] by total myeloablative chemotherapy and total body irradiation treatment for acute myeloid leukemia, followed by 2 allogeneic stem cell transplants using cells from a donor who was homozygous for CCR5 Δ32, rendering the donor cells resistant to R5-tropic HIV-1 infection. Now, more than 10 years after stopping his anti-HIV medications, the Berlin patient remains free from viral rebound, and ultrasensitive assays have repeatedly failed to detect definitive evidence of viral persistence [ 5 ].
Unfortunately, other cases of HIV-1-infected patients undergoing allogeneic stem cell transplants (with cells from CCR5 wild-type donors) have not had durable remissions from HIV-1 rebound following analytic treatment interruption (ATI), with 2 such cases experiencing viral rebound at 12 and 32 weeks post-ATI [ 6 ]. Most likely, allogeneic stem cell transplantation (allo-SCT) in these 2 patients significantly decreased, but did not fully eliminate, latently HIV-infected cells, so that viral rebound ignited by persisting viral reservoirs ultimately occurred. However, this interpretation does not exclude the possibility that allogeneic hematopoietic stem cell transplants may, at least in certain cases, induce a more profound or near-complete elimination of viral reservoirs, to enable a long-term drug-free remission of HIV-1 infection.
To explore that possibility, we took advantage of the opportunity to study viral and immune dynamics in an HIV-1 positive patient who, following treatment with prolonged suppressive ARV therapy, developed acute lymphoblastic leukemia and underwent allo-SCT with concurrent ARV therapy. Herein we report a comprehensive analysis of viral and immune parameters occurring after allo-SCT, before and after an ATI.
Methods
Study participant
A formal prospective analysis plan was not in place for this study prior to onset. Study visits were determined by routine clinical care, and acquisition of research samples followed previously approved protocols as follows. Following informed consent, and Mayo Clinic Institutional Review Board approval (protocol number 13–005646), the patient underwent leukapheresis on day −11 pre-transplant and days +142, +265, and +888 post-transplant. Leukapheresis was performed on Fenwal Amicus apheresis systems (version 3.1; Fenwal, Lake Zurich, IL, US) using peripheral venous access. ATI was performed under IRB protocol number 15–001678. The patient provided verbal informed consent to reporting and publication of his case history.
Peripheral CD4 T cell counts were measured as previously described [ 7 ]. Plasma HIV-1 viral load was measured using the COBAS AmpliPrep/COBAS TaqMan HIV-1 Test, version 2.0 (Roche Molecular Systems, Branchburg, NJ, US). Clinical HIV-1 proviral DNA was measured using the Amplicor HIV-1 DNA Test, version 1.5 (Roche Diagnostics, Indianapolis, IN, US). Presence of HIV-1 antibodies in serum was confirmed using GS HIV-1 Western Blot (Bio-Rad Laboratories, Redmond, WA, US).
Quantitative viral outgrowth assays
Resting CD4 T cells were isolated by negative selection and verified to be CD4 + , CCR7 + , CD27 + , CD8 − , CD25 − , HLA-DR − , and CD11b − . Quantitative viral outgrowth assays (QVOAs) were then performed as previously described [ 8 ]. Cultures were analyzed on day 15 for the presence of p24 in the culture supernatant, and the frequency of infection of resting CD4 T cells was determined using maximum likelihood estimates (expressed as number of infectious units per million resting CD4 T cells, similar to what we and others have previously reported).
Total HIV DNA and RNA by quantitative PCR
Total HIV-1 DNA in CD4 T cells was measured by real-time PCR as previously described [ 9 ]. Cell-associated HIV-1 RNA was measured using the Roche AmpliPrep Kit (detection limit is 20 copies). RNA was isolated using the RNeasy Mini Kit (Qiagen, Venlo, Netherlands) per manufacturer’s protocol.
Total HIV-1 DNA by droplet digital PCR
Droplet digital PCR (ddPCR) analysis was performed as previously described [ 10 ]. Briefly, peripheral blood mononuclear cell (PBMC) DNA was extracted using the Qiagen QIAamp DNA Blood Maxi Kit according to manufacturer’s instructions. Approximately 24,000 ng of total PBMC DNA was assayed by HIV-1 gag [ 11 ] and HIV-1 pol [ 12 ] duplex ddPCR using previously published TaqMan assays and Bio-Rad QX200 reagents. RPP30 DNA cell normalizer was measured in a separate ddPCR reaction. ddPCR droplet data were acquired and analyzed by Bio-Rad QuantaSoft software and are expressed as HIV copies per 1 million cells; limits of detection and 95% confidence intervals were calculated based on Poisson statistics and total number of droplet events analyzed. In total, 24 HIV-1 DNA ddPCR replicates, with approximately 1,000 ng per ddPCR replicate, and >250,000 droplet events were analyzed.
Integrated HIV-1 DNA in CD4 T cell subsets
CD4 T cell subsets (e.g., TN, TCM, TTM, and TEM) were sorted based on the expression of CD45RA, CCR7, and CD27, as described previously [ 13 ]. Sorted cells were subjected to proteinase K digestion, and the frequency of cells harboring integrated HIV-1 DNA was determined as previously described [ 14 ].
In situ hybridization
After deparaffinization with xylene, and rehydration through graded ethanols, tissue sections were treated with HCl, triethanolamine, digitonin, and 4 mcg/ml proteinase K, as previously described [ 15 ]. After acetylation with acetic anhydride and dehydration, tissue sections were hybridized at 45°C overnight with a 35 S-labeled riboprobe and 0.5 mM aurintricarboxylic acid in the hybridization mix. After extensive washes and ribonuclease treatment, tissue sections were dehydrated, coated in Kodak NTB emulsion diluted with 10% glycerol and 0.1 M ammonium acetate, exposed at 4°C for 7–14 days, and developed and fixed as previously described [ 15 ].
Microchimerism
Highly sensitive allele-specific PCR assays targeting HLA and insertion–deletion polymorphisms unique to the patient or donor were used to determine levels of host microchimerism in blood (the proportion of residual host PBMCs after hematopoietic stem cell transplantation), as previously described [ 16 , 17 ]. The microchimerism assay is highly specific and sensitive to a single copy of target DNA, allowing detection of host cells present as a very low proportion of the PBMC population, depending on the number of cells surveyed [ 17 ].
Immunophenotyping by flow cytometry
PBMCs were stained with selected monoclonal antibodies labeled with defined combinations of fluorescent dyes. Cells were then washed, fixed, washed again, and analyzed on a Fortessa flow cytometer, using standard protocols. Data were analyzed using FlowJo software (Treestar, Ashland, OR, US).
HIV antibody measurement
The gp41-detecting Limiting Antigen (LAg)–Avidity EIA (Sedia Biosciences, Portland, OR, US) was performed as previously described [ 18 , 19 ]. In brief, assay controls and HIV-positive specimens were diluted 1:101 in specimen diluent, and 100 μl of calibrator, controls, or specimens was added to antigen-coated plates and incubated. Plates were washed 4 times with 1× wash buffer to remove unbound antibodies. A pH 3.0 buffer was added to each well to dissociate low-avidity antibodies. Plates were developed, and the optical density (OD) was read using a spectrophotometer (microplate reader; Molecular Devices, Sunnyvale, CA, US). Raw OD for each specimen was normalized using the calibrator OD on each plate as a ratio, such that normalized OD = OD of specimen/median OD of calibrator.
Viral sequencing
Genomic DNA was extracted from indicated cell populations using the Qiagen DNeasy Blood & Tissue Kit and diluted to single-genome levels based on Poisson distribution statistics of HIV-1 gag amplification results. Subsequently, single-genome viral gene amplification was performed using Invitrogen Platinum Taq (Invitrogen, Carlsbad, CA, US) and nested primers spanning near full-length HIV-1 (HXB2 positions 638–9632). Primers were previously published [ 20 ] except for a modified nested forward primer: 5′-GCGCCCGAACAGGGACYTGAAARCGAAAG-3′. PCR products were visualized by agarose gel electrophoresis and subjected to Illumina MiSeq sequencing. Resulting short reads were de novo assembled and aligned to HXB2. Integrity of full-length sequences was determined using an automated in-house pipeline written in R scripting language [ 21 ]. Presence/absence of APOBEC-3G/3F-associated hypermutations was determined using Los Alamos HIV Sequence Database Hypermut 2.0 [ 22 ]. Multiple sequence alignments were performed using MUSCLE [ 23 ]. Genetic distances between sequences were examined using Clustal X–generated neighbor joining algorithms [ 24 ]. For the analysis of plasma HIV-1 sequences, plasma HIV-1 RNA was transcribed to cDNA using standard procedures, diluted to single genomes, and subjected to nested PCR with primers annealing to env (first round primers: 5′-CACCGGCTTAGGCATCTCCTATGGCAGGAAGAA-3′ and 5′- CATTGGTCTTAAAGGTACCTGAGG-3′; second round primers: 5′-AGAAAGAGCAGAAGACAGTGGCAATGA-3′ and 5′-TTTTGACCACTTGCCACCCAT-3′) and pol (first round primers: 5′-TGTACTGAGAGACAGGCTAATTTTT-3′ and 5′-AAACTCCCACTCAGGAATCCAGGT-3′; second round primers: 5′-AGACAGGCTAATTTTTTAGGGAAGAT-3′ and 5′-CACTCAGGAATCCAGGTGGCTT-3′). Subsequently, PCR products were processed by Sanger sequencing; sequence alignments were performed using MUSCLE.
Results
In June 2013, a 55-year-old HIV-1 positive man was referred to Mayo Clinic for evaluation of B-lineage acute lymphoblastic leukemia. Pre-transplant HIV-1 history is described in Table 1 . Briefly, he had been first diagnosed with HIV-1 infection in 1990 and believed his infection occurred in 1982. At the time of diagnosis, his CD4 count was >500 cells/μl (reference range 365–1,437), his plasma HIV-1 RNA viral load was approximately 400 copies/ml, and he did not receive ARV therapy. In 1999, when his CD4 count had declined to approximately 300 cells/μl and his HIV-1 viral load had increased to 10,000 copies/ml, he was started on ritonavir-boosted indinavir and zidovudine/lamivudine. In 2004, he took a drug holiday. When his HIV-1 viral load had increased to approximately 10,000 copies/ml in 2009, he initiated ritonavir-boosted atazanavir and tenofovir/emtricitabine. His regimen was changed to raltegravir and tenofovir/emtricitabine in April 2013 to avoid potential drug–drug interactions with anticipated chemotherapy (as noted below). He tolerated these medications with excellent adherence, and at the time of presentation for leukemia evaluation, his anti-HIV-1 Western blot was positive, with a plasma HIV-1 viral load of 107 copies/ml and a CD4 count of 293 cells/μl (37% of CD3 + cells).
10.1371/journal.pmed.1002461.t001
Table 1 Pre-transplant HIV laboratory test results and antiretroviral treatment history.
Time point
CD4 T cell count (cells/μl)
HIV-1 RNA (copies/ml)
HIV therapy
1990 (HIV diagnosis)
>500
400
None
1999
300
10,000
AZT/3TC, IDV/rtv
2004
>500
Undetectable
Therapy stopped
2009
>500
10,000
TDF/FTC ATV/rtv
6/2013
293
107
TDF/FTC, raltegravir
9/2013 (began chemotherapy for leukemia)
183
25
TDF/FTC, raltegravir, etravirine
10/2013 (allogeneic stem cell transplant on Oct 6)
Detected
TDF/FTC, raltegravir, etravirine
3TC, lamivudine; ATV, atazanavir; AZT, zidovudine; FTC, emtricitabine; IDV, indinavir; rtv, ritonavir; TDF, tenofovir.
In March 2013, he experienced the insidious onset of progressive light-headedness and fatigue, associated with a white blood cell count of 3,400 cells/mm 3 (reference range, 4,500–11,000) with 52% circulating blasts, hemoglobin 80 g/l (reference range, 135–175), and platelets of 47,000 cells/mm 3 (reference range, 140,000–440,000). A bone marrow biopsy was 90% cellular with 96% blasts. Flow cytometry showed multiple B-lineage markers including CD20, CD79a, and intranuclear terminal deoxynucleotidyl transferase. Cytogenetics were normal, and fluorescence in situ hybridizations for BCR-ABL and MLL rearrangement were both negative. Cerebrospinal fluid examination showed leukemic blasts, and a CT scan showed mild mediastinal adenopathy up to 14 mm in short-axis dimension, while the spleen was enlarged, at craniocaudal height of 17 cm. The patient received treatment with rituximab and hyper-CVAD (cyclophosphamide, vincristine, doxorubicin, dexamethasone) alternating with high-dose methotrexate and cytarabine beginning in April 2013. Repeat bone marrow biopsy in May 2013 showed 4% circulating blasts and 9% bone marrow blasts, prompting a third cycle of rituximab and hyper-CVAD; a repeat bone marrow biopsy in July was normocellular, with a cellularity of 40%. No morphologic features of acute leukemia were noted. In August 2013, he underwent a fourth cycle of hyper-CVAD with high-dose methotrexate and cytarabine.
In an effort to attain full HIV viral suppression, the patient’s ART regimen was intensified due to persistent low-level viremia (plasma HIV-1 viral load ranging from 90 to 107 copies/ml). In anticipation of myeloablative chemotherapy, his ARV regimen was modified to include etravirine 200 mg twice daily in addition to his current ART regimen of raltegravir and co-formulated tenofovir/emtricitabine. Due to a history of prior gastric bypass, concern for poor drug absorption in the setting of his low-level viremia, and reported decreased raltegravir exposure while on etravirine [ 25 ], the raltegravir dose was empirically increased to 600 mg twice daily, where satisfactory 3-hour peak levels were demonstrated (1.11 μg/ml; reference range, 0.67–3.54 μg/ml).
In October 2013, the patient underwent a fludarabine/melphalan reduced-intensity conditioning treatment prior to an HLA-matched, ABO-matched allogeneic peripheral blood stem cell transplant with infusion of 4.47 × 10 6 CD34 + cells/kg from a CCR5 wild-type donor. Donor and recipient characteristics are listed in Table 2 . The patient was placed on tacrolimus and full-dose methotrexate for graft-versus-host disease (GVHD) prophylaxis, and acyclovir, atovaquone, and anidulafungin antimicrobial prophylaxis, with plans to restart trimethoprim/sulfamethoxazole prophylaxis after engraftment, and voriconazole after liver function tests normalized. The patient remained on stable uninterrupted ARVs after peripheral blood stem cell transplantation (PBSCT).
10.1371/journal.pmed.1002461.t002
Table 2 Donor and recipient characteristics prior to allogeneic stem cell transplantation.
Characteristic
Donor
Recipient
HLA type
A*03,24
A*03,24
B*07,27
B*07,27
Cw*02,07
Cw*02,07
DRB1*04,04
DRB1*04,04
DRw*53,53
DRw*53,53
DRB4*01,01
DRB4*01,01
DQ*07,08
DQ*07,08
DQB1*03,03
DQB1*03,03
CCR5 genotype
Wild-type
Wild-type
Cytomegalovirus (IgG)
Positive
Positive
Epstein–Barr virus (IgG)
NA
Positive
Toxoplasma (IgG)
NA
Positive
NA, not available.
In January 2014, the patient discontinued his GVHD prophylaxis, and in February 2014, he developed progressive diarrhea, which prompted a diagnostic colonoscopy (day +133 post-transplant); biopsy and pathology revealed mildly increased crypt cell apoptosis in the colon and ileum (consistent with GVHD), and in situ stains for Epstein–Barr virus, adenovirus, and cytomegalovirus were negative. The patient was treated with loperamide for symptomatic management. In March 2014, trimethoprim/sulfamethoxazole was stopped because of low platelets, and in May 2014, the patient was admitted with fever and shortness of breath, and was diagnosed with Pneumocystis jirovecii pneumonia, which was treated with high dose trimethoprim/sulfamethoxazole.
An ATI was started according to an IRB-approved protocol on day +784 post-transplantation (1 December 2015) ( Fig 1A ). HIV-1 remained persistently undetectable by multiple measures (see below). Plasma HIV-1 RNA was monitored every 2 weeks for 12 weeks, then every 4 weeks thereafter, and remained undetectable (limit of detection at 20 copies/ml, COBAS AmpliPrep/COBAS TaqMan HIV-1 Test, version 2.0). However, at day 288 of the ATI, the patient experienced asymptomatic viral rebound, with a plasma HIV-1 RNA viral load of 60 copies/ml. The plasma HIV-1 RNA viral load rose to 283 copies/ml on ATI day 289 and to 1,640 copies/ml on ATI day 293, prompting reinstitution of ARV therapy, according to the clinical ATI protocol. Resistance testing by viral genotype revealed no mutations associated with ARV drug resistance. The patient denied risk factors for new HIV exposures. Reinstitution of ARV therapy resulted in suppression of detectable viral replication after 4 weeks, and the patient’s hematologic malignancy remains in full remission at the time of this publication.
10.1371/journal.pmed.1002461.g001
Fig 1
HIV-1 RNA monitoring in the peri-transplant period.
(A) CD4 T cell count and plasma HIV-1 RNA were measured in the pre- and post-transplant period. (B) Cell-associated HIV-1 RNA was measured in isolated CD4 T cells from leukapheresis samples on the days indicated. ATI, analytic treatment interruption.
HIV-1 RNA monitoring
HIV-1 RNA was detectable in plasma at days −119, −35, +20, and +56 of the peri-transplant period (107, 25, <20, and <20 copies/ml, respectively; lower limit of detection 10 copies/ml and lower limit of quantification 20 copies/ml); beginning 91 days after transplant (day +91), plasma RNA remained undetectable until day +1,072, which was 288 days following initiation of the ATI ( Fig 1A ). Cell-associated HIV-1 RNA was measured in isolated CD4 T cells sampled pre- and post-transplant ( Fig 1B ). Cell-associated HIV-1 RNA increased at day +142 (515 copies/1.5 μg RNA) compared to day −11 (397.5 copies/1.5 μg RNA), but was reduced at day +888 (17 copies/1.5 μg RNA) during the ATI.
HIV-1 DNA monitoring
We next estimated the size of the HIV-1 reservoir over time by measuring HIV-1 DNA in CD4 T cells at multiple time points before and after transplantation using multiple methods. The patient underwent leukapheresis on day −11 pre-transplant and on days +142, +265, +436, and +888 post-transplant to provide cells for analysis.
Total HIV-1 DNA in CD4 T cells was measured by quantitative real-time PCR pre-transplant and on days +142 and +888 post-transplant ( Fig 2A ). Total HIV-1 DNA decreased from 722 copies/million CD4 T cells pre-transplant to 28 copies/million CD4 T cells at day +142 post-transplant, representing a 96% reduction in HIV-1 DNA. We questioned whether this reduction in HIV-1 DNA in PBMCs was merely secondary to dilution by virtue of replacing HIV-1-DNA-containing recipient cells with uninfected donor cells. Microchimerism evaluation revealed that approximately 8% of DNA in circulating CD4 T cells on day +142 was of recipient origin, indicating that HIV-1 total DNA may have decreased further than expected by hemodilution alone, possibly by preferential loss of HIV-1-DNA-containing cells. Microchimerism analysis of cells from day +265 revealed that 0.0013% of CD4 T cell DNA was of recipient origin. By day +888, total HIV-1 DNA was below the limit of detection (<5 copies/million cells).
10.1371/journal.pmed.1002461.g002
Fig 2
HIV-1 DNA monitoring in the peri-transplant period.
(A) Total HIV-1 DNA was measured by quantitative PCR (qPCR) in isolated CD4 T cells from the days indicated. (B) Integrated HIV-1 DNA was measured in sorted bulk CD4 T cells and CD4 T cell subsets from the days indicated. (C) Total HIV-1 DNA was measured by digital droplet PCR (ddPCR) in isolated CD4 T cells from the days indicated. Indicated are point estimates (bars show 95% confidence intervals of the estimates), including for results that were below the lower limit of detection (undetectable [UD]). ATI, analytic treatment interruption.
We further analyzed which CD4 T cell subsets contained residual HIV. In these studies we measured integrated HIV-1 DNA in sorted memory CD4 T cell subsets from day −11 and day +142 ( Fig 2B ). We observed a reduction in the frequency of cells harboring integrated HIV-1 DNA between pre-transplant and day +142 post-transplant in the 3 memory CD4 T cell subsets (central memory, transitional memory, and effector memory), with the greatest reduction occurring within the transitional memory CD4 T cell subset (from 277 copies/10 6 cells to undetectable).
Although most proviral DNA is not replication competent, measurement of HIV-1 DNA permits the largest dynamic range to quantify reduction in HIV-1 reservoir size [ 26 ]. We regularly measured HIV-1 DNA over time, using ddPCR and 2 different target primer pairs ( Fig 2C ). By day +265 post-transplant, HIV-1 DNA measured using gag primers decreased to undetectable, while HIV-1 DNA measured using pol primers remained detectable at low levels (30.3 copies/10 6 cells pre-transplant to 2.3 copies/10 6 cells at day +265 post-transplant). By day +436, HIV-1 pol amplification of HIV-1 DNA was undetectable, whereas HIV-1 gag amplification of HIV-1 DNA was barely detectable, at 2.7 copies/10 6 cells.
HIV-1 quantitative viral outgrowth assay monitoring
Of the multiple proposed ways to measure the HIV-1 reservoir size, nucleic amplification approaches have been criticized because they measure both replication-competent viruses as well as replication-incompetent defective viruses [ 27 ], which may represent the majority of the measured viruses [ 28 ]. Thus, we opted to measure replication-competent HIV-1 by QVOA on CD4 T cells [ 29 ]. For this assay, 35 × 10 6 CD4 T cells from day −11 pre-transplant were cultured at 5 × 10 6 cells per well and stimulated with αCD3/CD28 antibodies to reactivate any virus present, yielding 2 of 7 wells with detectable p24 antigen production, for an estimated 0.0673 infectious units per million cells (IUPM) ( Fig 3A ). A similar viral outgrowth assay using 85 × 10 6 CD4 T cells from day +142 post-transplant yielded 0 of 17 wells producing p24 antigen, for an estimated IUPM of <0.0121. Another viral outgrowth assay performed on cells from day +888 using >400 × 10 6 cells (while the patient was aviremic during ATI) yielded 0 positive wells, for an estimated IUPM of <0.00235, or less than 1 cell carrying replication-competent HIV-1 per 425 million CD4 T cells.
10.1371/journal.pmed.1002461.g003
Fig 3
HIV-1 reservoir measurement in the peri-transplant period.
(A) Replication-competent virus in isolated peripheral resting CD4 T cells was estimated by quantitative viral outgrowth assay (QVOA) on the days indicated. (B) In situ hybridization for HIV-1 DNA in colon tissue samples obtained on day +133 after transplantation. ATI, analytic treatment interruption.
HIV-1 In situ hybridization
We used in situ hybridization to asses HIV reservoir size in tissue sections from the diagnostic colon biopsy on day +133 that revealed GVHD. There were no HIV positive cells, and no follicular dendritic cell signal, in a total of 105 biopsy sections ( Fig 3B ).
HIV-1 sequencing
To more closely examine changes in residual viral reservoirs over time, we conducted single-genome sequencing assays of near full-length proviral HIV-1 DNA, using a recently described experimental approach [ 30 ]. Immediately prior to transplantation, we obtained a total of 23 proviral DNA sequences, of which 2 were sequence-intact, corresponding to a frequency of 0.4 intact, near full-length sequences per million PBMCs ( Fig 4A ). This is lower than previously reported in HIV-1 positive individuals undergoing suppressive ART during chronic infection [ 30 ]; however, the patient had previously received chemotherapy for his malignancy, which may account for this low number. Notably, these 2 intact viral sequences were identical, and likely derived from a single HIV-1-infected cell clone. Identical proviral sequences were also observed within the pool of defective viral DNA products, consistent with clonal expansion [ 30 – 32 ] of cells harboring replication-deficient proviral HIV-1 DNA ( Fig 4A ).
10.1371/journal.pmed.1002461.g004
Fig 4
Single-genome, near full-length HIV-1 sequencing in the patient.
(A) Diagram summarizing all HIV-1 DNA sequences retrieved from peripheral blood mononuclear cells (PBMCs) at indicated time points (T1: day −11; T2: day +144; T6: day +1078). Asterisks indicate clusters of completely identical proviral sequences. Color coding reflects presence of intact or defective sequences. (B) Phylogenetic tree including all near full-length sequences from indicated time points. Viral tropism (R5 versus nonR5) and geno2pheno false positive rate (FPR) percentage are included for each sequence. (C and D) Phylogenetic trees for viral env (HXB2 positions 6271–6889) and pol (HXB2 positions 2131–2780) sequences amplified from indicated plasma or PBMC samples collected at T1 or T6. A proviral sequence from the patient’s HIV-1-infected, ART-treated partner is also included. Plasma sequences are denoted in red text.
On day +142 after transplantation, we detected 2 hypermutated viral sequences after sampling 577,017 PBMCs, while no viral sequences were detected on days +265, +436, and +888 after transplantation, after analyzing a total of 22,250, 648,217, and 704,583 PBMCs, respectively. At day +1,087 after transplant, when plasma viral rebound was noted, we detected a single near full-length, intact proviral sequence in a total of 3,737,467 analyzed PBMCs, corresponding to an extremely low viral reservoir size in peripheral blood. No defective viral sequences were observed at this time. Interestingly, this viral sequence showed remarkable phylogenetic distance to proviral sequences isolated prior to transplantation and had predicted R5 tropism, while pre-transplant intact sequences were non-R5 tropic ( Fig 4B ). Correspondingly, we noted that env and pol sequences from rebounding plasma virus were phylogenetically closely related to the contemporaneous proviral DNA sequence, but exhibited considerable phylogenetic distance to proviral HIV-1 DNA detected prior to transplantation ( Fig 4C and 4D ). A sequence of the patient’s HIV-1-infected ART-treated partner was phylogenetically clearly unrelated to rebounding HIV-1 plasma sequences, making sexually transmitted superinfection an extremely unlikely explanation of the patient’s viral relapse ( Fig 4C and 4D ). Together, these data suggest that the rebound viremia originated from a viral variant that was not detected in the peripheral blood compartment at any earlier time point, possibly implicating reactivation of an archived provirus harbored by one or more cellular or anatomical reservoirs that were distinct from CD4 T cells circulating prior to transplantation.
Major drug resistance mutations defined by the Stanford HIV Drug Resistance Database ( https://hivdb.stanford.edu ) algorithm were not detected in any of the HIV sequences from the patient (inclusive of all time points and all of PBMC- and plasma-derived HIV RNA and DNA). Cytotoxic T lymphocyte (CTL) escape mutation analyses were not performed, given that HIV-1-specific CTLs were only very weakly detected in the patient (see below).
Cellular immune responses
HIV-1-specific CD8 T cell responses can effectively restrict HIV-1 replication by MHC class I–restricted cytolysis and represent an important correlate of antiviral immune protection in individuals with natural control of HIV-1, specifically when restricted by HLA-B27 [ 33 , 34 ], an MHC class I allele present in the recipient and the donor of the hematopoietic stem cells in this case. To analyze HIV-1-specific T cell responses in our patient, we stimulated PBMCs with pools of peptides corresponding to individual HIV-1 gene products, followed by quantification of antigen-induced intracellular cytokine production. These results showed barely detectable HIV-1-specific CD8 T cell responses at all analyzed time points ( Fig 5A ). A similar observation was made for HIV-1-specific CD4 T cell responses, most of which also remained under the threshold of detection by flow cytometry ( Fig 5A ). The total number of CD4 cells was significantly expanded prior to transplantation at the expense of CD8 T cells, but CD4:CD8 T cell ratios improved during the subsequent disease process, towards an age-appropriate naïve and memory cell distribution [ 35 ] in both the CD4 and CD8 T cells ( Fig 5B ). Notably, expression of cellular activation markers and immune checkpoints on total CD4 and CD8 T cells was strongly upregulated 144 days after transplantation, the time point associated with his diagnosis of GVHD ( Fig 5C ). Corresponding to these findings, we observed that CD25+ CD127− FoxP3+ regulatory CD4 T cells were infrequently detected prior to transplantation, followed by a rapid increase of regulatory T cell (Treg) frequencies during the post-transplantation period; expression of immune checkpoint and activation markers on Tregs was most obvious at the time of clinical GVHD ( Fig 5D ).
10.1371/journal.pmed.1002461.g005
Fig 5
Dynamics of CD4 and CD8 T cell responses in the described patient.
(A) Proportions of CD4 (upper plots) and CD8 (lower plots) T cells specific for the indicated HIV-1 gene product. Color coding reflects time of sample collection. HIV-1-specific T cell responses were identified based on antigen-specific IFNγ secretion (left plots) or IL-2 secretion (right plots). (B) Proportions of total CD4 and CD8 T cells, and indicated T cell subsets, within CD4 and CD8 T cells. (C) Heatmaps reflecting the longitudinal evolution of the proportions of indicated CD4 (left plot) and CD8 (right plot) T cell subsets expressing PD-1 or CD38/HLA-DR. (D) Proportions of FoxP3+ regulatory T cells with indicated phenotypic characteristics. ATI, analytic treatment interruption; Tcm, central memory T cells; Teff, terminally differentiated T cells; Tem, effector memory T cells; Treg, regulatory T cells; Tscm, T memory stem cells.
Innate immune cells
Innate immune cells can modulate antiviral immune defense and HIV-1 immune activation by a variety of mechanisms [ 36 ]. To analyze change in the innate immune system during the patient’s treatment course, we focused on CD3− CD56+ natural killer (NK) cells, arguably the most important effector component of the innate immune system [ 37 ]. Before transplantation, the total number of NK cells was severely diminished, and consisted predominantly of CD56dim CD16− cells, while CD16+ CD56− NK cells, previously associated with improved cytotoxic function [ 37 , 38 ], made smaller contributions ( Fig 6A ). This relative distribution of NK cell subsets persisted during the subsequent disease course, although the total number of NK cells increased to normal levels. NK cell activation markers, in particular NKG2D, were again most strongly expressed at the time of GVHD on all NK cell subsets ( Fig 6B ); similar but less obvious trends were also noted for expression of NKp46 and NKp30 on NK cells ( S1 Fig ). Remarkably, CD11c+ myeloid dendritic cells were most frequently detected immediately prior to transplantation, and subsequently declined to levels more typically observed in ART-treated HIV-1-infected patients ( Fig 6C ); relative proportions of CD14+ monocytes and plasmacytoid dendritic cells remained relatively stable throughout the entire observation period, as did expression levels of co-stimulatory and dendritic cell maturation markers on these dendritic cells and monocytes ( Fig 6D ).
10.1371/journal.pmed.1002461.g006
Fig 6
Longitudinal changes in innate immune cells.
(A) Proportion of total NK cells and indicated NK cell subsets. Color coding reflects time of sample collection. (B) Spider diagram demonstrating longitudinal changes in proportions of NK cell subsets with indicated phenotypic characteristics. (C) Proportions of CD14+ monocytes, HLA-DR+ CD11c+ lin− myeloid dendritic cells, and HLA-DR− CD123+ plasmacytoid dendritic cells at indicated time points during treatment course. (D) Heatmap showing longitudinal evolution in the proportions of monocytes, conventional dendritic cells, and plasmacytoid dendritic cells expressing CD80, CD83, or CD86. ATI, analytic treatment interruption; cDC, conventional dendritic cells; mDC, myeloid dendritic cells; mono, monocytes; NK, natural killer; pDC, plasmacytoid dendritic cells.
B cell immune responses
Proportions of total CD19+ B cells and non-switched memory B cells were smallest prior to transplantation, but levels normalized during the post-transplantation disease course ( S2 Fig ). Consistent with these data, but unlike the Berlin patient [ 5 ], quantitative levels of anti-HIV antibodies did not change significantly in the first 100 days after transplantation ( Fig 7A ). However, at later time points after transplant, our patient demonstrated declining levels of anti-HIV antibodies, as demonstrated by the decreasing number and intensity of anti-HIV bands on Western blot from day −119 to day +888 ( Fig 7B ).
10.1371/journal.pmed.1002461.g007
Fig 7
HIV-1 antibody assessment in the peri-transplant period.
(A) HIV-1 antibodies in serum were quantified by the limiting antigen assay in the pre-transplant and early post-transplant period. (B) Anti-HIV-1 Western blot analyses were performed on the days indicated. “Strong pos.” and “Weak pos.” represent internal positive controls for the assay. ATI, analytic treatment interruption; Txp, transplant.
Discussion
To date, the only described cure of an adult with HIV-1 is the Berlin patient, who was cured of HIV-1 following treatment for acute myeloid leukemia that included induction chemotherapy and anti-thymoglobulin treatment, followed by 2 allo-SCTs from a donor with a homozygous CCR5 Δ32 mutation [ 4 ]. Two Harvard patients who underwent reduced intensity conditioning and allo-SCT, and had significant reductions in the latent viral reservoir, eventually had virologic rebound off ART; these cases are cautionary examples that near eradication of the reservoir may not be sufficient to achieve even a functional cure [ 6 ]. We present extensive host and virologic studies on an additional HIV-1 positive individual who underwent allo-SCT. Table 3 compares clinical features between these 4 cases of prolonged ARV-free HIV-1 remission after allo-SCT. Results from our patient confirm that HIV-1 burden (as measured by total HIV-1 DNA and integrated HIV-1 DNA) can decline significantly after allo-SCT, but this is not necessarily accompanied by cure of HIV-1 infection.
10.1371/journal.pmed.1002461.t003
Table 3 Clinical features of previous selected patients.
Clinical feature
Mayo patient
Berlin patient
Harvard patient A
Harvard patient B
Years of HIV-1 before Tx (years of ART)
23 years (9 cumulative years)
>10 years (4 years)
Lifelong (3–4 years)
20 years (7 years)
Viral load prior to Tx
23 copies/ml
Undetectable
Undetectable
Undetectable
Donor CCR5 genotype
Wild-type
CCR5 Δ32 homozygous
Wild-type
Wild-type
Donor HLA
A*03,24; B*07,27; Cw*02,07
A*0201; B*0702,3501; Cw*0401,0702; DRB1*0101,1501; DQB1*0501,0602
A*0201,2301; B*4403,5101; Cw*0202,0401
A*02,24; B*08,1517; Cw*07,07
Recipient HLA
A*03,24; B*07,27; Cw*02,07
A*0201; B*0702,3501; Cw*0401,0702; DRB1*0101,1501; DQB1*0501,0602
A*0201,2301; B*4403,5101; Cw*0202,0418
A*02,24; B*08,1517; Cw*07,07
Conditioning regimen
Rituximab, cyclophosphamide, vincristine, doxorubicin, dexamethasone, methotrexate, cytarabine (4 cycles)
Cytarabine, gemtuzumab, rabbit anti-thymocyte globulin, whole-body radiation
Gemcitabine, navelbine, doxorubicin, busulfan, fludarabine
Busulfan, fludarabine
GVHD episodes (sites and grade)
Mouth and colon grade 1
Skin grade 1
Skin, eye, liver
Skin, liver, oropharynx
Duration of ART from Tx to ATI
3.2 years
0 (stopped 1 day before SCT)
4.3 years
2.6 years
Time from ATI to HIV-1 RNA rebound
288 days
>10 years
84 days
219 days
Hematologic malignancy
Acute lymphoblastic leukemia, B lineage, with myeloid features
Acute myeloid leukemia
Nodular sclerosing Hodgkin lymphoma
Diffuse large B cell lymphoma; mixed cellularity Hodgkin disease
ATI, analytic treatment interruption; GVHD, graft-versus-host disease; SCT, stem cell transplant; Tx, transplant.
The major barrier to HIV-1 cure is the latent viral reservoir, composed largely of resting memory CD4 T cells that carry stably integrated, replication-competent HIV-1 DNA [ 39 , 40 ]. The mechanisms by which HIV-1 persists are multifactorial (reviewed in [ 41 ]). After therapy with effective ARV medications, plasma HIV-1 RNA becomes undetectable; however, cellular and anatomic reservoirs of HIV-1 persist. The half-life of the resting HIV-1-infected CD4 T cell is estimated to be approximately 44 months [ 42 ], which predicts that over 60 years of ARV therapy would be required to eradicate the reservoir of HIV-1 in resting T cells, provided fully suppressive HIV therapy can be achieved and maintained. However, the existence of chronic or intermittent low-level viral replication replenishes the viral reservoir by infection of additional CD4 T cells [ 43 ], some of which become latently infected resting memory CD4 T cells. In addition, other mechanisms may exist to contribute to maintenance of latently infected T cells, including homeostatic proliferation and clonal expansion [ 13 ]. Therefore, therapies that target HIV-1 replication are alone insufficient to eradicate HIV-1, and other interventions that target and eradicate the latent viral reservoir will be needed to cure HIV-1.
It remains unknown what components of the Berlin patient’s treatment were responsible for his HIV-1 cure. Possibilities include 2 courses of myeloablative chemotherapy eradicating the HIV-1 reservoir, CCR5 Δ32 donor cells being resistant to HIV-1 reinfection, and a “graft-versus-HIV” effect. Accordingly, a variety of approaches are being evaluated to recapitulate this only HIV-1 cure, including PBSCT, gene therapy to knockdown CCR5 or other host factors required for HIV-1 replication, and chimeric antigen receptors expressed in autologous CD8 T cells designed to kill HIV-1-infected cells (reviewed in [ 44 ]).
In experimental models in which rhesus macaques infected with simian HIV underwent myeloablative conditioning and autologous stem cell transplantation, despite significant reductions in viral reservoir size, viral rebound occurred shortly after ART interruption [ 45 ]. A greater than 10,000-fold reduction of the HIV-1 reservoir in a host would be required to prevent HIV-1 rebound after discontinuing combination ART, according to stochastic modeling estimates [ 46 ], a goal that would be difficult to attain. The HIV-1 reservoir is also difficult to measure reliably, as currently available assays are insufficiently sensitive to detect such low levels of virus above background signals in the assays themselves. An upper limit estimate of the magnitude of HIV-1 reservoir reduction in our patient would be approximately 200-fold, which is consistent with a delay in viral rebound, assuming stochastic reactivation of a reduced number of latently infected cells [ 46 , 47 ].
Both the donor and the recipient involved in our case had wild-type CCR 5. The donor for the Berlin patient was genotypically CCR5 Δ32, which has inspired the exploration of gene therapy approaches for HIV-1 eradication, including knock-down of host proteins required for HIV-1 replication, such as CCR5, as well as studies to identify donors for hematopoietic stem cell transplantation who are homozygous for CCR5 Δ32. In a pilot study, autologous CD4 T cells modified by CCR5 knock-down had a prolonged half-life compared to unmodified cells, demonstrating the feasibility of the approach, yet the functionality of the reinfused cells was not assessed, and approaches to replacing all potential target cells will need to be addressed for this approach to have an impact on effecting a cure [ 48 ].
A recent report of an HIV-1-infected patient who underwent allo-SCT from a homozygous CCR5 Δ32 donor further confounds our understanding as that recipient experienced virologic rebound associated with a shift in viral tropism from CCR5-tropic to X4-tropic [ 49 ]. Thus, CCR5 Δ32 transplants alone may not be sufficient for HIV-1 cure.
As the Berlin patient’s post-transplant course was complicated by GVHD, it has been hypothesized that allogeneic responses post-transplant may exert a graft-versus-HIV effect by killing residual recipient lymphocytes, including latently infected cells. In fact, our patient experienced grade I GVHD of the bowel that was managed symptomatically in an attempt to promote this effect. Coincident with clinical GVHD in our patient, immunologic studies of PBMCs at day +142 revealed increased expression of activation markers in multiple cell lineages, including NK cells, CD4 T cells, and CD8 T cells. Also coincident with clinical GVHD, cell-associated HIV-1 RNA was detectable at day +142, at levels comparable to pre-transplant levels ( Fig 1B ), despite total and integrated HIV-1 DNA being reduced and viremia being undetectable at the same time ( Fig 2 ). Altogether, these data suggest that GVHD caused polyclonal and generalized immune activation and consequent viral production from latent sources. However, in the absence of demonstrable HIV-specific CD8 T cells ( Fig 5A ), and with waning B cell immunity to HIV-1 (Figs 7 and S2 ), GVHD may not have had an effective graft-versus-HIV effect, but instead may have promoted HIV-1 persistence by stimulating subclinical replication, potentially in tissue sanctuary sites, although this is speculative.
There are 3 main limitations to our study. First, we did not have access to archived samples from prior to presentation for leukemia evaluation. Therefore, we were unable to characterize long-term trends in HIV reservoir size and phylogeny before allo-SCT. Second, we primarily sampled blood cells, with only 1 lymphoid tissue sample analyzed. Therefore, we were unable to characterize the HIV reservoir contained in the lymphoid tissue. In addition, it is likely that the immunologic changes noted in Figs 5 and 6 were the result of the pre-transplant conditioning regimen, the transplant procedure, and/or GVHD, and not necessarily due to underlying HIV infection or clearance thereof. Finally, since this is a single case description, it is unclear if the findings are applicable to other HIV positive patients undergoing allo-SCT.
Despite these limitations, our case clearly illustrates that allo-SCT in the setting of ART-suppressed HIV-1 infection can significantly reduce the HIV-1 reservoir size, in this case to a level that was sufficiently low that viral rebound did not occur for 288 days following treatment interruption. It is noteworthy that once virus rebound did occur, the proviral sequence was phylogenetically different from the viral sequences identified in the peri-transplant period, and may have originated from sanctuary tissue sites harboring archived viral species seeded during the extensive HIV-1 disease process preceding the patient’s oncologic history. Most researchers in the HIV-1 cure field believe that successful cure of HIV will involve a combination of approaches that act by different mechanisms to synergistically eradicate viral reservoirs. Our data suggest that allo-SCT can profoundly reduce HIV-1 reservoir size, but incompletely, and raise the hypothesis that coupling allo-SCT with other viral reservoir reduction approaches might eventually enable a cure or long-standing remission of HIV-1 infection.
Supporting information
S1 STROBE Checklist
STROBE statement.
(DOCX)
S1 Fig
Detailed phenotypic characterization of NK cells, CD4 T cells, and CD8 T cells.
Heatmaps for (A) NK cells, (B) CD4 T cells, and (C) CD8 T cells reflect proportions of cells with indicated phenotypic properties at given time points.
(DOCX)
S2 Fig
Phenotypic characteristics of B cells in the described patient.
(A) Longitudinal evolution of total B cells and indicated B cell subsets. Phenotypic classification was determined as follows: memory non-switched: CD27+ IgD+; memory IgM-only: CD27+ IgD− IgM+; memory switched: CD27+ IgD− IgM−; plasmablast: CD27high IgD− CD38high; transitional T1-T2: CD27− IgD+ CD10+ CD38high; memory double-negative: CD27− IgD−. (B) Heatmap reflecting the longitudinal proportion of B cells with indicated phenotypic characteristics.
(DOCX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The global population is ageing rapidly due to a decline in fertility and an increase in life expectancy [ 1 ]. In fact, healthy bodies can contribute to their families and the society positively [ 2 ]. In 2016, guidance policy on promoting healthcare services goods is proclaimed by State Council of China, National Commission of China and other national ministries. Hence, local governments, scientific research institutions and healthcare industries has put much attention on healthcare service goods. Basic healthcare service goods covers diagnosis and treatment of common diseases, frequently-occurring traditional Chinese and western medicine, rational drug use, guidance of medical treatment path and referral appointment. In fact, promoting healthcare service goods is not only an important task of deepening the reform of the medical and health system, but also an important way to better safeguard the health of the people under the new situation. Healthcare service goods belong to public goods, which can be identified as goods with huge positive externalities. Public goods make it possible to enjoy the benefits for others although they do not pay for these goods, which is different from the property of private goods significantly. Furthermore, healthcare service goods are beneficial to extend expected lifetime and promote regional health development level. At present, the promotion of healthcare service goods is at the initial stage, and China medical insurance system has not covered healthcare service goods due to the system’s imperfection, it partially explain why patients are not willing to purchase healthcare service goods. In other word, it is necessary for policy makers to promote healthcare service goods by an effective intervention mechanism. As we all know, conventional hospital resource is not enough to meet the expectation of residents’ demand in China, however, the importance of healthcare service goods does not get a lot of attention, so establishing an intervention mechanism of healthcare service goods is extremely important and necessary. The rapid growth of China's economy has improved people’s living standards dramatically [ 3 ], with the growth of various health demand in China, healthcare service goods have emerged in large cities in recent years. Personalized health package, chronic disease management service goods, postpartum recovery service products, or vaccines used for prophylactic vaccination are well-known examples of healthcare service goods. In fact, the usage and promotion of healthcare service goods is beneficial to actual customers and majority of local residents remarkably, so it is essential for the administrative sector to regulate the supply chain of healthcare service goods. There are notable examples of administrative sector’s intervention, such as the intervention of governments and private firms in health aspects to improve nutrition behavior [ 4 ], the intervention of donors in malaria drugs to encourage the channel to improve access to these drugs [ 5 ], the intervention of the public sectors in a two-tier health system to maximize the total weighted patient welfare [ 6 ], and the intervention of the government in demand and supply sides to investigate the relative effectiveness for the influenza vaccine supply chain [ 7 ]. For the above cases, the key role of an administrative sector is to design an intervention mechanism from perspective of social benefits. Then it is crucial and indispensable for us to establish an intervention mechanism based on social welfare maximization.
In this paper, we consider a research problem that regulates the supply chain system for healthcare service goods by an intervention mechanism, and the supply chain system is composed of a product provider and an administrative sector. In recent years, the rapid growth of China's economy has improved people’s living standards dramatically, but conventional hospital resource is not enough to meet the residents’ demand expectation in China. Promoting healthcare service goods is an important task of deepening the reform of the medical and health system, and many local governments are promoting healthcare service goods through subsidy strategy. In China, the administrative sector refers to the health commission, the commission of development and reform, the finance department, the civil administration department, the department of human resource and social security and the government department. For an administrative sector, the objective of the intervention mechanism is to maximize social welfare utility instead of maximizing expected profit. Then we should incorporate the objective into the supply chain system when designing the intervention mechanism based on social welfare maximization. Two intervention strategies can be applied to the intervention mechanism for an administrative sector, one strategy is to invest its capital in the demand-growth strategy, such as media publicity, physician training, demand investigation and direct development. Another strategy is to provide rebates or subsidies for the customers. In practice we have to consider the customers’ ability of their willingness to pay, so rebates or subsidies play a key role in making the goods more extensive and attractive [ 8 ]. We can easily find that intervention mechanism research of healthcare service goods in the literature, for example, Juergen et al. design an intervention strategy to achieve a higher efficiency of the payoff and production of the public goods including healthcare goods [ 9 ], Hamed Mamani et al. consider a subsidy that leads to a socially efficient level of coverage, and derive a tax-subsidy combination that is revenue neutral, but achieves the same effect [ 10 ]. Yoko Ibuka et al. find that an increase in the subsidy amount by 1,000 yen (10 USD) leads to a one percentage point increase in the vaccination rate among the elderly in Japan [ 11 ]. In our paper, a joint intervention mechanism using both demand-growth strategy and subsidy strategy for healthcare service goods is proposed, so to some extent, the key problem is to optimalize the allocation of the administrative sector’s budget between demand-growth strategy and subsidy strategy.
Bilevel programming is used to obtain the optimal allocation of the administrative sector’s budget between the two intervention strategies. In fact, bilevel programming can be regarded as a particular class of hierarchical mathematical program [ 12 , 13 ]. In a bilevel programming problem, the objective function of the upper-level problem is identified as the upper-level function, similarly, the objective function of the lower-level problem is identified as the lower-level function. The lower-level variables are constrained to be the solution of the lower-level problem, whereas the remaining variables are regarded as the upper-level variables and parametrize the lower-level problem [ 14 ]. In general, a bilevel programming problem contains two levels of optimization tasks [ 15 ], the upper-level makes optimal decision based on his objective first, then the lower-level chooses his optimal decision given the upper-level’s action [ 16 , 17 ]. In our paper, the administrative sector is the upper-level decision maker whose objective is to attain social utility maximization, whereas the product provider is the lower-level decision maker whose objective is to attain expected profit maximization, so it is scientific and reasonable to use bilevel programming model. More specifically, the administrative sector determines the optimal budget allocation between demand-growth strategy and subsidy strategy, then the product provider determines the order quantity. The response functions of investment in demand-growth strategy are usually supposed to be concave [ 18 – 20 ]. In other words, expected demand will increase with the money investment in demand-growth strategies. Finally, we explore a specific case study to demonstrate the effectiveness of our model.
The rest of the paper is organized as follows. In Section 2, we provide a literature review related to our research problem. In Section 3, we establish its mathematical formulation according to the research problem. In Section 4, we explore two benchmark approaches. In Section 5, we present the special case. In Section 6, we conduct numerical analysis to obtain the key results. In Section 7, we summarize conclusion remarks and highlight future directions of this research.
Literature review
In this paper, we consider a research problem that regulates the supply chain system for healthcare service goods by an intervention mechanism, and the supply chain system is composed of a product provider and an administrative sector, the objective of this paper is to facilitate the healthcare service goods to be adopted widely by establishing an intervention mechanism based on social welfare maximization. Therefore in this section, we outline four streams of literature according to the research problem.
First, the paper is related to the economics literature studying contract design for healthcare service goods and other public-interest goods. Regarding the contract design literature of healthcare service goods, Hamed Mamani et al. [ 21 ] propose a contract mechanism to reduce the inefficiency in the allocation of influenza vaccines. The proposed contract based on epidemic model reduces the overall financial burden of infection globally and improves the total number infected by seasonal influenza outbreaks. N. Shamsi G. et al. [ 22 ] develop a specific option contract for proactively provisioning required vaccine doses from two suppliers (a main and a backup). For the model in this paper, its aim is to minimize the procurement and social costs using the SIR epidemic model. In addition, there are some literature studying contract design for other public-interest goods. Wenhui Zhou et al. [ 23 ] discuss two types of contracts that specify the subsidy for energy-saving products with the government’s budget constraint, and the optimal design of the contracts is given under two government objectives: minimizing the total energy consumption and minimizing the average energy consumption. Fei Ye et al. [ 24 ] design a coordination contract in a random yield environment to ensure the sustainability of biofuel production and improve the performance of the biofuel supply chain, then over-production risk-sharing contract, under-production risk-sharing contract and mixed contract are examined. Zhaofu Hong et al. [ 25 ] study several cooperation contracts for a green product supply chain, the environmental responsibilities of a manufacturer and a retailer are considered in a two-echelon supply chain, the result shows that the cooperation between the manufacturer and the retailer may not always profitably benefit all partners. Given the characteristic of healthcare service goods in China, it is more appropriate to use intervention mechanism for healthcare service goods.
Second, the paper is related to the stream of operations research dealing with intervention mechanism of healthcare service goods and other public-interest goods. Regarding health-related products, Dan Yamin et al. [ 26 ] establish an epidemiological game model to find the optimal incentive for vaccination and the expected vaccination coverage. Elodie Adida et al. [ 27 ] consider how a socially optimal vaccine coverage can be induced through the central policy-maker’s subsidy to both consumers and the vaccine manufacturer. In fact, subsidy is a frequently-used intervention method. Some papers find a positive correlation between the subsidy and the production. For example, Huaying Gu et al. [ 28 ] investigate a electric vehicle manufacturer's optimal production strategy under subsidy and battery recycling when the market demand is uncertain. The results indicate that increased subsidy promotes the electric vehicle manufacturer's optimal production quantity and expected utility. Chunlin Luo et al. [ 29 ] consider two manufacturers in a symmetric duopoly setting how to produce the traditional and public interest products under a government’s subsidy scheme, and the result shows that a higher subsidy can increase the sale of the public-interest product. Maxime C. Cohen et al. [ 30 ] indicate that government subsidies offered directly to consumers impact the manufacturer’s production, but the policy makers should attach importance to demand uncertainty when designing consumer subsidies, otherwise they will not attain the expected adoption target level. Furthermore, Chaogai Xue et al. [ 31 ] study the decision-making of government subsidy on supply chain for straw power generation, and discuss the changes of members’ profits and supply chain’s profits under different subsidy circumstances. Bo Li et al. [ 32 ] consider which subsidy strategy is more efficient for environmental-friendly products in a dual-channel supply chain. Different from this stream of literature, we consider demand-growth and subsidy simultaneously.
Third, the paper is broadly related to ever-increasing literature that studies multi-echelon supply chain problem. As a matter of fact, two-echelon and three-echelon are common multi-echelon types of supply chain. For two-echelon supply chain problem, Yi Yuyin et al. [ 33 ], Dua Weraikat et al. [ 34 ], R.B.O. Kerkkamp et al. [ 35 ] and T. Maiti et al. [ 36 ] are some good examples. All these above studies consider a two-echelon supply chain model that is comprised of one manufacturer/supplier and one retailer. These studies indicate that both tax and subsidy policies can facilitate the sustainability of the supply chain, furthermore, designing a cooperation mechanism between the manufacturer and the retailer has an important effect on the supply chain. For three-echelon supply chain problems, B.C. Giri et al. [ 37 ] and Jian Li et al. [ 38 ] are the two examples that studies three-echelon supply chain that is comprised of a supplier, a manufacturer and a retailer. Different coordination strategies and detailed analyses are discussed. The paper of Songsong Liu et al. [ 39 ] study the optimal profit distribution in the supply chain consisting of active ingredient plants, formulation plants and markets. Y.N. Wu et al. [ 40 ] consider three levels of information sharing in a three-echelon supply chain consisting of a manufacturer, a distributor, and a retailer, and then derive the optimal inventory policy under each level of information sharing. Compared to this stream of literature, our paper considers a common two-echelon healthcare service goods supply chain setting composed of a product provider and an administrative sector.
Forth, there is a stream of literature of bilevel programming research. Walter J. Gutjahr et al. [ 41 ], Saemeh Aghajani et al. [ 42 ], Yue Zheng et al. [ 43 ] and S.M. Alizadeh et al. [ 44 ] introduce a bilevel programming model to solve the problem. In general, bilevel programming model is identified as an efficient mathematic method to solve the hierarchical decision-making problem with two different decision objectives. In the bilevel programming model, the decision-maker at the upper level optimizes his/her objective function under a set of constraints first, and then the decision-maker at the lower level optimizes his/her objective function taking into consideration of the upper decision-maker’s action [ 45 ]. In our paper, the administrative sector is the upper-level decision maker whose objective is to attain social utility maximization, whereas the product provider is the lower-level decision maker whose objective is to attain expected profit maximization.
In summary, although the above literature have enriched our understanding of the impact of intervention mechanism on supply chain or procurement, the existing literature has not studied intervention mechanism that maximizes social welfare for healthcare service goods, so solving this problem is important and crucial to facilitate the healthcare service goods to be adopted widely. To the best of our knowledge, Ece Zeliha Demirci, Lulu Shao and Huiping Ding [ 46 – 48 ] are closet to our research in supply chain and mechanism design. Demirci and Erkip [ 46 ] study the intervention problem for public-interest goods by using bilevel programming model, but they do not consider consumer’s willingness behavior. Shao, et al. [ 47 ] formulate a utility model composed of a population of consumers who make utility maximizing choices and manufacturers who set an optimal pricing, then optimal subsidies or optimal price discount rates can be found for policy makers. Ding, et al. [ 48 ] explores the collaborative mechanism that motivates supply chain firms to collectively invest in environmental technology and produce environmental friendly products. Due to public attribute of healthcare service goods, intervention mechanism should take into account the maximization of social welfare. With the growth of various health demand in China, Chinese government is promoting healthcare service goods by using subsidy strategy, so it is more scientific and reasonable to consider demand-growth strategy and subsidy strategy jointly. Our research work differs from the above three papers in three dimensions: (1) a method to determine the willingness price of healthcare service goods in China; (2) for healthcare service goods, the administrative sector’s budget allocation between demand-growth strategy and subsidy strategy is explored; (3) due to the characteristic of the healthcare service goods in China, an intervention mechanism considering demand distributions based on social welfare maximization is developed.
Model
In this section, we establish a mathematical model to design a maximizing social welfare intervention mechanism for healthcare service goods in China. The aim of the intervention mechanism is to expand the adoption of the healthcare service goods. The problem for a common setting composed of a product provider and an administrative sector that regulates the supply chain system for healthcare service goods by an intervention mechanism is considered. The main goal of the product provider is to maximize its expected profit, whereas the main goal of the administrative sector is to improve the healthcare service goods’ availability and adoption, hence promoting social welfare.
First, given the fuzziness and uncertainty of evaluation indicators in Table 1 , we establish a variable fuzzy set model to get the willingness price, which can help customers buy a healthcare service goods at a lower price that they are willing to pay. To the best of our knowledge, we are the first to study the willingness price for healthcare service goods in China. Then, we formulate a bilevel programming model to study the intervention problem, to some extent, the key problem is to optimalize the allocation of the administrative sector’s budget between demand-growth strategy and subsidy strategy.
10.1371/journal.pone.0214655.t001
Table 1 Healthcare service goods perceptive satisfaction evaluation indicator system.
healthcare service goods perceptive satisfaction evaluation indicator system
basic medical care
general clinic therapy
Chinese medicine clinic therapy
simple clinic (used to make up a prescription only)
conventional family diagnosis
health record maintenance management
basic public health
hypertension patient management
hyperglycemia patient management
hyperlipidemia patient management
diabetes patient management
coronary disease patient management
perceptive value
disease prevention
health state improvement
doctor-patient trust
3.1 Willingness price
In general, the public-interest goods price that the customers are willing to pay is lower than the cost price of the goods [ 49 ]. Let p w and p c denote the willingness price and the cost price respectively, and the willingness price will be divided into five levels, they are 0.85 p c , 0.75 p c , 0.65 p c , 0.55 p c , 0.45 p c respectively. The customers always determine the willingness price on the basis of their perceptive satisfaction evaluation of the healthcare service goods. In other words, the level of willingness price depends on the product perceptive satisfaction level, and there is a positive correlation between the level of willingness price and the perceptive satisfaction evaluation. For example, if the product perceptive satisfaction is regarded as the highest level, then the willingness price will be 0.85 p c accordingly.
3.1.1 Perceptive satisfaction evaluation indicator system
The perceptive satisfaction evaluation indicator system of healthcare service goods is established by following the principle of scientificity, systematicness and operability. According to the indicator system connotation and current situation in China, the evaluation indicator system of perceptive satisfaction should include basic medical care, basic public care and perceptive value. The detailed indicator system is formulated in Table 1 .
3.1.2 Variable fuzzy set method
The variable fuzzy set method is based on relative difference function, then the subordination relationship between evaluation objects and standard levels can be acquired by subordination information [ 50 ]. In the variable fuzzy set model, we suppose that u is the arbitrary element of fuzzy set U , the arbitrary element u has a relative membership degree μ A ( u ) with the attractive interval A , μ A ( u ) ∈ [0, 1]. The arbitrary element u has a relative membership degree μ Ac ( u ) with the exclusive interval A , μ Ac ( u ) ∈ [0, 1]. The arbitrary element’s relative difference coefficient D A ( u ) for the attract interval A is as follows:
D A ( u ) = μ A ( u ) - μ A c ( u )
(1)
μ A ( u ) - μ A c ( u ) = 1
(2)
We can obtain the relative difference coefficient D A ( u ) according to Eqs 8 and 9 , relative difference coefficient μ A ( u ) is as follows:
μ A ( u ) = [ 1 + D A ( u ) ] / 2
(3)
On the continuous membership number axis ( Fig 1 ), we suppose that X 0 = [ a , b ] refers to the arbitrary element’s attractive interval, X = [ c , d ] refers to a range interval including X 0 .
10.1371/journal.pone.0214655.g001
Fig 1
The position relation between zone [ a , b ], [ c , d ] and point x , M .
In Fig 1 , [ c , a ] and [ b , d ] refer to the arbitrary element’s exclusive interval, M refers to the point of relative membership μ A ( u ) = 1 located on attractive interval [ a , b ]. We suppose that x is the arbitrary point in the interval, if x is located on the left of point M , its relative difference function is Eq 11 ; however, if x is located on the right of point M , its relative difference function is Eq (12) .
{ D A ( u ) = ( x − a M − a ) β x ∈ [ a , M ] D A ( u ) = − ( x − a c − a ) β x ∈ [ c , a ]
(4)
{ D A ( u ) = ( x − b M − b ) β x ∈ [ M , b ] D A ( u ) = − ( x − b d − b ) β x ∈ [ b , d ]
(5)
In general, β = 1, it means that relative difference function is linear model. Then we can get the relative difference coefficient μ A ( u ) by putting Eqs 4 and 5 into the Eq 3 , so single-factor fuzzy matrix R is obtained. We suppose that n refers to evaluation indicators, m refers to evaluation levels, so the variable fuzzy set evaluation model is as follows:
{ u h ′ = [ 1 + ( d g h d b ) α ] - 1 d g h = { ∑ i = 1 n [ ω i ( 1 - μ A ( u ) i h ) ] p } 1 p d b = { ∑ i = 1 n [ ω i μ A ( u ) i h ) ] p } 1 p
(6)
Where, u h ′ is the relative membership degree that is not normalized for level h , h refers to evaluation degree, h = 1, 2,⋯, m ; d gh is the generalized weighted distance between the relative membership degree and the left limit point; d b is the generalized weighted distance between the relative membership degree and the right limit point; μ A ( u ) ih refers to the indicator’s relative membership degree for the level h ; α refers to variable optimization criterion parameter, α = 1,2; ω i refers to the weight of the evaluation indicator i . The relative membership degree u h ′ is normalized as follows:
u h = u h ′ ∑ h = 1 m u h ′
(7)
H = ∑ h = 1 m u h ⋅ h
(8)
Where, u h refers to the normalized relative membership degree for the level h , H refers to the evaluation object’s level. The type of willingness price depends on the product perceptive satisfaction level H . If the product perceptive satisfaction is regarded as the first level, then the willingness price will be 0.90 p c accordingly.
3.2 Intervention mechanism based on bilevel programming
In order to model the problem that regulates the supply chain system for healthcare service goods by means of an intervention mechanism, we assume a hierarchical decision process with two levels of decision. In constructing bilevel programming model for intervention mechanism problem, the following notations in Table 2 will be used.
10.1371/journal.pone.0214655.t002
Table 2 Notations used in the bilevel programming model.
Parameters
D
the healthcare service goods’s demand
P w
the willingness price
P w + s
the product provider’s revenue from per unit sold
c
the cost of each healthcare service goods
Q T
the healthcare service goods’ target amount formulated by the administrative sector
θ
the monetary value (RMB) per unit sold
v
the salvage value for each unsold goods
pdf f B d ( ∙ )
the probability density function of demand, and the demand distribution depends on B d
cdf F B d ( ∙ )
the cumulative distribution function of demand
Decision variables
upper level (the administrative sector )
B
the total budget that is used for intervention mechanism
B d
the budget amount that is allocated to investment in demand-growth strategy
B s
the budget amount that is allocated to subsidy strategy
s
the subsidy available to provide for each customer
lower level (the product provider)
Q
the quantity amount of healthcare service goods
Objective functions
upper level (the administrative sector )
u ( Q , B d )- B
it refers to social welfare function
lower leve l (the product provider)
E { P ( Q )| B d }
it refers to the product provider’s expected profit
In our model, the product provider’s problem is similar to a newsvendor problem. The demand distribution depends on the budget amount that is allocated to investment in demand-growth strategy. According to the relevant research in recent years, it is assumed that the cumulative distribution function of demand is monotonously increasing. The monotonicity of F B d ( ∙ ) implies that Q will increase with the increase of fractile. Especially, we assume that the cumulative distribution function of demand at a given value is a decreasing function of B d , so we can confirm that as B d increase, so does Q .
The bilevel programming model of the intervention mechanism problem is as follows:
M o d e l I : max s , B d , B s , B u ( Q , B d ) − B
(9)
s u b j e c t t o B d + B s ≤ B
(10)
s E [ { min ( Q , D ) | B d } ≤ B s
(11)
s ≥ c - P w
(12)
s ≥ 0
(13)
Q ≥ Q T
(14)
B d , B s , Q T ≥ 0
(15)
max Q E { P ( Q ) | B d }
(16)
where E { P ( Q ) | B d } = ∫ 0 Q [ ( P w + s ) x + v ( Q - x ) - c Q ] f B d ( x ) d x + ∫ Q ∞ ( P w + s - c ) Q f B d ( x ) d x refers to the expected profit of the product provider.
In model I , Eqs 9 to 16 refer to the administrative sector’s problem, and Eq 16 refers to the product provider’s problem. Eq 10 ensures that the sum of B d and B S can’t be higher than the total budget B . Eq 11 illustrates that the total subsidy amount is lower than the budget amount allocated to subsidy strategy. Eq 12 indicates that the subsidy per unit product is higher than c − P w , so the profit per unit product is greater than 0. Eq 14 highlights that the product quantity Q should be higher than the target amount Q T . In recent years, the Chinese government have attached great importance to the popularization of healthcare service goods, and the government at all levels has formulated clear targets. Eqs 13 and 15 guarantees that r , B d , B r , Q T are non-negative.
In fact, bilevel programming can be regarded as a particular class of hierarchical mathematical program. The upper-level objective function ( Eq 9 ) is identified as the administrative sector’s problem, with the objective of social welfare maximization, while the lower-level objective function ( Eq 16 ) is identified as the product provider’s problem, with the objective of its expected profit maximization. In our paper, we aim to establish an intervention mechanism based on social welfare maximization, and the product provider make an optimal decision under the dominant objective of the administrative sector by establishing a bilevel programming model.
Remark
We can easily find that Eq 11 can obtain optimal solution only when sE [{min( Q , D )| B d } = B s . In addition to this, it is optimal only when the total budget B is equal to the summation of B d (the budget amount that is allocated to investment in demand-growth strategy) and B r (the budget amount that is allocated to subsidy strategy).
The expected profit of the product provider E [ P ( Q )] is concave in Q for a given B d , B s and s , which indicates that the product provider has a unique solution. Then we can determine that the administrative sector can achieve his objective by maximizing social welfare due to the uniqueness of R [ B d , B s , s ]. E [ P ( Q )] is concave in Q for a given B d , B s and s , which implies that the product provider problem’s solution can be replaced by its first-order condition.
Note that the product provider’s problem is a newsvendor problem essentially, Model I can be written as the following single-level mathematical formation:
M o d e l I I : max s , B d , B s , Q u ( Q , B d ) − B d − B s
(17)
s u b j e c t t o s E [ { min ( Q , D ) | B d } = B s
(18)
s ≥ c - P w
(19)
s ≥ 0
(20)
Q ≥ Q T
(21)
B d , B s , Q T ≥ 0
(22)
F B d ( Q ) = P w + s - c P w + s - v
(23)
In general, a bilevel programming model is a challenging problem because it is difficult to calculate and obtain optimal solution, however, we obtain an easier solution method by translating a two-level model ( Model I ) into a single-level model ( Model II ). Now Model II is a nonlinear program, in which the objective function is not linear.
In the following section, we consider a specific form of the administrative sector’s social welfare function and the mean demand function. In practice, the administrative sector attaches great importance to the number of adopters for healthcare service goods in China, so it is scientific and reasonable to use θ times expected sales volume to quantify social welfare, then the social welfare objective function of the administrative sector can be regarded as a linear problem, which is convenient for us to do the following analyses. According to the related literature that studies the relationship between demand and the budget investment in demand-growth strategy B d , we assume that the response function of B d is increasing and concave, then the mathematic form is as follows:
μ ( B d ) = μ ∞ - d ( 1 + m B d ) c
(24)
where, m , d , c > 0.
In Eq 24 , we can determine that mean demand has a positive correlation with B d , in other words, if B d increases, then the mean demand will increase, but with a diminishing rate monotonically. The specific social welfare function is linear form, so we use θ times expected sales volume to quantify social welfare, the mathematical program can be written as follows:
M o d e l I I I : max s , B d , Q ( θ − s ) E m i n { ( Q , D ) | B d } − B d
(25)
s u b j e c t t o F B d ( Q ) = P w + s - c P w + s - v
(26)
s ≥ c - P w
(27)
s E [ { min ( Q , D ) | B d } = B s
(28)
Q ≥ Q T
(29)
B d , Q T ≥ 0
(30)
Next, we analyze specific demand distributions in model III . Based on related literature and the actual situation of demand distribution, it is reasonable to consider exponential and lognormal distributions to represent healthcare service goods’ demand distributions induced by B d . Then we consider Model III , in which demand distributions depends on B d , and the mean distribution follows Eq 24 . We analyze two situations: for the first situation, the variation coefficient is constant; for the second situation, the variation coefficient is related to B d .
The first situation: The variation coefficient is constant
First, we discuss that the demand distribution is exponential or lognormal when the variation coefficient is constant, and it can be divided into two situations: ( i ) μ ( B d ) for exponential distribution; ( ii ) μ ( B d ) for lognormal distribution, the analysis procedure is as follows:
According to Eq 26 , we can obtain the following Eqs 31 and 32 ,
1 - F B d ( Q ) = c - v P w + s - v
(31)
Q = F B d - 1 ( P w + s - c P w + s - v )
(32)
Then, the expected sales amount E [{min( Q , D )| B d } can be expressed as follows:
E [ { min ( Q , D ) | B d } = ∫ 0 F B d - 1 ( P w + s - c P w + s - v ) x f B d ( x ) d x + c - v P w + s - v F B d - 1 ( P w + s - c P w + s - v )
(33)
According to Eqs 40 and 32 , the administrative sector’s social welfare function can be expressed as follows:
u ( B d , s ) = ( θ - s ) { ∫ 0 F B d - 1 ( P w + s - c P w + s - v ) x f B d ( x ) d x + c - v P w + s - v F B d - 1 ( P w + s - c P w + s - v ) } - B d
(34)
Next, we analyze the above equation’s first order condition in regard to s , the mathematical formulation can be written as follows:
∂ u ( B d , s ) ∂ s = − { ∫ 0 F B d − 1 ( P w + s − c P w + s − v ) x f B d ( x ) d x + c − v P w + s − v F B d − 1 ( P w + s − c P w + s − v ) } + ( θ − s ) { F B d − 1 ( P w + s − c P w + s − v ) f B d [ F B d − 1 ( P w + s − c P w + s − v ) ] 1 f B d [ F B d − 1 ( P w + s − c P w + s − v ) ] c − s ( P w + s − v ) 2 + − ( c − v ) ( P w + s − v ) 2 F B d − 1 ( P w + s − c P w + s − v ) + c − v P w + s − v 1 f B d [ F B d − 1 ( P w + s − c P w + s − v ) ] ( c − v ) 2 ( P w + s − v ) 2 } = 0
(35)
We can obtain the following mathematical formulation by simplifying Eq 35 , the mathematical formulation can be written as follows:
∫ 0 F B d - 1 ( P w + s - c P w + s - v ) x f B d ( x ) d x + c - v P w + s - v F B d - 1 ( P w + s - c P w + s - v ) = θ - s f B d [ F B d - 1 ( P w + s - c P w + s - v ) ] ( c - v ) 2 ( P w + s - v ) 3
(36)
We assume that μ ( B d ) follows exponential distribution, the theorem of exponential distribution in mathematical formulation form can be written as follows:
F ( x ) = { 1 − e − γ x x ≥ 0 0 x < 0
(37)
Hence,
e - γ x = 1 - F ( x ) ⇒ ln ( e - γ x ) = ln ( 1 - F ( x ) )
(38)
So,
x = - 1 γ ln ( 1 - F ( x ) )
(39)
Note that
1 γ = μ ( B d ) , 1 - F ( x ) = c - v P w + s - v
(40)
We can obtain the following mathematical formulation by substituting Eq 40 into Eq 36 , it can be written as follows:
μ ( B d ) c - v P w + s - v ln ( c - v P w + s - v ) + μ ( B d ) P w + s - c P w + s - v - μ ( B d ) c - v P w + s - v ln ( c - v P w + s - v ) = ( θ - s ) μ ( B d ) c - v ( P w + s - v ) 2
(41)
Then, Eq 49 is obtained by simplifying Eq 41 ,
s 2 + 2 ( P w - v ) s + ( θ v + c v - P w c - P w v - θ c ) = 0
(42)
The solution of the above quadratic equation with one unknown is as follows:
s = - P w + P w + θ ( c - v ) + P w ( v - P w ) + c ( P w - v )
(43)
For Eq 43 , it is obvious to find that the optimal subsidy for each customer is independent of B d for exponential distribution.
We assume that μ ( B d ) follows lognormal distribution, the analysis procedure is similar to Eqs 31 to 43 . It can be expressed that the optimal subsidy for each customer is independent of B d for lognormal distribution when the variation coefficient is constan t .
For the first situation, we can find that the optimal subsidy for each customer is independent of the budget amount that is allocated to investment in demand-growth strategy and mean demand when the demand distribution is exponential or lognormal. It is interesting that the fact under the first situation is different from the traditional idea, maybe a large number of people would hold the idea that the optimal subsidy for each customer is related to the planning stage. In other words, the optimal fractile value is constant for the product provider, which means that there is no relation between the optimal fractile and the demand parameter μ ( B d ).
The second situation: The variation coefficient is related to B d for lognormal distribution
For the second situation, we assume that the variation coefficient is related to B d for lognormal distribution. To be more specific, if B d increases, then the variation coefficient will decrease. This assumption is in accordance with relevant literature and practical situations. Similar to Eq 24 , the mathematical formulation for the variation coefficient of the lognormal distribution can be written as follows:
c v ( B d ) = c v m i n + c v e l i ( 1 + m B d ) c
(44)
Where, m , c > 0, we assume that if B d (the demand-growth strategy) increases, then the variation coefficient will decrease, but with a monotonically diminishing rate. cv eli implies that a portion of variation coefficient can be eliminated. Eq 44 indicates that the lognormal distribution for healthcare service goods approaches a limited distribution with a mean of μ ∞ and a variation coefficient of cv min as B d (the demand-growth strategy) approaching infinity.
Next, we need to analyze the dependence relationship between the optimal subsidy for each customer and B d (the demand-growth strategy) considering the model III , the mathematic formulation of mean demand ( Eq 24 ) and the mathematic function of variation coefficient ( Eq 37 ). The specific analyses procedure is similar to that of Eqs 31 to 43 . Specifically, it manifests that there is dependence relationship between the optimal subsidy and B d (the demand-growth strategy). In other words, variability coefficient will decrease as dependence relationship between the optimal subsidy for each customer and B d (the demand-growth strategy) increases, which means that the optimal subsidy for each customer turns into a mathematic function of B d , that is to say, the optimal subsidy is relevant to the planning stage that can’t be ignored.
Benchmark approaches
In this section, we introduce two benchmark approaches of intervention mechanism that are general in practice. These two benchmark approaches are used to assess the performance of our proposed model. The first benchmark approach only consider the subsidy that is only determined by the administrative sector, and for the second benchmark approach, the optimal subsidy for each customer is independent of customer demand. Different from these two benchmark approaches, the optimal subsidy of the intervention mechanism is determined by the customer and the product provider jointly in this paper.
4.1 Benchmark approach 1
For benchmark approach 1, the subsidy is only determined by the administrative sector. In other words, decisions about intervention tools are not made by the administrative sector and the product provider jointly. In addition, B d (the demand-growth strategy) is not considered in benchmark approach 1. The aim of benchmark approach 1 is to analyze how the healthcare service goods supply chain system operates with a preset subsidy amount only, in other words, benchmark approach 1 does not consider the effect of demand-growth strategy on supply chain system for healthcare service goods. The mathematic formulation of benchmark approach 1 is written as follows:
B e n c h m a r k a p p r o a c h 1 : ( θ - s ) m i n ( Q , D ) - B s
(45)
s u b j e c t t o F ( Q ) = P w + s - c P w + s - v
(46)
s E m i n ( Q , D ) = B s
(47)
4.2 Benchmark approach 2
For the first situation, we can find that the optimal subsidy for each customer is independent of the budget amount that is allocated to investment in demand-growth strategy; whereas for the second situation, there is dependence relationship between the optimal subsidy for each customer and B d , which means that the optimal subsidy for each customer turns into a mathematic function of B d , that is to say, the optimal subsidy is relevant to the planning stage that can’t be ignored. For benchmark approach 1, we consider the subsidy that is preset only by the administrative sector and B d (the demand-growth strategy), but the decisions are not made jointly. The subsidy may be optimal in the first situation, but it is not optimal in the second situation. The aim of benchmark approach 2 is to analyze how the healthcare service goods supply chain system operates with B d (the demand-growth strategy) and a preset subsidy amount that is determined by the administrative sector only, not determined by the administrative sector and the product provider jointly. The mathematic formulation of benchmark approach 2 is written as follows:
B e n c h m a r k a p p r o a c h 2 : max B d , B s , Q ( θ − s ) E m i n ( Q , D ) − ( B d + B s )
(48)
s u b j e c t t o s E { min ( Q , D ) | B d } ≤ B s
(49)
F B d ( Q ) = P w + s - c P w + s - v
(50)
Case study
In this section, we introduce the Wudang personalized health package to be served as our case study. Apparently, personalized health package belongs to healthcare service goods.
Wudang is located in the south of China. The case and related data derive from the information center of this district. In 2014, Nation Health Commission of China announced that it would regard Wudang District as pilot zone of basic health reform, so Wudang personalized health package can be a representation example. As is known, conventional hospital resource is not enough to meet the residents’ expectation, so personalized health package provided by related health corporations, community health center and other entities is important to meet the residents’ increasing health demand. In the light of personalized health package’s important implications on the residents’ health and happiness, it has received great attention from government departments, corporation entitities and research institutes. Especially, all levels of governments pay high attention to the promotion of personalized health package, and plenty of government departments have expressed explicit target of personalized health package. In fact, Wudang District faces a dilemma that limit the promotion of personalized health package, furthermore, it does not achieve the target.
In this paper, we aim to establish an intervention mechanism to promote the social welfare. Wudang personalized health package include wearable device, tradition Chinese medicine service package, physician service, tele-medicine and fitness product for each customer. Under the current policy, the government of Wudang District have provided subsidies for personalized health package. A recent personalized health package survey conducted by a local institute implies that 69% of the customers regard subsidy as a key factor when purchasing personalized health package. The budget amount that is allocated to investment in demand-growth strategy including the media publicity, the physician training, the demand investigation and the development, which is expected to attract more customers and improve the efficiency of personalized health package. Unfortunately, the district faces a development dilemma and the performance of current intervention mechanism is not efficient. So we introduce an intervention mechanism, in which the decision is determined by the administrative sector and the product provider jointly.
Numerical analysis
In Wudang District, the price of personalized health package that the customers are willing to pay is lower than the product’s cost price. Let p w denotes the willingness price, and the willingness price will be divided into five types, they are 0.85 p c , 0.75 p c , 0.65 p c , 0.55 p c , 0.45 p c respectively. In general, the customer determines the willingness price on the basis of his/her perceptive satisfaction evaluation for the personalized health package. In other words, the type of willingness price depends on the product perceptive satisfaction level. Based on Eqs 1 to 8 , we calculate that H = 3, which implies that P w = 0.65 p c .
6.1 Basic parameters
We use exponential distribution and lognormal distribution to express personalized health package demand and conduct numerical analysis respectively. Personalized health package demand is expected to follow exponential distribution in the early stage, and personalized health package demand is expected to follow lognormal distribution in the subsequent stage, the demand distribution is in accordance with the practice in China. We consider the mean demand function given in Eq 24 , and the administrative sector’s social welfare function given in Eq 25 .
We collect relative data from the government information center of Wudang District and actual investigation. For the Wudang personalized health package, the cost of each healthcare service goods is RMB369, then P w = 0.65, P c = RMB239.8, the top subsidy for each customer who purchase personalized health package product is RMB145, θ = BMB591, v = RMB155. We consider the basic period refers to the time range from 2015 to 2017, and the long period refers to the time range from 2015 to 2023. The basic period situation can be seen in Tables 3 and 4 , whereas the long period situation can be seen in Tables 5 and 6 . For each table, it includes types of intervention mechanism, social welfare, expected profit, subsidy for each customer, expected sales, subsidy amount, and quantity. We use the KNITRO 9.0 Software to calculate the solution of the nonlinear problem.
10.1371/journal.pone.0214655.t003
Table 3 The solutions of the basic period given in exponential distribution.
Intervention mechanism
Social welfare (× 10 5 )
E [ P ( Q )] (× 10 5 )
s
Expected sales (× 10 3 )
B d
B s (× 10 3 )
μ ( B d ) (× 10 3 )
Q (× 10 3 )
JM
13.9
10.0
139
2.91
0
39.6
21.3
3.02
Ben. 1
13.6
9.6
139
2.90
0
45.4
21.3
3.11
Ben. 2
13.6
9.6
139
2.90
0
45.4
21.3
3.11
10.1371/journal.pone.0214655.t004
Table 4 The solutions of the basic period given in lognormal distribution.
Intervention mechanism
Social welfare (× 10 5 )
E [ P ( Q )] (× 10 5 )
s
Expected sales (× 10 3 )
B d (× 10 5 )
B s (× 10 3 )
μ ( B d ) (× 10 3 )
Q (× 10 3 )
cv = 0.8
JM
23.26
19.35
133
4.59
2.32
39.6
12.8
4.52
Ben. 1
17.91
18.56
149
4.32
0
45.4
11.3
4.63
Ben. 2
19.29
17.26
145
4.13
1.92
45.4
10.1
4.21
cv = 1.0
JM
21.13
18.35
133
4.26
2.01
37.1
12.3
4.18
Ben. 1
15.10
15.12
149
4.11
0
42.5
11.0
4.29
Ben. 2
17.03
16.98
145
3.60
1.65
42.5
9.81
3.91
cv = 1.2
JM
19.89
17.91
133
4.12
1.95
36.9
10.91
4.01
Ben. 1
13.55
15.39
149
3.99
0
40.11
9.82
4.15
Ben. 2
16.30
14.99
145
3.45
1.32
40.09
8.62
4.32
10.1371/journal.pone.0214655.t005
Table 5 The solutions of the long period given in exponential distribution.
Intervention mechanism
Social welfare (× 10 5 )
E [ P ( Q )] (× 10 5 )
s
Expected sales (× 10 3 )
B d
B s (× 10 3 )
μ ( B d ) (× 10 3 )
Q (× 10 3 )
JM
112
82
136
23
16.1
322.6
161.3
24.1
Ben. 1
110
78
136
25
0
345.4
161.3
25.1
Ben. 2
109
78
136
25.1
11.2
345.4
161.3
25.2
10.1371/journal.pone.0214655.t006
Table 6 The solutions of the long period given in lognormal distribution.
Intervention mechanism
Social welfare (× 10 5 )
E [ P ( Q )] (× 10 5 )
s
Expected sales (× 10 3 )
B d (× 10 5 )
B s (× 10 3 )
μ ( B d ) (× 10 3 )
Q (× 10 3 )
cv = 0.8
JM
238.9
213.35
133
56.9
20.2
320.6
161.8
52.2
Ben. 1
195.1
188.46
149
45.2
0
325.4
131.3
46.3
Ben. 2
205.9
201.26
145
47.3
16.3
325.4
160.1
49.1
cv = 1.0
JM
185.3
158.95
133
50.3
18.5
317.1
155.3
48.8
Ben. 1
132.0
138.92
149
41.1
0
292.7
121.0
43.9
Ben. 2
157.3
136.18
145
43.2
15.5
301.5
153.9
45.1
cv = 1.2
JM
149.9
139.61
133
50.2
14.1
310.9
160.7
49.1
Ben. 1
118.5
99.59
149
37.9
0
270.1
120.2
38.5
Ben. 2
126.3
114.25
145
40.5
10.9
289.9
149.2
45.2
6.2 Result analysis
Tables 3 to 6 imply that demand uncertainty has an important impact on the social welfare value, then we discuss three specific aspects: (1) subsidy for each customer; (2) social welfare improvement compared with the two benchmark approaches; (3) expected profit improvement compared with the two benchmark approaches.
(1) First, we analyze the subsidy variation between joint mechanism and two benchmarks, the result can be seen in Table 7 .
10.1371/journal.pone.0214655.t007
Table 7 Subsidy variation between three mechanisms.
cv
Basic period
Long period
Exponential distribution
Lognormal distribution
Exponential distribution
Lognormal distribution
0.8
139
133
136
133
1.0
139
149
136
144
1.2
139
145
136
145
In Table 7 , we can find that the subsidy is optimal under the exponential situation, but it is not optimal under the lognormal situation, because the optimal subsidy for each customer turns into a mathematic function of B d when the demand follows lognormal distribution.
(2) Next, we analyze the social welfare improvement compared to the two benchmark approaches, the result can be seen in Table 8 .
10.1371/journal.pone.0214655.t008
Table 8 Social welfare improvement compared to two benchmark approaches.
cv
Basic period
Long period
Exponential distribution (Ben.1, Ben.2)
Lognormal distribution (Ben.1, Ben.2)
Exponential distribution (Ben.1, Ben.2)
Lognormal distribution (Ben.1, Ben.2)
0.8
0.02%
29.8%, 20.6%
0.18%
22.4%, 16.0%
1.0
39.0%, 24.1%
40.3%, 17.8%
1.2
46.8%, 22.0%
26.5%, 18.7%
In Table 8 , we can find that when the demand follows exponential distribution, the welfare difference between the joint intervention mechanism and benchmark approaches is tiny. However, when the demand follows lognormal distribution, the social welfare of the joint intervention mechanism established by us is higher than the two benchmark approaches. Especially, benchmark 1 is current policy, and we find that the social welfare for the joint intervention mechanism has a more apparent improvement than current policy, which implies that our intervention mechanism is effective.
(3) Finally, we analyze the expected profit improvement compared to the two benchmark approaches, the result can be seen in Table 9 .
10.1371/journal.pone.0214655.t009
Table 9 Expected profit improvement compared to two benchmark approaches.
cv
Basic period
Long period
Exponential distribution (Ben.1, Ben.2)
Lognormal distribution (Ben.1, Ben.2)
Exponential distribution (Ben.1, Ben.2)
Lognormal distribution (Ben.1, Ben.2)
0.8
4%
4.30%, 12.1%
5%
13.2%, 5.80%
1.0
13.8%, 14.1%
14.4%, 16.7%
1.2
16.4%, 19.5%
37.5%, 20.0%
In Table 9 , we can find that when the demand follows exponential distribution, the expected profit difference between the joint intervention mechanism and benchmark approaches is tiny. However, when the demand follows lognormal distribution, the expected profit of the joint intervention mechanism established by us is higher than the two benchmark approaches. Especially, the expected profit improvement will increase with the increase of cv .
Conclusion
In this paper, we consider a research problem that regulates the supply chain system for healthcare service goods by an intervention mechanism, and the supply chain system is composed of a product provider and an administrative sector. Healthcare service goods belongs to public goods, so we should not regard the expected profit as the sole objective. Different from pre-existing research, we establish a supply chain intervention mechanism based on social welfare maximization for healthcare service goods. In specific, we analyze the relationship between the optimal subsidy for each customer and the B d (the budget amount that is allocated to investment in demand-growth strategy). We attempt to analyze the problem by using variable fuzzy set method and bilevel programming model.
The first contribution of our study is that the intervention mechanism for healthcare service goods can generate more abundant social welfare than the two benchmark approaches that are used generally in practice. The Wudang personalized health package case study implies that our mathematic model that is applied to intervention mechanism for healthcare service goods is scientific and effective. Compared with the two benchmark approaches, our joint intervention mechanism can help the administrative sector to achieve the target and increase social welfare. The second contribution is that the evaluation model that is used to obtain the willingness piece is pivotal for the intervention mechanism. Furthermore, B d (the demand-growth strategy) plays a key role in case study. Especially, the optimal subsidy for each customer is a mathematic function of B d under the second situation. We explore a perceptive satisfaction evaluation indicator system for healthcare service goods, and the indicator system include basic medical care, basic public care and perceptive value mainly. Besides, two intervention strategies composed of demand-growth strategy and subsidy strategy are used to the combination of intervention mechanism jointly.
In fact, tax credit also plays an important role in promoting the adoption of healthcare service goods, but we do not consider tax credit in our paper. As is known, the demand of the healthcare service goods may be influenced by regional policy, educational level, consumption structure and medical service convenience, which should be analyzed deeply in the future.
Supporting information
S1 File
Supporting information file (data).
(XLSX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Hand-foot-mouth disease (HFMD) is a common acute infectious disease, which is featured by fever, painful sores in the mouth, and a rash with blisters on the hands, feet and buttocks [1] , [2] . The dominant strain is Coxsackievirus A16 (CA 16) and Enterovirus A71 (EV 71). HFMD occurs mainly among children under 5 years of age, who are at the greatest risk. The infection is typically mild and self-limiting with few complications [3] . However, severe cases with complications of central nervous system, always caused by EV 71 [4] , would occasionally lead to neurological sequelae or subsequent quick death.
Since the 1970s, epidemics of HFMD have been reported worldwide in Hungary, southeastern Australia, United States, England, Wales, Malaysia and Singapore [5] – [10] . In China, after its first emergence in Shanghai in 1981 [11] , several sporadic cases were reported in Beijing, Tianjin, Jilin, Guangdong and other provinces [12] – [18] . In 2008, an unprecedented large-scale epidemic of HFMD broke out first in Fuyang of Anhui province [19] . The persistent outbreaks sounded the alarm bell to the Chinese government. Therefore, on May 2, 2008, the Ministry of Health of the People's Republic of China added HFMD to its category ‘C’ of notifiable diseases, which means all the cases must be reported through a national web-based system named China Information System for Diseases Control and Prevention (CISDCP) built in 2004 [20] .
Statistics show that in 2008, among the category ‘C’ of notifiable diseases [21] , the number of reported cases of HFMD ranked second after other infectious diarrhea but the reported deaths ranked first. However, both the reported cases and deaths of HFMD had always ranked the top among the category ‘C’ of notifiable diseases from 2009 to 2012 [22] – [25] . To tackle the growing threat, the government and the public health officials were aware of the importance of early detection of outbreaks, early recognition, early intervention in the disease and commencing aggressive therapy to minimize the impact exerted by HFMD.
Recently, researchers are interested in forecasting the incidence of infectious disease, using the liner time series forecasting models such as seasonal auto-regressive integrated moving average (ARIMA) models [26] – [30] . However, most real time series always contain nonlinear structures, from which liner models cannot yield adequate results [31] . To fit the nonlinear structures, nonlinear models such as artificial neural networks (ANNs), bilinear, auto-regressive conditional heteroskedasticity models (ARCH) performed better than liner models. Among them, ANNs have flexible nonlinear function mapping capability, which can approximatecontinuous measurable function with arbitrarily expected accuracy [32] , so that nonlinear structures can performed well [33] . However, when it comes to real time series that contain liner and nonlinear structures, neither linear nor nonlinear models seems to be satisfactory. To solve this problem, this paper attempts to use hybrid models combining liner and nonlinear models to improve the prediction accuracy by taking the advantages of both models. These hybrid models have been found to be viable contenders to various traditional time series models [31] , [34] – [37] .
Considering the variety of influencing factors on HFMD, this paper proposes a new hybrid model combining seasonal ARIMA and nonlinear auto-regressive neural network (NARNN) to predict the incidence of HFMD in Shenzhen. The aim of this paper is to describe the future trend of this disease and to achieve the early detection and early warning by mathematical method.
Method
Setting
Shenzhen is the first – and one of the most successful – Special Economic Zones (SEZs) in China as well as the largest manufacturing base in the world. Because of this, Shenzhen becomes the largest migrant city in China with a population of roughly ten million in 2010 Census. About six million are migrant workers who return their homes on weekends or festivals, and live in factory dormitories during the workdays. Therefore, it's difficult to obtain the exact annually average statistical population. Instead, this paper used the numbers of incidence cases as an evaluation indicator.
Data resource
This paper used the incidence cases of HFMD during January 2008 to August 2012 as the training data, the data during September 2012 to November 2012 as the validation set, and the data during December 2012 to May 2013 as the forecasting set. All the data were obtained from CISDCP mentioned above. The CISDCP has two important improvements compared with the previous reporting system [20] . The first is that diseases are reported in real time, which allows public-health officials to have real-time information and also helps to reduce the under-reporting of infectious diseases. The second improvement includes the availability of case-based reporting instead of aggregate reporting, which immediately helps public-health officials comprehensively identify the characteristics, nature and location of a particular disease outbreak or clusters of cases. In addition, a series of measures have been taken to improve the quality of data reporting, such as annual field audits, national training conferences on data entry, routine quality checking and quality reporting of data at varied levels of medical institutions. A crucial reason for the research to consider is that the data is viewed as being of high quality with respect to accuracy, comparability, timeliness, relevance and usability under the category ‘C’ of notifiable diseases management for HFMD.
Statistical method
The Auto-regressive Integrated Moving Average Models
Given a stationary time series of data , an auto-regressive moving average (ARMA) model is respectively composed by two parts, an auto regressive (AR) part of order p and a moving average (MA) part of order q.
An AR model of order p, denoted by AR , is given by
an MA model of order q, denoted by MA , is given by
an ARMA model of order p and q, denoted by ARMA , is given by where is the random error term assumed to be independent and referred as a white noise identically distributed with a mean of zero and equal variance [38] . It's usually interpreted as external effect that the model can't explain. Stationarity occurs in a time series when the mean value of the series remains constant over the time series. Frequently, differencing is needed to achieve stationarity in the model. It is denoted by ARIMA , where d is the value of differencing orders. In addition, a top priority of the model building is to identify the appropriate model order [31] . Box and Jenkins [38] proposed to identify the order of the model by the autocorrelation function (ACF) and the partial autocorrelation function (PACF) as the basic tool. If monthly data were used in the analysis, periodicity of series would be shown, which was more likely to lead to useful forecasting and should cover at least 2 periods [29] . Seasonal terms are also incorporated into ARIMA model, which are denoted by seasonal ARIMA , where S is often referred as the value of per period. The residuals, the differences between each observation and prediction according to the model, should also be inspected, ideally small and show no secular or seasonal trend [29] .
The Artificial Neural Networks
ANNs consist of a large number of highly connected nonlinear simple unit and store information in the connections between units by self-learning and self-organizing [39] . The commonest type of ANNs is the single hidden layerback-propagation (BP) neural network, which is a kind of multilayered feedforward neural network. The studied process of the BP neural network is formed by two parts: signal forward-propagating and error signal reverse dissemination, and the input single spreading from the input layer, captured by the hidden layer passing on to the output layer. When expected output value can't be obtained from the output layer, the process turns to error signal reverse dissemination stage, and with the back-propagation of the error is repeated, the error signal reduces and the correct response rate rises [40] . In this paper, NARNN, which is a dynamic neural network based on the BP neural network with the feedback layers to approximate the non-linear function [39] , is applied. The main differences in the design process occur because the inputs to the dynamic network are time sequences so it is good at time series forecasting.
The hybrid models with seasonal ARIMA and NARNN
In this section, the hybrid model building procedures will be described step by step in detail. In the seasonal ARIMA model stage, the main aim is to extract the linear information. In the beginning, one or more abnormal observations (AO) based on the real events, which could explain the reasons of abnormalities, would be found. Each abnormal observation is replaced with a missing value, and filled by the function of SAS ‘expand’ procedure. Next in the identified step, ACF and PACF could be detected visually by examining a regression line scatterplot, so that the AR and MA components and s would be identified with possible values. If the data is non-stationary, regular differencing or seasonal differencing is needed then, and the value of d and s would be the orders of differencing. Augmented Dickey-Fuller Unit Root (ADF) test is used to identify whether the series after differencing is stationary or not. Once the orders are specified, estimation of the model parameters by least squares estimation is thus straightforward. The parameters with significant statistical difference are reserved and the others are excluded. The next step of model building is the diagnostic checking of model adequacy. This is basically to check whether the model assumptions are satisfied or not. If the model were not adequate, a new tentative model would be identified. The steps of parameter estimation and model verification are not stopped until the new tentative model is satisfied. The autocorrelation and partial autocorrelation of residuals help to verify whether the series of residuals to be the white noise by using Boxing-Ljung Q-test. In the last step, the one with the lowest Bayesian Information Criterion (BIC) value is chosen to be the best-fitted model. P<0.05 is considered statistically significant. The training set is used to build the model and get 3 steps forecasting for validation, and the performance is evaluated by validation set. After that, the observations of the training set and validation set are utilized to build a new model again, repeating the same modeling procedures. All ARIMA modeling is implemented via SAS9.2 system.
In NARNN stage, the main aim is to model the nonlinear and probable linear relationships existing in the residuals of linear modeling and the original series. To model the NARNN, it is generally best to start with the neural network time series tool (ntstool) in the MATLAB, which can automatically generate command-line scripts in accordance with the demand of the research. First, the target series is inputted to obtain the command-line script and the next 6 months data set for multi-step-ahead prediction. Then the division of data using the provided divider and function, which divides the data into contiguous blocks, respectively 80% of the target series for training, 10% for validation and 10% for testing, is set up. In the last step, the number of hidden neurons and feedback delays are adjusted by trial and error, based on the error autocorrelation plot, the time series response plot and the Mean Square Error (MSE) of training and testing data, to select the optimal model. With this tool, the other important parameters are set as the defaults, such as the tan-sigmoid transfer function in the hidden layer, the linear transfer function in the output layer and the Levenberg-Marquardtal training-algorithm.
Based on the adjusted residuals, the expected monthly incidence cases of HFMD can be obtained. The eventual predictions are the sum of seasonal ARIMA predictions and adjusted residuals. It is , where denotes the predictions of linear model and denotes the residuals adjusted by nonlinear model.
Results
Cases distribution and demographic characteristics
During the study period, the reported cases of HFMD in Shenzhen increased every year with a slight rise in 2009 and a dramatic rise since 2010 ( Table 1 ). The amounts of severe cases and fatal cases in each year were parallel with the reported cases except a significant decrease in 2012. The male predominance was found in each year, and the proportion of male to female remained stable. The age distribution was similar every year, and children <5 years old were under greatest risk, especially those between 1 to 3 years old. The annual number of the cases occurring among different child-care centers showed little change. The majority of patients were home-cared, with kindergarten-cared and school-cared as the second and third respectively. The results of laboratory diagnosis showed that the first dominant strain of the epidemic in Shenzhen was EV 71, and there was a significant increasing trend of other virus supplanting Cox A16 to be the second dominant strain.
10.1371/journal.pone.0098241.t001 Table 1
Case distribution and demographic characteristics of HFMD in Shenzhen from January 2008 to November 2012.
2008
2009
2010
2011
2012*
Number of cases
7149
9121
23288
24838
30021
Number of severe cases
4
37
92
150
55
Number of fatal cases
1
5
6
6
2
Gender
Male
4574(63.98%)
5802(63.61%)
14564(62.54%)
15818(63.68%)
18763
Female
2575(36.02%)
3319(36.39%)
8724(37.46%)
9020(36.32%)
11258
Age
0∼
635(8.88%)
1037(11.37%)
2518(10.81%)
3143(12.65%)
3943(13.13%)
1∼
1083(15.15%)
1529(16.76%)
6657(28.59%)
7692(30.97%)
9219(30.71%)
2∼
1718(24.03%)
2462(26.99%)
4988(21.42%)
5087(20.48%)
5793(19.30%)
3∼
1398(19.56%)
1892(20.74%)
4208(18.07%)
4386(17.66%)
5130(17.09%)
4∼
1132(15.83%)
1060(11.62%)
2220(9.53%)
2241(9.02%)
3022(10.07%)
>5
1183(16.55%)
1141(12.51%)
2697(11.58%)
2289(9.22%)
2914(9.70%)
Form of child care
Home care
4633(64.81%)
6886(75.50%)
17147(73.63%)
19002(76.50%)
22343(74.42%)
Kindergarten care
2078(29.07%)
1827(20.03%)
5033(21.61%)
4850(19.53%)
6553(21.83%)
School care
364(5.09%)
279(3.06%)
878(3.77%)
779(3.14%)
888(2.96%)
Type of pathogen
Number of laboratory diagnosis
157
86
169
142
55
EV 71
107(68.15%)
64(74.42%)
111(65.68%)
107(75.35%)
38(69.09%)
CA 16
50(31.85%)
21(24.42%)
57(33.73%)
16(11.27%)
5(9.09%)
Others
0(0.00%)
1(1.16%)
1(0.59%)
19(13.38%)
12(21.82%)
Trend and seasonality of HFMD epidemic
The monthly numbers of reported cases of HFMD were graphically shown in Figure 1 with an increasing trend, a clear yearly periodicity and significant fluctuations in its yearly mean. During the study period, the highest peaks of seasonal periodicity occurred in April and remained high until July; the second small peaks appeared during September to November. From November to February of the following year, the incidence of HFMD was at a low level until the next epidemic started.
10.1371/journal.pone.0098241.g001 Figure 1
Series of observations of HFMD in Shenzhen.
Series 1 shows the observations of the training set (from January 2008 to August 2012). Series 2 shows the observations of training set without the abnormal observation (AO). Series 3 shows the series 2 achieving stationary after one regular differencing and one seasonal differencing (d = 1, s = 12). Series 4 shows the validation set (from September 2012 to November 2012).
Results of the validation set
In our opinion, due to the upgrade of HFMD to the category ‘C’ of notifiable diseases by the Ministry of Health of the PRC in May 2008, and the inclusion in CISDCP, the value in May 2008 is considered as an AO, which could be explained by increasing efforts in detecting and reporting HFMD. The differencing series appears stationary with a same mean value and variance over time ( Figure 1 ). This suggests that it would be appropriate to consider an order d = 1 and S = 12 in the fitted model given by seasonal ARIMA . We could get the same conclusion by graphing the ACF and PACF ( Figure 2 ). After the steps of parameter estimation and model verification, the model with order (2, 0) is best fitted to the data ( Table 2 ). The final mathematical form of the seasonal ARIMA model is ARIMA .
10.1371/journal.pone.0098241.g002 Figure 2
Autocorrelation function (ACF) and partial autocorrelation function (PACF) plotted against time lags.
A and B show ACF and PACF of the training set. C and D show ACF and PACF of the training set after one order of regular differencing and one order of seasonal differencing (d = 1, s = 12). After differencing, Most of the correlations fall around zero within their 95% confidence intervals (95%CI, U95: upper limit of 95%CI, L95: lower limit of 95%CI) except the one at the first lag, which is indicated the series would achieve stationary.
10.1371/journal.pone.0098241.t002 Table 2
Parameter estimation and model verification of seasonal ARIMA model with minimum BIC Value (3, 0) = 13.732.
Lag
Parameter
Estimate
P-value
Q(18) **
P-value
Estimate'
P-value
Q(18)' **
P-value
0
MU
−6.780
0.941
13.04
0.599
-
-
-
-
1
AR 1,1
0.093
0.552
-
-
-
-
2
AR 1,2
−0.387
0.013 *
−0.406
0.008 *
18.16
0.379
3
AR 1,3
−0.313
0.096
-
-
-
-
* Parameter estimation was considered statistically significant (P<0.05). **Box-Ljung test at lag 18 for the series of residuals.
The optimum NARNN this paper proposed to apply to forecast the residuals series produced by best-fitted seasonal ARIMA model has 12 hidden units and 4 delays. The MSE of training, validation, and testing data subsets are 8.2037×10 4 , 1.0477×10 6 , and 1.9939×10 6 respectively. The correlations of prediction errors fall within the 95% confidence limits around zero, therefore the model is adequate for the data ( Figure 3 ).
10.1371/journal.pone.0098241.g003 Figure 3
Time series response and error autocorrelation plot in training, testing and validation set of NARNN.
A shows the time series response in training, testing and validation set. B shows the error autocorrelation plot in training, testing and validation set. In B, all the correlations fall within the 95% confidence limits around zero except the one at zero lag, which is indicated the model would be adequate.
After model building, the predictions of the validation set are thus obtained ( Figure 4 ), which are very close to the observations and indicate that the proposed model was fitted and made good performance.
10.1371/journal.pone.0098241.g004 Figure 4
Series of the predictions of the validation set.
Series 1 shows the observations of training set without the abnormal observation. Series 3 shows the validation set. Series 2 shows the predictions of training set obtained by hybrid model combined with ARIMA and NARNN with 12 hidden units and 4 delays, and Series 4 shows the predictions of validation set obtained by the same model. Each prediction is very close to each observation.
Results of forecasting set
The best-fitted hybrid model of all the observations is with a combination of ARIMA and NARNN with 15 hidden units and 5 delays, and the predictions of forecasting set are shown in Figure 5 . It is easy to find out that each prediction obtained from the hybrid model is very close to each observation. The value of February 2011 is a negative value (−181.166) ( Table 3 ), and the value of the observation at the same time is 84, which indicates that if observations were extremely small, the corresponding predictions would be negative. Therefore, the trough of the epidemic in 2013 would occur in January. In addition, compared with previous years, the expected incidence cases in 2013 would rise rapidly, and the first peak would begin in February ( Table 2 ). The second peak would occur in April and the highest in May. The amount of total expected cases would be higher than any other previous years.
10.1371/journal.pone.0098241.g005 Figure 5
Series of predictions of all the observations.
Series 1 shows the observations of the training set without the abnormal observations and the validation set. Series 2 shows the predictions of series 1 and the expected cases of forecasting set (from December 2012 to May 2013) obtained by ARIMA and NARNN with 15 hidden units and 5 delays. There is a significantly increasing trend in first half of 2013.
10.1371/journal.pone.0098241.t003 Table 3
Expected incidence cases and observations in the corresponding period from 2010 to 2013 in Shenzhen.
Month
Observations
Expected cases
2010
2011
2012
2013
December *
504
631
1154
−965.03
January
658
276
271
−1879.58
February
402
84
426
4138.26
March
1415
252
1843
1858.17
April
3481
675
3659
4061.86
May
4267
3702
5684
6163.16
Total
10727
5620
13037
16221.45 **
*December in the previous year. **We made the assumption that the expected cases in January and February were zero.
Discussion
During the five-year study period, HFMD cases in Shenzhen increased every year. However, during the H1N1 pandemic in 2009, preventive measures such as massive use of face masks, school closures and reduction of outdoor activities could partly explain why there were the lower HFMD incidence cases, which was similar with the situation in Guangzhou [41] . The numbers of severe cases and fatal cases dropped significantly in 2012, which indicated that the measures on controlling severe and fatal cases were playing their roles. The found of male predominance was similar with the results of Zeng's study in Shanghai [42] and Zhu's study in mainland China [43] , which could be attributed to more restlessness and more opportunities for boys to contract the disease compared with girls. Patients with HFMD were mostly aged 1∼3 years old, which was similar with previous study [3] , [42] – [45] . The majority of patients were home-care which was different from situations in Shanghai [42] . A possible reason is that the proportion of migrant workers in Shenzhen substantially outnumbered that in Shanghai and migrant parents don't have enough time to take care of their children. Moreover, they are generally less educated and under poorer living conditions or financial status, which makes them the high-risk group of infection. The most commonly isolated enterovirus of HFMD cases were types associated with EV 71, which was similar with situations in mainland China [46] . However, the reported cases of infection by other enterovirus increased significantly, and public health officials should alert the outbreaks of new types of HFMD. The seasonal distribution in Shenzhen was similar to the southern region of China [46] .
The new hybrid model is applied to forecast the incidence cases of HFMD, and the results show that the combination model could be an effective way for prediction. In previous studies, multivariate SARIMA models with some meteorological variables were built to achieve a better predicting performance of statistical methods [41] . Nevertheless, the epidemic of HFMD is influenced not only by meteorological factors [41] , [47] – [50] , but also by total population and population density, rural versus urban living, literacy, enterovirus positivity, sanitary conditions and population susceptibility, even other unknown factors [46] . The hybrid model, whose aim is to reduce the risk of using an inappropriate model and which obtain more accurate results, takes advantages of the unique strength of seasonal ARIMA and NARNN in linear and nonlinear modeling. In such a model, the seasonal ARIMA model fits the non-stationary linear component, whilst the neural network model fits nonlinearity [31] .
When the epidemics or outbreaks occur, infectious disease should be investigated and the causes of them fully understood. However, this is impossible under any circumstances for scientific and practical reasons. Referring to experience, the usefulness of forecasting expected cases of HFMD performs not only in detecting outbreaks or providing probability statements, but also in giving decision makers a probable trend of the variability of future observations that contains both historical and recent information. As Table 3 shows, the comparison implies that the pandemic would increase earlier than the previous years, and the incidence population would come to a bigger size. The expected incidence trend is proposed to provide them with predicting future trend, early forecasting and detecting peak time and scale of HFMD outbreaks when observations significantly exceed standard thresholds, and evaluating effectiveness of health measures when the observations are lower than the forecasting trend, so that they can improve surveillance, make prevention and control strategies, and allocate health resources. In the practice, the key point is to keep the forecasts at hand completely, ideally on display, and to write in new observations to update the data as soon as they become available [29] , especially when new control measures are taken, or else, the model would have no chance to help detect epidemic or outbreaks sooner than otherwise possible. At the same time, the method is simple and easy to get started, which can be applied availably to field epidemiological investigation, for it only requires investigators to have a computer.
However, some flawed parts may affect the outcomes. The foundation of model building is the data reported to CISDCP, and the quality of reported cases everyday directly influence the forecasting performance of the model. Good quality of data collection by this system has been demonstrated by a recent data quality inspection report, apart from the following problems [51] . The reported data of HFMD, collected retrospectively when doctors investigate patients, would be inaccurate. The reported data would not be comprehensive because some mildly affected patients may not go to any medical institution for treatment, and those patients are not reported to the system. Lastly, a minor increase in the number of reported cases would occur due to the enhancive consciousness of the importance of reporting HFMD in medical institutions by HFMD-related policies. In addition, the hybrid models are combined with the liner model and the nonlinear model, thus they have the disadvantages of both linear and nonlinear models. Since the ANN models belong to the blackbox type of models, it may limit the model's ability to extrapolate beyond its training data as well as the implementation of subjective initiatives by operators in ANN analysis.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Identifying environmental conditions underlying the division of species into smaller units is central for understanding ecological and evolutionary processes and for the conservation management of biodiversity. In highly mobile species that are distributed across continuous environments with few barriers to dispersal, it is expected that persistent gene flow will stifle genetic differentiation and speciation. Nevertheless, there is growing recognition that gene flow can be limited even in the absence of geographical barriers, both in terrestrial and aquatic environments [1] , [2] . A detailed knowledge of how landscape characteristics structure populations has therefore become an important focus of molecular ecological research [3] , leading to the emerging field of landscape genetics [3] , [4] . This multidisciplinary approach aims to complement genetic data with lines of evidence from other areas such as spatial statistics and landscape ecology in order to understand the effects of the landscape on the spatial distribution of genetic diversity [3] , [5] , [6] . Although extensively applied in terrestrial systems, this approach has been used less frequently in the marine environment [4] ; but see [7] , [8] .
The study of connectivity in marine systems can be challenging due to the absence of obvious barriers to dispersal and generally large population sizes of marine organisms that often resist genetic divergence, leading to low statistical power to detect population structure [8] , [9] . Therefore, the use of an integrative approach such as the one used in landscape genetics (or ‘seascape genetics’ when applied to the marine environment) has provided valuable insights into which factors may be shaping genetic structure in the world's oceans [7] , [10] . Biogeographic barriers and environmental variables such as ocean currents, upwelling, variation in sea surface temperature and salinity are some of the factors that have been proposed to explain genetic diversity and structure in marine organisms [9] , [10] , [11] . However, most of these studies have been conducted in organisms with larval dispersal. In active marine dispersers such as sharks and dolphins, where dispersal potential is dependent upon individual vagility, the interplay of environmental features and genetic structure has remained largely untested (but see [12] ). Although differences in salinity, temperature and productivity levels have been suggested to explain genetic discontinuities in dolphins [13] , [14] , [15] , [16] , a direct relationship between such oceanographic features and genetic structure has only been recently evaluated for two coastal dolphin species with limited distribution: the franciscana ( Pontoporia blainvillei ) [12] and the humpback dolphin ( Sousa chinensis ) [17] . These authors found that heterogeneity in chlorophyll concentration, water turbidity and temperature likely influenced the occurrence of genetically distinct populations of these species along the coast of Argentina and in the Western Indian Ocean, respectively.
In this study we use as model a highly mobile, widely distributed cetacean species belonging to the genus Delphinus , the short-beaked common dolphin. Common dolphins occur in all oceans from tropical to temperate waters. Two species and four subspecies are currently recognized: the short-beaked common dolphin, Delphinus delphis Linnaeus, 1758, distributed in continental shelf and pelagic waters of the Atlantic and Pacific Oceans; the long-beaked common dolphin, Delphinus capensis Gray, 1828, distributed in nearshore tropical and temperate waters of the Pacific and southern Atlantic waters; D. d. ponticus Barabash, 1935, restricted to the Black sea; and D. c. tropicalis van Bree, 1971, restricted to the Indian Ocean [18] . However, due to discordance between morphological and genetic characters, the phylogenetic relationships and taxonomy within the genus, particularly in regard to the specific status of the long-beaked form, are still under debate (Amaral et al. unpublished data; [19] ).
Short-beaked common dolphins are known to occur in large groups of dozens to hundreds of individuals. Although their social structure is still poorly understood, individuals seem to group irrespective of genetic relationships, with possible gender and age segregation [20] . However, there is a gap in knowledge if these findings are representative for common dolphins in other geographic regions. The movements of common dolphins are thought to be largely determined by those of their potential prey (e.g. [21] ) and their diet varies between locations and seasons [21] , [22] . Nonetheless, they generally depend on small, mesopelagic shoaling fishes such as scombroids and clupeoids, and squids [21] , [22] . It has been suggested that short-beaked common dolphins often prefer specific water masses [15] , [23] , [24] and in the Eastern Tropical Pacific they occur preferentially in upwelling-modified waters [23] .
Genetic studies conducted so far have shown significant genetic differentiation among populations inhabiting different oceans and different coasts of the Atlantic Ocean [19] , [25] . However, within each side of the Atlantic Ocean, no genetic structure has been detected, suggesting a lack of strong dispersal barriers in these areas [25] , [26] . Within the Pacific Ocean, results from regional studies have reported fine-scale (≤1000 kms) population genetic structure in short-beaked common dolphins occurring off the USA coast (Chivers et al. unpublished data), off the Eastern [15] Australian Coast and around New Zealand (Stockin et al. unpublished data). Particular oceanographic characteristics, such as ocean currents and temperature and salinity differences have been pointed out as likely factors limiting movement of short-beaked common dolphins (Chivers et al. unpublished data; [15] , [27] ). However, a direct evaluation of the influence of oceanographic variables on the genetic structure of this species has never been carried out.
Our aim is to assess the relative influence of key oceanographic variables on population subdivision of short-beaked common dolphins at a range of medium to large spatial scales, including within ocean basins and across oceans. To achieve this aim we have sampled populations inhabiting the Atlantic, Pacific and Indian Oceans and used remote sensing data under a seascape genetics approach. The global distribution, high mobility, and putatively close association of short-beaked common dolphins with water masses, makes them an excellent model species to test for interactions between variation in environmental factors and genetic structure, contributing towards an understanding of ecological processes affecting population connectivity in the sea.
Methods
Ethics Statement
This study was conducted according to relevant national and international guidelines. No ethics approval was considered necessary because the animals were not handled directly. Permissions for collecting samples were obtained separately in countries where it was required (Macquarie University Animal Ethics Committee, Australia; Southwest Fisheries Science Center Ethics Advisory Committee, USA; Institute for Nature Conservation and Biodiversity, Portugal; and Department of Conservation, New Zealand). CITES permits numbers used to export/import samples were: 07US168545/9, 08US198270/9, 2009-AU-550713, 2009-AU-57-1209, 10NZ000011, PT/CR-0060/2009, PT/LE-0043/2009, PT/CR-005372009, PT/CR-0054/2009, PT/CR-0055/2009, PT/CR-0056/2009, PT/CR-0057/2009, PT/CR-0058/2009, PT/CR-0059/2009.
Sampling
We used samples from seven oceanic regions ( Figure 1 ): the Northeast Atlantic (NEATL), n = 75; the Central Eastern Atlantic (CEATL), n = 29; the Northwest Atlantic (NWATL), n = 38; the Northeast Pacific (NEPAC), n = 40; the Southwest Pacific, n = 35 (encompassing eastern Australian waters, SWPAC_AUS) and n = 39 (encompassing New Zealand waters, SWPAC_NZ) and the Southeast Indian Ocean (southern Australian waters, SEIND), n = 27 ( Table 1 ). All tissue samples were obtained from either stranded animals (103 samples) or from skin biopsies (178 samples) collected from free-ranging dolphins. Tissues were stored either in ethanol or in 20% DMSO/saturated NaCl.
10.1371/journal.pone.0031482.g001
Figure 1
Oceanic regions sampled.
Map showing sampling locations for the short-beaked common dolphin populations analysed in this study. (NEPAC – Northeast Pacific; NWATL – Northwest Atlantic; CEATL – Central eastern Atlantic; SEIND – Southeast Indian Ocean; SWPAC_AUS – Southwest Pacific Australia; SWPAC_NZ – Southwest Pacific New Zealand).
10.1371/journal.pone.0031482.t001 Table 1
Genetic diversity measures of 14 microsatellite loci for the short-beaked common dolphin populations analysed in this study.
Region
N
N a
A r
H
E
H
O
F
IS
NE Atlantic (NEATL)
75
10.500
8.371
0.789
0.774
0.020
CE Atlantic (CEATL)
29
8.214
7.511
0.739
0.687
0.072
NW Atlantic (NWATL)
38
9.286
8.184
0.785
0.745
0.051
NE Pacific (NEPAC)
40
11.643
9.424
0.784
0.730
0.069 *
SW Pacific Australia (SWPAC_AUS)
35
10.643
8.485
0.782
0.726
0.073 *
SW Pacific New Zealand (SWPAC_NZ)
39
10.500
9.130
0.792
0.697
0.121 *
SE Indian (SEIND)
25
7.571
7.163
0.700
0.696
0.006
Total/Mean
281
9.765
8.324
0.767
0.722
N - sample size; N a - mean number of alleles; A r - allelic richness; H E - expected heterozygosity; H O - observed heterozygosity; F IS - inbreeding coefficient.
*value statistically significant at P <0.05.
DNA extraction and microsatellite genotyping
Genomic DNA was isolated from skin or muscle using a standard proteinase K digestion and two phenol-chloroform and one chlorofom-isoamyl extractions followed by ethanol precipitation [28] for samples originated from stranded animals or, alternatively, using a salting-out protocol [29] for samples originated from biopsies. DNA quality and concentration was verified using Thermo Scientifc NanoDrop 1000 Spectrophotometer (Thermo Fisher Scientific Inc.). Samples from NEPAC and NWATL were provided as DNA by the Southwest Fisheries Science Center, Marine Mammal and Turtle Research Sample Collection (SWFSC-NOAA, La Jolla, CA).
All samples were genotyped at 14 polymorphic microsatellite loci: 7 tetranucleotide (Tur4_80, Tur4_87, Tur4_92, Tur4_105, Tur4_141, Tur4_142; [30] and Dde59 [31] and 7 dinucleotide (Dde66, Dde70; [31] ), KW2, KW12 [32] , EV1 [33] , MK6 and MK8 [34] . The forward primer for each primer pair was labelled with a M13 tag [35] . Fluorescent dyes were also labelled with the M13 tag. Amplification reactions contained 50–100 ng DNA, 1× GoTaq® reaction buffer (Promega), 2.5 mM MgCl 2 , 0.2 mM dNTPs, 0.1 µM of each primer and 1 U GoTaq® Taq DNA polymerase (Promega). The thermal cycler profile for the tetranucleotide loci and Dde66 and Dde70 consisted of initial denaturation at 94°C for 3 min followed by a touchdown profile for 5 cycles with the annealing temperature starting at 63°C and decreasing 2°C per cycle, followed by 30 cycles with an annealing temperature of 53°C, and a final extension step at 72°C for 10 min. The tetranucleotide loci were amplified in multiplex after optimization. For the remaining dinucleotide loci, conditions followed the original publications. All reactions included both positive and negative controls. Following amplification, samples were mixed with an internal size standard (LIZ 500) and run on an ABI 3130 Genetic Analyzer. The GeneMapper v.4.1 software (Applied Biosystems, CA) was used for sizing of allele fragments.
Data analysis
Genetic diversity
The program Micro-checker v.2.2.3 [36] was used to check for the presence of genotyping errors such as scoring errors due to stuttering, large allele dropout or evidence for null alleles. Departures from Hardy-Weinberg Equilibrium were tested for each population using the Fisher exact test in Genepop v.4.0 [37] . Genepop was also used to test for linkage disequilibrium between loci. Samples were grouped into 7 putative populations according to their geographical origin as described above. Genetic diversity measures such as mean number of alleles per locus and observed ( H O ) and expected ( H E ) heterozygosities were calculated in Arlequin v.3.5.1 [38] and allelic richness ( A R ) calculated using FSTAT v.2.9.3 [39] .
Genetic differentiation
Three different measures of population differentiation were used: the fixation index F ST , estimated using FSTAT [39] ; the analogous R ST , estimated using Genepop v.4.0 [37] ; and the statistic Jost's D [40] , estimated using SMOGD v.1.2.5 [41] . The latter has been shown to provide a more accurate measure of differentiation when using highly polymorphic microsatellite loci [40] . Additionally, we tested for a mutation effect on genetic structure by randomly reassigning allele sizes while keeping allele identity the same [42] . The test was conducted in spagedi v.1.3 through 10,000 permutations. R ST values significantly larger then F ST values indicate that mutation, in addition to drift and gene flow, has contributed to frequency differences among samples, which in some cases can be interpreted as phylogeographic signal [42] .
In order to visualize relationships among putative populations based on genetic variation, we performed a principal component analysis (PCA) on a table of standardised allele frequencies using the adegenet and ade4 packages in R [43] . In addition, we performed an analysis of nonmetric multidimensional scaling (MDS, [44] ) on each of the genetic distance matrices using the primer computer package [45] .
An analysis of molecular variance, AMOVA [46] was conducted in Arlequin to assess population structure. Different hierarchical levels were tested, considering differences occurring between populations in different oceans and within the same ocean basin.
A Bayesian approach to identify the number of populations ( K ) present in the dataset was implemented in the program STRUCTURE v.2.3.3 [47] , [48] . The admixture and the correlated allele frequencies models were implemented since we expect that allele frequencies in the different populations are likely to be similar due to migration or shared ancestry. Sampling locations were used as prior to help detect population structure [49] . Ten independent runs of K between 1 and 8 were run with 400 000 “burn in” and 4 million MCMC replicates. The maximum log-likelihood values from all runs corresponding to each given K were checked for consistency and averaged. The K with the highest averaged maximum log-likelihood was considered the most likely number of clusters that better explains our dataset. CLUMMP v.1.1.2 [50] was used to summarize parameters across 10 runs and distruct v.1.1 [51] was used to produce the corresponding graphical output.
Isolation by distance
Isolation by distance (IBD) was evaluated using a Mantel test implemented in the program IBDWS v.3.16 [52] . Genetic distance matrices given by F ST /(1− F ST ) were regressed against the logarithm of geographical distances following a two-dimensional model [53] . R ST and Jost's D values were also used. Geographic distances were measured in Google Earth by using set points and measuring either straight-line distance across oceans, or the shortest geographical distance along continental margins. The set points were chosen so as to represent the middle point of the area of distribution where the samples were collected.
Environmental predictors of genetic structure
Three different oceanographic variables were used as predictors of the observed genetic differences between short-beaked common dolphin populations. These were night-time sea surface temperature (SST, °C), chlorophyll concentration (CHL, mg/m 3 ) and water turbidity measured as diffuse attenuation coefficient at 490 nm (KD490, m −1 ). These variables, here obtained from remote sensing data, have been previously related to habitat heterogeneity [54] and associated with genetic differences in other dolphin species [17] . Furthermore, the oceanographic variables chosen have a wide geographic coverage through remote sensing, making them ideal for a global approach. Seven oceanic regions, corresponding to the sampling areas for short-beaked common dolphins, were used for the extraction of these oceanographic variables to assess association with patterns of genetic differentiation. Polygons were defined considering the possible range of common dolphins within that oceanic region, with the last side being the coastline. For NWATL the area was defined between 46°N, 38°N and 57°W; for CEATL between 34°N, 32°N and 16°W; for NEATL between 60°N, 35°N and 0°; for NEPAC between 45°N, 25°N and 108°W; for SWPAC_NZ between 32°S, 44°S and 180°W; for SWPAC_AUS between 26°S, 44°S and 156°E; and for SEIND between 31°S, 37°S and 140°E. In order to account for possible influence of area choice in the final results, areas restricted to where samples from free-ranging animals originally came from or from published distributional data were considered and re-analysed. Since no differences were found in the final results, only analyses including the areas defined above are presented, which account for a possible wider ranging distribution of common dolphins. Monthly averaged data of the three variables, with a 4 km spatial resolution was obtained from Ocean Color Web ( http://oceancolor.gsfc.nasa.gov/ ) for the period from July 2002 to October 2010 and processed using MATLAB software ( www.mathworks.com ). Data collected during this time period provide a characterization of the oceanographic features for each region and are robust to inter-annual oscillations (Supplementary Material, Figure S1 ). Data analysis included the construction of temperature, chlorophyll and turbidity maps for each region, where each pixel of the map corresponds to the eight-year average value for a 4 km grid. These maps were visually inspected to detect geographical areas of environmental heterogeneity. Monthly averages for each oceanic region were then statistically analysed using a paired t-test to detect differences among those regions. Total averages for the 8 year-period for each factor and each sampled region were subsequently used to examine environmental and genetic associations (details below). Environmental distances were calculated as pairwise differences in mean temperature, chlorophyll and turbidity between regions. Pairwise F ST , R ST and Jost's D were used as genetic distances.
All analyses were carried out at different spatial scales: at a large scale, all oceans included; each ocean considered in separate, i.e. all populations within the Atlantic and all populations within the Pacific Ocean and the population in the Southeast Indian Ocean; and at a medium scale, the North and Central Atlantic populations (hereinafter referred to as North Atlantic) and the South Pacific and Southeast Indian Ocean populations (hereinafter referred to as South Indo-Pacific).
Seascape genetics
Associations between genetic and environmental factors were examined using a hierarchical Bayesian method implemented in GESTE [55] , which estimates individual F ST values for each local population and then relates them to environmental factors via a generalized linear model. Here we used 10 pilot runs of 1,000 iterations to obtain the parameters of the proposal distribution used by the MCMC, and an additional burn-in of 5×10 6 iterations with a thinning interval of 20. The model with the highest posterior probability is the one that best explains the data [55] .
Additionally, we used the BIOENV procedure of [56] as implemented in primer v.5 [45] and as described in [57] to examine which predictor variable would provide the best model to explain the population genetic structure observed in the data. This procedure calculates the value of Spearman's rank correlation coefficient (ρ) between a genetic distance matrix (response matrix) with a distance matrix calculated as the Euclidean distance among one or more predictor variables. It then calculates the value of ρ using every possible combination of predictor variables until it finds the “best fit”, corresponding to the combination of predictor variables whose Euclidean distance matrix yields the highest value of ρ [56] . We used three different response matrices corresponding to F ST , R ST and Jost's D distance matrices to identify the best one, two or three-variable fits.
Mantel tests [58] were also used to test for correlations between the pairwise genetic and environmental distances. Partial Mantel tests were used to control the effect of geographical distances in these potential correlations. These tests were performed using the package vegan in R.
Results
Genetic Diversity
In total 281 short-beaked common dolphin samples were genotyped at 14 microsatellite loci ( Table 1 ). Results from Micro-Checker and the Fisher exact test suggested deviations from Hardy-Weinberg equilibrium (HWE) in 4 loci. Two of these (Tur91 and Tur80) showed deviations in only one population each and were therefore included in subsequent analyses, whereas the other two (Tur141 and Dde66) showed deviations in 4 and 2 populations, respectively. These deviations are due to a deficit of heterozygotes (significant F IS values, Table 1 ). To test whether results would be affected by the inclusion of these two loci, estimates of genetic variability and differentiation were carried out with and without them. Since no major differences in results were observed (data not shown), all 14 loci were used in subsequent analyses. These deviations are likely not related with the fact that some samples originated from strandings and others from biopsies. In fact, it has been recently shown that no apparent differences occur when testing population structure in common dolphins using samples originated from carcasses or from free-ranging dolphins [59] .
Levels of genetic diversity, given by mean number of alleles, allelic richness and expected and observed heterozygosities were high for most populations ( Table 1 ). Significant F IS values were obtained for populations from NE Pacific and SW Pacific Australia and New Zealand, which can be due to the presence of population sub-structure (i.e. Wahlund effect). In fact, this is known to be the case for common dolphins inhabiting those regions ( [15] , [27] ; Stockin et al. unpublished).
Genetic differentiation
Pairwise F ST and R ST comparisons showed significant levels of differentiation among all putative populations ( Table 2 ), although the extent of that differentiation differed for each index. Jost's D values tended to be higher than F ST and R ST values. R ST also tended to be higher than F ST . Since R ST is based on allele size, the differences observed indicate that mutation, in addition to drift or gene flow may be affecting the differentiation between these populations. This result was confirmed using spagedi . The overall R ST value was significantly higher than the overall F ST value ( P = 0.042).
10.1371/journal.pone.0031482.t002 Table 2
Pairwise fixation index values obtained between short-beaked common dolphins populations for 14 microsatellite loci.
a) F ST
NEATL
CEATL
NWATL
NEPAC
SWPACAUS
SWPACNZ
NEATL
CEATL
0.0150*
NWATL
0.0051*
0.0151*
NEPAC
0.0313*
0.0439*
0.0284*
SWPACAUS
0.0267*
0.0464*
0.0228*
0.0117*
SWPACNZ
0.0268*
0.0471*
0.0239*
0.0211*
0.0137*
SEIND
0.0680*
0.0896*
0.0716*
0.0663*
0.0473*
0.0386*
b) R ST
NEATL
CEATL
NWATL
NEPAC
SWPACAUS
SWPACNZ
NEATL
CEATL
0.0099*
NWATL
−0.0026
0.0069*
NEPAC
0.0341*
0.0434*
0.0335*
SWPACAUS
0.0122*
0.0280*
0.0059*
0.0114*
SWPACNZ
0.0373*
0.0671*
0.0336*
0.0720*
0.0668*
SEIND
0.0430*
0.0656*
0.0419*
0.0976*
0.0497*
0.0923*
c) Jost's D
NEATL
CEATL
NWATL
NEPAC
SWPACAUS
SWPACNZ
NEATL
CEATL
0.0082
NWATL
0.0119
0.0103
NEPAC
0.1136
0.1422
0.1090
SWPACAUS
0.0687
0.1142
0.0673
0.0293
SWPACNZ
0.0921
0.1398
0.0814
0.0234
0.0135
SEIND
0.1479
0.1795
0.1670
0.1542
0.0835
0.0736
a) F ST ; b) R ST and c) Jost's D .
Taken as a whole, the fixation indices showed high levels of differentiation between short-beaked populations inhabiting different ocean basins. The SEIND and NEPAC populations showed the highest levels of differentiation when compared with all other short-beaked populations. Contrasting to the inter-ocean basin differentiation, lower levels of differentiation were observed between short-beaked populations inhabiting the same ocean basins.
The first two principal components of the PCA analysis explained 84.35% of the variance in allele frequencies among putative populations ( Figure 2 ). The first principal component shows a clear separation between populations inhabiting the Indo-Pacific and the Atlantic Oceans. The second principal component further shows some structure within the Indo-Pacific region, with the SEIND and NEPAC populations appearing separated from the SWPAC_AUS and SWPAC_NZ populations.
10.1371/journal.pone.0031482.g002
Figure 2
Principal component analysis.
Principal component analysis (PCA) performed on a table of standardised allele frequencies based on 14 microsatellite loci of the short-beaked populations analysed in this study.
Non metric MDS analyses using the three different genetic indices also show a clear separation from populations inhabiting the Atlantic, the Pacific and Indian oceans, with the exception of the analysis using R ST , which grouped the NEPAC population with Atlantic ones ( Figure 3 ). The analyses using F ST and Jost's D show a closer proximity among the short-beaked populations inhabiting the North Atlantic, and also of the populations inhabiting the Pacific Ocean.
10.1371/journal.pone.0031482.g003
Figure 3
Non-metric MDS.
Non-metric MDS plots of short-beaked common dolphin populations on the basis of genetic distances using a) F ST , b) R ST or c) Jost's D . Stress values are indicated.
Results obtained in STRUCTURE using the correlated allele frequency model resulted in a peak of maximum ln P ( K ) at K = 3 ( Figure 4 , Supplementary Table S2 ). These clusters correspond to populations inhabiting the three ocean basins: the Atlantic (including the NEATL, NWATL and CEATL populations), the Pacific (including the NEPAC, SWPAC_AUS and SWPAC_NZ populations) and the Indian Ocean including the SEIND population ( Figure 4 ).
10.1371/journal.pone.0031482.g004
Figure 4
Number of clusters found for short-beaked common dolphin populations.
Results from the program STRUCTURE showing individual assignment values for K = 3. Each colour depicts the relative contribution of each of the three clusters to the genetic constitution of each individual.
The AMOVA analysis showed that the highest levels of differentiation were obtained when populations were divided by eastern versus western regions within ocean basins ( F CT = 0.03425, P <0.0001) ( Table 3 ).
10.1371/journal.pone.0031482.t003 Table 3
Analysis of hierarchical variance (AMOVA) results obtained for the short-beaked common dolphin populations.
Source of variation
%variation
F -statistics
P
Among ocean basins
2.71
F CT = 0.02710
0.0000
Among groups within populations
1.35
F SC = 0.01386
0.0000
Within populations
95.94
F ST = 0.04058
0.0000
Among regions
1.92
F CT = 0.03425
0.0001
Among groups within populations
1.5
F SC = 0.01532
0.0000
Within populations
96.58
F ST = 0.03425
0.0000
Isolation by distance
The relationship between geographic and genetic distance was only observed when populations inhabiting all oceans were considered in the analysis and when F ST and Jost's D values were used ( Table 4 ). This relationship was not detected when R ST values were used, nor when finer spatial scales were considered.
10.1371/journal.pone.0031482.t004 Table 4
Summary results for Isolation by Distance tests conducted for all short-beaked common dolphin populations in all oceans, for North Atlantic populations only, for Pacific populations only, and for South Indo-Pacific populations only.
P
r (slope)
R
2
All oceans
Fst
0.0196
0.0502
0.1560
Rst
0.9072
−0.0657
0.0416
Jost's D
0.0091
0.1240
0.4660
North Atlantic
Fst
0.4995
−0.0211
0.2010
Rst
0.8351
−0.0239
0.4210
Jost's D
0.3316
0.0068
0.7740
Pacific
Fst
0.3364
0.0573
0.0483
Rst
0.6241
−0.0840
0.0024
Jost's D
0.3328
0.1410
0.1150
South Indo-Pacific
Fst
0.3310
0.0984
0.7860
Rst
0.4980
0.1209
0.1130
Jost's D
0.3321
0.2137
0.8760
Values in bold were statistically significant ( P <0.05).
Oceanographic predictors
Data on sea surface temperature (SST), chlorophyll concentration (CHL) and water turbidity (KD490) was gathered for the seven oceanic regions where short-beaked common dolphins were sampled: NEATL, CEATL, NWATL, NEPAC, SWPAC_AUS, SWPAC_NZ and SEIND ( Figure 5 ). Paired t-tests showed significant differences in the 8 year average values of SST between most regions with exception of the comparison between NEATL and NWATL, between NEPAC and SWPAC (both AUS and NZ), and between NEPAC and SEIND, where differences were not statistically significant ( P <0.01, see Supplementary Material, Table S1 ). In the SST maps, all regions are heterogeneous, having regions of colder and warmer waters ( Figure 5 ). Nevertheless, NEATL and NWATL regions are dominated by colder waters when compared with other regions, which are dominated by warmer waters, such as SWPAC_AUS and SWPAC_NZ. Significant differences were not detected in mean CHL values between NEPAC and SWPAC (both AUS and NZ) and between NEPAC and SEIND, as well as among SEIND, SWPAC_AUS and SWPAC_NZ. All other comparisons were significant. Despite this, in the CHL maps, clear differences can be seen among the regions located in the Pacific Ocean. Chlorophyll concentrations are higher in the NEPAC region closer to the coast when compared to the SWPAC_AUS and SWPAC_NZ regions. Regarding turbidity mean values, these were only not significant in the comparisons among SWPAC_AUS, SWPAC_NZ and SEIND ( Table S1 ). Patterns seen in the maps are similar to the ones obtained for the CHL maps ( Figure 5 ).
10.1371/journal.pone.0031482.g005
Figure 5
Oceanographic predictors for each oceanic region.
Regional maps showing 8-year average values for sea surface temperature (SST), chlorophyll concentration (CHL) and water turbidity (KD490) on the left and standard deviation values on the right for the oceanic regions where the short-beaked common dolphin populations analysed in this study were sampled: a) Northwest Atlantic; b) Central eastern Atlantic; c) Northeast Atlantic; d) Northeast Pacific; e) Southwest Pacific New Zealand; f) Southwest Pacific Australia; g) Southeast Indian.
Seascape genetics
Hierarchical Bayesian analyses implemented in GESTE identified the model including the constant as the best one in all spatial scales considered ( Table 5 ). The second best model for all analyses was the one including KD490, though the third and fourth models (including CHL and SST) all had very similar posterior probability values. Higher posterior probabilities were obtained when medium spatial scales were analysed. Positive signals of the regression coefficients were obtained for the association between CHL and genetic differentiation in the Pacific Ocean and South Indo-Pacific Ocean populations, and for the association between KD490 and genetic differentiation in the Pacific Ocean populations ( Table 5 ). Regarding SST, positive signals of the regression coefficients were obtained for all populations across all oceans, for the North Atlantic populations, and for the South Indo-Pacific populations ( Table 5 ). Therefore, genetic isolation of populations within the Pacific Ocean increases with differences in CHL and KD490 among regions, whereas genetic isolation of populations within the Atlantic Ocean increases with differences in SST among regions. In the South Indo-Pacific region, both CHL and SST increase genetic isolation among populations. The percentage of variation that remained to be explained (indicated by sigma values) was however moderate ( Table 5 ).
10.1371/journal.pone.0031482.t005 Table 5
Posterior probabilities of the four most probable models for the GESTE analysis of environmental associations with genetic structure (population specific F ST ) of short-beaked common dolphins.
Model
Factors included
P
Coefficient
Mean
Mode
95% HPDI
All Oceans
1
Constant
0.702
α 0
−3.02
−3.01
−3.60; −2.43
σ
0.591
0.378
0.125; 1.319
2
Constant, SST
0.067
α 0
−3.01
−2.99
−3.61; −2.33
α 1
0.13
0.12
−0.52; 0.73
σ
0.708
0.422
0.125; 1.70
3
Constant, CHL
0.0649
α 0
−3
−3
−3.66; −2.36
α 2
−0.13
−0.11
−0.69; 0.56
σ
0.679
0.367
0.123; 1.501
5
Constant, KD490
0.0707
α 0
−3.03
−3.05
−3.60; −2.32
a3
−0.1
−0.1
−0.80; 0.53
σ
0.694
0.4
0.113; 1.726
Pacific
1
Constant
0.628
α 0
−3.08
−3.12
−4.02; −1.97
σ
1.094
0.701
0.173; 2.88
2
Constant, SST
0.092
α 0
−3.1
−3.16
−4.30; 2.02
α 1
−0.04
−0.12
−1.26; −1.10
σ
1.42
0.695
0.198; 4.102
3
Constant, CHL
0.0991
α 0
−3.04
−3.1
−4.16; −1.61
α 2
0.13
0.06
−1.07; 1.25
σ
1.63
0.713
0.140; 4.47
5
Constant, KD490
0.104
α 0
−3.04
−3.17
−4.16; −1.85
α 3
0.14
0.16
−1.10; 1.23
σ
1.534
0.68
0.199; 4.601
North Atlantic
1
Constant
0.496
α 0
−3.25
−3.33
−4.52; −2.05
σ
1.14
0.677
0.097; 3.27
2
Constant, SST
0.101
α 0
−3.22
−3.28
−4.59; −1.61
α 1
0.29
0.31
−0.97; 1.9
σ
1.557
0.774
0.114; 4.876
3
Constant, CHL
0.1
α 0
−3.22
−3.3
−4.46; 1.63
α 2
−0.25
−0.25
−1.55; −1.08
σ
1.547
0.783
0.135; 5.112
5
Constant, KD490
0.103
α 0
−3.19
−3.32
−4.45; −1.65
α 3
−0.27
−0.29
−1.85; −1.11
σ
1.694
0.86
0.134; 5.4
South Indo-Pacific
1
Constant
0.501
α 0
−2.95
−3
−4.26; −1.63
σ
1.481
0.825
0.146; 4.305
2
Constant, SST
0.0946
α 0
−2.87
−3.1
−4.25; 0.95
α 1
0.14
0.19
−1.52; 1.64
σ
2.246
1.195
0.163; 7-064
3
Constant, CHL
0.0969
α 0
−2.93
−2.99
−4.43; −1.06
α 2
0.08
0.13
−1.70; 1.65
σ
2.331
0.933
0.169; 7.64
5
Constant, KD490
0.171
α 0
−2.96
−3.07
−4.27; −1.61
α 3
−0.54
−0.59
−1.84; 0.91
σ
1.678
0.765
0.124; 5.344
SST – sea surface temperature; CHL – chlorophyll concentration; KD490 – sea water turbidity measured as diffuse attenuation coefficient at 490 nm; α – regression coefficient; σ – estimate of the variation that remains unexplained by the regression model; HPDI – highest probability density interval.
The BIOENV procedure found strong positive correlations between oceanographic predictors and genetic differentiation for the analyses conducted at medium spatial scales ( Table 6 ). For the populations within the Atlantic Ocean and within the South Indo-Pacific, CHL and KD490 showed stronger correlation with genetic distance. For the larger spatial scales considered (across all oceans and within the Pacific Ocean), a strong negative correlation between CHL and KD490 with rank genetic distance was found ( Table 6 ).
10.1371/journal.pone.0031482.t006 Table 6
Results of the BIOENV procedure, showing the best fit obtained, for all short-beaked common dolphin populations, North Atlantic populations only, Pacific populations only, and South Indo-Pacific populations only, in the case of one, two and three predictor variables for each genetic distance matrix.
Number
Spearman's
Variables
Number
Spearman's
Variables
variables
rho
chosen
variables
rho
chosen
All Oceans
North Atlantic
Fst
Fst
1
−0.341
CHL
1
1
KD490
2
−0.356
CHL, KD490
2
1
CHL, KD490
3
−0.227
SST, CHL, KD490
3
0.5
SST, CHL, KD490
Jost's D
Jost's D
1
−0.366
CHL
1
−0.5
KD490
2
−0.374
CHL, KD490
2
−0.5
CHL, KD490
3
−0.31
SST, CHL, KD490
3
−1
SST, CHL, KD490
Rst
Rst
1
−0.713
CHL
1
1
SST
2
−0.703
CHL, KD490
2
1
SST, CHL
3
−0.573
SST, CHL, KD490
3
1
SST, CHL, KD490
Pacific
South Indo-Pacific
Fst
Fst
1
−0.314
CHL
1
1
KD490
2
−0.371
CHL, KD490
2
−0.5
CHL, KD490
3
−0.029
SST, CHL, KD490
3
−0.5
SST, CHL, KD490
Jost's D
Jost's D
1
−0.314
CHL
1
1
KD490
2
−0.714
CHL, KD490
2
0.5
CHL, KD490
3
−0.714
SST, CHL, KD490
3
−1
SST, CHL, KD490
Rst
Rst
1
0.029
CHL
1
0.5
KD490
2
0.086
CHL, KD490
2
0.5
SST, KD490
3
−0.2
SST, CHL, KD490
3
0.5
SST, CHL, KD490
SST – sea surface temperature; CHL – chlorophyll concentration; KD490 – sea water turbidity measured as diffuse attenuation coefficient at 490 nm.
Mantel tests and Partial Mantel tests between genetic and environmental distances were not statistically significant for any comparison, even considering different spatial scales (results not shown). Failures of these tests to detect relationships between genetic and environmental data have been previously described [60] , [61] and could explain the unsuccessful use with our datasets.
Discussion
We used a seascape approach to investigate the interaction between a set of oceanographic variables and population structure in a highly mobile, widely distributed top marine predator, the short-beaked common dolphin. We show that sea surface temperature, chlorophyll concentration and water turbidity seem to be important factors in explaining the observed patterns of genetic structure in these dolphins, more than geographical distance alone, particularly when medium spatial scales were considered.
Genetic structure
The overall global pattern of genetic structure obtained here supports previous studies [19] : higher levels of differentiation were obtained across large geographical scales, between different ocean basins, and lower levels were obtained when medium geographical scales were considered, within the same ocean basin. While results from STRUCTURE showed a clear differentiation between ocean basins, the AMOVA analysis resulted in higher F CT estimates for partitioning of short-beaked populations among regions within each ocean basin. The low levels of divergence found between populations inhabiting the same ocean basin may have affected the power of the program STRUCTURE to detect such differentiation, even using recently developed algorithms that account for weak differentiation [49] . Nonetheless, the PCA and the NMDS plots also indicate some level of differentiation within ocean basins, which seems to be stronger among the Pacific Ocean populations. Multivariate analysis does not require strong assumptions about the underlying genetic model, such as Hardy-Weinberg equilibrium or the absence of linkage disequilibrium [43] . The high levels of differentiation found for the SEIND population (southern Australia) were surprising given the comparatively shorter distance separating this population from the Southwest Pacific populations (off New South Wales, southeastern Australia), even considering that the region where the SEIND population was sampled (off South Australia) falls into a different biogeographic region (see [62] to the one of the SWPAC_AUS population. Such high differentiation was also reported by [27] when comparing individuals from this region to individuals from southeastern Tasmania (Southwest Pacific) – in that case oceanographic features affecting the distribution of target prey were suggested to be the likely explanation for the genetic differentiation found. Our study corroborates this previous finding (see below).
Isolation by distance
A pattern of isolation by distance was only observed when large spatial scales were considered, indicating that the stronger genetic differentiation observed in short-beaked common dolphins from different oceans may be an effect of geographic distance. Isolation by distance has been reported for other cetacean species, such as in the harbour porpoise [63] and in bottlenose dolphins [64] . Conversely, when medium geographic scales were considered (i.e. within each ocean basin), no isolation by distance effect was detected, and genetic differentiation could be explained by oceanographic variables. This pattern has also been described for common dolphins at small geographical scales, along the eastern Australian coast [15] , for bottlenose dolphins in South Australia where a temperature and salinity front coincides with the boundary between two distinct genetic populations [13] , and for pilot whales, where ecological factors, such as SST, were more important in explaining genetic structure than geographic separation [14] . In franciscana and humpback dolphins, environmental factors were also more important in explaining genetic structure than distance at small geographical scales [12] , [17] .
Oceanographic predictors
All oceanographic variables tested, CHL, KD490 and SST, showed an association with population genetic structure in short-beaked common dolphins. These associations were strongest at the medium spatial scales considered. In the Pacific Ocean, CHL and KD490 were the environmental predictors that were most strongly associated with increased genetic isolation in short-beaked common dolphins. Conversely, in the Atlantic Ocean, SST was the strongest predictor associated with population divergence. Although no significant statistical differences in the 8-year average values of CHL and KD490 were detected among regions in the Pacific Ocean, a visual inspection of the regional maps shows heterogeneity in these variables among regions ( Figure 5 ). Heterogeneity in SST, CHL and KD490 is also seen among Atlantic Ocean regions, although our results suggest that only SST seems to explain genetic differentiation of short-beaked common dolphins in this area. Marine productivity and SST are important variables for habitat occupancy and dispersal in cetaceans [65] , [66] and have been shown to influence population structure in Franciscana [12] and in humpback dolphins [17] . Here, we suggest that they are also important drivers of population structure in common dolphins. A direct causality is however difficult to establish. For example, it has been suggested that ecological factors such as prey behaviour rather than inherent sensitivity to environmental factors, could account for the relationship between SST and population structure in pilot whales [14] , [66] , [67] . Similarly, differences in prey distribution and abundance between regions rather than SST differences themselves are suggested to account for genetic differentiation of bottlenose dolphins in South Australia [13] and short-beaked common dolphins in southern [27] and southeastern Australia [15] . We suggest that a similar process may account for the patterns obtained in this study. Since dolphins feed high in the food chain, a statistical association with oceanographic variables that do not directly affect the individuals, but rather affect their prey, is expected to be weak [23] . This could also explain the fact that analyses performed in GESTE did not result in a single best-chosen model and that the percentage of variability that remained to be explained in the data was moderate.
Chlorophyll concentration, water turbidity and SST are routinely used to map ocean primary productivity (e.g. [68] ). Due to the bottom-up processes that control marine ecosystems [69] , these variables have been related to prey distribution and abundance, and to the occurrence of top marine predators (e.g. [70] , [71] ). Distribution and abundance of prey has been suggested as the main factor dictating seasonal migrations in several species of delphinids, including short-beaked common dolphin (e.g. [21] ). Moreover, short-beaked common dolphins feed primarily on small mesopelagic schooling fish such as sardines and anchovies [21] , [22] . These fishes are filter feeders and occur in association with nutrient rich waters (e.g. [72] ), and could explain the dolphins' preference for certain oceanographic conditions.
We further suggest that a behavioural mechanism such as specialization for local resources could also explain the patterns observed. Resource specialization is a common mechanism driving population structure in delphinds [73] . Moreover, dietary segregation is known to occur in short-beaked common dolphins. In the Bay of Biscay, Northeast Atlantic Ocean, common dolphins inhabiting neritic and oceanic waters feed on different prey species [74] . Feeding specialization leading to local adaptation has also been suggested as driving speciation of the short and long-beak forms [19] and as important triggers for the process of population divergence and speciation in the genera Tursiops and Stenella [75] , [76] . Perhaps the best studied example within delphinids are killer whales ( Orcinus orca ), where resource partitioning and foraging specializations of sympatric populations occurring in the North Pacific have lead to the evolution of distinct lineages [77] . Short-beaked common dolphins could therefore be locally adapted to the existent prey species and only move within certain regions following prey migration. Seasonal migrations are known to occur in the Northeast Pacific [78] and Southwest Indian Ocean [79] . Further investigation is however required to support this hypothesis.
There are also other factors that may account for population divergence in common dolphins that were not assessed in this study. Fine-scale oceanic processes, for example, have recently been suggested to affect connectivity in common dolphins [15] . A proper assessment of its direct relationship with genetic structure requires knowledge on hydrodynamic modelling and will certainly be the aim of forthcoming studies. Demographic and historical processes can also contribute to population structure and should also be integrated in future analyses.
Implications for conservation and management
The results presented here are of particular importance for marine conservation management and design of marine protected areas (MPA). MPAs are usually designed to protect coastal regions that are either important habitats, as part of the marine ecosystem, or biodiversity hotspots [80] . Marine predators are often used as indicators for MPA design, because their protection aids in protecting the more complex environments they use [81] , [82] , [83] . Although several studies have described the distribution and occurrence of cetacean species in relation to different habitat variables (e.g. [84] , [85] , [86] ), only a few have found a direct correlation between oceanographic variables and population structure [12] , [17] . In this study, by showing how marine productivity correlate with population structure in short-beaked common dolphins, we highlight the importance of using seascape genetic studies to inform MPA design in relation to distribution and genetic connectivity of charismatic and ecologically important megafauna. Furthermore, we highlight how such an approach can track the biological effects of ongoing climate-change and prevent the loss of top marine predators [87] .
Supporting Information
Figure S1
Annual fluctuation of oceanographic predictor values. Annual average values for (a) sea surface temperature, (b) chlorophyll concentration and (c) water turbidity for the different oceanographic regions.
(PDF)
Table S1
Mean pairwise difference between average values of a) sea surface temperature (SST), b) chlorophyll concentration (CHL) and c) water turbidity (KD490) obtained for each oceanographic region where short-beaked common dolphins were sampled for this study, with significant values of paired t-tests indicated in bold.
(XLS)
Table S2
Individual runs for the Bayesian analysis implemented in the program STRUCTURE with a burn-in phase of 4×10 5 and 4×10 6 MCMC replicates. The log-likelihood of the data (LnP(D)) for each run and an average across 10 runs for each K are shown. The K with the highest averaged maximum log-likelihood was considered the most likely number of clusters that better explains our dataset (in bold).
(XLS)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Pneumococcal disease is a leading cause of childhood morbidity and mortality worldwide [ 1 ]. Pneumococcal disease is preceded by pneumococcal nasopharyngeal carriage, and the severity of pneumococcal pneumonia is associated with bacterial load (density) of pneumococci in the nasopharynx [ 2 , 3 ]. Public health interventions to prevent pneumococcal disease can be improved by identifying the factors associated with pneumococcal carriage. Determining factors associated with higher pneumococcal carriage density could aid estimation of pneumococcal pneumonia prevalence in childhood pneumonia studies [ 2 ].
In low- and middle-income countries, risk factors for pneumococcal nasopharyngeal carriage vary [ 4 – 8 ] and few studies have investigated factors associated with the density of pneumococcal carriage in healthy populations [ 2 , 9 – 11 ]. Common factors positively associated with pneumococcal carriage in low- and middle-income countries include indigenous ethnicity, passive smoking; co-colonisation with Haemophilus influenzae , childcare attendance, poverty, acute respiratory infection, living with young children, and being under five years old [ 4 , 8 , 12 ]. In studies from low- and middle-income countries, higher pneumococcal density has been positively associated with the symptoms of upper respiratory tract infection, presence of a febrile acute respiratory infection in children [ 13 ], rainy season, severe pneumonia, viral co-infection, radiologically confirmed pneumococcal pneumonia, and encapsulated serotypes (compared with non-encapsulated serotypes) [ 2 , 9 – 11 , 13 – 16 ].
Pneumococcal conjugate vaccines (PCV) reduce vaccine-type carriage and disease [ 17 , 18 ]. However, serotype replacement with non-vaccine-type carriage occurs following the introduction of PCV which can result in non-vaccine-type disease [ 19 , 20 ]. In the post-PCV era, it is largely unknown what the impact of PCV is on the risk factors for pneumococcal carriage and density in low- and middle-income countries in the Asia-Pacific region.
In 2012, Fiji introduced the ten-valent PCV (PCV10). Six years before PCV10 was introduced, factors associated with pneumococcal carriage in healthy 3–13 month old Fijians included indigenous iTaukei (iTaukei) ethnicity and having symptoms of an upper respiratory tract infection (URTI)[ 4 ]. iTaukei ethnicity was also associated with higher median pneumococcal density among 17 month old Fijians [ 21 ]. Fiji provides an opportunity to investigate factors associated with pneumococcal carriage and density in the post-PCV10, in a tropical, upper middle-income setting. As part of a PCV10 impact evaluation on pneumococcal carriage, we previously reported that the prevalence of vaccine-type carriage declined in both iTaukei and Fijians of Indian Descent (FID) three years after the introduction of PCV10, but carriage of non-vaccine-type carriage increased in iTaukei infants and toddlers [ 22 ]. The aim of this study is to determine the factors associated with carriage and density of pneumococci (overall, vaccine-type, and non-vaccine-type) in Fiji up to three years following PCV10 introduction.
Materials and methods
Setting
The majority of the Fijian population (81%) lives on Viti Levu, the bigger of Fiji’s two main islands. Greater than half (56.8%) of the population are iTaukei and 37.8% are FID. This study was conducted in Suva, and the surrounding areas, where over one-third of the population lives [ 23 ]. PCV10 was introduced nationally in October 2012 to be administered at 6, 10, and 14 weeks of age, with no catch-up campaign. The national coverage of the third dose of PCV10 one, two, and three years post-introduction was 84.9%, 84.9%, and 89.0%, respectively [ 24 – 26 ].
Cross-sectional carriage studies
The design and methods for these cross-sectional studies have been described elsewhere [ 22 ]. Briefly, four annual cross-sectional carriage surveys were undertaken: one pre-PCV10 (2012), and then annually thereafter (2013–2015). Purposive quota sampling achieved a sample proportionate to the national iTaukei: FID (3:2) and rural: urban (1:1) ratios [ 4 , 23 , 27 ]. Each year, approximately 500 participants were recruited into each of the following four groups: infants 5–8 weeks (infants), toddlers 12–23 months (toddlers), children 2–6 years (children), and their parents/guardians (caregivers). Age groups were selected in order to best answer the primary research question regarding impact of PCV10 on pneumococcal carriage prevalence in Fiji, and as described previously, were based on those most likely to benefit from direct and indirect effects of PCV10; those likely to have the highest pneumococcal carriage prevalence; those age-eligible for PCV10 vaccination; and those most likely to transmit or be in contact with transmitters of pneumococci [ 22 ]. This analysis used the same study population.
Participants were recruited from two health centres in the Greater Suva area, and from surrounding communities. Eligibility criteria included age or being a caregiver, and for non-infant participants, that they had lived in the area for at least three months. Those with an axillary temperature ≥37.0°C were excluded. For the pre-PCV10 survey, any receipt of PCV10 was an exclusion criterion for all participant groups. For subsequent surveys, only infants who had ever received PCV10 were excluded.
Study staff interviewed caregivers and recorded individual level participant characteristics on data collection forms. Variables collected included: self-reported ethnicity, sex, residential location, antibiotic use in the fortnight preceding the survey, exposure to household cigarette smoke, coryza, symptoms of allergic rhinitis, cough, ear discharge, number of children less than five years in the household, and weekly family income. Caregivers reported their own, and their participating child’s or children’s ethnicity, according to options defined by the investigator (iTaukei, FID, or other) and recognized by Fijian population[ 23 ]. PCV10 vaccination status for infants and children was obtained from written records. As PCV10 was unavailable privately, caregivers were assumed to be PCV10 unvaccinated. A binary variable for symptoms of URTI was derived from the presence of one or more of the following: coryza, allergic rhinitis, cough, or ear discharge. A binary variable for low family income was derived, defined as family income on / above or below the basic needs poverty line (<FJ$175 per week) [ 28 ].
Trained study nurses collected nasopharyngeal samples using flocked nylon swabs (COPAN FLOQSwabs TM ), which were transported and stored according to standard methods, as described previously [ 22 , 29 ]. Microbiological analyses were undertaken at the Murdoch Children’s Research Institute in Melbourne, Australia as described previously [ 22 ]. In brief, pneumococci were detected using real-time quantitative-polymerase chain reaction targeting the lytA gene, with molecular serotyping by microarray [ 29 – 31 ]. Laboratory staff were blinded to participant characteristic data. Detection of any pneumococci in swab samples, including non-encapsulated lineages, was defined as overall pneumococcal carriage. Detection of serotypes included in PCV10 (serotypes 1, 4, 5, 6B, 7F, 9V, 14, 18C, 19F, and 23F) was defined as PCV10 carriage, and detection of serotypes not included in PCV10, including non-encapsulated lineages, was defined as non-PCV10 carriage. Detection by microarray of a PCV10 serotype and a non-PCV10 serotype from the same swab sample was recorded as positive for both PCV10 and non-PCV10 serotype carriage [ 22 ]. Any detection of a serotype by microarray was considered positive, regardless of relative abundance [ 22 ]. Non-encapsulated lineages were classified based upon previously described genetic variants [ 32 ]. We determined pneumococcal density only for pneumococcal positive samples, and reported it in genome equivalents per ml (GE / ml). Participant characteristic data were double entered, and validated, in an EpiData 3.1 database [ 33 ]. Microbiological outcome data were entered into Microsoft Excel (Excel 2013) and merged with characteristic data in Stata 15.1 [ 34 ].
Statistical analyses
Participant characteristics were summarised using counts and percentages. We built logistic and quantile regression models to investigate the factors associated with pneumococcal carriage and density, respectively. Empirical univariable results (p<0.2) and a priori selection informed variable choice for multivariable models. Factors assessed empirically included residential location, participant sex, two or more children under five years living in the household, low family income, exposure to household cigarette smoke, and recent antibiotic use. Variables selected a priori included PCV10 vaccination, survey year, ethnicity, participant group, and URTI symptoms [ 4 , 14 ]. Interaction terms for ethnicity with PCV10 vaccination status, and with survey year, were assessed to account for the potential differential effect of PCV10 vaccination, or number of years post-PCV10 introduction, on pneumococcal carriage and density by ethnicity. We also assessed potential interaction between ethnicity and other socio-demographic factors in the models. Interaction terms were included as indicated, with significance level set at P <0.05. Estimates of the association of participant characteristics with carriage and density were reported as odds ratios and differences in medians, respectively, with 95% confidence intervals (95% CI) and P -values. Pneumococcal density data were log 10 transformed prior to analyses, and analyses of pneumococcal density were restricted to pneumococcal carriers. Only 14/38 participants who identified as “other” ethnicity had pneumococcal positive samples, so were excluded from inferential analyses. Merged data were cleaned and analysed in Stata 15.1 [ 34 ].
Ethics statement
This study was carried out in accordance with the protocols approved by the Fijian National Health Research and Ethics Review Committee (201228), and The University of Melbourne Health Sciences Human Ethics Sub-Committee (1238212). Study staff discussed the study with caregivers, and written informed consent was completed prior to any study procedures. Participants were not offered any incentive to participate.
Results
Participant characteristics
There were 8,109 participants, with characteristics shown in Table 1 . The overall vaccination rate of 13.6% reflects the pooled participant group, most (85.8%) of which were not age-eligible to receive PCV10. Similar numbers of people participated per survey year and by participant group [ 22 ]. Few had used antibiotics in the preceding two weeks. Forty-eight participant swab samples were excluded from microbiological analysis due to insufficient volume, sample loss, or labelling errors. A further 61 pneumococcal positive samples were excluded from serotyping due to biological reasons or technical issues. Of the 8,061 participants for whom swab sample results were available, 30.5% tested positive for pneumococci. Among the 8,000 serotyped samples, PCV10 carriage was uncommon (8.9%), and 23.9% of participants carried non-PCV10 pneumococci. Carriage of non-encapsulated pneumococci was rare (390 / 8,000, 4.9%). Density data was unavailable for one pneumococcal carrier. Overall carriage median density was 5.0 log 10 GE/ml (4.2–5.7), while those for PCV10, non-PCV10 carriage, and non-encapsulated lineages were 4.9 log 10 GE/ml (4.1–5.6), 4.9 log 10 GE/ml (4.1–5.7), and 4.3 log 10 GE/ml (3.7–4.9), respectively [ 22 ].
10.1371/journal.pone.0231041.t001
Table 1 Characteristics of participants in four annual cross-sectional community nasopharyngeal carriage surveys, 2012–2015, Fiji (n = 8,109 a ).
Characteristics
Summary statistic
PCV10 vaccinated b , n (%)
1105 (13.6)
Survey year, n (%)
Pre-PCV10 (2012)
2025 (25.0)
1 year post-PCV10 (2013)
2042 (25.2)
2 years post-PCV10 (2014)
2022 (24.9)
3 years post PCV10 (2015)
2020 (24.9)
Ethnicity, n (%)
Fijian of Indian Descent
3236 (39.9)
iTaukei
4835 (59.6)
Other
38 (0.5)
Participant group, n (%)
Infants (5–8 weeks)
2006 (24.7)
Toddlers (12–23 months)
2004 (24.7)
Children (2–6 years)
2052 (25.3)
Caregivers
2047 (25.3)
Residential location, n (%)
Rural
3944 (48.6)
Urban
4165 (51.4)
Female sex, n (%)
4683 (57.8)
Two or more children under five years in the household, n (%)
n = 8106
4004 (49.4)
Low family income c , n (%)
n = 7831
4599 (58.7)
Symptoms of URTI, n (%)
2092 (25.8)
Exposure to household cigarette smoke, n (%)
4353 (53.7)
Antibiotic use in past fortnight, n (%)
n = 8105
357 (4.4)
Pneumococcal carriage, n / N (%)
Overall d
n = 8061
2456 (30.5)
PCV10 serotypes e
n = 8000
713 (8.9)
Non-PCV10 serotypes f
n = 8000
1915 (23.9)
Non-encapsulated pneumococci g
n = 8000
390 (4.9)
Pneumococcal density h , n, median log 10 GE/ml (IQR)
Overall
2455, 5.0 (4.2–5.7)
PCV10 serotypes
713, 4.9 (4.1–5.6)
Non-PCV10 serotypes
1915, 4.9 (4.1–5.7)
Non-encapsulated pneumococci
390, 4.3 (3.7–4.9)
Abbreviations: PCV10, ten-valent pneumococcal conjugate vaccine; URTI, upper respiratory tract infection; IQR, interquartile range.
a Unless otherwise specified
b Two doses of PCV10 given before the age of 12 months, or one or more doses of PCV10 given at or after 12 months of age[ 35 ]
c Weekly family income below the basic needs poverty line (<FJ$175 per week)[ 28 ]
d Any pneumococci, including non-encapsulated lineages
e Pneumococcal serotypes included in PCV10 (serotypes 1, 4, 5, 6B, 7F, 9V, 14, 18C, 19F, and 23F)
f Pneumococcal serotypes not included in PCV10, including non-encapsulated lineages
g Includes carriage of any of the following non-encapsulated lineages: NT, NT1, NT2, NT2/NT3b, NT3a, NT3b, NT4a, NT4b
h Only includes participants who were carriers of indicated pneumococcal types
Factors associated with overall carriage
ITaukei ethnicity, young age (infant, toddler, and child participant groups vs caregivers), urban residence, living with two or more children under five years, low family income, and URTI symptoms were positively associated with overall carriage ( Table 2 ). Survey year was negatively associated with overall carriage. There was evidence of protection from PCV10 vaccination against overall carriage, however the confidence interval crossed the null value. There was evidence of an interaction between survey year and ethnicity (global P <0.001), but no evidence of an interaction between PCV10 vaccination status and ethnicity (global P = 0.880).
10.1371/journal.pone.0231041.t002
Table 2 Unadjusted and adjusted odds ratios of overall pneumococcal carriage in association with participant characteristics in four cross-sectional carriage surveys pre-PCV10 (2012) and annually thereafter (2013–2015) in Fiji (n = 8,023).
Exposure
Overall carriage a n/N (%)
Unadjusted odds ratio (95% CI)
P -value
Adjusted odds ratio (95% CI)
P -value
PCV10 vaccination status
<0.001
0.065
Not vaccinated
2027 / 6931 (29.3)
ref
ref
Vaccinated b
415 / 1092 (38.0)
1.48 (1.30–1.69)
0.82 (0.66–1.01)
Survey year
<0.001
<0.001
Pre-PCV10 (2012)
708 / 2001 (35.4)
ref
ref
1 year post-PCV10 (2013)
655 / 2033 (32.2)
0.87 (0.76–0.99)
0.67 (0.51–0.88)
2 years post-PCV10 (2014)
433 / 1997 (21.7)
0.51 (0.44–0.58)
0.49 (0.36–0.66)
3 years post PCV10 (2015)
646 / 1992 (32.4)
0.88 (0.77–1.00)
0.62 (0.46–0.83)
Ethnicity
<0.001
<0.001
Fijian of Indian Descent
496 / 3218 (15.4)
ref
ref
iTaukei
1946 / 4805 (40.5)
3.74 (3.34–4.18)
2.74 (2.17–3.45)
Participant group
<0.001
<0.001
Caregivers
193 / 2035 (9.5)
ref
Ref
Infants (5–8 weeks)
516 / 1974 (26.1)
3.38 (2.82–4.04)
4.15 (3.40–5.06)
Toddlers (12–23 months)
845 / 1986 (42.6)
7.07 (5.95–8.40)
8.88 (7.13–11.07)
Children (2–6 years)
888 / 2028 (43.8)
7.43 (6.26–8.83)
8.48 (6.99–10.29)
Residential location
<0.001
<0.001
Rural
1070 / 3911 (27.4)
ref
ref
Urban
1372 / 4112 (33.4)
1.33 (1.21–1.46)
1.45 (1.30–2.57)
Participant sex
<0.001
0.300
Male
1167 / 3385 (34.5)
ref
ref
Female
1275 / 4638 (27.5)
0.72 (0.65–0.79)
1.06 (0.95–1.19)
Number of children < 5 years living in the household c
<0.001
<0.001
Less than two
955/4067 (23.5)
ref
ref
Two or more
1487 / 3953 (37.6)
1.96 (1.78–2.16)
1.42 (1.27–1.59)
Family income level d
<0.001
<0.001
Not low
801 / 3193 (25.1)
ref
ref
Low
1523 / 4558 (33.4)
1.50 (1.35–1.66)
1.44 (1.28–1.62)
Symptoms of URTI
<0.001
<0.001
Not present
1529 / 5955 (25.7)
ref
ref
Present
913 / 2068 (44.2)
2.29 (2.06–2.54)
1.77 (1.57–2.01)
Household cigarette smoke
0.105
0.555
No exposure
1099 / 3729 (29.5)
ref
ref
Exposure
1343 / 4303 (31.2)
1.08 (0.98–1.19)
0.97 (0.87–1.08)
Antibiotic use in previous fortnight e
0.350
Not used
2325 / 7667 (30.3)
ref
Used
115 / 352 (32.7)
1.11 (0.89–1.40)
Abbreviations: CI, confidence interval; PCV10, ten-valent pneumococcal conjugate vaccine; URTI, upper respiratory tract infection.
a Any pneumococci, including non-encapsulated lineages
b Two doses of PCV10 given before the age of 12 months, or one or more doses of PCV10 given at or after 12 months of age[ 35 ]
c Data on number of children under five years living in the household were missing for three participants, of whom none were pneumococcal carriers
d Weekly family income below the basic needs poverty line (<FJ$175 per week)[ 28 ]; data on family income were missing for 272 participants, of whom 118 were pneumococcal carriers
e Data on antibiotics use were missing for four participants, of whom two were pneumococcal carriers.
Factors associated with PCV10 carriage
iTaukei ethnicity, young age (infant, toddler, and child participant groups), urban residence, living with two or more children under five years, low family income, symptoms of URTI, and exposure to household cigarette smoke were positively associated with PCV10 carriage ( Table 3 ). PCV10 vaccination and survey year were negatively associated with PCV10 carriage. There was no evidence of an interaction between PCV10 vaccination status ( P = 0.902) or survey year ( P = 0.171) and ethnicity with regard to PCV10 carriage.
10.1371/journal.pone.0231041.t003
Table 3 Unadjusted and adjusted odds ratios of PCV10 pneumococcal carriage in association with participant characteristics in four cross-sectional carriage surveys pre-PCV10 (2012) and annually thereafter (2013–2015) in Fiji (n = 7,962).
Exposure
PCV10 carriage a n / N (%)
Unadjusted odds ratio (95% CI)
P -value
Adjusted odds ratio (95% CI)
P -value
PCV10 vaccination status
0.043
0.002
Not vaccinated
629 / 6875 (9.2)
ref
ref
Vaccinated b
79 / 1087 (7.3)
0.78 (0.61–0.99)
0.58 (0.41–0.82)
Survey year
<0.001
<0.001
Pre-PCV10 (2012)
275 / 1975 (13.9)
ref
ref
1 year post-PCV10 (2013)
216 / 2022 (10.7)
0.74 (0.61–0.89)
0.74 (0.60–0.91)
2 years post-PCV10 (2014)
102 / 1987 (5.13)
0.33 (0.26–0.42)
0.40 (0.30–0.53)
3 years post PCV10 (2015)
115 / 1978 (5.8)
0.38 (0.30–0.48)
0.46 (0.35–0.61)
Ethnicity
<0.001
<0.001
Fijian of Indian Descent
147 / 3206 (4.6)
ref
ref
iTaukei
561 / 4756 (11.8)
2.78 (2.31–3.36)
2.70 (2.21–3.30)
Participant group
<0.001
<0.001
Caregivers
41 / 2029 (2.0)
ref
ref
Infants (5–8 weeks)
121 / 1946 (6.2)
3.21 (2.24–4.61)
3.60 (2.45–5.30)
Toddlers (12–23 months)
277 / 1972 (14.1)
7.92 (5.67–11.07)
9.76 (6.67–14.37)
Children (2–6 years)
269 / 2015 (13.4)
7.47 (5.34–10.44)
7.65 (5.32–11.00)
Residential location
0.001
0.001
Rural
304 / 3880 (7.8)
ref
ref
Urban
404 / 4082 (9.9)
1.29 (1.11–1.51)
1.34 (1.13–1.58)
Participant sex
<0.001
0.772
Male
352 / 3358 (10.5)
ref
ref
Female
356 / 4604 (7.7)
0.72 (0.61–0.84)
0.98 (0.82–1.16)
Number of children < 5 years living in the household c
<0.001
0.040
Less than two
283 / 4035 (7.0)
ref
ref
Two or more
425 / 3924 (10.8)
1.61 (1.38–1.88)
1.20 (1.01–1.43)
Family income level d
<0.001
0.004
Not low
207 / 3175 (6.5)
ref
ref
Low
463 / 4518 (10.3)
1.64 (1.38–1.94)
1.31 (1.09–1.57)
Symptoms of URTI
<0.001
<0.001
Not present
435 / 5903 (7.4)
ref
ref
Present
273 / 2059 (13.3)
1.92 (1.64–2.26)
1.42 (1.19–1.70)
Household cigarette smoke
<0.001
0.031
No exposure
283 / 3689 (7.7)
ref
ref
Exposure
425 / 4273 (10.0)
1.33 (1.14–1.56)
1.21 (1.02–1.43)
Antibiotic use in previous fortnight e
0.012
0.367
Not used
663 / 7610 (8.7)
ref
ref
Used
44 / 348 (12.6)
1.52 (1.09–2.10)
0.84 (0.58–1.22)
Abbreviations: CI, confidence interval; PCV10, ten-valent pneumococcal conjugate vaccine; URTI, upper respiratory tract infection.
a Pneumococcal serotypes included in PCV10 (serotypes 1, 4, 5, 6B, 7F, 9V, 14, 18C, 19F, and 23F)
b Two doses of PCV10 given before the age of 12 months, or one or more doses of PCV10 given at or after 12 months of age [ 35 ]
c Data on number of children under five years living in the household were missing for three participants, of whom none were PCV10 pneumococcal carriers
d Weekly family income below the basic needs poverty line (<FJ$175 per week) [ 28 ]; data on family income were missing for 269 participants, of whom 38 were PCV10 pneumococcal carriers
e Data on antibiotics use were missing for four participants, of whom one was a PCV10 pneumococcal carrier.
Factors associated with non-PCV10 carriage
iTaukei ethnicity, young age (infant, toddler, and child participant groups), urban residence, living with two or more children younger than five years, low family income, and URTI symptoms were positively associated with non-PCV10 carriage ( S1 Table ). As with overall carriage, there was evidence of an interaction between survey year and ethnicity, as the two ethnic groups had differential odds of non-PCV10 carriage ( P <0.001), but no evidence of interaction between PCV10 vaccination status and ethnicity ( P = 0.856).
Factors associated with overall pneumococcal density
Toddler and child participant groups, and symptoms of URTI were positively associated with density of overall pneumococcal carriage ( Table 4 ). There was evidence of an association between iTaukei ethnicity and overall pneumococcal density, however the confidence interval included the null value. Although the adjusted median difference in overall pneumococcal carriage density increased in the first two years after the introduction of PCV10, this was not sustained into the third year ( Table 4 ). There was no indication of an interaction between PCV10 vaccination status ( P = 0.864) or survey year ( P = 0.347) with ethnicity.
10.1371/journal.pone.0231041.t004
Table 4 Unadjusted and adjusted differences in medians of overall pneumococcal carriage density in association with participant characteristics in four cross-sectional carriage surveys pre-PCV10 (2012) and annually thereafter (2013–2015) in Fiji (n = 2,441).
Exposure
Density of overall pneumococcal carriage a (log10 GE/ml) n, median / IQR
Unadjusted mean difference (95% CI)
P -value
Adjusted mean difference (95% CI)
P -value
PCV10 vaccination status
0.408
0.210
Not vaccinated
2026, 4.9 (4.2–5.8)
ref
Ref
Vaccinated b
415, 4.9 (4.0–5.7)
-0.06 (-0.21, 0.08)
-0.13 (-0.35, 0.08)
Survey year
0.004
0.001
Pre-PCV10 (2012) b
707, 4.9 (4.2–5.7)
ref
ref
1 year post-PCV10 (2013)
655, 5.2 (4.3–5.9)
0.28 (0.13, 0.43)
0.33 (0.18, 0.47)
2 years post-PCV10 (2014)
433, 5.0 (4.1–5.8)
0.11 (-0.06, 0.27)
0.18 (0.00, 0.36)
3 years post-PCV10 (2015)
646, 4.9 (4.0–5.7)
0.00 (-0.15, 0.15)
0.01 (-0.17, 0.18)
Ethnicity
0.012
0.053
Fijian of Indian Descent
495, 4.8 (4.1–5.7)
ref
ref
iTaukei
1946 5.0 (4.2–5.8)
0.17 (0.04, 0.31)
0.14 (0.00, 0.28)
Participant group
0.003
0.008
Caregivers
193, 4.7 (4.0–5.4)
ref
ref
Infants (5–8 weeks)
515, 4.9 (4.1–5.7)
0.17 (-0.06, 0.39)
0.17 (-0.06, 0.40)
Toddlers (12–23 months)
845, 5.0 (4.1–5.8)
0.33 (0.11, 0.55)
0.32 (0.08, 0.55)
Children (2–6 years)
888, 5.0 (4.3–5.8)
0.34 (0.12, 0.55)
0.34 (0.12, 0.55)
Residential location
0.612
Rural
1070, 5.0 (4.1–5.7)
ref
Urban
1371, 5.0 (4.2–5.7)
0.03 (-0.08, 0.14)
Participant sex
0.597
Male
1166, 5.0 (4.2–5.7)
ref
Female
1275, 5.0 (4.2–5.8)
0.03 (-0.08, 0.14)
Number of children < 5 years living in the household
0.708
Less than two
954, 5.0 (4.2–5.8)
ref
Two or more
1487, 5.0 (4.2–5.7)
-0.02 (-0.13, 0.09)
Family income level c
0.943
Not low
801, 5.0 (4.1–5.7)
ref
Low
1522, 5.0 (4.2–5.8)
0.00 (-0.11, 0.12)
Symptoms of URTI
<0.001
<0.001
Not present
1528, 4.9 (4.1–5.6)
ref
Ref
Present
913, 5.2 (4.3–5.9)
0.31 (0.19, 0.42)
0.28 (0.16, 0.40)
Household cigarette smoke
0.540
No exposure
1098, 5.0 (4.1–5.7)
ref
Exposure
1343, 5.0 (4.2–5.8)
0.03 (-0.08, 0.15)
Antibiotic use in previous fortnight d
0..084
0.752
Not used
2324, 5.0 (4.2–5.7)
ref
ref
Used
115, 5.2 (4.5–5.7)
0.22 (-0.03, 0.48)
0.04 (-0.23, 0.30)
Abbreviations: CI, confidence interval; PCV10, ten-valent pneumococcal conjugate vaccine; URTI, upper respiratory tract infection.
a Density of overall, including non-encapsulated, pneumococci
b Two doses of PCV10 given before the age of 12 months, or one or more doses of PCV10 given after 12 months of age [ 35 ]
c Weekly family income below the basic needs poverty line (<FJ$175 per week) [ 28 ]; data on family income were missing for 118 pneumococcal carriers
d Data on antibiotic use were missing for two pneumococcal carriers.
Factors associated with PCV10 pneumococcal density
Symptoms of URTI were positively associated with median density of PCV10 carriage ( S2 Table ). Although there was an initial increase in the adjusted median difference in PCV10 carriage density in the first year after PCV10 introduction, it was not sustained through the second and third year post-PCV10 introduction ( S2 Table ). There was no indication of an interaction between PCV10 vaccination status ( P = 0.170) or survey year ( P = 0.686) with ethnicity.
Factors associated with non-PCV10 pneumococcal density
Symptoms of an URTI and young age (infant, toddler, and child participant groups) were positively associated with median density of non-PCV10 carriage ( S3 Table ). As with density of overall and PCV10 pneumococcal carriage, the increase in density of non-PCV10 carriage in association with survey year was not sustained to the third year post-PCV10 introduction, and there was no evidence of interaction between PCV10 vaccination status ( P = 0.156) survey year ( P = 0.138) with ethnicity.
Discussion
This study provides new findings on the factors associated with overall, PCV10, and non-PCV10 pneumococcal carriage and density, in the years surrounding PCV10 introduction in a tropical, upper middle-income country in the Asia-Pacific region.
In this study, iTaukei ethnicity was an independent predictor of carriage (overall, PCV10, and non-PCV10), after adjustment for PCV10 vaccination status and survey year post-PCV10 introduction. We also checked the potential interaction between ethnicity and other socio-demographic variables, and found no evidence of interaction. Previously, we found that prior to PCV10 introduction, iTaukei ethnicity was associated with increased odds of overall pneumococcal carriage in children aged 3–13 months (aOR 2.81 [95% CI 1.76–4.49] P <0.001) [ 4 ]. The reason for this association is unknown. Further, we have reported that differences in social contact patterns by ethnicity partially account for higher prevalence of pneumococci among iTaukei, compared with FID, but that differences in carriage prevalence are also likely related to ethnic differences in host or environmental factors [ 36 ]. Few studies have described associations between pneumococcal carriage and indigenous and non-indigenous populations living in the same area, with similar access to healthcare, and with similar and high PCV coverage rates. Our findings are consistent with other studies comparing indigenous and non-indigenous populations in the same setting. In a pre- and very early post-PCV7 longitudinal study of 280 indigenous and non-indigenous children in remote Australia, who were followed from birth to 2 years, indigenous status was positively associated with pneumococcal carriage (OR 3.3 [95% CI 2.6–4.2] P <0.001)[ 37 ]. In Israel, a longitudinal study of 369 Bedouin and 400 Jewish children enrolled in a trial of PCV7, found Jewish children to have significantly lower odds of pneumococcal carriage, compared with Bedouin children (aOR 0.14 [0.10–0.21]) [ 38 ]. In contrast, a post-PCV13 cross-sectional study of 352 children aged less than six years in Greenland, found indigenous ethnicity was not associated with pneumococcal carriage (aOR 0.7 [95% CI 0.3–1.5] P = 0.32) [ 39 ]. Likewise, a cross-sectional study in Alaska post-PCV7 involving 1,275 children aged 3–59 months, found no association between indigenous ethnicity and overall or PCV7 carriage (OR 1.0 [95% CI 0.8–1.3] and (OR 1.1 [95% CI 0.75–1.6], respectively) [ 40 ]. Unlike our study, however, both the Greenland and Alaska studies occurred in high income settings, which may comparatively reduce the impact of ethnicity based socio-environmental differences that might be related to pneumococcal carriage. Further, the Greenland study included few non-Inuit participants, and may have been underpowered for analysis by ethnicity [ 39 ].
Other factors associated with pneumococcal carriage in this study, were largely consistent with studies from the pre-PCV10 era. For example, we observed young age, residential location, living with young children, low family income, and symptoms of URTI to be risk factors for all types of carriage [ 4 , 8 , 12 ]. Similarly, the majority of studies assessing factors associated with pneumococcal carriage after the introduction of PCV into national immunization schedules have reported age, poverty or proxies of poverty, number of young children living in the household, and symptoms of URTI as positively associated with pneumococcal carriage [ 41 – 45 ]. Previous studies have heterogenous findings regarding exposure to cigarette smoke and recent antibiotic use and their associations with pneumococcal carriage [ 5 , 11 , 46 – 53 ]. We found exposure to household cigarette smoke was a risk factor, but only for PCV10 carriage. However, levels of exposure to cigarette smoke require detailed monitoring, which was not incorporated in this study. We also found no association between antibiotic use and pneumococcal carriage, which may reflect very low prevalence of recent antibiotic use in our sample.
In this study, PCV10 vaccination status and survey year were protective against overall and PCV10 carriage, but were not associated with non-PCV10 carriage. We found only three studies undertaken after PCV was introduced into national immunization programs that assessed factors associated with pneumococcal carriage, and included PCV vaccination status as a variable [ 16 , 54 , 55 ]. In a cross-sectional study of 361 children under five years of age in Jimma, Ethiopia, the odds of overall pneumococcal carriage increased in association with having siblings under five years old (aOR 1.80 [95% CI 1.17–2.77] P = 0.008), and malnutrition (aOR 2.07 [95% CI 1.24–3.44] P = 0.005), but PCV vaccination was not associated with a decrease in carriage (three doses aOR 1.08 [95% CI 0.60–1.89] P = 0.82, one or two doses aOR 1.06 [95% CI 0.40–2.83] P = 0.90) [ 54 ]. This may have been due to serotype replacement and capsular switching of pneumococci by recombination, such that the immune pressure from PCV selected for non-vaccine serotypes [ 54 ]. However, in a cross-sectional study of 1,668 healthy children aged 12–23 months in Brazil, the odds of vaccine-type carriage declined in association with three (aOR 0.073 [95% CI 0.026–0.204] P <0.001) and four (aOR 0.027 [95% CI 0.007–0.113] P <0.001) doses of PCV10, and increased in association with day care attendance (aOR 2.358 [95% CI 1.455–3.821] P <0.001) and colonization with H . influenzae (aOR 2.454 [95% CI 1.529–3.939] P = 0.0006) [ 55 ]. Similarly, in pre and post-PCV13 pneumococcal carriage surveys involving 999 infants aged 5–8 weeks and 1,010 toddlers aged 12–23 months in Lao PDR, two or three doses of PCV13, compared with zero or one, was protective against PCV13 carriage among toddlers (aOR 0.60 [95% CI 0.44–0.83] P = 0.002) [ 16 ].
Although our findings are consistent with the Brazilian and Lao PDR studies regarding PCV being protective against PCV carriage, broader comparisons between these and other studies is hampered by the heterogeneity of settings, populations sampled, and the factors and definitions used. For example, in our study participants were community based, healthy, from four different age groups, did not attend childcare, and we did not include co-colonisers or malnutrition as exposures. Comparatively, the studies from Ethiopia, Brazil, and Lao PDR included child and infant participants only, including those attending childcare / school, and those suffering malnutrition, pneumonia, sinusitis, and otitis media.
Our observations regarding no association between PCV10 and non-PCV10 carriage may be due a lack of selection pressure towards an overall increase in non-PCV10 carriage early post-PCV10 introduction, due to serotype replacement occurring only in iTaukei infants and toddlers, rather than across all participant groups, as described previously [ 22 ]. In contrast, increases in non-vaccine type carriage have been reported following the introduction of PCV7 in England, PCV10 in Kenya, and PCV13 in Malawi and The Gambia [ 56 – 59 ]. However, it should be noted that the studies from Kenya, England, The Gambia, and Malawi were in vastly different contexts from our study, one notably in a high-income setting, rendering comparisons difficult.
Pneumococcal density has previously been found to be positively associated with microbiologically confirmed pneumococcal pneumonia, and could be used to improve estimates of pneumococcal pneumonia prevalence in childhood pneumonia studies [ 2 ].Our study contributes to the understanding of factors associated with pneumococcal carriage density. We found that symptoms of an URTI were associated with increased median density of carriage (overall, PCV10, and non-PCV10), consistent with cross-sectional carriage surveys from Peru, Lao PDR, and Indonesia [ 9 , 11 , 13 , 16 ]. We found PCV10 vaccination was not associated with pneumococcal density, and that although differences in median density of all types of pneumococci increased one to two years following PCV10 introduction, this was not sustained into the third year. There are relatively few risk factor studies describing the association between pneumococcal density and PCV vaccination status. Those that do, have heterogeneous findings. A double-blind, randomized controlled trial of PCV13 and Hepatitis A vaccine (control arm) in adults, using the Experimental Human Pneumococcal Challenge model, found that pneumococcal density in the PCV13 arm was significantly lower compared with the control arm ( P <0.0001) [ 60 ]. In a cluster-randomized trial of PCV7 and Meningococcal C vaccine (control arm) in rural Gambia, density of PCV7 pneumococcal carriage was lower in PCV7 villages compared with control arm villages [ 61 ]. However, this was not attributed to PCV7, as density of non-PCV7 carriage also declined in both vaccine and control villages [ 61 ]. Similarly, a decline observed in both PCV13 and non-PCV13 pneumococcal density in Laotian infants and toddlers was attributed to secular trends rather than PCV13 directly [ 16 ].
Limitations to our study should be noted. Firstly, because participants were recruited from Greater Suva and the surrounding areas, generalizability to the wider Fijian population may be limited. The non-random sampling method may have introduced sample selection bias, such results may not be generalizable to the Fijian population. However, the purposive quota sampling method rendered the sample representative of the Fijian population with regard to ethnicity and residential location, which are associated with pneumococcal carriage [ 4 , 27 ]. The cross-sectional nature of this observational study precludes causal, associations from being drawn between participant characteristics and pneumococcal carriage or densities. Finally, we did not collect data on co-colonizing bacterial or viral species, and are therefore unable to investigate the association of such factors with pneumococcal carriage or density, which have been identified as risk factors in other studies [ 9 – 11 ].
These limitations notwithstanding, this study documents the factors associated with pneumococcal carriage and density post-PCV10 introduction in an upper middle-income country. Introduction of PCV10 was negatively associated with the odds of overall and PCV10 pneumococcal carriage in Fiji. However, iTaukei ethnicity remains positively associated with pneumococcal carriage in Fiji, despite high and similar PCV10 coverage rates across iTaukei and FID populations. Further research is warranted regarding the factors underlying the observed ethnicity based differences in pneumococcal carriage, and whether the impact of PCV10 on pneumococcal disease incidence differs by ethnicity in Fiji.
Supporting information
S1 Table
Unadjusted and adjusted odds ratios of non-PCV10 pneumococcal carriage in association with participant characteristics in four cross-sectional carriage surveys pre-PCV10 (2012) and annually thereafter (2013–2015) in Fiji (n = 7,962).
(DOCX)
S2 Table
Unadjusted and adjusted differences in medians of PCV10 pneumococcal carriage density in association with participant characteristics in four cross-sectional carriage surveys pre-PCV10 (2012) and annually thereafter (2013–2015) in Fiji (n = 708).
(DOCX)
S3 Table
Unadjusted and adjusted differences in medians of non-PCV10 pneumococcal carriage density in association with participant characteristics in four cross-sectional carriage surveys pre-PCV10 (2012) and annually thereafter (2013–2015) in Fiji (n = 1,905).
(DOCX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Leptin is a pleiotropic hormone that decreases appetite and increases energy expenditure [1] , [2] , [3] , [4] , [5] , [6] . Obese, leptin deficient obob mice have profoundly decreased locomotor activity [5] , [7] . Treatment of obob mice for three weeks with pharmacologic doses of leptin increases locomotor activity and substantially decreases adiposity [5] , but this late effect of leptin on activity could be secondary to reversal of the obesity as opposed to a direct effect of leptin on this behavior [5] .
While total activity in obob mice is decreased, food intake is increased. Humans with complete leptin deficiency display aggressive food seeking behavior [8] , [9] , [10] that is rapidly suppressed by treatment with leptin [11] , [12] , [13] , [14] , [15] .
There are different types of locomotor activity that can be assayed in mice including home cage activity (HCA), running wheel activity (RWA), and food anticipatory activity (FAA), each of which has distinct behavioral significance.
In mice, we tested the hypothesis that leptin increases total locomotor activity but inhibits food anticipatory activity. Using two independent assays of locomotor activity, we studied both a novel Tet-off transgenic mouse model in which circulating leptin can be acutely and non-invasively suppressed and leptin deficient obob mice. We analyzed effects of changing leptin on both total locomotor activity and on food anticipatory activity (FAA). Leptin was administered systemically and into a lateral cerebral ventricle. The time course of changes in activity was compared with changes in body weight to determine if leptin altered activity independent of changes in body weight. Finally, we evaluated the role of melanocortin-3 receptors in leptin-induced influences on FAA in obob mice [16] .
Results
Leptin deficient obob mice show decreased total locomotor activity
As previously reported [5] , [7] , obob mice showed profoundly decreased locomotor activity compared to wild type controls using both HCA and RWA assays ( Fig. 1 A – B ).
10.1371/journal.pone.0023364.g001
Figure 1
Locomotor responses to leptin in obob mice.
( A and B ) 24 hour running wheel activity (RWA) and home cage activity (HCA) in obob and wild type (WT) mice. ***p≤0.001 (A, WT n = 11 and obob n = 12; B, WT = 9 and obob n = 6). ( C ) Effect of leptin (150 ng/h) on 24 h RWA in obob mice. RWA decreased (*p≤0.05) with time and vehicle (PBS) in the obob mice (n = 6 for each group). In contrast, RWA activity increased (***p≤0.001) during leptin. ( D ) Effect of leptin (150 ng/h) on 24 h HCA in WT and obob mice . Leptin increased HCA in obob mice (**p≤0.01), but not in WT mice (n = 6 for each group).
Leptin produces early increases in locomotor activity in obob mice before substantial weight loss
In obob mice, systemic infusion of leptin (150 ng/h) produced marked early increases in RWA and HCA ( Fig. 1 C–D and 2 ). Recent studies from our laboratory in leptin deficient obob mice, indicate that infusion of leptin, 150 ng/h, restores physiologic plasma levels of leptin (∼5 ng/ml) in adult wild type mice [17] . Leptin infusion at 150 ng/h increased (p≤0.01) running wheel activity by +122% within 2 days ( Fig. 2 A – B ) at a time when there was a statistically significant but small decrease in body weight averaging −5% ( Fig. 2 B ). The increase in activity occurred during the dark phase ( Fig. 2A ). Activity returned to baseline during the recovery period following cessation of leptin administration ( Fig. 1 C ). A supraphysiologic dose of leptin (750 ng/h) did not produce further increases in activity in obob or wild type mice ( Figure S1 ). Food intake tended to decrease more during leptin 750 ng/h than during leptin 150 ng/h but activity tended to increase more during leptin 150 ng/h showing a divergence of the effects of leptin on activity and food intake ( Figures S1 and S2 ). Infusion of a lower dose of leptin (25 ng/h) that only partially restores physiologic levels of leptin [17] , [18] led to progressive increases in RWA in obob mice whereas vehicle did not increase activity ( Fig. S3 ). Leptin (150 ng/h) did not increase locomotor activity in wild type mice ( Fig. 1 D ).
10.1371/journal.pone.0023364.g002
Figure 2
Time course of leptin-induced increases in activity.
( A ) Representative actogram of RWA in an obob mouse before and during leptin (n = 6). The increase in RWA began by the 2nd day of leptin. ( B ) Summary data comparing the time course of increases in RWA vs decreases in body weight. Leptin caused significant(p≤0.01) and substantial (+122%) increases in RWA by day 2. At this time, there was a significant (p<0.001) but small (−5%) decrease in body weight.
These findings suggest that leptin-induced changes in locomotor activity are seen only with changes in plasma leptin levels within the physiological range.
Central neural mediation of leptin effect on locomotor activity
To determine if the effect of leptin on locomotor activity is mediated by the central nervous system, we infused leptin directly into a lateral cerebral ventricle (ICV) at 5 ng/h. ICV infusion of leptin in obob mice produced progressive early increases in total locomotor activity (HCA) beginning on the 1 st day (+28±8%; p≤0.05) and increasing to +69±23% on day 3 ( Fig. 3 A–B ), whereas body weight did not decrease, when compared to controls, at this time ( Fig. 3 B ). There were further increases (p≤0.01) in HCA between the second and third weeks of ICV leptin at a time when there were significant (p≤0.05) decreases in body weight ( Fig. 3 B ). ICV infusion of vehicle (PBS) did not increase locomotor activity in obob mice ( Fig. 3 A ).
10.1371/journal.pone.0023364.g003
Figure 3
Activity responses to cerebroventricular (ICV) administration of leptin.
( A ) ICV leptin (5 ng/h) increased HCA during weeks 1 (*p≤0.05) and 2 in obob mice. (**p≤0.01) ( B ) Temporal data showing that ICV leptin produced significant (p≤0.05) increases in HCA between days 1 and 4 whereas body weight did not decrease significantly during this time (PBS group n = 3; leptin group n = 4).
These findings suggest that the effects of leptin on spontaneous activity are mediated by a central neural action.
Generation and characterization of leptin, tetracycline-off transgenic mice
Generation of transgenic (Tg) mice chronically over-expressing leptin has been previously reported [19] , [20] , [21] , [22] . The Tet regulatory systems have been used extensively in cell culture systems and in mice [23] . In our mouse model, the expression of the Tg source of leptin can be turned off non-invasively by administering a tetracycline antibiotic (doxycyline or DOX) in the food. We used two classes of mouse lines: one that controls tTA expression in the liver indicated as LAP-tTa [24] and the other containing the TRE-hleptin components ( Fig. S4 A ). LAP-tTa/TRE-hleptin (Tg) mice showed a “skinny” phenotype [20] . The characterization of the line is described in Supporting Information S1 and Fig. S4 .
Both transgenes were then crossed in the obob background to generate LAP-tTa/Tre-hleptin/ obob (Tg obob ) mice, a strain in which the absence of a functional endogenous leptin gene allows the study of the acute effect of leptin deprivation in mice that are not perturbed or manipulated other than for the administration of Dox ( Fig. 4 A ). After the administration of doxycyline in the food, serum hleptin levels were suppressed (<15 pg/ml) within 24 h ( Fig. 4 B ) and accompanied by a progressive increase in body weight ( Fig. 4 C – D ).
10.1371/journal.pone.0023364.g004
Figure 4
Effects of acute leptin suppression on activity in Tet-off hyperleptinemic transgenic obob mice.
( A ) Strategy followed to obtain transgenic mice chronically over-expressing hleptin on an obob background. LAP-tTa/TRE-hleptin/ obob mice are skinny since hleptin is over-produced until doxycycline (DOX) is administered. ( B ) In Tg obob mice, plasma hleptin levels were suppressed 1, 3, 6, and 13 days after beginning chronic administration DOX in the food at concentrations of 0.1, 0.5, and 2 g/kg. After DOX suppression, hleptin can be turned on again by switching back to regular chow. The “recovery” time necessary to document detectable hleptin in the plasma (1M = 1 month, 2M = 2 months) is a function of the DOX concentration in the food and the duration of the administration as shown after a 13 days of DOX. ( C ) Administration of DOX to LAP-tTa/Tre-hleptin/ obob at 8 weeks of age was accompanied by an increase in body weight. ( D ) The top panel compares an 8 week old LAP-tTa/Tre-hleptin/ obob mouse ( left ) and a littermate obob control ( center ) before DOX. The bottom panel shows the same mice 5 weeks after beginning DOX. ( E–F ) Effect of acute leptin suppression on activity in Tg obob mice. ( E ) After beginning DOX, a steady decrease in HCA is observed in Tg obob mice (n = 5) as shown by linear regression analysis, becoming significant (p≤0.05) after 7 days of DOX and continuing through the end of the experiment (p<0.005),(n = 9 for wild type and n = 7 for obob ). ( F ) Leptin suppression also significantly decreased RWA in Tg obob (n = 5) although at a slower rate becoming significant at day 20 (p<0.05). (n = 11 for wild type and n = 12 for obob mice.)
Abrupt leptin suppression in skinny, hyperleptinemic transgenic mice decreases total locomotor activity
In hyperleptinemic, “skinny” Tg obob mice, baseline HCA was normal compared to wild type controls showing that leptin treatment can normalize the activity of an obob mouse and confirming that supra-physiological levels of leptin do not augment activity above physiologic levels ( Fig. 4 E ). In wild type mice, HCA and RWA did not decrease in the first two weeks after doxycycline. In contrast, the abrupt suppression of leptin occurring within 24 hours following addition of doxycycline decreased HCA in Tg obob mice as shown by linear regression analysis ( Fig. 4 E ), becoming significant (p≤0.05) 7 days after suppression and continuing through the end of the experiment (p≤0.005). This decrease in activity occurred entirely during the dark phase. Leptin suppression significantly (p≤0.05) decreased RWA in the Tg obob mice although at a slower rate ( Fig. 4 F ). These data indicate that acute leptin suppression reduces activity mirroring our data that leptin replacement in obob mice acutely increases activity.
Pair feeding increases locomotor activity in obob mice
To determine if the increased activity during leptin treatment was influenced by its effect to decrease food intake, we pair fed a group of obob mice treated with vehicle to the food intake of obob mice treated with leptin 150 ng/h ( Figure S2 ). Pair fed obob mice treated with vehicle also showed a substantial increase in activity ( Fig. 5 A ). However, the pattern of activity was distinctly different from that observed after leptin administration ( Fig. 5 B – C ). With leptin administration, the increase in activity occurred entirely during the dark phase ( Fig. 5 B ). In contrast pair fed obob mice showed increased activity that began during the light phase and peaked during the early dark phase when food was first provided ( Fig. 5 C ). This pattern of increased activity during pair feeding in obob mice is suggestive of food anticipatory activity.
10.1371/journal.pone.0023364.g005
Figure 5
Comparison of the effects of leptin vs pair feeding on RWA in obob mice.
(A) As shown previously, leptin (150 ng/h) increased RWA (***p≤0.001). Pair fed obob mice (not treated with leptin) also displayed striking increases in activity (***p≤0.001). However, the temporal pattern of activity during pair feeding was strikingly different from that seen during leptin (n = 6 each group). The leptin-induced increase in activity occurred entirely during the dark phase (B), whereas the increased activity during pair feeding occurred substantially during the light phase and peaked during the early dark phase time at which food was provided ( C ).
Food anticipatory activity (FAA) is augmented in obob mice and suppressed by leptin
We next performed a set of behavioral experiments to assess FAA. In these experiments, availability of food was restricted progressively to the first 8, 6, and 4 hrs of the onset of the dark phase and then shifted to 4 hrs in the middle of the light phase ( Fig. S5 ). With this protocol, FAA is expressed as the ratio of the activity in the three hours before the availability of food divided by total 24 hr activity and expressed as percent. When access to food was restricted to 4 hrs per day, FAA was markedly augmented in obob mice ( Fig. 6 A ) compared to wild type mice ( Fig. 6 B ). It is noteworthy that this increase in FAA in the obob mice is observed in the absence of leptin indicating that other nutritional signals are responsible for this behavioral change. Leptin 150 ng/h abolished FAA in obob mice ( Fig. 6 A ,), but not in wild type mice ( Fig. 6 B ). In obob mice, in the same protocol in which leptin abolished FAA ( Figures 6 A,C and S6 ) it increased total home cage and running wheel activity. This inhibitory effect of leptin on FAA in obob mice and its contrasting effect to increase total activity are also shown in the actograms of Fig. S6 . Leptin did not increase total activity in wild type mice. In other words, in leptin deficient obob mice, physiologic doses of leptin abolished FAA, but increased total locomotor activity.
10.1371/journal.pone.0023364.g006
Figure 6
Food anticipatory activity (FAA) is augmented in obob mice and suppressed by leptin.
( A–B ) FAA measured as HCA in obob and wild type (WT) mice (n = 6 each group) treated with vehicle (PBS) or leptin (150 ng/h). FAA was markedly augmented (p≤0.001) in obob mice compared with WT mice. Leptin virtually abolished (p≤0.001) FAA in obob mice, but did not attenuate FAA in WT mice. ( C ) FAA measured as RWA was pronounced in obob mice and abolished by leptin administration (150 ng/h). (n = 6 each group).
As shown in Fig. S7 , food intake decreased during the FAA protocol in the obob mice that displayed augmented FAA. Importantly, leptin suppressed FAA in obob mice ( Fig. 6 A,C ) despite an even lower food intake which would have been expected to increase FAA.
Role of melanocortin 3 receptors in augmented FAA in obob mice
We next tested whether the effect of leptin on FAA was abrogated in the absence of the melanocortin MC3 receptor. We were interested in testing the role of MC3R because of a previous report showing that melanocortin 3 receptor knockout mice (MC3R −/−) have decreased food anticipatory activity [16] . We tested whether the effect of leptin on FAA was reduced in MC3R −/− obob double mutant mice. Body weight and food intake of MC3R −/− obob mice were not significantly different from control obob mice. In contrast to previous observations that MC3R −/− mice have decreased FAA [16] , MC3R−/− obob knockout mice had enhanced, not attenuated, FAA ( Fig. 7 A ) while their total locomotor activity did not differ from obob mice ( Fig. S8 A ). Furthermore, similar to its effects in obob mice, leptin inhibited FAA ( Fig. 7 B ) in MC3R−/− obob mice and increased total locomotor activity ( Fig. S8 B ). These data indicate that MC3R do not mediate the influence of leptin on food anticipatory activity or total locomotor activity.
10.1371/journal.pone.0023364.g007
Figure 7
Role of melanocortin-3 receptors (MC3R) in augmented FAA in obob mice.
( A ) Leptin deficient, obob mice with deletion of melanocortin-3 receptors (MC3R−/− obob , n = 12 ) had increased FAA (p≤0.05) compared with that in obob mice (n = 5). ( B ) FAA in MC3R−/− obob mice (n = 7) treated with leptin (150 ng/h) vs FAA in control mice treated with PBS (n = 6). Leptin abolished (p≤0.01) FAA in the MC3R−/− obob mice.
Discussion
While Pelleymounter, et al [5] reported that treatment of obob mice with high doses of leptin for three weeks increased locomotor activity, this was observed at a time when body weight had decreased by approximately 40% making it difficult to determine if the increased activity reflected reversal of obesity or a direct effect of leptin. Our finding that both systemic and cerebroventricular administration of leptin produced early increases in activity in obob mice before substantial decreases in body weight provides evidence for a direct effect of leptin at least partially independent of reversal of obesity. The concept of a direct effect of leptin on activity independent of adiposity is supported by the decreases in activity with abrupt leptin suppression in hyperleptinemic, skinny, Tet-off transgenic obob mice. This result in the Tet-off transgenic mice, obtained non-invasively and without manipulation in food availability, expands on a previous study from our laboratory [25] . In the previous study, there was a 50% reduction in locomotor activity when a leptin deficient state was produced by a combination of abrupt leptin withdrawal and restriction of food intake. This decrease in activity was not accompanied by weight gain, given the restricted food intake.
A striking observation in this study was the robust increase in food anticipatory activity (FAA) in obob mice with food restriction. This is consistent with previous reports that Zucker obese rats (affected by a loss of function mutation in the leptin receptor) have augmented food anticipatory or seeking activity compared to lean control rats [26] , [27] . This augmented FAA is by definition leptin independent since obob mice have a loss of function mutation in the leptin gene. The nature of the relevant signal(s) for the potentiated FAA in the obob mice is unknown. Food intake is regulated by numerous short and long term hormonal and metabolic signals. The robust FAA in obob mice provides an assay for the identification of signals and neural pathways mediating this phenomenon independent of leptin. Based on previous observations that the absence of melanocortin-3 receptors (MC3R) abrogates FAA in MC3R deficient mice, we postulated that MC3R might mediate the augmented FAA in obob mice. To our surprise, in a new double mutant MC3R −/− obob mouse model, the augmented FAA was preserved and not attenuated by deletion of MC3R.
Leptin replacement in obob mice abolished FAA while increasing total locomotor activity. The contrast between the effects of leptin on FAA and total locomotor activity has implications for understanding the different components of locomotor activity by showing that they can be regulated independently by leptin. The leptin-induced suppression of FAA in obob mice is consistent with observations from humans by two groups of investigators [11] , [12] . Using functional magnetic resonance brain imaging, these investigators demonstrated that in patients with congenital leptin deficiency, treatment with leptin reduced brain activation in regions linked to hunger while enhancing activation in regions linked to satiety. In addition, in patients with common human obesity, brain imaging indicated that hunger associated with diet-induced obesity is related to relative leptin deficiency and is reversed by leptin reconstitution [28] .
Morton et al [5] , [29] recently reported that subcutaneous leptin acutely increased spontaneous physical activity in obob mice fed ad lib . In contrast, leptin suppressed activity in fasted obob and wild type mice. Morton et al concluded that leptin is a physiological regulator of spontaneous physical activity, but that the nature of leptin's effect on activity is dependent on food availability [5] , [29] . However, these investigators did not assess effects of leptin on food anticipatory activity and did not assess leptin actions non-invasively. Our study extends the observations of Morton et al [29] in several ways. In addition to demonstrating that administration of leptin promptly increases activity in obob mice, we demonstrated that acute non-invasive leptin suppression in hyperleptinemic, skinny, tet-off transgenic mice decreases locomotor activity. We also found that cerebroventricular administration of leptin promptly increased activity in obob mice supporting a central neural mechanism of leptin regulation of activity. Finally, we also performed experiments to evaluate food anticipatory activity (FAA) by reducing the time but not the amount of food availability. FAA was augmented in obob mice. In this protocol, leptin suppressed FAA while increasing total 24 hr locomotor activity. These observations indicate that leptin produces opposing effects on total locomotor activity and FAA within a 24 hr protocol.
The increase in total locomotor activity with cerebroventricular infusion of leptin in our study raises the question of the neural pathways and mediators involved in the locomotor actions of leptin. In leptin receptor (LRb) null mice, unilateral arcuate restoration of LRb substantially improves locomotor activity while only modestly decreasing body weight [30] . In dbdb mice with loss of leptin receptor function, restoration of LRb only in pro-opiomelanocortin (POMC) neurons virtually normalizes locomotor activity with only modest decreases in body weight [31] . POMC neurons are expressed mainly in the arcuate nucleus. Since POMC neurons are expressed mainly in the arcuate nucleus, these two studies [30] , [31] support a key role for the arcuate nucleus in the locomotor actions of leptin and are consistent with the divergence of locomotor and body weight effects of leptin.
Mice with constituitive activation of signal transducer and activator of transcription 3 (STAT3) in agouti-peptide related (AgRP) neurons are lean and have increased locomotor activity [32] . Conversely, obese mice with genetically engineered disruption of LRb-STAT3 signaling have reduced locomotor activity although not to the low levels observed in dbdb mice with complete loss of LRb signaling [33] .
While restoration of physiologic levels of leptin increased total locomotor activity, a higher supraphysiologic dose did not produce further increases. These findings are consistent with reports that locomotor activity does not increase when normoleptinemic, wild type control mice are given supplemental leptin [5] , [34] . This suggests that leptin-induced increases in locomotor activity are seen only with changes in leptin within physiological levels.
Leptin is a highly pleiotropic hormone. Decreases in leptin, which are sensed as starvation, trigger a constellation of adaptive responses including increases in appetite, decreases in energy expenditure, infertility, reduced immune function, and development of a euthyroid sick state [3] . We suggest that decreases in total locomotor activity and augmentation of food anticipatory behavior during decreases in physiologic levels of leptin would contribute to protection from starvation during famine.
In summary, these studies provide several insights into the behavioral actions of leptin. First, acutely and non-invasively suppressing leptin levels in a novel Tet-off hyperleptinemic transgenic obob mouse model decreases activity. Second, leptin replacement in obob mice promptly increases activity before substantial decreases in body weight. Third, the leptin-induced increases in locomotor activity in obob mice occur with cerebroventricular in addition to systemic administration, implicating a central neural mechanism. Fourth, food anticipatory activity (FAA) is markedly augmented in leptin deficient mice and is abolished by leptin. Fifth, deletion of melanocortin-3 receptors in obob mice does not attenuate the augmented FAA produced by leptin deficiency and does not alter leptin-induced regulation of FAA or total locomotor activity. Sixth, leptin produces opposing effects on total locomotor activity and food anticipatory activity.
Materials and Methods
Mice
Male C57/BL6 or obob mice 8 weeks old from Jackson Laboratories (Bar Harbor, ME) were used. Mice were individually housed with food and water available ad libitum , and a 12∶12 h light dark cycle (lights off at 1900h). Mice were fed standard rat chow (Lab Diet 5053). All procedures were approved by The Rockefeller University Institutional Animal Care and Use Committee, and followed the Public Health Services Policy on Humane Care and Use of Laboratory Animals.
Administration of Leptin
For systemic administration, Alzet® osmotic pumps were filled with phosphate buffered saline (PBS) or murine leptin (Amylin) at 150 or 750 ng/h and were incubated in saline at 37°C overnight prior to implantation. Subcutaneous implantation of the pumps was performed under isofluorane anesthesia and at the end of the procedure the animals were returned to their home cages and behavior recorded. For cerebroventricular administration, a cannula was placed in a lateral cerebral ventricle under isofluorane anesthesia using a stereotaxic apparatus. Cannulae were implanted 0.75 mm lateral to the midline, 0.1 mm posterior to the bregma, and 3.5 mm ventral to the surface of the skull. The cannula was connected to an Alzet subcutaneous osmotic pump filled with PBS or murine leptin at 5 ng/h.
Running Wheel Activity
Animals were placed in individual running wheel cages [dimensions of cage: 32×17×14 cm and dimensions of wheel: 25 cm in diameter (Mini Mitter, Bend, OR)] with food and water available ad libitum . RWA was recorded in 1 minute bins, and later summed into 3 h or 24 h totals.
Generalized Arousal: Home Cage Activity
Individually housed mice were placed in the arousal testing chamber. The chamber consists of a 3 dimensional infra-red photobeam system (AccuScan Instruments), where the beams are placed at 1 cm intervals. Voluntary motor activity is measured as general voluntary activity in numbers of beam breaks in the vertical axis [vertical activity (VACTV)] and horizontal axis [horizontal activity (HACTV)] and also as total distance traveled in cm (Total Distance). All these variables were collected in 1 min intervals using Versamax software (AccuScan Instruments), and were later summed into 3 h or 24 h totals.
Leptin Tet-off Transgenic Mice
The generation of these mice is described in detail in the SI Appendix. In synthesis, LAP-tTA C57Bl6J mice were purchased from Jackson laboratories (Bar Harbor, ME) and crossed with TRE-hleptin Tg mice generated in C57Bl6J background by pronuclear injection. The function of the Tre-hleptin construct was tested in vitro and in vivo ( Fig. S4 B ). LAP-tTa/Tre-hleptin mice were then crossed in an obob background to generate LAP-tTa/Tre-hleptin/ obob . Only male mice were studied.
Hleptin serum levels in Tg mice were measured by a specific human leptin Elisa (R & D systems). Blood sampling was performed by orbital bleeding under isofluorane anesthesia. Animals were maintained on autoclaved standard chow (Lab Diet 5053). Doxycycline was mixed to the standard chow at the concentration of 0.1, 0.5 and 2 g/kg (Bio-Serve).
Food Anticipatory Activity ( Figure S5 )
Mice were housed in individual cages (either running wheel cages or shoebox cages for arousal - as above) under a 12∶12 h light∶dark cycle for 1 week, with food and water available ad libitum . Animals were then implanted with osmotic pumps containing PBS or leptin (150 ng/h) and activity was recorded for 1 week. Following this period, food was removed at dark onset (1400h), and food availability was gradually decreased in the following manner: day 1 - food available from 1400h to 2200h, day 2 - food available from 1400h to 2000h and from day 3 to day 10 food was available from 1400h to 1800h. On day 11, food availability was moved to the middle of the light period (0600h to 1000h), and this feeding schedule was maintained for an additional 5 days. Food was then provided ad libitum for 48 h, followed by no food for 48 h, and finally ad libitum food conditions for 48 h. RWA or HCA was recorded daily and was summed into 24 h total activity amounts, as well as the ratio of the activity displayed in the 3 h period preceding food presentation to the 24 h total activity (both preceding the dark onset feeding time and the middle of the light feeding period). Using this food anticipatory activity (FAA) paradigm with 48 h and refeeding periods, we are able to demonstrate that the mechanism behind the FAA is a food entrainable oscillator with a periodicity of 24 h rather than a hunger-derived, non-temporally regulated arousal signal.
Generation of MC3R−/− obob mice
MC3R−/− mice were purchased from Jackson Laboratory and crossed with ob/+ mice to obtain double heterozygote for the MC3R and ob gene mutation. Double heterozygote mice were then crossed to obtain MC3R−/− obob and obob controls.
Food Anticipatory Activity in MC3R−/− obob mice
MC3R −/− obob and obob littermate control mice were housed in individual running wheel cages under a 12∶12 h light∶dark cycle for 2 weeks, with food and water available ad libitum . To detect food anticipatory activity food was removed during the light phase (0600h), and food availability was gradually decreased in the following manner: day 1 - food available from 0600h to 1400h, day 2 - food available from 0600h to 1200h and from day 3 to day 14 food was available from 0600h to 1000h. On day 15 mice were left with food ad libitum for 4 weeks. Running wheel activity was recorded daily and was summed into 24 h total activity amounts, as well as the ratio of the activity displayed in the 3 h period preceding food presentation to the 24 h total activity. After the recovery period, mice were implanted, under general anesthesia, with subcutaneous (Alzet) pumps filled with leptin (150 ng/h) or PBS. Six days after surgery mice were presented again with a protocol to detect food anticipatory activity. Again food was removed during the light phase (0600h), and food availability was gradually decreased in the following manner: day 1 - food was available from 0600h to 1400h; day 2 - food was available from 0600h to 1200h; and from day 3 to day 14 food was available from 0600h to 1000h. On day 14 mice were left with food ad libitum .
Statistical Analysis
Comparisons between groups of animals were made using a one-way analysis of variance (ANOVA) followed by unpaired Student's t -test post-hoc and using linear regression analysis. Statistical analyses were conducted using SigmaStat Software version 3.1, Systat, Chicago, IL and GraphPad Prism 5.
Supporting Information
Figure S1
Effect of a supraphysiological dose of leptin on locomotor activity. 24 hour running wheel activity (RWA) in obob mice treated with leptin, 150 ng/h (n = 6) and 750 ng/h (n = 6). As shown previously in Figure 1 , leptin 150 ng/h significantly (***p≤0.001) increased RWA (left). In contrast, a supraphysiologic dose of leptin (750 ng/h) did not significantly increase RWA (right).
(TIF)
Figure S2
Food intake during vehicle, leptin, and pair feeding. As expected, food intake did not decrease during vehicle. The decrease in food intake tended to be greater during leptin 750 ng/h than during leptin 150 ng/h. Food intake in the pair fed group was matched to food intake in the group treated with leptin 150 ng/h. n = 6 each group. ***p≤0.001. *p≤0.05.
(TIF)
Figure S3
Effect of leptin (low dose) on activity of obob mice. RWA during infusion of a very low dose of leptin (25 ng/h; n = 7) or vehicle (n = 5) in obob mice. The low dose of leptin produced a progressive increase in RWA over four weeks.
(TIF)
Figure S4
Development of a hyperleptinemic Tet-off transgenic mouse line. Schematic of the transgenes used for generating Tet-off Tg obob mice: one line that controls the tetracycline trans-activator expression in the liver indicated as LAP-tTa ( left ) and the other containing the tetracycline responsive element together with human leptin cDNA ( right ). ( B ) The function of the Tre-hleptin construct was tested in vitro after transient transfection of the corresponding plasmid in a pK-15 Tet-off cell line ( left ) or co-transfecting it with an rtTA plasmid in HEK293T (Tet-on) cells ( right ). Our results show that, in vitro , the production of human leptin from the construct used to generate Tg mice can be efficiently induced or repressed in Tet-on and Tet-off settings respectively. ( C ) The specificity of the expression of the tTa in the liver was confirmed crossing the mouse line with a reporter EGFP mouse line and measuring fluorescence in the tissue extracts (Sk muscle = skeletal muscle, Cerebral c. = cerebral cortex, Fat = epididymal fat). ( D ) Pronuclear injection of the Tre-hleptin construct in C57Bl6 background resulted in 10 founder lines. ( E ) Lap-tTa/Tre-hleptin (Tg, n = 5) mice showed a “skinny” phenotype as indicated by the lower body weight curve compared to the littermates controls (WT, n = 14), while mice carrying the Tre-hleptin gene alone (Tre-Lept, n = 6) did not show body weight significantly different from the wild type mice. ( F ) The pattern of expression of the human leptin transgene was tested, using real time PCR, in different tissues (L = liver, AT = adipose tissue, K = kidney) in Tg mice, Tre-hleptin and controls.
(TIF)
Figure S5
Schematic of food anticipatory activity (FAA) protocol. Schematic of the protocol for study of FAA. Clear bars represent periods when food was withdrawn. Shaded bars represent periods when food was available.
(TIF)
Figure S6
Representative actograms of FAA activity in obob mice and the effect of leptin treatment. ( A ) Representative actograms of RWA in an obob mouse treated with vehicle or ( B ) leptin, 150 ng/h, before and during exposure to the FAA protocol described in Fig. S5 . As indicated to the right of the actograms, the outlines show the period of food availability during the FAA protocol. The obob mouse treated with vehicle expressed marked FAA activity (increase in RWA activity in the light phase before feeding time in the dark phase). During the same protocol, leptin treatment in an obob mouse completely suppressed FAA while increasing RWA during the dark phase.
(TIF)
Figure S7
Food intake of obob mice during FAA protocol. Food intake (plotted as mean every two days ± SEM) in obob mice treated with vehicle and compared to the leptin treated group. As expected, in the group of obob mice treated with vehicle, food intake decreased, during the initial phase of the FAA protocol.
(TIF)
Figure S8
Role of melanocortin-3 receptors (MC3R) in locomotor activity of obob mice. ( A ) Total RWA was not significantly attenuated in obob mice with deletion of MC3R (MC3R−/− obob , n = 12) compared to obob (n = 5) . ( B ) Leptin increased (p≤0.05) total RWA in the MC3R−/− obob mice (n = 7) compared to controls treated with vehicle (n = 6).
(TIF)
Supporting Information S1
(DOC)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Staphylococcus aureus is an aggressive and versatile pathogen that is responsible for a wide array of diseases ranging from pyogenic skin infections to complicated life-threatening diseases such as bacteremia, central nervous system infections, and endocarditis [ 1 , 2 , 3 , 4 ]. Treatment of S . aureus infections is a great challenge because of the ability of the organism to develop or acquire antibiotic resistance. A widespread use of methicillin and other semi-synthetic penicillins has led to the emergence of methicillin-resistant S . aureus (MRSA) strains that have become prevalent both in the hospitals and the community throughout the world [ 5 , 6 ]. Infections by MRSA strains cause higher mortality and require longer and more expensive medical care than infections caused by methicillin-sensitive S . aureus [ 5 ].
Host phagocytic cells play key roles in determining the extent of bacterial infections. The phagocytic cells induce a respiratory burst and produce superoxide anion that serves as a precursor to generate additional reactive oxygen species (ROS) such as hydrogen peroxide (H 2 O 2 ), hydroxyl radical, singlet oxygen, and hypochlorous acid. These highly reactive species lead to the oxidation of DNA, lipids and proteins. S . aureus produces antioxidant enzymes such as superoxide dismutases, catalase, alkyl hydroperoxide reductases, etc. to defend itself from the ROS [ 7 ]. However, the ROS and other oxidizing conditions still cause damage to cellular macromolecules. The ROS oxidize the sulfur atom of protein-bound methionine residues, resulting in methionine sulfoxide (MetO) that typically leads to loss of protein function. MetO are reduced back to methionine by methionine sulfoxide reductase (Msr) enzymes that restore normal protein functions [ 8 ]. Oxidation of methionine results in two diastereomeric forms of MetO, R -MetO and S -MetO, which are reduced by two different Msr enzymes. MsrB is specific for R -MetO whereas MsrA is specific for S -MetO [ 9 , 10 ].
Msr proteins have also been shown to contribute to the virulence of bacterial pathogens [ 11 , 12 , 13 , 14 , 15 ]. Absence of Msr enzymes reduces the ability of bacterial cells to adhere to eukaryotic cells that probably impacts colonization of the host [ 13 , 14 , 16 , 17 ]. In the absence of the Msr enzymes, the integrity of the bacterial surface proteins is compromised and this deficiency may contribute to the reduced bacterial adherence to eukaryotic cells [ 13 , 14 , 16 , 17 ]. In addition, reduced Msr activity impacts bacterial survival within phagocytic cells [ 12 ].
In S . aureus chromosome, there are three msrA genes ( msrA1 , msrA2 and msrA3 ) and one msrB gene [ 18 ]. The msrA1 and msrB genes are co-transcribed in S . aureus and their expression is induced specifically in response to cell wall-active antibiotics [ 19 ]. The expression of msrA1/msrB occurs at much higher levels in S . aureus relative to the expression levels of msrA2 or msrA3 genes [ 20 ].
In view of multiple msrA and msrB genes in S . aureus ; with potential roles in virulence [ 12 , 21 ] and oxidative stress tolerance [ 18 , 22 ], mutations were generated in each of the msrA and msrB genes. Subsequently, three unique msr mutants were constructed by combining the individual mutants that included an msrB mutant (lacks ability to reduce R -MetO), a triple msrA mutant ( msrA1 , msrA2 , msrA3 ; lacks ability to reduce S -MetO), and a quadruple msrAB mutant ( msrA1 , msrA2 , msrA3 , msrB ; lacks ability to reduce either R - or S -MetO). These mutants were used to determine the precise roles of Msr proteins in survival of S . aureus under a variety of stress conditions. The presented data suggest that MsrA2 and MsrA3 play little or no role in staphylococcal protection from oxidative stress or in mice. However, the role of the msrA1/msrB locus is complex. While lack of MsrA1 increases the sensitivity of S . aureus to oxidative stress and host immune defense, the lack of MsrB, to some extent, is actually beneficial to the bacterial organism under these conditions.
Materials and Methods
Ethics statement
Animal studies were approved by the A.T. Still University-Kirksville College of Osteopathic Medicine’s Animal Care and Use Committee (IACUC protocol # 166).
Bacterial strains, plasmids, antibiotics and growth conditions
The bacterial strains and plasmids used in this study are shown in Table 1 . S . aureus cells were grown in tryptic soy broth (TSB) or tryptic soy agar (TSA) and Escherichia coli cells were grown in Luria-Bertani broth or Luria-Bertani agar. Plasmids in E . coli cells were maintained by adding ampicillin at 100 μg ml -1 , kanamycin at 20 μg ml -1 , erythromycin at 15 μg ml -1 and tetracyclin at 10 μg ml -1 , when required. S . aureus mutant strains were cultured with kanamycin at 100 μg ml -1 , erythromycin at 15 μg ml -1 and tetracyclin at 10 μg ml -1 , when required.
10.1371/journal.pone.0117594.t001
Table 1 Bacterial strains used in this study.
Strains
Characteristics
Reference
S . aureus RN4220
A restriction minus derivative of S . aureus strain 8325–4
[ 54 ]
SH1000
S . aureus strain 8325–4 with functional RsbU
[ 25 ]
SH1000: msrA1
SH1000 with mutation in the msrA1 gene (Kan R )
This study
SH1000: msrA2
SH1000 with mutation in the msrA2 gene (Tet R )
This study
SH1000: msrA3
SH1000 with mutation in the msrA3 gene (Erm R )
This study
SH1000: msrA1 - B
SH1000 with mutation in the msrA1-msrB genes (Kan R )
This study
SH1000: msrB
SH1000 with mutation in the msrB gene (Kan R )
This study
SH1000: msrA
SH1000 with mutation in the msrA1 , msrA2 and msrA3 genes (Kan R , Tet R , Erm R )
This study
SH1000: msrAB
SH1000 with mutation in the msrA1 , msrA2 , msrA3 , and msrB genes (Kan R , Tet R , Erm R )
This study
BB270
A homogeneous methicillin resistant S . aureus
[ 26 ]
BB270: msrA1
BB270 with mutation in the msrA1 gene (Kan R )
This study
BB270: msrA2
BB270 with mutation in the msrA2 gene (Tet R )
This study
BB270: msrA3
BB270 with mutation in the msrA3 gene (Erm R )
This study
BB270: msrA1 - B
BB270 with mutation in the msrA1-msrB genes (Kan R )
This study
BB270: msrB
BB270 with mutation in the msrB gene (Kan R )
This study
BB270: msrA
BB270 with mutation in the msrA1 , msrA2 and msrA3 genes (Kan R , Tet R , Erm R )
This study
BB270: msrAB
BB270 with mutation in the msrA1 , msrA2 , msrA3 , and msrB genes (Kan R , Tet R , Erm R )
This study
SH1000+pCU1
SH1000 with plasmid pCU1 (Cam R )
This study
SH1000: msrA +pCU1
SH1000: msrA with plasmid pCU1 (Kan R , Tet R , Erm R , Cam R )
This study
SH1000: msrAB +pCU1
SH1000: msrAB with pCU1 (Kan R , Tet R , Erm R , Cam R )
This study
SH1000: msrA + msrA1
SH1000: msrA with pCU1- msrA1P - msrA1 (Kan R , Tet R , Erm R , Cam R )
This study
SH1000: msrAB + msrA1
SH1000: msrAB with pCU1- msrA1P - msrA1 (Kan R , Tet R , Erm R , Cam R )
This study
SH1000: msrAB + msrB
SH1000: msrAB with pCU1- msrA1P - msrB (Kan R , Tet R , Erm R , Cam R )
This study
Erm R , erythromycin resistant; Kan R , kanamycin resistant; Tet R , tetracycline resistant; Cam R , chloramphenicol resistant
DNA manipulations
Plasmid DNA was isolated using the Qiaprep Miniprep kit (Qiagen Inc). Chromosomal DNA was isolated using a DNAzol kit (Molecular Research Center) from lysostaphin-treated S . aureus cells according to the manufacturer’s instructions. All restriction and modification enzymes were purchased from Promega. PCR was performed using a Peltier Thermal Cycler-200 system (MJ research). DNA manipulations were carried out using standard procedures. Oligonucleotide primers ( Table 2 ) were obtained from Eurofins.
10.1371/journal.pone.0117594.t002
Table 2 Oligonucleotide primers used in this study.
Oligo
Sequence (5’→3’)
P1
ATCAATTACCTTGGCACCTACC
P2
GGATCCTGACTTGATGCCTGGATATG
P3
GGATCCAACTGAAGGAGAAGTTGTG
P4
AAGCTTGGTCTTGATTGCTTGTTGC
P5
GGATCCTGACACATTCAGCATAACCA
P6
AAGCTTCAGATGCACATTCATGTGA
P7
GCTGCTTACAAACATTTCGA
P8
GGATCCGAACGACGTAAAGACAGAGA
P9
GCTAACGTCATTGAATATG
P10
GGAAGTAACCTCTGGATCA
P11
ATCGTACTAAGGTCTAATG
P12
CTTGGTGATAGTCTTCGGCT
P13
ATGGTAGTTGTTTATGTAG
P14
CTCCTCTGAAAATCACTTGT
P15
GTTACACAAGAAAACGGCA
P16
TCATCATCGTGTTTTGGG
P17
AGGATGTTTCTGGTGCATGG
P18
GACACAACTTCTCCTTCAGT
Construction of msr mutants in S . aureus
Construction of the msrA1 [ 22 ], msrA2 [ 18 ], and msrB [ 23 ] mutants has been described previously.
To construct a mutation in msrA1 and msrB genes simultaneously, flanking regions (left of msrA1 and right of msrB ) were PCR amplified and ligated. Briefly, primer pairs P1 and P2 were used to amplify a 1449 bp DNA fragment (starting 1364 nt upstream of the msrA1 start codon and going downstream). Another set of primers P3 and P4 were used to amplify an 841 bp DNA fragment (starting 156 nt downstream of the msrB stop codon and going further downstream). These two fragments were ligated in vector pTZ18R [ 24 ] which simultaneously engineered a unique Bam HI site between the ligated fragments to which a 1.7 kb kanamycin-resistance cassette was cloned. This fragment was used to construct a deletion mutant ( msrA1 - msrB ) in S . aureus utilizing the methodology described previously for the construction of individual msrA1 and msrA2 mutants [ 18 , 22 ].
To construct an msrA3 mutant, primers P5 and P6 were used to amplify a 1084 bp DNA fragment upstream of msrA3 (containing 151 nt of the 5’-end of the msrA3 gene and going upstream). Another set of primers, P7 and P8, were used to amplify a 1047 bp msrA3 downstream fragment (containing 153 nt of the 3’-end of the msrA3 gene and going downstream). These two fragments were ligated together in vector pTZ18R to generate a unique Bam HI restriction site between the fragments (lacking a significant portion of the msrA3 gene, from nucleotide position 152–321 with respect to msrA3 start codon) to which a 1.4 kb erythromycin-resistance cassette was cloned. The above construct was used as a suicidal plasmid to construct a mutation in the msrA3 gene utilizing a method described previously [ 18 , 22 ].
For in vitro and in vivo studies, the S . aureus strain SH1000 [ 25 ], which is a sigB positive derivative of the S . aureus strain RN450, was used. Since most MRSA strains are naturally resistant to tetracycline and or erythromycin, a S . aureus MRSA strain BB270 [ 26 ] (sensitive to kanamycin, erythromycin and tetracycline) was used to combine msr mutations for antibiotic resistance studies. The individual msr mutants were combined in these two S . aureus strains to generate a triple (mutant of msrA1 , msrA2 , and msrA3 genes) and a quadruple mutant (mutant of msrA1 , msrA2 , msrA3 , and msrB ).
Determination of Msr activity
Msr activity in the cell free extract of the wild-type and the msr mutants of S . aureus was determined using 200 μM of Dabsyl-MetO and 20 mM DTT in 50 mM Tris-HCl (pH 7.5) and incubation at 37°C for 30 min, as previously described [ 18 ].
Growth kinetics of the wild-type S . aureus and its isogenic msr mutant under stress
Mid-exponential phase cultures (OD 600 = 0.6) were diluted 50-fold in a nephelo culture flask (Wheaton) containing 50 ml fresh TSB with a flask-to-medium volume ratio of 6:1. Oxidative and antibiotic stress conditions were imposed by the addition of H 2 O 2 and oxacillin in TSB to appropriate concentrations. Bacterial growth was subsequently monitored by incubating the flask in a shaking incubator (250 rpm) and measuring turbidity of the liquid culture.
Determination of the sensitivity of msrA mutants to oxidants and cell wall inhibitors
The minimum inhibitory concentrations (MICs) for the wild-type and different msr mutant strains of S . aureus were determined as previously described [ 27 , 28 ]. In addition to H 2 O 2 , the following oxidizing agents were used in MIC determination studies: cumene hydroperoxide that acts as an intracellular source of reactive oxygen species [ 29 ], N-ethylmaleimide that oxidizes thiols and increases disulfide bonds in proteins [ 30 ], sodium nitroprusside that serves as a nitric oxide donor [ 31 ]; methyl viologen (paraquat) that generates superoxide [ 31 ].
Determination of staphyloxanthin production in msr mutants
S . aureus wild-type and its isogenic msr mutant strains were grown at 37˚C for 18 h in TSB. Cells were harvested and washed twice with sterile water and the levels of staphyloxanthin in these cells were quantified as described previously [ 28 , 32 ].
Phagocytic killing of S . aureus msr mutant
The promyelocytic HL-60 cells (obtained from American Type Culture Collection) were grown in Iscove’s Modified Dulbecco’s Medium (IMDM) (ATCC) with 10% fetal bovine serum (Fisher) and treated with 1.3% dimethyl sulfoxide (Fisher) for 5 days to induce their differentiation into neutrophil-like cells. The differentiated neutrophils were used for phagocytic killing using a method described previously [ 33 , 34 ]. In brief, the neutrophils (1X10 6 cells) were added with S . aureus cells (2.5X10 6 CFUs) (MOI 1:2.5) in a 24-well plate. The plate was centrifuged at 4000 rpm for 10 min and incubated in a CO 2 incubator at 37˚C for 1 h. The supernatant was gently aspirated and the neutrophils were lysed by the addition of IMDM containing 0.025% Titron X-100. The number of surviving bacteria was enumerated by making serial dilutions and plating of this lysate on TSA plate.
Adherence of msr mutant to A549 lung epithelial cells
Adherence of S . aureus SH1000 strain and its isogenic msr mutants was determined by infection of lung epithelial cells as described previously [ 35 , 36 ]. In these experiments, a mixture of msr mutant and wild-type (60:40 ratio) S . aureus was used to infect the monolayers of A549 cells. The ratio of the mutants cells adhered to the A549 cells after 1 h was enumerated and compared to the ratio of the mutants in the mixture used in these adherence assays.
Complementation of triple and quadruple mutants
For complementation studies, the triple mutant was complemented in trans with msrA1 and the quadruple mutant was complemented in trans with either msrA1 or msrB gene. The msrA1 and msrB coding regions were cloned immediately downstream of a previously described construct, pCU1- msrA1 P [ 18 ]. The resulting constructs pCU1- msrA1P - msrA1 or pCU1- msrA1P - msrB was transferred into S . aureus RN4220 by electroporation and subsequently transduced into the triple or quadruple mutants. For comparative studies, wild-type SH1000, and the triple and quadruple mutants were also transformed with the empty plasmid pCU1 [ 37 ].
Levels of Protein A in S . aureus cells
Total protein was extracted from lysostaphin treated S . aureus cells, separated by SDS-PAGE, and transferred to a nitrocellulose membrane. The membrane was blocked with 5% skimmed milk and incubated with rabbit antibodies conjugated to horseradish peroxidase (Bio-rad). The membrane was visualized for Protein A using an Opti-4CN substrate kit (Bio-Rad).
Hemolysis by msr deficient S . aureus
To visualize the hemolysis, 5.0 μl of the overnight cultures of the wild-type S . aureus SH1000 and the msr mutants were spotted on TSA plates with 5% sheep blood agar and the plate was incubated at 37°C for 48 h.
Survival of wild-type and msr mutants in a murine systemic infection model
Wild-type and msr mutants were mixed together and then tested in a murine systemic infection model to determine if these mutations had an effect on the ability of the organism to survive in vivo as described previously [ 28 , 36 ]. A 0.5 ml mixture of the wild-type and msr mutant cells (~1X10 8 CFU, approximately 40:60 ratio of the wild-type and mutant) was injected into the peritoneal cavity of Swiss white Hla (ICR)CVF female mice (16–20 g) (Hilltop Lab Animals, Inc.) and the fraction of mutants surviving in the spleen and liver of infected mice was enumerated relative to wild-type S . aureus as described previously [ 28 , 36 ].
Localization of S . aureus MsrA1 and MsrB
To determine the localization of MsrA1 and MsrB in S . aureus , wild type S . aureus SH1000 culture was grown in TSB to an OD 600 = 0.3 and treated with 1.2 μg ml -1 oxacillin for 2.5 h to induce the synthesis of these proteins as described previously [ 19 , 22 ]. Bacterial cells were harvested and the cytosolic and the cytoplasmic membrane fractions were prepared as described previously [ 38 ], separated by 15% SDS-PAGE and subjected to western blot analysis for the presence of MsrA1 and MsrB.
Statistical analysis
Data were analyzed with a paired t -test using a statistical analysis computer program (R for Windows, version 3.0.2, The R Foundation for Statistical Computing). Statistical significance was set at p ≤.05.
Results
Construction of the msr mutants
We previously reported the construction and findings of the msrA1 , msrA2 , and msrB mutants where the phenotypes of the mutant strains were restored by complementation of the mutated genes in trans [ 18 , 22 , 23 ]. The msrA1 and msrB genes in S . aureus are the first and second of a four-gene operon [ 18 , 22 ]. Also, the msrA1 mutant produced a significantly higher level of MsrB relative to wild-type S . aureus [ 18 ]. In this study, a mutant was created where the entire msrA1 and msrB gene segments were deleted from the bacterial chromosome and replaced with a kanamycin-resistance cassette to generate an msrA1 - msrB null mutant. An msrA3 deletion mutant was also constructed. Subsequently, the three msrA ( msrA1 , msrA2 , msrA3 ) mutants were combined to generate an MsrA-deficient triple mutant. In addition, the msrA2 and msrA3 individual mutants were combined with an msrA1-msrB mutant to generate a quadruple mutant. These mutations were verified by PCR using primer pairs flanking the region that had been deleted in the mutants and replaced by larger antibiotic resistance cassettes ( Fig. 1 ).
10.1371/journal.pone.0117594.g001
Fig 1
Confirmation of mutations in msr genes.
Primers from the regions flanking the site of the antibiotic-resistance cassette were used in the PCR. A larger PCR product was observed when genomic DNA from the mutant (even-numbered lanes) was used compared to when wild-type genomic DNA was used (odd-numbered lanes) as a template because of the insertion of a larger antibiotic-resistance cassette. Primers P9 and P10 were used to verify mutation in msrA1 (Lanes 1 & 2), P11 and P12 for mutation in msrA2 (Lanes 3 & 4), P13 and P14 for mutation in msrA3 (Lanes 5 & 6), P15 and P16 for mutation in msrB (Lanes 7 & 8), and P17 and P18 for mutation in msrA1 - B (Lanes 9 & 10). Lane M—DNA ladder.
Msr activity in wild-type and msr mutants of S . aureus
Cell-free protein extracts from the wild-type and msr mutant cultures were used to determine Msr activity using dabsyl-MetO as a substrate. The Msr activity in various mutants was normalized against the enzymatic activity in the wild-type S . aureus SH1000 and the data are shown in Table 3 . The data demonstrate that the MsrA2 and MsrA3 contribute little to the enzymatic activity in S . aureus cells ( Table 3 ). An increase in Msr activity in the msrA1 mutant is because of a higher production of MsrB in this mutant [ 18 ]. Further, MsrB is responsible for most of the enzymatic activity (~83%) in wild-type S . aureus SH1000 ( Table 3 ). There was no enzymatic activity noted in the quadruple msrAB mutant ( Table 3 ).
10.1371/journal.pone.0117594.t003
Table 3 Methionine sulfoxide reductase activity levels in different msr mutants relative to wild-type S . aureus strain SH1000.
Strain
Percent total activity
Wild-type SH1000
100
SH1000: msrA1
218
SH1000: msrA2
106
SH1000: msrA3
93
SH1000: msrB
17
SH1000: msrA1-B
19
SH1000: msrA
123
SH1000: msrAB
0
Oxidative and antibiotic stress tolerance of the msr mutants
In growth kinetic experiments, the mutants specifically lacking MsrA1 or all three MsrA proteins showed slightly slower growth rate in TSB at 37°C ( Fig. 2 ). Deletion of msrA2 , msrA3 , msrB , or msrA1-msrB had no apparent effect on the growth of the mutant cell compared to the growth of the wild-type S . aureus SH1000 ( Fig. 2 ). When the Msr deficient mutants were cultured in TSB supplemented with 4.4 mM H 2 O 2 , the S . aureus strains lacking MsrA1 failed to grow ( Fig. 3A ). In the case of the combinatorial mutants, no growth was recorded for the triple msrA mutant even after 16 h in TSB with 8.8 mM H 2 O 2 ( Fig. 3B ). The amount of H 2 O 2 was raised to 8.8 mM in growth studies utilizing the combinatorial mutants to assess the resistance of MsrB-deficient S . aureus relative to other strains ( Fig. 3B ). The S . aureus strains that lacked MsrB ( msrB , msrA1 - msrB and the quadruple msrAB mutants) were moderately resistant to the presence of H 2 O 2 in these growth experiments ( Fig. 3B ). The MsrB-deficient strains of methicillin-resistant S . aureus BB270 demonstrated better growth even in the presence of a cell wall-active antibiotic, oxacillin ( Fig. 3C ). In the MIC studies, the S . aureus strains deficient in MsrB were more resistant to H 2 O 2 ( Table 4 ). A similar increase in resistance to oxacillin and other cell wall-active antibiotics was observed in the case of MsrB-deficient S . aureus ( Table 5 ). The strains that lacked MsrA1 were susceptible to oxidative stress conditions and the S . aureus strain that lacked all three MsrA proteins (the triple msrA mutant) showed most sensitivity to oxidants ( Table 4 ). No such increase in sensitivity was noted in MsrA-deficient S . aureus to cell-wall active antibiotics ( Table 5 ).
10.1371/journal.pone.0117594.g002
Fig 2
Growth curve of the wild-type S . aureus strain and its derivative msr mutants in TSB.
Values indicate the average of two independent experiments.
10.1371/journal.pone.0117594.g003
Fig 3
Growth of the wild-type S . aureus strain and its derivative msr mutants in the presence of H 2 O 2 or oxacillin.
(A) growth in the presence of 4.4 mM H 2 O 2 , (B) growth in the presence of 8.8 mM H 2 O 2 (C) growth in the presence of 400 μg ml -1 of oxacillin. Values indicate the average of two independent experiments.
10.1371/journal.pone.0117594.t004
Table 4 Susceptibilities of S . aureus parental strain SH1000 and its derivative msr mutants to oxidants. MIC values indicate average mM concentrations of three independent experiments.
Strains
H 2 O 2
CHPO
NEM
SNP
Paraquat
Wild-type SH1000
1
9.5
0.625
250
125
SH1000: msrA1
0.5
4.75
0.313
7.81
125
SH1000: msrA2
1
9.5
0.625
250
125
SH1000: msrA3
1
9.5
0.625
250
125
SH1000: msrB
2
9.5
0.625
250
125
SH1000: msrA1-B
0.5
9.5
0.625
125
125
SH1000: msrA
0.25
2.38
0.313
1.95
31.25
SH1000: msrAB
0.5
9.5
0.625
250
125
Abbreviations: H 2 O 2 , hydrogen peroxide; CHPO, cumene hydroperoxide; NEM, N-ethylmaleimide; SNP, sodium nitroprusside.
10.1371/journal.pone.0117594.t005
Table 5 Susceptibilities of S . aureus BB270 parental and msr mutant strains to cell wall-active antibiotics.
Strains
Oxacillin
D-cycloserine
Bacitracin
Wild-type BB270
200
75
50
BB270: msrA1
200
75
50
BB270: msrA2
200
75
50
BB270: msrA3
200
75
50
BB270: msrB
400
150
100
BB270: msrA1-B
300
100
75
BB270: msrA
200
75
50
BB270: msrAB
200
75
50
MIC values (μg ml -1 ) indicate average of three independent experiments.
Production of staphyloxanthin pigment in msr mutants
Of the seven msr mutants used in this study, production of staphyloxanthin pigment was highest in the msrB mutant strain ( Fig. 3 ). The level of staphyloxanthin was lower in MsrA1-deficient strains ( Fig. 3 ). The MsrA1-deficient S . aureus has been shown to produce a much higher level of MsrB [ 18 ]. Increased pigmentation in MsrB-deficient S . aureus and reduced pigmentation in cells producing a higher level of MsrB suggests that the MsrB protein suppresses the production of staphyloxanthin in S . aureus . Production of staphyloxanthin in msrA2 and msrA3 mutants was not affected relative to wild-type S . aureus ( Fig. 4 ).
10.1371/journal.pone.0117594.g004
Fig 4
Production of staphyloxanthin in the wild-type S . aureus strain SH1000 and its derivative msr mutants.
-The bottom panel shows the color of the bacterial cell pellet from 50 ml overnight grown cultures. The amount of the staphyloxanthin pigment produced by these cells was quantified and is shown as A 462 . Values indicate the average of three independent experiments ± standard deviation (* significant at p ≤.05).
Phagocytic killing of the S . aureus msr mutant cells
Polymorphonuclear cells utilize oxygen-dependent bactericidal pathways in the phagolysosomes. The impact of Msr deletion was investigated on staphylococcal survival in differentiated polymorphonuclear cells. In these studies, the S . aureus strains with a non-functional MsrA1 showed increased susceptibility to the polymorphonuclear cells ( Fig. 5 ). The survival of the msrA2 , msrA3 , msrB or msrA1-B mutants of S . aureus was comparable to the wild-type S . aureus SH1000 in these assays ( Fig. 5 ).
10.1371/journal.pone.0117594.g005
Fig 5
Survival of the wild-type S . aureus strain SH1000 and its derivative msr mutant cells exposed to polymorphonuclear (PMN) cells.
PMN cells were infected (MOI 1:2.5) with wild-type S . aureus SH1000 and its isogenic msr mutants for 1 h at 37˚C and then plated on TSA for enumeration. Values indicate the average of three independent experiments ± standard deviation (* significant at p ≤.05).
Role of Msr proteins in adherence of S . aureus to lung epithelial cells
The mixture that was used in adherence assays was biased for an msr mutant (~60%) relative to the wild-type S . aureus SH1000 (~40%). In experiments investigating the adherence of this mixture to A549 cells, the MsrA1-deficient mutants ( msrA1 , msrA and msrAB ) showed significantly reduced adherence ( Fig. 6 ). Deficiency of MsrA2, MsrA3, or MsrB did not impact the adherence of the S . aureus cells to A549 cells ( Fig. 6 ).
10.1371/journal.pone.0117594.g006
Fig 6
Adherence of the wild-type S . aureus strain SH1000 and its derivative msr mutant cells to A549 human lung epithelial cells.
A total of 5X10 5 bacterial cells were used in these assays. The left light bar in each panel represents the ratios of the msr mutant relative to wild-type SH1000 in the mixture used to infect the A549 cells. The right dark bar in each panel represents the ratios of the msr mutant in the mixture that adhered to the A549 cells after 1 h of incubation. Values indicate the average of three independent experiments ± standard deviation (* significant at p ≤.05).
Protein A levels in msr mutants
Staphylococcal surface protein, Protein A, contributes to bacterial adhesion, virulence, and biofilm formation. In Western blot analysis involving total protein extract from wild-type S . aureus SH1000 and the derivative msr mutant cells, an apparent 55 kDa protein specific to Protein A was detected ( Fig. 7 ). Individual msr gene deletions had no appreciable impact on the levels of Protein A in S . aureus ( Fig. 7 ). However, the protein A-specific band was significantly lighter in the lane corresponding to the triple msrA mutant ( Fig. 7 , Lane 7).
10.1371/journal.pone.0117594.g007
Fig 7
Western analysis of the levels of staphylococcal Protein A in wild-type S . aureus strain SH1000 and its derivative msr mutants.
Top panel shows the total protein profile of wild-type and the msr mutant strains after SDS-PAGE suggesting that similar amounts of protein were used in the analysis of Protein A. The bottom panel shows the reactivity of Protein A (arrow) in each lane.
Hemolytic pattern of msr mutants
In qualitative assays, the S . aureus strain that lacked all three MsrA proteins showed a relatively smaller zone of beta-hemolysis relative to other strains ( Fig. 8 , Spot 7). Another interesting observation was the presence of a significantly reduced secondary zone of hemolysis for the triple mutant ( Fig. 8 , Spot 7).
10.1371/journal.pone.0117594.g008
Fig 8
Hemolytic pattern of the wild-type S . aureus strain SH1000 and its derivative msr mutants after 48 h of growth on 5% sheep blood agar plates.
(1) Wild-type SH1000, (2) SH1000: msrA1 , (3) SH1000: msrA2 , (4) SH1000: msrA3 , (5) SH1000: msrB , (6) SH1000: msrA1 - B , (7) SH1000: msrA , (8) SH1000: msrAB .
Hemolysis, phagocytic survival and adherence of complemented triple and quadruple mutants
The triple SH1000: msrA mutant showed a defective pattern in hemolysis but its complementation with the msrA1 gene in trans was shown to restore the level of hemolysis shown with the wild-type SH1000 ( Fig. 9A , Spot 3). In phagocytic killing assays, the triple SH1000: msrA and the quadruple SH1000: msrAB mutants were more sensitive than the wild-type SH1000. In complementation experiments, when triple and quadruple mutants were complemented with the msrA1 gene in trans , these strains showed phagocytic resistance that was comparable to wild-type SH1000 ( Fig. 9B ). However, complementation of the quadruple mutant with the msrB gene in trans did not restore the phagocytic resistance in these strains ( Fig. 9B ). Similarly, in adherence experiments, complementation with msrA1 gene in trans , restored the defect in adherence that was initially seen in case of the triple or quadruple mutants ( Fig. 9C ). Complementation with msrB , on the other hand, had no appreciable effect on the adherence of the quadruple effect ( Fig. 9C ).
10.1371/journal.pone.0117594.g009
Fig 9
Hemolysis (A), phagocytic survival (B) and adherence (C) of complemented triple and quadruple mutants.
A. Hemolysis on 5% sheep blood agar plates. (1) SH1000+pCU1, (2) SH1000: msrA +pCU1, (3) SH1000: msrA + msrA1 . B. PMN cells were infected (MOI 1:2.5) with wild-type, mutants, and the complemented strains for 1 h at 37˚C and then plated for enumeration. (1) SH1000+pCU1, (2) SH1000: msrA +pCU1, (3) SH1000: msrAB +pCU1, (4) SH1000: msrA + msrA1 , (5) SH1000: msrAB + msrA1 , (6) SH1000: msrAB + msrB . Values indicate the average of three independent experiments ± standard deviation (* significant at p ≤.05). C. The left light bar in each panel represents the ratios of the mutant or the complemented strain relative to SH1000-pCU1 in the mixture used to infect the A549 cells. The right dark bar in each panel represents the ratios of the mutant or the complemented strain in the mixture that adhered after 1 h of incubation. (1) SH1000: msrA +pCU1, (2) SH1000: msrAB +pCU1, (3) SH1000: msrA + msrA1 , (4) SH1000: msrAB + msrA1 , (5) SH1000: msrAB + msrB . Values indicate the average of three independent experiments ± standard deviation (* significant at p ≤.05).
Survival of msr mutants in mice
To elucidate the role of Msr in virulence of S . aureus , Swiss white female mice were injected with a bacterial mixture of wild-type S . aureus SH1000 and its derivative seven msr mutants (40:60 ratio of wild-type to mutant). The data suggest that the msrA1 mutant of S . aureus had a lower survival rate in mice. Post infection, the fraction of msrA1 mutants in spleen and liver was lower at 8 h and declined even further at 24 h in these tissues relative to their fraction in the mixture that was injected into the mice ( Fig. 10 ). Loss of MsrA2, MsrA3, or MsrB had little to no effect on the survival of S . aureus in mice ( Fig. 10 ). The triple msrA mutant showed the highest decline in spleen and liver tissues with time suggesting some roles for MsrA2 and MsrA3 under MsrA1-deficient conditions ( Fig. 10 ). Although, there is a slight growth defect in the msrA1 and triple msrA mutants as shown in Fig. 2 , when cultured at 37°C in vitro , it is highly unlikely that there was much of a growth of the wild-type or the mutant bacteria in mice during our experiments that lasted only 24 h. Most of the bacteria that were injected were cleared in mice with time, as we recovered fewer bacteria after 8 h and far fewer bacteria after 24 h. It is indeed the lack of MsrA1 that significantly reduced the survival of S . aureus in mice.
10.1371/journal.pone.0117594.g010
Fig 10
Survival of the wild-type S . aureus strain SH1000 and its derivative msr mutant cells in mouse.
Approximately 1.0X10 8 CFUs (predominantly mutant cells) were injected intra-peritoneally into mice. Three mice were sacrificed at 8 and 24 h post-injection. Closed circles represent the fraction of wild-type SH1000 and the open circles represent the msr mutant bacteria recovered from murine spleens (A) and murine livers (B) at 8 and 24 h post infection. Values at 0 h indicate the fraction of msr mutants and isogenic wild-type cells in the injected inoculum. Values indicate the average of three independent experiments ± standard deviation (* significant at p ≤.05).
Localization of msr protein
Localization was only investigated for MsrA1 and MsrB proteins because these two proteins have been shown to be expressed in S . aureus at a significantly higher level relative to MsrA2 and MsrA3 in S . aureus [ 20 ]. In addition, findings of this study suggest that the lack of MsrA1 or MsrB has a pleiotropic effect on S . aureus cells. Experiments utilizing anti-MsrA1 and anti-MsrB rabbit polyclonal antibodies demonstrated that the MsrA1 protein is distributed equally between the cytosolic and the membrane components in S . aureus ( Fig. 11 , Lanes 1 and 2). However, the MsrB protein appears to be predominantly a cytosolic protein and only a minor fraction of this protein is targeted into the bacterial membrane ( Fig. 11 , Lanes 3 and 4).
10.1371/journal.pone.0117594.g011
Fig 11
Distribution of MsrA1 and MsrB in the cytosolic and membrane fractions in S . aureus SH1000.
Lane 1 indicates the cytosolic fraction and Lane 2 indicates the membrane fraction.
Discussion
Numerous investigations in recent years have led to an increased interest and understanding of the biology of the methionine sulfoxide reductases. The reasons underlying this interest are because of a remarkable conservation across prokaryotes and eukaryotes, the importance in oxidative stress, and the novel protein repair functions of these enzymes. The two distinct Msr proteins, MsrA and MsrB, share no sequence homology. Orthologs of msrA and msrB show great variation in their genetic organization in bacterial chromosomes. In some bacterial species, the genes encoding MsrA and MsrB are located adjacent to each other and co-transcribed, and in others, the msrA and msrB genes are transcriptionally fused [ 17 , 39 ]. In addition, many bacterial species have multiple copies of these msrA and msrB genes distributed randomly in the bacterial chromosome and some are present even on plasmids [ 39 ].
S . aureus produces three different MsrA proteins (MsrA1, MsrA2 and MsrA3) and one MsrB protein. MsrA1 and MsrB production in S . aureus are induced by cell wall-active antibiotics. In the presence of these antibiotics, the cell wall is likely destabilized and the oxidizing agents have easy access to bacterial membrane and cytosolic compartments. In response, the staphylococcal cells produce a higher level of MsrA1 and MsrB; however, oxidative stress has not been shown to induce the synthesis of these proteins in S . aureus . In addition to these four Msr proteins, there is an additional gene ( fRMsr ) in S . aureus (SACOL1768 in S . aureus strain COL) that codes for a protein that reduces the free methionine sulfoxide. Although the structural and biochemical properties of this protein have been determined [ 40 ], its physiological relevance is unclear. The extent of expression of S . aureus fRMsr is also not clear. The fRMsr gene in S . aureus may be expressed at a very low level since there was no detectable Msr activity in the msrAB quadruple mutant ( Table 3 ).
Studies with the individual msr gene mutants make it clear that the MsrA2 and MsrA3 contribute little to cellular Msr activities, play a little to no role in protecting S . aureus from oxidative stress and neutrophils, and have no impact on bacterial survival in mice. Using promoter fusion experiments, we have previously shown that msrA2 and msrA3 are expressed at significantly lower levels compared to the expression of the msrA1 - msrB locus in S . aureus [ 20 ]. We also measured the relative transcript levels of the msrA2 and msrA3 relative to the transcript level of msrA1 in S . aureus . The expression level of msrA2 was 8–10 log lower compared to msrA1 , and the msrA3 -specific transcript was almost absent in a qRT-PCR assay (data not shown). In a previous study, with promoter reshuffling, we showed that the MsrA2 protein was as effective as MsrA1 in protection from oxidative stress when its expression level was raised in S . aureus [ 18 ]. The msrA3 gene may also be under the influence of a weaker promoter compared to the strength of the promoter that drives the transcription of the msrA1 - msrB genes in S . aureus .
In contrast to MsrA2 and MsrA3, lack of either MsrA1 or MsrB showed pleiotropic effects in S . aureus . The lack of MsrA1 increased the sensitivity of S . aureus to oxidative stress. Studies with a triple msrA mutant, which lacked all three MsrA proteins and therefore had no apparent capability to reduce S -MetO, showed a further increase in bacterial sensitivity to oxidants compared to only MsrA1-deficient S . aureus . This phenomenon suggests that, even though MsrA2 and MsrA3 are present at very low levels in S . aureus , they may be somewhat relevant in protecting S . aureus under MsrA1-deficient conditions. The triple msrA mutant also showed reduced hemolysis and increased susceptibility to neutrophil-mediated killing. This observation was expected given that MsrA deficiency in several organisms leads to enhanced vulnerability to oxidative stress [ 18 , 41 , 42 , 43 ]. In addition, the msrA1 gene was up-regulated in neutrophils [ 12 ]. Within the neutrophils, the staphylococcal two-component regulatory system VraSR contributes to the msrA1 up-regulation [ 12 ].
The MsrA1-deficient strains showed reduced pigmentation compared to the wild-type S . aureus . It has been previously shown that the staphyloxanthin pigment plays an important role in the protection of S . aureus from oxidants and neutrophils and regulates bacterial membrane fluidity and virulence [ 44 , 45 , 46 , 47 ]. In this study, the MsrB-deficient S . aureus strains were more pigmented and more resistant to H 2 O 2 and cell wall-active antibiotics. One possible explanation for this phenomenon is that an increased pigmentation in the MsrB-deficient S . aureus may contribute to an impermeable membrane that restricts the oxidants and antibiotics. In turn, this change may minimize damage to cellular components under these adverse conditions. We also noted that the MsrA1-deficient S . aureus or S . aureus that was deficient in all three MsrA proteins were less adherent to human lung epithelial cells and showed reduced survival in mouse spleen and liver. The quadruple msrAB mutant of S . aureus also showed reduced adherence to A549 cells and survival in mouse tissues. Furthermore, the complementation experiments with the triple and quadruple mutants provide evidence that it is the MsrA1 not MsrB that is critical for staphylococcal adherence to eukaryotic cells and its resistance to the killing by phagocytic cells.
With respect to the role of the Msr proteins, it is well documented that these enzymes contribute to the ability of a pathogen to adhere to host tissue, evade immune system, form biofilms, survive inside macrophages, and resist oxidative killing [ 14 ].MsrA protein contributes to cell wall integrity and maintenance of adhesion properties in Streptococcus gordonii [ 48 ]. Msr proteins have also been shown to affect adherence properties of pathogenic Neisseria [ 17 ]. In S . gordonii , the MsrA enzyme was shown to maintain the integrity of bacterial adhesins during oxidative stress [ 49 ]. The current study confirms the role of Msr proteins, particularly the MsrAs in the adherence of S . aureus to human cells. The MsrA1-deficient S . aureus , the triple msrA and the quadruple msrAB null-mutants, all showed reduced adherence to lung epithelial cells. The role of Msr proteins in virulence of the bacterial pathogens is also well documented. Both MsrA and MsrB contributed to the enzymatic defenses of Mycobacterium tuberculosis from reactive oxygen species [ 50 ]. In Pseudomonas aeruginosa , inactivation of either msrA or msrB or both reduced virulence and increased its killing by oxidants [ 51 ]. In Campylobacter jejuni , the single msrA or msrB mutants showed no growth defect, but the msrA - msrB double mutant showed increased sensitivity to oxidative stress conditions [ 31 ]. Mutation in the msrA or msrB gene in Enterococcus faecalis resulted in increased sensitivity to H 2 O 2 . In addition, an msrA msrB double mutant showed further increase in sensitivity suggesting that the effect of mutations were additive [ 15 ]. In a later study, however, the msrA and msrB mutants were shown to behave differently; the msrA mutant was more sensitive to oxidative stress conditions whereas the msrB mutant showed stimulated growth under similar conditions [ 52 ]. In Salmonella Typhimurium, deletion of msrA increased bacterial susceptibility to H 2 O 2 and reduced its virulence, but a mutation in msrB had no apparent phenotype [ 11 ]. In Mycobacterium smegmatis also, MsrB was shown to have a limited role in protection from oxidative stress conditions [ 53 ].
Thus, the role of MsrB protein in defense from oxidative stress is questionable in many bacterial species. It is possible that under oxidative stress the majority of the oxidized methionine is S -MetO and the MsrB protein has no activity against this epimer. This may be the reason why the MsrA-deficient bacteria showed a high sensitivity to conditions that impose oxidative stress. MsrB of S . aureus , seems to some extent, counterbalance the effect of MsrA1. For example, lack of MsrA1 reduces pigmentation and this may be due to previously shown higher level of MsrB in MsrA1-deficient S . aureus [ 18 ]. However, when MsrB is absent, the bacterium responds by increasing pigment production as a potential compensatory mechanism.
In summary, among the four Msr enzymes produced in S . aureus , MsrA2 and MsrA3 contribute little to the enzymatic activity and bacterial defense from oxidative stress. MsrA1 and MsrB have opposing roles in pigment production and resistance from oxidative stress. MsrA1 seems to be equally distributed between the cytosolic and membrane components but the MsrB appears to be predominantly cytosolic. Regulation of msrA1 - msrB locus is currently under investigation because of its significant role in S . aureus physiology and virulence.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Breast cancer is still the most prevalent cancer type in women, accounting for 29% of all cancer cases in women in 2013. In addition it is the second leading cause of cancer related death in women [1] . Adjuvant treatment decisions are based on various national and international guidelines and tools [2] , [3] , [4] .
In addition to the known relevant clinicopathological prognostic factors like e.g. large tumor size (>2 cm), lymph node metastasis or HER2 gene amplification [5] , [6] , genomic multigene assays can be used as additional tools to assist treatment decisions and to avoid under- or overtreatment by estimation of the biological tumor behavior [7] , [8] . These multigene signatures were discussed on the 13 th St Gallen International Breast Cancer Conference 2013 [9] . In this context, especially the identification of patients with ER-positive HER2-negative breast cancer with intermediate or high risk of recurrence defined by conventional clinicopathological features but low risk defined by multigene assays seems to be important for therapy decisions regarding chemotherapy and endocrine therapy. Furthermore, selected prognostic gene expression arrays that can be used in daily practice are listed in the WHO 2012 blue book [10] .
The most widely used assays in this regard are OncotypeDX® (Genomic Health, Inc., Redwood City, CA, USA) and Mammaprint® (Agendia BV, Amsterdam, The Netherlands) [11] , [12] .
The validated EndoPredict assay (EP) is a novel tool to predict the risk of metastases of patients with estrogen receptor positive (ER positive), HER2 negative breast cancer treated with endocrine therapy alone. Both risk scores were validated in two large independent clinical trials (ABCSG-6: n = 378, ABCSG-8: n = 1324) [13] . This assay can be performed on formalin-fixed paraffin-embedded tissue [13] . It provides additional prognostic information to standard pathological factors including ki67 and improves risk classification from common clinical guidelines [14] . In a recent review [15] , the EndoPredict assay has been assigned the level of evidence 1 according to Simon et al. [16] , this level of evidence is identical e.g. to the Oncotype DX® recurrence score.
The EndoPredict assay has been designed to integrate genomic and clinical information and therefore includes clinico-pathological factors such as tumor size and nodal status. From the view of diagnostic molecular pathology, the EndoPredict assay is an example of a new generation of molecular assays, since it is the first multigene expression assay which is very suitable for decentralized testing in specialized molecular pathological laboratories as shown by a round-robin trial [17] . Analytical performance characteristics and the robustness of the test in a molecular-pathological laboratory has been published [18] .
The test has been introduced in Germany as a new diagnostic tool in August 2011. The Charité University Hospital has been the first diagnostic molecular pathology laboratory that established the EndoPredict assay in routine diagnostic and has performed a large series of EP assays during the first year.
In this project, we investigated the performance of this test in clinical practice, and performed a retrospective evaluation of the impact of this new test on treatment decisions in breast cancer.
Methods
Study Population and Clinicopathological Parameters
Within one year (August 2011– July 2012), we received a total of 168 diagnostic requests to conduct the EndoPredict assay at the Institute of Pathology at the Charité University Hospital in Berlin, Germany. EndoPredict assays from 167 patients could be successfully performed, for one single sample the RNA extraction was not possible. The formalin-fixed paraffin-embedded (FFPE) tissue samples derived from female patients with primary invasive estrogen receptor (ER) positive, HER2 negative breast cancer. The median age at time of diagnosis was 54 years (range: 30–78 years), the median age in the subgroup with therapy data was 55 years (range: 30–76 years). The clinicopathological data (tumor size, pT, nodal status, grading, Ki67) were extracted from the pathological reports. Concerning Ki67, the cutoff point was extracted from the St. Gallen guidelines [2] . Table 1 gives an overview on these factors of the patients. For retrospective evaluation of treatment decisions, a questionnaire was sent to the clinical partner. The questionnaire consisted of two questions each with two possible answers regarding the treatment decisions (endocrine therapy alone vs. endocrine therapy together with chemotherapy) before and after the EndoPredict – test.
10.1371/journal.pone.0068252.t001 Table 1
Patients characteristics.
Characteristic
number of all patients
%
subgroup with therapy data
%
All
167
100
130
100
Tumor size (mm)
pT1a
1
0.6
1
0.8
pT1b
17
10.2
15
11.5
pT1c
68
40.7
51
39.2
pT2
67
40.1
53
40.8
pT3
14
8.4
10
7.7
Nodal status *
pN0
103
62.1
81
62.3
pN1
59
35.5
47
36.1
pN2a
2
1.2
1
0.8
pN3a
2
1.2
1
0.8
Histological grade *
G1
18
11.3
15
12.2
G2
113
71.1
89
72.4
G3
28
17.6
19
15.4
Hospital
intern
65
38.9
46
35.4
extern
102
61.1
84
64.6
Tumor proliferation ∗
Ki67<14%
67
54.9
50
52.6
Ki671≥14%
55
45.1
45
47.4
* = Not all of the data were available for all patients.
Ethical Statement
For this study, only existing data from routine diagnostic procedures were used, which were performed with informed consent as part of routine patient care. No additional tissue-based evaluations were performed. Therefore, no ethics committee approval was needed, based on the legal requirements in Berlin (Landeskrankenhausgesetz § 25.1, version 18.09.2011) which allow evaluation of existing diagnostic data.
RNA Extraction and Assessment of the EndoPredict Score
The invasive cancer was verified on a hematoxilin-eosin-stained slide before RNA was extracted, in case of a tumor area <30%, a macrodissection was done before RNA extraction. Usually, one 5 µm slide was used. In case of a lower tumor content, more than one 5 µm slide was used. The extraction of total RNA was carried out using a fully automatic method as described previously [19] , [20] , [21] , alternatively manually according to the same protocol. The EndoPredict assay analyzes the expression levels of eight genes of interest (BIRC5, UBE2C, DHCR7, RBBP8, IL6ST, AZGP1, MGP and STC2) as well as three reference genes [13] . PCR was performed as described before [17] . R elative expression levels of each gene of interest as well as EP and EPclin scores were calculated as described previously 13 using a web-based implementation (http://forschung.medizin.uni-mainz.de/epreport/) to process analytical PCR results into test results. Referring to this, the EPclin score combines the EP score with tumor size and nodal status resulting in a molecular-clinicopathological hybrid score . Finally, samples were classified as low or high risk of distant metastasis according to the predefined cutoff value of 5 (molecular risk score EP) respectively 3.3 (integrated molecular and clinical risk score EPclin) [13] .
Statistical Methods
The statistical analysis was done using SPSS Statistics Version 18 (IBM, Armonk, USA). The correlation between EP score and tumor grade was analyzed using the Jonckheere-Terpstra test for trends. The correlation between EP and proliferation activity was analyzed using the Wilcoxon-Mann-Whitney test. The graphics were generated with GraphPad Prism 5 (GraphPad Software, Inc., La Jolla, CA, USA).
Results
Test Performance and Distribution of Risk Groups
During one year, we routinely performed EndoPredict tests from a total of 167 patients in our molecular pathology laboratory.
Regarding the molecular EP class, samples from 56 patients (33.5%) had a low-risk, whereas 111 patients (66.5%) showed a high-risk gene profile ( Fig. 1A ). After integration of the clinicopathological factors (tumor size, nodal status) the combined clinical and molecular score (EPclin) resulted in a low-risk group of 77 patients (46.4%), while 89 (53.6%) had a high-risk EPclin score. For one patient, the combined clinical and molecular score (EPclin) could not be analyzed due to the unknown nodal status. The EPclin-based estimated median 10-year-risk for metastases with endocrine therapy alone was 11% for the whole cohort. The estimated median risk for the EPclin low group was 7%, for the EPclin high risk group it was 19%.
10.1371/journal.pone.0068252.g001 Figure 1
Distribution of EP class and EP clin class.
Distribution of EP class and EP clin class of all included EndoPredict assays (A). Distribution of EP class and EP clin class of a subgroup of patient for which therapy decision data were available (B).
The median handling time averaged three days (range: 0 to 11 days). 59.3% of the test could be performed in three or less than three days. The reasons for delay included missing clinical information as well as the need for repeated RNA extraction (8 cases). For one case, an additional paraffin block had to be requested from the local pathologist ( Fig. 2 ).
10.1371/journal.pone.0068252.g002 Figure 2
Test performance of the EndoPredict assay during one year.
167 samples could successfully analysed during one year.
Comparison of EndoPredict with Standard Clinical Parameters
The median of the EP score increased from 4.5 to 5.9 and 8.1 for tumors with histopathological grade G1, G2 and G3 (p = 7.6E-06). Further, the EP score had a higher median of 6.9 in tumors with high Ki67-index compared to 4.9 in slowly proliferating tumors (p = 1.5E-06). Figure 3 gives an overview of the distribution of the molecular risk score EP depending on the histological grade as well on the Ki67-index.
10.1371/journal.pone.0068252.g003 Figure 3
Distribution of the molecular risk score EP related to the histological grade and mitotic index.
Distribution of the molecular risk score EP related to the histological grade (A) as well as to the mitotic index (B). The continuous line revealed the median, the dotted line highlighted the cutoff point of the molecular risk score EP. The cutoff point of ki67 was extracted from the St. Gallen guidelines [2] .
Evaluation of the Impact of EndoPredict on Changes in Therapy Decisions
The information on treatment decisions was retrospectively collected using a standardized questionnaire. This therapy information was available from 130 (77.8%) of the 167 patients. Of those patients, 62 (47.7%) had a low risk combined clinical and molecular score (EPclin), the remaining 68 patients (52.3%) were EPclin high risk ( Fig. 1B ).
Before the EndoPredict assay, 47 patients (36.2%) had been scheduled for endocrine therapy alone. In contrary, for 83 patients (63.8%) a combination of endocrine therapy and chemotherapy had been planned. After the results of the EndoPredict assay were available, the number of patients with endocrine therapy alone was increased to 67 (51.5%) and only 62 patients (47.7%) were scheduled for a combination therapy of chemotherapy and endocrine therapy. For one patient the therapy decision after the EndoPredict assay was unknown because of the prior desire of the patient for another therapy.
Comparison of pre- and post-test therapy decisions showed a change of therapy in 37.7% of patients. In detail, for 16 patients (12.3%) it was decided to administer an additional chemotherapy based on the results of the EndoPredict assay. On the other hand, the therapy of 25.4% of patients (n = 33) was reduced to endocrine therapy alone. In 73 patients (56.2%) no change of therapy resulted from the EndoPredict result. Additionally, 8 patients did not agree to the recommendation of the tumor board after the EndoPredict assay. Figure 4 gives an overview about all changes in therapy decisions. Figure 5 depicts the changes of therapy decisions depending on the molecular risk score EP and the combined clinical and molecular score (EPclin).
10.1371/journal.pone.0068252.g004 Figure 4
Changes in therapy decisions.
Changes in therapy decisions regarding the decision before and after the EndoPredict assay.
10.1371/journal.pone.0068252.g005 Figure 5
Therapy decision related to the molecular risk score EP and the combined clinical and molecular score (EPclin).
Association between the molecular risk score EP, the combined clinical and molecular score (EPclin) and therapy decision (A). The group of patient’s desire for other therapy was excluded. The dotted vertical line marks the cutoff values of the molecular risk score EP, the broken horizontal line marks the cutoff value of the combined clinical and molecular score (EPclin). Additionally, the therapy decisions related to the combined clinical and molecular score (EPclin) are shown (B). The broken horizontal line marks the cutoff value of the combined clinical and molecular score (EPclin), the continuous line indicates the median.
Discussion
Our study demonstrates that the EndoPredict assay can be reliably performed in a routine molecular pathology laboratory in daily practice. The test could be successfully performed in 99% of all samples. More than 50% of tests were performed in three or less days.
This is the first analysis of changes in therapy decisions based on the EndoPredict assay. In over one third (37.7%) the results of the EndoPredict assay lead to a change of planned therapy. For a quarter of patients (25.4%) the originally planned chemotherapy could be omitted based on the result of the multi-gene assay.
Our results are in line with those observed for the prognostic Oncotype® DX 21-gene assay [22] , [23] . Similar to the current study, they observed a change of treatment recommendations in about thirty percent (31.5% resp. 32%). Comparable with our results, most of the changes were caused by reduction from chemotherapy plus endocrine to endocrine therapy alone (22.5% resp. 21%). A change from endocrine to chemo-endocrine therapy was observed in 3.4% [22] to 11% [23] . Therefore, despite the limitation of our study that therapy changes were only retrospectively assessed, the results are comparable with other reports. Recently, it was shown by Blohmer et al. according to other published studies that the treatment decision using the OncotypeDX® on adjuvant therapy leads to a reduction of costs as compared to costs without this molecular test [24] . The EndoPredict assay can be successfully done in decentral molecular pathology laboratories [17] with a median handling time of three days.
Furthermore, the results of the OncotypeDX® are divided in three groups: low Recurrence Score (RS), intermediate RS and high RS, whereas the EndoPredict results in only two groups (low risk, high risk). In contrast to most of the other multigene assays, the EndoPredict assay includes the relevant clinicopathological factors tumor size and nodal status which are known to be essential for assessing the biological behavior of breast cancer. The interpretation of the test results was in most cases straightforward, test results near the cutoff point were intensively discussed with the clinicians and the patients. From the point of view of a clinician having to decide whether to recommend additional adjuvant chemotherapy to a patient or not, this dichotomization of risk is helpful.
Recently, it was shown by Blohmer et al. according to other published studies that the treatment decision using the OncotypeDX® on adjuvant therapy leads to a reduction of as compared to costs without this molecular test [24] .
One limitation of our study is the retrospective assessment of the therapy changes. In addition, the EndoPredict test was a completely new test and the clinicians did not have any experience with the test at the time our analysis started. Therefore in the beginning the clinicians were not prepared to change therapy based on the test, which might underestimate the changes in therapy once the test is fully integrated in the diagnostic workup.
As a conclusion, our results show that the EndoPredict assay could be routinely performed in a decentral molecular pathology laboratory. The results of this multi- gene assay markedly change treatment decisions supporting clinical utility of this new diagnostic method. Based on the comprehensive clinical and analytical validation data and our results from clinical routine diagnostics, we suggest that implementation of this test could be very helpful as an additional tool for treatment decisions in breast cancer patients in clinical practice.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Scientific research can be a cutthroat business, with undue pressure to publish quickly, first, and frequently. The resulting race to publish ahead of competitors is intense and to the detriment of the scientific endeavor. Just as summiting Everest second is still an incredible achievement, so too, we believe, is the scientific research resulting from a group who have (perhaps inadvertently) replicated the important findings of another group. To recognize this, we are formalizing a policy whereby manuscripts that confirm or extend a recently published study (“scooped” manuscripts, also referred to as complementary) are eligible for consideration at PLOS Biology .
Being scooped is loosely defined as when two independent groups studying the same system produce the same or similar results, and one group publishes their work first. Being scooped is often considered to devalue the second, complementary study; many journals will reject it citing lack of novelty. However, there is a self-evident benefit to publishing complementary research, and at PLOS Biology , we consider that two papers from two groups independently identifying the same phenomenon in parallel increase the confidence in the results of the work.
This new policy, acknowledging the value of complementary studies, therefore addresses the current concern regarding the reproducibility, or lack thereof, of scientific findings. Currently, the gold standard for demonstrating that an article is based on solid results is a replication study. These studies are generally conducted after publication and are considered critically important for supporting and advancing scientific theories. We argue that the “organic” replication of a complementary study is even better than a post-hoc and often costly replication study for supporting conclusions. There are other efforts underway to improve reproducibility and encourage replication, such as the Reproducibility Project: Cancer Biology ( https://osf.io/e81xl/ ), as well as endeavors to implement high-quality reporting. With consideration of complementary research, PLOS Biology will support and promote scientific reproducibility and replication.
By formalizing this policy and providing a venue for complementary studies, PLOS Biology ensures high visibility for well-supported, significant research findings. We wish to recognize both the value of validating results and the researchers undertaking the work. Highlighting replication studies will ultimately prove positive for the public perception of science.
Although we are only now articulating our editorial policy regarding complementary research officially, we have implemented this policy on a case-by-case basis previously. Under our newly codified policy, authors of a complementary study have six months from the publication or posting (to a preprint server) of the first article to submit their manuscript to PLOS Biology . We hope that authors will use these six months to fully support and potentially extend the results of the first article. Complementary research submitted beyond the six-month period may still be considered, depending on individual circumstances. All submissions must still meet our editorial requirements for depth of study and potential impact.
By these means, we hope to promote replication and to provide a high-quality venue for these complementary studies. We welcome feedback from the community on this policy and our other efforts to strengthen the scientific literature. Please write to the editors at [email protected] .
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
During the natural process of ageing, some cognitive abilities, such as vocabulary and implicit memory, do not seem to decline and may even improve. Conversely, other abilities, such as attention, executive functions, processing speed and working memory, progressively deteriorate over time [ 1 ]. This long process is accelerated by the presence of neurodegenerative diseases. Furthermore, memory deficits are a central symptom in Alzheimer’s disease, but semantic and verbal abilities also are impaired [ 2 ]. This leads to poor quality of life due to reduced daily life functioning and increased disability. Fewer years of education, the presence of the APOE E4 allele, traumatic brain injuries, obesity, diabetes, metabolic syndrome, hypertension and unhealthy diet are considered risk factors of cognitive decline [ 3 , 4 ].
The protective effect of physical activity also has been investigated. In its report on reducing the risks of cognitive decline and dementia [ 5 ], the World Health Organization (WHO) concluded, based on low to medium quality evidence, that physical activity (PA) has a modest protective effect on cognition and that this effect might be due to aerobic exercise. Baumgart et al [ 4 ] summarized the results of the independent evaluation by the Alzheimer’s Association of systematic reviews and meta-analysis, longitudinal and cross-sectional studies and randomized controlled trials (RCTs) on risk factors for cognitive decline, and found that PA—even as light as walking—is associated with a reduced risk of cognitive impairment and/or improved cognitive functions. However, they did not determine the optimal PA duration, type and intensity. Conversely, a Cochrane review [ 6 ] found no evidence in the available data from RCTs that aerobic PA brings cognitive benefit to cognitively healthy older adults. Although several studies suggest that PA could improve cognitive function and reduce cognitive decline [ 7 ], others found no evidence of a neuroprotective effect [ 8 ]. Moreover, few cohort studies evaluated the different PA types, particularly household activities [ 9 ]. Therefore, the debate is open on the magnitude of PA effects, the cognitive domains that might most benefit, and the most suitable PA types.
In this study, our objective was to analyze the association between two PA types and the decline over time of different cognitive domains in a large prospective cohort of older adults.
Materials and methods
Study design
The Three-city (3C) study is a multi-site community-living cohort of 9,294 participants aged 65 years and over. Participants were recruited between 1999 and 2001 from the electoral rolls of three French cities: Bordeaux, Dijon, and Montpellier. The aim was to study the impact of cardiovascular factors on the risk of dementia [ 10 ]. A face-to-face interview and a clinical examination were performed at inclusion and at every follow-up visit, at 2 (wave 1, W1), 4 (W2), 7 (W3), 10 (W4), 12 (W5), 15 (W6), and 17 (W7) years after inclusion ( S1 Table ).
The PA questionnaire was introduced at W3 for the Montpellier center and at W4 for the Bordeaux center (2515 participants). Therefore, the analyzed sample included only participants who completed this questionnaire, and the baseline evaluation refers to W3 and W4 for Montpellier and Bordeaux, respectively. The study protocol was approved by the University Hospital of Kremlin-Bicêtre Ethics Committee. Each participant signed an informed consent form.
Cognition assessment
Five cognitive tests were administered by neuropsychologists at each visit to evaluate different cognitive domains: global cognitive function was assessed with the French version of the Mini-Mental State Examination (MMSE) [ 11 ], visual working memory by the Benton Visual Retention Test (BVRT) [ 12 ], psychomotor speed and executive functions by the Trail Making Test Part A and B (TMTA and TMTB) [ 13 ], verbal fluency by the Isaacs’s Set Test (IST total score [ 14 , 15 ] that corresponds to the sum of the number of words generated in four semantic categories—animals, colors, cities, fruits—in 30 seconds), and verbal episodic memory by the Free and Cued Selective Reminding Test (FCSRT) [ 16 ]. For this analysis, the FCSRT "free recall score" (total number of words retrieved at the three free recall trials) and "total recall score" (total number of words retrieved at the three free and cued recall trials) were used; both scores range from 0 to 48 [ 17 ].
Physical activity assessment
PA was assessed with the self-report Voorrips questionnaire [ 18 ] that is designed to estimate PA in older adults. This questionnaire includes three parts: household/transportation activities, leisure time activities, and sport activities. The household/transportation activity part consists of ten questions (four to five possible scores for each item) about housework, preparing meals, shopping, and transportation (car, public transportation, bicycle, walking). The total sum (divided by 10) constitutes the first sub-score used in the present analysis. The leisure time and sport activity parts include questions on the type of activity, number of hours per week, and number of months per year. The activity types are associated with intensities that are determined according to the activity energetic costs. Sitting unloaded activities, which are mainly cognitive, were excluded from the scoring in the present study. Therefore, only standing and sport activities were considered. All leisure time and sport activities were pooled in the leisure/sport activity sub-score (intensity* number of hours per week* number of months per year). As the appropriate activity level thresholds for the household/transportation and leisure/sport sub-scores were previously determined in the 3C cohort [ 19 ], participants could be classified in three groups according to their sub-scores: <1.6, [1.6;2] and >2 for household/transportation activities, and 0,] 0;8.18] and >8.18 for leisure/sport activities.
Dementia diagnosis
At baseline and at each follow-up visit, all participants recruited in Montpellier were routinely examined by a neurologist at each follow-up visits. In Bordeaux, after an extensive cognitive and functional evaluation by a neuropsychologist specifically trained in dementia diagnosis, only participants with suspicion of dementia were examined by a neurologist. For both centers, a panel of independent neurologists expert in dementia reviewed all the existing information on the participants with suspected dementia at each visit, and a consensus on the diagnosis was obtained according to the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV), revised criteria and etiology [ 20 ]. Participants with prevalent dementia at baseline (W3 Montpellier and W4 Bordeaux) were excluded from the present analysis.
Baseline socio-demographic and clinical variables
Socio-demographic variables included sex, age, study center, and education level (<6 years, 6–11 years, and >11 years). Lifestyle variables were: consumption of alcohol (0.1–36 g/day, >36 g/day) and of fruits and vegetables (less than twice per day). Health status variables included diabetes (declared, treated, or glycaemia ≥7 mmol/L), body mass index (BMI, Kg/m 2 in three classes), self-reported cardiovascular diseases (stroke, angina pectoris, myocardial infarction, cardiac and vascular surgery), hypertension (treated, or blood pressure ≥160/95 mmHg) and depression [Center for Epidemiologic Studies Depression Scale (CES-D [ 21 ]) scores ≥16, or antidepressant treatment: Anatomical Therapeutic Chemical code N06A]. Participants with at least one ε4 allele were defined as APOE e4 carriers. Benzodiazepine use and a hierarchical disability indicator were also included. The disability indicator was calculated by combining the Rosow and Breslau Mobility Scale, Lawton-Brody Instrumental Activity of Daily Living (IADL) scale, and Katz Activity of Daily Living (ADL) scale [ 22 ].
Statistical methods
Baseline characteristics were compared between included and excluded participants using the chi-square and Wilcoxon rank-sum tests.
To keep in the analysis participants with missing baseline data, multiple imputations were carried out using a fully conditional specification method with discriminant function for the nominal variable [ 23 ]. The missing values represented 2% of all data for 28 variables. The SAS PROC MI procedure [ 24 ] with the FCS DISCRIM option for categorical variables and five imputations was used.
The participants’ characteristics were compared according to their household/transportation and leisure/sport activity sub-score group, and differences were tested with the Chi-square and Wilcoxon rank-sum tests. Associations between baseline PA groups and changes in cognitive performances were analyzed using linear mixed regression models. Separate models were performed for each cognitive test with each cognitive score as dependent variable (baseline and follow-up measures). As several cognitive scores had a skewed distribution, they were transformed using the most adapted functions: (15-BRVT) 1/2 , (30-MMSE) 1/2 , natural logarithm of the TMT scores, and (48-FCSRT total recall score) 1/2 . The effect of each potential confounding variable (main-effect term and interaction-with-time term) was also examined separately with a linear mixed model adjusted for sex, age, center, and education level. Covariates were selected if significant (p<25%) for at least one term and one cognitive test. First, the relation with the three PA sub-score groups was adjusted for age at baseline, sex, study center and education level, and their interaction with time for age and study center (minimally adjusted model). Then, the model was additionally adjusted for the selected covariates (consumption of fruits and vegetables, alcohol, BMI, hypertension, diabetes, cardiovascular disease, APOE carrier, depression and benzodiazepine use) and their time interactions if significant (for BMI, hypertension, diabetes, APOE carrier, depression). Interactions between PA groups and sex, age and educational level were also tested. The mixed model analyses were carried out using the five imputed datasets. The estimates were then combined using the SAS MIANALYZE procedure. Additional sensitivity analyses were performed to address possible reverse causality. Specifically, models were restricted to participants without IADL/ADL limitations at baseline and also to those free of dementia at all follow-up visits. All statistical analyses were carried out with SAS, version 9.4.
Results
Among the 2515 participants at baseline, 244 were not included because of dementia at baseline and 124 because they were confined at home. In addition, 29 participants were excluded because of missing baseline cognitive tests, 276 because they did not have any follow-up visit, and 145 due to absence of the two baseline PA sub-scores. Finally, 1697 participants were included in the analysis. Excluded individuals were older (p<0.0001), less likely to be women (p = 0.0126), with lower education level (p = 0.0034), less fully independent (p<0.0001), and with more depressive symptoms (p = 0.0002).
Sample description
At baseline, the median age was 79.7 years (IQR 76.9–83.0), 63.5% of participants were women, and 17.5% had at least one APOE e4 allele. Moreover, 10.1% consumed fruits and vegetables less than twice per day and 10.6% were obese. Concerning comorbidities, 68.7% had blood hypertension, 10.4% diabetes, 16.3% had cardiovascular diseases, 17.9% had depression, and 18% took benzodiazepines. When participants were divided in three groups according to their PA sub-scores, being older, dependent or obese or having hypertension was significantly associated with lower PA scores both for household/transportation and leisure/sport activities. Diabetes and cardiovascular diseases were only associated with lower household/transportation PA scores, while low educational level, low fruit and vegetable consumption, depression, and taking benzodiazepines were only associated with lower leisure/sport PA scores. Regarding sex distribution, the percentage of women was significantly higher in the group with the highest household/transportation PA sub-score and lower in the group with the highest leisure/sport PA sub-score. Alcohol consumption was associated with lower household/transportation PA sub-score and higher leisure/sport PA sub-score ( Table 1 and S2 Table ). All baseline cognitive scores were significantly better in participants with higher PA sub-scores, except for the BVRT and FCSRT total recall scores that did not differ in the three household/transportation (p = 0.06) and leisure/sport activity PA groups (p = 0.76), respectively.
10.1371/journal.pone.0252500.t001
Table 1 Participants’ characteristics according to their level of household and transportation-related physical activities, n = 1697.
Baseline household and transportation activities
≤1.6
] 1.6–2.0]
>2.0
Chi 2 test p value
n = 622
n = 580
n = 495
%
%
%
Sex, female
44.53
66.72
83.64
< .0001
Education level
0.3088
<6 years
23.67
22.11
22.06
6–11 years
25.44
30.74
29.96
>11 years
50.89
47.15
47.98
Hierarchical disability indicator
< .0001
Fully independent
34.38
42.04
51.4
Mild disability
41.01
46.48
44.3
Moderate to severe disability
24.61
11.48
4.3
Fruit and vegetable consumption
0.71
less than twice per day
10.78
10.16
9.26
Alcohol
0.0095
0
29.45
31.09
36.84
1–36 g/day
65.21
65.8
60.73
> 36 g/day
5.34
3.11
2.43
Body mass index
< .0001
Normal (<25)
43.79
56.24
63.43
Overweight (25–30)
42.16
33.74
29.75
Obese (≥30)
14.05
10.02
6.82
Treated hypertension or blood pressure ≥160/95 mm Hg
76.05
67.41
60.81
< .0001
Diabetes
14.31
10.55
5.3
< .0001
Cardiovascular disease
22.03
15.69
9.7
< .0001
Depressive symptoms (CES-D ≥16 or treatment)
16.67
17.77
19.46
0.4893
Benzodiazepine use
20.26
16.55
16.77
0.1752
APOE4 allele
14.14
16.82
22.27
0.0021
Median (IQR)
Median (IQR)
Median (IQR)
Wilcoxon test
Age
81 (74–87)
80 (73–86)
78 (72–84)
< .0001
MMSE score
28 (26–30)
28 (26–30)
29 (27–31)
0.03
BVRT score
12 (9–15)
12 (10–14)
12 (10–14)
0.06
TMTA (in seconds)
51 (24–78)
47 (23–71)
46 (23–69)
0.0002
TMTB (in seconds)
103 (36–170)
94 (40–148)
98 (46–150)
0.01
IST score
45 (30–60)
48 (34–62)
48 (35–61)
0.0002
FCSRT "free recall score"
25 (15–35)
26 (18–34)
26 (18–34)
< .0001
FCSRT "total recall score"
46 (41–51)
47 (43–51)
47 (44–50)
0.0003
BVRT: Benton Visual Retention Test, FCSRT: Free and Cued Selective Reminding Test, IQR: interquartile range, IST: Isaacs Set Test, MMSE: Mini-Mental State Examination, TMTA or TMTB: Trail Making Tests A or B.
Physical activity and cognitive decline over the 8 years of follow-up
The median (IQR) follow-up time was 7.9 years (7.5–8.1). The results of the model adjusted for sex, age, center and education level ( Table 2 ) showed that at baseline, participants in the intermediate household/transportation PA group had better BRVT, TMTA, TMTB, IST and FCSRT-free recall scores (but TMTA and FCSRT-free recall scores were better only in the highest sub-score group). Similarly, intermediate and higher leisure/sport PA sub-scores were associated with better TMTA, TMTB, IST and FCSRT-free recall scores. The interaction with time showed that the cognitive performance assessed by the TMTB and IST decreased slower over time in participants in the highest household/transportation PA group and for TMTB also in the intermediate group. Conversely, no significant effect over time was found for the leisure/sport activities.
10.1371/journal.pone.0252500.t002
Table 2 Association of physical activity with cognitive changes (minimally adjusted model * ).
MMSE
BVRT
TMTA
TMTB
IST
FCSRT free recall score
FCSRT total recall score
sqrt(30-MMSE)
sqrt(15-BVRT)
ln(TMTA)
ln(TMTB)
sqrt(48- FCSRT)
N = 1691
N = 1621
N = 1569
N = 1527
N = 1657
N = 1594
N = 1593
beta (SE)
P value
beta (SE)
P value
beta (SE)
P value
beta (SE)
P value
beta (SE)
P value
beta (SE)
P value
beta (SE)
P value
Household/transportation activities
] 1.6–2.0]
-0.009*(0.039)
0.823
-0.063 (0.031)
0.039
-0.051 (0.020)
0.011
-0.047 (0.022)
0.037
1.596 (0.542)
0.003
1.092 (0.366)
0.003
-0.079 (0.064)
0.214
>2.0
-0.064 (0.043)
0.138
-0.039 (0.033)
0.235
-0.051 (0.022)
0.021
-0.011 (0.024)
0.649
1.053 (0.590)
0.074
0.861 (0.396)
0.030
-0.061 (0.069)
0.375
Household/transportation activities * time
] 1.6–2.0]
0.002 (0.008)
0.831
0.005 (0.006)
0.430
-0.005 (0.004)
0.142
-0.009 (0.004)
0.031
0.064 (0.076)
0.401
0.041 (0.071)
0.568
-0.007 (0.013)
0.610
>2.0
0.011 (0.009)
0.210
0.004 (0.007)
0.539
-0.007 (0.004)
0.067
-0.010 (0.004)
0.018
0.168 (0.079)
0.033
0.025 (0.072)
0.733
-0.004 (0.013)
0.736
Leisure/sports activities
] 0–8.18]
-0.054 (0.037)
0.146
-0.006 (0.029)
0.840
-0.065 (0.019)
0.001
-0.043 (0.022)
0.047
1.885 (0.514)
0.0002
1.092 (0.344)
0.002
-0.072 (0.061)
0.233
>8.18
-0.048 (0.047)
0.315
-0.001 (0.037)
0.976
-0.012 (0.024)
< .0001
-0.053 (0.027)
0.047
2.400 (0.658)
0.0003
1.133 (0.438)
0.010
-0.017 (0.077)
0.828
Leisure/sports activities * time
] 0–8.18]
-0.003 (0.008)
0.671
-0.001 (0.006)
0.918
-0.002 (0.004)
0.523
-0.005 (0.004)
0.204
-0.078 (0.074)
0.290
-0.104 (0.067)
0.120
0.019 (0.012)
0.128
>8.18
-0.004 (0.010)
0.662
-0.006 (0.007)
0.438
-0.001 (0.004)
0.732
-0.008 (0.005)
0.089
-0.074 (0.091)
0.415
-0.038 (0.083)
0.647
-0.001 (0.015)
0.951
*Models adjusted for age , sex , study center , education level , and time by age and time by study center interactions .
SE standard error, sqrt: square root, ln: natural logarithm.
BVRT: Benton Visual Retention Test, FCSRT: Free and Cued Selective Reminding Test, IST: Isaacs Set Test, MMSE: Mini-Mental State Examination, TMTA or TMTB: Trail Making Tests A or B.
The second model ( Table 3 ), adjusted for all selected covariates and their interaction with time, if significant, showed that the PA sub-scores were no longer significantly associated with the baseline BRVT and TMTB scores. The interaction between household/transportation PA groups and time remained significant for the TMTB (p = 0.03) and IST (p = 0.009) scores, suggesting a slower decline over time, and was borderline significant for the TMTA scores (p = 0.06). For instance, according to the parameters estimated by this multi-adjusted model, in 80-year-old participants from Bordeaux without comorbidity (obesity, hypertension, diabetes, depression) and without ApoE allele 4, the mean IST decline was 0.90 word per year in the low-level and 0.69 word per year in the high-level household/transportation PA group. For TMTB, in the same participants, the mean time in seconds to perform the test was multiplied by 1.049 and by 1.039 each year in the low-level and high-level household/transportation PA groups, respectively. For a baseline TMTB score of 100 seconds, this would correspond approximatively to an increase of 4.9 and 3.9 seconds per year, respectively. Again, no significant effect over time was found for the leisure/sport activities.
10.1371/journal.pone.0252500.t003
Table 3 Associations of physical activity with cognitive changes (multi-adjusted model * ).
MMSE
BVRT
TMTA
TMTB
IST
FCSRT free recall score
FCSRT total recall score
sqrt(30-MMSE)
sqrt(15-BVRT)
ln(TMTA)
ln(TMTB)
sqrt(48- FCSRT)
N = 1691
N = 1621
N = 1569
N = 1527
N = 1657
N = 1594
N = 1593
beta (SE)
P value
beta (SE)
P value
beta (SE)
P value
beta (SE)
P value
beta (SE)
P value
beta (SE)
P value
beta (SE)
P value
Household/transportation activities
] 1.6–2.0]
0.009 (0.039)
0.826
-0.054 (0.031)
0.078
-0.045 (0.020)
0.024
-0.038 (0.022)
0.092
1.293 (0.539)
0.017
0.992 (0.365)
0.007
-0.069 (0.064)
0.285
>2.0
-0.039 (0.044)
0.369
-0.026 (0.034)
0.447
-0.041 (0.022)
0.063
0.00004 (0.025)
0.999
0.632 (0.595)
0.288
0.705 (0.400)
0.078
-0.052 (0.071)
0.466
Household/transportation activities * time
] 1.6–2.0]
-0.001 (0.008)
0.907
0.005 (0.006)
0.463
-0.006 (0.004)
0.120
-0.009 (0.004)
0.035
0.095 (0.076)
0.215
0.054 (0.072)
0.450
-0.010 (0.013)
0.442
>2.0
0.007 (0.009)
0.462
0.003 (0.007)
0.645
-0.007 (0.004)
0.060
-0.009 (0.004)
0.032
0.210 (0.081)
0.009
0.054 (0.075)
0.473
-0.011 (0.014)
0.429
Leisure/sports activities
] 0–8.18]
-0.027 (0.038)
0.468
0.024 (0.030)
0.427
-0.052 (0.019)
0.007
-0.021 (0.022)
0.342
1.383 (0.522)
0.008
0.847 (0.349)
0.015
-0.066 (0.062)
0.289
>8.18
-0.004 (0.048)
0.941
0.042 (0.037)
0.263
-0.082 (0.024)
0.001
-0.019 (0.027)
0.470
1.612 (0.669)
0.016
0.792 (0.447)
0.076
0.0004 (0.079)
0.996
Leisure/sports activities * time
] 0–8.18]
-0.003 (0.008)
0.747
-0.003 (0.006)
0.688
-0.002 (0.004)
0.563
-0.005 (0.004)
0.236
-0.068 (0.075)
0.366
-0.111 (0.068)
0.101
0.021(0.013)
0.086
>8.18
-0.003 (0.010)
0.805
-0.007 (0.008)
0.341
-0.001 (0.004)
0.870
-0.008 (0.005)
0.116
-0.066 (0.094)
0.483
-0.066 (0.086)
0.444
0.005 (0.016)
0.752
*Models adjusted for age, sex, study center, education, depression, alcohol, BMI, benzodiazepine, diabetes, hypertension, APOE 4 allele, consumption of fruits and vegetables (less than twice per day), cardiovascular disease, and time by age, study center, depression, BMI, diabetes, hypertension and APOE4 allele interactions.
SE standard error, sqrt: square root, ln: natural logarithm.
BVRT: Benton Visual Retention Test, FCSRT: Free and Cued Selective Reminding Test, IST: Isaacs Set Test, MMSE: Mini-Mental State Examination, TMTA or TMTB: Trail Making Tests A or B.
The interactions of sex, age, and education with the household/transportation and leisure/sport PA groups were not significant, including the interaction with time.
Sensitivity analyses
When only participants without IADL/ADL limitations at baseline were included (n = 1461), the interaction between household/transportation PA groups with time remained significant for the TMTB and IST scores in the multi-adjusted models. Moreover, the interaction between household/ transportation PA scores and time became significant also for the TMTA scores (β(SE) = -0.01(0.004) and p = 0.014)). Whatever the cognitive function analyzed, no significant interaction with time was observed for the leisure/sport PA groups.
When the incident cases of dementia during the follow-up were excluded, the results for the interaction with time were similar for the TMTB [β(SE) = -0.01(0.008), p = 0.054] and IST scores [β(SE) = 0.15(0.08), p = 0.061].
Discussion
In this prospective cohort of community-dwelling participants older than 72 years of age, the decline over time in the TMTB (executive functions) and IST (verbal fluency) was significantly slower in participants who reported moderate or high level of household/transportation-related PA, whereas the TMTA (psychomotor speed) performance change was almost significant. At baseline and in the multi-adjusted models, the two PA sub-scores (household/transportation and leisure/sports activities) were also positively associated with better TMTA, IST and FCSRT ("free recall") scores. During the follow-up, the increase in the time required to perform the TMTB was almost one second shorter per year, while the decline in the number of generated words in the IST was reduced by approximately 0.2 words per year in the high-level household/transportation activity group compared with the low-level household/transportation activity group. These longitudinal results remained significant after adjustment for a large number of variables, including diabetes, cardiovascular diseases and hypertension, and also in the sensitivity analyses that excluded participants with incident dementia or moderate to severe disability.
Our study has several limitations. First, in our aged sub-sample of the 3C cohort (median = 80 years), the level of leisure and sports activities was relatively low. Although this was representative of PA in this age category, 40.4% of them did not perform any activity of this type, and this may preclude the detection of a significant PA effect. Second, their low PA level could be also a consequence of comorbidities or frailty. However, the effect of household/transportation-related PA persisted after adjustment for many different comorbidities and risk factors. Third, low PA could also be a consequence of a pre-dementia state. However, the long follow-up (8 years) and the sensitivity analysis after exclusion of participants with incident dementia limited this possibility. Fourth, PA was measured using a self-report questionnaire that might be susceptible to information bias (misreport of activity or frequency and duration overestimation). However, the Voorrips questionnaire has been specially designed for older adults and has been validated for light and high activities [ 25 ]. Finally, each cognitive test was analyzed separately and this could raise the issue of test multiplicity that might increase the false positive rate. However, each cognitive test assessed a specific cognitive domain, and PA might influence these cognitive domains in different ways.
Our study has several strengths. First, this was a longitudinal and multi-center study with a large sample size (n = 1697). Cognition was assessed with different tests that allowed evaluating global cognition and specific cognitive domains (working memory, executive functions, verbal fluency, verbal episodic memory). PA was assessed using the Voorrips questionnaire that is specific for older adults and allows analyzing two PA types: household/transportation activities and leisure/sports activities.
Many studies, systematic reviews, and meta-analyses have evaluated the link between PA and cognition. The systematic review by Cunningham et al. [ 26 ] found a reduction in the risk of cognitive decline (by 26% for moderate and by 33% for high PA). Conversely, the systematic review by Kikkert et al. [ 27 ] did not find PA as protector of cognitive decline. A meta-analysis of five randomized trials [ 28 ] concluded that PA was not associated with cognitive decline. These discrepancies may be due to the heterogeneity of the methods used. Some studies [ 29 – 35 ] used logistic regression analysis, and therefore did not consider the effect of time on the occurrence of cognitive disorders. Others included only women [ 32 , 36 ] or only men [ 37 ]. In some studies [ 33 – 35 , 38 ], the very short follow-up period did not allow reaching any conclusion. In other analyses [ 39 – 44 ], contradictory results were obtained due to differences in study design/statistical methods, cognitive test(s) used, or PA type analyzed. Therefore, it is necessary to develop models that take into account time and comorbidities, and with a sufficiently long follow-up, as done in our cohort. The cognitive domains evaluated and the tests used also might explain the result heterogeneity. Two indicators of global cognition may show different results. For instance, in the study by Willey et al [ 45 ] PA was significantly linked to global cognition defined by the MMSE score, but not with the Telephone Interview for Cognitive Status that is more sensitive to memory changes [ 46 ]. Noticeably, two studies [ 42 , 47 ] showed a possible PA protective effect particularly on episodic memory and language.
Our study did not detect a positive effect of sports and leisure activities on cognition, as demonstrated by a recent study [ 8 ]. Conversely, it found a positive effect of domestic activities on verbal fluency and executive functions during the 8 years of follow-up, in agreement with previous findings [ 29 , 48 , 49 ]. Newson et al [ 49 ] showed that in older adults, lifestyle activities, such as household, domestic and social activities, influence some cognitive domains, particularly speed of processing and picture naming. Angevaren et al [ 48 ] associated PA intensity with processing speed; they defined PA by leisure activities (walking, cycling, housekeeping, gardening). It has been proposed that PA influence some cognitive functions, such as processing speed, more than memory and mental flexibility that require also knowledge and experience [ 49 ]. Household and transportation activities are very important for older adults because they represent a large part of their daily activities. These regular, non-intensive activities may have a stronger protective effect on cognition than sport activities that are far less frequent. In agreement, it has been reported that light, regular aerobic exercise increases neurogenesis and neuroplasticity and improves cardiovascular function and its associated influence on the cerebrovascular system [ 50 ].
The prevalence of older adults reporting insufficient PA varies widely around the world, between rural and urban societies, and between men and women, and is increasing in high-income countries [ 51 ]. This might lead to differences in the statistical power of the studies, and could explain some inconsistent previous findings. The type of PA practiced also is changing in Western countries, with less time spent performing domestic activities and more time dedicated to leisure activities. However, these changes are not uniform across countries due to cultural norms and social habits. For example, the proportion of people performing resistance training or strength exercises is lower in South-eastern than in Nordic European countries [ 52 ]. Therefore, the results of our study highlighting the importance of domestic activities should be confirmed in other older populations.
Conclusion
We found a slower decline of cognitive functions, particularly executive functions and verbal fluency, over a 8-year follow-up period, in ≥72-year-old people who performed moderate to high household/transportation activities. Conversely, we did not detect any association with leisure and sports activities. These results remained stable after adjustment for potential confounders. Our study shows the importance of considering the PA type using a specific questionnaire that includes also domestic activities. Continuing to participate in domestic activities and to use adapted transport could allow older adults to better maintain their cognitive capacities.
Supporting information
S1 Table
3C cohort design.
(PDF)
S2 Table
Participants’ characteristics according to their leisure and sport activities level, n = 1697.
(PDF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The omicron variant of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) led to a surge in cases among young people during 2022 when several countries lifted or alleviated their restrictions [ 1 ]. Although clinical trials showed that the BNT162b2 and mRNA-1273 vaccines had acceptable safety profiles in adolescents and reduced the risk of infection in the short term [ 2 , 3 ], these trials were conducted before the emergence of omicron. Limited sample sizes also hindered the evaluation of the level of protection against severe Coronavirus Disease 2019 (COVID-19), such as hospitalisation, which could be estimated using large-scale observational studies. These studies also offer the possibility of investigating any potential links between vaccination and rare serious adverse events, such as myocarditis [ 4 , 5 ].
Currently, there is limited data on vaccine effectiveness (VE) against severe COVID-19 caused by the omicron variant among adolescents. Two case–control studies found that 2 doses of the BNT162b2 mRNA vaccine had about 80% VE against COVID-19 hospitalisation or death in adolescents during the omicron era [ 6 , 7 ]. However, given their study design, it is difficult to determine how common severe disease is during the omicron era. In addition, little is known about whether certain groups of adolescents should be prioritised for vaccination because of a higher risk of severe COVID-19, and whether vaccination has a similarly protective effect in such risk groups. Moreover, although a third dose, also known as a booster dose, may increase protection against symptomatic omicron infection in adolescents [ 8 , 9 ], the risk of severe COVID-19 after a third dose relative to after the second dose is unclear.
In the present study, we used Swedish nationwide registers to evaluate, among adolescents, (1) the risk of hospitalisation from any cause following monovalent COVID-19 mRNA vaccination, and (2) the effectiveness of monovalent COVID-19 mRNA vaccination against hospitalisation due to COVID-19 and risk factors for COVID-19 hospitalisation during an omicron predominant period.
Methods
Study design and cohorts
This nationwide, retrospective cohort study was approved by the Swedish Ethical Review Authority (number 00094/2021). There was no prospective written analysis plan for the present study, and the construction of the models and analyses were data driven. This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline ( S1 STROBE Checklist).
The cohort considered for inclusion was compiled by Statistics Sweden ( www.SCB.se ); the government agency in charge of nationwide statics in different areas and covering the total population of Sweden. The Public Health Agency of Sweden provided data for all individuals born 2003 to 2009 who were given at least 1 dose of monovalent COVID-19 mRNA vaccine or had a documented SARS-CoV-2 infection until March 2022 ( N = 692,419). To vaccinated individuals, Statistics Sweden matched 1 control individual on birth year, sex, and municipality. Controls could not have received a first dose of vaccine at the date when the corresponding vaccinated individual had received 2 doses of vaccine. The cohort was updated with vaccination data and SARS-CoV-2 infections until 2 June 2022. Because 1 control could be matched to several vaccinated individuals, the total eligible cohort consisted of 832,273 individuals, of whom 645,355 had been vaccinated with at least 1 dose, 600,721 had been vaccinated with 2 doses, and 60,391 had been vaccinated with at least 3 doses ( Fig 1 ). In a first set of analyses, named the safety analysis, all diagnoses set during hospitalisation were evaluated in all individuals given at least 1 dose of vaccine during follow-up ( N = 645,355) as compared to individuals never vaccinated during follow-up ( N = 186,918). In a second set of analyses, VE against COVID-19 hospitalisation was evaluated by comparing individuals given at least 2 doses of vaccine ( N = 501,945) to individuals never vaccinated during follow-up ( N = 157,979), excluding all individuals with a previous documented SARS-CoV-2 infection ( Fig 1 ). Data on individuals vaccinated against COVID-19 and data on documented SARS-CoV-2 infections were collected from the Swedish Vaccination Register and the SmiNet register, respectively. Both these registers are managed by the Public Health Agency of Sweden and all healthcare providers in Sweden were obliged to report to these registers according to law [ 10 , 11 ].
10.1371/journal.pmed.1004127.g001
Fig 1
Description of the study cohort.
Outcomes
In the safety analysis, based on all diagnoses set in the cohort during inpatient hospital stay until 5 June 2022, results are presented for all-cause hospitalisation and for 30 different diagnoses. The 30 specific diagnoses were selected based on their incidence and general interest given previous reports of links between certain diagnoses and COVID-19 mRNA vaccination [ 4 , 12 , 13 ]. Only the first diagnosis was evaluated for each individual, and individuals with this diagnosis at baseline were therefore excluded from the prospective analyses.
In the VE analysis, the primary outcome was a main diagnosis of COVID-19 (thus, “due to” COVID-19 rather than “with” COVID-19), set during inpatient hospital stay from 1 January 2022 until 5 June 2022. The secondary outcome was documented SARS-CoV-2 infection of any severity from the SmiNet register, from 1 January 2022 until 28 February 2022, in line with the changes made to testing guidelines in Sweden. For these 2 outcomes, only cases that occurred 7 days after the second dose of vaccine and onwards were counted to ensure the full effect of vaccination. The start of 1 January was selected on the basis that the omicron variant was first documented in Sweden on 29 November 2021 [ 14 ], and by early January 2022, it represented >90% of sequenced cases ( S1 Table ).
Data on all diagnoses used in this study were obtained from the National Patient Register [ 15 ] using 10th revision International Classification of Disease codes, starting from 1 January 2016 earliest until 5 June 2022 latest. The definition and diagnostic codes for all diagnoses considered in the safety analysis and for the analysis of risk factors for COVID-19 hospitalisation are shown in Table 1 . The positive predictive value for diagnoses set within the National Patient Register differs but has generally been found to be between 85% and 95%, although sensitivity is often lower [ 15 ]. Finally, data on death due to COVID-19 or influenza, defined as death within 30 days after the diagnosis, were obtained from the National Cause of Death Register [ 16 ].
10.1371/journal.pmed.1004127.t001
Table 1 Baseline characteristics of the individuals at date of the first dose of vaccine and in never vaccinated individuals.
Vaccinated with ≥1 dose ( N = 645,355)
Never vaccinated ( N = 186,918)
p
Age, years (standard deviation)
15.4 (1.9)
14.8 (2.0)
< 0.001
Female sex
318,983 (49.4)
85,798 (45.9)
< 0.001
Born in Sweden
567,298 (87.9)
139,213 (74.5)
< 0.001
Mean baseline date
10 October 2021
10 October 2021
< 0.001
Previous SARS-CoV-2 infection
91,934 (14.2)
20,896 (11.2)
< 0.001
Previous diagnoses (ICD-10 code)
Any diagnosis
25,686 (4.0)
7,515 (4.0)
< 0.001
Gastroenteritis (A09)
125 (0.019)
37 (0.020)
0.32
Sepsis (A41)
45 (0.007)
17 (0.009)
0.54
Erysipelas (A46)
25 (0.004)
9 (0.005)
0.93
Bacterial infection unspecified (A49)
21 (0.003)
8 (0.004)
0.29
Mononucleosis (B27)
224 (0.034)
38 (0.020)
< 0.001
Virus infection, unspecified (B34)
114 (0.018)
34 (0.018)
0.59
Chronic lymphocytic leukemia (C91)
30 (0.005)
13 (0.007)
0.07
Iron deficiency anemia (D50)
93 (0.014)
40 (0.021)
0.16
Thrombocytopenia (D69)
44 (0.007)
14 (0.007)
0.19
Agranulocytosis (D70)
24 (0.004)
8 (0.004)
0.93
Alcohol dependency (F10)
975 (0.15)
313 (0.17)
0.48
Depressive episode (F32)
881 (0.14)
211 (0.11)
< 0.001
Anxiety state, unspecified (F41)
586 (0.09)
185 (0.10)
0.38
Allergy or anaphylactic shock (T78)
177 (0.027)
58 (0.031)
0.35
Anorexia nervosa (F50)
571 (0.088)
107 (0.092)
< 0.001
Epilepsy (G40)
318 (0.049)
113 (0.060)
0.004
Otitis media (H66)
22 (0.003)
7 (0.004)
0.11
Myocarditis (I40)
50 (0.008)
19 (0.010)
0.75
Pericarditis (I30)
26 (0.004)
10 (0.005)
0.62
Sinusitis (J01)
39 (0.006)
17 (0.009)
0.13
Tonsillitis (J03)
166 (0.026)
38 (0.020)
0.67
Chronic tonsillitis (J35)
208 (0.032)
58 (0.031)
0.58
Upper respiratory infection (J06)
64 (0.010)
18 (0.010)
0.87
Pneumonia (J15 and J18)
151 (0.023)
40 (0.021)
0.65
Peritonsillar abscess (J36)
118 (0.018)
37 (0.020)
0.82
Appendicitis (K35)
2,071 (0.32)
544 (0.29)
< 0.001
Crohn’s disease (K50)
107 (0.017)
39 (0.021)
0.15
Cutaneous abscess (L02)
72 (0.011)
29 (0.016)
0.09
Nephritis (N10)
252 (0.039)
53 (0.028)
0.10
Traumatic brain injury (S06)
1,212 (0.19)
293 (0.16)
< 0.001
Combined diagnoses
Cerebral palsy/development disorders (G80, F73, F82, F84, Q02)
23,093 (3.6)
8,307 (4.4)
< 0.001
Selected infections (A49, J03, J15, J18, J35)
21,208 (3.3)
7,158 (3.8)
< 0.001
All data are shown as number (percentage) unless stated otherwise.
All previous diagnoses were main diagnoses set during inpatient hospital stay from 1 January 2020 earliest until 5 June 2022 latest as obtained from the National Patient Register, with the exception of the diagnoses listed under “combined diagnoses,” for which diagnoses set from 1 January 2016 and later was used and including secondary outpatient care.
Statistical analysis
In the safety analysis, diagnoses were evaluated before and after the first dose of vaccine in 645,355 individuals given at least 1 dose of vaccine and in 186,918 controls who were unvaccinated during the follow-up time. The baseline date in vaccinated individuals was the date of vaccination with the first dose. In the controls, a baseline date was randomly assigned based on the mean baseline date and standard deviation among the vaccinated individuals (10 October 2021 ± 54 days). Student t tests and chi-squared tests were used to compare the prevalence of different variables at baseline. To estimate hazard ratios (HRs) for all-cause hospitalisation and for the 30 different diagnoses during follow-up, Cox regression models were used. Individuals were censored on the date of the diagnosis of interest, death, or end of follow-up (5 June 2022), whichever came first.
In the VE analysis, the proportional hazards assumption was not met; hence, logistic regression was used to estimate odds ratios (ORs) for the primary outcome of COVID-19 hospitalisation (from 1 January 2022 until 5 June 2022), and for the secondary outcome of a SARS-CoV-2 infection (from 1 January 2022 until 28 February 2022). The ORs obtained were used to estimate VE as 1 minus the OR × 100. In all analyses, the first model was unadjusted and the second model was adjusted for baseline date, age, sex, and whether the individual was born in Sweden or not. Data underlying these covariates were retrieved from Statistics Sweden [ 17 ]. To investigate whether VE differed by the covariates, interaction analyses were performed using product terms created by multiplying the variable coding for vaccination status at baseline (vaccinated/unvaccinated) by each respective covariate, which was added to the logistic regression model. Given that the interaction term was statistically significant (p < 0.05) for the baseline date, VE was also estimated in subgroups according to this covariate. The number needed to vaccinate (NNV) with 2 doses to prevent 1 case of COVID-19 hospitalisation during follow-up was estimated as the inverse of the absolute risk difference between the groups (vaccinated/unvaccinated).
Finally, a sensitivity analysis using a negative control outcome was conducted to explore the potential risk of bias due to unmeasured confounding [ 18 ]. Here, a logistic regression model was performed, comparing individuals given at least 2 doses of COVID-19 mRNA vaccine compared to never vaccinated individuals concerning the outcome of hospitalisation due to influenza from 1 January 2022 until 5 June 2022. All analyses were performed in SPSS v29.0 for Mac (IBM, Armonk, New York, USA), and Stata v16.1 for Mac (Statcorp, College Station, Texas, USA). A two-sided p -value < 0.05 or ORs/HRs with 95% confidence intervals (CIs) not crossing one were considered statistically significant.
Results
The total cohort comprised 832,273 adolescents born 2003 to 2009 (age 11.3 to 19.2 years), of whom 645,355 received at least 1 dose of COVID-19 mRNA vaccine and 186,918 never vaccinated individuals (controls). Almost 90% of the vaccinated individuals received BNT162b2 as a first dose, while the remaining received mRNA-1273. Baseline characteristics are shown in Table 1 . Individuals that were never vaccinated were slightly younger, more often of male sex and born outside of Sweden, and less often diagnosed with a previous SARS-CoV-2 infection before baseline (p < 0.001 for all). Concerning other diagnoses at baseline, differences between the groups were marginal ( Table 1 ).
Serious adverse events after a first dose of vaccine
In the safety analysis ( N = 832,273), there were a total of 19,580 all-cause hospitalisations among 14,266 individuals during follow-up. In vaccinated individuals, 1.69% ( N = 10,906) were hospitalised at least once, compared to 1.80% ( N = 3,360) in those never vaccinated (HR; 0.84, 95% CI [0.81, 0.88], p < 0.001; Table 2 ). There were marginal differences between the 2 groups in the 30 selected diagnoses ( Table 2 ), although statistically significant associations in favour of vaccination were observed with respect to the risk of sepsis, thrombocytopenia, alcohol dependency, peritonsillar abscess, and Crohn’s disease (p < 0.05 for all). None of the other associations were statistically significant after adjustment.
10.1371/journal.pmed.1004127.t002
Table 2 Risk of hospitalisation for any cause and for 30 selected diagnoses in vaccinated individuals vaccinated compared to never vaccinated individuals.
Vaccinated with ≥1 one dose ( N = 645,355)
Never vaccinated ( N = 186,918)
Unadjusted analyses
Adjusted analyses
Outcome
Number of cases (IR)
Number of cases (IR)
HR [95% CI]
p
HR [95% CI]
p
All-cause hospitalisation
10,906 (71.9)
3,360 (76.5)
0.94 (0.91, 0.98)
0.003
0.84 (0.81, 0.88)
< 0.001
Hospitalisation for selected diagnosis (ICD-10 code)
Gastroenteritis (A09)
58 (0.38)
20 (0.45)
0.85 (0.51, 1.42)
0.54
0.99 (0.56, 1.66)
0.89
Sepsis (A41)
5 (0.03)
7 (0.16)
0.20 (0.06, 0.64)
0.006
0.17 (0.05, 0.56)
0.003
Erysipelas (A46)
8 (0.05)
2 (0.05)
1.28 (0.27, 6.15)
0.76
0.99 (0.18, 5.41)
0.99
Bacterial infection unspecified (A49)
11 (0.07)
3 (0.07)
1.05 (0.29, 3.76)
0.94
0.83 (0.23, 3.05)
0.78
Mononucleosis (B27)
115 (0.76)
26 (0.59)
1.27 (0.83, 1.95)
0.27
0.78 (0.50, 1.21)
0.26
Virus infection, unspecified (B34)
42 (0.28)
7 (0.16)
1.74 (0.78, 3.86)
0.18
1.52 (0.67, 3.43)
0.32
Chronic lymphocytic leukemia (C91)
4 (0.03)
4 (0.09)
0.29 (0.07, 1.16)
0.08
0.36 (0.08, 1.50)
0.16
Iron deficiency anemia (D50)
38 (0.25)
14 (0.32)
0.79 (0.43, 1.46)
0.45
0.88 (0.47, 1.66)
0.70
Thrombocytopenia (D69)
8 (0.05)
10 (0.23)
0.18 (0.10, 0.65)
0.004
0.21 (0.08, 0.58)
0.002
Agranulocytosis (D70)
1 (0.007)
1 (0.02)
-
-
-
Alcohol dependency (F10)
409 (2.69)
123 (2.80)
0.97 (0.79, 1.18)
0.74
0.75 (0.61, 0.93)
0.008
Depressive episode (F32)
363 (2.39)
102 (2.32)
1.03 (0.83, 1.29)
0.77
0.88 (0.70, 1.11)
0.28
Anxiety state, unspecified (F41)
341 (2.25)
75 (1.70)
1.40 (1.09, 1.80)
0.009
0.92 (0.71, 1.19)
0.51
Allergy or anaphylactic shock (T78)
64 (0.42)
23 (0.52)
0.81 (0.50, 1.30)
0.38
0.70 (0.43, 1.15)
0.16
Anorexia nervosa (F50)
197 (1.30)
47 (1.07)
1.21 (0.88, 1.67)
0.23
1.08 (0.78, 1.50)
0.64
Epilepsy (G40)
82 (0.54)
33 (0.75)
0.72 (0.48, 1.08)
0.12
0.89 (0.57, 1.41)
0.63
Otitis media (H66)
7 (0.05)
3 (0.07)
0.67 (0.17, 2.60)
0.57
0.70 (0.16, 2.97)
0.63
Myocarditis (I40)
86 (0.57)
19 (0.43)
1.31 (0.80, 2.15)
0.29
0.99 (0.60, 165)
0.98
Pericarditis (I30)
21 (0.14)
3 (0.07)
2.00 (0.60, 6.73)
0.26
0.88 (0.26, 3.02)
0.84
Sinusitis (J01)
15 (0.10)
10 (0.23)
0.43 (0.19, 0.96)
0.04
0.64 (0.26, 1.58)
0.33
Tonsillitis (J03)
90 (0.59)
11 (0.25)
2.34 (1.25, 4.38)
0.008
1.51 (0.80, 2.86)
0.21
Chronic tonsillitis (J35)
94 (0.59)
18 (0.41)
1.53 (0.92, 2.53)
0.10
1.15 (0.68, 1.93)
0.60
Upper respiratory infection (J06)
24 (0.16)
4 (0.09)
1.76 (0.61, 5.10)
0.30
1.72 (0.56, 5.27)
0.35
Pneumonia (J15 and J18)
68 (0.45)
16 (0.36)
1.26 (0.73, 2.17)
0.41
1.47 (0.82, 2.64)
0.20
Peritonsillar abscess (J36)
52 (0.34)
17 (0.39)
0.90 (0.52, 1.97)
0.72
0.56 (0.32, 0.98)
0.04
Appendicitis (K35)
665 (4.38)
174 (3.96)
1.11 (0.94, 1.32)
0.21
1.04 (0.87, 1.23)
0.69
Crohn’s disease (K50)
24 (0.16)
14 (0.32)
0.49 (0.26, 0.96)
0.04
0.46 (0.23, 0.92)
0.03
Cutaneous abscess (L02)
10 (0.07)
5 (0.11)
0.58 (0.20, 1.69)
0.32
0.59 (0.20, 1.69)
0.35
Nephritis (N10)
105 (0.69)
18 (0.41)
1.69 (1.02, 2.79)
0.04
1.22 (0.73, 2.03)
0.45
Traumatic brain injury (S06)
307 (2.02)
83 (1.89)
1.08 (0.84, 1.37)
0.56
1.09 (0.85, 1.41)
0.50
CI, confidence interval; HR, hazard ratio. IR = incidence rates per 1 million person-days of follow-up. Adjusted analyses were adjusted for age, sex, baseline date, and whether the individual was born in Sweden.
Vaccine effectiveness against COVID-19 hospitalisation
In the analysis of VE against COVID-19 hospitalisation, 501,945 individuals vaccinated with 2 doses and 157,979 never vaccinated controls were included. Between 1 January 2022 and 5 June 2022, a total of 47 individuals (7 per 100,000) were hospitalised due to COVID-19. Of these, 21 cases were among those vaccinated with 2 doses (0.004%), and 26 among unvaccinated (0.016%), resulting in a VE of 76% (95% CI [57, 87], p < 0.001) ( Table 3 ). The NNV with 2 doses to prevent 1 case of COVID-19 hospitalisation was 8,147. For those with a second dose of vaccine earlier than 15 November, the VE was 69% (95% CI [29, 87], p = 0.006), compared to 87% (95% CI [66, 95], p < 0.001) for those with a second dose of vaccine 15 November 2021 and later (p = 0.03 for interaction). When comparing individuals vaccinated with 3 doses ( N = 41,225) compared to those vaccinated with 2 doses ( N = 413,544), only 12 individuals in total (2.6 per 100,000 individuals) were hospitalised due to COVID-19 during follow-up (VE; 13%, 95% CI [−354, 84], p = 0.86). There were no deaths within 30 days of hospitalisation among the 261 individuals hospitalised due to COVID-19 in the total cohort (832,273) since the beginning of the pandemic in January 2020.
10.1371/journal.pmed.1004127.t003
Table 3 VE against COVID-19 hospitalisation from 7 days onwards after a second dose of vaccine as compared to never vaccinated individuals from 1 January 2022 until 5 June 2022, and by time since vaccination and subgroups.
Vaccinated with 2 doses
Never vaccinated
Unadjusted analyses
Adjusted analyses
Number of cases (%)
Number of cases (%)
VE [95% CI]
p
VE [95% CI]
p
Total cohort ( N = 659,924) a
21 (0.004%)
26 (0.016%)
75 (55, 86)
< 0.001
76 (57, 87)
< 0.001
Subgroups
Baseline date < 15 Nov 2021 ( N = 287,871) b
15 (0.007%)
14 (0.018%)
61 (19, 81)
0.01
69 (29, 87)
0.006
Baseline date > 14 Nov 2021 ( N = 372,053) c
6 (0.002%)
12 (0.015%)
86 (63, 95)
< 0.001
87 (66, 95)
< 0.001
Previously diagnosed with selected infections ( N = 21,981) d
4 (0.025%)
11 (0.187%)
87 (58, 96)
< 0.001
88 (58, 96)
< 0.001
Previously diagnosed with development disorders ( N = 25,832) e
6 (0.032%)
9 (0.124%)
74 (27, 91)
0.01
72 (20, 91)
0.02
CI, confidence interval; VE, vaccine effectiveness.
Adjusted models were adjusted for age, baseline date, sex, and whether the individual was born in Sweden.
a Of which 501,945 were vaccinated and 157,979 never vaccinated.
b Of which 210,747 were vaccinated and 77,124 never vaccinated.
c Of which 291,198 were vaccinated and 87,187 never vaccinated.
d Of which 16,083 were vaccinated and 5,898 never vaccinated.
e Of which 18,553 were vaccinated and 7,279 never vaccinated.
Risk factors for COVID-19 hospitalisation and vaccine effectiveness by subgroups
Of the 47 individuals hospitalised due to COVID-19 in the VE analysis, 28 (60%) had previously been hospitalised for another condition, compared to 66,483 (10%) of the individuals in the rest of the cohort. In addition, of those hospitalised, 15 (32%) had previously been diagnosed with an infection (bacterial infection, tonsillitis, and pneumonia), compared to 21,966 (3.3%) in the rest of the cohort, equal to an adjusted OR for COVID-19 hospitalisation of 14.3 (95% CI [7.7, 26.6], p < 0.001). The VE in this risk group was similar (VE; 88%, 95% CI [58, 96], p < 0.001), as in the total cohort, but with a slightly lower vaccination uptake (73.2% versus 76.2%). In addition, 15 (32%) of the individuals hospitalised due to COVID-19 had previously been diagnosed with cerebral palsy and/or different development disorders, compared to 25,817 individuals (3.9%) in the rest of the cohort, resulting in an adjusted OR for COVID-19 hospitalisation of 12.7 (95% CI [6.8, 23.8], p < 0.001). Again, VE in this subgroup was similar as in the total cohort (VE; 72%, 95% CI [20, 91], p = 0.02), but with a slightly lower vaccination coverage (71.7% versus 76.4%). The NNV with 2 doses to prevent 1 case of COVID-19 hospitalisation in the subgroup of individuals previously diagnosed with infections or development disorders ( N = 46,521) was 1,007.
Vaccine effectiveness against SARS-CoV-2 infection
The VE against SARS-CoV-2 infection of any severity was estimated among 488,441 individuals vaccinated with 2 doses compared to 153,882 never vaccinated controls. Between 1 January 2022 and 28 February 2022, there were 72,627 cases of confirmed SARS-CoV-2 infections. The VE varied by time since the last dose (p < 0.001; Table 4 ), with marginal VE in the total cohort (VE; 4%, 95% CI [3,4], p < 0.001), a low VE for individuals with a second dose no earlier than 1 November 2021 (VE; 27%, 95% CI [25, 29], p < 0.001), and a moderate VE for individuals with a second dose no earlier than 1 January 2022 (VE; 51%, 95% CI, [48, 55], p < 0.001).
10.1371/journal.pmed.1004127.t004
Table 4 VE against SARS-CoV-2 infection of any severity from 7 days onwards after a second dose of vaccine as compared to never vaccinated individuals from 1 January 2022 until 28 February 2022, and by baseline date.
Vaccinated with 2 doses
Never vaccinated
Unadjusted analyses
Adjusted analyses
Number of cases (%)
Number of cases (%)
VE [95% CI]
p
VE [95% CI]
p
Total cohort ( N = 642,323) a
55,101 (11.3%)
17,530 (11.4%)
1 (−1, 3)
0.23
4 (3, 4)
< 0.001
Subgroups by baseline date
≥1 September 2021 ( N = 611,401) b
51,497 (11.0%)
16,037 (11.3%)
3 (2, 5)
< 0.001
12 (10, 13)
< 0.001
≥1 October 2021 ( N = 477,924) c
32,613 (9.2%)
13,655 (11.0%)
18 (17, 20)
< 0.001
17 (15, 19)
< 0.001
≥1 November 2021 ( N = 402,682) d
27,213 (8.8%)
9,743 (10.5%)
18 (16, 20)
< 0.001
27 (25, 29)
< 0.001
≥1 December 2021 ( N = 241,350) e
12,547 (6.8%)
5,288 (9.1%)
27 (24, 29)
< 0.001
41 (39, 43)
< 0.001
≥1 January 2022 ( N = 105,630) f
2,087 (2.6%)
1,279 (4.8%)
47 (43, 50)
< 0.001
51 (48, 55)
< 0.001
CI, confidence interval; VE, vaccine effectiveness.
Adjusted models were adjusted for age, baseline date, sex, and whether the individual was born in Sweden.
a Of which 488,441 were vaccinated and 153,882 never vaccinated.
b Of which 469,627 were vaccinated and 141,774 never vaccinated.
c Of which 354,304 were vaccinated and 123,620 never vaccinated.
d Of which 309,804 were vaccinated and 92,878 never vaccinated.
e Of which 183,462 were vaccinated and 57,888 never vaccinated.
f Of which 79,161 were vaccinated and 26,469 never vaccinated.
Sensitivity analysis
The risk of hospitalisation due to influenza was evaluated in 600,721 individuals given at least 2 doses of COVID-19 vaccine compared to in 186,894 never vaccinated individuals. Between 1 January 2022 and 5 June 2022, a total of 47 individuals (6 per 100,000) were hospitalised due to influenza. The results showed that individuals vaccinated with 2 doses of vaccine did not experience a lower risk of hospitalisation due to influenza as compared to never vaccinated individuals (VE; −10%, 95% CI [−133, 48], p = 0.81), thus indicating an absence of important unmeasured confounding. Of the 117 individuals in the total cohort hospitalised for influenza since the start of the COVID-19 pandemic in January 2020, 1 individual died within 30 days of hospitalisation.
Discussion
In this nationwide study of more than 0.8 million Swedish adolescents, vaccination with at least 1 dose of monovalent COVID-19 mRNA vaccine was not associated an increased risk of hospitalisation for any cause, and vaccination with 2 doses was associated with lower risk of COVID-19 hospitalisation during an omicron-predominant period. However, the absolute risk of COVID-hospitalisation was extremely low, except in subgroups of adolescents that had previously been diagnosed with infections, cerebral palsy, or other development disorders.
Evidence on the safety and effectiveness of COVID-19 mRNA vaccination in adolescents during the omicron-predominant period is limited. In this study, vaccination was not associated with an increased risk of hospitalisation from any cause, or for any of the 30 selected diagnoses. This suggests that COVID-19 vaccination in adolescents is safe. These findings add to, and extend upon, those from a nationwide study in Scotland, reporting no association between vaccination and increased risk of hospital stay for 17 different diagnoses among adolescents [ 5 ]. The present study also estimated that 2 doses of mRNA vaccine had about 76% effectiveness against hospitalisation due to COVID-19. Similar estimates were reported in 2 case–control studies of adolescents conducted in the US and Brazil during an omicron-predominant period, where VE against COVID-19 hospitalisation or death was about 80% [ 6 , 7 ]. However, given their design, these studies were unable to determine how common severe COVID-19 is. Therefore, the very low absolute risk of severe disease observed in the present cohort study is an important finding for decision-making concerning the need for COVID-19 vaccination in adolescents. Overall, only 47 individuals, or 7 in 100,000, were hospitalised due to COVID-19 during a period of 5 months, and none of all adolescents hospitalised due to COVID-19 since the start of the pandemic died within 30 days of hospitalisation. Based on this very low risk of severe COVID-19 in the total cohort, the NNV with 2 doses to prevent 1 case during follow-up was more than 8,000. Interestingly, the absolute risk of being hospitalised due to COVID-19 and influenza was similar, despite that the infection pressure from SARS-CoV-2 during follow-up was more than 100 times higher than that of influenza [ 19 ].
Given the above, it is important to evaluate whether certain groups of adolescents are at higher risk of severe COVID-19, and if so, whether vaccination is associated with similar protection among these individuals. This study identified 2 different clusters of diagnoses that increased the risk of COVID-19 hospitalisation more than 10-fold. The first included previous infections (bacterial infection, tonsillitis, and pneumonia), and the second included cerebral palsy and other developmental disorders. An encouraging finding was that the VE in these subgroups was similar to that in the total cohort, and consequently, the NNV to prevent 1 case of COVID-19 hospitalisation was about 1,000 individuals. It is therefore of concern that vaccination coverage in these risk groups was somewhat lower compared to in the total cohort. Although we are unable to determine the underlying causes for this observation, factors such as recurrent infections interfering with the administration of the vaccine, and fear of adverse events, may have contributed. Taken together, these findings suggest that vaccination of adolescents during the omicron era should primarily be targeting those at high risk of severe disease, such as those with previous infections and development disorders.
Concerning the outcome of SARS-CoV-2 infection of any severity, the results indicated that VE from primary vaccination wanes within a few months, similar to observations made in a few other countries [ 7 – 9 ]. However, there is a lack of data exploring whether booster doses reduce the risk of severe COVID-19 as compared to primary vaccination [ 20 ]. The results for this comparison in the present study showed that COVID-19 hospitalisations during follow-up were once again extremely rare, implying that administration of booster doses to the general population of adolescents may not be warranted at the current stage of the pandemic. Based on these findings, it is of interest that the US and several countries in Europe are recommending booster doses to adolescents during the omicron era [ 21 , 22 ].
The present study has limitations that should be considered. Because of the observational design, conclusions based on the associations found should be made with caution. For example, despite that VE changed marginally before and after adjustment for covariates, there may be other factors that could have influenced the estimates. However, for the analysis of serious adverse events, we evaluated all diagnoses set during hospital stay both before and after the first dose of vaccine, thereby increasing the chance of detecting selection bias that could interfere with the results. In addition, the sensitivity analysis wherein influenza was used as a negative control outcome supported the lack of important confounding. Moreover, because COVID-19 hospitalisation in this population was very rare, it was not possible to estimate the VE of a first booster dose in this cohort with any form of precision. Finally, another limitation is that even though we excluded all individuals with documented prior SARS-CoV-2 infection, the estimates of VE could be underestimated if some of the individuals in the unvaccinated control group had acquired immunity from a prior SARS-CoV-2 infection that was either asymptomatic or undocumented. Strengths of this study include that all the registers used to obtain the data used in the present study have nationwide coverage with virtually zero loss to follow-up. Finally, the study cohort was based on the total population of Swedish adolescents, including 0.8 million individuals aged 11 to 19 years, which increases the possibility to generalise the findings to other countries with similar population structures.
In summary, monovalent COVID-19 mRNA vaccination was not associated with an increased risk for hospitalisation of any cause in adolescents, suggesting that they are safe to use. While vaccination was associated with a reduced risk of COVID-19 hospitalisation during the omicron era, the absolute risk among the general population of adolescents was extremely low. In contrast, in a large proportion of adolescents hospitalised due to COVID-19, certain risk factors were present, and the effectiveness of vaccination in these individuals was similar to in the total cohort. These results indicate that certain vulnerable subgroups of adolescents, rather than adolescents in general, should be prioritised for vaccination.
Supporting information
S1 STROBE Checklist
STROBE Checklist.
(DOCX)
S1 Table
Type of SARS-CoV-2 genotypes based on whole genome sequencing in Sweden during the follow-up period of the present study (week 1–22, 2022).
Data publicly available at the Public Health Agency of Sweden ( https://www.folkhalsomyndigheten.se/smittskydd-beredskap/utbrott/aktuellautbrott/covid-19/statistik-och-analyser/sars-cov-2-virusvarianter-av-sarskild-betydelse/ ).
(PDF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The ease and the rate of genome sequencing are higher today than ever before. Conversely, experimental techniques for protein structure determination are characterized by a much slower rate, entailing that three-dimensional (3D) structure for many potential drug targets will likely be not experimentally solved but predicted instead. For this reason, modeled structures obtained by computational techniques, once validated, will represent an irreplaceable reservoir for modern drug design and development. In this context, in the last 10–15 years, protein kinases have become particularly interesting drug targets for pharmaceutical industry. In cancer research only, over 50% of the current projects are indeed estimated to focus on kinase targets [1] .
There are around 500 members of the protein kinase superfamily encoded by the human genome, whose degree of similarity in the catalytic domain poses many challenges to develop really specific inhibitors targeting the ATP cavity [2] . Still, this similarity is the property that can be also exploited for structural modeling. In turn, such 3D knowledge will be important to predict sensitivity to ATP competitive inhibitors and represents the rationale for the development of more specific compounds (not only type I inhibitors, but also type II inhibitors and type III or allosteric inhibitors) [3] . Importantly, the predictive value of a reliable 3D structure will be also a useful tool to rationally modulate a possible second-line therapy when resistance arises.
Mitogen-activated protein kinases (MAPKs) regulate evolutionarily conserved signaling pathways affecting all essential cellular functions. For this reason, abnormalities in MAPKs signaling also play a critical role in the development and progression of cancer [4] . Extracellular signal-regulated kinase 8 (ERK8, MAPK15) is the last identified member of the MAPK family [5] . It is a proline-directed serine/threonine kinase featuring the distinctive Thr-Xaa-Tyr (TXY) motif in the activation loop [6] , whose post-translational modifications appears to be performed through autophosphorylation [7] . Still, its activity can be further modulated by serum, DNA-damage and human oncogenes [5] , [8] , [9] . Importantly, ERK8 has been implicated in cell transformation [10] , in the protection of genomic integrity [11] , and has been described as a potent regulator of telomerase activity [12] and of the autophagic process [13] . Consequently, it has been proposed as a novel therapeutic target for cancer. Ultimately, ERK8 has been also reported to stimulate the activity of the JUN proto-oncogene [10] and to reduce the activity of different nuclear receptors [14] , [15] . Specific ERK8 inhibitors would thus represent useful tools for the study of its still poorly characterized signaling pathways and could confirm the clinical potential of ERK8 targeting for cancer therapy.
With the aim of developing a 3D structure of ERK8, we took advantage of the similarity of its ATP-binding domain to other MAPKs for structural modeling. Once obtained, we successfully confirmed the reliability of our model by applying a structure-based virtual screening protocol that allowed us to identify molecular scaffolds able to inhibit ERK8 kinase activity. Importantly, we confirmed the binding of such molecules to the ERK8 ATP binding pocket both by ATP competition assays and by using the first reported ERK8 drug-resistant “gatekeeper” mutant. Overall, our experimental data clearly sustain the predictive ability of the generated model for the ERK8 kinase domain and promise its utility in a drug-design perspective.
Materials and Methods
Homology Modeling
All the primary sequences were obtained from UniPROT protein sequence database [16] . Sequence similarity searches were carried out using BlastP [17] . Based on previous homology modeling studies on protein kinases [18] , sequence alignment was performed by CLUSTAL W [19] with a gap open penalty of 10 and a gap extension penalty of 0.05. Also other parameters were kept at their default values. The alignment was also tested with the standard protocol of the T-Coffee method [20] ( Fig. S1 ). The crystal structure of FUS3, ERK2, p38α and CDK2 were from the Protein Data Bank [21] ; entries ID: 2B9F [22] , 1ERK [23] , 1P38 [24] , 1HCK [25] .
The kinase domain of ERK8 (residues 12-345) was obtained using Modeller 9v5 package [26] . The best protein model was chosen on the basis of the DOPE (Discrete Optimized Protein Energy) assessment method as implemented in Modeller. Cartoons were prepared with Pymol software (The PyMOL Molecular Graphics System, Version 1.2r3pre, Schrödinger, LLC).
Molecular Dynamics simulations
The AMBER10 program [27] was used for relaxing protein coordinates by means of energy minimization and molecular dynamics simulations. The ff99 and the GAFF force fields were used for protein and organic ligands, respectively. A rectilinear box of TIP3P water molecules buffering 12 Å was added to solvate the Mg ++ ERK8/ADP complex, while Cl − counterions were added to neutralize the system. Energy minimization was performed by using a combination of the steepest descent and the conjugate gradient algorithms. First, the solvent was energy minimized while keeping solute's coordinates frozen. Then, the solute alone was energy minimized prior to relax the whole solvated complex. The system was then heated to 300 K for 100 ps and the density was equilibrated for 100 ps before producing MD trajectories. The SHAKE algorithm was used to restrain bonds involving H atoms. A time-step of 1 fs was used. The unbound approach for Mg ++ ions was applied. Mg ++ and ADP were restrained by a force constant decreasing from 25 to 0 kcal/mol·Å 2 during the first 1 ns of MD simulation. Unrestrained trajectories were produced for 1.5 ns ( Fig. S3 ).
Pharmacophores
Ligand-based pharmacophore models were generated by using the “Common Feature Pharmacophore Generation” protocol of Discovery Studio 2.5 (Discovery Studio, v2.5. San Diego: Accelrys Software Inc., 2009) on a training set of molecules known to inhibit ERK8 [28] . With this method, two different pharmacophore models were generated, composed of 4 and 5 features, respectively.
Structure-based pharmacophore models were generated by following two different approaches: 1) the Ligandscout 3.0 software [29] was used to derive a pharmacophore model accounting for the interactions performed by ADP toward Mg ++ loaded ERK8, starting from the representative structure obtained by MD simulations (the refined ERK8 structure). Features accounting for the interaction of phosphate groups with the Mg ion were manually removed; 2) the GRID software (GRID 22; Molecular Discovery Ltd.) [30] was used to explore possible molecular determinants involved in ligand binding within the ERK8 active site, as generated by MD simulation. Starting from the representative MD structure, coordinates of ADP and Mg ++ were removed. Selected probe atoms were: O (sp2 carbonyl oxygen), DRY (the hydrophobic core), C3 (methyl CH3 group), C1 = (sp2 CH aromatic or vinyl), N1 (neutral flat NH, e.g. amide), O1 (alkyl hydroxyl OH group), N: (sp3 N with lone pair). Points of energy minimum were converted into pharmacophoric features according to the GBPM approach [31] .
Molecular Docking
Docking calculations were performed by using the GOLD program, version 4.1.2 [32] . The Chemscore scoring function [33] , [34] was used, as it is able to finely reproduce the binding mode of ADP described in the refined ERK8 structure. Default parameters of the Genetic Algorithm (GA) were used, whereas the efficiency was set at 150%. Twenty runs for each ligand were generated by the GA, allowing the early termination after 5 consecutive runs having a root mean square deviation (rmsd) below the tolerance limit of 1.5 Å. The binding site was centered on the side chain of leucine 144 and had a radius of 16 Å.
Expression Vectors
Bacterial expression vector pGEX4T3-ERK8 was generated by subcloning the ERK8 cDNA, obtained by restriction enzyme digestion from the already described pCEFL-HA-ERK8 [8] , into the pGEX4T3 vector in frame with the GST tag. ERK8 gatekeeper mutants were generated using the QuikChange site-directed mutagenesis kit (Stratagene), using pGEX4T3-ERK8 as a template. pGEX4T3-ERK8 kinase dead (KD) was generated subcloning ERK8 K42R cDNA (kindly provided by M. Abe) into the pGEX4T3 vector. The identity and integrity of all the vectors were confirmed by DNA sequencing.
Bacterial expression of GST-fusion proteins
The BL21 Lys strain of Escherichia coli ( E. coli ) was transformed with the pGEX4T3 vectors encoding for the full-length form of ERK8 protein and of the different mutants. Bacterially expressed GST-fusion proteins were purified as previously described [35] . The baculovirus-expressed GST-fusion ERK8 protein is from Carna Biosciences, Inc.
Reagents
Ro-318220 is from Calbiochem; radiolabeled [γ- 32 P] ATP is from PerkinElmer; screened compounds are from Asinex.
In vitro Kinase Assay
Purified GST-fusion protein (50 ng/sample) was incubated 30 min at 30°C in kinase buffer [25 mM HEPES (pH 7.6), 0.1 mM Na 3 VO 4 , 20 mM β-glycerophosphate, 2 mM DTT, 20 mM MgCl 2 ] with 5 µg/sample of MBP (Sigma) as generic substrate, 2.5 µCi [γ- 32 P]ATP and unlabeled ATP (final concentration 5 µM) [28] , unless differently indicated. Candidate compounds dissolved in dimethyl sulfoxide (DMSO) were added as needed (an equal volume of DMSO was added to control samples). Reaction was stopped adding 5X Laemmli buffer and resolved by SDS-PAGE. Dried gels were then exposed to phosphorimager (Typhoon 8600, Molecular Dynamics) and 32 P incorporation on MBP was estimated by densitometry (ImageQuant TL Software, GE Healthcare).
Alternatively, for samples to be analyzed by liquid scintillation counter, reaction mixture was spotted (after incubation at 30°C for 30 minutes in the same kinase buffer described above) on Whatman (P81) 3MM paper and immediately immersed in 1% phosphoric acid to terminate the reaction [36] . After four washes with fresh phosphoric acid, scintillant solution was added (BioFluor, PerkinElmer) and 32 P incorporation of triplicates was measured with a β-counter scintillator (TRI-CARB 2000, Packard). All the assays were replicated twice and the means of the replicates were calculated. For Coomassie staining, before adding radiolabeled ATP, an aliquot of each sample prepared for kinase assay was loaded on SDS-PAGE gel, stained with SimplyBlue SafeStain (Invitrogen), and revealed using the Odyssey Infrared Imaging System (Li-Cor Biosciences).
Western Blots
Proteins in storage buffer (50 mM Tris pH 8.0, 0.5 mM DTT, 0,5 mM EDTA and 10% glycerol) were loaded on SDS-PAGE gel, transferred to Immobilon-P PVDF membrane (Millipore), probed with anti-ERK8 primary antibody (custom preparation) and secondary antibody HRP-conjugated anti-rabbit IgGs (Santa Cruz Biotechnology), and revealed by enhanced chemo-luminescence detection (ECL Plus; GE Healthcare).
Results
Homology Modeling
ERK8 is a 544-amino acid protein with a typical MAPK catalytic domain and a peculiarly long C-terminal extension [5] . The degree of conservation of the catalytic domain, as compared with other MAPKs, is adequate to predict its 3D structure by means of homology modeling procedure, whereas the C-terminal region does not share a significant sequence identity with any known protein, thus impeding any attempt of modeling the whole protein structure with this approach. Therefore, we generated a 3D structure of the ERK8 kinase domain by homology modeling procedure, in the perspective of obtaining a reliable model to next screen for novel ERK8-directed ATP competitive scaffolds. We selected FUS3 ( S. cerevisiae MAPK) and ERK2 X-ray structures as templates, featuring a percentage of sequence identity with ERK8 kinase domain of 45% and 44%, respectively (similarity is 64% and 62%, respectively) ( Fig. 1A ). This degree of homology guaranteed a high-quality model structure [37] . In this context, while we used ClustalW as our leading program to align all the sequences, in order to get a higher confidence on the ability to correctly align motifs and domains, we also tested the multiple sequence alignment with T-Coffee ( Fig. S1 ), another widely used method. As expected, we obtained almost complete identity of the two final alignments in the core catalytic unit.
10.1371/journal.pone.0052011.g001
Figure 1
ERK8 kinase domain model.
(A), Multiple sequence alignment between ERK8 and the selected templates FUS3 and ERK2. Numbering is referred to human ERK8 cDNA sequence as defined in Uniprot accession number Q8TD08. Consensus code: “yellow” indicates positions which have a single, fully conserved residue; “green” indicates conservation between groups of strongly similar properties; “blue” indicates conservation between groups of weakly similar properties. Gatekeeper residue is in bold and indicated by a full black circle. The TEY activation motif is in red (activation loop spans from the DFG motif to the APE motif, residues 155–187). The region in square brackets has been substituted (starting from the position indicated with the dashed red line) with the alignment highlighted in the bottom square that includes p38α. (B), Model of the ERK8 kinase domain (residues 12–345 of the full-length 1–544 protein) obtained by means of homology modeling protocol. Conserved kinase domain features are indicated, β-sheets colored in yellow, α-helices colored in red, loops colored in green, TEY activation motif colored in blue. (C), Superimposition of the same ERK8 model (grey) with the ERK2 template (purple). (D), Evolution of ERK8 structure with the MD refinement. Superimposition of the ERK8 model (grey), used as MD input, with the representative final structure (the refined ERK8 structure) (cyan) obtained after the simulation. (E), Superimposition of the refined ERK8 model (cyan) with the ERK2 template (purple).
While featuring high homology with most of FUS3 and ERK2 kinase domains, the last thirty residues of ERK8 kinase domain did not share such high homology with the selected proteins and also provided lower level of agreement between the two methods of alignment ( Fig. 1A and Fig. S1 ). Therefore, we decided to include a third template, namely p38α, which shares higher homology with ERK8 in this particular segment obtaining the final alignment reported in Figure 1A . Based on this alignment, we next performed homology modeling using Modeller 9v5 [26] . Out of the twenty models that we generated, the top five showed a backbone rmsd lower than 1.5 Å. Accordingly, the model having Mg ++ and ADP within the catalytic active site, endowed with the lower DOPE score was selected as our ERK8 model. The structure showed the overall MAPK topology with an N-terminal and a C-terminal lobe forming the catalytic groove ( Fig. 1B ). Superimposition of the modeled ERK8 structure with corresponding atoms of ERK2 ( Fig. 1C ) and FUS3 ( Fig. S2 ) templates resulted in a rmsd for Cα atoms of <1 and 1.2 Å, respectively. This model was further refined by means of standard in silico procedures to optimize protein-ligand complex structure for ligand design approaches [38] .
Molecular Dynamics Simulation
The structure of ERK8 generated by homology modeling was relaxed by means of a Molecular Dynamics (MD) simulation. With respect to our previous approach to protein kinases study [18] , where the homology model was relaxed by means of energy minimization, here we further used MD simulations to better account for protein flexibility [39] . In detail, the system was first solvated and neutralized by the addition of counterions to resemble a physiological situation. Next, energy minimization was performed to solve bad contacts prior to heating and equilibrating the system, and eventually generating unrestrained MD trajectories for 1.5 ns. As expected for kinase proteins in complex with a ligand, the system was highly stable during the time of simulation [18] , [39] , [40] , especially within the ATP binding site ( Fig. S3 ). Therefore, we considered 1.5 ns a sufficient amount of time to refine this protein structure before performing a structure-based virtual screening. The average structure was calculated, whereas the frame with the lowest rmsd with respect to the average structure was considered representative for the system, and used for structural considerations, as well as for further studies ( Fig. 1D ). Psi and Phi angles of protein residues, taken from the representative structure, were plotted over a Ramachandran map showing that 99% residues were in the most favorable or additional allowed regions, whereas only three residues (1%) were in disallowed regions. However, these latter residues were significantly far from the ATP binding site. The representative structure resulting from MD simulations (hereafter referred to as refined ERK8), that showed a <2 Å rmsd for Cα atoms in comparison to corresponding atoms in ERK2 ( Fig. 1E ), was first validated by self-docking ADP within the catalytic site by using the GOLD docking program (see below). The docking-based binding pose of ADP was very similar (rmsd <1.5 Å) to that found in the refined ERK8, and to the crystallographic ligand pose within the active site of the template FUS3 ( Fig. S2 ) [22] .
Structure-based pharmacophores and molecular docking
Pharmacophore models represent a useful tool to filter large compound libraries on the basis of steric and electronic requirements. In this virtual screening campaign, the generation of structure-based pharmacophores was allowed by the availability of the refined ERK8 structure generated by homology modeling and MD, as reported above ( Fig. 2 ). Pharmacophores were built by using two different strategies: one – the Ligandscout software [29] was used to transform the ADP interaction pattern within the catalytic site of the refined ERK8 strucure into a pharmacophore model, which was composed of three hydrophobic, one H-bond acceptor, one H-bond donor and 19 exclude volume features ( Fig. 3A , left panel). Two – the GRID-based pharmacophore modeling approach (GBPM) [30] , [31] was applied to generate a pharmacophore model which accounted for the space regions endowed with the lowest interaction energy for representative probe atoms within the catalytic site of ERK8 ( Fig. 3A , right panel). This model was composed of two hydrophobic, one H-bond donor, two H-bond acceptor and one excluded volume features. After format conversion, necessary to allow compatibility with Discovery Studio (Accelrys), pharmacophore models were used as 3D queries to filter the Asinex library of compounds (about 400,000 small molecules) using a procedure previously described [41] . The 15,527 compounds that survived to the structure-based pharmacophore filtration were submitted to the following docking step.
10.1371/journal.pone.0052011.g002
Figure 2
Flowchart of the in silico protocol.
Computational steps applied to select all the hit compounds to be tested in vitro . In each set the percentage of success rate refers to the ratio between the number of active molecules and the number of tested molecules in the following experimental screening: purified GST-ERK8 protein (50 ng/sample) was used in kinase assays. Candidate compounds were dissolved in dimethyl sulfoxide (DMSO) and tested at fixed concentration of 50 µM (an equal volume of DMSO was added to control samples). Reactions were resolved by SDS-PAGE and 32 P incorporation on MBP was estimated by densitometry. Molecules were classified as active when the residual kinase activity was less than 50% in comparison to control samples.
10.1371/journal.pone.0052011.g003
Figure 3
Pharmacophore models.
(A), Left panel, Structure-based pharmacophore generated from the Mg ++ loaded ERK8/ADP complex (coordinates were taken from the refined ERK8 structure) by using the Ligandscout software. Right panel, Structure-based pharmacophore generated by the GRID-based pharmacophore modeling approach, starting from the ligand-bound refined structure of ERK8. Features code: HYD = hydrophobic; HBA = H-bond acceptor; HBD = H-bond donor; AROM = aromatic ring; grey spheres are excluded volumes. (B), The two ligand-based pharmacophores generated with the training set of 18 different inhibitors active towards ERK8 (from Bain J, et al., 2007). Features code same as above.
Indeed, we reckoned that the ability to identify ATP competitive compounds by molecular docking, relying on the kinase domain model of ERK8, might represent a good approach to corroborate the structure itself. Accordingly, the refined ERK8 structure was used as a rigid receptor in a docking-based hit identification procedure. The GOLD program (version 4.1.2) [32] was used with the Chemscore scoring function [33] , [34] . Docked compounds were sorted on the basis of their score and the three top-ranking poses of the first 250 molecules were visually inspected within the catalytic active site of ERK8. Molecules showing a clear overlapping between the three poses (rmsd <1.0 Å) were deemed top priority. Based on the predicted binding mode as well as on chemical diversity criteria, the most promising 25 virtual hits were selected for further experimental investigation ( Fig. 2 ).
Ligand-based pharmacophores
Beside the structure-based virtual screening approach, a ligand-based strategy for hit identification was also attempted, based on the activity data of 18 pan-kinase inhibitors ( Fig. S4 ) toward ERK8 [28] . In general, although results of structure- and ligand-based approaches are not superimposable, combination of different strategies could enhance the probability to identify chemically diverse molecular scaffolds, especially when performing the first round of a computer-aided drug design approach toward a given target system.
Two ligand-based common feature pharmacophore models were generated by using Discovery Studio 2.5, starting from known ERK8 inhibitors [28] . The first model was composed of two hydrophobic, one H-bond acceptor and one H-bond donor features, whereas the second model was composed of two hydrophobic, one aromatic ring, one H-bond donor and one H-bond acceptor features ( Fig. 3B ). Notably, these pharmacophores are significantly different from each other and from structure-based models. Ligand-based pharmacophores were used as 3D queries to filter the Asinex database, the resulting compounds were inspected for chemical diversity, and the selected 32 putative hits were submitted to experimental investigation ( Fig. 2 ).
Experimental Screening
To assess the effect of in silico -selected compounds on ERK8 catalytic activity, we generated an N-terminal, GST-tagged form of the full-length protein in E. coli . In a classical kinase assay with radiolabeled ATP, bacterially expressed GST-ERK8 was found to be constitutively active (data not shown), as previously described [7] . We, therefore, monitored ERK8 ability to phosphorylate a typical substrate, Myelin Basic Protein (MBP) [28] in the presence of each candidate compound at a fixed concentration of 50 µM. As a positive control of ERK8 catalytic inhibition, we used the ATP competitive inhibitor Ro-318220. This molecule, originally developed as a protein kinase C (PKC) inhibitor [42] , has been already proven to potently inhibit ERK8 [7] , [12] , [13] . By this approach, we tested the 32 compounds obtained from the ligand-based virtual screening and the 25 coming from the structure-based virtual screening. In order to classify the analyzed compounds in a binary fashion, we set a threshold at 50% of residual kinase activity with respect to control samples containing no inhibitors: molecules capable of an effect equal or higher than the threshold were labeled as “active”, “not active” otherwise. The structure-based derived subset was populated by a significant percentage (20%) of “active” molecules as compared to the ligand-based approach that showed only 3% of success rate ( Fig. 2 ). Interestingly, besides the “active” molecules, a high percentage of the total tested compounds showed at least a partial ability to decrease ERK8 activity ( Table S1 ). This finding, therefore, indicates that the modeled 3D kinase domain of the refined ERK8 has a reliable structure and that, in this specific case, the structure-based approach is to be preferred over the ligand-based one to identify molecular scaffolds potentially interesting for further optimization.
In vitro characterization of selected scaffolds
ITT45, ITT53 and ITT57 ( Fig. 4A–B and Table S1 ), the most potent identified inhibitors of ERK8 catalytic activity, were selected among “active” molecules also considering their significant chemical diversity. Indeed, the Tanimoto similarity indexes calculated by using ECFP_6, MDLPublicKeys and FCFP_6 sets of fingerprints were below 0.1, 0.5 and 0.2 each other, respectively. Moreover, to the best of our knowledge, this is the first time that these molecular scaffolds are proposed as ERK inhibitors. The three compounds were submitted to an in-depth in vitro characterization aimed at fully validating the refined ERK8 structure and, consequently, the overall computational protocol. For this purpose, we confirmed the inhibitory effect of the selected compounds, as found during the preliminary screening, showing a calculated percentage of residual kinase activity of 36±7%, 35±8% and 32±5% for ITT53, ITT45 and ITT57, respectively ( Fig. 4C and Fig. 4D ).
10.1371/journal.pone.0052011.g004
Figure 4
Effect of selected molecular scaffolds on bacterial and eukaryotic GST-ERK8.
(A), Molecular structure of selected compounds. (B), Binding mode of each compound as obtained after the molecular docking step. The ITT molecules are showed as sticks and colored by atom type. ERK8 protein structure is represented by secondary structure cyan elements. (C), Samples of GST-ERK8 from E. coli with the indicated concentration of inhibitors were subjected to kinase assay. Reactions were resolved by SDS-PAGE and 32 P incorporation on MBP was estimated by densitometry (upper panel). Coomassie staining verified that equal amounts of substrate were loaded (lower panel). (D), The average results of three independent experiments done in duplicate ± SD are plotted. (E), Samples of GST-ERK8 Bac with the indicated concentration of inhibitors were subjected to kinase assay. Reactions were resolved by SDS-PAGE and 32 P incorporation on MBP was estimated by densitometry (upper panel). Coomassie staining verified that equal amounts of substrate were loaded (lower panel). (F), The average results of three independent experiments done in duplicate ± SD are plotted.
Although widely used as a source of protein kinases for studying their activity and response to inhibitors [28] , it is well established that E. coli -based bacterial expression systems do not necessarily ensure a correct post-translational processing of heterologously expressed proteins, often required for fully and correctly controlled activity. In this regard, the baculovirus-based insect expression system has been used to produce recombinant proteins, because insect cells can perform correct post-translational modification of heterologous proteins [43] . Based on this information, we next decided to test our compounds on GST-ERK8 produced from baculovirus-infected insect cells. GST-ERK8 purified through this system (hereafter named GST-ERK8 Bac ) was then used in classical kinase assays to test the three scaffolds. ITT53 and ITT57 efficacy was almost equivalent to that observed with bacterially expressed protein (i.e., 32±6% and 26±6% of residual kinase activity, respectively), whereas, surprisingly, no effect was obtained by using ITT45 ( Fig. 4E and Fig. 4F ). This observation showed the importance of confirming the results obtained from bacterially-produced ERK8 also by using an eukaryotic expression system and also suggested a potentially specific role for post-translational modifications in the regulation of ERK8 activity. Based on this evidence, we decided to perform additional characterization only on ITT53 and ITT57, which gave best chances to affect fully active ERK8. Dose/response curves were then carried out on GST-ERK8 Bac and estimated half-maximal inhibitory concentration (IC 50 ) values for ITT53 and ITT57 were 27 µM (95% confidence interval 8–84 µM) and 17 µM (95% confidence interval 6–48 µM), respectively ( Fig. 5A ). Next, we tested ITT53 and ITT57 in a competition assay with different concentrations of ATP. As reported in Figure 5B , increasing ATP concentrations induced a decrease in the inhibitory action of ITT53 and ITT57. This profile was compatible with a competitive mechanism with the natural substrate, and indirectly supported the reliability of our ERK8 kinase domain model that has been used to select these ATP binding pocket small molecule inhibitors. As an additional control, the ATP competition assay profile of ITT53 and ITT57 displayed the same characteristics of the positive control Ro-318220 ( Fig. 5B ), already described as an ATP competitive kinase inhibitor [42] . Altogether, these results indicate that the two molecules are able to inhibit the kinase activity of the full-length protein expressed both in E. coli and in a eukaryotic system. The IC 50 values represent a valuable starting point for molecules obtained from a first “ in silico - in vitro ” screening process. They also show a behavior that is compatible with a competition mechanism with the natural substrate ATP.
10.1371/journal.pone.0052011.g005
Figure 5
In vitro characterization.
(A), Dose/response curves for ITT53 and ITT57 on GST-ERK8 Bac . Results are reported as residual MBP phosphorylation levels compared with the control (DMSO). The average results of two independent experiments done in triplicate ± SD are plotted with the curve-fitting PRISM software (GraphPad). The concentration of drug that inhibited activity by 50% (IC 50 ) is shown. (B), ITT53, ITT57 and Ro-318220 ATP competition assay on GST-ERK8 Bac . Inhibition values are reported as percentage of residual MBP phosphorylation levels (i.e., residual kinase activity) compared with the control (DMSO). Results for the two indicated concentrations of ITT53, ITT57 and Ro-318220 (top, middle, bottom panel, respectively) at four different ATP doses were plotted. The average results of two independent experiments done in triplicate ± SD are plotted.
Binding Mode Analysis
The mechanism of action assessed by means of the previously described ATP competition assay was a solid indication that the selected molecules occupy the ATP cavity, as predicted. Moreover, being the demonstration of the binding mode particularly important to solidly validate our computational assumptions, we decided to deal with this issue also by an alternative approach. We took advantage of the so-called “chemical genetic analysis”, first performed by Shokat and co-workers [44] through the generation of engineered kinases sensitive to inhibitors and ATP analogues, which are not equally effective on wild type proteins. Indeed, a single residue in the ATP binding pocket of protein kinases, termed the “gatekeeper”, has been shown to control sensitivity to a wide range of small molecule inhibitors [45] – [47] . Therefore, we hypothesized that perturbing the ATP cavity of ERK8 by mutating its “gatekeeper” residue would affect the affinity of the ITT53 and ITT57 molecules only in case they bind to this specific region. This would confirm an ATP competitive mechanism of action for these hit compounds, as predicted by calculations and previously suggested with in vitro assays ( Fig. 4B and Fig. 5B ). Our analysis showed that the gatekeeper position of ERK8 (residue 92) is occupied by a phenylalanine (Phe, F) ( Fig. 6A ), differently from the templates used for the modeling (a glutamine for both FUS3 and ERK2) but similarly, for example, to members of the CDK family. Indeed, superimposition of our model to a high resolution CDK2 X-ray structure confirmed the local structural similarity in the gatekeeper surroundings between these two proteins ( Fig. 6B ), further supporting the reliability of our computational model. Based on this analysis and on previous studies of the gatekeeper position of CDK2 (CDK2_F80G) [48] , [49] , we decided to generate an ERK8 gatekeeper mutant substituting the Phe92 with a smaller glycine (Gly, G) residue. Unexpectedly, ERK8_F92G showed an almost complete loss in catalytic activity ( Fig. 6D ). Indeed, thanks to the broad use of the chemical genetic analysis, evidences about the ability of many kinases to tolerate dramatic mutations at the gatekeeper position have accumulated but, more recently, it has also become clear that many kinases do not tolerate such perturbations [50] , [51] . These kinases (roughly 30% of tested kinases), that undergo loss of catalytic activity and/or cellular function upon introduction of space-creating gatekeeper mutations, have been classified as “intolerant”, unlike the “tolerant” ones that, in these conditions, are able to maintain their catalytic functions [51] . Although the selected structural templates FUS3 and ERK2, and the “gatekeeper-related” CDK2 protein all belong to the group of “tolerant” kinases, the ERK8_F92G loss of activity suggests that this protein could, therefore, represent a new “intolerant” kinase. In order to better investigate this aspect, we generated other three gatekeeper mutants substituting the phenylalanine with chemically different residues, namely another small amino acid, alanine (Ala, A), and two bulkier amino acids such as isoleucine (Ile, I) and tyrosine (Tyr, Y). All the generated single-point ERK8 mutants in the gatekeeper position, F92G, F92A, F92I, and F92Y, were expressed as N-terminal GST-tagged proteins in E. coli with the same procedure applied to the wild type protein. Purified fusion proteins were also controlled for correct identity via western blot analysis using a specific anti-ERK8 antibody, also demonstrating the expected molecular weight as compared to the wild type and kinase dead (ERK8_KD) proteins ( Fig. 6C ). Interestingly, all but the ERK8_F92I mutant showed a barely detectable basal activity in the kinase assay when it was measured in comparison to the wild type protein ( Fig. 6D ). These results support the classification of ERK8 as a new “intolerant” kinase to the mutation of the gatekeeper residue to glycine or alanine.
10.1371/journal.pone.0052011.g006
Figure 6
Gatekeeper mutants.
(A), Multiple sequence alignment of gatekeeper region among different members of the MAPK and CDK families of kinases. The position corresponding to the gatekeeper residue is highlighted. (B), Superimposition of the refined ERK8 structure (cyan) and CDK2 (magenta) X-ray structure. (C), Western Blot control of GST-fusion proteins from E. coli . Each lane was loaded with 100 ng of purified protein. ERK8_KD sample (lane 6) is a point mutant on the conserved lysine (Lys, K) in position 42 to arginine (Arg, R). (D), Representative kinase assay blot of gatekeeper mutants (200 ng/sample of purified protein) (upper panel). Reactions were resolved by SDS-PAGE and 32 P incorporation on MBP was estimated by densitometry. Coomassie staining verified that equal amounts of substrate were loaded (lower panel). Quantification of kinase activity in comparison to WT, as scored by MBP phosphorylation, from three independent experiments is reported in the lower panel.
Being our goal the confirmation of the selected scaffolds binding mode and also the evaluation of the predictive value of ERK8 model, we next decided to focus our analysis only on the partially active ERK8_F92I mutant. Therefore, we generated the ERK8_F92I model with the same computational protocol applied to ERK8 wild type (WT). Then, we chose to analyze the interaction pattern of both modeled structures with one of our selected compounds. To this purpose, both ERK8_WT and ERK8_F92I, in complex with ITT57, were relaxed during 2 ns of unrestrained MD. The ITT57 delta energy of binding was calculated by means of the MMPBSA approach [52] , showing that the substitution of the Phe92 aromatic ring with the aliphatic chain of isoleucine is associated to a higher delta energy of interaction. From a structural point of view, analysis of representative MD structures ( Fig. 7A ) revealed that the overall binding mode of ITT57 within the catalytic site is the same in both protein structures, although the methoxyphenyl moiety seems to be less buried into the lipophilic pocket of the ERK8_F92I mutant than into the wild type one. These differences, together with the possible enthalpy gain coming from a pi-pi stacking interaction between the Phe92 gatekeeper residue of ERK8_WT and ITT57, would suggest a less profitable interaction of ITT57 with the ERK8_F92I mutant than with the WT. Therefore, based on this in silico analysis, we could predict a decreased efficacy of ITT57 on the ERK8_F92I mutant. We next performed the conclusive experimental test to assess the binding mode of ITT53 and ITT57 by comparing ERK8_F92I and ERK8_WT inhibition profiles, this approach also giving us the possibility to challenge the theoretical prediction about ERK8_F92I increased resistance to ITT57. As a control, we also tested, in the same kinase assay, the Ro-318220 ATP competitive inhibitor [42] . As expected, we observed that the inhibitory activity of Ro-318220 was affected by the presence of the F92I mutation in ERK8, in particular being more active on the wild type than on the mutated ERK8 protein ( Fig. 7B and Fig. S5 ). Similarly, ITT53 and ITT57 molecules were also significantly more active on ERK8_WT than on ERK8_F92I, proving that this kind of perturbation at the gatekeeper position gives rise to an ERK8 protein more resistant to inhibition triggered by small molecules that occupy the catalytic pocket. More importantly, this result both emphasizes the predictive ability of our model and confirms the competition assays data, ultimately showing that the selected scaffolds bind ERK8 in the ATP cavity.
10.1371/journal.pone.0052011.g007
Figure 7
A resistant ERK8_F92I mutant confirms the predicted ATP pocket-binding mode.
(A), Representative structures from MD simulation of the complex between ITT57 and both ERK8_WT (left panel) and ERK8_F92I mutant (right panel). The residue at position 92 is labeled and showed as sticks. The ITT57 ligand is showed as sticks. Protein residues and ligand atoms are colored by atom type. (B), GST tagged ERK8_WT and ERK8_F92I proteins (200 ng/sample) were used in kinase assays in presence of the indicated concentrations of ITT53, ITT57 and Ro-318220 molecules. Using the paper-spotted kinase assay technique, we quantified and normalized the activities of the WT and of the mutant protein. MBP phosphorylation levels were evaluated by β-counting protocol of triplicates and results expressed as percentage of residual kinase activity compared with control samples. Significance (p-value) was obtained by one-way ANOVA test. Asterisks were attributed for the following significance values: p<0.05 (*), p<0.01 (**), p<0.001 (***).
Discussion
Over the past decade, protein kinases raised as the pharmaceutical industry's most popular drug targets, especially in the field of cancer. In particular, interest in MAPKs has recently exploded [53] , [54] . Also, chemical inhibitors, both the clinically relevant ones and many other molecules that do not reach the latest stages of approval procedures, have become invaluable reagents for studying the physiological functions of their target protein. However, the ATP binding sites with which most inhibitors interact, are highly conserved throughout the large kinase family, raising an acute specificity problem. Indeed, although this may sometimes be an advantage when it comes to clinical effect, most commercially available chemical inhibitors of protein kinases are poorly specific [28] .
Knowledge about ERK8 targets and downstream effectors and, ultimately, about its biological functions is still limited. Among currently available data suggesting a role for this kinase in normal and aberrant cell proliferation [10] , the recent observation about ERK8 being a potent regulator of telomerase activity [12] clearly shows the possibility of a beneficial use for pharmacological ERK8 inhibition. More specifically, it is shown that ERK8 pharmacological inhibition results in a significant decrease in telomerase activity, which tumor cells often activate to bypass replicative senescence and gain unlimited proliferation ability. In particular, Ro-318220 has been used to confirm the ERK8-dependent telomerase activity in transformed cell lines [12] . Altogether, these data strongly support the clinical potential of ERK8 inhibition.
The gap of knowledge to fill up about ERK8 signaling and the absence of a crystal structure led us to apply an in silico protocol to generate a 3D model of the ERK8 kinase domain and to screen, first in silico and then in vitro , molecular scaffolds to validate the ERK8 structure itself. As described above, whereas the structure-based derived subset showed a significant percentage (20%) of active molecules, the ligand-based approach only showed a 3% success rate. Although disappointing, such a low success rate indicated that the ERK8 inhibition observed with the other set of molecules was not a random effect based on an intrinsic high propensity of ERK8 for chemical inhibition.
Next, the structure-based model provided a valuable guidance in the screening of ATP competitive inhibitors, allowing us to identify a high percentage of in vitro active compounds (though with different effectiveness), despite the limited number of screened molecules. This result, together with the confirmation of their ability to target the ATP binding pocket, demonstrates that the generated ERK8 model is a reliable tool for the screening of novel inhibitors. These scaffolds, for the first time specifically selected towards the ERK8 ATP pocket, in turn can be worthy of further chemical optimization to increase and finely tune their potency.
Ultimately, this work also led us to the identification of an ERK8 drug-resistant mutant, namely ERK8_F92I, that not only proved the expected binding mode of our molecules, but could pave the way to the use of synthetic ATP analogues for the identification of ERK8 substrates [55] . Thanks to the study of gatekeeper mutations, we also demonstrated that space-creating mutations (i.e., glycine or alanine replacing the natural bulky phenylalanine) almost completely abolish kinase activity. Hence, we propose ERK8 as another member of the so-called “intolerant” group of kinases, as defined by Shokat and co-workers [51] . Altogether, our results suggest that the generated model will be an important resource for the identification of specific inhibitors for the ERK8 MAP kinase.
Conclusions
A well-developed body of knowledge identifies different MAPKs as critical regulators of cell proliferation and human cancer. Several recently developed pharmacological inhibitors targeting MAPKs have been effective in animal models and have therefore advanced to clinical trials for the treatment of inflammatory diseases and cancer. Still, although the specific ERK8 member of this family has been proposed as a novel potential therapeutic target for cancer, the lack of its experimental structure currently limits the possibilities to efficiently look for pharmacological compounds specifically targeting this kinase. As a consequence, the development of new therapeutic strategies based on molecular and pharmacological intervention on ERK8 functions is currently impaired. We believe that our results show that the 3D ERK8 model we generated is a reliable tool to be exploited in a drug-design perspective. Consequently, future work about this atypical MAPK will largely benefit of the identification of a sufficiently specific inhibitor to dissect its signaling functions and to further validate its potential as a novel therapeutic target in cancer treatment.
Supporting Information
Figure S1
T-Coffee multiple sequence alignment. Multiple sequence alignment between FUS3, ERK2 and ERK8 obtained with T-Coffee software (standard protocol). Consensus code: “yellow” indicates positions which have a single, fully conserved residue; “green” indicates conservation between groups of strongly similar properties; “blue” indicates conservation between groups of weakly similar properties. The TEY activation motif is in red.
(DOC)
Figure S2
ERK8 model and FUS3. (A), Superimposition of the ERK8 model (grey) with the FUS3 template (blue). (B), Superimposition of the refined ERK8 structure (cyan) with the FUS3 template (blue). (C), ADP binding mode within the catalytic pocket. In green sticks is showed the crystallographic binding mode of ADP within the FUS3 template. In grey sticks is showed the ADP binding mode in the ERK8 structure refined by molecular dynamics. In magenta sticks is showed ADP binding mode as obtained by self-docking ADP toward the refined ERK8 structure by the GOLD docking program. Heteroatoms are colored by atom types.
(DOC)
Figure S3
Stability of MD. Root mean square deviation (rmsd) of each frame with respect to the first frame of unrestrained MD, over time.
(DOC)
Figure S4
Ligand-based approach: the training set. List of compounds used to generate the two ligand-based pharmacophores (from Bain J, et al., 2007).
(DOC)
Figure S5
Kinase Assay of GST-tagged ERK8 proteins on autophosphorylation. (A), Representative kinase assay blot of WT and different ERK8 mutants (200 ng/sample of purified protein) (upper panel). Reactions were resolved by SDS-PAGE and 32 P incorporation on GST-ERK8 proteins themselves was estimated by densitometry. Quantification of kinase activity in comparison to WT, as scored by autophosphorylation, from three independent experiments is reported in the lower panel. (B), GST tagged ERK8_WT and ERK8_F92I proteins (200 ng/sample) were used in kinase assays in presence of the indicated concentrations of ITT53, ITT57 and Ro-318220 molecules. Using the paper-spotted kinase assay technique, we quantified and normalized the activities of the WT and of the mutant protein. Autophosphorylation levels were evaluated by β-counting protocol of triplicates and results expressed as percentage of residual kinase activity compared with control samples.
(DOC)
Table S1
Experimental Screening results for the structure-based selected molecules. Ranking of all the molecules obtained with the structure-based approach: the percentage (in brackets) of residual kinase activity is reported for all the compounds. (a) residual kinase activity with respect to control samples containing no inhibitors (b) ratio between the number of active molecules and the number of tested molecules (c) success rate obtained for threshold activity up to 55% (d) success rate obtained for threshold activity up to 65%.
(DOC)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Nitric oxide synthases (NOSs) are the enzymes responsible for ·NO production, a biological signaling molecule involved in the control of cardiovascular, immune and nervous system physiology [1] . Neuronal NOS (nNOS), is larger than both its endothelial (eNOS) and inducible (iNOS) counterparts, mostly due to a ∼300 amino acid N-terminal extension containing a PDZ domain (residues 14-105) [2] , [3] . The association of this N-terminal sequence with other neuronal proteins determines nNOS enrichment at post-synaptic densities [4] , [5] . Peptide library as well as yeast two-hybrid screens revealed that the PDZ module of nNOS displays a clear binding preference for cellular proteins with C-terminal acidic amino acids at -2 and -3 positions. In fact, proteins with a -Gly-(Asp/Glu)-X-Val C-terminus were proposed as tight binders of nNOS PDZ domain [6] , [7] . Soon afterwards, a protein referred to as CAPON (C-terminal PDZ ligand of nNOS), displaying a C-terminal -Glu-Ile-Ala-Val motif and highly enriched in the brain was reported to bind to the PDZ domain of nNOS [8] . In a similar fashion, the acidic C-terminus of other neuronal proteins such as melatonin receptor (-Val-Asp-Ser-Val), phosphofructokinase-M (-Glu-Ala-Ala-Val) and NIDD (-Glu-Asp-Ile-Val) have been reported as ligands of the PDZ domain of nNOS [9] – [11] . In addition, the nNOS beta hairpin that extends the preformed PDZ domain mediates the formation of PDZ/PDZ dimers of nNOS/PSD-95 and nNOS/α1-syntrophin in neuronal cells [12] , [13] . The postsynaptic density protein PSD-95 binds to the C-terminus of ionotropic N-Methyl-D-Aspartate (NMDA)-type of glutamate receptors (NMDARs) through PDZ1 and to nNOS through PDZ2 hence forming a ternary complex in neurons [14] , [15] . Therefore, nNOS activation is enhanced after physiological or pathological NMDARs stimulation leading to ·NO production [16] – [18] . We have previously reported that in cortical neurons and brain, NMDARs also associate with kinase D interacting substrate of 220-kDa (Kidins220) [19] , a protein also known as ankyrin-repeat rich membrane spanning (ARMS). Kidins220/ARMS is a neuronal enriched transmembrane protein identified as the first substrate of protein kinase D1 (PKD1) [20] and as a downstream effector of neurotrophin receptors [21] .
Protein kinase D1 (PKD1) belongs to a family of phorbol ester/diacylglycerol-stimulated Ser/Thr kinases constituted by two additional members, PKD2 and PKD3 [22] . PKDs play multiple roles in different cell types and tissues, from primary cellular functions such as protein traffic, adhesion, migration, proliferation, survival and death, to complex processes such as immune regulation, cardiac hypertrophy, angiogenesis and cancer [22] . In addition, PKD1 has been involved recently in specific neuronal functions such as axon formation, sorting of dendritic proteins and dendritic arborization [23] – [25] . All PKD isoforms bear a cysteine-rich domain (CRD) that binds diacylglycerol and phorbol esters, an autoinhibitory pleckstrin homology domain (PH), followed by the catalytic domain [22] . Importantly, we discovered a unique distinctive type I PDZ-binding sequence or PDZ-ligand at the very C-terminal end of PKD1 and PKD2 that is absent in PKD3 [26] . In PKD1, kinase activation results in autophosphorylation of Ser 916 located at -2 position within its PDZ-binding motif (-Val 915 - Ser 916 -Ile 917 -Leu 918 ), which in turn controls Kidins220/ARMS transport and localization at the neuronal plasma membrane [26] , [27] . These previous results led us to propose a model where the negative charge of the incorporated phosphate at this position in active PKD1 could mimic an acidic residue that could change the binding affinity of its PDZ-ligand for different PDZ proteins, regulating this way Kidins220/ARMS traffic [26] . Knowing that nNOS and PKD are spatially enriched in postsynaptic densities and dendrites, and that nNOS PDZ domain binds preferentially PDZ-ligands bearing acidic residues at -2 position, we hypothesized that the phosphorylated PDZ-binding motif of active PKD1 could be a bona-fide binding partner for the PDZ domain of nNOS. Herein, we have explored whether PKD1 activation could result in a direct interaction with nNOS and also if nNOS could be a substrate for PKD1, analyzing the functional consequences. Our studies show that PKD1 activation enhances its association with nNOS and favors their subcellular colocalization. However, contrary to our initial hypothesis, this association is independent of its PDZ-ligand but depends on the PH domain of PKD1. In addition, we demonstrate that PKD1 activates nNOS by phosphorylating the activatory residue Ser 1412 , leading to increased ·NO production, hence establishing a novel role of PKD in the regulation of ·NO synthesis.
Materials and Methods
Ethics Statement
Animal procedures were approved by “Consejo Superior de Investigaciones Científicas” - CSIC Ethics Committee and performed in compliance with European Directive 2010/63/EU. Animals used were kept to a minimum, they were sacrificed by deep anesthesia, and all efforts were made to minimize suffering.
Cell Lines, Reagents and Antibodies
HEK293T, COS-7, and PC12 cells were obtained from American Type Culture Collection ATCC (Manassas, VA, USA). Phorbol-12, 13-dibutyrate (PDBu), 8-Br-cGMP, L-N G -nitroarginine methyl ester (L-NAME), N-Methyl-D-aspartate (NMDA), glycine, cytosine β-D-arabino furanoside (AraC), poly-L-lysine, L-laminin, Protein A/G-Sepharose, 2′,5′-ADP–Sepharose, adenosine 2′(3′)-monophosphate mixed isomers, DNA single stranded from salmon testes for hybridization, and 5-Bromo-4-chloro-3-indolyl β-D-galactopyranoside (X-Gal) and 4,5-Diaminofluorescein diacetate (DAF2-DA) were from Sigma Co. (St. Louis, MO, USA). Nerve growth factor was from Alexis Corp. (San Diego, CA, USA). Ni-NTA resin was from Qiagen (Chatsworth, CA, USA). L-Arginine and Gö6976 were purchased from Calbiochem (Merck Millipore, Darmstadt, Germany). [γ 32 P]-ATP (370 MBq/ml) was from PerkinElmer, Inc. (Boston, MA, USA). Mouse monoclonal anti-Myc, anti-GST and rabbit polyclonal antibodies recognizing total PKD1/2 and phospho-Ser 916 were from Cell Signaling Technology (Beverly, MA, USA). Anti-β-tubulin I monoclonal antibody was purchased from Sigma and rabbit polyclonal anti-neuronal specific enolase (NSE) from ICN Biomedicals (Costa Mesa, CA, USA). We produced an antibody against nNOS immunizing rabbits with purified rat nNOS following standard procedures. Rabbit polyclonal anti-nNOS-phospho-Ser 1412 was purchased from Upstate-Merck Millipore (EMD Millipore Corporation, Billerica, MA, USA). Mouse monoclonal antibody recognizing total VASP and rabbit polyclonal antibody anti-VASP-phospho-Ser 239 were from Santa Cruz Biotechnology (Santa Cruz, CA, USA). Rabbit polyclonal anti-GFP was obtained from Invitrogen-Life Technologies (Carlsbad, CA, USA). Horseradish peroxidase-conjugated anti-rabbit and anti-mouse secondary antibodies were from General Electric (Fairfield, CT, USA). Oligonucleotide primers were from Invitrogen-Life Technologies (Carlsbad, CA, USA). All other reagents were from standard suppliers or as indicated in the text.
Identification of PKD1-phosphorylated residue in nNOS by mass spectrometry or MALDI TOF/TOF
In vitro kinase reactions after phosphorylating nNOS by a recombinant protein containing the active catalytic domain of PKD1 fused to GST (GST-PKD1-cat) were digested with trypsin and analyzed by HPLC followed by MALDI TOF/TOF and peptide fragmentation and de novo sequencing in the Proteomic Studies Unit (Unidad de Proteómica; Facultad de Farmacia Parque Científico de Madrid, Universidad Complutense de Madrid, Madrid, Spain) following standard procedures. MALDI-TOF MS analysis was performed in a 4800 Proteomics Analyzer MALDI-TOF/TOF mass spectrometer (Applied Biosystems, MDS Sciex, Toronto, Canada). The MALDI-TOF/TOF operated in positive reflector mode with an accelerating voltage of 20000 V. Selected peptides, were subjected to MS/MS sequencing analyzes using the 4800 Proteomics Analyzer (Applied Biosystems, Framingham, MA). Suitable precursors from the MS spectra were selected for MS/MS analysis with CID on (atmospheric gas was used) 1 Kv ion reflector mode and precursor mass Windows ± 4 Da. The plate model and default calibration were optimized for the MS/MS spectra processing. De novo sequencing from fragmentation spectra of peptides was performed using De novo tool software (Applied Biosystems), tentative sequences were manually checked and validated.
Yeast two hybrid screens
We used plasmids containing GAL4 binding domain that were confronted with plasmids containing the GAL4 activation domain as previously described [28] . Double transformants were plated in Leu − /Trp − /His − SD plates in the presence of 12 mM 3-amino triazole (TDO plates) as well as in Leu − /Trp − /His + . Interacting proteins expressed within the same yeast resulted in colonies that could rescue growth in the absence of His. These colonies were subsequently screened in the X-Gal assay. Blue colonies corresponded to a positive interaction whereas white colonies corresponded to absence of interaction. The complete PKD1 active catalytic domain (Gly-557 to Leu-918) or shorter C-terminal sequences (Pro-591 to Leu-918) containing the PDZ-binding motif of wild-type PKD1 or the phospho-mimetic mutant PKD1-Ser 916 Glu (PKD1 S916E , described in Sanchez-Ruiloba et al. [26] ) were PCR-amplified using primers carrying NdeI/EcoRI sites and subcloned into pGBKT7, in frame with the DNA-binding domain of GAL4. PKD1 baits were used to perform one to one yeast two hybrid assays against two nNOS constructs that were PCR-amplified using primers carrying EcoRI/SalI sites and subcloned into pGAD, in frame with the activation domain of GAL4: one shorter including the nNOS PDZ domain (aa 1–102) and a longer one (aa 1–131) that includes the C-terminal extension peptide of the nNOS PDZ domain that represents a relatively independent structural unit in mediating the interaction between nNOS and PDZ domain-containing proteins including PSD-95 and α1-syntrophin [12] . As controls both nNOS constructs were confronted with α1-syntrophin: the long nNOS construct as positive control and the short construct that lacks of the β-hairpin “finger” as a negative control. The C-terminus of rat CAPON (sequence ELGDSLDDEIAV) was cloned in the pGBT9 plasmid between the EcoRI and SalI sites.
Cell culture and transfection
HEK293T or COS-7 cells were cultured in Dulbecco's modified Eagle's medium (DMEM; Invitrogen-Life Technologies; Carlsbad, CA, USA), supplemented with 10% (v/v) foetal calf serum, and 2 mM glutamine at 37°C in a humidified atmosphere containing 5% CO 2 . HEK293T cells were seeded at 60% confluence for transfection using Lipofectamine2000 reagent (Invitrogen-Life Technologies; Carlsbad, CA, USA), according to the manufacturer's specifications, and collected for processing 48 h later. Cell were transfected with empty vector pEFBOS-GFP or containing GFP fused to PKD1 wild-type (PKD1), kinase-inactive (the single mutant Asp 733 Ala; PKD1ki), constitutively active (the double mutant Ser 744/748 Glu; PKD1ca), PDZ-ligand mutants (PKD1-Ser 916 Glu/PKD1 S916E ; PKD1-Ser 916 Ala/PKD1 S916A , or PKD1 lacking its PDZ-ligand/PKD1 ΔSIL ) and deletion mutants lacking the PH-domain (PKD1 ΔPH ) or the CRD domain (PKD1 ΔCRD ) that have been used previously [26] , [29] , [30] . Expression vectors for Myc-tagged wild-type rat nNOS (nNOS) and the point mutant nNOS-Ser 1412 Ala (nNOS S1412A ) were kindly provided by Dr. G. A. Rameau and Dr. E. B. Ziff [31] . When required, HEK293T cells were treated with PDBu (200 nM) for 15 min, 8-Br-cGMP (100 µM) for 30 min or L-NAME (100 µM) for 24 h, as specified in the text. PC12 cells were cultured at 37°C in Dulbecco's modified Eagle's medium (DMEM; Invitrogen-Life Technologies; Carlsbad, CA, USA) supplemented with 7.5% fetal calf serum, 7.5% horse serum, and 2 mM glutamine in a humidified atmosphere containing 5% CO 2 . Cells were treated with nerve growth factor (75 ng/ml) for 2 days post-transfection. For transfection and immunofluorescence, HEK293T and PC12 cells were seeded at 50–60% confluence on poly-L-lysine (10 µg/ml)-coated glass coverslips. Cells were transfected as above and 48 h later cells were treated with PDBu (200 nM) for 15 min, fixed and processed for immunofluorescence.
Cultures of primary cortical neurons
Cultures of dissociated E19 rat cortical neurons were prepared from the cerebral cortex of 19-day-old Wistar rat embryos as described [26] . Rats were obtained from the animal care facility at the Instituto de Investigaciones Biomédicas ‘Alberto Sols’ (CSIC-UAM, Madrid, Spain). Briefly, meninges were removed from the embryonic brains, and cortices were dissected. Tissue was resuspended in minimal essential medium (MEM; Invitrogen-Life Technologies; Carlsbad, CA, USA) complemented with 10% fetal calf serum, 10% horse serum, 0.6% glucose, 16 µg/ml gentamicin, and 2 mM glutamine. Cells were counted and seeded on laminin (4 µg/ml) and poly-L-lysine (10 µg/ml)-covered dishes at a final concentration of 5×10 5 and incubated at 37°C in an atmosphere of 5% CO 2 . Neurons grown in vitro for 14 days (DIV14) were pretreated with Gö6976 (5 µM) for 1 h and left unstimulated or stimulated with the NMDAR agonist NMDA (50 µM) and its coagonist glycine (10 µM) for 5 min.
Immunofluorescence and Confocal Microscopy
For immunofluorescence cells grown on coverslips were fixed for 10 min in 4% paraformaldehyde in phosphate-buffered saline at room temperature. After blocking (5% bovine serum albumin for 30 min) cells were incubated with the corresponding primary antibodies for 1 h at room temperature, and immunoreactivity was detected with the suitable fluorophore-conjugated secondary antibody before mounting in slides with ProLong (Invitrogen-Life Technologies; Carlsbad, CA, USA). Images are single sections of z-series acquiring each channel in a sequential mode using an inverted Zeiss LSM710 confocal microscope with a 63X/1.40 Plan-Apochromatic objective. Pictures were processed with ZEN 2009 light Edition (Carl Zeiss MicroImaging) and Adobe CS3 Extended (Adobe Systems Inc., CA) software.
Protein extracts, immunoprecipitation and immunoblot analysis
Preparation of lysates and immunoprecipitation assays were performed as described previously [26] . Briefly, rat brain or cells were lysed in radioimmunoprecipitation assay buffer (25 mM Tris-HCl, pH 7.6, 1% Triton X-100, 1% sodium deoxycholate, 0.1% SDS, 150 mM NaCl, 2 mM EDTA, 2 mM dithiothreitol) with protease and phosphatase inhibitors for 30 min at 4°C, and lysates were then centrifuged for 20 min at 14,000 rpm. When needed, Myc-nNOS or Myc-nNOS S1412A were immunoprecipitated with anti-Myc antibody during 4 h at 4°C. Equal amounts of total lysates or equivalent volumes of immunocomplexes were analyzed by SDS-PAGE followed by transfer to nitrocellulose filters and immunoblot. Membranes were blocked in TBST (20 mM Tris-HCl, pH 7.6, 137 mM NaCl, 0.05% Tween 20) plus 5% low-fat milk powder and incubated with the different primary and secondary antibodies in blocking solution and immunoreactive bands were visualized by enhanced chemiluminescence (ECL; PerkinElmer, Inc., Boston, MA, USA).
Cloning and expression of full-length nNOS and the two independent heme-oxygenase and reductase domains
Using rat nNOS as a template, the N-terminal half of nNOS comprising the heme-oxygenase domain (residues 1 to 759) or the C-terminal half of nNOS comprising the reductase domain (residues 715 to 1429) were amplified and NdeI and XbaI sites were introduced at the 5′and 3′end respectively. The PCR bands were digested with NdeI/XbaI and ligated in the corresponding sites of 6His-pCWori [32] and verified by automated DNA sequencing. Each of the two halves of nNOS included the calmodulin binding sequence, since calmodulin binding assists protein folding and total yield [33] . Full-length nNOS expression and purification in this vector has been already described [34] . In general, protein expression and purification was performed in BL21 cells in coexpression with a calmodulin chloramphenicol-resistant pACYC plasmid as previously described [32] . Full-length nNOS and its reductase domain were purified using a Ni-NTA affinity resin followed by a 2′,5′-ADP sepharose whereas the heme-oxygenase was purified using only the Ni-NTA resin as previously described [32] , [33] .
Cloning and expression of recombinant active catalytic domain of PKD1 fused to GST
The C-terminal region of PKD1 (Ser558-Leu918; PKD1cat) containing the full-length catalytic domain was amplified using as template pBS-PKD1 using oligonucleotides 5′ ( 5′-AAA AAG CAG GCT CCG GAT CCA ACT CAC ACA AAG ATA-3′ ) and 3′ ( 5′-AGA AAG CTG GGT TTT TGA CAG ATT AGA GGG GAT GGA-3′ ). The PCR product was cloned in pDONR201 by a recombination reaction with BP clonase (GATEWAY system, Invitrogen-Life Technologies; Carlsbad, CA, USA), to generate the construct pENTR-PKD1cat. After automated sequencing, PKD1cat was subcloned in pDEST15 using LR clonase. This vector for procaryotic expression generates PKD1cat fused to glutathione S-transferase (GST; GST-PKD1-cat) of approximate molecular weight of 65 kDa that was purified following standard methods and stored at −20°C. This protein is constitutively active since it lacks the regulatory autoinhibitory domain.
Pull-down assays using recombinant pleckstrin homology domain of PKD1 fused to GST
Preparation of GST-PH domain of PKD and pull-down assays has been described previously [35] . Brain extracts or purified nNOS were incubated for 2 h at 4°C with either GST (control) or GST-PH fusion proteins pre-adsorbed onto glutathione-agarose beads and the presence of nNOS was analyzed by Western blot.
In Vitro Kinase Assay
PKD or nNOS were immunoprecipitated from cultures of primary rat cortical neurons DIV14, and PKD phosphorylation activity was determined performing similar in vitro kinase assays as described previously [26] . nNOS phosphorylation by active catalytic domain of PKD1 (GST-PKD1-cat) was also analyzed using this type of assays. Briefly, PKD or nNOS immune-complexes or purified full-length nNOS or heme-oxygenase and reductase domains mixed with GST-PKD1-cat were resuspended in kinase buffer (30 mM Tris-HCl, pH 7.6, 10 mM MgCl 2 , and 2 mM dithiothreitol), and subjected to an in vitro kinase assay for 30 min at 30°C in the presence of 100 µM final concentration of [γ 32 P] ATP or non-radioactive ATP. Samples were analyzed by SDS-PAGE and Ponceau staining, autoradiography or immunoblot as indicated in the text.
Results
PKD1 interacts with nNOS through its PH domain but not its PDZ-ligand
Proteins known to bind to nNOS PDZ domain must present a hydrophobic amino acid such as Val, Leu or Ile at the final position together with an acidic residue at position -2 or - 3 [6] , [7] . The C-terminus of PKD1 possesses a -VSIL motif in which the Ser residue (Ser 916 ) becomes autophosphorylated in the active enzyme [26] , [27] . Therefore, we reasoned that the negative charge of the phosphate incorporated at Ser 916 in active PKD1 could convert this domain in a bona-fide PDZ-binding motif for the PDZ domain of nNOS. To check this idea, and given that nNOS and PKD are spatially enriched in postsynaptic densities and dendrites, we first examined the possible association of both enzymes in mature neurons performing co-immunoprecipitation assays. Cultured primary rat cortical neurons grown in vitro for 14 days (DIV14) were lysated and PKD and nNOS were immunoprecipitated using specific antibodies ( Figure 1A ). These immunoprecipitates were used to perform an in vitro kinase assay (IVK) in the presence of [γ- 32 P]-ATP before being resolved in SDS-PAGE gels and transferred to a nitrocellulose membrane. This filter was first subjected to Western blot analysis to examine the presence of PKD or nNOS in the immunoprecipitates. As shown in figure 1A , nNOS was not present in PKD immunoprecipitates while a band that could correspond to PKD was detected in nNOS immunoprecipitates. It is noticeable that immunoprecipitated PKD subjected to IVK migrated more slowly, indicating its hyperphosphorylated state compared to its signal in neuronal total lysates. Next, this membrane was exposed to obtain an autoradiography image ( Figure 1A , IVK, bottom panel). We clearly observed a radioactive band corresponding to autophosphorylated PKD in both immunoprecipitates. Importantly, nNOS immunoprecipitates showed an additional radioactive band of an apparent molecular weight similar to that of nNOS (160 kDa). When this autoradiography was overlapped with immunoblots developed with PKD or nNOS antibodies we could observe that both signals matched completely. This result demonstrates that endogenous nNOS is able to co-immunoprecipitate endogenous PKD from mature neuronal lysates and suggests that nNOS could be a PKD substrate.
10.1371/journal.pone.0095191.g001 Figure 1
PKD1 association with nNOS is enhanced by kinase activation in a PDZ-ligand independent manner.
( A ) PKD and nNOS were immunoprecipitated from cultured primary rat cortical neurons DIV14. These immunoprecipitates were used to perform an in vitro kinase assay (IVK) in the presence of [γ- 32 P]-ATP before being resolved in SDS-PAGE gels together with neuronal total lysates (TL) and transferred to a nitrocellulose membrane. Filter was first incubated with specific antibodies to determine the presence of PKD or nNOS in the immunoprecipitates by Western blot analysis (WB). This method detected a signal that could correspond to PKD in nNOS immunoprecipitates. Note that PKD showed a slower migration after IVK compared to total lysates (TL) indicative of its hyperphosphorylated state. This membrane was then exposed to obtain an autoradiography image of the IVK. Autophosphorylated PKD was present both in PKD and nNOS immunoprecipitates ( 32 P-PKD) and an additional radioactive band corresponding to nNOS was also detected. Radioactive bands and immunoblot signals matched completely after overlapping autoradiography and ECL films, indicative of the association of endogenous PKD with endogenous nNOS in primary cortical neurons DIV14. ( B ) HEK293T cells were transfected with Myc-nNOS (nNOS) and wild-type GFP-PKD1 (PKD1wt). Before lysis and 48 h after transfection, cells were untreated (−) or treated (+) with 200 nM PDBu for 15 min as indicated. Total lysates were subjected to immunoprecipitation with Myc antibody (Ip: Myc). Immunocomplexes were separated by SDS-PAGE and PKD1 and nNOS presence was analyzed by immunoblot using anti-GFP and anti-Myc antibodies, respectively. Expression levels of nNOS and PKD1wt in total lysates are also shown. Note that PDBu treatment enhances the formation of nNOS/PKD1 complexes. ( C ) A phospho-mimetic mutant GFP-PKD1-Ser 916 Glu (PKD1 S916E ) and a non-phosphorylatable mutant GFP-PKD1-Ser 916 Ala (PKD1 S916A ) within PKD1 PDZ-binding motif were transfected alone (-) or together with Myc-nNOS (nNOS). Total lysates were subjected to immunoprecipitation with Myc antibody (Ip: Myc). Immunocomplexes were separated by SDS-PAGE and PKD1 and nNOS presence was analyzed by immunoblot using anti-GFP and anti-Myc antibodies, respectively. Expression levels of nNOS and PKD1 mutants in total lysates are also shown. Note mutants in the PDZ-ligand of PKD1 are able to associate with nNOS similarly to PKDwt under non-stimulated conditions shown in panel B ( D ) Myc-nNOS (nNOS) was cotransfected into HEK293T cells together with wild-type GFP-PKD1 (PKD1wt) or a mutant where the PDZ-ligand had been deleted (PKD1 ΔSIL ), and 48 h after transfection cells were untreated (−) or treated (+) with PDBu as in panel B. Analysis of nNOS immunoprecipitates showed that PKD1/nNOS complexes are still formed after deletion of PKD1 PDZ-ligand. Representative images from three independent experiments are shown in panels A, B, C and D. ( E ) Active catalytic domain of PKD1 (aa 557–918), presenting phosphorylated Ser 916 in its PDZ-ligand (ERVpSIL), or two shorter C-terminal fragments containing the non-phosphorylated PDZ-binding motif of wild-type PKD1 (ERVSIL) or a phospho-mimetic mutant PKD1-Ser 916 Glu (ERVEIL) were cloned in pGBKT7 and used as baits in a yeast two hybrid assay using as prey the PDZ domain of nNOS (without or with the β-hairpin “finger” extension) cloned in pGAD. A positive interaction was detected by the ability of the yeasts to grow in the absence of histidine and to metabolize the X-Gal substrate. No direct interaction between any of PKD1 baits tested was found. The C-terminus of CAPON was used as a positive control of binding to both nNOS constructs whereas α-syntrophin was used as a control of protein known to bind to nNOS only in the presence of the β-hairpin.
To further support PKD-nNOS association, we analyzed if we could detect the association of PKD and nNOS by co-immumoprecipitation and Western blot using epitope-tagged versions of both proteins transfected into mammalian cells. Here, we also examined how PKD1 activation could affect their association. HEK293T cells were transfected with Myc-nNOS together with GFP-PKD1 wild-type (PKD1wt) and 48 h later were left untreated or treated with the phorbol ester PDBu in order to activate PKD ( Figure 1B ). After immunoprecipitating nNOS using an anti-Myc antibody, the presence of GFP-PKD1 in the immunocomplexes was assessed detecting GFP signal by immunoblot. These experiments showed that PKD1wt was present in nNOS immunoprecipitates and that this result was clearly enhanced after PDBu treatment ( Figure 1B ). This result confirmed the association of these two enzymes and indicated that PKD1 activation potentiates the formation of PKD1/nNOS complexes.
Because PKD activation leads to Ser 916 autophosphorylation within its PDZ-binding motif, we further examined the contribution of this phosphorylation to the association of PKD1 with nNOS. To this end, HEK293T cells were cotransfected with Myc-nNOS together with a phospho-mimetic mutant GFP-PKD1-Ser 916 Glu (PKD1 S916E ) or a non-phosphorylatable mutant GFP-PKD1-Ser 916 Ala (PKD1 S916A ). Co-immunoprecipitation analysis performed as above showed that mutation of this residue did not alter the association of PKD1 with nNOS under basal conditions and in the absence of PDBu ( Figure 1C ). Importantly, the association of these two PKD1 mutants to nNOS was comparable to that of their wt counterpart under resting non-stimulated conditions, indicating that mutations mimicking or abolishing phosphorylation of Ser 916 were not regulating this process. Additional cotransfection experiments using a PKD1 mutant lacking its PDZ-ligand (PKD1 ΔSIL ) rendered a similar result in which PKD1 was able to co-immunoprecipitate with nNOS preferentially after kinase activation by PDBu treatment ( Figure 1D ). These results suggested that the PDZ-binding motif of PKD1 was dispensable for the formation of a complex with nNOS.
In order to complement these studies, and to definitely rule out the possible participation of PKD PDZ-ligand on its association with nNOS, we tested the putative interaction between the autophosphorylated C-terminus of PKD1 and nNOS PDZ domain performing a yeast two-hybrid assay. Initially, we employed the complete active catalytic domain of PKD1 (aa 557–918), presenting the autophosphorylated PDZ-stretch (-ERVpS 916 IL) at its very C-terminal end, and used it as bait to screen its binding to the PDZ domain of nNOS ( Figure 1E ). Contrary to our initial prediction, we found that this PKD1 construct failed to form a complex with either nNOS PDZ domain (residues 1–102) or an nNOS construct that included also the beta-hairpin motif (residues 1–131) ( Figure 1E ). We obtained a similar result using two shorter constructs of PKD1 displaying a wild-type non-phosphorylated motif (-ERVSIL) or a phospho-mimetic sequence (-ERVEIL) ( Figure 1E ). Control experiments showed that both nNOS constructs bound tightly to the PDZ-ligand of CAPON ( Figure 1E ), a protein with an acidic residue at the -3 position known to bind nNOS PDZ domain [8] . In addition, full-length α-syntrophin could associate only to the nNOS construct that included the beta-hairpin extension ( Figure 1E ), in agreement with the PDZ/PDZ domain interaction of these two molecules/proteins previously reported [12] . These results reflect that, albeit nNOS PDZ domain constructs used in this assay are functional, the presence of a negative charge on Ser 916 at -2 position within PKD1 PDZ-ligand could not convey on this kinase the ability to bind to PDZ domain of nNOS. These data are also in agreement with immunoprecipitation experiments and further demonstrate that, contrary to our initial hypothesis, the PDZ-binding motif of PKD1 was dispensable for the association of the kinase with nNOS, even though there was a clear association between these two enzymes.
We continued examining the participation of other PKD1 domains that could be mediating PKD and nNOS association transfecting into mammalian cells GFP-PKD1 mutants where the PH or CRD domains had been deleted (PKD1 ΔPH and PKD1 ΔCRD , respectively). HEK293T cells transfected with these mutants together with nNOS for 48 h were untreated or treated with PDBu for 15 min to activate PKD1. Cellular lysates were immunoprecipitated with an anti-Myc antibody to detect PKD1 and nNOS co-immunoprecipitation. As shown in figure 2A , PKD1 without the CRD domain still associated to nNOS, however, nNOS/PKD1 complexes formation was absolutely hampered when PKD1 lacked its PH domain. This result is particularly important because we have shown that PKD1 ΔPH mutant is constitutively active and autophosphorylates at the PDZ-ligand Ser 916 [36] , and further supports the data obtained so far in yeast and mammalian cells, indicating again that phosphorylation of the PDZ binding motif is not involved in nNOS/PKD1 association. Given that PKD2 and PKD3 isoforms contain conserved PH domains, we also checked their possible association with nNOS performing transient transfections in HEK293T cells and coimmunoprecipitation analysis. Results showed that full-length PKD2 weakly interacted with nNOS (see Figure S1 ), whereas PKD3 did not (not shown), suggesting a higher preference of nNOS for binding to PKD1. Globally, from these experiments we can conclude that the presence of the PH domain of PKD1, but not its PDZ-ligand (phosphorylated or not), is absolutely required for the association of this kinase with nNOS.
10.1371/journal.pone.0095191.g002 Figure 2
PKD1 PH domain mediates nNOS interaction.
( A ) HEK293T cells were transfected with Myc-nNOS (nNOS) together with wild-type GFP-PKD1 (PKD1wt) or mutants lacking the PH domain (PKD1 ΔPH ) or the cysteine-rich domain (PKD1 ΔCRD ), and treated (+) or not (−) with PDBu 48 h later. Total lysates were subjected to immunoprecipitation with Myc antibody (Ip: Myc) to immunoprecipitate nNOS. The presence of PKD1 and nNOS in immunocomplexes and total lysates was analyzed by immunoblot using anti-GFP and anti-Myc antibodies, respectively. Note that deletion of the PH domain in PKD1 hampers the formation of nNOS/PKD1 complexes. ( B ) Different concentrations of purified nNOS (0.5, 1 and 5 µg) were incubated with 8 µg of immobilized GST or GST-PKD-PH (GST-PH) proteins. Pull-down complexes were run together with 0.1 µg of purified nNOS in SDS-PAGE gels and nNOS was detected by Western blot. ( C ) Rat brain extracts (500 µg) were incubated with GST or GST-PH and the presence of nNOS in pull-down samples and in brain total lysates (TL) was determined by Western blot as above. Loading of proteins is shown by Ponceau staining. Results are representative of three independent experiments.
Finally, in order to test whether PKD1 PH domain could be mediating a direct interaction with nNOS, we performed pull-down assays using recombinant GST-PH protein. Figure 2B shows that the PH domain of PKD1 alone was able to interact with purified nNOS. Furthermore, this domain also pulled-down nNOS from brain extracts ( Figure 2C ). These data indicate that PKD1 and nNOS interact directly through the PH domain of the kinase.
PKD1 activation potentiates its colocalization with nNOS
Depending on cell context and stimulation conditions PKD can be targeted to different intracellular locations such as the cytosol, plasma membrane, Golgi apparatus, or nucleus (for review, see [22] ). In many cell types, including neural PC12 cells, PKD is mainly cytosolic and treatment with phorbol esters or receptor stimulation provokes a rapid recruitment of the enzyme to specific plasma membrane domains [29] , [30] , [37] . In addition, subcellular targeting of nNOS is also critical for the regulation of its function [4] . Given that PKD1 activation enhances its association with nNOS, we examined whether it could also promote their intracellular colocalization. To this aim, Myc-nNOS together with GFP-PKD1wt were transfected into HEK293T and nerve growth factor-treated PC12 cells. Two days after transfection cells were left untreated or stimulated with PDBu for 15 min to activate PKD1, then fixed and immunostained using an anti-Myc antibody, and analyzed by confocal microscopy. Immunofluorescence images from both cell types showed that under resting conditions PKD1 and nNOS presented a major cytoplasmic distribution and low colocalization ( Figure 3 ). However, after phorbol ester stimulation PKD1 translocated to certain subdomains of the plasma membrane where it significantly co-localized with nNOS ( Figure 3 ). This result reinforces that PKD1 activation potentiates its association and subcellular colocalization with nNOS.
10.1371/journal.pone.0095191.g003 Figure 3
Activation of PKD1 increases its colocalization with nNOS.
HEK293T and nerve growth factor treated PC12 cells were cotransfected with Myc-nNOS and wild-type GFP-PKD1 (GFP-PKD1wt). To activate PKD1, cells were left untreated or stimulated with PDBu for 15 min, immunostained using an anti-Myc antibody and analyzed by confocal microscopy. See that PKD1 activation enhances its colocalization with nNOS in both cell types. Results are representative of three independent experiments. Confocal microscopy images correspond to single sections. A magnified detail of the merge images in PDBu treated cells is depicted. Scale bar , 20 µM.
PKD1 phosphorylates nNOS at activatory Ser 1412 in vitro and in live cells
Our initial in vitro kinase and immunoblot analysis suggested that immunoprecipitated nNOS from neuronal extracts could be phosphorylated by PKD ( Figure 1A , IVK). Sequence analysis of nNOS revealed that the heme-oxygenase domain displays one consensus site for PKD1 phosphorylation (IKRFG-pS 374 -K) [38] . Therefore, we further investigated whether nNOS was a PKD substrate performing in vitro kinase assays using radioactive [γ 32 P]-ATP and purified enzymes. As shown in figure 4A , recombinant full-length nNOS incubated with the purified active catalytic domain of PKD1 (PKD1-cat) rendered a clear radioactive band at 160 kDa, indicative of nNOS phosphorylation. Autophosphorylated PKD1 catalytic domain was also detected as a radioactive band of 65 kDa ( Figure 4A ).
10.1371/journal.pone.0095191.g004 Figure 4
Identification of Ser 1412 in nNOS as the unique site targeted by PKD1 phosphorylation.
( A ) Purified full-length rat nNOS (1429 amino acids, accession number P29476) was phosphorylated by purified active catalytic domain of PKD1 (PKD1-cat) in an in vitro kinase assay (IVK) using [γ- 32 P]-ATP. The image shows a representative IVK autoradiography out of three independent assays performed. ( B ) nNOS phosphorylated in vitro by PKD1 as in (A), but using non-radioactive ATP, was digested with trypsin, and the resulting peptides analyzed by HPLC coupled to MALDI-TOF/TOF. The MS/MS spectra of the tryptic nNOS peptide 1408 LRSEpSIAFIEESKK 1421 (Mass, 1715,851 Da) is shown. The “y-ion fragment series” and the “b-ion fragment series” are indicated on the top. Fragmentation of the precursor reveals unambiguously that Ser 1412 is the phosphorylation site. No other phosphopeptides could be detected among the over 200 peptides resolved by HPLC coupled to MALDI-TOF/TOF analysis. ( C ) Consensus motif for PKD phosphorylation and sites of phosphorylation in several PKD substrates (Kidins220; Slingshot-SSH1; Cortactin) and PKD1 C-terminal autophosphorylation motif. The phosphorylatable Ser (pS) is at position P(0), residue at P(-3) is typically occupied by a basic residue (Arg/Lys) and a hydrophobic amino acid is characteristic of P(-5) (preferentially Leu/Val/Ile). Although nNOS Ser 374 fulfilled the criteria of putative PKD consensus phosphorylation motif, it was not found to be phosphorylated by the kinase. Instead C-terminal nNOS Ser 1412 was identified as a phosphorylated site by mass spectrometry and represents an atypical consensus sequence for PKD since residue at P(-5) is occupied by an Arg.
In order to identify the residues phosphorylated by PKD1 within nNOS, a similar in vitro kinase reaction, performed with non-radioactive ATP, was digested with trypsin and subsequently subjected to HPLC and peptide fragmentation by MALDI TOF/TOF ( Figure 4B ). Of the several hundred nNOS-derived peptides that were obtained, the only significant phosphopeptide that was clearly identified corresponded to LRSESIAFIEESKK (residues L1408-K1421 of rat nNOS - Accession number P29476; 1429 aa). De novo sequencing of an eluted tryptic peptide with a mass of 1715,85 Da revealed that it corresponded to sequence LRSE(pS)IAFIEESKK and the phosphorylated residue was unambiguously assigned to the Ser residue present at the fifth position (pS; b 5 in Figure 4B ). This analysis allowed us to identify accurately nNOS Ser 1412 (Rat LRSE-pS 1412 -IAFIEESKK, residue that in human sequence corresponds to Ser 1417 ) as the serine phosphorylated by PKD1 ( Figure 4B ). In this context, it must be mentioned that according to crystallographic data, Ser 1412 is located within the nNOS C-terminal α-helix and its phosphorylation is known to activate the enzyme, inducing a conformational change that increases the NADPH-derived electrons from the reductase towards the heme-oxygenase domain [39] .
Since in silico analysis of nNOS sequence did not predict this serine was in a consensus context for PKD1 phosphorylation, we compared the amino acid sequences of known phosphorylation sites within several other PKD1 substrates and searched for homologies with the one we had just identified ( Figure 4C ). PKD1 substrates typically present a hydrophobic residue such as Leu or Ile at -5 position, together with a basic residue such as Lys or Arg at -3 position [38] . Interestingly, C-termini of both PKD1 and nNOS partially fail to fully meet this requirement, since an acidic Glu residue is present at -3 position in PKD1 while a basic Arg residue is present at -5 position in nNOS. However, these C-terminal sequences are indeed bona fide PKD1 substrates. In the case of PKD1, C-terminal Ser 916 can be not only autophosphorylated but also trans-phosphorylated by other active PKD1 molecules [26] . In the case of nNOS C-terminus, PKD1 phosphorylates Ser 1412 both in vitro and in living cells as we demonstrate herein (see data below). Furthermore, various amino acids present at nNOS PKD1 phosphorylation sequence are identical to those present in other known PKD1 substrates. In addition to the conserved Arg residue at position -3, nNOS phosphorylation sequence displays a Ser at -2 position (as in the case of slingshot-SSH1 [40] ) and a Glu at -1 position (as in the case of cortactin [41] ) ( Figure 4C ).
In order to corroborate that Ser 1412 was phosphorylated by PKD1 in nNOS we used a commercially available phospho-specific antibody recognizing this phospho-site (nNOS-pSer 1412 ). We performed in vitro kinase assays as above followed by immunoblot analysis. The nNOS-pSer 1412 antibody only detected purified nNOS when it had been pre-incubated with active PKD1 catalytic domain in the presence of ATP ( Figure 5A ). In addition we carried out a similar assay using the two independent domains to show that PKD1 was able to phosphorylate Ser 1412 in the full-length protein and in the reductase domain ( Figure 5B ). To further validate our in vitro data and to test whether nNOS was also a substrate of PKD1 in vivo we transfected HEK293T cells with Myc-tagged wild-type nNOS (nNOS) or the non-phosphorylatable mutant nNOS-Ser 1412 Ala (nNOS SA ) together with a constitutively active mutant of PKD1 fused to GFP (PKD1ca). Levels of ectopically expressed GFP-PKD1ca or mutated or wild-type Myc-nNOS were similar in all cellular total lysates. In agreement with our mass spectrometry results, after immunoprecipitating nNOS with anti-Myc antibodies we detected that Ser 1412 was phosphorylated in vivo only in cells that had been cotransfected with constitutively active PKD1 ( Figure 5C ). Accordingly, no signal was detected when nNOS-Ser 1412 Ala mutant was used ( Figure 5C ). Given that glutamate stimulation of NMDAR in primary cultured cortical neurons results in nNOS Ser 1412 phosphorylation [31] we finally examined whether PKD1 activation could occur downstream the activation of these type of glutamate receptors and control nNOS phosphorylation in this particular site ( Figure 5D ). Cortical neurons DIV14 were incubated with the NMDAR agonist NMDA and its co-agonist glycine for 5 min (named from now on as treatment with NMDA). Some neurons were pretreated for 1 h with Gö6976, an inhibitor that is frequently used to inhibit PKD [42] , [43] . Importantly, immunoblot analysis of neuronal lysates showed increased levels of both active PKD phospho-Ser 916 and nNOS phospho-Ser 1412 after NMDAR stimulation, an effect that was significantly blocked by preincubation with the inhibitor ( Figure 5D ). Altogether our data show that PKD1 specifically phosphorylates Ser 1412 in nNOS both in vitro and in vivo .
10.1371/journal.pone.0095191.g005 Figure 5
nNOS is phosphorylated by PKD1 at the activatory residue Ser 1412 within the reductase domain.
( A ) Purified wild-type full-length nNOS was phosphorylated by purified active catalytic domain of PKD1 fused to GST (PKD1-cat active) by in vitro kinase assays (IVK) using non-radioactive ATP. nNOS phosphorylation by PKD1 at Ser 1412 was detected by immunoblot using a phospho-specific antibody recognizing phospho-Ser 1412 within nNOS reductase domain (nNOS-pSer 1412 ). ( B ) In vitro phosphorylation of PKD1-cat occurs specifically at Ser 1412 both when full-length nNOS or its reductase domain are used as substrates, but not when the heme-oxygenase domain is used as substrate. Phosphorylation was determined using anti phospho-Ser 1412 antibodies. Loading of the different recombinant nNOS (full-length, or FL, heme-oxygenase or reductase domains) is shown by Ponceau staining. ( C ) HEK293T cells were cotransfected with either pEFBOS-GFP vector alone (−) or mutant active PKD1 (PKD1ca) and Myc-tagged nNOS or its phosphorylation deficient mutant Myc-nNOS-Ser1412Ala (nNOS SA ). Two days later cells were lysed and total lysates were incubated with an anti-Myc antibody (IP: Myc). Detection of phosphorylated nNOS at Ser 1412 or total nNOS in the immunocomplexes was determined by immunoblot analysis using a phosphospecific antibody (nNOS-pSer 1412 ) or a total nNOS antibody. Levels of nNOS and GFP-PKD1ca in total lysates are also shown. ( D ) Primary cultures of rat cortical neurons grown in vitro for 14 days were untreated or treated with 50 µM NMDA plus 10 µM glycine (NMDA) for 5 min, pre-incubated or not for 1 h with the inhibitor Gö6976 (5 µM). Detection of active PKD (PKD-pSer 916 ), total PKD, phosphorylated nNOS at Ser 1412 or total nNOS in the lysates was determined by immunoblot analysis. Signal for the neuronal specific enolase (NSE) was used as loading control. Representative blots from three independent experiments are shown. ( E ) Primary cultures of rat cortical neurons grown in vitro for 14 days were untreated or treated with 50 µM NMDA plus 10 µM glycine (NMDA) for 5 min, pre-incubated or not for 1 h with the inhibitor Gö6976 (5 µM). Detection of active PKD (PKD-pSer 916 ), total PKD, phosphorylated nNOS at Ser 1412 or total nNOS in the lysates was determined by immunoblot analysis. Signal for the neuronal specific enolase (NSE) was used as loading control. ( E ) Endogenous nNOS was immunoprecipitated from cultured primary rat cortical neurons DIV14 untreated or treated with NMDA as above. These immunoprecipitates were analyzed for the presence of PKD and nNOS by Western blot. Total lysates from these neurons were run in parallel and the corresponding proteins were detected by the indicated antibodies.
In figure 1A , we have already shown that endogenous PKD and nNOS specifically co-immunoprecipitated in lysates from mature cortical neurons in culture DIV14. Since PKD activation by PDBu increased its association with nNOS in transfected cells, we next examined whether this interaction could be also enhanced by NMDAR stimulation in neurons. We therefore analyzed PKD and nNOS co-immunoprecipitation under basal conditions or after stimulation with NMDA and found that NMDA treatment only slightly enhanced the association of both enzymes ( Figure 5E ). This result could be in part due to the basal activity presented by PKD in neurons in culture as detected by the kinase autophosphorylation signal phospho-Ser 916 .
PKD1 activity controls nNOS activation and NO synthesis
So far our data show that PKD1 activation enhances the formation of a complex with nNOS and PKD1 phosphorylates nNOS at Ser 1412 . It has been reported previously that nNOS activity is regulated through the concerted action of several protein kinases and phosphatases [4] , [44] , [45] . In fact, various signaling pathways result in the activation of protein kinases (such as Akt/PKB or PKA) that converge in the phosphorylation of Ser 1412 , activation of nNOS and increased ·NO synthesis [31] , [46] , [47] . Hence, our findings indicate that PKD1 might be a newly identified activatory partner of nNOS. Our next goal was to demonstrate that active PKD1 was in fact stimulating nNOS enzymatic activity and inducing the production of ·NO. As a first approach we used DAF2-DA (4,5-Diaminofluorescein diacetate), a reagent that is used to detect and quantify low concentrations of nitric oxide when loaded into cells [48] . Transfection of COS-7 cells with wild-type nNOS resulted in a modest increase in ·NO synthesis and DAF2-DA fluorescence levels, probably due to the absence of any added calcium ionophores ( Figure 6A , quantification graph represented on the right). However, ·NO production increased significantly in cells where nNOS had been cotransfected with a constitutively active mutant of PKD1 (PKD1ca). As a positive control, we transfected COS-7 cells with iNOS, an isoform that binds Ca 2+ /calmodulin irreversibly and induces the release of high amounts of ·NO ( Figure 6A ). In addition, we also examined the effects of PKD inhibition in ·NO synthesis by co-expressing wild-type nNOS and PKD1 in COS-7 cells. DAF2-DA fluorescence signal obtained in untreated cells or after PDBu stimulation was blocked when cells were pretreated with Gö6976 ( Figure 6B , quantification graph represented on the right). These results clearly show that PKD1 activity increases the synthesis of ·NO by nNOS in living cells.
10.1371/journal.pone.0095191.g006 Figure 6
PKD1 activity controls ·NO production and downstream cGMP/PKG signaling.
( A ) COS-7 cells were transfected with full-length wild-type nNOS in the absence or presence of PKD1ca. In a different well, iNOS was also transfected and served as a control of large amounts of released ·NO. 48 h after transfection cells were washed with medium and incubated with 25 µM of the fluorescent ·NO sensor DAF2-DA. Cell-released ·NO was allowed to react with DAF2-DA for at least 4 hours. Subsequently, the monolayer was extensively washed with medium and the fluorescence was detected between 505 and 525 nm using an excitation wavelength of 488 nm. A minimum of three large monolayer fields of over 400 cells were captured. A representative field is shown for each of the four conditions (left panels). Fluorescence was quantified through pixel to pixel intensity determination and signal corresponding to cells transfected with the empty vector was subtracted from each condition to represent the plot on the right. Data are mean ± S.D. for three determinations. *, p <0.05 in relation to non-transfected cells. ( B ) COS-7 cells were transfected with full-length wild-type nNOS and wild-type PKD1. In a different well, iNOS was also transfected as a positive control. Two days after transfection cells were incubated with DAF2-DA as before, then preincubated or not with Gö6976 (5 µM) and treated or not with PDBu (200 nM) for 15 min. A representative field for each of the six conditions is shown (left panels) and fluorescence was quantified as above (right panel). ( C ) HEK293T cells were cotransfected with either pEFBOS-GFP vector alone (−) or kinase inactive GFP-PKD1 (PKD1ki) and Myc-nNOS. Cells were pre-treated or not with the nNOS inhibitor L-N G -nitroarginine methyl ester (L-NAME; 100 µM) for the last 24 h and 24 h after transfection. Two days after transfection cells were lysed and total lysates were analyzed by immunoblot. Detection of phosphorylated VASP (VASP-pSer 239 ) as a doublet of 45 kDa and 50 kDa was used as a measurement of downstream signaling activated by ·NO production. As a positive control, VASP-pSer 239 phosphorylation was triggered by incubating the cultures with the cGMP homologue 8-Br-cGMP (100 µM) for 30 min. Levels of total VASP, nNOS, GFP and GFP-PKD1ca expression in total lysates are also shown. Note that total VASP appears as a doublet which upper band is absent in unstimulated cells. Signal for α-tubulin was used as loading control. Graph on the right represents the quantification of the immunoblot signals corresponding to the two VASP-pSer 239 bands (45 kDa plus 50 kDa) normalized to α-tubulin levels and expressed relative to the values obtained in cells expressing nNOS in the absence of L-NAME (arbitrarily assigned a value of 100%). Representative results from three independent experiments are shown. Note that co-transfection of inactive PKD1 abrogates nNOS-induced phosphorylation of VASP-pSer 239 levels to the same extent as L-NAME inhibitor, indicating that nNOS activation, production of ·NO and stimulation of cGMP/PKG signaling pathway is under the control of PKD activity.
As a read out of ·NO synthesis we also decided to determine the levels of vasodilator-stimulated phosphoprotein (VASP) phosphorylated at Ser 239 (VASP-pSer 239 ), which has been suggested to represent a biochemical marker of ·NO levels in intact cells [49] . Released ·NO is able to induce cGMP production and protein kinase G activation that ultimately phosphorylates this residue in VASP, an effect prevented by preincubation with NOS inhibitors. Protein kinase G activation promotes VASP phosphorylation mainly in Ser 239 , but also in other two residues [50] , [51] . Unphosphorylated VASP and VASP-pSer 239 migrate in SDS-PAGE gels with an apparent molecular weight of 45 kDa and additional phosphorylations in either one or both of the other residues produces a shift up to 50 kDa. Therefore, depending on the activation of this signaling cascade, the antibody recognizing VASP-pSer 239 will detect a double band in immunoblot analysis. Importantly, if the phosphorylation of the protein decreases significantly, total VASP will mainly be detected as a single band of 45 kDa. Changes in the intensity of VASP-pSer 239 band or in the mobility of the protein correlate with the degree of phosphorylation and consequently of the activation/inactivation of this pathway by ·NO. To determine the influence of PKD1 activity on this parameter, we transfected HEK293T cells with Myc-nNOS and kinase inactive GFP-PKD1ki (PKD1ki) ( Figure 6C ). Before preparing cellular extracts, cells were pre-treated or not with the nNOS inhibitor L-N G -nitroarginine methyl ester (L-NAME; 100 µM) for the last 24 h. As a positive control, we triggered VASP-pSer 239 phosphorylation by incubating the cultures with the cGMP homologue 8-Br-cGMP (100 µM) for 30 min. The immunoblot image and its quantification analysis showed that VASP-pSer 239 levels were almost undetectable in untreated and that the doublet signal only appeared after 8-Br-cGMP treatment ( Figure 6C ). Regarding total VASP, in 8-Br-cGMP stimulated cells a doublet was clearly visible which upper band was hardly detectable in unstimulated cells. When nNOS was expressed VASP-pSer 239 signal was clearly potentiated, an effect that was partially blocked by the inhibitor L-NAME. Transfection of kinase inactive PKD1 had no effect on VASP-pSer 239 or total VASP, being their signal very similar to control cells ( Figure 6C ). Noteworthy, when inactive PKD1ki was co-expressed with nNOS, VASP phosphorylation at Ser 239 was significantly decreased compared with that of cells expressing nNOS alone ( Figure 6C , quantification graph represented on the right) and similar to that obtained in cells pretreated with L-NAME. The reduction in the signal of VASP-pSer 239 and the lack of effect of nNOS inhibitor L-NAME that confers kinase inactive PKD1 demonstrates that the stimulation of this pathway is greatly hampered when PKD1 activity is compromised. Importantly, we also detected an increase in VASP-pSer 239 signal as readout of ·NO release and signaling in mature neuronal cultures after NMDAR stimulation, an effect blocked by inhibiting PKD with Gö6976 ( Figure 5D ). In summary, our data demonstrate that there is a direct correlation between PKD1 activity, nNOS phosphorylation and activation, and ·NO production (see model in Figure 7 ).
10.1371/journal.pone.0095191.g007 Figure 7
Scheme model of PKD1/nNOS complex formation, nNOS activatory phosphorylarion and NO synthesis.
Activation of PKD1 enhances the association of PKD1 with nNOS. The PH domain of PKD1 mediates a direct interaction with nNOS that is independent of PKD1 PDZ-ligand. Active PKD1 autophosphorylates at Ser 916 and phosphorylates the activatory residue Ser 1412 within nNOS C-terminal α-helix, leading to the stimulation of nNOS activity and enhancement of ·NO production. CRD, cysteine rich domain; PH, pleckstrin homology domain; C, catalytic domain; PDZ-L, PDZ-ligand.
Discussion
Among the three enzymes involved in ·NO synthesis (eNOS, nNOS and iNOS), the neuronal isoform nNOS is the only one bearing a PDZ domain. It is widely accepted that PDZ domains present selective interaction with specific PDZ-ligands. Particularly, the PDZ domain of nNOS was proposed long time ago to display a preference for PDZ-binding motifs bearing acidic residues at -2 or -3 position [6] , [7] . The discovery by our group of a PDZ-ligand autophosphorylated in Ser 916 at -2 position in active PKD1, therefore presenting a negatively charged phosphorylated residue at this site [26] , prompted us to hypothesize that this motif could be interacting directly with the PDZ domain of nNOS. We demonstrate here there is a spatial and physical association of PKD1 and nNOS that is potentiated by activation of the kinase. Unexpectedly, our experiments performed in yeast and mammalian cells, show that the association of these two enzymes occurs independently of the PDZ-ligand of PKD1 and the phosphorylation state of Ser 916 within this motif. Instead, the PH domain of PKD1 is absolutely required for its association with nNOS, mediating a direct interaction of both enzymes. Several years ago we discovered that the PH domain of PKD1 is autoinhibitory since point mutations or complete deletion of this domain render a constitutively active kinase [36] . Despite of its highly active state, and the consequent autophosphorylation at Ser 916 within the PDZ-ligand, PKD1 mutant lacking the PH domain (PKD1 ΔPH ) is unable to associate with nNOS. Importantly, α-syntrophin, which interaction with nNOS through its PDZ domain was first identified [14] , bears a PH domain needed to target nNOS to the sarcolemma in vivo , in addition to its PDZ domain [52] , [53] . These observations suggest that there might be a common molecular mechanism by which PH domains may play a critical role in the regulation of nNOS associations with protein complexes and/or subcellular compartments.
The PH domain of PKD1 was first identified to mediate direct interactions preferentially with protein kinase C (PKC) novel isoforms, PKCη and PKCε [35] . These PKCs participate in the classical pathway of PKD1 activation (induced by phorbol esters or diacylglycerol production - downstream membrane receptor's activation) by phosphorylating activation loop Ser 744 and Ser 748 [54] , which in turn results in a release of autoinhibition by the PH domain [55] . PKCδ, another member of the novel PKC subfamily, participates in oxidative stress-induced PKD activation by molecular mechanisms that involve an initial activation of Abl and Src tyrosine kinases [56] , [57] . In this alternative pathway, Src-mediated Abl activation leads to Tyr 463 phosphorylation within PKD1 autoinhibitory PH domain and provokes a molecular switch that allows Src-mediated Tyr 95 phosphorylation, the formation of a complex with PKCδ facilitating activation loop phosphorylation and correlated PKD1 activation [56] , [57] . Despite the pathway involved, a conformational change and a relief of PH domain autoinhibition accompany PKD activation. This novel open conformation may present a more accessible PH domain and favor the association of active PKD with different protein complexes, as we show here to occur with nNOS. Similarly to nNOS, the kinase activity and the PH domain of PKD are critical for apoptosis signal regulating kinase 1 (ASK1) interaction and activation [58] . However, we show here that PKD directly phosphorylates and activates nNOS whereas there are no evidences for ASK1 being a PKD substrate or of stimulation of ASK1 activity by direct PKD phosphorylation.
Despite the list of PKD substrates is increasing, very little is known about the biological significance of their association to the kinase. For most PKD substrates, like Hsp27 [59] , troponin [60] , snail [61] , slingshot [62] , RIN1 [63] , [64] , CERT [65] , oxysterol binding protein [66] or sphingosine kinase [67] , there are not association studies available. In the case of rhotekin [68] and phosphatidylinositol-4 kinase III-β [69] the association studies gave negative results. Interestingly, some PKD substrates have been shown to associate with the kinase, such as Kidins220 [20] , HDAC5 [70] , E-cadherin [71] , β-catenin [72] , CREB [73] and cortactin [41] . However, the effect of PKD activation on substrate association was only specifically addressed before for Kidins220 [20] and HDAC5 [70] that form complexes with the kinase independently of its activation state. Therefore, nNOS is the first identified substrate which interaction with PKD is clearly enhanced after activation of the kinase.
In addition, we demonstrate here that active PKD1 phosphorylates nNOS in the activatory Ser 1412 in vitro and in vivo in living cells, stimulating its enzymatic activity and increasing ·NO production. In this context, it must be mentioned that this is an atypical site for PKD phosphorylation. This kinase usually recognizes a consensus motif presenting a hydrophobic residue such as Leu/Val/Ile at -5 position and Arg/Lys at -3 position referred to the phosphorylatable Ser residue [38] . Remarkably, nNOS displays Arg residues at both positions (RLRSES 1412 ). Regardless of this fact, our data show that PKD1 is an activatory partner of this isoform by phosphorylating Ser 1412 . This residue is part of a motif positioned at the very C-terminal end of the reductase module and recent crystallographic data indicate that it adopts a helical conformation [39] . This α-helix is known to function as a physical “lid” in the reductase domain that impedes proper electron transfer [74] , [75] . The phosphorylatable oxygen atom of Ser 1412 is directed toward the negatively charged flavin mononucleotide-binding domain residues Glu 916 and Asp 918 [39] . It has been suggested that this arrangement thus rationalizes a mechanism for phosphorylation-induced NOS activation. The electrostatically-induced conformational change would be then mediated by the repulsion of the negative charge of the newly added phosphate group. Hence, phosphorylation of nNOS at Ser 1412 activates electron transfer from the reductase domain towards the oxygenase domain of nNOS thus augmenting ·NO synthesis and cGMP formation [39] .
Similarly, it has been reported that nNOS is also phosphorylated on Ser 1412 by Akt/PKB [31] , [76] , cyclic AMP-dependent protein kinase (PKA) [46] and AMP-activated protein kinase (AMPK) [47] in neurons and in skeletal muscle. Not only in nNOS, but also in eNOS, the equivalent serine residue (Ser 1179 and Ser 1177 in bovine and human eNOS, respectively), located at the C-terminus (presumably an α-helix as well), was early recognized as a phosphorylation site. This serine residue is immersed within a consensus Akt/PKB-dependent phosphorylation consensus sequence (RXRXX(S/T)X, herein RLRSE S I in nNOS and RIRTQ S F in eNOS). We and other authors reported the phosphorylation of eNOS by Akt/PKB [77] – [79] and AMPK [80] both in vitro and in a cellular environment. Furthermore, the activity of at least six protein kinases (Akt/PKB, AMPK, PKA, cGK-I/PKG, Chk1 and CaMKII converge on the same eNOS activatory Ser 1179 , and in all cases protein phosphorylation correlates with activation and increased ·NO synthesis (for a recent review see Dudzinski and Michel [81] ). In the case of eNOS, mimicking the phosphorylation of Ser 1179 by introducing an acidic residue in recombinant purified enzymes directly enhances enzyme activity and alters the sensitivity of the enzyme to Ca 2+ , rendering its activity maximal at sub-physiological concentrations of this cation [82] . Importantly, we have also identified that PKD1 induces eNOS phosphorylation on Ser 1179 in vitro and in endothelial cells stimulated with vascular endothelial growth factor (unpublished data).
Our data also show that activated PKD1 and substrate nNOS colocalize and are able to co-immunoprecipitate. This is not unprecedented, since the homologous eNOS can be co-immunoprecipitated with several protein kinases, such as Akt/PKB [83] , AMPK [84] and Chk1 [85] , known to phosphorylate the equivalent serine at the activatory C-terminus upon activation. The proximity of activated PKD1 and nNOS observed in transfected neural PC12 cells differentiated with nerve growth factor suggests that these two enzymes might be together at neuronal post-synaptic densities. Supporting this possibility, we also show that the stimulation of glutamate NMDAR with the specific agonist NMDA in primary cortical neurons in culture results in an activation of PKD1 that parallels nNOS phosphorylation at Ser 1412 which is blocked after pharmacological inhibition of PKD. Other authors have reported previously that protein kinase Akt/PKB activation in response to NMDARs stimulation also results in nNOS phosphorylation at Ser 1412 and increases its enzymatic activity [31] . It would be of high interest to follow this line of research in the future in order to establish the role of PKD1 activation downstream NMDAR stimulation and its relation to ·NO production both under physiological and pathological conditions. It is well established that whereas the small quantities of nitric oxide formed during synaptic transmission modulate neuronal signaling, excess of nitric oxide mediates neurotoxicity in pathological situations, such as ischemic stroke or neurodegeneration. Hence, since little is known about the various kinases that regulate nNOS function in vivo , PKD inhibitors might be useful drugs in cases of nitric oxide associated neurotoxicity.
In conclusion, herein we reveal that PKD1 interacts with nNOS and phosphorylates Ser 1412 enhancing this way nNOS activity and ·NO production. Considering that the corresponding serine in eNOS is subjected to similar mechanisms of phosphorylation-dependent activation, PKD emerges as a common regulator of the enzymatic activity of constitutive NOS isoforms and ·NO synthesis. These novel findings add to the list of biological relevant roles of PKD a crucial one in the regulation of ·NO synthesis and the plethora of physiological and pathological processes where this mediator is involved.
Supporting Information
Figure S1
Association of PKD2 with nNOS.
(TIF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Rising awareness about sustainability and associated development of green chemistry lead to the growing popularity and demands for commercial products with minimal environmental impact, with present emphasis on the origin of material resources, bioaccumulation, biodegradability, and greenhouse gas and waste production [ 1 ]. Sustainable chemistry has become a key focus area for chemical companies [ 2 ], which motivates pharmaceutical, cosmetic and food industry to employ more extensively biocatalysis as an environmentally friendly and effective alternative to traditional synthetic preparation [ 3 – 6 ]. Additionally, green chemistry provides an excellent opportunity for the demonstration of the important benefits of modern chemistry, often associated with disastrous poisoning accidents, global pollution, and exploitation of natural resources.
During the past decade, enzymes have proved to be an economical and sufficiently stable replacement for various chemical catalysts used commercially for both low-cost bulk products (ethanol by fermenting sugars), and for fine chemicals with high added value and purity (enantiomerically pure chiral compounds, optically active pharmaceuticals, plant protecting agents, and fragrances) [ 7 ]. Since massive commercialization in the 1990s, enzymes have been used extensively in various daily applications such as waste treatment (degradation of plastic, hydrolysis of lignocellulose), cleaning or bioremediation. However, industrial applications bring additional requirements on enzyme functionality due to environmental conditions significantly different than their natural sources, such as the presence of artificial substrates, organic solvents, inhibitors, non-optimal pH, elevated temperature, high ionic strength, which could result in both premature enzyme degradation, and loss of enzyme activity [ 8 ]. Therefore, comprehensive knowledge about enzyme long-term stability under various conditions is crucial for the effective enzyme use in the free or immobilized form [ 5 , 9 ]. Since enzymes are prone to lose a significant part of their activity before use, long-term storage stability is an additional critical factor that must be considered and addressed during the development of new products [ 4 , 5 , 9 – 12 ].
Alliinase (EC 4.4.1.4), an enzyme found in garlic and other plants of genus Allium in exceptionally high concentrations [ 13 ], is responsible for the formation of bioactive compound allicin from stable precursor alliin. This reaction serves as a self-defence mechanism of Allium plants initiated when a cellular structure is compromised, and the initially separated enzyme and the substrate are mixed. A combination of the short half-life, high reactivity and non-specificity to particular proteins [ 14 ] is the reason most bacteria, and other pathogens cannot deal with allicin mode of action and develop effective defence mechanisms. It has been reported that the development of bacterial resistance against potent, but highly reactive and unstable allicin is more than 1000 times slower compared to other antibiotics [ 15 ]. In order to harness the enzymatic production of allicin or its analogues on an industrial scale without significant loss of enzyme activity [ 16 – 20 ], its response to various environmental conditions accompanying enzyme extraction, purification, and storage must be examined and comprehensively detailed.
It has been shown by Weiner et al. that alliinase monomer contains 10 cysteine residues, of which eight form four S-S bridges responsible for the properly folded active state, and two are present in free thiol groups [ 21 ]. Purified alliinase stored in buffer solution without the presence of any antioxidizing agents is prone to air oxidation of cysteine, followed by the formation of inactive S-S bridged alliinase clusters consisting of four, or more enzyme subunits [ 22 ]. There are several commonly used substances which can affect enzymatic behaviour in different ways depending on the nature of target protein, i.e. positively or negatively. In general, additives such as salts, polyols, sugars or inert polymers, such as polyethylene glycol, can greatly enhance the stability without the need of chemical or genetic manipulation of the target protein [ 23 ]. Chhabria et al. reported purification of alliinase from soil bacterium Cupriavidus necator and found that the alliinase activity is enhanced by several metal ions, e.g., Ca 2+ , Mg 2+ , Zn 2+ , Fe 2+ , Mn 2+ ), but inhibited by others, e.g., Cu 2+ , Cs + , Li + , or Hg 2+ [ 24 ]. It was also shown that the addition of K + , Na + , and EDTA had no significant effect on the alliinase activity. Moreover, the effect of additives for alliinases of different origin may vary significantly. For example, it was reported that alliinase extracted from garlic is stimulated in the presence of EDTA [ 25 ]. The positive effects of glycerol, sucrose and sorbitol, osmolytes known as chemical chaperones, and mannose-specific lectin (Allium sativum agglutinins I) on alliinase stabilization were recently presented by Shin et al. [ 26 ].
The present study provides a comprehensive insight into the alliinase stability and activity in the solution and lyophilized state. The influence of commonly used buffers, additives, i.e., chloride salts, such as NaCl, KCl, CaCl 2 , and MgCl 2 , antioxidants (ascorbic acid and DTT), polyols and their combinations in various concentrations, on the enzyme activity and stability over an extended period of time (resistance to denaturation and oxidation), were investigated and discussed. The long-term stability of the enzyme in a lyophilized form stored under controlled conditions, i.e. oxygen presence, temperature and humidity, was investigated providing valuable insight into the problematics of enzyme use in dry powder formulations. Finally, antibacterial activity upon addition of substrate against Gram-positive ( S . epidermidis ) and Gram-negative ( E . coli and P . putida ) bacterial strains was investigated using static and dynamic in vitro assays.
Materials and methods
Chemicals
The following chemicals were purchased from Sigma-Aldrich: (±)-L-alliin (primary reference standard), Ethylenediaminetetraacetic acid (EDTA), Pyridoxal 5′-phosphate hydrate (PLP), Dithiothreitol (DTT), N[Tris(hydroxymethyl)methyl]glycine (Tricine), Polyethylene glycol (PEG) 6000, Bovine serum albumin (BSA), Nutrient Broth (NB), Kanamycin sulfate, Fluorescein diacetate, Propidium iodide (PI), Polyethylenimine (PEI). Glycerol, Sodium phosphate dibasic heptahydrate, Sodium phosphate monobasic decahydrate, Potassium dihydrogen phosphate, Phosphate-buffered saline (phosphate buffer, potassium chloride and sodium chloride), citric acid, hydrochloric acid, ascorbic acid, tryptone, yeast extract, sodium chloride, potassium chloride, calcium chloride, magnesium chloride, lithium chloride, and potassium hydroxide were purchased from Penta (Czech Republic). Magnesium nitrate hexahydrate was purchased from Fisher Scientific. L-Lactate dehydrogenase (LDH) from rabbit muscle and β-Nicotinamide adenine dinucleotide (NADH) disodium salt grade II (approx. 98%) were purchased from Roche Diagnostic (Germany). Garlic was purchased from Farmers’ Market, Zahradnictví Riegelová, country of origin Czech Republic.
Alliin synthesis
Alliin was synthesized by a method previously published by Stoll and Seebeck [ 27 ]. In the first step, L-cysteine (50 g) was dissolved in an aqueous solution of ammonium hydroxide (2M, 1200 mL) and alkylated with allyl bromide (75.0 g) at 0°C. After 40 minutes, the reaction was terminated, and volume reduced using a rotary evaporator. The raw product (L-deoxyalliin) was filtered, washed with ethanol, vacuum dried and recrystallized from 2:3 water/ethanol mixture. In the second step, L-deoxyalliin (53.0 g) was suspended in 530 mL of water at 25°C and mixed with 37 mL of hydrogen peroxide (30% w/w). After two days, the volume was reduced by rotary evaporator, 400 mL of acetone was added, and the mixture was stirred for 2 hours. Then, the crude (±)-L-alliin was filtered, repeatedly washed with a mixture of acetone/water (5:1) and vacuum dried. The purity of synthetic alliin (≥ 90%) was determined by HPLC (Agilent 1260 Infinity) using commercially available (±)-L-alliin as standard.
Extraction and purification of alliinase
The crude alliinase was obtained from fresh garlic stored at cold and dark prior to its use. All operations regarding enzyme extraction were carried out at 4°C, unless stated otherwise. The isolation protocol was followed, as described by Mallika with some minor modifications [ 28 ]. The garlic was peeled and homogenized by hand mixer in the ratio of 1:1.5 (w/v) in buffer (pH 6.5, 20 mM sodium phosphate buffer, 10% (v/v) glycerol, 5 mM EDTA, 5% (w/v) NaCl and 20 μM PLP). The presence of PLP (cofactor of alliinase), EDTA (chelating agent of heavy metals) and glycerol (inhibitor of enzyme unfolding and aggregation) is recommended to stabilize the enzyme during extraction steps [ 25 ]. The garlic slurry was squeezed through four layers of cheese cloth followed by vacuum filtration via 5 μm pore glass filter. The fractional precipitation of proteins was performed with PEG 6000 (25% w/w), followed by centrifugation at 30,000 g for 30 min. Next, the yellowish pellet was collected and resuspended in 20 mL of deionized water with 20 μM PLP. The solution was centrifuged again at 30,000 g for 30 min, and the supernatant was passed through at 0.45 μm syringe PVDF filter to remove residual tissue fragments. Finally, the purified solution was quickly frozen in liquid nitrogen and lyophilized (Sentry 2.0, SP Scientific, USA) at -90°C under chamber pressure of 1 μbar for 4 days. The dry product was stored in a freezer at -20°C prior to use.
Characterization of alliinase
The characterization of the enzyme was carried out by using sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) under reducing conditions (in the presence of β-ME). SDS-PAGE was carried out using 10% (w/v) polyacrylamide gel at a constant voltage of 20 V/cm of gel length, in which broad-range molecular mass standards (the protein marker VI (10–245), AppliChem, USA) were run simultaneously. The protein content was measured by the UV/VIS spectrophotometry at λ = 595 nm, according to the Bradford method using BSA (0 to 10 μg protein/mL) as an external standard [ 29 ].
Enzyme assay
The specific activity of the enzyme alliinase was determined via coupled NADH-dependent pyruvic acid reduction in the presence of lactate dehydrogenase (LDH) [ 30 ]. Pyruvic acid, released as a co-product during allicin formation ( Fig 1 ), was reduced, in the presence of LDH, to lactic acid and NADH oxidized to NAD + . The time-dependent disappearance of the NADH was measured by the UV/VIS spectrophotometry at λ = 340 nm. The decrease in absorbance was proportional to the reaction rate and thus to the enzyme activity. The typical alliinase assay solution (0.1 mL) contained 20 mM alliin, 20 μM PLP, 200 mM Tricine-KOH buffer (pH 8.0), 0.8 mM NADH, LDH (2.5 μL, 550 units/mg) and lyophilised alliinase at indicated concentrations. Both NADH, LDH and alliin were added in excess to ensure alliin conversion was the rate-limiting step of the assay.
10.1371/journal.pone.0248878.g001
Fig 1
A) enzymatic formation of allicin from alliin; B) the coupled reaction of pyruvic acid, a co-product of reaction alliin with alliinase, with NADH and LDH used for UV-VIS spectroscopic assay.
The Michaelis–Menten kinetic constants, i.e., the maximum velocity (V max ) and Michaelis–Menten constant (K m ) of the purified alliinase in respect to the mixture of alliin diastereomers, were determined experimentally using alliin initial concentration in the range of 0.5 to 50.0 mM. Kinetic constants of Michaelis-Menten equation (V max , K m ) were calculated by non-linear least-square regression using software SigmaPlot 11 (Systat Software Inc., Chicago, IL).
Effect of temperature and pH on alliinase activity
The influence of temperature on the specific activity of purified alliinase incubated in Tricine-KOH buffer (200 mM, pH 8) was studied in detail for the range of temperatures 20°C to 42°C using a mixture of alliin diastereomers (20 mM) as a substrate.
The time-dependent stability of solubilized alliinase was studied at various temperatures covering the typical range for the storage and use, i.e., -20°C, 4°C, 25°C and 37°C. The samples stored at -20°C were unfrozen before each experiment. Sodium azide (0.02% w/v) was used as an antimicrobial agent to prevent bacterial contamination during the long-term incubation.
The effect of pH on alliinase activity was determined by the incubation of the enzyme in a series of buffers covering a pH range from 1 to 8 at 25°C for 10 min. The alliinase activity was determined by enzyme assay performed at a given pH. Hydrochloric acid-potassium chloride (50 mM, pH 1 to 2), citrate-phosphate (50 mM, pH 3 to 5) and sodium phosphate (100 mM, pH 6 to 8) buffers were used, and their effect was determined. The highest observed enzyme activity was taken as the reference value of 100%, and the other experimental datapoints were rescaled accordingly. The relative expression of enzyme activity was preferred where possible, since it allowed the comparison of enzymes prepared in multiple batches, where the absolute value of enzyme activity may vary depending on garlic origin, enzyme content and purity of extracted alliinase.
The influence of various buffers of the same pH on alliinase activity was studied using sodium phosphate (100 mM, pH 6–8), potassium phosphate (100 mM, pH 6–8), phosphate-buffered saline (200 mM, pH 7.4) and Tricine–KOH (200 mM, pH 8) buffers.
Effect of additives on alliinase stability
The effects of chloride salts (NaCl, KCl, CaCl 2 , MgCl 2 ) at different concentrations (5 mM, 50 mM, 100 mM and 500 mM) and stabilizers (ascorbic acid (4 mM), EDTA (100 mM), DTT (5 mM), glycerol (0.1 to 10 v/v%)) on alliinase stability in Tricine-KOH (200 mM, pH 8) buffer were investigated for incubation times up to 24 hours at 25°C. The activity of the purified alliinase without any additives was taken as a reference value, i.e., 100%.
Stability of lyophilized enzyme
The sample of the enzyme (lyophilized powder, 0.05 g) in 1.5 ml polypropylene microtube was placed upright in a glass beaker. The beaker was kept sealed at 25°C or 37°C in the thermostatic oven to ensure constant conditions. For an inert atmosphere, the beaker was placed into a desiccator, and nitrogen gas was repeatedly passed through to purge all oxygen and water vapours. The saturated salt solution (LiCl, Mg(NO 3 ) 2 and NaCl) was used to maintain the desired relative humidity [ 31 ] with respect to the theoretical values shown in Table 1 . A sample of the powdered enzyme was analyzed for the remaining enzyme activity every 24 hours.
10.1371/journal.pone.0248878.t001
Table 1 Conditions for testing of enzyme stability at 25°C and 37°C.
Temperature [°C]
Relative humidity [%]
N 2
LiCl
Mg(NO 3 ) 2
NaCl
water
25
0
11.30 ± 0.27
52.89 ± 0.22
75.29 ± 0.12
100
37
0
11.23 ± 0.21
49.17 ± 0.47
74.77 ± 0.13
100
Antibacterial assays
Escherichia coli K12 (EC43) and Pseudomonas putida were chosen as model Gram-negative bacteria and Staphylococcus epidermidis as a Gram-positive bacterium. Bacterial strains were kindly provided by Dr Zdeněk Knejzlík (Academy of Science of the Czech Republic, E . coli K12 DSMZ-German Collection of Microorganisms and Cell Cultures GmbH) and prof. Dan V. Nicolau (Department of Bioengineering, McGill University, P . putida –ATTC, Manassas, USA; S . epidermidis —Carolina Biological Supply Company, Burlington, USA). All bacterial strains are classified as low biological risk organisms (BSL-1).
Cultivation of bacteria
The cultivation of Gram-negative bacteria ( E . coli and P . putida ) at 37°C used Luria-Bertani medium (LB). LB medium was prepared by dissolution of 20 g of LB salt mixture (10 g Tryptone/Peptone, 5 g yeast extract, 5 g NaCl) in 1000 mL of deionized H 2 O adjusted to pH 7.2–7.4 with 1 M NaOH. For the preparation of a solid medium, 1.5 g of agar per 100 mL of liquid LB medium was added. S . epidermidis was cultivated at 30°C in nutrient broth (NB) medium (8 g/L). For the preparation of a solid NB medium, 1.5 g of agar per 100 mL of liquid NB medium was added.
Disk diffusion assay
The antimicrobial effect of allicin was measured by a disk diffusion assay on an agar plate inoculated by bacterial suspension (OD 600 = 0.2). Different concentration of enzyme solution (0.02 mg/mL up to 10 mg/mL) was prepared using Tricine-KOH buffer (200 mM, pH 8). The mixture of the enzyme (10 μL) and substrate (10 μL, 100 mM) was added onto filter paper discs (6 mm in diameter) placed on the agar plates incubated upside down at 37°C for 24 hours. The antibiotic kanamycin (50 mg/mL, 20 μL) was used as a positive control. A solution of pure alliin (labelled as S, 20 μL, 100 mM) or alliinase (labelled as E, 20 μL, 10 mg/mL) was used as a negative control. The sensitivity of bacterial strains to formed allicin was determined by measuring the diameter of inhibition zones using ImageJ software [ 32 ]. All experiments were performed in triplicates for each concentration. If the inhibition zones were absent, the sample was regarded as ineffective against the respective bacterial strain. The inhibition zone for kanamycin positive control was taken as a reference value, i.e. 100%.
Cell viability analysis
Live/Dead assay . Confocal laser scanning microscopy was used to image the proportions of live and dead cells using Live/Dead assay kit (Sigma-Aldrich) based on the in vivo staining by fluorescein diacetate (FDA) and propidium iodide (PI) fluorescent dyes. The green fluorescent FDA dye permeated the intact membrane of the cells binding to nucleic acids, whereas the red fluorescent PI dye can enter only the non-viable cells with a damaged cell membrane. The bacterial suspension (300 μL) stained according to the manufacturer protocol (1 μL of FDA with 1 μL of PI per 1 mL of bacterial suspension) was incubated with the reaction mixture of alliin (100 μL, 100 mM) and alliinase (100 μL, 1 mg/mL). Aliquots withdrawn at the specific times (0, 10, 20 and 30 min) were examined by inverted confocal microscope IX83 (Olympus), and the images were analyzed using ImageJ software. The green and red channels were split, converted to 8-bit colour depth, thresholded and binarised. The particle count corresponded to the total number of cells stained with green, and red fluorescent dyes, respectively. The survival fraction was calculated as the ratio of viable cells vs overall cell count. All samples were plated in a triplicate under sterile conditions.
Plate count method . The bacterial suspension (300 μL) was incubated with the reaction mixture of alliin (100 μL, 100 mM) and alliinase (100 μL, 1 mg/mL). Samples of bacterial suspension (10 μL) with, or without alliin/alliinase were taken at the specific times (0, 10, 20 and 30 min) followed by 1000x dilution with cultivation medium (LB or NB) to suppress allicin effect. A prepared bacterial suspension was plated on Petri dishes, and a total count of emerging colonies after 24h incubation was monitored using ImageJ.
Analysis of bacterial morphology
The effect of in-situ formed allicin on bacterial morphology was examined by FEI Quanta 450 Environmental Scanning Electron Microscope (FE-ESEM). A glass coverslip serving as bacterial support was cleaned with acetone and isopropanol, activated by oxygen plasma (Harrick Plasma), and incubated in 1 w/v % PEI solution to enhance bacterial adhesion. The modified coverslips placed separately in a 6-well plate were incubated for 5 min in bacterial culture (OD 600 = 0.2), followed by addition of reaction mixture of alliin (100 μL, 100 mM) and alliinase (100 μL, 1 mg/mL). The coverslips withdrawn at the specific times (0, 10, 20 and 30 min) were immediately dehydrated with incrementally graded ethanol series (starting with 20 v/v % to anhydrous ethanol). Solvent dehydration followed by critical point drying (CPD, Leica EM CPD300) ensured the preservation of delicate sample morphology. The dehydrated samples were coated, prior SEM measurements, with a platinum layer to increase the conductivity of sample and SEM image clarity.
Statistical evaluation
Enzyme activities were expressed as means of 6 or more measurements and their respective standard deviation (SD). Data were analyzed using one-way ANOVA followed by Tukey and Dunnett tests to compare the influence of particular additives on enzyme activity. Values of P < 0.01 were considered to be statistically significant. Statistical analyses were performed using software SigmaPlot 11.
Results
Characterization of alliinase
The purified garlic alliinase, analyzed by 10% reducing SDS-PAGE showed a 51 kDa single band corresponding to alliinase monomer ( Fig 2A ), which is in agreement with previously published reports [ 33 – 35 ]. SDS-PAGE analysis revealed that alliinase obtained by PEG precipitation was of high purity without a significant presence of the accompanying proteins or degradation products. The lyophilized alliinase was found to have a protein content of 88 ± 4% per weight, accompanied by non-protein substances (polysaccharides, salts, PEG) and moisture. Kinetic parameters V max , K m and turnover number k cat for alliinase following Michaelis-Menten kinetics with respect to the mixture of alliin diastereomers were calculated as 18.9 ± 0.3 mM∙s -1 ∙mg -1 , 4.45 ± 0.36 mM and 193 s -1 , respectively ( Fig 2B ). The calculated values of kinetic parameters correspond to previously reported data [ 36 ].
10.1371/journal.pone.0248878.g002
Fig 2
A) 10% SDS-PAGE of isolated garlic alliinase; B) kinetic analysis of alliinase catalyzed reaction.
The reaction mixture (0.1 mL) contained alliinase (5 μg/mL) and various concentrations of alliin (0 to 50 mM), the reaction temperature was 25°C, Tricine-KOH (pH 8) buffer was used for the assays.
Effect of pH, temperature and buffer on alliinase stability and activity
In general, both pH and temperature are significant factors governing enzyme activity and stability. The alliinase activity was tested for buffers from pH 1 to 8 and a narrow pH optimum at 7.0 was observed ( Fig 3A ), which corresponds to results reported in the literature [ 24 , 37 ]. Bellow pH 5.0 and above pH 8.0, a sharp decrease in the enzyme activity was observed [ 38 , 39 ]. LDH concentration in the reaction mixture was chosen such that the conversion of pyruvic acid was significantly faster compared to its formation. Moreover, the activity of LDH under various conditions was examined separately to prevent underestimation of alliinase activity caused by the slower conversion of pyruvic acid and NADH oxidation in the second reaction step. LDH has proven to be stable over a broad range of buffer composition, temperature and pH. It should be noted that under acidic conditions (pH ≤ 4) both instability of NADH and lower LDH activity may affect data interpretation, although the measured activity of alliinase at pH 4 was already negligible.
10.1371/journal.pone.0248878.g003
Fig 3
Effect of pH (A) and type of used buffer (B) on the alliinase activity at 25°C.
In previously published studies, various buffers were used for alliinase extraction, purification and storage, e.g., Na-PB [ 22 , 24 , 34 ], K-PB [ 30 ], Na/K-PB [ 40 , 41 ], PBS [ 21 , 24 ] and Tricine [ 42 – 44 ]. However, the character of various buffers may play different roles concerning the conformational, colloidal and interfacial stability of enzyme [ 45 ]. Here, a direct comparison of the most used buffers and their respective role in alliinase activity is demonstrated. The results for sodium (Na-PB) and potassium (K-PB) phosphate buffers (pH 6 to 8), PBS (pH 7.4) and Tricine-KOH (pH 8) used as the reference are shown in Fig 3B . Enzyme activity in both phosphate buffers at pH 6 was comparable. Enzyme incubated in K-PB at pH 7 showed 11% lower enzyme activity compared to Na-PB. This difference was even more evident at pH 8, as the enzyme activity in K-PB was compared to Na-PB and Tricine-KOH more than 3.6-fold and 6.5-fold lower, respectively. The highest measured activity for all investigated buffers, i.e., 1.7-times higher compared to Tricine-KOH, was observed for PBS (pH 7.4), which can be explained by different pH and presence of dissolved monocations Na + and K + .
Time dependant enzyme activity was studied for Na-PB, PBS and Tricine-KOH at 25°C ( Fig 4A ). Alliinase was pre-incubated in a corresponding buffer (Na-PB, PBS and Tricine-KOH) for three hours, and the alliinase activity was measured at specified times and the identical conditions. The alliinase activity decreased in all tested buffers with time. After 3 hours, PBS and Tricine-KOH buffer preserved 67% and 61% of the initial enzyme activity, respectively, whereas alliinase stored in the Na-PH buffer showed only 38% of its initial activity.
10.1371/journal.pone.0248878.g004
Fig 4
A) Time-dependent enzyme stability and activity of alliinase in Na-PB, PBS and Tricine-KOH buffer. All results are normalized to alliinase activity measured at 0 h in Tricine-KOH buffer; B) effect of temperature on alliinase activity in Tricine-KOH buffer.
The alliinase activity was also studied at various temperatures within the range 20°C to 42°C in Tricine-KOH buffer ( Fig 4B ). Alliinase and alliin, pre-incubated separately for 5 min at the desired temperature, were mixed, and alliinase activity was measured. The bell-shaped dependence of enzyme activity on temperature showed the highest activity between 35°C to 40°C. Alliinase activity decreased rapidly over 40°C, which is in agreement with the findings reported elsewhere [ 46 – 48 ]. Nevertheless, for various bacterial alliinases, it has been shown that optimal temperature could be slightly shifted. Yutani et al. found that for alliinase extracted from Ensifer adhaerens the optimal temperature was 30°C [ 49 ]. In the case of immobilized alliinases optimum shifts to the higher temperatures, which is given by higher thermal stability of immobilized enzyme compared to its dissolved form [ 37 ].
Time-dependent alliinase stability in Tricine-KOH buffer at -20°C, 4°C, 25°C and 37°C is summarised in Table 2 . It was observed that the activity of samples stored at -20°C decreased after two days by 32%. This significant decrease was most probably caused by repeated freeze-thaw cycles before every enzyme assay. The samples stored in the fridge at 4°C and 25°C showed only 16% and 4% of the initial activity. Finally, the dissolved enzyme stored at 37°C was almost completely deactivated. Based on these findings, it can be concluded that alliinase in a solution cannot be effectively stored for a long time, unless frozen or without the presence of additional stabilizers, as discussed further.
10.1371/journal.pone.0248878.t002
Table 2 Effect of time and storage temperature on alliinase activity in Tricine-KOH buffer.
temperature [°C]
-20
4
25
37
day 1
90.5 ± 2.5
36.3 ± 0.4
10.4 ± 1.5
5.8 ± 0.1
day 2
67.7 ± 4.4
16.0 ± 2.2
3.6 ± 2.6
1.8 ± 1.3
Note: 100% corresponds to the initial enzyme activity at 25°C.
Influence of additives on enzyme activity—Operational stability
Effect of chlorides of mono- and divalent cations
Enzymes can be stabilized by increasing their concentration in the solution and by modulation of the ionic strength [ 4 ]. Several salts are frequently used to enhance the stability of the enzymes, but not all are equally effective in the stabilization of a specific protein. Therefore, the effect of NaCl, KCl, CaCl 2 and MgCl 2 on the alliinase activity was investigated in 5, 50, 100 and 500 mM concentrations in Tricine-KOH buffer. All salts except MgCl 2 proved to have a noticeable positive effect on enzyme stability after 3 hours even at lowest concentrations (5 mM) compared to pure Tricine-KOH buffer without any additives ( Fig 5 ). The most noticeable effect was observed for Na + (50 to 500 mM) and Mg 2+ cations (100 and 500 mM) resulting in substantially higher enzyme activity after 3 hours compared to initial activity in pure Tricine-KOH. The considerable increase of activity can be attributed to the vital role of dissolved salts in the stabilization of enzyme conformation by neutralization of protein charges or formation of salt bridges by divalent cations or to the fact that these ions can offer protection to thiols or other functional groups against oxidation by salting-out dissolved oxygen [ 4 ]. Increasing the concentration of mono or divalent cations from 100 mM to 500 mM did not show any further enhancement in terms of the alliinase activity.
10.1371/journal.pone.0248878.g005
Fig 5
Effect of salts on the initial activity of alliinase (A) and 3 h (B) incubation at 25°C.
All results are normalized to alliinase activity measured at 0 h (25°C, Tricine-KOH buffer, no additives). Means followed by the same letter (a–f) were not significantly different (P>0.01, ANOVA, Tukey’s test).
Effect of additives on alliinase stability
In most cases, the actual mechanism of stabilization of enzymes by altering their surrounding microenvironment is difficult to predict because of its complexity. Therefore, a complete acquaintance of additives with respect to alliinase is needed in order to assess their stabilization effect and role played in the retention of enzyme activity. The effects of ascorbic acid (antioxidant and reducing agent), glycerol (cosolvent), EDTA (chelating agent) and DTT (a strong reducing agent preventing oxidation of thiol groups and formation of disulphides) were investigated and discussed ( Fig 6A ).
10.1371/journal.pone.0248878.g006
Fig 6
A) The effect of additives on enzyme stability at 25°C; B) comparison of the individual and combined effect of 4 mM ascorbic acid (AA) and 50 mM NaCl on enzyme stability.
All results are normalized to alliinase activity measured at 0 h (25°C, Tricine-KOH buffer) without additives; * results did not differ statistically from the reference at given time (P>0.01, ANOVA, Dunnett’s test).
Ascorbic acid, which acts as a radical scavenger, and donor and acceptor in electron transfer reactions [ 50 ] increased enzyme activity significantly. The presence of ascorbic acid in the concentration 4 mM resulted in 2.9-fold and 7.2-fold increase in enzyme activity compared to reference Tricine-KOH buffer, after 3 and 24 hours at 25°C, respectively.
It has been reported that the addition of polyols to an aqueous solution of enzyme improves strengthening of the hydrophobic interactions among non-polar amino acid residues, resulting in more compact protein conformations [ 51 ]. The addition of glycerol (0.1, 1.0, 10.0% v/v) was tested to assess the stabilization of alliinase. However, no relationship between the overall glycerol content and observed alliinase activity has been found. In all cases, the presence of glycerol preserved the activity of alliinase for 3 hours, which was in all cases higher than the initial reference value. After one day, alliinase in the presence of glycerol showed, on average, 4.5-times higher activity compared to glycerol-free Tricine-KOH buffer.
Addition of EDTA, metal chelator preventing metal-induced oxidation of thiol groups, resulted in lower initial activity (measured at the 0 h). Although, during the first three hours, the observed values were 2-times higher than those of the reference sample. After 24 hours, the enzyme showed lower activity than the sample with ascorbic acid. However, enzyme activity was still 4.7-times higher than the reference value.
Next, DTT (dithiothreitol) and its effect on the stabilization of the enzyme was investigated. DTT showed a significant, but short-lived positive effect for the first 3 hours, while the activity after 24 h was comparable to the reference sample.
Finally, the combined effects of the most beneficial additives and salts, i.e., ascorbic acid (4 mM) and NaCl (50 mM) on the enzyme activity were examined ( Fig 6B ). The reference sample without any additives showed a typical decrease in the enzyme activity, which can be described by first-order kinetics. On the other hand, the presence of NaCl or ascorbic acid showed non-monotonous time-dependence of enzyme activity with a peak at 2 h and significantly higher activity compared to the reference. When both ascorbic acid and NaCl were present at the same time, alliinase showed lower activity at 1 h followed by the highest enzyme activity recorded after 2, 3 and 6 hours, which was more than 8x higher than for the reference sample.
The effect of alliin enantiomers
Synthetically prepared alliin (M w = 177.22 g/mol) exists as a mixture of two diastereomers, (2R)-2-amino-3-[(S)-prop-2-enylsulfinyl] propanoic acid (natural alliin), (2R)-2-amino-3-[(R)-prop-2-enylsulfinyl] propanoic acid, because of the asymmetry of the sulfoxide group. Therefore, the maximal velocities V max concerning natural L-(+)-alliin and diastereomixture of L-alliin were investigated. The value of V max for the mixture of diastereomers at 25°C was in average 52% lower compared to the natural alliin, which is given by a higher selectivity of alliinase toward naturally occurring substrate [ 52 , 53 ].
Storage stability of lyophilized alliinase
The formulation into a solid dry product is considered the optimal strategy for preservation of the enzyme activity and to achieve an acceptable shelf life. It is well-known that the presence of moisture may have a serious impact on the stability of lyophilized proteins [ 54 , 55 ]. However, to the best of our knowledge, the stability of lyophilized dry alliinase concerning temperature and various levels of relative humidity has not been yet explored. In order to study the adverse effect of water absorption, a dry lyophilized form of the enzyme was exposed to pre-set relative humidity (RH) levels, i.e., 0, 11, 50, 75 and 100% RH, and temperatures 25°C and 37°C, for one week, during which period the samples were periodically monitored for remaining enzyme activity. The corresponding values of equilibrium relative humidity are listed in Table 1 . The results showed that one-week storage under inert nitrogen atmosphere (0% RH) at 25°C (half-life t 0.5 = 43 days) and 37°C (t 0.5 = 32 days) preserved a high activity (>70%), which is due to the absence of both oxygen and moisture. The observed drop in activity after the first day compared to the activity of the lyophilized powder stored at -20°C (reference value of 100%) is attributed to the moisture uptake and thermal/oxidative stress associated with sample heating and handling. Enzyme activity and t 0.5 were reduced significantly with the increase of RH and storage temperature, as is summarised in Fig 7A and 7B . For instance, increasing the RH value from 0 to 11, 50, 75 and 100% resulted in a reduction of t 0.5 from 43 to 11, 5, 4 and 2 days, respectively, measured at 25°C. For both temperatures holds that activity of the enzyme stored at X% RH is approx. 2-times higher than of the same enzyme stored at X+20% RH. Therefore, for the long-term storage, the purified alliinase in the lyophilized form should be kept at conditions reducing water activity (frozen state) or under an oxygen-free atmosphere and sufficiently low RH level.
10.1371/journal.pone.0248878.g007
Fig 7
A) Effect of storage conditions on the activity of lyophilized alliinase; B) calculated half-life of alliinase vs relative humidity and storage temperature.
The initial activity of the lyophilized powder stored at -20°C was taken as a reference value.
Antibacterial potency of the alliin-alliinase system
Antibacterial assay of produced allicin was studied by a disk diffusion method against E . coli , P . putida and S . epidermidis as inhibition growth zones emerging after 24h of incubation period in the vicinity of the application sites ( Fig 8A ). Glycosidic antibiotic kanamycin (ATB) was used as a reference (50 mg/mL, 20 μL) for comparison of antibacterial effect for specific samples and bacterial strains [ 51 ]. Control samples, 100 mM alliin (S) and 5 mg/mL alliinase (E) applied separately, did not cause the formation of inhibition zones, i.e. no observable antibacterial effects. The results summarised for all studied bacterial strains and various alliinase amounts (from 0.001 to 10 mg/mL) in the presence of 100 mM alliin are shown in Fig 8B . Even the lowest tested concentration of alliinase (enzyme to substrate mass ratio = 5.6x10 -5 ) showed detectable inhibition zones for all studied bacterial strains. Higher enzyme content resulted in the formation of larger inhibition zones, which can be explained by the higher alliin conversion. Gram-positive S . epidermidis showed the largest inhibition zones with respect to allicin, followed by Gram-negative E . coli and P . putida (in that order).
10.1371/journal.pone.0248878.g008
Fig 8
A) Example of disk diffusion susceptibility testing of E . coli , P . putida and S . epidermidis (numbers correspond to alliinase concentration in mg/mL), ATB—kanamycin, E—enzyme, S—substrate; B) influence of alliinase concentration (0.001 to 10 mg/mL) in combination with 100 mM alliin on the diameter of inhibition zones.
Error bars represent standard deviation based on three independent experiments.
A series of additional measurements with finer concentration steps was conducted to evaluate the contribution of both the enzyme and the substrate against E . coli , which represents bacterial strain standing in the middle of the susceptibility spectrum of studied bacteria. Similarly to results shown in Fig 8 , the fixed concentration of alliin (100 mM) was mixed with alliinase solution of concentration 0.03 to 10 mg/mL. The diameter of the inhibition zone for 50 mg/mL kanamycin was used as the reference value for comparison of observed antibacterial effects of the produced allicin. Fig 9A illustrates that increasing alliinase amount provided a larger zone of inhibition, which is in agreement with results presented in Fig 8 . However, the application of alliinase concentration higher than 0.6 mg/mL did cause only marginal improvement of the inhibition of bacterial growth which indicates that most of the alliin was converted.
10.1371/journal.pone.0248878.g009
Fig 9
Effect of alliinase (A) and alliin (B) concentration on the inhibition of E . coli .
The inhibition zone for kanamycin (50 mg/mL, 20 μL) served as a reference value in both cases; x-axis in logarithmic scale; error bars represent standard deviation based on three independent experiments.
The influence of substrate concentration on the diameter of inhibition zones for E . coli was investigated. Inhibition zones corresponding to a mixture of alliin (in the concentration range of 10 to 800 mM) and 5 mg/mL alliinase are summarized in Fig 9B . It was confirmed, that with a higher initial concentration of alliin, the diameter of observed inhibition zones increased.
Bactericidal potency of in-situ formed allicin
The Live/Dead cell assay based on the propidium iodide and fluorescein staining was used to study the time-resolved antibacterial effect of allicin. Fig 10A presents the time evolution of bacterial viability immediately upon allicin addition (0 min), after 10, 20 and 30 min. The negative control bacterial samples without allicin, i.e., without alliin or enzyme, remained highly viable over 30 min. When both the enzyme and the substrate were present in the bacterial sample (enzyme—0.2 mg/mL, alliin—20 mM), the fraction of living cells exposed to allicin dropped after 30 min to 16% ( E . coli , S . epidermis ) and 23% ( P . putida ). The cell viability half-life (based on first-order kinetics) for E . coli , P . putida and S . epidermidis , was calculated as 12, 15 and 14 min, respectively.
10.1371/journal.pone.0248878.g010
Fig 10
Cell viability assay of bacterial suspension (300 μL) incubated for 0 to 30 min with a mixture of alliin (100 μL; 100 mM) and alliinase (100 μL; 1 mg/mL).
Ten fields or more were analyzed in triplicates; the scale bar corresponds to 50 μm; the total number of bacteria at a given time was taken as 100%.
The allicin bactericidal effect assay was verified in parallel by plate count method of viable bacteria ( Fig 11 ). The results of the plate count method were in agreement with Live/Dead assay, and it was revealed that alliin/alliinase had the strongest bactericidal effect on S . epidermidis , followed by E . coli and P . putida , which is in agreement with results summarised in Fig 8 .
10.1371/journal.pone.0248878.g011
Fig 11
Bactericidal effect of the allicin evaluated by the plate count method.
Left—agar plates with bacterial colonies after plating of the sample; Right—quantification of plate count method. The total count of emerging colonies from the untreated sample was taken as 100%.
Allicin-induced cell damage
The impact of allicin on the morphological changes of E . coli , P . putida and S . epidermidis was investigated using SEM analysis ( Fig 12 ). Gram-positive S . epidermidis changed morphology immediately upon allicin formation, which was indicated by the presence of cell fragments and intracellular matter enveloping a cluster of cells with typical cocci shape. After 30 min, no intact cells, and only cell fragments were observed. In contrast to S . epidermidis , Gram-negative P . putida presented small, but apparent morphological changes upon the addition of allicin (0 min), with wrinkles and small bumps observed on the cell body. After 30 min, most cells were disrupted, but some intact cells were still visible. Finally, E . coli showed no visible morphological changes immediately after allicin addition. After 30 min, some disrupted cells were found, but compared to the other two bacterial strains, the allicin-induced damage was relatively small. The results show clearly the difference in the dynamic of allicin action against Gram-positive and Gram-negative bacterial strains.
10.1371/journal.pone.0248878.g012
Fig 12
Cell morphology of E . coli , P . putida and S . epidermidis before (no allicin—The left column), right after (time 0 min—Middle column) and after 30 min upon allicin addition (the right column); Arrows show the first appearance of morphological changes.
Conclusion
In this comprehensive study, the isolated influence of numerous factors on operational and storage stability of purified garlic alliinase is presented and discussed. It was demonstrated that alliinase incubated in Tricine-KOH or PBS buffers, instead of phosphate buffers generally used in alliinase-alliin related studies, showed improvement in both short-term stability and activity. However, alliinase incubated in pure Tricine-KOH buffer was stabilized only for several hours. Therefore, the effect of chloride salts (NaCl, KCl, CaCl 2 or MgCl 2 ) on the stability of alliinase solution stability was explored. The comparison showed that NaCl had among studied salts the most noticeable stabilizing effect. In addition to salts, the contribution of other additives (antioxidants, osmolytes, chelating and reducing agents) was verified, from which ascorbic acid showed a superior stabilizing effect. The simultaneous presence of both NaCl and ascorbic acid stabilized alliinase in the solution for 6 hours at 25°C, which in the context of non-immobilized alliinase represents a considerable improvement, and an eight-times higher activity compared to Tricine-KOH buffer alone. The storage stability of lyophilized alliinase was examined for predefined relative humidity (RH) levels and storage temperatures (25°C and 37°C). It was demonstrated that with every 20% of RH, alliinase activity was reduced by 50% at specified times. Finally, alliinase activity and in-situ formation of antibacterial allicin were monitored by static disk diffusion method and dynamic viability assays showing the progress of allicin-induced cell damage on selected Gram-positive and Gram-negative bacterial strains.
Thorough knowledge regarding the time-dependent response of purified alliinase in the non-immobilized form to various factors, i.e. pH, temperature, type of buffer, and the presence of a variety of additives, significantly contribute to the development of formulations containing stabilized alliinase for the synthesis of allicin or its chemical analogues. The potential use of environmentally friendly and time-proved Allium chemistry in human or veterinary medicine (treatment of bacterial infections, reduction of blood pressure) or agriculture (plant protection, urease inhibitors in fertilizers) is great, and we believe, that the findings presented here will contribute towards the development of nature-based products harnessing the volatile nature of highly potent allicin.
Supporting information
S1 Raw images
(TIF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Reactive oxygen species including hydrogen peroxide, superoxide anion, and singlet oxygen are generated as by-products of cellular metabolism primarily in mitochondria, and play a predominant role in many pathological conditions, including immune suppression, photo-carcinogenesis, and photo-aging [ 1 – 4 ]. Excessive generation of reactive oxygen species in the skin is a major contributor for various cutaneous pathologies [ 5 ]. Using antioxidants to prevent oxidative skin damage appears to be a promising approach. Naringenin (5,7-dihydroxy-2-(4-hydroxyphenyl)chroman-4-one, C 15 H 12 O 5 , MW 272.3) a flavanone found in many citrus fruits, has been proven to possess anti-inflammatory, antioxidant, and free radical scavenger properties [ 6 , 7 ]. Furthermore, previous studies [ 8 , 9 ] reported that naringenin can increase the tyrosinase activity and melanin content, demonstrating naringenin can be used to prevent oxidative skin damage. Nevertheless, naringenin is a poor water-soluble compound and has minimal oral bioavailability (approximately 5.8%) owing to its largely hydrophobic ring structure [ 10 – 12 ]. Therefore, the purpose of this study was to design a naringenin formulation for topical administration.
In recent years, nano-scale structures such as microemulsions, ethosomes, liposomes and solid lipid nanoparticles have attracted increasing attention because they can provide a better chance for adhesion to biological membranes while delivering therapeutic drugs in a controlled manner. Moreover, nano-scale structures are capable of increased drug loading, sustained release, and the promise of tissue-specific targeting [ 13 – 22 ]. Liposomes are microscopic vesicles with an aqueous core surrounded by one or more outer shell(s) composed of phospholipids in a bilayer. They can incorporate a variety of hydrophilic and hydrophobic drugs, improve the accumulation of the drug at the administration site, and reduce side effects [ 23 – 26 ]. Hence, liposomes have been widely used as safe and effective drug vehicles in topical treatment of disease [ 27 – 30 ]. Modified liposomes such as elastic liposomes were first described by Cevc and Blume [ 31 ]. They consist of phospholipids and a single chain surfactant such as deoxycholate, sodium cholate, Tween 80 or Span 80, which can destabilize the lipid bilayers and provide greater flexibility compared to the liposome itself [ 32 – 34 ]. Numerous studies have demonstrated that elastic liposomes could provide potentially deeper permeation of drugs compared to conventional liposomes [ 15 , 35 , 36 ]. Thus, the present work was aimed at the development of an effective elastic liposome for naringenin topical application. With this purpose, different elastic liposome formulations were prepared. The vesicle size, surface charge and encapsulation efficiency were determined. The permeation properties of drug from these delivery systems through rat-excised skin were evaluated and compared with those of a saturated drug aqueous solution. The stability of formulation and skin irritation caused by drug-loaded elastic liposomes were also evaluated for assessing the clinical utility of elastic liposomes.
Materials and Methods
Materials
Naringenin and hesperetin were purchased from Tokyo Chemical Industry (Tokyo, Japan). Polyoxyethylene sorbitan monooleate (Tween 80) and propylene glycol (PG) was from J. T. Baker (Phillipsburg, USA). Epikuron-200 (containing more than 92% of phosphatidylcholine and others of lysoPC, phosphatidic acids, and triglycerides.) was acquired from Cargill, Inc. (Minnetonka, Minnesota, U.S.). Cholesterol and paraformaldehyde were purchased from Sigma-Aldrich (St. Louis, Missouri, USA). All other chemicals and solvents were of analytical reagent grade.
Naringenin-loaded elastic liposomes preparation
In order to easily evaluate the effect of components, a two-factor three-level factorial design [ 37 ] was used to prepare different naringenin-loaded elastic liposome. Each formulation (4 mL) contained 20 mg naringenin and 340 mg other ingredients of cholesterol of 5~15%, Tween 80 of 10~20% and Epikuron-200. Cholesterol and Tween 80 were set as formulation factors. The compositions of all different elastic liposome are listed in Table 1 .
10.1371/journal.pone.0131026.t001
Table 1 The composition and physicochemical characteristics of naringenin-loaded elastic liposomes.
Cholesterol
Tween 80
Size
PI
Zeta
EE
code
mg
code
mg
(nm)
mV
%
F1
+1
52
+1
68
155.0±1.9
0.16±0.03
-16.1±0.7
9942
F2
+1
52
0
52
177.7±3.5
0.19±0.05
-13.4±0.3
99.43
F3
+1
52
-1
34
ND-
F4
0
34
+1
68
123.7±1.8
0.16±0.02
-11.0±3.7
99.52
F5
0
34
0
52
138.3±1.7
0.13±0.03
-12.3±1.2
99.56
F6
0
34
-1
34
176.1±1.8
0.22±0.02
-15.6±1.5
99.53
F7
-1
17
+1
68
ND-
F8
-1
17
0
52
147.3±2.9
0.11±0.03
-8.5±1.5
99.59
F9
-1
17
-1
34
132.3±1.3
0.17±0.02
-2.2±2.4
99.64
Each liposome formulation contained 20 mg naringenin and 340 mg of other ingredients of epikuron, cholesterol and Tween 80 in 4 mL aqueous phase.
PI: polydispersity index; EE: Encapsulation efficiency.
ND: no data because the size was over detected level.
The accurately weighed amounts of naringenin, epikuron, Tween 80 and cholesterol were placed in a round-bottom flask, and the mixture was dissolved in 4 mL mixture solvent of chloroform-methanol at a ratio of 1: 1. The organic solvent was removed by rotary evaporation under reduced pressure at 65°C, and then the solvent traces were removed by maintaining the lipid film under a vacuum overnight. The deposited lipid film was hydrated with 4 mL aqueous solution by a probe-type sonicator (UP50H,Hielscher Ultrasonics, Teltow, Germany) for 10 min at 50 W [ 34 ]. The final concentration of drug-loaded liposomes was 5 mg/mL.
Physicochemical characterizes of elastic liposomes determination
The mean vesicle size and zeta potential (surface charge) of the naringenin-loaded elastic liposomes were measured using Zetasizer 3000HSA (Malvern Instruments, Malvern, UK) with a helium-neon laser with a wavelength of 633 nm at room temperature. A 1:80 dilution of the drug-loaded liposome was made using double-distilled water for the measurements. The size values were given as a volume distribution. Analysis time was kept at 60 sec, and average size and zeta potential of the vesicles were determined.
Encapsulation efficiency determination
The entrapment percentage of naringenin loaded in elastic liposomes was determined by an ultracentrifugation method. The sample was centrifuged at 120,000 rpm for 1 h at 4°C in a Hitachi CS150GXL ultracentrifuge (Tokyo, Japan) to separate the incorporated drug from the free form. There were no vesicles in the supernatant after Zetasizer examination. The supernatant was analyzed by high-performance liquid chromatography (HPLC) [ 38 ] to determine the drug encapsulation percentage of the total naringenin load. The percentage encapsulation efficiency of naringenin in elastic liposome was calculated as: Each experimental was performed in triplicate, and the data reported is the mean value.
Encapsulation efficiency(%) = Drug content Total drug loaded × 100 %
Skin permeation and drug deposition studies
The experimental protocol was approved by the Institutional Animal Care and Use Committee of Kaohsiung Medical University (Kaohsiung, Taiwan). The Committee confirmed that the permeation experiment followed the guidelines as set forth by the Guide for Laboratory Fact lines and Care. The in vitro skin permeation and skin deposition studies of naringenin-loaded elastic liposomes (5 mg/ mL) and control group (the saturated aqueous solution of drug and drug dissolved in 10% Tween 80 of 5 mg/ mL) were conducted by using a modified transdermal Franz diffusion cell. The effective diffusion area of the diffusion cell and receptor cell volume were 3.46 cm 2 and 20 mL respectively. The abdominal skin was excised from a Sprague-Dawley rat weighing 275–300 g, and then mounted on the receptor chamber with the stratum corneum side facing upward to the donor chamber. Samples of 1 mL were placed in the donor chamber and occluded by parafilm. Twenty mL of pH 7.4 phosphate buffer saline containing 40% PG (drug solubility of 941.50 ±3.54 mg/mL) was placed in the receiver chamber. The temperature of receiver medium was maintained at 37±0.5°C by thermostatic pump and was constantly stirred at 600 rpm by a magnetic stirrer during the experiment. At determined intervals, i . e ., 1, 2, 3, 4, 6, 8, 10, and 24 h, one milliliter of receptor medium was withdrawn via the sampling port and was quantified for naringenin level by a modified HPLC method [ 38 ].
At the end of the skin permeation experiments, the donor phase was removed and the rat skin was washed with deionized water to remove the residual naringenin on the skin surface. Then, the skin was dried with cotton wool and the stratum corneum (SC) was removed from the rest of the skin by tape-stripping the skin with 11 adhesive cellophane tapes (Scotch Book Tape no. 845, 3M, St Paul, MN) [ 39 , 40 ]. The first tape was discarded. The other stripping tapes were put in glass tubes containing 2.0 mL of methanol and then shaken horizontally for 1 h. The solution was filtered through a 0.45 mm membrane (Sartorius, Goettingen, Germany). Naringenin in filtrate was determined by HPLC. The epidermis was separated from the dermis with heat application at 80°C for 3 min and the help of forceps [ 41 ]. Then, the epidermis and dermis were separately cut into small pieces to extract the drug content present in the skin with methanol. The resulting solution was centrifuged for 10 min at 8533 g, and filtered through a 0.45 mm membrane. The filtrate was analyzed for naringenin by HPLC.
Chromatographic condition
Hitachi L-7100 series HPLC system and a LiChroCART RP-18e column (125×4 mm I.D., particle size 5 μm) were used in this study. The detection wave was set at 281 nm. The mobile phase of 0.5% triethylamine (adjusted to pH 3.0 by acetic acid) containing 28% acetonitrile was delivered at a rate of 1.0 mL/min. Hesperetin of 200 μg/mL was used as internal standard. The method was successfully validated with coefficient of variation (CV, %) of 3.72%, relative error (RE, %) of 7.57% and a determination coefficient (r) of 0.9998. The limit of quantitation was 0.1 μg/mL.
Skin irritation determination
The experimental protocol was approved by the Institutional Animal Care and Use Committee of Kaohsiung Medical University (Kaohsiung, Taiwan). The committee confirmed that the permeation experiment followed the guidelines as set forth by the Guide for Laboratory Fact Lines and Care. The hair on the abdomen of the rat was shaved before the skin was randomly divided into three study groups. A glass ring with 2.54 cm 2 was adhered to the abdomen skin. The experimental naringenin-loaded elastic liposome (F4), aqueous water (normal control group) and 0.8% paraformaldehyde (positive control group) of 0.5 mL were loaded into the glass ring and left for 24 h [ 28 , 42 ]. Then, the tested skin regions were excised and fixed in 4% buffered formaldehyde solution for 24 h. The skins were then embedded in paraffin and sliced transversely. The sections were rehydrated stepwise, stained with hematoxylin and eosin, and observed using an optical microscope (Nikon Eclipse Ci, Tokyo, Japan).
Stability determination
The naringenin-loaded elastic liposome was stored in dark-brown bottles for protection from light. The stability of drug-loaded liposome formulation (F4) was evaluated via physicochemical properties and drug content at 4°C. The physical stability was evaluated by mean vesicle size and zeta potential measurement over a three-month period.
Data analysis
All experimental measurements were performed in triplicate. Result values were expressed as the mean value ±standard deviation. Statistical analysis of differences between the experimental formulations was performed using ANOVA test provided by Winks SDA 6.0 software (Texasoft, Cedar Hill, TX, USA). The post hoc Newman—Keuls test was used to check individual differences between groups. A 0.05 level of probability (p < 0.05) was taken as the level of significance.
Results and Discussion
Typically, liposomes are composed of neutral phospholipids, which are biocompatible molecules. Cholesterol is often added to improve mechanical stability of the bilayer and decrease leakage of the encapsulated materials. Cholesterol has been shown to increase mechanical strength of membrane, affect its elasticity, and increase the packing density of lipid via the “ordering and condensing” effect [ 43 – 46 ]. Some studies have indicated that the cholesterol content might be the crucial factor for the effective delivery of liposome-entrapped drugs into the skin [ 47 , 48 ].
Conventional liposomes are reported to remain confined to the upper layer of the SC and to accumulate in the skin appendages, with minimal penetration to deeper tissues, due to their large vesicle size and lower flexibility of membrane [ 14 , 49 – 51 ]. Tween 80 is a single chain surfactant. It can act as an “edge activator” and destabilize the lipid bilayers of traditional liposomes, and this then provides greater flexibility of membrane [ 36 , 52 , 53 ]. However, liposomes in the present high concentration of Tween 80 are unstable; hence, the effect of concentration of cholesterol and Tween 80 on the physicochemical characteristic and skin permeation capacity of liposome was investigated in this study. Naringenin-loaded elastic liposomes containing different amounts of cholesterol and Tween 80 were prepared.
Vesicle size, zeta potential and encapsulation efficiency
The average vesicle size, zeta potential (surface charge) and encapsulation efficiency of experimental formulations are listed in Table 1 . Except for formulations F3 (cholesterol was at a high level and Tween 80 was at a low level) and F7 (cholesterol at low levels and Tween 80 at high levels), the average vesicle size of all formulations ranged from 123.7 to 177.7 nm. The polydispersity index values of the elastic liposomes were obtained in a range of 0.11–0.22, showing homogenous size distribution in all formulations. From Table 1 , it can be found that the average size of elastic liposomes tended to become large when formulated with higher levels of cholesterol. On the contrary, the average size tended to diminish, when formulated with higher levels of Tween 80. A possible explanation may be that the edge activator, Tween 80, destabilizes the lipid bilayers of liposomes, thus resulting in smaller vesicles [ 14 ]. However, the smallest size was obtained at elastic liposomes containing a medium level of cholesterol and a high level of Tween 80.
The surface charges of experimental formulations ranged from -2.2 to -16.1 mV. The drug-loaded elastic liposomes had lower surface charge when a high level of cholesterol was incorporated. The results agreed with a previous study, which reported that increasing the level of cholesterol in a phospholipid membrane decreases surface charge in the physiological environment [ 54 ].
Numerous studies have reported that the elastic liposome could significantly increase solubility of hydrophobic and hydrophilic compounds [ 15 , 55 ]. Naringenin is a poor water-soluble compound; its solubility in water was 41.76 ±0.51 μg/mL. In this study, 5 mg/mL of drug was loaded into the elastic liposomes. The encapsulation efficiency of all experimental formulations was larger than 99%, indicating that elastic liposomes should be a good carrier for naringenin.
Skin permeation and drug deposition
The cumulative amount transported through rat skin was plotted as a function of time, and the linear regression analysis was used to determine the permeation rate (flux) and permeation mechanism of drug. The result showed that the permeation profiles followed a zero-order model (R 2 > 0.9915). The permeation rate, cumulative amount at 24 h and deposition amount in three skin layers including SC, epidermis and dermis layer after 24 h treated with naringenin-loaded elastic liposomes with different levels of cholesterol and Tween 80 are presented in Figs 1 and 2 . The saturated aqueous solution and 5 mg/mL of drug dissolved in 10% Tween 80 solution were used as control groups to evaluate the enhancement effect of formulations. The permeation rate and cumulative amounts at 24 h were 0.25±0.1 μg/h/cm 3 and 4.8±2.6 μg/cm 3 for saturated aqueous solution, 0.37±0.15 μg/h/cm 3 and 14.4±3.1 μg/cm 3 for 10% Tween 80 solution, and 0.25±0.05~0.76±0.21 μg/h/cm 3 and 6.4±0.8 ~16.5±3.4 μg/cm 3 for elastic liposomes ( Fig 1A and 1B ). The permeation rate and cumulative amount were increased 1.5-fold and 3.0-fold by using permeation enhancer (Tween 80), indicating Tween 80 was an effective penetration enhancer (p<0.05) [ 56 , 57 ]. When elastic liposomes were used as carrier vehicles, the enhancement ratios were 1.0~3.0-fold for permeation rate and 1.3~3.5-fold for cumulative amount. The result showed that the composition proportions of elastic liposomes would affect the enhancement degree of drug transportation through skin. An appropriate composition proportion of formulation of F1 with high-level cholesterol and Tween 80 could obtain the highest enhancement effect. Its enhancement effect was similar to that of 10% Tween 80 solution. The result agreed with previous studies, which reported that liposome-like vesicles and/or penetration enhancer-containing vesicles could improve the transportation through skin of the drug [ 27 – 29 ].
10.1371/journal.pone.0131026.g001
Fig 1
The permeation rate and transdermal amount at 24 h of naringenin-loaded elastic liposomes through rat skin.
10.1371/journal.pone.0131026.g002
Fig 2
The skin deposition amount of naringenin-loaded elastic liposomes after 24 h treatment.
The deposition amounts of drug in SC, epidermis, dermis layers and total deposition amounts were 0.5±0.1, 1.8±0.5, 0.3± 0.1, and 2.6± 0.7 μg/cm 3 for the saturated aqueous solution-treated group and 4.4±1.2, 9.3±2.7, 2.2±1.6, and 15.9±2.5 μg/cm 3 for the 10% Tween 80 solution-treated group. The result showed that naringenin was deposited in different skin layers, particularly in the epidermis layer. Using elastic liposomes as carrier also showed similar results ( Fig 2 ). Hence, the value of total deposition amount was used to evaluate the efficacy of formulations. In comparison with the two control groups, the total deposition amount increased about 6.1-fold when 10% of Tween 80 was used as permeation enhancer (p<0.05) [ 56 , 57 ]. When elastic liposomes were used as drug carrier vehicles, the total deposition amounts (19.0±3.7 ~30.7±13. μg/cm 3 ) further increased 1.2~1.9-fold compared with the Tween 80-treated group, indicating that the elastic liposomes had even more potential for naringenin skin transportation (p<0.05). Furthermore, the enhancement efficiency of these elastic liposomes on the deposition amount was more than that on the permeation amount, indicating that the present liposomes were more suitable for topical skin target of naringenin. This finding is in agreement with published studies that reported use of liposomes as carrier may produce higher drug concentrations in the skin layers and lower systemic concentration [ 14 , 15 , 36 , 58 ]. The enhancement mechanism of skin delivery includes but is not limited to: the small size of vesicles; the cell membrane-like structure of the vesicles having good biocompatibility; formation of drug reservoirs in the skin; specific liposome-skin interactions; and the elastic vesicles being able to squeeze through intercellular regions of the SC under the influence of the transepidermal water-activity gradient. [ 14 , 51 , 59 ].
In evaluation of the effect of physicochemical characteristics and compositions of liposomes on the permeation parameters, a non-significant relationship was found between the average size vs permeation rate and cumulative amount ( p> 0.05). A possible explanation may be that elastic liposomes being flexible should be able to more easily squeeze through the skin without being affected in size. On the contrary, the total deposition amount in skin increased by increasing the amount of Tween 80 ( Fig 2 ). The result is consistent with previous studies [ 14 , 55 , 58 , 60 ], which reported surfactant “Tween 80” can be inserted into the lipid bilayers of liposome thereby creating a “softened” and “flexible” bilayer membrane, thus facilitating elastic liposome rapid distribution into the skin retaining the drug on, in, and below the skin barrier. In cases of elastic liposomes with different added amounts of cholesterol, it was found that the formulation with medium level(s) of cholesterol showed highest skin deposition. It is possible that excess cholesterol resulted in a rigid membrane, and insufficient cholesterol resulted in a looser membrane, both making less-permeable conditions [ 46 , 47 ]. However, the naringenin-loaded elastic liposome with high-level Tween 80 and medium-level cholesterol showed highest total deposition of 30.7±13.7 μg/cm 3 , which was 11.8-times that of saturated aqueous solution.
Skin irritation
Formulations might elicit primary skin irritation. Rat skin irritation experiments were conducted to assess the potential irritant effects of the developed elastic liposome formulation. The 0.8% v/v aqueous solution of paraformaldehyde was used as a standard irritant [ 28 , 42 ]. As shown in Fig 3B , a slight edema exfoliation of the stratum corneum, and collagen dissociate was caused by application with paraformaldehyde. Non-significant edema and erythema was found in tested elastic liposome ( Fig 3C ) and aqueous solution control ( Fig 3A ), when compared to the positive control group, indicating that the experimental elastic liposome formulation appeared to be safe for transdermal delivery.
10.1371/journal.pone.0131026.g003
Fig 3
The microstructure of a rat abdominal skin section, viewed under a light microscope.
Stability
After three months of storage at 4°C, non-aggregation and creaming were observed. The size and zeta potential of drug-loaded liposome had slight change from 123.7 to 128.1 nm and -11.0 to 12.7 mV respectively, showing non-significant difference. The residual drug content of tested drug-loaded elastic liposome was 98.89±3.90%, indicating that the formulation was stable.
Conclusions
The naringenin deposition amounts in SC, epidermis, dermis and total skin were significantly increased by using elastic liposomes as drug carrier when compared to the saturated aqueous solution and Tween 80-treated groups. The added contents of cholesterol and Tween 80 showed significant influence on the physicochemical properties and permeation capacity of naringenin from elastic liposomes. The naringenin-loaded elastic liposome with high-level Tween 80 and medium-level cholesterol showed highest deposition of the drug in skin, which was 11.8-times that of the saturated aqueous solution. It has also been demonstrated that naringenin-loaded elastic liposome was stable after three months of storage and produced less skin irritation than that of the standard irritant group. The result suggests that elastic liposome is a promising carrier for naringenin topical application.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Despite advances in clinical practice and research, falls remain the most common adverse event in hospitals. More than 240,000 in-hospital falls occur each year in England and Wales with falls being the most commonly reported safety incident in National Health Service hospitals [ 1 , 2 ]. Falls prevention programs for hospitalised older people are multifaceted, reflective of the complex causal pathway for falls. With increased complexity comes increased risk of implementation failure. Implementation of falls prevention programs can be influenced by several factors including environmental and contextual issues; staff knowledge, beliefs and attitudes; organisational culture and climate; staff workloads; and access to appropriate equipment and resources [ 3 ]. An understanding of these factors can inform the development of an implementation plan that addresses the barriers and enablers to the implementation of the intervention.
There is limited information about the barriers and enablers to the implementation of falls prevention in acute hospitals. Two survey based studies implemented across five acute care hospitals in Singapore showed that nurses perceived the greatest barriers to implementation of fall prevention practices to be: staff and patient education; lack of motivation in staff; availability of support staff; and access to facilities and equipment [ 4 – 5 ]. These barriers were also reported in a recent Cochrane review of 11 RCTs [ 6 ]. Other barriers reported in the review included: leadership support at the organisational and unit level; engagement of front-line staff in program design; pilot-testing to identify potential barriers to implementation; provision of data about falls; and changes in nihilistic staff attitudes about falls prevention were associated with successful implementation of inpatient falls prevention programs in hospitals [ 6 ].
Tailoring the implementation of falls prevention programs to the local context optimises implementation. The 6-PACK falls prevention program is nurse-led ( Box 1 ) and was developed as part of continuous quality improvement activities at an Australian acute hospital. An evaluation reported that fall-related injuries appeared to reduce following the implementation of the program [ 7 ]. This led to a multi-centre randomised controlled trial (RCT) to further establish the efficacy of the 6-PACK program [ 8 , 9 ]. Whilst 6-PACK intervention components are required to remain fixed in an RCT, the implementation of the program was tailored to the local context of the intervention wards to ensure implementation was optimised.
Box 1. The 6-PACK program
The 9 item fall-risk tool [ 10 ] is updated for each patient each shift by their treating nurse. Patients identified as high falls risk receive:
A ‘falls alert’ sign positioned above their bed, and one or more of the following interventions:
Supervision of patients in the bathroom
Ensuring patients’ walking aids are within reach
A toileting regime
A low-low bed
A bed/chair alarm
The COM-B model was developed by condensing concepts from 19 frameworks of behaviour change identified in a systematic review by Michie and colleagues [ 11 ]. The COM-B model demonstrates human behaviour (B) as the interaction between physical and psychological capabilities (C) that utilise social and environmental opportunities (O) via motivators (M) that are reflective (‘thinking’ with the head) or automatic (‘thinking’ with the heart). It has been widely adopted in implementation and health services research [ 12 , 13 ].
The aim of this study was to use the COM-B model to identify the perceived barriers to, and enablers of, implementation of the 6-PACK program from the perspectives of nurses and senior staff to inform the implementation plan. Specifically we sought to identify physical and psychological factors (capability); environmental and social contexts (opportunity); and reflective and autonomic processes (motivation) that are perceived to be barriers or enablers of the successful implementation of the 6-PACK program. In addition, we sought to gain insights into what strategies could be applied to optimise successful implementation of the 6-PACK program in the RCT.
Materials and methods
Design
A multi-centre mixed methods study. This study was part of the 6-PACK project that incorporated a three-year research plan: 1) Studies of current falls prevention practice and moderators (pre-implementation) [ 14 ]; 2) A cluster RCT testing 6-PACK effectiveness ( S1 Appendix ), including economic [ 15 ] and program evaluations (implementation); and 3) An assessment of sustainability of practice change and outcomes (maintenance). The study reported here forms part of the pre-implementation stage.
Participants and setting
Detailed information about participants, recruitment and data collection are reported elsewhere ( S2 Appendix ). In brief, this study involved staff from 16 medical and 8 surgical wards participating in the 6-PACK RCT. Nurses were invited to complete the survey and participate in focus groups. Key informant interviews were conducted with senior staff (Nurse Unit Managers (NUMs), senior physicians, Directors of Nursing (DONs) and clinical services, falls prevention leaders and senior personnel involved in quality, safety and risk management).
Nurse survey
The 42 item survey was developed with items related to beliefs about falls; current falls prevention practice; 6-PACK program components; best practice guidelines and key recommendations; and reporting practices were included. Participants indicated their level of agreement using a five point Likert scale ranging from strongly disagree to strongly agree. Seven items related to the COM-B domains: one to capability, three to opportunity and three to motivation ( Table 1 ).
10.1371/journal.pone.0171932.t001
Table 1 Mapping of survey, focus group and interview questions to the COM-B domains [ 11 ].
Survey
Focus group
Interview
Questions/Statements
Capability: The individual’s psychological and physical capacity to engage in the activity concerned.
✓
✓
What strategies would you recommend we use when implementing the 6-PACK program? Why?
✓
✓
What learning can we take from other program implementation experiences on your ward? What were some of the barriers? What would you do differently next time? What worked well?
✓
You can’t stop older people from falling.
✓
✓
Do you believe falls can be prevented? What interventions do you feel are most important?
Opportunity : The factors that lie outside the individual that make the behaviour possible or prompt it.
✓
✓
Who are the critical people that need to be involved in falls prevention activities at your hospital?
✓
✓
What strategies/factors would you consider to be essential to sustaining programs like the 6-PACK? Please explain.
✓
What falls prevention activities are currently occurring/or planned for the hospital? Do you perceive these activities to be complementary or inhibitory to the 6-PACK implementation on the intervention wards? Please explain.
✓
Who should we involve in the processes of implementing the 6-PACK this hospital? What do you see their role will be? How do you rate the relative importance of these individuals or group in terms of making the implementation successful?
✓
Who do you anticipate may be obstructive/resistive to the implementation of 6-PACK? Why? (Knowledge, beliefs and skills? Attitudes and opinions? Conflicting demands?) What strategies do you recommend to better engage these people? (Incentives and motivators?) What strategies do you recommend to inform/approach/involve key staff in the change process?
✓
What system level barriers do you feel may exist to implementing the 6-PACK program? E.g. Equipment and staffing resources, communication, leadership and teamwork, environmental constraints (e.g. budgets, redevelopments, restructuring)
✓
Leadership and supervision for falls prevention practice.
✓
An active falls prevention leader is essential for falls prevention programs to be successful on my ward.
✓
This feedback [about how I use falls prevention interventions] helps me use falls prevention interventions more effectively.
Motivation: Reflective and automatic mechanisms that activate or inhibit behaviour. Includes habitual processes, emotional responding, as well as analytical decision-making.
✓
✓
What effect do you feel audit, feedback and reminders will have on the effectiveness of the 6-PACK program implementation? Can you provide examples of when these have been effectively used previously?
✓
There are more important things I should do than falls prevention interventions for my high falls risk patients.
✓
Incident reporting provides us with a way of measuring how we are going with patient falls.
✓
It is not my responsibility to stop patients from falling.
Focus groups and key informant interviews
Discussion guides for the focus groups and key informant interviews based on the COM-B framework were developed to elicit ward nurse and senior staff views on barriers and enablers to implementing the 6-PACK program ( Table 1 ). Focus groups and key informant interviews at each hospital were scheduled and conducted. Senior staff nominated by the DON at each hospital received a letter of invitation to participate in an interview from the research team. The perspectives of senior staff were sought to understand hospital practices, policies and the organisational context influencing falls prevention interventions.
Data analysis
Descriptive statistics were calculated for survey responses using Stata MP v13 statistical software. Analysis of interview and focus group data was continuous with deductive coding being applied for the three COM-B domains and emerging themes explored and tested for applicability and consistency. Three researchers independently coded and recoded transcripts using Nvivo (QSR International 2012), continually working back and forth between data sources in a process of open, axial and thematic coding [ 16 , 17 ]. Discrepancies were resolved by discussion and consultation with the investigator team as required. Quantitative and qualitative data were analysed separately with a process of triangulation applied at the interpretation stage of the analysis whereby findings from each component were considered to determine whether findings were convergent, complementary or contradictory [ 18 ].
Ethics
This study was approved by Monash University Human Research Ethics Committee–CF11/0229–2011000072 and each of the relevant hospital ethics committees. Participants were given verbal information about the study and asked to sign consent forms if they were interested in participating.
Results
Study participants
Overall, 702 surveys were distributed with 420 (60%) returned. The majority of respondents were registered nurses (74%); staff working on medical wards (75%); and staff with at least one year of experience at the hospital (74%). Twelve focus groups involving 96 nurses and 24 interviews with senior staff (SS) were conducted. Six DONs, seven NUMs, one Clinical Risk Coordinator, one Quality and Safety Manager, one clinical program nurse manager, and eight nursing educators participated in the interviews.
Each of the COM-B domains and arising sub-themes are described below ( Table 2 ) in the context of barriers and enablers to the implementation of the 6-PACK program. Implementation strategies suggested by the participants have also been described and summarised in Table 3 .
10.1371/journal.pone.0171932.t002
Table 2 Mapping of barrier and enabler themes to COM-B domains.
COM-B domain
Theme
Capability
Barrier
• Management of complex patients (N) • Belief that falls are inevitable (N) • Ward layout (N)
Enabler
• Training and education (N and SS) ■ Face-to-face education ■ Case study based teaching
Opportunity
Barrier
• Lack of resources (N)
Enabler
• Use of falls data (SS) • Feedback on progress (N) • Competition (SS) • Leadership (SS)
Motivation
Barrier
• Lack of ownership (SS) • Complacency (SS, N)
Enabler
• Goal to reduce falls (SS) • Engaging staff in falls prevention • Emotional impact of patient falls (N) • Improved patient outcomes (SS) •Audit, reminders and feedback (N and SS)
N = nurses, SS = senior staff
10.1371/journal.pone.0171932.t003
Table 3 Strategies to optimise successful implementation of the 6-PACK program
COM-B domain
Rationale
Implementation strategy
Capability
• Improve knowledge and skills • Support attitudinal change • Model new behaviours
• Regular practical face-to-face education and training for nurses (ward walk arounds, small interactive group sessions)
Opportunity
• Provide and discuss data • Inform about progress
• Provision of falls data • Leadership and champions (ward champions, Nurse Unit Managers) • Provision of equipment • Newsletters and posters communicating progress, achievements and stories
Motivation
• Reinforce key strategies for falls prevention • Troubleshoot and provide support • Demonstrate commitment to project
• Compliance audits • Reminders and feedback • Reward and recognise change in practice and leadership
Capabilities
Management of complex patients
Implementing falls prevention interventions was viewed as difficult, particularly when treating patients with complex health issues. A nurse’s ability to manage multiple risks including pressure areas, medications, nutrition and falls was described as a “ daunting balancing act” .
When you’re looking at those more elderly, confused, aggressive patients, it’s weighing up between the falls risk versus the medication management to keep them settled…it is a balancing act. (SS3, Hospital (H) 3)
Belief that falls are inevitable
Many nurses reported that they were unable to prevent falls, despite feeling they had knowledge in falls prevention. They identified a number of patient characteristics that they perceived were associated with high falls risk and not amenable to falls prevention interventions.
We’ve got patients on the ward who are in the high visibility area, on low-low beds, have the pressure sensor, [yet] they are still falling… I don’t think falls can be prevented. (Nurse, H3)
We’ve got dementia patients… You can do as much as you can, and [falls are] still just going to happen…I don’t think falls can be prevented. (Nurse, H3)
Only 46% of nurses responding to the survey disagreed with the statement ‘You can’t stop older people from falling’, while 23% were undecided. This suggests discord in beliefs regarding the inevitability of falls. Senior staff were less likely than nurses to accept the inevitability of falls based on patient characteristics, and emphasised the need to minimise the impact of falls.
[Falls] should be preventable. We shouldn’t have them. I think it’s about changing that perception and that belief, and that awareness, [to be] that actually any fall is wrong, it shouldn’t have happened. (SS2, H6)
Ward layout
The layout of the ward was often perceived by nurses as a hindrance to surveillance. Single rooms made it difficult for nurses to physically move efficiently from one patient to another.
Sometimes we have four [high risk] patients in three different rooms, it’s a disaster…how do you get to look at everyone at the same time? (Nurse, H5)
Say if you’re stuck in a room or a bathroom with someone and someone else buzzed…you mightn’t see that [patient] for 20 minutes because you’re in doing a massive dressing or [something else]. (Nurse, H1)
Training and education
Improving knowledge and skills through training and education sessions were identified as enablers to falls prevention practice. Survey data indicated only 32% of nurses felt they received useful training from falls prevention leaders. Senior staff valued e-learning methods as they believed that information could be conveyed efficiently.
I put a module of falls strategies on e-Learning so nurses can access information on falls. (SS1, H5)
Nurses raised issues such as lack of access to computers and “no time to get to the computers ” as a barriers to e-learning education. Nurses specified that although e-learning was convenient, practical and hands-on training on the ward with case studies was preferable to increase their capabilities in falls prevention.
We all prefer face-to-face learning rather than e-learning…I think you learn more with real-life situations. (Nurse, H5)
Senior staff valued ongoing feedback and case review as an effective means of enhancing falls prevention knowledge.
I educate the staff every month about falls we’ve had…I explain strategies that could have been improved. I go and speak to staff who have been involved in a fall and find out why a strategy wasn’t put in place, what were the obstructions to that, and the circumstances around it. (SS3, H3)
In addition to discussions on the delivery mode for education, staff raised suggestions for education content. Nurses identified the specific need for education on the treatment of delirium and management of patients with cognitive impairment. Senior staff highlighted the need for training on how to connect fall-risk tool scores to appropriate interventions.
Opportunity
Access to resources
A key barrier identified in the implementation of the 6-PACK program was access to resources. One of the interventions of the 6-PACK program was to put high risk patients on a low-low bed, however there was not a sufficient number of beds available on the wards for nurses to use. This was further complicated by a lack of tracking systems of where the beds are within a hospital.
We have about 12 low-low beds, which is not sufficient. There is no system of tracking where the Low-Low bed is. The poor nurse has to ring Environmental Services or six different wards to see if they have a Low-Low bed. (SS2, H2)
If I’ve identified someone as a high-falls risk, I’ve got to put an intervention in place, [but] we don’t have the resources [equipment] to do that. (Nurse, H3)
Use of data to drive practice change
Senior staff highlighted the need to ensure that nursing staff understood the extent of the problem of falls on the wards. This involved presenting data on the trends and benchmarking of ward falls across wards.
Here’s our data, this is what we’re looking like and your patient safety boards…I think that’s really valuable because it puts your performance up there to be seen as a trend; they can be benchmarking against themselves. (SS3, H1)
The majority of nurses surveyed (75%) agreed with the statement ‘incident reporting provides us with a way of measuring how we are going with patient falls’ . Providing this feedback on progress in falls prevention to nurses was seen as an opportunity to encourage and promote practice change.
Participants were asked if using data to promote competition between wards would encourage falls prevention action. While senior staff believed “a bit of competition between wards” was a good idea, nurses were less positive as they felt ward experiences would vary due to different patient characteristics.
Oh, I don’t think it would make any difference. We’ve all got different patients. (Nurse, H1)
Leadership
Leadership, including the establishment of champions for falls prevention was identified as a key enabler for practice change. Leaders were identified by staff as playing a critical role in providing guidance and support to those less experienced, and to develop and promote standardised practices in terms of implementing falls prevention interventions. Nurses were either neutral (35%) or agreed (42%) that there was strong leadership support for falls on their ward and that their supervisors have assisted them when issues of falls have been raised (64%). Senior staff reported that the NUM has a critical role in falls prevention.
The NUMs are important players in [falls prevention]…to educate staff and support them about the right techniques. (SS3, H3)
NUMs were also seen as vital in ensuring the sustainability of the program.
[NUMs] are going to be the drivers, not just from the beginning but in six months’ time when it’s implemented. (SS1, H3)
Champions were identified as a practice change strategy for other projects including infection control, pain management and wound care. They were able to provide a link between committees, senior management and the ward staff and provide education and support while on the wards.
The falls champion on that ward will play a very active role in delivering the education and doing the assessments …because that links back to the Falls Committee. (SS3, H4)
Senior staff emphasised that the key to a successful champion is finding staff who have “the passion for falls and wants to make a difference to patient care” and willing to push the agenda of falls prevention on the wards. One staff member described champions as ‘ resource people ’.
Motivation
Lack of ownership
A perceived barrier to the implementation of the 6-PACK program was a lack of ownership for falls prevention in some hospitals.
Who drives falls? Nobody owns falls. (SS3, H2)
The majority of nurses (80%) believed that they were responsible for falls prevention. Senior staff agreed that nurses were primarily responsible but recognised the value of multidisciplinary input into falls prevention.
It’s everyone’s responsibility to work together to reduce falls. But I suppose primarily it comes back to nurses as they’re there with the patient 24/7. (SS1, H1)
Complacency
Reflecting on previous and current falls prevention practice, staff recognised that one barrier to practice change was complacency. Complacency was often discussed in relation to the completion of fall-risk tools. Prior experience of staff suggested that complacency in completing these tools could be an issue with nurses stating “ we all just go tick , tick , tick , tick” .
Staff just tick the same boxes that were done yesterday without really assessing…There’s that difficulty of just that complacency of ticking the same boxes…that doesn’t give you the best outcome. (SS1, H3)
To address issues of complacency, audits , reminders and feedback were suggested by staff.
Better to be reminded to do this, and reminded all the time. (Nurse, H6)
The other thing that we have a gap in is that we don’t do regular auditing…It’s about the audits and the feedback that’s given. (SS1, H1)
Falls prevention goals and commitment
An enabler to falls prevention was a commitment to falls prevention by senior staff demonstrated through provision of resources (equipment and staff) as well as clearly articulated goals. Participants believed this provided motivation and was also a source of pride and achievement when progress was being made.
So it’s pride in falls, reduction in falls. Commitment by staff. And it’s commitment by management…if they’re going to have the need for low-low beds or whatever you need, [they will get it]. Implementation care is paramount. (SS2, H3)
Engaging staff in falls prevention
As highlighted by one senior staff participant, staff engagement is important and can be facilitated through ‘ engaging hearts and minds’— both the emotional and logical aspects of falls prevention. Nurses described feeling ‘ guilty’ , ‘ stressed’ and ‘ distressed’ when a patient under their care experienced a fall. They also described the ‘ worry’ experienced if a patient suffered a fall-related injury. The emotional impact of a patient fall was seen as something that could be a motivating factor. A senior staff member at one hospital highlighted that nurses responded to interventions that emphasised the benefit to the patient . This also had implications for sustaining the project long term.
If you always promote it as best for the patient and patient focused you’ll get staff on-board, and continuing to help drive the program. You’ve got to be able to sell it to them…first of all say this is going to be so much better for your patient outcomes. (SS1, H1)
Discussion
This study identified a number of implementation targets, particularly in the areas of motivation and opportunity. These included education and training to address skills, knowledge and beliefs of nurses and developing systems to encourage falls prevention practice such as audits, reminders and feedback, provision of equipment and facilitating a culture of falls prevention through leadership and champions. Previous studies have also reported the above enablers [ 4 – 6 ]. Unlike prior research, this study details differences between nurses and senior staff beliefs regarding falls prevention. Learnings from this study were used to develop an implementation plan for the RCT [ 8 ].
The belief in the inevitability of falls is consistent with findings from other studies [ 6 , 19 ]. Although survey results suggest nurses thought falls could be prevented, nurses in focus groups identified patient groups where they believed falls could not be prevented. There was disagreement between nurse and senior staff perspectives as to whether in-hospital falls could be prevented. Incongruity between nurses’ and senior staff perceptions of the inevitability of falls has implications for the success of a falls prevention program. If nurses do not believe falls can be prevented, it may be difficult to implement interventions that aim to prevent falls. Senior staff recommended that education and training was required to increase nurse confidence and knowledge in how to prevent falls and to utilise the resources provided effectively.
Education was identified as a strategy to improve capabilities. However, implementation did raise some practical challenges. While both senior staff and nurses valued face to face case studies, senior staff favoured e-learning due to convenience and efficiency. Carefully designed e-learning packages can be effective in disseminating best practice education and have the potential to reach less accessible night and casual staff [ 20 ]. However, if a model of e-learning was adopted it would be important to ensure nurses have adequate access to computers and that these packages address aspects of falls prevention that are of greatest concern to nurses.
A motivator identified by senior staff was to harness the emotional impact of falls, for example through ‘story telling’ of falls incidents at handover. Case studies with patient stories and experiences of falls may also prove powerful in highlighting the need to address in-hospital falls in education sessions. Communicating clearly the patient benefits of the 6-PACK program was also seen to be a strategy to enhance engagement by staff.
A challenge to motivation is complacency in falls prevention practice. The acute setting is a crowded landscape of patient safety initiatives that can compete for the attention and time of nurses. Previous research has described the phenomenon of ‘missed care’ or ‘unfinished care’ where nurses can find it difficult to achieve all their tasks in caring for a patient. This can lead to adverse patient events such as falls [ 21 ]. To promote continuing engagement in strategies and to assist in care prioritisation, senior staff and nurses highlighted the importance of regular audits, reminders and feedback. Audits, reminders and feedback are generally an effective approach in guiding the implementation of an intervention [ 22 ]. Providing data to demonstrate the extent of the problem of falls on wards and to benchmark progress was another strategy identified by participants. Incident reporting has also been identified as a useful approach to change the attitudes, perceptions and practice of staff and promote engagement in patient safety initiatives [ 23 ].
The majority of falls prevention programs are focused on nurses and nursing interventions with falls often considered a nursing sensitive patient outcome [ 24 ]. However, a barrier to motivation identified is the lack of ownership for falls prevention. Senior staff stated that falls prevention should involve a multidisciplinary team approach and is everyone’s responsibility. Conversely, a recent study in Australian hospitals reported that doctors perceived time limitations as a major barrier to their involvement in falls prevention and acknowledged that medical priorities were more important for them [ 25 ]. While 6-PACK is a nurse led program, it does not discourage involvement of other clinicians. Indeed, nurses are a critical link between the patient and other care team members and often are responsible for communicating on progress and changes in patient status. This importance of the nurse role in multi-disciplinary management of falls should be communicated to staff in training.
The opportunity domain examined factors outside of the individual which enable or prompt falls prevention action. The key themes related to opportunity included lack of availability of resources, provision of falls data and leadership for falls prevention. The lack of availability of falls prevention equipment, such as low-low beds, has previously been described [ 19 , 24 ]. Leadership is both an opportunity and motivation strategy and was recognised as important by both nurses and senior staff. NUMs and champions were identified as key individuals in the implementation and sustainability of falls prevention interventions. The need for leadership and champions has been reported as an important implementation strategy in the literature [ 6 , 26 , 27 ].
Limitations and future research
The 6-PACK program is a nurse delivered intervention and therefore the focus of this research was to seek the perspective of nursing staff. The perspectives of other health professionals (doctors, allied health professionals such as physiotherapists, occupational therapists) involved in direct patient care were not captured in this study. Further research to explore whether the barriers and enablers identified by nurses and senior staff are also identified by other hospital staff is required. The wards that participated in this study volunteered to take part in the 6-PACK RCT which may have introduced selection bias. This may have impacted on the results with participants being more likely to recognise the importance of falls prevention practice.
Conclusions
This study identified barriers and enablers to the implementation of the 6-PACK program corresponding to the constructs of capability, opportunity and motivation. Barriers identified included beliefs that falls could not be prevented; limited knowledge on falls prevention in patients with complex care needs (e.g. cognitive impairment); lack of resources; and lack of ownership in falls prevention efforts. Enablers included education and training, particularly face to face case study based approaches; improved leadership; using data to drive practice change; and use of reminders, audits and feedback. Successful falls prevention program implementation in acute hospital wards are likely to require a multifaceted, planned approach that includes: regular practical face-to-face education and training for nurses to modify skills and established beliefs; provision of equipment; audit, reminders and feedback; leadership and champions; and the provision of falls data.
Supporting information
S1 Appendix
6-PACK programme to decrease fall injuries in acute hospitals: cluster randomised controlled trial (published article).
(PDF)
S2 Appendix
Development of an implementation plan for the 6-PACK falls prevention programme as part of a randomised controlled trial: protocol for a series of preimplementation studies (published article).
(PDF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
No-fault vaccine injury compensation programmes are established to compensate individuals who experience a rare vaccine-related injury due to the inherent risk of vaccination (e.g. intussusception in an infant following vaccination with a well manufactured and administered rotavirus vaccine, or a life-threatening anaphylactic reaction following any vaccine) [ 1 – 3 ]. These programmes do not require the injured party or their legal representative to prove negligence or fault by the vaccine provider, health care system or the manufacturer prior to compensation. They serve to waive the need for accessing compensation through litigation processes, which are often viewed as an adversarial approach requiring establishment of fault by at least one party prior to compensation [ 2 ]. The term ¨no-fault¨ implies a measure put in place by public health authorities, private insurance companies, manufacturers and other stakeholders to compensate individuals inadvertently harmed by vaccines [ 4 ]. In 1961, Germany was the first country to implement a no-fault compensation programme that covered vaccine injuries [ 2 ]. This stemmed from the 1953 supreme court ruling to compensate people injured with compulsory smallpox vaccination [ 2 ]. The drive to implement no-fault compensation programmes in most jurisdictions increased with reports of adverse events following immunisation with diphtheria-tetanus-whole cell pertussis in the 1970s [ 2 ].
However, with continued improvements in reporting and investigation of vaccine safety events, including in low- and middle-income settings, WHO Member States are identifying and documenting events that have scientific evidence of causal association to vaccination [ 5 ]. This is accompanied by increasing interest for national no-fault compensation policies related to vaccine injuries [ 6 – 9 ]. As of 2010, compensation schemes for vaccine-related injuries had been identified and characterized in nineteen out of WHO’s 194 Member States [ 2 ]. At the time, these programmes were exclusively implemented in high-income countries. Previous reviews have described the characteristics of existing programmes based on the six common elements identified by Evans in 1999 including administration and funding, eligibility, process and decision making, a standard of proof, elements of compensation, and litigation rights [ 1 , 2 ]. We conducted a global survey of the status of vaccine injury no-fault compensation programmes (complemented by triangulation of information from multiples sources) with the aim to update the inventory of such programmes and evaluate and update their characteristics to forecast the next segment of adopters and guide policy formulation.
Materials and methods
Initially, a landscape analysis and scoping review of published and unpublished literature were conducted to update the inventory of countries that have implemented vaccine injury no-fault compensation programmes (scoping review protocol not registered). Published data was supplemented with official documents accessed from government websites (where available). Structured literature search was done using PubMed, Excerpta Medica dataBASE (EMBASE), Cumulative index to Nursing and Allied Health Literature (CINAHL) and Global Online Access to Legal Information (GOALI) using the following predefined keywords: vaccine injury AND compensation programs; AEFI AND compensation; vaccine AND injury AND no-fault compensation; vaccine damage payment; and vaccine liability claims ( S1 File ). Using a lower cutoff period of 31 Dec 2009 to supplement on previous reviews, 41 articles published in English with relevant information were reviewed ( Fig 1 ). This descriptive analysis of the characteristics of WHO member states with no-fault compensation programmes implemented is published elsewhere [ 10 ].
10.1371/journal.pone.0233334.g001
Fig 1
PRISMA flow diagram indicating structured literature search to address a descriptive analysis of current policies and practices of no-fault compensation programmes for vaccine injuries.
In addition, all WHO Member States were approached, and screening was done using several methods to identify those with a no-fault compensation programme for vaccine injuries. We approached several professional networks including immunization programme focal points in WHO Regional or Country Offices and local Ministry of Health in Member States. Screening for programmes was also conducted amongst current and past members of the Global Advisory Committee on Vaccine Safety (GACVS) [ 11 ], and conference attendants (Global Immunization Meeting [ 12 ], Vaccine Safety Net meeting and International Conference of Drug Regulatory Authorities [ 13 ]). During the same period, the WHO Immunization, Vaccines and Biologicals Department repository and a global survey of national immunization technical advisory groups collected information on the presence of systematic compensation programmes for vaccine injuries. This data was used to triangulate the presence or absence of programmes for compensating vaccine injuries. For each country with a no-fault compensation programme for vaccine injuries identified through our approach, an expert with in-depth knowledge of the no-fault scheme was identified through colleagues in WHO country offices or National Immunization Programme Focal Points within the Member States, and invited to complete a structured online survey ( S4 File ). The questionnaire was created using a data collection tool which is based on LimeSurvey (Version 2.06+ Build 151215) to collect data on the structure, perceived benefits and operational challenges of existing programmes. The survey was designed by the Global Vaccine Safety team at the WHO Headquarters in Geneva, Switzerland. It was piloted and further refined before being administered. An independent scientific committee consisting of selected members of the GACVS, and additional immunization experts validated the survey and oversaw the conduct of the study to ensure scientific rigor. An email was sent out to participants ( S2 File ) inviting them to complete the online survey and responses were received from 03 July to 31 September 2018. To ensure data accuracy, survey respondents were encouraged to submit supporting documents. The survey tool was also made available in French, an official WHO language ( S5 File ).
A scientific review of the protocol was conducted in collaboration with the academic and scientific committee from the University of Siena, Master of Vaccinology and Pharmaceutical Clinical Development programme. The study was granted exemption from full ethical review by the WHO Ethics Committee since the study involved human subjects participating in their professional capacity (as staff or affiliates of WHO regional or country offices or Ministry of Health) and sharing information available in the public domain. Informed consent was sought from all participants prior to collecting any study-related data.
Results
All 194 WHO member states were screened for the presence of no-fault compensation programs for vaccine injuries. We received feedback from 151 countries who responded to an initial screening step to determine the presence of a no-fault compensation programme. From these responses, we identified 25 member-states implementing no-fault compensation programmes ( Fig 2 ) that met the predefined definition [ 10 ]. Through the survey and other data sources (i.e. government documents where available) we evaluated 23 existing programmes (including two from Japan) based on the six common elements reported in previous reviews [ 1 , 2 ]. Regional distribution and characteristics of implementing countries are described separately [ 10 ].
10.1371/journal.pone.0233334.g002
Fig 2
Member States screened for existence of vaccine injuries no-fault compensation programmes and number of programmes evaluated.
* 19 countries responded to survey; Japan provided information for two programmes ** Latvia, Nepal and Viet Nam.
The number of countries implementing no-fault compensation programmes for vaccine injuries has increased steadily from 19 in 2010 to 25 in 2018. As compared to previous decades there is, however, no acceleration in the number of countries. In recent years and for the first time, a low and a lower-middle-income country, Nepal and Viet Nam respectively, have instituted such programmes [ 14 , 15 ]. Table 1 .
10.1371/journal.pone.0233334.t001
Table 1 No-fault compensation programme for vaccine injuries distributed by countries and continents.
Continent
Number of countries
Countries
Africa
0
None
America
2
United States, Canada
Asia
6
China, Japan, South Korea, Viet Nam, Nepal, Thailand
Europe
16
Austria, Denmark, Finland, France, Germany, Hungary, Iceland, Italy, Luxembourg, Norway, Russia, Latvia, Slovenia, Sweden, Switzerland and United Kingdom
Oceania
1
New Zealand
Administration and funding
Administration
Fifteen (65%) of the no-fault compensation programmes for vaccine injuries are administered at the central government level. Germany, Italy, Republic of China and the Province of Quebec in Canada are the only jurisdictions implementing the compensation programme at the province level (17%). Finland and Sweden are the only countries where programmes are administered by the insurance sector [ 16 ].
Since its establishment in 1970, the programme in Switzerland was administered at the cantonal level (each of 26 states that compose the confederation). In 2016, the Swiss compensation policy was amended, and the administration of the programme is done by the central government. In Italy, the programme was decentralized in 2001 to be administered at province level in regions with the ordinary statute but remained run by the central government in regions with special statute. In 2014, the programme in the People’s Republic of China was amended requiring all 31 provinces to implement compensation mechanisms for vaccine injuries [ 4 ]. Administration of the Chinese programme involves all levels of government: filing of claims and causality assessment of events is done at district or county level; operational procedures for compensation are set at province level and general vaccine injury compensation policies including definitions of what constitutes a vaccine injury are determined at the central government level. The programme in Japan is also implemented at all levels of government.
Funding
Fifteen (65%) of the programmes are government funded including those being implemented in low- (Nepal) [ 14 ] and lower-middle-income settings (Viet Nam) [ 15 ]. The programmes in Finland and Sweden are funded by the insurance sector financed by contributions from pharmaceutical companies marketing their products in these jurisdictions. Although administered at the government level, the programme in Norway is also funded by a special insurance organization, the Drug Liability Association. In Latvia, the Treatment Risk Fund is funded through contributions from medical institutions [ 17 ], hence it also acts as professional indemnity insurance. In China, Japan and the Republic of Korea, there are two different programmes covering injuries arising from vaccines listed in the national immunization programme (NIP) and non-NIP vaccines [ 4 , 18 , 19 ]. These programmes are funded differently, with government funding NIP vaccines, and pharmaceutical companies or market authorization holders funding non-NIP vaccine injuries. The USA programme is funded by a flat-rate tax of 0.75 USD on each disease prevented in each vaccine dose (e.g., 2.25 USD for measles mumps rubella vaccines, and 0.75 USD for Haemophilus influenza type B vaccine) [ 2 , 20 ]. New Zealand has an Accident Compensation Corporation (ACC) which compensates for vaccine injuries under a general compensation for accidents and treatment injuries. The ACC is funded from the contribution of general taxation, and levies collected from employee earnings, businesses, vehicles licensing and fuel [ 21 ].
Eligibility
Vaccines
Thirteen programmes (57%) compensate for injuries arising from registered and recommended vaccines for children, pregnant women or adults (e.g. influenza vaccines) and for special indication (e.g. travel or occupation) within the jurisdiction. Five (22%) of the programmes cover injuries arising from mandatory or vaccines pro-actively recommend by law only including in France, Hungary, Italy, Slovenia and Japan (for injuries arising from NIP listed vaccines category A which are administered to achieve basic herd immunity). The programmes in the United Kingdom and Province of Quebec in Canada [ 22 ] cover for injuries arising from vaccines against specific diseases of infections as listed in their legislation.
Timelines of injury and vaccination
Timelines vary considerably from programme to programme. In the United Kingdom, claims can only be filed when the child is two years old. For adults, whichever is the latest of the following dates: either on or before their 21st birthday (or if they have died, the date they would have reached 21 years old), or within 6 years of vaccination. In the USA, the Province of Quebec, Denmark, Italy and Norway, the programmes compensate for injuries that occur within three years of vaccination or initial appearance of symptoms of the vaccine injuries. In case of death, the USA programme compensates if it occurs within two years of vaccination and not more than four years from the initial date of symptoms of the vaccine injury that led to death. In Denmark and Norway, the maximum interval between the occurrence of vaccine injury and filing a claim is 10 and 20 years respectively. In Switzerland, claims can be submitted up to when one is 21 years old for childhood vaccines or within five years of vaccination. Similarly, the programmes in Japan (for non-NIP vaccines) and the Republic of Korea have a five years window for filing claims. Finland and France have a 10 years window for filing a claim, in China, this varies by province. The programmes in Austria, German, Hungary, Japan (NIP vaccines), Luxembourg, Slovenia, Sweden, and New Zealand do not have specified timelines between the occurrence of a vaccine injury and filing a claim.
Injured party
Fifteen programmes (65%) compensate all individuals who experience an eligible injury arising from a vaccine administered within their jurisdiction. In Denmark, Slovenia and China, only citizens are eligible for compensation. Whilst in the Province of Quebec in Canada, Germany, Italy and Japan (programme for NIP listed vaccines), only province residents who experience a vaccine injury are eligible for compensation.
Types of injuries covered
All countries implementing no-fault compensation programmes have a threshold of eligibility for vaccine injury and these include: injuries resulting in financial loss or permanent or significant injury (i.e. medical disability), serious health damage or death, severe injuries exceeding normal post-vaccination reactions, severe disability secondary to vaccination against a specified disease in the legislation, serious adverse events following immunization (AEFI) or disability as per predefined criteria. In the Republic of Korea, compensation may be considered for any vaccine injury whose treatment cost beyond USD 260 (300,000 Korean Won). Although the schemes studied are primarily designed to compensate for inherent risks of vaccination (“no-fault”) Injuries arising from negligence (i.e. vaccine quality defects or immunization errors) are also covered under the schemes of twelve of the 23 programmes (52%) studied. In the remaining jurisdictions, injuries arising from negligence are handled separately either under a medical malpractice indemnity cover or through civil litigation.
Process and decision making
Process
In all the compensation programmes, the process is initiated by the injured party or their legal representative filing a claim with a special administrative unit handling vaccine injury compensation. In New Zealand, this process is initiated by the healthcare worker reviewing the injury, who then notifies the ACC. Eighteen of the programmes (78%) are purely administrative in nature with a unit consisting of health officials or an insurance organization that processes claims operating under a pre-set legislation. Five of the programmes (22%) in Austria, Finland, Hungary, and the USA have an approach that either combines both administrative and civil litigation processes or are considered a judicial review in Denmark. The national vaccine injury compensation programme in the USA involves a special court that deliberates on claims and makes the final decision on compensation for injuries pre-listed, or upon examination of an expert witness for non-listed injuries, an approach like civil litigation.
Decision making
In the purely administrative programmes, a group of medical experts reviews individual cases of vaccine injuries filed for compensation and make the decision based on available evidence. Once a decision to compensate or refuse compensation is made, the recommendations of the expert group are forwarded to the programme for action. In jurisdictions with both administrative and legal approaches, the final decision on compensation is made by legal experts (e.g. Austria, Denmark, Hungary, USA). In Finland and Sweden, compensation decisions are based on civil (tort) liability laws. Decisions making process varies amongst programmes ranging between 10 days to five years depending on the nature and complexity of the claim.
Standard of proof
All programmes reviewed require standard of proof showing a causal link between vaccination and injury. As described by Looker et al, most compensation programmes adopt the “balance of probabilities” approach which assumes that it is “more likely than not” that the vaccine caused the injury considering its nature, the consistency of time interval from vaccination, the existing medical evidence establishing an association between the injury and the vaccine including other supporting information available [ 2 , 23 ]. In sixteen of the programmes (69%), the standard of proof is based on a causal association to vaccination based on standard causality assessment. In the rest of the programmes, the standard of proof is as determined by a selected group of experts. The USA compensates injuries that are listed on the vaccine injury table occurring within pre-defined timelines [ 24 ]. Claimants with injuries not listed on the vaccine injury table are required to prove that vaccine caused the injury by presenting necessary medical records or opinions which may include testimonies from expert witnesses. In China, the standard of proof is based on epidemiological causation and the regulation excludes from compensation injuries that are deemed coincidental to vaccination [ 4 ]. In Switzerland, the causality assessment of vaccine injuries is subjected to methodological approval by a group of experts. This group of experts is equivalent to a national immunization technical advisory group (NITAG), a group of experts that provides scientific recommendations for evidence-based immunization policy and programme decisions [ 22 , 25 ]. In the Canadian Province of Quebec, the standard of proof is based on the existence or lack thereof of a probable causal link between the injury and the vaccine as determined by three independent medical experts appointed by the province, injured party and a third nominated by the initial two medical experts.
Elements of compensation
In all programmes, once a final decision has been reached, claimants are compensated with either (or a combination of): a lump-sum of money; monetary compensation calculated based on medical care costs and expenses, loss of earnings or earning capacity; or monetary compensation calculated based on non-monetary criteria e.g. pain and suffering, emotional distress, permanent impairment or loss of function. Other benefits include disability pension, survival pension, or death benefits. In the province of Quebec, the amount of compensation is determined based on rules and regulations as prescribed in the Automobile Insurance Act and is identical to the compensation offered to victims of automobile accidents. In Viet Nam, compensation for disability arising from a vaccine injury is equivalent to 30 months base salary or calculated based on lost or reduced income with a standardized formula [ 15 ]. The programme in Switzerland offers compensation equivalent up to USD 70,000 aimed at covering costs related to vaccine injuries that are not covered by other third-party benefits [ 25 ].
In twelve of the programmes (52%), the amount of compensation is calculated on a case by case basis and the final amount paid out depends on the extent of the injury. In ten of the programmes (44%), the compensation amount is standardized. Compensation amounts also vary across existing compensation programmes, and across provinces in countries implementing decentralized compensation programs e.g. China [ 5 ].
Litigation rights
In fifteen (65%) jurisdictions, claimants are required to file a vaccine injury claim with the compensation programme, but still maintain the right to pursue civil litigation against the vaccine manufacturer or health care professionals if they can prove there was a fault (i.e. vaccine quality defect). In Canadian Province of Quebec, Denmark, Hungary, New Zealand, Slovenia and Sweden, vaccine injury claims can only be filed with the compensation programme (26%).
In the USA, claimants forego their right to file for a civil claim once they have accepted compensation from the national Vaccine Injury Compensation Programme (VICP) [ 2 ]. The characteristics of the existing programmes are summarised in Table 2 below.
10.1371/journal.pone.0233334.t002
Table 2 Characteristics of existing no-fault compensation programmes for vaccine injuries.
VICP element
Programme attribute
Number of countries (N = 23 programmes * )
Admin
Central Government only
15 (65%)
Provincial Government
3 (13%)
Insurance sector
2 (9%)
Combination of the above
3 (13%)
Funding source
Government only
15 (65%)
Other sources **
8 (35%)
Eligibility: vaccines
Registered/recommended vaccines
13 (57%)
Mandatory vaccines
5 (22%)
Based on diseases listed in legislation
2 (9%)
Non-NIP vaccines ***
1 (4%)
No information
2 (9%)
Eligibility: injured party
All injured by a vaccine administered within jurisdiction
15 (65%)
Country citizens only
3 (13%)
Province residents only
4 (17%)
No information
1 (4%)
Process and decision making
Purely administrative process
18 (78%)
Combination of administrative and civil litigation processes
5 (22%)
Standard of proof
Causal association to vaccination
16 (69%)
As determined by a group of experts
5 (22%)
No information
2 (9%)
Compensation
Standardized compensation
10 (44%)
Case by case basis
12 (52%)
No information
1 (4%)
litigation rights
Vaccine injury compensation scheme alone
6 (26%)
Both vaccine compensation schemes and tort law or civil claims are allowed ◊
15 (65%)
No information
2 (9%)
*22 jurisdictions evaluated with 2 programmes from Japan resulting in 23 programmes evaluated.
** Other sources include: Pharmaceutical company contribution i.e. the USA, China for non-NIP vaccine injuries, Japan for non-NIP injuries; Insurance: Finland, Norway, and Sweden have special insurance funds where all pharmaceutical companies in their jurisdiction contribute towards. France complements Gov. funding with national health insurance, Latvia has treatment risk fund.
***China, Republic of Korea, Japan—separate system for non-NIP vaccines (detailed information available only for Japan).
◊ Limited in some jurisdiction i.e. USA
Benefits of vaccine injury compensation programmes
The benefits most referred to in existing no-fault compensation programmes were: fair compensation for individuals inadvertently injured by a vaccine meant for public good and increasing confidence in public vaccination programmes. Most respondents did not consider sustenance of vaccine supply, protecting manufacturers from liability and stabilization of vaccine prices as benefits of their programmes ( Fig 3 ). Compensation programmes were seen by respondents to enhance the legal basis of mandatory vaccination systems (in member states where such laws existed) and a sign of the government`s commitment towards immunization programmes.
10.1371/journal.pone.0233334.g003
Fig 3
Perceived benefits of no-fault compensation programmes for vaccine injuries.
Challenges of vaccine injury compensation programmes
The most notable operational challenge of the existing programmes noted by respondents was lack of public awareness of programme existence, strict requirements for standard of proof that vaccine caused injury, and long timelines for filing claims and receiving compensation ( Fig 4 ). Despite the lack of awareness ranking as a high challenge for most programmes, few participants indicated programme accessibility as an operational challenge. Most participants did not consider their programmes to be overwhelmed by the number of claims filed. One jurisdiction cited challenges with having a non-standardized calculation of compensation amount and inadequate programme funding.
10.1371/journal.pone.0233334.g004
Fig 4
Operational challenges of no-fault compensation programmes for vaccine injuries.
Discussion
As countries expand vaccine use and strengthen their safety surveillance and investigative capacity, occasional severe vaccine reactions are identified [ 5 , 26 ]. Subsequently, the question of fair and equitable compensation of identified vaccine injuries is more frequently raised. Previous reviews have shown that compensation programmes were perceived as interventions to offer equitable access to benefit for the injured party and lessen the financial burden for vaccine manufacturers [ 1 , 2 ]. For the injured party, usually, those with adequate resources would afford to access litigation procedures creating inequity to accessing compensation [ 27 , 28 ]. For vaccine manufacturers, an increase in litigation cases and substantial amounts paid out in compensation led to most players exiting the market and a subsequent significant decline in vaccine supply [ 29 ].
Proponents of this administrative approach argue it is less adversarial, more economical, and reduces the need to allot blame whilst maximizing opportunity for those with genuine vaccine injuries to access fair compensation [ 2 , 30 ]. Therefore, according to such proponents, since vaccinations are usually recommended and sometimes required and enforced by global and local authorities to control infectious diseases [ 31 ], no-fault compensation programmes are warranted and should be considered a social responsibility of each government, and global or national health authorities towards those injured by vaccines.
This evaluation has identified no-fault compensation programmes being implemented in Nepal and Viet Nam, a low and lower-middle-income country respectively. The general principles guiding the implementation of no-fault compensation programmes in these settings remains the same as identified in programmes implemented in high-income countries [ 1 , 2 ]. The implementation of these programmes based on six common elements (administration and funding, eligibility, process and decision making, standard of proof, elements of compensation, and litigation rights) indicates the feasibility of developing policy guide for countries to adopt. However, the diversity of actual programme implementation across Member States supports the need for developing compensation policies adjusted to local requirements, economic capacity and legal structures. A potential limitation of our study is that, as the focus was on programmes dedicated to vaccine injuries, other broader compensation mechanisms that could include vaccines–such as that from New Zealand–could have been missed.
No-fault compensation programmes are considered by some to increase adequacy and fairness of compensation as they provide clear legal guidance on how to access compensation for vaccine injuries [ 28 ]. However, implementation of such a compensation system should be considered simultaneously with the implementation of a well-established, comprehensive national social welfare system [ 4 ]. This has been thought to increase the efficiency of compensation programmes.
Advocacy for the implementation of no-fault compensation programmes should be approached cautiously to avoid distorting the public perception of vaccine safety and undermining confidence in immunization programmes. Sufficient country capacity for adverse event investigation and causality assessment should also be considered before considering a compensation programme. Most of the implementing countries surveyed in this article have not assessed the positive impact of no-fault compensation programmes on their vaccination programmes. However, there is no published data that suggests a negative impact of vaccine injury compensation programmes on immunization programmes. Unlike previous publications that have placed emphasis on the protection of vaccine manufacturers from liability, sustaining vaccine supply and stabilizing vaccine prices as benefits of compensation programmes [ 1 , 2 , 6 ], our findings suggest that these programmes are increasingly being considered for fair compensation of injured party and maintain confidence in immunization programmes. This perception has the potential to encourage countries to implement compensation programmes for the benefit of the population and not perceived as merely protecting the interests of vaccine manufacturers.
From the identified operational challenges for compensation programmes, policy formulation should include clear and appropriate communication strategies to ensure public awareness of compensation procedures, while not raising undue concerns and confidence issues in vaccination. Efficient systems should focus on developing policies that would allow programmes to process claims within an acceptable turnaround time, reduce bureaucratic challenges, have standardized procedures to ensure equity and fairness, and have dedicated funding mechanism to ensure programme sustainability.
Despite not having data from all main survey respondents, our approach of triangulating information from multiple sources allowed us to have a general picture of 88% of the existing programmes. This study did ask national respondents to state how they implement their programme in a structured way. This enriches our findings as it provides first-hand information from reliable sources which may be missed by literature review and grey literature searches alone. Data on policies and practices of no-fault compensation programmes from Iceland, Russia, and Thailand was not available at the time of documenting the results. This resulted in some incompleteness of the current evaluation. However, as each of the programmes is implemented uniquely in the context of country economic capacity and legal systems, the available information will be useful in guiding policy reviews and formulation for the next set of adopters. An important aspect to implementing VICP is understanding the main drivers for countries implementing compensation policies. Despite knowing motivations for implementing countries to adopt VICP policies, drivers for new implementers especially in less resourced settings remain undocumented. Our findings did not elaborate on this aspect and this remains an important area to explore as motivation for implementing VICP are likely to be different in varying socio-economic settings.
Conclusion
As countries expand their use of vaccine and strengthen their vaccine safety surveillance and investigative capacity, occasional severe vaccine reactions are identified. Subsequently, the question of fair and equitable compensation of identified vaccine injuries is more frequently raised. Findings from this study demonstrate that interest in this issue is no longer limited to high-income countries. They also demonstrate the diversity of approaches that have been selected so far, thereby justifying the development of global guidance documents. The current absence of evidence related to the impact of such programmes on vaccine confidence and clarification on the purpose (justice, ethical requirement in case of mandatory vaccination, reduced litigations among others) will have to be clarified in elaborating such guidance.
Disclaimer
The authors alone are responsible for the views expressed in this article and they do not necessarily represent the views, decisions or policies of the World Health Organization or of the other institutions with which they are affiliated.
Supporting information
S1 File
Search strategy_landscape-analysis.
(DOCX)
S2 File
Profile of survey respondents.
(DOCX)
S3 File
PRISMA-ScR checklist.
(DOCX)
S4 File
English questionnaire.
(PDF)
S5 File
French questionnaire.
(PDF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
With the increasing attention paid to the global energy and environmental issues, quite a lot of projects have been developed. Some focus on reducing size and weight of vehicle bodies, some centre around eco-driving training, and some others are centered on developing some intelligent transportation systems (ITS), which are able to adjust vehicles' speeds based on the preceding vehicles' information [ 1 ]. The adaptive cruise control (ACC) system is one of the most favorable ITS, and the existing studies based on microscopic simulations have also indicated that using the ACC system may be a possible solution to improve road traffic efficiency, fuel economy and emission performance of traffic flow [ 2 – 17 ], since the delay time of the controller in them is smaller than the human reaction time [ 18 ].
Good properties of traffic flow with the ACC system all rely on high performance control strategies depending on not only the properties of individual vehicles but also their interactions. Lee [ 19 ] introduced a memory function into the linear GHR model to store the information of relative speed during CF, which assumes that a driver reacts to the relative speed of the preceding vehicle over a period of time, rather than in an instant value. Zhang [ 20 ] developed a continuum macroscopic model arising from a CF model with driver memory, and found that driver memory in CF behaviors can lead to viscous effects in continuum traffic flow dynamics. Sipahi etal [ 21 ] analyzed the stability analysis of a constant time-headway driving strategy with driver memory effects modeled by distributed delays. Tang et al. [ 22 ] proposed an extended optimal velocity (OV) model with consideration of driver’s memory in the ACC strategy and found that utilizing driver memory to design the ACC strategy can improve the stability of traffic flow. Yu and Shi respectively explored the effects of the velocity changes with memory [ 23 ], the relative velocity changes [ 24 , 25 ] and the relative velocity fluctuation [ 26 ] on the dynamics and fuel economy of the corresponding traffic flow. Yu and Shi [ 27 ] put forward an improved car-following model considering headway changes with memory step of 1s and found that considering headway changes with memory in designing the ACC strategy can improve the stability and fuel economy of traffic flow. Yu and Shi [ 28 ] explored the effects of vehicular gap changes with memory on traffic flow in cooperative adaptive cruise control strategy. However, the above studies did not analyze the effects of the vehicle gap with different memory steps on the dynamics, fuel economy and emission performance of the simulated traffic flow in the ACC strategy as well as the nth car’s optimal velocity function. To explore the effects of vehicle gap changes on the dynamics, fuel economy and emission performance of the simulated traffic flow in the ACC strategy as well as the optimal velocity function, this study explores the close relation of vehicle gap changes and the host car's behaviors with the measured data, develop a optimal velocity function and incorporates an improved car-following model considering vehicle gap changes[ 28 ] to capture the operations of the ACC traffic flow system and then utilizes the VT-Micro model to estimate fuel consumption and exhaust emissions.
Data collection and mining analysis
Here, we select the signalized intersection of Jingshi Road/Shanshi East Road of Jinan in China. This intersection is located in the downtown area and on the major arterial, and it can completely meet the needs of real-time data acquisition.
We only record and analyze the movements of the CF vehicles on the middle three through lanes. The recording time is from 2:00 PM to 5:00 PM on December 11, 2013. The CF data in seconds are extracted by using the frame differential method, the raw data are preprocessed utilizing a linear transformations technique. The measured CF data are obtained and some are listed as shown in S1 , S2 and S3 Tables.
The gray correlation degree [ 29 , 30 ] is a quantitative value of the correlation between the behavior factors. Higher is the value of the gray correlation degree, more relevant are the main-factor and sub-factor. The interacting car-following process can be regarded as a nonlinear stochastic system. The paper uses the gray correlation analysis theory to test whether or not vehicle gap changes with memory greatly affect the following car’s behaviors as well as the nth car’s optimal velocity function. The corresponding gray correlation degrees are obtained and listed as shown in Tables 1 and 2 .
10.1371/journal.pone.0200110.t001
Table 1 Results of gray correlation analysis.
Time steps
d 21
v 1
Δv 21
Δx m21
δ = 1
0.4466
0.5227
0.9125
0.9176
δ = 2
0.5017
0.5760
0.9288
0.8675
δ = 3
0.6439
0.7137
0.9612
0.8794
where δ is the memory time step.
10.1371/journal.pone.0200110.t002
Table 2 Results of gray correlation analysis.
Time steps
δ = 1s
δ = 2s
δ = 3s
d 21
0.9595
0.9161
0.9616
Δx m21
0.8741
0.8642
0.8845
In Table 1 , it can be obviously found that Δv 21 and Δx m21 are more similar with a 1 than d 21 and v 1 , and the similarities of Δv 21 and Δx m21 are much the same, and that Δx m21 have significant effects on the host car's behaviors.
In Table 2 , it can be obviously found that d 21 is more similar with v 1 than Δx m21, and the similarities of d 21 and Δx m21 are all greater than 0.86, and that d 21 and Δx m21 are highly relevant to the following car's velocity.
The related models
The CF model considering vehicle gap changes
Yu and Shi [ 27 ] put forward an improved car-following model considering headway changes with memory step of 1s and found that considering headway changes with memory in designing the ACC strategy can improve the stability and fuel economy of traffic flow. However, the above study did not analyze the effects of the vehicle gap with different memory steps on the dynamics, fuel economy and emission performance of the simulated traffic flow in the ACC strategy as well as the nth car’s optimal velocity function. Thus, we develop the optimal velocity function and incorporate the improved CF model considering vehicle gap changes to capture the operations of the ACC traffic flow system. And the improved CF model is expressed as
x ¨ n ( t ) = κ [ V ( Δ x n ( t ) ) − v n ( t ) ] + λ Δ v n ( t ) + γ [ Δ x n ( t ) ‑ Δ x n ( t ‑ δ ) ]
(1)
where x n ( t ) is the position of car n at the time t ; V (.) is the optimal velocity function; Δ x n ( t ) and Δ v n ( t ) are the relative distance and the relative velocity between car n and car n+1 at the time t ; [Δ x n +1 ( t )-Δ x n +1 ( t - δ )] is vehicle gap changes with different time steps; κ , λ and γ are respectively sensitivity parameters.
Based on the above results of Table 2 , it can be obviously found that d 21 and Δx m21 are highly relevant to the following car's velocity. Therefore, the n th car’s optimal velocity function can be developed as follows:
V ( Δ x n ( t ) ) = V 1 + V 2 tanh ( C 1 ( Δ x n ( t ) − l ) − C 2 ) + V 3 [ Δ x n ( t ) − Δ x n ( t − δ ) ]
(2)
where V 1 , V 2 , V 3 , C 1 and C 2 are respectively sensitivity parameters, l is the car's length.
The proposed model can be reduced to the full velocity difference model when δ = 0s.
The VT-Micro model
Ahn [ 31 ] applied the data transformation technique to develop a VT-Micro model. The VT-Micro model was proposed as a statistical model consisting of linear, quadratic and cubic combinations of speed and acceleration levels using the measured data collected at the Oak Ridge National Laboratory and the Environmental Protection Agency. The calibrated model can provided a perfect fit for all measures of effectiveness tested, which can be expressed as:
ln ( M O E e ) = ∑ i = 0 3 ∑ j = 0 3 ( K i , j e × v i × ( d v d t ) j )
(3)
where MOE e is car’s instantaneous fuel consumption rate or exhaust emission rate, K i , j e is the model regression coefficient for MOE “e” at speed power “i” and acceleration power “j” for negative accelerations, v is the instantaneous speed(m/s), dv / dt is the instantaneous acceleration(m/s 2 ).
Simulation analysis of the traffic flow evolution
In this section, numerical simulations under the periodic boundary condition are carried out to analyze the traffic flow evolution process in the ACC strategy influenced by vehicle gap changes with different memory steps. The initial conditions are set as follows: 70 cars are distributed on the ring road with the length L = 1050m uniformly. The initial disturbance is supposed that the initial gap between car 69 and car 70 is 5 m, the initial gap between car 69 and car 68 is 15 m, and the others are 10 m. The memory steps are respectively set as 0s, 1s, 2s, 3s and 4s for comparative analysis, γ = 0.1, the parameters of the optimal velocity function are obtained by calibrating with the measured data, and the other parameters are adopted as same as those in the research study [ 7 ].
First, we explore the effects of vehicle gap changes with different memory steps in the ACC strategy on the velocity evolution process. Fig 1 illustrates velocity distributions obtained at the time steps of t = 100s, 300s and800s, where the different curves stand for velocity distributions of 70 cars simulated by the CF model, which consider vehicle gap changes with different memory steps.
10.1371/journal.pone.0200110.g001
Fig 1
Velocity distributions of 70 cars simulated by the CF models with different memory steps.
From Fig 1 , it can be obviously found that the velocities of all vehicles fluctuate around the initial velocity v0 = 4.6647m/s between the minimum and maximum caused by the initial small disturbance, however, the difference between them cannot be intuitively distinguished.
To distinguish the distinction caused by vehicle gap changes with different memory steps in the ACC strategy explicitly, the standard deviations of velocity distribution are obtained and listed in Table 3 .
10.1371/journal.pone.0200110.t003
Table 3 The standard deviation of velocity distribution.
Time
δ = 0s
δ = 1s
δ = 2s
δ = 3s
δ = 4s
t = 50s
0.0838
0.0335
0.0275
0.0592
0.2236
t = 100s
0.1900
0.0364
0.0213
0.0720
1.1893
t = 200s
1.1146
0.0525
0.0165
0.1243
5.0470
t = 300s
3.5854
0.0814
0.0143
0.2300
5.4468
t = 500s
4.9357
0.2114
0.0119
0.7966
5.6530
t = 800s
4.9681
0.9770
0.0101
2.0353
5.6792
t = 1000s
5.4117
2.4033
0.0094
2.3837
5.6806
t = 2000s
5.5313
3.8565
0.0075
2.6272
5.6807
t = 3000s
5.5312
3.8565
0.0066
2.2357
5.6807
From Table 3 , it can be obviously found that the velocities of all vehicles fluctuate around the initial velocity caused by the initial small disturbance, that the standard deviations of velocity distribution first gradually descend, and then ascend with the increase of the memory steps, and that the range of all vehicles' velocity fluctuation is smallest when the memory step is 2s.
Then, we explore the impacts of vehicle gap changes with different memory steps in the ACC strategy on the headway evolution process. Figs 2 , 3 and 4 illustrate hysteresis loops simulated by the 10th, the 30th and the 50th cars respectively.
10.1371/journal.pone.0200110.g002
Fig 2
Hysteresis loops from the 10th car simulated by CF model.
10.1371/journal.pone.0200110.g003
Fig 3
Hysteresis loops from the 30th car simulated by CF model.
10.1371/journal.pone.0200110.g004
Fig 4
Hysteresis loops from the 50th car simulated by CF model.
As can be seen from Figs 2 , 3 and 4 , the sizes of the hysteresis loops all first gradually descend, and then rise with the increase of the memory steps, and the size of hysteresis loops is smallest when the memory step is 2s, which is in accordance with the results of the above velocity evolution analysis.
The analysis of the above stop-and-go charts and hysteresis loops prove that vehicle gap changes with different memory steps have significant and different effects, that the stability of the traffic flow is optimal when the memory step is 2s.
Fuel economy and emissions estimation
Vehicles ' driving behaviors have been seen to offer considerable potential methods for reducing fuel consumptions and exhaust emissions [ 32 – 36 ]. Whether considering vehicle gap changes with different memory steps in design of the ACC strategy can affect fuel economy and emission performance of the simulated traffic flow need to be further investigated. The VT-Micro model is employed to explore the impacts of vehicle gap changes with different memory steps on the fuel consumptions and emission performance on the basis of the above numerical simulations under the periodic boundary condition.
First, we study the impacts of vehicle gap changes with different memory steps in the ACC strategy on the each car’s instantaneous fuel consumption and the total fuel consumptions of the whole CF system. The parameters of the VT-micro model can be obtained from the literature [ 34 ] as shown in the S4 Table .
Fig 5 depicts the fuel consumption rate of each car simulated by the CF model considering vehicle gap changes with different memory steps at the time steps of t = 50s, 100s and 300s.
10.1371/journal.pone.0200110.g005
Fig 5
The fuel consumption rate of 70 cars simulated by different CF models.
From Fig 5 , it can be found that all vehicles' instantaneous fuel consumption fluctuate around the initial value between the minimum and maximum caused by the initial disturbance, that the fluctuation range of the instantaneous fuel consumption fluctuation gradually descends firstly, and then ascends, and that the fluctuation range is smallest when the memory step is 2s, which is in accordance with the results of the above traffic flow evolution process.
Table 4 lists the total fuel consumptions of the whole CF system respectively simulated by the CF model considering vehicle gap changes with different memory steps during different period.
10.1371/journal.pone.0200110.t004
Table 4 The total fuel consumption of the whole CF system.
time
δ = 0s
δ = 1s
δ = 2s
δ = 3s
δ = 4s
t = 50s
4182. 5
4182.13
4182.12
4182.34
4183.85
t = 100s
8366.06
8363.60
8363.51
8364.16
8414.2
t = 200s
16814.0
16726.7
16726.3
16729.0
18843.5
t = 300s
26357.1
25090.4
25089.0
25098.7
30488.0
t = 500s
48235.3
41825.1
41814.3
41935.2
54178.9
t = 800s
81416.6
67118.8
66902.4
68575.6
89822.9
t = 1000s
103236
85087.4
83627.7
87371
113593
t = 2000s
213871
187626
167254
182893
232444
t = 3000s
324499
291853
250881
276045.6
351294.7
As can be seen from Table 4 , the total fuel consumptions of the whole CF system first gradually descends, and then rises with the increase of the memory steps, and that the fuel economy of the traffic flow is optimal when the memory step is 2s.
Next, we implement further study on the impacts of vehicle gap changes with different memory steps on exhaust emissions. Tables 5 , 6 and 7 respectively list the total CO, HC and NO X emissions of the whole CF system.
10.1371/journal.pone.0200110.t005
Table 5 The total CO emissions of the whole CF system.
time
δ = 0s
δ = 1s
δ = 2s
δ = 3s
δ = 4s
t = 50s
28880.2
28872.6
28872.3
28876.8
28907.3
t = 100s
57784.5
57735.2
57733.4
57746.5
58838.5
t = 200s
117330
115464
115454
115510
217360
t = 300s
217812
173205
173175
173373
505371
t = 500s
672301
288835
288615
291120
1143110
t = 800s
1503243
466396
461775
500251
2129430
t = 1000s
2078864
616127
577214
672007
2788909
t = 2000s
5466525
2176317
1154410
1707450
6086978
t = 3000s
8858945
3940903
1731607
2701972
9385067
10.1371/journal.pone.0200110.t006
Table 6 The total HC emissions of the whole CF system.
time
δ = 0s
δ = 1s
δ = 2s
δ = 3s
δ = 4s
t = 50s
2948.31
2947. 80
2947.78
2948.08
2950.15
t = 100s
5898.24
5894.92
5894.80
5895.69
5970.8
t = 200s
11916.85
11789
11788.8
11793
23546
t = 300s
21503.82
17685
17682.7
17696
65236
t = 500s
75906.65
29485
29471
29640
158635
t = 800s
187013
47465
47152
49832
304796
t = 1000s
266891
61749
58940
65765
402578
t = 2000s
765711
216273
117879
162772
891599
t = 3000s
1265854
395536
176817
257630
1380627
10.1371/journal.pone.0200110.t007
Table 7 The total NO X emissions of the whole CF system.
time
δ = 0s
δ = 1s
δ = 2s
δ = 3s
δ = 4s
t = 50s
4342.59
4341.962
4341.955
4342.37
4345.17
t = 100s
8686.88
8682.77
8682.64
8683.81
8778.59
t = 200s
17512.1
17364.7
17363.9
17368.8
22814.6
t = 300s
28461
26048
26045
26062
41915
t = 500s
57302
43425
43408
43621
82849
t = 800s
102533
69808.
69451
72502
145129
t = 1000s
132264
89361
86814
93852
186721
t = 2000s
287482
214985
173625
205007
394709
t = 3000s
442729
345332
260437
311763
602698
From Tables 5 , 6 and 7 , it can be obviously found that the total CO, HC and NO X emissions of the whole CF system all gradually descend firstly, and then ascend with the increase of the memory steps during different period of time steps, which is in accordance with the results of the above fuel economy analysis.
Conclusions
The data mining analysis shows that vehicle gap changes with different memory steps have obviously different effects on the host car's behaviors. The analysis results of several numerical simulations indicate that considering vehicle gap changes with different memory steps in designing the control strategy for the ACC system can improve the stability and fuel economy, as well as emissions performance of the simulated traffic flow. However, there are some limitations in this paper as follows:
We only obtain the measured CF data with three memory steps for the time being, due to the limitation of visual angle.
The communication delay and the time delay in the controller of the ACC system are not considered explicitly for the present.
In the further work, the image fusion technologies are used to collect the measured CF data with more memory steps, the communication delay and the time delay in the controller are considered to develop an effective CF model to study the effects on the dynamic, fuel economy and exhaust emission performance of the corresponding traffic flow by analytical analysis and numerical simulations.
Supporting information
S1 Table
Partial measured CF data (δ = 1s).
(DOC)
S2 Table
Partial measured CF data (δ = 2s).
(DOC)
S3 Table
Partial measured CF data (δ = 3s).
(DOC)
S4 Table
The related coefficients in Eq (6).
(DOC)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Non-suicidal self-injury (NSSI) is broadly defined as the direct and deliberate damage to one's own body tissue in the absence of suicidal intent [ 1 ]. Common methods of NSSI include cutting, scratching, burning, biting, hitting, and skin picking [ 2 ], but do not include forms of socially sanctioned self-injury, such as tattoos, ritual scarification, or piercings [ 3 , 4 ]. Engaging in these behaviours typically begins during early adolescence [ 5 ] with prevalence rates up to 17% in community samples [ 6 ] and between 40–80% in clinical samples [ 7 ]. NSSI is also a robust risk factor for predicting suicidal thoughts and behaviors [ 8 , 9 ]. Diagnostically, NSSI currently appears as a symptom of Borderline Personality Disorder (BPD) in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition [ 3 ]; however, the prevalence of engagement in NSSI behaviour is far higher in adolescents than are rates of BPD [ 6 , 7 ], suggesting that NSSI can both co-occur with BPD and exist independently of it. Consequently, it has been proposed that NSSI may be a diagnostic entity of its own, and it has been included in the third section of the DSM-5 as a condition requiring further study [ 3 ]. This highlights the need for greater research into this behaviour.
Although several motivations have been identified for engaging in NSSI (i.e. NSSI functions), many models of this behaviour postulate that it primarily serves an intrapersonal function, such as regulating emotions by reducing aversive states, like sadness or anxiety [ 10 – 12 ]. However, secondary motivations for this behaviour have also been identified and involve interpersonal functions, like interpersonal influence or peer-bonding, that can directly or indirectly socially reinforce NSSI [ 11 , 12 ]. For example, interpersonal stress, perceived criticism, and social rejection have been found to be common triggers of NSSI. Hence, a sensitivity to interpersonal stress, high emotional distress, and in some cases chronic romantic stress [ 13 ] are thought to play a role in the development and maintenance of NSSI, with self-injury being used as a means to both regulate general affective states as well as regulate emotional responses to social experiences [ 14 ].
A recent hypothesis postulates that violations of social rules may contribute to the interaction difficulties and social rejection experiences reported by many individuals who engage in NSSI [ 15 ]. For instance, interpersonal interactions are highly governed by rules that guide emotional behaviour [ 16 ]. These rules are perceived as normative by interaction partners, and even minor violations of the rules can create problems for people during interactions [ 17 ]. In particular, it has been suggested that individuals who engage in NSSI may be violating the expected social rules of interaction through nonconformity in facial mimicry [ 15 ]. This nonconformity may contribute to the increased interpersonal challenges and social rejection experiences reported by this population, which are known to trigger and maintain self-harming behaviour [ 18 , 19 ]
In conversation, automatically and subtly mimicking the facial expression of an interaction partner (also known as micro-expressions or automatic facial expressions), is a naturally occurring and important social phenomenon that contributes to liking and rapport building between individuals [ 20 , 21 ]. It is defined as the imitation of facial expressions of another person through the activation of corresponding facial muscles in the observer [ 22 ], often arising in response to an emotion. The movement of facial muscles mirroring the expression of another can be either visible (overt) or non-visible (covert), and typically occurs at an unconscious level [ 17 ]. In instances where emotion mimicry is covert, and movement of the skin is not easily perceptible, muscle activation can be measured using Electromyography (EMG). In 1982, Dimberg used EMG equipment to systematically demonstrate that individuals responded with elevated activity in the zygomaticus major muscle and corrugator supercilii muscle in response to happy and angry facial expressions respectively [ 23 ]. Dimberg’s study [ 23 ] was one of the first to suggest that individuals tend to automatically and covertly mimic the observed facial expressions of others. Although the activity of many different facial muscles can be measured, that of the zygomaticus major is primarily thought to correspond with positive emotional expressions, whereas the response of the corrugator supercilii is primarily thought to correspond with negative emotions [ 24 – 26 ]. The zygomaticus major is responsible for stretching the lips to create a smile, whereas the corrugator supercilii lowers the brows in response to emotions such as fear, anger, and sadness [ 27 ].
Following this early research, the mimicry-eliciting effects of other facial emotions, such as happiness, anger, sadness, fear, disgust, and surprise have been examined [ 28 ]. Results continued to confirm that happy and angry expressions produced greater EMG activity in the zygomaticus major and corrugator supercilii respectively compared to neutral expressions. The corrugator supercilii has also been found to consistently show relaxation in response to happy faces[ 28 – 31 ]. While evidence primarily exists for the emotion mimicry effect in corrugator supercilii and zygomaticus major muscles in response to anger and happiness [ 22 , 32 ], other research has found corrugator supercilii responses to fearful [ 33 – 35 ], sad [ 28 , 36 , 37 ], disgusted [ 28 , 38 , 39 ] and surprised faces [ 40 , 41 ]. However, these results are somewhat inconsistent across the literature and further investigation is continuing into the mimicry-eliciting effects of these emotional expressions. Due to the exploratory nature of this research, and to add to the above literature, our study opted to present all six basic emotions to investigate if group differences could be observed in the activity elicited in these two facial muscles. Furthermore, research has found that the intensity of facial mimicry can be modulated by stimulus type. For instance, facial mimicry is more pronounced in response to dynamic facial expressions compared to static ones [ 42 , 43 ].
In the past, mimicry reactions were thought to be an automatic response based on a link between perception and behaviour. Perception of a specific facial expression was thought to automatically evoke the same expression in the perceiver [ 44 ]. However, current researchers believe that emotion mimicry is more complex than this. For example, Hess and Fischer consider emotion mimicry to be a case of embodied simulation, in which a person understands the affect of another by simulating the affect in themselves [ 22 ]. This process would allow the perceiver to simulate the emotion on a motor, somatosensory, and affective level, to deduce its meaning and reward value [ 45 , 46 ]. This more complex view of facial mimicry is supported by data showing that it can vary across contexts, as one’s own emotional state can have an influence on emotion mimicry. For example, being in a good relationship [ 47 ] or belonging to a social group [ 36 , 48 ] has been found to increase facial mimicry, whereas having a negative attitude toward someone can inhibit it [ 48 – 50 ]. Not surprisingly, mimicking behaviours appear to be a key factor in successful interactions, as this process supports empathy and emotional understanding, as well as eliciting liking and rapport in conversational partners [ 20 , 21 ]. However, the imitation of positive emotions appears to be more likely to lead to liking and affiliation from others than the imitation of negative ones such as anger and disgust, whose mimicry can be viewed more negatively by others [ 51 ]. Nonetheless, the act of mimicking emotions, both positive and negative, may still help build relationships through affective empathy, which means to feel what others are feeling [ 52 ].
In contrast, suppressing the mimicry of emotions, also known as expressive suppression, has been found to have negative effects on interpersonal functioning. Expressive suppression has been linked to increased physiological responding in individuals interacting with suppressors [ 53 ], decreased liking and rapport, as well as decreased willingness to form a friendship with suppressors [ 53 ]. The mechanisms that underly the link between expressive suppression and impaired social functioning are not well articulated. However, one hypothesis is that those engaged in expressive suppression may also have impairments in perceiving the emotions of others. Emotion mimicry is thought to facilitate emotion recognition by having facial muscles function as a feedback system for an individual’s own experience of emotions [ 20 , 21 , 23 ]. Indeed, blocking facial mimicry has been shown to produce less accurate identification of happiness [ 45 , 53 – 55 ], as well as slower recognition of happy, sad and fearful emotion stimuli [ 56 ]. However, it is important to note that some studies have found no link between the degree of mimicry and emotion recognition ability [ 47 , 57 ]. Expressive suppression has also been identified as an emotion regulation strategy in several psychological disorders, including posttraumatic stress [ 58 , 59 ], eating disorders [ 60 , 61 ], and depressive symptomatology [ 62 , 63 ]. This research has shown that expressive suppression can act to paradoxically increase the experience of emotion within the suppressor [ 53 , 62 ].
Although substantial research exists on automatic facial expressions in individuals diagnosed with schizophrenia or depressive disorders, very little is known about individuals diagnosed with non-psychotic disorders. A systematic review conducted by Davies et al., revealed altered emotion expression across all disorders examined, with the exception of anxiety disorders [ 64 ]. In depression, decreased facial expression was mainly evident for positive affect, and in eating disorders decreased facial expression was observed in response to both positive and negative stimuli. Overall, the data included in the review pointed toward decreased facial emotion expressivity in individuals with different non-psychotic disorders [ 64 ]. Interestingly, a bias toward attenuated expression was also reported for people in remission and those at risk for mental illness, particularly schizophrenia [ 65 , 66 ]. In two BPD studies that explored facial emotion expression with EMG and observational coding [ 64 ]. Results from both studies reported an attenuation of positive and negative facial expressions in the BPD groups [ 67 , 68 ]. Two other studies examining a BPD population found no group differences for zygomaticus activity during positive emotions, but increased corrugator activity for negative emotions when pictures of disgust, anger and sadness were displayed [ 69 , 70 ]. However, activity of the levator labii (a muscle of the upper lip related to disgust) was noted to not differ between groups.
As demonstrated by the aforementioned literature, a crucial aspect of social interaction and emotion regulation is engaging in facial mimicry during social interactions, which encourages empathy and functions as a social catalyst between individuals. Because mirroring facial expressions of emotion is a robust effect in healthy individuals, and generally decreased facial emotion expressivity has been documented in individuals with different non-psychotic disorders, it is of interest to determine if blunted or incongruent affect in facial mimicry is observed in individuals with a history of NSSI (HNSSI). If so, altered facial expressivity may be a contributing factor to difficulties with emotion regulation and social rejection, which can trigger or perpetuate the cycle of self-harm [ 15 ].
Although no known previous EMG studies have examined emotion mimicry with a history of NSSI population, it is known that individuals with different non-psychotic disorders generally display decreased facial emotion expressivity [ 64 ]. Given reports of relationship problems frequently experienced by those who engage in NSSI [ 71 , 72 ], and our previous findings of deficits at the perceptual level for recognizing impoverished fear stimuli [ 73 ], it is of interest to explore whether this group displays similar attenuation in facial emotion mimicry that may contribute to social interaction problems. Additionally, this study seeks to investigate whether there is a correlation between facial emotion mimicry and motivations for engaging in NSSI. Emotion suppression has been identified as an emotion regulation strategy in several psychological disorders that paradoxically results in an increased emotional experience and negative relationship outcomes. Hence, it is hypothesized that those who more strongly endorse emotion regulation and social motivations for engaging in NSSI behaviours may also show less emotion mimicry responses. Investigating the mimicry behaviours of individuals who have engaged in NSSI will help further develop research into factors that contribute to the social problems observed in this population, an underdeveloped area of research in the literature. It will also help inform treatment strategies for clinicians working with the population, particularly because interpersonal difficulties are known triggers and maintaining factors for emotion dysregulation and self-injury.
Materials and methods
Participants
The study sample was composed of young adults (between 17 to 24 years of age) recruited from an undergraduate subject pool at the University of Ottawa. The University of Ottawa Ethics Review Board approved this research study. A total of 60 participants (30 HNSSI and 30 control participants) were recruited. Refer to Table 1 for demographic characteristics of the sample by group. No differences were found between HNSSI and the control groups for age, sex or ethnicity. Not surprisingly, the HNSSI group had higher rates of past comorbid depression and anxiety compared to the control group.
10.1371/journal.pone.0243860.t001
Table 1 Participant demographics.
Variable
HNSSI group ( n = 30)
Control group ( n = 30)
P
Mean Age: years
18.87 ± 1.31
18.87 ± 1.31
1
Sex: male
13% (4)
17% (5)
0.72
Ethnicity: White
57% (17)
33% (10)
0.64
Past Diagnosis:
Depression
30% (9)
3% (1)
<0.01
GAD
47% (14)
7% (2)
<0.01
PTSD
3% (1)
0% (0)
-
OCD
3% (1)
0% (0)
-
Other
3% (1)
0% (0)
-
None
43% (13)
93.3% (28)
<0.01
Eligibility criteria
Pre-screening questions were used to identify a subset of the university population who had a history of NSSI but reported no history of a BPD diagnosis. In order to be included in the HNSSI group, a participant had to have engaged in intentional self-inflicted injury to the surface of his or her body at least 5 times in their lifetime, with the expectation that the injury would lead to only minor or moderate physical harm (i.e., no suicidal intent). Pre-screening questions asked, “Have you ever intentionally self-inflicted damage to the surface of your body to cause bleeding, bruising, or pain (e.g., cutting, burning, stabbing, and/or hitting), without the intent to kill yourself? Please note that this does not include ear piercing, tattooing, circumcision, or cultural healing rituals.” Potential responses included Never; Once; 2–4 times; 5 or more times. Only individuals who responded “Never” or “5 or more times” were screened in to participate as controls or HNSSI respectively. Pre-screening exclusion criteria for both HNSSI and control groups included self-injury only once or 2–4 times, as well as a self-reported diagnosis of Borderline Personality Disorder. HNSSI individuals were also excluded if their self-injury status was unclear based on their subsequent responses to the Ottawa Self Injury Inventory (OSI) or the Inventory of Statements About Self-Injury (ISAS).
All HNSSI participants reported having engaged in intentional self-inflicted injury to the surface their body at least 5 times or more within their lifetime. Nearly half of the HNSSI participants 47% ( n = 14), reported thinking about self-injuring within the past month and 27% ( n = 8) had actually engaged in the behavior within the month. Additionally, 70% ( n = 21) of the HNSSI participants reported thinking about self-injuring within the past 6 months and 37% ( n = 11) of the sample reported actually engaging in self-injury within that period.
Measures
Socio-demographic questionnaire. This demographic questionnaire collected standard participant information such as age, gender, primary language, ethnicity, education, and current or past mental health diagnosis, health conditions, as well as medications.
The Ottawa self injury inventory
This questionnaire (OSI—Functions 1.1) assessed self-injurious behaviours and their functions [ 74 ]. It is a 33-item self-report measure designed to identify the psychosocial functions of NSSI. It addressed cognitive, affective, behavioural, and environmental aspects of self-injury and requires approximately 20 minutes to complete. Example questions included “why do you think you started” and “if you continue, why do you still self-injure?” and example responses include “to release unbearable tension” or “to punish myself”. Answers were provided on a 5-point scale (0 = never a reason, 2 = sometimes a reason, 4 = always a reason). This scale provided cumulative scores for the subscales of internal emotional regulation (0 to 32), external emotional regulation (0 to 12), social influence (0 to 36), and sensation seeking (0 to 12). The OSI has also been shown to be valid and reliable with excellent internal consistency scores of 0.67 to 0.87 in a university sample of young adults [ 75 ].
The inventory of statements about self-injury
This questionnaire (ISAS—Section II) assessed an individual’s reasons for engaging in self-injurious behaviours [ 76 ]. Section II of this questionnaire was administered as a 39-item self-report measure that assessed an individual’s reasons for engaging in self-injurious behaviours on a scale from 0 (“not relevant”) to 2 (“very relevant”). Questionnaire items begin with “when I self-harm, I am…” and example responses include “causing pain so I will stop feeling numb” or “creating a physical sign that I feel awful”. Based on previous research [ 77 ], 13 functions of NSSI were identified through this questionnaire. The scores on these 13 functions were summed (ranging from 0 to 6) to create separate factors that index interpersonal functions of NSSI (i.e., autonomy, interpersonal boundaries, interpersonal influence, peer-bonding, self-care, revenge, sensation seeking, toughness; Cronbach's alpha = 0.94) and intrapersonal functions (i.e., affect-regulation, anti-dissociation, anti-suicide, marking distress, self-punishment; Cronbach's alpha = 0.84). The interpersonal functions and intrapersonal functions factors show moderate correlation ( r = 0.40).
Procedure
Consent
Prior to the task, and after having the study described to them verbally, participants read and signed an informed consent.
Questionnaires
All participants completed the socio-demographic questionnaire. Two additional questionnaires, the Ottawa Self Injury Inventory (OSI) and the Inventory of Statements About Self-Injury (ISAS), were completed by the HNSSI group only. If participants screened into the HNSSI group did not report any self-injury behaviours on one or both of the OSI or ISAS questionnaires, there were removed from the study sample. Additionally, if HNSSI participants reported suicidal ideation or severe harm, by endorsing Q3, Q4, Q5, or Q6 on the OSI, a suicide protocol was implemented. Furthermore, NSSI participants were all given a list of resources in the event they wished to seek further psychological support. Questionnaires were administered prior to starting the experiment to identify any participants actively engaged in suicidal ideation, and to administer the appropriate suicidal assessment protocol to these vulnerable participants. Moreover, approximately 40 minutes elapsed between completion of the questionnaires, attaching EMG sensors, and the beginning of study administration. This lengthy time between questionnaire completion and study commencement mitigates any concerns regarding emotional arousal arising from completion of the questionnaires that might otherwise have affected participants’ reactions to the study.
Instructions
During the emotion mimicry task, participants’ automatic and unconscious reflections of emotional expressions were recorded via EMG measurements as they passively viewed images of emotionally expressive faces. However, to record these reactions successfully, only partial disclosure of the research intent was revealed to participants during consent and throughout testing. Participants were not explicitly informed that their emotion mimicry was being recorded with EMG electrodes. Instead, in accordance with the procedures of Achaibou, Pourtois, Schwartz, and Vuilleumier [ 78 ], and Hermans, van Wingen, Bos, Putman, and van Honk [ 79 ] participants were left blind to the exact purpose of the EMG recordings, being told only that the facial electrodes were to monitor ‘‘physiological changes” such as “skin conductance” and “ocular activities”. Participant information remained confidential and anonymous after collection and all data were labelled using a 5-digit code.
EMG set-up
The non-invasive technique of facial electromyography (fEMG) was used to evaluate and record the physiological properties of participants’ facial muscles at rest and while contracting. A Bionex™ bio-potential amplifier instrument and its electrodes (Model# 50-371102-00) were used. Facial muscle activity was measured using five Ag/AgCL miniature electrodes with a diameter of 4 mm that were externally attached to facial muscles in order to record changes in electrical potentials originating in the muscles over time and in reaction to the stimuli. Each participant’s skin was cleaned with alcohol and rubbed with an abrasive exfoliating paste to remove oils and skin residue (Lemon Prep TM ). Pairs of electrodes were placed along the zygomaticus major and the corrugator supercilii muscle regions of the left side of the face, as recommended by Fridlund and Cacioppo [ 80 ]. A ground electrode was also placed on the upper half of the forehead (See Fig 1 for electrode placement).
10.1371/journal.pone.0243860.g001
Fig 1
Electrode placement for measuring facial EMG.
Guidelines from Fridlund and Cacioppo (1986) were used to determine correct electrode placement. This image is modified from the Karolinska Directed Emotional Faces database, with “AF06NES” displayed. Reprinted from Lundqvist, D., Flykt, A., & Öhman, A. under a CC BY license, with permission from the Karolinska Institutet original copyright (1998) [ 81 ].
Impedances were kept below 10 kΩ using a conductive EMG-gel (Signa Gel). Electromyography Analysis Software (Model 60-0103-3.1) from MindWare Technologies, LTD. (Gahanna, Ohio) was used to record the electromyography data. The EMG signal was acquired at a rate of 1000 Hz and stored on a password protected laboratory computer.
Emotion mimicry task
Participants were asked to passively observe a series of dynamically morphing emotional face stimuli that changed from a neutral expression to one of happiness, sadness, anger, fear, disgust or surprise. While observing the morphs, participant's facial muscle activity was recorded through electromyography. Before stimulus presentation, the experimenter instructed the participants to restrict movement as much as possible, to watch the screen continuously, and to look directly at the morph stimuli. Participants were placed in a chin rest to help reduce excessive facial and body movement. The task began with participants recording a baseline measurement of a neutral, relaxed facial expression for 20 seconds while observing a fixation cross. A series of 6 practice trials, one for each emotion expression, were then provided to ensure participants were comfortable and understood the instructions to passively observe the morphing stimuli. Following this training period, participants were presented with the experimental stimuli as their facial muscle responses were recorded. A total of 144 novel facial morphing videos were presented, each separated by a 3 second inter-trial interval fixation cross stimuli. After presenting the videos, participants were again instructed to maintain a neutral expression for 20 seconds to ensure proper electrode contact was maintained throughout the experiment. Finally, following completion of the observational task, participants were instructed to produce their maximum contraction of each muscle (corrugator supercilii and zygomaticus major) individually, for three separate trials, resulting in 6 maximum contractions, each held for 4 seconds. If a participant produced a poor contraction, an additional trial was recorded. The entire task including instructions, practice and wrap-up, took approximately 15 minutes for participants to complete.
Emotion mimicry stimuli
During stimulus presentation, 150 still images were compressed, using temporal interpolation, into a video of 1.5 seconds duration to produce a seamless progress from 0 (neutral) to 100% (prototypic expression) in 0.5 seconds, followed by a 1 second hold of the emotion expression at full intensity ( Fig 2 ). This was done to create a realistic simulation of emotion expression dynamics and to evoke maximum facial mimicry in participants [ 42 ]. To create these morphing videos, a set of 24 identities (equal numbers of Caucasian male (12) and female (12) coloured photos) were selected from the Karolinska Directed Emotional Faces database [ 81 ] based on subjective image quality. The images were of amateur actors aged 20 to 30 years old with no beards, mustaches, earrings, or eyeglasses, and no visible makeup. Stimuli were pre-processed using the MATLAB image processing toolbox so that they all had equal overall lightness and color composition. A video of each of the 6 basic emotions was produced for each identity. Hence, all participants viewed 24 unique morphing videos per each of the 6 emotions (happy, sad, anger, fear, disgust, and surprise) resulting in the total of 144 facial morphing videos. These were viewed in random order.
10.1371/journal.pone.0243860.g002
Fig 2
Emotion mimicry stimuli.
Example stimuli intended to create a realistic simulation of dynamic emotion expressions where 150 still images were compressed into a seamless 1.5 second video progress from 0 (neutral) to 100% (prototypic expression) in 0.5 seconds then followed by a 1 second hold at full emotion intensity. These images were modified from the Karolinska Directed Emotional Faces database, with “AF01NES” and “AF01ANS” displayed. Reprinted from Lundqvist, D., Flykt, A., & Öhman, A. under a CC BY license, with permission from the Karolinska Institutet original copyright (1998) [ 81 ].
Debriefing
Upon study completion, participants were given a debriefing form describing the true purpose of the EMG electrodes. Participants were then provided with the opportunity to provide complete informed consent, thus allowing for inclusion of their data in the study if they agreed, or withdrawal of their data from the study if they did not. All participants opted to have their data included. All subjects were also asked to indicate if they were aware of the true intent to record their facial muscle responses and of any (voluntary or involuntary) facial mimicry that they perceived. The HNSSI ( M = 0.47, SD = 0.51) and control groups ( M = 0.57, SD = 0.51) did not significantly differ in their level of awareness of the true purpose of the facial muscle recordings t (58) = -0.77, p = 0.45. Similarly, the HNSSI ( M = 0.43, SD = 0.51) and control groups ( M = 0.57, SD = 0.51) did not differ in their self-perception of facial muscle mimicry t (58) = -1.03, p = 0.31.
EMG data reduction
Off-line analyses of the EMG activity were conducted with MATLAB. The raw signal was filtered with a band-pass filter between 30 Hz to 500 Hz with a 60 Hz notch filter to account for the frequency of the connected power supply and any external electrical activity. Subsequently, the data was rectified by implementing a Hilbert rectification to conserve all signal energy and convert the signal to positive polarity [ 82 ]. A baseline correction was then applied by subtracting mean signal amplitude from the 2 s prior to stimulus presentation for each epoch. An epoch rejection algorithm was then applied to the data using 3 SD from the mean as a rejection criterion and a 50-ms shift. Visual inspection of the data was used to confirm the effectiveness of the artifact rejection process. The rejection rate was < 5% of cases for both the corrugator and zygomaticus signals. The EMG signal was smoothed using a 250ms moving average, and EMG activity over the 24 trials for each of the six emotion conditions were collapsed and averaged for each participant. This allowed us to compare EMG activity across time during the 1500ms stimulus presentation. The processed EMG data was then imported into SPSS 24 for further statistical analysis.
Data analysis
Power analysis
A priori power analyses were conducted using GPower 3.1.9.2 to determine the sample size required for adequate power to detect predicted differences. Several prior studies have used EMG recordings to elicit emotion mimicry reactions in various non-psychiatric populations. A meta-analysis, by Davies et al. [ 64 ] examining facial reactions to emotionally expressive face stimuli, reported Cohen’s D effect sizes based on the activation of the zygomaticus muscle in response to emotion stimuli. These Cohen’s D values varied between 0.06 and 0.78, depending on the study and population tested [ 83 – 87 ]. Likewise, the Cohen’s D effect sizes reported in the meta-analysis for the activation of the corrugator muscle varied widely, between 0.08 to 0.66, depending on the study [ 84 , 86 , 88 ]. Due to the variability in the effect sizes reported above, a medium effect size f = 0.25 (Cohen’s D = 0.5) was selected. Additional parameters entered into the analysis were a power of 0.8, an α of 0.05, and a mixed factorial design with a 2 (NSSI and Control, between subjects) x 6 (happy, sad, anger, fear, disgust, and surprise, within subjects) structure. The power analysis yielded an estimated minimum total sample size of 20 participants (10 control and 10 NSSI).
A second a priori power analysis was calculated using GPower 3.1.9.2 to determine the minimum sample size needed to detect a possible correlation between emotion mimicry, as measured by fEMG amplitude values, and motivations for engaging in NSSI, obtained by the OSI and ISAS questionnaires. Given that little research has been previously conducted to inform the selected effect size, a medium Cohen’s effect size for Pearson’s r (Correlation ρ H1 = 0.31) was selected with a power of 0.8 and α of 0.05 for a two-tailed correlational analysis. This calculation yielded a total sample size of 79 NSSI participants. Given that a smaller than predicted sample size was recruited due to population limitations, it is possible that the study may not have achieved sufficient power to observe the hypothesized effects for the second analysis. With our samples size of 30 participants, α of 0.05 and power of 0.8, GPower 3.1.9.2 predicted the minimum correlations size that could be detected as significant would be 0.485.
Results
Emotion mimicry
To analyze between-groups differences in the degree of facial EMG elicited in participants in response to observing the various emotional expression stimuli, a series of 12 planned contrasts were conducted between the HNSSI group and the control group, one for each of the 6 emotions presented and each of the 2 muscles whose activity was measured [ 89 ]. The error term for these analyses were taken from a series of 2 (Group: NSSI vs. controls) x 6 (Time bin: 0-250ms, 251-500ms, 501-750ms, 751-1000ms, 1000ms-1250ms, 1251- 1500ms) x 6 facial expression categories (happy, sad, anger, fear, disgust, surprise) mixed ANOVAs, one for each of the 2 facial muscles (refer to Tables 2 and 3 ).
10.1371/journal.pone.0243860.t002
Table 2 Calculation of error term from mixed factorial ANOVA for corrugator supercilii EMG activity.
Source
df
MS
F
η 2 p
p
Group
1
5.304 e-5
3.400 e-5
5.877 e-7
0.995
Emotion
2.56
10.02
11.27
0.163
> 0.001
Emotion x Group
2.56
1.766
1.988
0.033
0.128
Error
148.48
0.888
Time
2.282
1.94
4.205
0.068
0.013
Time x Group
2.282
0.103
0.222
0.004
0.829
Error
132.33
0.461
Emotion x Time
7.498
1.096
5.925
0.093
> 0.001
Emotion x Time x Group
7.498
0.258
1.396
0.024
0.200
Error
434.90
0.185
10.1371/journal.pone.0243860.t003
Table 3 Calculation of error term from mixed factorial ANOVA for zygomaticus major EMG activity.
Source
df
MS
F
η 2 p
p
Group
1
0.034
0.155
0.003
0.695
Emotion
3.979
0.101
0.515
0.009
0.724
Emotion x Group
3.979
0.119
0.603
0.010
0.660
Error
230.77
0.197
Time
2.849
0.078
2.390
0.040
0.074
Time x Group
2.849
0.062
1.906
0.032
0.134
Error
165.22
0.032
Emotion x Time
4.336
0.054
0.674
0.011
0.623
Emotion x Time x Group
4.336
0.040
0.494
0.008
0.755
Error
251.51
0.081
In order to obtain the error term for the planned contrasts conducted between the HNSSI and the control groups for the corrugator supercilii muscle [ 89 ], a 2 (Group: NSSI or Control) × 6 (Time bin: 0-250ms, 251-500ms, 501-750ms, 751-1000ms, 1000ms-1250ms, 1251- 1500ms) x 6 (Facial expression category: happy, sad, anger, fear, disgust, surprise) mixed factorial ANOVA was conducted. Mauchly’s test indicated that the assumption of sphericity had been violated, and therefore the degrees of freedom were corrected using Greenhouse-Geisser estimates of sphericity (ε < 0.75).
The error term for the planned contrasts conducted between the HNSSI and the control groups for the zygomaticus major muscle [ 89 ], a 2 (Group: NSSI or Control) × 6 (Time bin: 0-250ms, 251-500ms, 501-750ms, 751-1000ms, 1000ms-1250ms, 1251- 1500ms) x 6 (Facial expression category: happy, sad, anger, fear, disgust, surprise) mixed factorial ANOVA was conducted. Mauchly’s test indicated that the assumption of sphericity had been violated, and therefore the degrees of freedom were corrected using Greenhouse-Geisser estimates of sphericity (ε < 0.75).
Thus, for each of the two facial muscles, six planned contrasts were performed, one for each emotion across time. Considering that only a small number of planned contrasts were conducted relative to the total number of possible comparisons, the alpha level was not adjusted and remained at α = 0.05 [ 89 ]. Effect sizes for contrasts were measured using correlation coefficient r [ 89 ]. The results of these contrast analyses are reported in the text.
The planned contrasts revealed attenuated EMG activity in the corrugator supercilii within the HNSSI group in response to the anger stimuli (F [1, 434.89] = 7.04, p < 0.008, r = 0.126). Likewise, the response of the corrugator supercilii was analyzed during presentation of happy facial stimuli. The typical mimicry behaviour of the corrugator supercilii is to decrease activity in response to happy faces [ 28 ]. Planned contrasts were again used to compare EMG activity of the corrugator supercilii between groups and also revealed attenuated EMG activity of the HNSSI group throughout the presentation of the happy stimuli (F [1, 434.89] = 15.84, p < 0.000, r = 0.187), as indicated by the negative polarity of the signal representing a decrease in activity, or relaxation, compared to baseline.
These results illustrate that the HNSSI group did not exhibit as strong a mimicry response in the corrugator supercilii muscle when observing angry and happy emotion stimuli, as compared to the control group in response to the same stimuli ( Fig 3 ).
10.1371/journal.pone.0243860.g003
Fig 3
Facial EMG activity of control and HNSSI participants in the corrugator supercilii muscle.
Positive values indicate increased activity compared to baseline and negative values indicate decreased activity compared to baseline, signaling muscle relaxation. The HNSSI participants showed less corrugator supercilii activity in response to viewing angry facial stimuli and less of an expected decrease in activity of the corrugator supercilii in response to viewing happy faces.
The EMG activity of corrugator supercilii was also analyzed in response to the presentation of the remaining four emotions (sadness, fear, surprise and disgust), but no statistically significant results were observed. Similarly, the response of the zygomaticus major was compared between the HNSSI and control groups in response to viewing all six emotions. Little mimicry was observed in this muscle in response to the stimuli presented and no statistically significant differences were observed between the HNSSI and control groups.
Emotion mimicry and NSSI functions
Correlational analyses were conducted to determine if zygomatic and corrugator EMG activity in response to viewing facial expressions of emotion were linked to the functions reported by HNSSI participants for engaging in the behaviour. In particular, scores for the internal emotion regulation and social influence functions of the OSI questionnaire and scores for the affect regulation and peer bonding functions of the ISAS questionnaire were analyzed for their degree of correlation with participant’s overall zygomatic and corrugator EMG activity in response to viewing happy and angry facial stimuli. Correlation results for the OSI questionnaire showed that scores of emotion regulation as a motivation for engaging in NSSI were positively related to stronger contraction of the corrugator muscle in response to viewing happy facial expressions (r = 0.375, p = 0.04). Recall that the typical behaviour of the corrugator supercilii in response to viewing happy faces is to relax. Hence, this response shows greater incongruent facial mimicry than would be typically expected with greater endorsement of emotion regulation as a motivation for self-injury. In contrast, the social influence scale of the OSI was negatively correlated with zygomatic EMG muscle activity in response to viewing happy facial expressions (r = -0.382, p = 0.04). This result suggests that greater endorsement of the social influence motivation is associated with less mimicry. Likewise, peer bonding as a motivator for engaging in NSSI, as measured by the ISAS, showed a trend toward negatively correlating with EMG activity of the corrugator muscle in response to viewing angry facial expressions (r = -0.361, p = 0.05). These findings provide preliminary evidence that engaging in NSSI as a means to socially influence or bond with others may be linked with reduced mimicry of the zygomatic and corrugator facial muscles.
Conclusion
To our knowledge, the present study is the first to investigate the facial mimicry responses of an HNSSI population in reaction to observing realistic depictions of emotional facial expressions. Several findings emerged from this study. First, attenuated EMG activity was observed in the corrugator supercilii in our HNSSI sample in response to observing angry and happy facial expressions of emotion. Second, we found evidence in support of the hypothesis that different motivations for engaging in NSSI are associated with different facial reactions to observing facial expressions of emotion. Specifically, the more strongly individuals with a history of NSSI endorsed self-injury as a means of social influence, the less mimicry they exhibited in response to happy and angry stimuli in the zygomaticus major and the corrugator supercilii, respectively. Furthermore, in HNSSI individuals, the degree to which emotion regulation was endorsed as a motivation for self-injury was found to be correlated with an incongruent mimicry response of the corrugator facial muscle in response to happy stimuli.
These findings support the hypothesis that social interaction difficulties of those who have engaged in NSSI may be related to implicit violations of expected social rules through facial mimicry nonconformity. That is, the non-verbal signs of facial emotion mimicry induced by facial stimuli are atypical in HNSSI participants compared to controls, and this divergence could play an important role in the disturbed social interactions reported by many individuals who have engaged in self-injury [ 90 – 92 ]. The first diverging mimicry response observed in the HNSSI group was the attenuated response in the corrugator supercilii muscle when viewing angry faces. The expected contraction of this muscle was significantly less compared to the control group. Additionally, in response to viewing happy faces, the corrugator muscle of the HNSSI group showed significantly less of an expected relaxation compared to the control group. Recall that a relaxation of the corrugator is a consistently observed response in this muscle when viewing happy faces [ 28 – 31 ]. The contraction and relaxation of the corrugator in response to these two emotions are typically observed in the emotion mimicry literature, yet the HNSSI group showed less of a response.
One possible explanation for the finding of attenuated facial mimicry in HNSSI individuals stems from Linehan’s (1993) theory as to the etiology of this behaviour. According to Linehan, the creator of dialectical behavior therapy for BPD, people engage in self-injurious behaviours because they do not appropriately recognize or manage negative emotions that arise from aversive mental states or external events. This inability to manage negative emotions is hypothesized to be the result of both a biological disposition and an emotionally invalidating environment, in which a child’s emotions (particularly negative emotions) were neither recognized nor validated by their parents. Consequently, children whose emotions were either ignored or even punished, likely failed to develop adaptive strategies to regulate their feelings and learned to use NSSI as a way to restore their emotional arousal to a tolerable level [ 93 ]. These children may have also learned to hide their overt facial expressions, causing the degree to which they also reflexively mimic expressions or produce micro-expressions to become suppressed later on.
Based on this theory one can posit that, because facial expressions communicate intent, individuals raised in an emotionally invalidating environment might have learned to adopt a more consistently neutral facial expression as a means to protect themselves from rejection by being less “visible” to their interaction partners, and by being perceived as less engaged in interactions. That is, since showing signs of emotion leaves a person vulnerable, particularly if the emotion displayed is not reciprocated or is dismissed by others as invalid or inappropriate, reduced facial expression mimicry may be a way for HNSSI individuals to protect themselves from signs of social rejection or scrutiny [ 94 ].
While the above explanation works well with regards to reduced mimicry of negative stimuli (e.g., angry faces), it is less clear why individuals with a history of NSSI would also mimic positive emotions less (e.g., happy faces) by relaxing the brow. Mimicking negative emotions tends to be viewed more negatively by others [ 51 ], so suppressing one’s tendency to reflect them makes sense as a measure for avoiding rejection or negative judgment. However, mimicking of positive emotions tends to be viewed positively and to increase feelings of liking and affiliation from others [ 20 , 21 ], so it is less clear why those with a history of NSSI would exhibit reduced mimicry in this case. One possibility is that these individuals have taken on a general strategy of not displaying any emotion, so as to hide all information about their emotional status from others. This strategy might be considered adaptive to hide one’s negative emotional states. This in turn may explain the trend in our results showing that peer bonding as a motivator for NSSI is associated with less EMG activity in the corrugator muscle in response to viewing angry facial expressions. However, in a situation where exhibiting a positive emotion may be socially expected, not doing so could easily be interpreted by others as reflecting an overall negative state. Again, our results found that greater endorsement of social influence as a motivator for engaging in NSSI was associated with less activity in the zygomatic muscle in response to viewing happy facial stimuli. In short, HNSSI individuals may have learned to hide their reflection of all emotions as a form of self-protection.
Considered through this lens, it becomes clear how such a self-reinforcing pattern of reduced spontaneous mimicry of facial expressions could contribute to ongoing social interaction difficulties frequently reported by individuals with a HNSSI [ 69 , 95 ]. This hypothesis might be taken to predict attenuated responses to all the emotions presented. However, our study showed only the typically observed results of mimicry effects in the corrugator supercilii muscles in response to angry and happy emotion stimuli [ 22 ]. Recall that the majority of studies have found elevated activity in the zygomaticus major and corrugator supercilii muscles in response to happy and angry facial expressions respectively [ 22 ], and many studies have found relaxation of the corrugator supercilii in response to happy faces [ 28 – 31 ]. Although some studies report corrugator activity to other emotions presented, we did not observe these differences. The fact that corrugator activity to emotions other than happiness and anger is only observed in some studies and not others may indicate that these effects are weak, which may in turn explain why we did not observe them. To investigate the above hypothesis further, other muscle groups that may better index the mimicry responses of disgust, sadness, fear and surprise should be examined. As such, future investigations into mimicry differences are encouraged to record activity in the frontalis for fear and surprise, the levator labii for disgust, and the mentalis for sadness. Recordings in these muscle groups may prove to be a better index for the mimicry response of these emotions and could further highlight possible group differences like those seen in the corrugator.
Another potential explanation for our results may be that some individuals with HNSSI have a greater tendency to exhibit mixed or ambiguous facial emotion expressions. Staebler et al. [ 70 ], found that a sample of participants with a diagnosis of BPD, as compared to healthy controls, tended to more often exhibit a blend of basic emotional expressions (e.g., flexing both the zygomaticus major and the corrugator supercilii) when faced with situations inducing feelings of inclusion or rejection. If this finding generalizes to the motivations for why people engage in NSSI, then it may be the case that those who more strongly endorsed emotion regulation as a motivation for NSSI in our sample were more likely to have an incongruent muscle response of contracting the corrugator muscle, as opposed to relaxing it, when viewing happy stimuli because their muscles reflected an experience of mixed emotions in response to signs of social inclusion. This hypothesis could be a suggestion for future research, as a greater number of research participants would be needed to fully investigate this theory. Also note that this explanation is not mutually exclusive to one based on incongruent mimicry. Indeed, it could well be that our results reflect some combination of attenuated and incongruous mimicry on the part of individuals with HNSSI. Either of these kinds of behaviours, or any combination of them, would be expected to lead to difficulties with social interactions.
Like all studies, several limitations of the current study need to be considered in interpreting its results. First, due to the cross-sectional design of this study, and the analysis using correlational methods, we cannot ascertain causation from the results. Second, the self-reported data collected from participants is inherently vulnerable to response bias. Since self-harm is a sensitive topic, participants may have been reluctant to report honestly about their self-injuring behaviors or motivations for engaging in the behaviour. Also, some participants may have felt stigma for endorsing interpersonal functions, such as eliciting attention from others, and thus may have under-reported interpersonal reasons as a motivation for self-injury. Third, our sample was collected from university students, which limits the generalizability of these results to other populations. Fourth, as NSSI is highly co-morbid with other disorders, it is pervasive in the literature that studies investigating NSSI include participants who have current or past co-morbid disorders. Although our sample size did not allow for examination of the potential confounding variable of comorbidity, it will be important for future studies to describe any possible influence of comorbid disorders on HNSSI emotion mimicry responses. Fifth, because this study is the first of its kind, our data need to be replicated in an independent and larger sample before they can be considered definitive.
Since a history of self-injury is an important clinical marker for increased risk of suicide, it is especially pertinent to better understanding its etiology [ 96 ]. Social dysfunctions are known to play an important role in initiating NSSI [ 97 ], and continued difficulties with social interactions often maintain the behaviour [ 19 ]. However, research into fully understanding the factors that contribute to these social interaction difficulties in NSSI is still in its infancy. Results from this study provide support for the recommendation to place greater emphasis on appropriate emotion expression and social skills training when working with this population as a complement to existing treatment approaches for NSSI in clinical practice.
Supporting information
S1 Fig
(JPG)
S2 Fig
(JPG)
S3 Fig
(JPG)
S1 Data
Correlational analysis publication data.
(XLSX)
S2 Data
EMG publication data.
(XLSX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Mutation is the ultimate source of genetic variation, and mutation rates can have a significant impact on evolutionary rate [ 1 – 3 ]. The intraspecific variability in mutation rate in many viruses and bacteria indicates that mutation rates have been optimized by natural selection [ 4 – 13 ]. Given that most mutations are deleterious, the burden of excess mutational load will select against strains with abnormally high mutation rates [ 14 – 17 ]. This principle led Sturtevant to ask, “Why does the mutation rate not evolve to zero?” [ 18 , 19 ].
A large body of theoretical and experimental work suggests that the selective pressure for higher mutation rates is due to either the physicochemical cost of maintaining a lower one or a selective advantage from an increased supply of beneficial mutations [ 20 – 23 ]. Many have argued for the adaptive benefit of high mutation rates in pathogenic microbes, which often exist in dynamic environments and are subject to host immune pressure [ 7 , 24 , 25 ]. However, direct selection of a variant with a higher mutation rate will only occur if it has been advantageous in the past, and in many cases, it has been difficult to separate the causes of a higher mutation rate from its consequences [ 19 , 26 ].
RNA viruses are ideal systems for studying the selective forces that act on mutation rates. While interspecies mutation rates range from 10 −4 to 10 −6 errors per nucleotide copied [ 4 ], studies of antimutators and hypermutators suggests that fidelity can only vary by several-fold within a species [ 27 ]. The severe burden of mutational load exerts a strong downward pressure on mutation rates, and hypermutator strains are attenuated in vivo [ 9 , 28 – 31 ]. Given the short generation times and remarkable fecundity of many RNA viruses, a small kinetic cost to higher fidelity should result in strong selection against antimutators [ 32 , 33 ]. However, the observed attenuation of antimutator RNA viruses in vivo has led many to argue for the adaptive benefit of high mutation rates, as genetic diversity provides a rich substrate for a virus’s evolution in the face of varying intrahost environments [ 7 , 10 , 34 – 38 ]. This concept is central to viral quasispecies theory, which generally proposes a link between genetic diversity and viral fitness [ 24 , 25 ].
Here, we define the selective forces that shape viral mutation rates by studying an antimutator variant. The 3D G64S mutant of poliovirus was selected after serial passage in ribavirin, an RNA virus mutagen. The RNA-dependent RNA polymerase (RdRp, 3D) of this variant contains a single glycine to serine substitution [ 5 – 7 ]. The basal mutation rate of 3D G64S is reported to be approximately 20% to 30% that of wild-type (WT) virus. While the 3D G64S mutant is attenuated in poliovirus receptor (PVR) transgenic mice, the relative importance of replicative speed and fidelity to this phenotype is not clear [ 7 , 36 ]. Biochemical assays of 3D G64S suggest a physicochemical cost of high fidelity, but as in other systems, its contribution to overall fitness remains unquantified [ 6 , 19 , 39 ].
Results
We measured the relative fitness of 3D G64S by direct competition over serial passage by quantitative reverse transcription polymerase chain reaction (qRT-PCR) ( Fig 1A ). Here, the fitness of 3D G64S is 0.78 ± 0.01 ( n = 3 replicates) relative to WT. This is a moderate fitness defect, falling in the 64th percentile in a dataset of 8,970 fitness values obtained for point mutants of poliovirus under similar conditions [ 16 ] (e.g., human epithelial cells [HeLa] multiplicity of infection [MOI] 0.1, 8-hour infection cycle, and 6 passages; Fig 1B ). We also measured the relative growth properties of WT and 3D G64S using a plaque-size assay, which measures the growth, burst size, and spread of individual viruses in the absence of competition [ 40 – 42 ]. The distribution of clonal plaque sizes was significantly different ( p < 0.005; unpaired t test with Welch correction; n = 272 WT and n = 220 3D G64S plaques) and consistent with a moderate fitness defect in 3D G64S ( Fig 1C ). In contrast with prior work, we were able to detect a significant replication defect for 3D G64S by one-step growth curve, but only with rigorous synchronization, more frequent time points, and larger numbers of replicates ( Fig 1D ). This replication defect was not specific to HeLa because we observed a similar lag for 3D G64S in a 3T3 cell line that we derived from mouse embryonic fibroblasts (MEFs) from PVR mice ( Fig 1E ). These data demonstrate that the fitness defect of 3D G64S is largely attributable to its slower replicative kinetics and is consistent with biochemical assays on purified RdRp [ 6 , 39 ].
10.1371/journal.pbio.2006459.g001
Fig 1
A speed–fidelity trade-off in the poliovirus RdRp.
(A) Relative fitness of 3D G64S as measured by direct competition. The amount of each virus at each passage was compared to the input and expressed as the difference in the log 10 ratio in RNA genomes for 3D G64S (open circles) relative to WT over time. The slope of the dotted lines are the relative fitness values, 0.78 ± 0.01, n = 3 replicates. (B) Cumulative distribution function of fitness values for SNVs of poliovirus as determined in [ 16 ]. “*” indicates the relative fitness (0.78) and percentile (64th) of 3D G64S . (C) Plaque size of clones from WT ( n = 272; black) and 3D G64S ( n = 220; grey) virus populations. Box plots show median, 25% and 75% quartiles, and 1.5× interquartile range. *** p ≤ 0.005; t test with Welch’s correction. (D) Single-cycle growth curve for WT (filled circles, black line) and 3D G64S (open circles, grey line) in HeLa. Data are mean ± standard deviation ( n = 5 replicates). *** p < 0.005; unpaired t test comparing WT and 3D G64S separately for each time point. (E) Single-cycle growth curve for WT (filled circles, black line) and 3D G64S (open circles, grey line) in 3T3 cell line derived from MEFs of PVR transgenic mice. Data are mean ± standard deviation ( n = 5 replicates). ** p < 0.01; *** p < 0.005; unpaired t test comparing WT and 3D G64S separately for each time point. (F) Relative fitness of 3D G64S (open circles) as measured by competition assay (see panel A) in the presence of varying concentrations of ribavirin. Note that the baseline relative fitness of 3D G64S (y-intercept) is lower than the fitness reported in panel A because the assays were performed under different experimental conditions (see Methods ). (G) Mutation rate in mutations per nucleotide per strand copied for WT (filled circles) and 3D G64S (open circles) in the presence of varying concentrations of ribavirin, as determined by Luria Delbruck fluctuation test. All plotted data can be found in S1 Data . HeLa, human epithelial cells; MEF, mouse embryonic fibroblast; PVR, poliovirus receptor; RdRp, RNA-dependent RNA polymerase; SNV, single-nucleotide variant; WT, wild-type.
The reduced mutation rate and replicative fitness of 3D G64S suggest a trade-off between speed and fidelity in RNA virus replication. Here, the fitness gain from increased replicative speed is offset by a reduction in fitness due to increased mutational load. We derived a quantitative model of this trade-off (see S1 Text Model 1) by measuring the replicative fitness ( Fig 1F ) and mutation rate ( Fig 1G , S1 Table ) of WT and 3D G64S under exposure to an exogenous mutagen, ribavirin [ 43 ]. WT and 3D G64S had equal fitness at approximately 150 μM ribavirin. Based on these data, our model indicates that WT incurs a fitness cost of 0.137 from mutational load alone. Therefore, any fitness benefit of the high baseline mutation rates in WT would presumably need to offset this cost. In 3D G64S , the cost of mutational load is reduced to 0.037.
If viral RdRp are constrained by a speed–fidelity trade-off, selection for increased replicative speed (r-selection) will increase mutation rate. We subjected the 3D G64S-1nt point mutant (A6176G) to r-selection over serial passage by infecting cells at low multiplicity and harvesting progeny at 4.5 hours (midexponential phase of replication). The 3D G64S point mutant reverted to WT within 15 passages in 5 independent lineages ( Fig 2A ). We only observed partial reversion at passage 15 in a subset of 24-hour control lineages, in which virus populations underwent twice as many cellular infection cycles per passage and experienced reduced r-selection. We next asked whether r-selection would lead to genetic compensation of the fidelity phenotype in 3D G64S-3nt , which has all 3 positions in the codon mutated to minimize reversion. After 50 passages of r-selection, we identified fixed and polymorphic single-nucleotide variants (SNVs) by next-generation sequencing of all r-selected and control (24-hour passage) populations of 3D G64S and WT ( Fig 2B ).
10.1371/journal.pbio.2006459.g002
Fig 2
R-selection leads to increased mutation rates.
(A) A point mutant of 3D G64S (GGT gly to AGT ser ) was introduced into a poliovirus genome that is marked with a nearby point mutation that ablates an AccI restriction site. Viruses were serially passaged every 4.5 hours (r-selected) or every 24 hours (control) for 15 passages. Chromatograms show the codon for position 64 (either GGT gly or AGT ser ). Gel image of AccI restriction digest of all passage 15 populations showing that the reversion occurred in the parental backbone and was not due to contamination with WT virus, which retains the AccI site. (B) WT and a “locked in” version of 3D G64S (GGT gly to UCA ser ) were subjected to r-selection (3.5–4 hours and 4–4.5 hours, respectively) or control (24-hour) passages for 50 passages as described in the text. Heatmap shows all mutations identified at >0.025 frequency in ≥2 out of the 20 total lineages, colored by log frequency. Diagram at left shows regions of the poliovirus genome. (C) Fitness of indicated variants relative to WT as determined by competition assay. Each symbol is a replicate competition assay, and exact p -values for the key comparisons are provided in the main text. (D) Mutation rate of indicated variants in mutations per nucleotide per strand copied as determined by Luria Delbruck fluctuation test. Each symbol is a replicate fluctuation test, and exact p -values for the key comparisons are provided in the main text. (E) In vitro kinetics of purified RdRp. Purified RdRp (2 μM), primer template (1 μM), and ATP were incubated, and samples were quenched at the indicated time points (schematic). The kinetics of complex assembly and single-nucleotide incorporation are expressed as μM extended template (y-axis) over time (x-axis). Representative data are shown. Complete data from replicates can be found in S1 Data 2E. All plotted data can be found in S1 Data . RdRp, RNA-dependent RNA polymerase; WT, wild-type.
Unbiased hierarchical clustering of SNVs by type and frequency indicates that the viruses explored distinct mutational pathways in adapting to either r-selective or control passaging regimes. Within the r-selected group, WT and 3D G64S lineages clustered together, and we noted a larger number of SNVs within the coding region for the RdRp across the 5 3D G64S populations. We found that a number of distinct SNVs increased viral fitness when introduced into the ancestral WT backbone. For example, the WT-VP4 S22G had a fitness of 2.62, and its presence in all r-selected and control lineages suggests that it mediates adaptation to HeLa cells (see SI, S1 Data 2C). In contrast, a mutation in the viral helicase found only in r-selected populations—2C V127L (fitness 1.41–1.67 in WT and 1.11 ± 0.02 in 3D G64S , SI S1 Data 2C)—would be more likely to have a general effect on replicative speed.
To identify compensatory mutations, we focused our subsequent analysis on nonsynonymous mutations in the RdRp that were found predominantly in r-selected populations, shared among multiple lineages, and more frequent in 3D G64S than in WT. Two mutations—U6261C/3D I92T and A6813G/3D K276R —met these criteria, and their frequencies at passages 30 and 50 suggest that the I92T mutation may have arisen first. The 3D I92T mutation, which was found in both r-selected WT (3 out of 5) and 3D G64S (5 out of 5) lineages, did not change either fitness or mutation rate appreciably in the 3D G64S background ( Fig 2C and 2D ). The r-selected K276R substitution, which was found in 3D G64S lineages (4 out of 5) and not in WT populations, decreased overall fitness in both WT (0.92 ± 0.03; p = 0.0031 versus WT; t test) and 3D G64S (0.55 ± 0.03; p = 0.005 versus 3D G64S ; t test). It had no detectable effect on mutation rate in either background. The G64S/I92T/K276R triple mutant had a significant increase in fitness (0.7637; p = 0.0012 versus 3D G64S ; t test) and mutation rate (8.71 × 10 −6 s/n/r; p = 0.0120 versus 3D G64S ; t test) compared to 3D G64S and each double mutant. Therefore, direct selection for replicative speed led to indirect selection of mutations that together increase the poliovirus mutation rate, with sign epistasis among G64S, I92T, and K276R in the RdRp.
To gain mechanistic insight into the interactions among these 3 mutations, we analyzed the kinetics of single-nucleotide incorporation and misincorporation by purified RdRp. The 3D G64S;I92T RdRp exhibits an assembly defect relative to 3D I92T when incubated with purified primer template and ATP ( Fig 2E and [ 6 ]). The K276R mutation partially compensates for this assembly defect in the 3D G64S;I92T background, resulting in a 1.5- to 2-fold increase in incorporation of the correct nucleotide (A opposite U). This interaction is dependent on G64S because K276R reduced RdRp activity in the 3D I92T background. While some poliovirus mutators exhibit altered kinetics of nucleotide misincorporation for G opposite U [ 30 ], the kinetics of 3D G64S;I92T;K276R were similar to those of 3D G64S;I92T ( S1 Fig ).
We further examined the relationship between RdRp speed and fidelity using a second poliovirus antimutator. The 3D K359R RdRp has slower polymerization kinetics and higher fidelity relative to WT, and the 3D K359H RdRp has similar characteristics [ 44 ]. In 2 separate experiments, we infected HeLa cells with 3D K359H virus and recovered mutants after 1 or 2 passages. We observed 2 missense mutations in the 3D gene. Amino acid changes I331F and P356H were identified together in 1 experiment, and the P356H change was identified alone in the second. We introduced each of these amino acid substitutions alone or in combination into the 3D K359H RdRp. In all cases, we observed an increase in the elongation rate ( k pol) of the viral polymerase (a surrogate for speed) as well as the nucleotide misincorporation rate ( k obs misincorporation for mutant relative to WT) in vitro ( Table 1 ).
10.1371/journal.pbio.2006459.t001
Table 1 Kinetic parameters for nucleotide incorporation and misincorporation for purified RdRp. The k pol for the correct nucleotide measures the speed of polymerization in vitro. The k pol,corr /k pol,incorr is an in vitro surrogate for fidelity, as it measures the relative rates of incorporation for the correct and incorrect nucleotides. A higher ratio indicates higher fidelity.
RdRp
Correct Nucleotide
Incorrect Nucleotide
k pol (s −1 )
K d,app (μM)
k pol (s −1 ) [× 10 −3 ] a
k pol,corr /k pol,incorr
WT
37 ± 3
180 ± 40
4.2 ± 0.6
8,800
K359H
4.0 ± 0.2
250 ± 20
0.10 ± 0.02
40,000
I331F K359H
8.5 ± 0.1
170 ± 10
0.74 ± 0.13
11,000
P356S K359H
9.8 ± 0.3
110 ± 10
0.78 ± 0.09
13,000
I331F P356S K359H
17 ± 1
120 ± 10
4.9 ± 0.8
3,500
a Determined at saturating concentrations of nucleotide substrate.
Abbreviations: RdRp, RNA-dependent RNA polymerase; WT, wild-type.
The adaptability of WT and high-fidelity viruses have generally been compared using assays that measure the acquisition of drug resistance, the reversion of an attenuating point mutation, or escape from microRNA in a limited number of replication cycles [ 5 – 7 , 34 , 36 ]. In these experiments, mutations come at little cost, and the assays essentially quantify the beneficial mutation rate. To capture better the impact of both deleterious and beneficial mutations on adaptability, we measured the fitness gain of WT and 3D G64S over 20 passages in HeLa. While our WT strain is “culture-adapted,” we found that it was far from a fitness peak; both WT and 3D G64S increased their fitness 10-fold in approximately 40 cellular infection cycles (20 passages, Fig 3A , S2 Fig ). The difference in the rate of fitness gain between WT and 3D G64S lineages was small but statistically significant (0.025 per passage, WT > 3D G64S ; mixed linear effects model, p = 0.0129).
10.1371/journal.pbio.2006459.g003
Fig 3
Adaptability of WT and 3D G64S over 20 passages in HeLa (A) or 12 passages in a 3T3 cell line derived from mice transgenic for the PVR (B).
Fitness values (≥3 replicate competition assays for each data point) were determined for populations from every fifth passage (panel A) or every fourth passage (panel B), and the adaptability in the top panels was expressed as the slope of the regression for log10 fitness over time for each of 5 independent lineages of WT (filled circles) and 3D G64S (open circles, blue) for each cell line. The bottom panels show all the data from the 5 lineages together with the regression of log10 fitness over time. Exact p -values for the difference between the slopes for WT and 3D G64S on HeLa (0.0129) and PVR-3T3 (0.0013) were derived from a mixed linear effects model (see Methods ). All plotted data can be found in SI, S1 Data . HeLa, human epithelial cells; PVR, poliovirus receptor; WT, wild-type.
We examined adaptation to a completely distinct environment by repeating the experiment on our PVR-3T3 cell line [ 45 , 46 ]. In this alternative species and cell type, we actually observed greater fitness gain in the high-fidelity 3D G64S variant relative to WT (0.121 per passage; mixed linear effects model, p = 0.0013). The larger fitness gain in 3T3 cells may reflect a larger supply of beneficial, compensatory mutations given the lower baseline fitness of 3D G64S in these cells. These data suggest that, despite its 2-fold reduction in mutation rate, 3D G64S is not mutation limited and that, in the absence of strong selection, any adaptive benefit of a higher mutation rate is countered by the fitness cost of increased mutational load (see Fig 1F and 1G and associated model above).
We next compared the phenotype of WT and 3D G64S viruses in vivo, where the ability to generate genetic diversity may allow a virus to escape host immune restriction and to replicate better in a range of environments. The available data have suggested that the attenuation of 3D G64S and other high-fidelity variants in experimental models is attributable to differences in the genetic diversity of the infecting population [ 7 ]. In this earlier work, viral diversity in the inoculum was manipulated using exogenous mutagen—2 passages in ribavirin and 5-fluorouracil followed by propagation in the absence of mutagen over 2 additional passages. The diversity of the mutant spectrum was assessed by Sanger sequencing of the 5’ noncoding and capsid-coding regions of viruses from 24 plaques.
Given the difficulty in controlling the type and level of genetic diversity in a population for correlative studies, we instead generated 5 independent stocks each of WT and 3D G64S and compared their diversity through next-generation sequencing. Using an internally benchmarked analysis pipeline that dramatically reduces false positive variant calls (see [ 47 ] and Methods ), we identified no variants at greater than 0.1% frequency. Therefore, we can exclude any significant differences in standing genetic diversity between our WT and 3D G64S populations. Even at extremely high MOIs, variants present at <0.1% are unlikely to complement each other or to cooperate reproducibly in a cellular context or in vivo [ 48 ].
The absence of mutational diversity in our replicate WT and 3D G64S stocks is important because poliovirus populations are subject to stringent bottleneck events, which further restrict intrahost diversity [ 49 – 51 ]. Work with barcoded RNA viruses suggests that the serial bottlenecks between the infecting population and the terminal population colonizing the central nervous system (CNS) are quite stringent [ 52 , 53 ], and we used published data to quantify the aggregate bottleneck size encountered by poliovirus in transgenic mice [ 46 , 49 , 50 ]. Maximum likelihood optimization of a simple probabilistic model estimated an aggregate bottleneck size of 2.67 between the inoculum and the CNS ( Fig 4A , S1 Text Model 2). Therefore, the population that causes eventual disease in these mice is derived from no more than 2 to 3 viruses in the infecting population. In the setting of tight bottlenecks, many mutations will increase in frequency due to genetic drift as opposed to positive selection.
10.1371/journal.pbio.2006459.g004
Fig 4
In vivo phenotype of WT and 3D G64S .
(A) Maximum likelihood optimization of a simple binomial model (see S1 Text Model 2) estimated an average inoculum to CNS bottleneck size of 2.67 (lambda 2.44; 95% CI 1.39–3.82) based on experimental data for 4 barcoded poliovirus populations [ 49 ]. Shown are outputs of 10,000 simulations of the model (number of mice with 1, 2, 3, or 4 barcodes represented in the CNS). Each simulation represents 27 mice, and each mouse has a bottleneck size drawn from a zero-truncated Poisson with an average lambda of 2.43 (blue) or 10 (magenta). Line is actual data from [ 49 ], the shaded regions represent the area occupied by 95% of the simulations, and the dark shaded regions represent the interquartile range of the simulations. (B) Survival curves showing mice with paralysis-free survival over time for groups infected intramuscularly with 10 5 pfu (left; n = 12 per virus), 10 6 pfu (center; n = 18 per virus), and 10 7 pfu (right; n = 18 per virus) of WT (black) or 3D G64S (dashed blue). * p < 0.05; *** p < 0.001 by log rank test. (C) Viral titer in brain and spinal cord 5 days post intravenous inoculation with 10 7 pfu of WT (filled circles) or 3D G64S (open circles). * p < 0.05; ** p < 0.005 by Mann Whitney U test; n = 7 mice in each group (out of 8 that were infected, 1 mouse in each group had titers below the limit of detection, dotted line). (D) Histogram of frequencies of intrahost SNVs identified in the spinal cords of 12 mice from panel C (7 infected with WT and 5 infected with 3D G64S ). Black, synonymous or noncoding; blue, nonsynonymous. (E) Survival curves showing mice with paralysis-free survival over time for groups ( n = 43 per virus combined from 2 experiments) infected intramuscularly with 10 6 pfu of 3D G64S (dashed blue) or 3D G64S ;2C V127L (orange). ** p < 0.005 by log rank test; actual p -value 0.0012. (F) Survival curves showing mice with paralysis-free survival over time for groups ( n = 43 per virus combined from 2 experiments) infected intramuscularly with 10 6 pfu of 3D G64S (dashed blue) or 3D G64S;I92T;K276R (pink). * p < 0.05 by log rank test; actual p -value 0.0411. All plotted data can be found in S1 Data . CNS, central nervous system; SNV, single-nucleotide variant; WT, wild-type.
We infected groups of PVR mice intramuscularly with both WT and 3D G64S populations. Both viruses were able to access the CNS efficiently through this route over a range of doses ( Fig 4B , S2 Table ), but there was a clear delay in the 3D G64S group ( p = 0.0239, p < 0.001, and p < 0.001 for 10 5 , 10 6 , and 10 7 pfu inocula, respectively). This lag persisted even when we infected with doses 20-fold higher than the median lethal dose (LD 50 ) of 3D G64S ( S3 Table ). The difference in 3D G64S attenuation compared to prior work [ 7 ] does not appear to be due to the mouse model used because other studies in cPVR mice have reported results similar to those presented here [ 36 , 54 ]. Both viruses spread to the CNS and replicated to high titers after intravenous inoculation, although WT titers in the brain and spinal cord were marginally higher at 5 days post infection ( Fig 4C , p = 0.0012 for brain and p = 0.0221 for spinal cord, Mann Whitney U test). We characterized the mutations present in the CNS populations of 12 spinal cords of intravenously infected mice ( Fig 4D , 7 WT and 5 3D G64S ). Most mutations were rare, none were shared among mice, and there was an excess of synonymous or noncoding variants relative to nonsynonymous ones. These data are consistent with random sampling of the infecting population as opposed to positive selection.
We also examined the impact of the r-selected mutations on virulence. The 2C V127L mutation, which conferred a fitness of 1.11 in the 3D G64S background and does not appear to affect fidelity ( S3 Fig ), restored virulence to nearly WT levels ( p = 0.0012, log rank test; compare 3D G64S to 3D G64S ;2C V127L in Fig 4E ). In contrast, the triple mutant, 3D G64S;I92T;K276R —which replicates with a WT mutation rate and a marginally increased fitness of 0.7637—was only slightly more virulent than the high-fidelity variant 3D G64S ( p = 0.0411, log rank test; compare 3D G64S to 3D G64S;I92T;K276R in Fig 4F ). Therefore, restoration of replicative speed restored virulence in 3D G64S , but compensation of the fidelity phenotype did not.
Discussion
We used a well-studied antimutator variant of poliovirus to identify the selective forces that optimize a pathogen’s mutation rate. Using 3 different assays, we identified a significant fitness cost to higher fidelity and directly link this cost to viral replication kinetics. Our quantitative model of the speed–fidelity trade-off suggests that selection for replicative speed has pushed viral mutation rates to a level that imposes a significant fitness cost at baseline due to lethal or highly deleterious mutations. Consistent with the trade-off model, direct selection for increased replicative speed led to indirect selection of polymerases with higher mutation rates. The genetic interactions are quite strong because the 2 compensatory mutations exhibited reciprocal sign epistasis. The speed–fidelity trade-off in the poliovirus RdRp appears to be a generalizable phenomenon because compensatory mutations that increased the replicative speed of the 3D K359H antimutator also increased its mutation rate ( Table 1 ). Given the structural similarity among viral RdRp, the polymerases of other RNA viruses are likely to be subject to the same speed–fidelity trade-off, and we predict that the molecular mechanisms governing polymerase kinetics and mutation rate will be similar.
Trade-offs are essentially constraints that force one parameter to change with another. In this case, viral mutation rate changes with replicative speed [ 55 ]. Similar trade-offs are central to the kinetic proofreading hypothesis, which posits a close relationship between the error rates of biosynthetic processes and the kinetics of their component reactions [ 56 ]. Studies of DNA replication and protein translation suggest that these systems optimize speed over accuracy as long as the error rates are within a tolerable range [ 57 , 58 ]. We find a similar phenomenon in viral RdRp, in which the WT generates an extraordinary amount of mutational load, largely because of the benefit in replicative speed.
Failure to consider evolutionary trade-offs can lead to teleological errors, in which the consequences of a process (e.g., increased genetic diversity) are misinterpreted as a cause (e.g., direct selection for a higher mutation rate [ 19 , 23 , 26 ]). Similarly, we find that the proposed link between within-host genetic diversity and virulence is confounded by the fact that faster replicating viruses are both more virulent and have higher mutation rates. The high mutation rates of RNA viruses and the highly deleterious fitness effects of mutations ensure that most genetic diversity is extremely rare and unlikely to be consistently maintained in the face of intrahost and interhost bottlenecks [ 53 ]. We do not dispute that virus populations will harbor minority variants, that a subset of these mutations may be adaptive or beneficial to the virus, and that some may be virulence determinants. However, the observation of genetic diversity is not in and of itself evidence that selection has optimized mutation rates for the future benefit of novel mutations. Indeed, our data show little adaptive benefit to a marginally increased mutation rate and identify no plausible mechanism whereby the observed increase in rare genetic diversity can influence pathogenesis. We suspect that RNA viruses are subject to other trade-offs of evolutionary significance, perhaps between polymerase speed and recombination rate or between recombination rate and polymerase fidelity. Here too it will be important to define the selective forces at play, thereby separating the causes from the consequences.
Methods
Ethics statement
The University of Michigan Institutional Animal Care and Use Committee reviewed and approved the protocols for all mouse studies described in this manuscript (Protocol ID PRO00008088). The University’s Animal Welfare Assurance Number on file with the NIH Office of Laboratory Animal Welfare (OLAW) is A3114-01. Information on humane endpoints used, the length of each experiment, the numbers of animals used and subsequently euthanized ( S2 and S3 Tables), the frequency of monitoring, and animal welfare considerations are described below (“Infection of transgenic mice” section).
Cells and viruses
A low passage stock of HeLa cells (<2 weeks in culture), previously obtained directly from ATCC (CCL-2), was kindly provided by Mary O’Riordan (University of Michigan). Except where noted, these cells were used for all experiments in this study and maintained in minimal essential media (MEM; Invitrogen 11090), supplemented with 10% fetal bovine serum (Gibco or Hyclone), 1x penicillin-streptomycin (Invitrogen 15140–148), 1x sodium pyruvate (Invitrogen 11360), 1x MEM alpha nonessential amino acids (Invitrogen 11140), and 1x glutamine (Invitrogen 25030). A second stock of HeLa of unknown passage history was obtained from Michael Imperiale (University of Michigan). These cells were only used for plaque assays to titer stocks and were maintained in Dulbecco’s modified Eagle’s media (DMEM; Invitrogen 11965) supplemented with 10% fetal bovine serum and 1x penicillin-streptomycin. PVR-3T3 cells are described below and were maintained in DMEM supplemented with 10% fetal bovine serum, 1x penicillin-streptomycin, and 1x glutamine. In all cases, cell lines were maintained for no more than 30 passages at a time. WT poliovirus and all mutants were generated from plasmid pEW-M, a Mahoney clone originally obtained from Eckard Wimmer (SUNY-Stonybrook) [ 59 ].
Generation of PVR-3T3 cells
C57/BL6 PVR-Tg21 (PVR) mice [ 45 , 46 ] were obtained from S. Koike (Tokyo, Japan) via Julie Pfeiffer (UT Southwestern) and maintained in specific pathogen-free conditions. Primary MEFs were derived from PVR mice. Day 13.5 embryos were harvested and washed in phosphate-buffered saline (PBS). The heads and viscera were removed, and the body was minced with a sterile razor blade, trypsinized, and homogenized by pipetting with a 10 ml serological pipette. Cells were plated in DMEM supplemented with 10% fetal bovine serum, 1x penicillin-streptomycin, and 1x glutamine. An immortalized cell line was derived from PVR MEFs following the 3T3 protocol [ 60 ]. Briefly, freshly thawed MEFs were plated in 30 T25 flasks at a density of 3.8 × 10 5 cells per flask in complete DMEM. Every third day, cells in each flask were trypsinized, counted, and transferred to fresh flasks at a density of 3.8 × 10 5 cells per flask. As the cellular population began to increase (passages 13–15), cells were expanded into larger vessels and ultimately frozen down at passage 17.
Site-directed mutagenesis
All mutations were introduced into either pEW-M or subclones using overlap extension PCR [ 61 ]. The presence of the desired mutation and the absence of additional mutations were verified by Sanger sequencing of the amplified insert and, in some cases, the entire genome.
In vitro transcription, transfection, and viral stocks
Viral RNA was generated by in vitro transcription of the corresponding plasmid clone using T7 RNA polymerase, and virus was recovered following RNA transfection of HeLa. For transfections, 2.6 × 10 5 HeLa were plated per well in a 12-well plate the day prior. One microgram of RNA was mixed with 4 μl TransIT mRNA transfection reagent (Mirus 2225) and 100 μl OptiMEM (Invitrogen 31985), incubated according to the manufacturer’s protocol and applied to cells. Passage 0 virus was harvested at 100% CPE (within 24–48 hours). Passage 1 stocks were generated by passaging 100 μl of passage 0 virus on fresh cells and were titered by either plaque assay or TCID 50 . Passage 2 and 3 stocks were generated by passaging at an MOI of 0.01. For all stocks, cells were subjected to 3 freeze–thaw cycles and the supernatants clarified by centrifugation a 1,400 × g for 4 minutes. These supernatants were stored at −80 °C in aliquots to limit the number of subsequent freeze–thaw cycles.
Competition assay for viral fitness
Competition assays were performed essentially as described in [ 17 , 42 ]. For the experiment in Fig 1A , HeLa cells were plated in 12-well plates, at a density of 2.6 × 10 5 per well the day prior to infection. Cells were infected at a total MOI of 0.1 with an equal TCID 50 of WT and 3D G64S . Three replicate wells were infected with each pair of viruses in 250 μl for 1 hour with occasional rocking. After 1 hour, the inoculum was removed and 1 ml fresh media applied. Passage 1 virus was harvested after an additional 7 hours (8 hours since infection). The titer of the passage 1 virus was used to calculate the dilution factor necessary to maintain an MOI of 0.1 for the subsequent 5 passages. RNA was harvested from each passage using Trizol (Ambion 15596026). Random hexamers were used to prime cDNA synthesis with 1/10 of the RNA. Each cDNA was analyzed by qRT-PCR using 3 different primer and/or probe sets with duplicate PCR reactions for each sample and primer set. The first set—COM2F 5’ CATGGCAGCCCCGGAACAGG 3'‘ and COM2R 5’ TGTGATGGATCCGGGGGTAGCG 3’—was used to quantify total viral genomic RNA in an SYBR green reaction (Power SYBR Green PCR Master Mix; Thermo 4368708). Two custom TaqMan probes (Applied Biosystems) were used to quantify the number of WT and 3D G64S genomes. Duplicate wells were averaged, and relative amounts of WT and 3D G64S RNA were determined by normalizing the cycle thresholds for each of these probes to those of the COM primer set (ΔCt = Ct Virus − Ct COM ). The normalized values for virus passages 1–6 were then compared to passage 0 to obtain a ratio relative to P0 (ΔΔCt = ΔCt PX − ΔCt P0 ). This relative Ct value was converted to reflect the fold change in the ratio (Δratio = 2 −ΔΔCt ). The change in ratio of the mutant relative to the change in ratio of the WT as a function of passage is the fitness ([Δlog ratio Mut − Δlog ratio WT ]/time). Competition assays in ribavirin ( Fig 1F ) were performed in the exact same manner except that serum-free media were used in both drug and mock passages.
For all other competition assays ( Fig 3 ), we compared the experimental virus (e.g., WT P4, 3D G64S P8, etc.) to a tagged WT reference (Tag8). We plated 2.6 × 10 5 cells per well (either HeLa or PVR-3T3) in 12-well plates. Infections were performed at an MOI of 0.05 in 250 μl complete media for 1 hour. After an hour, the media were aspirated and fresh 1 ml growth media applied. All passages were for 24 hours. The dilution factor between passages required to maintain this MOI was 400 for HeLa competitions and 350 for PVR-3T3 competitions. All RNA harvests for these competitions were performed in 96-well plates using Purelink Pro 96 Viral RNA/DNA kits (Invitrogen 12280), and cDNA synthesis was performed as above. In addition to the COM primer set (see above), we used primer pairs Tag8 seq.tag 5’ TTCAGCGTCAGGTTGTTGA 3’ + Rev. WT seq.tag 5’ CAGTGTTTGGGAGAGCGTCT 3’ and WT seq.tag 5’ AGCGTGCGCTTGTTGCGA 3’ + Rev. WT seq.tag 5’ CAGTGTTTGGGAGAGCGTCT 3’ to quantify the Tag8 reference and test samples, respectively. Note also that in these competitions the regressions were fit through passages 1–4 and excluded P0 as slight deviations from a 1:1 ratio of the two viruses in the inoculum can skew the slope when fit through this data point.
Plaque-size assay
Plaque assays were performed on subconfluent monolayers (7.5 × 10 6 on day of infection) in 10 cm dishes. The amount of virus applied to each plate was determined empirically to ensure well-spaced plaques (approximately 30 per 10 cm dish). Plates were stained with crystal violet at 72 hours post infection. Each plate was scanned individually at 300 dpi using a flat-bed scanner. Sixteen-bit image files were analyzed using ImageJ. Brightness, contrast, and circularity thresholds for plaque identification were set using uninfected plates.
Single replication cycle growth curve
The day prior to infection, 4 × 10 5 HeLa cells were plated in 12-well plates with 45 wells per virus (9 time points and 5 replicates per time point). Cells were infected at an MOI of 1 in 150 μl volume, and the infections were synchronized by incubation on ice for 1 hour with occasional rocking. At 1 hour, the inocula were aspirated, each well was washed twice with ice-cold PBS, and 1 ml of fresh, prewarmed growth media were applied to all wells. One set of 5 wells was immediately frozen as the t = 0 hours time point. All other plates were returned to the incubator, and a set of 5 wells was removed and frozen at t = 1.5, 2, 2.5, 3, 3.5, 4, 5, and 7 hours. All samples were titered by TCID 50 . The growth curve on PVR-3T3 cells was performed using a similar protocol, except that 5 × 10 5 cells were plated the day prior and the time points were t = 1, 2, 3, 4, 5, 6, 7, and 8 hours.
Measurement of viral mutation rates
Mutation rates were measured by Luria-Delbruck fluctuation test, which in this case quantifies the rate at which the poliovirus 2C protein acquires the necessary point mutations to permit viral growth in 1 mM guanidine hydrochloride [ 62 – 64 ]. Each fluctuation test was performed with 29 replicate cultures in 48-well plates. A total of 65,000 HeLa cells per well were plated the day prior to infection. In all cases, the media were changed to serum-free media 3 hours prior to infection. For infections in ribavirin, this serum-free media also included drug at the specified concentrations. Each well was infected in 200 μl volume with 1,000 to 4,000 pfu per well depending on the virus and experimental condition. Five independent aliquots were also saved for subsequent titering (see N i below). For infections in ribavirin, the infection media also included drug at the specified concentrations. Infected cells were incubated for 7 hours and then frozen. The lysed cells and media were harvested following 3 complete freeze–thaw cycles and transferred to a microcentrifuge tube. The empty wells were rinsed with 300 μl of complete growth media and combined with the initial 200 μl lysate. This 500 μl lysate was clarified by centrifugation at 1,400 × g for 4 minutes. Twenty-four wells were titered by plaque assay with 1 mM guanidine hydrochloride in the overlay (see P 0 below). Five wells were titered by standard plaque assay without guanidine hydrochloride (see N f below). The mutation rate, μ 0 , was estimated from these data using the P 0 null-class model: μ 0 = −lnP 0 /(N f -N i ), where P 0 was the fraction of the cultures that yielded no guanidine-resistant plaques, N f was the average number of pfu in the absence of guanidine, and N i was the average number of pfu in the inoculum. As described in [ 65 ], μ 0 can be converted to the mutation rate in nucleotide units by correcting for the mutation target (number of mutations leading to the scored phenotype, T) and the number of possible mutations at each target site (constant, 3) using the equation μ = 3μ 0 /T. The number of distinct mutations that could yield the guanidine-resistant phenotype was determined empirically by isolating and sequencing the entire 2C open reading frame for 15 guanidine-resistant plaques derived from WT virus and 15 guanidine-resistant plaques derived from WT virus treated with 200 μM ribavirin. In each case, we found 6 mutations that mediated resistance, although there were 7 total among 30 plaques ( S1 Table ).
Mutagen sensitivity assay
HeLa cells were plated the day prior to infection at a density of 2.6 × 10 5 cells per well in a 12-well dish. On the day of infection, monolayers were pretreated with 0 to 600 μM ribavirin in serum-free media for 3 hours, then infected with virus at an MOI of 0.1 (50,000 pfu) for 60 minutes. The cells were washed once in PBS and incubated in ribavirin for an additional 24 hours. Viral supernatants were harvested by freeze–thaw as above and titered by tissue culture infectious dose.
R-selection through serial passage
For each passage, HeLa cells were plated the day prior to infection in 6-well plates at a density of 7.25 × 10 5 cells per well, yielding 1.2 × 10 6 cells on the day of infection. Infections were initiated with passage 3 stocks of either WT or 3D G64S , and each passage was performed at an MOI of 0.5 (6 × 10 5 TCID 50 units) in 1 ml of media for 1 hour with occasional rocking. After 1 hour, the inoculum was aspirated, the cells were washed twice with PBS, and 2 ml of fresh growth media were applied. For the first 15 passages, WT and 3D G64S viruses were harvested at 4 and 4.5 hours, respectively. For passages 16 through 50, WT and 3D G64S viruses were harvested at 3.5 and 4 hours, respectively. Control populations were infected in the same manner except that viruses were harvested at 24 hours post infection. There were 5 r-selected WT lineages, 5 r-selected 3D G64S lineages, 5 control WT lineages, and 5 control 3D G64S lineages. Viruses were titered at every fifth passage to maintain an MOI of approximately 0.5.
Selection and identification of second-site suppressors of RdRp variant 3D K359H
HeLa cells were transfected by electroporation with viral RNA transcript, added to HeLa cell monolayers, and incubated at 37 °C. After 2 days, the media were passaged onto a separate monolayer of HeLa cells. Upon cytopathic effect, viruses were harvested by 3 repeated freeze–thaw cycles, cell debris was removed by centrifugation, and viral supernatants were titrated. In this time frame, the titer increased approximately 40-fold (from 5.1 × 10 5 pfu/mL to 2.1 × 10 7 ). Viral RNA was isolated with QIAamp viral RNA purification kits (Qiagen) according to the manufacturer’s instructions. The 3Dpol cDNA was prepared from purified viral RNA by RT-PCR and sequenced. The I331F and P356S substitutions were identified together in 1 experiment, and the P356S substitution was identified in a second.
In vitro assays of RdRp function
All mutations were introduced into the pET26Ub-PV 3D [ 66 ] or pSUMO-PV-3D [ 67 ] bacterial expression plasmids using either overlap extension PCR or QuickChange Site-Directed Mutagenesis. The presence of the desired mutations and the absence of additional mutations were verified by DNA sequencing. PV 3Dpol RdRps were expressed and purified as described previously [ 66 , 67 ]. The sym-sub assays used to measure assembly and elongation kinetics of purified RdRp on a defined template were performed as described in [ 6 , 30 , 68 ]. All assays had 1 μM primer/template and 2 μM enzyme.
Adaptability of WT and 3D G64S
For HeLa cells, adaptability was measured using the 24-hour passage control lineages from the r-selection experiment. The fitness values of WT and 3D G64S populations were measured by competition assay, as above, using samples from passages 0, 5, 10, 15, and 20. For PVR-3T3 cells, serial passages were performed as follows. Cells were seeded in 6-well plates at a density of 7.6 × 10 5 cells per well the day prior to infection, yielding approximately 1 × 10 6 the day of infection. Serial passage lineages were initiated with passage 1 stocks of either WT or 3D G64S , and each passage was performed at an MOI of 0.5 in 1 ml for 1 hour. After 1 hour, the inoculum was aspirated, the cells were washed twice with PBS, and 2 ml of fresh growth media were applied. Viruses were harvested at 24 hours and titered every fourth passage to ensure an MOI of 0.5. There were 5 replicate lineages of WT and 3D G64S .
Infection of transgenic mice
Six- to 9-week-old mice were used for all experiments. The age range and distribution of males and females in each group for each experiment are reported in S2 Table . On the day of each infection, a general health exam was performed on all animals by university veterinary technical staff, and animals were assigned unique ear tag identifiers. Study animals were housed in BSL2 conditions. Females were housed together by group. Males from the same litter were housed together and were separated by group as needed. Enrichment was provided when any animal became single-housed due to fighting or other indication. Animals were fed 5LOD rodent chow. Because hindlimb paralysis was an expected outcome of this study, moist chow placed on floor of cage and/or diet gel was provided as a supplement. Water was provided ad libitum. Temperature and humidity were monitored and recorded daily by husbandry staff on general housing room sheets. Alternating 12-hour light and dark cycles were in place as per standard housing.
For survival analyses, mice were infected intramuscularly with 50 μl to each hindlimb for the total dose of 100 μl. Mice were observed twice daily for lethargy, hunched posture, scruffy fur, paralysis, or decreased mobility and were euthanized when they exhibited bilateral hindlimb paralysis. Over 90% of all assessments were performed by members of the university veterinary technical staff, who were blinded to the hypotheses and expected outcomes of the studies. All surviving animals were euthanized after 12 days. These endpoints were also used to calculate the PD50 using the Spearman-Karber method ( S3 Table ).
For tissue distribution analyses, mice were infected intravenously via tail vein with 100 μl and observed twice daily as above. Mice were euthanized for severe morbidity (as above) or on day 5, the conclusion of the experiment. Whole organs were isolated from all mice and homogenized in PBS using a Bead Beater. The homogenates were clarified by centrifugation at 15,800 × g for 4 minutes in a microfuge and the supernatant extracted with chloroform. Half of this supernatant was titered by TCID 50 . RNA was extracted from the remainder using Trizol.
Next-generation sequencing
We amplified poliovirus genomes as 4 overlapping cDNA by RT-PCR. RNA was harvested from either cell-free supernatants or tissues as above and was reverse-transcribed using the SuperScript III First Strand Synthesis System for RT-PCR (Invitrogen 18080) and a mixture of random hexamers and oligo dT primer. The 4 genomic fragments were amplified using primer pairs: WFP37 FORWARD BASE 1 5’ TTAAAACAGCTCTGGGGTTGTACC 3’ + WFP41 REVERSE BASE 2434 5’ GCGCACGCTGAAGTCATTACACG 3’; WFP39 FORWARD BASE. 1911 5’ TCGACACCATGATTCCCTTTGACT 3’ + WFP42 REVERSE BASE 4348 5' AATTTCCTGGTGTTCCTGACTA 3'; WFP13 FORWARD BASE 4087 5' ATGCGATGTTCTGGAGATACCTTA 3' + WFP43 REVERSE BASE 5954 5' CCGCTGCAAACCCGTGTGA 3'; WFP40 FORWARD BASE 5545 5' TTTACCAACCCACGCTTCACCTG 3' + WFP33 REVERSE BASE 7441 5' CTCCGAATTAAAGAAAAATTTACCCC 3'. The thermocycler protocol was 98 °C for 30 seconds, then 30 cycles of 98 °C for 10 seconds, 68 °C for 20 seconds, and 72 °C for 3 minutes, followed by a single cycle of 72 °C for 5 minutes, then 4 °C hold. For each sample, amplification of all 4 fragments was confirmed by gel electrophoresis, and equal quantities of each PCR product were pooled. Seven hundred and fifty nanograms of each cDNA mixture were sheared to an average size of 300 to 400 bp using a Covaris S220 focused ultrasonicator. Sequencing libraries were prepared using the NEBNext Ultra DNA library prep kit (NEB E7370L), Agencourt AMPure XP beads (Beckman Coulter A63881), and NEBNext multiplex oligonucleotides for Illumina (NEB E7600S). The final concentration of each barcoded library was determined by Quanti PicoGreen dsDNA quantification (ThermoFisher Scientific), and equal nanomolar concentrations were pooled. Residual primer dimers were removed by gel isolation of a 300 to 500 bp band, which was purified using a GeneJet Gel Extraction Kit (ThermoFisher Scientific). Purified library pools were sequenced on an Illumina MiSeq with 2 × 250 nucleotide paired-end reads. All raw sequence data have been deposited at the NCBI short-read archive (Bioproject PRJNA396051, SRP113717).
Variant detection
Sequencing reads that passed standard Illumina quality-control filters were binned by index and aligned to the reference genome using bowtie [ 69 ]. SNVs were identified and analyzed using DeepSNV [ 70 ], which relies on a clonal control to estimate the local error rate within a given sequence context and to identify strand bias in base calling. The clonal control was a library prepared in an identical fashion from the pEW-M plasmid and was sequenced in the same flow cell to control for batch effects. True positive SNVs were identified from the raw output tables by applying the following filtering criteria in R: (i) Bonferonni-corrected p < 0.01, (ii) average MapQ score on variant reads >20, (iii) average phred score on variant positions >35, (iv) average position of variant call on a read >62 and <188, and (v) variant frequency >0.001. We only considered SNVs identified in a single RT-PCR reaction and sequencing library for samples with copy number ≥10 5 genomes/μl supernatant or in 2 separate RT-PCR reactions and sequencing libraries for samples with copy number 10 3 to 10 5 genomes per μl (e.g., in tissue studies). Our strategy for variant calling as well as our benchmarked sensitivity and specificity are described in [ 47 ], and all code can be found at https://github.com/lauringlab/variant_pipeline .
Statistical analysis
No explicit power analyses were used in designing the experiments. In most cases, we used 5 biological replicates. In a few cases, we used fewer (3) or more (7) where the variance was either sufficiently low or high. The number of replicates, the statistical tests used, and the relevant p -values are reported in each figure legend or the main text ( Fig 2C and 2D only). All replicates within the dynamic range of each assay are reported (i.e., no replicate experiments were excluded). Data on the relative adaptability of WT and 3D G64S populations were analyzed with a 3-level linear mixed-effects model estimating a random slope and intercept of time nested within each fitness measurement replicate (measID), nested within each lineage replicate (repID). Virus was included as a fixed effect. Models were fit with the R package lme4; all code for this model can be found at https://github.com/lauringlab/speed_fidelity .
Supporting information
S1 Data
Plotted data for all figures.
(XLSX)
S1 Text
Models of the speed–fidelity trade-off (model 1) and within-host bottlenecks (model 2).
(DOCX)
S1 Fig
In vitro assay of polymerase-mediated single-nucleotide incorporation.
(PDF)
S2 Fig
Plots of fitness versus passage in adaptability experiment.
(PDF)
S3 Fig
Mutagen sensitivity of 2C-V127L variants.
(PDF)
S1 Table
Mutations conferring resistance to 1 mM guanidine.
(DOCX)
S2 Table
Number, age, and sex of mice used in all experiments.
(DOCX)
S3 Table
Raw data for calculation of LD 50 . LD 50 , median lethal dose.
(DOCX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The way humans represent numbers has been a recurrent subject of interest both in science and education. Evidence for a close connection between numerical and spatial representations dates back to the 19 th century, when Galton described how people visualised numbers and number ranges, which took sometimes very elaborate spatial forms [1] . Since these early observations, there has been extensive behavioural evidence for the relation between numbers and space (for reviews, see [2] – [4] ).
One classical demonstration of the number-space association is the so-called SNARC ( S patial N umerical A ssociation of R esponse C odes) effect. Dehaene and colleagues [5] , [6] described that, during a binary classification task on single Arabic digits, adult participants were systematically faster to respond to small numbers with their left hand and to large numbers with their right hand. Classically, the SNARC effect is thought to reveal a spatial code in which numbers are represented horizontally [7] and from the left to the right (in Western participants) [6] according to their magnitude (mental number line hypothesis). This code is supposed to be activated automatically each time Arabic digits are processed - even in classification tasks that do not require access to number magnitude, such as in parity judgment tasks [6] . More recently, an alternative dual processing model has been proposed by Gevers and colleagues [8] – [10] . According to this account numerical magnitudes are associated with a verbal-spatial code ( [11] – [13] , see also [14] ) such that the verbal concepts of “small”/“large” are associated with the concepts of “left”/“right” respectively [11] , [15] . Whereas the mental number line theory proposes that number-space interactions affect semantic representations, the latter model situates the origin of the SNARC effect at response selection stages [9] , [16] , [17] . A third framework, elaborated by Fias and colleagues, challenges the role that cardinal long-term memories are thought to play in number-space associations [18] , [19] . This working memory (WM) account postulates that the SNARC effect depends on serial position in WM [18] , [19] , such that numbers (or any items) forming the beginning/end of a WM sequence are associated with the left/right side of space, respectively ( [19] , see also [20] ).
Despite this rich theoretical context and although it has often been replicated (e.g. [21] , [22] , see also [23] , [24] ), the SNARC effect is also characterized by high inter-subject variability that is still poorly understood. According to Wood, Nuerk, & Willmes [25] , [26] , the proportion of participants showing a SNARC effect varies between 65% and 75%. Individual variability with respect to SNARC effects has been attributed to relatively unspecified sources such as “individual differences in implicit mental representation of numbers, which differs from left to right representation” [27] , but the more concrete roles of finger counting habits [28] – [30] , response speed [10] , [24] , [31] and gender [32] have also been described. Inconsistent results in number-space associations appear to be a signature of all tasks assessing the interaction of numerical and spatial representations. The finding that numbers orient visuo-spatial attention according to their magnitude (small numbers to the left, large numbers to the right) (e.g. [33] ) is for instance known to be highly vulnerable to task manipulations and context [34] – [36] and could not be replicated in all studies [37] . Consequently, in order to better understand the nature of spatio-numerical associations it will be necessary to further investigate which factors influence them and how they lead to individual differences in number-space interactions.
When attempting to explain the individual variance observed in SNARC tasks, it seems particularly interesting to consider the role of mathematical proficiency. In SNARC experiments, participants are indeed required to perform basic numerical tasks, such as parity or magnitude judgments. Data from the individual differences as well as the neuropsychological literature both in children and adults demonstrate that performance in basic number tasks is systematically related to participants’ math proficiency. Recent studies exploring individual differences in number approximation tasks highlight the relationship between mathematical abilities and approximate number sense (for adults see [38] – [41] ; for children see [42] – [45] ). In children, further evidence comes from number line estimation tasks [46] – [49] and number comparison tasks [50] – [52] . Accordingly, the strength of number-space interactions (i.e. SNARC effects) associated with these basic number tasks might also be affected by individual differences related to math expertise.
Additionally, the neuropsychological literature repeatedly demonstrated that children and adults with developmental dyscalculia differ from their normally achieving peers in basic numerical processing tasks [53] – [58] . These differences are thought to arise either from differences in number magnitude representations [55] , [59] – [61] , in accessing number magnitude representations [58] , [62] – [64] or in abilities depending on more domain general factors such as working memory or inhibition [65] – [71] .
Given these results we hypothesized that the access of numerical representations and/or domain general factors such as working memory might modulate the way numerical and spatial information is integrated, leading to differential SNARC effects according to math proficiency. Young adults with high proficiency levels in mathematics might for instance display weaker SNARC effects because they have a better access to numerical representations allowing them to retrieve parity status information more easily. This would in turn decrease the interference of task irrelevant factors, such as digit magnitude in parity judgment SNARC tasks (see also [72] ). Concretely, parity status for one-digit Arabic numbers is thought to be retrieved from associative semantic networks [67] , [73] along with other arithmetic properties [6] via a spreading activation process [65] , [74] . Access to number representations has been shown to be modulated by math proficiency [58] , [62] – [64] , [75] , [76] , with lower math proficiency leading to less automatic access to number representations. Less automatic access, on the other hand, is known to result in more executive load [77] , which in turn has been shown to prevent prioritization of target processing as well as inhibition of distractor processing [78] . From this point of view, math proficiency might well contribute to inter-individual variability in parity judgment SNARC tasks.
Numerical proficiency was already evoked in previous discussions of individual variability in SNARC tasks (e.g. [6] , [31] , [79] ), but comprehensive studies remain scarce. Former studies exploring the influence of math proficiency and math training on the quality of participant’s SNARC effect revealed tendencies towards a modulatory impact; however, they lacked statistical power to draw firm conclusions concerning their influence [6] , [79] . So far only one study made a consequent effort to collect data in a large population (n = 71) of university students from study fields with two different math-level requirements [31] . Contrary to their expectations the authors could, however, not observe any systematic relation between math proficiency and the strength of SNARC effects, calling for more work on this topic.
The present study aimed to further investigate the influence of mathematical proficiency on number-space interactions as indexed by the SNARC effect, controlling for more domain general factors such as processing speed and visuo-spatial working memory. In order to explore a large range of mathematical proficiency levels, we formed three different math proficiency groups by recruiting university students from study fields with math-heavy vs. math-light curricula, including also students reporting specific math difficulties (i.e. Math Expert group, Control group, Math Difficulties group). To prevent potentially confounding gender-effects the groups were balanced for gender as far as possible, leading to gender-matched groups of problem-free participants. To complement the math proficiency characteristics indicated by the study field background, we also assessed participants’ arithmetic skill level with respect to the basic arithmetic operations. Using a parity judgment task, we expected that participants with more efficient numerical processing skills would display weaker SNARC effects. Concretely, we hypothesized that during a parity judgment, task-irrelevant magnitude information - and the associated spatial code leading to SNARC effects - would interfere less in adults that are highly trained and proficient with numbers. In contrast, adults that are weaker in mathematics or report specific difficulties in this domain should reveal stronger SNARC effects, because the irrelevant magnitude information (and the associated spatial representations) interferes more strongly with the parity judgment task. Moreover, if the impact of mathematical proficiency levels on the individual differences in SNARC effect strength is specific, we further expected that more general factors such as processing speed and working memory cannot explain the relationship between math proficiency and SNARC effect strength.
Methods
Ethics statement
In accordance with the National Ethics Committee (CNER) approving the present study, all participants gave written informed consent prior to participating.
1. Participants
A total of 95 university students took part in the present study in exchange for payment. In order to control for equal gender and math proficiency level distribution, the students were recruited on behalf of their study fields. Only students of curricula with either a clear predominance (Math Expert group) or absence of explicit daily number and mathematics use were included in the study. Within the latter group we further distinguished students reporting average math abilities (Control group) from those reporting specific difficulties with mathematics (Math Difficulties group).
1.1. Math Expert group
The Math Expert group (ME) included 38 students with a study field having a strong numerical demand (e.g. mathematics, engineering and sciences); 19 were female, 5 were left-handed and their mean age was 23.2 years (SD = 2.5, range = 19–29 years). None of the participants reported specific difficulties with mathematics and/or had a diagnosis of a learning disability.
1.2. Control group
The Control group (CON) included 38 university students enrolled in a field of study with no explicit use of mathematics (e.g. literature, linguistics and law), 20 participants were female, 1 was left-handed, and their mean age was 23.1 years (SD = 3.1, range = 18–33 years). None of the participants reported specific difficulties with mathematics and/or had a diagnosis of a learning disability.
1.3. Math Difficulties group
The Math Difficulties group (MD) included 19 university students enrolled in a field of study with no explicit use of mathematics (e.g. literature, linguistics and law) and who reported experiencing specific difficulties with numbers. Seventeen were female, 1 was left-handed and their mean age was 24.8 years (SD = 3.8, range = 20–31 years). None of these participants had been diagnosed with a learning disability, but all reported having difficulties with numbers specifically.
2. Materials & Procedure
The computerized tasks were programmed in E-prime (Version 2.0.8.79; [80] ) and administered using a Lenovo ThinkPad 61 Tablet Laptop with a 12.1 in. color monitor (1024×768 Pixels), in a quiet room.
2.1. Arithmetic, working memory and processing speed assessment
Arithmetic competency. To assess individual arithmetic proficiency, both timed and untimed arithmetic tests were administered to the participants.
Untimed arithmetic paper-pencil test: “Arith”. We used the battery of arithmetic operations developed by Shalev et al. [81] and adapted by Rubinsten & Henik [58] . This battery assesses proficiency in arithmetical operations including addition, subtraction, multiplication and division problems, ranging from number facts (5 problems/operation) over complex arithmetic (8 problems/operation) and decimals (4 addition and 4 subtraction problems) to fractions (5 problems/operation). Instead of errors (cf. [58] ) we scored 1 point per correct problem, hence participants could reach a maximum score of 80 points.
Timed computerized arithmetic task: “FastMath”. The timed computer-based calculation task was developed and described in detail by Mussolin and colleagues [56] . The participants were asked to solve addition, multiplication and subtraction problems on one- or two-digit Arabic numbers. During each trial an arithmetic problem was presented centrally in Times New Roman font, pt. size 50, along with two possible response propositions presented below. Participants had to press the key (“A” or “L” on a standard QWERTZ keyboard) on the side corresponding to the correct response. The experiment consisted of three blocks of 20 problems, one per arithmetic operation. Order of stimuli presentation and position of the correct response were randomized across trials.
Visuo-spatial working memory (VSWM). This paper-pencil test is based on the Visual Pattern Test [82] , [83] and provides a measure of the spatial-simultaneous working memory span. Participants were presented a series of matrices, progressively increasing in size, where half of the cells were filled in black. After the presentation phase, the participants had to reproduce the memorized patterns of filled squares in a blank matrix. The highest number of correctly recalled filled squares was taken as measure of VSWM span.
Processing speed (GPS). Both general and numerical processing speed measures were obtained in all participants.
General processing speed (GPS). To assess GPS, participants performed a speeded matching to sample task (see also [56] ). In each trial, a shape was presented centrally on the screen and just below the same shape was displayed with a new shape. Participants simply had to press the key on the side corresponding to the matching shape. Twenty trials of this task were performed at the end of the timed computerized arithmetic task.
Parity judgment reaction times (PJ-RT). The SNARC effect was evaluated using a parity judgment task. During this task participants’ response times (and accuracy scores) are recorded in order to compute the SNARC effect (for details cf. 2.4). However, the response times collected in this task can also be used to assess participants’ processing speed for this specific numerical task. This information concerning the response times during parity judgment (PJ-RT) complements the above-mentioned indication on participants’ general processing speed in a non-numerical task.
2.2 Descriptive information on the group composition
Details of the descriptive information concerning the three populations (number of men and women, mean age, number of left-handers) are given in Table 1 . The mean ages did not differ significantly across the three groups ( F (2,92) = 2.4, p >0.05), nor did the number of left-handers ( χ 2 (2) = 3.2, p >0.05). The first two groups matched closely for gender ( χ 2 (1) = 0.05, p >0.05), but due to the composition of the third group (MD) overall the number of men/women differed significantly between the three groups ( χ 2 (2) = 9.2, p <0.05).
10.1371/journal.pone.0085048.t001 Table 1
Descriptive information and mean performance for the three groups in the general assessment tasks.
ME
CON
MD
F
Mean (SD)
Mean (SD)
Mean (SD)
Descriptive Information
N
38
38
19
Gender (M/F)
19/19
18/20
2/17
χ 2 (2) = 9.2 *
Age (in years)
23.2 (2.5)
23.1 (3.1)
24.8 (3.8)
2.4; η 2 = 0.05
Handedness (R/L)
33/5
37/1
18/1
χ 2 (2) = 3.2
Arithmetic
Arith (ACC)
92.4 (6) a ***
82.8 (13)
72.7 (19) b *
16.31 *** ; η 2 = 0.27
FastMath (ACC)
94 (3) a †
92.2 (5.3)
90.5 (5.6)
4.26 * ; η 2 = 0.08
FastMath (RT)
2688 (891) a *
3234 (1124)
4850 (2168) b ***
9.91 *** ; η 2 = 0.27
zArithmetic
1.27 (1.2) a **
–0.14 (2.1)
–2.26 (2.9) b **
17.09 *** ; η 2 = 0.30
Response Speed
GPS
715 (273)
837 (679)
1086 (792)
2.29; η 2 = 0.05
PJ-RT
535 (68) a *
575 (79)
619 (75) b *
8.47 *** ; η 2 = 0.16
Visuo-spatial working
Memory span
9 (1.9) a *
8.1 (1.7)
7.8 (1.7)
3.5 * ; η 2 = 0.07
Note. Standard deviations are shown in parentheses; RTs are given in ms and ACC in percent; significant differences are indicated by * p<0.05; ** p<0.01; *** p<0.001; († p = 0.06) . A significant contrast between ME and CON is indicated by “a” followed by the level of significance; a significant contrast between MD and CON is indicated by “b” followed by the level of significance. Welch’s F is indicated in case the homogeneity of variances assumption was violated.
Concerning arithmetic tests, we found a significant group effect for all measures, confirming the distinct mathematical proficiency levels of the three groups. In the untimed Arith battery, planned comparisons indicated that the participants of the CON group made significantly more mistakes than the participants of the ME group ( p <0.001), but significantly less mistakes than the participants of the MD group ( p <0.05). Regarding the accuracy in the timed FastMath test, there was a similar trend between the ME and the CON group ( p = 0.06) but no difference between the CON and the MD group. On the other hand when considering the speed with which the participants solved the arithmetic problems in the FastMath test, the participants of the CON group were significantly slower than the participants of the ME group ( p <0.05), but significantly faster than the participants of the MD group ( p <0.01).
In order to have a single arithmetic proficiency measure, we computed the composite score zArithmetic from the normed values of the arithmetic tests available: zArithmetic = zArith (ACC) + zFastMath (ACC) – zFastMath (RT). As expected, there was a significant effect of group on this composite score, with the participants of the CON group obtaining a lower value than the participants of the ME group ( p <0.01) and a higher value than the participants of the MD group ( p <0.01). These different composite scores reflect the fact that the CON group was slower and more error prone in arithmetic problem solving than the ME group, but faster and more correct than the MD group.
There was no group effect in general processing speed as assessed by the GPS task ( p >0.05). In contrast, the three groups differed significantly in PJ-RT, with ME participants being significantly faster than CON participants ( p <0.05) and CON participants significantly faster than MD participants ( p <0.05) to judge the parity of single digits. The main effect of group was also significant in the VSWM task, because the ME had a larger visual short-term memory span than the CON group ( p <0.05). The 2 weaker math groups (CON and MD) on the other hand achieved similar results ( p >0.5). However, all three groups were within one standard deviation of the mean of the normative data of the VSWM task (9.08 ± 2.25, see [82] ).
2.3. Experimental Task: SNARC (computerized)
In order to assess the participant’s SNARC effect, they were administered a parity judgment task on single digits. The design of this task was adapted from Dehaene et al. [6] . During the parity judgment task, the participants had to judge whether a centrally presented Arabic digit was odd or even. Each trial started with an empty black-bordered transparent square on a white background (sides 100 pixels, border 2 pixels). After 300 msec, one of ten possible stimuli (Arabic digits 0, 1, 2, 3, 4, 5, 6, 7, 8 or 9) presented in black on a white background in font Arial pt. size 48, appeared at the center of the square and remained for 1300 msec. The intertrial interval consisted of a blank screen of 1300 msec. The stimuli were presented in a pseudo-random order, no number appeared twice in a row, and the correct response could not be on the same side more than three times consecutively. Responses were given by pressing the “A” or the “L” key of a standard QWERTZ keyboard.
Each participant completed two blocks, one in each mapping (in one block “A” was assigned to “odd”, in the other one “A” was assigned to “even”); block order was counterbalanced across participants. Each block started with 12 to 20 training trials, depending on response accuracy. An accuracy threshold of 70% correct answers had to be reached in order to proceed directly after 12 training trials to the experimental trials, if the threshold was not reached, another 8 training trials were administered before the experimental trials started. The experiment itself consisted of 180 trials, 90 trials per block; each number was presented 9 times per block.
Participants all started with the SNARC task, then “FastMath”, followed by “GPS”, the Arith paper-pencil test and then the VSWM task. The participants were part of a larger project including additional behavioral measures, not reported here and administered at the end of the testing for the present study.
2.4. Statistical Analyses
Prior to data analyses, error trials (with respect to the parity judgment) were removed from the data (5.78% of all trials). A univariate ANOVA revealed that the three participant groups did not differ in error rates ( F (2, 92) = 2.4; p >0.05). Reaction times (RTs) longer or shorter than 2.5 standard deviations from the individual mean were considered outliers and removed (2.67% of all trials).
In order to control for possible biases of parity status on lateralized RT (Markedness Association of Response Codes effect-MARC, see [84] ), we conducted a repeated measures ANOVA with Parity status (odd, even) and Response side (left, right) as within subjects variables and Group as a between subjects factor. There was no interaction between Parity status and Response side ( F (1, 92) = .56; p >.4; η 2 = .006) and no interaction of a MARC effect with Group ( F (2, 92) = .36; p >.7; η 2 = .008), hence we did not further investigate MARC effects.
For almost two decades, studies investigating the SNARC effect used regression analysis methods for repeated measures data following Lorch and Myers [85] as suggested by Fias and colleagues [21] . This method implies calculating mean RTs for each digit and response side (left/right) and for each individual subject separately. Individual RT difference scores (dRT) are then computed by subtracting for each digit the mean RT of left-sided responses from the mean RT of right-sided responses. The resulting dRT scores are submitted to a regression analysis, using the magnitude of individual stimuli numbers as predictor variable. Negative regression weights (slopes) reflect SNARC effects in the expected direction (faster left/right-sided RT for small/large digits respectively).
Recently, the habit of only using regression slopes to determine the strength of the SNARC effect has been questioned [86] , [87] . These authors argue that even though the Lorch and Myers regression method allows testing the significance of the linear relation between numbers and dRT (i.e. the expected RT difference between right and left hands for a given change of magnitude), it does not provide an estimate of the correlation between the dRT and number magnitude. Hence it is reasoned that individual slopes should not be interpreted as effect sizes of the SNARC effect [87] . Additionally, Tzelgov and colleagues propose to use magnitude as the predictor variable instead of individual numbers in order to avoid MARC effects (see [84] ).
Taking into account these recent methodological criticisms, we will analyze hypothesized group effects in the SNARC effect by conducting a repeated measures ANOVA on dRT with Magnitude as a within subject and Group as a between subjects factor as suggested by Pinhas and Tzelgov and colleagues [86] , [87] . In order to avoid bias induced by possible MARC effects, we first collapse RT to an even and an odd digit, resulting in 5 Magnitude categories: Very small (0, 1), Small (2, 3), Intermediate (4, 5), Large (6, 7) and Very large (8, 9) for each subject and response hand separately. We then compute for each subject dRTs for each Magnitude category. In this approach, a SNARC effect is revealed by a significant main effect of Magnitude associated with a significant linear trend. Additionally, this method provides us with an effect size of the linear trend. In the present study, the group factor differentiates between the three experimental groups (ME, CON, MD).
Additionally, we computed regression slopes (SNARC slope) with individual numbers as proposed by Fias and colleagues [21] since they a) directly reflect the interaction between numerical magnitude and response side b) highlight the inclination and direction of lateralized RT effects associated with the underlying hypothetical number line and c) permit direct comparison with the slope results reported in previous studies on the SNARC effect over the last 17 years. To complete our analyses, we report individual effect size measures of the number-space interaction, which comprise information on the scattering of the data points around the linear regression slope. Thus, we computed individual correlation analyses between dRT and Magnitude, yielding individual Pearson’s r , which were then transformed to z-scores using a Fisher transformation in order to have individual (and normally distributed) measures that could be correlated with the other variables (e.g. arithmetic scores).
According to our hypotheses, the ME group, which is expected to have the weakest SNARC effect, should have the least negative SNARC slope compared with the other two groups, whereas the MD group should have the most negative SNARC slope, reflecting the most pronounced number-space interaction. Similar results are also expected when computing the relation between individual slopes and arithmetic abilities. In contrast, we did not have any specific hypotheses regarding the impact of arithmetical proficiency on individual SNARC effect sizes since they rather reflect the scattering of data points around the linear regression than the shape (inclination and direction) of the number-space interaction itself.
Results
1. The SNARC effect: group contrast analysis
Following the approach suggested by Pinhas, Tzelgov and colleagues [86] , [87] , we conducted a repeated measures ANOVA on dRT with Magnitude (Very small, Small, Intermediate, Large and Very Large) as a within subjects variable and Group as a between subjects variable. This analysis revealed a significant main effect of Magnitude ( F (4, 368) = 39.9; p <0.001; η 2 = 0.303) associated with a significant linear trend ( F (1, 92) = 129.8; p <0.001; η 2 = 0.59), meaning that there was a significant SNARC effect in the entire sample. An interaction between Magnitude and Group confirms our hypothesis that the SNARC effect differed between the experimental groups ( F (8, 368) = 2.36; p <0.05; η 2 = 0.05). Evaluating the math proficiency groups separately, the analysis reveals a significant SNARC effect in every group (ME: main effect of Magnitude F (4, 148) = 7.99; p <0.001; η 2 = 0.18; associated linear trend F (1, 37) = 21.2; p <0.001; η 2 = 0.37; CON: main effect of Magnitude F (4, 148) = 16.37; p <0.001; η 2 = 0.31; associated linear trend F (1, 37) = 54.9; p <0.001; η 2 = 0.60; MD: main effect of Magnitude F (4, 72) = 16.8; p <0.001; η 2 = 0.48; associated linear trend F (1, 18) = 57.3; p <0.001; η 2 = 0.76; see also Figure 1 ). (These results remained the same when RT for « 0 » and « 5 » were excluded from the analyses (F(6, 276) = 3.43; p<0.01; η 2 = 0.07; see [84] , [88] for the special status of « 0 »; however, these results need to be interpreted with caution due to the reduced set-size on which they are computed. Additionally, they are based on a post-hoc simulation not reflecting actual experimental settings known to critically influence SNARC effects, i.e. [6] .)
10.1371/journal.pone.0085048.g001 Figure 1
dRT (in ms) as a function of Magnitude category by group.
Lines represent the linear fits on group data. A negative relation indicates the presence of a SNARC effect. The inset depicts linear trend effect sizes per group.
As mentioned in the methods section, in addition to the group analyses, we computed individual SNARC effect measures. Accordingly, we analyzed our results following Fias and colleagues [21] , obtaining regression slopes in order to allow comparison with SNARC studies published previously. We also computed individual effect size measures as described in the methods section. The regression analyses of individual digits on dRT revealed a significant negative (unstandardized) slope in the ME group (B = –5.25, one-tailed comparison of B to zero: t (37) = 3.92; p <0.001, effect size: –0.46), in the CON group (B = –8.82, one-tailed comparison of B to zero: t (37) = 7.15; p <0.001, effect size: –0.77) and in the MD group (B = –13.23, one-tailed comparison of B to zero: t (18) = 9.38; p <0.001, effect size: –1.11) respectively.
A one-way ANOVA on SNARC slopes and SNARC effect sizes with group as a between-subjects factor revealed that the groups differed significantly with respect to the SNARC effect (slopes: F (2,92) = 7.12, p <0.001; η 2 = 0.13; effect sizes: F (2,92) = 4.01, p <0.05; η 2 = 0.08). There was a significant linear trend, (slopes: F (1,92) = 14.2, p <0.001, η 2 = 0.13; effect sizes: F (1,92) = 8.01, p <0.01; η 2 = 0.08), indicating that the strength of the SNARC effect increased from ME over CON to MD. In other words, the SNARC increased with decreasing math proficiency of the groups. Planned contrasts revealed that the ME group had a significantly weaker SNARC effect than the CON group (slopes: t (92) = 2.04, p <0.05 (one-tailed), r = 0.21; effect sizes: t (92) = 1.64, p = 0.05 (one-tailed), r = 0.17), and the CON group had a weaker SNARC effect than the MD group, (slopes: t (92) = 2.07, p <0.05 (one-tailed), r = 0.21; effect sizes: t (92) = 1.44, p = 0.08 (one-tailed), r = 0.15). Reflecting the group differences in SNARC effect sizes, 74% of the participants in the ME group had a negative SNARC slope, 89% of the participants in CON group, and 100% of the participants in the MD group.
Our findings were confirmed by an ANCOVA including GPS, PJ-RT and VSWM span as covariates, yielding a main effect of group on SNARC slopes ( F (2,89) = 3.7, p <0.05; η 2 = 0.08) and effect sizes ( F (2,89) = 3.3, p <0.05; η 2 = 0.07), but no effect of either GPS, PJ-RT or VSWM (all p s>0.14). These analyses confirm that the strength of spatio-numerical interactions was significantly modulated by mathematical proficiency groups, even when controlled for potential confounds due to differences in processing speed (GPS, PJ-RT) or visuo-spatial working memory (VSWM span).
2. The SNARC effect: individual correlation analysis
To investigate the SNARC effect on an individual level, Pearson correlation analyses were conducted (see Table 2 ). The correlation analyses revealed that GPS and PJ-RT were related to each other and both correlated negatively with the zArithmetic score. Hence, participants that were faster in the speeded matching to sample task were also faster to do parity judgments. Moreover, they performed better in the arithmetic tests. In contrast to GPS, PJ-RT was additionally related to VSWM, participants that were faster in doing parity judgments had a better VSWM span. VSWM span was also related to the zArithmetic score, revealing that participants with a larger VSWM span also did better in the arithmetic tests.
10.1371/journal.pone.0085048.t002 Table 2
Correlations between different variables (N = 95).
1.
2.
3.
4.
5.
1. SNARC slope
2. SNARC effect size
.68 **
3. zArithmetic
.28 **
.17 ♯
4. VSWM
.12
.15
.39 **
5. GPS
–.19
–.04
–.27 **
–.17
6. PJ-RT
–.30 **
–.08
–.37 **
–.31 **
.38 **
Note. # p <0.1; ** p<0.01.
The SNARC effect measures of slope and effect size were related to each other; the steeper the participant’s slope, the more important his or her effect size. Furthermore, SNARC slopes correlated positively with the zArithmetic score. (The relation between the SNARC slope and zArithmetic remained similar when RT for the stimuli « 0 » and « 5 » were not included in the analyses: SNARC slope : r = .19 ; p<.07 ; SNARC effect size : r = .11 ; p>.1.) This finding illustrates that participants scoring lower in the arithmetic measures also had more negative slopes, meaning more pronounced SNARC effects. SNARC effect sizes and zArithmetic scores were marginally related as well, participants scoring better in arithmetic displayed slightly more important effect sizes. In contrast, participants who scored higher in the arithmetic tests had relatively weaker SNARC effects (i.e. less negative slopes). Moreover, the SNARC slope was related to the speed with which participants performed parity judgments (i.e. PJ-RT). Hence, the slower the participants were to decide whether a digit was odd or even, the steeper their slope. Interestingly, PJ-RT did not correlate with the SNARC effect size. Additionally, participant’s SNARC effects related neither to GPS, nor to their VSWM span; confirming the results obtained in the ANCOVAs of the group analysis.
In order to confirm that the present findings were not exclusively driven by the population reporting specific difficulties with numbers (MD group), we conducted additional correlation analyses excluding this group. A total of 76 participants (ME and CON group members) were included in the analyses, of which roughly half (N = 39) were female. Pearson’s correlation analyses confirmed the previous findings by showing that SNARC slopes were positively related to arithmetic proficiency ( r = .25, p <0.05) but neither to VSWM span ( r = .06, p >.5) nor to general processing speed ( r = –.15, p >.1). In contrast to the SNARC slope the relation between SNARC effect size and arithmetic proficiency scores did not reach significance ( r = .13, p >.1). (When RTs to the stimuli « 0 » and « 5 » were dropped from the analyses on the reduced sample size correlation coefficients were: slope : r = .16, p>.1 ; effect size : r = .08, p>.4)
To fully understand the significance of the two SNARC-related measures reported in the present study (i.e. slope and effect size), we investigated the individual relations of each SNARC measure to the variables of interest when the respective other SNARC measure was held constant. Consequently, we conducted two partial correlation analyses. In the first analysis, we investigated the relationship between SNARC slope, PJ-RT and zArithmetic when controlling for SNARC effect-size, whereas in the second analysis we investigated the relationships between SNARC effect size, PJ-RT and zArithmetic when controlling for the SNARC slope measure (see Table 3 ).
10.1371/journal.pone.0085048.t003 Table 3
Partial correlation analyses controlling for SNARC effect size (A) or SNARC slope (B).
(A)
1.
2.
(B)
1.
2.
Effect size
1.zArithmetic
Slope
1.zArithmetic
2.PJ-RT
–.37 **
2.PJ-RT
–.32 **
3.Slope
.22 *
–.34 **
3.Effect size
–.02
.19 #
Note. # p <0.1; * p <0.05; ** p<0.01.
The partial correlation analyses showed that whereas the relation between zArithmetic and SNARC slope remained significant when controlling for effect size, the marginal relation between zArithmetic and SNARC effect size disappeared when controlling for SNARC slope. Additionally, whereas the previously reported relation between SNARC slope and PJ-RT remained when controlling for effect size, the relation between SNARC effect size and PJ-RT reversed when controlling for SNARC slope.
3. The SNARC effect: multiple regression analyses
Finally, we conducted multiple regression analyses in order to investigate relations between the SNARC effect (slope and effect size) and each predictor variable when the effects of the other predictors are held constant. Specifically, we were interested to see whether arithmetic proficiency explained variance of the SNARC effect when GPS, PJ-RT and VSWM capacities were statistically controlled for. Consequently, the following predictors were entered: GPS, PJ-RT, VSWM and zArithmetic. The results show that zArithmetic and PJ-RT were the only predictors that explained a marginally significant amount of variance of the slope of the SNARC effect (see Table 4 ).
10.1371/journal.pone.0085048.t004 Table 4
Results of the regression analysis with SNARC slope as dependent variable.
B
SE
β
t
p
(Constant)
5.84
8.14
.72
.48
GPS
−.001
.001
−.06
−.52
.61
PJ-RT
−.02
.01
−.22
−1.89
.06
VSWM
−.12
.48
−.03
−.25
.80
zArithmetic
.66
.39
.19
1.70
.09
Note . R 2 = .13; adj. R 2 = .09; F(4,90) = 3.21, p<0.05.
Considering SNARC effect size, the regression model failed to reach significance ( F (4, 90) = 0.88; p >.1).
Together, the results of the regression analyses confirm the importance of zArithmetic in the observed variability of the SNARC slope. Furthermore, they confirm the importance of PJ-RT in the observed variability of the slope of the SNARC effect reported by previous studies (i.e. [31] ). (Note that the regression models did not reach significance when RT data to the stimuli « 0 » and « 5 » are dropped from the analyses.)
Discussion
The present study aimed to investigate whether mathematical proficiency levels affect the strength of number-space interactions as indexed by the SNARC effect. We recruited three groups of university students differing starkly in their mathematical level. Analysis of their arithmetical performance confirmed that students from mathematical study orientations (ME) were more proficient in arithmetic than their study colleagues from non-mathematical orientations. Moreover, within the latter student population those reporting specific difficulties in mathematics (MD) were even less proficient than their colleagues who did not relate specific math problems (CON). Confirming our hypothesis, we observed a main effect of group on the SNARC effect, revealing significantly different number-space interactions in the three groups. Indeed, the CON group displayed a weaker SNARC effect than the MD group, but a stronger SNARC effect than the ME group. Critically, when controlling for general processing speed, parity judgment reaction time or visuo-spatial working memory, the effect of group on the SNARC effect remained. Correlation analyses pertaining to individual performance levels confirmed the group findings and revealed a significant relation between the slope of the SNARC effect and arithmetic scores. Participants scoring lowest in the arithmetic tests displayed the most important SNARC effects (i.e. most negative slopes) and vice versa for participants scoring highest in arithmetic. In line with previous findings, SNARC slopes (but not effect sizes) also related to response times in parity judgment [10] , [31] . In contrast, there was no relation between the strength of the SNARC effect and general processing speed or visuo-spatial working memory span, excluding these general accounts for the systematic relationship observed between number-space interactions and mathematical skill level.
The present findings confirm the first indications in the literature [6] , [79] that math proficiency modulates the strength of the SNARC effect. However, they contrast with the recent findings of Cipora and Nuerk [31] , who failed to find systematic relations between math proficiency and SNARC slopes. As mentioned by Cipora and Nuerk [31] , there are a few differences between their study and ours, such as the inclusion of the Arabic digit “0” in our study and different school and language contexts [89] , [90] . Cipora and Nuerk [31] also cite the inclusion of a low-skill group as a potentially influencing difference. Our analyses, however, showed that there is a significant group difference when ME and CON are contrasted directly. Moreover the correlation results remain largely unchanged when the MD group is not included. Whereas we balanced the ME and CON groups in number and in gender, this was not the case in the study of Cipora and Nuerk [31] . In their study, only 25% of all participants were in the math group, the other 75% were included in the non-math group. Additionally, their groups were not balanced in gender, with only 28% of female participants in the math group and 83% of female participants in the non-math group. Since there is evidence for stronger SNARC slopes in male participants [32] , [51] , gender effects might have masked potential math proficiency effects in this population. Indeed, although the impact of math abilities on the SNARC slope was observed robustly and coherently in all analyses of the present study it was only characterized by small to medium effect sizes. Furthermore, both in our and Cipora and Nuerk’s [31] studies the SNARC slope correlated significantly with PJ-RT. When considering that this measure not only reflects participants’ processing speed, but also their numerical ability to judge digit parity, this common finding further supports the existence of a systematic link between mathematical skill level and the SNARC effect.
Finally, it should be mentioned that several other studies which used math abilities as a covariate when investigating SNARC effects also failed to find significant relationships between math proficiency and the strength of numbers-space associations ( [32] , [91] in adults and [92] in children). However, neither of these studies had made specific efforts to sample participants from a very large range of math proficiency levels. In addition, when math proficiency was assessed the math scores relied on performance in a mixture of arithmetical and other mathematical tasks, precluding a direct comparison with the present approach.
Besides the traditionally reported interaction and slope measures which inform on the presence of significant SNARC effects and characterize their shape, we completed our analyses by indicating also effect size measures of the SNARC effect [86] , [87] . In the group analysis the SNARC effect sizes of the three math-proficiency groups decreased linearly with math abilities. This was in line with the observation that a smaller proportion of ME participants (i.e. 74%) had negative SNARC slopes compared to participants of the CON (89%) and MD (100%) groups (hence there was less variance in the presence of SNARC effects in the MD group than in the ME group). In contrast, the correlation and regression analyses indicated that individual arithmetic scores explain considerably less variance of participants’ SNARC effect sizes than of their SNARC slopes. This observation indicates that arithmetic abilities relate to the inclination and direction of the number-space association, rather than the amount of scattering around the linear trend relating lateralized response time to digit magnitude. To explain the negative relationship we expected and observed between the strength of SNARC effects and math proficiency levels, we will discuss distinct (but potentially complementary) hypotheses currently proposed in the literature investigating individual differences in typical and atypical mathematical functioning.
Participants that are more proficient with numbers might have stronger associations between numerical facts than participants with less numerical exposure. In line with data from typical as well as atypical math development indicating that higher math proficiency leads to a more automatic access to number representations [58] , [62−64] , [75] , , students from the MD group could have less automatic access to numerical representation than those of the CON group, who themselves would access number semantics (i.e. parity status) less easily than their colleagues of the ME group.
Less automatic access to semantic (number) representations results in more executive load [77] , which in turn has been shown to prevent prioritization of target processing as well as inhibition of distractor processing [78] . In other words, if access to numerical representations was less automatic in MD or CON group students, they should experience higher executive loads when retrieving the parity of a given numeral. Higher loads then make it harder to (a) prioritize the parity judgment and (b) inhibit the spatially coded magnitude information that was activated in parallel. As a consequence, the response of mathematically less skilled participants would be more influenced by the task-irrelevant digit magnitude and their SNARC effects would be stronger than those of the ME participants. The proposal that inter-individual differences would be caused by differences in prioritizing parity information and inhibiting irrelevant magnitude information would be in line with the theoretical framework of the dual processing model [8] , [9] , [12] . This view is supported by behavioral and imaging studies that have found the SNARC effect to be localized at response related stages, as opposed to representational stages [8] , [9] , [16] , [17] . Specifically, using event-related potentials (ERPs) to investigate the functional locus of the SNARC effect during parity judgment, Keus and colleagues [17] only found evidence for the SNARC effect in response-locked ERPS, but not in stimulus-locked ERPs. Furthermore they found evidence that the SNARC effect is localized at response selection stages that take place prior to response preparation and execution stages. Additional support for the dual processing model is provided by the correlations between parity judgment response times and the slope of SNARC effect that were observed here and in previous studies [10] , [31] . On the other hand, as in Cipora and Nuerk [31] general processing speed assessed in an independent task did not relate to SNARC effects.
Consistent with the proposal that SNARC effects are localized at response selection stages, another hypothesis implicating domain-general factors might explain the findings of the present study. Working memory and inhibition deficits have repeatedly been proposed to be related to arithmetic proficiency in healthy adults [65] as well as in participants suffering from DD [66] – [71] . According to these findings, MD and CON participants should in general have a higher sensitivity to activation-based interference [65] , [66] and lower capacities to inhibit irrelevant information [70] , [71] . Following the above-mentioned principles, these executive difficulties would again lead to less efficient inhibition of task-irrelevant magnitude information (and consequently larger SNARC effects) when mathematically weaker students perform a parity judgment task. A general finding supporting this latter hypothesis is the increase of the SNARC effect with age [24] , [93] and declining general inhibition capacities [94] . In this recent study we assessed the influence of cognitive inhibition abilities on the strength of SNARC effects in younger and elderly participants and thus observed a significant correlation between Stroop and SNARC effects. In contrast, the present study indicates that visuo-spatial working memory capacity does not influence the strength of the SNARC effect. Indeed we observed a significant difference between the VSWM spans of the math groups. But despite this generic group effect, visuo-spatial working memory span did not correlate with individual SNARC slopes. This finding also mirrors our recent observation that individual differences in SNARC effect strength cannot be explained by differential performance levels in a verbal working memory task (i.e. backwards digit recall; [94] ). Whereas the “number access” hypothesis points to specific number treatment difficulties (which in turn weaken distracter inhibition), this last hypothesis points to a domain-general process. Of course, a theoretical framework combining above-mentioned factors is another possibility to provide a comprehensive explanation of the present findings.
In line with the hypotheses that inhibition processes play an important role in the strength of SNARC effects (see also [8] , [24] , [52] ) math anxiety might also contribute to inter-individual differences in number-space associations. Math anxiety negatively influences arithmetical performance [95] , [96] by affecting working memory performance [97] . It also decreases attentional control, which in turn diminishes inhibition capacities [98] . Whereas we tried to minimize the effects of math anxiety by administrating the simple parity judgment task first [95] , [96] , [99] , we cannot definitely rule out that math anxiety might have influenced the results. Consequently it would be interesting to consider math anxiety as a possible variable impacting on the strength of the SNARC effect in future studies.
A final crucial consideration to be taken into account are possible inter-individual differences in the strategies used, as the use of different cognitive strategies could lead to differential SNARC effects in the three groups. In parity judgment tasks, the SNARC effect can for instance be associated with visuo-spatial as well as with verbal-spatial coding [11] , [100] . Depending on their proficiency and training in mathematics, subjects could have employed different strategies to solve the task (see also [101] ). Accordingly a training study by Delazer and colleagues [102] showed a shift of activation in the parietal lobe from the intraparietal sulcus (IPS) to the left angular gyrus (AG) after extensive training of complex multiplication problems. These findings suggest a shift from quantity-based processing to more automatic retrieval ( [102] ; see also the triple-code model of [103] ). A differential study by Grabner and colleagues [104] showed that in healthy individuals, differing only in their mathematical competencies, higher achievers showed more left AG activation during single digit multiplication than their lower achieving peers. These findings were interpreted as high achievers relying more strongly on verbal strategies than low achievers ( [104] , see also [105] , [106] ). Similarly more trained subjects (i.e. ME) supposedly solve parity judgment employing more verbal strategies (similar to automatic fact retrieval) associated with left AG activation while less trained subjects (i.e. CON and MD) might rely more on quantity-based processes, thus activating more the IPS and the neighboring superior parietal regions critically involved in number-space interactions [107] – [110] . To date there are no studies investigating how the use of different strategies would modulate the strength of SNARC effects. To address these questions, future studies should explore how math proficiency levels influence dual task SNARC paradigms such as used by Van Dijck et al. [101] or Herrera and colleagues [111] .
Conclusion
The present study shows that the frequently reported inter-individual variance observed in the strength (and presence) of the SNARC effect is linked to mathematical proficiency. Participants that are more proficient in math have weaker SNARC effects in the classical parity judgment task. These findings could not be explained by general factors related to general processing speed, parity judgment reaction times or visuo-spatial working memory. We propose that they reflect individual differences concerning the access of numerical representations, as well as vulnerability to interference of irrelevant information.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
COVID-19 as a chronic worldwide stressor
Coronavirus disease caused by the SARS-CoV-2 virus (the COVID-19 pandemic) has affected many aspects of daily human life, leading to increased distress and serious mental health issues in society [ 1 – 3 ]. The threat of epidemics to physical health, as well as the related restrictive measures, can act as triggers of distress [ 4 ], thus the COVID-19 pandemic can be considered a chronic worldwide stressor [ 5 ]. Stress is widely studied due to its considerable impact on people’s wellbeing [ 6 ]. In the current study, we focused primarily on the psychological aspect of stress—that is, the subjective feeling of distress and the coping response. This approach is in line with Lazarus and Folkman’s definition of stress—that is, the mismatch between perceived external challenges and the individual’s perceived ability to overcome them [ 7 ].
Coping strategies in response to stress
Empirical findings suggest that the psychological reaction demonstrated in response to a stressor plays a bigger part in adaptation than the magnitude of the external stressor itself [ 7 , 8 ]. Different coping strategies can be employed in response to stressors. Coping comprises the individual’s mental and behavioral attempts to manage, diminish, or endure stress [ 9 ]. Three main types of coping have been identified: task-focused, emotion-focused, and avoidance coping [ 10 ]. Task-focused coping takes the form of a proactive response to a stressful event; it seeks an optimal resolution to the problem and is associated with a sense of control [ 11 ]. Avoidance coping refers to an individual’s engagement in cognitive, behavioral, and often maladaptive activities in an attempt to divert their attention away from the stressor and deny, minimize, or otherwise avoid dealing directly with the stressful situation [ 12 ]. Within this framework, behaviors such as going out with friends, alcohol and substance use, gambling, video gaming, or even compulsive buying can be considered as means of avoidance coping. On the other hand, these behaviors may also be motivated by many other factors (e.g., temporary promotions [ 13 ]) or may serve other functions than avoidance, such as escapism [ 14 ] or seeking social support [ 15 ].
Emotion-focused coping entails responses that are focused on the self, such as emotional responses to a stressful situation, self-preoccupation, and fantasizing/daydreaming [ 16 ]. However, the wide scope of emotion-focused coping has often generated ambiguous results in terms of whether the emotion-focused coping is adaptive or not [ 17 ]. Although emotion-focused coping can be considered adaptive (e.g., positive reappraisal, the seeking of social support) when it actively facilitates emotional processing and expression, which may help to mitigate the individual’s negative reactions to the stressor, it can be considered maladaptive (e.g., denial, self-blame, or counterfactual thinking) when it encourages passivity and avoidance and therefore does not facilitate acceptance or problem solving [ 17 ]. In the current study, we adopted the latter conceptualization of emotion-focused coping, as it appears to be closely related to maladaptive behavioral outcomes such as compulsive buying [ 18 ].
Coping strategies during COVID-19
Like many other aspects of life, the coping strategies employed during the COVID-19 pandemic have changed substantially. For instance, although seeking social support has been identified as an adaptive way to cope with distress [ 19 , 20 ], this coping strategy was not readily available during social distancing. The increased rate of reported loneliness indicates that video calls may not be able to fully substitute in-person social interactions, especially in the long term [ 21 ]. Similarly, a recent study found that lockdown measures, fewer interpersonal interactions, and voluntary self-quarantine were associated with a higher incidence of mental health problems, underscoring the detrimental effects of weakened social support [ 22 ]. Moreover, it is likely that the lack of in-person social support, coupled with uncertainty about the future and fear of the disease, have all contributed to the use of coping strategies that are less adaptive as the pandemic has unfolded [ 4 ]. In line with this, the empirical results indicate that avoidance and emotion-focused coping were highly associated with symptoms of depression and anxiety during COVID-19, whereas task-focused coping was not [ 23 ].
Compulsive buying during COVID-19
Compulsive buying involves a preoccupation with buying or urges to buy that are experienced as intrusive and uncontrollable [ 24 ]. Compulsive buying is associated with negative affectivity and is likely to result in social, personal, and/or financial difficulties [ 25 ]. According to a meta-analysis, the prevalence of compulsive buying behavior is 4.9% in adult representative samples, and even higher among younger adults and women [ 26 ]. The cognitive-behavioral model of compulsive buying [ 25 ] suggests that internal and external triggers, such as negative emotions, depression, and anxiety, may increase compulsive buying. These factors, together with distress, have become pertinent during the COVID-19 pandemic [ 2 , 4 , 5 , 22 , 27 ] and the empirical results confirm that vulnerability to compulsive buying has increased during COVID-19 due to distress. More specifically, experiencing COVID-19 symptoms and fearing COVID-19 were weakly but significantly associated with compulsive buying [ 28 , 29 ]. In line with this, empirical evidence indicates that compulsive buying, together with other maladaptive behaviors such as alcohol and substance use, gambling, smoking, and binge eating, also steadily increased during the first six months of the pandemic [ 30 ]. However, further investigation is needed to ascertain the role of mediating factors between distress and compulsive buying during COVID-19.
In summary, most of the population worldwide have been experiencing elevated distress in their daily lives due to COVID-19 and the related restrictive measures [ 4 , 5 ]. Chronic stress is known to contribute to the increased use of maladaptive behaviors, including compulsive buying, that may help to divert the individual’s attention away from the stressor, at least in the short term [ 31 ]. People who tend to use maladaptive coping strategies may be at increased risk of developing maladaptive behaviors [ 32 – 34 ], thus investigating the relationship between COVID-related distress, coping strategies, and maladaptive behavioral outcomes such as compulsive buying is of the utmost importance during the COVID-19 pandemic.
Aims of the current study
Our aim was to examine the associations between stress, coping, and compulsive buying during the COVID-19 pandemic. More specifically, in the current study we hypothesized that COVID-related distress would be positively associated with (online) compulsive buying, and that emotion-focused coping, but not task-focused coping, would mediate this association. Avoidance coping was not analyzed in the study, since compulsive buying can be considered as a means of avoidance in itself, while many other activities that may serve as means of avoidance (e.g., meeting friends/family or going out for leisure activities) were restricted by the social distancing orders in place at the time, thus asking whether people were engaging in such activities did not seem rational. The hypothesized associations are presented in Fig 1 .
10.1371/journal.pone.0274458.g001
Fig 1
The proposed mediation model.
We also explored whether these associations changed as the pandemic evolved. Our previous analysis of this dataset revealed that COVID-related distress among participants oscillated initially, until around day 80 of the pandemic (i.e., around June 1, 2020), when the number of COVID-19 cases started to rise continuously, as did the participants’ COVID-related distress. This steady growth halted at around day 140 of the pandemic (i.e., around July 17), when the number of new confirmed cases peaked in the U.S. Although reported distress began to decline after this peak, the rate of decrease in COVID-related distress was slower than the rate of decrease in the number of COVID cases [ 30 ]. At the same time, compulsive buying tendencies were continuously on the rise during our sampling period. Based on these events, we split our data collection into three periods: T1 = day 14-day 80 of the pandemic (03/26/2020–06/01/2020); T2 = day 81–140 of the pandemic (06/02/2020–07/17/2020); T3 = day 141–206 of the pandemic (07/18/2020-10/02/2020). Our goal was to examine whether the mediation models were invariant among these three time periods—that is, whether the associations between COVID-related distress, coping, and compulsive buying were similar over time. To our knowledge, this is the first study to examine whether coping acts as a mediator between distress and compulsive buying during COVID-19.
Methods
Sample and procedure
Parts of this dataset have been published in two recent studies [ 30 , 35 ]. Data were collected via Amazon MTurk during the first six months of the COVID-19 pandemic, between March 26, 2020, and October 2, 2020, starting 14 days after the official declaration of the start of the pandemic in the U.S. (Proclamation on Declaring a National Emergency Concerning the Novel Coronavirus Disease [COVID-19] Outbreak, issued on March 13, 2020). Every three days, a new cohort of 25 participants were asked to fill out self-report surveys, excluding individuals who had participated earlier. Inclusion criteria were being 18 years of age or older and logging in from a U.S.-based IP address. Four attention check items were hidden among the regular survey items (e.g., “Please check ‘true’ here.”). Subjects with more than one attentional error were excluded. Furthermore, participants with a maximum score (10 out of 10) on the Marlowe–Crowne Social Desirability Scale [ 36 ] were also excluded. The recruitment process is presented in Fig 2 .
10.1371/journal.pone.0274458.g002
Fig 2
Study participant recruitment process.
The final sample of 1,430 was 40% female (N = 572) with a mean age of 36.6 years (SD = 11). The study procedures were in accordance with the Declaration of Helsinki. The informed consent of participants was obtained before the surveys were filled in. The ethical approval of the Institutional Review Board was obtained prior to data collection (2020-15R3).
The study was pre-registered prior to data collection ( https://osf.io/m5kw9 ). The data, materials, and scripts of analysis are available in open access repositories: https://osf.io/qdhp4/ and https://github.com/anikomaraz/shopping_covid19 .
Measures
Subjective socioeconomic status (SES) was assessed with the help of the item How wealthy do you think you are compared to others ? Participants were asked to respond on a 7-point Likert scale, where 1 corresponded to Among the poorest , and 7 corresponded to Among the richest .
COVID-related distress was assessed with one item: How stressed do you feel about the current situation caused by the coronavirus outbreak ? Participants were asked to respond on a 10-point Likert scale ranging from 1 ( Not at all stressed ) to 10 ( Very stressed ).
The short form of the Coping Inventory for Stressful Situations (CISS) [ 37 ] is used to assess coping strategies in response to a difficult or stressful situation. Participants were asked to indicate on a 5-point Likert-type scale how they normally react when they encounter a stressful situation, ranging from Not at all to Very much . Shortened from the original CISS [ 38 ], the short form of the CISS is a 21-item self-report survey that consists of three subscales, each comprising seven items: task-focused coping, emotion-focused coping, and avoidance coping via social support or giving oneself treats. Emotion-focused coping, as measured by the CISS, is considered maladaptive due to items related to self-blame or counterfactual thinking, which do not facilitate problem solving and may generate passivity and avoidance [ 17 ]. The avoidance coping subscale was not used in the current study due to the fact that most of the activities described in the items in this subscale were restricted during the pandemic (e.g., Go out for a snack or meal ; Spend time with a special person ), and due to the obvious overlapping of the item Buy myself something with compulsive buying as the outcome measure. Cronbach’s α for task-focused coping and emotion-focused coping in the current sample was 0.828 and 0.883 respectively.
Participants who reported making an offline purchase in the previous seven days were asked to fill out the BERGEN Shopping Addiction Scale (BSAS) [ 37 ], which measures offline compulsive buying. The BSAS consists of 28 items that target the Diagnostic and Statistical Manual of Mental Disorders, 5 th Edition (DSM-5) [ 39 ] addiction criteria of salience, mood modification, conflict, tolerance, relapse, withdrawal, and problems. Participants were asked to respond using a 7-point Likert-type scale (1 = Strongly disagree , 7 = Strongly agree ) to items assessing their shopping behavior in the previous 30 days. Cronbach’s α for the current sample was 0.986.
Online compulsive buying was measured via the Compulsive Online Shopping Scale (COSS) [ 40 ]. The COSS was adapted from the BSAS [ 41 ] by adding the word “online” to items (e.g., I felt bad if for some reason I was prevented from shopping/buying things online). The subscales, the instruction regarding the previous 30 days, and the alternative responses thus corresponded to those described above in case of the BSAS. The minimum and maximum values are identical to the BSAS, ranging from 28 to 140. The COSS was only administered to participants who reported making any purchase online in the previous seven days. Cronbach’s α for the current sample was 0.984.
Statistical analysis
Descriptive statistics and reliability testing were carried out using IBM SPSS 24 software (2016). Structural equation modeling was performed in MPlus software Version 8 [ 42 ] with multiple linear regression estimation to test whether the relationship between COVID-related distress and compulsive buying was mediated by emotion-focused and task-focused coping. Gender, age, and subjective SES were controlled for in the model, as previous results for this sample revealed that compulsive buying (both online and offline) differed among high, average, and low SES groups [ 43 ] and, according to a meta-analysis, female gender and younger age may be associated with stronger compulsive buying behavior [ 26 ]. Coping measures were entered as latent variables (i.e., defined by their individual items), and the outcome measures were entered as parceled variables (i.e., defined by the total scores of the seven subscales). The model was initially estimated for the whole sample (Model 1), and time was subsequently used as a grouping variable (Models 2 & 3). The data were split into three phases, based on major events related to the pandemic in the U.S. that were also reflected in the level of distress and compulsive buying behavior within this sample [ 30 ]. These trends are illustrated in Fig 3 .
10.1371/journal.pone.0274458.g003
Fig 3
Changes in the number of new COVID cases in the U.S., the level of COVID-related distress, and compulsive buying throughout the period of data collection.
Note. COSS = Compulsive Online Shopping Scale. T1 = day 14 –day 80 of the pandemic (03/26/2020–06/01/2020); T2 = day 81 –day 140 of the pandemic (06/02/2020–07/17/2020); T3 = day 141 –day 206 of the pandemic (07/18/2020–10/02/2020). Lines were smoothed to reduce noise in the presentation of the data. Source for the number of new COVID cases in the U.S.: https://www.ecdc.europa.eu/en/publications-data/download-todays-data-geographic-distribution-covid-19-cases-worldwide .
Our goal was to examine whether the mediation models were invariant between the three time periods—that is, whether the associations between COVID-related distress, coping, and compulsive buying were similar over time. We therefore carried out two multigroup mediation analyses, where time was used as a grouping variable. In Model 2, path coefficients were estimated freely, whereas in Model 3 they were constrained to be equal among the three time periods. To test whether the mediations were invariant across time, Model 2 and Model 3 were compared in terms of their chi-square statistics [ 44 ]. The proportion mediated—that is, the unstandardized indirect effect divided by the unstandardized total effect [ 45 ]—was calculated for each mediation path in T1, T2, and T3. The models were estimated with both online and offline compulsive buying as their outcome measures.
Results
The means and standard deviations of the assessed variables, as well as their correlations, are presented in Table 1 for the total sample and for T1, T2, and T3 separately.
10.1371/journal.pone.0274458.t001
Table 1 Descriptive statistics and Pearson correlations of the assessed measures.
Total sample (N = 1,430)
T1 (N = 543)
T2 (N = 429)
T3 (N = 455)
Measure
Mean (SD)
2.
3.
4.
5.
6.
7.
Mean (SD)
2.
3.
4.
5
6
7.
Mean (SD)
2.
3.
4.
5.
6.
7.
Mean (SD)
2.
3.
4.
5.
6.
7.
1. CISS task
26.28 (4.97)
-.15 **
-.01
-.08 **
-.07 *
.07 **
.10 **
26.72 (5.54)
-.27 **
-.05
-.15 **
-.13 *
.095 *
.17 **
26.34 (4.33)
-.12 *
-.05
-.04
-.08
.13 **
-.003
25.69 (4.75)
.08
.12 *
.06
.06
.04
.06
2. CISS emot.
21.28 (6.81)
.46 **
.57 **
.59 **
.18 **
-.13 **
19.28 (7.32)
.46 **
.50 **
.48 **
.086 *
-.21 **
22.45 (6.45)
.45 **
.63 **
.65 **
.20 **
-.03
22.55 (5.90)
.44 **
.49 **
.57 **
.16 **
-.04
3. COVID distress
6.56 (2.42)
.40 **
.42 **
.12 **
-.03
6.16 (2.58)
.35 **
.34 **
-.011
-.01
6.75 (2.36)
.43 **
.48 **
.22 **
-.03
6.84 (2.21)
.38 **
.43 **
.10 *
-.03
4. COSS
75.38 (32.81)
.96 **
.50 **
-.17 **
59.12 (27.96)
.94 **
.35 **
-.22 **
83.71 (32.99)
.97 **
.56 **
-.09
86.01 (30.49)
.96 **
.45 **
-.11 *
5. BSAS
74.24 (33.92)
.48 **
-.12 **
57.92 (29.75)
.37 **
-.17 **
83.32 (33.45)
.52 **
-.03
83.70 (31.94)
.45 **
-.06
6. SES
4.24 (1.36)
-.08 **
3.86 (1.24)
-.10 *
4.53 (1.35)
-.06
4.43 (1.40)
-.02
7. Age
36.62 (11)
37.82 (11.57)
36.23 (10.73)
35.50 (10.42)
Note. CISS task = Coping Inventory for Stressful Situations task-focused coping subscale; CISS emot. = Coping Inventory for Stressful Situations emotion-focused coping subscale; COSS = Compulsive Online Shopping Scale; BSAS = Bergen Shopping Addiction Scale; SES = Subjective socioeconomic status.
*Correlations are significant at p < .05.
**Correlations are significant at p < .01.
Notably, there was a significant moderate negative correlation between task-focused and emotion-focused coping in T1, which decreased over T2 and then disappeared in T3. The difference in these correlation coefficients was significant between T1 and T2 (z = -2.41, p < .01), T1 and T3 (z = -5.6, p < .01), as well as T2 and T3 (z = -2.97, p < .01).
There were no significant gender differences in any of the variables. The gender differences in the assessed measures are reported in S1 Table of the Supporting Information.
In our first model, we tested whether the relationship between COVID-related distress and online compulsive buying was mediated by emotion-focused and task-focused coping with the help of structural equation modeling for the entire sample. Age, gender, and subjective SES were controlled for in the mediation model, as there were tendency-level gender differences in emotion-focused coping, and gender and age were significantly correlated to most of the assessed variables. Model 1 demonstrated excellent model fit (χ² = 1117.115, df = 258, RMSEA = 0.048 [0.046–0.051], SRMR = 0.043, CFI = 0.952, TLI = 0.945). As hypothesized, emotion-focused coping mediated the relationship between COVID-related distress (β = 0.47) and compulsive buying (β = 0.47); proportion mediated was 0.32. The explained variance of compulsive buying was 0.56. There was a weak but significant direct path between COVID-related distress and compulsive buying (β = 0.11). However, task-focused coping was not significantly associated with either online compulsive buying or COVID-related distress, thus it was excluded from further analyses.
In Model 2, time was used as a grouping variable to compare the relationship between the assessed variables throughout the different phases of the pandemic. The first period (T1; N = 543) ran from day 14 (the beginning of data collection) to day 80 of the pandemic; the second period (T2; N = 429) started on day 81 and ended on day 140; and the third period (T3; N = 455) ran from day 141 to day 206 (the end of data collection). To assess invariance across these timeframes, two models were estimated. In Model 2, betas were estimated freely, whereas in Model 3, all the path coefficients were constrained to be equal among T1, T2, and T3. Both Model 2 (χ² = 1119.377 [χ² T1 = 482.380, χ² T2 = 297.892 , χ² T3 = 339.105] , df = 420, RMSEA = 0.059 [0.055–0.064], SRMR = 0.049, CFI = 0.951, TLI = 0.947) and Model 3 (χ² = 1,146.918 [χ² T1 = 496.037, χ² T2 = 306.679, χ² T3 = 344.201], df = 426, RMSEA = 0.060 [0.056–0.064], SRMR = 0.058, CFI = 0.949, TLI = 0.947) demonstrated excellent model fit. The Satorra–Bentler Scaled Chi-Square Difference Test [ 44 ] significantly decreased between Model 2 and Model 3 (TRd = 26.02, Δdf = 6, p < .01), revealing that the models were not invariant across the three time periods. In order to explore when exactly the changes occurred, we conducted post hoc analyses in which path coefficients were constrained one by one, and we compared these to Model 2 (where all betas were estimated freely) using the Satorra–Bentler Scaled Chi-Square Difference Test. These analyses revealed that the path between emotion-focused coping and compulsive buying changed substantially throughout T1, T2, and T3 (TRd = 15.42, Δdf = 2, p < .01). It was moderate during T1 (β = .39, p < .001), increased during T2 (β = .55, p < .001), then decreased during T3 (β = .44, p < .001). The other paths were invariant (TRd = 1.05, Δdf = 2, p = .59 for COVID-related distress and emotion-focused coping; TRd = 3.42, Δdf = 2, p = .18 for COVID-related distress and compulsive buying).
As shown in Fig 4 , there was a weak but significant path between COVID-related distress and compulsive buying. The relationship between COVID-related distress and compulsive buying was mediated by emotion-focused coping across the three time periods—that is, higher COVID-related distress was associated with a more frequent use of emotion-focused coping, which in turn was associated with more compulsive buying online. The standardized indirect effect was 0.19 (p < .001), 0.25 (p < .001), and 0.20 (p < .001) for T1, T2, and T3 respectively. Proportion mediated for COVID-related distress, emotion-focused coping, and online compulsive buying was 59.45% across the three time periods. Subjective SES, age, and gender were controlled for in the model, where higher subjective SES was significantly associated with more compulsive buying (β T1 = .355, p < .001, β T2 = .424, p < .001, β T3 = .387 p < .001). The total explained variance of online compulsive buying was 39.7% (p < .001) in T1, 64.1% (p < .001) in T2, and 51.6% (p < .001) in T3.
10.1371/journal.pone.0274458.g004
Fig 4
The multi-group mediation model with online compulsive buying as the outcome and its standardized path coefficients for T1, T2, and T3, respectively (Model 2).
Note. *p < .01. **p < .001. (χ² = 1119.377 [χ² T1 = 482.380, χ² T2 = 297.892 , χ² T3 = 339.105] , df = 420, RMSEA = 0.059 [0.055–0.064], SRMR = 0.049, CFI = 0.951, TLI = 0.947).
We repeated these analyses with offline compulsive buying as the outcome measure, obtaining almost identical results (time invariance). The multigroup mediation model for offline compulsive buying and its fit indices are available in S1 Fig in the Supporting Information.
Discussion
Our findings suggest that the relationship between stress and compulsive buying is mediated by maladaptive coping during the COVID-19 pandemic. Furthermore, our results indicate that interindividual differences in the tendency to employ maladaptive emotion-focused coping strategies such as self-blame, counterfactual thinking, or becoming very upset in response to COVID-related distress may further exacerbate stress and give rise to maladaptive behavioral outcomes such as compulsive buying. Compulsive buying may be a source of stress itself, which may require further coping efforts on top of the already stressful pandemic, resulting in a vicious circle [ 46 ]. This highlights that the response to a stressor may play a significant role in adaptation [ 7 , 8 ] and underscores the importance of psychoeducation/counselling on adaptive ways of coping, especially during challenging times such as the COVID-19 pandemic.
The mediation model was not invariant across the three time periods, indicating that the strength of the associations between COVID-related distress, emotion-focused coping, and compulsive buying was not the same across T1, T2, and T3. Specifically, the relationship between emotion-focused coping as the mediator and compulsive buying as the outcome was moderate during T1, became stronger during T2, and then decreased again during T3. One reason for this may be that buying tendencies have varied throughout the pandemic—for instance, people initially hoarded certain products either in an attempt to avoid catching the virus (e.g., hand sanitizers, face masks) or due to panic about potential supply shortages (e.g., food, toilet paper) [ 47 ]. However, the connection between COVID-related distress, emotion-focused coping, and compulsive buying persisted beyond the first period of panic. Moreover, the associations between emotion-focused coping and compulsive buying grew stronger in T2, when these initial buying behaviors became less salient. This suggests that there is an emotional aspect to compulsive buying, rather than it being hoarding behavior or simply overbuying. In fact, stress and buying tendencies fluctuated, but were generally on the rise during the time of data collection [ 30 ], while emotion-focused coping increased from T1 to T2 and stagnated for the rest of our sampling period. The increased tendency to employ an emotion-focused, often self-blaming coping strategy may reflect the fact that by this time (day 80 –day 141, June 2 to July 17, 2020) people had become exhausted by the demands and the unfamiliar circumstances generated by the pandemic. Compulsive buying played the strongest role in terms of coping with distress during the period when distress was on the rise (T2), as indicated by the largest proportion of variance explained within this period and the fact that the path between emotional coping and compulsive buying was strongest during this time.
Another interesting finding that suggests a potential change in coping profiles is that the correlations between task-focused and emotion-focused coping changed substantially across the three time periods. Initially, there was a moderate negative association, which decreased over time and then disappeared. This may suggest that, immediately after the outbreak of the pandemic, people tended to commit to one or another coping strategy, whereas these boundaries became less articulated as time passed. Task-focused coping was not associated with COVID-related distress, which may reflect the notion that perceived stress regarding the pandemic fosters emotion-focused rather than action-oriented, task-focused coping strategies, or that task-focused coping strategies may be more effective in reducing COVID-related stress. However, we need to be cautious when interpreting the changes that occurred over time. Given that our study has a cross-sectional design, we cannot draw conclusions about intraindividual changes in coping styles. Specifically, based on our results it is not possible to determine whether the observed differences between T1, T2, and T3 lie in the differences in the recruited samples or truly reflect different phases in people’s reactions to the unfolding pandemic. Our results raise the question of whether the pandemic, as a persistent stressor, fosters changes in coping profiles over time, but this question can only be answered by a longitudinal study.
In the same vein, one might argue for a reversed model, assuming that maladaptive emotion-focused coping may be associated with elevated stress, which, in turn, may lead to more compulsive buying. We tested this alternative multigroup model, in which emotion-focused coping was the predictor and COVID-related distress was the mediator, for both online and offline compulsive buying. Although these alternative models also demonstrated excellent model fit, we argue that our proposed model makes more sense conceptually. Emotion-focused coping was assessed as a trait-like person-level variable, whereas stress was assessed as a state-level variable, as we were interested in the time-varying effects of stress throughout the initial phase of the pandemic. We therefore posit that it is more plausible for a state-like variable (COVID-related distress) to trigger a trait-like variable (emotion-focused coping), which in turn affects the behavioral outcome, than vice versa.
The strengths of this research include the fast reaction to the pandemic (data collection started two weeks after the outbreak of the pandemic in the U.S.); data collection that was spread over time, which enabled us to compare our mediation model in different phases of the pandemic; and the large sample size. However, certain limitations must be taken into consideration. We collected the self-reported data cross-sectionally only, thus the data are prone to response bias and do not allow us to draw causal conclusions. Furthermore, we assessed general coping styles in a stressful situation rather than the actual coping strategies employed in response to the pandemic. Although it is reasonable to assume that general coping styles are likely to be adopted across different stressful situations, the fact that we did not assess other coping strategies used in response to COVID-related distress is a limitation of our study. Another limitation is that compulsive buying itself is sometimes considered as a means of avoidance coping [ 48 ], making it hard to distinguish our outcome variable from our mediators.
In summary, our results indicate that maladaptive emotion-focused coping, including self-blame and counterfactual thinking, mediated the relationship between COVID-related distress and compulsive buying behavior. In other words, interindividual differences in coping may play a substantial role in the occurrence of maladaptive behavioral outcomes such as compulsive buying. At the time of the COVID-19 pandemic, when various coping strategies (e.g., social support, going out to engage in leisure activities) are hampered, people may seek other activities to disengage from COVID-related distress. Compulsive buying may represent one such attempt, especially when maladaptive emotion-focused coping is used. It is therefore important to promote active coping strategies that are more adaptive in terms of stress reduction via psychoeducation, especially during challenging times such as the COVID-19 pandemic. Such strategies might include finding new hobbies or interests that can be practiced indoors, or performing socially distanced outdoor activities (e.g., going biking or taking walks in the neighborhood) as a means of distraction from COVID-related distress.
Supporting information
S1 Table
Gender differences of the assessed measures.
Note. CISS task = Coping Inventory for Stressful Situations task-focused coping subscale; CISS emot. = Coping Inventory for Stressful Situations emotion-focused coping subscale; COSS = Compulsive Online Shopping Scale; BSAS = Bergen Shopping Addiction Scale; SES = Subjective socio-economic status.
(DOCX)
S1 Fig
The multi-group mediation model with offline compulsive buying as the outcome and its standardized path coefficients for T1, T2 and T3, respectively.
Note. *p < .01. **p < .001. χ² = 955.107 [χ² T1 = 377.307, χ² T2 = 255.708, χ² T3 = 322.092], df = 420, RMSEA = 0.052 [0.048–0.056], SRMR = 0.044, CFI = 0.961, TLI = 0.960. Standardized indirect effect was 0.188 (p < 0.001), 0.253 (p < 0.001) and 0.218 (p < 0.001) for T1, T2 and T3 respectively. Proportion mediated for COVID-related distress, emotion-focused coping and offline compulsive buying was 57.5% across the three time periods. The total explained variance of online compulsive buying was 39.1% (p < .001) in T1, 66.3% (p < .001) in T2, and 53.2% (p < .001) in T3.
(DOCX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The haematopoietic system is highly complex, extremely active and very important in health and diseases. Two key components of the system are the haematopoietic stem cells (HSCs) and their bone marrow microenvironment. HSCs exist in a relatively quiescent state and are the source of all the differentiated blood cells. They execute long-term self-renewal and multi-lineage differentiation functions [1] , [2] . The bone marrow microenvironment is referred as the HSC niche, a microenvironment where different cell types and extracellular matrix molecules dictate stem cell self-renewal and progeny production [3] . It has been appreciated that bone marrow mesenchymal stem cells (BM-MSCs) provide a structural scaffold for haematopoiesis [4] . Besides reticular fibroblasts, macrophages, adipocytes and endothelial cells, osteoblasts are also part of the stromal cell support system in the bone marrow. Osteoblastic cells regulate the HSC niche [5] . HSC differentiation occurs in direct proximity to osteoblasts within the bone marrow cavity. HSCs reside in medullar niches, mainly in the endosteum where osteoblasts and stromal cells provide HSCs with signals for maintaining the homeostatic quiescent state [5] , [6] .
The polycomb protein Bmi1 ( B lymphoma M o-MLV i nsertion 1) was identified originally as an oncogenic partner of c-Myc in murine lymphomagenesis. Bmi1 is required for self-renewal and postnatal maintenance of HSCs [7] and also important for neural stem cells from the central and peripheral nervous systems [8] , [9] . Mice lacking Bmi1 display defects in haematopoiesis and development of the central and peripheral nervous systems [10] . It has been reported that progressively impaired haematopoiesis in the bone marrow of Bmi1 -/- mice results in decreased cell numbers and replacement of large areas of haematopoiesis in the bone marrow by adipocytes [10] . Our previous studies indicate that Bmi1 maintains self-renewal of BM-MSCs by inhibiting the expression of p27, p16, and p19 and alters the cell fate of BM-MSCs by enhancing osteoblast differentiation and inhibiting adipocyte differentiation at least in part by stimulating expression of the lsyine deacetylase Sirt1 [11] . These results indicate that Bmi1 deficiency results in both defects in haematopoiesis and osteoblastic bone formation; however, it is unclear whether defects in haematopoiesis caused by Bmi1 deficiency are associated with impaired bone marrow microenvironment for haematopoiesis.
Parathyroid hormone (PTH) is a peptide comprised of 84 amino acids and is the main regulator of calcium and phosphate homeostasis. It is secreted from the cells of the parathyroid glands and acts through a class B G-protein–coupled receptor, PTH receptor (PTHR). In fact, the N-terminal 34 amino acids of mature PTH are sufficient for activation of signaling through PTHR [12] . PTH1-34, a recombinant human parathyroid hormone analog containing the 34 residues, is currently used as an anabolic drug to treat osteoporosis [13] . PTH increases osteoblast production rate and inhibits apoptosis of osteoblasts, thereby leading to a rapid increase in skeletal mass as well as improvement of bone micro-architecture and strength [14] . Our previous study demonstrated that PTH1-34 administration significantly increased cortical and trabecular bone mass with augmented osteoblast number and activity [15] . Furthermore, available evidence also suggests that PTH stimulates haematopoiesis [5] , [16] – [18] . Osteoblasts in transgenic mice expressing a constitutively active form of PTHR only in the osteoblast lineage support accumulation of twice more HSCs than normal [5] . PTH not only exerts anabolic action by stimulating osteoblastic bone formation, but also upregulates haematopoiesis by improving bone marrow microenvironment. These findings raise the important question whether PTH administration is able to rescue haematopoietic defects from Bmi1 deficiency though improving the bone marrow microenvironment. To answer this question, Bmi1 -/- mice were treated with PTH1-34 and compared with vehicle-treated Bmi1 -/- and wild-type mice. The results indicate the administration improves the microenvironment and rescues haematopoietic defects in Bmi1 -null mice, thereby revealing a potential value of PTH1-34, an anabolic drug for osteoporosis, for repairing haematopoietic defects.
Materials and Methods
Mice and genotyping
Bmi1 -/- mice (129Ola/FVB/N hybrid background), kindly provided by Professor Anton Berns, The Netherlands Cancer Institute, [10] had been backcrossed 10 to 12 times on the C57BL/6J background. Genotypes of the mutant mice were determined by PCR analysis as described previously [8] . The 1-week-old wild type and Bmi1 -/- mice received daily injections of vehicle or of PTH1-34 (80 μg/kg) subcutaneously for 3 weeks. For DAPT (N-[N-(3,5-difluorophenacetyl)-l-alanyl]-S-phenylglycine t-butyl ester) blocking assay, the 4-week-old Bmi1 -/- mice received daily injections of PTH1-34 (80 μg/kg) subcutaneously or injections of PTH1-34 plus injections of DAPT (10 μmol/kg) intraperitoneally for 2 days, the same age wild type and Bmi1 -/- mice received daily injections of vehicle as controls. All animal experiments were carried out in compliance with and approval by the Institutional Animal Care and Use Committee of Nanjing Medical University (Approval ID 2008-00318).
Skeletal radiography
Femurs were removed and dissected free of soft tissue. Contact radiographs were taken using a Faxitron Model 805 radiographic inspection system (Faxitron Contact, Faxitron, Germany; 22 kV and 4-minute exposure time). X-Omat TL film (Eastman Kodak Co., Rochester, NY, USA) was used and processed routinely.
Micro-computed tomography (µCT)
Femurs obtained from 2-week-old mice were dissected free of soft tissue, fixed overnight in 70% ethanol, and analyzed by µCT with a SkyScan 1072 scanner and associated analysis software (SkyScan, Antwerp, Belgium) as described previously [19] . Briefly, image acquisition was performed at 100 kV and 98mA with a 0.9-degree rotation between frames. During scanning, the samples were enclosed in tightly fitting plastic wrap to prevent movement and dehydration. Thresholding was applied to the images to segment the bone from the background. 2D images were used to generate 3D renderings using the 3D Creator software supplied with the instrument. The resolution of the µCT images is 18.2 μm.
Histology
Tibiae were removed and fixed in PLP fixative (2% paraformaldehyde containing 0.075M lysine and 0.01M sodium periodate) overnight at 4°C and processed histologically as described previously [20] . Proximal ends of tibiae were decalcified in EDTA glycerol solution for 5 to 7 days at 4°C. Decalcified tibiae were dehydrated and embedded in paraffin, after which 5 μm sections were cut on a rotary microtome. The sections were stained with hematoxylin and eosin (H&E) or histochemically for total collagen [21] or alkaline phosphatase activity (ALP) [22] , or immunohistochemically as described below.
Immunohistochemical staining
Osterix, type I collagen, osteopontin, PTHR, Jagged1 and Notch1 were determined by immunohistochemistry as described previously [20] , [23] . Polyclonal rabbit Anti-osterix (Abcam, USA), polyclonal goat anti-type I collagen (Santa Cruz, USA), rabbit anti-mouse osteopontin (Millipore, USA), monoclonal anti-PTHR (Millipore, USA), rabbit anti-mouse Jagged1 (Santa Cruz, USA) and rabbit anti-mouse activated Notch1 polyclonal antibody (Unconjugated, Abcam, USA) were employed.
Quantitative real-time PCR
RNA was isolated from mouse long bones using Trizol reagent (Invitrogen, Inc., Carlsbad, CA, USA) according to the manufacturer's protocol. Reverse-transcription reactions were performed using the SuperScript First-Strand Synthesis System (Invitrogen), as described previously [24] . Real-time PCR was performed using a LightCycler system (Roche, Indianapolis, IN, USA) as described previously [25] . The conditions were 2 μL of LightCycler DNA master SYBR Green I (Roche), 0.25 μM of each 50 and 30 primer and 2 μL of sample and/or H 2 O to a final volume of 20 μL. The MgCl 2 concentration was adjusted to 3 mM. Samples were amplified for 35 cycles with a temperature transition rate of 20°C/s for all three steps, which were denaturation at 95°C for 10 seconds, annealing for 5 seconds and extension at 72°C for 20 seconds. SYBR green fluorescence was measured to determine the amount of double-stranded DNA. To discriminate specific from nonspecific cDNA products, a melting curve was obtained at the end of each run. Products were denatured at 95°C for 3 seconds, and then the temperature was decreased to 58°C for 15 seconds and raised slowly from 58 to 95°C using a temperature transition rate of 0.1°C/s. To determine the number of copies of target DNA in the samples, purified PCR fragments of known concentration were serially diluted and served as external standards for each experiment. Glyceraldehyde-3-phosphate dehydrogenase ( Gapdh ) was used as the internal control for each reaction. The relative amount of mRNA was normalized to Gapdh mRNA. The primer sequences used for the real-time PCR were as described [25] , [26] .
Western blot analysis
Proteins were extracted from long bones or cells and quantitated by a kit (Bio-Rad, Mississauga, Ontario, Canada). Protein samples of 30 μg were fractionated by SDS-PAGE and transferred to nitrocellulose membranes. Immunoblotting was carried out as described previously [25] using antibodies against Runx2 (MBL International, Woburn, MA), peroxisome proliferator-activated receptor γ (Ppar-γ, E-8, Santa Cruz, CA, USA), PTHR (clone 3D1.1, Millipore), insulin-like growth factor 1 (IGF-1, clone Sm1.2, Millipore), Jagged1 (Santa Cruz, USA), activated Notch1 (Abcam, USA) and β-tubulin (Santa Cruz, CA, USA). Bands were visualized using enhanced chemiluminescence (ECL, Amersham) and quantitated by Scion Image Beta 4.02 (Scion Corporation, Bethesda, MD, USA).
Complete blood count (CBC)
Each mouse was bled by retro-orbital puncture for blood cell counts. Blood (20 μl) was collected and mixed with 180 μL Cell-Dyn buffer immediately. Complete blood count was analyzed with a Cell Dyn 3700 counter (Abbott Laboratories, Ill, USA). Two blood samples of each mouse were collected for CBC analysis. The numbers of neutrophils and platelets from all animals were averaged, and the data are presented as means ± standard deviations.
Flow cytometry
For analysis of HSCs, BM cells were stained with PE-conjugated anti-Sca1 (BioLegend), PE-Cy5.5-conjugated anti-c-Kit (eBioscience), and Alexa Fluor 488-conjugated Mouse Lineage Mixture Antibodies (Invitrogen). The HSCs were defined as Sca-1 + c-Kit + Lin - and the HPCs as Sca-1 + c-Kit + Lin + . All analyses were performed on a FACSCalibur (BD Biosciences).
Computer-assisted image analysis
After HE staining or histochemical or immunohistochemical staining of sections from six mice of each genotype, images of fields were photographed with a Sony digital camera. Images of micrographs from single sections were digitally recorded using a rectangular template, and recordings were processed and analyzed using Northern Eclipse image analysis software as described previously [23] , [27] .
Statistical analysis
Data from image analysis are presented as mean ± s.e.m. Statistical comparisons were performed by use of a two-way ANOVA, with P <0.05 considered to be significant.
Results
Effect of PTH1-34 on the length of long bones and trabecular bone volume in Bmi1 -/- mice
To determine whether skeletal growth retardation and osteopenic phenotype caused by Bmi1 deficiency were improved by PTH administration, we treated 1 week old Bmi1 -/- and wild-type mice subcutaneously with vehicle or PTH1-34 at 80 μg/kg per day for 3 weeks. Long bones from vehicle-treated wild-type and Bmi1 -/- mice and PTH1-34-treated Bmi1 -/- mice were analyzed at 4 weeks of ages by radiography and μCT. The lengths of tibiae were shorter in vehicle-treated Bmi1 -/- mice than in their wild-type littermates ( Figs. 1A–B ). Radiolucency was greater in Bmi1 -/- mice relative to wild-type mice ( Fig. 1A ). From 3D reconstructed longitudinal sections of the proximal ends of tibiae, it was clear that epiphyses were smaller and trabecular bone volumes were lower in Bmi1 -/- mice than the wild-type mice ( Fig. 1C ). The length of tibiae was not increased, whereas the trabecular bone volume increased significantly in Bmi1 -/- mice by PTH1-34 administration, but had not reached the normal levels as vehicle-treated wild-type mice ( Figs. 1A–C ). Consistent with μCT analysis, histological analysis demonstrated that trabecular bone volume was reduced significantly at 4 weeks of age in vehicle-treated Bmi1 -/- mice when compared with their wild-type littermates ( Figs. 1D–E ). The volume increased significantly in Bmi1 -/- mice upon PTH1-34 administration, but had not reached normal levels as vehicle-treated wild-type mice ( Figs. 1D–E ). These results demonstrated that osteoporotic phenotypes caused Bmi1 deficiency was reversed partially by PTH1-34 administration.
10.1371/journal.pone.0093864.g001 Figure 1
Effect of PTH1-34 on the length of long bones and trabecular bone volume in Bmi-1 -/- mice.
Representative radiographs, (B) quantitation of the length of tibiae, (C) 3-dimensional reconstructed longitudinal sections of micro-CT scanning images and (D) micrographs of paraffin sections of the tibiae stained with Siries Red for total collagen from 4-week-old vehicle-treated wild-type (WT) and Bmi-1 -/- mice (KO) and PTH1-34-treated Bmi-1 -/- mice (KO+PTH), magnification, ×50. (E) Quantitation of trabecular bone volume relative to tissue volume (BV/TV, %) in metaphyseal regions. For each genotype, n = 6; *: p <0.05, **: p <0.01, ***: p <0.001, compared to vehicle-treated WT mice; ###: p <0.001 compared to vehicle-treated Bmi-1 -/- mice.
Effect of PTH1-34 on osteoblast bone formation in Bmi1 -/- mice
To determine whether the increased trabecular bone volume in Bmi1 -/- mice by PTH1-34 administration was associated with the improvement of osteoblastic bone formation, the number of osteoblasts, ALP activity in osteoblasts, deposition of type I collagen and osteopontin in the bone matrix and expression of osterix and PTHR in osteoblasts were examined by HE staining, histochemical staining for ALP and immunostaining for type I collagen, osteopontin, osterix and PTHR. At 4 weeks of age, the osteoblast number ( Figs. 2A and B ), ALP-positive areas ( Figs. 2C and D ), type I collagen ( Figs. 2E and F ), osteopontin ( Figs. 2G and H ), osterix ( Figs. 2I and J ) and PTHR ( Figs. 2K and L ) were reduced significantly in vehicle-treated Bmi1 -/- mice compared with their wild-type littermates. These parameters all improved significantly in Bmi1 -/- mice upon PTH1-34 administration, but did not reach the levels comparable to their wild-type littermates ( Fig. 2 ). It was noted that PTHR was localized in osteoblasts and stromal cells in bone marrow, but not in bone marrow haematopoietic cells ( Fig. 2K ). We also examined expression of genes and proteins important for bone formation. RNA and proteins were isolated from long bones for real-time RT-PCR and Western blots. The results showed that the transcript levels of ALP and osteocalcin and the protein levels of Runx2, PTHR and IGF1 were all down-regulated significantly in Bmi1 -/- mice compared with their wild-type littermates ( Figs. 3A–F ). Importantly, PTH1-34 administration upregulated the levels of the transcripts and proteins in Bmi1 -/- mice although not to the normal levels ( Figs. 3A–F ), indicating that the defects in osteoblastic bone formation caused by Bmi1 deficiency were partially repaired by PTH1-34 administration.
10.1371/journal.pone.0093864.g002 Figure 2
Effect of PTH1-34 on osteoblastic bone formation in Bmi-1 -/- mice.
Representative micrographs of paraffin-embedded sections for tibial metaphyseal regions from 4-week old vehicle-treated wild-type (WT) and Bmi-1 -/- mice (KO) and PTH1-34 treated Bmi-1 -/- mice (KO+PTH) stained (A) histologically with hematoxylin & eosin (HE, ×400), (C) histochemically for alkaline phosphatase (ALP, ×200), immunohistochemically for (E) type I collagen (ColI, ×200), (G) osteopontin (OPN, ×200), (I) osterix (x400) and (K) PTHR (x400). (B) Osteoblast counts (#/mm 2 ), (D) ALP-positive areas, (F) type I collagen- or (H) OPN- or (J) osterix- or (L) PTHR-immunopositive areas were measured by computer-assisted image analysis. For each genotype, N = 6; *: p <0.05, **: p <0.01, ***: p <0.001, compared to vehicle-treated WT mice; #: p <0.05, ##: p <0.01, ###: p <0.001 compared to vehicle-treated Bmi-1 -/- mice.
10.1371/journal.pone.0093864.g003 Figure 3
Effect of PTH1-34 on expression of markers for osteoblastic bone formation in Bmi-1 -/- mice.
(A–B) Real-time RT–PCR was performed on humerus extracts from 4-week-old vehicle-treated wild-type (WT) and Bmi-1 -/- mice (KO) and PTH1-34 treated Bmi-1 -/- mice (KO+PTH) for determining the expression of (A) alkaline phosphatase (ALP) and (B) osteocalcin. The expression is calculated as a ratio to the GAPDH mRNA level and shown relative to the levels in vehicle-treated WT mice. (C) Western blots of femur extracts from 4-week-old vehicle-treated WT and Bmi-1 -/- mice and PTH1-34-treated Bmi-1 -/- mice for expression of Runx2, PTHR and IGF-1. β-actin was used as loading control for Western blots. (D-F) Runx2, PTHR and IGF-1 protein levels relative to the β-actin level were assessed by densitometric analysis and presented relative to the levels in vehicle-treated WT mice. For each genotype, n = 6; *: p <0.05, **: p <0.01, ***: p <0.001, compared to vehicle-treated WT mice; #: p <0.05, ##: p <0.01, ###: p <0.001 compared to vehicle-treated Bmi-1 -/- mice.
Effect of PTH1-34 on bone marrow haematopoietic cells and adipocytes in Bmi1 -/- mice
To assess directly whether haematopoietic defects caused by Bmi1 deficiency could be improved by PTH administration, the numbers of bone marrow haematopoietic cells and adipocytes were counted on HE-stained sections of diaphyseal regions in long bones. At 4 weeks of age, the number of bone marrow haematopoietic cells decreased significantly, while the number of bone marrow adipocytes increased dramatically in 4-week old Bmi-1 -/- mice when compared with their wild-type littermates ( Figs. 4A–C ). When PTH1-34-treated Bmi-1 -/- mice were compared with the vehicle-treated counterparts, the number of bone marrow haematopoietic cells increased significantly, while the number of bone marrow adipocytes was reduced significantly ( Figs. 4A–C ). We also assessed whether the alterations of bone marrow adipocytes were associated with PPARγ expression changes. Proteins were isolated from long bones and Western blots were performed. Results showed that the expression level of PPARγ was upregulated significantly in Bmi1 -/- mice compared to their wild-type littermates ( Figs. 4D and E ). PTH1-34 administration significantly downregulated PPARγ expression in Bmi1 -/- mice ( Figs. 4D–E ). To determine whether alterations of bone marrow haematopoietic cells in Bmi1 -/- mice by PTH administration were associated with effects on the HSC and HPC populations, the fraction and numbers of HSCs and HPCs in the bone marrow were analyzed in 4-week old mice by fluorescence-activated flow cytometry. Results showed that the fractions of HSCs (Sca-1 + c-Kit + Lin - ) and HPCs (Sca-1 + c-Kit + Lin + ) were reduced insignificantly ( Figs. 4F–H ), however, total numbers in the femur bone marrow were reduced significantly in Bmi1 -/- mice when compared to their wild-type littermates ( Figs. 4I–J ). The fractions of HSCs and HPCs, as well as the total numbers of both HSCs and HPCs, increased significantly or remarkably in the femur bone marrow of PTH1-34-treated Bmi1 -/- mice ( Figs. 4F–J ). These results indicated that PTH administration reverses the low haematopoietic cell number (including HSCs and HPCs) and the high adipocyte count in the bone marrow due to Bmi1 deficiency.
10.1371/journal.pone.0093864.g004 Figure 4
Effect of PTH1-34 on the bone marrow cellularity in Bmi-1 -/- mice.
(A) Representative micrographs of paraffin-embedded sections of tibial diaphyseal regions from 4-week-old vehicle-treated wild-type (WT) and Bmi-1 -/- mice (KO) and PTH1-34 treated Bmi-1 -/- mice (KO+PTH) stained with hematoxylin and eosin (HE, ×400). (B) The one marrow cell number and (C) adipocyte number relative to tissue area were measured by computer-assisted image analysis. (D) Western blots of femur extracts from 4-week-old vehicle-treated WT and Bmi-1 -/- mice and PTH1-34-treated Bmi-1 -/- mice for determination of PPARγ expression. β-actin was used as the loading control. (E) The PPARγ level relative to the β-actin level was assessed by densitometric analysis and shown relative to the levels in the vehicle-treated WT mice. (F) Representative graphs of flow cytometry analysis for hematopoietic stem cells (HSCs) and hematopoietic progenitor cells (HPCs) in the bone marrows from 4-week-old vehicle-treated wild-type (WT) and Bmi-1 -/- mice (KO) and PTH1-34- treated Bmi-1 -/- mice (KO+PTH). (G-H) Fractions of Sca-1 + c-kit + Lin − HSCs and Sca-1 + c-kit + Lin + HPCs in the bone marrows. (I–J) The numbers of Sca-1 + c-kit + Lin - HSCs and Sca-1 + c-kit + Lin + HPCs in each femur were assessed and presented relative to the levels in the vehicle-treated WT mice. For each genotype, n = 6; *: p <0.05, **: p <0.01, ***: p <0.001, compared to vehicle-treated WT mice; #: p <0.05, ##: p <0.01, ###: p <0.001 compared to vehicle-treated Bmi-1 -/- mice.
Effect of PTH1-34 on the peripheral blood cellularity in Bmi1 -/- mice
To assess whether Bmi1 deficiency resulted in abnormalities of peripheral blood cells and PTH administration could rescue their possible abnormalities, we analyzed peripheral blood from 4-week old vehicle-treated wild-type and Bmi1 -/- mice and PTH1-34-treated Bmi1 -/- mice by use of a hematological analyzer. Specifically, the number of white blood cells, lymphocytes, granulocytes, red blood cells, platelets and the fraction of lymphocytes were evaluated. At 4 weeks of age, the number of white blood cells and lymphocytes in peripheral blood decreased significantly in vehicle-treated Bmi-1 -/- mice when compared to the wild-type littermates ( Fig. 5A–B ). PTH1-34 administration partially reversed the deficits in Bmi-1 -/- mice, but not to the normal levels as wild-type mice ( Figs. 5A–B ). The numbers of granulocytes in peripheral blood was slightly reduced insignificantly in vehicle-treated Bmi-1 -/- mice when compared to the wild-type littermates ( Fig. 5C ) and increased dramatically in PTH1-34-treated Bmi-1 -/- mice ( Fig. 5C ). The numbers of red blood cells and platelets in peripheral blood were not altered significantly in both vehicle-treated and PTH1-34-treated Bmi1 -/- mice ( Figs. 5D–E ). The fraction of lymphocytes decreased significantly in both vehicle-treated and PTH1-34-treated Bmi1 -/- mice ( Fig. 5F ). These results demonstrated that the numbers of white blood cells and lymphocytes and the fraction of lymphocytes decreased significantly in Bmi1 -deficient mice and that PTH administration increased the numbers of these cells significantly but had minimal impact on the fraction of lymphocytes.
10.1371/journal.pone.0093864.g005 Figure 5
Effect of PTH1-34 on the peripheral blood cellularity in Bmi-1 -/- mice.
The peripheral blood from 4-week-old vehicle-treated wild-type (WT) and Bmi-1 -/- mice (KO) and PTH1-34 treated Bmi-1 -/- mice (KO+PTH) were analyzed by a hematological analyzer, to determine the number of white blood cells (A), lymphocytes (B), granulocytes (C), red blood cells (D), platelets (E) and the fraction of lymphocytes (F). For each genotype, n = 6; *: p <0.05, ***: p <0.001, compared to vehicle-treated WT mice; ##: p <0.01, ###: p <0.001 compared to vehicle-treated Bmi-1 -/- mice.
Effect of PTH1-34 on Notch signaling molecules in Bmi1 -/- mice
To gain insights into the underlying mechanisms, we investigated whether the improvement of haematopoietic defects occurred in Bmi1 deficient mice by PTH administration is associated with activated Notch signaling, which is crucial for haematopoietic stem cells [28] . Expression of the Notch ligand Jagged1 and the Notch intracellular domain (NICD) was thus examined by immunohistochemistry and Western blots. At 4 weeks of age, the number of Jagged1-positive cells, the percentage of Notch intracellular domain (NICD)-positive bone marrow cells and protein expression levels of Jagged1 and NICD in bone tissue were decreased significantly in vehicle-treated Bmi1 -/- mice when compared to the wild-type counterparts ( Figs. 6A–G ). Administration of PTH1-34 increased the levels significantly in Bmi1 -/- mice ( Figs. 6A–G ). To further demonstrate whether PTH administration activated Notch signaling, the 4-week-old Bmi1 -/- mice were injected daily with PTH1-34 alone or with both PTH1-34 and DAPT, a Notch inhibitor, for 2 days, RNAs were isolated from mouse long bones, the expression of Notch1 and Jagged1 was examined at mRNA levels by real-time RT-PCR. The gene expression levels of Notch1 and Jagged1 were down-regulated in vehicle-treated Bmi1 -/- mice and up-regulated in PTH-treated Bmi1 -/- mice, however, the up-regulation of PTH on both Notch1 and Jagged1 was blocked by the DAPT administration. These results indicated that the Notch pathway is inhibited by Bmi1 deletion, but activated by PTH administration.
10.1371/journal.pone.0093864.g006 Figure 6
Effect of PTH1-34 on Notch signal pathway-related molecules in Bmi-1 -/- mice.
(A–B) Representative micrographs of paraffin-embedded sections of tibiae from 4-week-old vehicle-treated wild-type (WT) and Bmi-1 -/- mice (KO) and PTH1-34- treated Bmi-1 -/- mice (KO+PTH) stained immunohistochemically for the Notch ligand Jagged1 (A) and Notch intracellular domain (NICD, B), magnification, ×400. (C) The number of Jagged1-positive cells relative to bone surface (#/mm 2 ) and (D) the percentage of NICD-positive bone marrow cells were measured by computer-assisted image analysis. (E) Western blots were performed on the long bone extracts for expression of jagged1and NICD. β-actin was used as loading control for Western blots. (F) jagged1 and (G) NICD protein levels relative to the β-actin level were assessed by densitometric analysis and presented relative to the levels in vehicle-treated WT mice. (H) Real-time RT–PCR was performed on long bone extracts from vehicle-treated wild-type (WT) and Bmi-1 -/- mice (KO), PTH-treated Bmi-1 -/- mice (KO+PTH) and PTH and DAPT-treated Bmi-1 -/- mice (KO+PTH+PAPT) for determining the expression of Nortch1 and jagged1. The expression is calculated as a ratio to the GAPDH mRNA level and shown relative to the levels in vehicle-treated WT mice. For each genotype, n = 6; *: p <0.05, **: p <0.01, compared to vehicle-treated WT mice; #: p <0.05 compared to vehicle-treated Bmi-1 -/- mice.
Discussion
It was previously shown that defects in haematopoiesis in Bmi1 -null mice are due to severely impaired HSC self-renewal [7] , but it is unclear whether the defects are associated with an impaired bone marrow microenvironment. It was reported that PTH not only exerts bone anabolic action by stimulating osteoblastic bone formation, but also promotes haematopoiesis by improving the bone marrow microenvironment [5] , [16] – [18] . In the present study, we have investigated whether PTH administration rescues the haematopoietic defects caused by Bmi1 deficiency though improving the haematopoietic microenvironment. Our results demonstrate that administration of PTH1-34, a drug currently used for treating osteoporosis, partially rescues haematopoietic defects in Bmi1 -deficient mice by improving the bone marrow microenvironment. These results imply that the haematopoietic defects caused by Bmi1 deficiency are not only because of reduced HSC self-renewal, but also due to an impaired haematopoietic microenvironment.
Consistent with the previous findings [11] , we found here that compared to the wild-type counterparts, Bmi1 -deficient mice displayed i) defects in osteoblastic bone formation, including decreased trabecular bone volume, osteoblast number and ALP- or type I collagen-positive areas; ii) down-regulated ALP, osteocalcin and Runx2 expression in the bone; and iii) increased adipocytes and up-regulated PPARγ expression in the bone marrow. In addition, we found that Bmi1 deficiency resulted in down-regulated expression of the PTHR and the IGF-1 in the bone. Our results also demonstrated that PTHR was localized in osteoblasts and stromal cells in bone marrow, but not in bone marrow haematopoietic cells. PTHR gene was localized in osteoblasts demonstrated by in situ hybridization [29] . Previous studies indicated that PTHR is a crucial mediator of bone-forming action of PTH as targeted expression of the constitutively active PTHR led to increased osteoblast function in trabecular bone and at the endosteal surface of cortical bone [30] and IGF-1 is required for the anabolic effect of PTH on bone formation as PTH had little effects on IGF-1-null mice [31] . Thus, the bone anabolic action of Bmi1 on osteoblastic bone formation is also associated with the regulation of PTHR and IGF-1.
Some of our findings appear to be slightly different from those in a previous report [7] . The previous study examined the alterations of bone marrow haematopoietic cells and peripheral blood cells in 2-month old Bmi1 -/- and wild-type mice. The results showed that Bmi1 -/- mice have significant less HSC frequency in the bone marrow and an average 10-fold less total HSCs when the total numbers of bone marrow cells were taken into account [7] . It was also found that in the peripheral blood, there were a normal number of myeloid cells but a smaller number of lymphocytes [7] . We examined 1-month old mice. The results showed that Bmi1 -/- mice had insignificant fewer HSCs and HPCs in the bone marrow and 4-5-fold less total HSCs and HPCs when the total numbers of bone marrow cells were taken into account. Although the amount of granulocytes, red blood cells and platelets in the peripheral blood was not altered significantly, the numbers of lymphocytes and some other white blood cells were decreased dramatically in Bmi1 -deficient mice ( Fig. 5 ). The differences between this and the previous study could be due to that the severity of haematopoietic defects was lighter in 4-week-old Bmi1 -/- mice than 2-month-old Bmi1 -/- mice as previous report [7] . Related to this, we found that bone marrow cells were less, but adipocytes were more in 4-week old Bmi1 -/- mice than 2-week-old ones [11] . These results indicate that Bmi1 deficiency results in progressive defects in haematopoiesis as the mice develop and age.
Importantly, we found that PTH1-34 administration partially reversed premature osteoporosis occurred in Bmi1 -deficient mice. PTH1-34 administration increased trabecular bone volume, osteoblast number and activity, up-regulated ALP, osterix, osteocalcin, Runx2, PTHR and IGF1 expression in the bone, and reduced the number of adipocytes and PPARγ expression in the bone marrow. Runx2 is essential for the differentiation of osteoblasts from mesenchymal precursors [32] – [34] . Osterix, which acts downstream of Runx2, is a zinc-finger-containing transcription factor essential for embryonic osteoblast differentiation and bone formation [35] . PPARγ is a critical transcription factor involved in adipogenic differentiation [36] . As reported previously [11] , our results showed that the Runx2 and osterix levels were downregulated, whereas the PPARγ level was upregulated in the bone tissues from Bmi1 -/- mice. More importantly, PTH1-34 administration upregulated the Runx2 and osterix protein levels and downregulated the PPARγ levels in the bone tissues from Bmi1 -/- mice. In osteoblasts, the binding of PTH to PTHR activates adenyl cyclase and phospholipase, leading to formation of cAMP and a subsequent increase in intracellular calcium concentration as well as activation of PKC, promoting osteoblastic bone formation [30] . Intermittent treatment with PTH induces an increase in IGF-1 expression in the bone tissues from both mice and rats [37] , [38] . The anabolic effect of PTH depends on the expression of IGF-1 as PTH had no effect in IGF-1 null mice and was unable to induce important target genes for osteoblastic bone formation [31] . The current study not only demonstrated that PTHR and IGF-1 protein levels were downregulated in bone tissues from Bmi1 -/- mice, but also showed that both proteins were upregulated in the bone tissues upon PTH1-34 administration. Consequently, the administration increased osteoblastic bone formation at least in part by stimulating osteoblast differentiation and inhibiting adipocyte differentiation through PTHR and IGF-1.
Studies using genetically altered animal models that could activate or destroy osteoblastic cells suggest that osteoblasts contribute to the HSC niche [5] , [39] , [40] . On the other hand, PTH mediated activation of osteoblasts resulted in a significant expansion of the HSC pool and led to the realization that targeting cells of the osteoblastic lineage is a potential therapeutic approach to enhance stem cell-based therapies [5] , [16] – [18] . Base on that PTH1-34 administration increased osteoblastic bone formation in Bmi1 -deficient mice, we assessed whether PTH1-34 administration could also rescue defects in haematopoiesis caused by Bmi1 deficiency. Our results revealed that the total numbers of bone marrow cells, the fraction of Sca-1 + c-kit + Lin + HPCs relative to the total number of bone marrow cells, and the numbers of white blood cells and granulocytes in peripheral blood all increased significantly after PTH1-34 administration. Although the fraction of Sca-1 + c-kit + Lin - HSCs relative to the total number of bone marrow cells did not increased significantly, total HSCs were increased 2.7-fold, and total HPCs increased 3-fold when the total numbers of bone marrow cells were taken into account. This reversal of haematopoietic defects was consistent with the increased osteoblast number and activity in PTH-treated Bmi1 deficient mice. Furthermore, we also found that osteopontin-positive bone matrix area, the number of Jagged1-positive cells, percentage of NICD-positive bone marrow cells and protein expression levels of Jagged1 and NICD in bone tissue were decreased significantly in vehicle-treated Bmi1 -/- mice, and all increased significantly in Bmi1 -/- mice by the administration of PTH1-34, whereas the up-regulation of PTH on both Notch1 and Jagged1 gene expression was blocked by the Notch inhibitor DAPT administration. Osteopontin is an important component of the HSC niche in which it participates in HSC location and acts as a physiologic-negative regulator of HSC proliferation [41] . In mice in which the PTHR was activated in osteoblastic cells only, osteoblastic cells were increased in number and produced high levels of Jagged1, the activated NICD was increased in the HSC fraction in vivo, and Notch inactivation by DAPT blocked HSC expansion in vitro [5] . Results from the present study indicate that PTH administration partially rescues haematopoietic defects in Bmi1 deficient mice by improving haematopoietic microenvironment and activating the Notch pathway.
In conclusion, this study shows that PTH administration increased osteoblastic bone formation and partially repaired the haematopoietic defects in Bmi1 -deficient mice. The results indicate that haematopoietic defects caused by Bmi1 deficiency are not only because of reduced HSC self-renewal, but also because of impairment in the haematopoietic microenvironment. This study also reveals a potential value of PTH1-34, an anabolic drug for osteoporosis, for repairing haematopoietic deficiency.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
In vitro differentiation of specific cell types from human pluripotent stem cells (hPSCs) allows for molecular and functional analysis of cells that are otherwise inaccessible. This holds special promise in neurodegenerative diseases such as amyotrophic lateral sclerosis (ALS), where ethical and technical constraints prevent access to human spinal motor neurons [1] . Using protocols based on normal developmental pathways, it has proven possible to generate spinal motor neurons from both mouse and human embryonic stem cells (ESCs) [2] – [6] . These are an important source of new mechanistic insights into the developmental requirements of wildtype motor neurons in both species. Moreover, successful specification of motor neurons from human induced pluripotent stem cells (hiPSCs) has opened novel avenues for mechanistic analysis of neuronal cell death and drug testing in motor neuron disease models [1] , [4] – [8] . Yet our knowledge of the survival requirements of human motor neurons remains limited.
Cultured motor neurons from rodent embryos served as the basis for identification of the neurotrophic factors responsible for keeping motor neurons alive during development [9] – [11] and the same factors significantly retard motor neuron death in animal models of ALS [12] . In parallel, motor neurons cultured from mouse models of ALS shed light on the mechanisms underlying neurodegeneration [13] . All these discoveries required the purification of motor neurons from the complex environment of the spinal cord. This approach allowed for identification of factors that act directly on motor neurons, significantly facilitated direct quantification of motor neuron survival, and opened the door to biochemical studies that would not have been possible in mixed cultures. Although this might be considered a reductionist approach, conclusions about both survival factors and cell death mechanisms were subsequently validated in vivo [14] – [21] , demonstrating that the advantages of motor neuron purification outweigh concerns about the artificial nature of the assay. It is therefore important to extend such approaches to human motor neurons. However, standard protocols for hPSC differentiation generate mixed populations of spinal neurons of which motor neurons constitute a minority, and to date survival of purified motor neurons has necessitated generally co-culture with other cell types [22] – [25] . There is consequently a need for a robust survival assay based on purified human motor neurons.
Another challenge is that absolute numbers of motor neurons generated from hESC/hiPSCs by standard procedures are relatively low. During embryonic development in rodents, motor neurons are produced from a short-lived pool of committed ventral spinal progenitors expressing OLIG2, which are rapidly exhausted or converted to oligodendroglial progenitors [26] , [27] . However, in contrast to mouse motor neurons, which are produced during a brief period between embryonic days 9 and 12, the period of human motor neuron generation spans approximately twenty days [28] , [29] . This raises the possibility that agents that enhance proliferation of motor neuron progenitors might be used to increase the yield of human motor neurons in culture.
Here we have developed techniques that allow us both to amplify stem cell-derived motor neurons and to perform survival assays in the absence of other cell types. We first report that there is indeed significant ongoing motor neuron generation in cultures of differentiated hESCs. To exploit this so as to increase yield, we therefore screened for compounds that increase the number of motor neurons when applied over this period. We report that the ROCK inhibitor Y-27632 stimulates the proliferation of OLIG2-expressing progenitors, and increases the yield of motor neurons up to four-fold. Using amplified motor neurons from the Hb9::GFP hESC line, we next defined conditions for a robust survival assay using FACS-sorted motor neurons, and used it to demonstrate potent activity for three known neurotrophic factors as well as Y-27632 itself. These approaches should be of general interest for the preparation of human motor neurons on a large scale and for functional and biochemical studies of molecular processes controlling motor neuron genesis, survival and degeneration.
Results
Ongoing motor neuron generation in cultures of differentiated hESCs
To determine whether hESCs differentiated in vitro to a mixed spinal cord identity exhibit prolonged motor neurogenesis as in the fetal human spinal cord, we first examined changes in numbers of hESC-derived motor neurons (hESC-MNs) in mixed spinal cultures over a 15-day period using an hESC reporter line that expresses green fluorescent protein (GFP) under the control of the motor neuron-specific murine homeobox gene 9 (Hb9) promoter [23] . We and others previously showed using a range of other markers and functional assays that GFP-positive neurons generated from this line possess many properties of postmitotic motor neurons [6] , [23] , [30] . Motor neurons were differentiated from hESCs using a standard protocol involving exposure of embryoid bodies (EBs) to retinoic acid (RA) and recombinant sonic hedgehog protein (SHH) (see Methods ) [4] , [6] . After 31 days, EBs were dissociated and cryopreserved to allow multiple experiments to be performed on identical aliquots; however, similar data were obtained using fresh, unfrozen cells (not shown). Cell suspensions were thawed and plated in 96-well plates and automated counts of live motor neurons, defined as GFP + neurons with significant neurite outgrowth (SNO, total neurite length >75 µm), were performed ( Figures 1A and 1B ) [31] – [33] . In standard culture medium without neurotrophic support motor neuron numbers decreased over the first 7 days, reaching a plateau that was maintained until day 31+13 ( Figure 1C and 1D ). This did not reflect a loss of reporter expression since a similar decrease was seen when motor neurons were identified by staining for endogenous HB9 (not shown). In contrast, when the medium was supplemented with four neurotrophic factors [NTFs; brain-derived neurotrophic factor (BDNF), ciliary neurotrophic factor (CNTF), glial cell line-derived neurotrophic factor (GDNF) and insulin-like growth factor 1 (IGF-1) at 10 ng/mL] in addition to the cAMP-elevating compounds forskolin (F; 10 µM) and isobutylmethylxanthine (I; 100 µM), after an initial decrease in motor neuron numbers by day 31+7, there was a subsequent increase in the number of hESC-MNs, which reached nearly starting levels by day 31+13 ( Figure 1C and 1D ).
10.1371/journal.pone.0110324.g001 Figure 1
Ongoing birth of motor neurons in hESC-derived cultures is stimulated by neurotrophic factors.
(A) Live fluorescent human motor neurons derived from the Hb9::GFP reporter line at day 31+13 after growth with a cocktail of neurotrophic factors (NTFs). (B) Automated quantification of fluorescent cells with significant neurite outgrowth (SNO) using the Neurite Outgrowth module of MetaMorph software; cells counted are identified with a red overlay. Motor neurons were considered to have significant neurite outgrowth when their overall neurite length exceeded 75 µm (scale bar). (C) Representative image of immunostained Hb9::GFP hESC-motor neuron cultures at day 31+13 after growth with a cocktail of neurotrophic factors (NTFs). Scale bar = 50 µM. (D) Number of cells with significant neurite outgrowth (SNO) when grown with (red bars) or without (blue bars) neurotrophic factors, expressed as a percentage of numbers at day 31+1. The increase in motor neuron numbers after day 31+7 in NTF-supplemented cultures suggests ongoing neurogenesis. Surviving fluorescent GFP-positive motor neurons with SNO shown as mean ± s.e.m., n >5 (t-test, ***p<0.001, *p<0.05). (E) BrdU-positive Hb9::GFP-positive motor neurons (arrows) at day 31+15 confirming the presence of newborn human motor neurons in culture. Scale bar = 50 µM. (F) The percentage of Hb9::GFP-positive motor neurons that were BrdU-positive at day 31+15 is not changed by NTFs but (G) total numbers of BrdU-positive motor neurons are increased with NTFs. Bars indicate mean ± s.e.m., n = 3 (t-test, *p<0.05; n.s. = not significant).
This late increase in human motor neuron numbers could potentially be explained by ongoing genesis of motor neurons. To assess overall generation of new-born motor neurons we cultured cells with or without NTFs in the continuous presence of the mitotic label 5’-bromo-2’deoxyuridine (BrdU, 2 µM) and counted GFP-positive cells that had incorporated BrdU ( Figure 1E ). After 15 days, ∼60% of all Hb9::GFP cells were positive for BrdU in both conditions ( Figures 1F and 1G ), but cultures supplemented with NTFs contained 4-fold higher absolute numbers of new-born hESC-MNs ( Figure 1G , p<0.05). Together, these results demonstrate that human motor neurons are generated over extended periods of culture and that the yield of motor neurons can be increased by treatment with neurotrophic factors.
Screening for small molecules able to increase the yield of human motor neurons
Neurotrophic factors are costly culture supplements and have pleiotropic effects on neural development [16] , [34] , [35] . To exploit the high rate of neurogenesis in hESC-MN cultures in a more targeted manner to increase motor neuron yields, we sought to identify available reagents with similar activity. We therefore performed a small-scale screen of 160 bioactive compounds selected from a collection of drug-like chemicals and examined their effect on total motor neuron numbers. We reasoned that the assay might capture two types of compounds, those that increase motor neuron survival and/or others that increase motor neurogenesis. Both types of compounds would be of interest as they could be applied to increase the overall yields of motor neurons derived from hESCs.
Compounds (10 µM in quadruplicate wells) were added on the day of seeding and motor neurons were counted at day 31+13, the time point at which the greatest differences in human motor neuron numbers between control and NTF-supplemented cultures were observed ( Figure 1C , p<0.001). Most compounds showed no effect, and a significant number resulted in lower motor neuron numbers than the negative control condition ( Figure 2A ). In contrast, two compounds increased motor neuron numbers by>1.4 fold compared to basal conditions ( Figure 2A ). The most significant increase (1.9-fold) was induced by the Rho kinase (ROCK) inhibitor Y-27632 ( Figure 2A , Y-27632 vs. No NTFs, p<0.05).
10.1371/journal.pone.0110324.g002 Figure 2
The ROCK inhibitor Y-27632 increases human motor neuron numbers in hESC-derived motor neuron cultures.
(A) Screening of 160 compounds for their potential to increase the number of human motor neurons in hESC cultures at day 31+13. Compounds were tested in quadruplicate at a single concentration (10 µM). Values are plotted as mean fold difference in motor neuron numbers relative to the negative control condition (No NTFs). The Rho-kinase (ROCK) inhibitor Y-27632 was the compound showing the highest capacity to increase the number of human motor neurons. (B) Y-27632 increases the number of fluorescent hESC-motor neurons in mixed cultures in a dose-dependent manner. Cells were cultured in the absence of neurotrophic factors and in the presence of increasing concentrations of Y-27632. Values shown as mean ± s.e.m., n = 4. (C) Representative images of hESC-motor neuron cultures at day 31+13 grown under neurotrophic factor deprivation (No NTFs), neurotrophic factor supplementation (NTFs + F + I) and Y-27632 (10 µM). Scale bar = 25 µM. (D) Time-dependent increase in the number of motor neurons in the presence (green) but not absence (blue) of Y-27632 (10 µM), with a peak effect at day 31+9. Values shown as mean ± s.e.m., n >5 (t-test, *p<0.05; **p<0.01). (E) Y-27632 also increases the total number of cells in culture. Mean ± s.e.m., n = 3. (F) Hb9::GFP-positive neurons continue to express motor neuron markers HB9 and ISL1 after treatment with Y-27632 for 9 days. Scale bar = 50 µM. (G) Supplementation of cultures with Y-27632 (red line) leads to increased numbers of human motor neurons expressing endogenous ISL1 at day 31+9. Mean ± s.e.m., n = 3 (**p<0.01).
The EC 50 for Y-27632 at day 31+13 was 1.9 µM with a maximum effect of ∼5-fold ( Figure 2B ; p<0.05), even greater than that of neurotrophic factors ( Figures 2A and 2C ; Y-27632 vs. NTFs + F + I, p<0.05). To optimize the time window for the effects of Y-27632, we next studied the kinetics of human motor neuron generation with or without Y-27632 at its optimal concentration (10 µM). We focused on its effects when added post-dissociation at day 31. Maximum numbers of GFP-positive neurons, representing a ∼5-fold increase in motor neuron numbers over basal levels ( Figure 2D , p<0.05) were reached at day 31+9, which was therefore adopted as the standard time point for all subsequent experiments.
To exclude the possibility that Y-27632 might have affected the fidelity of HB9 reporter expression, we checked day 31+9 cultures using direct immunostaining for the motor neuron markers ISL1 and HB9; both showed a high degree of overlap with the GFP reporter ( Figure 2F ). Moreover, Y-27632 induced a nearly 4-fold increase in absolute numbers of hESC-MNs expressing endogenous ISL1 ( Figure 2G , p<0.05). Furthermore, to exclude the possibility that the class of motor neurons generated was altered with respect to standard differentiation protocols, we quantified the fraction of GFP-positive neurons expressing FoxP1, a marker for limb-innervating motor neurons, or Lhx3, a marker of medial motor neurons [30] , [36] , [37] . Comparable numbers of each class were generated and the ratio was not significantly affected by amplification with Y-27632 (p>0.05; not shown). One potential risk of this amplification procedure was that Y-27632 might dilute out motor neurons by stimulating the generation of other cell types. However, this did not appear to negatively affect the outcome: across the many different batches of hES-MNs analyzed in this study, the final abundance of motor neurons ranged from 5% to 45% of total cells, making it important that all treatment groups be compared to controls from the same batch. Although we did not exclude the batches with lower abundance, the value of 45% motor neurons is among the highest reported, demonstrating that expansion did not lead to excessive motor neuron depletion. Thus using a small-scale drug testing approach we were able to identify a compound, Y-27632, which can significantly increase motor neuron numbers in differentiated hESC cultures.
Y-27632 enhances proliferation of motor neuron progenitors in both hESC- and hiPSC-derived cultures
To better understand the level at which Y-27632 exerts its effect, we next examined the expansion of motor neuron progenitors (pMNs), using OLIG2 as a marker [26] . Treatment with Y-27632 increased the number of OLIG2-positive cells ∼3.6-fold compared to controls by day 31+9 ( Figures 3A and 3B , p<0.05) comparable to the ∼3.3-fold increase in DAPI-stained cells over controls over the same period [ Figure 2E , p>0.05; Ratio DAPI/OLIG2 = 6.8∶1 (CONTROL) vs. 6.2∶1 (Y-27632)]. Application of BrdU from day 31 to day 31+9 led to nuclear labeling of 86% of OLIG2-positive cells, indicating that they are actively proliferating progenitors ( Figures 3C and 3D ). Accordingly, 74% of GFP-positive motor neurons on day 31+9 were BrdU-positive, indicating that they were born during the period of Y-27632 treatment ( Figures 3E and 3F ). Similar percentages were observed using fresh, unfrozen motor neuron preparations (not shown). Therefore Y-27632 non-selectively enhances cell proliferation in hESC-derived cultures, resulting in a ∼3.5-fold increase in the number of motor neuron progenitors that is likely to contribute significantly to the observed increase in postmitotic hESC-MNs.
10.1371/journal.pone.0110324.g003 Figure 3
Y-27632 enhances proliferation of motor neuron progenitors in hESC- and hiPSC-derived motor neuron cultures.
(A) Y-27632-supplemented cultures contain increased numbers of OLIG2-positive cells at day 31+9. Scale bar = 50 µM. (B) Time-dependent increase in numbers of OLIG2-expressing progenitors in the presence of Y-27632. Data normalized to control at day 31+1; mean ± s.e.m., n >5 (t-test, **p<0.01). (C) OLIG2 progenitors at day 31+9 stained for BrdU. Scale bar = 25 µM. (D) Percent of OLIG2 precursors that are BrdU-positive at day 31+9 (mean ± s.e.m., n = 4). (E) Hb9::GFP-expressing motor neurons at day 31+9 stained for BrdU. Scale bar = 25 µM. (F) Percent motor neurons that are BrdU-positive at day 31+9 (mean ± s.e.m., n = 4). (G) The total number of cells in culture is increased at day 31+9 following Y-27632 treatment of hESC RUES1 and hiPSC18c. Values are mean ± s.e.m., n ≥3 (t-test, *p<0.05). (H) Numbers of OLIG2 precursors increase significantly at day 31+9 following Y-27632 treatment of hiPSC 18c. Values are mean ± s.e.m., n ≥3 (t-test, *p<0.05). (I) Numbers of motor neurons identified by staining for endogenous HB9 increase significantly at day 31+9 following Y-27632 treatment of hESC RUES1 and hiPSC 18c. Values are mean ± s.e.m., n ≥3 (t-test, *p<0.05). (J) Cultures from healthy control hESCs (RUES1) or hiPSCs (18c) immunostained for the motor neuron marker HB9 and the pan-neuronal marker β-III tubulin. Y-27632 increases the number of motor neurons in each case. Scale bar = 25 µM.
To determine whether Y-27632 was a generally effective treatment for pluripotent stem cell lines, we performed similar experiments using an additional hESC line, RUES1; and a hiPSC line, 18c, derived from a healthy control subject [6] . Total numbers of DAPI-stained cells and OLIG2-positive progenitors were quantified as above after 31+9 days. Significant increases in both DAPI-positive and OLIG2-positive cells were observed following Y-27632 treatment using hiPSC 18c ( Figures 3G and 3H , p<0.05) and for DAPI using RUES1 ( Figure 3G ). To detect motor neurons in the absence of a reporter we performed immunostaining for HB9, to label motor neurons, and β-III tubulin, to label all neurons ( Figures 3I and 3J ). Automated image analysis of such cultures revealed a 2- to 4-fold increase in motor neuron numbers ( Figure 3I , p<0.05). Y-27632 is therefore a useful tool for both hESCs and clinically relevant hiPSC lines.
Design of a robust survival assay for purified human motor neurons
Our overall goal was to study the trophic requirements of human motor neurons. Bulk day 31 cultures were therefore dissociated and grown in the presence of Y-27632 for 3 days or 9 days before FACS analysis, leading to a ∼2-fold increase in the total yield of motor neurons after 3 days ( Figure 4A ; p<0.01) and a nearly 4-fold increase after 9 days ( Figure 4B ; p<0.01). For all subsequent experiments, expanded human motor neurons from the day 31+3 time point were used.
10.1371/journal.pone.0110324.g004 Figure 4
FACS-sorting of amplified cultures yields a pure preparation of viable human motor neurons.
(A) Y-27632 supplementation for 3 days leads to a 1.8-fold increase in motor neuron yield judged by FACS analysis. Data normalized to controls without Y-27632. Values are mean ± s.e.m., n>5 (t-test, **p<0.01). (B) Nine-day treatment with Y-27632 gives a ∼5-fold increase in motor neuron yield as compared to controls without Y-27632, as quantified by flow cytometry. Values are mean ± s.e.m., n>5 (t-test, **p<0.01). (C) FACS purification of Hb9::GFP motor neurons expanded with Y-27632 for 3 days. Representative FACS gating used to retrieve an almost pure (>95%) population of human motor neurons. (D) FACS-purified motor neurons at day 31+3+1 stained for GFP (green), and a combination of HB9 and ISL1 (“pan-MN”; white nuclei).>95% of the FACS-purified cells in culture are Hb9::GFP positive. Scale bar = 25 µM. (E) Even following FACS sorting, some contaminant cells were able to proliferate and form colonies that interfered with survival assays (left panel). Uridine/Fluorodeoxyuridine (U/FdU) (each at 1 µM) successfully prevented the proliferation (right panel).
Given the ongoing neurogenesis in mixed cultures, it was first necessary to find conditions in which expanded postmitotic neurons could be studied in isolation. Direct treatment of mixed cultures with mitotic inhibitors did not produce satisfactory results: cytosine arabinoside (AraC) proved toxic for human motor neurons, while even the less toxic uridine/fluorodeoxyuridine (U/FdU) led to clumping of neurons on remaining islands of non-neuronal cells (not shown). Motor neurons were therefore FACS-sorted ( Figure 4C ) and seeded on polyornithine/laminin-coated coverslips in medium containing a cocktail of NTFs plus the c-AMP elevating compounds forskolin and IBMX. Using FACS conditions involving a slow sorting rate and a wide nozzle, the seeded motor neurons rapidly developed robust neurite outgrowth ( Figure 4D ). To estimate their purity, we performed immunostaining using a combination of antibodies to HB9 and ISL1 (“pan-MN”) [30] . At day 31+3+1 (differentiation + expansion + days post-FACS),>95% of the neurons were Hb9::GFP-positive, and reporter expression showed strong overlap with HB9/ISL1 staining ( Figure 4D ). Despite this high degree of enrichment, colonies of proliferating progenitors were occasionally observed ( Figure 4E ); sorted motor neurons were therefore cultured in the presence of the antimitotic drug U/FdU ( Figure 4E ). The new protocol therefore provides a robust and abundant source of highly purified hESC-MNs.
To develop a survival assay based on neurotrophic factor deprivation [31] , [38] , [39] , FACS-sorted motor neurons were seeded in 96-well plates and stained using the vital dye calcein-AM. This had the advantage that it stained cell bodies and neurites more intensely than live imaging of GFP, which was no longer required to identify motor neurons. Numbers of surviving hESC-MNs were counted in whole culture wells in an automated manner using MetaMorph ( Figure 5A ). We first asked whether the survival of purified motor neurons was dependent on trophic support in these conditions. At day 31+3+7, motor neuron survival was enhanced ∼2.5-fold by a cocktail of NTFs (BDNF, CNTF, GDNF, IGF-1, each at 10 ng/ml) with F (10 µM) plus IBMX (100 µM) ( Figure 5B ), similar to published results using cultures of primary rodent motor neurons [39] – [41] . We tested forskolin and IBMX alone and found that they showed only slight innate neurotrophic activity (not shown).
10.1371/journal.pone.0110324.g005 Figure 5
Purified human motor neurons show a potent response to known neurotrophic factors.
(A) Whole-well imaging of live motor neurons labeled with calcein-AM captured using the Plate Runner (left two panels). Surviving human motor neurons were counted in whole culture wells in an automated manner using MetaMorph (red tracing, right two panels). Scale bar = 200 µm. (B) Y-27632-expanded motor neurons show enhanced survival in the presence of a cocktail of neurotrophic factors. Values shown as mean ± s.e.m., n>5 (t-test, ***p<0.001). (C) GDNF, (D) BDNF and (E) CNTF alone (blue lines) enhance the survival of expanded FACS-purified human motor neurons. The addition of F+I significantly potentiates the survival-inducing activity of GDNF at high concentrations. Values shown as mean ± s.e.m., n>4 (t-test, *p<0.05; **p<0.01; ***p<0.001). Asterisks on individual points represent significance of difference with No-NTF control (white rectangle in the curve); asterisks on bars represent significant differences between a given concentration of NTF and the corresponding value for NTF + F+ I. (F) The cocktail of neurotrophic factors (NTFs) enhances the survival of expanded FACS-purified human motor neurons in a dose-dependent manner in the presence of 10 µM forskolin plus 100 µM IBMX. Values shown as mean ± s.e.m., n≥5 (t-test, **p<0.01; ***p<0.001).
This provided an opportunity to better characterize the effects of known neurotrophic factors on hESC-MNs. Doses of GDNF, BDNF, CNTF and IGF-1 ranging from 2 pg/mL to 10 ng/mL were first tested alone for their effects on survival at day 31+3+7 ( Figure 5D to 5G ). Except for IGF-1 (not shown), which showed no survival promoting effect alone, each neurotrophic factor provided significant support for human motor neuron survival with EC 50 values as follows: 2 pM for BDNF, 2 pM for GDNF and 1 pM for CNTF. These are slightly higher than the most potent EC 50 values reported for the same factors on primary rodent motor neurons (BDNF, EC 50 = 1 pM [38] ; GDNF, EC 50 = 0.2 pM [31] ; CNTF, EC 50 = 0.1 pM [18] , [42] ); this may reflect differences related to species, human stem cell origin or batch of neurotrophic factor. Since the effects of neurotrophic factors on rat motor neurons in defined media was reported to depend on intracellular cAMP levels [40] , [41] , we also tested the effects of inclusion of forskolin and IBMX (F+I). The neurotrophic activity of each factor tested appeared to be increased in the presence of F+I, though this effect was only significant for single points at the highest concentration of GDNF ( Figure 5D-F ). To determine whether different neurotrophic factors were potentially acting on different subsets of motor neurons in the cultures, we next performed a dose-response analysis for a combination of all factors with a fixed concentration of F+I ( Figure 5H ). The maximum number of motor neurons maintained in culture was not significantly greater than that with BDNF, CNTF or GDNF alone (with F+I). This suggests that essentially all viable motor neurons are maintained by optimal doses of these single factors, at least after 7 days in culture. Therefore, like their rodent counterparts, human motor neurons show an exquisitely sensitive response to multiple neurotrophic factors.
Lastly, to evaluate the ability of the newly developed human motor neuron survival assay to detect novel neurotrophic compounds, we determined whether the beneficial effect of Y-27632 on human motor neuron numbers, in addition to its effect on progenitor proliferation, might also reflect a survival effect. To exclude effects on cell attachment we first verified that the presence of the drug did not affect hESC-MN numbers after 24 hours ( Figure 6A ). After 7 days in culture, Y-27632 had a clear dose-dependent survival effect ( Figures 6B and 6C ), though to a more modest extent than neurotrophic factors. The EC 50 for the Y-27632 survival effect on motor neurons was 2 µM, similar to the value for motor neuron expansion. Thus Y-27632 not only promotes proliferation of motor neuron progenitors but also functions as a motor neuron survival factor. The fact that the fold-increase in survival was lower than that induced in long-term treatment of mixed cultures ( Fig. 2B ), likely reflects the absence of proliferation and/or other cell types.
10.1371/journal.pone.0110324.g006 Figure 6
Y-27632 is also a survival factor for human motor neurons.
(A) The plating efficiency of FACS-purified human motor neurons after 24 hours is not increased in the presence of Y-27632. (B) Y-27632 enhances the survival of FACS-purified human motor neurons in a 7-day survival assay. Scale bar = 200 µM. (C) Dose-dependent effects of Y-27632 on human motor neuron survival, expressed relative to the basal condition (0 µM). Values shown as mean ± s.e.m., n≥5 (t-test, *p<0.05; **p<0.01).
Discussion
Human embryonic and induced pluripotent stem cells (hESCs and hiPSCs) represent a powerful tool for studying human development, disease modeling and drug discovery. However, one major limiting factor for prospective drug screens is the efficiency with which the affected cell types can be generated and, in the case of motor neurons and many other neuronal classes, the absence of a validated survival assay. Here, we took advantage of our observation of ongoing motor neuron generation in hESC-derived cultures to devise a new method for amplification of motor neuron progenitors to increase motor neuron yields. In addition, using optimized conditions for FACS sorting of neurons expressing the Hb9::GFP reporter, we developed a robust assay for survival factors acting directly on postmitotic motor neurons, and used it to show that human motor neurons respond in a potent manner to both known and novel neurotrophic molecules.
The ongoing neurogenesis in human motor neuron cultures that we describe contrasts with the short ∼24-hour period of motor neuron production in differentiated mouse ES cell cultures [43] . This is likely to reflect normal biological differences in the development of motor systems in rodent and human embryos, since human motor neurons are produced over an extended three-week period in vivo [28] , [29] . We first exploited this to screen for compounds that would further amplify the precursor population, identifying Y-27632 as the most active compound in a screen which, like higher-throughput assays, was carried out at a single concentration. Exactly how Y-27632 is achieving this may involve multiple mechanisms, but we considered three potential modes of action for Y-27632 in increasing numbers of motor neuron progenitors. First, it could act by blocking differentiation of progenitors to motor neurons. This seems unlikely since the numbers of OLIG2-positive precursors and motor neurons increased in parallel. Second, it might specifically promote the generation of OLIG2-positive precursors. Since total DAPI numbers increased in parallel, such a selective effect seems unlikely. We therefore believe Y-27632 acts by shortening cell cycle time for dividing precursors as a whole, leading to expansion – but not enrichment – of motor neuron progenitors and a subsequent increase in motor neuron yield [44] – [46] .
Nevertheless, the ongoing neurogenesis also provides a potentially serious confound for interpretation of experiments examining changes in motor neuron numbers in mixed cultures. In studies that do not take this into account, it is possible that an increase of motor neuron numbers attributed to improved survival may instead reflect an effect on neurogenesis. To overcome this issue, we FACS-purified motor neurons derived from the Hb9::GFP hESC line and cultured them alone in the presence of a mitotic inhibitor U/FdU to inhibit proliferation of any remaining progenitors. This is in some ways analogous to the approach recently reported by Yang et al. [24] , except that to block proliferation they used cytosine arabinoside, which was cytotoxic in our hands. Moreover, their cell survival experiments, performed over a 20-day period, required a mouse astrocyte monolayer as substrate, whereas our cultures contained essentially only motor neurons.
Using this essentially pure preparation of postmitotic motor neurons we showed that three known neurotrophic factors potently enhance human motor neuron survival, and that their action is potentiated when endogenous levels of cAMP are increased. Therefore, in this respect, the human stem cell-derived neurons closely resemble rodent motor neurons both in primary culture and in vivo . Since dependence on trophic factors is acquired over time during embryogenesis [47] , this also suggests that the human motor neurons have reached a stage of maturation comparable to those in the mid-embryonic period in mice.
Y-27632 has been shown to have contrasting biological effects in different systems, ranging from pro-proliferative effects on hESCs and hiPSCs [48] , [49] to anti-proliferative effects on cancer cells [50] , cord blood-derived CD34 + hematopoietic progenitor cells [51] , hepatic stellate cells [52] and smooth muscle cells [53] . While it is neuroprotective for primary mouse Purkinje cells [54] , retinal ganglion cells [55] and murine hippocampal slice cultures [56] and growth-promoting for corticospinal tract axons [57] , [58] and adult optic nerve [59] , Y-27632 is not protective for hiPSC-derived dopaminergic neurons [60] . Our study extends others which suggest that Y-27632 exhibits generally beneficial effects on motor neurons. A recent report documented an increase in the lifespan of an intermediate mouse model of SMA following administration of fasudil, another ROCK inhibitor [61] . Even though the compound was not able to halt motor neuron loss in the ventral horn of the spinal cord, positive effects on the maturation of the neuromuscular junction and muscle fiber size were reported [61] . More recently, fasudil was reported to extend survival – and reduce motor neuron death - in a mouse model of amyotrophic lateral sclerosis [62] . Therefore, the neurotrophic properties of Y-27632 described here for cultured human neurons likely reflect a mechanism of action that is conserved across species and in vivo .
In summary, our study defines conditions for systematic assays of neurotrophic factors and survival-promoting compounds for human motor neurons. We show that the technique can be extended to human iPSC-derived motor neurons and therefore in principle to comparisons between cells derived from ALS patients and controls: we and others recently derived Hb9::GFP or Hb9::RFP reporters for different ALS-iPSC lines [63] . Importantly, in agreement with our earlier studies on the expression of specific markers, electrophysiological characteristics and development following transplantation [6] , [30] , we show that the neurotrophic dependence of human stem cell-derived motor neurons has reached a state of maturity comparable to that of primary embryonic motor neurons in vitro and in vivo . Although more still needs to be done before they can be considered to reflect the properties of the postnatal spinal cord, this validates their use as a human model for analyzing multiple aspects of motor neuron development and pathology.
Materials and Methods
Cell lines
All the human ES and iPSC lines have been reported in an earlier publication [6] . The iPS cell lines were derived by retroviral transduction of OCT4, SOX2, and KLF4 in dermal fibroblasts. All pluripotent cell lines were characterized by conventional methods and grown under standardized conditions as described below.
Ethics Statement
The work performed with human motor neurons derived from hESCs and hiPSCs has been approved by Columbia University ESCRO committee (Embryonic Stem Cell Research Oversight committee). Patient fibroblasts for generating human iPS lines were collected with written informed consent under IRB approval AAAC1257 from Columbia University Medical Center.
Growth of hPSC lines
We used an HB9::GFP reporter hESC line [23] , the wild-type hESCs line RUES1 and hiPSC line 18c [6] . All cell cultures were maintained in a humidified incubator at 37°C and 5% CO 2 . Human ESCs and hiPSCs were grown on a pre-gelatinized tissue culture flask on a monolayer of irradiated CF-1 mouse embryonic fibroblasts (MEFs; GlobalStem) plated at 15,000–18,000 cells/cm 2 in hPSC medium [DMEM/F12 (Invitrogen), 20% knockout serum replacement (Invitrogen), 1 mM L-glutamine (Gibco), 100 µM non-essential aminoacids (Gibco) and 100 µM β-mercaptoethanol (Sigma-Aldrich)] supplemented with 20 ng/ml recombinant human basic fibroblast growth factor (bFGF; R&D Systems). Medium was changed every day for the duration of the expansion and lines were passaged every 4–6 days using dispase (Gibco) at 1 mg/mL in hPSC medium for 30 minutes at 37°C.
Differentiation of hESCs and hiPSCs into motor neurons
hESCs and hiPSCs were allowed to reach 75%–90% confluency. Then, colonies were treated with dispase (1 mg/mL) to separate colonies from the MEF layer. After 30 minutes, cells were washed off the flask using hPSC medium and collected in a 50 mL Falcon tube. Colonies were allowed to settle by gravity and then medium was aspirated. Fresh hPSC medium was added to the cells. This step was repeated three times to wash away all the remaining dispase. Settled colonies were then mechanically dissociated into small 10- to 15-cell chunks using a P1000 tip by performing up and down movements in a 1 mL volume. Cell aggregates were transferred to low adherence T75 flasks in hPSC medium with 20 ng/mL bFGF and 20 µM Y-27632 (Ascent) for the first 24 hours. At day 1, cells for all experiments were supplemented with hPSC medium containing 20 ng/mL bFGF, 20 µM Y-27632, 10 µM SB431542 (Sigma-Aldrich) and 0.2 µM LDN193189 (Stemgent). The medium was changed daily from day 2 to day 4. At day 5, embryoid bodies (EBs) were switched to medium composed of 50% hPSC medium and 50% neural induction medium [NIM; DMEM/F12 (Invitrogen), 1% N2 supplement (Invitrogen), 1 mM L-glutamine (Gibco), 100 µM non-essential aminoacids (Gibco) and 2 µg/mL heparin (Sigma-Aldrich)] supplemented with 10 µM SB431542, 0.2 µM LDN193189, 10 ng/mL recombinant human brain-derived neurotrophic factor (BDNF; R & D Systems), 0.4 µg/mL ascorbic acid (Sigma-Aldrich) and 1 µM retinoic acid (Sigma-Aldrich). At Day 7 cells were switched to 100% NIM, keeping the same medium supplementation. Every other day between days 9 and 21, NIM supplemented with 10 ng/mL BDNF, 0.4 µg/mL ascorbic acid, 1 µM retinoic acid and 200 ng/mL recombinant C25II modified sonic hedgehog protein (SHH; Invitrogen) was added to the EBs. At day 22, cells were cultured with 50% NIM and 50% neural differentiation medium [NDM; Neurobasal (Invitrogen), 1% N2 Supplement (Invitrogen), 1 mM L-Glutamine (Gibco) and 100 µM Non-Essential Aminoacids (Gibco)] supplemented with 2% B-27 supplement (Invitrogen), 0.4 µg/mL ascorbic acid, 1 µM retinoic acid, 200 ng/mL SHH (Invitrogen), 10 ng/mL BDNF, 10 ng/mL recombinant human ciliary neurotrophic factor (CNTF; R & D Systems), 10 ng/mL recombinant human glial cell line-derived neurotrophic factor (GDNF; R & D Systems) and 10 ng/mL recombinant human insulin-like growth factor 1 (IGF-1; R & D Systems). Between days 24 and 31, the bulk medium was switched to 100% NDM and the EBs grown under the previous medium supplementation. After 31 days of differentiation the EBs were dissociated and the resulting neuronal cultures cryopreserved. Briefly, the EBs were collected in a 50 mL Falcon tube and then washed twice with PBS without Ca 2+ and Mg 2+ (Invitrogen) to eliminate residual media. The EBs were then incubated at 37°C in pre-warmed 0.05% Trypsin-EDTA (Invitrogen) for 5–10 minutes. Lastly, fetal bovine serum (Invitrogen) supplemented with 100 µg/mL deoxyribonuclease I (DNAse I, Sigma-Aldrich) was added to stop the trypsin reaction and the cells were spun for 5 minutes at 400× g . The cells were resuspended in 1 mL of complete trituration and wash medium [CTWM, PBS without Ca 2+ and Mg 2+ , 25 mM Glucose (Sigma-Aldrich), 4% L-15 dialyzed BSA (Sigma-Aldrich), 100 µg/mL DNAse I, 1% N2 supplement, 2% B27 supplement, 600 mM magnesium chloride (Sigma-Aldrich), 500 nM EDTA (Sigma-Aldrich) and 2% FBS] and subsequently mechanically triturated using a P1000 tip. The resulting cell suspension was filtered using a 40 µM cell strainer (BD Falcon) to eliminate large residual clumps and centrifuged for 5 minutes at 400× g . The cells were then resuspended in NDM supplemented with 2% B27, 0.4 µg/mL ascorbic acid, 25 µM glutamate E (Sigma-Aldrich), 25 µM β-mercaptoethanol (Millipore), 0.1 µM retinoic acid, 10 ng/mL BDNF, 10 ng/mL CNTF, 10 ng/mL GDNF and 10 ng/mL IGF-1. These cells were counted and then prepared for cryopreservation using 2x Freezing Media (Millipore). Vials of 5–10 million cells/mL were prepared to be used in further experiments.
Coating of 96-well plates
All survival and proliferation studies were performed in 96-well plates (Greiner Bio-One) coated with polyornithine (Sigma-Aldrich) and mouse laminin (Invitrogen). Briefly, 100 µg/mL polyornithine (Sigma-Aldrich) in cell culture water was added to the wells for at least 2 hours then aspirated and the wells rinsed once using water. Coating was completed by adding overnight 15 µg/mL mouse laminin in L15 medium (Sigma-Aldrich) supplemented with 7.5% sodium bicarbonate (Gibco). In studies involving FACS-sorted cells, a concentration of 1000 µg/mL polyornithine was used for coating.
Studies involving mixed hPSC-derived motor neuron cultures
All proliferation/survival studies involving mixed hPSC-derived motor neuron cultures were started from previously cryopreserved vials. After quickly thawing the vials in a 37°C water bath, cells were resuspended in NDM medium supplemented with 2% B27 Supplement. Cells were then spun at 400× g for 5 minutes. The supernatant was gently aspirated and cells resuspended in 10 mL of NDM with 2% B27. A 4% BSA protein cushion was then layered under the cell suspension and the cells spun at 400× g for 5 minutes, with low acceleration and deceleration. Afterwards, cells were resuspended in basal medium (BM) [Custom Clear Neurobasal (Invitrogen), which omits phenol red and riboflavin to allow live fluorescent imaging in the presence of a significantly attenuated auto-fluorescent background; 1 mM L-glutamine and 100 µM non-essential aminoacids, 2% B27, 0.4 µg/mL ascorbic acid, 25 µM glutamate E, 25 µM β-mercaptoethanol, 0.1 µM retinoic acid] and counted using a hemocytometer. Finally, cells were resuspended at the final desired seeding concentration of 32,000 cells/well and 100 µL was added to each well. Cells were allowed to attach at 37°C for 2 hours before addition of supplements at 3x concentration in 50 µL of BM.
Screening for small molecules with the potential to increase numbers of human motor neurons in culture
From a collection of drug-like chemicals from the Microsource and Tocris collections, two plates containing a total of 160 compounds were selected. Each compound was tested at 10 µM. Basal medium to dilute compounds from original stocks was M-199 (without phenol red; Invitrogen) with 5% DMSO (100% anhydrous, Fisher Scientific), freshly prepared. Survival in BM was used as negative control (trophic factor deprivation). Survival in BM supplemented with a cocktail of NTFs [BDNF, CNTF, GDNF and IGF-1] plus the cAMP-elevating compounds forskolin (F; 10 µM; Sigma-Aldrich) and isobutylmethylxanthine (I; 100 µM; Sigma-Aldrich) was the positive control (trophic factor supplementation). Cells were seeded at 32,000 cells/well in 150 µL and compounds added in a 15 µL volume (the final DMSO concentration of 0.45% did not adversely affect motor neuron survival when added alone, not shown). The same volume of M-199 with 5% DMSO was added to the negative control wells. NTFs + F + I were also added in 15 µL of M-199 with 5% DMSO in positive control wells. For each set, compounds were tested in quadruplicate by creation of 4 test plates. In each plate, each control condition (positive and negative) had six replicate wells. Readouts were performed on day 31+13. After quantification of the total number of surviving cells with significant neurite outgrowth (see Results ), data were plotted as mean fold difference as compared to numbers in the negative control condition. Plates were rejected when the mean difference in cells numbers between positive and negative control was lower than 1.3 fold. Validation of the most active compounds was performed by serial dose response studies.
Immunocytochemistry
Neuronal cultures were pre-fixed by adding one volume of 4% paraformaldehyde diluted in phosphate-buffered saline 1x (PBS1x/4%PFA) for 2 minutes at room temperature. Then, cells were fixed with PBS1x/4%PFA for 30 minutes at 4°C. After fixation, cells were washed with PBS1x three times for 5 minutes and then permeabilized and quenched for at least 30 minutes using PBS1x with 0.1% Triton-X (PBSTX-0.1%) supplemented with 100 mM glycine and 0.1% Sodium Azide (Sigma-Aldrich). Cells were blocked in PBSTX-0.1% containing 10% donkey serum (Sigma-Aldrich) and 0.1% sodium azide (Sigma-Aldrich) (blocking solution) for one hour. After blocking, cells were incubated overnight at 4°C with primary antibodies diluted in the blocking solution. Primary antibodies used in this study were the following: rabbit anti-GFP (1∶3000, Abcam), mouse anti-ISL1 (1∶200, DSHB, 39.4D5), guinea-pig anti-ISL1 (1∶2000, courtesy of Susan Brenner-Morton, Jessell laboratory at Columbia University), mouse anti-HB9 (1∶100, DSHB, MNR2 81.5C10-c), chicken anti-β-III Tubulin (TUJ1, 1∶1000, Neuromics), rabbit anti-Olig2 (1∶1000, Millipore) and rat anti-BrdU (1∶150, Serotec). Cells were washed five times with PBSTX-0.1% for 5 minutes. Antigens were visualized by incubating for 60–75 minutes at room temperature with the appropriate secondary antibodies (DyLight 488, 549 and 649 conjugated, 1∶1000, Jackson ImmunoResearch). Lastly, neuronal cultures were again washed five times with PBSTX-0.1% for 5 minutes and incubated in a solution containing DAPI (1∶50000, Sigma-Aldrich) for 15 minutes. Cells were washed once with PBSTX-0.1% and then imaged.
BrdU incorporation studies
5-bromo-2-deoxuridine (BrdU) incorporation studies were performed to analyze cell proliferation. Neuronal cultures were incubated with BrdU (2 µM; Sigma-Aldrich) for the full duration of culture until fixation with PFA. The standard protocol for immunochemistry described above was followed for other antigens besides BrdU. Then, to detect BrdU incorporation, cells were again pre-fixed for 2 minutes at room temperature and fixed with PBS1x/4% PFA for 15 minutes at 4°C. They were then washed with PBS1x three times for 5 minutes. Cells were then incubated with pre-warmed (37°C) 2 M HCl in distilled water for 10 minutes at 37°C, light protected. Lastly, the HCl was aspirated and cells were incubated in 0.15 M boric acid in distilled water for 2 minutes at room temperature. Cells were then washed three times with PBS1x for 5 minutes and blocked for 1 hour using blocking solution. Finally, cells were incubated overnight at 4°C with rat anti-BrdU primary antibody (1∶150, Serotec) in blocking solution. In order to correct for any non-specific background staining, the same procedures were performed on samples incubated or not with BrdU.
FACS purification and motor neuron survival studies
HB9::GFP reporter hESC-derived motor neurons were grown in polyornithine/laminin-coated T75 flasks prior to FACS purification in order to maximize the amount of cells retrieved after the procedure. After the expansion period, the medium was aspirated and cells washed once with PBS without Ca 2+ and Mg 2+ to eliminate residual medium. The cells were then incubated at 37°C in pre-warmed 0.05% Trypsin-EDTA for 5 minutes. DNAse I-supplemented FBS was used to stop the trypsin reaction. Cells were collected and centrifuged for 5 minutes at 400× g . Cells were resuspended in complete trituration and wash medium [CTWM, PBS without Ca 2+ and Mg 2+ , 25 mM Glucose (Sigma-Aldrich), 4% L-15 dialyzed BSA (Sigma-Aldrich), 100 µg/mL DNAse I, 1% N2 supplement, 2% B27 supplement, 600 mM magnesium chloride (Sigma-Aldrich), 500 nM EDTA (Sigma-Aldrich) and 2% FBS], filtered through a 40 µM Cell Strainer (BD Falcon) and centrifuged for 5 minutes, at 400× g . The cells were then resuspended in 750–800 µL of CTWM and transferred to a sorting tube (BD Falcon). Cells were sorted based on GFP expression using a BD FACS Aria II sorter (Becton Dickinson) configured with a 100 µm ceramic nozzle and operating at 20 psi for no longer than 30 minutes. Purified cells were collected in a tube containing CTWM. After collection cells were spun for 5 minutes at 400 xg and resuspended in Basal FACS Medium [Basal Medium with Clear Custom Neurobasal, supplemented with 1 µM uridine/fluorodeoxyuridine (U/FdU; Sigma-Aldrich), 100 Units/mL Penicillin (Invitrogen), 100 µg/mL Streptomycin (Invitrogen) and 100 µg/mL Normocyn (InvivoGen)]. After cells were counted using a hemocytometer, they were resuspended at the final seeding concentration of 2000 cells/well and added to the wells in 100 µL. Medium supplements were added to the cells at 3x concentration in 50 µL of Basal FACS Medium after the cells were allowed to incubate at 37°C for 2 hours in order to attach to the bottom of the plate. Readouts were performed after 7 days.
Calcein live imaging
To facilitate the imaging of FACS-sorted GFP-positive cells, we used the Calcein-AM Red-Orange (Invitrogen) vital dye at 2.5 µM concentration. Briefly, cells were incubated with the dye for 5 minutes and then extraneous fluorescence was quenched using 5 mg/mL hemoglobin before image acquisition using the Plate Runner (Trophos).
Image acquisition and quantitative image analysis
Image acquisition was performed using either a Carl Zeiss Observer Z1 epi-fluorescence Microscope (Carl Zeiss Inc.; acquisition of 12 images per well at 10x magnification) or the whole well imaging device Plate Runner (Trophos). Automated quantitative image analysis of fluorescent surviving hESC-MNs and stained neuronal cultures was performed using the MetaMorph Software V7.6 (Molecular Devices). The Neurite Outgrowth application in the software was employed to quantify fluorescent human motor neurons that have neurite outgrowth above a certain threshold, reducing the number of false-positive cells such as non-viable neurons included in the analysis. Quantitative analysis of stained hPSC-derived human motor neuron cultures was performed using the Multi-Wavelength Cell Scoring application. For a specific marker, positive cells were selectively identified as having clear signal intensity above local background. Intensity thresholds were set blinded to sample identity. In a given experiment the same parameters were used in all images analyzed. Parameters were only minimally adjusted across different experiments.
Statistical Analyses
All quantitative data were analyzed using IBM SPSS Statistics 19 (IBM SPSS). For each set of data a double statistical evaluation was performed: A) for each condition/time point mean values were compared using one-way ANOVA statistical evaluation followed by Tukey HSD Post-hoc test; B) possible interactions between time and condition were assessed using two-way ANOVA statistical evaluation. In cases involving only one time point and a two-group comparison, p value was determined using Student's t-test. Differences were considered to be significant when p<0.05.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
There are several errors in this article:
The subtitle of this article is incorrect. The correct subtitle is “Virus Threat to Plant Biodiversity”.
Table 5 is incorrect. The common name of Hovea elliptica is missing. The authors have provided a corrected version below.
10.1371/journal.pone.0091224.t005 Table 5
Native plants in which infection with introduced or unidentified viruses was detected by ELISA tests on samples
Species
Common name
Site location
No. of plants tested (grouping)
No. of positive samples (% infection)
AMV
TuYV *
BYMV
CMV
Potyvirus
Tospovirus
Luteovirus
2001 Samples
Caesalpiniaceae
Cassia sp.
-
Calingiri
1 (1)
-
0
1
-
1
0
0
Fabaceae
Bossiaea eriocarpa
Common brown pea
Bindoon
85 (5)
-
0
-
-
0
1 (1)
-
Bossiaea ornata
Broad-leaf brown pea
The Lakes
5 (1)
-
1
-
-
0
0
-
Daviesia nudiflora
-
Quairading
40 (1)
-
1 (3)
-
-
0
0
0
Gompholobium sp.
-
Badgingarra
7 (7)
-
0
-
-
1
0
-
Hovea elliptica
Tree hovea
Mt Barker
10 (10)
-
0
-
-
1
0
0
Kennedia eximia
-
Bindoon
30 (5)
-
0
3 (13)
-
3 (13)
0
0
Kennedia prostrata
Scarlet runner
Brookton
15 (5)
-
0
2
-
2
0
0
Leptosema aphyllum
Ribbon pea
Carnamah
21 (7)
-
0
-
-
2 (14)
0
0
Goodeniaceae
Damperia sp.
-
Woodanilling
10 (1)
-
1 (10)
-
-
0
0
1 (10)
2009 Samples
Asparagaceae
Chamaescilla corymbosa
Blue squill
Kings Park
12 (1)
0
0
-
0
12 (100)
0
-
Droseraceae
Drosera sp.
Sundew
Wooroloo
6 (6)
-
0
1
0
1
0
-
Fabaceae
Hovea elliptica
Tree hovea
Wellard
1
-
-
-
0
0
1 **
-
Haemodoraceae
Anigozanthos sp.
Kangaroo paw
Manjumup
60 (10)
0
0
-
0
3 (8)
0
-
Anigozanthos manglesii
Mangles kangaroo paw
Wooroloo
1
-
-
-
1
0
0
-
Hemerocallidaceae
Caesia micrantha
Grass lily
Kings Park
30 (1)
-
-
-
-
14 (47)
-
-
Juncaginaceae
Triglochlin sp.
Arrowgrass
Helena River
20 (1)
-
-
11(55)
-
-
-
-
Triglochlin sp.
Arrowgrass
Guildford
20 (1)
-
-
13 (65)
-
-
-
-
Triglochlin sp.
Arrowgrass
Kings Park
18 (1)
18 (100)
-
18 (100)
Triglochlin sp.
Arrowgrass
Not recorded
50 (1)
-
0
-
-
47 (94)
Orchidaceae
Caladenia paludosa
Common swamp spider-orchid
Kings Park +
2
-
-
2
0
2
0
-
Cymbidium canaliculatum
Black orchid
Kings Park +
1
-
-
0
0
1
0
-
Dendrobium sp.
-
Kings Park +
1
-
-
0
0
1
0
-
Diuris longifolia
Common donkey orchid
Kings Park +
3 (1)
-
-
2
0
3
0
-
Diuris longifolia
Common donkey orchid
Kings Park +
46 (1)
-
-
12 (26)
0
25 (54)
0
-
Diuris micrantha
Dwarf bee orchid
Kings Park +
1
-
-
1
0
1
0
-
Microtis sp.
Onion orchid
Kings Park +
1
-
-
1
0
1
0
-
Thelymitra sp.
Sun orchid
Kings Park +
1
-
-
1
0
1
0
-
For an explanation of virus acronyms see Table 1, - = Not tested, * = TuYV detected by BWYV polyclonal antibodies. ** = Tospovirus positive sample tested negative for TSWV and Impatiens necrotic spot virus (INSV), + = Native orchid collection. Samples were either tested individually or grouped (in 5’s-10’s) before testing. When sufficient grouped samples were present, percentage infection was calculated using the formula of Gibbs and Gower [71]. All orchid samples also tested for Cymbidium mosaic virus and Odontoglossum ringspot virus by ELISA, but none contained them.
In the Discussion section, the second sentence of the seventh paragraph is incorrect. The correct sentence is, “Previously, seven introduced viruses were reported naturally infecting native plants in the SWAFR (see Introduction).”
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Cardiac fibrosis (CF) is frequently associated with cardiac hypertrophy, atrial fibrillation, and ventricular arrhythmias, and heart failure (HF). It is characterized by the excessive deposition of extracellular matrix (ECM) molecules [ 1 ]. The fibrotic ECM molecules increase left ventricular (LV) stiffness and trigger various molecular signaling pathways, collectively resulting in the development of HF [ 2 ]. They also impair mechanoelectric coupling of cardiomyocytes, thus increasing the risk of arrhythmias [ 3 ]. Cardiac fibroblasts (FBs) play a central role in the development of CF. In response to various insults, FBs proliferate, migrate to the site of insults, and transdifferentiate into myofibroblasts (MFBs) that actively secrete fibrotic ECM molecules [ 4 – 6 ]. The formation of FBs during CF involved multiple mechanisms. For example, resident FBs are triggered to proliferate, and endothelial cells are substantially transformed to FBs via the endothelial-mesenchymal transition (EndMT).
Cytokine-like 1 (Cytl1) is a secreted protein first identified in CD34 + hematopoietic cells. It is expressed abundantly in cartilaginous tissues, including mouse inner ear and human articular cartilage[ 7 , 8 ], and it functions in the chondrogenesis and cartilage homeostasis [ 9 , 10 ]. Cytl1 also plays a role in the development and metastasis of neuroblastoma cells [ 11 ]. A comparative modeling study indicated that Cytl1 adopts an IL8-like chemokine fold, similar to the one present in monocyte chemoattractant protein 1 (MCP-1, also known as CCL2). Therefore, Cytl1 might be functionally related to MCP-1 that is known to be involved in the pathogenesis of CF [ 12 – 14 ]. We found that Cytl1 expression was highly elevated in mice with severe CF associated with pressure overload, myocardial infarction (MI), and ischemia-reperfusion (I-R) injury ( Fig 1 ). This led us to pursue the role of Cytl1 in CF.
10.1371/journal.pone.0166480.g001
Fig 1
Cytl1 is upregulated under pathological conditions.
Three groups of WT mice were subjected to TAC for 6 wks (TAC), ligation of coronary artery for 4 wks (MI), or ligation of coronary artery for 30 min followed by reperfusion for 24 hrs (I/R). Hearts were harvested and qRT-PCR was performed to determine the transcript levels of Cytl1. n = 3 for sham, n = 3 for TAC, n = 4 for MI, n = 4 for I/R. * p < 0.05, ** p < 0.01.
In this study, we found that CF was significantly attenuated in cytl1 KO mice upon pressure overload. By contrast, adeno-associated virus (AAV)-mediated overexpression of cytl1 resulted in the development of CF in vivo . Further in vitro experiments suggest that Cytl1 induces CF likely through activating the TGF-β-SMAD signaling pathway. Collectively, our results show that Cytl1 is a pro-fibrotic molecule in the heart. Cytl1 may serve as a therapeutic modality for CF.
Materials and Methods
Materials
Recombinant human TGF-β2 was purchased from PeproTech and used at a final concentration of 10 ng/ml. SB-431542 [ 15 ], an antagonist of the TGF-β receptor 1, was obtained from Sigma-Aldrich, dissolved in DMSO at a concentration of 10 mM and used at a final concentration of 10 μM. Recombinant human MCP-1 (CCL2) was purchased from R&D Systems and used at a final concentration of 20 ng/ml. CAS 445479-97-0 [ 16 ], an antagonist of CCR2, was obtained from Millipore and used at a final concentration of 6 nM.
Animals
The mice were maintained under controlled conditions, and all animal experiments were performed with the approval of the Animal Care Committee of the Gwangju Institute of Science and Technology. The generation of Cytl1 KO mice was previously described [ 10 ]. For surgical models, male mice at 8 wks of age (23–28 g) were anesthetized with 0.5–0.7 ml of a 1x Avertin solution (a mixture of 2,2,2-tribromoethanol and tert-amyl alcohol) administered via intra-peritoneal injection. The mice were ventilated with a tidal volume of 0.1 ml and a respiratory rate of 120 breaths per minute (Harvard Apparatus).
Transverse aortic constriction (TAC)
TAC was performed as previously described [ 17 ]. A longitudinal incision of 2 to 3 mm was made in the proximal sternum to allow visualization of the aortic arch, and the transverse aorta was ligated between the innominate and left common carotid arteries with an overlaid 27-gauge needle. The needle was then immediately removed, leaving a discrete region of constriction. Same-operated animals underwent the same surgical procedures, except that the ligature was not tied.
Myocardial infarction (MI)
The thorax was opened under sterile conditions through a left intercostal thoracotomy and the heart was approached under direct visualization. The animal was slightly rotated to the right to enhance visualization of the left ventricle, and the left auricle was slightly retracted to fully expose the left main coronary artery system. The left anterior descending coronary artery was ligated approximately 2 mm below the tip of the normally positioned left auricle using a 7–0 silk suture. Ischemia was confirmed by discoloration of the ventricle.
Ischemia-reperfusion (I-R)
Myocardial I-R was induced as previously described [ 18 ]. Briefly, the left anterior descending coronary artery was ligated using 7–0 silk sutured approximately 2 mm below the level of the tip of the normally positioned left auricle. Polyethylene (PE) 10 tubing with a diameter of 1 mm was placed on top of the vessel, and the suture was tied. After 30 min of occlusion, reperfusion was established by cutting the knot and removing the PE10 tubing. The chest wall was closed using 5–0 suture. Mice were sacrificed, and their hearts were removed after 24 h of reperfusion.
Histological analysis
The paraffin-embedded heart was cross-sectioned at a thickness of 6 μm. The sections were stained with 0.1% Sirius Red solution (Sigma-Aldrich) for 1 h, and observed under an Axiophot microscope (Carl Zeiss). The fibrotic area was calculated using MetaMorph software (Molecular Devices).
Quantitative real-time (qRT)-PCR
Total RNA was isolated with TRI Reagent (Sigma-Aldrich). Reverse transcription was performed using ImProm II reverse-transcriptase (Promega) with an oligo-dT primer. PCR was performed using an ABI PRISM Sequence Detector System 7500 (Applied Biosystems) with SYBR Green (Takara) as the fluorescent dye and ROX (Takara) as the passive reference dye. The primers used for qRT-PCR were as follows: α-SMA, 5’- ATCGT CCACC GCAAA TGC-3’ and 5’-AAGGA ACTGG AGGCG CTG-3’ ; Collagen 1, 5’–CGAAG GCAAC AGTCG CTTCA-3’ and 5’-GGTCT TGGTG GTTTT GTATT CCAT -3’ ; Cytl1, 5’- CCACC TGCTA CTCTC GGATG-3’ and 5’-CCTCG GGAAT TGGGT CTTC-3’ ; TGF-β2, 5’-TTGCT TCAGC TCCAC AGAGA-3’ and 5’-TGGTT GTAGA GGGCA AGGAC-3’ ; TNF-α, 5’-CATCT TCTCA AAATT CGAGT GACAA-3’ and 5’-TGGGA GTAGA CAAGG TACAA CCC-3’ ;
Western blot analysis
Heart tissue lysates were prepared by homogenization in lysis buffer (50 mM Tris-HCl, 150 mM NaCl, 0.25% Triton X-100, pH 7.4) supplemented with a protease inhibitor cocktail (Boehringer Mannheim). Approximately 50 μg of protein from each sample was separated by SDS-PAGE and transferred to a PVDF membranes (Schleicher & Schuell). The membranes were blocked with 5% non-fat milk and then incubated with primary antibodies at 4°C overnight. Antibodies for TGF-β2 and CD31 were purchased from Abcam, SMAD7 from Invitrogen, vimentin from Santa Cruz Biotechnology, Cytl1, HA, and α-SMA from Sigma-Aldrich, GAPDH, phospho-SMAD2, SMAD2/3, and VE-cadherin from Cell Signaling. The membranes were incubated with secondary antibodies conjugated to horseradish peroxidase (Jackson ImmunoResearch) and then developed with a chemiluminescent substrate (Perkin Elmer).
Recombinant AAV production and injection
Self-complementary AAV (serotype 9) constructs were generated using the pds-AAV2-EGFP vector and the mouse cytl1 cDNA. The recombinant AAV was produced by transfecting 293T cells as previously described [ 19 ]. The AAV particles in the cell culture media were precipitated with ammonium sulfate and purified by ultracentrifugation on an iodixanol gradient. The particles were then concentrated using a centrifugal concentrator. The AAV titer was determined by qRT-PCR and SDS-PAGE. AAV-VLP or AAV-Cytl1 (5 x 10 10 viral genome) was injected into the tail vein of C57/BL6 mice, and the phenotype of the heart was examined after 8 wks.
Cell culture
Human coronary artery endothelial cells (HCAEC) were cultured in EBM-2 bullet kit (Lonza). Adult primary cardiac FBs were isolated as previously described [ 20 ] and cultured in DMEM supplemented with 1% glucose (Gibco BRL).
Recombinant adenovirus production
The AdEasy XL (Stratagene) was used to generate the recombinant adenovirus. Ad-Cytl1 was produced as previously described [ 17 ]. The viral titer was determined by the tissue culture infectious dose method. HCAECs and cardiac FBs were infected with Ad-Cytl1 for 48 h at a multiplicity of infection (moi) of 10–50.
Fluorescent immunostaining
HCAECs and primary cardiac FBs were cultured on 12-chamber slides (Nunc). The cells were fixed with 4% paraformaldehyde for 15 min and blocked with 3–5% BSA for 30 min. The cells were then incubated with Hoechst, anti-CD31 (Santa Cruz), or anti-α-SMA antibody (Sigma-Aldrich) for 1 h 30 min, followed by incubation with FITC-conjugated secondary antibodies for 45 min. The cells were observed under a fluorescence microscope (Olympus).
Statistics
The data were analyzed by Student’s t -test or one-way ANOVA, followed by the Bonferroni post-hoc test using StatView software (version 5.0, SAS). The data were expressed as mean ± SD. A p -value of <0.05 was considered statistically significant.
Results
Cytl1 is upregulated under pathological conditions
Prominent CF was observed in mice with pressure overload induced by TAC for 6 weeks, MI induced by ligation of coronary artery for 4 weeks, or I-R induced by ligation of coronary artery for 30 min followed by reperfusion for 24 h. qRT-PCR showed that the expression level of cytl1 was significantly elevated in these failing hearts ( Fig 1 ) suggesting a role of Cytl1 in CF.
CF is attenuated in cytl1 KO mice
Wild type (WT) and cytl1 KO mice were subjected to TAC for 6 weeks. By Picrosirius staining, extensive CF in both interstitial and perivascular areas of the heart was observed in WT mice, but not in cytl1 KO mice ( Fig 2A ). CF is typically accompanied by the increased expression of pro-fibrotic, fibrotic ECM, and pro-inflammatory markers such as TGF-β2, collagen 1, and TNF-α, respectively. qRT-PCR showed that the mRNA levels of these proteins increased significantly in WT mice; however, this fibrotic response was significantly attenuated in cytl1 KO mice ( Fig 2B ). The TGF-β-SMAD signaling pathway is known to be involved in CF. This signaling pathway was activated in WT mice, as shown by the increased levels of TGF-β2 and phosphorylated SMAD2, and the decreased SMAD7 level. However, it remained unaltered in cytl1 KO mice ( Fig 2C ). These data demonstrate that Cytl1 plays a critical role in CF via the TGF-β-SMAD signaling pathway.
10.1371/journal.pone.0166480.g002
Fig 2
CF is attenuated in cytl1 KO mice.
WT and cytl1 KO mice were subjected to TAC for 6 wks and the extent of fibrosis in the heart was analyzed. (A) Picrosirius staining of heart cross-sections from WT and cytl1 KO mice subjected to TAC. Fibrotic areas in the interstitial and perivascular areas were quantified using MetaMorph software (right panels). (B) Quantification of the mRNA levels of several fibrotic markers (TGF-β2, collagen 1 and TNF-α) by qRT-PCR. (C) Activation of the TGF-β signaling pathway was investigated by western blotting. GAPDH served as the loading control. n = 3–5 for each experimental group. * p < 0.05, ** p < 0.01.
AAV-mediated overexpression of Cytl1 induces CF
We generated a recombinant AAV (serotype 9) that expresses Cytl1 under the control of the CMV promoter (AAV-Cytl1). A two- to three-fold overexpression of Cytl1 was observed in the heart of WT mice at 8 wks after the virus (5 x 10 10 viral genome) delivery ( Fig 3C ). As assessed by Picrosirius staining, the overexpression of Cytl1 significantly induced CF in both interstitial and perivascular areas of the heart ( Fig 3A ). qRT-PCR showed that the mRNA levels of TGF-β2, collagen 1, and TNF-α increased significantly in mice that received AAV-Cytl1 compared to those that received the control virus ( Fig 3B ). Western blotting results showed that Cytl1 activated the TGF-β-SMAD signaling pathway ( Fig 3C ). These data suggest that Cytl1 induces CF via activation of the TGF-β-SMAD signaling pathway.
10.1371/journal.pone.0166480.g003
Fig 3
AAV-mediated overexpression of Cytl1 induces CF.
Control virus or AAV-Cytl1 (5 × 10 10 viral genome) was injected into the tail vein of WT mice, and the phenotype of the heart was examined after 8 wks. (A) Picrosirius staining of heart cross-sections from WT mice injected with control virus or AAV-Cytl. Fibrotic areas in the interstitial and perivascular areas were quantified using MetaMorph software (right panels). (B) Quantification of the mRNA levels of several fibrotic markers (TGF-β2, collagen 1 and TNF-α) by qRT-PCR. (C) Activation of the TGF-β signaling pathway was investigated by western blotting. GAPDH served as the loading control. n = 3–5 for each experimental group. * p < 0.05, ** p < 0.01.
Cytl1 induces EndMT
Endothelial cells contribute significantly to CF through the EndMT. Thus, we tested whether Cytl1 affects this process using a recombinant adenovirus that expresses HA-tagged Cytl1 (Ad-Cytl1). TGF-β2 triggered the EndMT in human coronary artery endothelial cells (HCAECs) as shown by the downregulation of the endothelial marker CD31 and the upregulation of the mesenchymal marker, α-smooth muscle action (α-SMA). These effects of TGF-β2 were completely blocked by co-incubation with SB431542, an antagonist of the TGF-β receptor 1 ( Fig 4A ). These immunostaining experiments revealed that Ad-Cytl1 induced EndMT, which was further supported by the decrease of the endothelial markers CD31 and VE-cadherin and the increase of the mesenchymal marker vimentin ( Fig 4B ). The effects of Ad-Cytl1 were completely blocked by co-incubation with SB431542, implying that Cytl1 exerts it pro-EndMT effects via TGF-β signaling ( Fig 4A ). This conclusion was supported by a western blotting result, in which treatment of Ad-Cytl1 alone increased the levels of TGF-β2 and phosphorylated SMAD2, and decreased the SMAD7 level ( Fig 4C ). qRT-PCR showed that TGF-β2 increased the expression of TGF-β2 itself, collagen 1, and α-SMA, but not that of Cytl1, whereas Ad-Cytl1 increased the expression of Cytl1 itself, TGF-β2, collagen 1, and α-SMA. The effects of TGF-β2 and Cytl1 were blocked by co-incubation with SB431542 ( Fig 3C ). Collectively, these data suggest that Cytl1 induces the EndMT in HCAECs via activation of the TGF-β-SMAD signaling pathway. However, TGF-β2 does not regulate the expression of Cytl1.
10.1371/journal.pone.0166480.g004
Fig 4
Cytl1 induces EndMT.
HCAECs were treated with TGF-β2 (10 ng/ml) or Ad-Cytl1 (50 moi) for 48 h. In selected experiments, the cells were pretreated with the TGF-β receptor 1 antagonist SB431542 (10 μM). (A) The cells were immunostained with antibodies against the endothelial marker CD31 and the MFB marker α-SMA. (B) Lysates of HCAECs infected with Ad-Cytl1 were analyzed by western blotting. CD31 and VE-cadherin served as endothelial cell markers, and vimentin served as the MFB marker. (C) Activation of the TGF-β signaling pathway was investigated by western blotting. GAPDH served as the loading control. (D) Quantification of the mRNA levels of several fibrotic markers (TGF-β2, Cytl1, collagen 1 and TNF-α) by qRT-PCR. n = 3–5 for each experimental group. * p < 0.05, ** p < 0.01.
Cytl1 induces transdifferentiation of FBs to MFBs
FBs that expanded and originated from diverse sources are transdifferentiated to MFBs. The MFBs were contractile due to the increased expression of α-SMA. Both TGF-β2 and Ad-Cytl1 induced the expression of α-SMA in adult primary rat cardiac FBs, which was completely blocked by co-incubation with SB431542 ( Fig 5A ). Ad-Cytl1 activated the TGF-β-SMAD signaling pathway in cardiac FBs, as shown by western blotting ( Fig 5B ). qRT-PCR showed similar effects of TGF-β2 and Ad-Cytl1 on the mRNA levels of TGF-β2, Cytl1, collagen 1, and α-SMA in cardiac FBs similar to HCAECs ( Fig 5C ). These data indicate that Cytl1 induces transdifferentiation of FBs to MFBs via activation of the TGF-β-SMAD signaling pathway.
10.1371/journal.pone.0166480.g005
Fig 5
Cytl1 induces transdifferentiation of FBs to MFBs.
Adult primary cardiac FBs were treated with TGF-β2 (10 ng/ml) or Ad-Cytl1 (50 moi) for 48 h. In selected experiments, the cells were pretreated with SB431542 (10 μM). (A) The cells were immunostained with an antibody against the MFB marker α-SMA. (B) Activation of the TGF-β signaling pathway in primary cardiac FBs infected with Ad-Cytl was investigated by western blotting. GAPDH served as the loading control. (C) Quantification of the mRNA levels of TGF-β2, Cytl1, the MFB marker collagen 1 and α-SMA by qRT-PCR. n = 3–5 for each experimental group. * p < 0.05, ** p < 0.01.
Cytl1 functions independently of CCR2
A comparative modeling study indicated that Cytl1 might exert its function by activating CCR2, the MCP-1 receptor [ 21 ]. Purified MCP-1 induced the expression of β-SMA in adult rat cardiac FBs, which was completely blocked by co-incubation with CAS 445379-97-0, a CCR2 antagonist ( Fig 6A ). These data illustrate that CCR2 exists in cardiac FBs and is involved in CF. By contrast, the effects of Ad-Cytl1 were not affected by CAS 445379-97-0 ( Fig 6A ). These findings were confirmed by qRT-PCR ( Fig 6B ). These data suggest that Cytl1function in cardiac FBs is not mediated by CCR2.
10.1371/journal.pone.0166480.g006
Fig 6
Cytl1 functions independently of CCR2.
Adult primary cardiac FBs were treated with CCL2 (20 ng/ml) or Ad-Cytl1 (50 moi) for 48 h. In selected experiments, the cells were pretreated with the CCR2 antagonist CAS445679-97-0 (6 nM). (A) The cells were immunostained with an antibody against α-SMA. (B) Quantification of the mRNA levels of TGF-β2, Cytl1, collagen 1 and α-SMA by qRT-PCR. n = 3–5 for each experimental group. * p < 0.05, ** p < 0.01.
Discussion
Fibrosis is defined by the excessive accumulation of fibrous connective tissue (i.e., ECM components such as collagen and fibronectin) in and around damaged tissues, which can lead to permanent scarring, organ malfunction and, death [ 22 ]. CF is a hallmark of cardiac remodeling associated with various heart diseases, including hypertension, myocardial infarction, and HF. Cardiac FBs are the most prevalent cell type in the heart and they regulate cardiac function. During CF, however, they undergo proliferation and transdifferentiation to MFBs [ 23 ].
Cytl1 was first identified in CD34 + cells derived from bone marrow and cord blood [ 24 ]. This protein contains a putative signal peptide at its amino terminus and is secreted. With no known functions identified yet, this protein was defined as cytokine-like 1. Our group previously showed that Cytl1 is abundantly expressed in cartilage [ 9 ]. Follow-up studies using cytl1 KO mice revealed that Cytl1 is required for the maintenance of cartilaginous homeostasis [ 10 ]. No gross morphological defects in heart, liver, spleen, and thymus were observed in cytl1 KO mice. CD3, CD19, CD11c, and F4/80 are markers for T cells, B cells, monocytes, and macrophages, respectively. The expression levels of these markers were not significantly altered in cytl1 KO mice as assessed by qRT-PCR, suggesting that immune responses are not affected by Cytl1 at baseline ( S1 Fig ). In this study, we provide compelling evidence that supports the role of Cytl1 in CF via the regulation of TGF-β-SMAD signaling. In particular, we found that Cytl1 acts directly on endothelial cells and FBs, triggering the EndMT and transdifferentiation of FBs to MFBs, the two critical processes that occur during CF.
A molecular modeling study suggested that Cytl1 contains a chemokine-fold similar to that of MCP-1 and other features necessary for signaling through the chemokine receptor CCR2 [ 25 ]. Many CXC-type chemokines, including MCP-1 (CCL2), CCL7, CCL8, and CCL13, are cognate ligands of CCR2 [ 26 ]. While preparing this manuscript, the Han group reported that Cytl1 attracts monocytes/macrophages in in vitro chemotaxis assays. They also showed that Cytl1 exerts this chemotactic activity through a direct binding to CCR2 [ 21 ]. These findings illustrate remarkable biochemical and functional similarities between Cytl1 and MCP-1. However, we observed that antagonizing CCR2 did not inhibit the activity of Cytl1, whereas it completely blocked the activity of MCP-1 for the transdifferentiation of FBs to MFBs. This raises the possibility that more than two different receptors for Cytl1 exist and that a yet-to-be identified receptor(s), but not CCR2, mediates the pro-fibrotic activity of Cytl1 in the heart. This hypothesis remains to be tested in future studies. In addition, we found no evidence to support a possibility that Cytl1 contributes to CF through facilitating immune responses as MCP-1 does. The expression level of markers for diverse immune cells including CD3, CD19, CD11c, and F4/80 was unaltered in cytl1 KO mice ( S2 Fig ) or unaffected by Cytl1 overexpression ( S3 Fig ).
Previous studies suggested that CF plays a causative, but not a secondary, role in myocardial dysfunction [ 27 ]. Consistent with this hypothesis, we observed that myocardial function was significantly preserved in cytl1 KO mice compared to that in WT mice upon pressure overload ( S4 Fig ) and myocardial infarction ( S5 Fig ). In addition, myocardial function was prominently deteriorated in mice that received AAV-Cytl1 compared to those that received control virus ( S6 Fig ). Collectively, these data suggest that abrogation of the Cytl1 signaling might lead to the preservation of myocardial function along with the prevention of CF. Therefore, Cytl1 might provide a novel strategy for the treatment of heart diseases such as CF and HF.
Supporting Information
S1 Fig
No significant chagnes were observed in gross morphology and the expression levels of immune cells-specific markers in cytl1 KO mice.
(A) Heart, liver, spleen, and thymus were harvested from WT and cytl1 KO mice. (B) Whole body weight and weights of organs were measured. (C) Cytl1 is predominantly expressed in non-myocyte cells, as assessed by qRT-PCR. (D) No differences were observed in the expression level of immune cell-specific markers including CD3 (T cells), CD19 (B cells), CD11c (monocytes), and F4/80 (macrophages) in the hearts of WT and cytl1 KO mice. WT, n = 3; cytl1 KO, n = 3. ** p < 0.01, * p < 0.05.
(TIF)
S2 Fig
No significant changes were observed in the expression levels of immune cells-specific markers in cytl1 KO mice under pressure overload.
WT and cytl1 KO mice were subjected to TAC for 6 weeks. (A) Immunohistochemistry showed no differences in the expression of CD11b/c (monocytes) and F4/80 (macrophages) between WT and i cytl1 KO mice. (B) qRT-PCR showed no significant differences in the expression levels of CD3 (T cells), CD19 (B cells), CD11c, and F4/80. WT, n = 3; cytl1 KO, n = 4.
(TIF)
S3 Fig
Cytl1 overexpression did not affect the expression of markers for immune cells.
Injection of AAV-Cytl1 through tail vein induces prominent CF. Hearts were obtained from mice injected with control or AAV-Cytl1. (A) Immunohistochemistry showed no difference in the expression of CD11b/c (a marker for monocytes) and F4/80 (a marker for macrophages) in these mice. (B) Western blotting showed no difference in the expression of CD11b/c and F4/80. AAV-Cytl1 induced about 50% increase in the expression level of Cyt1. (C) qRT-PCR showed no difference in the expression of CD3 (a marker for T cells), CD19 (a marker for B cells), CD11c, and F4/80. Con. n = 3, AAV-Cytl1 n = 3.
(TIF)
S4 Fig
Contractile performance was preserved in cytl1 KO mice after TAC.
Echocardiography was performed at 1, 2, 4 and 6 wks after TAC. LVIDD, left ventricular inter-dimension at diastole; LVIDS, left ventricular inter-dimension at systole; FS, fractional shortening. * p < 0.05.
(TIF)
S5 Fig
Contractile performance was preserved in cytl1 KO mice after myocardial infarction.
Echocardiography was performed at 2 and 4 wks after coronary artery ligation. LVIDD, left ventricular inter-dimension at diastole; LVIDS, left ventricular inter-dimension at systole; FS, fractional shortening. * p < 0.05, ** p < 0.01.
(TIF)
S6 Fig
Contractile performance deteriorated after AAV-mediated overexpression of Cytl1.
Echocardiography was performed at 4 and 8 wks after tail vein injection of AAV-Cytl1. LVIDD, left ventricular inter-dimension at diastole; LVIDS, left ventricular inter-dimension at systole; FS, fractional shortening. ** p < 0.01.
(TIF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Somatic cells can be reprogrammed to a pluripotent state via ectopic expression of the transcription factors Oct4, Sox2, Klf4, and c-Myc, thereby generating induced pluripotent stem cells (iPSCs) [ 1 , 2 ]. Pluripotent stem cells (PSCs), including iPSCs and embryonic stem cells (ESCs), can be classified into two categories: ‘naïve’ and ‘primed’ PSCs [ 3 – 5 ]. Naïve PSCs correspond to ICM of blastocysts, and are similar to mouse ES cells (mESCs), whereas primed PSCs correspond to the epiblast at the postimplantation stage, as represented by mouse epiblast stem cells (mEpiSCs) and human ES cells (hESCs). Naïve PSCs exhibit some distinctive characteristics, such as a compact and dome-shaped morphology, the ability to be passaged as a single cell, dependence of proliferation on the leukemia inhibitory factor (LIF)–Jak/Stat signaling pathway, two active X chromosomes (X a X a ), and specific expression of REX1, ESRRβ, and STELLA [ 3 , 4 ]. In addition, naïve PSCs are more frequent in homologous recombination [ 6 ], and more efficient in repopulating the ICM region after aggregation or injection into host blastocysts; this feature allows them to develop into chimeras that, in turn, can transmit mutated or modified genes to subsequent generations via the germline [ 3 , 7 ]. On the other hand, primed PSCs have a flattened morphology, basic fibroblast growth factor (bFGF)-dependent proliferation, an inactivated X chromosome (X a X i ), and a relatively limited capacity to produce chimeras [ 8 ]. Recent studies demonstrated that primed human PSCs can be converted into a naïve state using certain chemical compounds (a GSK3β inhibitor and a MEK/ERK inhibitor (2i), and the protein kinase A pathway agonist forskolin) [ 9 ]. Thus, naïve PSCs are a feasible potential source of material for production of PSC-derived offspring in domestic species.
Naïve iPSCs and PSCs in large animals, such as cattle, could be applied to genetic improvement of domestic animals as well as preclinical models for human regenerative medicine and disease. To date, however, it has been quite difficult to establish naïve iPSCs and PSCs in mammals other than rodents; most of the iPSCs reported in monkeys [ 10 , 11 ], pigs [ 12 – 15 ], and rabbits [ 16 ] have been of the primed type. In cattle, Sumer et al. [ 17 ] reported the generation of bovine iPSCs (biPSCs), but these cells exhibited characteristics of primed iPSCs. The major reason for this is that the optimal culture conditions for generation and maintenance of iPSCs in various mammalian species have not been determined. Recently, Tsukiyama et al. [ 18 ] proposed an efficient system using piggyBac (PB) transposition of doxycycline (Dox)-inducible transcription factors, which allows the expression of introduced reprogramming factors to be controlled by the presence or absence of Dox, for screening of culture conditions for generation of iPSCs. The PB transposon vector [ 19 – 21 ] is a non-viral and safe vector system, in which integrated constructs (including reprogramming factors) can be removed by the re-expression of transposase (PBase). On the other hand, the source of somatic cells for iPSCs significantly affects reprogramming efficiency [ 22 , 23 ]. For example, mouse neural progenitor cells can be more efficiently reprogrammed than fibroblasts, and iPSCs can be established from these cells by expression of only one or two exogenous factors, due to their endogenous expression of Sox2 and c-Myc [ 23 ]. However, neural progenitor cells are not easily accessible, and are not available in large quantities. By contrast, amnion-derived cells (ADCs) can be readily obtained from the placental tissue after delivery. The amnion, derived from the epiblast as early as the eighth day after fertilization, is a thin membrane-lined cavity filled with fluid that protects the developing fetus. Moreover, in both human and mouse, reprogramming of ADCs into iPSCs is more efficient and faster than that of fibroblasts [ 24 – 26 ]. Therefore, ADCs represent an ideal cell source for the generation of iPSCs.
In this experiment, we attempted to establish biPSCs from bovine amnion-derived cells by introducing Dox-inducible PB vectors expressing the mouse reprogramming factors ( Oct3/4 , Klf4 , Sox2 , and c-Myc ) and culturing cells in the presence of certain chemical compounds (2i and forskolin). Two different types (primed and naïve) of biPSCs appeared under different culture conditions, and we characterized the pluripotent properties of the resultant biPSCs both in vitro and in vivo .
Materials and Methods
Ethics statements
All cattle were fed grass silage-based diet ad libitum . All procedures involving the care and use of animals were approved by the Animal Research Committee of NARO institute of Livestock and Grassland Science.
Isolation of bovine amnion-derived cells (bADCs)
Bovine amnion layer was harvested from a female Japanese black cattle fetus at 50 days of gestation at the National Institute of Livestock and Grassland Science, Japan. The amnion was mechanically peeled away from the chorion and the allantois, divided into small pieces with fine surgical scissors, and dissociated by incubating for 2 hours at 37°C with 0.3% collagenase (Wako) in Dulbecco’s modified Eagle’s medium (DMEM, Invitrogen) containing 10% fetal bovine serum (FBS, JRH Biosciences). After collagenase digestion, the cell suspension was kept at room temperature for 5 min, and then poured through a cell strainer; the filtered suspension was then centrifuged at 200 g for 5 min. The precipitated cells were cultured in DMEM (Sigma) containing 10% FBS, penicillin (Sigma), and streptomycin (Sigma). When the cells reached confluence, they were cryopreserved in liquid nitrogen until use.
Preparation of bovine LIF
Total RNA was isolated from bovine fetal fibroblasts (bFFs) using TRIzol RNA Isolation Reagents (Invitrogen). DNase I (Takara) was added to the RNA preparation to avoid genomic DNA contamination. For reverse transcription, ReverTra Ace (Toyobo) and Random Primer (Invitrogen) were used. Polymerase chain reaction (PCR) was then performed to amplify the bovine LIF (bLIF) cDNA sequence (GenBank accession no. NM_173931.1) using the KOD-Plus- Neo kit (TOYOBO) and specific primers: sense, 5’- GGA GAG CTC CAC CAT GAA GGT CTT GGC GGC AGG -3’; reverse, 5’- AAG GCT AGC CTA GAA GGC CTG GGC CAG CA -3’. The amplified cDNA fragment (630 bp) was inserted into the pCAGGS vector [ 27 ]. Specifically, after digestion of the vector with Sac I and Nhe I (Takara Bio), the cDNA was inserted downstream of the chicken β-actin–based promoter (CAG) in pCAGGS to create the bLIF expression vector. This vector was transfected into COS-7 cells using Lipofectamine LTX (Invitrogen), and the conditioned medium was collected, filtered, and stored at -20°C until use. The conditioned medium was diluted 1:1000 for biPSC culture.
Culture of bADCs and biPSCs
bADCs were maintained on collagen-coated (Nitta Gelatin) dishes in somatic cell medium consisting of DMEM containing 10% FBS, 50 ng/ml epidermal growth factor (EGF, Calbiochem), penicillin, and streptomycin. The cells were dissociated enzymatically with TrypLE Select (Invitrogen) for further propagation.
Primed- and naïve-type biPSCs were generated from bADCs as described below. Primed-type biPSCs resembling hESCs were maintained in primed iPSC medium (piPSC medium) consisting of Dulbecco's Modified Eagle Medium/Nutrient Mixture F-12 (DMEM/F12, Invitrogen) containing 20% Knockout Serum Replacement (KSR, Invitrogen), 2 mM L-glutamine (MP Biomedicals), 1× MEM nonessential amino acids (NEAA, Invitrogen), 0.1 mM 2-mercaptoethanol (2-ME, Wako), penicillin, and streptomycin supplemented with 2.0 μg ⁄mL doxycycline (Dox, Sigma), 5 ng/mL human basic fibroblast growth factor (bFGF, Wako or ReproCELL), and bLIF-conditioned medium (1:1000 dilution), prepared as described above. Cells were subcultured every 7 days by physical splitting using a pulled Pasteur pipette, and maintained on 35-mm-diameter cell culture dishes (IWAKI) on a feeder layer of 3–5 × 10 5 cells SNL cells [ 28 , 29 ] inactivated with mitomycin C (Sigma).
Naïve-type biPSCs were maintained in naïve iPSC medium (niPSC medium), which consisted of piPSC medium supplemented with 1 μM MEK/ERK inhibitor (PD) (PD0325901, REAGENTS DIRECT), 3 μM GSK3β inhibitor (CH) (CHIR99021, REAGENTS DIRECT), and 10 mM forskolin (FK) (REAGENTS DIRECT) in the absence of bFGF. Cells were subcultured every 4 days by enzymatic dissociation using TrypLE Select and maintained on an SNL feeder layer.
Medium lacking bLIF and containing 20 μM JAK Inhibitor I (Calbiochem #420099) was used to test the influence of the LIF signal pathway on self-renewal and survival of biPSCs. Cells were maintained on an SNL feeder layer, and cell number was counted.
All the cultures were maintained in a humidified incubator at 38.5°C with 5% CO 2 in air.
Reprogramming of bADCs
bADCs were plated at 1.25 × 10 5 cells per 35-mm dish in culture medium without antibiotics and incubated overnight. Cells were then transfected using Lipofectamine LTX. Briefly, equal amounts (0.4 μg) of hyPBase vector (pCAG-hyPBase [ 30 ]), PB vectors with reprogramming factors (PB-TET-OKS and pPB-TET-c-Myc [ 18 , 20 , 31 ]), rtTA PB vector (PB-CAG-rtTA Adv, Addgene), and/or TagRFP PB vector (pPBCAG-TagRFP-IH [ 18 ]), 2 μL of Plus reagent (Invitrogen), and 10 μL of Lipofectamine LTX transfection reagent were diluted and mixed in 400 μL of Opti-MEM medium (Invitrogen). DNA–lipid complex was then added to the culture dish. The culture medium was changed 6 hours after transfection. One day after transfection, the culture was supplemented with 2.0 μg ⁄mL Dox. Four days after Dox addition, cells were dissociated with TrypLE Select, 1 × 10 5 cells were reseeded on an SNL feeder layer, and the medium was replaced with piPSC or niPSC medium lacking 2i and forskolin. Eight days after the addition of Dox, 2i and forskolin were added to niPSC medium. Fourteen days after the addition of Dox, primary colonies were mechanically picked and isolated (in the case of primed-type biPSCs) or chemically dissociated (in the case of naïve-type biPSCs) and transferred onto an SNL feeder layer in 48-well or 4-well plates. The medium was changed every 1–2 days, depending on cell growth ( Fig 1 ).
10.1371/journal.pone.0135403.g001
Fig 1
Reprogramming of bovine amnion-derived cells (bADCs) into iPSCs using Dox-inducible PB vectors.
(A) Timeline for the establishment of primed-type biPSC lines. (B) Timeline for the establishment of naïve-type biPSC lines.
The original vectors (PB-TET-MKOS, PB-CAG-rtTA Adv, and pCX-OKS-2A) were obtained from Addgene (plasmids 20959, 20910, and 19771, respectively [ 20 , 31 ]). The empty PB vector and the c-Myc PB vector (pPB-TET-c-Myc) were kind gifts from Dr. Hitoshi Niwa at the RIKEN Center for Developmental Biology. The hyPBase vector (pCMV-hyPBase) was a kind gift from Dr. Keisuke Yusa at Sanger Institute. To generate pCAG-hyPBase, pCMV-hyPBase was cloned into blunt-ended pCAGGS.
Alkaline phosphatase and immunofluorescence staining
Alkaline phosphatase staining was performed using the Vector Alkaline Phosphatase Substrate kit (Vector). For immunofluorescence analysis, cells were fixed with PBS containing 3.7% paraformaldehyde for 10 min at room temperature. After washing with PBS, cells were blocked with 5% bovine serum albumin (Sigma) and 0.1% Triton X-100 (Sigma) for 45 min at room temperature, and then incubated overnight at 4°C with primary antibodies against OCT3/4 (1:25, SC-5279, Santa Cruz Biotechnology), NANOG (1:250, AB5731, Millipore), glial fibrillary acidic protein (GFAP, 1:100, Z0334, DAKO), actin smooth muscle (ASM, 1:1000, MS-113-P0, Thermo), or alpha-fetoprotein (AFP, 1:100, MAB1368, R&D Systems). Alexa Fluor 594 goat anti-mouse IgG or IgM (1:500, Invitrogen), Alexa Fluor 594 goat anti-rabbit IgG (1:500, Invitrogen), Alexa Fluor 488 goat anti-mouse IgG or IgM (1:500, Invitrogen), and Alexa Fluor 488 goat anti-rabbit IgG (1:500, Invitrogen) were used as secondary antibodies. Nuclei were stained with 1 μg ⁄mL Hoechst 33342 (Sigma).
Reverse transcription PCR
Total RNAs of cells were prepared using TRIzol reagent. DNase I was added to preparations to avoid genomic DNA contamination. For reverse transcription, ReverTra Ace and Random Primer were used. PCR was carried out with ExTaq (Takara). An example PCR condition was as follows: 94°C for 2 min, followed by 35 amplification cycles (94°C, 20 s; 60°C, 30 s; 72°C, 30 s). The reaction was terminated by an elongation step at 72°C for 7 min. Primer sequences are shown in S1 Table .
Differentiation of biPSCs in vitro
To produce embryoid bodies (EBs), established biPSCs were harvested by trypsinization (in case of naïve-type biPSCs) or physical splitting (in case of primed-type biPSCs), and then transferred to MPC-treated round-bottom dishes (Nunc) in Iscove’s modified Dulbecco’s medium (IMDM, Invitrogen) containing 15% FBS, 2 mM L-glutamine, 1 mM sodium pyruvate, 1× NEAA, 0.1 mM 2-mercaptoethanol, penicillin, and streptomycin supplemented with 2.0 μg/mL Dox. After 3 days of culture, the medium was changed to fresh medium without Dox. After an additional 3 days of culture, floating cell masses were transferred onto gelatin-coated dishes and cultured in 10% FBS DMEM for another 6 days. The resulting cell culture was analyzed by immunocytochemistry and reverse-transcription PCR, as described above.
Karyotype analysis
Karyotype analysis of the established biPSC lines (pbiPSCs at 65 passages and nbiPSCs at 10 passages) was performed using KaryoMAX Colcemid Solution (Invitrogen) at the time of subculture. For each cell line, 20 metaphase plates were counted.
Generation of chimeric fetuses from naïve-type biPSCs
RFP-expressing naïve-type biPSCs were generated for the analysis of chimera production. bADCs were transfected with a constitutive TagRFP-expressing PB vector (pPBCAG-TagRFP-IH) simultaneously with pCAG-hyPBase PB-TET-OKS, pPB-TET-c-Myc, and PB-CAG-rtTA Adv, and RFP-expressing colonies were picked and propagated.
For aggregation of biPSCs with bovine embryos, in vitro fertilized host embryos were prepared. In vitro maturation, fertilization, and culture were performed as described previously [ 32 ]. Sixty hours after fertilization, 8- to 16-cell stage embryos were collected and transferred to acid Tyrode solution (pH 1.5) until the zona pellucida was completely dissolved. Microwells (WOWs; Vajta et al., 2000) were prepared in four-well dishes (Nunc) by making microspots with an aggregation needle (BLS) filled with 50 mL of modified SOF medium containing 1.5 mM glucose and 10% KSR and supplemented with 2.0 μg/mL Dox, bLIF-conditioned medium (1:1000 dilution), 1 μM PD0325901, 3 μM CHIR99021, and 10 mM forskolin, and then covered with 400 μL of paraffin oil. Subsequently, the zona-free 8- to 16-cell stage embryos were transferred into the WOWs, and 20–30 naïve-type biPSCs were also transferred and co-cultured at 38.5°C in an atmosphere consisting of 5% CO 2 , 5% O 2 , and 90% N 2 . Five days after initiation of co-cultivation, the aggregated embryos were collected. Three aggregated embryos were then transplanted into each uterine horn of a Japanese black cow at 7 days after heat. Transplantation was performed using a catheter (Misawa Medical Industry).
Pregnancy was diagnosed by rectal palpation or ultrasonography. At 90 days after transplantation, the pregnant cow was sacrificed by an overdose of sodium pentobarbital; the uteri were isolated, dissected, and rinsed with PBS. Fetuses were then isolated, rinsed with PBS, and separated by individual organs/tissues including the brain, skin, lung, liver, stomach, small intestine, large intestine, heart, kidney, spleen, muscle, gonad, amnion, and placenta. Genomic DNA (gDNA) was isolated from the tissues by using TRIzol reagent. Genomically integrated Oct3/4-2A-Klf4 sequences in these samples was detected by PCR analysis of isolated gDNA using O-2A-K (Tg) primers ( S1 Table ). PCR products from all tissues were confirmed by DNA sequencing. Frozen tissue sections isolated from the chimera were prepared using OCT compounds, and the tissue sections were immunostained with anti-Vasa (1:200, ab13840, Abcam) and/or anti-RFP (1:500, MM-0172-P, MédiMabs) primary antibodies, and Alexa Fluor 488 goat anti-rabbit IgG (1:500, Invitrogen) and/or Alexa Fluor 594 goat anti-mouse IgG (1:500, Invitrogen) secondary antibodies.
Statistical analysis
Each experiment was repeated at least three times. The values were analyzed using a t-test. p values < 0.05 were considered to be statistically significant.
Results
Generation of bovine iPSCs
bADCs in culture originally exhibited a heterogeneous population consisting of epithelial and fibroblastic cells ( Fig 2A ). After transfection with Dox-inducible piggyBac vectors containing reprogramming factors, the bADCS were cultured in piPSC medium, which was similar to medium generally used for maintenance of human ES cells. Around 8 days after Dox addition, primary colonies appeared, and flattened colonies (human ES cell-like) emerged at day 14 ( Fig 2B ). The efficiency of colony appearance at day 14, relative to the total number of cells reseeded onto the SNL feeder layer at day 5, was 0.01%. These colonies (pbiPSCs) were then picked mechanically and transferred onto a fresh SNL feeder layer. pbiPSCs could be stably propagated and subcultured every 7 days over 70 passages in piPSC medium ( Fig 2C ).
10.1371/journal.pone.0135403.g002
Fig 2
Phase-contrast images of biPSCs established in two different culture conditions.
(A) bADCs. (B) Primary colonies appearing in primed cell-culture medium. (C) Established primed-type biPSCs. (D) Colonies converted from the primed to naïve state. (E) Primary colonies appearing in niPSCs medium. (F) Established naïve-type biPSCs. (A)–(C), (E), scale bars = 500 μm. (D), (F), scale bars = 100 μm.
Addition of 2i and forskolin to culture medium can support naïve characteristics of human iPSCs [ 9 ]. In this study, pbiPSCs were dissociated enzymatically and transferred to niPSC medium containing KSR, bLIF, 2i, and forskolin. These cells proliferated and formed mouse ES cell-like colonies with 3-dimensional morphology ( Fig 2D ). Cells converted from the primed to the naïve state (pnbiPSCs) were maintained in niPSC medium by at least 15 rounds of trypsinization and single-cell dissociation. Furthermore, when the transfected cells were directly cultured in niPSC medium from 8 days after Dox addition, when primary colonies appeared, nbiPSCs colonies with dome-shape and compact morphology emerged at day 14 ( Fig 2E ). The efficiency of colony appearance at day 14 against the total number of cells, which were reseeded on the SNL feeder layer at day 4, was 0.01%. After passage of nbiPSCs by trypsinization and reseeding onto a fresh SNL feeder layer, cells could be maintained for at least 15 passages by single-cell dissociation every 4 days ( Fig 2F ). When naïve-type biPSCs (pnbiPSCs and nbiPSCs) were transferred to piPSC medium, they reverted to flattened hESC-like colony morphology, and could be maintained in this medium (retaining the flattened morphology) for over 10 passages.
To test which culture compounds were essential for maintaining naïve characteristics, pnbiPSCs were cultured in different culture conditions, in the presence or absence of GSK3 inhibitor (CH), Mek/Erk inhibitor (PD), and/or forskolin (FK) ( S1 Fig ). In the absence of all of these compounds ( S1A Fig ), or in the presence of FK alone ( S1D Fig ), colonies became more flattened. In the presence of PD, a subset of colonies retained the three-dimensional morphology ( S1B Fig ), whereas in the presence of CH, most of the colonies had the naïve-type morphology ( S1C Fig ). Cell number was elevated in the presence of FK ( S1I Fig ).
Characterization of primed-type and naïve-type biPSCs
Both primed-type and naïve-type biPSCs were positive for alkaline phosphatase (AP) activity ( Fig 3A, 3B and S2A, S2H Fig ). In the karyotype analysis, 19 out of 20 (95%) pbiPSCs or nbiPSCs with metaphase plates had the normal number (2n = 60) of chromosomes, even after they were propagated over 65 passages ( Fig 3C–3E ); one pbiPSC had 59 chromosomes, and one nbiPSC had 61 chromosomes.
10.1371/journal.pone.0135403.g003
Fig 3
Characterization of biPSCs.
(A) Alkaline phosphatase activity in primed-type iPSCs (pbiPSCs). (B) Alkaline phosphatase activity in naïve-type iPSCs derived from pbiPSCs (pnbiPSCs). (C) Karyotyping image of pbiPSCs at passage 65. (D) Karyotyping image of nbiPSCs at passage 10. (E) Proportion of cells with the indicated number of chromosomes (n = 20). (F)–(I) OCT3⁄4 (F, OCT3⁄4 staining; G, Hoechst staining) and NANOG (H, NANOG staining; I, Hoechst staining) expression in pbiPSCs. (J)–(M) OCT3⁄4 (J, OCT3⁄4 staining; K, Hoechst staining) and NANOG (L, NANOG staining; M, Hoechst staining) expression in pnbiPSCs. (A), (B), (J)–(M), scale bars = 100 μm. (F)–(G), scale bars = 500 μm.
Immunocytochemistry analysis revealed that biPSCs expressed exogenous and/or endogenous OCT3/4 and NANOG ( Fig 3F–3I, 3J–3M , and S2B–S2G, S2I–S2L Fig ). RT-PCR analysis also revealed that these cells expressed the pluripotency-related genes including OCT3/4 , SOX2 , NANOG , E-CADHERIN ( CDH1 ), as well as naïve mouse iPSC–specific genes such as REX1 , ESRRβ , STELLA , LIFR , and SOCS 3 ( Fig 4 ). On the other hand, FGF5 , and OTX2 , a marker gene for primed human iPSCs, was only detected in pbiPSCs ( Fig 4 ).
10.1371/journal.pone.0135403.g004
Fig 4
Endogenous and exogenous expression of genes specific to undifferentiated ESCs in biPSCs.
mRNA expression was evaluated by reverse-transcription polymerase chain reaction (RT-PCR). pbiPSCs (P5), primed-type iPSCs at passage 5; pbiPSCs (P50), primed-type iPSCs at passage 50; pnbiPSCs, naïve-type iPSCs at passage 3 converted from primed-type iPSCs at passage 48; nbiPSCs, naïve-type iPSCs cultured under naïve medium from primary culture; bADCs, bovine amnion-derived cells; SNL feeder, SNL feeder cells; vector, plasmid DNA of PB vectors; bACT, bovine β-ACTIN specific for cattle; Uni-ACT, universal β-ACTIN that reacts with both cattle and mice.
To investigate the signaling pathways required for the maintenance of self-renewal of biPSCs, naïve-type biPSCs were cultured with JAK inhibitor in the absence of bLIF for 4 days. Under these culture conditions, the number of cells was significantly reduced, by more than 60% in pnbiPSCs ( Fig 5A, 5B and 5C ) and 50% in nbiPSCs ( S2M and S2N Fig and Fig 5C ). By contrast, judging by their appearance, the proliferation of pbiPSCs was not affected in this culture condition ( S2O and S2P Fig ).
10.1371/journal.pone.0135403.g005
Fig 5
Naïve-type features of iPSCs.
(A) pnbiPSCs cultured for 4 days in niPSC medium in the presence of JAK inhibitor. (B) pnbiPSCs cultured in the presence of DMSO. (C) The number of cells cultured in the presence of JAK inhibitor or DMSO (*p < 0.05). (D) XIST expression evaluated in pbiPSCs, but not in pnbiPSCs. Immunocytochemistry images of methylation status at H3K27me3 sites (E, pbiPSCs; F, Hoechst staining; G, pnbiPSCs; H, Hoechst staining). Arrowheads indicate puncta of H3K27me3. (A), (B), scale bars = 500 μm.
Next, we examined X-chromosome inactivation states in the established biPSC cell lines. RT-PCR analysis revealed that XIST mRNA was not expressed in pnbiPSCs, but it was expressed in pbiPSCs and female bADCs (used as a positive control) ( Fig 5D ). On the other hand, histone H3K27 trimethylation signals associated with X chromosomes inactivation were observed in pbiPSCs, but not in pnbiPSCs ( Fig 5E–5H ).
When biPSCs were cultured in the absence of Dox, they readily changed their morphology and differentiated ( S3A and S3B Fig ). RT-PCR analysis revealed that in the absence of Dox, pbiPSCs and pnbiPSCs no longer expressed exogenous transgenes and endogenous pluripotent genes, such as OCT3/4 , ESRRβ , and STELLA , but instead began to express endogenous c-MYC ( S3C Fig ).
Differentiation potential of biPSCs
When biPSCs were induced to differentiate in low-adhesion culture dishes for 6 days, they formed embryoid bodies ( Fig 6A, 6E and S4A, S4B and S4M Fig ). After they were cultured for another 6 days, they differentiated into all three germ layers, as assessed by immunocytochemistry using specific markers for each layer: GFAP for the ectoderm, ASM for the mesoderm, and AFP for the endoderm ( Fig 6B–6D, 6F–6H and S4C–S4K, S4N–S4S Fig ). We also assessed the in vitro differentiation potential of pnbiPSCs by RT-PCR using specific markers such as VIMENTIN ( VIM ) for ectoderm, BMP4 for mesoderm, and AFP for endoderm ( S4L Fig ).
10.1371/journal.pone.0135403.g006
Fig 6
Differentiation potential of biPSCs in culture.
(A) Embryoid body formation of pbiPSCs grown for 6 days in low cell-adhesion dishes. Immunocytochemical staining for markers for the three germ-layer in differentiated cells derived from pbiPSCs. α-fetoprotein (B, endoderm), actin smooth muscle (C, mesoderm), and glial fibrillary acidic protein (D, ectoderm) were used as markers. (E) Embryoid body formation by pnbiPSCs. Immunocytochemical staining for α-fetoprotein (F), actin smooth muscle (G), glial fibrillary acidic protein (H). (A), (E), scale bars = 500 μm. (B)–(D), (F)–(H), scale bars = 100 μm.
To test whether naïve-type biPSCs had the potential to associate with the normal development of bovine embryos, we generated naïve-type biPSCs expressing Tag-RFP and aggregated them with 8- to 16-cell stage embryos. In preliminary experiments, pbiPSCs were allowed to aggregate with 8- to 16-cell stage embryos by two different ways of cell dissociation. When we trypsinized pbiPSCs and aggregated them with embryos, the cells were scattered around and didn’t aggregate with embryonic cells. On the other hand, when we mechanically split pbiPSCs and aggregated with embryos, cell clumps were formed outside embryos, however, didn’t make aggregation. Meanwhile, naïve-type biPSCs developed normally into blastocysts, to the same degree as aggregation performed only with 8- to 16-cell stage embryos ( Table 1 ). Judging from RFP fluorescence, pnbiPSCs and nbiPSCs were successfully incorporated into the ICM region of blastocyst stage embryos ( Fig 7A ). In addition, we sometimes observed incorporation into both the ICM and TE regions (13.9% and 34.4% in the case of pnbiPSCs and nbiPSCs, respectively; Table 1 ). By contrast, we did not observe incorporation of bADCs expressing RFP into the embryos ( Table 1 ).
10.1371/journal.pone.0135403.g007
Fig 7
Production of chimeric fetuses from bovine embryos using the aggregation method.
(A) Naïve-type biPSCs expressing Tag-RFP were aggregated with host at the 8- to 16-cell stage of in vitro fertilized embryos. (B) Chimeric fetuses at day 90 of gestation derived from aggregated embryos. (C) PCR analysis using transgene-specific primers for genomically integrated Oct3/4-2A-Klf4 sequences in 14 tissues. Genomic DNA isolated from pnbiPSCs was used as a positive control. H 2 O was used as a negative control (buffer alone for RT-PCR). Immunofluorescence analysis showing the distribution of pnbiPSC-derived cells (RFP-positive with red signals) in the small intestine (D), placenta (E), gonad (F, VASA-positive cells with green signals; arrowheads indicate the portion that is double-positive for RFP and VASA), and kidney (G) of the chimeric fetus. Nuclei were stained with DAPI (blue).
10.1371/journal.pone.0135403.t001
Table 1 Aggregation of pnbiPSCs into in vitro fertilized embryos, and their development in vitro .
Chimeric blastocysts (%)
Donor cells
No. of aggregated embryos
No. of blastcysts developed (%)
No. contributed to ICM
No. contributed to both ICM and TE
No. contributed to TE
pnbiPSCs
60
38 (63.3)
31 (86.1)
5 (13.9)
0 (0)
nbiPSCs
60
35 (58.3)
21 (65.6)
11 (34.4)
0 (0)
bADCs *
60
35 (58.3)
0 (0)
0 (0)
0 (0)
Embryo only
60
34 (56.7)
0 (0)
0 (0)
0 (0)
*RFP vector was introduced
Chimeric embryos formed by aggregation of 8- to 16-cell bovine embryos and pnbiPSCs were subsequently transferred to the uteruses of a surrogate mother in order to trace their developmental capacity in vivo . Three chimeric embryos were transferred into each uterine horn, and the resultant fetuses were recovered at 90 days after transplantation ( Fig 7B ). All of the fetuses developed normally. The fetuses were then isolated and separated by individual organs/tissues representing ectoderm (brain, skin), endoderm (lung, liver, stomach, small intestine, large intestine), mesoderm (heart, kidney, spleen, muscle), germline (gonad), fetal membrane (amnion) and trophectoderm (placenta). PCR analysis of integrated Oct3/4-2A-Klf4 sequences were performed in each fetal tissue. Eleven out of 14 tissues (brain, skin, lung, stomach, small intestine, large intestine, spleen, muscle, gonad, amnion, and placenta) showed evidence of a pnbiPSC contribution ( Fig 7C ).
PCR analysis revealed the contribution of pnbiPSCs to chimeric fetuses. In addition, we performed immunofluorescence analysis on frozen tissue sections of small intestine, kidney, placenta, and gonad. Red fluorescence from TagRFP-expressing vector was observed in the small intestine, placenta, and gonad ( Fig 7D, 7E and 7F ). By contrast, fluorescence was faint or absent in the kidney ( Fig 7G ). These immunofluorescence results were coincident with those of PCR analysis. Moreover, in the gonad, some of the cells were double-positive for RFP (red) and VASA (green), a primordial germ cell marker ( Fig 7F , arrowheads).
Discussion
Since the first generation of iPSCs in mouse [ 1 ], a large number of studies of iPSCs have been performed in both mouse and human. Despite numerous attempts, however, very few studies have reported the generation of iPSCs from somatic cells or ESCs from preimplantation embryos in other species. In cattle, Cao et al. [ 33 ], Ozawa et al. [ 34 ], and Furusawa et al. [ 35 ] reported the generation of bovine embryonic stem cell–like cells, and Sumer et al [ 17 ] reported the generation of primed biPSCs, but also described the difficulty of derivation and maintenance of these cells. Most iPSCs reported in non-rodent species were primed iPSCs [ 10 – 17 ], which have a limited capacity to produce chimeras relative to naïve iPSCs [ 3 , 4 , 8 ]. Here, we report the generation of two different types (i.e., primed and naïve) of biPSCs established from bADCs by introducing Dox-inducible PB vectors. Both types had characteristics of pluripotent stem cells; in particular, the naïve-type biPSCs exhibited several characteristics of pluripotent cells, comparable to those of naïve mouse iPSCs, and contributed to the ICM of bovine host blastocysts and chimeric fetuses.
Sumer et al. [ 17 ] reported previously that generation of biPSCs by retroviral delivery and the ectopic expression of POU5F1 , SOX2 , KLF4 , c-MYC , and NANOG , and demonstrated that ectopic expression of NANOG is necessary for the generation and maintenance of biPSCs from bovine fetal fibroblasts. In this study, we were able to generate biPSCs without ectopic expression of NANOG , possibly because the bADCs used in this study expressed intrinsic endogenous NANOG ( Fig 4 ). Therefore, ectopic expression of NANOG may not be necessary for production and maintenance of biPSCs. In addition, studies on human and mouse iPSCs have suggested that reprogramming via introduction of transcription factors in ADCs is more efficient and faster than in fibroblasts [ 24 – 26 ] probably because ADCs in mouse and human express high endogenous levels of Klf4 , c-Myc , and Ronin , which support proliferation and self-renewal of iPSCs [ 25 ]. When we used bovine embryonic fibroblast cells and transfected them with the same combination of transcription factors, iPSC-like colonies appeared, but no stable iPSC lines were established (data not shown). Moreover, when we cultured bADCs in low-adhesion culture dishes for 6 days, they formed EB-like cell masses and differentiated a part of cell types representing each germ layer after culturing on gelatin-coated dishes for another 6 days. A previous report [ 36 ] showed that a subpopulation of human ADCs exhibited stem cell-like characteristics and an ability to differentiate into various cell types in tridermic layer. In our study, bADCs also exhibited the expression of NANOG as well as several lineage specification markers such as VIM (ectoderm marker) and BMP4 (mesoderm marker). Thus, bADCs have stem cell-like characteristics and represent an appropriate cell source for generation of iPSCs in cattle.
The choice of reprogramming components, and the order of these factors in the vector, also affect the efficiency of reprogramming of somatic cells and the generation of iPSCs. Okita et al. [ 31 ] reported that expression of Oct3/4, Klf4, and then Sox2 (OKS) in that order in polycistronic vectors improves efficiency of reprogramming and facilitates the generation of murine iPSCs. Tsukiyama et al. [ 18 ] reported that cell reprogramming induced by the combination of OKS and c-Myc (M) vectors (OKS + M) was more efficient than that induced by polycistronic MKOS vectors. When we used MKOS vectors instead of OKS + M vectors, no iPSC-like colonies appeared (data not shown), whereas OKS + M vectors could generate iPSC-like colonies with an efficiency of 0.01%. Therefore, the OKS + M vector combination is also useful for generation of iPSCs in cattle.
The biPSCs reported by Sumer et al. [ 17 ] appeared in the primed state, as judged by their morphology and the use of a culture medium containing FGF. In this study, when the primed-type biPSCs were passaged and maintained in the niPSCs medium containing bLIF, 2i, and forskolin, naïve-type biPSCs appeared. These cells exhibited several hallmarks of naïve PSCs, such as the expression of naïve marker genes (including the STAT3 target SOCS3 [ 37 ]) LIF-dependent proliferation [ 3 , 4 ], and reactivation of the X chromosome (X a X a ) [ 3 , 4 , 9 ]. In humans, primed iPSCs can be converted into naïve iPSCs by the addition of 2i and forskolin to the culture medium [ 9 ]. In our experiment, once naïve-type cells appeared, only the addition of bLIF + GSK3β inhibitor could maintain them in this state ( S1C Fig ), indicating that the continuous expression of transcription factors in naïve-type cells could substitute for the effects of Mek/Erk inhibitor and forskolin. On the other hand, our primed-type biPSCs exhibited a flattened morphology, LIF-independent proliferation, inactivation of the X chromosome (X a X i ), and the expression of primed marker genes such as FGF5 ( Fig 4 ); however, they also expressed naïve marker genes such as REX1 , ESRRβ , STELLA , LIFR , and SOCS3 ( Fig 4 ). Recent reports describe intermediate states of cells of murine ESCs or iPSCs that share characteristics with the primed- and naïve-cell states [ 38 , 39 ], even in primed culture conditions [ 39 ]. These intermediate cells express both primed marker genes, such as Fgf5 , and naïve marker genes such as Rex1 , Esrrβ , and Stella . Tsukiyama et al. [ 38 ] also showed that intermediate cells do not depend on the LIF-Jak/Stat pathway for proliferation. These data are in agreement with our findings; hence, our primed-type biPSCs should be considered as similar to intermediate cells.
In the absence of Dox, biPSCs readily changed their morphology and no longer expressed exogenous transgenes and endogenous pluripotent genes. Instead, they began to express endogenous c-MYC ( S3C Fig ). These results indicated that transgene expression could be controlled by the removal of Dox, and that continuous transgene expression is necessary to maintain biPSCs in a pluripotent state. Optimal culture conditions for the establishment and maintenance of ESCs vary among species [ 40 – 42 ]. Most of the iPSC lines established in non-rodent species depend on continuous transgene expression to maintain their pluripotency [ 12 – 14 , 17 , 43 ]. More recently, naïve human ESCs or iPSCs can be established without transgene expression in the presence of LIF, FGF2, and TGFβ1 (a member of the TGF-β superfamily, and the same as activin A) and inhibitors against four signaling pathways (ERK1/2, GSK3β, JNK, and p38) for stable cell propagation [ 44 ]. Therefore, further studies are required in order to determine the optimal culture conditions for maintaining established biPSCs in the absence of Dox. This study provides a model for generating authentic naïve-type biPSCs under the control of transgenes.
The biPSCs we established exhibited many features of pluripotency; however, they did not form teratomas in nude mice (BALB⁄c nu/nu, data not shown). The difficulty of developing mature teratomas from naïve-like human iPSCs has been described previously [ 6 , 45 , 46 ], and the specific strain of immunodeficient mouse affects teratoma formation [ 47 ]. On the other hand, biPSCs could form embryoid bodies and differentiate into all three germ layers in vitro . In addition, they were incorporated into ICM, and sometimes into TE, after aggregation with 8- to 16-cell stage embryos. These observations led us to examine the potential of biPSCs to contribute to chimeric fetuses. Although primed-type biPSCs could propagate stably for more than 70 passages, naïve-type biPSCs were somewhat difficult to maintain, and could be propagated for only 10 to 15 passages; therefore, pnbiPSCs (derived from pbiPSCs) were employed for aggregation and transplantation. Aggregated embryos transplanted into the uteruses of surrogate mothers successfully produced three chimeric fetuses. Although the contribution of pnbiPSCs to some tissues was faint or absent, these cells had the potential to differentiate into all three germ layers, trophectoderm, and potentially germline in vivo . Incorporation into the trophectoderm has not been observed in mice, but has been frequently observed in pigs [ 48 ]. This difference is possibly due to interspecies diversity in OCT3/4 regulation, resulting in changes in Cdx2:Oct4 ratios and the relative timing of trophectoderm commitment observed not only in cattle but also in human, pig, and rabbit blastocysts [ 49 ]. Our study provides the first demonstration that biPSCs can contribute to chimeric fetuses and differentiate into all tissues, including extraembryonic tissues. The next step to be achieved is the production of adult chimeras and offspring.
In summary, we generated two different types of biPSCs from bADCs using Dox-inducible PB vectors. Our results show for the first time that biPSCs meet the criteria for naïve iPSCs and have the capacity to contribute to chimeric fetuses, including the germ cell lineage. These cell lines should facilitate the optimization of culture conditions for generation and and maintenance of bovine iPSCs that have the potential to generate chimeric offspring.
Supporting Information
S1 Fig
Appearance of naïve-type biPSCs cultured under different conditions.
pnbiPSCs cultured only with bLIF (A), bLIF+CH (B), bLIF+PD (C), bLIF+FK (D), LIF+CH+PD (E), bLIF+PD+FK (F), bLIF+FK+CH (G), and bLIF+FK+CH+FK (H). (I) Numbers of growing cells after cultivation in different conditions for 4 days. (A)–(H), scale bars = 500 μm.
(TIFF)
S2 Fig
Characterization of biPSCs.
(A) Alkaline phosphatase activity in nbiPSCs. OCT3⁄4 (B, RFP-positive image; C, OCT3⁄4 staining; D, Hoechst staining) and NANOG (E, RFP-positive image; F, NANOG staining; G, Hoechst staining) expression in nbiPSCs expressing RFP. (H) Alkaline phosphatase activity in bADCs. OCT3⁄4 (I, OCT3⁄4 staining; J, Hoechst staining) and NANOG (K, NANOG staining; L, Hoechst staining) expression in bADCs. (M) nbiPSCs cultured in the presence of JAK inhibitor for 4 days. (N) nbiPSCs cultured in the presence of DMSO. (O) pbiPSCs cultured in the presence of JAK inhibitor for 4 days. (P) pbiPSCs cultured in the presence of DMSO. (A)–(L) scale bars = 100 μm. (M)–(P) scale bars = 500 μm.
(TIF)
S3 Fig
biPSCs cultured in the absence of Dox (- Dox).
(A) Phase-contrast image of pnbiPSCs cultured in the absence of Dox for 4 days. (B) Phase-contrast image of pbiPSCs cultured in the absence of Dox for 7 days. (C) Endogenous and exogenous gene expression in biPSCs cultured in the absence of Dox. (A), (B), scale bars = 500 μm.
(TIFF)
S4 Fig
Differentiation potential of nbiPSCs expressing RFP.
(A) Embryoid body formation of nbiPSCs grown in low cell-adhesion dishes for 6 days. Glial fibrillary acidic protein (C, RFP-positive image; D, glial fibrillary acidic protein; E, Hoechst staining), actin smooth muscle (F, RFP-positive image; G, actin smooth muscle; H, Hoechst staining) and α-fetoprotein (I, RFP-positive image; J, α-fetoprotein; K, Hoechst staining) were used for the markers. (L) Gene-expression profile of pnbiPSCs after embryoid body differentiation. EB, Embryoid body; EB dif, EB-derived cells cultured for an additional 6 days on a gelatin-coated dish; VIM, VIMENTIN. (M) Embryoid body formation of bADCs grown in low cell-adhesion dishes for 6 days. Glial fibrillary acidic protein (N, glial fibrillary acidic protein; O, Hoechst staining), actin smooth muscle (P, actin smooth muscle; Q, Hoechst staining) and α-fetoprotein (R, α-fetoprotein; S, Hoechst staining) were used for the markers. (A), (B), (M) scale bars = 500 μm. (C)–(S) scale bars = 70 μm.
(TIF)
S1 Table
Primer sequences.
(XLSX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Disrupting the ion channels responsible for depolarisation and repolarisation of cardiomyocytes by gene mutation or off-target drug effect increases the risk of sudden arrhythmic death [ 1 ]. This pillar of safety pharmacology is principally studied using animal explant material and manual approaches requiring direct contact with the cell. Improvements to increase human relevance and experimental throughput are currently being evaluated in the Comprehensive In Vitro Proarrhythmia Assay (CiPA) initiative [ 2 ]. This multi-disciplinary industry, academia, and regulatory collaboration is testing strategies (e.g. automated patching platforms, in silico modelling), and substrates (e.g. human stem cell derived cardiomyocytes (hSC-CM) with the higher through-put methods that have been used to study repolarization changes (e.g. Multi-electrode array, or optical recordings of voltage or calcium) in response to known drugs or mutations [ 3 ].
The time taken for ventricular cells to depolarise and repolarise can be measured at the bedside by the QT interval on an electrocardiogram. Abnormal QT duration reveals inherited (Long/Short QT) or acquired disease states. Since medications were shown to cause QT prolongation, and sudden death [ 4 ], evolution of safety pharmacology and statutory medicines regulatory frameworks [ 5 ] limit such entities reaching the clinic. Detection of QT prolongation pre-clinically requires measurements reflecting the action potential duration (APD). Since the APD is influenced by beat frequency [ 6 ] methods to control the cell are also needed. Normally this is achieved with electrical stimulation and patch clamp. These require contact with the cell and with the dish, constraining throughput. Automated platforms offer modest improvement generating 150–3000 data points a day [ 7 ]. Adopting approaches to measure, and control, cells with light may eliminate the complexity required of patching, or the variability and toxicity of electrical field stimulation [ 8 ] and thus liberate the need for contact which constrains scalability.
Optogenetic tools, in place of electrical stimulation, can impose control on whole hearts [ 9 , 10 ], monolayer cultures of neonatal rodent cardiomyocytes [ 11 – 13 ] or hSC-CM’s [ 14 , 15 ]. Since genetically encoded indicators can be used to phenotype voltage or calcium transients in hSC-CM’s with less toxicity than chemical dyes [ 16 , 17 ] we considered combining approaches into an all-optical-all-genetic single cell assay as previously reported for cultured single neurons using compatible control/indicator pairs for voltage [ 18 ] or calcium [ 19 – 21 ]. In contrast to genetically encoded calcium indicators, equivalent voltage tools have a smaller dynamic range and signal to noise ratio, require cofactor (eg retinal) addition which may have biological consequences in SC-CM’s [ 13 , 22 ], and high power illumination which may limit observation duration. Whilst the calcium transient duration (CTD) is only a surrogate for the voltage changes that defines the APD; this integrated response from multiple currents and pumps closely reflects the cellular APD [ 23 ]. Hence we asked if optical control with calcium imaging may be possible, and useful, in single cells directly avoiding the need for monolayers to make single cell measurements.
Materials and methods
Construct and virus production
ChETA TC -5xMyc linked by a 2A peptide to R-GECO1 was codon optimised for human cells and synthesised (Genscript, Picastaway, NJ, US) and subcloned into the pDUAL backbone for commercial Adenoviral production (Vector Biolabs, Malvern, PA, US). Sequences are provided in S1 Appendix .
Primary cardiomyocyte isolation
This investigation was approved by the Animal Welfare and Ethical Review Board at the University of Oxford and conforms to the UK Animals (Scientific Procedures) Act, 1986, incorporating Directive 2010/63/EU of the European Parliament. Adult guinea pigs were obtained from Harlan UK and sacrificed by cervical dislocation. Guinea pig left ventricular myocytes were isolated using an enzymatic digestion technique. Animals were culled via cervical dislocation and the heart was submitted to retrograde perfusion with a Ca 2+ free buffer for 3 minutes followed by a 1mg/ml Collagenase type II solution (250 units/mg, Worthington Biochemical Corporation) for 9 minutes. The digested tissue was subsequently mechanically agitated in enzymatic solution for an additional 5 minutes and the cells were collected by centrifugation at low speed (500 rpm).
ChETA TC -myc and R-GECO imaging in fixed cells
Adult ventricular cardiomyocytes were infected at an MOI of 5 as for live imaging and kept in storage solution in a humidified incubator at 37°C, 5% CO2. At 48 hours an aliquot was taken and paraformaldehyde to a final concentration of 4% was added, at 10 minutes, cells were pelleted (200rpm, 2min) in a benchtop centrifuge, washed with 1xPBS and resuspended in PBS. Cells were spun (Cytospin, Thermo Scientific) onto glass slides, and ringed with a PAP pen (Sigma). Permeabilisation with 0.1% Triton-X-100 in Tris Buffered saline for 10 minutes at room temperature was followed by blocking (0.2% albumin in permeabilisation buffer) for 20 minutes. Primary antibodies (mouse monoclonal 9E10 anti-myc (Santa-Cruz), and rabbit polyclonal anti-DsRed (Clontech) were diluted 1:200 in blocking buffer. Three hours after primary incubation cells were washed in permeabilisation buffer, and counter stained with Alexa-488 anti-mouse, and Alexa-568 anti-rabbit fab fragment secondaries (Invitrogen), nuclear counterstaining was with Topro3 (Invitrogen), for an hour before washing and mounting (Vectashield, Vector labs). Images were acquired on a Leica SP5 confocal microscope with a 63x oil immersion lens. hSC-CM’s were plated onto 0 thickness coverglass, infected at an MOI of 5. Cells were fixed and stained 48 hours after infection as above.
Live cell imaging microscope
An inverted IX81 frame (Olympus, Japan), with a custom 7 LED array (Cairn Research, Faversham, UK), automated stage (Prior Scientific, Cambridge, UK), lens turret, and filter wheel is housed in a custom heated, humidified chamber (Solent Scientific, Segensworth, UK) with image collection on two C-1900 EMCCD cameras (Hamamatsu, Japan) mounted on a beam splitter (Photometrics, UK). Single camera mode image acquisition through an Olympus PlanApo N 60x oil objective lens (NA 1.42) using CellR software (Olympus) was used. A 30mW 405nm blue diode laser FRAP module (Olympus) was used at 10% power for cell stimulation, controlled through CellR. Software conflicts allowed a maximum programmed stimulation rate within CellR of 2.4Hz (144 beats per minute), hence factors of 2.4Hz were used in the stimulation series.
Cardiomyocyte optical control and phenotyping
Primary cardiomyocytes were seeded onto coverslips and mounted in a Ludin chamber (Life Imaging Services GmbH, Switzerland) with Tyrode’s buffer. Electrical stimulation was achieved with a custom modification to the RC-37WS electrode (Warner Instruments) and myopacer unit (Ionoptix). Optical stimulation was achieved with the 405nm laser light at a power density of 1.17mW/mm 2 . Uninterrupted recording of the calcium transient was achieved with a 5msec 568nm LED illumination pulse. The RFP filter set (DS/FF01-560/25-25, T565lpxr dichroic mirror, and ET620/60 emission filter) was used to observe R-GECO.
Stem cell derived cardiomyocyte optical control and calcium imaging
Human iPS derived cardiomyocytes and maintenance media were purchased from Axol Bioscience, (Cambridge, UK), and handled according to manufacturer’s recommendations at 37°C, 5% CO2 in a humidified incubator. After a week in culture they were singularised and re-plated onto Fibronectin (0.5%)/Gelatin (0.1%) coated 24-well MatTek plates (Ashland, MA, US) at 10,000 cells/well. Viral transduction at an MOI of 5 for 24 hours preceded imaging by 2 days. Results shown are obtained from a minimum of three separate batches of cells.
Stem cell derived cardiomyocyte and chemical calcium dye imaging
iPS derived cardiomyocytes were plated out as above, and then loaded with 5μM Fluo-4-AM (Thermo-Fisher, UK) at room temperature for 20 min, free dye was washed off by media replacement with pre-heated culture media, followed by imaging with C-1900 EMCCD (Hamamatsu, Japan) camera using a 5msec 488nm LED illumination pulse. The GFP filter set (DS/FF02-485/20-25, T495lpxr dichroic mirror, and ET525/50 emission filter) was used for Fluo-4 observation.
INS-1 pancreatic β-cell optical control and calcium imaging
INS-1 cells cultured in RPMI supplemented with 10% FCS, 50μM β-mercaptoethanol, 1mM Sodium pyruvate, 10mM HEPES, 2mM Glutamine and 0.5% penicillin/streptomycin (all Invitrogen) were plated onto 24-well MatTek plates (Ashland, MA, US) at 50,000 cells/well. Cells were incubated at 37°C, 5% CO2 in a humidified chamber for 48 hours prior to viral transduction. Viral transduction for 24 hours at an MOI of 5 preceded imaging by 2 days. Imaging was performed as described above for cardiomyocytes, but with a 10x Olympus UPlanFLN lens (NA 0.3) as cells are more abundant, and lower frequency of activation (0.1–0.3Hz) and lower acquisition frame rate (10Hz) as calcium transient duration is longer. 10 μM Gliclazide was added to Tyrode’s buffer (Sigma-Aldrich- T2145) and imaged from 20 minutes post addition.
Image processing & statistics
Raw image data was extracted using CellR, and processed in Excel (Microsoft) MATLAB (Mathworks) and OriginLab7.5 (Origin), unfiltered traces are shown. Numerical data is presented as mean +/- standard deviation. Raw movies were exported to Fiji for processing for publication, movies were compressed and time-stamped using VideoMach (Gromada.com). Samples were compared by Students T-test, significance values P<0.05 is shown by *, or P<0.005 by **where applicable.
Chemicals
Were purchased from Sigma-Aldrich (Dorset, UK) diluted in DMSO, and diluted in culture medium to final concentrations as stated in the text. Cells were observed 30 minutes after drug addition.
Results and discussion
Optical control and calcium imaging in primary adult cardiomyocytes
Although not currently possible, changes to hSC-CM differentiation and culture aspire to produce increasingly adult ventricular phenotypes, hence initial method development was undertaken in primary ventricular cardiomyocytes. We felt the brief in vitro survival of adult cells should bias tool selection to the brightest of the indicator/control tool combinations previously tested in neurons. An adenovirus was made containing the optical control tool, ChETA TC [ 24 ] engineered for large photocurrents and rapid inactivation, together with the brightest red shifted calcium indicator R-GECO [ 21 , 25 ]. To prevent unbalanced expression of two transgenes in a single cell [ 26 ] a rapid self-cleaving 2A peptide [ 27 ] was used ( S1 Fig ). Guinea pig ventricular cardiomyocytes survive 72 hours once isolated allowing 48 hours for reporter production ( Fig 1A ) after infection. Electrical stimulation confirmed the ability of the calcium indicator to report in this cell-type ( S2 Fig ). It was also possible to control the cells optically with blue light stimulation and observe the triggered calcium transient by fluorescent red emission ( Fig 1B and 1C , S1 Movie ). Laser illumination enables control of the size, intensity, and duration of the stimulating light spot. Discrete observation and control windows in single cells can therefore be made ( Fig 1D ) overcoming imaging artefacts previously tackled by syncytial approaches where optical control, and reporter tools are expressed in different cell populations [ 14 , 28 ].
10.1371/journal.pone.0174181.g001
Fig 1
Expression and function of ChETA TC /R-GECO in primary adult cardiomyocytes.
(A) Guinea pig ventricular cardiomyocytes were fixed and stained to reveal ChETA TC (green) and R-GECO (red) two days following infection. The magnified region demonstrates regional staining differences compatible with successful 2A peptide cleavage, and membrane localisation of the optical control tool. The panel on the right shows the red fluorescence emission from R-GECO in a living cardiomyocyte; ROI’s show the stimulated regions for Fig 1D. (B) Pseudo coloured image of the single cell calcium transient revealed by R-GECO, stimulation was at t = 1sec (C). Optical stimulation with 405nm light pulses (blue tabs) every second generates synchronised calcium transients in single adult ventricular cardiomyocytes. (D) Subcellular observation and control windows in single cells. ROI’s as shown in Fig 1A were stimulated with 405nm light singly or in combination as documented in the text above the trace. Raw intensity changes for the ROI’s are shown across the experiment. Scale bar = 10μm.
Optical control and calcium imaging in stem-cell derived cardiomyocytes
The growth of human genomes represented in iPS repositories, and the ease of hSC-CM differentiation, provides a bridge between computer models and patients [ 29 ]. Existing differentiation strategies produce cells with APD and cycle length spanning an order of magnitude [ 30 ] particularly at low cell density [ 31 ]. This variability may hinder identification of small effect sizes relevant to human health, for example a QT interval of 450msec is normal, whereas 500msec is pathological [ 32 ].
Although adult cardiomyocytes do not spontaneously depolarise, the hSC-CM’s do (those used here beat at 0.52Hz +/- 0.15Hz). Some cells have lower (<0.2Hz) rates of spontaneous activity. We reasoned this group might represent an electrically consistent population for further testing as spontaneous activity would not break through an applied optical pacing regime. Programmed optical stimulation at lower frequencies ( Fig 2A–2C , S2 Movie ) gave CTD90 values (Tables 1 and 2 , Fig 2E ) similar to the reported 1Hz APD90 (0.521 +/- 0.069 sec) for these cells. At higher stimulation frequencies loss of signal to noise, and CTD shortening was apparent ( Fig 2C–2F ) similar to electrical stimulation using chemical dyes ( S3 Fig , and [ 17 ]). Immunofluorescence suggests that although membrane localisation of ChETA TC occurs by 48 hours, the development of T-tubule like invaginations in the hSC-CM membrane does not occur ( Fig 2g ). Importantly calcium sequestration with BAPTA (1mM) blocked significant fluorescent responses at 1.2Hz ( S4 Fig ) suggesting photoactivation artefact from R-GECO [ 21 ] can be avoided. Vehicle addition had no clear effect ( S5 Fig ). Since lower triggering frequencies minimise potential phototoxicity, and photoactivation artefact, while maximising signal to noise, low frequency stimulation at 0.3Hz and 0.6Hz was used in further work.
10.1371/journal.pone.0174181.g002
Fig 2
Expression and function of ChETA TC /R-GECO in single stem-cell derived cardiomyocytes.
hSC-CM’s were singularised and replated at low density before stimulation at 0.3Hz, 0.6Hz, 1.2Hz, or 2.4Hz with 405nm light, and simultaneous calcium transient visualisation. (A&B) Averaged response is shown by the heavy line, individual cell response by the thin line for 0.3Hz, and 0.6Hz stimulation. (C) Pacing up to 2.4Hz was possible, average results of single calcium transients are shown. (D-F) From the transient it is possible to measure the half maximal width (CTD50), and the 90% transient duration (CTD90), slopes of activation and decay can also be estimated. Individual cell responses are plotted with sample means represented by black bars. (G) Expression analysis of ChETA TC and R-GECO in hSC-CMs does not show the same striated pattern seen in primary cells as these cells lack T-tubule invaginations. Significance values are indicated by * (P = <0.05), and ** (P<0.005) respectively. Scale bar = 10μm.
10.1371/journal.pone.0174181.t001
Table 1 0.3Hz hSC-CM stimulation results following drug addition summary.
CTD50 (s)
CTD90 (s)
Calcium change (%)
Rising Slope
Decay time τ (s)
Baseline
0.32+-0.09
0.54+-0.18
38.12+-12.96
1.53+-0.47
0.16+-0.04
Flecainide (0.5 μM)
0.79+-0.07**
1.32+-0.34**
37.28+-17.13
0.72+-0.34*
0.30+-0.01**
Dofetilide (5nM)
0.34+-0.05
0.69+-0.11*
23.83+-8.59
1.03+-0.41*
0.18+-0.01*
Cisapride (30nM)
0.50+-0.14
0.97+-0.31**
32.56+-6.21
2.08+-1.10
0.31+-0.06*
Nifedipine (100nM)
0.23+-0.06**
0.50+-0.11
7.53+-0.44**
-
0.22+-0.04*
Significance values are indicated by
P = <0.05*,
P<0.005**
10.1371/journal.pone.0174181.t002
Table 2 0.6Hz hSC-CM stimulation results following drug addition summary.
CTD50 (s)
CTD90 (s)
Calcium change (%)
Rising Slope
Decay time τ (s)
Baseline
0.28+-0.03
0.60+-0.07
32.13+-10.1
3.12+-2.12
0.14+-0.02
Flecainide (0.5 μM)
0.81+-0.09**
1.18+-0.20**
38.0+-17.06
0.66+-0.35**
0.43+-0.22**
Dofetilide (5nM)
0.32+-0.02**
0.57+-0.12
28.91+-15.91
2.63+-1.64*
0.20+-0.05**
Diltiazem (1μM)
-
-
3.18+-4.2**
-
-
Verapamil (1μM)
0.26+-0.06
0.42+-0.07**
9.6+-3.1**
1.00+-0.37**
0.20+-0.04**
Significance values are indicated by
P = <0.05*,
P<0.005**
Drug testing can both be done by paired measurements before and after compound addition using the same cell as its own internal control [ 14 , 16 ]; or as an unpaired assessment compared to a reference population ( S6 Fig , and approximately half of the patient disease models summarised in [ 3 ]). The first approach might improve sensitivity to small drug effects, whereas the second approach is simpler but vulnerable to intrinsic variability in a sample which may preclude this option. We find that the cell selection strategy combined with optical stimulation limits baseline variability and makes the second approach feasible, although either is possible ( S6 Fig ).
Application of flecainide (sodium (INa) channel inhibitor), dofetilide (potassium (hERG) channel inhibitor), cisapride (serotonin receptor inhibitor with off-target hERG block), or the voltage dependent L-type calcium channel inhibitors verapamil (has off target hERG inhibition), nifedipine, and diltiazem were tested at clinically relevant concentrations ( Fig 3 , Tables 1 and 2 ). All calcium channel blockers suppress the triggered intensity change of the calcium indicator suggesting this alone may identify such compounds. However loss of indicator brightness represents a weakness in this strategy as although the anticipated CTD shortening with nifedipine was seen, the opposite trend due to off-target hERG block by verapamil was missed even though this was detected in parallel patching studies. Cisapride and dofetilide ( Fig 4A, 4C and 4D ) both cause dose-dependent prolongation of CTD, with early ( Fig 4B1 ) and after ( Fig 4B2 ) depolarisation transients at higher doses. Flecainide demonstrated a complex response, reducing the activation slope of the calcium transient, and also prolonging it. hERG inhibition by flecainide [ 33 ] at this clinical dose is reported, at higher doses this effect is less marked ( Fig 4E and 4F ) reflecting the integrated sum of other off-target effects.
10.1371/journal.pone.0174181.g003
Fig 3
Small molecule ion channel inhibitors applied to hSC-CMs.
hSC-CMs were plated out at low density and exposed to clinically relevant doses of known small molecule inhibitors of the sodium channel NaV1.5 (flecainide 0.5μM), the potassium channel KCNH2/hERG (dofetilide, 5nM, and cisapride, 30nM), and the voltage gated calcium channel CACNA1.2 (nifedipine, 100nM; verapamil 1μM, diltiazem 1μM). Traces obtained at 0.3Hz, and 0.6Hz are shown for the annotated small molecule as before with the average response shown by the thick line. Analysis of the data is presented in Tables 1 and 2 .
10.1371/journal.pone.0174181.g004
Fig 4
Detailed analysis of dofetilide, and flecainide response in optically controlled hSC-CMs.
(A) Averaged response of single hSC-CMs to a Dofetilide dose series. (B) Unfiltered calcium traces from single cells showing dofetilide triggers proarrythmic early after depolarisations (EAD) (B1) and delayed after depolarisations (DAD) (B2). These cells were excluded from the averaged response in(A). (C-D) Log dose response for the CTD90, and decay half time in response to dofetilide is shown. Cells showing EAD, or DAD like perturbations to the calcium transient were excluded. (E) Average dose response to Flecainide is shown, with prolongation of rise time, and calcium transient duration at lower doses which normalises at higher doses. (F) CTD50 changes for the different flecainide doses are shown graphically. Significance values are indicated by * (P = <0.05), and ** (P<0.005) respectively.
Optical control and calcium imaging in pancreatic beta-cells
The need for simultaneous calcium transient phenotyping and control is not a unique cardiomyocyte problem suggesting the method may be useful in other excitable cell types. The beta-cell of the endocrine pancreas uses a glucose triggered voltage change to produce a calcium transient enabling insulin release by exocytosis. Glucose causes ATP levels to rise, increasing the inhibition of an ATP dependent potassium channel K ATP reducing repolarisation, and causing calcium release. Augmentation of insulin release by chemical inhibition of K ATP using the sulphonylurea class of drugs has been a mainstay of Type II Diabetes management for 60 years [ 34 ]. Optical control and calcium imaging has been tried in this model [ 35 , 36 ] but although insulin accumulation could be measured biochemically the underlying calcium transients were either not detected, or drift upward as the excitation spectrum of the calcium indicators (Fluo-4-AM, and Fura2-AM respectively) overlaps with the blue-green activation spectrum of the optical control tool causing unintended ChR2 activation during imaging.
As proof of concept that simultaneous optical control and calcium imaging could be improved the immortalised rat insulinoma beta-cell model (INS-1) was infected and imaged as above in the presence or absence of the sulphonylurea gliclazide at low (3mM) and activating (9mM) glucose concentrations. In the absence of optical control, a chaotic pattern of random calcium activation is apparent ( S3 Movie , S7 Fig ). At 3mM glucose 45.7 +/- 0.28% of cells have a calcium transient 10% greater than baseline over a 2minute interval, which increases to 71.4 +/- 0.05% at 9mM (p = 0.01); this is accompanied by an increase in transient frequency (1.06 +/- 0.17 to 2.88 +/- 0.3, p<0.001) and increase in peak transient intensity (1.21 +/-0.01 to 1.41 +/- 0.02, p<0.001, n = 110 at 3mM, and 91 at 9mM respectively) but variability in the sample makes extraction of transient parameters challenging.
By contrast optical control of the calcium transient in beta-cells is possible ( S3 Movie ) across the range of frequencies observed in spontaneous samples ( S7c Fig ), at 3mM ( Fig 5A–5C ) and 9mM glucose ( Fig 5D ) producing coordinated datasets from which parameters equivalent to the cardiomyocyte can be derived ( Fig 5E , Table 3 ). When exposed to 10μM gliclazide prolongation of CTD50 and CTD90 was observed at 9mM glucose, at 3mM glucose only a change in CTD50 was seen, peak calcium intensity increases in both 3mM and 9mM glucose, as summarised in Table 3 and Fig 5F and 5G .
10.1371/journal.pone.0174181.g005
Fig 5
Application of optical control and calcium imaging in pancreatic beta-cells.
(A) Intensity traces of the calcium transient visualised by R-GECO in INS-1 beta-pancreatic cells triggered by optical stimulation at frequencies 0.1–0.3Hz, the averaged responses of 10 cells is shown in the black trace. (B) A pseudo-coloured calcium transient time-series taken from movie 3 at 60x. The cells indicated by arrows in the top left of the image were controlled optically, other activity in the field of view is spontaneous. Scale bar = 20μm. (C) Extracted calcium traces at 3mM glucose following 0.1Hz, 0.2Hz, and 0.3Hz activation, the averaged result shown in black. (D) The effect of glucose concentration and stimulation frequency in INS-1 cells determined by CTD90 analysis of calcium intensity traces following optical stimulation. (E) Decay constants of calcium transients in INS-1 cells in 3mM glucose stimulated at 0.1–0.3Hz are shown. (F) CTD50 analysis for INS-1 cells stimulated at 0.1Hz in 3mM and 9mM glucose are shown at baseline, following 1% DMSO or 10μM Gliclazide addition. (G) CTD90 analysis for INS-1 cells stimulated at 0.1Hz in 3mM and 9mM glucose are shown at baseline, or following 10μM Gliclazide addition. Significance values are indicated by * (P = <0.05), and ** (P<0.005) respectively.
10.1371/journal.pone.0174181.t003
Table 3 Summary of INS1 calcium transient changes due to glucose and gliclazide at 0.1 Hz.
CTD50 (s)
CTD90 (s)
Calcium change(%)
Decay time τ(s)
3mM glucose baseline
0.85+-0.18
2.76+-0.37
16.00+-0.92
0.79+-0.23
3mM glucose + 10μM gliclazide
1.32+-0.49*
3.07+-0.82
69.60+-17.25**
2.53+-2.39*
9mM glucose baseline
1.06+-0.19
3.71+-0.50
20.00+-7.29
1.11+-028
9mM glucose + 10μM gliclazide
1.57+-0.20*
4.62+-0.79*
36.71+-11.78**
1.65+-0.69
Significance values are indicated by
P = <0.05*,
P<0.005**
Discussion
The principle limitations of the method in its current form is that although it identifies that cellular behaviour is altered, unlike patching, it is unable to distinguish effects of individual ion currents through which effects might be occurring. As such it is not a direct replacement for patching. Similarly at the moment transgenes are delivered by virus. This means that primary cardiomyocytes infected in vitro are often used towards the end of their natural life as time for gene expression is required with inevitable loss of quality. This can be offset by in vivo infection, and delayed cell isolation, but at the expense of additional procedures such as direct myocardial injection which limits the general utility of the approach. Furthermore, viral infection produces a host cell response that may alter the overall performance of any cell. This should be possible to overcome either via transgenic approaches of gene targeted knock-in or perhaps use of alternative viral strategies engineered to elicit less host cell reaction.
Improving the maturity of stem-cell derived cardiomyocytes will lead to pacing dependence. As these tools are genetically encoded, and equivalents have already been utilised in transgenic models [ 9 , 37 ], genome knock-in lines with constitutive and homogeneous expression may become useful adjuncts to this approach. Indeed, further refinement identifying particular cardiomyocyte subtypes by targeting promoters active in the atrium or ventricle, as demonstrated for lentiviral promoter:transgene combinations in SC-CMs [ 38 ] may be possible.
Methodological improvements may arise by combining hyperpolarising and depolarising optical control tools to completely suppress intrinsic activity independently of innovations in differentiation or cell maintenance. Although this would increase the number of cardiomyocytes suitable for study, compatible indicators are currently limiting for this approach [ 39 ]. Alternatively, developments allowing multi-parameter (voltage, calcium and contraction) measurements under optical stimulation would enable creation of tools to explore the interaction between these aspects of cardiac physiology in health and disease states and how those changes can be influenced by small molecules.
Conclusion
This method provides a contactless all genetic single cell assay with temporal stimulation control over the physiological range of cardiomyocytes and pancreatic beta-cells. Anticipated small molecule effects were detected during brief experimental periods on low cell numbers in two model systems. The approach is quick, simple, and can be applied to microscopes with conventional blue/green/red imaging capabilities, using a single virus and isolated cells.
Supporting information
S1 Appendix
Sequence information.
Human codon optimised ChETA TC is shown in blue, with a 5xMyc epitope (EQKLISEEDL) tag in grey, the P2A peptide linker (GSGATNFSLLKQAGDVEENPGP) is in bold and underlined. Human codon optimised R-GECO1 is highlighted in Red. Restriction enzyme sites introduced for cloning purposes are in plain text.
(DOCX)
S1 Fig
Overview.
An adenovirus containing a fusion of the Channel Rhodopsin variant ChETA TC with an N terminal Myc tag linked to the red fluorescent calcium indicator R-GECO via a self-cleaving 2A peptide was made. Following infection gene synthesis, autocatalytic cleavage and chromophore maturation occurs, producing two functional gene products. The minimum collective time for these processes is 24 hours, and we performed experiments at 48hours. ChETA TC , a light sensitive voltage channel, allows excitable cells to be depolarised by light in the blue-green range of the visible spectrum. We used 405nm light in these experiments. Calcium transients in excitable cells can be generated by R-GECO which requires green excitation light, producing red emissions. From the evoked calcium transient, it is possible to extract various parameters, as indicated.
(EPS)
S2 Fig
Electrical stimulation of primary cardiomyocytes with R-GECO to visualise the calcium transient.
A raw unfiltered calcium transient from a single adult guinea pig ventricular cardiomyocyte controlled with 0.5Hz electrode stimulation confirms R-GECO function in primary cardiomyocytes, normalised emission intensity is shown over time.
(EPS)
S3 Fig
Electrical stimulation of hSC-CM’s with chemical dye based visualisation of the calcium transient.
(A) Averaged single cell calcium transient traces from hSC-CM’s loaded with the Fluo4 calcium dye and stimulated electrically at 0.5Hz, 1Hz, and 2Hz. CTD50 and CTD90 values extracted from the raw traces to generate (S3A) are shown in (S3B) and (S3C). Significance values are indicated by * (P = <0.05), and ** (P<0.005) respectively.
(EPS)
S4 Fig
Calcium sequestration with BAPTA prevents visualisation of a dynamic response to optical stimulation.
Photoactivation of R-GECO has been documented in response to 488nm light previously. The same phenomenon occurs to a lesser extent with 405nm light. To ensure the triggered responses visualised here are due to calcium release and not imaging artefact, in addition to showing that cells paced electrically report a calcium transient ( S2 Fig ), and that cells stimulated optically at higher frequency show reduction (imaging artefact would cause an increase) in signal amplitude, CTD50, and CTD90 ( Fig 2C–2E ) we used the intracellular calcium sequestration agent BAPTA. The two traces during optical stimulation at 1.2Hz obtained from the same cell before (red line) and after (black line) BAPTA addition are shown. If significant photoactivation of R-GECO were occurring this would be expected to be seen as increased signal in the absence of calcium.
(EPS)
S5 Fig
Calcium transient duration with vehicle controls evoked by optical stimulation.
(A) Averaged calcium transients obtained by optical stimulation at 0.3Hz in the presence of DMSO at 0.1% and 0.001% are shown. (B) CTD50 and (C) Decay half times are shown graphically and numerically in the table.
(EPS)
S6 Fig
Paired versus unpaired small molecule assessment.
(A) single hSC-CM’s can be phenotyped at baseline, and then restudied following compound addition. Raw data traces of stimulation at 0.3Hz before and after 0.5 μM flecainide addition are shown. (B) & (C) CTD50, and CTD90 data extracted from paired (P) or unpaired (U) experiments using 0.5μM flecainide and 0.3Hz optical stimulation in single hSC-CM’s is shown. Significance values are indicated by * (p<0.05) and ** (p<0.005) respectively. Pairwise comparison of cells at baseline and then after drug addition limits the impact variability between individual cells may produce. However it increases data storage, data processing requirements, slows down throughput and may be more vulnerable to phototoxicity. An alternative strategy compares drug exposed cells to a reference population. This is more useful when cells show a consistent behaviour. We find that the cell selection strategy combined with optical stimulation enables either approach, even at the lowest (0.3Hz) stimulation frequency where variability is greatest. In the paired experiment CTD90 rises from 0.82 +/-0.13s to 1.72 +/- 0.37s (p<0.005), in the unpaired experiment it rises from 0.535+/-0.18s to 1.32 +/- 0.34s (p<0.005).
(EPS)
S7 Fig
Spontaneous calcium transients in INS-1 cells.
INS-1 cells were infected with ChETA TC and R-GECO and imaged at 2 days in 3mM or 9mM glucose. Cells were imaged at 10Hz for 2 minutes and single cell traces extracted for analysis. A change 10% over baseline was regarded as definite activity. The number of spikes, and the maximum intensity were enumerated. (A) 5 traces for each condition are shown. The most active, and the largest transient traces are shown in addition to 3 other traces with behaviour close to the group average. 53% of the cells at 3mM had no detectable activity, compared to 28% at 9mM. (B) F/Fo intensity, and (C) event rate over 2min are plotted. Significance is indicated by * (p<0.01) and ** (p<0.001).
(EPS)
S1 Movie
Calcium transient imaging in an optically stimulated Adult ventricular cardiomyocyte.
Isolated Adult ventricular myocytes were infected with the ChETA TC and R-GECO and stimulated with 405nm light pulses at 1Hz with continuous visualisation of the red calcium transient. A Pseudo-coloured movie of a grey-scale image is provided. Time stamp = seconds.
(MP4)
S2 Movie
Calcium transient imaging in an optically stimulated hSC-CM.
hSC-CMs were infected with the ChETA TC and R-GECO and singularised. Cells were stimulated with 405nm light pulses at 0.3Hz with continuous visualisation of the red calcium transient. A Pseudo-coloured movie of a grey-scale image is provided. Time stamp = seconds.
(MP4)
S3 Movie
Calcium transient imaging in INS-1 cells with discrete optical stimulation.
INS-1 cells were infected with ChETA TC and R-GECO and visualised at 2 days. Cells in the top left corner marked by an asterix * were stimulated with 405nm light pulses at 0.1Hz, other cells show a random activation. A pseudo-coloured movie of a grey-scale image is provided. Time stamp shows hours:minutes:seconds:milliseconds, the 10Hz acquisition is presented at 50 frames per second.
(MP4)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The order Carnivora is composed of 296 extant species [ 1 ] currently ranged into two suborders: the Caniformia which includes nine families, namely the Canidae (dog-like species), Ailuridae (red panda), Mephitidae (skunks and stink badgers), Mustelidae (weasels, badgers, martens, otters, etc.), Odobenidae (walrus), Otariidae (eared seals), Phocidae (earless seal), Procyonidae (raccoons, coatis, kinkajous, etc.) and Ursidae (bears); and the Feliformia which is represented by seven families, namely the Felidae (cat-like species), Eupleridae (Malagasy carnivorans), Herpestidae (mongooses), Hyaenidae (hyenas), Nandiniidae (African palm civet), Prionodontidae (Asiatic linsangs) and Viverridae (civets and genets). The oldest known fossils of Carnivora have been found in the late Paleocene; they belong to the extinct families Miacidae and Viverravidae, and have small body-size, comparable to extant weasels and martens [ 2 – 4 ]. The timing of the emergence of the crown carnivorans and their relationships to Paleocene and Eocene fossils are still unresolved. However, there is consensus for supporting two higher fossil taxa: the Carnivoraformes, which are composed of the crown group plus the stem family Miacidae, which is probably paraphyletic; and the Carnivoramorpha, which groups the Carnivoraformes and the Viverravidae [ 4 , 5 ].
The phylogeny of several carnivoran families has been extensively studied based on mitochondrial and nuclear data, including the Felidae, the Mustelidae and the Ursidae [ 6 – 8 ], while other families remain poorly studied, specifically the Eupleridae, Herpestidae, Mephitidae, Procyonidae and Viverridae. The phylogeny and timescale of diversification of the Carnivora have been studied by Eizirik et al. [ 9 ] using a molecular supermatrix of 7,765 base pairs (bp) containing 14 nuclear genes for 50 species, which represents less than 17% of the species diversity of the order Carnivora. The evolutionary history of carnivorans has also been inferred using a supertree approach by Nyakatura and Bininda-Emonds [ 10 ]. These two studies were based on different methods and data, and different fossils were used as calibration points. Although several nodes showed similar ages in the two studies, such as the most recent common ancestor (MRCA) of the Carnivora (59.2 Mya versus 64.9 Mya), Arctoidea (42.6 Mya versus 47.8 Mya) and Pinnipedia (24.5 Mya versus 22.4 Mya), some nodes were highly discordant, including the MRCA of the Caniformia (48.2 Mya versus 61.2 Mya), Feliformia (44.5 Mya versus 53.2 Mya), Feloidea (Felidae + Prionodontidae) (33.3 Mya versus 52.9 Mya) and Canidae (7.8 Mya versus 16.3 Mya). Most other studies to date have focused on either the Caniformia [ 8 ] or the Feliformia [ 11 ] and the timing of diversification in some families appears highly elusive or uncertain in the absence of a molecular timescale based on a high diversity of species from all carnivoran families.
With the development of next-generation sequencing (NGS) technologies, the number of mitochondrial genomes available in the international nucleotide databases has considerably increased during the last decade [ 12 ]. For the order Carnivora, there are currently more than 2,400 complete mitogenomes, and some of these were sequenced from Pleistocene fossils, such as polar bear [ 13 ], giant short-faced bears [ 14 ], cave lion [ 15 ] or saber-toothed cats [ 16 ]. This notably large and diversified dataset offers an excellent opportunity to better understand the evolutionary history of the order Carnivora, as problematic sequences and taxonomic issues can be more easily detected, and more importantly, as many fossils can be included as calibration points for estimating divergence times. The latter aspect is particularly relevant considering the paleontological record of Carnivora has been significantly improved over the last 10 years, with the discovery of several key fossils [ 17 – 21 ].
Here we analysed complete mitochondrial genomes to reconstruct phylogenetic relationships among carnivorans and to estimate divergence times. We sequenced 43 mitogenomes using various methods (Sanger sequencing of PCR products, NGS of long PCR products or Illumina shotgun sequencing). We also assembled eight mitogenomes from Sequence Read Archive (SRA) data. The 51 new mitogenomes belong to several families, namely Canidae (2), Eupleridae (6), Felidae (4), Herpestidae (12), Hyaenidae (1), Mustelidae (10), Otariidae (2), Phocidae (1), Prionodontidae (1), Procyonidae (3), Ursidae (1), Viverridae (7) for Carnivora, as well as Tapiridae (1) for the order Perissodactyla, which was included as an outgroup. These new mitogenomes were compared to all mitogenomes available for Carnivora in public repositories. At the intraspecific level, we selected only mitogenomes that were separated by more than 1% of nucleotide divergence and excluded those suspected to be of low-quality or from misidentified taxa. Our final alignment includes 220 taxa, which represent 2,442 mitogenomes. Our main objective was to build one of the largest time-trees of Carnivora estimated using a selection of fossil calibration points in order to provide further insights into the evolution of this broadly distributed and morphologically diverse order of mammals.
Material and methods
DNA extraction, amplification, sequencing and mitogenome assembly
Total DNA was extracted from cells, muscle or skin samples using the DNeasy Blood and Tissue Kit (Qiagen, Hilden, Germany). Details on the 43 samples extracted for this study are given in S1 Appendix . The mitochondrial genomes were sequenced using one of the three following approaches: Sanger sequencing of about 20 overlapping PCR products (length between 700 and 2000 bp); NGS of five overlapping long PCR products of around 4–5 kb; and Illumina shotgun sequencing.
In the first approach, PCR amplifications were carried out as previously described [ 22 ] using the primers listed in S2 Appendix . The amplicons were then sequenced in both directions by Eurofins MWG Operon (Ebersberg, Germany). Genomes were assembled with electropherograms of overlapping amplicons using Sequencher 5.1 (Gene Codes Corporation, Ann Arbor, MI, USA).
In the second approach, five overlapping PCR products of around 4–5 kb were amplified as previously described [ 23 ], and were sequenced at the “Service de Systématique Moléculaire” (UMS CNRS 2700, MNHN, Paris, France) using a Ion Torrent Personal Genome Machine (Thermo Fisher Scientific, Waltham, MA, USA).
All PCR products generated for this study were overlapping in the 5’ and 3’ regions with at least two other PCR products. As a consequence, the authenticity of each sequence was validated by double checking after verification of the perfect nucleotide identity with the two other sequences overlapping in the 5’ and 3’ regions (length between 50 and 500 bp).
The third approach was based on Illumina shotgun sequencing. DNA samples were quantified with a Qubit® 2.0 Fluorometer using the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA). Libraries were prepared using the TruSeq® Nano DNA Library Prep kit (Illumina, San Diego, CA, USA) after pooling 150 ng of total DNA of 10 species belonging to distant taxonomic groups (i.e. different phyla, classes, orders or families). Libraries were sequenced at the “Institut du Cerveau et de la Moelle épinière” (Paris, France) using a NextSeq® 500 system and the NextSeq 500 High Output Kit v2 (300 cycles) (Illumina).
The NGS reads generated with either Ion Torrent or Illumina sequencers were assembled by baiting and iterative mapping approach on Geneious® 10.2.2 (Biomatters Ltd., Auckland, New Zealand) using available mitochondrial references, including cytochrome b, cytochrome c oxidase subunit I, 12S and 16S rRNA genes. The 43 new mitochondrial genomes generated for this study were annotated using MITOS [ 24 ] and deposited in GenBank under accession numbers MW257198-MW257240.
Eight mitochondrial genomes were assembled from SRA downloaded from NCBI for the following species: Bassaricyon neblina (SRX1097850), Bassariscus sumichrasti (SRX1099089), Helogale parvula (SRR7637809), Lontra canadensis (SRR10409165), Mirounga angustirostris (SRR10331586), Mungos mungo (SRR7704821), Prionodon linsang (ERR2391707) and Zalophus wollebaeki (SRR4431565).
Mitochondrial alignments
The 51 mitochondrial genomes assembled in this study were compared with other genomes available in the NCBI nucleotide database (see details in S1 Appendix ) for the different families. At the intraspecific level, we selected only mitochondrial haplotypes separated by more than 1% of divergence. Our final alignment includes 218 mitogenomes of Carnivora representing a large taxonomic diversity in the following 16 families (number of mitogenomes in parentheses): Ailuridae (2), Canidae (21), Eupleridae (7), Felidae (46), Herpestidae (15), Hyaenidae (4), Mephitidae (3), Mustelidae (39), Nandiniidae (1), Odobenidae (1), Otariidae (12), Phocidae (30), Prionodontidae (2), Procyonidae (6), Ursidae (21) and Viverridae (19). A tapir ( Tapirus terrestris ) and a pangolin ( Phataginus tricuspis ) were used to root the carnivoran tree, as these two taxa represent different Laurasiatherian orders, Perissodactyla and Pholidota, respectively, which were shown to be closely related to Carnivora in previous molecular studies [ 25 , 26 ].
The 220 mitochondrial genomes were aligned using AliView 1.22 [ 27 ]. Ambiguous regions for primary homology were excluded from the alignment. For that reason, the control region was completely removed, as well as some parts of rRNA and tRNA genes and a few nucleotides at the 5’- and 3’-regions of some protein-coding genes. To limit the impact of missing data, we also removed from the alignment all indels (insertions or deletions) detected in only one genome. The final alignment, named mtDNA , contains 220 taxa and 14,892 bp, representing 89% of the reference mitogenome for Canis lupus (NC_002008). Two other datasets were derived from mtDNA : (1) mtDNA-Tv (transversions only), in which the nucleotide G was replaced by A and the nucleotide T by C; and (2) PCG-mtDNA (10,809 nt), in which all regions other than protein-coding genes were removed (i.e. 12S and 16S rRNA genes and tRNA genes), as well as the ND6 gene (because it is located on the opposite strand of other protein-coding genes). The three datasets used in this study are available at https://osf.io/cfx8r/ .
Analysis of base composition
The alignment of the protein-coding genes of 220 mitochondrial genomes ( PCG-mtDNA dataset) was used to calculate the frequency of the four bases (A, C, G and T) at each of the three codon-positions ( S3 Appendix ). The 12 variables measured were then summarized by a principal component analysis (PCA) using the FactoMineR package [ 28 ] in R version 3.5.3 (from http://www.R-project.org/ ). The strand bias in nucleotide composition was studied at third codon-positions of the PCG-mtDNA dataset by calculating the relative frequencies of A and T nucleotides (AT3 skew = [A—T] / [A + T]) and the relative frequencies of C and G nucleotides (CG3 skew = [C—G] / [C + G]) [ 29 – 31 ].
Phylogenetic analyses
Two datasets ( mtDNA and mtDNA-Tv ) were analysed with probabilistic methods for tree reconstruction using the resources available from the CIPRES Science Gateway [ 32 ]. The Bayesian analyses were done with MrBayes 3.2.7 [ 33 ] using the two following models: GTR+I+G for mtDNA and JC69+I+G for mtDNA-Tv . The posterior probabilities (PP) were calculated using 10,000,000 Metropolis-coupled MCMC generations, tree sampling every 1000 generations and a burn-in of 25%.
To examine the phylogenetic signal within the mtDNA dataset, we also performed Bayesian analyses (with the same parameters) on 10 half-overlapping sub-datasets (i–x) of the about the same length (i.e., 2978 or 2980 bp), corresponding to the following positions: (i) 1–2978; (ii) 1489–4466; (iii) 2979–5956; (iv) 4467–7444; (v) 5957–8934; (vi) 7445–10422; (vii) 8935–11912; (viii) 10423–13400; (ix) 11913–14892; and (x) 13401–14892 + 1–1488. The use of half-overlapping sub-datasets (sliding window of ≈ 2978 bp) implies that all nucleotide sites of the total mtDNA alignment are represented twice in these Bayesian analyses. The lists of bipartitions obtained from Bayesian analyses of the 10 sub-datasets were transformed into a weighted binary matrix for supertree construction using SuperTRI v57 [ 34 ]. Each binary character corresponds to a node, which was weighted according to its frequency of occurrence in one of the 10 lists of bipartitions. In this manner, the SuperTRI method takes into account both principal and secondary phylogenetic signals. The SuperTRI bootstrap percentages (SBP) were obtained from PAUP 4* version 4b10 [ 35 ] after 1000 bootstrap replicates of the MRP (Matrix Representation with Parsimony) matrix of 3,398 binary characters (reconstructed under SuperTRI v57 [ 34 ]). Here, the SuperTRI analyses were conducted to test for phylogenetic signal along the mtDNA genome. If a robust node in the Bayesian tree (PP ≥ 0.95) is recovered with high SBP (≥ 95%) and repeated in most of the 10 Bayesian trees reconstructed from the half-overlapping sub-datasets of the mtDNA dataset, this signifies that the phylogenetic signal is present all along the mtDNA genome. If a node in the Bayesian tree is recovered with low SBP (< 95%) and repeated in less than five of the 10 Bayesian trees reconstructed from the half-overlapping sub-datasets of the mtDNA dataset, this indicates that the phylogenetic signal is weak or confined to a few fragments of the mtDNA genome. If there is a robust topological conflict between Bayesian and SuperTRI results, this suggests that at least one of the studied genomes was partially contaminated by a mitochondrial DNA sequence from another species or by a nuclear DNA sequence of mitochondrial origin (Numt). An example has been previously reported for the mitochondrial genomes of domestic goat [ 36 ].
Molecular dating
Divergence times were estimated on the CIPRES Science Gateway [ 32 ] using the mtDNA dataset and the Bayesian approach implemented in BEAST v.2.4.7 [ 37 ]. Twenty-one fossil calibration points were selected for molecular dating ( Table 1 ). Most of these were interpreted from the fossil record using maximum (Max) and minimum (Min) ages. We applied two strategies for fossil calibration: (1) a uniform distribution between Max and Min on the calibrated node ages; or (2) a log-normal distribution on the calibrated node ages using Min as offset, M = Max—Min / 4 and S = 0.926 (to match the 97.5% quantile to Max) (see details in Table 1 ). The second strategy relies on the fact that minimum ages are generally more accurate and reliable than maximum ages because younger fossils are more abundant and more accurately dated than older fossils as a consequence of taphonomic processes and dating methods [ 38 ]. We applied a GTR + I + G model for the mtDNA alignment (with a proportion of invariants of 0.425) and a relaxed-clock model with uncorrelated lognormal distribution for substitution rates. Node ages were estimated using a calibrated Yule speciation prior and 2.10 8 generations, with tree sampling every 2,000 generations and a burn-in of 25% generations. MCMC mixing efficiency and convergence were assessed using the ESS values (>200) in Tracer v.1.7.1. The chronogram was reconstructed with TreeAnnotator, which is included in the BEAST package [ 37 ].
10.1371/journal.pone.0240770.t001
Table 1 Maximum (Max) and minimum (Min) age calibrations (in millions of years ago, Mya) used for molecular dating analyses (with either uniform or log-normal prior distributions [U/L] on calibrated node ages interpreted from the fossil record).
Most recent common ancestor
Max
Distribution
Min
References for justification
Ferae
85
N, M = 75
65
www.timetree.org ; split Canis / Manis (13 studies)
Caniformia
Canidae
( Canis , Cuon , Lycaon )
12
U/L, M = 1.5
6
[ 39 ]
( Vulpes , Nyctereutes )
34
U/L, M = 6.25
9
[ 39 ]
Ursidae
34
U/L, M = 5.75
11
[ 17 , 40 , 41 ]
( Arctotherium , Tremarctos )
10
U/L, M = 5.75
3
[ 42 ]
Ursus maritimus † from Svalbard (GU573488)
0.11
N, M = 0.12
0.13
[ 13 ]
Pinnipedia
34
U/L, M = 3.75
19
[ 20 , 42 , 43 ]
Otarioidea
23
U/L, M = 2.00
15
[ 20 ]
Phocidae
23
U/L, M = 2.75
12
[ 20 , 44 ]
Mustelidae
(Guloninae … Mustelinae)
27.6
U/L, M = 2.825
16.3
[ 21 , 45 ]
(Lutrinae, Ictonychinae)
27.6
U/L, M = 3.85
12.2
[ 18 , 45 ]
Feliformia
Feloidea
34
U/L, M = 3.5
20
[ 46 ]
Felidae
20
U/L, M = 1.5
14
[ 46 ]
( Acinonyx , Puma )
14
U/L, M = 2.65
3.4
[ 46 , 47 ]
( Leptailurus … Profelis )
14
U/L, M = 2.5
4
[ 47 ]
( Caracal , Profelis )
14
U/L, M = 2.5
4
[ 46 ]
( Panthera , Neofelis )
14
U/L, M = 2.0125
5.95
[ 19 ]
Viverroidea
Viverridae
(Genettinae, Viverrinae)
34
U/L, M = 5.00
14
[ 47 ]
( Herpestidae , Eupleridae )
34
U/L, M = 3.375
20.5
[ 47 ]
( Helogale , Crossarchus )
20.5
U/L, M = 3.625
6
[ 47 ]
( Galerella … Cynictis )
20.5
U/L, M = 3.275
7.4
[ 47 ]
Hyaenidae
( Hyaena , Parahyaena )
9.5
U/L, M = 1.475
3.6
[ 47 ]
Abbreviations: L = Log Normal; M = Mean age; N = Normal; U = Uniform.
Results
Variation in base composition
The base composition (frequency of the nucleotides A, C, G and T) was analysed at the three codon-positions of the PCG-mtDNA dataset ( S3 Appendix ). The 12 variables measured for 220 taxa were summarized by a principal component analysis (PCA) based on the first two principal components (PC), which contribute 45.80% and 24.48% of the total variance, respectively ( Fig 1 ). The variables factor map shows that the variance can be explained by similar differences in base composition at the three codon-positions. Some taxa have a mtDNA genome characterized by a divergent base composition.
10.1371/journal.pone.0240770.g001
Fig 1
Variation in base composition of the mitogenomes of Carnivora.
The PCG-mtDNA dataset was used to calculate the frequency of the four bases (A, C, G and T) at each of the three codon positions, and the 12 variables measured were then summarized by a principal component analysis (PCA). The main graph represents the individual factor map based on 220 taxa. The families of Carnivora are highlighted by different colours. The small circular graph at the top left represents the variables factor map.
Most members of the Mustelidae are found near the middle of the graph, except Mellivora because its mtDNA genome contains a higher percentage of guanine (15.65% versus “mean in other Mustelidae” [MoM] = 13.18%) and a lower percentage of adenine (28.35% versus MoM = 30.95%). This trend is observed at each of the three codon positions, and is more marked at third positions ( S3 Appendix ): G1 = 22.19% versus MoM = 20.87%, G2 = 12.19% versus MoM = 11.84%, G3 = 12.58% versus MoM = 6.82%; A1 = 30.55% versus MoM = 31.95%, A2 = 19.13% versus MoM = 19.41%, A3 = 35.38% versus MoM = 41.51%.
Within Viverridae, the mitogenome of the otter civet ( Cynogale bennettii ) is characterized by a higher percentage of thymine (32.90% versus “mean in other Viverridae” [MoV] = 30.73%) ( Fig 1 ). This trend is observed at each of the three codon positions, and is more marked at third positions: T1 = 25.47% versus MoV = 23.79%, T2 = 42.71% versus MoV = 42.17%, T3 = 30.52% versus MoV = 26.21%.
Within Ursidae, the giant panda ( Ailuropoda melanoleuca ) shows a different base composition, with more adenines (30.63% versus “mean in other Ursidae” [MoU] = 29.55%) and thymines (31.02% versus MoU = 28.98%) than other species of the family. This trend is mainly explained by differences at third codon-positions, in which the giant panda shows more adenines (40.91% versus MoU = 38.46%) and thymines (28.52% versus MoU = 23.38%).
Most members of the Felidae, including Smilodon populator (Machairodontinae), are found near the middle of the graph, except Homotherium latidens (Machairodontinae) for which the mtDNA genome shows a very atypical base composition. At third codon-positions, Homotherium is characterized by a lower percentage of guanine (2.80% versus “mean in other Felidae” [MoF] = 6.25%) and a higher percentage of thymine (25.41% versus MoF = 20.79%). At second codon-positions, its mitogenome contains a lower percentage of thymine (40.80% versus “mean in other Felidae” [MoF] = 41.89%) and a higher percentage of guanine (12.58% versus MoF = 11.86%). At first codon-positions, the mtDNA of Homotherium shows a lower percentage of adenine (30.08% versus “mean in other Felidae” [MoF] = 32.07%), a higher percentage of guanine (22.65% versus MoF = 20.92%) and a higher percentage of guanine (23.07% versus MoF = 22.01%).
Within Canidae, the two extant tribes can be distinguished based on their nucleotide composition, as the mitogenomes of the Vulpini (fox-like canines) show more guanines (≥ 13.89% versus ≤ 13.28%) and less adenines (≤ 29.53% versus ≥ 29.83%) than those of the Canini (dog-like canines). This trend is mainly explained by a strong bias observed at third codon-positions, in which the Vulpini taxa have more guanines (G3 ≥ 8.39% versus ≤ 7.30%) and less adenines (A3 ≤ 37.18% versus ≥ 37.80%) than the Canini.
At third codon-positions, the Eupleridae and Hyaenidae have higher percentages of cytosine than other Feliformia (mean C3: Eupleridae = 36.78%; Hyaenidae = 35.55%; other Feliformia = 30.65%).
Pairwise mtDNA distances
Uncorrected pairwise distances were calculated on PAUP4* [ 35 ] using the mtDNA dataset ( S4 Appendix ). Intraspecific similarity distances are generally lower than 2%. However, there are several exceptions involving individuals from different geographic regions, which are or could be assigned to different subspecies or even species (see discussion): the extinct Japanese river otter ( Lutra lutra , LC050126) versus extant representatives of Lutra lutra (2.1%); Ursus arctos isabellinus versus other brown bear subspecies (2.3–2.4%); Ursus thibetanus laniger versus Ursus thibetanus mupinensis (2.3%); the three samples of Melogale moschata (2.2–2.9%); the three samples of Viverricula indica (2.5–3.1%); Canis lupus familiaris versus Canis lupus chanco (2.6%); the two samples of Paradoxurus hermaphroditus (3.0%); the two samples of Leopardus pardalis (2.9%); the two mtDNA lineages identified in Prionailurus bengalensis (3.6%); the two samples of Herpestes javanicus (4.5%), although BLAST searches in NCBI suggest that the sequence NC_006835 belongs in fact to Herpestes auropunctatus (see discussion); the two samples of Mungos mungo (6.3%), but we suggest that the sample SRR7704821 may belong in fact to Mungos gambianus (see discussion).
Interspecific distances are generally higher than 2% of similarity. However, several species show more similar mitogenomes: Martes martes and Martes zibellina (1.9%); Phoca largha and Phoca vitulina (1.9%); Mustela putorius versus Mustela eversmannii (1.2%) and Mustela nigripes (1.6%); Felis catus and Felis silvestris (0.7%); Zalophus californianus and Zalophus wollebaeki (0.5%); Urocyon cinereoargenteus and Urocyon littoralis (0.4%). Within Arctocephalus forsteri , there are two mtDNA lineages: 17 mitogenomes are similar to the mitogenome of Arctocephalus australis (MG023139) (1.3%), whereas the 28 other mitogenomes are more divergent (2.1%). Within Ursus arctos , we found six mtDNA lineages showing between 1.1% and 2.4% of nucleotide divergence, but brown bears sampled on the Alaskan ABC islands (northern portion of the Alexander Archipelago) have mitogenomes which are highly similar to those of Ursus maritimus (0.4–0.5%).
Phylogeny of Carnivora
The Bayesian tree of Fig 2 was reconstructed from the mtDNA alignment. The results show that most nodes of the tree were highly supported (PP ≥ 0.95) and were similarly recovered as monophyletic entities using two other methods (nodes highlighted with a filled back circle in Fig 2 ), i.e. in the Bayesian tree reconstructed using the mtDNA-Tv alignment and JC69+I+G model ( S5 Appendix ) and in the Bootstrap 50% majority-rule consensus tree reconstructed from the MRP matrix of the SuperTRI analysis ( S6 Appendix ). Many of these nodes show maximal support in the SuperTRI analysis, as they were found monophyletic in all 10 Bayesian trees reconstructed from the 10 half-overlapping sub-datasets of the mtDNA alignment. For all these nodes, the phylogenetic signal is therefore robust across all parts of the mtDNA alignment.
10.1371/journal.pone.0240770.g002
Fig 2
Phylogeny of Carnivora based on mitogenomes.
The Bayesian tree was reconstructed using the mtDNA dataset (220 taxa and 14,892 bp) and GTR+I+G model. The two outgroup species are not shown. Species names follow the classification of the IUCN [ 1 ]; the taxa written in red highlight the taxonomic issues discussed in the main text. The accession numbers of the 42 mitogenomes of Carnivora specially sequenced for this study are indicated in red. The eight mitogenomes here assembled from SRA data are shown in green. The blue circle associated to Mustela nudipes MH464792 indicates that the mitogenome was originally misassigned to Viverra tangalunga . Fossil species are followed by the symbol “†”. For each terminal taxon, the number of similar mitogenome(s) found in GenBank (pairwise distance < 1%) is indicated after the accession number. Dash branches indicate nodes supported by posterior probability (PP) < 0.95. Black circles indicate nodes that are also monophyletic in the two following trees: SuperTRI bootstrap 50% majority-rule consensus tree; and Bayesian tree obtained from the analysis of the mtDNA-Tv dataset and JC69+I+G model. Grey circles show nodes that are not found to be monophyletic with one of the two methods detailed above. White circles indicate nodes that are not monophyletic in both mtDNA-Tv and SuperTRI bootstrap consensus trees. No information was provided for the nodes highly supported by the SuperTRI analyses, i.e. which were found monophyletic in all the 10 Bayesian trees reconstructed from the 10 half-overlapping sub-datasets of the mtDNA dataset. For nodes less supported by the SuperTRI analyses, the number of Bayesian trees (< 10) showing the nodes is indicated.
Among the nodes supported by all three methods of tree reconstruction and at least seven of the 10 Bayesian SuperTRI analyses are the following taxa: order Carnivora; suborders Caniformia and Feliformia; infraorders Arctoidea, Cynoidea, Feloidea and Viverroidea; all the 14 families represented by at least two members in our analyses (i.e. all families except Odobenidae and Nandiniidae); the subfamilies Euplerinae, Felinae, Galidiinae, Genettinae, Guloninae, Helictidinae, Hemigalinae, Herpestinae, Ictonychinae, Lutrinae, Machairodontinae, Melinae, Monachinae, Mungotinae, Mustelinae, Paradoxurinae, Phocinae, Tremarctinae, Ursinae and Viverrinae; the genera Catopuma , Felis , Genetta , Herpestes , Leopardus , Lutra , Lynx , Martes , Meles , Mirounga , Monachus , Mungos , Panthera , Paradoxurus , Phoca , Prionailurus , Prionodon , Puma , Urocyon , Vulpes and Zalophus ; and the species Ailurus fulgens , Canis lupus , Civettictis civetta , Leopardus pardalis , Lutra lutra , Melogale moschata , Mustela sibirica , Panthera onca , Panthera pardus , Panthera tigris , Panthera uncia , Paradoxurus hermaphroditus , Prionailurus planiceps , Procyon lotor , Ursus maritimus , Ursus spelaeus , Ursus thibetanus and Viverricula indica . The results support the non-monophyly of four genera: (1) Arctocephalus , because Arctocephalus pusillus is divergent from Neophoca , Phocarctos and other species of Arctocephalus ; (2) Canis , because Canis adustus is more distantly related to other Canis species than to Cuon alpinus ; (3) Mustela , because Mustela frenata is the sister-species of Neovison vison , whereas all other species of Mustela are enclosed into a robust clade; and (4) Ursus , because Helarctos malayanus is closely related to Ursus americanus and Ursus thibetanus , whereas Ursus arctos , Ursus maritimus and Ursus spelaeus are grouped together. In addition, three species are not found monophyletic: (1) Arctocephalus forsteri is paraphyletic with respect to Arctocephalus australis ; (2) Prionailurus bengalensis is paraphyletic with respect to Prionailurus viverrinus ; and (3) Ursus arctos is paraphyletic with respect to Ursus maritimus .
BEAST chronogram inferred from the mtDNA alignment
Molecular estimates of divergence times are shown in the chronogram provided in Fig 3 . The ages were inferred with BEAST using the mtDNA alignment and the 22 calibration points detailed in Table 1 , including 21 fossil calibration points and a single molecular calibration point (used for the MCRA of Ferae). Two analyses were performed using either a uniform or log-normal prior distribution on the calibrated node ages (named “U approach” and “L approach”, respectively). The two BEAST chronograms show the same tree topology ( Fig 3 ), which is similar to the one reconstructed under MrBayes ( Fig 2 ). As expected, the chronogram inferred with the “L approach” show more recent estimates of divergence times (values highlighted in blue in Fig 3 ) than the chronogram inferred with the “U approach”.
10.1371/journal.pone.0240770.g003
Fig 3
A molecular timescale for carnivoran evolution.
The estimates of divergence time were calculated under BEAST v.2.4.7 using the GTR+I+G model on the mtDNA dataset. The asterisks show the 21 fossil calibration points used for molecular estimation (see Table 1 for details). The chronogram, mean ages (values in black) and associated 95% confidence intervals (grey bars) were inferred using a uniform prior distribution for fossil calibration points (“U approach”). For comparison, the values in blue are mean ages estimated using a log normal prior distribution for fossil calibration points (“L approach”; see main text for details and discussion). Species names follow the classification of the IUCN [ 1 ]; the name of the taxa written in red have been changed following our results which suggest these taxonomic changes (see the discussion for details).
Using the Geologic Time Scale v. 5.0 [ 48 ] for the correspondence between estimated divergence times and geologic epochs, the results suggest that the crown Carnivora divided into Caniformia and Feliformia during the early Eocene at around 52.7–46.7 Mya. Subsequently, the Canidae separated from other families in the early/middle Eocene at around 48.0–42.5 Mya, followed by the Ursidae and Pinnipedia in the middle/late Eocene at around 43.4–38.3 Mya and 41.4–36.5 Mya, respectively. The diversification of the Musteloidea began near the Eocene/Oligocene boundary at around 37.4–33.0 Mya, but the Mustelidae diverged from the Procyonidae in the early Oligocene at around 34.4–30 Mya. The family Phocidae separated from the Otarioidea in the Oligocene at around 27.2–22.9 Mya, whereas the separation between the Otariidae and Odobenidae occurred in the early Miocene at around 20.0–16.7 Mya. Within the Feliformia, the Nandiniidae diverged from other families at the Eocene-Oligocene transition at around 34.4–31.1 Mya. Then, the Feloidea and Viverroidea split in the Oligocene at around 29.7–27.0 Mya. The separation between the Felidae and Prionodontidae occurred in the Oligocene at around 27.4–24.7 Mya. Within the Viverroidea, the Viverridae diverged from other families in the Oligocene at around 28.2–25.8 Mya, followed by the Hyaenidae in the late Oligocene at around 24.5–22.9 Mya, whereas the split between the Eupleridae and Herpestidae took place in the Miocene at around 22.3–21.2 Mya. Within carnivoran families, the generic diversification occurred during the Miocene (all genera of Canidae, Herpestidae, Mephitidae, Procyonidae and Viverridae; most genera of Eupleridae, Felidae, Hyaenidae, Mustelidae, Phocidae and Ursidae) or more rarely during the Pliocene (most genera of Otariidae, excepting Callorhinus ; Aonyx / Lutrogale ; Caracal / Profelis ; Histriophoca / Pagophilus ; Hyaena / Parahyaena ; Hydrurga / Leptonychotes ; Galidictis / Mungotictis + Salanoia ; Helarctos / Ursus ) or Pleistocene ( Halichoerus / Phoca / Pusa ; Mungotictis / Salanoia ).
Discussion
Mitogenomic variations at the species level
In 88% of the species for which at least two mitogenomes were available (65 out of 74), intraspecific distances were found to be less than 2%. In nine species, we detected mtDNA distances greater than 2% ( S4 Appendix ). Such a high mitogenomic divergence suggests that the samples may belong to two or more distinct species, because of either imperfect taxonomy or species misidentification (human error). Other explanations are however possible, such as mtDNA introgression from another species or high levels of mtDNA divergence due to strong female philopatry.
In mammals, females are generally philopatric, spending their lives close to their birthplace, whereas males typically undertake longer-distance dispersals during their lives [ 49 – 51 ]. Indeed, because they have to take care of their young, females tend to stay in areas where they can predict risks, as well as the resources. With time, this behaviour can result in a high spatial genetic structure of mtDNA variation since the mitogenome is transmitted maternally. The impact of female philopatry on the mtDNA evolution is expected to be more important if the preferred habitat-type is delimited by strong geographic barriers representing higher risks associated with dispersal of females (and their young), for example, open areas for small forest fruit bats [ 52 ] or large rivers and mountains for giraffes [ 53 ]. Female philopatry may therefore result in the selection of highly divergent mtDNA haplotypes in geographically isolated maternal lineages, whereas gene flow can be still maintained through long-distance male dispersals. Female philopatry can be advanced to explain the high intraspecific mtDNA distances found between populations of bears of the species Ursus arctos (six lineages, 1.1–2.4%) and Ursus thibetanus (four linages, 1.0–2.3%). This hypothesis is supported by previous phylogeographic studies showing strong geographic structure of mtDNA variation [ 54 , 55 ] and by field data on brown bears indicating that natal dispersal distances are five times smaller for females [ 56 ].
In several other species, we found two or three mitogenomic haplotypes or haplogroups which are separated by more than 2%. In the Eurasian otter ( Lutra lutra ), Waku et al. [ 57 ] sequenced three mtDNA genomes from extinct river otters of Japan: the two mtDNA genomes of otters from Honshu are similar to those found in extant Eurasian otters (< 1%), whereas the genome of the otter from Shikoku Island differs by 2.1%, suggesting that it belongs to a distinct subspecies ( Lutra lutra nippon ) or species ( Lutra nippon ). Since female philopatry versus male biased dispersal was well attested by both radiotracking and genetic data in European otters [ 58 ], we consider that nuclear data are needed to solve this taxonomic issue.
The three mtDNA haplotypes available for the small-toothed ferret badger ( Melogale moschata ) show comparatively high nucleotide distances, i.e., between 2.2% and 2.9%. The genus Melogale includes five species [ 1 ]: Melogale cucphuongensis known from a few animals collected in northern Vietnam [ 59 , 60 ], Melogale everetti on Borneo, Melogale moschata which is widely distributed from North-east India to Taiwan through China and Indochina, Melogale orientalis on Java, and Melogale personata , which is found in mainland Southeast Asia. In agreement with the recent study of Rozhnov et al. [ 61 ], our phylogenetic analysis based on cytochrome b ( CYB ) sequences ( S7 Appendix ) shows that Melogale cucphuongensis is the sister-group of a clade composed of three divergent mtDNA lineages: (i) Melogale personata , as represented by 16 samples from Vietnam; (ii) a first mtDNA haplogroup of Melogale moschata , which contains five samples from Vietnam (including our MW257240 sequence) and the reference mitogenome (NC_020644, unknown origin); and (iii) a second mtDNA haplogroup of Melogale moschata , which contains a sample from Taiwan (subspecies subaurantiaca ) and a sample from China (MN400429, from Fei Zhou). Since the holotype of Melogale moschata was collected in the Guangdong Province of South China [ 62 ], our CYB tree suggests that the two samples from China and Taiwan represent the species Melogale moschata , whereas the samples from Vietnam currently assigned to Melogale moschata may belong to a new species, confirming that Vietnam is a key hot spot for the genus Melogale .
Three mitogenomic haplotypes were available for the small Indian civet ( Viverricula indica ), representing the three main mtDNA lineages previously identified [ 63 ], i.e. the subspecies V . i . indica from India, Madagascar (where it was introduced), Sri Lanka, West China and northern Indochina; V . i . rasse from southern Indochina to Java; and V . i . pallida , from East China to Taiwan. The mitogenomic distances are between 2.5% and 3.1%, suggesting that they may be treated as different species, but nuclear markers are needed to further investigate this taxonomic issue. Gaubert et al. [ 63 ] concluded that these three mitogenomic lineages diverged from each other in the Pliocene, between 3.2 and 2.6 Mya, but our dating estimates rather support that the divergences occurred in the Pleistocene, between 2.2–1.9 Mya and 1.6–1.4 Mya.
In the common palm civet ( Paradoxurus hermaphroditus ), the two mtDNA haplotypes differ by 2.9%, suggesting that they may come from two separate species. Phylogeographic studies based on both mitochondrial and nuclear sequences ( CYB , control region, intron 7 of FGB ) [ 64 , 65 ] supported the existence of three distinct species of Paradoxurus : Paradoxurus hermaphroditus (Indian and Indochinese regions), Paradoxurus musangus (mainland Southeast Asia, Sumatra, Java and other small Indonesian islands) and Paradoxurus philippinensis (Mentawai Islands, Borneo and the Philippines). Only the first two of these were represented in our mitogenomic study. Since they have overlapping distributions in mainland Southeast Asia, where it can be assumed that Paradoxurus hermaphroditus dispersed from the north of India while Paradoxurus musangus arrived from the south (see Fig 2 in [ 64 ]), the possibility of interspecific hybridization should be investigated in this region.
The two mitogenomes sequenced for the ocelot ( Leopardus pardalis ) differ by 3.0%, suggesting that this taxon may be split into two separate species, as proposed by Nascimento [ 66 ] based on a morphological analysis of 591 specimens of Leopardus spp. Using an alignment of the 5’-part of the mitochondrial control region, Eizirik et al. [ 67 ] have suggested a further division into four geographic groups. However, there is a 100-bp stretch of missing data in all their sequences, which is located in a region where we detected three motifs of 80 bp repeated in tandem in the mitogenome of Leopardus pardalis mitis and only two motifs in that of Leopardus pardalis pardalis . These repetitive sequences, named RS2 by Hoelzel et al. [ 68 ], were found in variable number in all families of Feliformia, from two in Prionodon pardicolor [ 69 ] to five in Nandinia binotata [ 70 ]. These variations in RS2 tandem repeats may pose serious problems of homology for DNA alignment. If RS2 repeats are removed from the alignment of Eizirik et al. [ 67 ], there is no robust signal for phylogenetic relationships within Leopardus pardalis ( S8 Appendix ). We recommend therefore to study the phylogeography of ocelot lineages with mitochondrial protein-coding genes, such as CYB and CO1 genes, or full mitogenomes. For taxonomy purpose, the analyses should be completed with nuclear markers, as Wultsch et al. [ 71 ] have shown evidence for female philopatry versus male-biased dispersal in ocelots sampled in Belize (Central America).
Two divergent mtDNA haplogroups were found for the leopard cat Prionailurus bengalensis (3.6%), corresponding to the Asian mainland leopard cat and Sunda leopard cat [ 72 , 73 ]. A species-level separation was confirmed by Y chromosome and whole-genome SNP studies [ 7 , 72 ]. Therefore, we follow Kitchener et al. [ 74 ] in recognizing two species of leopard cats: the mainland leopard cat Prionailurus bengalensis for Asian mainland leopard cats and the Sunda leopard cat Prionailurus javanensis for leopard cats from Java, Sumatra and Borneo. Previous molecular estimates have provided different ages for their divergence, between 2.67 and 0.92 Mya [ 7 , 72 , 73 ]. Although based on different calibration points, our estimate of 2.7–2.4 Mya corroborates that of Luo et al. [ 72 ]. This suggests that maternal linages were first isolated at the Pliocene/Pleistocene boundary, when the glacial/interglacial oscillations commenced. However, the mitogenomic tree does not support a sister-group relationship between the two species of leopard cats, as Prionailurus bengalensis is closer to Prionailurus viverrinus , the fishing cat, than to Prionailurus javanensis . This result contrasts with nuDNA trees showing a sister relationship of Prionailurus bengalensis and Prionailurus javanensis [ 7 , 72 ]. In agreement with the scenario proposed by Li et al. [ 7 ], such a discordance suggests that a mtDNA introgression occurred from Prionailurus bengalensis to Prionailurus viverrinus in the Pleistocene epoch, here dated at 2.1–1.8 Mya.
The two mitogenomes of the Javan mongoose ( Urva javanica , previously named Herpestes javanicus , see [ 75 ] for details) differ by 4.5% in our alignment. BLAST searches in NCBI show that the reference genome sequenced for Urva javanica (NC_006835) belongs in fact to its sister-species, Urva auropunctata (small Indian mongoose), previously named Herpestes javanicus or Herpestes auropunctatus [ 75 ]. In agreement with this, the sample was collected in Fiji where the small Indian mongoose is known to have been introduced [ 76 ].
Two highly divergent mtDNA haplotypes (6.3%) were detected in banded mongooses ( Mungos mungo ) from Halle Zoological Garden (Germany) and from the San Diego Zoo Institute for Conservation Research (USA). The geographic origins of these animals are unknown, but such a high pairwise distance suggests that the two mitogenomes may represent the two species currently described in the genus Mungos : the banded mongoose ( Mungos mungo ), which is widely distributed in sub-Saharan Africa; and the Gambian mongoose ( Mungos gambianus ), which is endemic to West Africa. BLAST searches in NCBI show that our MW257205 sequence shares 99.9% of CYB identity with a mongoose from Kenya (AY928674) [ 77 ], confirming that it belongs to the species Mungos mungo . By contrast, the SRR7704821 sequence shows only 93% of identity with Mungos mungo AY928674, suggesting that it may rather belong to Mungos gambianus . Morphologically, the two species are similar but Mungos mungo is characterized by 10 to 15 dark dorsal stripes that are absent in Mungos gambianus . These marked differences render the hypothesis of misidentification unlikely. As an alternative, a mitochondrial introgression from Mungos gambianus to Mungos mungo may have occurred either in captivity or in the wild, as the two species can be found in sympatry in West Africa [ 1 ].
Shallow mitochondrial phylogeny
The mtDNA distances calculated between closely-related species are generally higher than 2% (73% of the comparisons detailed in S4 Appendix ). However, there are several exceptions in the genera Arctocephalus , Felis , Martes , Mustela , Phoca , Urocyon , Ursus and Zalophus . Four main hypotheses can be advanced to explain the existence of similar mtDNA genomes in two putative species: species misidentification (human error), imperfect taxonomy (species synonymy), mtDNA introgression or recent speciation event during the Pleistocene [ 22 ].
Within Arctocephalus , two divergent mtDNA lineages (2.1%) were sequenced for the New Zealand fur seal ( Arctocephalus forsteri ) [ 78 ]. One of them was found to be more similar to the single mitogenome available for the South American fur seal, Arctocephalus australis (1.3%). The mitochondrial paraphyly of Arctocephalus forsteri suggests a mtDNA introgression from Arctocephalus australis at 0.8–0.7 Mya. Nuclear data should be sequenced to confirm this hypothesis and to further explore a possible taxonomic issue between South American and New Zealand fur seals.
Within Felis , the mitogenome of the domestic cat ( Felis catus ) is very similar to those of the wild cat, Felis silvestris (0.7%), including the African wild cat ( Felis silvestris lybica ), European wild cat ( Felis silvestris silvestris ) and Chinese desert cat ( Felis silvestris bieti ). These three subspecies and the domestic cat are either considered separate species [ 74 ] or subspecies of Felis silvestris [ 79 , 80 ]. Such a genome similarity between the domestic cat and wild cat was expected because the cat was domesticated from the Near Eastern wild cat Felis silvestris lybica [ 79 ] and can hybridize with wild forms [ 81 ]. Our results confirm that these taxa should be considered subspecies rather than species.
Within Martes , the pine marten ( Martes martes ) from western Europe and the sable ( Martes zibellina ) from Siberia and adjacent areas have quite similar mitogenomes (1.9% divergence), and our estimation of divergence time is 1.1–1.0 Mya, which is in agreement with Law et al. [ 8 ]. However, high levels of reciprocal mtDNA introgression have been detected in populations of northern Urals [ 82 ], where the two species are found in sympatry. In this hybrid zone, levels of gene flow should be further studied using nuclear markers to determine if introgression between the two taxa is asymmetric or not.
Within Mustela , several species show low levels of mitogenomic divergence (1.2-1-6%), including the western polecat ( Mustela putorius ) distributed in western Europe, the steppe polecat ( Mustela eversmanii ) found in eastern Europe and Asia, and the black-footed ferret ( Mustela nigripes ) endemic to North America. In addition, the mitogenome of the European mink, Mustela lutreola (MT304869, not included in our phylogenetic analyses) shares between 99.1% and 99.2% of identity with the four mitogenomes available for Mustela putorius . Our estimations indicate that these four closely related species of Mustela have diverged from each other after 1.1–1.0 Mya. However, Mustela putorius , Mustela eversmanii and Mustela lutreola can be found in sympatry in Europe [ 1 ], suggesting possible mtDNA introgression during the Pleistocene epoch. Cabria et al. [ 83 ] have shown evidence for mtDNA introgression from Mustela lutreola to Mustela putorius , resulting apparently from asymmetric interspecific hybridization, i.e. involving only females of European mink and males of polecat. Since the western and steppe polecats are known to occasionally hybridize in Eastern Europe [ 84 ], gene flow should be also studied between the two taxa.
Within Phoca , harbor seals ( Phoca vitulina ) and spotted seals ( Phoca largha ) are parapatric sibling species showing 1.9% of differences in our mitogenomic alignment. Although the two species are known to hybridize in captivity [ 1 ], they were found reciprocally monophyletic in previous mtDNA studies [ 85 , 86 ] and there is no evidence of mixed ancestry in wild populations found in sympatry (e.g. in Alaska [ 86 ]). The hypothesis of a recent Pleistocene speciation, at 1.1–1.0 Mya according to our estimation, is therefore the most likely explanation.
Within Urocyon , the island fox ( Urocyon littoralis ), restricted to six Channel Islands located off the coast of southern California (USA), and the mainland grey fox ( Urocyon cinereoargenteus ) share very similar mtDNA genomes (0.4%). In addition, the species Urocyon cinereoargenteus was found to be paraphyletic in the mitochondrial tree of Hofman et al. [ 87 ] due to the inclusive placement of Urocyon littoralis . Even if the island fox is approximately 25% smaller than the mainland grey fox [ 88 ], we suggest that the island fox should be rather treated as a subspecies of Urocyon cinereoargenteus . Nuclear genome comparisons are however needed to definitively clarify this taxonomic issue.
Within Ursus , bears living on the Alaskan ABC islands have mitogenomes which are very similar (0.4–0.5%) to those sequenced for extant polar bears and the fossil of Svalbard dated between 130 ka and 110 kya [ 13 ]. Two possible scenarios of mitochondrial introgression have been previously proposed: Hailer et al. [ 89 ] suggested that the ancestor of polar bears was introgressed by brown bears between 166 and 111 kya; whereas Hassanin [ 90 ] suggested that different populations of brown bears were introgressed by polar bears at two glacial periods of the Pleistocene, at 340 ± 10 ka in western Europe, and at 155 ± 5 ka on the ABC islands, and probably also in Beringia and Ireland based on ancient mtDNA sequences.
Within Zalophus , two species are currently recognized: the Californian sea lion ( Zalophus californianus ), which is found on the Pacific coasts of North America; and the Galápagos sea lion ( Zalophus wollebaeki ), which has been considered as a subspecies of Zalophus californianus by some authors [ 91 ]. Despite the low mitogenomic distance measured here between the two species (0.5%), Wolf et al. [ 92 ] have previously concluded that they are separate species because they were found reciprocally monophyletic with D-loop and CYB sequences and numerous private alleles were detected for both taxa at most of the 25 investigated microsatellite loci. Our molecular estimate of divergence time between these two species is 0.3–0.2 Mya, which is much more recent than previous estimations based on D-loop sequences, i.e. 0.8 Mya [ 93 ] and 2.3 ± 0.5 Mya [ 92 ].
Mitogenomic variations at the genus level
In mammals, most events of interspecific diversification at the genus level are more recent than 10 Mya and the great majority of them occurred during the Pliocene and Pleistocene epochs. This trend, previously reported in various genera of Cetartiodactyla [ 22 ] and Primates [ 94 ], is here confirmed in the following 25 carnivoran genera: Arctocephalus , Catopuma , Felis , Genetta , Leopardus , Lutra , Lynx , Martes , Meles , Melogale , Mirounga , Monachus , Mungos , Mustela , Panthera , Paradoxurus , Phoca , Prionailurus , Prionodon , Puma , Pusa , Ursus , Urva and Vulpes . Our analyses suggest however that two mustelid species separated from their congeneric representatives in the Middle Miocene: Martes pennanti at 14.2–12.4 Mya; and Mustela frenata at 13.4–11.8 Mya. In addition, the phylogenetic positions of these two species result in the polyphyly of the two genera Martes and Mustela , as previously found using both mtDNA and nuDNA markers [ 8 ]. Taken together, these results clearly indicate that Martes pennanti should be placed in its own genus Pekania , whereas Mustela frenata should be placed in the genus Grammogale . Indeed, the genus Grammogale was used to unite Mustela africana and Mustela felipei in a separate genus [ 95 ]. In addition, several molecular studies have shown that these two species fall into a robust clade with Mustela frenata and that they are the sister-group of Neovison vison [ 8 , 96 ]. We recommend therefore to include the four species Mustela africana , Mustela felipei , Mustela frenata and Neovison vison into the genus Grammogale .
Within the tribe Canini, our mitogenomic tree shows that Cuon alpinus (dhole) is the sister-group of all Canis species, excepting Canis adustus (side-striped jackal) which diverged from them at 8.7–7.4 Mya. The African wild dog ( Lycaon pictus ) occupies a more basal position, but this placement is only supported by 6/10 overlapping datasets in our SuperTRI analyses, indicating that an alternative position cannot be excluded. The mitochondrial phylogeny differs from published nuclear phylogenies [ 97 , 98 ] in which Canis adustus was found to be related to Canis mesomelas (black-backed jackal), but more divergent from the clade composed of other Canis species, Cuon and Lycaon . Following these results, in order to keep the genus Canis monophyletic, the species Canis adustus and Canis mesomelas should be placed in a different genus, which is Lupulella according to Viranta et al. [ 99 ]. The mito-nuclear discordance for the monophyly of the genus Lupulella was not discussed in previous studies. Our interpretation is that a mitochondrial introgression occurred at 6.2–5.2 Mya from an ancestor of Canis species to the lineage leading to Lupulella mesomelas . In agreement with the genomic study of Gopalakrishnan et al. [ 100 ], which concluded to pervasive gene flow among Canis species, our results suggest that interspecies hybridization has been also frequent in the early evolutionary history of canid genera in Africa.
Our chronogram in Fig 3 shows that no intergeneric divergence occurred during the Pleistocene epoch, except the split between the Malagasy euplerid genera Mungotictis and Salanoia , and the separation between the seal genera Halichoerus , Phoca and Pusa . A Pleistocene diversification was also found in previous molecular dating analyses of the Eupleridae and Phocidae [ 8 , 101 ]. In the absence of striking morphological feature for diagnosing these genera, we recommend synonymizing Halichoerus and Pusa with Phoca , as proposed in previous studies [ 102 ], and Mungotictis with Salanoia .
Changes in base composition and their impact on evolutionary rates
Mitogenomic rearrangements did not occur during the evolutionary history of the Carnivora. This deeply contrasts with other animal groups for which inversions of mitochondrial protein-coding genes and control region have resulted in many local or full reversals of base composition [ 30 , 31 ]. Despite this, five carnivoran species in our Bayesian tree ( Fig 2 ) have a longer terminal branch than other representatives of the same family or superfamily: Cynogale bennetti within the Viverridae, Homotherium latidens within the Felidae, Mellivora capensis within the Mustelidae, Nasua nasua within the Procyonidae and Odobenus rosmarus within the Otarioidea. In addition, two higher taxa of Caniformia show longer branches: Otarioidea and Musteloidea. Since branch length is proportional to the number of substitutions, these long branches indicate either an acceleration in the rates of substitution or alternatively an important change in the pattern of substitution [ 103 , 104 ]. As shown in Fig 1 , Cynogale , Homotherium and Mellivora show a divergent base composition. For Cynogale and Mellivora , we can therefore assume that a change in the pattern of substitution has taken place during their recent evolution. The case of Homotherium is more problematic. The mitogenome of this extinct genus was sequenced from a bone (YG 439.38) dated at >47,500 years [ 16 ]. Since ancient DNA molecules are known to exhibit a high rate of cytosine deamination, the sequences can contain artefactual C = >T and G = >A substitutions [ 105 ]. By comparing the mitogenomes of all felid species, this is exactly the pattern observed for third-codon positions, as Homotherium is characterized by less cytosines (28.94% versus 28.96–35.21% in other felid species), more thymines (25.41% versus 17.44–23.51%), less guanines (2.80% versus 4.69–8.41%) and more adenines (42.85% versus 38.41–42.79%). As a consequence, we conclude that sequencing errors introduced by DNA damage are the cause of the long branch of Homotherium in Fig 2 .
The mitogenomes of Nasua and Odobenus have a base composition that is not so divergent from their closest relatives, suggesting that an acceleration of substitution rates is the cause of their long branch in Fig 2 . As pointed out by Bromham [ 106 ], mammals evolving with faster evolutionary rates have generally a small body-size, but the cause of this body-size effect continues to be debated, because there are many possible mechanisms including shorter generation times, shorter lifespan, higher fecundity, larger litter size and higher metabolic rates. From this point, the walrus ( Odobenus rosmarus ) is intriguing because it is one of largest pinnipeds. Its field metabolic rate is 381.2 MJ/day, which is six to 32 times more than in other species of Pinnipedia [ 107 ]. Whereas most other pinnipeds hunt pelagic organisms, such as fish and cephalopod species, in the water column, the walrus feeds on benthic species and prefers molluscs, especially clams [ 108 ]. This special diet and associated behaviour, with long exposure in the cold coastal waters of the Arctic Ocean, may therefore explain the increased rate of mtDNA evolution detected in the walrus. Repetitive bottlenecks generated by climatic oscillations during the Pleistocene may be proposed as an alternative hypothesis. Nuclear genome comparisons between pinniped species will help to decipher between these two main hypotheses, as we can expect accelerated rates of substitution in both mtDNA and nuDNA genomes in case of repetitive bottlenecks.
Robust and reliable mitogenomic phylogeny for deep relationships within Carnivora
Previous molecular studies have shown that inferring deep phylogenetic relationships using mitogenomes can be problematic. An important issue concerns the reversals of strand mutational bias: detected in many animal phyla, such as Arthropoda, Chordata and Mollusca, it may be particularly misleading for phylogenetic reconstruction [ 30 , 31 ]. The mitogenome of mammals is not affected by this kind of bias, as its structure is highly conserved among the 27 mammalian orders. However, mutational saturation is a different issue: since the mtDNA evolves with higher mutation rates than the nuclear genome [ 109 ], it is more prone to multiple substitutions at the same site which leads, with time, to the disappearance of the phylogenetic signal [ 110 , 111 ]. Although mutational saturation may be particularly problematic for reconstructing Precambrian, Paleozoic and Mesozoic divergences, it is expected to have less impact for inferring more recent divergences, such as those during the Cenozoic. This explains why mitogenomic sequences have been largely used for inferring interfamilial and intrafamilial relationships in most mammalian groups, including Cetartiodactyla [ 22 ], Feliformia [ 11 ], Primates [ 94 , 112 ], Pholidota [ 113 ], Xenarthra [ 114 ], etc. In agreement with this view, all families and inter-familial levels shown in our mtDNA tree of Fig 2 received high support with all the three phylogenetic approaches (Bayesian trees of mtDNA and mtDNA-Tv datasets and SuperTRI analysis) indicating considerable stability of the topology. Interfamilial relationships are conformed to the nuclear tree published by Eizirik et al. [ 9 ], except the position of the Mephitidae as the sister-group of the Ailuridae. However, this relationship was also found by Law et al. [ 8 ] based on 46 genes (4 mitochondrial and 42 nuclear) and 75 species of Musteloidea.
Diversification of the Carnivora during the Cenozoic
The mtDNA phylogeny reconstructed here presents a good opportunity to study the evolution of the order Carnivora. Indeed, our dense taxonomic sampling allow us to include many fossils as calibration points for estimating divergence times. This point is particularly relevant as the fossil record of Carnivora has been significantly improved over the last 10 years, with the discovery of several key fossils [ 17 – 21 ]. As pointed out in Warnock et al. [ 115 ], the most effective means of establishing the quality of fossil-based calibrations is through a priori evaluation of the intrinsic palaeontological, stratigraphic, geochronological and phylogenetic data. We identified therefore 21 fossil calibration points ( Table 1 ), including one tip calibration (the mitogenome of a late Pleistocene fossil of Ursus maritimus [ 13 ]) and 20 well-constrained calibration points having both minimum and maximum age boundaries interpreted from fossils with known position with respect to extant taxa included in our study. Molecular dating analyses were performed using either a uniform (U) or log-normal (L) prior distribution on the calibrated node ages. The “L approach” used here considers that minimum ages are generally more accurate and reliable than maximum ages because younger fossils are generally more abundant and precisely dated than older fossils as a consequence of taphonomy and dating methods [ 38 ]. As expected, the divergence times estimated with the “L approach” are more recent than those estimated with the “U approach” ( Figs 3 and 4 ). The values calculated with the two approaches are given for all the nodes described below.
10.1371/journal.pone.0240770.g004
Fig 4
Comparison with published chronograms on Carnivora.
The mean divergence times were here estimated with two approaches for the prior distribution on the calibrated node ages: (1) a uniform distribution between maximum and minimum boundaries (“U approach”, blue histograms); and (2) a log-normal distribution (“L approach”, green histograms) (see main text for details). The results were compared with mean ages inferred in Eizirik et al. [ 9 ] (red histograms) and Nyakatura and Bininda-Emonds [ 10 ] (orange histograms).
The age estimated for the MRCA of Carnivora was 52.7 Mya with the “U approach” and 46.7 Mya with the “L approach”, and both are younger than previous estimates published by Eizirik et al. [ 9 ] (59.2 Mya) and Nyakatura and Bininda-Emonds [ 10 ] (64.9 Mya). However, our dating estimates fit well with the end of the warmest period of the Cenozoic era, from the Palaeocene–Eocene Thermal Maximum (PETM) at 56 Mya to the Early Eocene Climatic Optimum (EECO) at around 53–50 Mya [ 116 – 118 ]. In the fossil record, this period was marked by the first appearances of Primates, Perissodactyla and Cetartiodactyla in North America and Europe, as well as Carnivoraforms, a group formed by the crown group Carnivora plus the stem family ‘Miacidae’ [ 2 , 4 , 119 ]. During the early Eocene greenhouse world, rainforests spread on Earth from pole to pole [ 120 , 121 ], suggesting that early carnivorans were small arboreal species. Biogeographically, Carnivoraforms diversified in the three Laurasian continents during the early Eocene, with a possible origin in the late Paleocene of Asia [ 4 ]. Crown carnivorans originated in one of the three continents of the Northern Hemisphere. To know which one, new fossils need to be discovered in the middle Eocene to fill the gap between the Carnivoraforms of the early Eocene and the oldest carnivorans of the late Eocene [ 4 ].
The divergence times estimated for basal relationships in the Caniformia are similar to those published by Eizirik et al. [ 9 ] but much younger than those published by Nyakatura and Bininda-Emonds [ 10 ]: Caniformia = 48.0–42.5 Mya versus 48.2 Mya and 61.2 Mya respectively; Arctoidea = 43.4–38.3 Mya versus 42.6 Mya and 47.8 Mya respectively. For more recent nodes, our estimates are similar to the two previously cited studies: Musteloidea = 37.4–33.0 Mya versus 32.0 Mya and 35.7 Mya respectively; Mustelidae+Procyonidae = 34.4–30.3 Mya versus 27.4 Mya and 31.6 Mya respectively; and Pinnipedia = 27.2–22.9 Mya versus 24.5 Mya and 22.4 Mya respectively. By contrast, the ages inferred for the families are generally much younger in Eizirik et al. [ 9 ] ( Fig 4 ), a result potentially explained by a poor species sampling (two species of Ursidae, three species of Canidae, etc.) and the low phylogenetic signal of nuclear protein-coding genes for the most recent nodes.
The ages estimated for basal divergences in the Feliformia are much younger than in Eizirik et al. [ 9 ] and Nyakatura and Bininda-Emonds [ 10 ]: Feliformia = 34.4–31.1 Mya versus 44.5 Mya and 53.2 Mya respectively; Feloidea = 27.4–24.7 Mya versus 33.3 Mya and 52.9 Mya respectively; and Viverroidea = 28.2–25.8 Mya versus 37.4 Mya and 31.8 Mya respectively. These results may be explained by the use of different calibration points: Feloidea = 34–20 Mya versus >28.5 Mya in Eizirik et al. [ 9 ]; Felidae = 20–14 Mya versus >31.15 Mya in Nyakatura and Bininda-Emonds [ 10 ]; and Viverridae = 34–14 Mya versus >25.72 Mya in Nyakatura and Bininda-Emonds [ 10 ]. The minimum age constraints used by Nyakatura and Bininda-Emonds [ 10 ] for Felidae and Viverridae seem however problematic because they were not extracted from the fossil record, but from the supertree of Bininda-Emonds et al. [ 122 ]. By contrast, the minimum boundary of 28.5 Mya used for Feloidea in Eizirik et al. [ 9 ] was interpreted from early Oligocene fossils assumed to be the oldest felids according to McKenna and Bell [ 123 ]. In Meredith et al. [ 26 ], who used a similar minimum boundary for Feloidea (> 28.3 Mya), the divergence times estimated for Feliformia were found to be highly variable between their eight molecular dating analyses, with mean ages between 35.81 and 43.07 Mya. Three early Oligocene fossils were used as calibration points in this study: Proailurus and Stenogale , which were assumed to be stem Felidae; and Palaeoprionodon , which was supposed to be a stem Prionodontidae. However, the phylogenetic positions of these fossils remain uncertain. Indeed, the three fossil genera formed the sister-group of Feloidea + Viverroidea + Herpestides antiquus † in the phylogenetic analyses of Solé et al. [ 3 , 4 ]. In the classification of Hunt [ 124 ], Palaeoprionodon was included in the subfamily Prionodontinae with Prionodon and other extant viverrid genera, such as Genetta and Pioana . Although Prionodon was treated as a member of the Viverridae by Hunt [ 124 ] and previous authors, it was then found to be the sister-group of the Felidae by Gaubert and Veron [ 125 ] which was confirmed by all more recent studies [ 9 , 10 , 26 ] (see also Fig 2 ). In Nyakatura and Bininda-Emonds [ 10 ], Paleoprionodon was used as a calibration point for the Viverridae (based on Hunt and Tedford [ 126 ]), although it is suggested to be either close to Prionodon or a stem Feliformia [ 3 , 124 , 127 ].
Our dating estimates suggests that the basal split between Nandinia and other genera of Feliformia occurred at the Eocene/Oligocene transition (34 Mya), when a brutal and global cooling of 5°C resulted in the extinction of many taxa and the appearance of several modern mammal lineages [ 128 ]. Accordingly, Nandinia may be the descendant of forest-adapted ancestors, whereas the other lineage of Feliformia may have evolved in response to this important climatic change by adapting to more open vegetation.
Conclusion
Based on a large taxonomic sampling (220 taxa represented by 2,442 mitogenomes) and 21 fossil calibration points (most of them having both minimum and maximum age boundaries and a position based on recent fossil revisions and phylogenies), our study proposes a new time-scaled phylogeny for Carnivora and provides further insights into the evolutionary history of this order. The age estimates for the Carnivora and the two suborders Caniformia and Feliformia fit well with global changes corresponding to the appearance of other mammal lineages. Moreover, our phylogenetic results, although largely similar to other recent phylogenies, suggest several taxonomic changes that would need to be confirmed using nuclear data.
Supporting information
S1 Appendix
Origin of the sequences used in this study.
(PDF)
S2 Appendix
PCR primers used in this study.
(PDF)
S3 Appendix
Base composition.
(PDF)
S4 Appendix
Distances.
(PDF)
S5 Appendix
Bayesian tree reconstructed using the mtDNA-Tv dataset (220 taxa and 14,892 bp) and JC69+I+G model.
(PDF)
S6 Appendix
Bootstrap 50% majority-rule consensus tree reconstructed from the MRP matrix of the SuperTRI analysis.
(PDF)
S7 Appendix
Cytochrome b phylogeny of the genus Melogale .
(PDF)
S8 Appendix
D-loop phylogeny of the species Leopardus pardalis .
(PDF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The paraurethral ducts in men are small, blind channels lined with columnar epithelium [ 1 ]. These ducts run parallel to the terminal part of the urethra for varying distances and open near or within the lips of the external meatus [ 2 ]. The paraurethral ducts appear to be embryological remnants and are not visible to the naked eye [ 3 ]. Neisseria gonorrhoeae can infect the paraurethral duct. Such infection clinically manifests as an erythematous swelling of the external urethral orifice. An abscess may form at the center of the swollen area and perforate outward to form a pinhead-like ostium at the opening of the paraurethral duct. Pressure on the glans penis can result in purulent excretion from this ostium.[ 4 – 8 ] High-frequency ultrasound can be used to examine the morphological features of the paraurethral duct after N . gonorrhoeae infection [ 9 ].This inflammation resolved with ceftriaxone in some patients in the present study, eliminating both the erythematous swelling of the external urethral orifice and the purulent excretion from the ostium. However, the ostium did not close, and overflow of transparent liquid was still visible after squeezing the lesion. The liquid was negative for all pathogens examined, and pathological biopsy of the lesion indicated the presence of paraurethral duct dilatation [ 10 , 11 ]. To better understand the risk factors for paraurethral duct dilatation following the paraurethral duct infection by N . gonorrhoeae in men, we performed a prospective case-control study in which we compared the demographic, behavioral, and clinical data of patients with paraurethral duct infection by N . gonorrhoeae with and without dilatation of the paraurethral duct.
Patients and Methods
This study was approved by the Ethics Committee of the First People’s Hospital of Changshu City, Soochow University, Jiangsu Province, China. All participants provided written informed consent to participate in the study. We obtained written informed consent from the guardians of the minors included in the study.
Inclusion criteria
All patients initially had local erythematous swelling at the external urethral orifice, with an ostium at its center; pressure caused expression of purulent excretion from the ostium, and N . gonorrhoeae was confirmed to be the pathogen. The morphological features of the paraurethral duct were examined using high-frequency ultrasound. The inclusion criteria for the control group were as follows: after ceftriaxone therapy, the local swelling at the external urethral orifice was resolved, the purulent excretion had disappeared, and the ostium was closed; pressure did not cause expression of fluid from the site where the ostium had been located; and there was no relapse during the 3-month follow-up. Thus, the condition was considered cured. The inclusion criteria for the patient group were as follows: after ceftriaxone therapy, the local erythematous swelling at the external urethral orifice was resolved and the purulent excretion had disappeared, but the ostium was not closed and pressure caused expression of transparent fluid from the ostium. However, testing for pathogens in the transparent liquid still showed negative results. The ostium did not close during the 3-month follow-up, and a diagnosis of paraurethral duct dilatation was made.
Exclusion criteria
The exclusion criteria for both groups were as follows: local erythematous swelling was seen at the external urethral orifice, but the lesion did not perforate; two or more skin lesions were present; the lesion was not caused by N . gonorrhoeae ; and the clinical data were incomplete.
Laboratory tests
Paraurethral duct discharge was collected and examined under a microscope to detect intracellular Gram-negative diplococci within phagocytes. Specimens were also cultured to detect N . gonorrhoeae , Ureaplasma urealyticum , or other bacteria, and polymerase chain reaction was used to detect the DNA of N . gonorrhoeae , Chlamydia trachomatis , U . urealyticum , and herpes simplex virus type 1 or 2. Venous blood samples were taken and analyzed using the rapid plasma reagin test, Treponema pallidum hemagglutination assay, and human immunodeficiency virus antibody assay.
Data collection
The demographic, behavioral, and clinical data from patients with paraurethral duct infection by N . gonorrhoeae were collected. These data included age, marital status, educational background, source of infection, sexual orientation, sexual behavior pattern, condom use, prepuce condition, morphological features of the paraurethral duct infected by N . gonorrhoeae (diameter and length of the paraurethral duct), disease course of the paraurethral duct infected by N . gonorrhoeae (amount of time the patient had experienced symptoms before undergoing treatment), C . trachomatis infection of the paraurethral duct, whether the patient drank alcohol or engaged in sexual intercourse during paraurethral duct infection by N . gonorrhoeae , history of paraurethral duct infection by N . gonorrhoeae , and concomitant diseases.
Statistical analysis
All data analyses were performed using SPSS 13.0 for Windows (IBM Corporation, Chicago, IL, USA), and P< 0.05 was considered statistically significant. Continuous variables were analyzed using Student’s t -test, and discrete variables were compared using the chi-square test. Discrete variables that were not feasible for analysis by the chi-square test were analyzed using Fisher’s exact probability test. Multivariate logistic regression analysis was applied to analyze the risk factors for secondary paraurethral duct dilatation. The variables for logistic regression analysis were screened using a stepwise method, and the 95% confidence intervals were calculated.
Results
In total, 106 male patients with paraurethral duct infection by N . gonorrhoeae were diagnosed and treated in our department from October 1997 to May 2015. Serological tests for syphilis and human immunodeficiency virus were negative for all patients. All patients were treated with intramuscular ceftriaxone sodium (1 g once daily for 5 days). Patients with concomitant chlamydial urethritis were simultaneously treated with azithromycin (0.5 g once daily for 5 days). After treatment, paraurethral duct dilatation was found in 15 patients (14.15%) ( Fig 1 ). These 15 patients were included in the patient group, and 71 patients were included in the control group. The age of patients in the patient group was 28 to71 years (mean, 44.9 ± 12.0 years), and that of patients in the control group was 16 to70 years (mean, 30.8 ± 13.2 years; t = 3.8, P< 0.01). There were no significant differences in the other demographic and behavioral features between the two groups ( Table 1 ). In the patient group, the paraurethral ducts had a diameter of 0.7 to1.2 mm (mean, 1.1± 0.2 mm) and a length of 7.5 to11.2 mm (mean, 8.5 ± 1.2 mm). In the control group, the diameter and length were 0.7 to1.4 mm (mean, 1.2 ± 0.2 mm) and 7.1 to12.7 mm (mean, 8.4 ± 1.3 mm), respectively. The diameter (t = −1.74, P = 0.09) and length (t = 0.27, P = 0.79) were not significantly different between the two groups. The disease course ranged from 6 to 35 days (mean, 15.6 ± 6.5 days) in the patient group and from 1 to 16 days (mean, 4.9 ± 2.8 days) in the control group (t = 10.2, P< 0.01). The disease course for both groups was divided into three grades for logistic regression: ≤ 7 days, 8 to 14 days, and ≥15 days. In addition, C . trachomatis infection of the paraurethral duct, a history of paraurethral duct infection by N . gonorrhoeae and disease course were also significantly different between these two groups ( Table 1 ).
10.1371/journal.pone.0166355.g001
Fig 1
(A). A pinhead-like ostium was present at the 3 o’clock position on the left side of the external urethral orifice. ( B). An overflow of transparent liquid was visible after squeezing the lesion.
10.1371/journal.pone.0166355.t001
Table 1 Comparison of demographic, behavioral, and clinical data between patient and control groups.
Variables
Case group (n = 15) n (%)
Control group (n = 71) n (%)
χ2 value
P value
Age (years)
5.046
0.02
<45
6 (40.00)
50 (70.42)
≥45
9 (60.00)
21 (29.58)
Marital status
0.0983
0.76
Unmarried
4 (26.67)
19 (26.76)
Married
11 (73.33)
52 (73.24)
Education background
0.03
0.87
Middle or high school
6 (40.00)
30 (42.25)
College
9 (60.00)
41 (57.75)
Source of infection
0.54
Use of sex workers
14 (93.33)
68 (95.77)
Others
1 (6.67)
3 (4.23)
Sexual orientation
Heterosexual
15 (100)
71 (100)
Homosexual
0 (0)
0 (0)
Bisexual
0 (0)
0 (0)
Sexual behavior pattern
Genital/genital
15 (100)
71 (100)
Genital/oral
0 (0)
0 (0)
Genital/anal
0 (0)
0 (0)
Condom use
0.0076
0.98
Never
12 (80.00)
59 (83.10)
Sometimes
3 (20.00)
12 (16.90)
Prepuce condition
0.0451
0.71
Normal
12 (80.00)
58 (81.69)
Redundant prepuce
3 (20.00)
13 (18.31)
Drinking alcohol during the paraurethral duct infected by gonococcus
0.0841
0.78
Yes
2 (13.33)
5 (7.04)
No
13 (86.67)
66 (92.96)
Sexual intercourse during the paraurethral duct infected by gonococcus
0.2559
0.69
Yes
2 (13.33)
4 (5.63)
No
13 (86.67)
67 (94.37)
C . trachomatis infection in the paraurethral duct
7.490
0.01
Yes
4 (26.67)
2 (2.82)
No
11 (73.33)
69 (97.18)
Previous history of the paraurethral duct infected by gonococcus
5.610
0.02
Yes
4 (26.67)
3 (4.23)
No
11 (73.33)
68 (95.77)
Concomitant diseases
0.06
With diabetes
2 (13.33)
1 (1.41)
Without diabetes
13 (86.67)
70 (98.59)
Course of disease
<0.01
≤ 7 days
2 (13.33)
46 (64.79)
8 to 14 days
4 (26.67)
22 (30.99)
≥15 days
9(60.00)
3(4.22)
After step-by-step screening with logistic regression, the following risk factors were identified: age, disease course, C . trachomatis infection of the paraurethral duct, a history of paraurethral duct infection by N . gonorrhoeae and concomitant disease. The first four risk factors were significant risk factors for secondary paraurethral duct dilatation after multivariate logistic regression analysis ( Table 2 ).
10.1371/journal.pone.0166355.t002
Table 2 Results of logistic regression for identification of significant risk factors.
Parameter
Odds ratio
95%CL for OR
P-value
Age
2.46
1.33–6.39
0.03 *
Chlamydia trachomatis infection in the paraurethral duct
3.96
1.69–8.75
0.01 *
Previous gonococcus infection of paraurethral duct
2.19
1.22–7.48
0.03 *
Concomitant disease (diabetes)
1.12
0.65–1.79
0.46
Disease course a
6.25
2.76–13.26
0.00 **
*P<0.05
**P<0.01
a, amount of time the patient had experienced symptoms before undergoing treatment
Discussion
Infection of the paraurethral duct by N . gonorrhoeae is a localized complication in male patients with gonorrhea. In total, 7058 male patients with gonorrhea that was confirmed both clinically and by laboratory test results were treated in our department from October 1997 to May 2015, and 106 (1.50%) of these patients had accompanying paraurethral duct infection by N . gonorrhoeae . A prolonged disease course, sexual intercourse during gonorrhea, repeated squeezing of the penis, and a redundant prepuce may be risk factors for paraurethral duct infection by N . gonorrhoeae in male patients with gonorrhea [ 6 ]. Paraurethral duct dilatation is a sequela of paraurethral duct infection by N . gonorrhoeae and has characteristic clinical, histopathological, and high-frequency ultrasound findings. Wedge excision is an effective treatment for this condition [ 10 ].
In the current study, univariate analysis showed that age, disease course, infection of the paraurethral duct with C . trachomatis , and a history of paraurethral duct infection by N . gonorrhoeae were significantly different between the patient and control groups. Multivariate logistic regression analysis showed consistent results. A prolonged disease course and repeated infection in male patients with paraurethral duct infection by N . gonorrhoeae may cause the acute inflammation to become chronic. Research using rat models has shown that chronic prostate inflammation can cause fibrosis of the prostate and its surrounding tissues [ 12 ]. Additionally, controlling chronic prostate inflammation can, to some extent, reverse the progression of fibrosis [ 13 ]. Studies of the human prostate have reached similar conclusions [ 14 , 15 ]. Some authors have also found that urethral fibrosis reduces urethral compliance [ 16 ]. Therefore, the pathogenic mechanism of paraurethral duct dilatation may be as follows: a prolonged disease course causes acute inflammation to become chronic, and the paraurethral duct thus becomes more susceptible to fibrosis. As a result, the compliance of the paraurethral duct wall decreases or is even eliminated, and the dilated paraurethral duct cannot close by itself. Therefore, educating patients to receive timely treatment may be an effective intervention.
Paraurethral duct inflammation is most commonly caused by N . gonorrhoeae infection, but it can also be a consequence of infection with C . trachomatis or other pathogens [ 7 , 17 ]. Coinfection with N . gonorrhoeae and C . trachomatis in the paraurethral duct can exacerbate the inflammation because the synergetic effect between N . gonorrhoeae and C . trachomatis induces a high level of bacterial multiplication that aggravates the infection by N . gonorrhoeae [ 18 ].
The pathologic changes caused by repeated and persistent infection with C . trachomatis may result in tissue damage secondary to chronic inflammation [ 19 ]. Previous studies have confirmed that chronic C . trachomatis infection can induce inflammatory damage and fibrosis in genital ducts such as the oviduct [ 20 ]. Furuya et al.[ 21 ] also found that C . trachomatis infection can induce ectasia of the seminal vesicles, which have an epithelial structure similar to that of the paraurethral ducts. Thus, paraurethral duct fibrosis and dilation can be attributed to inflammation caused by C . trachomatis , and treating the C . trachomatis coinfection may be an effective intervention.
Age may contribute to the degeneration of physiological function in patients with paraurethral duct dilatation. One study showed that epithelial regression with senility was the major risk factor for deferent duct dilation and cystogenesis [ 22 ]. Thus, regression of the paraurethral duct epithelium may induce post-infection dilation of the paraurethral duct.
The incidence of paraurethral duct infection by N . gonorrhoeae is low in male patients, and paraurethral duct dilatation secondary to this disease is rare in the clinical setting. In our department, only 15 patients with complete data were identified within the past 18 years. Multicenter studies with larger sample sizes may identify more risk factors for paraurethral duct dilatation.
Supporting Information
S1 Dataset
Raw material of manuscript.
(XLS)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Traffic congestion in freeway network becomes more noticeable in modern societies, introducing negative effects for a sustainable development of the Intelligent Transport System (ITS), such as the declining public transport service level and the increasing traffic costs [ 1 ]. Variable speed limits (VSL) is a new mainline traffic control method and is widely applied on freeway mainlines to solve the traffic congestion problem, such as in Hadiuzzaman and Qiu [ 2 ]. The VSL values are the core of the VSL control, calculated by diverse traffic conditions involving traffic volume and traffic density. Instead of using static speed limits, many studies on the VSL control present a common result that the VSL control can enhance traffic safety, decrease drivers’ complain, and improve the obey rate of drivers. This has been investigated by Moghadam [ 3 ], Hegyi et al. [ 4 ] and Li et al. [ 5 ]. According to these studies, several views exist on the use of VSL control. Heydecker and Addison [ 6 ] emphasized the homogenization effect. In contrast, Abdel-Aty and Wang [ 7 ], Yang et al. [ 8 ], and Yang and Hegyi [ 9 ] mainly focused on the prevention of freeway traffic flow breakdown and maximization of traffic flow. To apply the VSL control strategies in the mainline area without considering the influence of traffic outflows from on-ramp, this approach cannot effectively solve the traffic congestion problem in merging region during peak periods[ 10 , 11 ]. The congestion may transfer to the upstream mainline and the related freeway network, if the VSL control or/and ramp metering in the problem path is applied alone during peak periods.
To overcome these problems mentioned above, a coordinated approach, involving ramp metering, VSL control, and route guidance, has been gradually paid more attention (see e.g. Wang and Papageorgiou[ 12 ], Liu et al.[ 13 ], and Pasqual et al. [ 14 ]).To mitigate the interference of the on-ramp flows and relieve congestion in freeway network, Pasquale et al.[ 15 ] and Kotsialo et al. [ 16 , 17 ] attempted to fit the freeway network control problem in the format of a discrete-time optimal control problem and solve it by using both integrating ramp metering and route guidance. Karimi et al. [ 18 ] proposed an integrated control approach, involving VSL control, ramp metering and route guidance. However, these studies only tested the efficiency of coordinating ramp metering and route guidance. Moreover, most traffic control methods pay little attention to the detailed analysis of the relationship between the traffic-controlled objective and route guidance objective, which are therefore not suitable in certain conditions (such as there is a specific problem path existed in the freeway network).
In this paper, we propose an optimization approach based on a bi-level programming optimal model to relieve the congestion in the freeway network, integrating VSL control, ramp metering, and route guidance. The contributions of this paper can be summarized as follows.
A coordinated optimization approach is proposed to relieve the traffic congestion in a freeway network on the basis of the bi-level programming model.
The upper-level programming model mainly handles the traffic loading distribution equilibrium problem and the total travel time in the network. Since the traditional definition of traffic load ignores some special conditions (e.g. bottleneck region and merging region) existing in the path, a novel traffic load definition is given as the ratio between the maximum traffic flow among each section and the minimum capacity of each section in the same path.
The lower-level programming model coordinates the VSL control and ramp metering. We consider a variety of situations, involving two kinds of common problem paths and comprehensive conditions. Furthermore, we select both travel time and traffic flow in each path as the control goal to establish the objective function.
The rest of this paper is organized as follows. Section 2 contains macroscopic traffic flow model and the extension model. The bi-level optimal model is formulated in detail in Section 3, while Section 4 contains some illustrative and experimental evaluations obtained by different control strategies. Finally, in Section 5, conclusions and some topics for further research are summarized.
Macroscopic traffic flow model
Traffic flow model on normal section
A foundation macroscopic traffic flow model described in detail by Messmer and Papageorgiou [ 19 ] and Kotsialos et al. [ 20 ] is applied in this paper to descript the traffic flow parameters. The macroscopic traffic flow model is used for the description of freeway traffic flow.
To introduce the operation principle of METANET model, we select one link as an example section. The discrete time step is denoted by T (typically, T = 10 s ). Take the link m as an example, which is divided into N m segments with the length of Δ l m , where Δ l m is given by the product of free-flow velocity v f , m , see Fig 1 . The discrete-time instant can be expressed by t = kT . The macroscopic traffic flow parameters are defined as follows. The traffic density ρ m , i ( k ) represents the number of vehicles in single line of segment i of link m during period k . The traffic speed v m , i ( k ) represents the average speed of the vehicles in segment i of link m during period k . The traffic volume q m , i -1 ( k ) represents the number of vehicles inflowing segment i of link m . q m , i ( k ) represents the number of vehicles outflowing of segment i of link m .
10.1371/journal.pone.0204255.g001
Fig 1
Discretized freeway link.
The macroscopic traffic flow model expresses the traffic parameters on freeway networks at a certain space-time domain. For each segment i of link m at each time step k , the following equations are applied.
q m , i ( k ) = ρ m , i ( k ) v m , i ( k ) λ m ,
(1)
ρ m , i ( k + 1 ) = ρ m , i ( k ) + T Δ l m λ m [ q m , i − 1 ( k ) − q m , i ( k ) ] ,
(2)
v m , i ( k + 1 ) = v m , i ( k ) + T τ { V [ ρ m , i ( k ) ] − v m , i ( k ) } + T Δ l m [ v m , i − 1 ( k ) − v m , i ( k ) ] v m , i ( k ) − υ T τ Δ l m ρ m , i + 1 ( k ) − ρ m , i ( k ) ρ m , i ( k ) + κ ,
(3)
V [ ρ m , i ( k ) ] = v f , m exp [ − 1 α m ( ρ m , i ( k ) ρ c r , m ) α m ] ,
(4)
where ρ cr , m represents the critical density of the link m (the density at which the traffic flow reaches the capacity of link m Q cap , m ) and α m represents the model parameter of Eq ( 4 ) which expresses a nonlinear relationship between the traffic density and speed. In addition, τ , υ and κ represent constant parameters determined by the traffic system, driver’s behavior, geometry characteristic of link m , etc. Furthermore, the average speed of the segment i in link m is shown in Eq ( 3 ) which is limited by the minimum velocity v min . A slack movement, which is considered in Eq ( 3 ), includes the expected velocity, the velocity changes caused by the inflow q m , i −1 ( k ), and speed increase (or decrease).
For seeking the reason that the mainstream speed decreases caused by ramp confluence, it is necessary to consider the traffic operation in merging region of the mainline and ramp. When there is an on-ramp on freeway, merging phenomena caused by on-ramp traffic flow and upstream mainline traffic flow appears in merging region and the range can be described by the term
− δ T q 0 ( k ) v m , 1 ( k ) L m λ m ( ρ m , 1 ( k ) + κ ) ,
(5)
where δ represents a constant parameter.
If there is a lane drop, the velocity change caused by weaving phenomena can be expressed as
− ϕ T ρ m , N m ( k ) v 2 m , N m ( k ) ( λ m − λ m + 1 ) L m λ m ρ c r i t , m ,
(6)
where ϕ represents the model parameter and λ m − λ m +1 represents the number of dropped lanes.
The extension model considering VSL
In order to describe the traffic flow evolution under VSL control, we should analyze the traffic characteristics under VSL control. The differences of the traffic operational mechanism whether the road traffic adopts VSL control are described as follows. If the VSL control is adopted, under the free-flow condition, the traffic speed is expressed by the limit speed v vsl instead of the free-flow speed v f , where the limit speed value has to be less than the free-flow speed value, whereas the traffic follows self-organization operating under no special control condition. When the traffic condition is congestion or traffic density is higher than the critical density, the traffic under VSL control keeps a steady and regular operation on account of drivers have to act up to the VSL values to their vehicles. When the traffic density is close to the jam density ρ J , the vehicles have to obey self-organizing operation instead of the control strategy.
According to the discussions above, the desired speed variation caused by the VSL control can be expressed as a function of traffic density. Therefore, the desired speed under the VSL control can be given by
V m , i [ ρ m , i ( k ) ] = min ( v f , m exp [ − 1 α m ( ρ m , i ( k ) ρ c r , m ) α m ] , η V v s l , m ( k ) + ( 1 − η ) v f , m exp [ − 1 α m ( ρ m , i ( k ) ρ c r , m ) α m ] ) ,
(7)
where v vsl , m , i ( k ) represents the speed limit value of segment i of link m during period k and η represents the parameter that drivers obey the VSL control [ 21 , 22 ]. Especially, if η = 1, all the drivers obey the VSL values; else η = 0, all the drivers do not follow the VSL values; else, 0 < η < 1, a part of drivers obey the VSL values. In addition, either on-line or off-line traffic parameters can be obtained through the formulas mentioned in this section, which is possible to establish the following integrated control model as a fundamental for traffic flow description.
Bi-level programming formulation of the freeway network
Bi-level programming optimal model
The bi-level programming optimal model is widely employed for the optimal control and decision problem [ 23 ]. Bi-level programming optimal theory enables the decision manager to achieve the optimal control system by analyzing the theoretical relationship between the upper-level programming system and the lower-level programming system. The bi-level programming optimal model can be shown as follows.
The upper-level programming model
min J O U [ u , k ] = ∑ k = 0 k p − 1 Φ ( x ( k ) , u ( k ) , v ( k ) , d ( k ) , k ) s . t . x ( k + 1 ) = F [ x ( k ) , u ( k ) , v ( k ) , d ( k ) , k ] H ( x ( k ) , u ( k ) , v ( k ) , d ( k ) , k ) ≤ 0 k = 1 , 2 , … , k p ,
(8)
where v ( k ) = v ( u , k ) is determined by the lower-level programming model.
The lower-level programming model
min J O L [ v , k ] = ∑ k ′ = 0 k p − 1 φ ( x ( k ) , u ( k ) , v ( k ) , d ( k ) , k ) s . t . x ( k + 1 ) = f [ x ( k ) , u ( k ) , v ( k ) , d ( k ) , k ] g ( x ( k ) , u ( k ) , v ( k ) , d ( k ) , k ) ≤ 0 k = 1 , 2 , … , k p ,
(9)
where J OU [ u , k ] and J OL [ v , k ] are the upper-level programming model and the lower-level programming model, respectively; u represents the decision vector of the upper-level decision makers, which is influenced by the objective vector of the lower-level decision vector v ; Φ and φ are arbitrary functions, and d represents the disturbance variable; H and g represent the constraint sets of the upper-level decision vector and the lower-level decision vector, respectively.
In this paper, the core concept of the optimal control approach includes two aspects. The first aspect is to alleviate the congestion in the problem path via VSL control and ramp metering. The other is to use route guidance to adjust the traffic flow distribution at the freeway network. The bi-level programming optimal model can analyze the decision problem of the subordinate construction and solve the coupling among different traffic control approaches that belong to different levels. Hierarchical partitioning between freeway traffic control and route guidance meets the demand and concept of a bi-level programming optimal approach. Specifically, in a coordination traffic system, the traffic load balancing and the total travel time minimum are selected as the optimal control objective functions of the upper-level programming system. The lower-level programming system establishes the optimal control model in paths, aiming to relieve the negative effects of the road traffic caused by traffic congestion. Furthermore, the coordinated control strategy proposed in this paper is shown as Fig 2 .
10.1371/journal.pone.0204255.g002
Fig 2
The proposed control system.
Formulation of the upper-level optimal model
The upper-level programming model involves two important factors, the expected benefits of drivers and the freeway network system. The travel time parameter attracts much attention among the expected benefits of driver, as it plays a significant role in the process of driver route selection. Hence, we select travel time as the evaluation index of traffic customer satisfaction and a sub-control objective of the upper-level optimal model. Furthermore, considering the traffic load in each path of freeway network, traffic managers generally balance the traffic distributions via traffic guidance. A change of the travel time can influence drivers’ route choice and traffic distribution in network because drivers pay more attention to the travel time. In this section, the first objective of the upper-level programming model mainly considers the travel time minimum, aiming to enhance the efficiency of the whole freeway network and improve the traffic load distribution equilibrium. The objective function of the upper-level programming of travel time can be presented as follows:
min S T T T = ∑ k ∈ K ∑ i = 1 p ′ F l i ( x l i ( k ) , u l i ( k ) , v l i ( k ) , d l i ( k ) , k ) ,
(10)
where S TTT represents the control objective function; l i represents one path that belongs to the path set Y = ( l 1 , l 2 ,…., l p ′ ); F l i ( x l i ( k ) , u l i ( k ) , d l i ( k ) , k ) represents the travel time of path l i during period k ; x l i ( k ) represents the state variable; u l i ( k ) represents the control variable; v l i ( k ) represents the control variable of the lower-level programming model; d l i ( k ) represents the disturbance variable.
In the upper-level programming model, the accuracy of the travel time calculated affects the traffic control effect. According to the difference of freeway type, the freeway road can be classified into four groups, which are the basic mainline section (BM) l 1 , the basic mainline section involving bottleneck (BMB) l 2 , the basic mainline section including ramp (BMR) l 3 and the basic mainline section including ramp and bottleneck (BMRB) l 4 . A simple example is shown in Fig 3 .
10.1371/journal.pone.0204255.g003
Fig 3
Road alignments: (a) basic mainline section; (b) basic mainline section involving bottleneck; (c) basic mainline section including ramp; (d) basic mainline section including both ramp and bottleneck.
If there are no ramps or bottlenecks on the freeway path as presented in Fig 3(A) , the traffic flow is mainly influenced by the external environment and internal disturbance. The travel time calculation model can be described by the total time of traffic flows through each path in the freeway network. The travel time of l 1 can be expressed by Eq ( 11 ).
F l 1 ( k ) = ∑ i = 1 N l 1 ρ l 1 , i ( k ) λ l 1 Δ x l 1 q l 1 , i ( k ) ,
(11)
where F l 1 ( k ) represents the travel time of l 1 ; N l 1 represents the number of elements; ρ l 1 , i ( k ) represents the traffic density; Δ x l 1 represents the length of l 1 ; λ l 1 represents the number of lanes; q l 1 , i ( k ) represents the traffic flow.
If there is a bottleneck on the freeway path without ramps, e.g. Fig 3(B) , we call basic mainline section involving bottleneck. We select a simple path as an example to describe the travel time of l 2 . In Fig 3(B) , the path can be divided into three main regions, including two mainline segments and one bottleneck segment. According to the basic method of the travel time calculation in Eq ( 11 ) and the change of segment lanes in l 2 , we can calculate the travel time of l 2 based on Eq ( 12 ).
F l 2 ( k ) = ∑ i = 1 N l 2 , 1 ρ l 2 , 1 , i ( k ) λ l 2 Δ x l 2 , 1 , i q l 2 , 1 , i ( k ) + ∑ i = 1 N l 2 , 3 ρ l 2 , 3 , i ( k ) λ l 2 Δ x l 2 , 3 , i q l 2 , 3 , i ( k ) + ∑ i = 1 N l 2 , 2 ρ l 2 , 2 , i ( k ) λ l 2 , 2 Δ x l 2 , 2 , i q l 2 , 2 , i ( k ) ,
(12)
where F l 2 ( k ) represents the total travel time of vehicles on l 2 ; N l 2 , 1 , N l 2 , 2 , and N l 2 3 represent the number of elements on segment 1, segment 2, and segment 3, respectively; ρ l 2 , 1 , i ( k ) and q l 2 , 1 , i ( k ) represent traffic density and traffic flow of the basic element i on the segment 1, segment 2, and segment 3, respectively; Δ x l 2 , i represents the length of element and λ l 2 represents the number of lanes.
Ramp, as an important part of the freeway network, connects the freeway network with other roads. According to the variety of ramps, the BMR can be divided into three groups, including on-ramp, off-ramp, and both on-ramp and off-ramp. We select a path (see Fig 3(C) ), involving one on-ramp and one off-ramp, as an example to build the calculation model of travel time for l 3 , which is presented in Eq ( 13 ).
F l 3 ( k ) = ∑ i = 1 N l 3 , 1 ρ l 3 , 1 , i ( k ) λ l 3 Δ x l 3 , 1 , i q l 3 , 1 , i ( k ) + ∑ i = 1 N l 3 , 2 ρ l 3 , 2 , i ( k ) λ l 3 Δ x l 3 , 2 , i q l 3 , 2 , i ( k ) + ∑ i = 1 N l 3 , 3 ρ l 3 , 3 , i ( k ) λ l 3 Δ x l 3 , 3 , i q l 3 , 3 , i ( k ) ,
(13)
where F l 3 ( k ) represents the total travel time of vehicles on l 3 ; N l 3 , 1 , N l 3 , 2 , and N l 3 , 3 represent the element number of segment 1, segment 2, and segment 3, respectively; ρ l 3 , 1 , i ( k ) , ρ l 3 , 2 , i ( k ) , and ρ l 3 , 3 , i ( k ) represent traffic density and traffic flow of the basic element i on the segment 1, segment 2, and segment 3, respectively.
The last case is the BMRB including ramp and bottleneck, and there are some complicated conditions in this region. One case of BMRB, including one on-ramp, one off-ramp, and one bottleneck, is considered in this paper, which can be seen in Fig 3(D) . Based on the characteristics of traffic composition and operation, this case can be divided into five segments, involving BM, BMB, and BMR. Therefore, the travel time of l 4 can be calculated with Eq ( 14 ).
F l 4 ( k ) = ∑ i = 1 N l 4 , 1 ρ l 4 , 1 , i ( k ) λ l 4 Δ x l 4 , 1 , i q l 4 , 1 , i ( k ) + ∑ i = 1 N l 4 , 4 ρ l 4 , 4 , j ( k ) λ l 4 Δ x l 4 , 4 , j q l 4 , 4 , j ( k ) + ∑ i = 1 N l 4 , 2 ρ l 4 , 2 , i ( k ) λ l 4 Δ x l 4 , 2 , i q l 4 , 2 , i ( k ) + ∑ i = 1 N l 4 , 3 ρ l 4 , 3 , i ( k ) λ l 4 , 3 Δ x l 4 , 3 , i q l 4 , 3 , i ( k ) + ∑ i = 1 N l 4 , 5 ρ l 4 , 5 , i ( k ) λ l 4 Δ x l 4 , 5 , i q l 4 , 5 , i ( k ) ,
(14)
where F l 4 ( k ) represents the travel time of l 4 ; N l 4 , 1 , N l 4 , 2 , N l 4 , 3 , N l 4 , 4 , and N l 4 , 5 represent the number of element from segment 1 to segment 5, respectively; ρ l 4 , i ( k ) and q l 4 , i ( k ) represent traffic density and traffic flow of element i , respectively; λ l 4 and λ l 4 , 3 represent the number of lanes of the basic segment and bottleneck segment, respectively.
From the perspective of decision-makers, the first objective of the upper-level programming model represents the travel time of freeway network, and the second objective represents the traffic loading distribution of the network. The managers balance traffic flow in the freeway network based on the traffic operation and demands. Since some special conditions considered in this paper, such as bottleneck region and merging region, it is difficult to use the traditional definition of the traffic load (TLD) to show the actual traffic load. In this paper, the TLD is defined as the ratio between the maximum traffic flow among each section and the minimum capacity of each section in path l i (see Eq ( 15 )). TLD is a dynamic variable with the traffic conditions of each path changes. The decision about the TLD is complicated and depends on both the traffic conditions and the road type.
T L D l i ( k ) = q l i , max ( k ) C l i , min ,
(15)
where q l i , max ( k ) represents the maximum traffic flow among sections; C l i , min represents the minimum capacity in path l i .
According to Eq ( 15 ), we can achieve the traffic load in path l i . The average value of TLD can be shown in Eq ( 16 ).
T L ¯ D ( k ) = 1 p ′ ∑ l i = 1 l p ′ T L D l i ( k ) ,
(16)
where T L ¯ D ( k ) represents the average of these paths l 1 , l 2 ,…, l p ′ .
The second objective function of the upper-level programming model S TTD is presented in Eq ( 17 ) based on Eq ( 15 ) and Eq ( 16 ).
S T T D = ∑ k ∈ K ( ∑ i = 1 p ′ ( T L D l i ( k ) − T L ¯ D ( k ) ) 2 ) 1 2 .
(17)
To equilibrium dimensions of the objective functions, weighting factors are introduced. Therefore, the upper-level optimal control objective function is shown in Eq ( 18 ).
min S O U = α T T T ∑ k ∈ K ∑ i = 1 p ′ F l i ( x l i ( k ) , u l i ( k ) , v l i ( k ) , d l i ( k ) , k ) + α T T D ∑ k ∈ K ( ∑ i = 1 p ′ ( T L D l i ( k ) − T L ¯ D ( k ) ) 2 ) 1 2 ,
(18)
where S OU represents the objective function of the upper-level programming model; α TTT and α TTD are the weighting factors of S TTT and S TTD , respectively.
The upper-level programming model should be restricted by the traffic system constraints to meet the demands of the optimal control. The basic constraint conditions can be shown as follows.
s . t . q l i , 1 ( k + 1 ) = ( ρ l i , 1 ( k ) + T Δ l l i , 1 λ l i , 1 [ q ( k ) ϖ l i ( k ) − q l i , 1 ( k ) ] ) v l i , 1 ( k + 1 ) λ l i , 1 q l i ( k ) = ∑ μ ∈ I n q i n , μ ( k ) ϖ l i ( k ) x l i ( k + 1 ) = φ l i [ x l i ( k ) , u l i ( k ) , v l i ( k ) , d l i ( k ) , k ] ∑ i = 1 p ϖ l i ( k ) = 1 ϖ l i , min ≤ ϖ l i ( k ) ≤ ϖ l i , max 0 ≤ ϖ l i , min ≤ ϖ l i , max ≤ 1 0 ≤ ∑ i = 1 p ′ q l i , N l i ( k ) ≤ Q o u t , n e t , c a p 0 ≤ q l i ( k ) ≤ Q l i − Δ Q l i ,
where ξ represents the compliance rate, 0 ≤ ξ ≤ 1; ϖ n o r , l i ( k ) represents nominal splitting of l i without guidance information; ϖ V M S , l i ( k ) represents the splitting rate choosing path l i under the guidance information; ∑ l i = l 1 l p ′ ϖ l i ( k ) = 1 .
Considering the driver compliance rates, the splitting rate ϖ l i ( k ) can be presented by Eq ( 19 ).
ϖ l i ( k ) = ( 1 − ξ ) ϖ n o r , l i ( k ) + ξ ϖ V M S , l i ( k ) .
(19)
In addition, the splitting rate ϖ l i ( k ) also can be shown as Eq ( 20 ).
ϖ l i ( k ) = q l i ( k ) / ∑ μ ∈ I n q i n , μ ( k ) ,
(20)
where I n is the connection segment set on the original note upstream; q l i ( k ) is traffic flow of l i .
According to the Eq ( 19 ) and Eq ( 20 ), we can get the traffic volume of l i that can be presented in Eq ( 21 ).
q l i ( k ) = ∑ μ ∈ I n q i n , μ ( k ) ( ( 1 − ξ ) ϖ n o r , l i ( k ) + ξ ϖ V M S , l i ( k ) ) .
(21)
Formulation of the lower-level optimal model
The bi-level programming optimal model focuses on the stability and optimization of each path in the test network. Building the upper-level programming model, we consider the problem of the network traffic balance and the traffic flow distribution of the freeway network. In the lower-level programming model, we adopt the VSL control method proposed by Ma et al. [ 11 , 24 ] to solve the traffic congestion in path l i . Therefore, the primary goal of the lower-level programming model is to ensure the optimization of traffic volume and the travel time in path.
We define the traffic condition of no-control as a control condition. Hence, the control condition parameter S W l i is given as
S W l i = { 0 n o n − c o n t r o l , 1 o t h e r w i s e ; .
(22)
Thus, the objective function of each path in the regional network is showed in this paper as Eq ( 23 ).
S l i = S W l i ∑ k ∈ K R l i ( x l i ( k ) , u l i ( k ) , v l i ( k ) , d l i ( k ) , k ) .
(23)
The basic constraint conditions of the lower-level programming model can be shown as follows.
s . t . q l i ( k ) = ∑ μ ∈ I n q i n , μ ( k ) ( ( 1 − ξ ) ϖ n o r , l i ( k ) + ξ ϖ V M S , l i ( k ) ) x l i ( k + 1 ) = f l i ( x l i ( k ) , u l i ( k ) , v l i ( k ) , d l i ( k ) , k ) g ( x l i ( k ) , u l i ( k ) , v l i ( k ) , d l i ( k ) , k ) ≤ 0 k = 1 , 2 , … , k p .
Application results
In this section, we illustrate the efficiency and robustness of the optimization approach proposed in this paper through two ways. First, we adopt the numerical analysis method to test the sensitivity of the optimization approach based on two different types of networks. In addition, a common and typical network is used as the test network for case study.
Sensitivity analysis
We design a series of numerical analysis to study the changes of total travel time obtained from the coordinated control (CC) proposed in this paper and independent control (IC) in previous researches, see [ 11 , 24 ], in two cases: with bottlenecks or merging regions. It is assumed that there are two paths in the freeway network with the same length 10 km. The path 1 has three lanes while the path 2 has two lanes. The bottlenecks locate at the last 1 km in the two paths respectively. The saturated traffic flow in one lane is 1800 veh/h/lane, and the speed of free traffic flow is 100 km/h. In the numerical analysis, the variable is the input of traffic flow, which is assumed as constant in one experiment. In the independent control, the splitting rates in the initial node are fixed with four settings (input traffic flow to path 1: input traffic flow to path 2) 3:1, 3:2, 2:1 and 1:1. In the proposed coordinated control, the splitting rates in the initial node are obtained by solving the objective function Eq ( 18 ). The numerical analysis results with two bottlenecks are shown in Fig 4 and the numerical analysis results with two merging regions are shown in Fig 5 .
10.1371/journal.pone.0204255.g004
Fig 4
Improvement in terms of saving total travel time with bottlenecks.
(a) Splitting rates in the initial node with 3:1. (b) Splitting rates in the initial node with 3:2. (c) Splitting rates in the initial node with 2:1. (d) Splitting rates in the initial node with 1:1.
10.1371/journal.pone.0204255.g005
Fig 5
Improvement in terms of saving total travel time with merging regions.
(a) Splitting rates in the initial node with 3:1. (b) Splitting rates in the initial node with 3:2. (c) Splitting rates in the initial node with 2:1. (d) Splitting rates in the initial node with 1:1.
As shown in the Fig 4 , the saving total travel time by coordinated control and independent control compared with no-control case are denoted with red and blue lines respectively. The total input traffic flows of mainline at the initial node are changing from 5400 veh/h to 8400 veh/h. Fig 4(A), 4(B), 4(C) , and 4(D) show the saving total travel time under different splitting rates in the initial node. According to the Fig 4 , it can be seen that, using the coordinated control, the total travel time can be saved greatly, and along with increase of total input traffic flow, the saving total travel time decreased gradually. Although, using the independent control method can improve the travel speed compared with no-control case, the improvement is relatively small, less than 10%. As shown in the Fig 4(A), 4(B) and 4(D) , the lines are not monotone decreasing. When the input traffic flow is relatively large, the two bottlenecks in both paths are activated. The saving total travel time is improved because of the developed the traffic efficiency. When the input traffic flow is relatively small, the improvement of saving total travel time is mainly contributed by proper dynamic splitting rates with coordinated control. Therefore, we can obtain that the coordinated control method performs better than independent control method. When the input traffic flow is small, the coordinated control method can obtain more proper splitting rates in the initial node dynamically and avoid activating the bottlenecks. In addition, when the input traffic flow is relatively large, both the bottlenecks activated, the coordinated control method can improve traffic efficiency greatly.
As shown in the Fig 5 , the x-axis refers to the total input traffic flow of mainline (the input traffic flow of on-ramp is constant with 800 veh/h and the off-ramp proportion of mainline traffic flow is 10%), while the y-axis means saving total travel time. The red line and blue line denote the saving total travel time using coordinated control and independent control compared with no-control case respectively. According to the numerical analysis results shown in the figure, the coordinated control can save more travel time than using independent control method. When the total input traffic flow is relative large, the difference of saving total travel time between the two control methods is relatively large with about 20%. With the increase of input traffic flow, the difference changes to be close and stable eventually with about 7%. Similar to the Fig 4 , the lines in the Fig 5(C) and 5(D) do not monotone decrease but increase a little in the middle. At the left part the coordinated control can save total travel time greatly because of obtaining proper splitting rates in the initial node and avoiding activating the bottleneck of merging region. When the total input traffic flow is relatively large, the saved total travel time is mainly contributed by increasing the traffic efficiency of merging region.
Experimentation design and test network
To demonstrate the effectiveness of the optimal control approach proposed in this paper, three cases involving the no-control, independent control, and coordinated control methods are applied in the experiment. In particular, the settings of the above scheme are designed as follows: the no-control case is a case without any additional traffic control; independent control case adopts the control approach from the previous researches which are [ 11 , 24 ]; the coordinated control case adopts the optimal control approach proposed in this paper. To achieve the purpose of this study, we select a simple and ubiquitous freeway regional network as test network, which is presented in Fig 6 .
10.1371/journal.pone.0204255.g006
Fig 6
Test freeway network.
There are two paths in the test freeway network: one represents the bottleneck path BMB, while the other represents the BMR, which are path 1 and path 2, respectively. During the peak period, the traffic congestion phenomenon appears in path 1 and path 2. Moreover, the congestion in path 1 is caused by the bottleneck and the congestion in path 2 mainly appears in the merging region. For ease of analysis and discussion, the upstream mainlines of the bottleneck in path 1 and merging region in path 2 are defined as mainline 1 and mainline 2, respectively.
Furthermore, we have the network parameters: the hypothetic test freeway network has a simple structure and the test time horizon of the test is 5 h. The mainline 1 has three lanes; the bottleneck region and path 2 have two lanes, respectively; ramp 1 and ramp 2 have 1 lane, respectively. The capacity of the test network is 1800 veh/h/lane. The freeway mainline that consists of the upstream mainline of point O and the downstream mainline of point D is composed of four traffic lanes with a legal speed limit of 100 km/h, and the on-ramp own per lane with a legal speed limit of 40 km/h. The model discrete time step length is T = 10s. The traffic flowing into the network is shown in Table 1 .
10.1371/journal.pone.0204255.t001
Table 1 Traffic flow input in test network.
Time
1H
2H
3H
4H
5H
Total flow (veh/h)
5000
7000
6500
7000
5000
Ramp flow (veh/h)
800
800
900
800
800
Results analysis
The studied regional network is formulated on a simulation platform. In this section, we show the simulation results under three different conditions: the no-control case, independent control case, and the coordinated control case. The evaluation results are presented in the following. Freeway network outflow is a major performance index for evaluation of the proposed control approach. Traffic outflows of the test network under the no-control case, independent control case, and the coordinated control case are shown in Fig 7 .
10.1371/journal.pone.0204255.g007
Fig 7
Outflows of the test network: (a) shows the traffic flow during the entire simulation time; (b) represents the detailed information of traffic flow during peak hours.
As can be seen in Fig 7 , the traffic outflow from the test network in the coordinated control case is higher than that under the no-control case and the independent control case. In the first hour, the bottleneck is not activated, and the vehicles can flow into the downstream directly without any delay. During peak hours, the traffic demand is larger than the capacity of freeway network, that the input traffic flow is larger than 7200 veh/h, which can be seen in Fig 7(B) . The bottlenecks are activated, where outflow is smaller than the capacity because of chaos of vehicles operation. Although the independent control is applied in the freeway networks in case 2, the outflow is still smaller than it with using coordinated control due to reasonable splitting rates given in case 3. Specifically, at around 1.4 h, a significant drop in the merging region outflow curve can be seen under the no-control case due to the traffic problem activation leading to the decline of total efficiency. In comparison, the traffic outflow from the network in the coordinated control case decreased slightly. Furthermore, compared with no-control case, the total outflows raise about 1900 vehicles in coordinated control case and the peak time reduces 0.34 h, while the independent control only reduces the peak time about 0.15 h.
Travel time is an important index to evaluate the efficiency of the test network and the test results are presented in Table 2 . The first column is average travel time of vehicles, and the second and third columns present absolute difference and relative difference of travel time compared with no-control case respectively. As shown in this table, compared with no-control case, we can see that the values of travel time have obvious improvement in independent control case and coordinated control case. In the independent control case, the saving average travel time is about 11 s, while in the coordinated control case performs much better saving average travel time about 44 s. In terms of relative difference, the coordinated control approach proposed in this paper can improve the traffic efficiency with 12.3%, while the improvement is 3.1% using independent control approach. Although both the control methods can improve the traffic efficiency, the coordinated control method can save more travel time compared with independent control method.
10.1371/journal.pone.0204255.t002
Table 2 Travel time in three test cases.
Cases
Average travel time (s)
Difference compared with no-control case (s)
Improvement compared with no-control case (%)
No-control case
357.8
-
-
Independent control case
346.6
11.2
3.1
Coordinated control case
313.9
43.9
12.3
Test results analysis of no-control
The traffic problem can be presented through the analysis of the network traffic operation under the no-control condition. Since the legal speed limit value is adopted in the no-control case, traffic in this simulated system obeys self-organization. The queues in the test network are shown in Fig 8 . Experimental results show that the queues in mainline 1 and ramp 2 appear at 1.2 h and the queue in mainline 2 appears at 1.4 h. Furthermore, the reason of the queue in mainline 1 is that the inflow of path 1 is close to the capacity of bottleneck during the peak period and the bottleneck is activated. The traffic congestions in mainline 2 and ramp 2 are caused by the traffic disturbance in the merging region during the peak period.
10.1371/journal.pone.0204255.g008
Fig 8
Queue procession in the test network.
Fig 8 shows an obvious difference of the starting time of queue in mainline 2. There is a certain queue time delay under the no-control case compared with ramp 2. The reason for this is that more rights of way are given to the mainline vehicles flowing into the merging region in the no-control case during the peak period. Fig 9 shows the density in mainline 1 and mainline 2 when the no-control approach is applied.
10.1371/journal.pone.0204255.g009
Fig 9
The traffic density of mainline 1 and mainline 2.
Figs 8 and 9 show the basic parameters extracted from the no-control case, which would provide further support for the analysis of traffic operation in the test network. At 1.7 h, the congestion in the off-ramp 1 and the existing traffic disturbance in the merging region result in a visible decreasing value of the outflow in the mainline 2.
Fig 10 shows the average travel time of path 1 and path 2 during the entire simulation time. By analyzing this information among Figs 8 , 9 and 10 , it becomes clear that during peak period, heavy traffic flows in both the merging region and the bottleneck region result in aggravated congestion on the freeway network. Consequently, during 1.4 h to 4.3 h, the average travel times that drivers spent in path 1 and path 2 are 333 s and 473 s, respectively.
10.1371/journal.pone.0204255.g010
Fig 10
Travel time.
Test results analysis of independent control
Fig 11(A) presents the VSL values of path 1 using independent control method. The x-axis refers to time and the y-axis means the limits values. At the beginning, the average traffic flow is lower than freeway networks capacity, so that the values of speed limits are relatively large without obvious influence to the mainline traffic flow. Later, along with the increase of input traffic flow, a set of low speed limits values are selected in order to maintain the efficiency of traffic flow at the bottleneck with high level.
10.1371/journal.pone.0204255.g011
Fig 11
Control parameters in case 2.
In the path 2, there is a merging region controlled by the VSL control integrated with ramp metering. Fig 11(B) and 11(C) present the values of variable speed limits and ramp metering rate respectively. From Fig 11(B) and 11(C) , we can see that at the beginning both the speed limits control and ramp metering are not triggered because of low input traffic flow. Along with the increase of input traffic flow, a set of low VSL values and ramp metering rates are selected, in order to maintain the traffic efficiency at the merging region. Especially, the ramp metering rates are quite small for providing more right of way to the vehicles on mainline, because of more vehicles on the mainline.
Using the independent control approach, the traffic flow is controlled by VSL control and ramp metering, and the control effects can be reflected by queue length, travel time, and traffic density.
First we can see the queue in the freeway network under the independent control from Fig 12 . All the queues on mainline 1, mainline 2, and on-ramp begin to increase at about 1.5 h. During peak period, the queue length on mainline 1 fluctuates around 1000 m, while the queue length on mainline 2 reaches between 1500 m to 2000 m, and the queue length at the on-ramp is a little smaller than 500 m.
10.1371/journal.pone.0204255.g012
Fig 12
Queue in the test work.
As presented in Fig 13 , the traffic flow density of path 1 in each cell is given. The x-axis refers to the number of cells, and the y-axis means the simulation time, while the z-axis denotes traffic density. It can be seen that the traffic density during off-peak hours is relatively small with about 50 veh/km/lane, while it reaches more than 200 veh/km/lane during peak period, and the density at cell 1 and cell 2 is much larger than it in backward cells. However, compared with the no-control case, the density distribution is more balanced in the cells and the maximum density is lower.
10.1371/journal.pone.0204255.g013
Fig 13
Density in each unit of mainline 1.
In Fig 14 , the traffic flow density of path 2 in each cell is given. Compared with the traffic density figure in path 1, the density values of path 2 during peak period in all cells are large, which means that the traffic demand of path 2 is higher than path 1. The traffic condition in path 2 is worse than it in path 1. These results are consistent with the conclusion of travel time comparison.
10.1371/journal.pone.0204255.g014
Fig 14
Density in each unit of mainline 2.
Fig 15 shows average travel time of vehicles on path 1, path 2, and mainline 3 (from on-ramp to the destination). Although the lengths of path 1 and path 2 are almost the same, the travel time of path 2 is larger than it of path 1. It means that the travel speed on path 1 is larger than it on path 2. The travel time of on-ramp increases greatly along with the increase of input traffic flow because the ramp metering rate is relatively low.
10.1371/journal.pone.0204255.g015
Fig 15
Travel time of path 1, path 2 and mainline 3.
Test results analysis of coordinated control
The coordinated control approach proposed handles the congestion in the freeway network. Particularly, the control strategy can be represented as the cooperation between path control and network control. The splitting rate as a significant parameter can be achieved by the upper-level optimal control model. As shown in Fig 16 , the splitting rate is fluctuating and ranges between 0.4 and 0.6.
10.1371/journal.pone.0204255.g016
Fig 16
Splitting rates in the initial node O.
In the lower-level programming model, the decision variables include the VSL values in path 1 and the VSL values and ramp metering rates in path 2, all of which are shown in Fig 17 . Furthermore, Fig 17(A) and 17(B) show the VSL values, with 10 step-lengths as scale variations. Fig 17(C) shows the RM rates in path 2. During peak period, the VSL values of path 2 are inferior to those of path 1, because ramp metering is adopted in path 2. Especially, when the traffic condition index is higher than the setting threshold, the mainline controllers may select a series of lower VSL values to strain the outflows of mainline 2.
10.1371/journal.pone.0204255.g017
Fig 17
Control parameters in case 3.
When the traffic flow in the upstream region of the bottleneck is higher than the bottleneck capacity, the bottleneck is promptly activated leading to a series of corresponding traffic problems in mainline 1. The queue and the density in mainline 1 are presented in Figs 18 and 19 , respectively.
10.1371/journal.pone.0204255.g018
Fig 18
Queue in mainline 1.
10.1371/journal.pone.0204255.g019
Fig 19
Density in each unit of mainline 1.
In Figs 18 and 19 , the queue and density of mainline 1 always maintain high values during peak period, indicating the existence of severe traffic congestion in mainline 1. The average queue and maximum queue are 639 m and 1020 m, respectively. According to Figs 8 , 12 and 18 , the queue in mainline 1 has a small extent improvement compared with the no-control case and independent control case in that a series of lower VSL values were adopted in the coordinated control case to relieve traffic congestion. Furthermore, in Figs 9 and 19 , the mainline 1 density in the coordinated control case is lower than that of the no-control case. The vehicles in the coordinated control case are in a dynamic queue with a relatively high speed, and the queue time shortens about 0.16 h compared with the no-control case. In addition, in Figs 13 and 19 , the density in mainline 1 under coordinated control case is higher than it under independent control case. The result is caused by unreasonable split rate in the independent control case. Although the density in mainline 1 under the coordinated control case is a little larger than that under independent control case, a large improvement is achieved in mainline 2.
Fig 20 shows the travel time in the coordinated control case. According to Figs 10 and 20 , the travel time in path 1 has a little improvement of 1.5% in coordinated control case compared with the no-control case. The reason is that the control goal of the coordinated control is enhancing the efficiency of the traffic network rather than improving a single path only. When traffic inflow continues to maintain a high level, the congestion caused by the traffic flow disturbance in merging region occurs in path 2, which is presented in Figs 21 and 22 .
10.1371/journal.pone.0204255.g020
Fig 20
Travel time in the coordinated control case of path 1.
10.1371/journal.pone.0204255.g021
Fig 21
Queue in mainline 2 and ramp 2.
10.1371/journal.pone.0204255.g022
Fig 22
Density in each unit of mainline 2.
According to Figs 8 , 9 , 21 and 22 , the following conclusions can be obtained in comparison with the no-control case.
The queue phenomenon in mainline 2 is alleviated and the queue time is reduced. The coordinated control case shortened the average queue length by 75% and the queue time by 0.27 h because of the improvement of the merging region traffic order.
The upstream density of mainline 2 is reduced, which means that the dynamic queue vehicles decreased in the coordinated control case. Furthermore, the range of queue in ramp 2 is 0–204 m and the average queue length is 69.5 m. A notable improvement of 86.2% of the queue in ramp 2 can be obtained, while reducing the queue time by 0.42 h.
In addition, compared with the independent control case, according to Figs 12 , 18 and 21 , there is a remarkably improvement on the queue in the test network under the coordinated control. Specifically, the extent of improvement about the queue, in mainline 1, mainline 2, and ramp 2, can be expressed by 12%, 53%, and 83% respectively. Seen the queue figures, the queue times in each road are reduced observably. As shown in Figs 10 and 23 , there is a large difference between the no-control case and the coordinated control case. Compared with the no-control case, the travel time in path 2 has a better performance than the traffic flow stability enhancement, and mainline 3 shows a strong improvement on travel time of 53%. According to Figs 15 and 23 , the travel time in path 3 under coordinated control case is improved about 44%, compared with the independent control case.
10.1371/journal.pone.0204255.g023
Fig 23
Travel time of path 2 and mainline 3.
Conclusions and future research
The majority of cities in the world still use route guidance and VSL control. To enhance the traffic efficiency and stability of the traffic network, we propose a novel coordinated control approach that adopts the bi-level optimal model combining VSL, ramp metering, and road guidance. The control approach mainly concerns the optimization and balancing between road network service level and traffic conditions of the traffic network. Specially, the bi-level optimal model proposed in this paper involves two parts: in the upper-level optimal model, we employ the route guidance as basic method concerning the travel time minimum and traffic equilibrium in the network level; and in the lower-level optimal model we pay more attention on the optimization of traffic volume and the travel time in each path of the traffic network.
Furthermore, a real and universal traffic network is selected to verify the proposed approach. The performance of the method is studied using the related traffic parameters, such as travel speed, travel time, and traffic volume. The results demonstrate that the bi-level optimal model proposed in this paper is capable to provide an efficient and flexible transportation environment. More specially, the control strategy that connects the traffic network with each path in the test network is successful in shortening the travel time and improving the traffic conditions. Specifically, in this paper, we consider three different types of traffic networks to evaluate the effect of the optimization control approach proposed in this paper. In detail, two different types of networks, including bottleneck region and merging region separately, are selected to test the sensitivity of the optimization approach. We get some good results using numerical analysis to compare the saving total travel time. Another type of traffic network includes both the bottleneck region and merging region which are set on different paths. Additionally, three cases involving the no-control, independent control, and coordinated control methods are applied in this integrated network, and the effectiveness of the control method is further proved through the simulation. Note that the application type of networks of the optimization control approach including the former test networks, but is not limited to these three types. The scope of application of the optimization method includes different combinations of path types involving common mainline path and the tested paths in this paper. However, some special situations may not been taken into account in this paper, such as the bad weather and the special road alignment.
Although the efficiency and the flexibility of the coordinated control approach are proved in this paper, more complex road networks have not been taken into consideration. In future work, we will apply this control approach proposed in this paper to the variety situation to illustrate the robustness. We also will consider other methods integrated in the control framework to enhance the efficiency.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Myelin facilitates rapid and energy efficient action potential conduction and when disrupted, as in demyelinating diseases such as Multiple Sclerosis, nervous system function can be severely compromised. In the mouse, myelination initiates between birth and weaning in both the CNS and PNS with peak accumulation of the mRNAs encoding myelin-specific proteins reached during the third postnatal week [ 1 , 2 ]. Adaptive changes in myelin volume also occur in the mature nervous system where they are thought to influence circuit properties [ 3 , 4 ]. Consequently, the mechanisms controlling myelin synthesis, including regulation of the genes encoding myelin related proteins, are the focus of intense investigation [ 5 ].
Myelin basic protein (MBP), a major component of myelin, is an intrinsically disordered protein susceptible to multiple post-translational modifications. While MBP is largely concentrated in the myelin sheath, it is implicated in a wide array of cellular functions [ 6 ]. Shiverer mice have a deletion in 3’ domain of the Mbp locus, lack MBP and are incapable of elaborating compact CNS myelin [ 7 – 9 ]. In contrast, only limited ultrastructural anomalies and a modest decrease in myelin sheath thickness are observed in their PNS [ 10 – 13 ]. Notably, a previously reported positive correlation between the accumulated level of Mbp mRNA and CNS myelin sheath thickness demonstrated that MBP is not only essential but is a limiting factor in CNS myelin production [ 14 , 15 ]. The Golli/Mbp locus also expresses Golli transcripts that initiate far upstream and incorporate various Mbp exons. Golli accumulates in diverse lineages both within and beyond the nervous system on cell-type specific developmental programs and has been shown to modulate Ca 2+ transients [ 16 – 22 ].
Many of the transcription factors (TFs) specifying myelinating cell lineages and/or controlling lineage maturation have been identified [ 5 , 23 – 38 ]. Further, chromatin remodeling and epigenetic changes associated with myelin gene expression have revealed numerous features of the landscape in which such regulatory components operate [ 33 , 34 , 39 – 44 ]. Most notably, the Mbp enhancers investigated here have been shown to reside within a super-enhancer domain providing an environment thought to support exceptionally high concentrations of transcription factors and the intimate association of regulatory components in distinct condensates [ 45 – 47 ].
Motivated by the critical role myelin plays in nervous system function and the essential and rate-limiting role of MBP in CNS myelin formation, we, and others, have characterized features of the mechanism controlling transcription of the Golli/Mbp locus [ 1 , 2 , 26 , 33 , 34 , 48 – 53 ]. Because Mbp is expressed by both myelinating cell types, accumulates in a well characterized post-natal developmental program and has an unusual association with the widely expressed overlapping Golli transcriptional unit, the Golli/Mbp locus represents a particularly rich target within which any higher order organization of transcriptional regulators might be revealed.
Previously, the lineage specificities and developmental programs conferred by Golli/Mbp -associated enhancers were assigned using reporter constructs [ 2 , 48 , 50 , 51 , 53 – 55 ]. However, in such preparations, enhancers are isolated from their normal chromatin environment and often ligated adjacent to each other or directly to a promoter creating novel spatial relationships that may impose, or diminish, higher-order interactions [ 56 ]. Therefore, we sought to characterize enhancer contributions in the fully integrated context of the endogenous Golli/Mbp locus. Using CRISPR-based gene editing we derived lines of mice bearing alleles deleted of one or more enhancers and these were assessed for Mbp and Golli mRNA accumulation (relative to Gapdh mRNA) at key stages of post-natal development. Each enhancer KO allele caused the greatest mRNA reduction in the lineage where the respective enhancer conferred autonomous targeting. Unexpectedly, some enhancer deletions also led to reduced expression in the lineage where they demonstrated no autonomous targeting capacity and we refer to such additional cryptic function as “stealth” activity. The extent to which the super-enhancer contributes to such activity remains to be determined. Finally, our observations suggest a model in which transcription factor mediated interactions give rise to chromatin looping that brings the Golli and Mbp promoters and their relevant enhancers into close proximity.
Results
Enhancer knockout (KO) lines
Five domains of high interspecies conservation (M1-M5) are located 5’ of the Mbp start site within a putative super-enhancer domain ( Fig 1A ). In previous studies, M1 (which encompasses the M1E enhancer and the contiguous Mbp proximal promoter), M3 and M5 were each shown to drive expression in oligodendrocytes while M4 drove expression in Schwann cells. M2 demonstrated no autonomous activity in either lineage [ 2 , 48 – 50 , 53 ]. In the present investigation we derived mouse lines homozygous for alleles deleted individually of M3, M4 or M5 or deleted of M3/M5 or M1E/M3/M5 combined. Additionally, to explore the potential function of an enhancer subdomain, the M3(225)KO allele bearing a partial deletion of M3 shown previously to upregulate reporter gene activity in both oligodendrocytes and Schwann cells was derived [ 49 , 50 ] ( Fig 1B ). At key stages of myelination, accumulation of Mbp mRNA in spinal cord and sciatic nerve was analyzed while accumulation of Golli mRNA was assessed in spinal cord, thymus and retina at multiple ages.
10.1371/journal.pgen.1008752.g001
Fig 1
Organization of the Golli/Mbp locus.
(A) Golli/Mbp locus with identified transcripts, CTCF binding signals and super-enhancer domain. (B) Mbp 5’ flanking sequence indicating five modules of high interspecies sequence conservation in selected vertebrates with demonstrated autonomous enhancer activity (adapted from UCSC browser; see full species comparisons in mouse genome assembly NCBI37/mm9). Only representative Golli/Mbp transcripts are shown. (C) The position and length of individual and combined enhancer deletions derived into individual mouse lines.
Mbp mRNA accumulation in spinal cord oligodendrocytes
Relative accumulation of Mbp mRNA was measured by analyzing whole tissue homogenates of spinal cord (CNS) and sciatic nerve (PNS) for oligodendrocytes and Schwann cells respectively. Oligodendrocytes initiate expression of Mbp as a terminal maturation event coincident with myelin sheath elaboration that initiates on widely different schedules in different CNS domains [ 1 ]. Consequently, we restricted our analysis to the cervical spinal cord where myelination initiates perinatally and oligodendrocyte numbers remain constant from at least P10 through P30 [ 57 ]. Thus, a close relationship between the mRNA levels observed and that realized within individual oligodendrocytes was expected. Samples were obtained at P7, a stage of maturation when significant myelin elaboration in cervical spinal cord has occurred [ 1 ]; at P14, when myelin acquisition nears peak levels; at P21, when the levels of myelin protein mRNAs are declining from peak levels; at P30, when mature myelin maintaining cells predominate and at P90 when mice are fully mature.
Wild type (WT)
In cervical spinal cord samples, Mbp mRNA was readily detectable at P7 and rapidly increased over the next week. Relative to P14, levels decreased to 85% by P21, 74% by P30 and 33% by P90; a developmental program consistent with prior investigations of multiple myelin gene expression programs and Mbp regulated reporters [ 1 , 2 , 48 – 50 , 53 , 58 ] ( Fig 2 , S1 Table ).
10.1371/journal.pgen.1008752.g002
Fig 2
Relative Mbp mRNA accumulation in cervical spinal cord.
M1E, M3 and M5 are major enhancers of Mbp in oligodendrocytes. X-axis = post-natal age in days and Y-axis = Mbp/Gapdh mRNA ratio. Dots represent values at P7, 14, 21, 30 and 90 and connecting lines are the predicted developmental programs calculated as trendlines (Excel). Error bars = SEM.
M5 KO lines
M5 function has not been investigated in transgenic preparations but its capacity to drive transcription in the oligodendrocyte CG-4 cell line was detected using transfected reporter constructs [ 51 ]. Although the M5 enhancer displays only limited interspecies conservation, it is associated with a ChIP-Seq binding profile for myelin regulatory factor (MYRF) that extends for 917 bp encompassing a repeat domain. Three M5 deletions were generated; the 829 bp deletion (chr18: 82707436–82708264 NCBI/mm9) in the M3M5KO and M1EM3M5KO alleles removed only the conserved non-repeat sequence; the 1014 bp deletion (M5KOΔ1kb) (chr18: 82707433–82708446 NCBI/mm9) extends 185 bp further 3’ to include a short repeat region, while the 3647 bp deletion (M5KOΔ3.6kb) (chr18: 82706381–82710027 NCBI/mm9) encompasses the conserved 829 bp as well as multiple flanking repeat domains. A similar reduction in Mbp mRNA accumulation was observed with the M5KOΔ1kb and M5KOΔ3.6kb alleles at P14 demonstrating that the deleted repeat sequences do not function, at least at this age ( S1 Table ).
M3KO, M5KOΔ3.6kb, M3M5 double KO and M3(225)KO
Oligodendrocytes in mice bearing either M3KO (chr18: 82718631–82719324 NCBI/mm9) or M5KOΔ3.6kb alleles accumulated Mbp mRNA in similar developmental programs that differed only modestly at P7 and P90 when the M5KOΔ3.6kb allele supported relatively higher accumulation ( Fig 2 , S1 Table ). Notably, while M3 and M5 share multiple TF binding peaks including those for Sox10 and OLIG2 [ 52 , 59 ], they differ for others including MYRF, present only in M5 [ 51 ], and ZFP24, present only in M3 [ 36 ]. The combined M3M5KO allele resulted in an additive reduction in Mbp mRNA accumulation with only a modest difference observed at P7. Typical of WT, M3KO and M5KOΔ3.6kb mice, those bearing the M3M5KO or M1EM3M5KO alleles demonstrated an approximately 2-fold upregulation from P7 to P14. In contrast, the rapid downregulation realized by P21 did not occur in the M3M5KO or M1EM3M5KO lines in which only a limited drop off was observed at P90. Although, the partially truncated M3(225) allele supported massive and continuous upregulation of reporter gene expression [ 50 ], the identical in situ truncation in M3(225)KO (chr18: 82719120–82719395 NCBI/mm9) led to equivalent or even modestly reduced Mbp mRNA accumulation relative to M3KO.
M1EM3M5 triple KO
The Mbp proximal promoter sequence extending to -300 bp 5’ of the transcription start site failed to drive reporter expression in oligodendrocytes [ 53 ]. However, a sequence extending further 5’ to -377 bp (M1) supported transient expression during primary myelination thus demonstrating that the 5’ 77 bp (referred to as the M1 enhancer, M1E) is required for oligodendrocyte-specific activity. The M1EM3M5 allele investigated here is deleted of M3 and M5 in addition to M1E (chr18: 82723587–82723666 NCBI/mm9) and the functional significance of M1E was inferred by comparison of the M1EM3M5KO and M3M5KO alleles. Notably, M1EM3M5KO mice demonstrated a developmental expression program that paralleled that of M3M5KO mice but at approximately half absolute levels ( Fig 2 , S1 Table ). Consistent with previous models in which reduced Mbp levels were shown to limit CNS myelin production [ 14 , 15 , 60 ], the M3M5KO mice exhibited pronounced CNS hypomyelination ( S1 Fig ). Further, M1EM3M5KO mice that accumulated Mbp mRNA to approximately 14% of WT at P21 demonstrated a shivering phenotype similar to that reported previously for transgenic mice that accumulated Mbp mRNA to 13.5% of WT at P18 [ 14 ].
M4KO
M4 drives reporter expression only in Schwann cells and, consistent with that autonomous activity, the M4KO allele (chr18: 82714831–82715271 NCBI/mm9) had no effect on Mbp mRNA accumulation in oligodendrocytes at P14 or P30. However, significantly reduced accumulation, albeit modest, was observed in oligodendrocytes at P90 ( Fig 2 , S1 Table ).
Mbp mRNA accumulation in sciatic nerve Schwann cells
WT and M4KO
In WT sciatic nerve Mbp mRNA was readily detectable at P4, reached a peak level at P14 and declined to 42% of the P14 value by P90. In previous reporter-based investigations, Mbp 5’ sequence extending to -8.9 kb (thus encompassing M1-M3) drove expression in oligodendrocytes but remained silent in Schwann cells. In contrast, 5’ sequence extending to -9.4 kb, and thus incorporating the full M4 conserved sequence, drove robust expression in both oligodendrocytes and Schwann cells [ 49 , 53 ]. Consistent with the strong Schwann cell specific programming observed with all M4 bearing reporter constructs, M4KO mice accumulated Mbp mRNA at much reduced levels, approximating 20% of WT at all ages ( Fig 3 , S2 Table ).
10.1371/journal.pgen.1008752.g003
Fig 3
Relative Mbp mRNA accumulation in sciatic nerves.
M4 is the major Mbp enhancer in Schwann cells. Loss of M3 and/or M5 alone resulted in lower accumulation at several ages while loss of M1E has no additional affect. X-axis = age in days and Y-axis = Mbp/Gapdh mRNA ratio. Dots represent post-natal days P4, 7, 14, 21, 30 and 90 and connecting lines are the predicted developmental programs calculated as trendlines (Excel). Error bars = SEM.
M3KO, M5KOΔ3.6kb, M3M5KO, M1EM3M5KO and M3(225)KO
In contrast to all M4 bearing reporter constructs, no M3 constructs driven through the Mbp proximal promoter (M1) expressed in Schwann cells. However, when M3 was ligated to a heterologous and minimal hsp promoter, transient Schwann cell expression was observed during a restricted period of preweaning myelin elaboration [ 53 ]. Thus, while M3 is not capable of productively engaging the Mbp proximal promoter in Schwann cells, it nonetheless must bind relevant TFs. Consistent with that interpretation, sciatic nerve demonstrates a binding peak for Sox10 and enrichment for H3K27ac over M3 [ 52 ]. Unexpectedly, all lines bearing an allele deleted of M3 demonstrated reduced Mbp mRNA accumulation in Schwann cells throughout development as did those bearing the M5KOΔ3.6kb allele, although in a more modest fashion. Further, the reduction seen in mice bearing the combined M3M5KO allele trended lower than either the M3KO or M5KOΔ3.6kb alleles at most ages and was strictly additive at P14. The accumulation program in M1EM3M5KO mice was comparable to that observed in M3M5KO mice while the partially deleted M3(225)KO allele had no effect ( Fig 3 , S2 Table ).
Golli mRNA accumulation
In the mouse, Golli transcription initiates approximately 80 kb upstream of Mbp and Golli isoforms variously incorporate specific Mbp exons as well as 264 bp of the Mbp proximal promoter [ 16 , 61 ]. The M1-M5 Mbp regulatory modules are located in the 20 kb 5’ of the Mbp start site and therefore within a Golli intron. In the oligodendrocyte lineage, Golli accumulates predominantly in progenitors and is observed only rarely in post-mitotic myelinating oligodendrocytes [ 62 ]. However, Golli is expressed by numerous other cell types, both within and beyond the nervous system, including neurons and T-cells where distinct lineage specific developmental programs are observed [ 16 , 62 – 66 ]. Therefore, the Golli mRNA values obtained from spinal cord likely arise from a combination of cell types that potentially changes with development. To determine levels of expression in nervous tissue devoid of oligodendrocytes we examined retina and for T-cells we examined thymus. Expression in Schwann cells was not determined as Golli accumulation was at the limit of detectability in sciatic nerve samples. Consistent with previous findings for optic nerve [ 49 ], Golli mRNA accumulation was reduced to approximately 10% of WT in all tissues examined from the lines in which M3 was deleted. The absence of the M1E sequence in the M1EM3M5KO allele had no further effect. In contrast, M5KOΔ3.6kb and M4KO lines demonstrated normal Golli mRNA accumulation in all tissues examined. Notably, in samples from M3(225)KO mice, Golli accumulation was normal at P14, indicating that the specific M3 subsequence deleted in this allele plays no role in Golli regulation at this age. However, at P30, Golli accumulation increased to 141% and 130% of WT in spinal cord and thymus respectively, indicating an age-specific putative repressor role for the deleted sequence (Figs 4 and 5 , S3 Table ). However, that relationship was reversed at P90 where accumulation reached only 65% of WT.
10.1371/journal.pgen.1008752.g004
Fig 4
M3 is a major Golli enhancer in spinal cord.
All alleles deleted of M3 demonstrate an approximately 90% reduction in Golli mRNA accumulation in spinal cord. In contrast, the partially deleted M3(225)KO allele supported accumulation ranging from a mild decrease to a modest increase at different ages. X-axis = age in days and Y-axis = Golli/Gapdh mRNA ratio. Dots represent post-natal days P7, 14, 21, 30 and 90 and connecting lines (both solid and dashed) are the predicted developmental programs calculated as trendlines (Excel). Error bars = SEM.
10.1371/journal.pgen.1008752.g005
Fig 5
M3 is also the major Golli enhancer in thymus and retina.
All alleles deleted of M3, but not the partially deleted M3(225)KO allele, reduced Golli accumulation to a similar extent. X-axis = age in days and Y-axis = % Golli/Gapdh mRNA. Error bars = SEM.
Discussion
In this investigation we sought to define the developmental programming conferred by different Mbp enhancers in the context of the endogenous Golli/Mbp locus. We expected such insight to lay the functional foundation upon which investigations capable of revealing mechanistic insights into myelin gene programming can be realized. Further, by contrasting such integrated activity with that realized through enhancers isolated in reporter constructs, we expected any higher order levels of functional interaction to be exposed. As Mbp expression follows a tightly scripted in vivo developmental program, we evaluated mRNA accumulation in both CNS and PNS at key stages of primary myelin elaboration through to the mature myelin maintenance phase.
Extensive characterization of the autonomous activity conferred by M1, M3 and M4 was achieved previously using randomly inserted reporter constructs [ 1 , 2 ]. In all cases, multiple expressing transgenic lines bearing different copy number inserts yielded similar lineage specificities. Subsequently, those initial investigations were complemented with controlled transgenic preparations in which single copy constructs were inserted in a predetermined orientation at a common and permissive chromatin site [ 48 – 50 , 53 ]. The latter approach also supported direct comparisons between the quantitative outputs conferred by different enhancer sub-sequences. Reporters driven through the Mbp promoter bearing all enhancer combinations investigated revealed identical targeting capacities in both types of transgenic preparations. These observations provided the required basis from which the autonomous regulatory capacity of enhancers could be compared with high confidence to their contributions in the context of the endogenous Golli/Mbp locus.
Here, mRNA analysis was performed on whole tissue samples thus making the assay susceptible to multiple potential variables including changes in the density and/or differentiation state of relevant cell populations, compensatory feedback mechanisms and secondary changes beyond the transcriptional level. However, these potential caveats are largely dispelled by previous observations. Firstly, shiverer mice, despite their Mbp null status and fully amyelinated CNS, have a normal density of mature oligodendrocytes from P10 through P30, although a notable increase is observed at P60 [ 57 ]. Consequently, in enhancer KO lines that demonstrate only reduced Mbp mRNA accumulation, rather than a total absence, oligodendrocyte density is expected to be normal at least through P30; the period in which the majority of the present observations were made. However, whether an increase in oligodendrocyte density occurs by P90 remains to be determined. In a similar manner, the accumulation of other myelin gene products in the amyelinating oligodendrocytes of Mbp null shiverer mice, including the prominent Plp transcript and protein, show little change from WT [ 67 , 68 ]. Further, the rate of initiation of their truncated Mbp transcript also tracks closely that of WT [ 69 ]. Thus, changes in cell density or major disruptions in myelin gene programming were not anticipated responses to the reduced Mbp mRNA accumulations imposed by the enhancer KOs. A similar relationship was expected to exist in the PNS; e.g., by P5 in WT mice, spinal root axons are heavily myelinated and Schwann cell proliferation has ceased [ 10 ]. Further, shiverer mice demonstrate only minor ultrastructural anomalies and mild thinning of PNS myelin sheaths [ 10 , 13 ]. Finally, the endogenous Mbp accumulation program observed in enhancer KO mice is reflected in the reporter accumulation programs observed in many transgenic lines that bear an intact Golli/Mbp allele [ 1 , 2 , 48 – 50 ]. Therefore, we interpret observations made on the different KO lines to reflect a valid estimate of the average Mbp mRNA accumulation realized by generally equivalent populations of oligodendrocytes and Schwann cells.
Expression of reporter genes vs. enhancer KOs
Extensive prior investigations exploiting reporter constructs revealed multiple features of Mbp enhancer function extending to their autonomous lineage specificities and functional capacities of their sub-sequences. However, reporter constructs are unlikely to reveal the full extent of the integrated regulatory activity conferred by individual enhancers as insertion site, copy number and multiple levels of chromatin organization, beyond that associated with the endogenous Golli/Mbp locus, could influence transcriptional efficiency. A notable level of such higher-order organization is the super-enhancer, a chromatin domain encompassing one or more transcriptionally active enhancers frequently associated with key lineage-specific or lineage specifying genes [ 54 , 70 – 73 ]. Super-enhancers are enriched in active chromatin modifications, bind master TFs, mediator and co-activators and their boundaries are often demarcated by duplicated CTCF sites. Association between CTCF binding factor and cohesion supports the formation of chromatin loops creating insulated enhancer bearing neighborhoods [ 70 ]. Further, recent evidence shows that super-enhancer components, brought into close proximity, can form liquid-like condensates in which transcriptional components are highly enriched [ 45 – 47 ].
Multiple data sets from human brain revealed a super-enhancer domain extending through and 35 kb upstream of the Mbp gene [ 52 , 71 ]. Mouse cortical samples revealed a similar domain of 24 kb enriched in active transcription marks such as H3K27ac and H3K4me1 and demarcated by paired CTCF binding sites within which the M1-M5 enhancers are located [ 55 , 74 , 75 ]. In the CNS, M1, M3 and M5 (but not M4) demonstrate H3K27ac enrichment and are bound by SOX10, a glial specifying transcriptional activator with a potential role in regulating the formation of super-enhancer domains [ 52 ]. Further, SOX10 has been shown to interact with mediator in both oligodendrocytes and Schwann cells [ 29 ]. Whether this super-enhancer domain forms exclusively in Mbp -expressing oligodendrocytes or also in Schwann cells remains to be determined. Nonetheless, in Schwann cells, the M1, M3, M4 and M5 enhancers are all enriched in H3K27ac and bind SOX10 [ 34 , 52 ]. Finally, multiple Golli -expressing tissues demonstrate variably sized super-enhancer domains elsewhere in Golli/Mbp locus.
The extent to which super-enhancers confer synergy on their constituent enhancers remains controversial [ 46 , 47 , 71 , 75 – 79 ]. Should inter-enhancer interactions occur within the Golli/Mbp super-enhancer, we expected them to be revealed by comparing the outputs of different enhancer KO alleles. In contrast to the synergy model, in oligodendrocytes the double M3M5KO allele gave rise to the precise reduction of Mbp mRNA predicted by combining the individual consequences of the M3 and M5 KO alleles. Such observations are consistent with an additive model in which super-enhancer activity equals the sum of its constituent enhancer activities.
In multiple reporter configurations, the M4 enhancer revealed strong and autonomous targeting activity only in Schwann cells [ 2 , 48 , 49 , 53 ]. In contrast, all contiguous 5’ sequences terminating before M4, but encompassing M3, failed to drive reporter expression in Schwann cells. As predicted by these observations, the M4KO allele resulted in a major reduction in Mbp mRNA accumulation in Schwann cells (~20% of WT). However, despite the failure of M3 to support reporter expression in Schwann cells, its absence in KO lines caused a significant reduction in Mbp mRNA accumulation in that lineage at all ages examined. We identify this additional cryptic activity arising from the otherwise autonomous oligodendrocyte specific M3 enhancer as “stealth” activity. Defined in this manner, stealth activity also was exhibited in oligodendrocytes by the otherwise autonomous lineage specific M4 Schwann cell enhancer, albeit only at P90. Although the reduction in mRNA accumulation observed in M5KOΔ3.6kb Schwan cells may reflect similar stealth activity, this cannot be confirmed as the autonomous in vivo targeting capacity of M5 is yet to be investigated.
We are not aware of previous experiments designed to reveal stealth enhancer activity such as that assigned here to M3 and M4. Thus, it remains to be determined if the stealth phenomenon is widespread, unique to Golli/Mbp or limited to enhancers within a common super-enhancer domain. However, that certain enhancers affect activity only through association with special “hub” enhancers has experimental support [ 76 , 80 ]. M3, and possibly M5, might participate in Schwann cells as non-hub enhancers engaging with M4, while in oligodendrocytes, M4 might act as a non-hub enhancer engaging with any of the oligodendrocyte enhancers. Notably, M3 and M5 both show H3K27ac enrichment and Sox10 binding in the PNS while M4 binds OLIG2 in immature oligodendrocytes [ 52 , 59 ]. Unique to the previously hidden non-autonomous stealth activity observed here is its origin from enhancers that drive robust fully autonomous expression only in a different lineage.
Reporter genes vs. partial enhancer module KOs
Insight into the location of TF binding and chromatin modifications associated with the sequences engaged in Mbp enhancer activity has been obtained from prior functional and ChIP-Seq analysis [ 33 , 34 , 36 , 51 , 52 , 59 ] ( S4 Table ). Complementing such investigations have been constitutive and conditional KO of the TFs and chromatin modifiers themselves [ 24 , 26 , 32 , 37 , 42 , 59 , 81 – 89 ]. However, KOs of oligodendrocyte or Schwann cell TFs often disrupt their maturation prior to expression of Mbp and myelin formation and/or their survival and cell identity thus making it difficult to assign a direct role for such TFs to Mbp transcription per se. An interesting example is MYRF, a presumptive master regulatory TF that, within the super-enhancer, only binds to M5 [ 51 ]. In Myrf KO mice, maturation of oligodendrocyte is arrested affecting the expression of multiple myelin genes and preventing Mbp expression [ 27 ]. However, the KO of M5 did not similarly prevent Mbp expression but resulted in only a ~40% reduction in its accumulation. Here we contribute to this analysis by investigating the function of alleles bearing partial deletions of M1 and M3. Absence of the 77 bp M1E sequence in the M1EM3M5KO allele reduced Mbp mRNA accumulation in oligodendrocytes beyond that imposed by the M3M5KO allele alone, demonstrating an enhancing role for the M1E domain ( Fig 2 ). In contrast, loss of M1E had no effect on Mbp mRNA accumulation in Schwann cells, demonstrating that it confers neither enhancing activity nor plays any role in the productive engagement of M4 with the remaining promoter components in that lineage ( Fig 3 ).
When M3 was ligated to M1 promoted reporters it failed to drive expression in Schwann cells. However, when ligated to a minimal hsp promoter it conferred transient pre-weaning reporter expression. In marked contrast, the partially deleted M3(225) sequence conferred constitutive expression at levels up to 50 fold higher in Schwann cells and up to 5 fold higher in oligodendrocytes [ 50 ]. In the context of the endogenous locus, the same M3 truncation (M3(225)KO) paradoxically had no effect on Mbp mRNA accumulation in Schwann cells and, relative to WT, reduced rather than enhanced accumulation in oligodendrocytes (Figs 2 and 3 ). The basis for this striking difference between reporter construct activity and that realized by the same deletion in the endogenous locus remains unknown but the experimentally imposed close proximity of M3(225) and the hsp promoter in reporter constructs may contribute to the observed upregulation.
Further demonstration that distinct enhancer sub-domains affect promoter-enhancer relationships was revealed by Golli expression. All alleles deleted of M3 reduced Golli mRNA accumulation to approximately 10% of WT in all samples examined (Figs 4 and 5 ). In marked contrast, the partially truncated M3(225)KO allele supported near normal Golli mRNA accumulation in spinal cord at P14, modest upregulation at P30 and only mild downregulation at P90. Notably, the upregulation at P30 was observed also in thymus. Thus, M3 sub-sequences differ in their capacity to engage the Mbp and Golli promoters, consistent with a model in which M3 functions through different TFs as a general “house-keeping” enhancer for Golli and a strong lineage-specific enhancer for Mb p.
A model accommodating enhancer targeting activities
The Golli/Mbp enhancer activities, and their functional interactions documented here, lead to a DNA-looping model compatible with much of the observed integrated output of the locus ( Fig 6A and 6B ). This model accommodates previously described differential TF binding by different Mbp enhancers [ 32 , 46 , 70 , 90 – 94 ]. Specifically, self-associating TF dimers, such as those formed by YY1 and SP1 are implicated in chromatin-looping and long-distance enhancer-promoter interactions [ 56 , 95 – 99 ]. This mechanism could accommodate the emergence of multiple regulatory programs such that a single enhancer can regulate two different promoters and the output of a single locus can evolve to match the unique requirements of the different cell types that myelinate the CNS and PNS.
10.1371/journal.pgen.1008752.g006
Fig 6
A tentative DNA-looping model for enhancer engagement.
(A) The predicted direct enhancer-promoter interaction for Mbp expression in oligodendrocytes (top with blue arrows) and Schwann cells (bellow with the red arrow). (B) The predicted relationship between M3 and the Mbp and Golli promoters. White boxes = enhancers and proximal promoters. Deleted sequences are highlighted in grey. Arrows = potential interactions. Red triangles = CTCF binding sites.
M1, M3, M4, M5 and the Golli proximal promoter all have predicted high affinity YY1 binding sites, while only M3, M3(225) and the Golli proximal promoter have similarly high affinity SP1 binding sites (Jaspar; relative score (rs) > 90%) [ 100 ]. However, lower affinity SP1 and YY1 motifs (80% < rs < 90%) exist in all. Accordingly, in oligodendrocytes, M3 and M5 interaction with the Mbp promoter could involve DNA-looping mediated by YY1 and/or SP1 dimerization ( Fig 6A ). While conditional KO of YY1 in oligodendrocytes leads to amyelination characteristic of MBP null shiverer mice, post-mitotic oligodendrocyte maturation unfortunately is blocked such that any direct role for YYI in Mbp transcription cannot be easily revealed in that model [ 88 , 89 ]. Notably, the M3(225)KO allele lacks the YYI motif and exhibits the same reduction seen in mice bearing the full M3KO mice at P30. However, higher levels of Mbp mRNA accumulation are observed at P14 and P90 and this partial restoration of enhancer-promoter interaction might be conferred through its retained SP1 motif ( Fig 6A ). These observations suggest that age-specific differences in capacity to promote inter-sequence interaction may exist. Indeed, the global methylation of DNA in both oligodendrocytes and Schwann cells is known to change during differentiation and myelination [ 101 ] and YY1 DNA binding is methylation sensitive [ 102 ]. In addition, during myelin synthesis in oligodendrocytes SP1 becomes phosphorylated via PKC/Erk [ 103 , 104 ]; a pathway shown to increase its DNA binding capacity in smooth muscle [ 105 ]. During this period, SP1 accumulation and binding to the Mbp promoter also increases [ 103 , 104 ].
Activity mediated in part by SP1 and/or YY1 dimerization provides a convenient model that appears to accommodate many of our observations on Mbp expression and an expansion of this looping model may also accommodate Golli programming ( Fig 6B ). While the Golli proximal promoter contains both YY1 and SP1 binding motifs, among the enhancers investigated here, only the M3 enhancer modulates Golli output and it uniquely contains a high affinity SP1 binding site [ 100 ]. Moreover, the truncated M3(225)KO allele that retains this SP1 binding site drives high Golli expression at all ages ( Fig 6B ).
Beyond insight into the functional organization of Golli/Mbp regulatory sequences, further aspects of oligodendrocyte biology are revealed by the developmental programs realized by the endogenous Mbp locus, relevant reporter constructs and the KO alleles reported here. As demonstrated by the capacity of oligodendrocytes to myelinate inanimate fibers in vitro , initial myelin elaboration can be supported entirely by oligodendrocyte intrinsic programming [ 106 ]. In contrast, it is widely recognized that myelin in the mature CNS demonstrates plastic changes potentially in response to neuronal activity [ 3 , 4 , 107 , 108 ]. Consistent with such developmental changes, numerous reporter constructs demonstrated maturation specific expression programs as observed here with the M3(225)KO and M4KO alleles [ 48 , 50 ]. These observations are consistent with a model in which the qualitative and/or quantitative features of the transcription factor repertoire evolves during maturation; a circumstance indicating the eventual necessity to evaluate mechanisms regulating myelin gene transcription in the in vivo developmental context.
Like WT mice, all enhancer KO lines showed an approximately 2-fold increase in Mbp mRNA accumulation in spinal cord between P7 and P14. Similar upregulation was observed in sciatic nerve samples. Notably, in the brains of amyelinated shiverer mice, expression of their truncated Mbp transcript increases in a similar manner during this period [ 69 ]. An increased density of oligodendrocytes or Schwann cells may contribute to that rise but if so, such contributions are expected to be small; the density of mature oligodendrocytes in spinal cord is unchanged between P10 and P30 [ 57 ] and Schwann cell proliferation terminates by P5 in spinal roots when myelination is well advanced [ 10 ]. However, regardless of what combination of enhancers are deleted, an early and seemingly uniform developmental rise is observed.
A rapid and pronounced decline in Mbp mRNA accumulation occurs in WT mice prior to weaning. However, this decline was not uniform in the enhancer KO lines. While M3KO and M5KOΔ3.6kb lines followed the WT program, the double M3M5KO and triple M1EM3M5KO lines did not. Rather, their accumulation levels at P14 remained unchanged through P30 after which they reduced only slightly. What accounts for this striking difference remains unknown but our combined evidence suggests that at least two different programs regulate normal Mbp expression; one responsible for its upregulation during the primary myelination period and another that maintains expression at maturity [ 3 , 108 ]. Indeed, this is consistent with observations from reporter studies in which regulatory sequences were shown to have different expression capacities in young vs. old mice [ 48 , 50 ].
While the present study characterizes the major regulatory sequences encompassed within the oligodendrocyte super-enhancer, other potential regulatory sequences located elsewhere in the Golli/Mbp locus have emerged from ChIP-Seq analysis and these may contribute to the residual expression observed in the enhancer KO lines ( S4 Table ).
This investigation provides insights into the complex regulatory mechanisms governing Golli/Mbp programming and lays the required functional framework from which the role of chromatin configuration and modification along with specific TF binding can be approached [ 76 ]. As the mouse models described here contain unique configurations of Golli/Mbp regulatory sequence they may themselves contribute to such investigations including the formation and function of super-enhancers and their potential role in stealth activity. This study further suggests the future requirement to assess the transcriptional mechanisms controlling myelin genes at key stages of in vivo maturation. Finally, while these enhancer-deleted mouse models are fully viable, they elaborate myelin sheathes of variably reduced thickness providing a unique opportunity to re-evaluate basic features of the axon-myelin relationship.
Materials and methods
Animals
All experiments were carried out in accordance with the guidelines of the Canadian Council on Animal Care. Protocol number 215–7668 approved by the McGill University DOW Facilities Animal Care Committee.
CRISPR design and gene editing
M1E, M4 and M5 sequences
The 422bp M4 enhancer targeted here was described previously [ 48 ]. M5 refers to the target of the MYRF ChIP carried out in rat [ 51 ]. Using the UCSC browser, this sequence was aligned with the mouse genome (chr18: 82707095–82708011 NCBI/mm9). For the purpose of generating M4 and M5 enhancer KOs, single guide RNAs (sgRNAs) were designed to target sequences flanking the conserved enhancer domain such that double strand breaks would be simultaneously introduced at both sides of the enhancer resulting in deletion of the intervening enhancer sequence. sgRNAs were designed (using the CRISPR Design http://crispr.mit.edu/ ) [ 109 ] to identify locus specific targets. To minimize the potential impact of inefficient sgRNAs, we designed 2 that bind in close proximity (for a total of 4 sgRNAs per enhancer deletion). The sgRNA target sequences used to generate the KO mice are indicated in S5 Table .
sgRNA design
The plasmid DR274 was a gift from Keith Joung (Addgene plasmid # 42250) [ 110 ]. DR274 was digested with BsaI which cuts twice between the T7 promoter and the gRNA scaffold, leaving sticky ends. For each of the targets listed above, two oligos, one for each strand, were ordered from IDT ( S6 Table ). They were annealed at 40uM each in NEB3 buffer. Each has one of the DR274 sticky ends so that they could be ligated into the plasmid using the NEB Quick Ligation Kit (M2200S). Each ligation mix was transformed into competent bacteria and kanamycin resistant clones obtained. 3 clones of each were sequenced in the relevant region and used to generate the sgRNA. To generate the sgRNA template for the M4 deletion, the PCR method using two long overlapping oligos was used ( S6 Table ) [ 111 ]. The MEGA shortscript T7 kit from Life Technologies was used to synthesize the sgRNA from the T7 promoter. The resulting sgRNA was tested for integrity on a Bioanlyzer at the McGill Genome Center.
The target-specific crRNAs ( S6 Table ) were hybridized to Alt-R® CRISPR-Cas9 tracrRNA, the Universal 67mer tracrRNA from IDT to generate the functional sgRNAs according to the manufacturer’s instructions. AltR1 and AltR2 are proprietary (IDT) modifications to increase the stability of these short RNAs.
Zygote manipulation, delivery of CRISPR components and transplantation into pseudopregnant mice
Zygotes were recovered mid-day from the oviduct of WT or M3KO C57Bl/6 mice [ 49 ] naturally mated to wmN2 transgenic mice [ 112 ]. The cumulus cells were removed by a short incubation in 1% hyaluronidase/M2 medium (Millipore) and moved into Advanced KSOM media (Millipore) at 37°C with 5% CO2. All zygote manipulation was done at room temperature and the media was kept under mineral oil. M4KO mice were generated by microinjection into zygote cytoplasm of 25ng/ul Cas9 mRNA (PNA Bio) and 12.5ng/ul of each of four sgRNAs. All M5 deletions were generated by zygote electroporation. Prior to electroporation the zygotes were moved to Opti-MEM (Life Technologies) and thinning of the zona was achieved by treating the zygotes with Acid Tyrode’s solution (Millipore) for 10 seconds and transferring them back into fresh Opti-MEM. Zygotes were electroporated according to the ZEN2 protocol [ 113 , 114 ] with a final concentration of 250ng/ul Cas9 mRNA (PNA Bio) and 300ng/ul sgRNA dissolved in TE pH7.5/Opti-MEM at a 1:1 ratio [ 113 ]. A 20ul drop of this mix containing the CRISPR reagents was prepared and the batch of 30–50 zygotes carried in less than 1ul of Opti-MEM were moved into this drop. The mix was transferred to a 1 mm electroporation cuvette purchased from BioRad and electroporation was carried out using a Bio-Rad Gene Pulser Xcell electroporator. Embryos were subjected to 1–2 pulses of 25–30 V according to the ZEN protocol [ 113 ]. After microinjection or electroporation embryos were cultured overnight in advanced KSOM media at 37°C with 5% CO2. After overnight incubation, embryos at the 2-cell stage were transplanted (bilaterally, approximately 15/mouse) into the fallopian tubes of CD1 female recipients rendered pseudopregnant by mating with B6C3F1 vasectomized males (purchased from Charles River).
Genotyping and breeding scheme
Pups were tail-biopsied at weaning for genotyping. Tail samples were digested at 55C overnight in lysis buffer (containing 100 mM Tris, pH 8.0, 5 mM EDTA, pH8.0, 200 mM NaCl, 0.2% sodium dodecyl sulfate (SDS) and 100ug/ul proteinase K) and genomic DNA was extracted. Genotyping initially was done using PCR with primers surrounding the sequence to be deleted. Upon detection of a desired, shorter-than-WT, band, the PCR product was sequenced at the McGill University and Génome Québec Innovation Centre and the existence of M4 and M5 deletions confirmed. Founder mice were mated to WT C57Bl/6 and the consequent progeny were genotyped by PCR for the deletion and LacZ (to detect the presence of a transgene at the HPRT locus, that exists within our donor colony and select against it). Mice carrying the enhancer deletion were mated to homozygosity while breeding out the transgene located on the X chromosome. In total, 2 lines of M4KO mice (identical sequencing results), 2 lines bearing single M5KOs (different deletion lengths), 1 line of M3M5 double KO and 1 line of M1EM3M5 triple KO were established.
Tissue samples
After the homozygous lines of mice were established, samples from 3–11 mice of both genders from WT and all KO lines were obtained at the ages indicated ( S1 , S2 and S3 Tables). The mice were anesthetized with a lethal dosage of Avertin and sciatic nerve, cervical spinal cord, retina and thymus samples were collected into RNAlater solution (Ambion) according to the manufacturer’s instructions and stored at -20°C.
RNA extraction and qRT-PCR
Total RNA extraction was done using Trizol (Life Technologies) and a Qiagen RNeasy MinElute Cleanup kit. RNA was eluted in nuclease free water and its concentration was measured using a spectrophotometer. The RT reaction was carried out using Superscript IV VILO Mastermix (Life Technologies) using 1ug of total RNA according to the manufacturer’s instructions and the resulting cDNA was stored at -80°C. A QuantStudio™ 7 Flex Real-Time PCR System (Life technologies) was used for qPCR in a 96-well plate. On the day of qPCR, the cDNA was diluted 20x and 40x for measuring Golli and 100x and 400x for measuring Mbp and Gapdh . Each sample was measured twice at the low dilution and once at the high dilution. To avoid inter-plate variability samples from WT and KO mice were measured together on individual 96-well plates. To measure Mbp and Gapdh in SN and cervical spinal cord, Taqman probes ( Mbp : Mm01266402_m1, Gapdh : Mm99999915_g1, Life technologies) were used. For Golli measurements in cervical spinal cord and thymus however, the SYBR green method was used (PowerUp SYBR green master mix, Life Technologies). Multiple primer sets were designed, tested and the optimal pair (2F: 5’ATTGGGTCGCCATGGGAAAC, 2B: 5’CCAGCCTCTCCTCGGTGAAT) was chosen. On each plate, 5 10-fold serial dilutions of a DNA standard were run in triplicate to generate a standard curve. Standards were prepared by amplifying a sequence larger than the measured amplicon. After standard PCR, the single band was purified from a gel with a NucleoSpin Gel and PCR cleanup kit (Macherey-Nagel) and its concentration determined. The efficiencies of reactions for both Taqman and SYBR green methods inferred from standard curves were 95–105%.
Data analysis
After qRT-PCR, sample measurements from multiple dilutions were averaged. Relative amounts of Mbp and Golli were calculated by dividing the average of each by the average of Gapdh for the same sample. The relative Mbp and Golli measurements of all samples of each mouse line were averaged and the standard error of the mean was calculated. To determine statistical significance, ANOVA with post-hoc Tukey test was performed using IBM SPSS Statistics 25 software. As analysis of sub-samples revealed no gender differences, male and female samples were combined throughout this investigation.
Light and electron microscopy
Mice lethally anesthetized with Avertin, were transcardially perfused with 2.5% glutaraldehyde + 0.5% paraformaldehyde in 0.1M sodium cacodylate buffer and cervical spinal cord and ON samples were collected. Samples were postfixed overnight at 4˚C followed by rinsing with 0.1M sodium cacodylate buffer. A second post fixation was done with 1% osmium tetroxide followed by rinsing with ddH2O. Samples were dehydrated by incubation in increasing concentrations of acetone: 30%, 50%, 70%, 80%. 90% and 3X100%. Infiltration was done with 1:1, 2:1, 3:1 (epon:acetone) followed by embedding in epon and overnight polymerization at 60˚C. 0.5 um sections were stained with Toluidine blue and cover slips were mounted with epon for imaging. Slides were imaged with 63X or 100X oil immersion objectives by light microscopy (Zeiss Axio Imager M1).
Supporting information
S1 Fig
Mice bearing the M3M5KO allele demonstrate CNS hypomyelination.
Electron micrographs were obtained from cross sections of the ventral medial cervical spinal cord from P90 WT and M3M5 mice ( A and B at 640x) and ( C and D at 3000x). The axon population in this domain ranges from small to large calibers. In the M3M5KO sample, axons of all calibers are typically ensheathed with compact myelin markedly thinner than normal although rare small calibre axons lacking compact myelin (*) were encountered.
(PDF)
S1 Table
Relative Mbp mRNA analysis in spinal cord of enhancer knock-out mice at P7, P14, P21, P30 and P90.
The values are presented as % ± standard error of the mean. “*” and “**” represent p-values ≤ 0.05 and ≤ 0.01 respectively. n(F:M) represents the n umber of F emale and M ale mice from each genotype analyzed at each age.
(PDF)
S2 Table
Relative Mbp mRNA analysis in sciatic nerve of enhancer knock-out mice at P4, P7, P14, P21, P30 and P90.
The values are presented as % ± standard error of the mean. “*” and “**” represent p-values ≤ 0.05 and ≤ 0.01 respectively. n(F:M) represents the n umber of F emale and M ale mice from each genotype analyzed at each age. “#” indicates that sciatic nerves from two mice were combined for each sample.
(PDF)
S3 Table
Relative Golli mRNA accumulation in spinal cord of enhancer knock-out mice at P7, P14, P21, P30 and P90.
The values are presented as % ± standard error of the mean. “**” represent p-values ≤ 0.01. n(F:M) represents n umber of F emale and M ale mice analyzed.
(PDF)
S4 Table
Oligodendrocyte and Schwann cell ChIP-Seq data relevant to the Golli/Mbp locus.
References are indicated by ().
(PDF)
S5 Table
sgRNA target sequences used to generate the KO mice.
(PDF)
S6 Table
Oligonucleotides used for CRISPR editing.
(PDF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The primary non-redundant function of interleukin-2 (IL-2) is as a mediator of peripheral T cell tolerance [1] . As well as the maintenance of peripheral tolerance, IL-2 has also been somewhat paradoxically described play an important role in driving the proliferative response of activated T cells and as a critical factor in the generation of an appropriate memory T cell response [2] . IL-2 exerts these pleiotropic effects through interacting with the heterotrimeric IL-2 receptor (IL-2R) complex comprised of α (CD25), β (CD122) and common γ (CD132) chains expressed on the surface of activated T cells [3] . The critical role for IL-2 in the maintenance of T cell tolerance is evident from studies on transgenic mice deficient for either the cytokine itself or its receptor [4] , [5] . These mice develop profound autoimmunity characterized by an uncontrolled expansion of auto-reactive T cells. Further analysis has revealed the mechanistic basis for these observations, and uncovered the critical role of IL-2R signalling in maintaining the competitive fitness of peripheral regulatory T cells (Tregs) [6] , and in the direct inhibition of Th17 cell responses [7] – [9] . These observations are further underscored by the strong genetic linkage between mutations at both the IL-2 and IL2RA/CD25 gene loci and several T cell mediated autoimmune diseases in humans [10] . However, precisely how such polymorphisms confer susceptibility to autoimmunity remains incompletely understood.
A number of specific CD25 alleles associated with autoimmune susceptibility occur in association with enhanced levels of the soluble form of the IL-2R alpha chain (sCD25) in the serum of patients [10] , [11] . However, the functional consequences of these observations are unknown. Numerous examples of soluble cytokine receptors have been described to exert immunomodulatory effects in vivo . These range from antagonistic (IL-1RII) or agonistic (IL-6R) effects on receptor signalling, to acting as ligand chaperones or carrier proteins (IL-4R) [12] . Although there have been several descriptions of sCD25 acting as an inhibitor of IL-2 induced T cell responses in vitro [10] , [13] , whether it plays a similar role in vivo has not been determined. sCD25 is known to be generated as a result of proteolytic cleavage, largely from the surface of activated T cells and levels of CD25 ‘shedding’ are directly related to the rate of proliferation of activated T cells [14] . The levels of systemic sCD25 in the steady state are known to be remarkably stable and as a consequence sCD25 has been used extensively as a biomarker reflecting inflammatory diseases and tumours characterized by T cell expansion [14] . However, whether these increased levels of sCD25 play any direct role in modulating disease has not been fully investigated.
In this study we demonstrate for the first time that sCD25 exacerbates experimental autoimmune encephalomyelitis (EAE). These effects are associated with the enhanced generation of Th17 type responses in the periphery and increased infiltration of both CD4+ Th1 and Th17 cell subsets into the central nervous system (CNS). Similar to monoclonal antibody mediated IL-2 neutralization, sCD25 also enhances Th17 responses in vitro and acts early during the Th17 developmental programme by inhibiting signalling downstream of the IL-2R through its ability to sequester local IL-2. These data identify a previously unappreciated role for sCD25 in the pathogenesis of autoimmune disease.
Materials and Methods
Mice
Female C57BL/6J mice (Charles River, Ireland) and IL-17AeGFP mice, on a C57BL/6 background, (Biocytogen, Worcester, MA, USA) aged between 6–8 weeks were utilised in experiments. Mice were housed under SPF conditions at the Institute for Molecular Medicine, St. James Hospital Dublin. All animal experiments were performed in compliance with Irish Department of Health regulations (license number B100/4272) and approved by the institutional ethical review board.
Materials
ELISA kits for mouse IL-17A, IFNγ, IL-2 and IL-22 were purchased from ebioscience (Hatfield UK). ELISA kit for sCD25 was purchased from R&D systems (Abingdon, UK) Recombinant murine sCD25His was purchased from R&D systems. Endotoxin levels in sCD25 were determined by LAL assay and found to be lower than 0.05 EU/µg of protein. These levels were found to exert no detectable levels of immune stimulation on primary macrophages in vitro . All antibodies used in this study from ebioscience unless otherwise stated. Recombinant mouse IL-6 was purchased from Becton Dickinson (Oxford, UK), IL-12p70 and TGFβ from ebioscience. Anti-IL-2 neutralizing antibody (JES6-1A12) was purchased from ebioscience.
Flow Cytometry
Cells were analysed for surface and intracellular protein expression using an LSR/Fortessa (BD). For intracellular staining, cells were stimulated with PMA 10 ng/ml and ionomycin 1 µg/ml for 6 h in the presence of Brefeldin A (Sigma) 5 µg/ml for the final 4 hrs. Fixation and permeabilisation was performed using FIX/PERM Kit (Dako) according to manufacturer's instructions. Intracellular staining was performed for IL-17A (17B7), IL-17F (eBio18F10), IFN-γ (XMG 1.2) and FoxP3 (FJK-16s) and RORγT (AFKJS-9) (ebioscience, UK). pSTAT5 (pY694) and pSTAT3 (pY705) staining was performed with Phosflow kit according to the manufacturer's instructions (Becton Dickinson, UK). Surface staining was performed for CD4 and CCR6. Cell surface staining for sCD25 was performed using anti-HIS (GG11-8F3.5.1) (Miltenyi Biotec).
EAE Induction and sCD25 treatment
EAE was induced according to manufacturer's instructions using active EAE induction kit EK-0113 (Hooke Labs MA. U.S.A). Mice were monitored daily for signs of disease with disease severity recorded as follows 0. Normal, 1. Limp tail, 2. Wobbly gait, 3. Severe hind limb weakness 4. Complete hind limb paralysis 5. Moribund or death. Recombinant sCD25 was administered immediately prior to immunization and every 12 hours thereafter for the first 3 days. Control mice were treated with PBS.
T cell isolation and differentiation
Naive CD4+CD62L+T cells from spleens of 8 week old mice were purified by magnetic bead separation (Miltenyi Biotec). Cells were activated with plate bound anti-CD3 (145-2C11) and anti-CD28 (37.51) both 5 µg/ml. For Th17 differentiation cells were cultured in the presence of TGF-β 5ng/ml, IL-6 10 ng/ml, anti-IFN-γ 10 µg/ml (XMG 1.2) and anti-IL-4 10 µg/ml (11B11). For Th1 differentiation cells were cultured with IL-12 10 ng/ml and anti-IL-4 10 µg/ml. iTreg cells were induced in the presence of TGF-β 5 ng/ml and rIL-2 (10 U/ml). After 72–96 hours supernatants were analysed by ELISA and cells were examined for intracellular cytokine/transcription factor expression by flow cytometry.
Results
sCD25 leads to exacerbated autoimmune disease and increased antigen-specific peripheral Th17 responses
Specific alleles at the CD25 gene locus, known to be associated with susceptibility to autoimmune diseases such as Multiple Sclerosis (MS), lead to increased levels of soluble CD25 in patient's serum [10] . Although such observations implicate sCD25 as having an important mechanistic role in disease pathogenesis, it is not clear how sCD25 may contribute to a loss of self tolerance. To determine what role, if any, sCD25 may play in autoimmunity we induced experimental autoimmune encephalomyelitis (EAE), a mouse model of MS, in the presence of exogenous recombinant sCD25 administered immediately prior to, and during the first 3 days after immunization. Increased levels of sCD25 during the early stages of antigen-specific T cell priming led to a significant exacerbation of disease symptoms during the onset and induction phase of the disease from day 10 through to the peak of disease after 18 days (P>0.01) ( Fig. 1A ). To examine the cells infiltrating into the CNS during EAE, IL-17-eGFP reporter mice were immunized with MOG, in the presence or absence of sCD25, and the expression of IL-17A or IFNγ was assessed at 15 days after induction of EAE. Although the relative percentages of infiltrating Th1 versus Th17 type cells was not altered ( Fig. 1B ), administration of sCD25 resulted in significantly increased numbers of both subsets in the spinal cords of treated mice at day 15 during disease induction ( Fig. 1C ). We also examined the effects of sCD25 administration on the generation of peripheral antigen specific T cell responses in vivo . Significantly, increased levels of sCD25 were found to result in increased antigen-specific T cell expression of IL-17A upon MOG antigen restimulation ex vivo 7 days after immunization (p>0.05) ( Fig. 2A &B ). Expression of IFNγ was not significantly affected ( Fig. 2A ). Furthermore, administration of sCD25 did not affect the levels or relative numbers of CD4+Foxp3+ regulatory T cells in immunized mice after 7 days, indicating that increased severity of EAE did not occur in association with any effects on Treg homeostasis ( Fig. 2C&D ). Together these data demonstrate that increased levels of sCD25 in vivo led to increased severity of EAE that occurs in association with enhanced generation of antigen specific Th17 responses in the periphery and increased numbers of both of CD4+ Th1 and Th17 cell subsets in the CNS. These observations are consistent with previous reports which demonstrate that administration of an anti-IL-2 neutralizing antibody leads to the spontaneous development of EAE-like symptoms in mice and also that treatment with recombinant IL-2 during the early stages of disease can offer significant protection from EAE [15] – [16] .
10.1371/journal.pone.0047748.g001
Figure 1
Exogenous sCD25 exacerbates autoimmunity.
( A ) MOG 33−55 immunized C57BL/6 mice developed clinical symptoms of EAE from day 12 after immunization with a peak of disease severity observed from day 19. Subcutaneous administration of recombinant sCD25 (25 µg/mouse) immediately prior to immunization and every 12 hours thereafter for 72 hrs resulted in a significant exacerbation in severity of symptoms during disease onset and induction. 6–7 mice used per group. ( B ) Mononuclear cells harvested from spinal cords of control (PBS) and sCD25 treated IL-17A-eGFP reporter mice (3 per group) on day 15 after immunization and analysed for expression of IL-17 and IFN-γ by CD4+ cells. ( C ) Cell numbers of CD4+IL-17+ and CD4+ IFNγ+ cells in spinal cords of IL-17A-eGFP reporter mice at day 15. Data representative of mean +/− std dev of 3 mice per group and 2 independent experiments.
10.1371/journal.pone.0047748.g002
Figure 2
sCD25 enhances peripheral antigen specific Th17 responses.
( A ) Levels of IL-17A and IFNγ secreted by draining lymph node cells of mice immunized and treated as above (3 per group), isolated after 9 days, and restimulated ex vivo with MOG 33−55 (10 µg/ml) for 72 hours. ( B ) Percentage of CD4+ IL-17A-eGFP+ve cells after treatment as above and ex vivo restimulation for 72 hours. ( C ) Percentage and ( D ) relative number ratios of CD4+FoxP3+:CD4+FoxP3- T cells in the draining lymph nodes of both sCD25 and control treated immunized mice. 3 mice per group were analysed. Data in D & G is representative of mean +/− standard deviation. Statistical Significance determined by unpaired student's t-test, * p≤0.05, **p≤0.01.
sCD25 enhances the development of Th17 cell responses
Conflicting studies have demonstrated both antagonistic and agonistic roles for sCD25 in the context of IL-2R signalling indicating that sCD25 could either promote or inhibit Treg responses and inhibit IL-2 mediated activation induced cell death in vitro [10] [17] . As sCD25 enhances the generation of peripheral autoimmune antigen-specific Th17 responses in vivo , we sought to determine how sCD25 might regulate these events by investigating the effects of sCD25 on the generation of Th17, Th1 and Treg responses in vitro .
sCD25 significantly enhanced the generation of Th17 type responses after 96 hours in vitro in a dose dependant manner and to a similar extent to an anti-IL-2 neutralizing antibody ( Figure 3A ). In contrast, no effects of sCD25 were observed on the generation of either Th1 or inducible Treg subsets ( Figure 3 A, B & C ). As Th17 cells are considered critical in driving the pathogenesis of EAE, these data are consistent with our earlier in vivo data ( Figure 1 ). We also examined what influence sCD25 may have on Th17 cell proliferation and survival as IL-2 is a well established T cell growth factor and we observed an increase in the overall numbers of infiltrating CD4+ T cells in the spinal cords of treated mice ( Figure 1C ), Although sCD25 resulted in an increase in the proportion of cells expressing IL-17A, it did not affect either the rate of cell division as determined by levels of CFSE dilution, or levels of survival/cell death, as measured by incorporation of 7-AAD, under these conditions ( Figure 3D ). However, consistent with the enhanced generation of a Th17 response, we also observed increased levels of phosphorylation of STAT3 ( Figure 3E ). sCD25 also led to a marginal but reproducible increased expression of both IL-17F and the Th17 associated chemokine receptor CCR6, while levels of IL-22 expression were not affected (data not shown). These data clearly demonstrate that sCD25 can enhance Th17 cell development in vitro and suggest a mechanism through which sCD25 may increase autoimmune disease severity.
10.1371/journal.pone.0047748.g003
Figure 3
sCD25 enhances Th17 cell responses in vitro .
( A&B ) Purified naive CD4+ T cells were activated under either Th17 or Th1 inducing conditions (as described in methods ) in the presence of a range of concentrations of sCD25 (20, 10, 5 or 1 µg/ml) or anti-IL-2 (10 µg/ml). Levels of IL-17A or IFNγ expression were determined after 96 hrs by ( A ) FACS and ( B ) ELISA. ( C ) Purified naive CD4+ T cells were activated under Treg inducing conditions, as described in methods , in the presence or absence of sCD25 (20µg/ml) and FoxP3 expression determined by FACS. ( D ) Naive CD4+ T cells were stained with CFSE (2.5 µM) prior to activation under Th17 conditions in presence or absence of sCD25 (20 µg/ml). After 96 hours, levels of intracellular IL-17A expression and CFSE dilution or 7AAD incorporation were determined by FACS. ( E ) Purified naive CD4+ T cells were activated under Th0, Th17 and Th17& sCD25 (20 µg/ml) conditions for 72 hours and levels of P-Stat3 (pY705) determined by FACS. All data are representative of 3 independent experiments. Statistical Significance determined by unpaired student's t-test, ∗ p≤0.05, **p≤0.01, ***p≤0.001.
sCD25 inhibits IL-2R signalling on Th17 cells through sequestering extracellular IL-2
Recent reports have demonstrated that CD4+ FoxP3+ve Tregs can enhance the generation of Th17 type responses early during the developmental programme by limiting the bioavailability of IL-2 through constitutive expression of CD25 on their cell surface [18] , [19] . Given these observations, we investigated whether sCD25 could act in a similar way. Strikingly, the presence of sCD25 led to a significant enhancement of IL-17A (11% vs 2%) expression early during Th17 cell differentiation (48 hours) indicating that sCD25 also mediated its effects early during Th17 development ( Figure 4A ). Consistent with this observation, sCD25 also led to increased expression of the Th17 instructive transcription factor RORγT ( Figure 4B ).
10.1371/journal.pone.0047748.g004
Figure 4
sCD25 acts early during Th17 development.
Purified naive CD4+ T cells from IL-17AeGFP reporter mice activated under Th17 inducing conditions in the presence of sCD25 (20 µg/ml) or anti-IL-2 (10 µg/ml) for 48 hrs and examined for levels of ( A ) IL-17A and ( B ) RORγT expression by FACS. All data are representative of at least 3 independent experiments.
As the effects of sCD25 were similar to those observed for IL-2 neutralization [9] , we examined whether signalling downstream of the IL-2R was affected in these cells. IL-2 is expressed early during the first 24 hours after TCR stimulation of CD4+ T cells and activation of Jak3-STAT5 dependent signal pathways in T cells during this time is considered to be largely driven by the autocrine effects IL-2. sCD25 significantly decreased levels of STAT5 activation in Th17 cells demonstrating its ability to inhibit signalling downstream of the IL-2R ( Figure 5A ). IL-2 dependent activation of STAT5 signalling is known to directly inhibit early programming events in the development of a Th17 response by blocking the induction of RORγT expression [9] . These data identify a novel mechanism whereby sCD25 enhanced the generation and development of proinflammatory Th17 responses through inhibiting the protolerogenic effects of IL-2R signalling.
10.1371/journal.pone.0047748.g005
Figure 5
sCD25 inhibits IL-2R signalling by sequestering secreted IL-2.
( A–D ) Purified naive CD4+ T cells activated for 24 hrs under Th17 inducing conditions in the presence or absence of sCD25 and examined for levels of ( A ) p-Stat5 pY694, ( B ) intracellular IL-2 expression ( C ) surface binding of sCD25-His and ( D ) surface expression of endogenous CD25 by FACS analysis. ( E ) Levels of secreted IL-2 detected by ELISA 24 hours after Th17 activation in the presence of decreasing doses of sCD25 (20,10,5,1 µg/ml). ( F ) Levels of IL-2/sCD25 complex in the supernatants of Th17 cells activated for 24 hours in the presence or absence of sCD25 (20 µg/ml). Data shown as mean +/− standard deviation from triplicate experiments. All data are representative of at least 3 independent experiments. Statistical Significance determined by unpaired student's t-test, ***p≤0.001.
To determine the precise mechanism through which sCD25 was mediating this inhibition we considered a number of possibilities. First, sCD25 may inhibit the levels of IL-2 expressed upon T cell activation (although IL-2 neutralization by monoclonal antibodies has previously been found to enhance IL-2 expression by inhibiting an auto-regulatory negative feedback loop [20] ). We observed no differences between the levels of IL-2 expressed on a per cell basis either in the presence or absence of sCD25 after 24 hours ( Figure 5B ). Second, sCD25 may exert its effects at the cell surface by acting to either inhibit appropriate assembly of the heterotrimeric receptor complex or inhibit IL-2 binding. To examine this possibility we used a His-tag on the soluble form of the receptor to discriminate between soluble and surface expressed forms of CD25. However, we were not able to detect any binding of sCD25 to the cell surface during the first 24 hours after activation ( Figure 5C ). In contrast, the presence of sCD25 did significantly inhibit the upregulation of endogenous surface CD25 expression ( Figure 5D ). This observation further indicated a role for sCD25 in inhibiting IL-2R signalling as IL-2 is recognised as an important mediator in driving surface CD25 expression early during T cell activation. Third, we investigated the possibility that sCD25 may act to sequester secreted IL-2 in the T cell microenvironment. Significantly, sCD25 inhibited the detection of secreted IL-2 by ELISA in a dose dependent fashion suggesting its ability to sequester secreted IL-2 ( Figure 5E ). The specific interaction between sCD25 and IL-2 was also demonstrated using a mixed ELISA approach of capture with an anti-IL-2 antibody followed by detection with an anti-CD25 antibody. Using this approach, significant levels of the IL-2/sCD25 complex were detected in the supernatants of CD4+ T cells activated under Th17 conditions in the presence of sCD25 ( Figure 5F ).
Together these data demonstrate the immunomodulatory activity of the soluble form of the IL-2R alpha chain in vivo for the first time and indicate that these effects are mediated by its capacity to sequester secreted IL-2.
Discussion
The expression of the heterotrimeric IL-2 receptor on the surface of T cells plays a pleiotropic role in directing T cell responses. One such critical non-redundant role is the maintenance of peripheral T cell tolerance. This occurs through its promotion of the induction and persistence of regulatory T cell subsets while also acting directly to inhibit the generation of Th17 type responses which are considered to be critical in driving autoimmune disease [1] . Such observations are thought to form the mechanistic basis for the close association between mutations at the CD25 gene locus and enhanced susceptibility to a number of autoimmune diseases in humans.
Similar to other cytokine receptors expressed at the cell surface, the individual chains of the IL-2R are also known to exist in soluble form in serum [21] , [22] . In particular, stable expression levels of soluble CD25 observed in healthy adults has underscored its clinical use as a biomarker for a variety of inflammatory conditions [23] . Unlike other soluble cytokine receptors, no evidence exists for an alternative splice form at the CD25 gene locus which encodes a specific soluble CD25 protein. Consequently, the generation of the soluble form of this receptor is thought to occur through proteolytic cleavage at the cell surface by as yet unidentified proteases [24] . Levels of sCD25 generation increase upon T cell activation in vitro and enhanced levels observed in vivo are thought to be directly related to the magnitude T cell mediated inflammatory responses. However, recent analysis of specific autoimmune susceptibility alleles at the CD25 gene locus has uncovered a direct association between increased disease susceptibility, disease severity and increased levels of sCD25 [10] , [11] . These studies indicate that sCD25 may play an important mechanistic role in driving disease pathogenesis.
As expression of all three chains of the IL-2R signalling complex on the cell surface are known to be required for efficient IL-2 binding and the subsequent activation of downstream signalling events [25] , whether sCD25 has any physiological relevance or is a mere by-product of T cell activation and expansion has remained controversial. Despite the lower affinity of CD25 for IL-2 when compared to the heterotrimeric IL-2R complex, sCD25 has been found to bind IL-2 efficiently and have immunomodulatory effects in vitro [10] , [26] . It is also possible that sCD25 may interact with an as yet unidentified accessory protein(s) in vivo to enhance its affinity for IL-2. Along those lines, it is noteworthy that soluble IL-1RII is known to have its affinity for IL-1α/β enhanced almost 100 fold through its interaction with soluble IL-1R Accessory protein [27] . Although monomeric sCD25 has a molecular weight in the region of 40 kDa, it has previously been found to be present as part of a protein complex with a molecular weight in the region of 100 kDa in the synovial fluid of rheumatoid arthritis patients [28] . Although the accessory proteins involved in this complex were not identified, it was found to efficiently inhibit IL-2 mediated responses in vitro . Furthermore, sCD25 has been demonstrated to exist in homodimeric form, although whether this alters its relative affinity for IL-2 is unknown [29] . Studies are ongoing to determine whether sCD25 exerts its immunomodulatory effects in EAE through either oligomerization or binding accessory proteins in vivo .
Numerous studies have previously investigated the role of sCD25 in modulating T cell responses in vitro . These reports have often led to conflicting results with sCD25 having been variously described to both inhibit and enhance T cell responses. To our knowledge, no previous studies have examined the role of increased sCD25 in the clinical severity of an auto-immune disease. As sCD25 has been previously examined with respect to multiple sclerosis in humans, we chose a murine model of this disease to examine in vivo effects of sCD25. While a number of groups have demonstrated the capacity of sCD25 to inhibit IL-2 mediated proliferation of CD8+ cytotoxic T cell lines [28] , [30] , it is noteworthy that Maier et al. also demonstrated that sCD25 could inhibit IL-2 mediated STAT5 phosphorylation in primary CD4+ T cells while enhancing responses through the inhibition of activation induced cell death [10] . Our study further extends these in vitro findings and demonstrates that sCD25-mediated blockade of IL-2 signalling modulates T cell responses towards a Th17 phenotype.
Given the established role of IL-2 in mediating Treg homeostasis in vivo [3] , it is surprising that we did not observe any effects on Treg subsets in the presence of sCD25 in this study. Although we did not specifically examine whether sCD25 affected the suppressive function of Tregs, levels of Foxp3 expression both in vitro and in vivo clearly indicate that sCD25 did not impact Treg survival or persistence. Similarly, previous reports have found that IL-2 can play a role in enhancing Th1 mediated responses but these were not affected by sCD25 at the concentrations used in this study [31] ( Figure 2A , 3B ). As Th17 responses are clearly elevated under these conditions, these data may reflect differing levels of sensitivity among pathogenic versus regulatory CD4+ T cell subsets towards the effects of IL-2 signalling. Alternatively, it is possible that Tregs, which express constitutively high levels of surface CD25 (and the heterotrimeric IL-2R) in comparison to Th17 cells, may be competitively less sensitive to sequestration of circulating IL-2 by sCD25. Studies are ongoing to determine how limiting doses of IL-2 may differentially impact CD4+ T cell responses and how sCD25 might influence these events.
It is clear from our in vivo studies that the ability of sCD25 to enhance Th17 responses on a per cell basis is only observed in the periphery and not at the site of inflammation where although percentages of both Th17 and Th1 cells are remarkably unaltered upon sCD25 treatment, both are present in significantly increased numbers ( Figure 1 ). Although this may reflect the effects of sCD25 on T cell expansion, this seems unlikely given that we observed no such effects in vitro ( Figure 3D ). A further possible explanation for this is an increased plasticity or inter-conversion between both subsets at the site of inflammation. However, this is unlikely given that the IL-17A eGFP mouse used in these studies allows the identification of cells which also have a legacy of IL-17A expression. As such, increased plasticity would be evident as an increase in the percentage of GFP+ve cells expressing IFNγ which was not detected. More likely, these data indicate that enhanced antigen-specific Th17 cells in the periphery can facilitate the infiltration of both pathogenic Th1 and Th17 cells to the site of inflammation as has been previously reported [32] .
In direct relevance to this study, the use of humanized anti-CD25 antibodies is showing considerable promise as a potential therapeutic for Multiple Sclerosis [33] . Although efficacy for this approach has been demonstrated in early clinical trials, the exact mechanism through which these antibodies inhibit disease remains obscure. It is noteworthy that these antibodies bind sCD25 and block its ability to sequester IL-2 [34] . As levels sCD25 are elevated among MS patients [10] , blockade of its immunomodulatory effects with anti-CD25 could conceivably play an important part in the mechanism of action of Anti-CD25.
Together these data demonstrate the immunomodulatory activity of the soluble form of the IL-2R alpha chain in vivo for the first time and indicate that these effects are mediated by its capacity to act as a decoy receptor for secreted IL-2. Although biochemical studies indicate that CD25 in isolation has a significantly lower affinity for IL-2 when compared to the heterotrimeric IL-2R complex, it has been demonstrated to bind IL-2 efficiently and its ability to suppress IL-2 mediated responses in vitro has been extensively reported [10] , [13] , [26] , [30] . The association between elevated levels of sCD25 found in the sera of autoimmune patients and the presence of specific susceptibility alleles at the CD25 gene locus offer perhaps the clearest indication that sCD25 plays a role in autoimmune pathogenesis [10] . Although whether elevated levels of sCD25 are causally linked to the pathogenesis of human autoimmune disease remains to be determined, our studies demonstrate that sCD25 can act to enhance Th17 cell responses and provide a novel mechanism which may explain these observations.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Bacteriophages of the T7 group consist of a large number of related phages that grow on Escherichia coli and other bacterial genera. Many coliphages of T7 family have been isolated. Based the on the promoter specificity of phage RNA polymerase and the efficiency of recombination between the phages, T7 group phages that grow on E. coli B are classified into three sub-groups [1] . T7 is the prototype of the largest subgroup, the T7-like phages. T3 is the representative of the second subgroup. The third subgroup, the BA phages, contain phages BA 14, BA 127, and BA 156 [1] .
The sequence of the genome of T7 (Studier strain) was published in 1983 [2] , and that of T3 (Luria strain,T3L) was completed in 2002 [3] . T7 has a linear genome of 39937 base-pairs (bp), while T3 has 38208 bp in the complete genome. It was suggested that they are genetically isolated and not naturally interbreeding populations [4] . Their base sequences display homology across the genomes. Complementation between some phage genes, such as genes 9 , 10 , and 11 , are possible [5] . However, T3 has a number of properties that are not found in T7. T3 is able to escape F plasmid-mediated exclusion, and grow on E. coli male strain and Shigella sonnei D 2 371-48, while T7 fails to do so [6] . Gene product (gp) 0.3 of T3 possesses an S-adenosyl-methionine hydrolase (SAMase) activity not present in the T7 gp 0.3 [7] . Despite the fact that RNA polymerases from T3 and T7 resemble each other, they can not efficiently transcribe the heterologous DNA [8] . T3 and T7 recombine at a low rate, probably due to mutual exclusion [9] , although hybrid T3/7 phages have been isolated [10] , [11] , [12] . The mechanism of recombination between two T7 group's phages remains unclear. Repair of the T7 double strand breakage (DSB) using a 2.1 kb PCR fragment as a donor resulted in the incorporation of a patch of the donor in vitro, suggesting a strand annealing model of repair [13] . However, a short fragment is especially susceptible to degradation and unwinding, while how a requisite repair patch will be produced for the full-length genome donor is not known. The repair does not require genes 3 (endonuclease I) and 4 (helicase) [13] , [14] , whereas earlier studies on genetic crosses of mutations in phage T7 have implicated the involvement of genes 2.5 (ssDNA binding protein), 3, 4, 5 (DNA polymerase), and 6 (exonuclease) in T7 genetic recombination [15] , [16] , [17] . It is a concern that in vitro experiments may overestimate the importance of the patch incorporation pathway [13] , [14] .
Two different T7 with different phage buoyant density have been reported, T7 Meselson (T7M) and T7 Luria (T7) [4] , [18] . Davison and Freifelder also found that these two strains differ in electrophoretic mobility [18] . Both phages were found to have equal DNA length [4] . Davis and Hyman constructed the heteroduplex of T7M (obtained from the Caltech stock) and T3L. Extensive partial sequence homology was observed by electron microscopy. The degree of formation of heteroduplex decreased with increasing formamide concentration. The sequence of T7M has not been reported.
Phages rely on specific hosts for propagation. Unraveling the host range is vital to understanding the interplay between host and virus in the microbial community and its evolutionary consequence. In this study, we show that the propagation of a T7M phage varies on different host strains of E. coli , not conforming to the features of T7. Restriction mapping and partial sequencing reveal the close homology of T7M to T3. A T3 and T7 hybrid phage, T3/7, displays yet a different growth pattern. T3 adsorbed inefficiently to a refractory host strain. T7, despite efficient adsorption, was excluded by several host strains. The T3/7 recombinant phage gains efficient adsorption and also a broader host range. Crucial elements rendering the crossover within a small region of homology between T3 and T7 phages were identified, and the recombination model was proposed.
Results and Discussion
Propagation on female strains
Plating efficiency of T7, T3, T7M and T3/7 on female strains of E. coli was studied. All four phages can grow efficiently on BL21 to a titer of approximately 1×10 10 ml −1 or above. Their plaque sizes are similar (∼0.4 cm). On infecting another female strain DH5α, the efficiency of plating (EOP) is 0.57 for T3, 0.67 for T7M, and about three-fold lower for T3/7 than that of T7M ( Table 1 ). In contrast, T7 has an EOP lower than 10 −6 . The sizes of plaques of T3 and T7M are similar for the phage growing on BL21 and DH5α (∼0.4 cm), but the size (∼0.2 cm) of T3/7 on DH5α is only half of that on BL21.
10.1371/journal.pone.0030954.t001 Table 1
Efficiency of plating of phages on E. coli female and male strains determined at 37°C.
EOP
T3
T7
T7M
T3/7
BL21
1
1
1
1
K91
<10 −6
(8±6)×10 −3
(2.7±2.4)×10 −3
(8.3±1.7)×10 −1
DH5α
(5.7±0.7)×10 −1
<10 −6
(6.7±2.3)×10 −1
(2.3±1.0)×10 −1
XL1-Blue
(5.1±0.5)×10 −1
<10 −6
(8.7±1.8)×10 −1
(6.4±2.5)×10 −1
Propagation on male strains
The ability of phages to infect the male K91 strain was tested. T3, T7, and T7M all displayed poor plating efficiency on K91. The EOP of T3 was less than 10 −6 . T7 and T7M displayed very small plaques on K91, ∼0.1 cm or smaller in diameter compared to ∼0.4 cm when propagated on BL21, with EOP approximately around 10 −3 relative to BL21 ( Table 1 ). Although plaques are quite clear at high phage concentrations (10 7 –10 9 phages/ml), the numbers of the plaques are not so consistent at phage concentrations below this range. In contrast, T3/7 propagated on K91 with approximately the same EOP and plaque size as on BL21 ( Table 1 ).
The plating efficiency of phages on another male strain, XL1-Blue, was also tested. T3 shows an EOP of 0.51, but EOP of T7 is smaller than 10 −6 . T7M has a high EOP of 0.87 on this strain, albeit the plaques (∼0.15 cm) are smaller than on BL21 ( Table 1 ). EOP of T3/7 on XL1-blue is 0.64, and the plaque sizes (∼0.3 cm) are closer to those on BL21.
Mapping of phage DNA
Restriction mapping of T7M and T3/7 was used to gain insight into the complicated growth phenomena. T3/7 DNA was digested by Avr II, Hpa I, Mbo I, Nde I, and Stu I, and the fragments were separated by electrophoresis ( Figure 1A ). Comparison to the sizes of restriction fragments generated by these enzymes on T3 and T7 ( Table S3 ) shows that the restriction pattern of T3/7 genome was closer to that of T3 than T7, suggesting that the major part of the genome is from T3. Alterations of several nucleotide positions (indicated by nt below) in the T3/7 genome relative to T3 were observed ( Figure 1C ). Mbo I digestion of T3/7 DNA missed the 1423 bp fragment, but showed extra ∼400 bp and ∼1100 bp bands, indicating changes around nt 34033 to nt 35456 that create an extra cutting site. Nde I digest of T3/7 DNA missed the 698 bp, 718 bp, 2390 bp, and 3574 bp bands, but at ∼1400 bp and ∼6000 bp two extra bands appeared, indicating base changes near nt 718 and nt 34330 which abolished the two restriction sites. Bands produced by Avr II and Stu I digestions were the same as predicted for T3.
10.1371/journal.pone.0030954.g001
Figure 1
Mapping of the T3/7 and T7M DNA.
The genomic DNA was digested by restriction endonucleases, and the fragments were observed by 1% (T3/7) or 0.8% (T7M) agarose gel electrophoresis. M: DNA marker. (A) Digestions of T3/7 DNA. Lane 1, Mbo I; lane 2, Hpa I; lane 3, Nde I; lane 4, Stu I; lane 5, Avr II. The markers are the same for the three slices of gels. (B) Digestion of T7M DNA. Lane 1, Hpa I. (C) Restriction sites in the T3/7 genome. Restriction sites are represented by vertical bars above the DNA. A: The T3 genome (black) and the region replaced by T7 DNA (in grey) in T3/7. The region of recombination is deduced from restriction mapping and sequencing. The two Mbo I sites in T7 that replace the single Mbo I site at T3 nt 34033 are shown by vertical bars. B to E show restriction sites in T3 genome (black); a dot above the bar indicates that the cutting site is present in T3 but not in T3/7. B: Hpa I sites, C: Mbo I sites, D: Nde I sites, E: Avr II cutting positions are indicated by bars, and the single Stu I site is shown by an arrow.
The Hpa I digestion was shown to provide positive identification of T7 and related phages [6] . Hpa I fragments of the T3/7 DNA resemble those of T3 except an extra ∼8300 bp band and loss of 2098 bp and 6191 bp bands ( Figure 1A , lane 2), indicating that the cutting site at nt 14303 was eliminated in T3/7 relative to T3 ( Figure 1C ). Unexpectedly, Hpa I digestion of the genomic DNA from T7M gave bands distinctly different from those predicted from T7 genome, but consistent with those of T3 ( Figure 1B , Table S3 ).
Partial nucleotide sequences of T7M phage DNA
The high similarity of T7M to T3 was further confirmed by partial sequencing of T7M DNA. Twenty two percent of the T7M genome was sequenced. The sequenced regions of T7M DNA span 10 different areas on the genome of T7M ( Table S4 ). All the sequences are more homologous to T3 than T7. Compared to the published T3 sequence, only two base substitutions at gene 10B were found ( Table S4 ). One was a T to C change at nt 22151, which caused an Thr for Ile substitution at residue 421 of gene 10B . The other was an alteration of T to C at nt 22169, which retains the same amino acid.
Partial nucleotide sequences of T3/7 phage DNA
Approximately 25% of the T3/7 genome was sequenced. It covers 16 different areas of the T3/7 DNA, encompassing the regions of possible changes as revealed by the restriction mapping and other randomly chosen regions ( Table S5 ). The results demonstrate that the T3/7 DNA is largely based on T3 sequence except that a fragment is exchanged to that of the T7 sequence. However, several regions outside of the exchanged segment showed sporadic nucleotide substitutions compared to the published T3 sequence. These were found in front of gene 0.3 , and in gene 1.1 , gene 2.5 , gene 5 , gene 6 , as well as gene 10B , and in the region between gene 10B and gene 11 . The important parts of the sequence are described below.
The left arm of T3/7 is derived from T3
The sequences of T3/7 nt 26 to 457, and nt 492 to 1008 reveal that the terminal repeat, the coli RNA polymerase promoters A0, −35 signal of A1, A2, A3, and the phage promoter φOL as well as the RNaseIII recognition site R 0.3 are all identical to those of T3. In this region only two nucleotides, located between A1 and A2, and between A3 and R 0.3 , respectively, differ from those of the T3 phage, indicating that the left arm of the phage genome is derived from that of T3. The base substitution between A3 and R 0.3 results in the absence of an Nde I site at nt 718 of T3/7. One salient difference between T3 and T7 is the presence of SAMase encoded by gene 0.3 ( sam , nt 901–1359 of T3) in T3, but not in T7 [7] . Partial sequence of the gene 0.3 of T3/7 indicates that sam is present in the T3/7.
Gene 1.2 for escaping F exclusion is present in T7M and T3/7
The presence of the F plasmid pif A protein elicits the F exclusion when T7 infects E. coli [19] , [20] . The T7 gp1.2 or gp10 is necessary for the exclusion, while T3 gp1.2 can prevent the F exclusion [20] . Between T3 and T7, the N terminus of gene 1.2 retains homology, while the distal half exhibits low sequence homology. The complete nucleotide sequences of gene 1.2 of the T7M and T3/7 phages are identical to that of the T3. Therefore, the male exclusion of T7M and the growth difference between T7M and T3/7 are not caused by gene 1.2 .
Amino acid substitutions occur in gp5 DNA polymerase of T3/7
Phage T7 gp5 is a DNA polymerase with polymerase, exonuclease, and thioredoxin binding domains (TBD) [21] . The host protein thioredoxin (Trx) binds to gp5 and stimulates its polymerase and 3′ to 5′ exonuclease activity [12] . The complete nucleotide sequence of T7M gene 5 is identical to that of T3 gene 5 . T3/7 phage gene 5 incurs three base substitutions (nt 13784, 13956, 14303) and residue changes, A248T, A305V, and N421D, compared with that of T7M. The mutation at nt 14303 results in the loss of a Hpa I site at this position ( Figure 1 ). Residues 248 and 421 are located in the polymerase domain, and residue 305 is on the TBD, although not in direct contact with Trx, as illustrated by the crystal structure [21] . Whether the substitutions impose any effect on the gp5 awaits further studies. However, the efficient growth of T3/7 on BL21 suggests that the substitutions do not significantly affect propagation of the phage.
An amino acid change in the capsid protein from T3 to T7M and T3/7
Genes 10A and 10B encode viral capsid proteins. Partial sequences of gene 10A and 10B of T7M and T3/7 show high similarity to the counterpart sequences of T3 except for a couple of variations ( Tables S4 , S5 ). Most of these changes do not alter the amino acids. However, the T to C substitutions at nt 22151 in both T7M and T3/7 changed the residue 421 of gp10B from Ile to Thr. A –1 frameshift during translation near the end of the coding sequence for the major capsid protein gp10A produces the minor form gp10B in wild-type T7 and T3 phages [2] , [22] . T3 gp10A contains 347 residues. Gp10B is 86 residues longer than gp10A. Phage particles normally contain both gp10A and gp10B, but gp10B is dispensable for phage growth [22] . Both T3/7 and T7M could propagate efficiently on BL21, indicating that the substitution of Thr for Ile421 in gp10B does not impair the phage growth. Nevertheless, the possibility of the substitution affecting the structure of the protein or the capsid awaits detail structural investigation.
Difference in the stem-loop structures of Tφ between phages
Compared to T3L, a G to A replacement at nt 22374 of the T3/7 DNA is located between gene 10B and gene 11 , while T7M retains a G at this position. After the coding sequence of gene 10B , a transcription termination signal, Tφ, for T3 RNA polymerase is located between nt 22352–22390. The RNA transcript can form a stem-loop structure at this terminator ( Figure S1A ). No base changes were observed when sequenced up to nt 22385 for the T7M terminator, and the T7M terminator is expected to have the same structure as that of the T3, which comprises a 17 bp stem and a 5-nucleotide loop. Substitution of A for G at nt 22374 in the T3/7 terminator results in one less base pairing at the top of the stem, and 2 more nucleotides in the loop ( Figure S1B ). Tφ in T7 has a 15 bp stem, and a 6-nucleotide loop ( Figure S1C ) and is inefficient in terminating transcription by T7 RNA polymerase [2] . Whether the stem-loop structural difference between these phage terminators has an effect on the termination efficiency requires further investigation.
A large fragment of DNA changed to the T7 sequence in phage T3/7
A large piece of sequence different from the T3 DNA was identified in T3/7. The sequencing results indicate that the sequences of T3/7 nt 33146–33331 and nt 35382–35936 are indistinguishable from those of T3. However, between these two regions (T3/7 nt 33332–35382), except for a fragment of identical sequence between T3 and T7 (T3 nt 35315–35395 is identical to T7 nt 37094–37174), T3/7 changes to the T7 sequence. This is demonstrated by the identity of T3/7 nt 33332–33638 and nt 34788–35329 to T7 nt 35124–35430 and nt 36580–37121, respectively ( Table S5 ), as well as the consistence of the mapping to T7 sequence rather than T3 sequence in this region. Two smaller Mbo I fragments replace the 1423 bp one, as the DNA exchange gives Mbo I sites at nt 33893 and nt 34296, in agreement with the T7 sequence, while eliminating the T3 Mbo I site at nt 34033 ( Figure 1A & C , Table S3 ). Moreover, T3 DNA has a Nde I site at nt 34330, yet the exchanged T7 fragment does not possess a equivalent site, resulting in a ∼6 kb fragment instead of two smaller ones ( Figure 1A & C , Table S3 ). The retention of the Mbo I site at T3 nt 35456, and the Hpa I sites at T3 nt 34116 and 35125 that have equivalent restriction sites in the T7 DNA, are all consistent with the sites of exchange ( Figure 1 , Figure 2 ).
10.1371/journal.pone.0030954.g002
Figure 2
Alignment of T3 and T7 sequences near the crossover regions seen in the T3/7 phage.
(A) Within gene 17 . (B) Within gene 18.5 . Sequences were aligned by ClustalW. The parent phages of the two sides of the crossover region (shown by arrows) are indicated on top of the alignment.
The T3/7 DNA sequence switches from T3 to T7 at nt 33332, and reverts from T7 to T3 sequence anywhere between nt 35302–35382, where the sequence is identical to both T3 and T7 [2] , [3] . The exchanged genes are part of gene 17 , complete genes 17.5 and 18 , and part of gene 18.5 and gene 18.7 .
The recombinant tail-fiber protein of T3/7
Tail fiber proteins participate in the adsorption of phages to the bacterial outer membrane. Bacterial lipopolysaccharide (LPS) was shown to be the receptor for phage T3 and T7 [23] . T3 and T7 tail-fiber proteins encoded by gene 17 consist of 558 and 553 residues, respectively. Each of the six tail-fibers is comprised of a trimer of gp17 [24] . The amino acid sequences of the N-terminal domain (residues 1–149) that interacts with the tail-tube have high homology between T3 and T7, only residues 48 and 118 are different, both incur an Ala to Thr substitution from T3 to T7. Whether the Ala and Thr exchanges have any effect on the interaction with the tail-tube is not clear. Nevertheless, T3/7 has Ala at positions 48 and 118, ensuring proper connection with the tail-tube of the T3 type in the hybrid phage. Sequences of gp17 beyond residue 149 contain antigenicity and the host range determinants, and display less homology (77% identity) between T3 and T7 [24] . Residues 150 to 167 are identical between gp17 of T3 and T7. T7 sequence in T3/7 commences from nt 33332, which resides in the codon for residue 167. Therefore, the antigenecity of T3/7 is expected to be similar to phage T7. The host range of the T3/7, however, is different from T7. The T3/7 hybrid can efficiently infect male E. coli including K91 ( Table 1 ), showing that the hybridization has broadened the host range compared to T3 and T7.
Efficiency of phage adsorption to bacteria
Adsorption of T3/7 to bacteria was compared to the T3 and T7 to investigate the effects of the alteration of tail fiber. Despite the low EOP of T7 on K91, DH5α and XL1-Blue, the efficiency of adsorption of T7 on these strains is higher than 95%, indicating the full competence of the T7 tail fiber in the receptor adsorption ( Table 2 ). The low EOP of T7 on K91 and XL1-Blue can be explained by F exclusion. However, the cause of the low propagation efficiency of T7 on DH5α remains to be investigated. A high adsorption efficiency of 98% but low EOP was also observed for T7M infecting K91, suggesting other factors are affecting the T7M propagation. The efficiency of adsorption of T3 is much worse than T7. The adsorption efficiency of T3 is 46% on XL1-Blue, and plunges to 29% on K91 and DH5α. The deficiency of adsorption of T3 to K91 may contribute to a portion of the low EOP. As T3 has the same low adsorption efficiency on DH5α, yet still exhibits an EOP of 0.51, other undiscovered components may also suppress the propagation of T3 on K91. Compared to T3, the recombinant T3/7 gains a substantial increment in adsorption to K91, DH5α, and XL1-Blue. The efficiency of adsorption to K91 rises from 29% for T3 to 88% for T3/7. The increase in EOP of T3/7 compared to T3 on K91 is more than five orders, among which the proportion conferred by the adsorption increment awaits further elucidation.
10.1371/journal.pone.0030954.t002 Table 2
Adsorption efficiency (%) of phages on E. coli female and male strains.
Adsorption efficiency (%)
T3
T7
T7M
T3/7
BL21
75±4
86±5
95±2
78±5
K91
29±9
96±4
98±1
88±2
DH5α
29±5
95±1
94±1
85±4
XL1-Blue
46±4
97±1
95±1
84±5
T7 gene 17.5 and gene 18 can substitute for those of T3 in the T3/7 genome
The gene 17.5 encodes a lysis protein, holin, of 67 residues [2] , [3] . Between T3 and T7, the nucleotide sequences of gene 17.5 have 87% identity. Only two amino acids are different between the T3 and T7 gp17.5. Substitution of T7 gp17.5 for T3 gp17.5 in T3/7 does not affect the lysis function since T3/7 lyses BL21 normally.
Gene 18 encodes a non-capsid DNA packaging protein A (terminase small subunit) [2] , [3] . Gp18 exhibits 93% identity between T3 and T7, with the carboxyl terminal residues 34–89 completely identical. The gp18 and gp19 complex, which has a prohead-stimulated, DNA-dependent ATPase activity, is required for DNA packaging [25] . The ability of T3/7 to grow efficiently in BL21 indicates that phage DNA packaging proceeds normally. The replacement of gene 18 in T3/7 does not impede the DNA packaging, showing that gp18 from T7 can complex with gp19 from T3 in packaging DNA in vivo. It was also reported that gp18 from T3 or T7 was able to complement the gene 18 − extract in packaging DNA of either phage into the prohead in vitro [25] .
N-termini of T3 gp18.5 and gp18.7 can be substituted by T7 counterparts in the T3/7 genome
Both genes 18.5 and 18.7 are involved cell lysis. Gene 18.5 encodes a protein of 143 residues for T7 and 147 residues for T3, with 90.5% identity. The protein is an analog of phage lambda protein Rz, which might be an inner membrane murein-specific endopeptidase [26] . Gene 18.7 overlaps with gene 18.5 , but starts from +1 frame. Gene 18.7 encodes an 83-residue protein analogous to lambda protein Rz1, an outer membrane lipoprotein involved in lysis [27] . The C-terminal 10 residues of gp18.7 interact with the C-terminal 50 residues of gp18.5 in T7 [28] . The gp18.5–gp18.7 complex was proposed to lead to fusion of the inner and outer membrane and the final process of bacterial lysis [29] . The T3/7 phage reverts to T3 sequence between nt 35315–35395, which is before the sequences of interaction domains for gp18.5 and gp18.7, thereby securing the proper protein complex formation and the subsequent lysis event. Despite the T3 to T7 DNA replacement causing T48A, A49K, N56D, E57A, and I58V substitutions in gp18.5, and two residues alterations, K19R and R21L, in gp 18.7, T3/7 formed normal plaques in the BL21, indicating that 86 N-terminal residues of T3 gp18.5 and 47 N-terminal residues of T3 gp18.7 can be substituted by their T7 counterparts without hampering the propagation of T3/7.
T3 gp7.3 is functional with the heterologous T3/7 tail fiber protein
The phage gene 17 tail fiber protein is the main determinant of adsorption and specificity toward bacteria. However, gene 7.3 is also required for T7 to form plaques on E. coli B and K12 strains [30] . Gp7.3 is a host specificity protein located on the tail. Without gp7.3 the virion can assemble, yet the assembled tail fiber does not adsorb to the bacteria [30] . The protein may assemble the tail and tail fiber in a proper state for adsorption.
Differences between gp7.3 of T7 and T3 were observed. Gp7.3 is 99 residues in T7 phage and 106 residues in T3 phage. Their amino acid sequences display significant variations with about 56% identity. Whether a heterologous gp7.3 and gp17 combination is functional is an interesting question. The sequencing data demonstrate that both T7M and T3/7 harbor the T3 gp7.3. The ability of T3/7 to infect B and K12 strains efficiently demonstrates that the T3 gp7.3 can function together with the T3/7 hybrid tail fiber proteins for adsorption and infection.
Crossover sites and advantages gained by the T3/7
T3/7 nt 33332 is the third base of the codon for Glu167 in gp17. Its substitution from G to A (nt 35124 of T7) starts the T7 sequence in T3/7. The sequence of nt 33324 to 33331 are identical between T3 and T7, and bases immediately in front of it vary between these two phages ( Figure 2A ). Therefore, the crossover in gene 17 happened within the short stretch of 8 nucleotides. The crossover back to the T3 sequence occurred in the nucleotides encoding gp18.5 residue 60 to 86 (T3 nt 35315–35395, equivalent to T7 nt 37094–37174, Figure 2B ).
The crossover sites reveal several advantages gained by the T3/7. It was shown that a T3/7 hybrid phage with heterologous terminal redundant regions can not be replicated effectively [31] . T3 and T7 package homologous DNA more efficiently than heterologous DNA. Gp19 is involved in the specificity for packaging the homologous DNA, and the DNA sequence responsible for recognition is located within 5% of the termini of the phage genomes [32] . Thus in the present T3/7 phage, the crossover in gene 17 can enhance adsorption to certain hosts, and that in gene 18.5 can assure that gene 19 and the right end terminal DNA retain the sequences of the T3 phage.
During transcription, phage RNA polymerase interacts with two phage proteins, gp3.5 (lysozyme) and gp19 [33] . The polymerase-gp3.5 complex pauses at the right end of the concatemer junction and interacts with gp18 and gp19 to initiate maturation and DNA packaging [34] . By returning back to the T3 sequence within gene 18.5 , T3/7 can retain the efficient interaction between the RNA polymerase and gp19.
Gp19 also binds to the prohead of phage. Residues 571 to 576 of gp19 form the core domain crucial for binding the prohead, and the C-terminal ten residues, 577 to 586, form the anchor domain for stable binding [35] . These amino acids display low identity between T3 and T7. Thus, the change back to gene 19 of T3 can also benefit T3/7 by allowing a stable complex between gp19 and the prohead.
In light of the above multiple functions of gp19, the reversion to T3 gene 19 sequence may be a prerequisite for the packaging and maturation of the hybrid phage.
Identification of four-way junction structures for Endo I cleavage
To understand the mechanism of recombination within the short 8 nucleotides, we identified that the nearby nucleotides can form four-way junction structures [36] ( Figure 3 ). Such structures are apt to be cut by phage endonuclease I (Endo I) [36] ; for instance in Figure 3A , the cutting sites are primarily between nt 33330 and 33331 for the sense strand and between nt 33346 and 33347 for the antisense strand.
10.1371/journal.pone.0030954.g003
Figure 3
Structures of four-way junctions in gene 17 of T3 and T7.
(A) T3 nt 33324–33353 (B) T3 nt 33317–33347 (C) T7 nt 35109–35139 (D) T7 nt 35082–35117. One strand is highlighted in grey; the other is not. Arrows indicate Endo I cutting sites.
The other crossover occurs within gene 18.5 . Search for four-way junctions for Endo I cleavages identified prominent areas around T3 nt 35315–35345 (sequence identical to T7 nt 37094–37124) and T7 nt 37163–37188 ( Figure 4 ). Endo I does not exhibit a clear minimal arm length requirement at a junction; a duplex arm of 4 base pairs or shorter was cleaved [36] , [37] . The cleavage is essentially structure specific and prefers non-crossover strands at branched junctions [37] .
10.1371/journal.pone.0030954.g004
Figure 4
Structures of four-way junctions in gene 18.5 of T3 and T7.
(A) T3 nt 35315–35345 or equivalently T7 nt 37094–37124 (B) T7 nt 37163–37188. One of the strands is highlighted in grey. Endo I cutting sites are shown by arrows.
Recombination models
The previous in vitro experiment shows efficient patch incorporation of a donor fragment in T7 DSB repair [13] . It suggests that a T7 fragment, which covers the exchanged area, can serve as a donor for T3 DSBs to yield the recombinant T3/7.
Separately, the short stretch of homology and special sequence in the crossover site also shed light on the recombination path. The identification of the four-way junctions in this region suggests a role of the Endo I cleavage in phage recombination. Seeing that the ends generated by Endo I cleavages of the 8 bp homology regions in both T3 and T7 can produce complementary pairing, a simplest model is proposed based on c leavages at e quivalent-sites of phage DNA followed by s trand a nnealing (CESA) ( Figure 5 ). The available sequences allow a closer examination of whether this and the following (see below) proposed mechanisms will produce sequences at the crossover sites matching those of the recombinant T3/7. Endo I cleavages at four-way junctions produce DSBs. After cutting the structures in the equivalent positions of T3 ( Figure 3B ) and T7 ( Figure 3C ), the 5′ protruding tail of T7 DSB will comprise of 8 bases that complement T3 DSB nt 33324–33331, followed by 9 bases in incomplete homology with T3 5′ overhang nt 33332–33340 ( Figure 5 ). Phage gp2.5 is a recombinase that can mediate single strand annealing [38] . The incomplete homologous T3 5′ overhang is displaced and cleaved by T7 (or T3) gp6 exonuclease [39] . The gap is refilled by T7 gp5/Trx or E. coli DNA polymerase I using the T7 tail as a template, and then the ends are ligated [40] , [41] .
10.1371/journal.pone.0030954.g005
Figure 5
The mechanism of endonucleolytic cleavages at equivalent sites and strand annealing (CESA).
T7 and T3 DNA are shown in blue and pink, respectively. Arrows indicate the 3′ end. The steps are: 1, cleavage of T3 and T7 genomes by Endo I at equivalent sites; 2, annealing of the DSBs at both ends of T3 genome with the T7 DSB from the middle of the genome; 3, displacement and removal of the nonhomologous region, then filling the gap and ligation of the ends.
The Endo I cutting sites ( Figure 4A ) in gene 18.5 are within a longer homologous region ( Figure 2 ), and therefore the 5′ protrusions of the DSBs generated from T7 and T3 genomes can be directly annealed through 15-base complementarity and then ligated ( Figure 5 ). T7 gp2.5 has a high efficiency in stimulating DNA annealing, much higher than T4 gene 32 protein and E. coli SSB protein [38] . T3 DNA ligase has an efficiency 5 to 6-fold higher than T4 DNA ligase for cohesive DNA fragments. In vitro ligation of 4.5×10 −8 M Hin dIII cut fragments using 0.1 nM T3 DNA ligase achieved 90% ligation in 30 min without gp2.5 [42] . Phage gp2.5 and ligase thus can promote the recombination reaction through the strand annealing mechanism.
Cutting at equivalent positions in two phages is not the only way to produce two DSBs capable of annealing. Alternatively, recombination can also proceed through c leavages at n onequivalent-sites of DNA followed by s trand a nnealing (CNSA) ( Figure 6 ). For instance, two four-way junction structures were identified to be in nonequivalent positions in gene 17 of T7 and T3, where the structure in T7 ( Figure 3D , T7 nt 35082–35117) lies upstream of that in T3 ( Figure 3A , T3 nt 33324–33353). Endo I cleavages of both structures result in DSBs that can be annealed after 5′ resections ( Figure 6 ). The T3 DSB 3′ tail (nt 33324–33330) can anneal with the T7 DSB 3′ nt 35116–35122 through the 7-base pair complementarities. The 5 nucleotides (T7 nt 35111–35115) at the 3′ end of T7 DSB are not completely homologous with T3 nt 33319–33323, and therefore are excised with the gap refilled. The 3′ to 5′ exonuclease activity and polymerase activity of gp5/Trx or E. coli DNA polymerase I can carry out the excision and gap-filling [40] , [41] . Similarly, within gene 18.5 , a T7 four-way junction structure ( Figure 4B , T7 nt 37163–37188) lies downstream of the position of the T3 four-way junction structure ( Figure 4A , T3 nt 35315–35345). Endo I cleavages of both structures and 5′ resections of the DSBs result in a T7 DSB with a 3′ protrusion ending at nt 37169 and a T3 DSB with a 3′ tail ending at nt 35338. The 3′ overhangs of the two DSBs are annealed due to the identical sequences between T7 nt 37117–37169 and T3 nt 35338–35390. The remaining gaps are filled and ligated. The frequency of phage recombination can be increased through the possibility of annealing DSBs generated by different Endo I cutting sites.
10.1371/journal.pone.0030954.g006
Figure 6
The mechanism of endonucleolytic cleavages at nonequivalent sites and strand annealing (CNSA).
Colors and arrows are the same as those of Figure 5 . The steps are: 1, production of DSBs by cutting at nonequivalent sites of T3 and T7 DNA; 2, 5′ resections of DSBs of T3 and T7; 3, annealing of the T3 DSBs with the T7 DSB; 4, removal of the nonhomologous nucleotides, filling the gap, and ligation.
Homologous recombination invoking strand transfer, such as double Holliday junction formation [43] , remains a possible pathway for the production of T3/7. A very low frequency of recombination is expected by utilization of the 8 bp homology for strand transfer, since strand transfer has a minimal requirement of 23–27 bp homology [44] . Nevertheless, except several mismatches, the 8 bp homology occurs in a context of sequences sharing longer homology that has a possibility in giving rise to strand transfer. However, in this case all the mismatches in front of the 8 bp match will need to be repaired selectively to maintain the sequences of T3 to yield the T3/7. A combination of strand annealing and strand transfer reactions is also possible in recombination. Small regions of microhomology have been suggested to be positions of genetic exchange through homologous recombination, although the detailed mechanism was not investigated [45] .
Possibility exists that despite the presence of the four-way junctions in the crossover sites of T3 and T7, cleavages were not performed at these positions. Inside the cells, the left-side and right-side DSBs of the T3 genome can be generated by other endonucleases to include the left and right crossover sites, respectively. Similarly a T7 DSB comprising the exchanged fragment can also be generated through cutting sites different from the Endo I sites. Subsequent exonuclease digestion of these DSBs can produce single-stranded regions that can then be annealed to form a recombinant phage. However, for generation of the T3/7 with DSBs cleaved farther away from the crossover sites, the frequency of a) extensive unwinding and exonucleolytic digestion to reach the crossover areas, b) annealing the left-side DSB of T3 with a T7 DSB at only the 8 bp homology, and removing a long overhang with other sequence homology or c) repairing all mismatched positions selectively to obtain a sequence conformable to that of T3 before the 8 bp homology is likely decreased.
Pseudo-palindromes and Endo I function in recombination
The in vitro patch incorporation for T7 DSB repair was found to require gp6 exonuclease and gp2.5, but was independent of Endo I, helicase, T7 DNA polymerase, and E. coli recombinase (RecA) [14] . The study used DSBs produced by Xho I cutting at an engineered site and a synthesized donor fragment with no restriction on the incorporation positions, and therefore the importance of Endo I may not be observed. Nevertheless, the in vitro incorporation of a patch of donor is in agreement with the phage recombination models deduced from the microhomology and four-way junction. When the phage genome serves as a donor, Endo I can generate various phage DNA fragments that then anneal directly or after 5′ resections, as proposed in CESA and CNSA, for repair or recombination. This also alleviates the difficulty of unwinding and degrading the long genome for a requisite patch. Under stressed conditions, phages strive by recombination, which eventually leads to the diversity of phages. Phage genomes were found to avoid palindromes [3] , yet as observed in this study, retention of pseudo-palindromes in phage may serve functions in recombination and repair.
Broadening the host range with some compromise
T7 and T3 manifest low plating efficiency on K91, yet the T3/7 phage raises its EOP to near 1. The acquisition of the T3/7 hybrid phage's ability to infect male K91 indicates that the host range is broadened. Interestingly, the EOP of T3/7 on a female strain DH5α is lower than T3 by 2.5 folds. The size of plague is also smaller than those formed by T3 on DH5α. This illustrates that while the host range is broadened, the efficiency of propagation toward some other strains may be compromised. Sequencing results demonstrate that the T7M phage is closely related to T3L. The deficiency in infecting the male K91 strain does not arise from the absence of T3 gene 1.2 or deterred adsorption. Complete genomic sequencing for T7M and T3/7 is underway to explore the possible explanation.
Applications of Endo I and pseudo-palindromes in genetic and phage manipulation
Our study suggests that phages may have preserved pseudo-palindromes for recombination of genes to survive harsh conditions. It is feasible to implement the same strategy in cloning or targeting a desired DNA sequence for genetic manipulation of cells and isolated DNA via short pseudo-palindromic sequences and Endo I cleavages. Endo I and pseudo-palindromes can also be employed to generate a novel hybrid phage with altered bacterial host range for phage therapy.
Materials and Methods
Materials
Restriction enzymes were purchased from New England Biolabs (Berverly, MA), Promega (Madison, WI) and Fermentas (Hanover, MD). Chemicals were from Sigma Chemical Co. (St. Louis, MO). T7M was obtained from ATCC (11303-B38). E. coli strain K91 and phage T3/7 were generously provided by Dr. M. Russel (Rockefeller University) [12] , [46] . Dr. S. R. Kushner (University of Georgia) kindly supplied E. coli SK3967 [47] . T3 and T7 were generous gifts of Dr. F.W. Studier (Brookhaven National Laboratory) [6] . Genotypes of strains used are listed in Table S1 .
Propagation of phages
The titers of T3/7 and T7M were determined on E. coli strain BL21. Propagation of phages follows the procedure described previously [48] . Briefly, E. coli was grown in T-medium until an OD 600 of 0.5, and infected with T3/7 or T7M that had been serially diluted. The infected cells were plated on agar plates, and incubated at 37°C overnight. The plaque numbers were counted. Efficiency of plating (EOP) was determined with respect to BL21.
Efficiency of Adsorption
The measurements of adsorption of bacteriophage to E. coli follow the method of Koike and Iida [49] with some modification. Bacterial culture was grown in T medium until OD 600 of approximately 0.5–0.6, and mixed with a phage suspension of a titer of about 3×10 9 ml to yield an MOI of 0.5. After 10 min incubation at room temperature, 0.1 ml of the mixture was diluted with 0.9 ml of saline containing 3.3% (v/v) chloroform, and the number of unadsorbed phage was determined by counting the plaque forming units (PFU). The adsorption efficiency was calculated from [1−(PFU of free phage after adsorption/original PFU in the bacteria-phage mixture)]×100%.
Purification of phage DNA
E. coli strain SK3967 carrying thioredoxin gene on a plasmid, pET/trxA, was grown to OD 600 of 0.5 to 0.6, infected with a phage, and then cultured until the cells were lysed (∼2.5 h). DNase I was added to a final concentration of 0.2 µg/ml. After incubation at 37°C for 15 min, NaCl was added to 2.5% (w/v). The cultures were centrifuged at 12000× g, 4°C for 15 min, and PEG8000 was dissolved into the supernatant to a final 10% (w/v). The solution was kept at 4°C for 8–12 hr. The phage was spun down by centrifugation at 12000× g, 4°C for 15 min, resuspended in 1.5 ml TE buffer (10 mM Tris, 1 mM EDTA, pH 8.0) with 1 M NaCl, and centrifuged again at 11500× g in room temperature for 10 min. A CsCl gradient was prepared by mixing 62.5% (w/v) CsCl and TE buffer in a ratio of 1∶0, 2∶1, 1∶1, and 1∶2, with a volume of 0.5 ml, 1 ml, 1 ml, and 1 ml, respectively. The supernatant was loaded onto the gradient, and centrifuged at 210000× g, 4°C for 2 hr. The phage was removed and dialyzed in 1 liter of 0.1 M Tris, 0.1 M NaCl, pH 8.0. DNA was extracted once by 90% phenol and twice by chloroform/isopropanol (24∶1), and then dialyzed in TE buffer overnight.
DNA sequencing
Fragments of phage DNA were amplified by PCR using a mix of Taq and Pfu DNA polymerase (Protech Technology Enterprise, Taiwan) and oligonucleotides upstream and downstream of the region to be sequenced as primers. Sequencing was performed by a Perkin-Elmer 377 DNA autosequencer or an ABI 3730XL autosequencer using an ABI prism Dye Terminator Cycle Sequencing Ready Reaction Kit (PE Applied Biosystems, Forter City, CA). The sequences of oligonucleotides used for sequencing are listed in Table S2 .
Sequence analysis
Sequence alignment was performed using ClustalW . Repeated sequences were identified via the EMBOSS PALINDROME software [50] and manual inspection.
Accession numbers
The sequence data of phages T7M and T3/7 were deposited to GenBank with accession numbers JF906059 and JF906060, respectively. We refer to accession numbers NC_001604.1 and AJ318471.1 for the sequences of T7 and T3, respectively.
Supporting Information
Figure S1
Stem-loop structures of phage terminator Tφ in T3, T3/7, and T7. The drawings show the structures in nt 22352–22390 for T3 (A) and T3/7 (B). From the sequenced T3/7 nucleotides in this region, nt 22352–22381, it can be inferred that the G to A replacement at nt 22374 of T3/7 increases the size of the loop of Tφ while reducing one base pair at the top of the stem. The structure of T7 terminator Tφ is shown in (C) for comparison.
(TIF)
Table S1
Bacterial strains.
(DOC)
Table S2
Sequences of primers. Primers 1–15 were used to sequence T7M DNA. Primers 5–18 were used to sequence T3/7 DNA. Gene 5 from both phages were cloned to a T vector by primers 19–20, and sequenced by primers 21–25.
(DOC)
Table S3
Restriction sites and fragment sizes of T3 and T7 DNA. The positions of restriction sites and sizes of fragments generated by restriction endonucleases on T3 and T7 DNA based on published sequences.
(DOC)
Table S4
Sequenced positions of phage T7M. The numbering of nucleotides (2 nd column), as well as the genes, promoters (φ), terminators (T), and RNase III sites (R) (3 rd column), follows that of phage T3L. If a gene is not completely sequenced, the number of nucleotides sequenced from 5′, 3′ or the middle (mid) of the gene is indicated inside the parenthesis. In all sequenced positions, only nt 22151 and nt 22169 are different from T3. Both incur a T→C change in gene 10B .
(DOC)
Table S5
Sequenced positions of phage T3/7 and changes relative to T3L. The numberings of nucleotides (2 nd column), genes, promoters (A, φ), terminators (T), and RNase III sites (R) follow those of phage T3L, except that due to the shorter exchanged T7 region, the numbering of T3/7 in column 2 is reduced by 13 compared to T3. For the regions changed to T7, the nucleotide numbering in the T7 genome are indicated in the fourth column, while the multiple base changes are not listed. In the fifth column, a slash indicates the mutation location between the two identities.
(DOC)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Prion diseases or transmissible spongiform encephalopathies are rare, late-onset, rapidly progressing and fatal infectious neurodegenerative disorders [ 1 ]. They include sporadic, genetic, variant or iatrogenic Creutzfeldt-Jakob disease (CJD) in humans, ovine scrapie, bovine spongiform encephalopathy, and chronic wasting disease of deer and elk [ 1 ]. The etiologic agent is termed prion and consists of the scrapie prion protein (PrP Sc ), a misfolded, aggregated, often partially protease-resistant conformer of the cellular prion protein (PrP C ) [ 2 ]. PrP Sc is postulated to exist in different conformations, corresponding to different prion strains, which differ in their biochemical properties, and cause different forms of the disease in susceptible host species [ 3 ]. Misfolding of PrP C into PrP Sc and accumulation of PrP Sc aggregates within the central nervous system (CNS) result in progressive neuronal loss and pathognomonic spongiform vacuolation, accompanied by intense astro- and microgliosis [ 4 ].
Neuroimaging and neuro-electrophysiological tests, analysis of 14-3-3 protein and other surrogate biomarkers of CNS damage in the cerebrospinal fluid (CSF) and, more recently, the demonstration of PrP Sc seeding activity in CSF or nasal brushings, can support the diagnosis of prion disease in the presence of a consistent clinical picture[ 5 ]. However, a definitive diagnosis requires the demonstration of PrP Sc in brain tissue, most often performed post-mortem with immunohistochemical-based techniques [ 5 ].
Experimental transmission, as well as acquired and iatrogenic cases of prion disease after known exposure to prions, indicate that prion infection undergoes a long incubation before its clinical onset [ 1 ]. Identification of patients with prion disease during the pre-clinical stage could be instrumental in minimizing the risk of iatrogenic diseases (for example by deferral of blood donors), and is a precondition to early intervention before the establishment of more advanced neurodegeneration once effective therapies will become available. Disappointingly, the development of validated diagnostic tests to identify pre-symptomatic patients with prion disease still remains an unmet medical need.
In an effort to identify novel biomarkers of prion diseases, we performed a global, microarray-based transcriptomic analysis in brains of pre-symptomatic, prion-infected mice [ 6 ]. This led to the identification of the serine/cysteine protease inhibitor Serpina3n as an early upregulated transcript during prion disease, and of urinary levels of its human homologue, α1-antichimotrypsin as a surrogate biomarker of prion infection [ 6 ]. In the present study, we have re-analyzed this microarray dataset using alternative computational methods to uncover additional candidate biomarkers.
Materials and methods
Ethical statement
Animal care and experimental protocols were in accordance with the “Swiss Ethical Principles and Guidelines for Experiments on Animals” and with the Swiss Animal Protection Law, and approved by the Veterinary office of the Canton Zurich (permits 85/2003, 171/2004, 30/2005, 41/2012 and 90/2013). All efforts were made to minimise animal discomfort and suffering.
All human tissue samples used in this project were irreversibly anonymized and collected before the year 2005. Approval by an Institutional Review Board is not required for the use of irreversibly anonymized samples collected before the approval of the Swiss Medical-ethical guidelines and recommendations (Senate of the Swiss Academy of Medical Sciences, Basel, Switzerland, May 23 rd 2006). The study using human CSF was approved by the Swiss Ethics Committees of Canton Zurich (KEK-ZH-Nr. 2012–0376, Amendment dt. 05.07.2014).
Patient populations
Cohort 1
This cohort comprised CSF samples from cases with different neuroinfectious, neuroinflammatory and neurodegenerative conditions. Leftovers from diagnostic CSF samples were obtained from the biobank of the Department of Neurology, University Hospital Zurich. Clinical data concerning diagnosis, gender, date of birth/death and lumbar puncture (LP) were collected from medical charts and autopsy reports obtained from the hospital’s databases PathoPRO and KISIM.
After the performance of desired diagnostic tests, CSF samples were spun down at 3000 rpm for 10 min, aliquoted and since then stored at -80° C.
Cohort 2
This cohort comprised CSF samples from cases with autopsy-proven, definitive Creutzfeldt-Jakob disease or Alzheimer’s disease, as well as cases with clinically diagnosed Alzheimer’s disease and from non-demented, control subjects.
Leftovers from diagnostic CSF samples from subjects with autopsy-proven, definitive Creutzfeldt-Jakob disease (Def CJD, n = 39) or Alzheimer’s disease (Def AD, n = 11) were obtained from the CSF biobank of the Swiss Referral Center for Prion Diseases (Nationales Referenzzentrum für Prionenerkrankungen, NRPE, located at the Institute of Neuropathology, University Hospital Zurich). The NRPE is in charge of the registration and surveillance of human transmissible spongiform encephalopathies in Switzerland. It provides, inter alia , p14-3-3 detection in CSF and autopsies of possible or probable CJD patients from all over the country. CSF samples from the NRPE-CSF biobank were sent ideally on dry ice from corresponding hospitals for p14-3-3 detection, temporarily kept at -20°C awaiting Western blot analysis and were finally stored at -80° C.
Definitive diagnosis of Creutzfeldt-Jakob disease was made by a board certified neuropathologist, according to current guidelines after histopathological examination and the detection of proteinase K resistant PrP sc in brain homogenates by Western blot analysis. Neuropathologic assessment of Alzheimer type brain pathology was made according to the NIA/AA criteria [ 7 ]. For all NRPE cases, date of CSF draw, date of death and autopsy, histopathological report and p14-3-3 results were available. Samples obtained from consecutive cases evaluated between 2004 and 2014 were included in this cohort. Cases with insufficient CSF leftover or with macroscopically haemorrhagic CSF (14 DEF CJD and 2 Def AD) were excluded from the study.
Leftovers from diagnostic CSF samples from subjects with clinically diagnosed Alzheimer’s disease (Clin AD, n = 26) and from non-demented, control subjects (Controls, n = 24) were obtained from the biobank of the Department of Neurology, University Hospital Zurich.
The clinical history of Alzheimer’s disease cases was extensively assessed and diagnosed according to strictly applied diagnostic criteria. Patients underwent a cranial MRI scan (all), neuropsychological evaluation (all), EEG (n = 22), Mini Mental Status (n = 24), CSF testing (all, n = 11 with neurodegenerative markers such as p-tau and amyloid β). Twenty-five patients had at least one additional follow up evaluation at the Clinic of Neurology after CSF withdrawal, with the clinical picture being still compatible with a diagnosis of AD. In three cases the clinical history was compatible with a diagnosis of mixed dementia caused by the coexistence of neurodegenerative and vascular changes. The group of clin AD was matched to the def AD and def CJD groups in terms of patients’ age and gender and duration of sample storage.
The non-demented, control group (controls, n = 24) was composed of patients who underwent clinical examination including lumbar puncture during the work-up of one of the following symptoms: headache, vertigo, sleep disorder. Fourteen patients underwent a cranial MRI showing no pathological changes. All patients received at least two clinical evaluations at different time points and no other neurological complaints including signs of cognitive impairment or neurodegeneration were reported. The control group was matched to the others based on the gender distribution and on the duration of sample storage.
Cohort 3
Archival brain tissue from patients with autopsy-proven, definitive Creutzfeldt-Jakob disease (n = 11) or Alzheimer’s disease (n = 8) were obtained from the autopsy biobank of the NRPE at the Institute of Neuropathology, University Hospital Zurich.
Mice
C57BL/6J male mice were purchased from Charles River. Mice were kept in a conventional hygienic-grade facility, housed in groups of 3–5 in type IIL cages under a 12 h light/dark cycle (light from 7 am to 7 pm) at 22±1°C, with unrestricted access to sterilized food (Kliba No. 3340, Provimi Kliba, Kaiseraugst, Switzerland) and water.
Prion inoculation
Mice were anesthetized with isofluorane and injected in the right hemisphere with 30 μl of 0.1% of RML6 (passage 6 of Rocky Mountain Laboratory strain mouse-adapted scrapie prions) or of 0.1% of non-infectious brain homogenate (NBH) from CD-1 mice as control [ 8 ]. Mice were monitored three times per week in the absence of any clinical sign of disease and every day after clinical onset. Clinical assessment and scoring were performed as previously described [ 9 ] with minor modifications ( S1 Table ). Actions were taken to minimize animal distress and suffering ( S1 Table ). Prion-infected mice were sacrificed at 27, 56, 82, 110, 123 and 137 days post-inoculation (dpi) or when they reached the terminal stage of prion disease at ca. 176 dpi. One additional group of control mice injected with NBH was sacrificed one week after the last prion-injected mouse reached the terminal stage, at 192 dpi. Euthanasia was performed through transcardial perfusion with PBS after deep anesthesia with ketamine and xylazine ( N -(2,6-Dimethylphenyl)-5,6-dihydro-4 H -1,3-thiazin-2-amine). Brain areas were dissected, snap frozen and kept at -80°C until further processing. Archival murine CNS samples from mice inoculated with RML and 22L rodent-adapted scrapie prions and control NBH-injected mice from previously published studies [ 6 , 8 , 10 ] were analyzed.
Experimental autoimmune encephalitis
Experimental autoimmune encephalitis was induced in 13–16 week old C57BL/6 mice through subcutaneous administration of 200 μg of MOG 35-55 peptide (MEVGWYRSPFSRVVHLYRNGK; Neosystem, Strasbourg, France) emulsified in complete Freund’s adjuvant (CFA) supplemented with 4 mg/ml Mycobacterium tuberculosis (DIFCO, Detroit, MI, USA), as previously described [ 11 ]. Mice received intraperitoneal injections with 200 ng pertussis toxin (Sigma, Switzerland) at the time of immunization and 48 hours later. Control mice received an identical regimen with the exception of MOG 35-55 peptide. Clinical assessment and scoring were performed as previously described [ 11 ] and actions were taken to minimize animal distress and suffering ( S2 Table ). Mice with experimental autoimmune encephalitis and control mice were sacrificed in parallel.
Genetically modified mice used in this study are listed in Table 1 .
10.1371/journal.pone.0171923.t001
Table 1 Genetically modified mice used in this study.
Line (abbreviation)
Genetic modification
Description of mice used
Ref.
Prnp ΔF (ΔF)
Transgenic mice expressing a truncated PrP C (32–134) on a B6129- Prnp ZrchI/ZrchI background
Adult mice with signs of neurodegeneration
[ 31 ]
B6-Tg/Thy1APP23Sdz (APP23)
Transgenic mice overexpressing the human APP with the Swedish double mutation at 670/671 (KM→NL)
1.5 year-old mice with histologic evidence of astrogliosis, microgliosis and amyloid β plaque
[ 32 ]
C57BL/6J- Prnp ZH3/ZH3 (ZH3)
Co-isogenic C57BL/6J mice lacking the cellular prion protein.
Adult mice
[ 18 ]
Reverse transcription and polymerase chain reaction
Total RNA was isolated using Trizol (Invitrogen AG, Switzerland) or the RNeasy Universal Plus Mini kit (Qiagen). Prior to cDNA synthesis, residual genomic DNA was removed by the DNA free-kit (Ambion, USA) or using the gDNA Wipeout buffer (Qiagen). cDNA was synthesized using the bulk first strand cDNA synthesis kit and Not I-(dT) 18 as primer (Amersham Biosciences Europe GmbH, Germany) or with Quantiscript Reverse Transcriptase (Qiagen). Control reactions omitting the reverse transcriptase enzyme were performed to verify successful removal of genomic DNA. All these procedures were performed according to the manufacturers’ protocols. Extracted RNA was analysed using a ND-1000 Spectrophotometer (NanoDrop).
Primer pairs were designed using Primer3 software ( http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi ) or using NCBI primer blast ( http://www.ncbi.nlm.nih.gov/tools/primer-blast/ ) or derived from [ 12 ].
For amplification we used the SYBR Green PCR Master Mix (Qiagen) or the Fast Start SYBR Green Master Mix (Roche) with 0.3–1 uM of each forward and reverse primers for the target of interest and cDNA (or no reverse transcriptase sample or water, as controls) as template.
The reaction was run on an ABI PRISM 7700 Sequence Detection System (Applied Biosystems) or on a ViiA 7 Real-Time PCR System (Life Technology) using default cycle conditions followed by a melting curve analysis. Samples were tested in technical triplicates. Expression of a subset of transcripts identified by microarray as upregulated during prion disease (which also included Cst7 ) was measured in mouse brain and normalized to Actb levels. Expression of Cst7 in different mouse organs was normalized by geNorm [ 13 ] to the levels of the following normalization genes: Gapdh , Eif2a , Utp6c , Hprt1 .
Northern blotting
Total RNA was isolated using Trizol (Invitrogen AG, Switzerland) according to standard procedures. Northern blotting was performed exactly as previously described[ 6 ]. Northern blotting for Cst7 was performed in parallel to Northern blotting for Serpina3n , Gfap and Rn18s as reported in [ 6 ] and hence blots for Serpina3n , Gfap and Rn18s are reproduced from [ 6 ].
Murine cDNAs for derivation of Cst7 probe was obtained from Open Biosystems ( www.openbiosystems.com ), accession number AA089317.
Uniform adjustment of contrast and levels on the acquired images and subsequent cropping were performed using Adobe Photoshop. Unprocessed Northern and Western blot images are provided in S1 File .
RNA sequencing
Data on Cst7 mRNA expression levels in different cell types of the mouse brain were retrieved from the RNA-sequencing-based transcriptome database for mouse brain cortex cells [ 14 ]. Data on CST7 mRNA expression levels in different cell types of the human brain were retrieved from the RNA-sequencing-based transcriptome database for human brain cortex cells [ 15 ]. Data on Cst7 mRNA expression levels during microglia development and upon LPS stimulation were retrieved from an RNA-sequencing-based transcriptome database for mouse microglia cells [ 16 ]. RNA sequencing on hippocampi and cerebella of prion-inoculated and control mice was performed essentially as described previously [ 17 , 18 ].
Antibody generation
Mouse monoclonal antibodies F010 and C067 against cystatin F were generated using mammalian cell display as previously described [ 19 ] with modifications. BALB/c mice were weekly immunized with his-tagged human recombinant cystatin F conjugated to Qβ virus-like particle (VPLs) [ 19 ] in alum four times. Splenocytes from immunized mice were harvested and stained with: FITC-labelled anti-mouse IgM/D, CD3, CD11c/b, CD4 and Gr-1; Qβ-bound human cystatin F, anti-Qβ serum (rabbit) and PE labelled anti-rabbit IgG; PE-TexasRed labelled anti-mouse CD19; Alexa 647 nm-labelled Qβ-VPLs (Cytos Biotechnology). Cystatin F-specific B cells (cystatin F/CD19 positive, Qβ-VPLs/IgM-/IgD-/CD3-/CD11c/b-/CD4-/Gr-1-negative) were sorted on a FACS Vantage SE flow cytometer (Becton Dickinson). Total RNA was isolated from Cystatin F-specific B cells and used to generate a Sindbis virus-based scFv library [ 19 , 20 ]. BHK cells were infected with the Sindbis virus-based single chain fragment variable (scFv) library and stained with Qβ-conjugated human cystatin F, anti-Qβ serum (rabbit) and PE labelled anti-rabbit IgG and a rabbit anti-sindbis virus serum and Cy5-labelled anti-rabbit IgG (Cytos Biotechnology), in the presence of propidium iodide. Infected BHK cells displaying cystatin F-specific scFv were individually sorted on 24-well plates containing BHK feeder cells. At least 48 h after sorting/plating, cells from wells with signs of infection were tested by flow cytometry for expression of cystatin F-binding scFv to identify viral clones encoding cystatin F-specific scFv. Fusion single chain fragment variable/fragment crystallizable (scFv-Fc) proteins were generated as previously described [ 19 ] and tested for binding to human recombinant cystatin F (Cytos Biotechnology) in ELISA assays using standard procedures. For expression as mouse IgG2aκ monoclonal antibodies, the heavy chain variable region (HCVR) and light chain variable region (LCVR) coding segments of the cystatin F-binding scFv molecules were PCR amplified using the following primer pairs (5’→3’, Microsynth): HCVR forward: CGA GGT GCA GCT GCT CGA GTC TGG GGC TGA GCT ; HVCR reverse; GAC AGA TGG GCC CGT TGT TTT GGC TGA GGA GAC ; LCVR forward: GAT ATT GAG CTC ACC CAG TCT CAA AAA TTC ATG; LCVR reverse: GCC ACC AGA GGA TTT GAT ATC CAG CTT GGT CCC.
Heavy chain and light chain encoding regions were cloned into a Epstein Barr Virus-based, pCB15 episomal expression vector allowing co-expression of immunoglobulin heavy and light chains under the cytomegalovirus promoter [ 19 ]. Expression vectors were transfected into HEK293T cells using Lipofectamin Plus (Invitrogen) and antibodies were purified from cell supernatants using protein A-Sepharose columns (GE healthcare).
Tissue processing for Western blotting and ELISA
Human brain tissues from the NRPE-autopsy bio-bank were snap frozen and stored at -80°C. Ten-twenty percent brain homogenates (w/vol) were prepared in 2% Sacrosyl (Sigma) and PBS, using lysing matrix tubes (MP Bio) and a Precellys 24 (Bertin Technologies). Samples were spun down to remove gross debris and supernatants were stored at -80° C until further analysis.
Mice were deeply anesthetized with ketamine‐xylazine, transcardially perfused with a solution of PBS containing heparin and dissected organs were snap frozen and kept at -80°C until further analysis. Ten percent (w/vol) homogenates were prepared in either RIPA buffer (25mM Tris-HCl pH 7.6, 150mM NaCl, 1% NP-40, 1% sodium deoxycholate, 0.1% SDS) + Complete Mini Protease Inhibitor Cocktail (Roche) or in PBS + sucrose 0.32 M (Sigma), using Stainless Steel Beads 5 mm (Qiagen) and Tissue Lyser LT (Qiagen). Samples were spun down to remove gross debris and supernatants were stored at -80° C until further analysis.
ELISA
Sandwich ELISA was performed on 96-well ELISA plates (F96 Cert. Maxisorp Nunc-immuno plate) coated with 5 μg/mL of the monoclonal mouse anti-cystatin F antibody F010 in PBS (pH = 7.2, Gibco). After overnight incubation, plates were washed 5 times with PBS supplemented with 0.1% (vol/vol) Tween-20 (PBST) using a 96-well plate washer (BioTek Instruments, ELx405). Nonspecific binding sites were subsequently blocked with 5% (w/vol) fat-free milk in PBST for 2h at room temperature. Recombinant cystatin F (AJ Roboscreen) was subjected to 1:2 serial dilution in sample diluent buffer (1% w/vol fat-free milk in PBST) to generate a standard curve with the following concentrations: 1000, 500, 250, 125, 62.5, 31.25, 15.6, 7.8 pg/ml. Blocking buffer was removed and standards/samples were loaded in either duplicates (undiluted CSF) or triplicates (cell lysates and tissue homogenates, diluted in sample diluent buffer). Plates were incubated for 2h at 37°C and washed 5 times. For detection of the captured antigen, plates were incubated with 0.75 μg/ml of the biotinylated monoclonal mouse anti-cystatin F antibody C067 for 1h at 37°C, then washed 5 times. Avidin-horseradish-peroxidase HRP (BD Pharmingen) diluted 1:1000 in sample diluent was loaded. After incubation of 45 min at 37°C plates were washed 5 times and developed with 3,3’,5,5’-tetramethylbenzidine substrate solution (high sensitivity TMB, BioLegend), for 10 min at room temperature in the dark. The reaction was stopped by adding stop solution, 0.18 M H 2 SO 4 . Plates were read at wavelength of 450 nm on a Versamax SNB plate reader (Molecular Devices).
Standard curves were generated using a four-parameter logistic (4PL) equation to calculate concentrations of unknown samples. Values of all 4 parameters were determined by SOFTmax PRO software (Molecular Devices).
To assess matrix effects in the assay, recombinant human cystatin F (Cytos Biotechnology) was spiked in pooled human CSF with the concentrations 1000, 500, 250, 125, 62.5, 31.25, 15.6 pg/ml and compared to the observed values with the expected protein levels diluted only in sample diluent buffer. We obtained measured levels differing up to 22% with respect to nominal values in the range between 62.5 and 1000 pg/mL. Below 62.5 the discrepancy was larger, possibly due to the presence of endogenous cystatin F in pooled human CSF used for spiking. Macroscopically haemorrhagic CSF samples were excluded, as preliminary results indicated extensively elevated cystatin F concentrations due to blood contamination.
Intra-assay variability was evaluated by spiking recombinant cystatin F into pooled human CSF at the following concentrations: 500, 62.5, 7.8 pg/ml. Validation samples were measured in 10 replicates. The coefficient of variation (%CV) was 2.5%, 6.2% and 10.5%, respectively.
To account for possible inter-assay variability, for measurements in mouse brain homogenates from different experiments, cystatin F levels are normalized to levels in NBH-injected control mice, whereas for measurements in human CSF samples 3 pre-defined CSF cases were employed as internal calibrators to pool data for the evaluation of cystatin F levels in CSF samples from cohort 2.
Western blotting
Total protein concentration of tissue homogenates was measured using the BCA Protein Assay (Pierce), according to the manufacturer’s instructions. For proteinase K (PK) digestion, PK (100 μg/ml, pH = 7, Roche) was added aiming for a final concentration of 25 μg/ml and incubated at 37°C for 30 min; the reaction was stopped by adding NuPAGE 4xLDS sample buffer (Invitrogen) with β-mercaptoethanol (Sigma-Aldrich) and boiling at 95°C for 10 min. Separation was performed on a NuPAGE 12% Bis-Tris gel (Invitrogen) and transferred onto a Protran Nitrocellulose Transfer Membrane (Whatman) using the NuPAGE Gel Electrophoresis System (Invitrogen) or the Mini Trans-Blot cell System (Bio-Rad). After blotting, the membrane was washed once with PBS supplemented with 0.1% (vol/vol) Tween-20 (PBST), blocked with 5% (w/vol) fat-free milk in PBST and decorated with mouse anti-PrP POM1 [ 21 ] (300 ng/ml in 1% (w/vol) fat-free milk in PBST) overnight at 4°C. After washing, the membrane was probed with rabbit anti-mouse IgG2a antibody (1:10’000 dilution, Zymed). Blots were developed using Luminata Western HRP Substrates (Millipore) and visualized using either a Stella (Raytest) or LAS-3000 (Fujifilm) luminescent image analyzer. After initial image acquisition, membranes were washed, reprobed with mouse anti-actin monoclonal antibody (1:10’000 dilution, Millipore) and the same procedure was followed for acquisition of the actin signal. Uniform adjustment of contrast and levels on the acquired images and subsequent cropping were performed using Adobe Photoshop. Unprocessed Western blot images are provided in S1 File .
Immunohistochemistry
Tissues were fixed in formalin, treated with concentrated formic acid to inactivate prions and embedded in paraffin. Tissue sections were subjected to deparaffinization through graded alcohols and heat-induced antigen retrieval in EDTA-based buffers. Stainings were performed on a NEXES immunohistochemistry robot (Ventana instruments) or on a BOND-III robot (Leica) using the following antibodies: 1:200 dilution of 3F4 (stock concentration 2 mg/ml, purified in house) for PrP staining; 1:3000 dilution of 4G8 (Signet) for Aβ staining; 1:1000 dilution of AT8 (Thermo Fisher Scientific) for staining of phosphorylated tau; 1:200 dilution of HPA040442 (Sigma) for cystatin F; 1:1000 dilution of rabbit anti-Iba1 (Wako). Immunoreactivity was visualized using an IVIEW DAB Detection Kit (Ventana) or Bond Polymer Refine Detection Kit (Leica). Haematoxylin and eosin staining was performed according to a standard protocol. For image analysis and acquisition, slides were scanned with NanoZoomer and images were obtained using the NanoZoomer Digital Pathology System (NDPview, Hamamatsu Photonics).
Statistical analysis
Microarray data from a previous study [ 6 ] were re-analyzed using the perfect match-mismatch model-based expression analysis algorithm ( www.dchip.org ). Differentially expressed genes were defined as having an absolute fold change ≥1.3, a confidence interval >90% and Student’s t test p value <0.05 when comparing expression values between RML-infected vs. NBH-injected control mice.
Statistical analysis was performed using Graphpad Prism software and SPSS. Significance between two groups was determined by unpaired Student’s t -test. Comparison between multiple groups was assessed by One-way ANOVA followed by Bonferroni’s or Dunnet’s Multiple Comparison Test or Fisher’s least significant difference (LSD) post-hoc test, as stated in corresponding figure legends. P-values <0.05 were considered as statistically significant. For the statistical analysis, a normalizing log-transformation of murine Cst7 mRNA levels and of mouse and human cystatin F values was applied. A multi-factorial analysis of co-variance (ANCOVA) of the log-transformed analysis variable cystatin F was performed on the data from Cohort 2 using SPSS. Categorical ANCOVA factors were diagnosis group and sex; continuous covariates were age of patient, age of CSF sample and total protein concentration. An additional term was included in the ANCOVA model to account for a possible interaction between sex and diagnosis.
Results
Microarray analysis of prion-infected mouse brains unveils Cst7 as a candidate biomarker
We had previously performed a genome-wide, microarray-based transcriptomic analysis on brains from C57BL/6J mice at 145 days after intraperitoneal injection with RML prions (or non-infectious brain homogenate, NBH, as control) [ 6 ]. Under this experimental paradigm, the onset of clinical signs occurs at around 180 dpi, whereas the terminal stage is achieved at around 200 dpi in prion-inoculated mice. Microarray data were initially analyzed using the perfect-match-only model-based expression analysis algorithm, which relies on hybridization intensities of the perfect match probes, 25mers containing the canonical sequence for the corresponding transcripts. This approach allowed the identification of 77 differentially expressed transcripts, including Serpina3n [ 6 ].
To identify additional transcripts differentially expressed in the early, pre-clinical phases of experimental prion disease, we have re-analyzed this dataset. We have taken advantage of the presence, within the employed microarray chip, of the mismatch probes. These are 25mer probes containing a deliberate mutation at position 13 and as such designed to measure non-specific or cross-hybridization [ 22 , 23 ]. We have therefore applied the perfect-match mismatch model-based expression analysis algorithm and focused on differentially expressed genes with absolute fold change ≥1.3, a confidence interval >90% and Student’s t test p value <0.05. By applying these filters, we identified 278 differentially expressed probes, corresponding to 259 genes, of which 201 genes were upregulated and 58 genes were downregulated in the brain of prion-infected vs. control mice ( Fig 1A , S4 Table ). Of note, the majority (≈78%) of these differentially expressed probes were not previously identified when using the perfect-match only algorithm. Remarkably, among the newly identified differentially expressed genes was Cst7 , which represented the gene undergoing the strongest modulation in brains of prion-infected mice, with an upregulation of ≈8.7 fold.
10.1371/journal.pone.0171923.g001
Fig 1
Microarray analysis identifies Cst7 as the most upregulated transcript in RML inoculated mice.
A Heatmap showing relative expression of 251 differentially expressed genes (absolute fold change ≥1.3, a confidence interval >90% and Student’s t test p value <0.05) between RML prion-inoculated mice (RML, n = 3, A to C) and control mice injected with non-infectious brain homogenates (NBH, n = 3, A to C). For each gene, relative expression is defined as the ratio between the actual gene expression of each mouse and the average gene expression from all 6 analyzed mice (the latter set as 1, black) and is reported as grades of blue (downregulation) or red (upregulation). Expression data were obtained through the employment of the perfect match mismatch model-based expression analysis algorithm of microarray data previously reported in [ 6 ]. The list of differentially expressed genes is reported in S4 Table . B Validation of the upregulation of 29 transcripts (including Cst7 ) in RML-inoculated mice (n = 3) compared to NBH-injected controls (n = 3). Bars (left y axis) depict mean and standard deviation of transcript fold change expression between brains of RML-inoculated mice and NBH-injected controls (after normalization to Actb levels) at 145 and 190 days post-inoculation (dpi), as assessed by RT-PCR. Red diamonds (right y axis) depict transcript fold change expression between brains of RML-inoculated mice and NBH-injected controls at 145 dpi as assessed by microarray.
Cst7 encodes for cystatin F, a secreted type-II cysteine protease inhibitor mainly expressed in immune cells [ 24 – 28 ]. The early and strong upregulation of Cst7 mRNA during prion disease, along with the fact that cystatin F can be secreted into the extra-cellular space, prompted us to investigate the role of cystatin F as a possible surrogate biomarker in prion diseases.
Levels of Cst7 transcripts during prion pathogenesis
To validate the upregulation of Cst7 transcripts during prion disease identified with the microarray, we performed quantitative RT-PCR analysis in brains of prion-infected versus control mice (the latter injected with NBH). We analyzed samples at 145 dpi, the same pre-symptomatic stage used for microarray analysis [ 6 ], and at 190 dpi, when prion-inoculated mice showed signs of disease and were close to the terminal stage. We measured levels of Cst7 , as well as levels of 28 additional transcripts which were found, to different extent, as upregulated in prion-infected mice by microarray analysis. Quantitative RT-PCR analysis confirmed the upregulation of all investigated transcripts at both time points ( Fig 1B ). Compared to microarray analysis, RT-PCR showed a higher magnitude of upregulation, which, in the case of Cst7 , was >40 fold in brains of prion-infected versus control mice ( Fig 1A and 1B ).
We then investigated the temporal pattern of Cst7 transcription in the CNS after prion infection, and its relationship to the accumulation of partially protease-K resistant PrP. Northern blotting indicated an initial upregulation of Cst7 transcripts already at 110 dpi ( Fig 2A ), when PrP Sc becomes detectable by Western blotting in whole brains of prion-infected mice ( S1 Fig ). The PrP Sc signal progressively increased over time ( Fig 2A ). Compared to Serpina3n and to Gfap , a marker of astrogliosis typically associated to prion disease, Cst7 mRNA upregulation occurred earlier during prion disease ( Fig 2A ).
10.1371/journal.pone.0171923.g002
Fig 2
Spatial and temporal pattern of Cst7 upregulation in the central nervous system during experimental prion pathogenesis.
A Northern hybridization analysis of Cst7 mRNA in whole brains of mice at different time points (from 100 to 190 days post-inoculation, dpi) after injection with either non-infectious brain homogenate (NBH) or RML prions. For comparison, expression levels of Serpina3n and Gfap mRNA are also shown. Hybridization for Rn18s , encoding 18S rRNA, is performed for normalization. Each lane denotes a pool of brain extracts from 3 mice. B Northern hybridization analysis of Cst7 mRNA in selected areas of the central nervous system (cb: cerebellum; sc: spinal cord; ob: olfactory bulb; cx: cortex) at different time points after injection with either NBH or RML prions. For comparison, expression levels of Serpina3n and Gfap mRNA are also shown. Hybridization for Rn18s , encoding 18S rRNA, is performed for normalization. Each lane denotes a pool of brain extracts from 3 mice. C Cst7 , Serpina3n and Gfap mRNA levels in hippocampus (hp) and cerebellum (cb) of RML-inoculated mice (n = 3) compared to NBH-injected controls (n = 3). Bars depict mean and standard deviation of transcript fold change expression between brains of RML-inoculated mice and NBH-injected controls at different time points post-inoculation (expressed as week, wk), or at the terminal stage (term), as assessed RNA-sequencing. Dashed line indicates equal levels between RML-inoculated mice and NBH-injected controls (fold change of 1). Data concerning Serpina3n , Gfap , Rn18s and PK-resistant PrP from panels A and B reproduced, in modified form [ 6 ].
We next analyzed the course of Cst7 upregulation in selected areas of the CNS. Cst7 transcripts were found to be upregulated in prion-infected spinal cords at 100 dpi ( Fig 2B ). By 140 dpi Cst7 mRNA levels were further increased in spinal cord and Cst7 upregulation became evident also in cerebellum and in cerebral cortex ( Fig 2B ). Again, Cst7 mRNA upregulation occurred earlier with respect to Serpina3n and Gfap mRNA levels ( Fig 2B ).
We then used RNA sequencing to assess the number of Cst7 mRNA transcripts in the hippocampi and cerebella of mice at various time points after intracerebral injection with either RML prions or NBH (as control). In the hippocampus, Cst7 showed a trend toward upregulation at 110 and 123 dpi (≈3 fold and 10 fold, respectively), which reached statistical significance at 137 dpi (≈79 fold, p<0.05, One-way ANOVA, Dunnett’s Multiple Comparison Test) and at the terminal stage of disease (≈168 fold, p<0.001, One-way ANOVA, Dunnett’s Multiple Comparison Test; Fig 2C , S5 Table ). At these two time points, the increase of Cst7 transcripts was significantly larger than that of Serpina3n and Gfap (for both time points p<0.0001, Two-way ANOVA, Bonferroni’s Multiple Comparison Test).
In the cerebellum, Cst7 was significantly upregulated at 110 dpi (≈8 fold, p<0.05, One-way ANOVA, Dunnett’s Multiple Comparison Test), as well as at the terminal stage (≈251 fold, p<0.001, One-way ANOVA, Dunnett’s Multiple Comparison Test) compared to NBH-injected mice ( Fig 2C , S5 Table ). At the terminal stage, Cst7 upregulation was significantly higher than that of Serpina3n and Gfap (p<0.0001, Two-way ANOVA, Bonferroni’s Multiple Comparison Test). In summary, Cst7 mRNA upregulation occurred earlier and was larger than that of Serpina3n and Gfap ( Fig 2C ). Collectively, these data indicate that in C57BL/6 mice intraperitoneally or intracerebrally inoculated with RML prions, early, progressive and robust upregulation of Cst7 mRNA occurs throughout the disease course.
Levels of Cst7 mRNA in other neurologic conditions
To study the cellular pattern of Cst7 expression within the mouse brain, we interrogated the RNA-sequencing-based transcriptome database for mouse brain cortex cells [ 14 ]. This analysis revealed that Cst7 ranked at the 48 th percentile of expression of all mouse genes and that Cst7 transcripts were significantly enriched in microglia/macrophages as compared to other cell types of the brain cortex ( S2 Fig ), in line with the notion that Cst7 /Cystatin F is mainly expressed in immune cells, including microglia [ 29 , 30 ]. Of interest, transcriptomic analysis of different cell types of the human brain [ 15 ] showed that CST7 ranked t the 43 rd percentile of expression of all human genes and that CST7 transcripts are mainly detected in microglia/macrophages and endothelial cells ( S2 Fig ). Next, we analysed Cst7 mRNA levels in mouse microglial cells during development and after an immune challenge with lipopolysaccharides using the RNA sequencing-based transcriptome database for mouse microglia cells[ 16 ]. This showed that Cst7 mRNA levels are stable during microglia maturation, whereas they show a trend toward increase upon exposure to lipopolysaccharides ( S2 Fig ).
In light of these data, we asked whether neurologic conditions with prominent neuroinflammation/microgliosis other than prion disease may result in the upregulation of Cst7 mRNA. Cst7 mRNA overexpression as assessed by Northern blotting was evident to different extents in the brains of terminally sick ΔF mice expressing a toxic, truncated PrP C molecule [ 31 ], in mice with experimental autoimmune encephalitis [ 11 ], a model of multiple sclerosis, and in aged TgAPP23 mice overexpressing human APP Swe/V717I [ 32 ], a model for Alzheimer’s disease (AD) ( S2 Fig ). This analysis showed that mild brain Cst7 overexpression occurs in other mouse models of neurologic conditions with neuroinflammation/microgliosis.
Cystatin F protein levels during experimental prion pathogenesis
We next aimed at studying cystatin F protein levels in mouse models of prion diseases. In naïve C57BL/6 mice, cystatin F levels oin spinal cord, cerebellum and forebrain were significantly lower than in thymus, in line with RT-PCR data of Cst7 mRNA in the same organs ( S3 Fig ).
However, compared to NBH-injected control mice, RML prion-inoculated, terminally-sick mice showed significantly higher levels of brain cystatin F ( Fig 3A and 3B ). We also studied cystatin F levels in brains of C57BL/6 mice at different time points after intracerebral challenge with the 22L inoculum, another strain of rodent-adapted scrapie prions. Compared to NBH-injected control mice, a trend toward increased cystatin F levels was evident in brains of 22L-inoculated mice already at 60 dpi, and at 90 dpi this difference reached statistical significance and was maintained throughout the disease course, until the terminal stage, which was reached at around 150 dpi ( Fig 3C ). Also, immunohistochemical analysis showed an increase in cystatin F protein expression in brains of terminally-sick, 22L-inoculated mice compared with NBH-injected controls, with a diffuse pattern compatible with the fact that the protein can be released in the extracellular space ( Fig 3D ).
10.1371/journal.pone.0171923.g003
Fig 3
Cst7 /cystatin F brain expression during prion pathogenesis.
A Western blotting analysis in brain extracts of terminally sick, prion-infected mice (RML), and of mice injected with non-infectious brain homogenate (NBH), showing the amount of partially protease K (PK)-resistant prion protein (PrP) (left panel). Identical protein extracts omitting proteinase K treatment were used for Western blotting with an anti-PrP and anti-actin antibody to verify equivalent loading in each lane (right panel). ZH3 denotes brain from a C57BL/6J- Prnp ZH3/ZH3 mouse as control. Each lane denotes one mouse. B Cystatin F levels in brain extracts from the same mice depicted in A, relative to levels in NBH-injected mice (set as 100%; ***, p<0.001, Student’s t test). C Cystatin F levels in brain extracts from mice injected with either non-infectious brain homogenate (NBH) or 22L prions at different time points (dpi, days post-inoculation; term, terminal prion disease, reached at approximately 150 pdi in this experiment), relative to levels in NBH-injected mice at 150 dpi (ns, p>0.05; **, p<0.01; ***, p<0.001; ****, p<0.0001, one-way ANOVA followed by Bonferroni’s Multiple Comparison test). D Histological analysis of cystatin F expression in cerebella of 22L-inoculated mice at the terminal stage of the disease and the relative NBH-injected controls. Three representative mice per each group are depicted. Scale bar: 100 μm.
Cystatin F protein levels in human Cerebrospinal Fluid (CSF) samples
To gain a first insight into cystatin F levels in neurologic conditions other than prion diseases, we analyzed a first cohort of patients with several neurodegenerative and neuro-infectious/inflammatory conditions. This cohort included patients with various forms of cognitive impairment (mild cognitive impairment, late-onset Alzheimer’s disease and fronto-temporal dementia n = 4), Parkinson’s disease (n = 2), bacterial and viral meningitis (n = 3 and 8, respectively), limbic encephalitis (n = 3), multiple sclerosis (n = 6) and neuroborreliosis (n = 3). Cystatin F was detected in all CSF samples, with the highest levels observed in samples from the bacterial meningitis groups ( S4 Fig , p<0.01 compared to dementia and to multiple sclerosis, p<0.05 compared to encephalitis, one-way ANOVA followed by Bonferroni’s post-hoc test).
We next explored the role of cystatin F as a potential biomarker for prion diseases. For this purpose, we analyzed CSF samples from patients referred to the NRPE for 14-3-3 analysis because of clinical suspicion of prion disease. Of these cases, we included only cases of autopsy-confirmed Creutzfeldt-Jakob disease or Alzheimer’s disease, whereas non-prion related diagnoses were excluded. Cases evaluated at the Neurology Department of the University of Zurich with available CSF samples and with a definitive autopsy-based diagnosis were also included. This resulted in 11 cases of definitive Alzheimer’s disease (“Def AD”) and 39 cases of definitive CJD (“Def CJD”). In addition, we also included 26 cases of Alzheimer’s disease diagnosed clinically based on stringent criteria (“Clin AD”), which were matched to the Def AD and Def CJD groups by patient age as well as gender, and sample age. We included 24 cases with no neurodegenerative condition (“Controls”), consisting of subjects undergoing lumbar puncture during the diagnostic work-up for headache, vertigo, or sleep disorders. This latter group was matched to the others based on the sample age.
No significant differences were observed in patient’s age and gender distribution as well as CSF sample storage time among Def CJD, Def AD and, as a result of the matching process, Clin AD ( Fig 4A and 4B ). Instead, subjects from the control group were significantly younger compared to the other groups but also in this case CSF sample storage time was similar thanks to the matching process ( Fig 4A and 4B ). For cases for which an autopsy was performed (namely Def AD and Def CJD groups), the time between lumbar puncture and the exitus was significantly shorter for the Def CJD group (p = 0.0051, Student’s t test), in line with the more rapid clinical course [ 33 , 34 ].
10.1371/journal.pone.0171923.g004
Fig 4
Cystatin F levels in cerebrospinal fluid from patients with Creutzfeldt-Jakob diseases, Alzheimer’s disease and controls.
Cystatin F levels were measured in patients with autopsy-confirmed definitive Creutzfeldt-Jakob disease (Def CJD) or Alzheimer’s disease (Def AD), clinically diagnosed Alzheimer’s disease (Clin AD) and non-demented controls (Controls). A Age and gender distribution (****, p<0.0001 against all other categories, one-way ANOVA followed by Bonferroni’s Multiple Comparison test). B Sample storage of archival cerebrospinal fluid (CSF). C Time interval between lumbar puncture (LP) and exitus for deceased patients with autopsy-confirmed diagnosis of neurodegeneration (Def AD and Def CJD; **, p<0.01, Student’s t test). D Cystatin F levels in CSF from all study subjects from cohort 2 (*, p<0.05, one-way ANOVA followed by Bonferroni’s Multiple Comparison test). E Sub-analysis of cystatin F levels in CSF from Def CJD subjects according to the PrP type[ 35 ]. F Sub-analysis of cystatin F levels in CSF from Def CJD subjects according to the results of p14-3-3 Western blotting analysis as performed at the time of CSF sample collection. In all panels: Dots: denote individual subjects; bars: mean; error bars: standard deviation.
The highest cystatin F CSF levels were found in the Def AD group and these values were significantly different compared to levels in the Controls (p<0.05, one-way ANOVA followed by Bonferroni’s Multiple Comparison test). Notably, no significant difference was seen between Def CJD and Controls (p = 1, one-way ANOVA), Def CJD and Def AD (p = 0.054, one-way ANOVA) and Def CJD and Clin AD (p = 0.3, one-way ANOVA). Analysis of covariance was performed on the log-transformed cystatin F levels taking into account possible influences of sex, patient’s age at the time of lumbar puncture and age of CSF sample. There was no association between diagnosis or gender and cystatin F levels (p = 0.239, ANCOVA). However, patient’s age was associated with cystatin F levels (p = 0.007, ANCOVA). Further analysis showed that only for the Def AD group cystatin F levels correlate with patient’s age (r = 0.667, p = 0.025).
Within the Def CJD group, there was no significant difference of cystatin F levels between PrP type 1 and type 2 cases [ 35 ] (p = 0.14, Student’s t test, Fig 4E ), nor between patients with 14-3-3 positive and negative results (p = 0.19, Student’s t test, Fig 4F ).
Collectively, these analyses indicate the lack of a significant increase of cystatin F levels in CSF samples of Creutzfeldt-Jakob disease cases compared to clinically relevant controls.
Cystatin F protein levels in human brain samples
We next aimed at comparing cystatin F levels in the brain parenchyma of Creutzfeldt-Jakob disease vs. Alzheimer’s disease patients. To this purpose, we examined a third cohort comprising 11 and 8 autopsy-confirmed cases, respectively. We focused on frontal cortex, a brain area commonly affected both by amyloid β pathology in Alzheimer’s disease and by partially PK-resistant PrP in Creutzfeldt-Jakob disease, as well as on cerebellum, which is rarely involved by amyloid β pathology in Alzheimer’s disease [ 7 , 36 , 37 ].
Cystatin F levels were not significantly different between patients with Creutzfeldt-Jakob disease and Alzheimer’s disease, both sampled from the frontal cortex and cerebellum ( Fig 5A ), with linear regression analysis indicating a correlation of cystatin F levels in the two regions of the same case ( Fig 5B ). Also, within each diagnostic group, no significant difference was observed between cystatin F levels in the frontal cortex and in the cerebellum ( Fig 5A ). Moreover, immunohistochemical analysis could not detect any cystatin F in the frontal cortex of these patients with Creutzfeldt-Jakob disease and Alzheimer’s disease, despite the presence of microgliosis as detected by Iba1 stainings ( S5 Fig ) and the presence of partially protease-resistant PrP in the former and amyloid β deposits and neurofibrillary tangles in the latter ( Fig 5C ). Conversely, cystatin F was detectable, mainly in infiltrating cells, in cases of encephalitis, as well as in other tissues known to express cystatin F such as bone marrow ( Fig 5C ). Collectively, these data indicate similar levels of cystatin F levels in brain parenchyma of patients with Creutzfeldt-Jakob disease and Alzheimer’s disease.
10.1371/journal.pone.0171923.g005
Fig 5
Cystatin F levels in brains from patients with Creutzfeldt-Jakob disease and Alzheimer’s disease.
A Cystatin F levels in frontal cortex (FCx) and cerebellum (Cb) of patients with autopsy-confirmed definitive Creutzfeldt-Jakob disease (Def CJD) or Alzheimer’s disease (Def AD). Dots: denote individual subjects; bars: mean; error bars: standard deviation. B Correlation of cystatin F levels in the two regions of each patient, denoted by a dot (R 2 0.73 for Def AD and 0.83 for CJD, linear regression analysis). C Histologic analysis of cystatin F expression in frontal cortex of the same cohort of patients. Regions with abundant pathology, including amyloid β (Aβ) plaques and neurofibrillary tangles (Tau, further magnified in the inset) in Alzheimer’s disease and spongiosis and partially protease resistant prion protein (PrP) in Creutzfeldt-Jakob disease, from representative cases are depicted. Non-demented subjects were included as control (Ctrl). Lower row, left: encephalitis, with cystatin F positive infiltrating cells (further magnified in the inset at center). Lower row, right: bone marrow, with cystatin F positive cells (further magnified in the inset). Scale bar: 100 μm (black) or 25 μm (blue).
Discussion and conclusion
A global transcriptomic analysis has identified Cst7 as a highly upregulated transcript in the brain of prion-inoculated mice in the preclinical stage of the disease. Further analyses validated this finding, showing that Cst7 upregulation occurs early during prion pathogenesis, parallels the appearance of partially PK-resistant PrP, and has a magnitude of change that is unmatched by other examined prion-induced transcripts. Cst7 encodes cystatin F, which is mainly expressed by immune cells, with microglia being the main cellular source of Cst7 expression within the brain. Importantly, upregulated Cst7 translated into higher cystatin F protein levels in the brains of prion-inoculated mice. All these findings, together with the observation that cystatin F can be secreted in the extracellular space and is found in biological fluids, formed the rationale for investigating the possible role of cystatin F as a biomarker of prion diseases.
Microglial Cst7 can be induced by lipopolysaccharide challenge, and moderately increased Cst7 levels were observed in different mouse models of neurodegenerative and neuroinflammatory conditions with microgliosis. Therefore, we first analyzed a cohort of CSF samples from subjects with different neuro-infectious, -inflammatory and -degenerative diseases. Significantly increased levels of cystatin F were observed in the CSF of patients diagnosed with bacterial and viral meningitis. In light of the cellular origin of cystatin F, this observation possibly reflects blood-brain-barrier leakage and associated CSF leucocytosis that is often observed in the course of bacterial and viral meningitis [ 38 , 39 ]. The fact that infectious conditions of the central nervous system are accompanied by significantly increased CSF cystatin F levels does not necessarily represent a limitation, as in the vast majority of the cases these conditions are not relevant in the differential diagnosis of CJD.
We next measured cystatin F levels in CSF samples from patients with rapidly progressive dementia and autopsy-confirmed Creutzfeldt-Jakob or Alzheimer’s disease, as well as in patients with clinically diagnosed Alzheimer’s disease and non-demented controls. Unexpectedly, we found no significant increase in CSF levels of cystatin F among Creutzfeldt-Jakob disease cases compared to all other diagnostic groups.
A certain degree of pre-analytical variation, especially in sample withdrawal, processing, storage and shipping, may play a role as a confounding factor [ 40 ], since our work was a retrospective study mainly based on external CSF samples sent to our reference laboratory in the context of the clinical work-up for suspected Creutzfeldt-Jakob disease. On the other hand, we found no correlation between CSF sample age and cystatin F levels, and the measured total protein levels were in line with the different diagnoses (data not shown). This finding essentially excludes the likelihood of substantial protein degradation.
Metabolism, biodistribution and the half-life of cystatin F in different biological fluids, as well as the contribution of blood-borne vs brain-borne cystatin F within the CSF remain largely unknown. Taking these aspects into consideration, we analysed cystatin F levels in autoptic brain tissue to elucidate whether there was any significant upregulation of this protein in the brains of patients with Creutzfeldt-Jakob disease. Remarkably, no significant differences were observed between Creutzfeldt-Jakob disease and Alzheimer’s disease brain tissue, in line with the results in CSF samples.
Alternatively, brain-derived cystatin F may be preferentially retained or degraded within brain parenchyma, without significantly diffusing into the CSF. Cystatin C belongs to the cystatin type II super family–like cystatin F–and has amyloidogenic properties [ 41 ], and the Leu68Gln variant ( CST3 p.Leu94Gln) causes hereditary amyloidosis with cystatin C amyloid deposition within the CNS [ 42 , 43 ]. Further, in vitro and in vivo studies in mice indicate that cystatin C co-aggregates with amyloid β and thus inhibits oligomerization and fibril formation in Alzheimer’s disease models [ 44 , 45 ]. Similarly, cystatin F could be sequestered by PrP Sc aggregates and accumulate in the brain instead of diffusing into the subarachnoid space and hence no elevated concentrations would be detectable in the CSF. To test for this hypothesis, we investigated cystatin F expression and distribution within brain tissue by immunohistochemistry. This analysis failed to detect a significant accumulation of cystatin F in brain areas with prominent PrP Sc aggregates in Creutzfeldt-Jakob disease cases or amyloid β plaques in Alzheimer’s disease cases. In fact, cystatin F was barely detectable in patients with Creutzfeldt-Jakob disease and Alzheimer’s disease, similarly to non-demented controls, whereas cystatin F expression could be detected in inflammatory infiltrating cells in cases of meningitis.
It is also possible that prion-induced upregulation of Cst7 is a prion strain-dependent phenomenon. This would not be unexpected, as prion strains transmit disease with distinctive phenotypes, which include neuropathological and biochemical changes [ 1 , 4 ]. For example, microglia activation, which plays a key neuroprotective role in prion diseases [ 46 ], was shown to vary considerably among different prion models [ 47 ]. However, substantial upregulation of Cst7 transcripts was detected in cultured murine microglia infected with the Creutzfeldt-Jakob disease-derived Fukuoka strain [ 48 ], as well as in brains of mice infected with RML, ME7, 139A and 301V rodent-adapted scrapie prions [ 49 – 52 ]. Furthermore, we observed upregulation of both Cst7 mRNA and cystatin F protein levels in brains of mice infected with RML and 22L prions.
These observations suggest that Cst7 upregulation is remarkably conserved upon exposure to different prion strains, at least in mouse models of prion diseases. Whether Cst7 upregulation occurs also in the context of field cases of animal prion disease has to be established. Of interest are the results of a recent study testing levels of selected transcripts in two CNS areas of a limited number of Creutzfeldt-Jakob disease cases, subtypes MM1 and VV2, and controls. Compared to controls, a trend towards slightly increased CST7 levels was found in the frontal cortex of patients with Creutzfeldt-Jakob disease subtype MM1, and in the cerebellum of patients with Creutzfeldt-Jakob disease subtype VV2, yet none of these changes reached statistical significance [ 53 ]. Conversely, there was a significant upregulation of microglial transcripts CD68 , ITGAM (encoding CD11b) and Aif1 (encoding Iba1) in frontal cortex of MM1 cases and in cerebellum of VV2 cases [ 53 ]. This observation, together with the results of our studies, suggests that the presence of microgliosis is not sufficient to cause a significant increase in cystatin F levels in patients with Creutzfeldt-Jakob disease.
One possible explanation for the divergent observations in experimental and human prion disease may lie in the intrinsic differences between mouse and human immune system, which is shaped by a complex interplay of genetic, epigenetic and environmental factors, and exhibits significant differences in development, homeostasis and response to challenge [ 54 ]. Multiple divergent molecular pathways in immune cells have been described, including for macrophages [ 55 ] and microglia [ 56 ]. Also, transcriptional changes upon different pro-inflammatory stimuli significantly diverge between mice and humans [ 57 ], and differences in selected inflammatory molecules have been reported between brains of Creutzfeldt-Jakob disease patients and brains of transgenic mice overexpressing the human prion protein and inoculated with Creutzfeldt-Jakob disease-derived prions [ 53 ]. In this context, it is of interest to note that Cst7 was among the most strongly upregulated transcripts in the hippocampus of young vs. old mice, ranking 2 nd and 12 th between 3 vs. 24 month-old mice and 3 vs. 29 month-old mice, respectively [ 58 ]. Conversely, its human homologue CST7 was not differentially expressed in the aging human brain (The Human Brain Transcriptome atlas [ 59 ]). Also, Cst7 transcripts are almost exclusively found in microglia/macrophages in mouse brain, whereas CST7 transcripts are present both in microglia/macrophages and endothelial cells in the human brain [ 14 , 15 ].
The robust overexpression of Cst7 /cystatin F in our investigated mouse models, together with the upregulation of Serpina3n /α1-anti-chimotrypsin in both mouse models and human cases of prion disease, suggests a role for cysteine protease inhibitors in prion pathogenesis. Cysteine proteases have been found to be involved in the clearance of partially PK-resistant PrP in vitro [ 60 – 62 ]. Also, upregulation of cysteine proteases has been reported in different cellular and animal models of prion diseases [ 47 , 50 , 63 – 65 ] and in brains of Creutzfeldt-Jakob disease patients [ 66 ]. Further studies may clarify whether changes in cysteine proteases, and inhibitors thereof, are directly involved in prion disease pathogenesis or instead represent reactive changes.
As cystatin F can be secreted in the extracellular space and is present in body fluids, we explored its potential as a biomarker of Creutzfeldt-Jakob disease, the prototypical prion disease in humans. However, we found no significant increase in cystatin F levels in either cerebrospinal fluid or brain parenchyma of patients with Creutzfeldt-Jakob disease compared to Alzheimer’s disease or non-demented controls. Our results demonstrate the existence of dramatic species differences between mice and humans in terms of brain cystatin F expression during prion disease. They also rule out cystatin F as a useful surrogate biomarker in Creutzfeldt-Jakob disease.
Supporting information
S1 Fig
Partially protease K-resistant PrP brain accumulation during prion disease.
Western blotting analysis showing the amount of partially protease K (PK)-resistant prion protein (PrP) in whole-brain extracts of prion-infected mice at various dpi (upper membrane). Identical protein extracts omitting proteinase K treatment were used for western blotting with an anti-actin antibody to allow for normalization and to verify equivalent loading in each lane (lower membrane). Each lane denotes a brain extract from a representative mouse from each group.
(TIF)
S2 Fig
Cst7 brain expression in physiologic and pathologic conditions.
A Expression levels of Cst7 in different cell types of the mouse brain based on RNA sequencing profiling of acutely purified cell populations from the mouse brain cortex. FPKM: fragments per kilobase of transcript per million mapped reads. The oligodendrocyte precursor population is reported to have a 5% contamination with microglia based on whole transcriptome profile. Dots: denote individual cell preparations; bars: mean; error bars: standard deviation. Data are from the mouse brain transcriptome database [ 14 ]. B Expression levels of CST7 in different cell types of the human brain based on RNA sequencing profiling of acutely purified cell populations from the human brain cortex. FPKM: fragments per kilobase of transcript per million mapped reads. Dots: denote individual cell preparations; bars: mean; error bars: standard deviation. Data are from the human brain transcriptome database [ 15 ]. C Expression levels of Cst7 in different developmental stages of mouse microglia and in relation to the expression of the early microglial marker Tmem119 or to the treatment with lipopolysaccharide (LPS). En indicates embryonal day n; Pn indicates post-natal day n. Dots: denote individual cell preparations; bars: mean; error bars: standard deviation. Data are from the mouse developmental microglia dataset, Bennett et al. [ 16 ]. C Northern hybridization analysis of Cst7 mRNA in whole brains of mice with different neurodegenerative/neuroinflammatory conditions, as compared with levels in brains of mice at 190 days post injection (dpi) with RML prions or non-infectious brain homogenate (NBH). ΔF: terminally sick mice expressing a toxic, truncated PrP C molecule [ 31 ]; TgAPP23: aged mice overexpressing human APP Swe/V717I [ 32 ]; EAE: mice with experimental autoimmune encephalitis induced by administration of MOG 35-55 peptide emulsified in complete Freund’s adjuvant (CFA) and receiving an intraperitoneal injection with pertussis toxin [ 11 ]; CFA: EAE control mice were only the injection of the MOG 35-55 peptide was omitted. Each lane denotes a pool of brain extracts from 3 mice.
(TIF)
S3 Fig
Cst7 /cystatin F expression in different organs in mice.
Correlation between cystatin F protein levels (x axis) and Cst7 mRNA levels (y axis) in thymus and different regions of the central nervous system (CNS) in adult C57BL/6J mice (n = 4 for all organs except n = 3 for spinal cord). Each dot denotes an individual organ or CNS area. Inset: magnification of the area delimited between the axis and the dashed lines.
(TIF)
S4 Fig
Cystatin F levels in cerebrospinal fluid from patients with different neurologic conditions.
Cystatin F levels were measured in patients with different neurological conditions (Bact. Meningitis, bacterial meningitis; Vir. meningitis, viral meningitis). Dots: denote individual subjects; bars: mean; error bars: standard deviation (*, p<0.05; **, p<0.01, one-way ANOVA followed by Bonferroni’s multiple comparison test).
(TIF)
S5 Fig
Microglia in brains from analyzed patients with Creutzfeldt-Jakob disease and Alzheimer’s disease.
Histologic analysis of microglia (IBA1) in frontal cortex of the same cohort of patients with Alzheimer’s disease (AD) and Creutzfeldt-Jakob disease (CJD) reported in Fig 5 . Non-demented subjects were included as control (Ctrl). Images from two representative cases per group are depicted. Scale bar: 100 μm.
(TIF)
S1 File
Uncropped and unmodified Western blot and Norther blot images.
(PDF)
S1 Table
Clinical assessment and scoring of wild-type mice inoculated with RML prions.
(DOCX)
S2 Table
Clinical assessment and scoring of wild-type mice after induction of experimental autoimmune encephalitis.
(DOCX)
S3 Table
List of primers used in this study.
(DOCX)
S4 Table
Differentially expressed probes as assessed by the perfect match-mismatch model-based analysis of microarray data.
(XLSX)
S5 Table
Expression data (as reads per kilobase per million mapped reads, RPKM) of Cst7 , Gfap and Serpina3n during prion disease as assessed by RNA sequencing.
(XLSX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
This article [ 1 ] cites a PLOS ONE article that was retracted on February 9, 2021 [ 2 , 3 ] because the sample used for genome analysis was found to have been misidentified as it may have been from a hybrid between Arctic charr and a related species, Dolly Varden. The genomic assembly reported in [ 2 ] is still available at GenBank, but it has been relabeled as an unclassified species in the Salvelinus genus, or Salvelinus sp . (GenBank accession: GCF_002910315.2). The draft Salvelinus sp . genome was used for the determination of SNP location for the array in [ 1 ].
Of the 15 sources of fish used in the design and testing of the genotyping array (Table 1 in [ 1 ]), the authors are confident that those involving Fraser, Nauyuk and Icelandic populations are pure Arctic charr based on related research [ 4 , 5 ] and geographical isolation of the populations from other species involved in potential hybridization. The Tree River strain is represented in 3 sources used in [ 1 ], and it is possible that these samples are hybrids of Arctic charr and Dolly Varden. The authors noted that this was previously discussed in [ 4 ].
A member of the Editorial Board advised that the SNP locations in the array may be approximations due to the misidentification of the sample used for the draft genome [ 2 ] and that the resolution of the array may be overrepresented due to the potential inclusion of Dolly Varden SNPs via the Tree River samples. They advised that a draft assembly using a confirmed pure Arctic charr sample could be used to evaluate the performance of the array.
The corresponding author agreed that an array based on only Arctic charr SNPs would provide higher resolution, and stated that subsequent analysis published in a PhD thesis [ 6 ] shows that the Salvelinus sp genome was useful as a reference for the location of SNPs within the Arctic charr genome. A second member of the Editorial Board indicated that the thesis supports these claims, and agreed with the comments of the first Editorial Board member.
The PLOS ONE Editors issue this Editorial Note to notify readers of the use of a potentially hybrid strain in the design of this array, and to provide a summary of the discussion regarding the potential implication of this on the efficacy and accuracy of the array.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
In wild type Escherichia coli cells, DNA double-strand break repair (DSBR) is mediated by the RecBCD pathway of homologous recombination. In this pathway, DNA is unwound by RecBCD and cleaved five nucleotides 3’ of the sequence known as Chi (5’-GCTGGTGG-3’) [ 1 ]. Following recognition of Chi, RecBCD continues to unwind the substrate and facilitates the loading of RecA onto the 3’ strand close to Chi. In vitro , the fates of the DNA strands between the DNA double-strand break (DSB) site and Chi and of the strand terminating 5’ at Chi depend on the ATP/Mg 2+ concentration ratio. Degradation of these strands increases in vitro as the ATP/Mg 2+ concentration ratio increases but the extent of degradation in vivo is unknown. Two recent reviews of the RecBCD pathway of recombination describe this reaction in detail and depict the “Chi modulated DNA degradation” and “nick at Chi” models for the initiation of recombination shown in Fig 1A [ 2 , 3 ]. Following the formation of a D-loop through the strand exchange activity of RecA, Holliday junctions are formed, migrated and resolved by the RuvABC complex resulting in the formation of a structure resembling a replication fork. Subsequently, PriA is recruited to this fork-like structure, and is required to initiate a cascade of protein binding that ultimately results in the loading of the primary replicative helicase, DnaB, to the lagging-strand template [ 4 ]. DNA synthesis then proceeds to replace any genetic information lost at the site of the DSB (see [ 5 ] for a recent review). RecG has been a mysterious player in these reactions.
10.1371/journal.pgen.1005799.g001
Fig 1
Current models for RecBCD action and Chi cleavage in the initiation of recombination and schematic depiction of the site of DSBR used in this work.
A. Current alternative hypotheses for the initiation of recombination by RecBCD [ 2 , 3 ]. RecBCD (pink figure) loads at the site of a DSB and translocates along the duplex DNA (i). During translocation, RecBCD either degrades both strands up to the recognition of a correctly oriented Chi site (with a preference for cleaving the 3’ terminal strand) or unwinds the duplex DNA without degrading it. Once a correctly oriented Chi site is recognised, the complex undergoes a conformational change and either up-regulates 5’ to 3’ cleavage while inhibiting 3’ to 5’ cleavage (ii) or nicks the 3’ terminal DNA strand and continues unwinding ( iv ). Both of these scenarios “Chi modulated DNA degradation” and “nick at Chi” lead to the formation ssDNA with a 3’ terminus, which is a substrate for the loading and polymerisation of RecA. These alternative hypotheses for the initiation of recombination lead to the formation of different structures of joint molecules (iii and v) and therefore to different biochemical steps following strand invasion and D-loop formation by RecA coated DNA. B. Map of the E . coli chromosomal depicting the two replichores and the site of DSBR used in this work. The chromosome of E . coli is drawn as a black line and the directions of replication of the left and right replichores are indicated by green and red arrows respectively. The regions of DSB induction in lacZ and of insertion of an ectopic terB site in ykgM-terB are shown in more detail. The palindrome and Chi arrays are shown by a black triangle and three coloured circles, respectively.
The observation that RecG not only plays a role in the RecBCD pathway of DSBR but also in the RecF and RecE pathways (activated in mutant strains of E . coli ) suggests that, like RecA, it plays a fundamental role in DNA repair and acts on a DNA substrate that is common to different recombination pathways [ 6 ]. Indeed, its importance in DSBR has been confirmed using both cleavage of a chromosomal I-SceI target site with the I-SceI enzyme [ 7 ] and cleavage of a hairpin DNA structure by SbcCD nuclease [ 8 , 9 ]. Despite the early genetic evidence for a function common to three recombination pathways [ 6 ], many different roles for RecG have been proposed. These range from the migration of Holliday junctions to facilitate their resolution [ 7 , 10 , 11 , 12 , 13 , 14 ], the promotion and opposition of RecA strand exchange [ 15 , 16 ], the reversal of replication forks [ 17 , 18 , 19 , 20 , 21 , 22 , 23 ], the processing of flaps generated when DNA replication forks converge [ 24 , 25 , 26 , 27 ] and the stabilisation of D-loops [ 9 ]. Understanding the role of RecG has not been facilitated by the fact that the existence or identity of a eukaryotic homologue or functional orthologue has not been reported until recently [ 28 ]. If SMARCAL1 is indeed the human functional orthologue of RecG, there is hope that more light will be shed on the function of this important protein.
Purified RecG protein is a helicase that can bind and unwind synthetic model Holliday junctions [ 12 ]. In vitro , RecG efficiently catalyses the re-pairing of template strands in substrates mimicking replication forks, in a reaction termed replication fork reversal or replication fork regression [ 18 , 19 , 21 , 22 , 23 ]. Interestingly, this RecG promoted reaction occurs preferentially on substrates mimicking replication forks with a new strand annealed to the lagging-strand template [ 20 , 21 ]. RecG also efficiently reverses a replication fork blocked at a DNA lesion in an in vitro replication system where the DNA polymerase and replicative helicase are associated with the DNA [ 29 ]. These studies have led to a current view that an important biochemical action of RecG in vitro is replication fork reversal [ 30 ]. However in live cells there is a lack of evidence for RecG mediated fork reversal in several in vivo fork reversal reactions (e.g. [ 31 ]). Some indirect results imply that RecG might reverse replication forks following UV irradiation [ 19 ]. However following UV irradiation, the chromosome fragmentation by RuvABC-mediated cleavage of Holliday junctions present at reversed forks, which can be detected in a recBC mutant, is hardly affected by RecG [ 32 ]. This does not support even the view that RecG has a specific role in reversing forks following UV irradiation. The discordance between the substantial amount of evidence for RecG catalysed fork reversal in vitro and the small amount of evidence in vivo raises an interesting question: what is the substrate for RecG in live cells?
A clue as to the nature of the RecG substrate in vivo comes from the observation that a class of suppressors of the recG recombination deficient phenotype carry mutations in the helicase domain of PriA [ 33 ]. This is consistent with an interaction between RecG and PriA during the processing of recombination intermediates. PriA is required for the re-start of replication forks, during chromosomal DNA replication, recombination and replicative transposition, via the loading of the DnaB helicase [ 4 , 34 , 35 , 36 ]. Both RecG and PriA are known to remodel replication fork substrates in vitro . RecG binds the parental double-stranded part of a replication fork and unwinds the new strands (see [ 30 ] for a recent review). It has a preference to unwind a model fork substrate with a 5’ new lagging-strand at the fork over a substrate with a 3’ new leading-strand at the fork [ 20 , 21 ]. RecG unwinds the 5’ new lagging-strand and pairs it to the 3’ new leading-strand to generate a reversed fork [ 18 , 19 , 21 , 22 , 23 , 29 ]. However, in a coupled reaction where RecG and PriA are both present, RecG unwinds the 5’ new lagging-strand until a recessed 3’ new leading-strand end is brought to the branch point of the fork whereupon PriA binds in a configuration that does not lead to unwinding of parental template strands by the PriA helicase or continued unwinding by RecG [ 37 ]. A replication fork with a 3’ end at the branch point is a favoured substrate for PriA binding through the combined action of its N-terminal 3’ end binding domain (3’DB), a parental-strand binding winged helix domain (WH) and the helicase domains (HD1 and HD2) thought to contact the lagging-strand [ 38 ]. The biochemical literature supports the idea most clearly presented by Masai and colleagues [ 35 ] that RecG remodels replication forks to permit the 3’ end binding mode of PriA at a stalled fork or D-loop promoting the hand-off reaction to DnaB via PriB, DnaT and DnaC [ 39 , 40 , 41 ]. In the absence of a 3’ new leading-strand at the fork, PriA alone cannot be stabilised in the configuration in which its helicase is inactive for unwinding the parental duplex [ 35 ]. Instead, PriA moves from 3’ to 5’ on the leading-strand template to unwind the parental duplex and on the lagging-strand template to unwind the 5’ new lagging strand [ 35 ].
We show here that in the absence of RecG, abnormal DNA synthesis proceeds outwards and away from a specific site of attempted DSBR. Also, we show that in the absence of RecG attempted DSBR occurs at sites known to block DNA replication forks. Furthermore, we demonstrate that the DNA loss associated with the unwinding of joint molecules observed in the absence of both RecG and RuvAB requires PriA helicase activity. These results have led us to conclude that in vivo RecG plays a critical role in directing DNA synthesis at D-loops through its remodelling of the DNA to promote the correct binding of PriA. In turn, this has led us to reconsider the RecBCD recombination pathway in bacteria and to propose a mechanism in which the presence of a 5’ terminal DNA strand at a D-loop plays a more prominent role than generally envisaged.
Results
Chromosomal marker frequency analysis (MFA) following induction of a DSB in the absence of RecG
We have used MFA by next generation genomic DNA sequencing to determine the DNA abundance profile in a recG deletion mutant following attempted DSBR at the site of an interrupted 246 bp palindrome (Pal + ) in the lacZ gene of the E . coli chromosome ( Fig 1B ), following expression of the hairpin endonuclease SbcCD [ 8 ]. In the absence of a DSB at lacZ , the MFA pattern observed in a Δ recG mutant was as previously published [ 27 , 42 ]. An excess of DNA reads was detected in the region of the chromosome between the unidirectional termination sites, terA and terB ( S1 Fig and S2 Fig ). Normalisation of the number of mapped sequencing reads in a Δ recG mutant to the number of mapped reads in a Rec + strain clearly revealed the excess of reads in the terminus region of the Δ recG mutant ( Fig 2A ). In the strain undergoing DSBR at the palindrome in lacZ , a similar pattern as in the strain that was not attempting DSBR was observed in the terminus region. However, there was also a loss of reads in the immediate vicinity of the DSB in lacZ followed by an excess of reads on both sides of this DSB ( Fig 2B , S1 Fig and S2 Fig ). The effect of attempted DSBR in the lacZ region is clearly visible ( Fig 2D ) when normalising the Δ recG Pal + dataset (induced DSBR, in the presence of a 246 bp palindrome at lacZ ) to the Δ recG dataset (no induced DSBR). Extra DNA accumulates on both sides of the DSBR site in a Δ recG mutant. This extra DNA, which is not observed in a Rec + strain ( Fig 2C ), extends back towards the origin for about 300 kb and towards the terminus for about 1 Mb. It has previously been demonstrated that UV irradiated recG mutant cells undergo excess DNA replication that is not associated with initiation of DNA replication from the origin ( oriC ) [ 25 , 26 ]. Our work now shows that this abnormal DNA synthesis occurs on both sides of a site-specific DSBR event directly linking the location of DNA synthesis to the location of DSBR. In order to confirm that there is increased divergent DNA replication from the site of attempted DSBR in lacZ in the abnormal direction towards the origin, we inserted an ectopic terB site 50 kb origin-proximal of the palindrome in the orientation predicted to block replication forks progressing back towards the origin. We detect a 9-fold increase in replication fork blockage at this ectopic terB site in a recG mutant over Rec + under conditions of DSBR at lacZ ( S3 Fig ) consistent with an increased level of divergent DNA replication in the recG mutant.
10.1371/journal.pgen.1005799.g002
Fig 2
MFA profiles of Δ recG mutants and Rec + strains of E . coli as a consequence of attempted DSBR at the lacZ locus.
The ratio of the normalized DNA copy number (or “relative enrichment”) of uniquely mapped sequence reads from exponentially growing cultures of the strains of interest are plotted along the y -axis against replichore-formatted genomic coordinates along the x -axis. The average relative enrichment of DNA in a Δ recG mutant to a Rec + strain is shown in the absence (A) or in the presence of an induced break at the palindrome (C), between Rec + strains and Δ recG mutants in the presence and in the absence of an induced break at the palindrome (C-D). The relative positions of the replication termination sites ( terB , terC and terA ), dif site and the location of palindrome are shown for each plot. The data are the averages of the two biological replicates shown individually in supporting Information S1 Fig and S2 Fig . Strains used were DL4184 (Rec + Pal + ), DL4201 (Rec + Pal - ), DL4311 (Δ recG Pal + ), and DL4312 (Δ recG Pal - ).
In the absence of RecG, attempted DSBR occurs at sites of replication fork arrest
We have previously developed a method for visualising attempted DSBR that relies on chromatin immunoprecipitation of RecA cross-linked to DNA, followed by whole genome sequencing (RecA ChIP-seq; [ 43 ]). RecA is bound to DNA at sites of attempted DSBR following its loading at Chi sites by RecBCD. The shape of the RecA binding profile is distinctive. Binding rises sharply to a maximum value close to the position of a correctly oriented Chi site and then decreases with a slow exponential decay. This binding profile coupled to the locations and orientations of the Chi sites can be used to identify the region of the chromosome in which a DSB has been generated. These characteristics can also be used to distinguish between one-ended and two-ended breaks and to determine the directionality of a one-ended break. As can be seen in Fig 3A–3D , attempted DSBR in the presence and absence of RecG occurs at the site of palindrome cleavage in the lacZ gene. As expected from the results obtained in a Rec + strain [ 43 ], RecA enrichment was observed on both sides of the break consistent with two-ended DSBR. In addition, three sites of attempted one-ended DSBR were specifically observed in the absence of RecG. The first of these was at a Chi site oriented appropriately if a replication fork proceeding from the site of the initial DSB in lacZ towards the origin of chromosomal replication generated a double-strand end at the closest ribosomal RNA operon ( rrnH ), 120 kb on the origin-proximal side of the DSB ( Fig 3C ). Because this replication fork would be proceeding in the reverse direction to normal chromosomal replication, it would have encountered the rrnH operon as it moved in the opposite direction to its transcription. Replication-transcription collisions of this kind are known to result in blocking of replication forks [ 44 , 45 , 46 , 47 , 48 ] and can generate one-ended DSBs [ 31 ]. It is worth noting that the rrnH operon itself is recognised by RecA but this recognition is independent of DSBR and independent of RecG. Furthermore, it bears no hallmarks of DSBR such as correlation with Chi sites or an asymmetric distribution (see [ 43 ] for further details of recombination independent RecA binding to rRNA genes). In a Δ recG mutant, RecA binding was also detected approximately 100 kb origin-distal to the DSB in lacZ in a Δ recG mutant ( Fig 3C ). This peak (which can also be detected at a low level in the Rec + data) most likely corresponds to the origin-distal end of the DSB at lacZ being processed at a long distance. The elevated processing at a distance in a Δ recG mutant may be caused by unwinding of joint molecules followed by re-invasion downstream of the first Chi array or simply from RecBCD enzymes that had failed to recognise the first Chi array. The second and third sites of Δ recG specific attempted DSBR ( Fig 3G and 3H ) were located at positions of correctly oriented Chi sites for DSBs generated at the replication termination sites terA and terB [ 47 , 48 ]. Again these events were one-ended, consistent with replication fork processing, and were oriented appropriately for replication forks proceeding outward (from terminus towards origin) and being blocked at the termination sites. These sites of one-ended DSBR were also the boundaries of the extra terminal DNA replication that has been detected in Δ recG mutants by MFA ( Fig 2 and S1 Fig ) and [ 27 , 42 ].
10.1371/journal.pgen.1005799.g003
Fig 3
Relative RecA ChIP-seq reads in the lacZ DSBR region (A-D) and terminus region (E-H) of the chromosome.
The raw data are shown in grey and smoothed data are shown in red. The smoothed data were plotted using a moving average filter with a 4 kb window. Red and green circles indicate Chi sites. Red Chi sites interact with RecBCD enzymes moving from right to left and green Chi sites interact with RecBCD enzymes moving from left to right. The closest Chi sites on either side of the DSB in lacZ were triple Chi arrays at 1.5kb from the palindrome, which have been used previously [ 9 ]. The positions of the rrnH operon, the palindrome at which a DSB is induced, termination sites ( terA , terB and terC ) and the site of resolution of chromosome dimers ( dif ) are all indicated. The direction of replication is indicated by green and red arrows for the left and the right replichore, respectively. A. and E. Rec + Pal + ; B. and F. Rec + Pal - ; C. and G. Δ recG Pal + ; D. and H. Δ recG Pal - . Strains used were DL4184 (Rec + Pal + ), DL4201 (Rec + Pal - ), DL4311 (Δ recG Pal + ), and DL4312 (Δ recG Pal - ).
PriA helicase is responsible for the unwinding of joint molecules in the absence of RecG and RuvAB
We have shown previously that intermediates of DSBR are lost in a Δ recG Δ ruvAB double mutant [ 9 ] and have hypothesised that the branch migration activities of RecG and RuvAB stabilise joint molecules. Since RuvAB is a complex known to branch migrate Holliday junctions and to facilitate their resolution by cleavage in the presence of RuvC (see [ 5 ]), we considered it likely that the stabilising activity of RuvAB is mediated by branch migration of Holliday junctions. This was confirmed by the observation that 4-way junctions accumulated in a Δ ruvAB mutant [ 9 ]. However, the branch migration activity of RecG implicated in stabilising the joint molecules was less clear. The fact that 4-way junctions accumulated in the presence of RecG in a Δ ruvAB mutant indicated that they were not migrated away from the region of joint molecule formation by RecG. Instead, this suggested that RecG might stabilise joint molecules by remodelling the nascent fork end of the D-loop to promote DNA synthesis from the invading 3’ end.
We have now tested whether the helicase activity of PriA is responsible for the DNA loss associated with destabilising joint molecules in the absence of RecG and RuvAB. The loss of DNA following induction of DSBR at lacZ was quantified by agarose gel electrophoresis and Southern hybridisation. The recovery of the 7.8 kb NdeI DNA fragment containing the DSB site in lacZ ( Fig 4A ) was compared to the recovery of the 10 kb NdeI cysN control fragment situated on the opposite side of the chromosome. As can be seen in Fig 4B and 4C , 40% of the DNA undergoing DSBR in a Δ recG Δ ruvAB mutant was lost from the lacZ region. This loss was prevented in a Δ recG Δ ruvAB priA300 mutant, in which the helicase activity of PriA is inactivated by the K230R mutation [ 49 ]. The nature of the intermediates accumulated in a Δ recG Δ ruvAB priA300 mutant was investigated by two-dimensional native-native agarose gel electrophoresis. In the absence of RecG and RuvAB, the priA300 mutation increased the recovery of X-spike intermediates in the 7.8 kb NdeI fragment containing the DSB site, consistent with the accumulation of 4-way junctions ( Fig 4E and 4F and S4 Fig ). Our data suggest that the helicase activity of PriA is responsible for the unwinding of D-loops in the absence of the stabilising activities of RecG and RuvAB. Since the priA300 mutation also suppresses the recombination deficiency of a recG mutant [ 50 ], we argue that it is RecG that prevents the unwinding activity of PriA helicase, suggesting that RecG is operating to facilitate the correct binding of PriA for DNA synthesis rather than D-loop unwinding.
10.1371/journal.pgen.1005799.g004
Fig 4
The priA300 mutation suppresses the loss of DNA around a DSB in the absence of RecG and RuvAB.
A. NdeI digestion map of the region surrounding the palindrome locus. NdeI cutting sites and the distance between them are marked with black vertical arrows and numbers (in kb), respectively. The palindrome is indicated by a black triangle, Chi arrays by three coloured circles, the lacZ probe by a blue line and the lacZ . distal probe by a red line. B. Southern blot of a 1% agarose gel probed with a lacZ fragment (top) and a cysN control fragment (bottom). Strains used were DL4184 (Rec + Pal + ), DL4260 (Δ ruvAB Δ recG Pal + ), DL5610 (Δ ruvAB Δ recG priA300 Pal + ), DL4201 (Rec + Pal - ), DL4313 (Δ ruvAB Δ recG Pal - ) and DL5611 (Δ ruvAB Δ recG priA300 Pal - ). C. Quantification of the total amount of DNA at and around the break site. These values were first normalised to the values for the cysN control fragment. Then these ratios for the Pal + strains were normalised to their Pal - controls. Finally, these ratios were normalised to the Rec+ ratio that was set to the value of 1. Error bars represent the standard error of the mean where n = 3. D. Schematic representation of the migration patterns of different species of branched DNA when separated on a two-dimensional native-native agarose gel. E. Two-dimensional native-native agarose gel electrophoresis. The DNA was detected using the lacZ . distal probe. Strains used were DL4243 (Δ ruvAB Pal + ), DL4257 (Δ ruvAB Pal - ), DL4260 (Δ recG Δ ruvAB Pal + ), DL4313 (Δ recG Δ ruvAB Pal - ), DL5610 (Δ recG Δ ruvAB priA300 Pal + ) and DL5611 (Δ recG Δ ruvAB priA300 Pal - ). F. Quantification of the DNA in the Y-arc and the X-spike normalised against the total branched DNA. Error bars represent the standard error of the mean where n = 3.
Discussion
In this work, we have made three principal observations pertaining to DSBR that is attempted in the absence of RecG. First, divergent replication occurs on both sides of the DSB ( Fig 2 ). Second, stalled replication forks are processed to generate double-strand ends (at an rrnH operon, where collision between transcription and a divergent replication fork is expected, and at replication termination sites terA and terB ) ( Fig 3 ). Third, the helicase activity of PriA unwinds joint molecules in the absence of both RecG and RuvAB ( Fig 4 ). We propose that RecG directs DNA synthesis at sites of DSBR and that this is mediated via the correct binding of the PriA. This proposal builds on the demonstration that RecG determines the correct binding of PriA in vitro [ 35 , 37 ] and reconciles a large body of literature describing the biochemical and genetic properties of both RecG and PriA.
RecG directs DNA synthesis at sites of DSBR
Previous work has demonstrated that an excess of oriC -independent DNA replication occurs in a recG mutant following UV irradiation [ 25 ]. We have shown that at an induced DSB in lacZ , and adjacent to sites of one-ended breaks in the terminus region, there is DNA over-replication that proceeds away from the direction of appropriate replication (the direction of reconstitution of a replication fork at a D-loop). This establishes that the over-replication observed following attempted DSBR in a Δ recG mutant is associated with the site of DSBR itself. Previous work has shown that the over-replication observed following UV irradiation of a recG mutant is suppressed in priA helicase mutants implicating PriA in the over-replication phenotype [ 25 ]. Given the biochemical evidence that RecG remodels the DNA at a replication fork for the appropriate binding of PriA [ 37 ], we have considered whether, in the absence of RecG, PriA might bind to direct DNA synthesis inappropriately. In order for PriA to load DnaB incorrectly at the site of a D-loop, we envisage that PriA would bind in its 3’ end recognition mode in an orientation appropriate for loading DnaB onto the strand ending 5’ at the D-loop. We only see this as possible if the strand ending 5’ at the D-loop extends further than the 3’ ended strand ( Fig 5 ).
10.1371/journal.pgen.1005799.g005
Fig 5
Model depicting the proposed action of RecG and PriA in the RecBCD recombination pathway.
A. Revised model for the RecBCD recombination pathway. (i) The RecBCD enzyme recognises and binds to a DNA double-strand end. (ii) RecBCD generates a substrate with a 3’ end adjacent to a Chi site (shown as an arrow pointing in the direction of recombination stimulation) and a 5’ overhang. Continued unwinding by the RecBCD enzyme coupled to RecA loading onto the strand ending 3’ close to Chi allows the invasion of a target duplex and the formation of a D-loop. (iii) RuvAB binds to the Holliday junction end of the D-loop and migrates the Holliday junction away from the DSB end enlarging the D-loop. When a preferred recognition site for RuvC is encountered, the Holliday junction is resolved by cleavage and ligation. (iv) The RecG protein binds to the replication fork with a 5’ extended strand generated from the other end of the D-loop and unwinds the 5’ end while reannealing the parental DNA strands. (v) The action of RecG hands off the junction to PriA that binds in the correct manner to initiate the loading of DnaB. (vi) DnaB is loaded onto the lagging-strand template. (vii) DNA replication proceeds in the correct direction to restore the DNA lost in the region of the DSB. B. Hand-off between RecG and PriA ensures the correct loading of DnaB. The region between the dotted vertical lines is enlarged to show the binding of PriA. (i) RecG binds to a replication fork with an extended 5’ strand and unwinds this end while re-winding the parental template strands. (ii) This unwound fork is now in the right conformation to be bound by PriA in the orientation to load DnaB correctly onto the lagging-strand template. (iii) The hand-off reaction from PriA to PriB, to DnaT to DnaC to DnaB ensures that the replisome is reassembled. (iv) The replisome is loaded correctly to ensure the restoration of the DNA lost during the resection of the break. The box shows the domains of PriA as determined by X-ray crystallography [ 38 ]. The N terminus of the protein encodes the 3’ end-binding domain (3’BD–red). This is followed by a winged-helix domain proposed to interact with the parental DNA duplex (WH–orange). This is followed by two helicase lobes (HL1 –blue and HL2 –green). This is followed by a cysteine-rich region proposed to act as a wedge during helicase action (CRR–purple). Finally the protein is completed by a C-terminal domain that loops back round to the 3’BD (CTD–yellow). The priA300 mutation is predicted to lie in the HL1 domain. C. Action of PriA in the absence of RecG. The region between the dotted vertical lines is enlarged to show the binding of PriA. (i) A replication fork substrate with an extended 5’ new end is available to bind PriA but is not specifically remodelled for this hand-off in the absence of RecG. (ii) and (v) Because the 3’ end is readily available, PriA remodels the fork to ensure that the 3’ end is bound by the 3’BD and the parental duplex is bound by the WH domain. In structure (ii), the PriA helicase domains (HL1 and HL2) bind correctly to the lagging-strand template and in structure (v) the helicase domains bind incorrectly to the new lagging-strand. The PriA(K230R) helicase (present in the priA300 mutant) retains only the ability to bind correctly. (iii) and (vi) DnaB is loaded via the hand-off mechanism from PriA to PriB to DnaT to DnaC to DnaB. In (iii) DnaB is loaded correctly to the lagging-strand template and in (vi) DnaB is loaded incorrectly to the new lagging-strand. (iv) and (vii) A replication fork is reassembled. In (iv) the replication fork is assembled in the correct orientation to restore the DNA lost in the early stages of recombination at the site of the DSB. In (vii) the replication fork is assembled in the incorrect orientation and replicates the DNA flanking the DSB site. The direction of translocation of DnaB is indicated by a tan arrow. D. Action of PriA in the absence of RecG and RuvAB. The region between the dotted lines is enlarged to show the binding of PriA. (i) A replication fork with an extended 5’ “new” end is available for binding by PriA. However, the Holliday junction associated with the fork is not resolved by RuvABC and the fork itself is not remodelled by RecG. (ii) PriA has difficulty to remodel the fork to allow binding in the 3’ end-binding mode because the presence of the Holliday junction interferes with the required movement of the arms of the fork. This results in a significant proportion of molecules being bound by PriA in its helicase mode where it unwinds the parental duplex arms. (iii) This causes over-winding of the parental arms of the fork and under-winding of the D-loop, resulting in its dissociation. The direction of translocation of PriA is indicated by a black arrow.
How far this 5’ strand extends back towards the DSB site requires further investigation as does the fate of the 3’ strand from the DSB site to Chi. One can envisage two general scenarios based upon the known biochemistry of RecBCD enzyme (see [ 2 , 3 , 5 ] for recent reviews) and the models presented in Fig 1A . In one scenario, degradation of the 3’ end from the DSB site to Chi occurs frequently and the 5’ strand is cleaved infrequently leading to a recessed 3’ end at Chi. Following Chi recognition, unwinding by RecBCD continues but, in the presence of RecA, the Chi-activated 5’-3’ nuclease is inhibited, retaining the extended 5’ end. This would require an extension of the “Chi modulated DNA degradation” model [ 2 ] (see Fig 1A ). In this new scenario, RecA loading would inhibit 5’ end cleavage by RecBCD after Chi recognition. In an alternative scenario, DNA from the DSB to Chi is unwound and the 3’ end is cleaved at Chi while the 5’ end remains intact. Following Chi recognition and cutting, unwinding continues and RecA is loaded to the 3’ strand. In this scenario, the unwound 3’ strand from the DSB site to Chi is somehow prevented from annealing to the 5’ strand. This might be accomplished by cleavage of the 3’ or 5’ stands before Chi by unknown nucleases (e.g. ExoI or RecJ) or by the binding of SSB to both unwound strands. This would require an extension of the “nick at Chi” model [ 3 ] to explain the fate of unwound strands between the DSB site and Chi. Previous studies have demonstrated that SSB attenuates RecBCD nuclease action and inhibits reannealing of strands unwound by RecBCD [ 51 , 52 , 53 , 54 ]. These actions of SSB are likely to promote the persistence of a protruding 5’ single-stand provided the Chi-activated 5’-3’ nuclease of RecBCD is not operating (e.g. because of the ionic conditions or because of RecA loading).
Our model is summarised in Fig 5A . We envisage that RecBCD enables loading of RecA to a 3’ single-strand generated by unwinding beyond the cleaved Chi site and that a joint molecule is formed that retains a 5’ tail. RuvABC migrates and resolves the Holliday junction at one end of this joint molecule allowing the formation of a replication fork with an extended 5’ end. This is the preferred substrate for RecG [ 20 , 21 ]. RecG binds and unwinds the 5’ end while reannealing the parental template stands of the fork but hands off to PriA before unwinding of the 3’ end can occur [ 37 ], thus preventing fork reversal. In Fig 5B we show how PriA is expected to bind to permit the loading of DnaB to the lagging-strand template. In Fig 5C we compare the two possible binding modes of PriA to a substrate with a 5’ new strand at the fork in the absence of a hand-off reaction from RecG. It can be seen that a simple rotation of strands coupled to displacement of the 5’ end can lead to alternative 3’ end-binding modes that predict either loading of DnaB onto the lagging-strand template (correct loading) or onto the new lagging-strand (incorrect loading). Because the 3’ end is available and PriA can manipulate the junction both binding modes involve recognition of the 3’ end and lead to DnaB loading rather than helicase activity.
In the absence of RecG, PriA helicase can unwind D-loops that have not been converted to replication forks by RuvABC
Joint molecules are formed through the action of RecBCD and RecA. We have previously proposed that in the absence of RuvAB and RecG these joint molecules are unstable because D-loops cannot be converted to replication forks by RuvABC action and because RecG is not present to carry out an unknown stabilising role [ 9 ]. We considered that this stabilising role could either be the migration of the Holliday junction away from the site of DSBR or the establishment of correct DNA synthesis from the site of the D-loop. Given the known suppression of the recG recombination defective phenotype by helicase mutants of PriA and our observation of inappropriate backward-directed DNA synthesis at sites of attempted DSBR in a Δ recG mutant we sought to test whether PriA helicase activity might unwind D-loops in the absence of RecG and RuvAB. Our data reveal that the helicase activity of PriA is indeed responsible for the DNA loss associated with destabilisation of joint molecules in a Δ ruvAB Δ recG mutant. Two possible modes of unwinding by PriA helicase that have been observed in vitro might be responsible for this. Unwinding of the 5’ end would directly unpair one of the D-loop double-strands, while unwinding the parental duplex strands would cause strand rotation that would unwind the D-loop (D-loop migration). We consider that the unwinding of the parental duplex and the consequent unwinding of the D-loop by strand rotation, required to minimise accumulation of positive supercoils (ahead of the D-loop) and negative supercoils (behind the D-loop) during its migration, is likely to be the critical activity of PriA helicase in this situation. This is because this action would result in ejection of both the 3’ and the 5’ ends from the D-loop, which would be needed to unwind the joint molecules. This action of PriA helicase requires an extended 5’ end at the replication fork side of the D-loop, to provide the single-stranded DNA region for PriA binding on the leading-strand template. This is consistent with our view that such an end is indeed present. We envisage that remodelling of the replication fork end of the D-loop is prevented in the absence of RuvAB by a persistent Holliday junction that tethers the two strands of the fork. This prevents the binding of PriA in the 3’ end-binding mode required for DnaB loading and leaves only the helicase mode of PriA binding available as shown in Fig 5D .
The role of RecG in terminus over-replication
As seen previously in a recG mutant [ 27 , 42 ], we observe DNA over-replication in the terminus region of the chromosome between the sequences terA and terB . This over-replication is eliminated in helicase mutants of PriA [ 27 ]. We show here that terminus over-replication in the absence of RecG is not influenced by attempted DSBR at lacZ but is associated with attempted DSBR at terA and terB as revealed by RecA binding at the positions of the first correctly-oriented Chi sites adjacent to these ter sites. We therefore propose that this over-replication is caused by a similar reaction to the backward replication from D-loops that we envisage happening at the DSBR event in lacZ . Because the DSBs at terA and at terB are one-ended and outward-facing, they do not arise from replication fork collision in the centre of the terminus region as envisaged in the model proposed by Lloyd and colleagues [ 24 , 25 , 26 , 27 , 42 ]. Furthermore, our demonstration of backward-directed replication at a site of attempted DSBR in lacZ and of one-ended DSBR at terA and terB do not fit with the model of Gowrishakar [ 55 ] that does not envisage replication initiation in the terminus region.
A depiction of how we envisage terminus replication in the absence of RecG is shown in Fig 6 . We propose that in the absence of RecG, a replication fork that has been blocked by collision with a Tus/ ter complex is no longer protected from incorrect binding of PriA helicase. This results in the deposition of DnaB on the newly synthesised strand ending 5’ close to ter and the establishment of a fork that moves back across the terminus region until it is stopped by encounter with another ter site. At this point, another backward-directed replication fork can be assembled and replication can copy the same region again in the opposite direction. In the meantime the ends generated by backward-directed replication will attempt recombination and so create more forks that can set up more backward-directed replication as well as forks that will collide with the original ter sites. This cascade of replication in the absence of RecG explains the DNA over-replication of the terminus region. The initial formation of replication forks blocked at the ter sites in a Δ recG mutant is likely to be contributed to by stable DNA replication as suggested previously [ 55 ].
10.1371/journal.pgen.1005799.g006
Fig 6
Model proposed for the over-replication of the terminus region between terA and terB in the absence of RecG.
(i) A replication fork is shown having traversed the terminus region in the direction from terA to terB where it is arrested. (ii) In the absence of RecG, PriA binds incorrectly at terB and causes a replication fork to be assembled that moves in the reverse direction towards terA . Whether the arrested fork is originally broken and repair is attempted prior to the assembly of the backward-directed replication fork is unknown. Whether or not the fork is broken, a single DNA end is generated adjacent to terB . (iii) The backward directed replication fork is blocked at terA and recombination of the DNA end with the intact duplex is attempted. (iv) The same process of assembly of a backward directed replication fork is set up at terA , this time moving towards terB . This game of ping-pong between terA and terB continues indefinitely with a preference for attempted DSBR events close to the ter sites but also with more internal sites derived from the D-loops generated from attempted recombination events. Replication forks can either start the process by being blocked at terB (as shown in (i)) or at terA . The combination of all the events occurring in the population results in the accumulation of DNA observed in a Δ recG mutant between terA and terB .
The precise molecular details of how PriA binds in the terminus region require further investigation. It is known that Tus protein blocks DNA synthesis initially leaving a recessed 5’ end of 50–100 nt [ 56 ]. It is possible that this is a poor substrate for the hand-off reaction from RecG to PriA but is converted to a good substrate via the action of 3’ to 5’ exonucleases, the absence of which can cause RecG independent replication in the terminus region [ 27 , 42 ]. Alternatively, Tus protein itself modifies the interaction of PriA with DNA in the absence of RecG.
Conclusion
We have shown that in the absence of RecG attempted DSBR at either the site of an induced two-ended DSB in lacZ , or at a site in which a replication fork is predicted to collide with a transcription bubble (at the rrnH operon), or at sites in which replication forks are expected to collide with the Tus/ter complex at ter sites, abnormal backward-directed DNA synthesis is observed. Furthermore, we have shown that D-loops that have not been acted upon by RuvABC or RecG are unwound by the helicase activity of PriA. These results strongly suggest that RecG acts at the replication fork end of a D-loop and possibly at a stalled replication fork to direct the correct loading of the DnaB replicative helicase through the correct binding of PriA. This conclusion is supported by the biochemical evidence that the action of RecG allows PriA to associate with a synthetic replication fork substrate with a recessed 3’ end in its 3’ end-binding mode in which it can promote the further hand-off reaction to DnaB rather than acting as a helicase [ 37 ]. This new understanding of the role of RecG reconciles many roles previously proposed. The synergistic action of RecG and RuvAB is explained by alternative modes of stabilising D-loops. The apparent contradiction that RecG strongly promotes replication fork reversal in vitro whereas little evidence for this reaction has been obtained in vivo is explained by the hand-off reaction from RecG to PriA, which captures a key DNA structure and prevents fork reversal in vivo . The single situation in which fork reversal has been proposed to occur in vivo is following UV irradiation [ 19 ]. It is possible that the extent of damage overwhelms the ability of PriA to capture all the precursors to fork reversal. There is no longer any need to propose a role for RecG in the processing of flaps hypothesised to occur at sites of convergent replication forks [ 24 , 25 , 26 , 27 ] as the fork collision model is not supported by the outward facing one-ended attempted DSBR that we infer at ter sites in the absence of RecG. Our new understanding also explains why RecG has a preference for action at a replication fork substrate with an extended 5’ end. This is indeed the substrate that we hypothesise normally to be present in a D-loop since we propose that the extended 5’ end is required for the inappropriate binding of PriA (in its incorrect 3’ end-binding mode) in the absence of RecG. It is also the structure that we hypothesise to be required for the incorrect binding of PriA (in its helicase mode) in the absence of RuvAB and RecG.
According to this view, RecG may be considered an early participant in the hand-off reaction from PriA to DnaB, which is required for the re-start of replication during DSBR. This pathway may be considered to run from RecG to PriA to PriB to DnaT to DnaC to DnaB [ 39 , 57 , 58 , 59 , 60 ]. Given that a pathway of replication restart from a DSB has not yet been identified in eukaryotic cells it will be interesting to know whether the potential human functional orthologue of RecG (SMARCAL1) opens a window on this important reaction in higher organisms.
Materials and Methods
Strains and oligonucleotide sequences used
All strains and oligonucleotide sequences used are listed in supporting information S1 and S2 Tables ( S1 Table : DNA oligonucleotide sequences used and S2 Table : Bacterial strains used).
Plasmid construction
The plasmid pDL4922 (Cm R Ts Suc s ) was created in order to introduce a terB site (5’-AATAAGTATGTTGTAACTAAAGT-3’) site in between the pseudogenes ykgM and eaeH of the E . coli chromosome to pause counter clockwise replication forks specifically. Primer pairs used for the cross-over PCR on BW27784 genomic DNA were ykgMterB-F1 /R1 and ykgMterB-F2/R2. These primers permit the insertion of a terB site between the two homology arms. This fragment was cloned in pTOF24 using PstI and SalI restriction enzymes [ 61 ].
The plasmid pDL4947 (Cm R Ts Suc s ) was created in order to introduce the priA300 mutation into the priA locus of the E . coli chromosome. The region was amplified from JJC1422 using priA300 .F and priA300 .R primers, digested using SalI and PstI and inserted into the temperature sensitive plasmid pTOF24.
Induction of DSBs
Overnight cultures were grown in 5ml of LB medium. The following day, cultures were diluted to an OD 600nm of 0.02 and grown shaking at 37°C to an OD 600nm of 0.2. Cultures were then re-diluted to an OD 600nm of 0.02 and grown shaking at 37°C to an OD 600nm of 0.2. Expression from the P BAD -sbcDC construct was induced by the addition of 0.2% arabinose to the culture medium. Cultures were then incubated at 37°C for 1 hour before samples were isolated.
Sample preparation for MFA by genomic DNA sequencing
DNA was isolated from cultures after 1 hour induction of sbcDC expression using the Promega Wizard ® Genomic DNA purification kit by following the manufacturer’s instructions. RNase treatment was carried out for 50 minutes and the DNA was re-hydrated overnight in TE (10 mM Tris (pH 7.4), 1 mM EDTA) at 4°C. To further eliminate potential RNA, 3 units of Riboshredder (RNase Blend) were added per sample according to the manufacturer’s instructions. Samples were purified by phenol/chloroform extraction and ethanol precipitation. The integrity of the DNA was verified by running the samples on a 0.8% agarose gel and the quantity of DNA was determined by Nanodrop analysis (Thermo Scientific) and by Qubit fluorometry (Life Technologies). Finally, construction of libraries and DNA sequencing was carried out on an Illumina HiSeq 2000 platform by Edinburgh Genomics, using the Illumina TruSeq DNA Sample Prep kit according to manufacturer’s instructions.
MFA data analysis
Paired-end raw datasets from an Illumina HiSeq 2000 sequencing platform (obtained from Edinburgh Genomics) were mapped against the genomic sequence of the reference strain ‘BW27784’ using BWA sequence aligner (version 0.7.11) and subsequently analysed using SAMtools (version 1.2). ‘BW27784’ is a modified version of E . coli K12 MG1655 (NC000913.3) including all published differences between the strains [ 62 , 63 ]. Replication profiles of exponentially growing cultures were calculated by normalizing to the number of uniquely mapped sequence reads (to correct for differences in depth of sequencing) and then to the normalised reads of a non-replicating stationary-phase wild-type culture (a Rec + strain without palindrome) to correct for differences in sequence-based recovery across the genome. An in-lab R-script (available on request) has been used to calculate the enrichment (normalised read depth) in 1 kb non-overlapping windows across the genome and a non-parametric smoothing method (LOESS, Local regression) has been applied to the data points of the replication profiles of each strain.
ChIP sample preparation
All ChIP experiments were performed with cells grown in exponential growth phase. RecA-DNA interactions were chemically cross-linked with formaldehyde (Sigma-Aldrich, at a final concentration of 1%) for 10 minutes at 22.5°C. Crosslinking was quenched by the addition of 0.5 M glycine (Sigma-Aldrich). Cells were collected by centrifugation at 1,500 x g for 10 minutes and then washed three times in ice-cold 1X PBS. The pellet was then re-suspended in 250 μl ChIP buffer (200 mM Tris-HCl (pH 8.0), 600 mM NaCl 4% Triton X, Complete protease inhibitor cocktail EDTA-free (Roche)). Sonication of crosslinked samples was performed using the Diagenode Bioruptor at 30 seconds intervals for 10 minutes at high amplitude. After sonication, 350 μl of ChIP buffer was added to each sample, the samples were mixed by gentle pipetting and 100 μl of each lysate were removed and stored as ‘input’. Immunoprecipitation was performed overnight at 4°C using 1/100 anti-RecA antibody (Abcam, ab63797). Immunoprecipitated (IP) samples were then incubated with Protein G Dynabeads® (Life Technologies) for 2 hours with rotation at room temperature. All samples were washed three times with 1 X PBS + 0.02% Tween-20 before re-suspending the Protein G dynabeads in 200 μl of TE buffer + 1% SDS. 100 μl of TE buffer were added to the input samples and all samples were then incubated at 65°C for 10 hours to reverse the formaldehyde cross-links. DNA was isolated using the MinElute PCR purification kit (Qiagen) according to manufacturer’s instructions. DNA was eluted in 100 μl of TE buffer using a 2-step elution. Samples were stored at -20°C.
ChIP library preparation for high-throughput sequencing
Input and ChIP samples were processed following NEB’s protocol from the NEBNext ChIP-Seq library preparation kit. Briefly, input and ChIP-enriched DNA were subjected to end repair to fill in ssDNA overhangs, remove 3’ phosphates and phosphorylate the 5’ ends of sheared DNA. Klenow exo- was used to adenylate the 3’ ends of the DNA and NEXTflex DNA barcodes (Bioo Scientific) were ligated using T4 DNA ligase. After each step, the DNA was purified using the Qiagen MinElute PCR purification kit according to the manufacturer’s instructions. After adaptor ligation, the adaptor-modified DNA fragments were enriched by PCR using primers corresponding to the beginning of each adaptor. Finally, agarose gel electrophoresis was used to size select adaptor-ligated DNA with an average size of approximately 275 bp. All samples were quantified on a Bioanalyzer (Agilent) before being sequenced on the Illumina® HiSeq 2000 by BGI International.
ChIP-Seq data analysis
50 bp single-end reads were mapped to the E . coli K12 ‘BW27784’ genome using Novoalign version 2.07 ( www.novocraft.com ). Novoalign uses the Needleman-Wunsch algorithm to determine the optimal alignment of reads. Before mapping, the 3’ adaptor sequences were removed using fastx_clipper and the data collapsed using fastx_collapser to remove identical sequence reads ( http://hannonlab.cshl.edu/fastx_toolkit/index.html ). Sequences were mapped with default parameters, allowing for a maximum of one mismatch per read. In order to report reads that have multiple alignment loci we specified the–r parameter as “Random”. PyReadCounters was used to calculate the overlap between aligned reads and E . coli genomic features [ 64 ]. The distribution of reads along the E . coli genome was visualized using the Integrated Genome Browser [ 65 ]. Full details of all scripts are available upon request.
The raw data are shown in grey and smoothed data are shown in red. The smoothed data were plotted using a moving average filter with a 4 kb window. The data have been normalised relative to the peak of RecA ChIP observed at the rrnH locus. This peak of RecA ChIP is independent of induced DSBR at lacZ , is independent of the recG genotype and does not have the characteristics of DSBR (it is not correlated with the positions of Chi sites and the binding is uniform across the gene). Whether or not this binding is of biological interest or represents a ChIP artefact remains to be determined. However, it usefully provides a way of approximately normalising reads between experiments. This normalisation cannot be considered absolute as this peak may itself be influenced by unknown factors that differ between experiments. We are therefore careful not to infer absolute levels of RecA binding between experiments.
DNA analysis by gel electrophoresis
Methods were adapted from [ 9 , 66 ]
(a) Isolation of chromosomal DNA in agarose plugs
After 1 hour of sbcCD induction, cells were harvested at 4°C and washed 3 times in TEN buffer (50 mM Tris, 50 mM EDTA, 100 mM NaCl, pH 8.0). Cells were re-suspended in TEN buffer to an OD 600nm of 6 or 80 for conventional agarose gels or native/native two-dimensional gels, respectively. The cells were then mixed with an equal volume of 2% (for conventional gels) or 0.8% (for two dimensional gels) of low melting point agarose (Invitrogen) prepared in TEN buffer and equilibrated to 37°C. The mix was poured into plug moulds (BioRad) and allowed to set for 1 hour. Plugs were treated in NDS solution (0.5 M EDTA, 10 mM Tris, 0.55 M NaOH, 36.8 mM lauroyl sarcosine; pH 8.0) supplemented with 1 mg/ml of proteinase K (Roche) for an overnight shaking at 37°C. Fresh NDS + proteinase K were added for a second overnight incubation. Following this treatment, plugs were stored at 4°C in fresh NDS. Before digestion of the DNA, a plug was washed in 1 x restriction buffer 6 times, replacing the buffer every hour. The plug was then placed in fresh 1 x restriction buffer, supplemented with the restriction enzyme and incubated rocking at 37°C overnight.
(b) Agarose gel electrophoresis
An agarose plug containing digested DNA was run on a 1% (w/v) agarose gel in 0.5 x TBE (44.5 mM Tris-borate, 1mM EDTA) at 2 V/cm for 12 hours at 4°C. The DNA was transferred to a positively charged nylon membrane (GE heathcare hybond+) by Southern blotting and cross-linked using UV-light.
(c) Native/native two dimensional agarose gel electrophoresis
An agarose plug containing digested DNA was run in the first dimension on a 0.4% (w/v) agarose gel in 1 x TBE (89 mM Tris-borate, 2 mM EDTA) at 1 V/cm for either 24 (for 4 kb fragment) or 36 hours (for 8 kb fragment) at 4°C. The lane was cut out, rotated 90°, and set in the second dimension agarose (1% in 1 x TBE supplemented with 0.3 μg/ml of ethidium bromide). The second dimension was run at 6 V/cm for either 10 (for 4 kb fragment) or 14 hours (for 8 kb fragment) at 4°C. The DNA was transferred to a positively charged nylon membrane (GE heathcare hybond+) by Southern blotting and cross-linked using UV-light.
(d) Radioactive detection of DNA
DNA was detected using 32 P α-dATP incorporated into a PCR fragment (using Stratagene Prime-It II random primer labelling kit). Probes were hybridised to membranes overnight at 65°C in 10 ml of Church-Gilbert buffer (7% SDS, 0.5 M NaH 2 PO 4 , 1 mM EDTA, 1% BSA). Membranes were washed for 15 minutes at 60°C in 2X SSC (1X SSC: 0.15 M NaCl, 0.015 M Na-citrate) supplemented with 0.1% SDS and then 30 minutes in 0.5 x SSC supplemented with 0.1% SDS. Labelled membranes were exposed to GE healthcare storage phosphor screens and scanned using a Molecular Dynamics Storm 860 phosphorImager scanner. Images were quantified using GE healthcare ImageQuant TL.
(e) Analyses of loss of DNA following Southern blotting
To quantify the loss of DNA, the data obtained from lacZ probing were normalised to the data obtained from the probing of the cysN control fragment, located on the opposite side of the chromosome. The background signal was subtracted and the data were normalised to the no palindrome control.
Supporting Information
S1 Table
DNA oligonucleotides used.
(DOCX)
S2 Table
Bacterial strains used.
(DOCX)
S1 Fig
Replication profiles of individual biological replicates of Δ recG mutants and Rec + strains.
Replication profiles of exponentially growing cultures of Rec + strains with (A) or without (B) the palindrome and a Δ recG mutant with (C) or without (D) the palindrome are shown. In each graph, log 2 of the normalized copy number of uniquely mapped sequence reads (log 2 DNA abundance) is plotted along the y -axis against replichore-formatted genomic coordinates along the x -axis. The directions of chromosomal replication are depicted with green and red arrows to indicate the left and right replichores, respectively. The relative positions of the replication termination sites ( terB terC and terA ), the dif site and the palindrome are shown for each plot.
(TIFF)
S2 Fig
Comparative analysis of replication profiles between biological replicates of recG mutants and Rec + strains of Escherichia coli .
Replication profiles across the genome of growing cultures of Δ recG mutants with and without the palindrome are shown in (A). The same has been shown for Δ recG mutants and RecG + strains with the palindrome in (B), and for Δ recG mutants and Rec + strains without a palindrome in (C). In all cultures the expression of SbcCD was induced for one hour prior to isolation of the DNA. In each graph, log 2 of the normalized copy number of uniquely mapped sequence reads (log 2 DNA abundance) is plotted along the y -axis against replichore-formatted genomic coordinates along the x -axis. The continuous and dotted lines represent biological replicates of the experiment. The directions of chromosomal replication are depicted either with a green arrow to indicate left replichore or a red arrow to indicate the right replichore. The relative positions of the replication termination sites ( terB , terC and terA ), the dif site and the location of the palindrome are shown for each plot. This analysis was carried out because of the notable difference in enrichment of mapped sequence reads on the two sides of the induced DSB in lacZ in the Δ recG mutant. All other duplicates correspond closely across their genome as do the two biological replicates with an induced DSB in lacZ in the recG mutant in the left replichore and the terminus region. The basis for the notable differences on the two sides of the induced DSB in the Δ recG mutant requires further investigation. Nevertheless, because both replicates show enrichment of sequence reads on both sides of the induced DSB we conclude that this particular behaviour is reproducible and we have presented the average relative enrichment in Fig 2 .
(TIFF)
S3 Fig
Divergent replication forks are elevated in a Δ recG mutant subjected to DSBs.
A. PvuII digestion map of the region 50 kb upstream of the palindrome locus. PvuII cutting sites and the distance between them are marked with black vertical arrows and numbers (in kb), respectively. The terB site and the ykgM . 3 probe are marked by a green shape and a blue line, respectively. B. 2-D native-native agarose gel electrophoresis. The DNA was detected using the ykgM . 3 probe. Some partial digestion products are visible on the gels. Strains used were DL5096 (Rec + lacZ :: 246 ykgM-terB ), DL5097 (Rec + lacZ + ykgM-terB ), DL6033 (Δ recG lacZ :: 246 ykgM-terB ), and DL6034 (Δ recG lacZ + ykgM-terB ). C. Quantification of the paused forks relative to the linear DNA. Proportion of signal at the terB over linear DNA was calculated. Then, the data obtained from palindrome containing strains were normalised to the data obtained from no palindrome control. Finally, the signal obtained from Rec + strain were subtracted from Δ recG sample. Error bars represent the standard error of the mean where n = 3.
(TIFF)
S4 Fig
Further quantification of X-spike and Y-arc intermediates.
Quantification of X-spike and Y-arc intermediates compared to linear DNA in the Δ ruvAB , Δ recG Δ ruvAB , and Δ ruvAB Δ recG priA300 strains subjected to DSBs (data from Fig 4E ). Error bars represent the standard error of the mean where n = 3.
(TIFF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
In many regions of the world, water deficits impose serious constraint on plant growth and crop productivity. Plant transpiration efficiency (TE) is critical to plant survival and has important implications for both carbon cycling and water balance. Plants have evolved a variety of ways of controlling TE; understanding this control is essential to underpin attempts to improve crop productivity with limited water availability. TE is affected significantly and variably by canopy characteristics and leaf anatomy (i.e. leaf thickness, mesophyll cell size and position, stomatal density) and activity (stomatal conductance). In Arabidopsis thaliana , ERECTA (ER) was demonstrated to regulate the development of leaf architecture, and be a major gene contributing to TE, ER was the major contributor to a locus for carbon isotopic discrimination (Δ) and was negatively related to transpiration efficiency [ 1 ]. Thus, understanding the ER genotypic variation of leaf traits will be valuable to in attempts to improve TE, photosynthesis and crop productivity.
ER is associated with numerous functions that affect plant development and TE [ 2 ]. The ER gene was first isolated from Arabidopsis thaliana and belongs to the receptor-like kinase family (RLKs) with an N-terminal extracellular domain and C-terminal intracellular kinase that transduces extracellular signals into the cells to control a wide range of physiological responses [ 3 , 4 , 5 ]. The role of ER has been examined by both forward and reverse genetic approaches. Mutations to ER in Arabidopsis conferred decreased TE, but ER complementation led to restoration of TE [ 1 ]. In transgenic tomato plants, the expression of a truncated ER protein from Arabidopsis (atΔKinase), increased the number of stomata per leaf, transpiration and photosynthetic rates [ 6 ]. Over-expression in Arabidopsis of the PdERECTA gene from Populus nigra L. (35S:PdERECTA) increased photosynthetic rate, whilst decreasing transpiration rate and thereby increasing water use efficiency (WUEi) [ 7 ]. Complete function loss of three ER -family genes ( ER , ER-LIKE1 ( ERL1 ) and ERL2 ) in Arabidopsis resulted in the generation of high-density stomatal clusters and a 50–200% increase of the stomatal index [ 8 ]. ER appears to play a central role in the epidermal cell differentiation signaling pathway, inhibiting stomatal development and leading to reduced stomatal density and conductance. Therefore, ER is a prime candidate gene for studying the natural diversity of TE and photosynthesis in crops.
Wheat is a major cereal crop in the world, and is cultivated in arid and semi-arid regions of the world, where water deficit and other environmental fluctuations limit its growth, development and yield. Since ER has been theorized to play a major role in plant development and TE for a number of species, this study investigates the multi-gene ER family in bread wheat and tests whether its expression correlates with transpiration efficiency (as evidenced by stomatal density, stomatal conductance, carbon isotope discrimination) and yield. The goal is to establish whether the TaER genes could be used in approaches to improve transpiration efficiency and yield in wheat.
Materials and Methods
Plant material, growth conditions and sampling
Forty-eight bread wheat varieties with diverse carbon isotope discrimination (CID) values were sown in October, 2013 in the experimental field at Northwest A&F University, Yangling, Shaanxi, China (N 34°10’, E 108°10’, 526 m elevation). Details on the 48 varieties are provided in S1 Table ; most of them are from the main wheat production areas of China, with two varieties (Drysdale and Quarrion) from Australia, which were characterized with low CID and high TE. Each variety was sown in 3 rows of 2 m length, with 25 cm between rows and 6.7 cm between plants. The wheat was grown without irrigation and dependent on the soil moisture and rainfall. The flag leaves of three plants of each variety were collected at heading (Z55) and grain-filling (Z73) stages [ 9 ], respectively. Epidermal samples taken from each leaf were used for monitoring the stomata density, and the remainder of each sample was quickly frozen in liquid nitrogen and stored at -80°C for preparation of total RNA.
Sequences identification and analysis of TaER genes
Two ER sequences from Triticum aestivum ( ER2 from chromosome 7B, JQ599261.2 and ER1 from chromosome 7D, JQ599260.2 [ 10 ]) were used to search for additional wheat TaER sequences with the BLASTN program on the URGI wheat genome database ( http://urgi.versailles.inra.fr/Platform ). Homologous sequences of TaER with at least 90% similarity were found on the chromosomes 6 and 7. These sequences were assembled with Geneious 6.0.5 software (Biomatters Ltd., USA), based on ER1 , ER2 and sequences from T . urartu and Ae . tauschii (progenitors of the A and D genomes, respectively) [ 11 , 12 ].
The phylogenetic tree of the six cDNA sequences was obtained using the Geneious Tree Builder software, and the amino acid sequences were deduced from the cDNA sequence of TaER genes. The functional domains of TaER were identified by a BLAST survey with the deduced proteins in SMART databank ( http://smart.embl-heidelberg.de ) and the Conserved Domain Search ( http://www.ncbi.nlm.nih.gov/Structure/cdd/wrpsb.cgi ).
RNA extraction and cDNA synthesis
Total RNA was extracted according to the manufacturer’s recommendation using RNeasy Plant Mini Kit (QIAGEN, Germany). RNA concentration and purity were tested by Gene Quant Pro spectrophotometry (Amersham Biosciences, USA) and agarose gel electrophoresis. cDNA was synthesized with oligo(dT) 12–18 by Super Script III Reverse Transcriptase (Invitrogen, USA) according to the manufacturer’s instructions. cDNA was stored at -20°C and then used for amplification.
Expression analysis of TaER by qRT-PCR
The six cDNA sequences of the two TaER genes were aligned to design specific primers for TaER1 or TaER2 , selecting regions with high similarity among the three copies of each gene but with diversity between TaER1 and TaER2 . These primers were used to differentiate the specific expression of TaER1 and TaER2 in the flag leaves of the 48 wheat varieties at heading (Z55) and grain-filling (Z73) stages. Three housekeeping genes of TaActin , TaSand and TaCell were used to standardize the background expression in wheat. Details of the primers used for TaER1 , TaER2 , TaActin , TaSand and TaCell were listed in S2 Table . Each cDNA sample was used for three technical replicates according to the specifications of the SYBR Premix ExTaq Kit (TaKaRa, China) using the real time PCR system ABI 7300 (Applied Bio systems, USA). The reaction system included 10μl 2×SYBR MIX, 0.3μl for each of the forward and reverse primer, 2μl (50ng) template cDNA, ddH 2 O up to 20μl. The amplification procedure included an initial step of 95°C for 20s, followed by 40 cycles of 95°C for 5s, 61°C for 30s. Data was analyzed using the formula:
N E = ( E X ) - C t , X ( E R ) - C t , R
Where NE is the relative expression of target gene, E is the primer efficiency, Ct value is collected where the fluorescence is above the threshold value, X indicates values from the target gene, R indicates the geometric mean of values from the three reference genes [ 13 , 14 , 15 ].
Measurement of stomatal density and flag leaf area
Leaf epidermal samples of the 48 wheat varieties were collected at the grain-filling (Z73) stage, from both the adaxial (top) and abaxial (bottom) surface of flag leaves from three plants. The leaf surface was brushed with 1 cm 2 of transparent nail polish for about 20 s, and covered with sellotape avoiding veins, if possible. The sellotape was then removed and placed on a microscope slide. Leaf epidermal samples were observed with a Zeiss Axiophot upright light microscope (Zeiss, Germany). Images were recorded using a QImaging Retiga Exi CCD digital camera (QImaging, Canada) and the MetaMorph Microscopy Automation & Image Analysis software (Molecular Devices, USA). An epidermal area free of debris was selected and oriented to allow as many stomata as possible inside the area of the image acquired (viewing area). Three images were collected from each slide. The total number of stomata in each image was counted and the average stomatal density (SD) of each flag leaf was estimated using the formula:
S D ( N o . m m - 2 ) = N u m b e r o f s t o m a t a ( N o . ) V i e w a r e a ( m m 2 )
The flag leaves of five plants in each plot were measured to investigate flag leaf area (FLA) at the grain-filling stage (Z73) using the formula:
F L A ( c m 2 ) = l e a f l e n g t h × l e a f w i d t h × 0.8
where leaf length and leaf width are the longest and widest dimensions of the flag leaf. [ 16 ]
Measurement of stomatal related traits
At heading (Z55) and grain-filling (Z73) stages, the photosynthesis rate (A), stomatal conductance (gs), transpiration rate (E) and instant water use efficiency (WUEi) of the flag leaf were measured with a photosynthesis system (Li-6400, USA). The leaves were wide enough to completely fill the chamber area. Conditions in the leaf chamber were as reference CO 2 concentration = 400 μmol mol -1 , PPFD = 1800 μmol m -2 s -1 , relative humidity 50–70% and block temperature = 20°C. Photosynthetic traits were measured for five plants of each of the 48 wheat varieties between 9:00 and 11:00 am in sunny and windless weather. After the measurements were concluded, samples were taken for TaER expression analyses.
Measurement of CID, BYPP and GYPP
At the Z75 developmental stage, the flag leaves of three plants were collected and dried to constant weight, ground into fine powder and sent for measurement of δ 13 C using an Isotopic-Ratio Mass Spectrometer (Delta-V advantage, Germany) in the Lab of Stable Isotopes, Chinese Academy of Forestry Sciences (Beijing). An additional standard, Pee Dee Belemnite (PDB), was also measured. The CID value (Δ) was estimated using the formula:
Δ ( ‰ ) = ( δ a - δ s ) ( 1 + δ s ) × 1000
Where δs is the δ 13 C of the sample, and δa is the δ 13 C of atmospheric CO 2 , δa = -8 ‰ [ 17 ].
After harvest (Z93), biomass yield per plant (BYPP) and grain yield per plant (GYPP) were also investigated as described by Chen et al. [ 18 ], using five replicate plants for each of the 48 wheat varieties.
Data analysis
Analysis of variance was used to assess variation of TaER expression, transpiration efficiency related traits, CID values and yield-related traits among the 48 wheat varieties. Varieties were grouped according to TaER expression following hierarchical cluster analysis of both TaER1 and TaER2 expression at the Z55 and Z73 developmental stages. Correlation analysis was performed between TaER expressions and these measured traits using the Pearson Product Moment Correlation test. All of the analyses used the SPSS Statistics Software version 19.0 (IBM SPSS Statistics, USA).
Results
Characterization of TaER genes in wheat
In previous work, the sequences of ER2 ( TaER_B1 ) and ER1 ( TaER_D1 ) were isolated by homology-based cloning with ER family genes, and localized on chromosomes 7B and 7D of bread wheat by using the nullisomic-tetrasomic lines of Chinese Spring [ 10 ]. A BLAST search using those cDNA sequences in the URGI databank revealed homologous sequences of TaER on the short arm of chromosome 7 and the long arm of chromosome 6. Copies of each gene, corresponding to the three bread wheat genomes (A, B and D,) were identified. TaER1 was located on the short arm of chromosome 7 in all three genomes and the genes were named TaER1_AS , TaER1_BS (equivalent to TaER_B1 ) and TaER1_DS (equivalent to TaER_D1 ). TaER2 was located on the long arm of chromosome 6 in all three genomes and the genes were named TaER2_AL , TaER2_BL and TaER2_DL . Alignment between TaER1 and TaER2 showed 75.4% identical nucleotides, while the three copies of TaER1 and TaER2 showed 94.4% and 95.9% identical nucleotides, respectively. Both TaER1 and TaER2 contained 27 exons and 26 introns, cluster analysis indicated that the copies of TaER genes were divided into two groups (corresponding to each of TaER1 and TaER2 ), and in each case the gene copies encoded by genome A were more similar to those encoded by genome D than to those encoded by genome B ( Fig 1 ).
10.1371/journal.pone.0128415.g001
Fig 1
Phylogenetic tree of TaER genes family.
Two genes ( TaER1 and TaER2 ) are encoded by each of the three genomes A, B and D (short arm of chromosome 7 for TaER1 and long arm of chromosome 6 for TaER2 ). Six copies of cDNA sequences of TaER were aligned, and phylogenetic links were obtained based on their homology. Numbers relate to the phylogenetic distance between each of the TaER gene copies.
Alignment of the deduced amino acid sequences of TaER revealed a predicted extracellular domain consisting of leucine-rich repeat (LRR) elements, a Serine/Threonine (Ser/Thr) protein kinase domain at the C-terminal, and a transmembrane region from amino acids 579 to 589, which shared the same conserved domains as other ER family proteins. Alignment of the predicted sequences of TaER copies showed the high homology in the kinase region but the variable extracellular and transmembrane regions. In TaER1, there were 93% identical amino acids across genomes with identical amino acids in the kinase region; in TaER2, there were 93.2% identical amino acids across genomes with 18 amino acids varied in the kinase region. The Ser/Thr kinase region showed 268 highly conserved residues and the 11 typical sub-domains were recognized by homology to other ER family proteins ( Fig 2 ).
10.1371/journal.pone.0128415.g002
Fig 2
Sequence alignment of TaER Ser/Thr kinase domain.
Red stars indicate different amino acid residues between chromosomes 6 and 7. The lack of black background indicates different amino acids among the genomes A, B and D. Roman numerals on top of sequences indicate the 11 subdomains of TaER protein kinases.
Expression of TaER1 and TaER2 in flag leaves of 48 wheat varieties
The expression levels of TaER1 and TaER2 in wheat flag leaves were higher at heading (Z55) stage than at grain-filling (Z73) stage, with similar expression patterns in both TaER1 and TaER2 at the two stages ( Fig 3 , S3 Table ). Significant variations were observed on the expression levels of TaER1 and TaER2 in the flag leaves at both stages among the 48 diverse bread wheat varieties ( Fig 3 , S3 Table ). Cluster analysis on the expression levels of two TaER genes, classified these varieties into three groups as: 5 varieties with high TaER expression in group I, 27 varieties with intermediate TaER expression in group II and 16 varieties with low TaER expression in group III ( Fig 4 , S3 Table ). The distribution of genotypes per group implied that there were relatively few genotypes with higher TaER expressions, but most genotypes with intermediate TaER expression across these wheat varieties.
10.1371/journal.pone.0128415.g003
Fig 3
Relative expression of TaER1 and TaER2 in flag leaves of 48 wheat genotypes at heading (Z55) and grain-filling (Z73) stage.
Group I: high TaER expression; Group II: intermediate TaER expression; Group III: low TaER expression. Uppercase letters represent significant differences among the three groups ( P <0.01). The reference genes were TaActin , TaCell and TaSand . Values are presented with the mean ± SD calculated from the formula [ 13 , 14 , 15 ]:
N E = ( E X ) - C t , X ( E R ) - C t , R
10.1371/journal.pone.0128415.g004
Fig 4
Cluster analysis of TaER expression in the 48 wheat genotypes ( P <0.01).
Group I: high TaER expression; Group II: intermediate TaER expression; Group III: low TaER expression. The numbers of Y axis present the code of the diverse wheat varieties ( S1 Table ), X axis presents the square of euclidean distance.
At both developmental stages, the expression levels of TaER1 and TaER2 were significantly different ( P < 0.01) among the three groups ( Fig 3 , Table 1 ). The expression of both genes was about 30–40% lower in group III compared to group I at both developmental stages. In each group, significant differences (P< 0.05) were also detected among different wheat varieties ( S3 Table ). The wheat variety Drysdale (No.42) showed the highest expression level of both TaER genes.
10.1371/journal.pone.0128415.t001
Table 1 TaER relative expression in the three groups of 48 wheat varieties
TaER genes
Development stage
Items
Grouping of 48 wheat varieties
Group I
Group II
Group III
TaER1
Heading(Z55)
Mean
0.68±0.004 A
0.57±0.004 B
0.46±0.005 C
Heading(Z55)
Minimum
0.66
0.50
0.38
Heading(Z55)
Maximum
0.70
0.61
0.50
Grain-filling(Z73)
Mean
0.56±0.005 A
0.46±0.004 B
0.38±0.005 C
Grain-filling(Z73)
Minimum
0.52
0.40
0.32
Grain-filling(Z73)
Maximum
0.58
0.53
0.46
TaER2
Heading(Z55)
Mean
0.60±0.002 A
0.49 ±0.003 B
0.39±0.004 C
Heading(Z55)
Minimum
0.59
0.44
0.33
Heading(Z55)
Maximum
0.62
0.56
0.44
Grain-filling(Z73)
Mean
0.57±0.006 A
0.44±0.003 B
0.34±0.004 C
Grain-filling(Z73)
Minimum
0.55
0.38
0.29
Grain-filling(Z73)
Maximum
0.61
0.51
0.39
Group I: high TaER expression; Group II: intermediate TaER expression; Group III: low TaER expression. Uppercase letters represent significant differences among the three groups ( P <0.01).
Correlation between TaER expression and TE related traits
The stomatal density of flag leaf (SD) and flag leaf area (FLA) were significantly different among the 48 wheat varieties at the grain-filling stage (Z73) ( S5 Table ). SD and FLA were also significantly different ( P < 0.01) among the three groups ( Fig 5A , Table 2 ). SD was the lowest and FLA was the highest in the group of varieties with high TaER expression (group I), within this group, the varieties Fengchan3 (No. 36) and Shijiazhuang8 (No. 3) had the lowest SD and the highest FLA. At both heading (Z55) and grain-filling (Z73) stages, negative linear correlations were found between TaER expression and SD; whereas positive linear correlations were detected between TaER expression and FLA ( Fig 5B , Table 3 ).
10.1371/journal.pone.0128415.g005
Fig 5
Correlation between TaER expression with SD of flag leaf and FLA.
A: Stomatal density (SD) of flag leaf and flag leaf area (FLA) of the 48 wheat genotypes; B: Regression analysis between TaER expression with SD of flag leaf and FLA. Group I: high TaER expression; Group II: intermediate TaER expression; Group III: low TaER expression. Uppercase letters represent significant differences among the three groups ( P < 0.01).
10.1371/journal.pone.0128415.t002
Table 2 Transpiration efficiency related traits and yield in the three groups of 48 wheat varieties
Traits
Grouping of 48 wheat varieties
Group I
Group II
Group III
SD (No.mm -2 )
49.24±0.15 A
53.37±0.18 B
57.79±0.28 C
FLA (cm 2 )
32.02±0.24 A
25.69±0.16 B
20.03±0.28 C
A (μmol.m -2 s -1 )
21.54±0.13 Aa
18.52±0.11 BCb
16.58±0.09 Cc
E (mmol.m -2 s -1 )
2.76±0.03 Aa
3.70±0.04 BCb
4.19±0.05 Cc
WUEi (μmol.mmol -1 )
6.79±0.07 Aa
5.67±0.04 BCb
4.95±0.05 Cc
CID (‰)
21.07±0.09 A
21.53±0.03 B
22.17±0.03 C
BYPP (g)
42.46±0.37 A
33.76±0.38 B
24.10±0.42 C
GYPP (g)
19.39±0.40 Aa
14.19±0.24 BCb
12.68±0.36 Cc
HI
0.46±0.04
0.43±0.01
0.54±0.02
Group I: high TaER expression; Group II: intermediate TaER expression; Group III: low TaER expression. Uppercase and lowercase letters represent significant differences among the three groups (uppercase P < 0.01; lowercase P < 0.05). SD: stomatal density (No. mm -2 ); FLA: flag leaf area (cm 2 ); A: photosynthetic rate (μmol. m -2 s -1 ); E: transpiration rate (mmol. m -2 s -1 ); WUEi: instant water use efficiency (μmol mmol -1 ); CID: carbon isotopic discrimination (‰); BYPP: biomass yield per plant (g); GYPP: grain yield per plant (g); HI: harvest index.
10.1371/journal.pone.0128415.t003
Table 3 Correlation coefficient between TaER expressions with transpiration efficiency related traits and yield.
Trait
TaER1
TaER2
Heading(Z55)
Grain-filling(Z73)
Heading(Z55)
Grain-filling(Z73)
SD
-0.697 **
-0.725 **
-0.422 **
-0.575 **
FLA
0.727 **
0.826 **
0.447 **
0.448 **
A
0.216 *
0.366 *
0.129 *
0.275 *
gs
-0.059
0.069
-0.157
0.096
E
-0.376 *
-0.486 *
-0.210 *
-0.304 *
WUEi
0.227 *
0.258 *
0.142 *
0.192 *
CID
-0.752 **
-0.793 **
-0.524 **
-0.578 **
GYPP
0.573 **
0.656 **
0.319 *
0.328 *
BYPP
0.651 **
0.658 **
0.489 **
0.494 **
HI
0.202
-0.095
-0.246
-0.335 *
Asterisks represent significant differences.
* P < 0.05;
** P < 0.01. SD: stomatal density (No. mm -2 ); FLA: flag leaf area (cm 2 ); A: photosynthetic rate (μmol. m -2 s -1 ); gs: stomatal conductance (mmol. m -2 s -1 ); E: transpiration rate (mmol. m -2 s -1 ); WUEi: instant water use efficiency (μmol. mmol -1 ); CID: carbon isotopic discrimination (‰); BYPP: biomass yield per plant (g); GYPP: grain yield per plant (g); HI: harvest index.
Variation in photosynthesis rate (A), stomatal conductance (gs), transpiration rate (E) and instant water use efficiency (WUEi) among the 48 wheat varieties ( S5 Table ) at grain-filling stage (Z73) was closely correlated with TaER expression levels in the corresponding groups ( Fig 6A , Table 2 ). Photosynthesis rate was higher, transpiration rate was lower and WUEi was higher in wheat varieties with high expression of TaER . The varieties Drysdale (No. 42) and Fengchan3 (No. 36) displayed the highest A and WUEi, and the variety Shijiazhuang8 (No. 3) had the lowest E.
10.1371/journal.pone.0128415.g006
Fig 6
Correlation between TaER expression and transpiration efficiency related traits.
A: Photosynthetic rate (A), transpiration rate (E) and instant water use efficiency (WUEi) of the 48 wheat genotypes; B: Regression analysis between TaER expression with A, E and WUEi. Group I: high TaER expression; Group II: intermediate TaER expression; Group III: low TaER expression. Upper and lowercase letters represent significant differences among the three groups (uppercase P < 0.01, lowercase P < 0.05).
Regression analyses confirmed that TaER expression was significantly and positively correlated with A and WUEi, while significantly and negatively correlated with E ( Fig 6B , Table 3 ), but was not significantly correlated with gs at heading (Z55) and grain-filling (Z73) stages. The significance of these correlations was stronger for TaER1 than for TaER2 .
Correlation between TaER expression and CID, BYPP or GYPP
The carbon isotope discrimination (CID) of flag leaf was determined at grain filling (Z73) stage, and the biomass yield plant -1 (BYPP) and grain yield plant -1 (GYPP) were determined after harvest (Z93) ( S6 Table ). CID, BYPP and GYPP were significantly different (P< 0.05) between groups of wheat genotypes characterized by different levels of TaER expression ( Fig 7A , Table 2 ). In general, cultivars with high TaER expression had low CID and high BYPP and GYPP, with group I varieties Drysdale (No. 42) and Fengchan3 (No. 36) showing the lowest CID and the highest BYPP. Correlation analysis indicated that CID was negatively correlated with TaER expression, but BYPP and GYPP were positively correlated ( Fig 7B , Table 3 ). At grain-filling (Z73) stage, TaER1 expression was more strongly correlated with CID, BYPP and GYPP than TaER2 ( Table 3 ).
10.1371/journal.pone.0128415.g007
Fig 7
Correlation between TaER expression with CID, BYPP and GYPP.
A: Flag leaf carbon isotope discrimination (CID), biomass yield per plant (BYPP) and grain yield per plant (GYPP) of the 48 wheat genotypes; B: Regression analysis between TaER expression with CID, BYPP and GYPP. Group I: high TaER expression; Group II: intermediate TaER expression; Group III: low TaER expression. Upper and lowercase letters represent significant differences among the three groups (uppercase P < 0.01, lowercase P < 0.05).
Correlation analysis among TE- and yield- related traits
Correlation analysis between the measured traits was performed in the 48 wheat varieties ( Table 4 ). The results showed that there were significant correlations (P< 0.05) among most traits, but gs did not correlate with SD, FLA or CID. Similarly, not all traits correlated with harvest index (HI), but significant correlations with HI were observed for A, gs and GYPP. GYPP was positively correlated with FLA, A, gs, WUEi, BYPP and HI, but negatively correlated with SD, E and CID.
10.1371/journal.pone.0128415.t004
Table 4 Correlation coefficient between transpiration efficiency related traits and yield.
FLA
SD
A
gs
E
WUEi
CID
BYPP
GYPP
HI
FLA
-0.543 **
0.467 **
-0.154
-0.625 **
0.637 **
-0.684 *
0.624 **
0.604 **
0.002
SD
-0.543 **
-0.578 *
-0.128
0.782 **
-0.650 **
0.775 *
-0.639 **
-0.539 **
-0.028
A
0.467 **
-0.578 *
0.217 *
-0.705 **
0.618 **
-0.600 **
0.556 **
0.348 *
0.369 *
gs
-0.154
-0.128
0.217 *
0.306 *
-0.260 *
0.038
0.316 *
0.246 *
0.325 *
E
-0.625 **
0.782 **
-0.505 *
0.306 *
-0.688 **
0.760 **
-0.557 **
-0.602 **
0.139
WUEi
0.637 **
-0.650 **
0.618 **
-0.260 *
-0.688 **
-0.699 **
0.481 **
0.620 **
0.003
CID
-0.684 *
0.775 *
-0.600 **
0.038
0.760 **
-0.699 **
-0.600 *
-0.608 *
0.201 *
BYPP
0.624 **
-0.639 **
0.556 **
0.316 *
-0.557 **
0.481 **
-0.600 *
0.324 *
-0.107
GYPP
0.604 **
-0.539 **
0.348 *
0.246 *
-0.602 **
0.620 **
-0.608 *
0.324 *
0.375 **
HI
0.002
-0.028
0.369 *
0.325 *
0.139
0.003
0.201
-0.107
0.375 **
Asterisks represent significant differences.
* P < 0.05;
** P < 0.01. SD: stomatal density (No. mm -2 ); FLA: flag leaf area (cm 2 ); A: photosynthetic rate (μmol. m -2 s -1 ); gs: stomatal conductance (mmol. m -2 s -1 ); E: transpiration rate (mmol. m -2 s -1 ); WUEi: instant water use efficiency (μmol. mmol -1 ); CID: carbon isotopic discrimination (‰); BYPP: biomass yield per plant (g); GYPP: grain yield per plant (g); HI: harvest index.
Discussion
ER has multiple effects on plant development, growth and physiology. Gene Ontology analysis revealed several pathways potentially mediated by ER [ 1 ]. Although most functional analyses have been performed in Arabidopsis , the presence of ER and ER-Like members in several crop species suggests that it may provide a novel breeding target. This study provided a detailed characterization of TaER genes in bread wheat and showed that a correlation exists between TaER expression and transpiration efficiency traits related to development and productivity of wheat.
TaER genes in wheat
Six copies of TaER , homologues of the well-characterized ERECTA ( ER ) genes in Arabidopsis , were identified in the genome of bread wheat (using data currently available in the public domain). The TaER1 and TaER2 genes were located on chromosomes (6 and 7) and occurred in the three wheat genomes (A, B and D). The six copies had similar predicted amino acid sequences, reflecting the relatively conserved evolutionary history of the ER family in bread wheat [ 2 ]. Cluster analysis suggested that sequences encoded by genomes A and D are more closely related than genome B. These results would suggest that the three copies maybe not be complementary, but likely have parallel functions. This hypothesis could be tested in future experiments by silencing the individual TaER genes.
TaER expression in wheat
Variations were observed in the expression levels of TaER1 and TaER2 among the 48 wheat genotypes. TaER1 and TaER2 had conserved amino acid sequences and probably overlapping functions. The 48 wheat varieties were clustered into three groups as high, moderate and low of TaER expression levels; these significant variations may regulate the change of agronomical traits during the wheat development. Most studies on the expression of ERECTA have been carried out in mutants of Arabidopsis and these did not provide evidence on the expressional diversity of this gene in a panel of germplasm. In this study, the intraspecific diversity in the expression of TaER genes has been shown in flag leaves of 48 wheat genotypes at two growth stages, which are sufficient to provide a potentially useful estimate of intra-specific phenotypic variability. This finding is in agreement with results for anatomical and physiological changes following overexpression of PdERECTA in Arabidopsis [ 7 ]. Moreover, as the total expression of the three copies in A, B and D genome of TaER1 and TaER2 were evaluated in this study, to clarify whether there were variations among the three copies, the expression patterns of the three copies in A, B and D genome for TaER1 and TaER2 were further evaluated in four genotypes with two for high expression and two for low expression of TaER1 and TaER2 , which showed highly similar expression patterns with that of TaER1 and TaER2 common expression, although there were variations among the three copies, especially for that of TaER1 at heading stage ( S4 Table ), this suggested that the total expression of TaER1 and TaER2 could be used to evaluate the diversity of TaER expression in different wheat varieties, and the variation seen within the 48 genotypes may be due to TaER expression diversity and that TaER is a critical factor in the wheat leaf development.
Effect of TaER on leaf anatomy
Compelling evidence already suggests that ER affects leaf development and stomata formation [ 19 ]. Here, SD and FLA varied greatly among the high, medium and low TaER expression groups within the 48 wheat genotypes, the expression of both TaER1 and TaER2 were negatively and positively correlated with SD and FLA, respectively. Since leaf anatomy is a primary determinant of water relations between the plant and the environment [ 20 ] and TaER expression is significantly correlated with leaf anatomy of SD and FLA traits, so we predict TaER similarly regulates the development of epidermal cells and affects stomatal density in wheat leaves. This could be investigated further in transgenic wheat plants in which TaER is either overexpressed or silenced.
Correlation of TaER expression with flag leaf traits was stronger at grain-filling (Z73) stage than that at heading (Z55) stage. It is possible that the development of leaf anatomy at heading stage is regulated by a combination of TaER in addition to other factors, such as TMM (TOO MANY MOUTHS) and EPF (EPIDERMAL PATTERNING FACTOR) [ 20 ]. The stronger correlation coefficients found for TaER1 than TaER2 indicates that the former has a stronger effect on the development of wheat leaf anatomy.
Association of TaER expression with transpiration efficiency related traits
The variation in SD and FLA probably led to the diversity in transpiration efficiency related traits seen among the 48 genotypes. In this study, TaER expression was positively correlated with A and WUEi, but negatively correlated with E. This is in agreement with previous studies in Arabidopsis . The link between TaER expression and photosynthesis rate offers another strategic target for breeding or biotechnological approaches to increase photosynthesis in wheat. However, there was only a weak correlation between TaER expression and gs. A possible reason might be that gs is not only determined by genetic variation but is also strongly influenced by environmental factors [ 21 ]. These results are consistent with TaER expression indirectly controlling photosynthesis.
As with SD and FLA, the correlation coefficients between TaER expression and transpiration efficiency related traits were higher at grain-filling (Z73) stage than that at heading (Z55) stage. At heading stage, the wheat organs were under rapid change, hence measured data were more variable. At grain-filling (Z73) stage, almost all of the vegetative organs were formed, measured data were less variable and correlation coefficients were higher. Moreover, TaER1 expression was more strongly correlated with the diverse transpiration efficiency related traits than TaER2 . These results support that TaER1 has a stronger effect on the regulation of wheat development than TaER2 . Silencing studies within the TaER family would provide more evidence for the functional effects of TaER1 and TaER2 .
Regulation of transpiration efficiency related traits and yield by TaER
CID provides a reliable integrated measurement of A, E and WUE, and is proposed as an indicator of transpiration efficiency in the long term, avoiding issues associated with instant measurements made under varying environmental conditions [ 21 ]. AtER was suggested as the major gene contributing 20–60% of CID variation in Arabidopsis . Masle et al. [ 1 ] showed that the gene product of ER influences significantly CID and WUE in Arabidopsis , with lower transcript levels relating to higher CID and lower WUE. Here, it was also found that genotypes with higher TaER expression levels had lower CID and higher WUEi of flag leaf in bread wheat. Correlation analyses also suggested that TaER expression was positively correlated with BYPP and GYPP. It is proposed that TaER expression could be used as a rapid and reliable criterion for selecting superior genotypes for wheat breeding.
Conclusions
Anatomical and physiological traits determining transpiration efficiency, such as stomatal density, photosynthetic and transpiration rates, carbon isotope discrimination and yield traits were measured to evaluate the correlation between TaER expression with transpiration efficiency related traits and yield in 48 wheat varieties. TaER expression was associated with higher flag leaf area (FLA), photosynthetic rate (A) and WUEi, lower stomatal density (SD), transpiration rate (E) and CID, higher biomass yield per plant (BYPP) and grain yield per plant (GYPP). In addition, TaER1 had a stronger effect at grain-filling (Z73) stage on the development of bread wheat than TaER2 . The functions of the two genes could be further investigated by independent silencing of the TaER genes.
Supporting Information
S1 Table
Name and planting region of the 48 wheat varieties.
(PDF)
S2 Table
Primer sequences used for expression analysis of TaER1 and TaER2 .
(PDF)
S3 Table
TaER relative expression in 48 wheat varieties.
(PDF)
S4 Table
Relative expression of three homologous copies in the A, B and D genome of TaER1 and TaER2 in 4 wheat varieties.
(PDF)
S5 Table
Transpiration efficiency related traits of 48 wheat varieties at grain- filling (Z73) stage.
(PDF)
S6 Table
Carbon isotope discrimination and yield-related traits of the 48 wheat varieties.
(PDF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Spatial variation in the strength and type of natural selection can generate local adaptation, whereby resident individuals possess higher relative fitness than individuals from other populations [1] , [2] . Biotic interactions have been a particular focus of studies of the presence and strength of local adaptation, and these studies have shown that host-parasite interactions (e.g. [3] ), plant-herbivore interactions (e.g. [4] ), and predator-prey interactions (e.g., [5] ) are all common sources of local adaptation. While abiotic factors as a driver of local adaptation have received less empirical attention, local adaptation has been documented with respect to factors such as geographic variation (e.g., [6] ), latitudinal gradients (e.g., [7] ), temperature (e.g., [8] ), and soil type [9] .
Here, we consider whether the New Zealand freshwater snail Potamopyrgus antipodarum possesses local adaptation to phosphorus availability. Potamopyrgus antipodarum is an important model system for the study of sexual reproduction (e.g., [10] ) and invasion biology (e.g., [11] , [12] ). This snail exhibits local adaptation to trematode parasites among lakes [13] , [14] , [15] , and among lake habitats [16] , [17] , and it also shows local adaptation in shell form among habitats differing in flow rate [18] , [19] , and predation by fish [18] . Invasive P. antipodarum also show population-specific responses in growth rate to different temperatures [11] suggesting local adaptation to temperature regime. The presence of distinct across-lake population genetic structure in P. antipodarum [20] increases the likelihood that across-population variation in the strength, direction, and/or efficacy of selection would generate local adaptation.
There is accumulating evidence that phosphorus availability is an important component of P. antipodarum ecology. Phosphorus often composes a relatively large fraction of organismal dry mass and is critical to growth and reproduction because the production of nucleic acids requires a relatively high input of phosphorus [21] . Two recent studies have demonstrated that growth in juvenile P. antipodarum is severely reduced by limited phosphorus [22] , [23] . One of these studies also demonstrated substantial genetic variation in the extent to which growth is reduced by low availability of phosphorus [23] , indicating that there is heritable variation for response to phosphorus limitation in P. antipodarum .
New Zealand lakes vary considerably in total phosphorus content [24] and phosphorus limitation of benthic algae (the main food source of P. antipodarum ) differs among lakes (Krist et al., unpublished), leading us to hypothesize that P. antipodarum might be locally adapted to the availability of phosphorus in its native New Zealand. In particular, we predict that local adaptation would reflect selection favoring phenotypes that minimize consequences of phosphorus limitation in low-phosphorus environments [25] .
To address whether P. antipodarum is adapted to local levels of phosphorus availability, we must first establish whether New Zealand populations differ in their responses to phosphorus limitation. To do this, we manipulated phosphorus content in the diets of juvenile P. antipodarum sampled from three New Zealand lakes. Because New Zealand P. antipodarum feature extensive ploidy variation [26] , and because ploidy level can affect response to phosphorus limitation through the material costs associated with producing phosphorus-rich nucleic acids [23] , [25] , we restricted our study to triploid individuals.
Materials and Methods
Potamopyrgus antipodarum were collected from rocks and aquatic vegetation with kick nets in January 2011 in the shallow regions of three New Zealand lakes (Brunner, Hawdon, Selfe) that have a high relative frequency of triploid snails. Potamopyrgus antipodarum is not an endangered or protected species, and necessary permits were granted by the New Zealand Department of Conservation and the Iowa Department of Natural Resources. The triploid component of lake populations of New Zealand P. antipodarum generally harbor very high genetic diversity [27] , [28] , [29] , meaning that even a single lake collection likely included many distinct triploid genotypes. After collection, all snails were transported to the University of Wyoming, where the experiment was conducted.
All of the snails in the experiment were <2.5 mm in shell length at the beginning of the experiment, below the 2.5–3.0 mm shell length threshold typically used to designate adult vs . juvenile P. antipodarum (e.g., [30] , [31] ). We used as many juvenile snails as were available from each lake, which ranged from 85 individuals (Brunner & Hawdon) to 96 individuals (Selfe), for a total of 266 snails at the start of the experiment.
From each lake, individual snails were randomly assigned to either the high or low phosphorus (“high P” or “low P”) diet treatment, such that about half of the snails from each lake sample were fed each diet treatment. We produced low-P (carbon:phosphorus ∼984; standard deviation (SD) for % carbon = 2.41, % phosphorus = 0.04) and high-P (carbon:phosphorus ∼239; SD for % carbon = 1.38, % phosphorus = 0.09) diets for P. antipodarum by manipulating the carbon:phosphorus ratio of the green alga Scenedesmus obliquus . The molar carbon:phosphorus ratios of the experimental diets were above (low P) or below (high P) the threshold elemental ratio [32] for phosphorus limitation, estimated at carbon:phosphorus ∼270 for P. antipodarum [22] . We have already demonstrated that growth of juvenile P. antipodarum is severely reduced by low-P S. obliquus , indicating that this diet treatment does cause phosphorus limitation in P. antipodarum [22] , [23] .
We fed each snail 0.0035 g dried algae (re-suspended in 1 ml of well water), equivalent to ad libitum food levels for juvenile P. antipodarum (high-food level in [33] ), 3 times per week. We measured shell length of all individuals every two weeks using an ocular micrometer on a dissecting microscope (Leica S6E). Snails were housed individually in round plastic cups (300 ml). We changed the water in each cup 2 times per week and cleaned the cups once a week. We housed the snails at 18°C on a 12-hour light and 12-hour dark cycle. After 81 days of the diet treatments, a similar duration to previous studies that have demonstrated clear effects of diet treatments on growth and reproduction in P. antipodarum (e.g., [34] , [35] ), we transferred all snails to the University of Iowa. We then measured final shell length to the nearest tenth of a millimeter under a dissecting microscope.
The New Zealand lakes we sampled contain both diploid sexuals, which are ∼50% males [36] , [26] , and polyploid asexuals, which are ∼2–5% males [26] , [37] . Because ploidy level can affect response to dietary phosphorus [23] and because male and female P. antipodarum grow at different rates (Neiman, unpublished data), we restricted our analysis to triploid females. However, because juvenile snails (<3.0 mm in shell length) cannot be sexed reliably (e.g., [38] ) and because ploidy determination in P. antipodarum requires sacrificing the individual (e.g., [26] , [37] ), we had to wait until the end of the experiment to sex each snail and determine ploidy. At this time, we determined the sex of each snail using presence (male) or absence (female) of a penis for each snail >3.0 mm in shell length. We then dissected each snail and removed and snap-froze head tissue of all snails without penises (females and juveniles) for ploidy determination via flow cytometry.
Following Neiman et al. [26] [37] , we prepared each sample for flow cytometry by grinding frozen head tissue in a solution containing 0.2 M Tris-HCl (pH 7.5), 4 mM MgCl 2 , 1% TritonX-100, and 4 ug/mL DAPI. This solution was filtered through a 70-micron nylon sheet and then run on a Becton Dickinson LSR II flow cytometer. We used the FL1 channel to assess the DAPI fluorescence (and thus the DNA content) of sample nuclei under a UV lamp. At the beginning and end of each flow cytometry run, we calibrated the machine with 20 µL of chicken red blood cells (Lampire Biological Labs, Pipersville, PA) treated and filtered as for the snail head tissue. We adjusted the gain of the flow cytometer so that the chicken standard peak was always centered on 80 FL1 units. Each standard and sample was run until a count of 10,000 events was achieved. We analyzed all results using FlowJo software (Version 8.8.7, Tree Star, Inc.).
We first used FlowJo software to confine data analysis for each flow cytometry sample to the peak of data points corresponding to intact nuclei of single cells in growth phase 1. We calculated the mean fluorescence for this peak region for each sample and standardized this mean by dividing it by the mean FL1 value of the chicken red blood cell standard used to calibrate that particular run. Because there is substantial population-level variation in genome size even within ploidy level [26] , we used a histogram of the distribution of these standardized FL1 values for each sample within each population to assign ploidy status. As previous analyses of ploidy in P. antipodarum have demonstrated [26] , [37] , [39] , flow cytometry readily distinguishes among diploid, triploid, and “tetraploid” (>3×) P. antipodarum .
We were able to use flow cytometry to determine the ploidy of all of the 266 snails in the experiment. Because we were focused on triploids, we excluded the 6 tetraploids and 83 diploids from subsequent analyses. We also omitted the one triploid male that we identified (out of the 88 adult triploid individuals sexed) for a total of 176 triploid individuals in our final dataset. Of these 176 triploid snails, 87 were >3.0 mm in shell length and had been confirmed by the absence of a penis as adult females; 57 snails were <3.0 mm in shell length, and were thus classified as juveniles. The other 32 snails were >3.0 mm in shell length but due to an unintended oversight were left unsexed. However, the overwhelming predominance of females (87 of 88 snails; 98.9%) amongst the adult triploids that we sexed means that all or nearly all of both the unsexed adult and juvenile triploid snails were also females (also see [26] , [37] ). We thus included all 176 of these snails in subsequent data analyses.
Paired samples t-tests indicated that there were significant ( p <0.05) increases in shell length between each consecutive set of shell length measurements ( p <0.002 for all comparisons; Fig. S1 ). Because this meant that snails were still experiencing rapid growth at the conclusion of the experiment, we used the total increase in shell length between the 81 days marking the beginning and end of the experiment as our primary response variable.
We used ln(final shell/initial shell length))/81 days to estimate specific growth rate (SGR) for each snail. We then used a two-way univariate ANOVA to determine whether the fixed factors of diet treatment and lake alone or in interaction affected specific growth rate and conducted Tukey posthoc pairwise comparisons to determine whether there were differences in SGR among lakes. We used a Fisher's exact test to compare the proportion of snails in the low-P vs . high-P treatments that crossed the 3.0 mm shell length threshold typically used to assign “adult” status in P. antipodarum (e.g., [30] , [31] ). All statistical tests were implemented with IBM SPSS Statistics (version 21.0) with the exception of the Fisher's exact test, for which we used http://graphpad.com/quickcalcs/contingency1.cfm .
Results
Diet treatment had a large effect on SGR of P. antipodarum ( F 1, 170 = 71.23, p <0.0001; Fig. 1 ), such that growth was ∼64% higher in snails fed the high-P diet (mean SGR = 0.008+/− 0.003 mm SD) than the low-P diet (mean SGR = 0.005+/− 0.002 mm SD). Significantly more snails attained the 3.0 mm shell length threshold in the high-P diet vs . the low-P diet treatment (Fisher's exact test, p <0.0001; Fig. 2 ), suggesting that the lower rate of growth experienced by P. antipodarum in the low-P diet treatment might also delay reproductive maturity (e.g., [22] ). Taken together, these data indicate that snails receiving the low-P diet experienced phosphorus limitation. Lake of origin also significantly affected SGR of snails ( F 2, 170 = 14.73, p <0.0001; Fig. 1 ). Posthoc pairwise Tukey comparisons indicated that this effect of lake was largely driven by the high SGR of snails from Lakes Brunner and Selfe relative to snails from Lake Hawdon ( Fig. 1 ).
10.1371/journal.pone.0085845.g001 Figure 1
Mean SGR of snails in the high-P vs . low-P treatments. Error bars are 95% confidence intervals.
Potamopyrgus antipodarum fed the low-P diet grew at a significantly lower rate than snails fed the high-P diet. There were also significant differences in SGR among lakes, driven by significantly higher SGR in Brunner (N = 41) and Selfe (N = 54) snails relative to Hawdon (N = 81; p <0.001 for both comparisons) and a marginally significant lake by diet treatment interaction ( F 2, 170 = 2.52, p = 0.083).
10.1371/journal.pone.0085845.g002 Figure 2
More snails attained the 3.0 P. antipodarum in the high-P vs . low-P diet treatment (Fisher's exact test, p <0.0001).
We also found a marginally significant lake of origin by diet treatment interaction ( F 2, 170 = 2.52, p = 0.083). Although we lacked statistical power to detect a significant interaction because of limited sample sizes in Lake Brunner [40] (20 and 21 snails per treatment in Lake Brunner, power = 0.60), Fig. 1 reveals large differences in the effect of the high-P vs . low-P diet on SGR among snails from different lakes. We thus compared the magnitude of the decrease in SGR from the high-P relative to the low-P diet for P. antipodarum across lakes, which represents the response of each population to phosphorus limitation. We found that the largest differences in SGR between the high-P and low-P diets were in snails from Lakes Hawdon and Selfe (Hawdon; mean SGR on high-P diet = 0.007, low-P diet = 0.003: Selfe; high-P diet = 0.009, low-P diet = 0.005) and the smallest difference was in snails from Lake Brunner (high-P diet = 0.008; low-P diet = 0.006; Fig. 1 ). This means that snails from Lakes Hawdon and Selfe experienced an approximately twofold increase in the severity of the response to phosphorus limitation relative to snails from Lake Brunner.
Discussion
Juvenile P. antipodarum fed a low-P diet grew at a substantially lower rate and reached an important body size threshold more slowly than juveniles fed the high-P diet. These growth rate responses likely translate into fitness differences because P. antipodarum females that grow more slowly reach reproductive maturity later than females that grow more rapidly [22] and because fecundity and the rate of achieving reproductive maturity are both positively correlated with body size in female P. antipodarum [22] , [41] , [42] .
This result is the first evidence for phosphorus limitation in field-collected P. antipodarum . We also demonstrated population-level variation in specific growth rate, and detected preliminary evidence that the sensitivity of P. antipodarum to phosphorus limitation may vary across lakes. This potential difference in response to phosphorus limitation may result from snails being locally adapted to the availability of phosphorus in their lakes of origin.
While studies measuring across-lake phosphorus availability and reciprocal transplant experiments between lakes that differ in phosphorus availability are needed to conclusively establish the presence (or absence) of local adaptation to phosphorus availability, the particularly severe responses of P. antipodarum from Lakes Hawdon and Selfe to limited phosphorus relative to Lake Brunner suggests the existence of the type of across-lake variation in sensitivity to limited phosphorus that would be expected under local adaptation. One testable prediction that follows from our results is that, under local adaptation, the availability of phosphorus in the diet of P. antipodarum from Lakes Hawdon and Selfe should be higher than for snails from Lake Brunner.
Evidence for other types of local adaptation in P. antipodarum (e.g., parasites [13] – [17] , flow rate [18] , [19] , temperature [11] ) combined with marked across-lake population structure of P . antipodarum in its native New Zealand [20] and substantial genetic variation in the extent to which growth of P. antipodarum is reduced by low availability of phosphorus [23] suggest that local adaptation to phosphorus availability is a plausible explanation for the variation in response to phosphorus limitation among lake populations of P. antipodarum that we found. In fact, our finding that SGR varies among lakes, despite the generally high diversity of asexual lineages that are expected within each lake, suggests that SGR may differ among lakes because of local adaptation.
Evidence for across-population variation in the response of P. antipodarum to phosphorus limitation is relevant to evaluating a hypothesis suggesting a potential connection between sensitivity to phosphorus limitation and the outcome of competition between P. antipodarum differing in ploidy level and/or reproductive mode [23] , [25] , [43] . In particular, we hypothesized that P. antipodarum with higher ploidy level should be more sensitive to phosphorus limitation because of the material cost of producing extra nucleic acids, which are rich in phosphorus [25] , [43] . In support of this hypothesis, we found that relative to triploids, tetraploid P. antipodarum suffered a much larger reduction in growth on a low-P relative to a high-P diet [23] . Because sexual P. antipodarum are diploid and asexuals are polyploid, this hypothesis may also contribute to understanding the distribution and maintenance of sexual reproduction. The next steps in evaluating this hypothesis require establishment of spatial variation in dietary phosphorus availability and in response of P. antipodarum to limited phosphorus. The data presented here provide preliminary evidence of the latter and set the stage for future studies that provide more rigorous tests of local adaptation to phosphorus availability in New Zealand populations of P. antipodarum .
Supporting Information
Figure S1
Mean SGR of snails in the high-P (closed circles) and low-P diets (open circles) for each population over 6 two-week intervals during the 81 days of the experiment. Error bars are 95% confidence intervals. SGR for snails from Lake Hawdon are missing for 4/26/2011 because snails were accidentally not measured on this date.
(EPS)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The protection of key areas for biodiversity at sea is not as widespread as on land and research investment is necessary to identify biodiversity hotspots in the open ocean. It is now globally accepted that spatially explicit conservation measures such as the creation of representative networks of marine protected areas (MPAs) is a critical step towards the conservation and management of marine ecosystems, as well as to improve public awareness [1] , [2] . Conservation efforts in ecologically rich and threatened ecosystems are especially needed, urgently for the Mediterranean Sea [2] . This marine ecosystem is particularly diverse showing both high degree of endemism (around 20–30%) [3] and high occurrence of threatened species [1] , covering only 0.3% of the global oceans while hosting 7% of the world’s marine species [3] . Moreover, species richness shows a NW to SE decreasing gradient for both invertebrates and vertebrates related to important environmental drivers such as salinity, temperature, and water circulation, with a highly heterogeneous distribution depending on the region considered (highest for vertebrates around Sicily, northwestern coastal and shelf areas) [4] . However, current protected areas (namely MPAs) do not constitute a representative network since most of them are located in shallow waters of the northern part of the basin and represent 3.8% of the total surface of the Mediterranean Sea [1] . This may be a consequence of the limited marine research efforts of several eastern and southern regions of the Mediterranean [4] , which could have delayed the implementation of protected areas in these biogeographic areas. Thus, ecologically important habitats of high conservation value (especially those identified in the southern and eastern Mediterranean), including pelagic habitats of highly mobile marine vertebrates [1] , should be protected.
The Mediterranean Sea has been exploited and modified for thousands of years (e.g. fisheries) and it hosts very large populations of pelagic top predators - resident and transient (e.g. tuna, swordfish, dolphins, whales and seabirds) [2] . Upper-trophic level predators have been suggested as good indicators of the ecosystem functioning [5] while they integrate marine food webs across space and time [6] . Therefore, they provide information on ecosystem changes (i.e. variations in prey availability may influence breeding population sizes [7] ). Seabirds are predators easy to monitor when visiting their land-based colony and the study of their at-sea distribution by using tracking devices is thus facilitated. They follow the distribution of their prey, which depends on abiotic and biotic factors interacting at different spatial and temporal scales [8] . Preys are patchily distributed, but different physical processes can retain prey at certain oceanographic features (e.g. eddies, river plumes, slope currents). Productivity at these oceanographic features can vary at different temporal and spatial scales and species can shift their distribution or habitat preferences due to environmental changes. However, year-to-year persistent productive areas are likely to be exploited recurrently by top predators over years [9] , [10] .
Here, we studied the at sea distribution of one of the most endangered Mediterranean seabirds, the Balearic shearwater Puffinus mauretanicus . Due to its small (ca. 3200 breeding pairs) and declining population (7.4% decrease per year [11] ), this pelagic species is currently listed as Critically Endangered on the IUCN Red List [12] . It breeds solely on the Balearic Islands, on Menorca (northern populations), Mallorca-Cabrera (central populations) and Eivissa-Formentera (southern populations). Present knowledge from at sea vessel-based observations of birds of unknown origin, suggests that this species has a coastal distribution over productive shelf areas [13] . Oceanographic features promoting areas of high productivity influences the distribution of small pelagic fish, which constitutes the main natural prey for Balearic shearwaters apart from exploiting fishing discards [14] . During breeding (March–June), Balearic shearwaters forage along the Iberian continental shelf where different mesoscale oceanographic features result in productivity hotspots [13] , [15] – [18] . They also forage around the breeding grounds in the Balearic archipelago (e.g. Menorca-Mallorca channel, South of Mallorca and the marine area surrounding Formentera and South of Eivissa) [13] , [16] , [18] . Furthermore, former satellite tracking data evidenced that Balearic shearwater from northern populations (Menorca) were visiting marine areas close to Algeria [16] . However, evidence was inferred from sparse locations of 3 tagged birds (out of 7) and another 3 tagged birds from central populations (Mallorca) foraged over the Iberian shelf and breeding sites, whereas there is no data available from southern populations (breeding in Eivissa-Formentera). Thus, more research is needed to confirm the importance of western North African waters for this critically endangered shearwater in the Mediterranean.
Our aim was to provide seminal information on the key marine areas of the southern population of Balearic shearwaters (the least known) to support conservation efforts by elucidating spatial connections between breeding and key marine areas. In a first step, we identify the main biogeographic areas where Balearic shearwaters from southern populations forage based on miniaturized satellite transmitters tracking data. Then, we develop species distribution models (SDM) in order to define those oceanographic features characterizing their key habitats in the western Mediterranean. We specially search for the persistence of identified oceanographic features across time series available in the study area. Finally, we discuss our results within the current conservation scenario and draw attention to the main threats for the species.
Materials and Methods
Tracking Data
Experiments comply with the current laws of the country, and were performed under the permission CAP03/2011 of the Direcció General de Biodiversitat ( Conselleria de Medi Ambient i Mobilitat, Govern de les Illes Balears ). The study was carried out on Illa de Conillera, Reserves Naturals des Vedrà, es Vedranell i els Illots de Ponent (west of Eivissa Island, western Mediterranean) during the 27 th and 28 th of May 2011. Tracking was conducted on adult Balearic shearwaters rearing large chicks in a cave where at least 10 nests were easily accessible. During this period, adults forage intensively in order to feed their chick while returning to the colony by night. In order to minimise our impact on chick provisioning and colony attendance, we decided to tag birds after they fed their chick while leaving the colony and capture them at the entrance of the cave just before they took off for the sea. Some non-breeding adults can also visit the colony at night at this period (as prospectors or early failed breeders) [19] . Therefore, the breeding status of the adults captured was only suspected as we did not directly observe chick feeding. We tagged 6 adults, 4 males and 2 females (sex was determined based on a species-specific discriminant function based on biometry measurements [20] ). Adults weighed on average 525 g (range: 480 – 630 g). We deployed Argos satellite transmitters PTT (Platform Terminal Transmitters) solar panel, three 5 g and three 9 g-devices, with a duty-cycle of 48-hour off period, with 10-hour on period (Microwave Telemetry, Columbia). PTTs were attached on the back feathers using solely Tesa Tape®. This method allows a rapid deployment of the tag and avoids a medium to long term impact on the bird, since tags are lost with feathers when birds moult at the end of the breeding season. The total mass of devices was below the recommended 3% threshold (range: 0.9–2%) [21] .
Previous to any analyses, we discarded those positions over land. Then, we used all Argos locations (accuracy classes A, B, 0, 1 to 3), after filtering positions above 70 km h −1 (McConnell, Chambers & Fedak 1992) which is the maximum GPS-based speed of the closely related Manx shearwater Puffinus puffinus [22] . Speed filtering led to the removal of the 27 % of the positions. We also explored an important feature of our tracking data, the duty cycle programme. Argos PTTs had a duty cycle programme (10 hrs on, 48 hrs off) that lead to alternating day-night transmission periods (see Text S1 ). After filtering the data, the number of locations per duty cycle ranged between 1 and 9 ( Text S1 ). Indeed, transmission period started around 5 am and 2 pm, while no transmission started between 9 am and 10 am ( Text S1 ). The number of 10-h on periods ranged between 8 and 12 with a range of 1.81 and 5.33 locations per duty cycle lasting between 0 and 10.09 hours ( Table 1 & Text S1 ).
10.1371/journal.pone.0035728.t001 Table 1
Summary of birds (sex and weight), PTT devices (weight and date of equipment) and transmission characteristics (date of first and last position, number of locations, percentage of discarded positions, number of active duty cycles and mean (range) number of locations per duty cycle).
Shearwaters Status
Sex
WB (g)
PTT Number
WD (g)
Date ofequipment
Date of first position
Date of lastposition
# LOC
LOC over CONT (%)
DISC LOC (%)
# DC
# LOC/duty cycle mean (min - max)
Breeder
Male
610
40872
9
27/05/2011 21∶00
29/05/2011 20∶22
29/05/2011 21∶13
2
1 (50.0)
NA
NA
NA
Breeder
Male (?)
480
40866
5
27/05/2011 21∶30
30/05/2011 2∶57
08/07/2011 6∶10
30
2 (6.7)
2 (7.1)
11
2.36 (1–6)
Breeder
Male
630
40895
9
27/05/2011 22∶30
29/05/2011 19∶27
26/06/2011 10∶11
95
23 (24.2)
8 (11.1)
12
5.33 (3–8)
Breeder
Female
555
40825
5
28/05/2011 21∶00
29/05/2011 16∶53
29/06/2011 10∶16
27
3 (11.1)
4 (16.7)
11
1.81 (1–4)
Breeder
Male
490
40870
9
29/05/2011 0∶00
29/05/2011 19∶24
23/06/2011 13∶24
73
15 (20.5)
9 (15.5)
11
4.45 (1–9)
Non-breeder
Female
520
40823
5
29/05/2011 0∶00
01/06/2011 4∶13
19/06/2011 22∶17
36
12 (33.3)
4 (16.7)
8
2.5 (2–4)
WB: weight bird. WD: weight device. LOC: location. CONT: continent. DISC: discarded. DC: duty cycle.
Important marine areas were identified generating density distribution maps using fixed kernel density using the ad hoc method of the ‘adehabitat’ package (i.e. bivariate normal kernel; smoothing factor h of 0.51) and a cell size of 0.0417° (to match the spatial resolution of the satellite imagery data) in R 2.12.2 [23] . The smoothing factor was chosen based on exploratory analysis comparing the bivariate normal kernel, the least-square cross validation and arbitrarily chosen values ( h = 1 and h = 2). The bivariate normal kernel method showed the best fit to our data of the western Mediterranean basin (see details in Text S2 ) and was further used for analyses. It is important to note that we explored the effect of the duty cycle on the kernel density estimates (see details in Text S3 ). The described duty cycle (10 hrs on, 48 hrs off) might provide locations in close proximity (during the 10 hr on-cycle) but then none for 2 days. Hence, the kernels may give more emphasis to the areas where the PTT transmitted, rather than properly reflect the proportion of time spent by the birds. We test this influence by using just one location per duty cycle (e.g. the centroid of all locations per duty cycle) and re-calculating the kernels to see whether they would change massively in size or location. Since kernel estimations did not change, we did not discard locations for further analysis. We also excluded the possibility of night-time activity periods heavily influencing the kernel density estimates (see Text S3 ).
Environmental Predictors
We selected the most biologically relevant environmental variables from known habitat selection of the species [13] , [24] . Oceanographic data were extracted corresponding to the central month of the study period (i.e. June 2011) for the western Mediterranean, from the Environmental Research Division, Southwest Fisheries Science Center and US National Marine Fisheries Service ( http://coastwatch.pfel.noaa.gov/coastwatch/CWBrowserWW180.jsp ). Dynamic oceanographic variables such as sea surface temperature (SST, °C, as a proxy of water mass distribution) and chlorophyll a concentration (CHL, mg m −3 , as a surrogate of marine productivity) were extracted from MODIS. We extracted dynamic variables from each location from the corresponding raster of June 2011. Bathymetry (static variable; BAT, m, as a proxy of coastal versus pelagic domains) was extracted once. All variables were aggregated to match the standard grid (0.0417° cell size). Additionally, we estimated their spatial gradients by estimating their Proportional Change (PC) within a surrounding 3×3 cell (0.0417°×0.0417°) grid using a moving window as follows: PC = [(maximum value − minimum value) * 100] / (maximum value). This dimensionless metric expresses the magnitude of change in each habitat variable, scaled to the maximum value [13] . The spatial gradients of chlorophyll a (CHLG) and sea surface temperature (SSTG) indicate the presence of frontal systems, whereas the gradient of bathymetry (BATG) reflects the presence of topographic features (e.g. shelf break or seamount). To account for the influence of central-place foraging shearwaters [25] , we included the distance between each grid cell and the colony (COLONY).
Identifying Oceanographic Features Driving Shearwater Distribution
Data preparation
Prior to modelling, strongly ‘correlated’ (Spearman rank correlation coefficient, |r s |>0.6) predictors were identified by estimating all pair-wise Spearman rank correlation coefficients. This exploratory step is necessary since correlation between predictors might produce spurious results. High correlation was found for BAT-BATG and SST-COLONY pair-wise correlation coefficients. Therefore, we excluded BATG and SST from further analysis, based on previous knowledge on habitat selection of the species [13] , [24] .
Species distribution modelling
Species distribution model was performed with Maximum Entropy (MaxEnt) modelling based on only presence data (version 3.3.3 ( http://www.cs.princeton.edu/~schapire/maxent/ [accessed 5 August 2011]) with default parameters (modelling script will be made available by email). It is considered one of the best modelling techniques [26] and uses background samples of the environment rather than absence locations to estimate environmental relationships. Background samples were drawn within the western Mediterranean (latitudinal range: 34.73°N–42.86°N; longitudinal range: 5.70°W–4.88°E) to facilitate a valid comparison with previous habitat models within the known foraging range of the species [13] , during the central months of the study period (i.e. June 2011). Although Maxent can fit complex relationships to environmental variables, we chose to fit only linear and quadratic relationships due to the difficult interpretation of other more complex relationships. Maxent is robust to small sample sizes [27] , with spatial positioning errors [28] , and to spatial mismatch between locations and environmental variables [29] , [30] . Moreover, it gives the probability distribution of maximum entropy taking into account data available on species distribution and the environmental conditions across the study area [31] , [32] .
To minimise the influence of any individual on the population-wide model, we randomly selected an equal number of locations for each bird based on a boostrapping procedure [30] . It is important to note that we developed habitat models for breeding birds and, in turn, we did not include locations of the suspected non-breeder. The number of locations per bird was determined by the minimum value after applying the speed filter (20 locations), totalling 80 locations for developing species distribution models. Therefore, we took 100 random draws for those birds that had more than 20 locations without replacement. Due to the duty cycle of the tracking devices, autocorrelation of satellite positions was limited and no more filtering was necessary [30] . We ran Maxent on the randomly selected satellite positions 100 times. We calculated the mean of the 100 Maxent predictions to obtain an average prediction and coefficient of variation of predictions [30] . In addition, we evaluated the contribution of the environmental variables to the Maxent model based on a jackknife procedure, providing the explanatory power of each variable when used in isolation.
Model evaluation
To assess the predictive performance of SDMs, we evaluated each Maxent prediction using the Area Under the receiver operating characteristic Curve (AUC) [33] . AUC evaluates how well model predictions discriminate between locations where observations are present and absent (i.e. the presence locations and background in our study). AUC can range from 0 to 1. An AUC of 0.5 indicates that model performance is equal to that of a random prediction, whereas values from 0.5 to 1 with the following model predictive performance classification: >0.9 excellent, 0.9–0.8 good, 0.8–0.7 reasonable, 0.7–0.6 poor and 0.6–0.5 unsuccessful [34] . We applied two cross-validation procedures running 100 replicates: (1) an internal validation using data only from Eivissa (i.e. using original data) and (2) an optimal validation using an independent dataset [35] .
In the case of the internal validation, we randomly selected 80 locations and trained the model, while the resulting model was tested on the remaining 79 locations (total sample size of 159 locations). Regarding the optimal solution, an independent tracking dataset was available for birds from northern and central populations (Menorca and Mallorca, respectively) tagged 10 years ago (June 1999–2000) [4] . We developed a SDM with data from Eivissa in predicting distribution patterns of birds from Menorca (the population with enough sample size). Working on two spatially distinct groups (Eivissa and Menorca) allowed us to assess the model transferability in time and space. SDMs were trained with data from Eivissa and the model was then used to predict the distribution of birds from Menorca.
For both evaluation procedures, the AUC was estimated for each simulation and the mean, upper and lower 95% confidence interval (CI) of the AUC were used as a cross-validation measure of the predictive performance of the models [36] . If the lower 95% CI limit did not include the 0.5 value, there was evidence that SDMs were able to accurately predict beyond training dataset.
Oceanographic Persistence at Key Marine Areas
Once key marine areas were identified, we intended to estimate the persistence of significant dynamic oceanographic features. After distribution modelling (see Results Section), chlorophyll a was found to be the most important dynamic environmental variable. Since the presence probability of Balearic shearwaters increased with this dynamic variable, we quantified its persistence across the study area during the annual mean to assess its predictability for key marine areas. We hypothesised that those marine areas consistently identified as productive (indicated by high concentration of chlorophyll a ) might be visited by Balearic shearwaters (and marine top predators in general) from one year to the next.
We extracted the longest time series of chlorophyll a (July 2002–July 2011). General persistence across the study area was analysed using a 3-step approach. Firstly, we obtained an average annual map and identified those pixels with high nutrient concentration (CHL >0.3 mg m −3 ) [37] . Then, we counted for each pixel in how many years (maximum of 8 corresponding to the CHL time series) the concentration corresponded to high or low concentration. In addition, we extracted monthly CHL values from the main key areas identified by kernel analysis to compare oceanographic conditions between geographically distant important marine areas.
Results
Satellite Tracking and Kernel Analysis
A total of 263 satellite positions were recorded from the 6-tagged shearwaters during the transmitter emission period (range: 0.04–39.13 days). Summary statistics of tracking devices can be found in Table 1 . One of the PTTs provided only 2 locations over the Iberian continental shelf. From the remaining 5 PTTs, we discarded between the 6.7 % and 33.3 % of locations that were over land from which we disregarded between the 7.1 % and 16.7 %, after applying the speed filter. Overall, we discarded the 13.1% of the total locations after filtering by the speed threshold. Tracking devices were lost when birds started moulting, which took place between the 19 th June 2011 and 8 th July 2011.
Breeders remained in the western Mediterranean basin, only one -suspected non-breeder-individual travelled to Portuguese coastal waters (PTT 40823, see Figure 1 ). After deployment on the 29 th May 2011 night, this shearwater foraged over the southern sector of the Ebro Delta 3 days later (1 st June 2011) and 2 days later (3 rd June 2011) was entering the Atlantic Ocean through the northern coast of the Strait of Gibraltar, travelling more than 400 km from Cape Palos to Málaga in only 8 hours. Then, the same bird reached the northern coast of Portugal (Aveiro) on the 5 th June 2011 after at least 2 days travelling 900 km and stayed until the last position on the 19 th June 2011 (see Figure 1 ).
10.1371/journal.pone.0035728.g001
Figure 1
Satellite tracking of Balearic shearwaters during the chick-rearing period of 2011 from a southern breeding population (Conillera Island, western Mediterranean).
Breeders stay in the western Mediterranean basin, whereas one non-breeder travels and stays in Portugal (violet points). Background values represent depth. Studied colonies are represented by white triangles.
Breeders commuted between the colony and several areas of shallow waters, the Iberian Peninsula and also the Algerian and Moroccoan waters ( Figure 1 ). Filtered positions were assigned to different biogeographic areas ( Figure S1 ): continental shelf around the colony (Eivissa, EIV), Iberian Peninsula (IBE), Algeria (ALG) and Morocco (MOR), as well as commuting between continental shelves. Commuting between Eivissa or Iberian and Algeria lasted on average 6 h and 20 min, mainly during daylight (n = 4; see examples in Figure S2 ). The Algerian continental shelf accounted for the highest number of locations, followed by the Iberian Peninsula and commuting (represented by the percentage of locations in each biogeographic area) locations. One bird (40895) strongly influenced this pattern since locations of this bird represented 75% of locations in Algeria. This individual also visited the eastern coast of Morocco. Kernel analysis identified two main key marine areas (indicated by the 50% UD) for southern Balearic shearwaters: marine area around Cape La Nao and Arzew Bay in the Iberian and Algerian continental shelves, respectively ( Figure 2 ).
10.1371/journal.pone.0035728.g002
Figure 2
Filtered locations (each shearwater represented by different colours) and density contours (95%, 75% and 50% UD represented by black, dark grey and light grey lines, respectively) resulting from kernel estimation of the distribution of southern Balearic shearwaters during the chick-rearing period of 2011.
Two main marine areas (indicated by the 50% UD) are identified: marine area around Cape La Nao and Arzew Bay in the Iberian and Algerian continental shelves, respectively.
Shearwater Distribution Modelling
Distribution models for southern Balearic shearwaters yielded reasonable model performance (AUC (CI 95%): 0.70 (0.66–0.73)). The variables that most contributed to explain southern shearwater distribution were BAT followed by COLONY, CHLG and CHL ( Figure 3 ). Regarding response curves, the relationship between presence probability and environmental variables differed as illustrated in Figure 4 . Shearwater presence probability decreased linearly with BAT, whereas non-linear relationships were found for remaining variables. Maximum presence probability was found at 200 km from the colony, lower CHL values, 20% of CHLG and 1% of SSTG. In summary, presence probability was higher over the continental shelf to a certain distance to the colony in association to frontal systems of productive areas.
10.1371/journal.pone.0035728.g003
Figure 3
Variable importance estimated by the jackknife test.
Bars indicate the explanatory power (in terms of gain) when only a single predictor is included in the model with 95% confidence intervals.
10.1371/journal.pone.0035728.g004
Figure 4
Response curve illustrating relationship of presence probability to environmental variables.
Higher values correspond to higher probability of occurrence. These curves show how the logistic prediction of a particular variable changes, keeping all other environmental variables at their average sample value. Grey lines show the output of the 100 iterations, while the black line represents the mean bootstrap value.
Model predictions matched observed patterns within the range of southern Balearic shearwaters and identified key marine areas beyond the training dataset in June 2011.
Shearwaters were found with higher probability along three continental shelves: Balearic Islands (i.e. breeding colonies), Iberian and Algerian coast (dark red areas in Figure 5a ). Those areas were consistently identified as important due to the low coefficient of variation in predictions ( Figure 5b ). Around breeding sites, all marine areas showed high presence probability excepting the North coast of Mallorca and Menorca. Along the Iberian continental shelf, presence probability was higher from the Ebro Delta until Cape Palos, as well as from Nador (Morocco) to Argel (Argelia) along the narrow continental shelf of North Africa.
10.1371/journal.pone.0035728.g005
Figure 5
Species distribution modelling output for southern Balearic shearwaters distribution in June 2011: (a) mean prediction and (b) coefficient of variation.
The position of the studied colony (white triangle), the two key marine areas (indicated by the 50% UD) around Cape La Nao and Arzew Bay, and the 200 m isobath (i.e. the limit of the continental shelf) are shown (black line). In addition, limits of current Protected Areas are placed by yellow polygons in the western Mediterranean [56] , as well as Spanish marine Important Bird Areas (mIBAs) for Balearic shearwaters (BASH) (dark green polygons), other Spanish mIBAs (light green polygons) and potentially important (PI) mIBAs in international waters (violet polygons) [18] .
Regarding model evaluation, the internal validation showed a reasonable model performance (AUC (CI 95%): 0.70 (0.67–0.74)), as well as the external validation (AUC (CI 95%): 0.67 (0.64–0.71). Thus, the SDM developed with data from Eivissa had the ability to predict distribution patterns from birds from Menorca.
Productivity Persistence at Key Marine Areas
We quantified the persistence of the most important dynamic variable (CHL) across the study area over 8 years ( Figure 6a ). We found that high CHL values (>0.3 mg m −3 ) occurred consistently along the coastal areas of the study area (bluer areas), with a break South of the Cape La Nao until the Alboran Sea. The central-southern parts of the western basin were the least persistent areas (clearer areas). Within the distribution range of Balearic shearwaters, areas with high presence probability matched highly predictable productive areas (compare Figure 5a and 6a ). Regarding the two core areas identified by kernel analysis (areas covered by red lines in Figure 6a ), CHL values in Algeria were on average 14% higher than in Iberia, especially after 2009 ( Figure 6b ). Seasonal patterns were similarly reproduced in both biogeographic areas.
10.1371/journal.pone.0035728.g006
Figure 6
Oceanographic persistence at key marine areas.
(a) Persistence of chlorophyll a values in the western Mediterranean during the breeding period (February–June) of Balearic shearwaters. Values were obtained from the MODIS sensor between July 2002 and July 2011. For each pixel, we counted in how many years (maximum of 8) the concentration corresponded to high nutrient concentration (>0.3 mg m −3 ). Bluer areas correspond to persistent productive areas (check number of years on the scale bar). The position of the studied colony (white triangle), and the two key marine areas (indicated by the 50% UD) around Cape La Nao and Arzew Bay are placed. Limits of the continental shelf (200 m isobath) are defined by the black line, whereas Economic Exclusive Zones are delimited by dotted white lines [57] . In addition, limits of current Protected Areas are placed by yellow polygons in the western Mediterranean [56] , as well as Spanish marine Important Bird Areas (mIBAs) for Balearic shearwaters (BASH) (dark green polygons), other Spanish mIBAs (light green polygons) and potentially important (PI) mIBAs in international waters (violet polygons) [18] . (b) Time series of chlorophyll a anomalies estimated for the two main marine areas based on kernel analysis. For each key marine area, we computed the chlorophyll a anomaly by estimating the global mean value and subtracting the global mean value to corresponding month.
Discussion
The Algerian Current: An Important Dynamic Feature for Balearic Shearwaters
Our tracking study clearly indicates that Balearic shearwaters do not only forage along the Iberian continental shelf, but also in more distant key marine areas located along the northern African coast, as suggested by early satellite tracking from Menorca [12] , [16] . In fact, birds from Eivissa (located closer to the North African coast than Menorca) recurrently commute to the northern African coast (especially the Algerian coast, but also Morocco) at the end of the breeding season. The reason why shearwaters from Eivissa do not only forage over the closest productive area along the Iberian continental shelf could be related to the marine productivity of the western northern African coast and the relative proximity from the colonies (300 km). The dynamic of the Atlantic inflow along this biogeographic area is characterised by the Algerian Current where mesoscale instabilities create zones of enhancement of productivity [38] . In this area, chlorophyll a concentration is higher at cyclonic eddies linked to higher zooplankton abundance [39] . Across the current, large horizontal gradients exist in hydrography (e.g. salinity) as well as in phytoplankton and zooplankton biomass and species composition [40] , [41] . Productivity is particularly high at the offshore band of the Algerian Current where high concentrations of nutrients and chlorophyll a are related to high zooplankton biomass levels and abundances [38] , [42] .
Few data are available on the distribution of the main prey of Balearic shearwaters along the Algerian coast (e.g. small pelagic fish [14] ). Anchovy Engraulis encrasicholus and sardine Sardina pilchardus represent the most important small pelagic fish target for Algerian fleet (in terms of biomass) [43] , but captures exhibited great fluctuations during the 1970–2008 period ( Figure S3 ). Since landing data on small pelagic fishes can be used as a proxy of their availability in the system [44] , we compared total landings of anchovy, sardine and sardinella Sardinella aurita for Algeria and Spain to compare prey availability between Iberian and Algerian continental shelves [45] . Overall, all three species seemed to be available in the system over the 1970–2008 period and slight differences might be due to differences in number of vessels and technological development.
The present study complements the broader ecological perspective of foraging movements of Balearic shearwaters providing information on a less studied area. From previous studies, we know that breeding Balearic shearwaters repeatedly exploit the same foraging grounds along the Iberian continental shelf in relation to predictable resources (i.e. mesoscale oceanographic features) [13] , [15] , [17] , [24] , [46] . Indeed, central-place foraging Balearic shearwaters continually commute between the less productive waters around the breeding colonies (Balearic Islands) and the highly productive waters of the shelf-slope areas of the Iberian Peninsula and the northern African coast ( Figure S2 ). Commuting seems to be a common type of movement of pelagic birds within temperate and polar regions, and might suggest that breeding seabirds ‘know’ where to find food, probably from previous experience [47] . Within a coarser temporal and spatial scale, prey patches are likely to be scattered within mesoscale features [47] shifting depending on physical and biological drivers (e.g. river run-off, small pelagic fish spawning, fishing) [48] .
Transboundary Conservation Target Areas
Balearic shearwaters appear to favour shelf areas with dynamic oceanographic features. While static systems can easily be protected by designating areas of interest, dynamic habitats (e.g. eddies, frontal systems, currents) are much more problematic for conservation management. By combining tracking studies and habitat modelling, now it is possible to define the location of key marine areas, provided that models show a reasonable predictive performance and the validation with additional observations support model predictions. Within the process of identifying ecologically important marine areas, once key areas are located the next step is their protection. For southern Balearic shearwaters, key marine areas appear to lie mainly under the jurisdiction of Spain and Algeria (i.e. Exclusive Economic Zone, see Figure 6a ) [12] .
Given Balearic shearwater threatened status, extending protective measures beyond the breeding sites to the marine environment should be a priority, and marine zoning is essential in this way to prioritise those areas that are preferentially used by the species. Key marine areas over the Iberian continental shelf (e.g. centred in Cape Creus, Ebro Delta and Cape La Nao) have been already identified as marine Important Bird Areas (IBAs) for the Balearic shearwater and other seabird species ( Figure 6a ) [18] , and are currently in the process of designation as marine Special Protection Areas (SPAs) by the Spanish government, in commitment with the European Union Bird Directive (2009/147/EC) ( http://www.magrama.es/es/biodiversidad/participacion-publica/PP_borrador_orden_zepa_marinas.aspx ). Once finally designated, these marine SPAs will automatically be considered as part of Natura 2000, the network of protected sites across the European Union. These areas would not only protect targeted study species, but also their underlying habitat. It contrasts highly with the southern Mediterranean waters. In the case of Balearic shearwaters, there was little information on the likely location of key marine areas along the North African coast. Thus, this is the first study identifying key marine areas for the conservation of Balearic shearwaters in the south-western Mediterranean Sea from Nador (Morocco) to Alger (Algeria). Southern Mediterranean key areas should be incorporated into a transboundary conservation initiative to constitute an effective network of protected sites to be considered in further Marine Protected Areas proposals such as Specially Protected Areas of Mediterranean Importance.
In these areas, it would first be important to obtain information on the interactions between shearwaters and the local fisheries, to identify potential threats such as by-catch. In a second step, conservation measures should include a compulsory fishery observer program to record potential bycatch level since direct observations are important to detect threat by fishing. Quantification of Balearic shearwater bycatch is sparse and (limited) observer programmes often have reported low bycatch rates, but when bycatch occurs up to 50 birds or more can get entangled in a single line, due to the gregarious behaviour of the species [49] , [50] . This is a serious and direct threat for Balearic shearwaters given the high overall mortality experienced by the species [11] .
Such dramatic incidental captures occurred in the Spanish Mediterranean bottom longline vessels [51] – [53] and Portuguese purse-seiners [12] , and most likely also occur in other areas. The puzzle still gets more complicated when considering interaction between fisheries: seabird bycatch in both pelagic and demersal Mediterranean longline vessels increases when trawlers are not operating (i.e. less discards are available) [47] , [52] , pointing towards the need of an integrated multi-fisheries management approach [12] .
Even if some information on bycatch is available, still more research is needed for the northern Mediterranean since systematic surveys are necessary in order to clearly identify the spatio-temporal window of those interactions. Perspectives are still worse for the southern Mediterranean, where there is no information available for bycatch [53] . In fact, due to the current severe conservation concern of the species the Spanish Government has now advised the inclusion of Balearic shearwater in the Agreement on the Conservation of Petrels and Albatrosses (ACAP) [54] , which is a multilateral agreement for the conservation of Southern Hemisphere species. In fact, it will be the first European species included in this agreement. Thus, we highlight the urgent need to assess the interaction of Balearic shearwaters with the longline fishery within their key marine areas not only during breeding in the western Mediterranean, but also during the post-breeding period along the Mediterranean and North Atlantic [55] .
Conclusion
Transboundary conservation efforts are needed for the critically endangered Balearic shearwater. This is the main conclusion of the present study which tracked for the first time breeding birds from the southern population. Highly mobile animals are not attached to administrative boundaries and it is therefore essential to overcome legal limitations to provide full protection to highly threatened species. Within this framework, our results match previous studies in identifying similar key marine areas for the species over the Iberian continental shelf and improve our understanding by identifying key marine areas for Balearic shearwaters in southern Mediterranean waters. Along this biogeographic area, kernel analysis identified Arzew Bay as key marine area for tagged birds. Thanks to SDMs, we additionally were able to predict Balearic shearwaters distribution beyond observed data identifying all close bays along the northern African coast from Nador (Morocco) to Alger (Algeria) as potential key marine areas. These southern Mediterranean key areas could be integrated into a supranational conservation initiative to develop a successful network of protected sites across Mediterranean waters. We therefore highlight the importance of tracking studies and the establishment of long-term studies in order to comprehend how the current changing environment will impact on the distribution of species of high conservation concern in the future.
Supporting Information
Figure S1
Biogeographic area visited by each breeding shearwater.
(DOC)
Figure S2
Examples of a Balearic shearwater commuting.
(DOC)
Figure S3
Time series of small pelagic fish captures in the western Mediterranean.
(DOC)
Text S1
Duty cycle transmission.
(DOC)
Text S2
Choosing a smoothing factor for kernel analysis.
(DOC)
Text S3
Effect of duty cycle on kernel estimation.
(DOC)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Atherosclerosis is a multifactorial disease affecting arterial blood vessels due to a chronic inflammatory response. One of the early stages of atherosclerosis is the endothelial adhesion/infiltration of white blood cells to the arterial sites, attracted by different pro-inflammatory molecules [ 1 , 2 ]. This accumulative process gradually promotes inflammation and, by doing so, accelerates the atherosclerotic progression.
Triggering Receptor Expressed on Myeloid cells-1 (TREM-1) is a member of the recently discovered immunoglobulin family [ 3 , 4 ]. It increases the inflammatory responses of the neutrophils and monocytes [ 5 ] and amplifies the production of pro-inflammatory cytokines (TNFα and IL-1beta) [ 6 ]. Furthermore, it has been reported that migrating neutrophils bind to TREM-1 [ 7 ], suggesting an important role of this receptor in the recruitment and mobilization of neutrophils and monocytes to the arterial intima [ 8 ]. Although TREM-1 was initially characterized for its role during the pathophysiology of septic shock [ 3 ], there are today increasing evidences that it is also implicated in other acute and chronic inflammatory diseases of non-infectious etiology, such as rheumatoid arthritis [ 9 ], atherosclerosis [ 10 ], acute myocardial infarction [ 11 ] and critical limb ischemia [ 12 ]. Activation of innate immunity and inflammatory cells recruitment and extravasation is a common feature of these diseases in which TREM-1 seems to be a central player [ 8 , 10 , 13 ]. Also, the serum levels of soluble TREM-1 (sTREM-1) correlate with the severity of these diseases [ 8 – 10 , 12 , 14 – 18 ]. Moreover, the genetic invalidation or pharmacological blockage of TREM-1 results in a reduced inflammatory state and improved outcome in animal models [ 19 , 20 ]. The mechanistic process by which TREM-1 participates in inflammatory cell extravasation during inflammatory diseases remains to be clarified. It is well known the important role that the adhesion molecules, and specially the selectins, play during this specific process [ 21 – 23 ]. During the last two decades, the importance of the adhesion molecules in the process of inflammatory cells adhesion and trans-endothelial migration, which contributes to pathological inflammation and thrombosis in many preclinical diseases including atherosclerosis, has been widely studied [ 21 , 24 – 26 ]. Thus, TREM-1 could be in close connection with the selectins.
TREM-1 gene polymorphisms have been associated with the development of acute inflammation and sepsis, but also with coronary artery disease. Indeed, the polymorphism rs4711668, which is located within the TREM-1 gene has been associated with severe coronary atherosclerosis in a Russian population [ 27 ]. However, it is still unknown whether the levels of sTREM-1 are genetically regulated, and no polymorphism has been yet reported to affect its expression levels. Furthermore, another interesting part of TREM-1 regulation is that although there exists a specific splicing variant for the soluble form of the protein (TREM-1sv) [ 28 ], some authors argue that the mechanism of shedding of the membrane bound TREM-1 by metalloproteases is the main contributor for the release of sTREM-1 [ 7 , 29 ].
In the present study, we investigated whether single nucleotide polymorphisms (SNPs) within or near the TREM-1 gene were associated with soluble TREM-1 (sTREM-1 and TREM-1sv) serum levels and with the expression levels of two TREM-1 splicing isoforms, more precisely, the membrane form (mbTREM-1) and the soluble form (TREM-1sv). Moreover, we investigated the associations between the significant in the above associations SNPs and levels of soluble E-selectin, L-selectin and P-selectin (sE-selectin, sL-selectin and sP-selectin), which are involved in the early stages of atherosclerotic processes.
Material and methods
Ethics statement
The samples are part of a human sample storage platform: the Biological Resources Centre ‘Interactions Gène- Environnement en Physiopathologie CardioVasculaire’ (BRC IGE-PCV—number BB-0033-00051) in Nancy, East of France. All participants gave a written informed consent. All the populations involved in this study were recruited in accordance with the latest version of the Declaration of Helsinki for Ethical Principles for Medical Research Involving Human Subjects. All the protocols were approved by the local ethics committees for the protection of subjects for biomedical research: the Comité Consultatif de Protection des Persones dans la Recherche Biomédicale (CCPPRB).
Study populations
The population enrolled in this study makes part of the STANISLAS Family Study (SFS) [ 30 ]. Participants were of French origin and were apparently in good health, not under lipid-lowering and/or cardiovascular drug treatment and free from chronic diseases. This cohort is a longitudinal family study designed to investigate factors related to cardio-vascular disease. The clinical data of the investigated individuals were obtained at the Centre for Preventive Medicine (CMP) of Vandoeuvre-lés-Nancy. Participants were of European descent and came from the Vosges and the South of Meurthe-et-Moselle, in the East of France. Among them, 30 unrelated individuals were used as discovery cohort for the selection of the TREM-1 SNPs and for their relation with TREM-1 levels. A population of 351 unrelated individuals was used as discovery cohort for the associations with the adhesion molecules levels.
An independent population (n = 80), available in the Biological Resources Centre ‘Interactions Gène- Environnement en Physiopathologie CardioVasculaire’ (BRC IGE-PCV, number BB-0033-00051), composed of unrelated adults of French origin was used as replication population for the results of associations of SNPs with TREM-1 and adhesion molecules levels.
During and after the data collection, authors had access to information that could identify individual participants.
Data collection and biological measurements
Body mass index (BMI) was measured using the Quetelet’s formula: weight divided by height squared (kg/m 2 ). Blood samples were taken from the individuals after an overnight fast. Plasma and serum samples for adhesion molecules and TREM-1 measurements were frozen at -80°C until analysis.
Plasma levels of E-selectin, L-selectin and P-selectin were measured with enzyme-linked immunosorbent assay (R&D Systems, Abington, UK). Serum levels of soluble TREM-1 were measured with a double antibody sandwich ELISA assay (Quantikine Human TREM-1 Immunoassay ELISA Kit; R&D Systems, Minneapolis, MN, USA) using the iMARK Microplate Absorbance Reader (Bio-Rad).
All molecules were measured in duplicate and according to manufacturers’ instructions.
TREM-1sv levels were measured using a home-made ELISA assay adapted from Duo-Set ELISA assay (R&D Systems, hTREM-1 Duo-Set ELISA assay) and using a monoclonal anti-TREM-1sv antibody (R&D Systems) as capture antibody. A TREM-1sv recombinant protein was used as standard protein for quantification [ 28 ]. The sensitivity of the detection of TREM-1sv was of 15pg/mL.
SNPs selection–Bioinformatics analyses
A bibliographic search of the SNPs within or near TREM-1 was performed. The HuGE navigator ( https://phgkb.cdc.gov/PHGKB/phgHome.action?action=home ) and the NCBI dbSNP database ( http://www.ncbi.nlm.nih.gov/projects/SNP/ ) were used for the selection of SNPs for the further investigations. Only SNPs with a minor allele frequency >5% according to HapMap were selected.
To point out if the significant SNPs are located in regulation zones (promoter, enhancer, silencers) of the TREM-1 gene and are linked to specific epigenetic profiles (acetylation/methylation), bioinformatics analyses were performed comparing different cell types.
The SNPs were localised on the Human genome (GRCh38.p7) using Ensembl browser [ 31 ]. Identification of potent regulation zones and establishment of epigenetic profiles according to cell types, were determined using also this browser. Regulation and epigenetic profiles have been obtained using 23 different cell types from 19 healthy and 4 cancerous individuals from Ensembl release 87.
The possible influence of the SNPs on the transcription factor binding site (TFBS) was investigated by using the transcription factor database TRANSFAC R.3.4 [ 32 ].
Genotyping
DNA was extracted from all participants, and relative biobanks have been constructed inside the BRC IGE-PCV. The SNPs were genotyped by Genoscreen ( http://genoscreen.fr ), using a Sequenom iPLEX Gold assay-Medium Throughput Genotyping Technology.
Gene expression investigations
Gene expression analysis was performed on peripheral blood mononuclear cells (PBMCs) of a subsample of 30 individuals from the SFS. PBMCs were isolated by centrifugation on a density gradient of Ficoll. The RNA was extracted from PBMCs with the MagNAPure automate, using the MagNA Pure LC RNA HP isolation kit protocol (Roche Diagnostics, France) and the RNA quality was measured by the NanoDrop 1000 Spectrophotometer (Thermo Scientific). Total RNA was used to generate first-strand complementary DNA (cDNA) with the C1000 Thermal cycler (Bio-Rad) and the cDNA Synthesis Kit (Bio-Rad). The reactions conditions were as follows: 30 minutes at 42°C, followed by 5 minutes at 85°C.
The mbTREM-1 and the TREM-1sv were investigated. The primers used for the amplification of mbTREM-1 were as follows: forward primer, 5’-GTGACCAAGGGTTTTTCAGG-3’ ; reverse primer, 5’-ACACCGGAACCCTGATGATA-3’ . The primers used for the amplification of TREM-1sv were as follows: forward primer, 5’-GTGGTGACCAAGGGGTTC-3’ ; reverse primer, 5’-AGATGGATGTGGCTGGAAGT-3’ . The reaction conditions were as follows: Initial denaturation of 94°C for 2 min, followed by 40 cycles of denaturation at 94°C for 30 seconds, annealing at 58°C for 20 seconds, extension at 70°C for 20 seconds. For the absolute quantification of the 2 splicing forms of TREM-1 , corresponding mRNA levels were normalized to the mRNA levels of beta 2 microglobulin (B2M) gene. The housekeeping gene was simultaneously amplified to check the RNA integrity and to verify that the same amounts of template were used in all cases. The absolute quantification of the 2 splicing forms of the TREM-1 and the housekeeping gene were performed using Lightcycler Carousel-Based System (Roche Diagnostics) and SYBR Green I reaction Kit. Negative controls were used for each reaction and the specificity of the products was confirmed by polyacrylamide gel electrophoresis as well as by melt curve analysis. All experiments were performed in duplicate.
No PBMCs were available for the replication population and therefore this analysis was performed only in the SFS individuals.
Statistical analyses
All investigated molecules (intermediate phenotypes) were log transformed in order to normalise their distribution before analyses. The normality of distribution was tested by Kolmogorov-Smirnov test. Hardy-Weinberg equilibrium was tested using the chi-square test. The SNPs-mRNA and soluble TREM-1 levels associations were assessed through linear regression adjusted for age, gender and BMI under three inheritance models (additive, dominant and recessive) and using the minor allele as the reference allele. The association of the most significant SNPs with selectins levels were tested using similar models. Association studies were performed using the Plink Software [ 33 ]. Populations’ characteristics were determined using the SPSS statistical software version 20.0 (SPSS, Inc, Chicago, Illinois). Bonferroni correction was applied in order to adjust the multiplicity of tests and avoid the type 1 error. Cut-offs were 0.05/ number of SNPs for the association of the selected SNPs with soluble TREM-1 levels, 0.05/ number of SNPs/2 for the associations of the significant SNPs with the 2 mRNA levels and 0.05/number of SNPs/number of selectins for the associations of the significant SNPs with the selectins’ levels.
Results
HuGE navigator and NCBI dbSNP database searches resulted to the selection of 10 SNPs located within or near TREM-1 with a minor allele frequency >5%. The 10 SNPs were genotyped in 30 individuals from the SFS. No significant deviation from Hardy-Weinberg equilibrium was observed for the studied polymorphisms ( Table 1 ).
10.1371/journal.pone.0182226.t001
Table 1 Characteristics of the polymorphisms studied in 30 individuals of SFS.
SNP
Minor allele
Minor allele frequency (MAF)
Chromosome
HWE P
rs3789204
A
0.265
6
0.667
rs7768162
A
0.317
6
0.912
rs7772334
T
0.460
6
0.948
rs728488
T
0.368
6
0.505
rs612399
C
0.377
6
0.989
rs2234246
T
0.416
6
0.763
rs2234237
A
0.086
6
0.632
rs13211886
T
0.089
6
0.350
rs6910730
G
0.098
6
0.825
rs9471554
C
0.313
6
0.425
Demographic and clinical characteristics of the studied individuals are summarized in the Table 2 .
Association studies between the 10 polymorphisms in TREM-1 and serum levels of sTREM-1/TREM-1sv
For this first decisional step, we used a discovery and a replication population. Demographic and clinical characteristics of the studied individuals included in each population are summarized in the Table 2 .
10.1371/journal.pone.0182226.t002
Table 2 Demographic and clinical characteristics of the individuals.
SFS sub-sample
SFS Total
Replication Population
Sample size [% female]
30 [47%]
351 [48.43%]
80 [52.5%]
Age (years) [S.D]
47.07 [5.3]
44.09 [4.3]
47 [8.1]
BMI (kg/m 2 ) [S.D]
26 [4.69]
24.96 [3.76]
26.7 [3.69]
sTREM-1 (pg/ml) [S.D]
213.47 [70.24]
-
352.38 [87.88]
TREM-1sv (pg/ml) [S.D]
-
-
39.08 [32.01]
L-selectin (mg/l) [S.D]
948.68 [293.2]
1018.65 [280.8]
888.19 [74.67]
P-selectin (mg/l) [S.D]
132 [31.94]
137.93 [43.45]
-
E-selectin (mg/l) [S.D]
44.93 [11.68]
53.56 [27.07]
-
CRP (mg/L) [S.D]
1.41 [1.45]
1.83 [2.95]
-
Association analyses were performed in 30 individuals of SFS (discovery population) for the 10 SNPs and the levels of TREM-1 in serum using the additive, dominant and recessive inheritance models (P-value cut-off was set at 0.05/10 = 0.005). Only for rs2234246 we found significant results. According to the additive model, the minor allele T of rs2234246 was associated with increased levels of sTREM-1 (p-value = 0.003, β = 0.3, data available in Table 3 and Fig 1 ). When the dominant model was used, this association was even stronger (p-value = 0.0003, β = 0.49), and the variance in the serum TREM-1 levels explained by the model is of 33% in the discovery population.
10.1371/journal.pone.0182226.g001
Fig 1
Mean values of sTREM-1 levels according to the different genotypes of rs2234246 (CC vs TC vs TT) in the discovery population.
Thin bars show standard errors. CC; Homozygous for the major allele of the rs2234246. TC; Heterozygous for the rs2234246. TT; Homozygous for the minor allele of the rs2234246. The significance between genotypes is showed as follows; N.S.; >0.05, * p<0.05, ** p<0.01, *** p<0.001.
10.1371/journal.pone.0182226.t003
Table 3 Statistical associations of the 10 polymorphisms studied with the serum levels of TREM-1 and mRNA levels according to the different inheritance models. P-value thresholds for TREM-1 protein levels are P<0.005 for the discovery population and P<0.05 for the replication population. P-value threshold for mRNA levels is <0.025.
Polymorphisms
Model
sTREM-1 (protein) discovery (N = 30)
sTREM-1 (protein) replication (N = 80)
TREM-1sv (mRNA) (N = 30)
mbTREM-1 (mRNA) (N = 30)
P-value
β
S.E
P-value
β
S.E
P-value
β
S.E
P-value
β
S.E
rs3789204
Additive
0.090
0.25
0.14
-
-
-
-
-
-
-
-
-
Dominant
0.090
0.25
0.14
-
-
-
-
-
-
-
-
-
Recessive
-
-
-
-
-
-
-
-
-
-
-
-
rs7768162
Additive
0.024
0.21
0.08
-
-
-
-
-
-
-
-
-
Dominant
0.037
0.28
0.12
-
-
-
-
-
-
-
-
-
Recessive
0.139
0.28
0.18
-
-
-
-
-
-
-
-
-
rs7772334
Additive
0.105
-0.17
0.10
-
-
-
-
-
-
-
-
-
Dominant
0.226
-0.20
0.16
-
-
-
-
-
-
-
-
-
Recessive
0.195
-0.23
0.17
-
-
-
-
-
-
-
-
-
rs728488
Additive
0.191
-0.15
0.11
-
-
-
-
-
-
-
-
-
Dominant
0.103
-0.38
0.22
-
-
-
-
-
-
-
-
-
Recessive
0.471
-0.11
0.16
-
-
-
-
-
-
-
-
-
rs612399
Additive
0.063
-0.21
0.11
-
-
-
-
-
-
-
-
-
Dominant
0.193
-0.20
0.15
-
-
-
-
-
-
-
-
-
Recessive
0.098
-0.38
0.22
-
-
-
-
-
-
-
-
-
rs2234246
Additive
0.003
0.30
0.09
0.0007
0.13
0.03
0.874
0.02
0.17
0.007
0.49
0.16
Dominant
0.0003
0.49
0.11
0.001
0.21
0.06
0.225
0.31
0.24
0.002
0.79
0.23
Recessive
0.440
0.17
0.22
0.017
0.15
0.06
0.215
-0.40
0.31
0.250
0.42
0.35
rs2234237
Additive
0.337
-0.15
0.15
-
-
-
-
-
-
-
-
-
Dominant
0.337
-0.15
0.15
-
-
-
-
-
-
-
-
-
Recessive
-
-
-
-
-
-
-
-
-
-
-
-
rs13211886
Additive
0.953
0.01
0.19
-
-
-
-
-
-
-
-
-
Dominant
0.953
0.01
0.19
-
-
-
-
-
-
-
-
-
Recessive
-
-
-
-
-
-
-
-
-
-
-
-
rs6910730
Additive
0.337
-0.15
0.15
-
-
-
-
-
-
-
-
-
Dominant
0.337
-0.15
0.15
-
-
-
-
-
-
-
-
-
Recessive
-
-
-
-
-
-
-
-
-
-
-
-
rs9471554
Additive
0.054
0.28
0.13
-
-
-
-
-
-
-
-
-
Dominant
0.054
0.28
0.13
-
-
-
-
-
-
-
-
-
Recessive
-
-
-
-
-
-
-
-
-
-
-
-
The SNP rs2234246 was then genotyped in 80 additional and independent individuals (replication population). Additive and dominant models confirmed the association between the minor allele T and sTREM-1 levels (p-values = 0.0007 and 0.0017 respectively, β = 0.13 and 0.22 respectively, data available in Table 3 and Fig 2 ). Threshold of significance was 0.05 (as only 1 SNP was tested). The TREM-1 variance explained by the model is of 13% in the replication population.
10.1371/journal.pone.0182226.g002
Fig 2
Mean values of sTREM-1 levels according to the different genotypes of rs2234246 (CC vs TC vs TT) in the replication population.
Thin bars show standard errors. CC; Homozygous for the major allele of the rs2234246. TC; Heterozygous for the rs2234246. TT; Homozygous for the minor allele of the rs2234246. The significance between genotypes is showed as follows; N.S.; >0.05, * p<0.05, ** p<0.01, *** p<0.001.
The serum TREM-1sv levels, generated by the mRNA splicing variant TREM-1sv, were also measured in 78 individuals of the replication population. The TREM-1sv levels were detected only in four of the 78 samples studied, allowing no statistical analysis.
Association studies between the rs2234246 in TREM-1 and mRNA expression levels (mbTREM-1 and TREM-1sv)
Association analyses were then performed between the rs2234246 SNP and the expression levels of the two different alternative splicing isoforms of TREM-1 (mbTREM-1 and TREM-1sv mRNAs) in PBMCs of the 30 individuals of the discovery population (threshold of significance was 0.05/1 SNP/2 phenotypes = 0.025). Interestingly, the SNP rs2234246 showed significant association with increased mRNA levels of the splicing form that codes for mbTREM-1 (additive model, p-value = 0.007, β = 0.49, data available on Table 3 ), whereas it was not associated with the levels of mRNA coding TREM-1sv. The TREM-1sv protein was only present in four samples out of the 78 samples studied, not showing increased expression in the presence of the SNP rs2234246.
Association studies of the rs2234246 polymorphism in TREM-1 with soluble selectins’ levels
The SNP rs2234246 was further genotyped in 351 individuals (discovery population, characteristics in Table 2 ). Association analyses were performed between the SNP rs2234246 and the sL-, sP- and sE-selectin levels (cut-off value for significance was set to 0.05/1 SNP/3 phenotypes = 0.016). Only the association with sL-selectin was significant (p-value = 0.011, β = 0.05) and explained a total of 2.1% of the variability of the sL-selectin. The minor allele T of the polymorphism was significantly associated with increased sL-selectin levels. No significant association was observed for sP- and sE-selectin levels.
The association between the minor allele T and sL-selectin levels was then confirmed in the 80 individuals of the replication population (p-value = 0.018, β = 0.03) (cut-off value for significance was set to 0.05, as only 1 SNP was tested) and association explained a total of 4.3% of the variability of the sL-selectin. All results are available in the Table 4 .
10.1371/journal.pone.0182226.t004
Table 4 Effects of the polymorphism rs2234246 located within the TREM-1 gene on the serum levels of the studied selectin molecules. MAF = Minor allele frequency of the rs2234246. Cutoff value of significance: 0.016 in discovery and 0.05 in replication population.
Population
Phenotypes
β
S.E
P-value
Discovery population (MAF: 0.498)
sL-selectin (mg/l) (n = 351)
0.05
0.02
0.011
sP-selectin (mg/l) (n = 311)
0.02
0.02
0.435
sE-selectin (mg/l) (n = 351)
0.01
0.04
0.685
Replication population (MAF: 0.487)
sL-selectin (mg/l) (n = 80)
0.03
0.01
0.018
Regulatory environment of TREM-1 rs2234246
Using bioinformatics analyses, we established the regulatory environment of SNP rs2234246 ( S1 Fig ). The rs2234246 polymorphism is located at 41276002 bp on the forward strand. It is positioned halfway between an open chromatin zone and a promoter flanking region. We can note that the open chromatin zone is active only in monocytes CD14+, which are present in PBMCs. Furthermore, the promoter flanking region is in an active state only in two cell types: the monocytes CD14+ and CD14+CD16-. It can be observed as well that the polymorphism is located in the 3’UTR region of the two mRNA splicing variants studied: mbTREM-1 (TREM1-001) and TREM-1sv (TREM1-002).
It is important to mention also that in the non-leukocyte cell types, the neighbourhood of TREM-1 gene show only repressed or inactive areas, suggesting that these areas are not subject to regulations (results not shown).
Epigenetic footprint of TREM-1 rs2234246
Using bioinformatics tools, we established the epigenetic profiles (methylation / acetylation) of rs2234246 according to the cell type involved ( S2 Fig ). A specific epigenetic pattern emerged in monocytes (CD14+ and CD14+CD16-), vein blood neutrophils, eosinophils and macrophages in particular by the presence of H3K36me3 and H3K4me1. This specific methylation pattern is not present in the other 18 cell types investigated. Especially the presence of H3K36me3 is interesting, as is considered as a hallmark of actively transcribed regions. Thus, we speculate that the genetic region where the SNP rs2234246 is located is prone of higher expression levels of the TREM-1 gene.
Potential transcription factor binding sites in the rs2234246 locus
The bioinformatics results showed also differences in the potential transcriptional factors binding the locus according to the different alleles of the SNP rs2234246.
When the minor allele is present (T), the specific potential transcription factors that are able to bind are AP4, LMO2 and TAL1, with a similarity score of 0.857, 0.791 and 0.773, respectively. Regarding the position weight matrix, we can see that the minor allele T is a highly conserved nucleotide in the TAL1 binding site, having a height of 2.0 bits. In the case of the transcription factor binding site LMO2 and AP4, the T allele is also important with a height >1.7 and >1.2 bits, respectively ( Fig 3 ).
10.1371/journal.pone.0182226.g003
Fig 3
Specific transcription factor binding sites for the minor allele T of the SNP rs2234246.
When the major allele is present (C), the specific potential transcription factors that are able to bind are REL, CAAT and NFY, showing a similarity score of 0.838, 0.801 and 0.771 respectively. The position weight matrix shows that the minor allele C is highly conserved in the case of the CAAT and NFY binding sites (height >2 bits and >1.7 bits, respectively). In the binding site of REL, the C allele is also a conserved nucleotide with a height >1.2 bits ( Fig 4 ).
10.1371/journal.pone.0182226.g004
Fig 4
Specific transcription factor binding sites for the major allele C of the SNP rs2234246.
The consensus sequence (fixed) of the transcription factor binding sites means: S = C or G, W = A or T, R = A or G, Y = C or T, K = G or T, M = A or C, N = any base pair. The input sequence was 38pb length, centred on the SNP of interest rs2234246. Only TFBS harbouring a nucleotide in its consensus sequence in coherence with the SNP of interest were selected.
Discussion
TREM-1 protein levels have been related to numerous diseases where inflammation and inflammatory cell extravasation play central roles, such as atherosclerosis [ 10 ], acute myocardial infarction [ 11 ], and critical limb ischemia [ 12 ]. All these diseases have an important genetic component [ 34 , 35 ], and despite all advances in treatment and prevention, they remain the leading cause of death worldwide [ 36 ]. This makes necessary the detection of genetic polymorphisms that could uncover novel metabolic pathways involved in the pathophysiology of these diseases, and therefore, improve their prevention and treatment [ 37 , 38 ].
After a bibliography search of the SNPs within or near TREM-1 , a total of 10 polymorphisms were selected and further analysed. Among them, we found that rs2234246, located in the mRNA 3’UTR region of TREM-1 , was associated with sTREM-1 protein levels. So far, this is the first polymorphism that has been reported to affect the sTREM-1 levels. The minor allele T of this polymorphism was strongly associated with increased sTREM-1 levels, both in the discovery and the replication populations (p-values = 0.003 and 0.0007 respectively), explaining a large percentage of its phenotypic variation (33% and 13% in the discovery and replication population respectively).
It has been widely reported that increased levels of sTREM-1 were correlated to the severity of above-mentioned inflammatory diseases and poor prognosis [ 8 – 10 , 12 , 14 – 18 ], and the most recent studies have showed that TREM-1 deletion or blockade is associated with up to 60% reduction of the development of atherosclerosis [ 10 ]. Thus, the T allele of rs2234246 may be considered as a risk factor, while the C allele may be a protective factor for these diseases. Given the high frequency of the polymorphism in the general population (about 50%), this result could be of high interest in further personalized medicine strategies for stratifying patients according to their risk to the above pathologies. In this context, it is important to note that a previous study has associated the minor allele of the SNP rs2234246 with an increased 3.1 odds ratio for septic shock [ 39 ]. This finding strengthens our hypothesis on the utility of the polymorphisms located in TREM-1 gene in risk stratification. Another SNP located also within the TREM-1 gene (rs4711668) has been associated with severe coronary atherosclerosis in a Russian population [ 27 ]. However, it is not known whether this polymorphism is affecting the TREM-1 protein levels and although it is close to our SNP of interest, the 2 SNPs are not in strong linkage disequilibrium between them (r 2 = 0.46).
We have also associated the T allele of the SNP rs2234246 with the expression levels of mbTREM-1 mRNA in PBMCs (p-value = 0.007 β = 0.49). This result is conferring functional properties to this polymorphism. However, we didn’t differentiate the monocyte subsets, and the expression of TREM-1 as well as the effect of the SNP rs2234246 could vary according the different monocyte subtypes. It is important to note that according to our bioinformatics analysis, the polymorphism is located in the 3’UTR region of the two mRNA splicing variants studied ( S1 Fig ). Interestingly, the polymorphism rs2234246 was related to an increase in the expression level of the mRNA coding mbTREM-1, but it was not related to the mRNA that codes TREM-1sv, nor to the TREM-1sv protein levels itself, which was not present in the serum. Two hypotheses have been proposed for the origin of soluble TREM-1: (1) splicing of different variants of alternative mRNA, which generates the TREM-1sv [ 40 ] and (2) shedding of mbTREM-1 by metalloproteases, which generates the sTREM-1. Our results, taking also into consideration that TREM-1sv protein levels were not present in the serum, support the hypothesis that the levels of sTREM-1 are controlled post transcriptionally by metalloproteases, rather than by alternatively spliced forms of RNA [ 7 , 29 ].
According to our bioinformatics-epigenetics results, the rs2234246 has a specific H3K36me3 and H3K4me1 methylation epigenetic patterns in two groups of monocytes (CD14+ and CD14+CD16-), which are present in PBMCs. The presence of H3K36me3 is especially interesting as it is considered to be a hallmark of actively transcribed gene bodies. Thus, we speculate that the genetic region where the SNP rs2234246 is located, is prone of higher expression levels of the TREM-1 gene [ 41 ]. However, this is only an assumption, as we were not able to perform experiments to confirm this effect. At the same time, the regulation profile of TREM-1 shows an open chromatin zone and active promoter flanking regions only in CD14+ monocytes. The open chromatin zones are functionally related to transcriptional activity where DNA is accessible for the binding of proteins such as transcription factors [ 42 ]. Moreover, the SNP rs2234246, by the fact that it is located in the 3’-UTR of the mRNA, which is rich in regulatory regions that post-transcriptionally could influence gene expression [ 43 , 44 ]. The effect of rs2234246 polymorphism on the sTREM-1 levels could be explained by several hypothetical mechanisms: (i) Post-transcriptional regulation of many pro-inflammatory mediators is controlled by adenosine and uridine-rich elements (AREs) [ 45 ]. AREs regions, located in the 3’-UTR of the mRNA can promote mRNA decay, affect mRNA stability, or activate translation. (ii) According to the TRANSFAC R.3.4. Database, the potential transcription factors binding the locus of the SNP rs2234246 could be different depending on the allele. The minor allele T is potentially associated with the matching of the transcription factors AP4, LMO2 and TAL1, while the major allele C is potentially associated with the matching of the transcription factors REL, CAAT and NFY. The type of transcription factors or even the affinity of those for the different polymorphisms of the rs2234246, could explain the changes in the expression levels of the TREM-1 gene. However, we didn’t have experimental data that could confirm these possible direct affinity changes depending on the different alleles of the rs2234246, thus further experiments are needed to confirm this assumption.
It has been previously suggested that the receptors of the TREM family are regulating the cellular adhesion of macrophages and neutrophils via the phosphorylation of DAP12, which leads to activation of calcium sensitive kinases [ 6 , 46 ]. At the same time, the number of studies supporting the importance of TREM-1 in the trans-epithelial migration of neutrophils and monocytes is increasing. Migrating neutrophils in septic patients have been found to bind to TREM-1 [ 7 ]. It has been previously demonstrated that TREM-1 is crucially involved in leukocyte recruitment after myocardial infarction and atherosclerosis [ 10 , 11 ]. We demonstrated a five-fold decrease in the number of recruited neutrophils when TREM-1 was inhibited pharmacologically [ 11 ]. Also, in acute respiratory infections, TREM-1 is required for the trans-epithelial migration of neutrophils into the lung [ 13 ]. Despite these arguments, so far, the process by which TREM-1 contributes to this trans-epithelial migration is unknown. In our study, we demonstrated for the first time that TREM-1 is regulating one adhesion molecule.
We found that the rs2234246 polymorphism is specifically correlated with increased plasma levels of L -selectin. The soluble L-selectin levels are thought to represent a homeostatic effort to limit excessive inflammation. Indeed, they are correlated with the severity of inflammatory diseases, including cardio-vascular diseases [ 47 , 48 ]. Although, once in a soluble form, they may have reduced adhesion, migration and trans-epithelial migration capacity, it can be postulated that TREM-1 increases inflammation in general, leading to L-selectin shedding as a downstream effect. Especially, L-selectin, has been showed to be important in the recruitment of monocytes and neutrophils to sites of acute and chronic inflammation [ 49 ]. The role of the selectins in inflammatory processes is well established, and the selectin-mediated adhesion and signalling contribute to different cardio-vascular diseases [ 21 ]. The selectins, have been shown to contribute to atherosclerosis [ 24 , 25 ] and arterial thrombosis [ 50 ]. According to our results, the minor allele T of the rs2234246 could act as a risk factor as it is correlated with an increase of 2.1–4.3% of sL-selectin [ 47 , 48 , 51 ]. It would be also interesting to further investigate the possible associations between the membrane bound L-selectin levels and rs2234246. Unfortunately, in this work we were not able to address this issue.
The conclusion from our bioinformatics and transcriptomic analyses supports the hypothesis that TREM-1 is involved in the trans-epithelial migration process of leukocytes, more specifically of monocytes, that could be effective through a higher level of inflammation, which could be observed with an overexpression of sL-selectin.
One of the advantages of our research is that we limited the potential confounders by using a homogeneous population coming from the region of Lorraine in France as a discovery population. Because of this, we have been able to reduce the enormous cardio-vascular-related heterogeneity of the population, by linking genes of interest to intermediate phenotypes and not to the disease. This approach can represent better the real biological pathways where the gene of interest is involved. Moreover, we replicated the results in an independent population, while simultaneously bioinformatics and bibliographical data also strengthen our results.
One limitation of our study is that we were not able to realize gene expression analysis in the PMNs and more specifically in neutrophils. As well as the monocytes, the neutrophils are also expressing TREM-1 protein. However, we were not able to have direct experimental data about the effect of the SNP rs2234246 in neutrophils due to lack of biological material. Future studies are needed to confirm these possible effects. Also, we couldn’t make the distinction among the different subgroup of monocytes and TREM-1 expression, and the effect of the SNP rs2234246 may be different in those subgroups. A more specific study including the monocyte subsets would significantly improve the biological relevance of the SNP rs2234246. In conclusion, our study led to the discovery of one polymorphism (rs2234246) strongly affecting sTREM-1 protein levels and associated to an increase in the levels of the mRNA coding mbTREM-1 in PBMCs. Since no association was established with splicing mRNA and TREM-1sv was not detected in the serum of the individuals, it seems that, the levels of sTREM-1 are controlled post transcriptionally by metalloproteases. Interestingly, we demonstrated for the first time that TREM-1 rs2234246 polymorphism can also modulate the sL-selectin levels via a higher inflammatory state, suggesting that TREM-1 acts in the trans-epithelial migration process of leukocytes and more specifically of monocytes through expression of sL-selectin.
Supporting information
S1 Fig
Regulation profile of the TREM-1 gene in different cell types expressing the protein TREM-1.
The polymorphism rs2234246 is located at 41276002 bp on the forward strand (vertical red line). It is positioned halfway between an open chromatin zone and a promoter flanking region. It can also be observed that the polymorphism is located in the 3’UTR region of the two mRNA splicing variants studied: mbTREM-1 (TREM1-001; ENST00000244709.8) and TREM-1sv (TREM1-002; ENST00000334475.10) and in an intron zone of another TREM-1 transcript: TREM1-006 (ENST00000589695.1) (VB: Vein blood).
(DOCX)
S2 Fig
Epigenetic profile of rs2234246 in leukocytes cell types: Specific histone methylation patterns (VB: Vein blood).
(DOCX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Clinical supervision is widely practiced in health and social care professions across the globe owing to its beneficial effects to patients, health professionals and organisations [ 1 , 2 ]. Operationally, clinical supervision, for post-qualification health professionals, is viewed as a process that provides quarantined time and an opportunity to further develop the supervisee’s skills and knowledge, within the context of an ongoing professional relationship, usually with an experienced practitioner (one-to-one supervision), or with peers (peer group supervision). The aim of clinical supervision is for the supervisee to engage in guided reflection on current practice in ways designed to develop and enhance that practice in the future [ 1 , 2 ]. This type of supervision involves reflective thinking, and discussion regarding professional development issues, caseload, clinical issues, and staff interpersonal issues. Issues in clinical supervision definition and terminologies are widely prevalent [ 2 ]. In this review, the following definition of clinical supervision has been adopted:
“ The formal provision , by approved supervisors , of relationship-based education and training that is work-focused , and which manages , supports , develops and evaluates the work of colleague/s ” [ 1 ].
Whilst efforts are growing to strengthen the evidence for clinical supervision, there is also criticism about a vast majority of evidence on supervision, as being proof by association or tentative [ 3 ]. While there is a growing evidence base for the impact of clinical supervision on patient outcomes such as reduced risk of mortality, reduced risk of complications and more effective care [ 4 – 7 ], and health professional outcomes such as being better supported in their roles [ 8 ], there remains a need to systematically review the evidence for the impact of clinical supervision of post-qualification health professionals, on organisational outcomes, to further strengthen the evidence base on clinical supervision.
Determining the impact of clinical supervision on healthcare organisations, however, is difficult given the challenges in defining organisational outcomes and the overlapping nature of patient, health professional and organisational outcomes. For example, improved patient outcomes (e.g. improved morbidity and mortality) can satisfy multiple targets for healthcare organisations, as can health professional outcomes (e.g. reduction in stress and burnout), which can reduce staff sick leave, a usual key performance indicator for organisations. In determining the organisational outcomes of interest for this review, we undertook a scan of the broader literature. A recent systematic review of leadership styles and outcome patterns for the nursing workforce and work environment, grouped the outcomes into six categories: staff satisfaction and job factors, staff relationships with work, staff health and wellbeing, relations among staff, organisational environment factors and productivity and effectiveness [ 9 ]. Another systematic review on the relationship between governance mechanisms in healthcare and health workforce outcomes considered staff turnover and job satisfaction [ 10 ]. Other organisational outcomes cited in the clinical supervision literature include improved teamwork [ 11 ] and job satisfaction [ 12 ]. In considering all this, organisational outcomes in the current review will reflect the well-being of health professionals resulting from clinical supervision, that lead to better outcomes for the organisations such as recruitment and retention, intent-to-stay, intent-to-leave, job satisfaction and quality of work life, burnout and absenteeism. Furthermore, despite the benefits of supervision, to date, no review has explored health professionals’ perspectives of, and the impact from, clinical supervision on organisational outcomes.
Therefore, as means of addressing these knowledge gaps, using a mixed methods design, this review aims to answer the following research questions:
What are the effects of clinical supervision of healthcare professionals on organisational outcomes?
What are healthcare professionals’ experiences, views, and opinions regarding clinical supervision as it relates to organisational processes and outcomes?
What can be inferred from the qualitative synthesis of healthcare professionals’ experiences/ views that can explain the effects of clinical supervision or inform its appropriateness and acceptability for health professionals?
Methods
This systematic review was conducted using Joanna Briggs Institute (JBI) methodology for mixed methods systematic review, specifically the convergent segregated approach to synthesis and integration [ 13 ]. The review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guideline [ 14 ] and was based on an a-priori published protocol [ 15 ].
Eligibility criteria
The review protocol indicated the inclusion of studies that focused on one-to-one clinical supervision rather than group supervision. However, during the screening of studies, it became apparent that there was a prevalence of studies that investigated both one-to-one and group supervision (which was facilitated by a supervisor, as opposed to peer supervision), and studies that did not specify the type of clinical supervision investigated. Given this challenge, and to reflect the reality of healthcare organisations utilising both these types of supervision regularly, the review team agreed to include any study on clinical supervision, regardless of the type (i.e. one-to-one or group). To be eligible, studies had to meet the following criteria: (1) investigated clinical supervision of qualified or registered health professionals (i.e. clinical supervision of post-qualified health professionals, where they engage in one-to-one or group supervision sessions that happen over a period of time); (2) used qualitative, quantitative or mixed-methods study design; (3) if a quantitative study, examined the effects of clinical supervision on organisational outcomes, such as staff retention and recruitment, intent to stay, intent to leave, job satisfaction and quality of work life, burnout, and absenteeism; (4) if a qualitative study, explored health professionals’ experiences, views, or opinions regarding clinical supervision as they relate to organisational outcomes.
Search strategy
As means of avoiding publication and location bias, the search strategy was developed to identify black (commercially published) and grey literature. Search terms were identified based on the key concepts relating to the intervention/phenomenon of interest, i.e. clinical supervision and outcomes of interest, i.e. organisational outcomes.
An initial limited search of PubMed and CINAHL was undertaken followed by analysis of text words contained in the title and abstract and the index terms used to describe the articles. The search strategy, including all identified keywords and index terms, was then adapted for each database. The search for published studies was performed from the date of inception until May 2020 in the following databases: CINAHL, Embase, PubMed, PsycINFO, and Scopus. These databases were chosen as they commonly include literature from health disciplines, a combination of discipline specific (e.g. CINAHL includes nursing and allied health literature) and multi-disciplinary (e.g. Scopus) and are routinely used in systematic reviews. The search for grey literature was undertaken in ProQuest Dissertations and Theses, Google Scholar and WorldWideScience.org . Reference lists of relevant studies were reviewed to identify additional publications. The search strategy for each database is shown in S1 Appendix .
Study selection
Following the search, all identified citations were collated and uploaded into EndNote X8.2 (Clarivate Analytics, PA, USA) [ 16 ] and duplicates removed. Two reviewers independently screened the titles and abstracts (LL and DS) against the inclusion criteria for the review. Potentially relevant articles were retrieved in full and assessed independently for eligibility by two other reviewers (PM and SK). Disagreements were resolved through discussion and consensus. Studies that did not meet the inclusion criteria were excluded and reasons for their exclusion are provided in S2 Appendix . Abstracts and full text articles did not require translation to another language to determine their eligibility. All full text articles reviewed contained sufficient information to determine their eligibility without the need for further clarification from authors. The PRISMA flow diagram of included studies is available in Fig 1 .
10.1371/journal.pone.0260156.g001
Fig 1
Flow diagram of included studies.
Quality assessment
All eligible studies were assessed for methodological quality by two independent reviewers (PM and DS for quantitative studies; PM and LL for qualitative studies using the relevant JBI critical appraisal tools [ 17 ]. These tools were chosen as they assist in assessing the trustworthiness, relevance and results of published studies and are widely used. Any disagreements that arose between the reviewers were resolved through discussion. All studies, regardless of the results of their methodological quality, underwent data extraction and synthesis.
Data collection
For the quantitative component, data were extracted from quantitative and mixed methods studies (quantitative component only) and included specific details about the supervisee and supervisor characteristics (sample size, profession), characteristics of the supervision (type, frequency, duration), study design, setting, clinical supervision characteristics, outcomes measured, and results related to the organisational outcomes. For the qualitative component, data were extracted from qualitative and mixed methods studies (qualitative component only) and included specific details about the supervisee and supervisor characteristics (sample size, profession, work experience), study design and methods, setting, and findings which included participants’ experiences of clinical supervision as they relate to organisational outcomes. Findings extracted from individual studies consisted of themes or subthemes reported by the authors. These findings were accompanied by a direct quotation representing a participant’s voice (i.e. illustration). Findings were also assigned one of three levels of credibility according to the following criteria: (1) unequivocal: findings accompanied by an illustration that is beyond reasonable doubt and therefore not open to challenge, (2) credible: findings accompanied by an illustration lacking clear association with it and therefore open to challenge, and (3) not supported: findings are not supported by data. The review team discussed the data extraction process, established standards and consistencies on how this should occur, and those with quantitative expertise (DS) and qualitative expertise (LL) lead the extraction process, with the primary reviewer (PM) acting as the additional reviewer for validation purposes.
Data synthesis and integration
A convergent segregated approach to synthesis and integration was applied [ 13 ]. This involved an initial independent synthesis of the quantitative studies and qualitative studies followed by the integration of findings from such syntheses using configurative analysis.
Quantitative data were analysed descriptively; meta-analyses were deemed not appropriate due to heterogeneity between studies in terms of clinical supervision interventions and participants. Odds ratios (OR) of dichotomous events and standardised mean differences (SMD) for continuous measures were calculated. For experimental studies OR were converted to SMD using an online calculator [ 18 ], to assist with interpretation of effect size. For observational analytical studies the correlational coefficient ( r ) was calculated in addition to OR and SMD. Effect size was determined using the following reference values for SMD: small 0.2, medium 0.5, large 0.8 [ 19 ]; OR: small 1.68, medium 3.47, large 6.71 [ 20 ]; and r : small 0.1, moderate 0.3, large 0.5 [ 19 ].
Qualitative synthesis was conducted using the meta-aggregative approach [ 21 ]. Meta-aggregation is aligned with the philosophy of pragmatism, focusing on the practicality and usability of the synthesised findings and generation of statements that are useful for informing actions in clinical practice [ 21 ]. This involved assembling and aggregating the extracted findings from individual studies, based on similarity in meaning, to generate a set of statements (i.e. categories) that represented that aggregation. These categories were then subjected to meta-synthesis to produce a set of synthesised findings. The development of categories and synthesised findings was conducted via a consensus process between the reviewers (LL and PM).
The findings of each single method synthesis were juxtaposed and examined for convergence/divergence and complementarity. To explore the relationship across individual syntheses, the findings were reviewed to determine whether they were supportive or contradictory and identify which aspects of the quantitative evidence were not explored in the qualitative studies and vice-versa. The clinical supervision interventions which had been investigated in the quantitative studies were further analysed in line with the experiences of participants in the qualitative studies to explain the impact of clinical supervision on the different organisational outcomes. However, due to the heterogeneity of the quantitative studies and the lack of well-conducted trials, and the limited qualitative studies, no clear cause and effect relationships can be determined, nor in-depth analysis can be made to explain the impact of clinical supervision.
Results
The database search yielded 1266 records. Eighty-five articles were retrieved for full text review following application of the eligibility criteria to title and abstract. Thirty-four fulfilled the inclusion criteria when applied to full texts. Three of these articles were duplicate publications, resulting in a yield of 31 studies. One article was identified through pearling of references in the included studies; hence the final yield was 32 studies ( Fig 1 ).
Study characteristics
Twenty-seven quantitative [ 22 – 48 ], two qualitative [ 49 , 50 ] and three mixed methods studies [ 51 – 53 ] were included in the review. Fifteen studies used a randomised controlled (n = 1) [ 22 ] or quasi-experimental design (n = 14) [ 23 – 35 , 51 ] to establish the effect of clinical supervision on organisational outcomes. Eight studies investigated the association between effectiveness of clinical supervision and organisational outcomes [ 37 – 43 , 52 ]. Eight studies investigated the association between the effectiveness of the supervisor and organisational outcomes [ 32 , 36 , 39 , 41 , 43 , 45 , 46 , 48 ]. Two studies used a cross sectional survey study design to measure perceptions of effect of clinical supervision on organisational outcomes [ 44 , 47 ]. Four studies [ 49 – 52 ] used a qualitative descriptive design, with either individual [ 49 – 51 ] or focus group [ 52 ], semi structured interviews as the method of data collection. The qualitative component of one study [ 53 ] applied the grounded theory methodology, using a qualitative questionnaire for data collection. Ten studies were published in the 1990s [ 23 , 25 , 26 , 30 , 34 , 45 – 49 ], six studies were published in the 2000s [ 24 , 28 , 35 , 39 , 41 , 43 ], and 16 studies were published in the 2010s [ 22 , 27 , 29 , 31 , 32 , 33 , 36 – 38 , 40 , 42 , 44 , 50 – 53 ] with seven of these published in the last 5 years [ 27 , 37 , 40 , 50 – 53 ].
Studies were conducted in hospital (n = 15) [ 22 – 26 , 29 – 31 , 34 , 37 , 40 , 41 , 44 , 47 , 53 ], community healthcare settings (n = 6) [ 32 , 36 , 38 , 39 , 48 , 52 ] and a combination of hospital and community healthcare settings (n = 11) [ 27 , 28 , 33 , 35 , 42 , 43 , 45 , 46 , 49 – 51 ]. Most studies were conducted in the mental health setting (n = 15) [ 25 – 28 , 30 , 34 , 37 , 39 , 42 – 44 , 46 – 48 , 53 ]. Health professionals who received clinical supervision included nursing (n = 23) [ 22 – 26 , 30 – 35 , 37 , 39 – 42 , 44 , 46 – 49 , 51 , 53 ], social work/psychology/counselling professionals (n = 10) [ 27 , 32 , 36 , 38 , 43 , 45 , 46 , 48 , 52 ], other allied health professionals (n = 4) [ 28 , 29 , 50 , 52 ] and medical professionals (n = 3) [ 22 , 33 , 48 ]. Seven studies were conducted in Sweden, [ 23 , 25 , 26 , 28 , 30 , 34 , 47 ] seven in Australia [ 32 , 38 , 40 , 43 , 50 – 52 ], seven in the United Kingdom [ 22 , 35 , 37 , 39 , 44 , 49 , 53 ], four in the United States of America [ 27 , 45 , 46 , 48 ], two in Finland [ 31 , 41 ] and one each in Norway [ 24 ], Israel [ 36 ], Africa [ 33 ], Denmark [ 42 ] and Italy [ 29 ]. Eight studies investigated only group supervision [ 23 – 26 , 30 , 31 , 42 , 47 ] four studies investigated only individual (one-to-one) supervision [ 22 , 27 , 38 , 50 ], 12 studies investigated both group and individual supervision [ 29 , 32 , 35 , 39 , 41 , 43 – 46 , 49 , 51 , 52 ] and eight studies did not state whether the supervision they investigated was group or individual [ 28 , 33 , 34 , 36 , 37 , 40 , 48 , 53 ]. The frequency and duration of supervision sessions were variable between studies, ranging from weekly to every three months, and 30 to 480 minutes. Frequency and duration of supervision were not reported in 16 [ 22 , 28 , 29 , 33 – 37 , 40 , 45 , 48 – 53 ] and 18 studies [ 24 , 28 , 29 , 33 – 38 , 40 , 44 , 45 , 48 – 53 ], respectively.
Five studies (two qualitative [ 49 , 50 ] and three mixed methods studies [ 51 – 53 ]) explored the clinical supervision experiences of healthcare professionals including its impact on clinical practice. Fifteen studies investigated the effect of supervision on burnout [ 22 , 25 , 27 – 32 , 35 – 37 , 39 , 41 , 42 , 48 , 52 ], 9 studies on other measures of well-being [ 22 , 24 – 26 , 30 – 32 , 42 , 44 ], 13 studies on job satisfaction [ 25 – 28 , 30 , 32 , 33 , 41 – 43 , 45 , 46 , 51 ], 9 studies on the work environment [ 23 – 26 , 31 , 34 , 35 , 38 , 47 ], and 3 studies on job retention [ 32 , 33 , 40 ]. There was a large diversity of outcome measures used with only four measures used in more than one study; the Maslach Burnout Inventory was used in 13 studies [ 25 , 27 , 29 – 32 , 35 , 37 , 39 , 41 , 42 , 48 , 52 ], and the Creative Climate Questionnaire [ 25 , 26 ], Tedium Measure [ 25 , 30 ] and Satisfaction with Nursing Care questionnaire [ 25 , 30 ] each used in two studies. Study characteristics can be found in Table 1 .
10.1371/journal.pone.0260156.t001
Table 1 Study characteristics.
Study
Design
Setting
Participants
Supervision
Outcomes (Quantitative) OR Interview questions (Qualitative)
(country)
Supervisee
Supervisor
Type
Frequency
Duration
Profession
Profession
Work Experience, mean
n
Begat 1997
Quantitative Quasi-experimental pre/post
Acute hospital medical wards (Sweden)
Nursing
Nursing
Group
Weekly—Fortnightly
90 minutes
Work environment a
11 to 18 years
n = 34
Begat 2005
Quantitative Quasi-experimental cross sectional
Acute hospital medical wards (Norway)
Nursing
N/S
Group
Fortnightly
N/S
Well-Being a
9 years
Work Environment
n = 71
WEQ
Ben-Porat 2011
Quantitative Cross sectional
Domestic violence and women’s shelters (Israel)
Social Work
N/S
N/S
N/S
N/S
Burnout
11 years
Burnout Questionnaire
n = 143
Berg 1994
Quantitative Quasi-experimental pre/post
Psychogeriatric hospital (Sweden)
Nursing
Nursing
Group
Fortnightly–every third week
120 minutes
Burnout
11 years
MBI
n = 39
Job Satisfaction
Satisfaction with Nursing Care
Well-being
Tedium Measure
Work Environment CCQ
Berg 1999
Quantitative Quasi-experimental pre/post
Psychiatric hospital (Sweden)
Nursing
Nursing
Group
Fortnightly
180 minutes
Job Satisfaction
14 years
SNCW
n = 22
Well-being
SOC
WRSI
Work Environment
CCQ
Berry 2019
Quantitative Cross sectional
Psychiatric hospital (UK)
Nursing
N/S
N/S
N/S
N/S
Burnout MBI
N/S
n = 137
Best 2014
Quantitative Cross sectional
Alcohol and drug community service (Australia)
Social Work/Psychology/Counselling
N/S
Individual
Fortnightly–monthly
N/S
Work Environment
56% > 10 years
Organizational Readiness for Change Assessment
n = 43
Cooper-Nurse 2018
Quantitative Quasi-experimental cross sectional
Mental health settings (USA)
Social Work/Psychology/ Counselling
N/S
Individual face-to-face +/- over phone/online
55% less than once per week
82% >30 minutes
Burnout
MBI
N/S
Job Satisfaction
n = 60
AJDI
Ducat 2016
Qualitative Qualitative descriptive
Rural and regional areas (Australia)
Social work/Nutrition/Dietetics/ Occupational Therapy/Physiotherapy/Speech pathology/Medical radiation/Psychology
N/S
Individual
N/S
N/S
Interview question
N/S
What effect has CS had on your practice (if any)?
n = 42
Edwards 2006
Quantitative Cross sectional
Community mental health (UK)
Nursing
N/S
Individual, group or combination
57% monthly
32% >60 minutes
Burnout
52% <5 years
MBI
n = 260
Eklund 2000
Quantitative Quasi-experimental cross sectional
Acute and community psychiatric care (Sweden)
Occupational Therapy
Occupational Therapy/Social Work/Psychology Nursing/Medical
N/S
N/S
N/S
Job Satisfaction
N/S
Job Satisfaction Questionnaire
n = 291
Fischer 2013
Quantitative Quasi-experimental cross sectional
Acute Hospital (Italy)
Physiotherapy
N/S
Individual or group
N/S
N/S
Burnout
13 years
MBI
n = 132
Gonge 2011
Quantitative Cross sectional
Psychiatric hospital wards and community mental health centres (Denmark)
Nursing
Psychiatry/ Psychology
Group
Every two months
90 minutes
Burnout
MBI
Job Satisfaction
N/S
CPQ
Well-being
CPQ
n = 145
SF-36
Hallberg 1994
Quantitative Quasi-experimental pre/post
Paediatric psychiatric ward (Sweden)
Nursing
Nursing
Group
Every third week
120 minutes
Burnout
MBI
Job Satisfaction
15 years
Satisfaction with Nursing Care
n = 11
Well-being
Tedium Measure
Hussein 2019
Quantitative Cross sectional
Acute hospital (Australia)
Nursing
N/S
N/S
N/S
N/S
Job Retention
1 year
Modified Nurse Retention Index
n = 87
Hyrkäs 2005
Quantitative Cross sectional
Acute hospitals (Finland)
Nursing
Nursing/Psychology
Individual or group
67% every three weeks or monthly
34% 60 minutes duration
Burnout
57% > 10 years
MBI
n = 569
Job Satisfaction
Minnesota Job Satisfaction Scale
Kavanagh 2003
Quantitative Cross sectional
Hospital and community mental health settings (Australia)
Social Work/ Psychology/Occupational Therapy/ Speech Therapy
N/S
Individual, group or combination
Monthly
120 minutes
Job Satisfaction
8 years
Hoppock Job Satisfaction Measure
n = 199
Koivu 2012
Quantitative Quasi-experimental cross sectional
Acute hospital medical and surgical wards (Finland)
Nursing
N/S
Group
Every 3 or 4 weeks
90 minutes
Burnout
15 to 17 years
MBI-GS
n = 304
Well-being
GHQ-12
Work Environment
QPSNordic
Livini 2012
Quantitative Quasi-experimental pre/post
Drug and alcohol service (Australia)
Nursing/Psychology/Social Work/Counselling
Nursing/Psychology
Individual, group or combination
2 to 8 sessions over 6 months
70 to 480 minutes
Burnout
MBI
Job Satisfaction
N/S
IJSS
n = 42
Well-being
Scales of psychological well-being
Long 2014
Quantitative Cross sectional
Mental Health Hospital (UK)
Nursing
N/S
Individual, group or combination
23% monthly
N/S
Well-being
28% > 7 years
BCS
n = 128
Love 2017
Quantitative Quasi-experimental cross sectional Qualitative Qualitative descriptive
Hospital and community maternity services (Australia)
Nursing
N/S
Individual, group or combination
N/S
N/S
Job Satisfaction NSWQ
17 years
Interview questions
n = 108
What can you tell me about your overall experience of CS? What, if any, benefits have you gained from CS? Has CS been of use to you in your practice and personal life?
McAuliffe 2013
Quantitative Quasi-experimental cross sectional
Obstetric care settings (Africa)
Nursing/Medical
N/S
N/S
N/S
N/S
Job Retention a
N/S
Job Satisfaction
Cohort 1 n = 540
Job Satisfaction Scale
Cohort 2 n = 541
Cohort 3 n = 480
McCarron 2017
Quantitative (Not included in the review) Qualitative Grounded theory
Psychiatric hospital (UK)
Nursing
N/S
N/S
N/S
N/S
No relevant outcomes
Cohort 1, 8.5 years n = 20
Cohort 2, 6.5 years n = 30
Interview questions
What has your experience of CS been? If you feel that your level of CS is inadequate, how do you think this impacts on you, your ability to do your job and patient care?
Nathanson 1992
Quantitative
Hospital and community services (USA)
Social work
Social work
Individual or group
N/S
N/S
Job Satisfaction a
50% ≤ 3 years
n = 196
Saxby 2016
Quantitative Cross sectional Qualitative Qualitative descriptive
Community health service (Australia)
Dietetics/Social Work/ Physiotherapy/Podiatry/ Occupational Therapy/ Psychology/Speech Therapy
N/S
Individual or group
N/S
N/S
Burnout
MBI
Job Retention
Intention to Leave Scale
57% > 10 years
Interview questions
How would you describe your experience of CS? What makes a CS effective? Any factors that reduce the effectiveness of CS? Can you give examples where CS has made a difference to: how services are delivered to clients? How workers cope with stresses in their job? how workers feel about where they work?
n = 82
Schroffel 1999
Quantitative Cross sectional
Mental health service (USA)
Social Work/Counselling/Nursing/ Psychology
N/S
Individual or group
Weekly
71% > 30 minutes
Job Satisfaction
16 years
JDI
n = 84
JIG
Severinsson 1996
Quantitative Cross sectional
Psychiatric hospital (Sweden)
Nursing
Nursing
Group
Weekly
90 minutes
Work Environment a
10 years
n = 26
Severinsson 1999
Quantitative Quasi-experimental cross sectional
Acute hospital (Sweden)
Nursing
N/S
N/S
N/S
N/S
Work Environment
N/S
Work Environment Measure
n = 158
Teasdale 2001
Quantitative Quasi-experimental cross sectional
Acute hospital and community health settings (UK)
Nursing
N/S
Individual, group or combination
N/S
N/S
Burnout
14 years
MBI
n = 211
Work Environment
Nursing in Context Questionnaire
Wallbank 2010
Quantitative Randomised controlled trial
Acute hospital obstetrics and gynaecology (UK)
Nursing/Medical
Psychology
Individual
N/S
60 minutes
Burnout
N/S
ProQol
n = 30
Well-being
IES
ProQol
Webster 1999
Quantitative Cross sectional
Community mental health services (USA)
Social Work/Medical/Psychology/ Counselling/Nursing
N/S
N/S
N/S
N/S
Burnout
N/S
MBI
n = 151
White 1998
Qualitative Qualitative descriptive
Community, medical ward, paediatric ward, management, School of Nursing, A&E department, gynaecology ward, GP unit, residential care (UK)
Nursing
Nursing
Individual or group
N/S
N/S
Interview
N/S
Questions
N = 12
N/S
a–outcome measure not validated; AJDI–Abridged Job Descriptive Index; BCS–Bradford Clinical Supervision Scale; CCQ–Creative Climate Questionnaire; CPQ–Copenhagen Psychosocial Questionnaire; GHQ–General Health Questionnaire; IES–Impact of Event Scale; IJSS–Intrinsic Job Satisfaction Scale; JDI–Job Descriptive Index; JIG–Job in General Index; MBI–Maslach Burnout Inventory; MBI-GS–Maslach Burnout Inventory-General Survey; SNCW–Satisfaction with Nursing Care and Work; NSWQ–Nursing Workplace Satisfaction Questionnaire; SF-36–36-Item Short Form Survey; ProQol–Professional Quality of Life Scale; QPSNordic–The Nordic Questionnaire for Psychological and Social Factors at Work; SOC–Sense of Coherence Scale; WEQ–Work Environment Questionnaire; WRSI–Work-related Strain Scale.
N/S–Not stated.
Methodological quality
The predominant methodological risk of bias for analytical cross-sectional cohort studies (n = 14) was the absence of strategies to deal with confounding factors [ 36 , 39 , 41 , 45 , 46 , 48 , 52 ]. For quasi-experimental studies (n = 14) it was unclear if participants received similar support interventions other than clinical supervision in 12 studies [ 23 – 29 , 31 , 33 – 35 , 51 ], outcome measurement was not performed both pre and post intervention (i.e. multiple time points) in nine studies [ 24 , 27 – 29 , 31 , 33 – 35 , 51 ], and it was unclear if participants were similar at baseline in seven studies [ 24 , 27 – 29 , 33 , 34 , 51 ]. The single randomised controlled trial [ 22 ] only met five of the 13 items; notably the method of randomisation was unclear and there was no between group statistical comparison. JBI Critical Appraisal Checklists can be found in S1 – S3 Tables.
The methodological quality of the five qualitative studies (including the qualitative component of mixed methods studies) was generally high. Two studies [ 51 , 52 ] scored 10 out of 10, while two other studies [ 49 , 50 ] scored eight out of 10, failing to account for the potential influence of the researcher on the research findings. One study [ 53 ] did not demonstrate congruity between their stated philosophical perspective and the research methodology used, nor was there congruence between their research methodology and their research question/objectives, methods of data collection and analysis and interpretation of results. The JBI Critical Appraisal Checklist can be found in S4 Table .
Impact of clinical supervision on organisational outcomes (quantitative findings)
1. Clinical supervision compared to control
Eleven studies, including 2,965 participants, evaluated the effect of clinical supervision on organisational outcomes by comparison to a control group that did not receive clinical supervision [ 22 , 24 , 25 , 27 – 29 , 31 , 33 – 35 , 51 ]. Eight studies included nursing professionals [ 22 , 24 , 25 , 31 , 33 – 35 , 51 ], one study included social work/psychology/counselling professionals [ 27 ], two studies included other allied health professionals [ 28 , 29 ] and two studies included medical professionals [ 22 , 33 ]. While individual studies found clinical supervision had a positive effect on organisational outcomes, there were variable results across studies for burnout (six studies, n = 776 participants) ( Fig 2A–2D ), job satisfaction (four studies, n = 2,020 participants), well-being (four studies, n = 444 participants), and workplace environment (five studies, n = 783 participants). Notably, a single randomised controlled trial (n = 30 participants) found that clinical supervision had a large effect on burnout ( Fig 2D ) and well-being [ 22 ]. Results from individual studies can be found in S5 Table .
10.1371/journal.pone.0260156.g002
Fig 2
A. Supervision vs. control: emotional exhaustion (burnout) SMD 95%CI. B: Supervision vs. control: depersonalisation (burnout) SMD 95%CI. C: Supervision vs. control: personal accomplishment (burnout) SMD 95%CI. D: Supervision vs. control: overall burnout SMD 95%CI.
2. Clinical supervision compared to within-group historical control (pre/post implementation)
Six studies, including 178 participants, evaluated the effect of clinical supervision on organisational outcomes by comparing post-implementation with pre-implementation [ 22 , 23 , 25 , 26 , 30 , 32 ]. Six studies included nursing professionals [ 22 , 23 , 25 , 26 , 30 , 32 ], one study included social work/psychology/counselling professionals [ 32 ] and one study included medical professionals [ 22 ]. While individual studies found clinical supervision had a positive effect on organisational outcomes, there were variable results across studies for burnout (four studies, n = 122 participants) ( Fig 3A–3D ), job satisfaction (four studies, n = 114 participants), well-being (five studies, n = 144 participants), and workplace environment (three studies, n = 95 participants). Results from individual studies can be found in S6 Table .
10.1371/journal.pone.0260156.g003
Fig 3
A: Pre- vs. post-supervision implementation: emotional exhaustion (burnout) SMD 95%CI. B: Pre- vs. post-supervision implementation: depersonalisation (burnout) SMD 95%CI. C: Pre- vs. post-supervision implementation: personal accomplishment (burnout) SMD 95%CI. D: Pre- vs. post-supervision implementation: overall burnout SMD 95%CI.
3. Association between effective clinical supervision and organisational outcomes
Eight studies, including 1,376 participants, investigated the association between effective clinical supervision and organisational outcomes [ 37 , 38 – 43 , 52 ]. Five studies included nursing professionals [ 37 , 39 – 42 ], three studies included social work/psychology/counselling professionals [ 38 , 43 , 52 ] and one study included other allied health professions [ 52 ]. There was preliminary evidence to suggest that effectiveness of clinical supervision may be negatively associated with burnout and positively associated with job retention ( Table 2 ). The association between effective clinical supervision and job satisfaction was unclear.
10.1371/journal.pone.0260156.t002
Table 2 Synthesis of studies investigating association between effectiveness of clinical supervision and organisational outcomes.
Outcome
Number of studies
Number of participants
Direction of association within study (number of studies)
Effect size
-
o
+
Burnout–Emotional Exhaustion
5 [ 37 , 39 , 41 , 42 , 52 ]
1,046
3
2
1
Small to moderate
Burnout–Depersonalisation
5 [ 37 , 39 , 41 , 42 , 52 ]
1,046
4
1
0
Small
Burnout–Personal Accomplishment
5 [ 37 , 39 , 41 , 42 , 52 ]
1,046
1
3
1
Moderate
Job Retention
2 [ 40 , 52 ]
152
0
0
2
Moderate
Job Satisfaction
3 [ 41 – 43 ]
836
1
0
2
Small
Well-being
1 [ 42 ]
136
0
0
1
U/A
N/A–not applicable; U/A–Unable to calculate.
Positive association for job retention, job satisfaction, and well-being indicates effectiveness of supervision is associated with better outcome.
Negative association for burnout indicates effectiveness of supervision is associated with better outcome.
Synthesis of five studies [ 37 , 39 , 41 , 42 , 52 ], including 1,046 participants, indicated that effectiveness of clinical supervision may be negatively associated with emotional exhaustion and depersonalisation, but not associated with personal accomplishment. Three studies found a small to moderate association with emotional exhaustion [ 39 , 42 , 52 ] and four studies found small association with depersonalisation [ 37 , 39 , 41 , 42 ].
Synthesis of two studies [ 40 , 52 ], including 152 participants, indicated that effectiveness of clinical supervision may be positively associated with job retention. Both studies found a moderate association with job retention.
Synthesis of three studies [ 41 – 43 ], including 836 participants, indicated that the association between effectiveness of clinical supervision and job satisfaction was unclear. Two studies [ 41 , 42 ] found a small positive association and one study [ 43 ] found a small negative association. Results from individual studies are available in S7 Table .
4. Association between effective supervisor and organisational outcomes
Eight studies, including 1,600 participants, investigated the association between effectiveness of the supervisor and organisational outcomes [ 32 , 36 , 39 , 41 , 43 , 45 , 46 , 48 ]. Five studies included nursing professionals [ 32 , 39 , 41 , 46 , 48 ], seven studies included social work/psychology/counselling professionals [ 32 , 36 , 43 , 45 , 46 , 48 , 52 ] and one study included medical professionals [ 48 ]. There was preliminary evidence to suggest that an effective supervisor may be negatively associated with burnout, and positively associated with job satisfaction ( Table 3 ).
10.1371/journal.pone.0260156.t003
Table 3 Synthesis of results: Association between an effective supervisor and organisational outcomes.
Outcome
Number of studies
Number of participants
Direction of association within study (number of studies)
Effect size
-
o
+
Effectiveness of Supervisor
Burnout–Emotional Exhaustion
3 [ 39 , 41 , 48 ]
901
2
0
1
Small
Burnout–Depersonalisation
3 [ 39 , 41 , 48 ]
901
2
1
1
Small
Burnout–Personal Accomplishment
3 [ 39 , 41 , 48 ]
901
0
3
0
U/A
Burnout–Overall
2 [ 32 , 36 ]
150
1
1
0
Large
Job Satisfaction
5 [ 32 , 41 , 43 , 45 , 46 ]
1128
0
0
5
Small to Large
Well-being
2 [ 32 , 36 ]
180
0
1
1
Large
U/A–Unable to calculate.
Positive association for job satisfaction, role competence and well-being indicates effectiveness of supervision is associated with better outcome.
Negative association for burnout indicates effectiveness of supervision is associated with better outcome.
Synthesis of three studies, [ 39 , 41 , 48 ] including 901 participants, indicated that an effective supervisor may be negatively associated with depersonalisation but not associated with personal accomplishment. Two studies found a small association with depersonalisation [ 39 , 48 ]. The association between an effective supervisor and emotional exhaustion was unclear, with two studies finding a small negative association [ 39 , 48 ] and one study finding a small positive association [ 41 ].
Synthesis of five studies [ 32 , 41 , 43 , 45 , 46 ], including 1128 participants, indicated that an effective supervisor may be positively associated with job satisfaction. Studies found a small to large association with job satisfaction. Results from individual studies are available in S8 Table .
Healthcare professionals’ experiences of clinical supervision as it relates to organizational processes and outcomes (qualitative findings)
Five studies, including two qualitative [ 49 , 50 ] and three mixed methods studies [ 51 – 53 ], explored the experiences of healthcare professional supervisees on clinical supervision as it relates to organisational outcomes. A total of 16 findings and their illustrations were extracted. Of the 16 findings, 14 were unequivocal and two were credible. The 16 findings were organised into four categories which were further deduced to two synthesised findings. Table 4 shows a summary of the qualitative findings.
10.1371/journal.pone.0260156.t004
Table 4 Summary of qualitative findings.
Synthesised Findings
Categories
Findings
Illustrations
Synthesised Finding 1
Category 1 . 1
Some respondents felt that inadequate supervision had no impact; however, others identified personal consequences in terms of stress and burnout, feeling unsupported and there being an impact on their work, the ward and clients. (UNEQUIVOCAL)
‘I feel my confidence is affected . ’ (RN 2016) (McCarron et al 2017, p. 153)
Supervision assisted them to manage the workplace stress and hence, reduce their risk of burnout. (UNEQUIVOCAL)
‘When I first started with my supervisor I was in a really bad place . . . . and I was sort of at the point of no return , so getting my clinical supervision organized and constantly every month , that gave me back my confidence . ’ (Saxby 2016, p. 175)
Adequate clinical supervision mitigates the risk of burn out and facilitates staff retention, while inadequate clinical supervision can lead to stress and burnout.
Adequate CS mitigates the risk of burn-out, while inadequate CS can lead to stress and burnout
Supervision was helpful for the worker to gain a greater understanding of the dynamics operating in the client interaction to ensure there were no negative impacts for the worker or the client. (UNEQUIVOCAL)
‘We’re exploring . . . . the impact of that particular case on myself as a worker . . . .. it seems to make it clearer and give me insight into different ways of looking at that particular person . ’ (Saxby 2016, p. 175)
Opportunity to debrief challenging events provided supervisees with validation of their feelings and consideration of different management strategies to reduce their distress. (UNEQUIVOCAL)
‘I was absolutely gob-smacked with this new reform that could be coming in and potentially what could happen to me in terms of where I’m going to be going or that type of thing , you know , it’s quite unsettling . . .. . . but just having that opportunity to debrief and face my concerns has been helpful . ’ (Saxby 2016, p. 175–176)
Category 1 . 2
The implementation of clinical supervision as evidence that the health service management ‘cared about’ her and her colleagues and valued and wished to retain their workers. (UNEQUIVOCAL)
‘Yeah , it’s supportive and I guess it’s an indication the organization does care about us enough to push that . . . .. and they want to keep their staff . ’ (Saxby 2016, p. 173)
Supervisees’ responses illustrated that supervision did enhance job satisfaction and reduce workers’ intention to leave. (UNEQUIVOCAL)
‘Now I feel like I can still cope with what’s going on and that to me was worth it because otherwise I would probably be packing shelves at Coles or something . So it’s given me back my self worth , just from supervision . ’ (Saxby 2016, p. 174)
Implementation of effective CS facilitates staff retention and reduces their intention to leave
The supervisor played an active role in encouraging staff to undertake career developing activities. (UNEQUIVOCAL)
‘I guess , encouragement , being encouraged to do something , maybe something that you didn’t think you were capable of . . . . Yes , my supervisor . . . she’s suggested I become a supervisor , so I’ve done that and I’m going to start doing that . Yes , she makes suggestions like that from a professional development point of view . ’ (Saxby 2016, p. 173)
Synthesised Finding 2
Category 2 . 1
Midwives felt the structure of a safe space for regular reflection offered them continual opportunities for self-development especially in terms of enhanced communication and improved working relationships. (UNEQUIVOCAL)
‘For me personally it has helped with dealing with conflict stuff , and people , or my own personal issues with other people without ever having to involve them , because it was me that was able to adjust things . ’ (Midwife 6) (Love et al 2017, p.278)
CS enhances team relationships through improved communication
Midwives used words such as ‘courage’, ‘confidence’ and ‘strength’ to describe how their CS sessions had fostered in them an improved ability to engage in difficult conversations at work. (UNEQUIVOCAL)
‘I would have just been left in limbo with that situation and that person I think . So it enabled me to actually look at the situation and address it with that person . ’ (Midwife 2) (Love et al 2017, p.278)
CS improves the work environment through boosting staff morale, motivation to work, staff well-being and team relationships
Category 2 . 2
Midwives described feeling more positive about the work environment with an increased desire to ‘give back’ to the unit. (CREDIBLE)
‘It really boosted morale and got people motivated . ’ (Midwife 3) (Love et al 2017, p.278)
Prominent valuable outcomes of clinical supervision at the level of organization were the strengthened relationships with work colleagues, which on occasion was reported as a challenge for senior staff, and increased staff morale. (CREDIBLE)
‘I think if you affect staff morale , that in turn has got to affect patient morale , because the staff has such a strong influence over the patients . . . If the staff member feels supported , feels as if they’ve got somewhere to go , feel that they are not on their own and not isolated , which is how I think people do feel perhaps without the [clinical supervision] session , then you can sometimes unwittingly take it out on patients , I think . So I think it definitely affects patient care . ’ (Staff Nurse) (White et al 1998, p. 190)
CS promotes staff morale, motivation to work and well-being
Enthusiasm, growth and organisational commitment were identified by supervisors and supervisees. (UNEQUIVOCAL)
‘ ... we did an evaluation just when we had our face to face meeting , she said that she’s more enthusiastic about her position , she’s more motivated , she’s more organised and she’s been encouraged to do more skills development activities . ’ (Ducat et al 2016, p. 32)
Supervision kept workers motivated, interested and engaged in their roles of delivering health care services. These features of supervision increased allied health workers’ sense of connection to the employing organisation and decreased their intention to leave. (UNEQUIVOCAL)
‘It’s made such a difference to me as a practitioner . It helps you stay really focused on why am I here and it helps you stay focused on the positives that you are getting all the time because they are easy to forget about . ’ (Saxby 2016, p. 171)
Receiving positive feedback was particularly valuable for workers (at the time of data collection) as they were experiencing high uncertainty in many areas including changes to their roles and the focus of the service. Feedback from supervisors provided reassurance, as well as a sense of stability amid the evolving occupational landscape. (UNEQUIVOCAL)
‘It’s quite a supportive relationship , so your skills and your experience are recognised and that’s quite important in the current environment when everything else is being questioned and changed all the time . ’ (Saxby 2016, p. 172)
Supervision increased staffs’ sense of connection to the employing organisation, enabling supervisees to feel that they individually had a place within the organisation and therefore a sense of belonging to something greater than their immediate and often atomized local environment. (UNEQUIVOCAL)
‘What it does bring is a sense of being connected to the broader organisation . To feel connected , it’s just to feel connected to , that somebody has a clue what I do , that somebody thinks it’s ok , that it’s not just me floating around here hoping like crazy , I’m doing something useful . . . .. like I’m out there and nobody knows where I am or what I’m doing and that total sense of no one having you back almost . . . .. That feeling for me , the word is connected , to something bigger . ’ (Saxby 2016, p. 173)
Improved evidence-based practice, best practice, patient safety and clinical governance were identified by managers, supervisors and clinicians. (UNEQUIVOCAL)
‘ ... and , we really do need to ensure that our clinicians are doing the best practice , that they are supported to develop the skills they need for the role they do , and to have someone to support them to do that , not just measure them against it . . . ’ (Ducat et al 2016 , p . 32)
Synthesised finding 1: Adequate clinical supervision mitigates the risk of burnout and facilitates staff retention, while inadequate clinical supervision can lead to stress and burnout
Health professionals indicated that if clinical supervision was adequate or if they felt supported the risk of experiencing burnout or leaving the workplace was less likely. Conversely, health professionals, who felt that their supervision was inadequate, reported that clinical supervision had no positive impact or can lead to stress and burnout if they felt unsupported. This synthesised finding was developed from two categories comprising of seven unequivocal findings.
○ Category 1 . 1 Adequate clinical supervision mitigates the risk of burnout , while inadequate clinical supervision can lead to stress and burnout .
Adequate supervision meant that health professionals experienced the opportunity to debrief challenging events with their supervisor and gain a better understanding of patient interactions which can be stressful, and cause burnout for some staff. However, participants who felt unsupported identified stress and burnout as the negative consequences. This category was supported by four findings:
Some respondents felt that inadequate supervision had no impact; however, others identified personal consequences in terms of stress and burnout, feeling unsupported and there being an impact on their work, the ward, and clients. (Unequivocal)
Supervision assisted them to manage the workplace stress and hence, reduce their risk of burnout. (Unequivocal)
Supervision was helpful for the worker to gain a greater understanding of the dynamics operating in the client interaction to ensure there were no negative impacts for the worker or the client. (Unequivocal)
Opportunity to debrief challenging events provided supervisees with validation of their feelings and consideration of different management strategies to reduce their distress. (Unequivocal)
○ Category 1 . 2 Implementation of effective clinical supervision facilitates staff retention and reduces their intention to leave .
Participants reported that clinical supervision was a reflection that the health organisation valued their staff. Participants also indicated that supervisors encouraged staff to pursue career developments. These experiences enhanced job satisfaction and reduced staffs’ intention to leave the healthcare organisation. This category was supported by three findings:
The implementation of clinical supervision as evidence that the health service management ‘cared about’ her and her colleagues and valued and wished to retain their workers. (Unequivocal)
Supervisees’ responses illustrated that supervision did enhance job satisfaction and reduce workers’ intention to leave. (Unequivocal)
The supervisor played an active role in encouraging staff to undertake career developing activities. (Unequivocal)
Synthesised finding 2: Clinical supervision improves the work environment through boosting of staff morale, motivation to work, staff well-being and team relationships
Health professionals indicated that clinical supervision was valuable, led to increased motivation and enthusiasm at work, and provided not only reassurance to staff but also a safe space for improved working relationships. This synthesised finding was developed from two categories comprising of seven unequivocal findings and two credible findings.
○ Category 2 . 1 Clinical supervision enhances team relationships through improved communication .
Participants (ie. midwives) felt that clinical supervision offered an opportunity to enhance their ability to engage in difficult conversations with their team which is key in effective working relationships. This category was supported by two findings:
Midwives felt the structure of a safe space for regular reflection offered them continual opportunities for self-development especially in terms of enhanced communication and improved working relationships. (Unequivocal)
Midwives used words such as ‘courage’, ‘confidence’ and ‘strength’ to describe how their clinical supervision sessions had fostered in them an improved ability to engage in difficult conversations at work. (Unequivocal)
○ Category 2 . 2 Clinical supervision promotes staff morale , motivation to work and well-being .
Participants reported that having a clinical supervisor to support them and provide valuable feedback made them believe that they had a place within their organisation, increased their morale and enthusiasm at work, and improved their overall perception of their work environment. This category was supported by six findings:
Midwives described feeling more positive about the work environment with an increased desire to ‘give back’ to the unit. (Credible)
Prominent valuable outcomes of clinical supervision at the level of organization were the strengthened relationships with work colleagues, which on occasion was reported as a challenge for senior staff, and increased staff morale. (Credible)
Enthusiasm, growth and organisational commitment were identified by supervisors and supervisees. (Unequivocal)
Supervision kept workers motivated, interested, and engaged in their roles of delivering healthcare services. These features of supervision increased allied health workers’ sense of connection to the employing organisation and decreased their intention to leave. (Unequivocal)
Receiving positive feedback was particularly valuable for workers (at the time of data collection) as they were experiencing high uncertainty in many areas including changes to their roles and the focus of the service. Feedback from supervisors provided reassurance, as well as a sense of stability amidst the evolving occupational landscape. (Unequivocal)
Supervision increased health professionals’ sense of connection to the employing organisation, enabling supervisees to feel that they individually had a place within the organisation and therefore a sense of belonging to something greater than their immediate and often atomized local environment. (Unequivocal).
Integration of quantitative and qualitative evidence
Quantitative and qualitative findings in this review have been largely complementary and supportive of each other, especially on the impact of clinical supervision on burnout, staff well-being, job satisfaction, job retention and workplace environment.
Burnout
Quantitative findings have provided preliminary evidence that effective clinical supervision and effective supervisor may be negatively associated with burnout. This was also supported by qualitative findings that showed that adequate clinical supervision mitigated the risk of burnout, and that inadequate clinical supervision lead to stress and burnout.
Staff well-being
Quantitative findings from a single randomised controlled trial showed a large effect on reducing burnout and enhancing well-being. Qualitative studies supported this, showing that effective clinical supervision improved staff well-being.
Job satisfaction
Although quantitative evidence from three studies showed that the association between effective clinical supervision and job satisfaction was unclear, evidence from four studies showed a positive association of an effective supervisor with job satisfaction. Qualitative findings supported this showing that effective clinical supervision strengthened team relationships and sense of belonging to the organisation, thereby enhancing job satisfaction. This was particularly true when the supervisor was effective, provided valuable feedback and encouraged staff to pursue career developments.
Job retention
Evidence from two quantitative studies showed a moderate positive association of the effectiveness of clinical supervision with job retention. Similarly, qualitative studies showed that adequate clinical supervision facilitated staff retention.
Workplace environment
Synthesis of quantitative evidence from 11 studies investigating the effect of clinical supervision, and six studies investigating post-implementation of clinical supervision with pre-implementation, showed variable results in regard to its effect on workplace environment. However, qualitative evidence highlighted that effective feedback from supervisors were considered valuable and improved supervisee perceptions of the work environment and their sense of belonging to the organisation.
In summary, both the quantitative and qualitative evidence highlight that effective clinical supervision and effective clinical supervisors may be associated with positive organisational outcomes, whereas, ineffective or inadequate clinical supervision and ineffective supervisors may have a negative impact on the well-being of the supervisee.
Discussion
This systematic review of 32 studies is the first known synthesis of quantitative and qualitative evidence to further our knowledge on the impact from, and experiences of, clinical supervision of post-qualification health professionals, on organisational outcomes. Quantitative findings indicate that clinical supervision can have variable effects on organisational outcomes. The effectiveness of both the clinical supervision and the supervisor appear to influence this effect; effective clinical supervision is associated with lower burnout and greater staff retention, and an effective supervisor is associated with lower burnout and greater job satisfaction. This is supported by the qualitative findings which show that healthcare professionals believe adequate clinical supervision can mitigate the risk of burnout, facilitate staff retention, and improve the work environment, while inadequate clinical supervision can lead to stress and burnout. Overall, qualitative synthesis highlights that the effectiveness of clinical supervision and supervisors can significantly influence the effect of clinical supervision on organisational outcomes.
Effective clinical supervision and effective supervisors may be pre-cursors for the realisation of beneficial effects of clinical supervision by healthcare organisations. This is consistent with a model of clinical supervision, for post-qualification health professionals, proposed by Gonge and Buss [ 42 ], where participation in effective clinical supervision (ie. prioritising supervision time) is a pre-requisite to beneficial clinical supervision. While clinical supervision has become increasingly mandated in many healthcare organisations, through standard policies and procedures, the subsequent challenge lies in its effective and consistent implementation and uptake. This can be achieved in several ways. Organisations can adopt/utilise evidence-informed clinical supervision frameworks to guide supervision, such as the one recently developed by Rothwell and colleagues [ 54 ]. This review by Rothwell and colleagues, based on evidence from 135 studies, encourages organisations to consider making supervision mandatory to increase the value placed on it, and provide protected time for supervisors and supervisees to engage with it. It also offers several practical strategies such as providing staff with both one-to-one and group supervision options, facilitating a person-centred supervision approach with clear boundaries, tasks, ground rules and record keeping processes, and provision of ongoing training to supervisors and supervisees [ 54 ]. Implementation and uptake of clinical supervision can be completed by building a positive organisational culture that supports engagement in and uptake of clinical supervision [ 54 ], which could be regularly monitored through routine evaluations. Such evaluations will be critical to identify and respond to what clinical supervision strategies have worked, or not worked, for whom and why. Based on our work in this field, we believe that the organisational context can have an important role, and there is no one-size fits all approach when it comes to supporting the implementation and uptake of clinical supervision within organisations.
Healthcare organisations also need to support clinical supervisors to build and foster positive supervisory relationships with their supervisees. This has commonly been reported to be the single most important factor that influences the effectiveness of clinical supervision [ 3 , 11 , 54 ], and requires investment of both time and resources. Supervisors and supervisees can also be guided by evidence-informed principles that facilitate effective clinical supervision. For example, Martin and colleagues [ 11 ] provide several practical recommendations for supervisors and supervisees to enhance the effectiveness of clinical supervision, such as the development of a supervision contract, undertaking sessions at an optimal length and frequency, utilising different modes including telesupervision, evaluating supervision, and working on skills and abilities such as open communication, flexibility, trust and availability to foster a positive supervisory relationship [ 11 ]. Health professionals can be provided with continuing professional development opportunities to upskill in evidence-informed supervision practices [ 3 , 55 ]. There is evidence from a longitudinal, multi-methods study to support the delivery of supervision training in various modes such as videoconference, online and blended modes, thereby catering to those that can’t access face-to-face training. In this study, participants knowledge and confidence in the provision of supervision increased after training, which was also sustained at three-months post-training across all the four modes. This success was attributed to the careful design and delivery of training across different modes, which maximised participant access to training [ 56 ].
This review found various methodological concerns across many studies reviewed, which is consistent with findings from a recent survey of 20 systematic reviews on clinical supervision reported between 1995 to 2019 [ 3 ]. Methodological concerns include predominance of ex post facto, cross-sectional, correlational designs, small sample sizes, over reliance on self-report measures, lack of psychometrically sound supervision measures, and lack of experimental and longitudinal designs [ 3 ]. Incomplete provision of information (on clinical supervision parameters) seems to continue to plague supervision research, as again found in the survey of supervision reviews [ 3 ], and in the systematic review reported here, making it hard to judge the full merit of the study or replicate it. There is a need for further rigorous high-quality studies in this area that use pluralistic research approaches where experimental investigation, randomisation, and data-driven case studies are used in conjunction with ex post facto, and cross-sectional designs [ 3 ]. Studies also need to better define the specifics of the clinical supervision intervention to allow replication and identification of the clinical supervision practices that are, or are not, effective for improving outcomes.
Limitations
The final review deviated from the protocol to also include group supervision, as many studies did not specify the type of supervision investigated. However, group supervision is commonly practiced in healthcare organisations and including these studies in this review likely improves the generalisability of our findings. Although the qualitative studies included were deemed to be of good quality, there were several shortcomings in the methodologies employed by the quantitative studies, especially the lack of randomised trials and absence of strategies to deal with confounding factors in cross-sectional studies. Although there were a variety of healthcare settings and health professionals represented in this review, the majority of included studies were conducted in mental health settings with nursing and/or mental health disciplines (i.e. psychology, counselling, and social work). This may limit the generalisability of the results to other disciplines and indicates the need for further research beyond mental health settings and nursing/mental health disciplines.
Conclusions
Clinical supervision can have a variable effect on healthcare organisational outcomes. This effect appears to be influenced by the effectiveness of both the clinical supervision provided and that of the clinical supervisor. This highlights the need for organisations to invest in high quality supervision practices if they wish to benefit from clinical supervision. Without such investment, there is a risk of policy-practice gaps in this area (i.e. while there may be policies to support clinical supervision in healthcare organisations, in practice it may not be implemented well). Ongoing further research, which grows the evidence base for high quality clinical supervision and helps to unpack the black box of clinical supervision practices that have the most effect on organisational outcomes, is required.
Supporting information
S1 Checklist
PRISMA checklist.
(DOC)
S1 Table
JBI critical appraisal checklist for randomised controlled trials.
(DOCX)
S2 Table
JBI critical appraisal checklist for quasi-experimental studies.
(DOCX)
S3 Table
JBI critical appraisal checklist for analytical cross sectional studies.
(DOCX)
S4 Table
JBI critical appraisal checklist for qualitative studies (including qualitative component of mixed methods studies).
(DOCX)
S5 Table
Results of studies investigating the effect of clinical supervision on organisational outcomes compared to control (no supervision).
(DOCX)
S6 Table
Results of studies investigating the effect of clinical supervision on organisational outcomes pre/post implementation.
(DOCX)
S7 Table
Results of studies investigating the association between effectiveness of clinical supervision and organisational outcomes.
(DOCX)
S8 Table
Results of studies investigating the association between an effective supervisor and organisational outcomes.
(DOCX)
S1 Appendix
Search strategy.
(DOCX)
S2 Appendix
Excluded studies.
(DOCX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Taenia solium , the pork tapeworm, is a parasite that imposes a substantial burden on the health and livelihoods of subsistence farming communities in developing countries of Asia, Latin America and sub-Saharan Africa, including Zambia [ 1 ]. This cestode is the causative agent of T . solium taeniosis/cysticercosis (TSTC), a neglected zoonotic disease complex that accounts for the highest global burden of foodborne parasitic diseases and acts as a leading cause of deaths from foodborne diseases [ 2 – 4 ]. It is a significant contributor to late-onset epilepsy in tropical regions around the world [ 5 , 6 ], and causes pig production losses, driving smallholder farmers into poverty or even destitution [ 7 , 8 ]. The importance of T . solium has been recognized by the World Health Organization, which listed TSTC as a major neglected tropical disease (NTD) and has targeted it for control [ 9 ].
The human definitive host acquires the intestinal tapeworm infection (taeniosis) by ingestion of raw or undercooked pork containing viable cysticerci. The adult worm produces infective eggs that are passed in the host’s stool, contaminating the environment. Pigs act as intermediate hosts after ingestion of these eggs and develop porcine cysticercosis (PCC), a tissue infection with larval cysts (cysticerci). Humans can act as dead-end intermediate host (human cysticercosis, HCC); with larval cysts commonly found in the central nervous system (neurocysticercosis, NCC) [ 1 ].
Several studies in Zambia reported high prevalences of PCC, HCC and taeniosis, confirming a high T . solium endemicity [ 10 – 14 ]. PCC antigen prevalences of 28.3%, 16.9% and 30.0% were found for the Southern, Eastern and Western Province of Zambia respectively when 1,691 pigs were examined [ 10 ]. Like in many other developing countries, there is an increasing demand for pork in Zambia, especially in rural areas [ 15 , 16 ] where pigs are mostly kept under smallholder conditions as a low input transitory activity with pigs roaming freely [ 17 ]. Moreover, pig slaughter often occurs informally in backyards or at illegal slaughter establishments with little or no controlled meat inspection [ 15 ]. This pork supply chain creates a favorable environment for the propagation of T . solium , which is further aggravated by a reluctance to build and use latrines by most Zambian rural households [ 18 ]. Addressing practices at the level of pork production and distribution may prove valuable T . solium control points. Indeed, previous PRECEDE-PROCEDE [ 19 ] assessments have identified pork producers and sellers as important target groups for T . solium health promotion strategies [ 20 , 21 ].
Lack of knowledge is an acknowledged barrier for control. Previous T . solium health education interventions in India [ 22 ], Kenya [ 23 ], Mexico [ 24 ], Tanzania [ 20 , 25 – 29 ] and Zambia [ 30 , 31 ] led to significant increases in knowledge among targeted stakeholders. Moreover, attitudinal and behavioral change were reported in different T . solium health education studies; including increased condemnation of infected meat, better confinement of pigs and improved sanitary/hygienic practices [ 20 , 22 – 25 , 27 – 30 , 32 ]. More importantly, health education proved able to reduce PCC [ 20 , 24 , 25 ] and HCC [ 32 ] incidence. An additional advantage of health education is that the promotion of good sanitation and hygiene also decreases the disease burden of many other sanitation-related pathogens [ 1 , 33 , 34 ], such as soil transmitted helminths [ 35 ] and diarrheal agents [ 36 ].
Different health education approaches have been tested; including the use of informational movies, street plays, songs, interactive discussions, leaflets, posters, banners, trained teachers and software programs [ 22 – 28 , 30 , 31 ]. ‘The Vicious Worm’ (TVW, https://theviciousworm.be/ ) is a free educational software program that provides evidence-based information on T . solium through illustrated short stories, videos, quizzes and scientific texts. It aims to raise awareness of T . solium and reduce risk behaviors such as open defecation, keeping pigs free-ranging and cooking pork inadequately. The program supports standardized education while three dedicated levels (village, town and city level) allow more targeted conveyance of information to different stakeholders (laypeople, professionals and policy makers) [ 37 ]. The software is currently available as computer program and mobile application in English and Swahili (Minyoo Matata) [ 37 , 38 ].
Previous studies found that health education with TVW provided significant short- and long-term knowledge uptake in Zambian primary school students [ 30 , 31 ] and Tanzanian medical and agricultural professionals [ 26 , 28 ]. Furthermore, TVW was well-received by study participants and exposure to the program was reported to lead to positive behavioral changes and effective knowledge dissemination [ 20 , 26 , 28 , 30 ]. Our study aimed to assess the effects of TVW health education on key actors in the Zambian pork supply chain.
Methods
Ethics statement
Ethical clearance was obtained from the Excellence in Research Ethics and Science Converge (ERES) Institutional Review Board of Lusaka, Zambia (Approval No. 2018-Dec-012) and the Ethical Committee of the University of Antwerp, Belgium (Approval No. B300201628043, EC UZA16/8/73).
The study was explained to the participants at the start of the educational workshop, before names and phone numbers (if applicable) were registered on an ID list to facilitate future contact. Written informed consent was sought from the individual participants. For illiterate participants, a thumbprint was used and a witness signed on their behalf. Participation did not involve any noteworthy risk, and all collected data was processed confidentially.
Attendance to the educational sessions was voluntary and participants were free to leave the study at any time. The workshops took place during working hours; hence permission was sought from the employers. Each participant was offered a free beverage and snack during the sessions, and a small monetary incentive for participation and lost working time.
Study area
The study was carried out at two study sites in Zambia, where the national pig production quadrupled between 2000 and 2018 (0.3 million to 1.2 million) [ 39 ].
The first study site was Lusaka district (Lusaka province), where pig slaughter and meat cutting often took place at an unofficial slaughter slab, before the pork was sold at the establishment or taken to markets in surrounding high density areas within the city. Most of these animals were village pigs from resource-poor farmers, largely originating from rural areas of the Southern Province of Zambia [ 11 , 40 , 41 ]. A previous study in 868 village pigs slaughtered at this slaughter slab, found a PCC prevalence of 64% [ 11 ].
The second study site was Katete district (Eastern province), where most households performed home slaughter [ 14 , 42 , 43 ]. A 2015 study in this district found a PCC prevalence of 46% (17/37) in randomly selected slaughter age pigs based on full carcass dissection, which is regarded as the golden standard detection technique [ 12 ]. Other studies in the district found very high prevalences of HCC/NCC [ 14 , 44 ] and indicated NCC as the single most important cause of epilepsy in the study area [ 44 ].
Study design
The cross-sectional study was conducted in March and April 2019. At both study sites, an educational workshop was organized with a subsequent follow-up session after three weeks.
The educational workshop included a brief study introduction; participant registration and signing of informed consent forms; a short survey on the participant’s general information (7 questions); a ‘pre’ questionnaire survey; a short break; an educational component using TVW; a dialogue addressing questions about TVW; and a ‘post’ questionnaire survey. After the workshop, USB sticks with TVW software were handed out to participants with access to computers, facilitating the continuation of their training and knowledge sharing.
The follow-up session consisted of a ‘3weekspost’ questionnaire survey and focus group discussions (FGDs). Additionally, observational data were collected at three time points per study site throughout the seven-week study period ( Fig 1 ).
10.1371/journal.pntd.0008790.g001
Fig 1
Simplified study timeline.
Observational data were collected throughout the seven-week study period.
For the Lusaka study site, pork supply chain workers were gathered at the unofficial Chibolya slaughter slab, where the observational data were also collected. For the Katete study site, pork supply chain workers from the area were gathered at a local lodge, while observational data was collected throughout the district.
Selection of study subjects
All professionals who were involved in the pork supply chain at the level of live trading, slaughtering, meat cutting and distribution in the study area were eligible to participate in the study. These selection criteria included professional pig traders, slaughterslab workers, butchers, and meat inspectors. Slaughterslab workers include the professionals who are responsible for the pig slaughter, the meat cutting, the selling of pork and the management at the slaughter slab facility. Professionals who reared pigs (e.g. pig farmers) and/or cooked pork (e.g. cooks), but were not active in live trading, slaughtering, meat cutting or distribution, were excluded. Each participant was assigned an individual code, enabling confidential data processing.
Questionnaires
The ‘pre’, ‘post’ and ‘3weekspost’ questionnaires consisted of 22 multiple choice questions about T . solium ( S1 File ): 19 quantitative and 3 qualitative questions, based on the previously validated questionnaire [ 45 ]. The quantitative questions aimed to test the participants’ knowledge about T . solium and were coded into six categories: T . solium in general, acquisition & transmission, PCC in live pigs, PCC in slaughtered pigs, taeniosis and HCC/NCC. Furthermore, the influence of participants’ characteristics on their baseline knowledge and knowledge uptake was assessed using these quantitative questions in combination with the 7 questions concerning the participant’s general information. There were two to six answer options per quantitative question with one correct answer. The qualitative questions were used to obtain information about the perceived PCC occurrence and the handling of PCC cases. During all three surveys, the same questions were asked, but the ‘post’ and ‘3weekspost’ questionnaires had a different question sequence compared to the ‘pre’ questionnaire. The questionnaire was written in English and a monochrome paper copy was handed to each participant, in addition to a pen. The questionnaire was projected in color on a white cloth surface while the moderator read the questions aloud in English and translated them in the local language (Nyanja-Chewa). The questions were repeated until they were clear to the participants.
Educational component using ‘The Vicious Worm’
Approximately one hour and a half were assigned to the educational component of the study, with most attention devoted to TVW’s ‘village level’, where scenes from an African village are depicted. In this level, information is presented in a simple manner, through illustrated short stories. Furthermore, the slaughterhouse section from the ‘town level’ was used to give more detailed and technical information about pig slaughter, meat inspection and good meat hygiene practices.
A fifteen-minute break with light refreshments was included between the pre-questionnaire and educational component. The English version of the software was projected on a white cloth as the moderator navigated through the program, reading (English) and translating (Nyanja-Chewa) the visible content.
Focus group discussions
A total of four FGDs were conducted in the local language (Nyanja-Chewa) after the participants had completed the ‘3weekspost’ questionnaire. Each group consisted of one moderator and nine to twelve study subjects. In the four FGDs, the participants’ opinion of the program was evaluated, as well as their beliefs, attitudes, and insights concerning T . solium . Furthermore, participants were given the opportunity to suggest improvements and alternatives to the health education tool, and express the needs and wishes of local pork supply chain workers to enhance T . solium control. Furthermore, video footage of seizures in a human (3 min. fragment) and two pigs (1 min. and 2 min. fragments) was projected on a white cloth surface. Afterwards, participants were asked whether they considered it useful and acceptable to use such footage in TVW to highlight the severity of the disease and the link between HCC and PCC. All discussions were video recorded to facilitate the transcription of the discussions involving several individuals at the same time. A backup audio recording was made with a dictaphone app on a smartphone to ensure full coverage of the FGDs.
Data management and statistics
The acquired quantitative and qualitative data were analyzed to assess knowledge uptake and short-term knowledge retention among key actors in the Zambian pork supply chain, and their attitude towards TVW. Furthermore, it inquired about possible improvements and alternatives to the software, the acceptability of using video footage demonstrating seizures in humans and pigs in TVW, and the needs and wishes of pork supply chain workers to enhance T . solium control.
Questionnaires
The questionnaire data was entered in an Excel spreadsheet (Microsoft Office Professional 2016) where questions (q) were assessed individually and by category (c) ( S1 Data ). Each quantitative question was assigned a score of 1 when answered correctly, or 0 when answered incorrectly. When questions were not answered or when the indications were unclear, NA was assigned. For each questionnaire, data were independently entered twice in separate excel files. Both entries were subsequently compared using Spreadsheet Compare (Microsoft Office Professional 2016) and mismatches were revised. To simplify the comparison of the answers at different time points, each question was assigned a code from q0 to q22, based on their sequence in the pre-questionnaire. Similarly, the 19 quantitative questions were assigned a second code from c1 to c6, to estimate the knowledge per category at each time point.
To assess knowledge uptake between the ‘pre’, ‘post’, and ‘3weekspost’ time points, we fitted generalized linear mixed models with logit link to the questionnaire results (number of correct answers over total number of completed answers). We used participant ID as a random effect to account for within-individual clustering. The period time point was included as a fixed effect. We used Tukey's all-pair comparisons to assess all possible pairwise differences between the three time points (i.e., post vs ‘pre’, ‘3weekspost’ vs ‘pre’, and ‘3weekspost’ vs ‘post’).
We also assessed the effect of covariates on baseline knowledge and on knowledge uptake (‘post’ vs ‘pre’). For the effect on baseline knowledge, we fitted similar models as above, but now using the specific covariate as fixed effect, and limiting the dataset to the pre responses only. For the effect on knowledge uptake, we fitted the above-mentioned models with addition of an interaction effect between period time point and the specific covariate. For all models, we reported odds ratios (OR), and corresponding 95% confidence intervals (CI) and p-values.
All analyses were performed in R [ 46 ], using the ‘lme4’ [ 47 ] and ‘multcomp’ [ 48 ] packages.
Focus group discussions
The recordings were transcribed and translated into English and each focus group was assigned a code for data analysis. Data analysis was performed using NVivo qualitative data analysis software [ 49 ]. All data was coded (172 codes with 290 references), which allowed classification and sorting of data, so that relationships and trends in the data could be examined. The major themes were separately identified using an inductive approach [ 18 , 26 ].
The data was coded into seven major themes: ‘Positive aspects of TVW’, ‘Possible improvements to TVW’, ‘Alternative methods to disseminate knowledge’, ‘Assessment of the seizure videos’, ‘Control hurdles’, ‘Behavioral change’ and ‘Extra’. The key points of the seven themes were summarized with supporting statements by participants ( S2 File ).
Observations
Observational data was collected at three time points per study site within the 7-week period. Pictures were taken with a smartphone and an observational checklist was filled out immediately after leaving the sites ( S3 File ).
Results
The data from 47 study subjects were included in the study, with 25 and 22 from Lusaka and Katete, respectively. In Lusaka, the study participants consisted of 10 pig traders (40%) and 15 slaughter slab workers (60%; 12 slaughterers, 2 board members and 1 cashier). In Katete, no conventional pork supply chain with slaughter slabs or slaughterhouses was present. All 22 participants from Katete performed all steps from acquisition of pigs to the selling of pork and offal and will be referred to as butchers.
The study group consisted of 43 males (91.5%) and 4 females (8.5%), with all female participants being part of the Lusaka group. The ages ranged from 22 to 52 years, with an average of 36 years.
The follow-up sessions were attended by 43 of the original 47 participants (Lusaka: 24/25, Katete: 19/22). The participants with access to computers (11/47) and who consequently received USB sticks at the end of the workshop were all from the Lusaka study site.
Questionnaires
On average, about one hour was needed to complete the questionnaires. The average baseline T . solium knowledge was 62% (Lusaka: 56%, Katete: 68%), with 11 participants scoring less than 50% (all from Lusaka group) and 5 participants scoring more than 75%. At baseline, 91% (43/47) of participants had heard about cysticercosis which is locally known as masese , mase or m’sokwe .
At the end of the workshop (‘post’), the overall average score increased significantly (+20%, p<0.001) to 82% (Lusaka: 81%, Katete: 84%), with 1 participant scoring less than 50% (Lusaka group) and 36 participants scoring more than 75%. When the knowledge was tested at the ‘3weekspost’ time point, there was no significant difference in the overall average score compared to the ‘post’ time point (+2.5%, p = 0.320). With an overall average score of 85% (Lusaka: 85%, Katete: 84%), the ‘3weekspost’ knowledge remained significantly higher compared to the ‘pre’ time point (p<0.001) ( Fig 2 ).
10.1371/journal.pntd.0008790.g002
Fig 2
Average knowledge scores over all six categories at each of the three time points.
Baseline knowledge
Averages per category ranged from 42–88% with five of the six category scores exceeding 50% ( Table 1 ). The least understood category of T . solium at baseline was ‘acquisition and transmission’ (42%), especially the acquisition of PCC (11%) and NCC/HCC (13%). Furthermore, knowledge about PCC prevention (17%) and the relation between PCC, HCC and taeniosis (cysticercosis-taeniosis: 28%; PCC-HCC: 44%) were lacking. In total, seven individual questions received less than 50% correct answers at the ‘pre’ time point.
10.1371/journal.pntd.0008790.t001
Table 1 Results from ‘pre’, ‘post’ and ‘3weekspost’ questionnaire surveys and associated knowledge changes between the different time points. P-values of the knowledge changes are listed between brackets.
EDUCATIONAL WORKSHOPS
FOLLOW-UP SESSIONS
Category
Pre (%)
Post-Pre
Post (%)
3weekspost-Post
3weeks post (%)
3weekspost-Pre
T . solium in general
53%
+36% (p<0.001)
89%
-2.0% (p = 0.904)
87%
+34% (p<0.001)
Acquisition & transmission
42%
+23% (p<0.001)
65%
-3.0% (p = 0.779)
62%
+20% (p<0.001)
PCC 1 in live pigs
53%
+33% (p<0.001)
86%
+5.0% (p = 0.309)
91%
+38% (p<0.001)
PCC 1 in slaughtered pigs
88%
-5.0% (p = 0.398)
83%
+13% (p = 0.003)
96%
+8.0% (p = 0.053)
Taeniosis
72%
+20% (p<0.001)
92%
+5.0% (p = 0.212)
97%
+25% (p<0.001)
HCC 2 /NCC 3
80%
+9.0% (p = 0.138)
89%
+2.0% (p = 0.943)
91%
+11% (p = 0.084)
Overall
62%
+20% (p<0.001)
82%
+3.0% (p = 0.320)
85%
+23% (p<0.001)
1 PCC: porcine cysticercosis, 2 HCC: human cysticercosis, 3 NCC: neurocysticercosis
Baseline knowledge was significantly higher (p<0.001) in the Katete study group compared to the Lusaka study group; with butchers scoring significantly higher than slaughter slab workers (p<0.001) and pig traders (p = 0.007). Furthermore, participants who had owned pigs scored significantly higher (p<0.001) than those who had not ( Table 2 ).
10.1371/journal.pntd.0008790.t002
Table 2 Association between participants’ characteristics and their baseline knowledge & knowledge uptake.
BASELINE KNOWLEDGE
KNOWLEDGE UPTAKE 1
Variable
Level (n)
OR (95% CI)
P-value
OR (95% CI)
P-value
Gender
Male (43)
1
1
Female (4)
-0.473 (-1.040; 0.094)
0.102
0.020 (-0.747; 0.786)
0.960
Age (years)
age ≤ 30 (10)
1
1
30 < age ≤ 40 (24)
0.341 (-0.037; -0.294)
0.077
-0.075 (-0.668; -1.255)
0.803
age > 40 (11)
0.145 (0.719; 0.585)
0.517
-0.601 (0.517; 0.052)
0.071
District
Katete (22)
1
1
Lusaka (25)
-0.531 (-0.817; -0.245)
<0.001
0.344 (-0.110; 0.799)
0.138
Job
Butcher (22)
1
1
Slaughter slab worker 2 (12)
-0.550 (-0.877; -0.870)
<0.001
0.068 (-0.438; 0.212)
0.792
Trader (10)
-0.503 (-0.224; -0.135)
0.007
0.842 (0.574; 1.471)
0.009
Level of education
Secondary school (18)
1
1
Primary school (26)
-0.071 (-0.404; -1.075)
0.674
-0.869 (-1.372; -0.939)
<0.001
None (3)
-0.412 (0.261; 0.250)
0.222
0.069 (-0.366; 1.078)
0.893
Pig ownership
No (27)
1
1
Yes (20)
0.533 (0.242; 0.824)
<0.001
-0.177 (-0.640; 0.287)
0.455
1 Knowledge uptake refers the knowledge change between the ‘post’ and ‘pre’ time points (educational workshop).
2 Slaughterslab workers include 12 slaughterers, 2 slaughterslab executives and 1 slaughterslab cashier.
n = number of participants
Knowledge uptake
Knowledge increases from baseline were found in all categories except ‘PCC in slaughtered pigs’ and ‘HCC/NCC’, both immediately and three weeks after the use of TVW ( Table 1 ). There were no categories with significant knowledge decreases and all six category scores exceeded 50% immediately and three weeks after the educational component.
Two individual questions received less than 50% correct answers at the ‘post’ and ‘3weekspost’ time points, both concerning NCC. Firstly, the acquisition of NCC remained poorly understood (‘post’: 30%, ‘3weekspost’: 33%) despite a small increase in correct answers. Secondly, the question regarding NCC’s transmissibility received a lot of incorrect answers (‘post’: 40%%, ‘3weekspost’: 12%) despite being relatively well answered at baseline (57%).
Knowledge uptake was significantly higher (p<0.001) in participants with a secondary school diploma than those with only a primary school diploma ( Table 2 ). Traders had a significantly higher (p = 0.007) knowledge uptake from baseline than butchers at the end of the educational workshop.
Perceived PCC occurrence
At the end of the educational workshop, half of the participants (23/46) indicated to detect PCC in at least one pig a week. All but one participant had observed PCC in a slaughtered pig and the perceived occurrence of PCC was higher in Katete than in Lusaka. During the short survey on the participant’s general information (7 questions), all participants from Katete stated that meat inspection was always performed. However, 52% (13/25) of the Lusaka participants answered that meat inspection was only performed when there was a meat inspector around. Moreover, three participants from Lusaka (12%, 3/25) replied that meat inspection was never performed.
Focus group discussions
Four FGDs were conducted and an average of 11 participants were involved in each discussion. The groups from Lusaka and Katete were referred to as L1/L2 and K1/K2, respectively. The average duration of an FGD was 30 minutes (20–40 minutes) including the demonstration of the video fragments (6 minutes in total).
All groups generally expressed a strong positive attitude towards TVW. The most mentioned positive aspects of the program over all four groups were the educative value, the simplicity/clarity, the good illustrations, and the potential to share the knowledge.
“I think that the program was very good because now we are protected by knowledge . Now we know how you get the tapeworm and how you get masese .” (Focus group K2)
The most mentioned potential improvements to TVW were increased accessibility (electronics needed), larger scale and addition of videos (not cartoons). The acquisition of NCC and the link with seizures was mentioned to need further clarification by both focus groups from Lusaka.
The most mentioned alternative methods to disseminate knowledge about T . solium were drama/sketch, radio, and booklets. In the K1 group, a participant argued that using a booklet would not be helpful because many people were illiterate. Other methods that were discussed included flyers, television, poems, and community meetings.
All groups unanimously agreed that the seizure videos were acceptable to be included in TVW, except in FGD L2 where one participant argued that “ the videos will lead to poor business as people will stop buying the pork .” In all FGDs it was stated that the videos would help to convey the importance of T . solium and would promote the adoption of control measures. Participants from L2, K1 and K2 recommended to properly teach people about T . solium before showing the videos to avoid misunderstandings that could lead to the shunning of pork. Other mentioned benefits of the inclusion of seizure videos were improved recognition of the disease, and the tackling of the misbelief that seizures are the result of witchcraft.
The FGDs also addressed control hurdles including lack of knowledge about T . solium , trade of infected pigs and/or pork, free-ranging pigs, inadequate meat inspection, inadequate sanitation, consumption of undercooked pork, unknown origin of vegetables, unavailability of drugs and difficulties to maintain a good relationship with the public. Based on the FGD data, the lack of knowledge was mostly concerning the acquisition and transmission of T . solium infections and the relation between PCC, HCC/NCC and taeniosis. The participants referred to popular misconceptions/misbeliefs including acquisition of PCC through consumption of beer brewing residues, and seizures in humans as a result of witchcraft that had to be treated by witch doctors. A participant from the K1 group also mentioned that PCC symptoms were sometimes falsely ascribed to African Swine Fever.
“In the villages everyone will just say that a person with epilepsy has been bewitched . Even we did not know that masese can affect people , but now we have learned . ” (Focus group K1)
During the FGDs we did not explicitly ask whether the participants had adjusted their attitude or behavior after the educational workshop to prevent the provocation of socially desirable answers. Nevertheless, many participants from all groups expressed attitudinal change, and several participants from L1 mentioned behavioral change since the educational workshop. This included the examination of the tongue to detect PCC in live pigs, not selling infected pigs/pork, sending a potential NCC patient to the hospital, thoroughly cooking pork, and teaching others. The L2 group also mentioned that the USB sticks had been used by several participants during the 3-week interval.
“I have taught at least 4 or 5 people since learning about the worm . I used what I learned , the knowledge I got from you , to teach them about masese . ” (Focus group L1)
All groups reported to have observed cysticerci in pigs and seizures in both humans and pigs, with a participant from L2 stating that they had seen many people having seizures. A participant from the K2 group also referred to the ubiquity of pig rearing, stating that almost everyone in the village had reared pigs.
Observations
Several breaches to good hygiene practices were observed at both study sites, such as inadequate cleaning/disinfection of hands/tools/clothes, lacking protective clothing, public access to production areas, absence of a cold chain, inappropriate waste, and pest control, etc.
Observed failures in meat processing hygiene included: slaughter of a conscious pig on a patch of dirt, carcass splitting and evisceration on a soiled wet floor, piling of meat and offal from different carcasses, inappropriate carcass and meat transportation using wheelbarrows, etc. Furthermore, there was no meat inspection or official inspection observed.
Discussion
This study indicates that educational workshops using ‘The Vicious Worm’ have highly significant positive effects on T . solium knowledge uptake and short-term retention in Zambian slaughter slab workers, butchers, and pig traders.
Baseline knowledge of T . solium was relatively high, especially in the Katete study group. However, the acquisition and transmission of the parasite, and the relation between PCC, HCC and taeniosis were not well comprehended; which is in line with previous studies of varying designs in Zambia and Tanzania [ 25 , 26 , 31 , 50 ]. This lack of knowledge constitutes a major problem as it impedes the attitudinal and behavioral changes that are vital to the control of T . solium [ 25 , 51 , 52 ]. The most prominent lack of knowledge at baseline concerned the acquisition of PCC and NCC, which were also found to be the least understood aspects among Zambian primary school students when a similar questionnaire was used [ 31 , 45 ].
The educational workshops using TVW significantly increased the participants’ knowledge of most aspects of T . solium . This included considerable knowledge increases concerning PCC and taeniosis prevention, which are vital to inspire implementation of preventive measures. However, the acquisition and transmissibility of HCC/NCC remained poorly understood after the use of TVW, which was also described by Hobbs et al. and Ertel et al. [ 26 , 31 ]. Moreover, the decrease in correct answers regarding the transmissibility of NCC after use of TVW was also found in Zambian primary school children [ 31 ]. Insufficient understanding of the T . solium lifecycle despite health education has been reported after diverse educational interventions among different target groups [ 22 – 24 , 26 , 27 , 29 , 31 ]. The most common misconceptions after the educational session were the acquisition of NCC through consumption of undercooked pork and the transmission of NCC through stool. This likely reflects the complexity of the parasite’s life cycle and calls for the clarification or simplification of the NCC aspect in future versions of the software. This notion is further strengthened by the FGDs in which two participants proposed to emphasize the NCC aspects during TVW health education.
Our study found a better baseline knowledge in pig owners, butchers, and participants from Katete; however, the latter two groups overlapped. The increase in knowledge was more prominent in participants with a secondary school diploma than those with only a primary school diploma. This suggests that further simplifications of TVW might be beneficial to ensure a similar knowledge uptake among people with less advanced educational levels. Furthermore, a bigger knowledge increase was found in pig traders than in butchers, with no significant difference between butchers and slaughter slab workers. The small number of female participants (4) and participants without a diploma (3) impeded their comparison with other groups.
The FGD data confirmed the lack of knowledge regarding the acquisition and transmission of the parasite and the relation between PCC, HCC/NCC and taeniosis in the communities. Furthermore, the FGDs exposed the persistence of certain misbeliefs concerning PCC and NCC. Locally porcine cysticercosis is called with the terms masese , m’sokwe and mase , which literally translates to ‘beer dreg’ due to the resemblance of beer brewing residues with white nodular cysticerci. Therefore, a popular misbelief suggested that pigs acquire PCC after the consumption of beer brewing residues, which pigs were sometimes fed. Another popular misbelief alleged that epilepsy patients were bewitched and could be healed by witch doctors. A 2010 study by Thys et al. found the same misbeliefs to be popular in communities from the Petauke district of the Eastern province of Zambia [ 17 ]. These misperceptions highlight the potential of health education in endemic areas, where communities are unable to take well-advised precautionary measures.
The study participants expressed a positive attitude towards TVW, describing the program as simple, clear, and educative. A similar positive attitude towards the program was observed amongst Tanzanian medical and agricultural professionals [ 26 ]. The FGD data also suggest that some of the participants who received a USB stick with the software voluntarily used the software between the initial and follow-up visits. Moreover, the participants appeared committed to share the knowledge they had acquired from TVW, with several participants stating to have done so during the three weeks after the initial visit. Previous studies also reported knowledge transfer from educated individuals to the community [ 23 , 28 , 30 ] but more research is needed to assess the efficacy and effectiveness of this information transfer.
During the FGDs, participants proposed potential improvements, including support for more languages, and the addition of videos. Videos depicting seizures in pigs and humans could also enhance the software as it could help to convey the importance of T . solium and promote the adoption of control measures. During the FGDs, the accessibility was indicated as a limitation of the tool as it requires a computer, smartphone, or tablet. Therefore, the organization of group workshops with a moderator might boost the dissemination of TVW’s information; these are quite easy to organize and only require a beamer, a computer, and a small generator in the absence of electrical supply.
The observed breaches to good hygiene practices and meat processing hygiene demonstrated the importance of the pork supply chain workers in the propagation of foodborne parasites like T . solium . Furthermore, the observed lack of meat inspection or any type of official control passes the responsibility for distribution and consumption of infected pork to non-trained individuals. Therefore, slaughter slab workers, pig traders, butchers and meat inspectors could serve as target groups for future educational interventions. In addition to tackling the lack of knowledge, it could lead to long-term behavioral changes that improve PCC detection and reduce the distribution of infected pork and pigs. Furthermore, these pork supply chain workers were in close contact with pig farmers and pork consumers with whom they could share knowledge, possibly inspiring the adoption of control measures. However, increased knowledge does not automatically translate into attitudinal and behavioral change and economic and/or sociocultural factors can override the perceived and rather long-term health risk [ 18 , 25 , 53 ]. Nevertheless, our preliminary FGD data suggest incipient attitudinal and behavioral changes.
The study would have benefited from a larger number of study subjects and a control group. Social desirability bias and unanswered questions (mainly ‘pre’ questionnaire) could also affect the validity and reliability of the results. Another limitation of the study is the reliance on written English for the questionnaires, which could potentially have contributed to misunderstandings. Assessments of long-term effects, the extent of knowledge transfer and behavioral change associated with the use of TVW were outside the scope of this study but would be useful in future studies.
The results demonstrated that TVW is an effective short-term health education tool for key actors in the Zambian pork supply chain. Translation into other languages and adaptations to different sociocultural and regional contexts could allow global implementation of the specific health education tool. However, the software could benefit from the addition of videos and more emphasis on the NCC aspect of TSTC. Moreover, we recommend a well-thought-out and unhurried teaching approach in future group workshops to prevent confusion. In conclusion, the highly significant effect on knowledge uptake and positive attitude towards the program call for consideration of TVW in future intervention programs.
Supporting information
S1 Data
Questionnaire scores per participant and per question.
Spreadsheet with assessment of the answers per question, and per category.
(XLSX)
S1 File
The Vicious Worm Question List.
A list of questions asked during the ‘pre’, ‘post’ and ‘3weekspost’ questionnaire survey.
(PDF)
S2 File
Key points FGDs.
Overview of key points that were discussed during the focus group discussions.
(PDF)
S3 File
Observation checklist.
The checklist used to collect observational data.
(PDF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
A good psychological state is an important index of health. Various psychological and psychosocial factors comprise what we call a “healthy psychology”, such as life satisfaction, optimism, self-esteem, and perception of social support. On the other hand, anxiety, stress, depression, and hostility reflect a less desirable psychological state which can affect one's health in various aspects [1] . So far, numerous studies have addressed the impact of psychological, psychosocial and personality factors on several health indices [2] , [3] . Well-established evidence suggests that these factors may influence the short-term and long-term course of diseases, as well as patients' recovery and survival [4] – [6] .
Scientists have been investigating the pathophysiological mechanisms through which psychology may influence the immunologic functions of human body, and several biobehavioral models have been proposed [7] . This field of medicine is called psychoneuroimmunology. More specifically, it has been stated that psychological variables could influence wound healing, thus surgical recovery, through direct and indirect paths. Directs paths include the effects of emotions on the stress hormones which regulate healing (cortisol, adrenaline, noradrenaline) [7] – [10] , whereas indirect paths involve the psychology-influenced choice of anesthetic (choice of anesthetic may be influenced by the patient's psychological status) and general preoperative state of health (e.g. smoking, alcohol intake, obesity) [7] .
Survival and long-term recovery after surgery can be influenced by numerous variables, apart from the preoperative psychological state; consequently its actual contribution is obscured when studying these outcomes. Thus, to examine the actual effects that psychological factors may have on surgical recovery, one should examine objectively measured, early postoperative outcomes. Despite the lack of evidence, one can assume that favorable early surgical recovery may be also associated with a shorter hospitalization and possibly long-term recovery. In this context, we sought to evaluate the existing evidence regarding the impact of psychological and psychosocial factors or interventions on the early postoperative outcomes of surgical patients.
Methods
Data sources
A systematic search of the literature regarding the association of psychological or psychosocial variables with the outcome of surgery was performed on PubMed, Scopus and PsycINFO databases (January 1970 to January 2010). The primary search was conducted with the following pattern: (psychological OR psychosocial OR personality) AND (surgery OR surgical OR postoperative*) AND (recovery OR outcome). Secondary searches used the following terms in various combinations: “psychological”, “psychosocial”, “personality”, “surgery”, “surgical”, “postoperative*”, “outcome*”, “recovery”, “prognosis”, “complication*”. We also sought to find potentially useful studies in the references of the relevant articles. We considered eligible all studies written in English, Spanish, German, French, Italian and Greek.
Study selection
Three investigators (MNM, IDG, KAP) independently searched the literature and examined relevant studies for potential inclusion in this review. Any disagreement was resolved by consensus in meetings with all investigators. To be considered eligible, a study should examine the association of preoperative psychological or psychosocial status or intervention with an early objectively measured postoperative outcome in surgical patients. We considered eligible observational as well as interventional original studies, regardless of blinding and randomization. One exclusion criterion was set regarding the populations examined: having a diagnosed mental illness. Apart from that, studies in healthy volunteers and patients requiring elective or emergent surgery were eligible for inclusion. Unpublished studies reported as abstracts in conferences were not included in this review [11] .
Data extraction
We extracted data regarding the design (prospective or retrospective), the statistical analysis (univariate or multivariate), the population characteristics (number of participants, type of surgery, controls), the quality of the study (randomization, blinding), the psychological or psychosocial variables or interventions, the factors controlled in each analysis, the early postoperative outcomes measured and the related methodology, and the results of each study. When a study used both univariate and multivariate analyses for the same outcomes, the multivariate analysis data were extracted.
Definition of psychological status and interventions
The psychological status was defined according to the definitions of each study. Most studies used the standardized psychometric scales to determine the patients' psychological status. No restriction was set in the psychological or psychosocial factors measured preoperatively. We included studies examining distress [12] , depression [13] , [14] , loneliness [13] , coping [12] , [15] , [16] , anger expression and control [17] , state and trait anger and hostility [14] , [18] , state and trait anxiety [15] , [18] , perceived stress [19] , [20] , perceived social support [14] , optimism [14] , [21] , intramarital relationships [22] , religiousness [14] , expectations about recovery [15] , health locus of control [15] , worry about the operation [15] , [16] , [19] , examinations-induced stress [23] , and demented relative-induced stress [24] . Perceived stress and demented relative-induced stress were considered equivalent to trait anxiety, whereas worry about the operation and examinations-induced stress were considered equivalent to state anxiety. “State” and “trait” are defined as the sudden subversion and the baseline of one's emotional equilibrium respectively.
Similarly, no restriction was set regarding the psychological or psychiatric intervention in the studied groups. Examined interventions included guided relaxation [17] , [25] , [26] , relationship-support or conflict visits [22] , and psychiatric consults [27] .
Examined outcomes
To be considered eligible for our review, any study should examine the effect of psychological factors on at least one objectively measured, early postoperative surgical outcome. We excluded outcomes that were reported to be measured later than a month postoperatively. Also excluded were outcomes that were assessed by patients (such as perceived recovery or quality of life) and outcomes which could be directly or indirectly affected by the preoperative psychological status of the studied population regardless of the surgery (e.g. we excluded the analgesic requirements and the ambulation, since a depressed patient is expected to require more analgesics and have a delayed ambulation due to diminished drive). Moreover, studies reporting solely non-surgical outcomes (e.g. outcomes unrelated to the site of the operation) were not included. Eligible outcomes included wound inflammation or healing (assessed by peroxide stimulation test [23] , [24] , cytokine levels in wound fluid [19] , or clinically by independent surgeons [13] , [15] , [17] , [18] , [22] , [25] ) and postoperative complications (wound infection, hematoma, postoperative fever etc) [12] , [14] , [16] , [18] , [20] , [21] , [26] , [27] . Any of these outcomes is defined as an index of recovery.
Numerous studies examined the effect of psychological factors on the hospital or ICU length of stay. The length of stay is a useful index of physical recovery; however, it is multiply determined and influenced by individual behaviors, health care usage and associated costs [14] . On those grounds, the length of stay was not considered an eligible outcome.
Results
A total of 16 studies were eligible for inclusion in the present review ( Figure 1 ). Thirteen studies were observational (comparing two groups with different psychological status) and 5 were interventional (comparing the study group, which received the psychological intervention, with a control group); two of the studies had both an observational and an interventional design (in one study, the authors both assessed the volunteers' preoperative psychological status and applied an intervention [17] ; in the other study they asked the volunteers to visit the clinic twice and performed two separate analyses [22] ). In seven studies there was some degree of blinding, 4 of the interventional studies were randomized and 6 studies did not report any blinding or randomization. The median quality of the interventional studies was 2. Most of the studies were prospective in design (exception: [21] ) and performed multivariate analyses (exceptions: [16] , [23] , [25] , [27] ).
10.1371/journal.pone.0020306.g001
Figure 1
Flow diagram for reviewed studies.
The included studies enrolled a total of 1473 individuals (1071 patients and 402 volunteers, 1347 in observational and 266 in interventional studies [two studies had both an observational and an interventional design]). Of the 1071 patients, 861 received cardiac surgery [14] , [18] , [21] , [26] , [27] , 173 general surgery [12] , [16] , [19] , [20] , [25] and 37 dental surgery [15] . Most of the procedures (1050 of 1071) were elective. Volunteers underwent standardized suction blister wounds [17] , [22] , punch biopsy wounds [23] , [24] , or oral mucosa wounds [13] . The examined outcome was assessed within the first postoperative month (8 of 16 studies did not report the time of assessment of outcomes but were included, since it was obvious that this was within the first postoperative month). Details on the measurement of each outcome are available in Table S1 .
Fifteen out of 16 studies reported a significant association between at least one index of psychological status and surgical recovery ( Table 1 ). The remarkable heterogeneity of the studies, along with the substantially different methodologies and definitions of outcomes did not allow for synthesis of the data with the methodology of meta-analysis. The only study showing no statistically significant association examined the effects of perceived life stress and cold pressor test on postoperative complications [20] . The scarce complications observed in this study did not allow for a direct effect of perceived stress to be shown, but even so, there was a significant association between reaction to cold pressor test and complications (p<0.05). Since reaction to cold pressor test is itself considered a measure of stress, this study also indirectly links stress with surgical outcomes.
10.1371/journal.pone.0020306.t001 Table 1
Impact of psychological variables on wound healing and postoperative surgical complications.
Psychological and psychosocial factors
Wound healing **
Complications **
No significant association
Positiveimpact
Negative impact
Positive impact
Negative impact
Trait anxiety ±
[19] b , [24]
[15] , [18] a , [19] b , [20] c
State anxiety ±
[19] b , [23]
[18]
a
[15] , [16] , [18] a , [19] b
Trait anger / hostility
[14] , [18] a
State anger
[18]
a
[18]
a
Anger expression (in-out)
[17]
Anger control
[17]
Coping (active - vigilant)
[12] , [16]
[15]
Perceived social support
[14]
Optimism
[21]
d
[14] , [21] d
Distress (depression, anxiety, and stress)
[12]
Depression
[12]
[14]
Loneliness
[12]
Intramarital hostility
[22]
Religiousness
[14]
Expectations about recovery (low pain expectations)
[15]
Health locus of control (external)
[15]
Relaxation intervention
[25]
e
[26]
[17] , [25] e
Support instead of conflict visit (couples)
[22]
Psychiatric interview
[27]
*In brackets are the references of the studies. An association was considered significant if p≤05. ± Demented relative-induced stress [24] and perceived life stress [19] , [20] were considered equivalent to trait anxiety. Worry about the surgery [15] , [16] , [19] and exam-induced stress [23] were considered equivalent to state anxiety.
**Significant association of psychological factors with at least one aspect of wound healing or complications.
a
State anger and anxiety were found to have no impact on clinical outcome (apart from complications) in this study.
b
State anxiety and perceived stress were associated with some, but not all, markers of wound healing.
c
Reaction to cold pressor test was significantly associated with postoperative complications.
d
Only prolonged ventilation (out of 7 complications) proved to have significant association.
e
Intervention was associated with less wound erythema, but not overall wound inflammation.
Including the above mentioned study, trait anxiety (or its equivalents, perceived stress and demented relative-induced stress) was examined in a total of 5 studies [15] , [18] , [19] , [20] , [24] , showing a negative impact on recovery in 2 of them. Specifically, one study found perceived stress to result in attenuated healing (p<0.05) [19] . In addition, stress related to caring for demented patients in women was associated with a slower healing process (p<0.05) [24] . In contrast, no association was reported in the remaining two studies (on clinical assessment of the oral wound after dental surgery [15] and on postoperative complications after cardiac surgery [18] .
State anxiety (or its equivalents, worry about the operation and examinations-induced stress) was examined in 5 studies [15] , [16] , [18] , [19] , [23] . One study reported an associated significant effect on postoperative complications after cardiac surgery (p<0.01) [18] . In two studies, increased state anxiety resulted in impaired wound healing after inguinal hernia repair (p<0.05) [19] and after a standardized wound in volunteers (p<0.001) [23] . However, the first study didn't find an association between psychological variables and clinical rating of outcome by two independent surgeons [18] . The remaining 2 studies did not report any association between state anxiety and indices of recovery; however, recovery was associated with other psychological variables in those studies (described below) [15] , [16] .
Four studies examined the effects of anger or hostility [14] , [17] , [18] , [22] . Anger expression (in versus out) was found to be unrelated to the outcomes, while anger control proved to result in faster wound healing (p<0.05) [17] . State anger had a significantly negative impact on complications (p<0.001), but no impact on clinical outcome [18] . Trait anger and hostility had no significant association with postoperative complications in two studies [14] , [18] , while intramarital hostility resulted in delayed wound healing (p<0.05) [22] . Decreased religious beliefs were found to increase complications (p<0.01) [14] , while low pain expectations and external health locus of control resulted in decreased facial swelling (p<0.05) and faster healing (p<0.05), respectively, in patients undergoing 3rd molar extraction [15] .
Coping (avoidance versus vigilance) was examined in 3 studies [12] , [15] , [16] , 2 of which demonstrated increased complications in vigilant patients (p<0.05) [12] , [16] . The third study found no association between coping and clinically assessed dental wound healing [15] . Subclinical depression, distress (defined as depression, anxiety and stress), and loneliness were studied in 3 articles [12] – [14] . Depression was found to have negative impact on healing (p<0.01) [13] , but no significant impact on complications in another study [14] . Distress was associated with increased postoperative infections (p<0.05) [12] , while loneliness had no effect on the speed of healing [13] .
Perceived social support, assessed in one study, was not found to influence postoperative complications [14] . Two studies examined the effects of optimism, which were proved to be minimal, since in one study there was no significant association with complications [14] and in the other, optimism only reduced prolonged ventilation (p<0.05), of the 7 complications measured [21] .
Psychological intervention of guided relaxation was assessed in three prospective randomized studies, which found it to have no [17] or minimal (only erythema was significantly increased, p<0.05) [25] effect on wound healing speed and wound inflammation, respectively, and a significant impact on the incidence of postoperative SVT (supraventricular tachycardia, p = 0.04), but not of severe SVT [26] . One study comparing the effect of a relationship-supporting versus a conflict-inducing psychiatric consult in couples observed slower healing after the conflict visit (p = 0.01) [22] . Another interventional study reported that one standard structured psychiatric interview before elective CABG (controls: no interview) in 33 patients resulted in less postoperative complications (p<0.05) [27] .
Discussion
Although the studies we reviewed were considerably heterogeneous, our findings suggest an association of preoperative psychological variables with early surgical recovery. Almost all studies found a psychological variable to have a statistically significant effect on one of the examined outcomes ( Table 2 ). However, most studies also stated psychological factors that did not influence recovery from surgery significantly. Furthermore, psychological variables that were found to be significantly associated with recovery in some studies were found to be unrelated in other studies. This may be attributed to the limited number of participants in each study, which does not allow for all associations to be demonstrated.
10.1371/journal.pone.0020306.t002 Table 2
Overview of psychological factors and interventions examined in the reviewed studies.
Factors associated with favourable recovery
Anger control [17] , low pain expectations [15] , external locus of control [15] , optimism * [21] , religiousness [14] , relaxation intervention * [25] , [26] , social support visit intervention [22] , psychiatric interview [27]
Factors associated with impaired recovery
Trait anxiety * [19] , [24] , state anxiety * [18] , [19] , [23] , depression * [12] , intramarital hostility [22] , state anger * [18] , vigilant coping * [12] , [16] , distress [12]
Factors not associated with recovery
Trait anxiety * [15] , [18] – [20] , state anxiety * [15] , [16] , [18] , [19] , trait anger/hostility [14] , [18] , state anger * [18] , anger expression [17] , vigilant coping [15] , perceived social support [14] , optimism * [14] , [21] , depression * [14] , loneliness [12] , relaxation intervention * [17] , [25]
In brackets are the references of the studies. An association was considered significant if p≤0.05.
*These factors were associated with some of the outcomes of the studies (but not all).
The effect of psychology on surgical recovery has been addressed in the past. Several studies have reviewed this association in various patient populations (cardiac surgery, dental surgery and others) and most of them acknowledge its influence [28] – [37] . A meta-analysis suggested that psychological preparation for surgery improves postoperative outcomes [28] , while another review concurred that psychosocial factors are predictive of surgical outcomes [29] . Others have examined the cost-effectiveness of psychological interventions, which was confirmed through lower complication rate and decreased hospital and ICU length of stay [31] . However, our review is the first to examine solely objectively measured, early postoperative outcomes.
Apart from the outcomes examined in this study, numerous trials have investigated the impact of psychological factors and interventions on the hospital length of stay. Interventions such as randomizing surgical patients to a room with view of either a natural setting or a brick wall [38] , or exposing the patients to therapeutic suggestions during general anaesthesia [39] proved to reduce hospital stay. However, the length of stay is a multiply determined measure, influenced by various factors irrelevant to our analysis; thus, such studies were not included.
This study should be interpreted in view of certain limitations. Initially, the vast variety of investigated psychological variables did not allow for a full search of the literature to be performed; however, we additionally searched the references of all relevant articles to minimize the chance of potentially missed studies. Moreover, the studies included suffer from heterogeneity in the populations, the factors controlled, the measuring of outcomes and the statistical analyses, a fact that prevents us from applying the findings of each to the general population. In particular, the differences between the studies' populations may account for part of the observed effects. For example, since volunteers undergoing standardized incisions or punch biopsy wounds are presumed to suffer less operation-related stress than patients undergoing major surgery, the effects of stress in the former group are considerably milder, and the association of psychological variables with surgical outcome is expected to be attenuated. In addition, the findings for many of the psychological variables are inconsistent across studies. Moreover, the psychometric tests used in some of the studies, despite their acknowledged credibility [40] , [41] , cannot substitute the gold standard of psychological evaluation, which is the clinical interview, and might pose risks if applied universally.
Our findings could bear important implications for clinical practice. The fact that groups exposed to psychological interventions had faster recovery or fewer complications, when compared to groups receiving standard care suggests that current standard care is far from optimal. Furthermore, since psychology has been shown to affect surgical outcomes, surgeons might consider arranging elective procedures in a time when the patient is in a better psychological state, or arrange for a psychological intervention to boost the patient's psychological status. In addition, in some cases, psychological variables might aid in the prediction of the surgical outcome.
In conclusion, although the significant heterogeneity of the available studies obscures the accuracy of our findings, an association between certain psychological variables and surgical recovery is suggested. Trait and state anxiety, state anger, active coping, subclinical depression, and intramarital hostility seem to complicate recovery, while dispositional optimism, religiousness, anger control, low pain expectations, and external locus of control can promote healing. Psychological interventions also appear to improve recovery. Loneliness and perceived social support, as well as anger expression, and trait anger didn't correlate significantly with the postoperative outcomes. Large randomized controlled trials and further analyses are needed to conclusively determine the potential contribution of psychological preparation of surgical patients to their postoperative recovery.
Supporting Information
Table S1
Characteristics of the reviewed studies.
(DOC)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Obesity is a public health problem that is increasing worldwide [ 1 ]. Recently published data from the ANIBES (Anthropometry, Intake, and Energy Balance in Spain) Study show that the prevalence of overweight (OW) and obesity (OB) in the Spanish population is 35.8 and 19.9%, respectively [ 2 ]. The figures regarding abdominal obesity (AO) are even more disturbing, affecting 58.4% of the population when waist to height ratio (WHtR) ≥0.5 is considered [ 2 ].
Obesity is the result of an imbalance between energy intake and expenditure where genetic, physiological and environmental factors are also involved [ 3 ], Apart from sex and age, which has shown to be related to obesity [ 2 ], other potential confounders of this association in adults include the socioeconomic status (SES) [ 4 , 5 ], physical inactivity [ 6 ], sleeping time [ 7 , 8 ] and smoking habits [ 9 ].
Because of the economic downturn that has affected Spain since 2008, unemployment has increased and household economic resources declined. It has been suggested that economic problems can affect not only the diet but also physical activity (PA), smoking habits [ 10 ] and sleep [ 11 ]. There is also some evidence of a relationship between the economic crisis in Spain and the health and lifestyle of the Spanish population [ 12 ], and this probably also had an impact on the factors associated with overweight and obesity.
The ANIBES Study (Anthropometry, Intake, and Energy Balance in Spain) was conducted from mid-September to mid-November 2013 in a representative sample of the Spanish population. The aim of this paper was to analyze the relationships between different socioeconomic and lifestyle factors, including PA, with the conditions of OW, OB and AO.
Materials and Methods
This study is part of the ANIBES study. Briefly, ANIBES Study was designed to carry out an accurate updating of food and beverage intake, dietary habits/behavior and anthropometric data of the Spanish population (9–75 years, n = 2009), as well as the energy expenditure and physical activity patterns. The overall design, protocol, and methodology of the ANIBES study have previously been reported in detail [ 13 ].
Study design and sampling procedure
The target population consisted of all inhabitants living in Spain (excluding the autonomous cities of Melilla and Ceuta in the north of Africa), aged 9 to 75 years and living in municipalities of at least 2,000 inhabitants. The sample for the ANIBES Study was based on the 2012 census data published by the INE (Instituto Nacional de Estadística/Spanish Bureau of Statistics) for gender, age, habitat size and region. The total sample size was calculated based on a 0.05 probability of Type I error (rejecting a null hypothesis when it is true) and 0.1 probability of Type II error (accepting a null hypothesis when it is wrong) in the main outcome of the study (energy intake). No previous pre-recruitment was considered to minimize the risk of bias in responses. The present paper focused on the adult population (18–64 years, n = 1655).
For the population sampling, the following variables were taken into account: age, sex, geographical distribution (Northeast, East, South, West, North-Central, Barcelona, Madrid, and the Balearic and Canary Islands), habitant size (rural, semi-urban, or urban populations of 2,000–30,000, 30,000–200,000, and over 200,000 inhabitants, respectively), and other factors such as rate of unemployment, rate of immigrant population, level of PA, and educational level. Geographical distributions were grouped into four different regions (South, Central, Atlantic and Mediterranean). Exclusion criteria included the following: living in an institutional setting (e.g. college, nursing home, hospital), following a therapeutic diet because of recent surgery or any medical prescription, potential participants with any transitory illness (e.g. flu, gastroenteritis, chicken pox) at the time fieldwork was undertaken, and individuals employed in areas related to consumer science, marketing or the media. However, individuals with the following conditions were considered eligible to be included: those following dietary advice such as for prevention of hypertension, diabetes, hypercholesterolemia or hyperuricemia, pregnant and lactating women, those with diagnosed allergy and/or food intolerance, and those suffering a metabolic disease such as hyperthyroidism or hypothyroidism.
The fieldwork for the ANIBES Study was three months from mid-September to mid-November 2013, and two previous pilot studies were from June to September, 2013. The final protocol was approved by the Ethical Committee for Clinical Research of the Region of Madrid (Spain). Written informed consent was obtained from all subjects. All data were collected by trained interviewers.
Anthropometric data
Weight, height, and waist circumference were measured using standardized procedures by well-trained interviewers to minimize the inter-observer coefficients of variation [ 14 ]. Weight was measured once with a Seca model 804 weighing scale (Medizinische Messsysteme und Waagen seit 1840, Hamburg, Germany; range 0.1–150 kg, precision 100 g). Height was assessed in triplicate using a Seca model 206 Stadiometer (range 70–205 cm, precision 1 mm). Waist circumference was measured in triplicate using a Seca 201 tape measure (Seca, Hamburg, Germany; range 0–150 cm, precision 1 mm).
General adiposity was assessed using body mass index (BMI) and AO by waist to height ratio (WHtR). BMI was calculated as weight to height squared (kg/m 2 ). Overweight and obesity were defined as BMI 25.0–29.9 kg/m 2 and ≥30 kg/m 2 , respectively [ 15 ]. WHtR was calculated as waist (cm)/height (cm). Abdominal obesity was defined as a WHtR ≥0.5 [ 16 ].
Socioeconomic factors
Each participant answered a questionnaire administered face-to-face that included the following questions: age in completed years (18–40/41–65 years), place of birth (no immigrant/immigrant) educational level according to years and type of education (primary or less/secondary/university), occupational status (unemployed/employed), and monthly family income (0–1000€/1001–2000€/>2000€/no answer).
Lifestyle factors
Physical activity was estimated based on the International Physical Activity questionnaire (IPAQ) [ 17 ]. Time spent in vigorous-intensity PA (VPA) was calculated and grouped into (a) <75 min/week, (b) 75–150 min/week, (c) 151–300 min/week, or (d) >300 min /week. Also, moderate vigorous-intensity PA (MVPA) was calculated and grouped into (a) <150 min/week, (b) 151–300 min/week, or (c) >300 min/week based on public health guidelines [ 18 ]. Smoking habits were grouped as smoker or non-smoker. Participants also described their frequency of viewing television as (a) never or almost never, (b) low frequency, (c) frequently, (d) quite often, or (e) very often, and reported their daily hours of sleeping as (a) <7, (b)≥7 and <8, or (c) ≥8 h).
Statistical analyses
Analyses were performed using SPSS version 22.0 (SPSS, Inc, Chicago, IL, USA). Data are presented as means, standard deviation, and percentages. The Kolmogorov-Smirnoff test was used to test if the variables followed a normal distribution and to decide between parametric or non-parametric analysis. Differences between sexes were performed using Student’s-t or Mann–Whitney test. When comparing proportions, the z -test was used. The associations of sex, age and socioeconomic and lifestyle factors (independent variables) and OW, OB or AO (dependent variables) were analyzed by logistic regression analysis to calculate odds ratios (OR). The 95% confidence intervals (CI) were calculated, and Wald’s test was used for comparison of the OR. Multivariate analyses were also used to examine the simultaneous effect of the different socioeconomic and lifestyle variables on the prevalence of OW, OB and AO. The level of significance was set at p <0.05.
Results
Table 1 shows personal, anthropometric, sociodemographic and lifestyle factors of the total sample and by sex. Some of these results have been described in previous publications [ 2 , 19 ] and are presented again for simple characterization. The final sample consisted of 1655 individuals (48.2% men). Over a third (35.8%) were OW and 19.9% were OB, with higher percentages in males. Also, 58.4% of the population had AO, and again the percentage of people with central obesity is significantly higher in men.
10.1371/journal.pone.0169027.t001
Table 1 Characteristics of the study population.
Total
Men
Women
N (%)
1655
798 (48.2)
857 (51.8)
Age (years) (Mean ± SD)
39.97±12.20
39.6±12.2
40.3±12.2
18–40 y (n,%)
883 (53.4)
435 (26.3)
448 (27.1)
41–65 y (n,%)
772 (46.6)
363 (21.9)
409 (24.7)
Weight (kg) (Mean ± SD)
74.2±16.48
82.4±15.34
66.6±13.62 *
Height (cm) (Mean ± SD)
167.7±9.35
174.5±6.95
161.3±6.37 *
BMI (kg/m 2 ) (Mean ± SD)
26.3±5.15
27.1±4.87
25.6±5.3 *
Overweight (n,%)
592 (35.8)
323 (40.5)
269 (31.4) *
Obesity (n,%)
329 (19.9)
181 (22.7)
148 (17.3) *
Waist circumference (cm) (Mean ± SD)
88.1±14.5
93.8±13.61
82.7±13.19 *
WHtR (Mean ± SD)
0.53±0.08
0.54±0.08
0.51±0.09 *
Abdominal obesity a (%)
966 (58.4)
516 (64.7)
450 (52.5) *
Region (n,%)
South
425 (25.7)
198 (24.8)
227 (26.5)
Central
379 (22.9)
197 (24.7)
182 (21.2)
Atlantic
281 (17.0)
137 (17.2)
144 (16.8)
Mediterranean
570 (34.4)
266 (33.3)
304 (35.5)
Habitat size b (n,%)
Rural
564 (34.1)
266 (33.3)
298 (34.8)
Semi-urban
561 (33.9)
284 (35.6)
277 (32.3)
Urban
530 (32.0)
248 (31.0)
282 (32.9)
Level of education (n,%)
Primary or less
443 (26.8)
212 (26.5)
231 (26.9)
Secondary
810 (48.9)
396 (49.6)
414 (48.3)
University
402 (24.3)
190 (23.8)
212 (24.7)
Family income (n,%)
0–1000 €
315 (19.0)
162 (20.3)
153 (17.9)
1000–2000 €
647 (39.1)
290 (36.3)
357 (41.7)
≥2000 €
303 (18.3)
151 (18.9)
152 (17.7)
No answer (%)
390 (23.6)
195 (24.4)
195 (22.8)
Immigrant population (n, %)
65 (3.9)
58 (7.27)
48 (5.6)
Rate of unemployment (n, %)
272 (16.4)
228 (28.6)
118 (13.8)
Smoking status (n,%)
Non smoker
1076 (65.0)
480 (60.1)
596 (69.5)
Smoker
579 (35.0)
318 (39.8)
261 (30.4)
Vigorous physical activity (min/wk) (Mean ± SD)
149.2±264.15
209±302
94±209 *
<75 min/wk (n,%)
1041 (62.9)
420 (52.6)
621 (72.5)
75–149 min/wk (n,%)
118 (7.1)
49 (6.1)
69 (8.1)
150–299 min/wk (n,%)
185 (11.2)
107 (13.4)
78 (9.1)
≥300 min/wk (n,%)
311 (18.8)
222 (27.8)
89 (10.4)
Moderate-vigorous physical activity (min/wk) (Mean ± SD)
565±509
524±513
603±503 *
<150 min/wk (n,%)
415 (25.1)
230 (28.9)
185 (21.6)
150–300 min/wk (n,%)
224 (13.6)
114 (14.3)
110 (12.9)
≥300 min/wk (n,%)
1013 (61.3)
452 (56.8)
562 (65.5)
Time watching TV
Never or almost never (n,%)
56 (3.4)
30 (3.7)
26 (3.0)
Low frequency (n,%)
204 (12.3)
88 (11.02)
116 (13.5)
Frequently (n,%)
311 (18.8)
140 (17.5)
171 (19.9)
Quite often (n,%)
651 (39.4)
318 (39.8)
333 (38.8)
Very often (n,%)
432 (26.1)
222 (27.8)
210 (24.5)
Sleep time (h/day) (Mean ± SD)
7.46±1.13
7.46±1.10
7.46±1.16
<7 h/day (n,%)
318 (20.6)
147 (19.8)
171 (21.3)
7–8 h/day (n,%)
506 (32.7)
259 (34.8)
247 (30.8)
≥8 h/day (n,%)
722 (46.7)
338 (45.4)
384 (47.9)
BMI: Body mass index, WHtR: Waist to height ratio
* Significant differences regarding sex
a Abdominal obesity defined by WHtR ≥0.5
b Habitat size: rural populations: 2,000–30,000; semi-urban: 30,000–200,000; urban population: over 200,000 inhabitants.
Most participants had a high school diploma (48.9%), an average monthly income between 1000 and 2000 € (39.1%), and 16% were unemployed. Approximately a third were smokers. Women spent more time on MVPA, and 1 out of 4 people spent less than 150 min/week on physical activities. By contrast, men spent more time weekly on VPA, and 62.9% of the population spent less than 75 min/week on them. The average sleeping time was 7.46 ± 1.13 h/day, and 46.7% sleep more than 8 h/day. Sixty five percent of the population indicated watching TV quite or very often.
The univariate analyses with sociodemographic and lifestyle variables and the risk of OW, OB and AO are shown in Table 2 .
10.1371/journal.pone.0169027.t002
Table 2 Association of sociodemographic and lifestyle factors with prevalence of overweight, general and abdominal obesity in Spanish adults. Logistic regression analysis.
Overweight
Obesity
Abdominal obesity
25≥BMI<30 kg/m 2
BMI≥30 kg/m 2
WHtR≥0.5
OR (95% CI)
p
OR (95% CI)
p
OR (95% CI)
p
Sex
Female
1
1
1
Male
1.80 (1.44–2.24)
0.000
1.83 (1.41–2.38)
0.000
1.65 (1.36–2.02)
0.000
Age
18–40 y
1
1
1
41–65 y
2.30 (1.84–2.87)
0.000
3.89 (2.95–5.11)
0.000
4.32 (3.49–5.34)
0.000
Region
South
1
1
1
Central
0.90 (0.66–1.22)
0.485
1.06 (0.72–1.55)
0.766
1.01 (0.76–1.34)
0.944
Atlantic
0.87 (0.62–1.22)
0.419
0.87 (0.57–1.33)
0.511
0.75 (0.55–1.02)
0.067
Mediterranean
0.87 (0.66–1.16)
0.343
1.30 (0.93–1.83)
0.130
0.86 (0.67–1.12)
0.266
Habitats size a
Rural
1
1
1
Semi-urban
0.94 (0.73–1.22)
0.659
1.38 (1.00–1.90)
0.051
0.91 (0.72–1.15)
0.423
City/town
0.75 (0.58–0.98)
0.037
1.13 (0.81–1.56)
0.466
0.87 (0.68–1.10)
0.241
Level of education
Primary or less
1
1
1
Secondary
0.60 (0.46–0.79)
0.000
0.43 (0.32–0.59)
0.000
0.42 (0.33–0.54)
0.000
University
0.43 (0.32–0.59)
0.000
0.29 (0.20–0.43)
0.000
0.33 (0.25–0.45)
0.000
Family income
0–1000 €
1
1
1
1001–2000 €
0.78 (0.57–1.06)
0.111
0.70 (0.49–0.99)
0.046
0.66 (0.49–0.88)
0.004
≥2000 €
0.78 (0.55–1.11)
0.169
0.43 (0.27–0.67)
0.000
0.50 (0.36–0.69)
0.000
No answer (%)
0.69 (0.49–0.97)
0.034
0.67 (0.46–0.99)
0.047
0.48 (0.35–0.65)
0.000
Immigrant population
No
1
1
1
Yes
0.96 (0.62–1.50)
0.858
0.98 (0.57–1.66)
0.932
1.09 (0.73–1.63)
0.667
Rate of unemployment
No
1
1
1
Yes
1.20 (0.92–1.57)
0.185
1.48 (1.09–2.02)
0.013
1.59 (1.24–2.05)
0.000
Smoking status
Non smoker
1
1
1
Smoker
0.99 (0.79–1.24)
0.943
0.84 (0.64–1.11)
0.215
1.01 (0.82–1.24)
0.913
Vigorous physical activity
<75 min/wk
1
1
1
75–149 min/wk
0.66 (0.43–1.00)
0.050
0.29 (0.15–0.55)
0.000
0.54 (0.37–0.79)
0.002
150–299 min/wk
0.73 (0.52–1.04)
0.081
0.42 (0.26–0.67)
0.000
0.59 (0.43–0.80)
0.001
≥300 min/wk
0.90 (0.68–1.18)
0.433
0.37 (0.25–0.55)
0.000
0.59 (0.46–0.76)
0.000
Moderate-vigorous physical activity
<150 min/wk
1
1
1
150–300 min/wk
0.69 (0.48–1.01)
0.058
0.64 (0.42–0.97)
0.036
0.63 (0.45–0.87)
0.006
≥300 min/wk
0.94 (0.73–1.23)
0.670
0.56 (0.42–0.76)
0.000
0.74 (0.58–0.93)
0.011
Time watching TV
Never or almost never
1
1
1
Low frequency
0.90 (0.47–1.69)
0.734
1.12 (0.39–3.22)
0.830
1.12 (0.62–2.04)
0.700
Frequently
1.14 (0.62–2.10)
0.677
2.01 (0.74–5.46)
0.170
1.50 (0.84–2.65)
0.168
Quite often
1.32 (0.73–2.38)
0.352
3.29 (1.25–8.65)
0.016
1.93 (1.11–3.34)
0.020
Very often
1.52 (0.83–2.78)
0.173
4.15 (1.56–10.99)
0.004
2.22 (1.26–3.89)
0.006
Sleep time
<7 h/day
1
1
1
7–8 h/day
0.81 (0.59–1.12)
0.202
0.56 (0.38–0.80)
0.002
0.53 (0.39–0.72)
0.000
≥8 h/day
0.69 (0.51–0.93)
0.016
0.49 (0.35–0.69)
0.000
0.45 (0.34–0.60)
0.000
BMI: Body mass index, WHtR: Waist to height ratio; OR: crude odds ratio. The reference category for overweight and obesity is BMI <25 kg/m 2 , and for abdominal obesity is WHtR <0.5.
a Habitat size: rural populations: 2,000–30,000; semi-urban: 30,000–200,000; urban population: over 200,000 inhabitants.
Being male and aged more than 41 years was significantly associated with an increased risk of being OW, OB or AO, while having high school or university education was associated with a lower risk in all cases. It is noteworthy that not answering the question of income was associated with a lower risk of any excess body weight. Regarding OW, the only other variables associated with a lower risk were living in a city and sleeping more than 8 h/day. On the other hand, regarding both OB and AO, general and central obesity, a higher family income, more time on VPA or MVPA, and sleeping more than 7 h/day were associated with a lower risk, while being unemployed, and watching TV quite or very often were associated with increased risk.
Multivariate analyses of all the variables studied are shown in Table 3 .
10.1371/journal.pone.0169027.t003
Table 3 Association of socio-demographic and lifestyle factors on prevalence of overweight, general and abdominal obesity in Spanish adults. Multivariate regression analysis.
Overweight
Obesity
Abdominal obesity
25≥BMI<30 kg/m 2
BMI≥30 kg/m 2
WHtR≥0,5
AOR (95% CI)
p
AOR (95% CI)
p
AOR (95% CI)
p
Sex
Women
1
1
1
Men
2.12 (1.62–2.78)
0.000
2.38 (1.70–3.35)
0.000
2.24 (1.73–2.91)
0.000
Age
18–40 y
1
1
1
41–65 y
2.21 (1.72–2.84)
0.000
3.12 (2.27–4.28)
0.000
4.11 (3.23–5.23)
0.000
Region
South
1
1
1
Central
0.80 (0.56–1.13)
0.203
0.81 (0.51–1.30)
0.388
0.89 (0.64–1.26)
0.516
Atlantic
0.78 (0.53–1.13)
0.186
0.76 (0.46–1.28)
0.305
0.64 (0.45–0.92)
0.017
Mediterranean
0.89 (0.65–1.23)
0.490
1.19 (0.78–1.80)
0.424
0.79 (0.58–1.07)
0.130
Habitat size a
Rural
1
1
1
Semi-urban
0.92 (0.69–1.24)
0.600
1.25 (0.85–1.83)
0.260
0.80 (0.61–1.06)
0.124
City/town
0.78 (0.57–1.05)
0.103
1.14 (0.77–1.71)
0.509
0.81 (0.60–1.08)
0.154
Level of education
Primary or less
1
1
1
Secondary
0.74 (0.54–1.01)
0.055
0.56 (0.38–0.81)
0.002
0.59 (0.44–0.80)
0.001
University
0.59 (0.41–0.85)
0.005
0.41 (0.25–0.65)
0.000
0.55 (0.39–0.78)
0.001
Family income
0–1000 €
1
1
1
1001–2000 €
1.00 (0.70–1.43)
0.998
1.11 (0.71–1.75)
0.643
0.88 (0.62–1.25)
0.467
≥2000 €
1.05 (0.69–1.59)
0.833
0.72 (0.41–1.26)
0.250
0.70 (0.47–1.06)
0.091
No answer
0.86 (0.58–1.29)
0.471
0.98 (0.60–1.62)
0.947
0.61 (0.41–0.90)
0.012
Immigrant population
No
1
1
1
Yes
0.92 (0.57–1.49)
0.740
0.74 (0.40–1.37)
0.339
1.00 (0.64–1.58)
0.989
Rate of unemployment
No
1
1
1
Yes
0.78 (0.56–1.08)
0.140
1.05 (0.71–1.56)
0.798
1.03 (0.76–1.41)
0.836
Smoking status
Non smoker
1
1
1
Smoker
0.81 (0.63–1.05)
0.112
0.60 (0.42–0.84)
0.003
0.72 (0.56–0.92)
0.009
Vigorous physical activity
<75 min/wk
1
1
1
75–149 min/wk
0.72 (0.44–1.16)
0. 178
0.43 (0.21–0.87)
0.019
0.73 (0.46–1.17)
0.190
150–299 min/wk
0.64 (0.42–0.96)
0.032
0.39 (0.22–0.69)
0.001
0.55 (0.37–0.81)
0.003
≥300 min/wk
0.66 (0.46–0.96)
0.029
0.31 (0.18–0.54)
0.000
0.48 (0.34–0.70)
0.000
Moderate-vigorous physical activity
<150 min/wk
1
1
1
150–300 min/wk
0.79 (0.52–1.21)
0.279
1.16 (0.70–1.92)
0.564
0.88 (0.60–1.31)
0.540
≥300 min/wk
1.24 (0.88–1.74)
0.219
1.07 (0.71–1.61)
0.735
1.10 (0.80–1.52)
0.561
Time watching TV
Never or almost never
1
1
1
Low frequency
1.16 (0.58–2.32)
0.681
1.51 (0.48–4.77)
0.481
1.83 (0.92–3.64)
0.084
Frequently
1.28 (0.66–2.49)
0.468
2.09 (0.71–6.17)
0.181
2.15 (1.12–4.15)
0.022
Quite often
1.53 (0.80–2.91)
0.197
3.33 (1.17–9.45)
0.024
2.76 (1.46–5.21)
0.002
Very often
1.74 (0.89–3.38)
0.105
4.92 (1.70–14.23)
0.003
3.22 (1.67–6.19)
0.000
Sleep time
<7 h/day
1
1
1
7–8 h/day
0.87 (0.62–1.23)
0.443
0.60 (0.39–0.93)
0.022
0.54 (0.39–0.76)
0.000
≥8 h/day
0.77 (0.56–1.08)
0.128
0.51 (0.34–0.77)
0.001
0.48 (0.34–0.66)
0.000
Abbreviations: BMI: Body mass index, WHtR: Waist to height ratio; AOR: Odds ratio adjusted for all other variables in the table. The reference category for overweight and obesity is BMI <25 kg/m 2 , and for abdominal obesity is WHtR <0.5.
a Habitat size: rural populations: 2,000–30,000; semi-urban: 30,000–200,000; urban population: over 200,000 inhabitants.
Sex and age remained statistically significant in all models. After adjusting for other variables, the only variables that were associated with a lower risk of OW were a higher educational level and spending more than 150 min/week in VPA. Educational level, time spent in VPA, sleeping time, and frequency of viewing television remained significantly associated with both OB and AO, while the family income, employed status, and MVPA were no longer significant. It is noteworthy that the no answer to the question about income level only maintains its association with AO, and that in the adjusted model living in the Atlantic region becomes a protective factor of AO. In contrast to the findings in the unadjusted model, being a smoker is associated with a lower risk of both OB and AO.
Discussion
Our results show that a higher level of education, smoking, more time spent in VPA, and sleeping more than 7 h/day are associated with a lower risk of OB and AB. While being male, older than 40 years, and watching TV quite or very often are associated with the risk factors. Aspects such as the municipal habitat size, family income, being an immigrant or unemployed, or time spent on MVPA are not associated with an increased risk of OW, OB and AO.
The associations of sex and age with OW, OB and AO in the ANIBES study have been previously described [ 2 ]. It has been suggested that sex itself (sex hormones affect the amount and distribution of body fat) is a factor influencing body composition, its oxidation and mobilization [ 20 ]. In our study, the increased risk to males may also be due to the different pattern of PA [ 19 ] or to different dietary habits of men and women. Other studies indicate that living in the Atlantic region is associated with a lower risk for AO, and have shown geographic differences in obesity prevalence in Spain [ 21 , 22 ].
Whenever possible, SES level is generally measured by occupation, educational level and income. Although they are not completely independent, analysis of these three SES dimensions together is of interest, but owing to limited availability of data, only a few surveys have included all of these together [ 23 ]. In our investigation, only educational level was inversely associated with both OB and AO. Our results agree with others and confirm that in developed countries the level of education is inversely associated with increased risk of OW and OB [ 5 ], and also with AO as defined by WHtR [ 24 ], waist circumference [ 22 , 25 , 26 ], or waist to hip ratio [ 27 ]. Educational level can exert its influence on health and body weight since it is related to knowledge about health and healthy lifestyles [ 27 ], including dietary habits and PA [ 25 ]. Furthermore, educational level is assumed to be stable throughout life and to partly reflect childhood socioeconomic conditions [ 25 ].
It is noteworthy that 23.6% did not answer the question of family income and this was associated with a lower risk of AO. Bearing in mind that the questionnaire was administered face to face and some people may really have not known their income, it is also possible that those with higher incomes did not want to declare them. This group also included those with the highest purchasing power.
Smoking was associated with lower prevalence of both OB and AO. This relationship has been confirmed in numerous studies that have shown that smokers have less weight or BMI than nonsmokers [ 9 ], although there are also studies that observed an inverse association [ 28 ]. However, the results of studies that have examined the relationship between smoking and AO are controversial, since some found no association [ 24 ], while others found lower [ 29 ] or higher risk in smokers [ 9 , 27 ]. Smoking could possibly be associated with lower weight and adiposity because nicotine acutely increases the levels of various neurotransmitters, suppresses appetite and consequently reduces food intake [ 30 ]. This process likely explains why smokers tend to decrease body weight, and why smoking cessation is frequently followed by weight gain. As a result, one of the barriers to quitting is the fear of gaining weight [ 31 ].
Physical activity is a key determinant of energy expenditure. Adults over 18 years should perform at least 150 min/week of moderate-intensity aerobic PA or at least 75 min/week of vigorous-intensity aerobic PA, or an equivalent combination of MVPA [ 18 ]. Recently, it has been shown that in Spain, 27% of the adult population did not meet international recommendations regarding PA, and 20.1% of adults never performed MVPA [ 19 ]. Our data suggest that VPA may have a greater effect on preventing obesity than PA of lower intensity, since MVPA was not associated with the prevalence of OB. In addition, more than 75 min/week of VPA was associated with a lower risk of OB, while more than 150 min/week (vs. less than 75 min/week) was associated with lower risk of OW or AO. It is possible that some people in our study who performed between 75 and 150 min/week of VPA were classified as OW, although actually they did not have excess body fat. Therefore, BMI alone is not an indicator of body composition and that 15.0% of our population who were OW had a WHtR <0.5 [ 2 ]. Regarding AO, our results suggest that performing less than 150 min/week may not be enough to prevent central adiposity, agreeing with other authors who indicate that it is necessary to devote between 150 and 250 min/week of moderate intensity PA to prevent weight gain effectively [ 32 ].
The sedentary behavior of watching television is the most commonly reported daily activity during leisure time [ 33 ]. In our study, the higher self-perceived frequency of watching television was associated with a higher risk of OB and AO. The association is higher for AO, since watching television "with some frequency" and more is a significant risk of OA, while the risk of OB was observed with watching television "quite frequently" and more ( Table 3 ). These results agree with those observed in other studies that have described the increased risk of OW and OB [ 34 , 35 ] or AO [ 35 , 36 ] in adults with increasing time watching television. Watching television may contribute to obesity via promotion of sedentary behavior and exposure to food-related commercials and other programs that encourage more eating [ 35 , 37 ].
In our study, frequency of watching television is a risk factor for OB and AO, independent of VPA and MVPA, as suggested by other studies in adults [ 34 , 36 ]. Mansoubi et al. [ 38 ] suggested that sedentary behaviors, like watching television, may displace time spent in light intensity activities that involve standing and light ambulation, but not in MVPA, which is likely to be more structured. Furthermore, it is worth mentioning that out of the different types of sedentary behavior (computer use, reading, listening to radio/music and other type of relaxation…), TV viewing seem to be the most consistently associated with adiposity markers in adults despite the co-existence of several other confounding factors (diet, smoking, PA, SES or genetic predisposition) [ 35 ]; Therefore, according with our results, it seems advisable to increase levels of total PA, which include light intensity PA, and decrease sedentary behavior including screen time to prevent AO and OB.
Sleep is also an important lifestyle factor that influences health. In our study, sleeping more than 7 h/day was associated with a lower risk of OB and AO, and the risk was even lower sleeping more than 8 h/day. These data agree with other studies that found an association between shorter sleeping time and the risk of OB [ 7 ] and AO [ 8 ]. However, results of other studies that have considered other indicators of AO are controversial [ 26 ]; the association between sleep and obesity may be more linear in young adults and weaker in older adults [ 39 ]. Some of the proposed mechanisms to explain the relationship between sleep and obesity suggest that lower levels of leptin and elevated ghrelin associated with shorter sleep [ 40 ] can stimulate appetite and cause weight gain [ 41 ].
This paper presents results of objective measures of weight, height and waist circumference along with indicators of socioeconomic status and lifestyle. The sample included 1655 individuals aged 18 to 65 years, and it is representative of the Spanish general population. Our data were collected in late 2013, providing new epidemiological data after 5 years of the economic downturn that began in 2008. Data collectors were trained rigorously to ensure the quality of data collection. Height, weight and waist were measured, not self-reported, which is a more accurate assessment procedure that strengthens the data. One of the strengths of our study is using WHtR to assess AO. Other studies in Spain have used WC to assess abdominal adiposity [ 22 ], but there is evidence that the WHtR may be a more accurate diagnostic tool for obesity-related chronic diseases than BMI or waist circumference because it more accurately characterizes adiposity [ 16 ]. There are few studies in Spain focused on the influence of SES level, PA or sleep on AO that include WHtR as an indicator of AO [ 24 ].
Our study has some limitations. This was a cross-sectional study. Therefore, the associations reported cannot be identified and/or can only be interpreted as hypothetical causal relations. Associations between SES level, PA or sleep and body weight may be due to influences of obesity leading to a lower SES (less access to educational benefits, employment, social relations, etc.), lower PA level, or sleep problems. No exact information about different sedentary behaviours was asked because when the study was designed there was not a validated questionnaire of sedentary habits in adults with good reliability. Furthermore, time expend watching TV was described very generally and the exact time spend on this activity was not recorded. This study has not analyzed the association of diet, which is the subject of another investigation. The questionnaire, with self-reported questions, might have introduced some recall bias. The PA assessed by the IPAQ might be overestimated, although it is a validated questionnaire [ 17 ].
Our results show that in Spain, being male, older than 40 years, and watching TV quite or very often are associated with higher risk of OB and AO, while a higher level of education, being a smoker, spending more time in VPA, and sleeping more than 7 h/day are associated with a lower risk. Strategies for preventing and reducing OB and AO should consider improving sleeping habits and PA (increasing time in VPA and decreasing time in very sedentary activities). In addition, the strategies should also pay more attention to the most vulnerable groups such as those less educated.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
There is a numerical error that appears in the Abstract, Table 1 , and the Results. The mean difference (95% confidence interval (CI)) of physical function and physical role in quality of life (QOL) incorrectly appears as 21.10 (95% CI: 6.57–35.63) and 44.40 (95% CI: 22.55–66.05), respectively. The correct mean difference (95% confidence interval (CI)) of physical function and physical role in quality of life (QOL) were 21.80 (3.18–40.42) and 44.30 (14.15–74.45). Please see the correct Table 1 below.
10.1371/journal.pone.0221224.t001
Table 1 Summary of Findings.
Rehabilitation compared with usual care in adult patients with sepsis
Patients or study population: adult patients with sepsis Setting: any Intervention: protocolized rehabilitation designed to either commence earlier, and/or more intensive than the care received by the control group Comparison: usual care
Outcome
Illustrative comparative risks* (95% CI)
Relative effect (95% CI)
No. of participants (studies)
Certainty of the evidence (GRADE)
Comments
Risk usual care
Risk rehabilitation
Quality of life SF-36 (at 6 months)
Mean difference [95% CI] of physical function and physical role were 21.80 [3.18–40.42] and 44.30 [14.15–74.45] respectively. These mean differences were significantly higher for those who received intervention.
-
30 (1 RCT)
⊕⊝⊝⊝ Very low a b c
ICU mortality
Study population
RR 2.02 (0.46 to 8.91)
75 (2 RCT)
⊕⊝⊝⊝ Very low b c
65 per 1,000
130 per 1,000 (30 to 575)
ICU length of stay
Median (interquartile range) of ICU length of stay was not statistically significantly different in both studies. Intervention vs. comparison: 12 (4–45) vs. 8.5 (3–36) days
-
50 (1 RCT)
⊕⊝⊝⊝ Very low a b c
Hospital length of stay
Hospital length of stay was not statistically significantly different in both studies. Intervention vs. comparison: 41 (9–158) vs. 45 (14–308) days and 30 (18–45) vs. 36 (26–78) days
-
75 (2 RCT)
⊕⊝⊝⊝ Very low a b c
Muscle strength MRC sum-score (at ICU discharge)
Mean difference [95% CI] of MRC sum-score was 4.6 [-2.69–11.89]. The mean difference was higher for those who received intervention.
-
42 (1 RCT)
⊕⊝⊝⊝ Very low a b c
Adverse events
Two studies reported no adverse events.
-
75 (2 RCT)
⊕⊝⊝⊝ Very low a b c
*The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; RR: risk ratio
GRADE Working Group grades of evidence High certainty : We are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty : We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty : Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect. Very low certainty : We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of the effect.
a Participants and personnel were not blinded.
b Number of participants was small.
c There were four ongoing studies.
Additionally, there are citation errors in the Methods section. The fourth sentence of the third paragraph should have cited reference 25 instead of 26. The third sentence of the fifth paragraph should have cited reference 20 instead of 18.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
A Brief Conceptual History of Modifier Genes
The concept of modifier genes originated with the solution to an early genetic mystery. As early workers tested Mendel's segregation ratios with observations on a wide range of traits and species, several cases of “inconstant inheritance” caused some to question the Mendelian basis of factors underlying these traits [1] , [2] . Three cases of particular note are Beaded and truncate wings in fruit flies and pigmentation in hooded rats. In each of these cases, the phenotype varied widely among offspring of parents with established genotypes. Mutant frequencies did not follow Mendelian ratios and—more troubling still—varied among derived lines carrying the same mutation. This led some workers to question whether genes were constant physical units or changed in properties during transmission [3] , [4] and was the last serious challenge to the theory of the gene and the generality of Mendelian inheritance. Resolution of this issue came with demonstrations that defined genetic backgrounds could account for the variation in truncate flies [5] and hooded rats [6] , [7] and more precisely, by linkage mapping, that discrete genetic loci account for the observed variation in Beaded flies [8] . The term “modifier gene” was coined to indicate a genetic variant identified by its impact on a conditioning mutation, with no obvious phenotype of its own. This is the usage we will follow here ( Box 1 ).
Box 1. What Is a Modifier Gene?
The term has been applied to several classes of genetic activities, some of which overlap. Genetic variations that alter the activity of a protein encoded by a second locus are one class. To the extent that the modifying effect is independent of allelic variation in the gene being modified, this usage is conceptually equivalent to any primary mutation whose phenotypic consequences include loss of interaction with its normal targets (as for a transcription factor and its target genes, or a protein kinase and its substrates). Interactions between mutations that were each previously recognized by their independent phenotypes are another class. While these genetic interactions can be revelatory, one need not invoke the term “modifier” to fully describe both individual and interaction effects, as each locus is identifiable without the other. Quantitative trait loci (QTL) are sometimes referred to as modifiers in the context of a major effect locus, which can be nearer the original meaning, depending on the structure of interactions in the QTL model. For the purposes of this review, we will use “modifier” to mean a genetic variant that is best recognized by its ability to change the phenotypic outcome of an independent “conditioning” variant at another locus. Modifier genes in this usage have no obvious phenotype of their own prior to discovery and are effectively silent—or at least quiet—with respect to the phenotype under study, in the absence of the conditioning mutation.
Early workers quickly appreciated that modifier genes are pervasive across experimental systems and organisms. Modifier genes need not be subtle and can have phenotypic effects as large as the initial conditioning mutation. These early observations contributed to the conceptualization of genetic pathways, prior to knowing the molecular identities of any components. This formal concept remains useful in analyzing genetic networks for sensitive nodes (genes) through which to manipulate mutant phenotypes in genetic disease or experimental biology.
Modifier Genes in Human Disease and Mouse Models: Ripe for Harvest
Modifier genes are also frequent in human disease and often invoked to explain divergent outcomes in genetic disorders with apparently equivalent cause. Among the best examples is cystic fibrosis (CF). CF patients homozygous for the Δ508 allele (approximately half of patients with European ancestry) present with a broad range of organ involvement and clinical severity, much of which is controlled by modifier genes [9] . Despite being among the most common Mendelian disorders and having an unusually common disease allele to minimize heterogeneity, molecularly defined modifiers have only very recently been reported [10] , [11] . Just as CF was one of the first major victories in positional cloning of disease genes, its modifier gene network is one of the first to see real progress apart from candidate genes and serendipitous studies. Comparison between modifier studies in human patients and mouse models of CF [12] – [15] will be especially interesting, given replication of at least one syntenic locus between species [16] . Among rarer disorders, some ciliopathy phenotypes (and related developmental abnormalities) may require or be modified by tri-allelic inheritance—requiring three alleles among two or more loci—among functionally related genes [17] – [21] . However, these examples are the exceptions. Most modifier effects in human genetic disorders are much less well understood, and the numbers of available subjects, allelic heterogeneity, and environmental factors that obscure modifier effects in smaller cohorts may limit systematic analyses for a large number of individually rare disorders. This increases the potential value of animal models for both candidate discovery and experimental validation of human modifier genes.
Identification of modifier genes in mouse models offers an opportunity to understand forms of plasticity in mammalian genetic architectures that could be exploited as a preclinical knowledge base in designing therapeutic strategies [22] . A very common finding in mouse models is phenotypic difference between strain backgrounds (e.g., 129 versus C57BL/6 in hybrid lines from gene-targeting experiments), but the sources of variation are not often pursued because of technical and resource constraints. Limited access to these modifier genes is a missed opportunity. Since modifiers are necessarily context dependent, any experimental organism is likely to model only a portion of any particular genetic network in disease. The accuracy of multigenic models likely depends on both the evolutionary plasticity of the network involved and the evolutionary distance between the model and human subjects. From this perspective, with rodents as a sister group to primates, mouse variants that have large effect sizes as modifiers while remaining phenotypically quiet on their own should be useful in identifying points in a genetic network where intervention might have higher therapeutic benefit than collateral cost.
The examples below argue that modifier genes are a field ripe for harvest, particularly with the recent arrival of several new community resources that improve experimental access to genetic variations among common strains. The situation is reminiscent of early days of positional cloning in many respects. Large numbers of loci have been reported and mapped ( Figure 1 ; Table S1 ), but few have been molecularly identified. Stories of unrewarded efforts have inhibited some investigators from pursuing otherwise tantalizing genetic effects. The two most important parallels, however, are that early successes—highlighting unexpected interactions—show the value of the approach and that new resources and emerging technologies are making the approach both more palatable and more cost effective. With new tools and a critical mass of encouraging results from several laboratories, this is a good time to consider both the value of the waiting crop and the machinery needed for harvest. Instructive examples to date include identification of genetic interactions from candidate genes, from mutagenesis screens, and from natural variants. Understanding modifier gene network architectures in mouse models may provide a powerful functional basis for inferring candidate interactions from human exome and genome sequences, particularly where sample size limits statistical power for independent discovery in clinical samples for Mendelian disorders.
10.1371/journal.pgen.1002644.g001
Figure 1
PubMed references to mouse modifier genes.
Growing interest in and recognition of modifier genes in mouse models is supported by increased publication rates in 5-year windows for the last 30 years. Mouse modifier papers have increased at a rate faster than the increase of total PubMed citations over the past 15 years.
Candidate Genes Are a Useful, but Limited, View of Interactions
Many of the genetic interactions known in mice come from direct tests of candidate interactions. Testing interactions first observed in other species can identify physiological context for gene pairs or networks that are conserved more deeply than the physiology or organ system for which they are most relevant to human disease. For observed binding partners, genetic interactions can test the biological relevance of likely physical contacts. Proposed interactions between mutations with similar phenotypes can also clarify points of convergence between previously separate pathways.
Candidate interactions are often based on homology to interacting genes in other species, including components of developmental signaling pathways, homeotic regulators, and transcription factor cascades in development. Some of these interactions may confirm modifier genes in the strict sense of Box 1 , but more often involve interactions between phenotypic null mutations available from other studies. While these are helpful in confirming functional conservation among pathways and defining mammalian contexts that might be relevant to disease, this paradigm is limited in its ability to identify levels of interaction unique to mammalian biology that might be expected from successive genome duplications, rescissions, and neo-functionalization in the lineage leading from ancestral vertebrates to primates and rodents (Euarchontoglires). Indeed, levels of redundancy among paralogous genes in mouse experiments often underscore important differences in trait architectures between mammals and other models. Moreover, focus on just the highly conserved rather than more plastic components of a genetic pathway or network might bias against finding genes that are more readily manipulated without damage to the organism.
Synthetic interactions between environmentally sensitizing alleles are another kind of candidate gene approach. For example, BALB/c and 129S1 mice are sensitive to adriamycin-induced nephropathy and show mitochondrial DNA (mtDNA) depletion in vulnerable organs after adriamycin treatment. Gharavi and colleagues identified a single amino acid substitution in the shared BALB/c and 129S1 allele of Prkdc , which encodes a DNA-activated protein kinase, as the sensitizing variant and asked whether this might be relevant to other mtDNA dysfunctions [23] . The human orthologue of Mpv17 , which encodes an inner mitochondrial membrane protein, is mutated in mtDNA depletion syndromes [24] , but the corresponding mouse mutation does not produce this phenotype. Papeta et al. showed that Prkdc potentiates mtDNA loss in Mpv17 -mutant mice: double mutant mice (but neither single mutant) suffer mtDNA loss and other features of adriamycin-induced nephropathy without adriamycin exposure, providing parallel gene×gene and gene×environment models.
The many successes in modeling the phenotypic consequences of predicted candidate gene interactions in an intact mammal should encourage us to consider what value might be obtained from screens that are less constrained by prior predictions—and therefore capable of identifying novel and unexpected interactions that might catalyze more rapid progress in the often complex genetic architectures relevant to disease.
Spontaneous Modifiers: Volunteers Lead the Way
An early success in using modifier genes to understand a genetic network in mice came from the dilute suppressor ( Mreg dsu ). This modifier arose spontaneously in a non-agouti , dilute stock, suppressing the coat color dilution but with no obvious phenotype of its own [25] . Importantly, dsu similarly suppressed pigmentation phenotypes in mutations at five of 11 classical coat color loci tested [26] , [27] , indicating a fundamental role in melanosome function. Molecular analysis revealed a spontaneous 11-kb deletion, creating a null allele in a vertebrate-specific gene, subsequently dubbed Melanoregulin [28] . Characterization of Melanoregulin in the context of its genetic suppressor activity revealed a previously unknown molecular function required for pigment granule transfer and other transactions among membrane-bound organelles [29] . This work remains an instructive example as few other strict-sense modifiers have been demonstrated to act on several different mutations and only one mouse modifier, Nxf1 Mvb1 , has been effective on a larger number.
Spontaneous mutations can also contribute to nominally wild-type inbred backgrounds. The Sodium channel modifier 1 ( Scnm1 ) locus was identified as a strain-dependent modifier of Scn8a med-J , a mutation in a neuronal sodium channel gene responsible for a range of neurological phenotypes [30] . Among several Scn8a mutations, the modifier is specific for the med J allele, an intronic single nucleotide variant that alters 5′ splice site usage. Positional cloning of the modifier identified a novel gene whose protein plays a direct and previously unsuspected role in pre-mRNA splicing [31] , [32] . The sensitizing modifier allele (C57BL/6) encodes a truncated protein (R187X), but is less severe than a subsequently generated null allele [33] . The interaction between Scnm1 and Scn8a med-J appears highly specific, as Scnm1 does not alter general RNA patterns in brain nor modify other tested mutations with similar defects [33] . Because the variant Scnm1 allele is only found in C57 and C58 strains, but not in other strains with otherwise similar Scnm1 haplotypes, it is proposed to have arisen as a spontaneous mutation in a progenitor stock rather than as a wild population variant [32] . The unexpected aspects of mammalian biology identified through the dsu and Scnm1 spontaneous modifiers should encourage further explorations of modifier genes.
Induced Mutations Allow Systematic Screens—with Dramatic Effects
Random mutagenesis to introduce and screen new variants has several advantages for identifying genetic interactions and trait architectures. Ethylnitrosourea (ENU) is especially efficient in mice [34] , [35] and in principle can both produce a range of alleles at a given locus and saturate a phenotype for simple loss-of-function alleles at most loci. High-throughput sequencing now rapidly identifies de novo mutations relative to a defined background [36] – [38] . With respect to modifier screens, mutagenesis should in principle identify both strict-sense modifiers and interacting genes with significant independent phenotypes, though the relative proportion may be difficult to predict.
Using epigenetically sensitive Agouti alleles and green fluorescent protein (GFP) transgenes that show variegated expression as phenotype reporters, Whitelaw and coworkers have recovered a substantial collection of ENU-induced Modifier of murine metastable epialleles dominant ( MommeD ) mutations. This approach creates an “outside-in” interaction network module, with multiple edges (interactions) converging on a single node (conditioning mutation or reporter gene) ( Figure 2A ). At least five of these mutations have been molecularly identified, encoding core components of chromatin regulatory complexes and a DNA methyltransferase [39] – [42] . The first recessive mutation identified from this screen, MommeR1 , is an amino acid replacement allele of transcription factor Foxo3a [43] . Each of these mutations has independent phenotypes in addition to the modifier activity, including recessive lethality for the dominant alleles and female infertility for Foxo3a .
10.1371/journal.pgen.1002644.g002
Figure 2
Modifier gene networks have directional edges between nodes.
Mouse modifier gene interactions can be diagrammed as nascent modules of an interaction network. (A) Identification of multiple modifier genes for a conditioning mutation through either mutagenesis or linkage analysis of strain variants can be represented as in “outside-in” network module, where one node (the conditioning mutation, beige circle) is a sink hub, acted on by each experimentally discovered modifier (light green circles). MommeD modifiers of the epigenetically sensitive A vy mutation are illustrated as an example. Direction of effect is indicated by the flared edges (connections) between nodes. (B) Validation of modifier mechanisms across independent mutations result in an “inside-out” module, with the shared modifier (green circle) acting as a source hub on several conditioning or test mutations. An incipient network around Nxf1 Mvb1 , based on shared genetic mechanism, is illustrated. Atrn mgL and several intronic ETn-induced mutations are not affected by Nxf1 Mvb1 variation (gray). Pitpna vb and Eya1 BOR have additional known modifiers (light green), adding modules to the incipient network.
Another instructive example comes from an ENU screen for dominant enhancers of dominant white spotting in a Sox10 lacZ /+ model of Waardenburg syndrome [44] . Among 230 pedigrees, Pavan and colleagues recovered three transmissible modifiers of spotting ( Mos1,2,3 ). Remarkably, these three induced mutations mapped to locations distinct from several previously identified modifiers of Sox10 and from each other, suggesting that Sox10 phenotypes are sensitive to perturbations at many positions in an extended genetic network. A collateral phenotype (cephalosyndactyly) and map position strongly suggested Gli3 as the Mos1 gene, which sequence and complementation analysis rapidly confirmed. Mos2 was subsequently identified as a null allele of the RNA regulator Magoh , with severe haploinsufficiency phenotypes in brain development [45] .
As illustrated in these examples, chemical mutagenesis allows efficient exploration of genetic networks unconstrained by prior hypothesis. Networks ascertained by this approach, in both examples, comprise modifier alleles that typically have independent deleterious phenotypes. Whether this is more often true of induced mutations or is a property of the specific networks tested, and whether milder alleles at the same genes might retain modifier effects without collateral phenotypes remains to be seen. Alternatively, networks based on variants in natural populations—or nominally wild-type strains—might highlight different genes and network connections, depending on biological trade-offs between modifier effect and collateral phenotypes in any gene's potential allele spectrum and the forces of selection through which a given variant has passed prior to discovery.
Natural Variants: Keys to Genetic Plasticity?
While inbred mouse strains are often referred to as nominally “wild type,” each represents a different sampling of wild (and a few laboratory-derived) alleles across the genome. Inbred strains in aggregate represent an abundant source of captive variation [46] , [47] and the vast majority of underlying genetic variants predate laboratory domestication [48] – [50] . This argues that the captive variants—on average—were sufficiently neutral to be sufficiently frequent among wild mice to be incorporated in the common laboratory strains, although some specific instances will be maladaptive by themselves or in combination with other variants. Modifier alleles among the nearly neutral majority of strain variants might be expected to favor, in comparison to mutagenesis experiments, either milder alleles or nodes in the relevant genetic networks with greater intrinsic plasticity—functional variation compatible with normal phenotype. Identification of network nodes with greater genetic plasticity may be especially useful for pointing out experimental and therapeutic approaches with higher likelihood to minimize collateral damage in modifying a specific condition. Three examples of early successes indicate some of the areas to which natural variants might contribute unique findings.
Studies in several laboratories have identified modifiers that alter Apc Min -dependent tumor phenotypes in a genetic cancer model [51] – [58] . Modifier of Min-1 ( Mom1 ) is the most well studied, beginning with its discovery by linkage analysis for intestinal tumor number nearly two decades ago [57] . Based on rough map location and strain distribution, the secreted phospholipase Pla2g2a was proposed as a positional and functionally variant candidate gene, with an apparent null allele caused by a single base insertion in Apc Min sensitive strains [59] . Mom1 is semidominant and analysis of tumor DNA suggested that its effects are not cell autonomous, consistent with its identification as a secreted enzyme. Complementation studies confirmed that Pla2g2a accounted for a large fraction of the variance in tumor number [60] (an additional component, Mom6 , was inferred to account for the full effect of the initial linkage [56] , adding as a grace note that linkage peaks may be driven both by variants of large effect and by regions containing more than one contributing variant). Elevated expression of human PLA2G2A was subsequently associated with survival in gastric cancer patients [61] , providing another example of cross-species validation. Mom1 alleles in mice are widely distributed across inbred strains, suggesting an early origin. Re-sequencing [49] , [62] and dense genotyping [48] data provide a partial answer. Review of published data shows that the Mom1 single base insertion is a derived allele that arose on a unique haplotype and that at least 2 Mb of this interval is shared by all characterized Mom1 S -allele strains. (Surprisingly, C57BL/6 and wild-derived MOLF/Ei strains share haplotype, 3630/3635 called variants, across 2.4 Mb around Pla2g2a , suggesting contamination of the MOLF/Ei stock with laboratory strains for this interval [63] .) While the origin of this Pla2g2a allele is not yet clear, the circumstantial evidence warrants further investigation as a possible wild variant.
The Heart failure modifier 2 , one of several mapped Hrtfm loci that modify the course of cardiomyopathy caused by overexpression of calsequestrin [64] , was recently identified as an allele of the troponin-interacting kinase gene, Tnni3k , by a combination of fine mapping, strain distribution, and transgenic studies [65] . Molecular studies in a cell culture model implicated a single intronic SNP as the functional variant, creating an alternate 5′ splice donor site that results in a shifted open reading frame in ∼70% of Tnni3k transcripts in calsequestrin-resistant strains. To illustrate potential origin of this strain variant, we tabulated both functionally tested alleles [65] and correlated sequence data from additional strains ( Table 1 ) [62] . This shows a broad distribution of each allele across accepted strain genealogies [66] for a region of modest haplotype diversity, suggesting an origin prior to development of laboratory strains and possibly wild-derived. Interestingly, subsequent work implicates the same Tnni3k variant in susceptibility to coxsackievirus-induced myocarditis [67] . If this variant is confirmed as a wild allele, it will be interesting to determine what factors might account for its accumulation despite the increased risk to cardiac challenges.
10.1371/journal.pgen.1002644.t001 Table 1
TNNI3 interacting kinase–Heart failure modifier 2 ( Tnni3k Hrtfm2 ) variant is widely distributed in laboratory mice.
Variant
Granby Farm-Derived a
Swiss Mice
China and Japan Stocks
Other Inbred
Wild Derived
Casq1-
resistant variant (derived) and haplotype
A/J, BALB/c, C3H, DBA, NZO
NOD, NZO
Share haplotype, variant not tested
I/Ln, MA/My, SEA
DDK, KK
BUB, RIIIS
Share haplotype w/o resistant variant
WSB (North America)
Casq1-s
ensitive variant (ancestral)
129P2, 129S1, AKR, C57BL/6, CBA
FVB
CAST (Asia), PWK (Europe)
a
Strain categories are taken from [66] ; Granby Farm-derived mice include both strains descended from Castle's mice and C57-related strains, both of which derived in part from A. Lathrop's stocks at Granby Farm. Strain-specific sequences from [62] . Inferred haplotypes from [48] .
The Modifier of vibrator 1 was originally detected as a semi-dominant modifier of neurodegeneration and premature death in vibrator mutant animals. The Pitpna vibrator mutation is an IAP-family endogenous retrovirus insertion into an intron of a phosphatidylinositol transfer protein gene, which reduces normal gene expression by competing with splicing of the adjacent exons [68] . Suppression by the CAST/Ei modifier allele occurs by elevating the level and proportion of normally spliced mRNA from the mutant gene. Positional complementation cloning identified the modifier as an amino acid substitution allele of mRNA nuclear export factor Nxf1 , suggesting a previously unexpected role for canonical mRNA export machinery in splicing fidelity in at least some contexts [69] . Importantly, this dose-responsive effect on alternative splicing has been demonstrated for six out of seven mutations caused by sense-strand IAP insertions in host gene introns, but none of 15 mutations with other classes of insertional events [70] . This creates an “inside-out” network module, with a modifier node acting on several different mutations, some of which have additional known modifiers ( Figure 2B ). Of special interest, the suppressing Nxf1 allele was found in a majority of wild castaneus isolates, confirming it as a natural variant, and haplotype analysis suggested directional selection, with a non-conservative amino acid replacement in Nxf1 as the youngest variant on the most common wild haplotype [69] .
Among modifiers not yet identified molecularly, the Modifier of dactylaplasia ( Mdac ) may impact several themes raised above, including selection, specificity, and epigenetic regulation. The functionally tested alleles are both well dispersed across inbred strains [71] , [72] , suggesting an early—and possibly wild—origin. The modifier acts on two different Dac alleles of Fbxw4 , caused by independent insertions of MusD-family retrovirus elements at different sites [73] , but has no obvious independent phenotype in reports from several different laboratories. Most interestingly, Mdac alleles alter both DNA methylation and the accumulation of inhibitory histone modifications on the Dac MusD elements [74] , raising the intriguing possibility that this locus might act directly in epigenetic mechanisms, without strong collateral phenotypes. This would be interesting both as a phenotypically quiet epigenetic modifier and potentially as a tool for titrating allele strength of other MusD/ETn-induced mutations with similar epigenetic profiles.
Identifying modifiers on the basis of natural variants may require more resources than de novo mutations (once generated), but have the added value of generally highlighting alleles, genes, or networks whose manipulation appears well tolerated by the organism and largely compatible with normal function, having been vetted by selective pressures. In the ultimate goal of finding pressure points through which to modulate disease networks, this may prove advantageous.
Going Forward: Tools for Accessing Strain Variation and Modifier Genes
How readily can we use modifier genes to predict sensitive pathways and nodes for therapeutic interventions? New tools and resources in mouse genetics should help to unclog the pipeline of modifier genes that have been detected but not molecularly defined. Maps of copy number variation [75] and comprehensive genome sequences of the most commonly used strains [62] , [76] will be enormously powerful. The combination of predictive sequence variants plus expression data from increasingly facile array and RNA-Seq methods should allow comprehensive consideration of candidate variants across even broadly defined loci. Because modifier genes require the presence of the conditioning mutation, strain resources built for genome-wide shuffling of alleles [77] may have less value for this application (in requiring many crosses to access the variation), while others such as chromosome substitution panels [78] , [79] may be useful for isolating individual modifier effects by limiting the need for de novo congenic strain construction. New tools for functional validation at fine scale are also essential to opening the pipeline of modifier gene studies in mice. Recent developments on this front include an increased diversity of strain-specific large-insert BAC libraries ( http://bacpac.chori.org/ , among others) for transgenic studies, larger-scale tools for engineering ES cells [80] – [82] , and most recently germline-potent ES cells from previously refractory strains [83] . These resources, some anticipated a decade ago [84] , [85] and some less expected, are only now widely available. This combination of tools and continuing innovations promises deeper insight into mammalian genetic architectures in health and disease, through the lens of modifier genes and mouse genetic diversity.
Supporting Information
Table S1
A curated list of mouse modifier genes. The attached spreadsheet collects several examples of modifiers and a few exemplar interactions that may strain the strict definition of the term. Entries were manually curated from entries in the Mouse Genome Database [86] , [87] with modifier, suppressor, or enhancer annotations; independent nested searches of PubMed for similar terms; authors' recollections; referrals from colleagues and suggestions from anonymous reviewers. We acknowledge that the list is incomplete and apologize to colleagues whose work we have missed by this limited search strategy. We have not independently reviewed the strength of linkage evidence for named loci. Molecularly identified modifier genes are emphasized with red text.
(XLSX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Imprecision in determining when and where dogs were first domesticated has vexed geneticists for the past 20 years and archaeologists for many decades longer. This has been particularly frustrating since dogs were certainly the first domesticated taxa, so understanding when and where our relationship with dogs began is crucial to comprehending the transition of humans from hunter-gatherers to farmers.
Genetic efforts to query the time and place of dog domestication have moved from mtDNA phylogeography through several generations of autosomal marker analysis and now enter an exciting new phase: the interrogation of whole genome sequences. Freedman et al. [1] present one of two recent papers (including Wang et al. [2] ) that generate and analyze multiple genomes of dogs and wolves. However, the approaches, sampling, and conclusions differ significantly between the two papers.
Dating the Divergence: Dogged by Mutation Rate Estimate Variation
Establishing the precise geography and timing of dog domestication using archaeology has been difficult for several reasons. Firstly, because wolves were once distributed across the entire Northern Hemisphere, zooarchaeologists have not been able to establish the wild or domestic status of fossil canid remains based solely on geographic location; thus cranial and dental characters have had to be used to differentiate domestic dogs from wild wolves. Despite uncertainty regarding natural morphological variation, the earliest appearance of dogs has been placed at about 15,000 years ago in Europe and the Far East. More recently, claims have been made that canid remains dated to about 30,000 years ago in Belgium, Ukraine, and Russia are either of early dogs or failed efforts at dog domestication; though some archaeologists remain unconvinced.
Geneticists first entered the fray in 1997 when, using mitochondrial control region fragments of dogs and wolves, Vila et al. [3] concluded that the two lineages diverged 135,000 years ago. Subsequent genetic studies have produced a wide range of estimates, often with large confidence intervals, and despite the generation of ever-larger data sets, date ranges have not yet begun to converge. For example, despite the fact that both Wang and Freedman generated high-coverage complete genomes from multiple distantly related dogs and wolves, they reach different conclusions about the date and population effects. Wang et al. [2] concluded that dogs and wolves diverged 32,000 years ago and that the domestication bottleneck was relatively mild, while Freedman et al. [1] , in their closely argued analysis, placed the wolf-dog bifurcation at 11,000–16,000 years ago and concluded that the domestication process resulted in a 16-fold reduction in population size ( Figure 1 ).
10.1371/journal.pgen.1004093.g001 Figure 1
Summary of the demographic model and sampling from Freedman et al. [1] .
Their critical inclusion of data from the Australian Dingo illustrates that high copy number in AMY2B is not a basal trait in dogs. The reciprocal monophyly of wild and domestic suggests that, despite the geographical diversity of sampling, descendants of the wolf population that contributed to dog domestication are not represented and may only be accessible using ancient DNA.
The primary reason for this disparity is reliance on molecular evolutionary rates that differ by an order of magnitude. As Freedman et al. [1] point out, little is known about the dog-specific mutation rate. By incorporating the entire range of published estimates, they demonstrate that the mutation rate is “the dominant source of uncertainty in dating the origin of dogs.” The use of the entire range of rates therefore results in a credible interval of the origin of dogs from 9,000–34,000 years ago, certainly in greater agreement with the archaeological estimates, but still lacking precision.
Did Dogs Originate Before or After Agriculture?
Though they may differ on whether the recently described 30,000-year-old canids were dogs, all zooarchaeologists support the contention that dogs were not only the first domestic animal, but that the appearance of dogs significantly predates the origins of domestic plants and early agriculture. They base this conclusion on the fact that the earliest dog bones found across the Old World from Europe to the Near East to the Kamchatka Peninsula have been reliably dated to several millennia prior to the first archaeological appearance of domesticated crops in the Near East and East Asia [4] .
A recent study of pooled resequenced whole genomes revealed that dogs possessed a seven-fold increase in the copy number of the AMY2B locus, a gene involved in amylase activity crucial to the digestion of starches. Based upon this observation, Axelsson et al. [5] concluded that the shift away from a more carnivorous diet was central and that the “development of agriculture catalysed the domestication of dogs.” In other words, the genomic evidence for copy number variation in dietary genes between dogs and wolves suggested that the archaeologists were wrong, and that dogs were domesticated not before, but after the origin of agriculture.
Freedman et al. [1] investigated this locus in their study and found not only that the AMY2B copy number increase was not fixed across all dogs (their Dingo possessed only two copies while the Saluki had 29), but also that the observed variation was polymorphic in nearly half of 20 wolves under investigation. These results suggest a more complex pattern of amylase copy number variation in dogs and wolves that reflects our long-standing relationship with dogs, but may not have resulted during early domestication.
Where Dogs Were Domesticated
Given the broad geographical range over which early dog remains have been discovered, archaeologists have been generally content to embrace the ambiguity of the zooarchaeological record and accept that there has not been sufficient evidence to support one or several geographic centers of dog domestication.
Many genetic studies have not been as reticent. For instance, though an early mitochondrial study concluded that dogs were domesticated just once in East Asia [4] , a subsequent analysis of African village dogs [6] cast doubt on this claim. A more recent study [7] using >48,000 single-nucleotide polymorphisms in wolves and dogs concluded that East Asian and Near Eastern wolf populations both contributed DNA to modern dog breeds. Though studies of nuclear markers have suggested diverse geographic origins for dogs, several authors continue to insist that all dogs descend from a single East Asian wolf population.
One reason for these discrepancies is likely to be the sustained admixture between different dog and wolf populations across the Old and New Worlds over at least the last 10,000 years. This has blurred the genetic signatures and confounded efforts at pinpointing the origins of dogs [7] . Another more intriguing reason stems from Freedman et al.'s conclusion that dog and wolf lineages are reciprocally monophyletic, suggesting that none of the modern wolf populations are related to the wolves that were first domesticated. In other words, the extinction of the wolves that were the direct ancestors of dogs has muddied efforts to pinpoint the time and place of dog domestication.
The sequencing of multiple complete, high-quality genomes of dogs and wolves is a significant step forward in the genetic hunt for the origins of our earliest domestic animal. The trick now is to extend the application of these methods to ancient remains: in effect, merging the materials and methods of both archaeology and genetics. By combining the expertise of both disciplines, not only might the extinct population of ancestral wolves be identified, but we will gain an enormous insight into the timing, location, and admixture patterns of dogs and wolves, thus revealing the complex origins of our first and best friend.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The retina ganglion cell layer (GCL) consists mainly of retinal ganglion cells (RGCs) and displaced amacrine cells. Over the years many efforts have been made to develop methods to objectively differentiate between RGCs and displaced amacrine cells. Several antigenic RGC markers, including Brn3a [ 1 – 3 ], Thy-1[ 4 ], neurofilament [ 5 ], and retrograde labeling have been viewed as good markers for RGCs. It was reported that a member of the RNA recognition motif family of RNA-binding proteins known as RNA-binding protein with multiple splicing (RBPMS), and its paralogue RBPMS2 (hermes), are expressed in RGCs in rats [ 6 – 10 ]. Recent studies have revealed that RBPMS can label all RGCs in normal retinas of mouse, rat, guinea pig, rabbit, and monkey [ 11 ]. In addition, RBPMS can also serve as a RGC marker for quantitative analysis in animal models of RGC degeneration induced by IOP elevation, optic nerve crush, and excitotoxicity [ 11 , 12 – 13 ]. The morphological and physiological properties of cat RGCs have been thoroughly investigated. However, whether RBPMS are a good marker for cat RGCs remains to be determined.
Although the neuromechanisms underlying glaucoma-induced RGC apoptosis remain controversial, increased levels of reactive oxygen species (ROS) are thought to play a crucial role in pathogenesis. A substantial body of evidence suggests that ROS are part of the signaling pathway in cell death after axonal injury. RGC axons within the globe are functionally specialized and are richly endowed with mitochondria. Mitochondria are important in the maintenance of cellular homeostasis as they are involved in numerous metabolic and physiologic functions. Mitochondria produce the energy required for nerve conduction in the unmyelinated part of ganglion cell axons. Thus, optic nerve injury-induced RGC apoptosis may at least partially be due to mitochondrial malfunction [ 14 – 15 ]. In this study, an optic nerve crush (ONC) model was used to examine RGC apoptosis. This ONC model has been widely used in studying the pathophysiology of glaucoma[ 16 – 19 ]. ALA is a disulfide compound found naturally in mitochondria that serves as the coenzyme involved in carbohydrate utilization necessary for the production of mitochondrial ATP. A substantial body of evidence shows that ALA is a superb antioxidant that enhances mitochondrial function [ 14 , 20 ]. ALA provides protection to the retina as a whole, and to ganglion cells in particular in ischemia-reperfusion injuries [ 21 ] and optic nerve crush [ 14 ]. Recent studies have revealed that ALA exerts a neuroprotective effect against oxidative stress in retinal neurons [ 22 – 23 ]. Despite such positive results, the effectiveness of ALA as a neural protectant in the retina has not been investigated.
The purpose of the current study was to determine: (1) whether RBPMS can be used as a selectively marker for RGCs in the cat retina; and (2) whether ALA can alleviate ONC-induced RGC injury.
Materials and Methods
Animals
Fourteen young adult domestic cats of both sexes with body weights of 2.2 to 3.5 kg were used in this study. Two cats were used for testing RBPMS antibody and twelve cats were used for study of the effect of ALA. Animals were purchased from a local research animal provider(Xinglong Institute for Experimental Animals, Beijing; registration number:110108600078158. This is an ordinary animal housing facility and it managed in keeping with national standards as described in ⟪Laboratory Animal–Requirements of Environment and Housing Facilities⟫ (GB 14925–2001). The care of laboratory animals and animal experimental guidelines used in this study conformed to the ⟪Beijing Administration Rule of Laboratory Animal⟫. Each animal was housed in an individual stainless steel cage (dimensions 100cm ×100cm ×113cm) under a 12-hour light/12-hour dark cycle. A feeding station, water fountain, litter box, scratching post and pet toys were provided in each cage. Room temperature was set at 18–21°C. Commercial cat food and water were provided ad libitum . Procedures were approved by the Institutional Animal Care and Use Committee (IACUC) at Peking University, and all procedures adhered to the ARVO Statement for the use of animals in ophthalmic and vision research.
Animal preparation and surgical procedures
Retrograde labeling
Two cats were anesthetized with intramuscular injections of ketamine (20 mg/kg) and xylazine (5 mg/kg) and placed in a stereotaxic instrument. Animals were kept warm with a heat lamp throughout the experiment. The superior colliculus (SC) was targeted as follows: a small hole was made using a dental drill (3 mm posterior to lambda; 2 mm to midline) and a 10ul Hamilton syringe was inserted vertically to a point 14 mm below the surface of the brain. A total of 1.5ul of RED Retrobeads (Lumafluor, Inc, Durham, NC) was slowly injected over 5 minutes and slowly withdrawn 10 minutes later. The wound was sutured and topical analgesic gel was applied three times a day for three days. Penicillin sodium (16000 U/kg) dissolved in sterile saline was then injected intramuscularly on a daily basis for three days to prevent infection. No animal in this study became ill or contracted an infection prior to the endpoint of the experiments.
Optic nerve crush (ONC)
Twelve cats underwent unilateral ONC. ONC surgery was performed following the procedure described by Weber and colleagues [ 24 ]. Briefly, cats were anesthetized as described above. The head of each animal was stabilized and the dorsal surface of the skull was shaved. Using sterile procedures, after the roof of the bony orbit was opened, the optic nerve sheath of the left eye was exposed in all treatment groups by blunt dissection of the overlying tissues without disturbing the nerve sheath or retinal artery. The left optic nerve was crushed 2 mm distal to the globe for 2 minutes with a 40 g vascular clamp (TKF-5-40, ARO Surgical Corp, Newport Beach, CA). The contralateral eye did not undergo any surgical procedure and served as a normal control. In all cases, the retinal blood supply remained grossly intact, as evaluated by direct ophthalmoscopic inspection. Observation of the fundus showed normal retinal vasculature. After removal of the clamp, penicillin sodium powder was applied topically, and the overlying tissue and skin were sutured and treated with betadine solution. The entire surgery required 30 minutes. Topical analgesic gel and penicillin sodium were applied as described above. Animal respiration rates and body temperature were monitored three times a day for three days after the surgery, and activity patterns were monitored 24 hours per day for seven consecutive days. One of the senior investigators (MPU) had a Peking University IACUC approved protocol in place (LA-2011-054) for the use of humane endpoints and early euthanasia for animals that became severely ill prior to the experimental endpoint. The criteria used to determine when to euthanize animals were based primarily on responses to pain. Responses to pain were judged by external physical appearance, clinical signs, unprovoked behavior, and behavioral responses to external stimuli. Two animals that experienced severe pain that could not be relieved by analgesic treatment for two consecutive days were euthanized.
Alpha-lipoic acid (ALA) application
Cats underwent ONC and were divided into two groups: five cats received no ALA treatment, and five cats received a single intravenous injection of ALA (15 mg/kg) (STADA, Germany) through the femoral vein immediately after ONC. Tissues from five cats with ONC alone, and from five cats with ONC and ALA treatment were analyzed.
Tissue Processing
After an interval of 5 days to allow for retrograde labeling surgery and 7 days for ONC surgery, the animals were deeply anesthetized as described above. The retina preparation process was similar to that previously reported [ 24 ]. Animals were euthanatized with an intravenous injection of pentobarbital sodium (>50 mg/kg) and perfused transcardially with 0.75 L of physiologic saline (0.9%), followed by 1.0 L of solution containing 4% paraformaldehyde (PFA) (#15710, Electron Microscopy Sciences, Ft. Washington, PA, USA) in 0.1 M phosphate buffered saline (PBS)(pH 7.4, Sigma-Aldrich, USA). The brains were placed in the same fixative for subsequent processing. The eyes were enucleated, and the lens and vitreous were carefully removed in 0.1 M cold PBS. The eyecups were fixed immediately by immersion in 4% PFA as described above for 60 min at room temperature. The retinas were then dissected from the retinal pigment epithelium, and 3–4 radial cuts were made to flatten the retina. The retina was then mounted on a nitrocellulose membrane filter (GSWP02500, Millipore, USA) and transferred to a 6-well cell culture cluster (#3516, Corning Incorporated, Corning, USA) for immunohistochemistry staining.
RNA binding protein with multiple splicing (RBPMS) antibody generation
A rabbit polyclonal antibody was generated against the N-terminus of the RNA Binding Protein Multiple Splice (RBPMS) polypeptide (RBPMS4-24), GGKAEKENTPSEANLQEEEVR by a commercial vendor (ProSci, Poway, CA). RBPMS is highly conserved among mammals and the polypeptide sequence used for immunization is identical in mouse, rat, monkey and humans (NCBI Protein Bank, http://www.ncbi.nlm.nih.gov/protein ). Rabbit sera were collected following immunization and affinity purified using a RBPMS polypeptide affinity column. The affinity purified antibody was shown to immunostain ganglion cells in mouse and rat retina [ 11 ]. To evaluate the specificity of the RBPMS immunostaining, a preabsorption control was performed with the rabbit antibody. Briefly, RBPMS antibody was diluted in 0.1 M PB containing 0.5% Triton X-100 and mixed with RBPMS polypeptide at a final concentration of 1 μg/ml for two hours at RT. No RBPMS immunostaining was present in tissue sections incubated with the rabbit antibody preabsorbed with RBPMS and processed by standard immunohistochemical techniques.
Antibodies and reagents
The primary antibodies used in this study were rabbit polyclonal antibody against RBPMS (1:1000) and goat polyclonal antibody against Choline Acetyltransferase (CHAT) (1:200, Cat# AB144P, Millipore, LOT#2070392, RRID:AB_2079751). The secondary antibodies were DyLight 549 anti-rabbit IgG, DyLight 488 anti-rabbit IgG (1:2000, Vector Labs, Inc., Burlingame, CA, USA), Alexa Fluor 488-conjugated AffiniPure Donkey anti-goat IgG (1:400, code: 705-545-147, Jckson Immuno Research, USA), and Dylight 549 conjugated Donkey anti-rabbit IgG (1:2000, code: 611-742-127, Rockland Immunochemicals Inc.,USA). 2%Triton X-100 (Sigma-Aldrich, USA) and 0.5% dimethylsulfoxide (Sigma-Aldrich, USA) in 0.1 M PBS (Sigma-Aldrich, USA) were used to facilitate antibody penetration. Blocking reagent (CAS-Block, Invitrogen Corp, US A) was used to suppress non-specific immunoreactions. The dilutions used to dissolve antibodies were 10% CAS-Block, 2% Triton X-100 and 0.5% dimethylsulfoxide in 0.1 M PBS.
Immunohistochemistry
The retinas were washed six times for 10 minutes with 0.1 M PBS on a shaker at room temperature, blocked for 48 hours at 4°C, and incubated in primary antibody solution for 7 days at 4°C. After rinsing six times as described above to remove excess primary antibody, the retinas were incubated in secondary antibody overnight at 4°C. After washing an additional six times as described above, the free floating retinas were mounted on slides, immersed in anti-fade mounting medium (#17985–11, Electron Microscopy Sciences, Hatfield, PA) and covered with a cover slip (#12-545G, Electron Microscopy Science, USA).
Imaging processing and quantitative morphological analysis
Entire whole-mount retinas were photographed with a 2.5× objective (3207×2415 μm²/microscope field), a 10×objective (872 ×65403BCm²/microscope field) and a 20×objective (436×327 μm²/microscope field) under fluorescent microscopy (BX51, Olympus, Japan). In each sampling field, RGCs were labeled for RBPMS immunoreactivity.
Photographic images were montaged with Adobe Photoshop CS5 graphic software (Adobe Systems, Inc., San Jose, CA, USA). The average soma densities of RBPMS expressing RGCs at different eccentricities were calculated with Photoshop. RBPMS labeled RGC counts were determined from retinal whole-mounts and were converted to cells per square millimeter. For the cell density versus spatial quantification analysis of normal retina whole-mounts, each retina was divided into superior, inferior, nasal and temporal quadrants. Twenty predefined areas (436×327 μm 2 /microscope field) with a separation of 1000 μm were analyzed from the central to the peripheral retina. Location of the area centralis (AC) was selected as the zero retinal eccentricity point. The twenty sampling areas were distributed along the vertical axis. To compare RGC survival under three different conditions (normal, ONC with or without ALA treatment), we divided the retina whole mounts as described above and selected only ten areas with six areas next to the AC in the superior quadrants and four areas next to the AC in the inferior quadrants. Ratios of total numbers of surviving RGCs in these ten areas of the ONC eye to those in the normal control eyes were obtained to estimate the survival rate of injured RGCs. Cell counting was carried out in a double-blinded manner. The final images were processed and saved in TIFF format at 300 dpi using Photoshop software. To analyze the alpha-RGC distribution pattern under these three different conditions, the AC was selected as the geometric center of the retina, and from this point the entire retina was divided into 8×8 grids, each grid covering a 2 mm×2 mm retinal area. The sampling area at the center of each grid was 872×654 μm 2 , and the density was determined by dividing the total number of RBPMS expressing RGCs identified within this area by 0.57. An isodensity map of retinal RBPMS expressing alpha cells was generated using the Contourf function of Matplotlib software [ 25 ]. The contour plot was generated by assigning a color code to each of the 64 grids according to its RBPMS expressing alpha cell isodensity value (density/mm 2 ) within a six color scale ranging from 0 (camel) to 60, representing a higher isodensity region (navy blue).
Statistics
All results are reported as mean ± standard error (SE). Statistical significance was calculated using the Students’ T-test for comparisons between two groups, and one-way ANOVA for multiple comparisons followed by the SNK test as a post hoc test with SPSS 13.0 for Windows Software (SPSS Inc., Armonk, NY, USA). Values of P<0.05 were regarded as significant in all comparisons.
Results
Retrograde labeling and double fluorescence immunohistochemistry
At least eight microscopic fields from each of two retinas were included for quantitative analysis. 100% of the cells labeled by microbeads ( Fig 1 ) were immunoreactive for RBPMS, while only 15.8± 0.9% (range 14.8% to 17%) of RBPMS positive cells were labeled by microbeads ( Table 1 ). No RBPMS positive cell expressed CHAT ( Fig 2 ) in the GCL.
10.1371/journal.pone.0160309.g001
Fig 1
Double labeling with RBPMS, and retrograde RGC labeling with Microbeads.
A: RBPMS positive cells in a normal retina. B: Microbead labeled RGCs. C: Co-localization of RBPMS (green) and Microbeads (red). Scale bar: 50 μm.
10.1371/journal.pone.0160309.g002
Fig 2
Double labeling with RBPMS and CHAT.
A: RBPMS positive cells in a normal retina. B: CHAT positive cells. C: Co-localization of RBPMS (red) and CHAT (green). Scale bar: 50 μm.
10.1371/journal.pone.0160309.t001
Table 1 Quantification of double labeling with microbeads and RBPMS in whole mount retinas.
Retina1
Retina2
% of RGCs (beads) are RBPMS-positive
100
100
% of RBPMS-positive cells are RGCs (beads)
15.8
15.9
( n = 2. eight images per retina and sixteen images in total)
The distribution patterns of RBPMS expressing RGCs in cat retinas
Retinal ganglion cells of various soma sizes and RBPMS-RGCs were distributed across the entire retina. Fig 3 shows a micrograph of typical immunocytochemically stained RBPMS-RGCs in a cat retina. We quantitatively analyzed the numbers and density of RBPMS-RGCs ( Fig 4G and 4H ). The density distribution reached a peak at the AC and gradually tapered down at the periphery of the retina ( Fig 4H ). This RBPMS-RGC distribution profile matches that of HRP labeled RGCs [ 26 ]. The numbers and distribution of RBPMS-RGCs were also quantitatively analyzed in whole mount retinas. One of the characteristics of RGC distribution in cats is that the density of RGCs change drastically within 1 mm of the center of the AC. Thus, we counted and verified this pattern of density distribution around the AC. The density of RBPMS-RGCs within 1 mm 2 of the center of the AC was 5610/mm 2 , while density increased to 8174/mm 2 when the sampling area was reduced to 0.0314 mm 2 . If the sampling area was reduced to 0.016 mm 2 , the density increased to 9043 mm 2 (n = 4). Thus, the peak density of RBPMS-RGCs varied with sampling area with respect to the distance from the center of the AC. We then counted the numbers of RBPMS-RGCs at ten superior and ten inferior locations along the vertical meridian. The RMBMS density distribution profile matched that of the HRP stained RGCs, as shown in Fig 3H . We next compared the numbers and density of RBPMS-RGCs under different experimental conditions. The top panels of Fig 4 illustrate whole mount retinas under different experimental conditions ( Fig 4A : Normal, Fig 4B : ONC, and Fig 4C : ONC with ALA treatment (ONC+ALA)). The distribution pattern of RBPMS-RGCs in a small area (encircled dashed square) of a normal whole mount retina is shown in Fig 3D , while Fig 4E and 4F illustrate RBPMS-RGCs after ONC and ONC+ALA, respectively. To quantitatively assess the impact of ONC on RGCs under these conditions, the number of RBPMS expressing RGCs was counted in ten sampling locations along the vertical meridian, six from the superior and four from the inferior retina as shown in Fig 4G . The total number of RBPMS-RGCs was counted and is summarized in Table 2 . In comparison with the control (1466 ± 20/mm 2 ), ONC resulted in a sharp reduction in the number of RPBMS-RGCs (1012 ± 38/mm 2 ) while ALA treatment reduced the impact of ONC on RBPMS-RGCs (1178 ± 27/mm 2 ).
10.1371/journal.pone.0160309.g003
Fig 3
Micrograph of RBPMS expressing RGCs from a normal cat retina.
Micrograph location: 2.5 mm temporal and 1.6 mm superior to the AC. Scale bar: 100 μm.
10.1371/journal.pone.0160309.g004
Fig 4
Number and density distribution of RBPMS-RGCs under different conditions.
A-C reveal retinal distribution patterns of RBPMS expressing RGCs under different experimental conditions. A: RBPMS expressing RGCs in a normal retina, B: ONC retina, C: ONC+ALA retina. Scale bar: 2 mm. D-F show magnified micrographs of RBPMS immunoreactive RGCs (Sampling location: 2 mm superior to the AC from A-C) under different experimental conditions: D: normal retina, E: ONC, F: ONC + ALA. Scale bar: 50 um. G shows a diagram of sampling sites along the vertical meridian on each retina. The open rectangle depicts retinal location where micrographs were taken. The filled rectangles show sampling locations at different eccentricities. The large filled circle represents the optic disk (OD) and the small filled circle at the intersection of horizontal and vertical meridians shows the AC. Panel H shows a comparison of RBPMS stained and HRP stained density profiles. The filled red triangles and solid line depict RBPMS. Open black circles and the dashed profile which is taken from Hughes (1981, his Fig 13), is shown for comparison. The open arrow indicates sample area dimension and the distance between centers of two sample sites. Each sampling area was measured as 436×327μm 2 . The distance between centers of neighboring sampling areas is 1 mm.
10.1371/journal.pone.0160309.t002
Table 2 Average density of RGCs after ONC with or without lipoic acid treatment.
NORMAL
ONC
ONC+ALA
Alpha-RGCs (cells)
68±4
29±4
41±4
total-RGCs (cells)
1466±20
1012±38
1178±27
(n = 5 for each group.)
ALA improves the distribution patterns of alpha ganglion cells in the crushed optic nerve of cats
One of the morphological characteristics of alpha cells is their gigantic soma. RBPMS expressing alpha cells (RBPMS-alpha cells) were thus defined by giant sized perikaryon (>20 μm at AC and >35 μm, in the visual streak at the peripheral retina), and alpha cells with soma size that is always larger than any of the RGCs at different eccentricities. As RBPMS antibody staining is present in the perinuclear zone of medium sized soma, and is found predominantly in the cytoplasm of large sized soma, the gigantic cell body of alpha cells is thus readily identifiable. Fig 5A–5C shows RBPMS-alpha cells. These cells were photographed at different eccentricities ( Fig 5A : 1 mm; Fig 5B : 5 mm; and Fig 5C : 10 mm). Fig 5D shows the distribution pattern of alpha cells in a normal retina. Each number represents the total number of alpha cells found within a sampling field of 1 mm 2 . The distance between the centers of two neighboring sampling fields was 1 mm. The density distribution of alpha cells was measured across the entire retina, and the peak density at the AC was 186/mm 2 . Along the horizontal meridian, the density was reduced to 61/mm 2 two millimeters from the AC on the nasal retina and to 37/mm 2 on the temporal retina. Along the vertical meridian, the density was reduced to 47/mm 2 two millimeters from the AC on the superior retina and 53/mm 2 on the inferior retina. Cell density gradually tapered down along the vertical and horizontal meridians. We defined the edge of the visual streak as the retinal region where the alpha cell density was greater than 20/mm 2 . The area encircled by the solid line represents the visual streak ( Fig 5D ). Because cat retinal alpha/Y cells are more susceptible to optic nerve injury than other RGCs [ 27 – 28 ], we first observed the impact of ONC on RBPMS-alpha cells. The top panels ( Fig 6A : normal, Fig 6B : ONC, Fig 6C : ONC+ALA) show images taken from the superior retina, 1 mm from the AC. We then compared RBPMS-alpha cells at a given retinal location (area enclosed by a dashed line). Panels D through F show enlarged micrographs of the sampled areas (256× 256 μm 2 ) under different experimental conditions ( Fig 6D : Normal, Fig 6E : ONC, and Fig 6F : ONC + ALA). Asterisks denote RBPMS-alpha cells. ONC completely eliminated RBPMS-alpha cells ( Fig 6E , 13/mm 2 ) compared to nomal retina (6D, 70/mm 2 ), while ALA treatment substantially prevented ONC-induced injury to these cells ( Fig 6F , 21/mm 2 ). We next investigated the global impact of ONC on RBPMS-alpha cells. As shown in the bottom panels ( Fig 6G–6I ), the isodensity profiles of RBPMS-alpha cells reveal different distribution patterns after ONC. It is evident that ONC resulted in a substantial shrinkage of the high density alpha cell distribution area ( Fig 6H ). Although ALA treatment did not change the alpha cell peak density at the AC (control: 186/mm 2 , ONC: 67/mm 2 , ONC+ALA: 69/mm 2 ), the sizes of both the high isodensity zones (>60/mm 2 ) and low density regions (0–10 mm 2 and 10-21/mm 2 ) were substantially recovered ( Fig 6 and Table 3 , data in S1 Table ). Finally, the survival rates of RBPMS-alpha and RBPMS-RGCs were analyzed ( Fig 7A and 7B ). Except for one superior retinal location (eccentricity = 5 mm), the density distributions at every measured site showed a sharp decline and ALA improved the density distribution of alpha cells ( Fig 7C ). As for the RBPMS-RGCs, results from three out of eleven measured sites show that ONC reduced the density, and ALA improved the local density distribution of RBPMS-RGCs ( Fig 7D ). Together, these results suggest that alpha cells are more sensitive to ALA treatment than other classes of RGCs after ONC.
10.1371/journal.pone.0160309.g005
Fig 5
RBPMS expressing alpha cells and their density distribution pattern.
A-C illustrate micrographs of RBPMS-alpha RGCs photographed at 1, 5, and 10mm from the AC. Scale bar: 20μm. Arrows depict alpha cells. D shows density distribution pattern of RBPMS-alpha RGCs from a normal cat retina. Each number represents the total number of alpha cells encountered within the sampling field. The size of sampling area was 1×1 mm 2 . The distance between centers of two neighboring sampling fields was 1 mm. The peak density at the AC was 186/mm 2 . The dashed circle in the center depicts the optic disk. The edge of the visual streak is defined as local alpha cell density greater than 20/mm 2 . The area encircled by the solid line represents the visual streak. Scale bar: 5 mm.
10.1371/journal.pone.0160309.g006
Fig 6
Number and density distribution of RBPMS-alpha RGCs under different experimental conditions.
A-C show retinal micrographs taken 1 mm temporal to the AC in the superior retina. A: Normal, B: ONC, C: ONC+ALA. Scale bar: 100 μm. D-F show magnified areas in the panels above (dashed line enclosed areas). Asterisks denote alpha cells. Scale bar: 100 um. Bottom panels provide alpha cell isodensity distribution maps under different experimental conditions. G: Normal (n = 3), H: ONC (n = 3), and I: ONC+ALA (n = 4). Color scale ranges from 0 (camel) to 60 or higher RGCs/ mm 2 (navy blue): 0–10 –camel; 11–20 –desert dune; 21–30 –light brown; 31–40 –brown; 41–60 –light blue, 61 and up—navy blue. Scale bar: 2 mm.
10.1371/journal.pone.0160309.g007
Fig 7
Histograms of survival rate and density distribution of RBPMS-alpha cells and RBPMS-RGCs under different experimental conditions.
A: Survival rate of alpha-RGCs, B: Survival rate of RGCs, C: Density distribution of alpha cells at different eccentricities, D: Density distribution of RGCs at different eccentricity. Mean ± SE (n = 5, *compared to normal P<0.001, # compared with ONC, #P<0.05,## P<0.01).
10.1371/journal.pone.0160309.t003
Table 3 Alpha cell isodensity distribution under different experimental conditions.
Density(cells/mm 2 )
NORMAL (%)
ONC (%)
ONC+ALA (%)
0–10
33.6±4.28
85.6±1.72
59.7±8.47 *
11–20
34.5±1.71
12.1±1.80
35.9±7.11 **
21–30
17.4±0.81
1.13±0.12
2.84±1.21
31–40
7.10±1.38
0.60±0.18
0.81±0.18
41–60
5.03±0.46
0.50±0.02
0.66±0.14
>60
2.35±0.39
0.06±0.01
0.10±0.02 *
(normal , n = 3; ONC , n = 3; ONC+ALA , n = 4 . Mean ± SE,
*compared to ONC, *, P<0.05
**, P<0.01)
Discussion
In this study fourteen cats were used, including two cats for demonstrating that RBPMS selectively labels all cat RGCs. Twelve cats were used to study the impact of ONC-induced neurodegenerative changes on RGCs and the neuroprotective effects of ALA, and two of the cats in this group were euthanized before the endpoint of the experiment. Retinas from remaining ten cats were divided into two groups (ONC alone and ONC+ALA) and were used for data analysis. One or two retinas in each group were partly damaged during the tissue processing, and were inadequate for some data analysis. The sample size of each group therefore varied.
Numerous efforts have been made to develop methods to selectively and reliably differentiate between RGCs and displaced amacrine cells. However, selective RGC markers remain to be identified. Evidence has demonstrated that RBPMS can selectively label retinal ganglion cells across several mammal species [ 7 , 11 – 12 , 29 ], but there has been no report of RBPMS labeling RGCs in cats. In this study, we used microbead retrograde labeling and RBPMS staining to determine whether cat RGCs can be labeled by RBPMS. 100% of retrograde labeled RGCs were RBPMS positive, illustrating that RBPMS label all RGCs in the cat retina. In this study, we observed that retrograde labeled RGCs are located mainly in a small part of the retina whole mount, which may explain why only 15.8% of RBPMS positive cells were labeled by microbeads from the SC. Tracer may not spread through the entire SC, and injecting tracer into SC may thus be insufficient for labeling all RGCs. We also tested CHAT which marks displaced amacrine cells and RBPMS double labeled retina whole mounts. No RBPMS positive cells were labeled with CHAT, showing that RBPMS is a reliable marker which can distinguish RGCs from displaced amacrine cells. In addition, we compared the total number and peak density of RBPMS immunoreactive RGCs with RGCs identified by classical cresyl violet or methylene blue staining [ 30 – 32 ] and HRP retrograde labeled RGCs [ 26 ]. Over the years, several attempts have been made to accurately determine the peak density of RGCs at the center of the AC. However, the density varies from less than 6000/mm 2 [ 30 ] to over 10000/mm 2 [ 26 , 32 ]. This variation may be due to differences among animals [ 31 ]. In addition, the actual size of the counting area may be crucial, as peak density drops sharply within the AC. For example, the density is less than 6000/mm 2 if the counting area is 1 mm 2 [ 30 ], whereas it is over 10000/mm 2 if the area is reduced to 0.0089 mm 2 [ 26 ]. Consequently, we found RBPMS-RGCs within 1 mm 2 of the center of the AC have a density of 5610/mm 2 , while the density increased to 8174/mm 2 when the area was 0.03144mm 2 , and if the sampling area was reduced to 0.016 mm 2 , the density further increased to 9043 mm 2 (n = 4). These numerical findings closely resemble previous observations (9040/mm 2 , [ 26 ], 8788/mm 2 , [ 32 ]). We thus confirmed that the peak density of RBPMS labeled RGCs matches previously reported results that were obtained using conventional staining methods. We next quantitatively analyzed alpha cells. As shown in Fig 4 , the total number of RBPMS-alpha cells was 5682. This number closely resembles the number derived by HRP staining (5600, [ 33 ] and is in agreement with results of both Wässle and colleagues [ 32 ] and Stone [ 31 ]. As for the peak density of RBPMS-alpha cells in the AC, our estimation of 186/mm 2 (n = 3) is comparable to previous reports (195/mm 2 , [ 32 ]; 186/mm 2 , [ 31 ]. We thus demonstrate that RBPMS can be a selective RGC marker for cat cells as well as for other mammalian species [ 11 ]. RBPMS expression in RGCs was first demonstrated by analyzing its transcription with RT-PCR and in situ hybridization [ 7 ]. These authors observed very few RGCs expressing RBPMS in axotomized retinas. Reduced RBMPS expression levels in RGCs can be observed in different optic nerve injury animal models [ 11 – 13 ]. Taken together, these results suggest that RBPMS can be used to identify cat RGCs.
There is a substantial body of evidence that large soma RGCs are vulnerable to injury in humans and monkeys [ 34 – 37 ]. In addition, numerous physiologic and morphologic results reveal that Alpha/Y cells are more sensitive to optic nerve injury than other RGCs in cats [ 27 – 28 , 38 – 40 ] and canines [ 34 ]. There is evidence that Beta/X cells are more sensitive to injury than Alpha/Y cells [ 41 – 42 ]. Nevertheless, there is a significant decrease in the soma, dendritic field size, and dendritic complexity of alpha cells, while beta cells display only a significant decrease in soma size [ 39 ]. It was reported that up to 55% of RGCs located on the periphery of the retina and 47% around the AC survived one week after ONC, and BDNF treatment drastically improved survival rates to 91% in both of these locations [ 40 ]. Contrary to this observation, we found substantial shrinkage of the topographic isodensity zones of RBPMS-alpha cells, especially in high density areas, and there was a 70% drop in the 10-21/mm 2 isodensity zone seven days after ONC. However, as shown in Fig 6 , the peak density area around the AC showed significant recovery (see Table 3 ), and two low density zones also recovered. For example, the 10-21/mm 2 zone had a full recovery. The full recovery of this low density zone could due to several factors. One possibility is based on the speculation that RGCs in high density zones consume more oxygen and are thus more sensitive to ONC, while RGCs in low density zones consume less oxygen and better tolerate ONC-induced injury. ALA is therefore an effective antioxidant in the short term, as it alleviated the stress of RGCs in low density zones. A second point of speculation is that the recovered 10-21/mm 2 zone occupied some of the areas that were high density zones prior to the injury. Indeed, the observed 0-10/mm 2 isodensity zone showed similar recovery trends. The greater retinal space occupying area of this zone may be due to massive RGC injury and recovery after ONC. Thus, to quantitatively and objectively evaluate the impact of ALA treatment, it would be important to verify visual functions of RGC undergoing recovery within different density zones. In this study, axotomy-induced cell apoptosis resulted in decreased numbers of RBPMS positive neurons after ONC. However, this may also due to axotomy-induced RBPMS expression level down regulation that is difficult to recognize. In fact we observed that a small portion of RGCs in the ONC model were weakly labeled with RBPMS. Research has shown that optic nerve crush can produce a transient shrinkage of RGCs in rats [ 43 ]. A possible explanation for the decrease in numbers of alpha cells is that there is cell shrinkage after axonal lesions.
Although the neuromechanisms underlying glaucoma-induced RGC apoptosis remain controversial, increased levels of ROS are thought to play a crucial role in its pathogenesis. A substantial body of evidence suggests that ROS are part of the signaling pathway in cell death after axonal injury. RGC axons within the globe are functionally specialized and are richly endowed with mitochondria. These mitochondria generate the high energy required for fast optic nerve conduction in the unmyelinated part of RGC axons. Thus, optic nerve injury-induced RGC apoptosis may be at least partly due to mitochondrial malfunction [ 14 ]. The present study shows that ALA treatment improved RGC survival rate after ONC, which is consistent with previous reports showing that BDNF induces protection [ 39 ]. However, after ischemia-reperfusion, the retinal BDNF mRNA level was not upregulated by ALA [ 21 ]. Therefore, these two substances may act through different neuromechanisms. Nevertheless, recent studies have revealed that ALA exerts a neuroprotective effect against oxidative stress in retinal neurons [ 22 – 23 ]. Despite such positive results, the effectiveness of ALA as a neural protectant in the retina remains to be determined.
Supporting Information
S1 Table
Raw data of alpha cell isodensity area under different experimental conditions.
(DOCX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The last decade has seen a resurgence of pertussis in high-income countries to levels experienced over half a century ago [ 1 ]. Possible reasons for this include waning immunity following acellular vaccination, antigenic divergence of circulating strains from vaccine antigens as well as increased ascertainment due to improved diagnostic tools [ 2 – 5 ].
Introduction of the whole cell (wP) pertussis vaccine in the 1940s greatly reduced the incidence of pertussis. wP has since been succeeded by acellular vaccines (aP), mainly in high-income countries. The South Africa National Expanded Programme on Immunization (EPI) replaced wP with a combination formulation aP (DTaP-IPV/HIB; Pentaxim®, Sanofi Pasteur) in 2009. The primary schedule comprises doses at six, ten and 14 weeks with a booster at 18 months of age [ 6 ]. The only available reliable data indicates that a decade ago vaccine coverage for the Western Cape Province, where this study was done, was 97% at six weeks, 90.8% at 10 weeks and 85.2% at 14 weeks. By 18 months of age coverage had declined to 58.7% [ 7 ]. There has been no change on policy over the period to change the situation. The current national schedule does not include vaccinating pregnant women.
Risk factors for pertussis include lack of immunization or impaired immune responses to vaccination. Laboratory studies suggest that immune responsiveness to pertussis vaccines may be impaired by both infection and intrauterine exposure to HIV even in HIV-uninfected children [ 8 – 11 ]. Another factor associated with reduced immune responses to pertussis vaccine is poor nutritional status, a common problem among children in low and middle-income countries (LMIC) [ 12 ]. Although household use of biomass fuels, indoor air pollution, and cigarette smoking have been associated with an increased risk of respiratory illness and bacterial carriage in children, it is unclear if these impact on the risk of pertussis [ 13 ].
With the resurgence of pertussis, adults and adolescents, who tend to exhibit milder and atypical symptoms of pertussis, are now recognized as important sources of pertussis in infants. In particular, household contacts pose the greatest risk to unvaccinated or partially vaccinated infants [ 14 – 16 ].
We aimed to investigate risk factors for Bordetella pertussis disease in a cohort of African children less than 13 years of age hospitalized with lower respiratory tract infection (LRTI) in a high HIV prevalence setting [ 17 ].
Materials and methods
Children less than 13 years of age admitted over a one-year period (07 September 2012 to 06 September 2013) for LRTI to the Red Cross War Memorial Children’s Hospital (RCH), Cape Town, South Africa, were prospectively screened for enrolment. The hospital provides services for children up to 12 years of age. Children with WHO-defined severe pneumonia (age specific tachypnoea or/and lower chest indrawing requiring hospitalization) or apnea were eligible to be included. A child was included if the legal guardian gave written consent and the child had not been in contact with a health care facility in the previous 48 hours to two weeks prior to screening for enrolment. Enrolment was limited to the first four qualifying children per working weekday.
History of symptoms of the presenting illness and information on current socio-demographic factors including type of housing, access to amenities such as tap water, electricity and toilet facilities, was taken from the caregiver. The mother’s level of education was recorded. Socio-economic status was categorized into quartiles on the basis of a validated weighted composite score used elsewhere that included asset ownership, employment and education [ 18 ]. The use of household biofuels, presence of smokers in the household and the number of people sharing the bedroom with the child were established. Information on day-care attendance was also collected.
The mother’s HIV status (and that of the primary caregiver if this was not the mother) was established. If the mother or caregiver was HIV infected the latest available CD4 count was recorded and used in the staging of HIV disease according to the Centre for Disease Control (CDC) classification [ 19 ]. History was taken on the presence and duration of recent primary caregiver respiratory symptoms as well as presence and numbers of other household members with similar symptoms.
The vaccination status of each child was verified using the national standardized immunization handheld record, the Road to Health Card (RTHC); specifically, the date and type of each vaccine was copied from the record. All children who missed any vaccinations that should have been received for age, where referred to an immunization catch-up program.
The weight of each child, as measured on admission, was used to evaluate nutritional status using World Health Organization (WHO) weight for age Z scores (WAZ). Mild under-nutrition was defined as ≤ -1 WAZ > -2, moderate under-nutrition ≤ -2 WAZ > -3 and severe under-nutrition WAZ ≤ -3 [ 20 ].
Each child was screened for HIV infection using an ELISA test (Architect HIV Ag/Ab Combo, Abbott Diagnostics, Wiesbaden). The diagnosis of HIV infection was made if both the ELISA and an HIV PCR test (COBAS AmpliPrep/COBAS Taqman HIV-1, Roche Molecular Diagnostics, Pleasanton, CA) were positive in children younger than 18 months. A positive ELISA was confirmed with a different ELISA test (Enzygnost Anti-HIV 1/2 Plus, Siemens/Dade Behring, Erlangen) in children older than 18 months to diagnose HIV infection. Children younger than 18 months who were ELISA positive, but PCR negative were classified as HIV exposed uninfected while older children were classified as HIV exposed uninfected if the mother was HIV infected at the time of the pregnancy but the child tested HIV negative. Caregivers who did not know their HIV status were counselled and offered HIV testing. Children or caregivers with HIV who were not accessing appropriate treatment were referred to public health facilities for further follow up and treatment of HIV.
Methods employed in the collection of respiratory specimens have been published [ 21 ]. Briefly, nasopharyngeal (NP) specimens from caregivers as well as paired NP and induced sputum (IS) specimens from children were tested by PCR for B . pertussis . The NP specimen was taken with a flocked nylon swab which was transported in a nucleic acid preservation medium (PrimeStore ® MTM, Longhorn Vaccines and Diagnostics, San Antonio, TX). The IS specimen was collected after the NP was taken from each child. All specimens were frozen at minus 80°C until they were thawed for batched molecular diagnostic testing.
A commercially validated duplex real-time PCR assay targeting the insertion sequence IS 481 for Bordetella and IS 1001 for Bordetella parapertussis (LightMix® Bordetella pertussis and parapertussis Kit, TIB MolBiol, Berlin, Germany) was used for screening all the respiratory specimens [ 22 , 23 ]. All specimens testing positive for IS 481 were further tested for the presence of insertion site hIS 1001 in order to exclude Bordetella holmesii (IS 481 +, IS 1001- , hIS 1001 +) before the diagnosis of B . pertussis infection was made [ 24 ].
Positive cases were offered macrolide treatment and the same offered to household contacts for prophylaxis.
Ethics
The study was approved by the Human Research Ethics Committee of the Faculty of Health Sciences of the University of Cape Town; Reference: 371/2011. Written informed consent was sought and received from the legal guardian for the participation of both the child and the guardian/caregiver in the study.
Analysis plan
The study aimed to investigate risk factors for pertussis as a secondary outcome and was thus not specifically powered to achieve this secondary outcome. The study sample was determined to attain a 3% precision above and below a point estimate risk of 5% for the primary outcome (prevalence of pertussis).
Categorical data are presented as percentages with 95% confidence intervals (CI). All continuous data are summarized as medians with interquartile ranges (IQR). A χ 2 test assessed strength of association between two categorical variables with a two-tailed cut-off significance set at p<0.05.
A causal model employing the current understanding of respiratory disease processes and pathogenesis of pertussis was constructed using a directed acyclic graph (DAG) to identify variables in the model required for minimal sufficient adjustment sets for estimating total independent effects of each assessed risk factor [ 25 ].
To adjust for potential confounders for each risk factor as identified by the DAG, a generalized linear Poisson regression model with robust error variance was used to estimate adjusted relative risks (aRR) and their 95% level of confidence in a multivariable analysis. For all analyses, Stata Statistical Software Release 13 (StataCorp LP, College Station, TX) was employed.
Results
Baseline characteristics of children
In total, 987 children hospitalized for acute LRTI were screened; 460 child-caregiver pairs were enrolled; Fig 1 . The median age of children was 8 (IQR 4–18) months with 41 (8.9%) younger than two months of age. The median duration of symptoms was 3 days (IQR 2–5 days); 173 (37.6%) received antibiotics prior to admission, Table 1 . Seven children received both ceftriaxone and penicillin while another one received both ceftriaxone and cotrimoxazole.
10.1371/journal.pone.0240717.g001
Fig 1
Enrolment flow diagram of study participants showing number of Bordetella pertussis positive children and caregivers.
10.1371/journal.pone.0240717.t001
Table 1 Baseline characteristics of all enrolled children and children with Bordetella pertussis .
Baseline character
All (N = 460)
B . pertussis + (n = 32)
n (%)
n (%)
Age
< 2 months old
41 (8.9)
6 (18.8)
≥ 2 months old
419 (91.1)
26 (81.20
Gender
Female
202 (43.9)
15 (46.9)
Male
258 (56.1)
17 (53.1)
Pertussis vaccine doses
0
28 (6.1)
6 (18.8)
1
57 (12.4)
6 (18.8)
2
58 (12.6)
5 (15.6)
≥ 3
308 (67)
15 (46.9)
Unknown
9 (2)
0 (0.0)
Nutritional status (WAZ) #
Normal nutrition
351 (76.3)
19 (59.4)
Mild under-nutrition
64 (13.9)
8 (25)
Moderate under-nutrition
33 (7.2)
5 (15.6)
Severe under-nutrition
12 (2.6)
0 (0.0)
HIV status
Unexposed Uninfected
349 (75.9)
19 (59.4)
Exposed Uninfected
92 (20)
10 (31.3)
Exposed Infected
19 (4.1)
3 (9.4)
Presenting symptoms
Cough
456 (99.1)
32 (100)
Apnoea
20 (4.5)
3(9.4)
Fever
288 (63.7)
14 (43.8)
Age-specific tachypnoea
286 (62.2)
19 (59.4)
Chest indrawing
376 (81.7)
28 (87.5)
Presence of a household smoker
162 (35.2)
11 (34.4)
Household use of biofuel
18 (3.9)
3 (9.4)
Pre-hospital antibiotic (n = 173) *
Penicillin
77 (44.5)
7 (21.9)
Ceftriaxone
99 (57.2)
10 (31.3)
Cotrimoxazole
4 (2.3)
0 (0.0)
Erythromycin
1 (0.4)
0 (0.0)
# Normal: WAZ > -1, Mild: ≤ -1 WAZ >-2, Moderate: ≤ -2 WAZ > -3, Severe: WAZ ≤ -3
WAZ = World Health Organization weight for age Z score.
* 8 children received more than one antibiotic
Ninety-two (20.0%) children were HIV exposed while infection was confirmed in 19 (4.1%) ( Table 1 ) of whom nine were on antiretroviral therapy, four with viral suppression. Of the 19 children, one was WHO HIV stage 1; another was stage 2 while 10 and seven were stage 3 and 4 respectively.
Most children 351(76.3%) were adequately nourished with weight for age Z score > -1. Only 60 (13.0%) of the children had never been breast-fed. The majority (n = 323, 70.2%) was breast-fed for the first four months of life and 77 (16.7%) for longer than four months.
Most children (n = 451; 98.0%) had their RTHC available and their immunization status, including the number of vaccine doses could be verified, Table 1 . Nineteen (4.2%) children were younger than six weeks and had as yet not received the first dose of pertussis vaccine. Of the 432 (95.8%) with immunization records and old enough to receive at least one vaccine dose, 369 (85.4%) had received expected doses for age.
The duration of the children’s symptoms was 3 (IQR 2–5) days. In 176 (38.3%) children, there was a history of another household member [median 1 (IQR 1–2)] with cough symptoms.
Confirmed Bordetella pertussis in children
NP specimens were obtained from all child participants. Four children were transferred or discharged out of the ward before an IS specimen could be obtained. For the remaining 456 children, IS was successfully obtained with no major adverse events, although two were later lost to container leakage.
PCR for IS 481 was positive in 17 NP specimens and 25 IS specimens. There was an overlap of positive NP and IS specimens in 10 participants giving a total of 32 (7.0%; 95% CI 4.8–9.7%) children with confirmed B . pertussis infection. B . parapertussis (IS1001+) was detected in seven (1.5%; 95% CI 0.6–3.2%) children. B . holmesii was excluded in all the positive specimens by the absence of hIS 1001 .
Caregiver baseline characteristics
All 460 primary caregivers took part in the study of whom 450 (97.8%) were mothers. For the remaining 10 children, the caregiver was the father in two (0.4%) instances, grandmother in five (1.1%) and another relative in the other three (0.7%) children. The caregiver was the mother in all 32 cases with confirmed pertussis. In 451(98.0%) of the recruited pairs, the caregiver slept in the same room as the child. The median age of the caregivers was 28 (IQR 24–33) years. In the week the child presented to hospital, 171 (37.2%) of the caregivers had respiratory symptoms. The symptoms, predominantly of an upper respiratory tract infection, were present in 10 (31.3%) of caregivers whose children had confirmed pertussis and in 161 (37.6%) whose children did not have pertussis; p = 0.466. Baseline characteristics of the caregivers are summarized in Table 2 .
10.1371/journal.pone.0240717.t002
Table 2 Caregiver characteristics by child’s Bordetella pertussis PCR status.
Baseline character
PCR negative (n = 428)
PCR positive (n = 32)
P value
n (%)
n (%)
Gender
Female
426 (99.6)
32 (100.0)
0.805
HIV status
Infected
98 (22.9)
13 (40.6)
0.024
Presenting symptoms
Cough
96 (22.4)
7 (21.9)
0.931
Runny nose
107 (25)
6 (18.8)
0.409
Wheeze
35 (8.2)
3 (9.4)
0.831
Fever
100 (23.4)
6 (18.8)
Caregiver B . pertussis NP
PCR positive
5 (1.2)
10 (31.3)
< 0.001
PCR negative
423 (98.8)
22 (68.7)
NP = nasopharyngeal swab specimen
HIV infection was present in 111 (24.1%) of caregivers of which 55 (49.5%) were on antiretroviral treatment.
Cigarette smoking was recorded in 162 (35.2%) of the households although only 33 (7.2%) of the caregivers were themselves smokers. The use of biofuels for cooking or heating was uncommon and reported in 18 (3.9%) households.
Confirmed Bordetella pertussis in caregivers
NP specimens were successfully obtained from all 460 caregivers. IS 481 was positive in 15 (3.3%; 95% CI 1.8–5.3%) of the caregivers, 10 in mothers of children with confirmed B . pertussis infection, Table 2 . All IS 481 positive specimens were negative for the hIS 1001 B . holmesii locus.
Caregivers with detected nasal B . pertussis were all mothers of enrolled children and all slept in the same bedroom with the child. No association was noted between presence of maternal symptoms and confirmed B . pertussis infection [4/171 (2.3%) symptomatic versus 11/289 (3.8%) asymptomatic; p = 0.392]. There was no difference in the duration of symptoms between caregivers with confirmed pertussis and those without: median 3 (IQR 2.5–5) days and 2 (IQR 2–5) days respectively; p = 0.513. B . pertussis was detected in 6 (5.4%) of HIV infected caregivers compared to 9 (2.6%) of those who were not; p = 0.214.
Effect of risk factors
Unadjusted and adjusted effects of factors on risk of pertussis disease in children are shown in Table 3 .
10.1371/journal.pone.0240717.t003
Table 3 Risk factors for confirme d Bordetella pertussis infection in study children.
Relative Risk (95% Confidence Interval)
Risk factor
Risk n/N (%)
Crude
Adjusted *
Age
≥ 2 months old
26/419 (6.2)
1
1
< 2 months old
6/41 (14.6)
2.36 (1.03–5.40)
2.37 (1.03–5.42)
Nutritional status
Normal
19/351 (5.4)
1
1
Mild under-nutrition
8/64 (12.5)
2.31 (1.06–5.05)
2.27 (1.01–5.09)
Moderate under-nutrition
5/33 (15.2)
2.80 (1.12–7.02)
2.70 (1.13–6.45)
Severe under-nutrition
0/12 (0.0)
NA
NA
HIV status
Unexposed uninfected
19/349 (5.4)
1
1
Exposed uninfected
10/92 (10.9)
2.00 (0.96–4.15)
3.53 (1.04–12.01)
Infected
3/19 (15.8)
2.90 (0.94–8.96)
4.35 (1.24–15.29)
Pertussis vaccine doses
None
5/28 (17.9)
1
1
One
4/57 (7.0)
0.39 (0.11–1.35)
0.39 (0.11–1.33)
Two
5/58 (6.9)
0.39 (0.11–1.33)
0.33 (0.09–1.19)
Three and more
19/308 (6.2)
0.34 (0.14–0.86)
0.28 (0.10–0.75)
Caregiver B . pertussis
PCR negative
22/455 (4.9)
1
1
PCR positive
10/15 (66.7)
13.48 (7.84–23.21)
13.82 (7.76–24.62)
Home cigarette smoking
No home smoker
21/298 (7.0)
1
1
Home smoker
11/162 (6.8)
0.96 (0.48–1.95)
0.98 (0.49–1.99)
Bio-fuel use
No bio-fuel
29/442 (6.6)
1
1
Use of bio-fuel
3/18 (16.7)
2.54 (0.85–7.57)
2.40 (0.73–7.91)
n/N (%) = stratum specific proportion and percent.
* Multivariable models adjusted for age, sex, HIV status, socio-economic status, breast-feeding and number of household members with cough. (HIV not adjusted for in model for HIV status as a risk factor) Risk ratio 95% confidence intervals that do not cross the null value of 1 are shown in bold
Clinical features of children with and without pertussis were similar except for fever which was present in 274 (64.0%) of children without pertussis compared to 14 (43.8%) in children with pertussis; p = 0.022. LRTI cases with confirmed B . pertussis had a median age of 8 months (IQR 2–21), similar to LRTI cases without pertussis [8 months (IQR 4–18)]; p = 0.43). However, the risk of pertussis was significantly increased in young infants less than two months of age; 14.6% versus 6.2%; aRR 2.37 (95% CI 1.03–5.42).
No association was found between household air pollution or smoking, and risk of pertussis even after adjusting for potential confounders.
Both HIV exposure and HIV infection were independently associated with an increased risk of confirmed B . pertussis infection with aRR 3.53 (1.04–12.01) and 4.35(1.24–15.29) respectively. The risk of B . pertussis declined with each extra dose of pertussis vaccine independent of age, although the reduction only became significant after completion of the 3-dose primary vaccine schedule; aRR 0.28 (95% CI 0.10–0.75).
Mild and moderate under-nutrition were also associated with an increased risk of pertussis in the adjusted model, however no cases occurred in severely under-nourished children, Table 3 .
Detection of maternal nasal B . pertussis was most strongly associated with an increased risk of pertussis in the children with aRR 13.82(7.76–24.62). HIV infected caregivers were more likely to have children with confirmed pertussis infection with 13/111 (11.7%) compared to 19/349 (5.4%) in HIV negative caregivers; p = 0.024.
Outcome
Supplemental oxygen was required by 12 (37.5%) children out of 32 with confirmed B . pertussis and 102 (23.8%) of the 428 who were negative for the bacterium; p = 0.084. Similarly, there were 3 (9.4%) children requiring high care or critical care admission in children with confirmed compared to 11 (2.6%) in the negative group; p = 0.66. The length of hospital stay was a median of 2 (IQR 1–4.5) days and 2 (IQR 1–4) days in B . pertussis positive and negative groups, respectively; p = 0.418.
Discussion
This study reports important, novel findings of significant increased risk of pertussis in children exposed to HIV in utero and in children with HIV infection as well as in children with poor nutritional status. In addition, the highest risk of pertussis-associated LRTI in hospitalized African children was in those whose mothers had B . pertussis detected in nasopharyngeal specimens, with more than 13 fold increased risk. This study also confirms known factors, namely, incomplete primary vaccination and early infancy as important risks for pertussis in an LMIC setting.
Sub-Saharan Africa, where this study was conducted, carries a high burden of HIV, including a large number of infected or exposed children. In the current study, after adjusting for potential confounding, HIV-infected children had a four-fold increase in the risk of pertussis, possibly due to reduced vaccine effectiveness due to both poor responses to vaccination as well as low persistence of immunoglobulin following vaccination [ 8 , 9 ]. The large proportion (more than 50% each) of HIV infected children not on antiretroviral therapy and lack of virologic suppression for those on treatment may have increased the risk in this population. A Nigerian study showed a 20-fold risk of pertussis in adolescents not yet initiated on antiretroviral therapy [ 26 ]. Other recent studies have reported an increased risk of pertussis in HIV infected individuals [ 27 – 31 ]. The quality and duration of immunity to pertussis in HIV infected children once they are started on antiretroviral therapy is uncertain [ 32 ]. The small number of HIV-infected children in our study made it impossible for us to investigate these aspects.
HIV-exposed, but uninfected, children are increasingly emerging as a group more susceptible to developing disease compared to unexposed children, due to successful implementation of prevention of mother to child transmission strategies with a reduction in vertically transmitted HIV [ 33 ]. This study identifies HIV exposure in utero as a significant important risk factor for pertussis, consistent with other reports that suggested an increased risk in infants, even if the findings were not statistically significant [ 28 – 31 , 34 ]. In our study a quarter of the mothers were HIV infected. The increased risk in HIV exposed uninfected children seems related to reduced immunoglobulin levels passively transmitted from the mother, increased exposure to pertussis in a HIV-household as well as possible impaired responses to vaccination that are not yet clearly understood [ 10 , 11 ].
Since the current data were collected, regimens for antiretroviral therapy (including PMTCT) have continued to evolve and improve [ 35 ]. These developments will hopefully serve to reduce the risks associated with HIV status observed in this study.
The high risk of pertussis-associated LRTI in children whose mothers had nasopharyngeal B . pertussis is consistent with studies showing that most infants acquire pertussis from an older sibling or parent. Consequently, attempts to protect young infants have advocated cocooning, which involves vaccinating household members, as well as antenatal and postnatal vaccination of mothers of neonates. Whereas cocooning does not seem cost-effective, antenatal vaccination of mothers has shown promising protection for infants with no added risk to either the mother or the pregnancy [ 36 – 40 ]. In our study, the risk of pertussis may be partially explained by the high proportion of HIV infected caregivers who exhibited a higher risk for nasopharyngeal carriage compared to HIV uninfected caregivers (5.4% vs 2.6%) although these findings were not statistically significant, most likely due to small numbers. Due to lack of data, we could not explore the impact of maternal HIV control on the risk of pertussis on both children with HIV in utero exposure as well as HIV infected.
The risk of B . pertussis infection independently decreased with each extra dose of vaccine received, but as observed in other studies, statistically significant reduction was only seen with completion of at least three doses [ 41 , 42 ]. This highlights the great risk pertussis poses to children in LMIC who, according to WHO, largely receive incomplete vaccination [ 43 , 44 ]. This risk is further increased by the high incidence of endemic childhood malnutrition [ 45 ]. The South African National Department of Health has commissioned a study to update data on vaccine coverage as well as to better understand factors behind the observed decline in vaccine coverage. The results are anticipated before the end of 2020.
The study is limited by low frequencies of pertussis in some subgroups. Even when the study possessed sufficient power to demonstrate statistically significant risk, the estimated magnitude had low precision in some instances. A further limitation is that the study was done in children hospitalized with LRTI so the generalizability of the results to children with less severe illness requires further study.
Conclusions
In this study we set out to investigate risk factors for pertussis in children with severe LRTI. Our findings indicate an urgent need for interventions in LMICs to address modifiable risk factors for pertussis-associated LRTI. Such interventions should include nutritional support and immunization. Immunization programs should be strengthened to ensure high levels of coverage for children with at least three vaccine doses and include catch-up immunization for missed doses. A key consideration is to prioritize vaccination of pregnant women, particularly those who are HIV infected, as maternal infection is the greatest risk for disease in infants [ 46 ].
Supporting information
S1 File
Minimal data set for effect models.
(PDF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The HIV epidemic has substantially increased the incidence of tuberculosis, and especially smear-negative tuberculosis (SNTB), in high prevalence resource-limited settings [1] – [3] . Tuberculosis is a curable infection yet remains the leading cause of death in sub-Saharan Africa [4] – [6] , in part due to delayed diagnosis. New diagnostic tests for tuberculosis are urgently required and are under evaluation, but currently sputum microscopy is the only widely available diagnostic test for tuberculosis [7] .
C-reactive protein (CRP) is a non-specific acute phase serum protein that is elevated in HIV seropositive and seronegative patients with tuberculosis [8] – [15] . Recently, CRP has been proposed as a candidate biomarker for active infection with Mycobacterium tuberculosis [16] . Point-of-care CRP testing has been shown to be of use in the clinical evaluation of respiratory tract infections in adults [17] – [19] and in the evaluation of fever in children [20] – [22] . Additionally, an elevated point-of-care CRP has been used as an indication to initiate antibiotic therapy [18] . The diagnostic utility of CRP in smear-negative tuberculosis suspects in a high HIV prevalence setting is unknown.
We evaluated serum CRP in sputum smear-negative tuberculosis suspects in the Umgungundlovu District, KwaZulu-Natal, South Africa.
Methods
Ethics statement
The study was approved by the Biomedical Research Ethics Committees of the University of KwaZulu-Natal and the University of Cape Town, and by the KwaZulu-Natal Department of Health. All participants gave written informed consent prior to enrolment.
Study setting
Tuberculosis suspects were enrolled between June 2005 and February 2007 from primary care clinics in the uMgungundlovu District, KwaZulu-Natal, South Africa. The district is representative of the high tuberculosis and HIV burden in southern Africa: in 2007 the HIV prevalence in public sector antenatal clinics was 40.4%, and the estimated 2006 annual incidence of tuberculosis was 1,094 cases/100,000 [23] , [24] .
Study population
Consecutive adults older than 18 years who gave informed consent were eligible for inclusion if they were being worked up by the primary care clinic as TB suspects. Inclusion criteria were one or more symptoms compatible with tuberculosis for >2 weeks (cough, weight loss, loss of appetite, haemoptysis, fevers and chills, drenching night sweats, fatigue, chest pain, haemoptysis, shortness of breath, swollen lymph nodes, abdominal swelling), and two or more sputum smears negative for acid-fast bacilli (AFB) or unable to produce sputum. Exclusion criteria were: missing baseline CRP, not completing eight weeks of follow-up, Karnofsky performance score <40, tuberculous meningitis, Pneumocystis jirovecii pneumonia, more than one week of antitubercular therapy, less than three months of antiretroviral therapy, or a flouroquinolone within the past 6 months.
Study procedures
Eligible participants were assessed for evidence of SNTB using a standardized protocol that included chest radiograph in all cases, and abdominal and pericardial ultrasound if indicated. These clinical and radiological data were used to assign participants either to antitubercular therapy or observation using pre-specified criteria. Participants were assigned to antitubercular therapy if clinicians identified a focal process compatible with tuberculosis either on physical examination or radiographic imaging using modifications of previously validated case definitions [9] .
At enrolment at least two relevant specimens were taken for mycobacterial culture. Sputum induction with hypertonic saline and an ultrasonic nebuliser was performed in all participants. All specimens sent for mycobacterial culture were stained for acid-fast bacilli using fluorescent microscopy and cultured in liquid culture media (BACTEC™ MGIT™ BD, Sparks, MD). Positive cultures were identified as Mycobacterium tuberculosis using the niacin test. All laboratory investigations were performed in accordance with manufacturers' instructions at accredited laboratories by technicians registered with the Health Professions Council of South Africa. External quality assurance for sputum smears was routinely undertaken by the National Health Laboratory Service. Technicians evaluating specimens for acid-fast bacilli and positive mycobacterial cultures did not have access to CRP results. Serum CRP was measured with the Olympus AU640 (normal range 0–8 mg/L) for the initial 182 participants; and the Dade Dimension RXL (normal range 0–5 mg/L) for the subsequent 182 participants.
Participants were reviewed 2, 4 and 8 weeks after enrolment. Patients in the observation group who had a positive AFB smear or mycobacterial culture, or who clinically deteriorated during follow-up were started on antitubercular therapy and reviewed for an alternative diagnosis. Participants in the treatment group who deteriorated were reviewed for an alternative diagnosis. At each visit data were collected on weight, haemoglobin concentration, Karnofsky performance score and symptoms score (number of tuberculosis-related symptoms rated ‘much better’ or ‘resolved’ dividing by the total number of tuberculosis-related symptoms present at enrolment).
Diagnosis of tuberculosis
Participants were classified into one of three groups: i) confirmed tuberculosis (at least one clinical specimen culture-positive for Mycobacterium tuberculosis or AFB with granulomata on histology); ii) possible tuberculosis (focal disease process compatible with tuberculosis detected on clinical examination or imaging, or a wasting syndrome, and clinical response to anti-tubercular therapy); iii) no tuberculosis (culture-negative, without a focal process or wasting and with a clinically stable course over an eight-week observation period, or an alternative diagnosis made, or no response to empiric antitubercular therapy). Research clinicians had access to the CRP result only after the decision whether or not to initiate antitubercular therapy had been made. Clinical response to antitubercular therapy was defined as meeting two or more criteria at week 8: i) weight gain >5%; ii) haemoglobin increase >1.0 g/dL; iii) increase in Karnofsky score >20 (or >10 if baseline score was 80 or 90); and iv) symptom score >0.5 (i.e., at least half of the symptoms much better or resolved). We have previously evaluated these criteria [9] .
Data analysis
Data were entered into a Microsoft Access database and analysed using Analyse-it for Microsoft Excel (version 2.11). In order to correct for the differing normal ranges CRP results were expressed as the quotient of the absolute CRP value divided by the upper limit of normal for the assay. In order to determine the most useful cut-off for the CRP quotient the performance characteristics were calculated for measurements that were either normal or raised (CRP quotient >1× upper limit of normal [ULN]), and either greater than or less than 2.5×, 5× and 10× ULN. Categorical data were compared between different groups using the Pearson chi-squared test, and odds ratio using Fisher's exact test. Accuracy of probability estimates were determined by calculating ninety five percent confidence intervals (95% CI). Distribution of continuous data was determined using the Shapiro-Wilk test, difference in medians between two groups using the Mann-Whitney test, and difference in medians between more than two groups using the Kruskal-Wallis test.
Results
Cohort characteristics
Five hundred and four patients were screened, and 364 participants evaluated. Participant flow and final diagnosis is summarized in Figure 1 . The clinical characteristics of the participants are shown in Table 1 . Two hundred (55%) participants were HIV seropositive, 39 (11.%) HIV seronegative, and 125 (34%) declined an HIV test. The median CD4 T-lymphocyte count was 143 cells/µL (interquartile range [IQR] 78; 248) in the 136 HIV seropositive participants who had the test during the eight week follow-up period.
10.1371/journal.pone.0015248.g001
Figure 1
Participant flow chart and final diagnosis.
* Reasons for exclusion (n) Not able to attend for regular review - determined during screening visit (28) No active symptoms (17) Alternative medical diagnosis made at screening (14) Karnofsky Score <40 (5) Pneumocystis pneumonia (4) Informed consent not obtained (3) Sputum smear positive (3) Already on antitubercular therapy (3) Other (6).
10.1371/journal.pone.0015248.t001 Table 1
Participant characteristics (n = 364).
Characteristic
Age, median (IQR), years
34.4 (29.3–42.1)
Men, n (%)
203 (57.5)
HIV seropositive, n (%) [of 239 who tested]
200 (83.7)
Prior tuberculosis treatment, n (%)
98 (26.9)
Cough >2 weeks, n (%)
341 (93.7)
Weight loss, n (%)
277 (76.1)
Anorexia, n (%)
256 (70.3)
Drenching night sweats, n (%)
248 (68.1)
Fatigue, n (%)
232 (63.7)
Fever and chills, n (%)
219 (60.2)
Chest pain, n (%)
207 (56.9)
Dyspnoea, n (%)
145 (39.8)
Haemoptysis, n (%)
40 (11.0)
Lymph node swelling, n (%)
34 (9.3)
Abdominal swelling, n (%)
4 (1.1)
Received antibiotic within 2 weeks of enrolment, n (%)
261 (71.7)
Karnofsky Performance Score, median (IQR)
70 (60–80)
Respiratory rate, median (IQR) breaths/minute
24 (20–30)
Temperature, median (IQR) °C
36.8 (36.0–37.5)
Resting heart rate, median (IQR)
100 (84–120)
Weight, median (IQR) kg
57.5 (50.6–64.8)
Body mass index, median (IQR) kg/m 2
21.2 (18.9–23.6)
Haemoglobin, mean (SD) g/dL
11.1 (2.3)
Tuberculosis diagnosis
One hundred and thirty five (37%) participants were classified as having confirmed tuberculosis (132 on culture and three on histology), and 114 (31%) had possible tuberculosis (including two who were originally assigned to the observation arm, deteriorated, and subsequently responded to antitubercular therapy). One hundred and fifteen participants (32%) were classified as not having tuberculosis: 68 remained stable without antitubercular therapy for eight weeks, 26 had no response to antitubercular therapy criteria, and 3 deteriorated on antitubercular therapy and another diagnosis was made. Of the 249 participants diagnosed with confirmed or possible tuberculosis 112 (45%) had pulmonary disease, 63 (25%) had both pulmonary and extrapulmonary disease and 74 (30%) had extrapulmonary disease only. Five (1.4%) of the participants were on antiretroviral therapy for >3 months (two with confirmed tuberculosis and three without tuberculosis).
CRP quotient by tuberculosis diagnosis
The median CRP quotient in the confirmed tuberculosis group was 15.4 (IQR 7.2; 23.3), 5.8 (IQR 1.4; 16.0) in the group with possible tuberculosis, and 0.7 (IQR 0.2; 2.2) in the group without tuberculosis ( p <0.0001). Three (2%), 23 (20%) and 68 (59%) participants had a normal CRP in these three groups respectively ( p <0.0001).
Of the 135 participants diagnosed with confirmed tuberculosis 116 (86%) were diagnosed with a focal disease process consistent with tuberculosis: 36 had pulmonary disease; 33 had both pulmonary and extrapulmonary disease; and 47 had extrapulmonary disease. The median CRP quotient in these three groups was 12.4 (IQR 7.1–17.4), 18.2 (IQR 12.6–27.3) and 21.6 (IQR 9.3–27.6) ( p = 0.01). The 19 participants with confirmed tuberculosis that was not diagnosed during the initial clinical evaluation had a median CRP quotient of 7.0 (IQR 2.4–16.2) compared to the median CRP quotient of 15.9 (IQR 9.4–24.5) in the 116 participants diagnosed with a focal process ( p = 0.0003).
Receiver operating curve characteristics and sensitivity/specificity decision plots for the group with confirmed tuberculosis and for the group with confirmed and possible tuberculosis are shown in Figure 2 . Performance of the screening CRP quotient at various levels above the upper limit of normal are shown in Tables 2 – 4 . The CRP quotient above the upper limit of normal had sensitivity 0.98 (95% CI 0.94; 0.99), specificity 0.59 (95% CI 0.50; 0.68), positive predictive value 0.74 (95% CI 0.67; 0.80), negative predictive value 0.96 (95% CI 0.88; 0.99), and diagnostic odds ratio 63.7 (95% CI 19.1; 212.0) in the confirmed tuberculosis group compared with the group without tuberculosis ( Table 2 ). The positive predictive value was highest in the analyses that combined the confirmed and possible tuberculosis group ( Table 3 ). Sensitivity and negative predictive value were maintained at the expense of the specificity and positive predictive value when the confirmed tuberculosis group was compared with the combined possible tuberculosis and no tuberculosis groups ( Table 4 ). Higher CRP quotients improved specificity at the expense of sensitivity.
10.1371/journal.pone.0015248.g002
Figure 2
Receiver operating curves and sensitivity/specificity decision plots.
A and B comparing participants with confirmed tuberculosis vs. those with no tuberculosis (n = 250); C and D combined confirmed and possible tuberculosis vs. those with no tuberculosis (n = 364); and E and F confirmed tuberculosis vs. those with possible tuberculosis and no tuberculosis.
10.1371/journal.pone.0015248.t002 Table 2
Performance of CRP as a screening test: Confirmed tuberculosis vs. those with no tuberculosis (n = 250).
CRP quotient
Sensitivity
Specificity
Positive likelihood ratio
Negative likelihood ratio
Diagnostic odds ratio
Positive predictive value
Negative predictive value
(95% CI)
(95% CI)
(95% CI)
(95% CI)
(95% CI)
(95% CI)
(95% CI)
>1×ULN
0.98
0.59
2.39
0.04
63.7
0.74
0.96
(0.94; 0.99)
(0.50; 0.68)
(2.29; 2.49)
(0.02; 0.07)
(19.1; 212.0)
(0.67; 0.80)
(0.88; 0.99)
≥2.5×ULN
0.95
0.77
4.19
0.07
62.6
0.83
0.93
(0.90; 0.98)
(0.69; 0.85)
(3.89; 4.52)
(0.05; 0.09)
(26.0; 150.5)
(0.76; 0.89)
(0.86; 0.97)
≥5×ULN
0.88
0.85
5.96
0.14
42.9
0.87
0.86
(0.81; 0.93)
(0.77; 0.91)
(5.30; 6.71)
(0.12; 0.16)
(20.6; 89.2)
(0.81; 0.92)
(0.78; 0.92)
≥10×ULN
0.69
0.93
9.90
0.33
29.6
0.92
0.72
(0.64; 0.77)
(0.87; 0.97)
(7.68; 12.77)
(0.32; 0.35)
(13.2; 66.3)
(0.85; 0.96)
(0.64; 0.79)
10.1371/journal.pone.0015248.t003 Table 3
Performance of CRP as a screening test: Combined group of confirmed or possible tuberculosis vs. those with no tuberculosis (n = 364).
CRP quotient
Sensitivity
Specificity
Positive likelihood ratio
Negative likelihood ratio
Diagnostic odds ratio
Positive predictive value
Negative predictive value
(95% CI)
(95% CI)
(95% CI)
(95% CI)
(95% CI)
(95% CI)
(95% CI)
>1×ULN
0.90
0.59
2.19
0.18
12.4
0.83
0.72
(0.85;0.93)
(0.50; 0.68)
(2.1; 2.3)
(0.16; 0.19)
(7.1; 21.5)
(0.77; 0.87)
(0.62; 0.81)
≥2.5×ULN
0.81
0.77
3.59
0.24
14.7
0.89
0.65
(0.76; 0.86)
(0.69; 0.85)
(3.32; 3.88)
(0.23; 0.25)
(8.6; 25.2)
(0.84; 0.92)
(0.57; 0.73)
≥5×ULN
0.71
0.85
4.84
0.33
14.4
0.91
0.58
(0.65; 0.77)
(0.77; 0.91)
(4.29; 5.45)
(0.32; 0.34)
(8.1; 25.9)
(0.86; 0.95)
(0.50; 0.65)
≥10 u ULN
0.53
0.93
7.56
0.51
14.8
0.94
0.48
(0.46; 0.59)
(0.87; 0.97)
(5.84; 9.79)
(0.50; 0.52)
(6.9; 31.7)
(0.89; 0.97)
(0.41; 0.54)
10.1371/journal.pone.0015248.t004 Table 4
Performance of CRP as a screening test: Confirmed tuberculosis vs. those with possible tuberculosis or without tuberculosis (n = 364).
CRP quotient
Sensitivity
Specificity
Positive likelihood ratio
Negative likelihood ratio
Diagnostic odds ratio
Positive predictive value
Negative predictive value
(95% CI)
(95% CI)
(95% CI)
(95% CI)
(95% CI)
(95% CI)
(95% CI)
>1×ULN
0.97
0.4
1.62
0.07
22.0
0.49
0.96
(0.92; 0.99)
(0.34; 0.47)
(1.45; 1.81)
(0.03; 0.20)
(7.8; 61.6)
(0.42; 0.55)
(0.90; 0.99)
≥2.5×ULN
0.95
0.56
2.17
0.09
23.6
0.56
0.95
(0.90; 0.98)
(0.50; 0.63)
(1.86; 2.53)
(0.04; 0.19)
(10.5; 52.7)
(0.49; 0.63)
(0.90; 0.98)
≥5 ×ULN
0.88
0.67
2.66
0.18
14.9
0.61
0.90
(0.81; 0.93)
(0.60; 0.73)
(2.19; 3.22)
(0.11; 0.28)
(8.3: 27.0)
(0.54; 0.68)
(0.85; 0.94)
≥10×ULN
0.69
0.80
3.43
0.39
8.8
0.67
0.81
(0.60; 0.77)
(0.74; 0.85)
(2.59; 4.55)
(0.30; 0.50)
(5.4; 14.3)
(0.58; 0.75)
(0.76; 0.86)
The two CRP assays used showed similar performance characteristics in subgroup analyses of the confirmed tuberculosis group vs. those without confirmed tuberculosis ( Figure S1 ).
The median CRP quotient was not affected by previous treatment for tuberculosis in the 135 participants with confirmed tuberculosis (16.5 [IQR 11.4; 21.2] and 14.2 [IQR 7.1; 23.4] in the participants with and without prior tuberculosis respectively, p = 0.58). Prescription of an antibiotic before enrolment did not significantly alter the median CRP quotient in subgroup analysis of participants with confirmed, possible and without tuberculosis ( p ≥0.57, data not shown).
The performance characteristics of the CRP quotient were similar when the 54 participants without week 8 data were included (15 [28%] with confirmed tuberculosis) in an intention to treat analysis ( Figure S2 ): for those with confirmed tuberculosis versus those without confirmed tuberculosis the area under the receiver operating curve was 0.81 (95% CI 0.77; 0.85) p <0.0001, and sensitivity/specificity decision plot intercept rate 0.75 and CRP quotient 7.4.
Effect of HIV status on CRP quotient
HIV status did not significantly influence the median CRP in the three diagnostic categories ( Table 5 ). Fifty five HIV seropositive patients were diagnosed with confirmed tuberculosis and had a CD4 T-lymphocyte count result. Forty had a CD4 count of <200 cells/µL and median CRP quotient of 19.9 (IQR 12.8; 25.1), and 15 had a CD4 count ≥200 cells/µL and median CRP quotient of 8.2 (IQR 6.5; 15.1) ( p = 0.001). Performance of CRP quotient by HIV serostatus are shown in Table S1 and S2 .
10.1371/journal.pone.0015248.t005 Table 5
Influence of HIV status on screening CRP.
HIV seropositive
Median CRP quotient (IQR)
Tuberculosis diagnosis
n (%)
HIV seropositive(n = 200)
HIV seronegative(n = 39)
p for CRP comparison
Confirmed tuberculosis
79 (89)
17.7
11.4
0.17
(n = 89)
(7.7; 23.7)
(3.9; 20.8)
Possible tuberculosis
61 (82)
4.4
6.2
0.69
(n = 74)
(1.4; 11.3)
(1.1; 20.3)
No tuberculosis
60 (79)
0.7
1.0
0.79
(n = 76)
(0.3; 2.0)
(0.2; 1.7)
p for trend
0.22
<0.0001
0.0005
Discussion
We have shown that CRP has high sensitivity for the diagnosis of tuberculosis in this well-defined ambulatory cohort of SNTB suspects. Importantly, the CRP quotient was not affected by HIV status, and in HIV seropositive participants with confirmed tuberculosis was significantly higher in those with advanced HIV disease. Overall, higher CRP quotients were associated with increasing specificity and positive predictive value at the cost of lower sensitivity and negative predictive values. However, the high negative predictive value of any elevated CRP (0.96 for participants with confirmed tuberculosis and 0.72 for all with tuberculosis) suggests that a normal CRP would be useful for ruling out tuberculosis.
Elevated CRP levels have been demonstrated in HIV infected patients with confirmed tuberculosis [8] – [11] and in patients with a clinical syndrome compatible with either pneumonia or tuberculosis [11] – [14] . Breen et al found that an elevated CRP detected 85% of proven tuberculosis cases in London, with a 29% HIV seroprevalence in the cohort [15] . This study develops these findings by focusing on the performance of CRP in ambulatory SNTB suspects in a resource-limited high HIV prevalence setting in sub-Saharan Africa. Recent data from Europe and the United Kingdom have demonstrated the feasibility and acceptability of using a point-of-care CRP assay in primary care settings. [18] – [22] . CRP may have a role in algorithms for the evaluation of SNTB in sub-Saharan Africa where the need for novel tuberculosis diagnostics is greatest. Screening CRP may be most effectively used in early in diagnostic algorithms in primary care settings to triage smear-negative tuberculosis suspects for onward referral for chest radiograph and clinician review.
This study has several limitations. Participant enrolment was from a single site in an area with exceptionally high prevalence of HIV and incidence of tuberculosis. The utility of CRP in the diagnosis of SNTB should be evaluated in other primary care settings in sub-Saharan Africa where the prevalence of tuberculosis and/or HIV is lower. Secondly, the research clinicians had access to the screening CRP result during the follow-up visits and this may have affected evaluation of the subjective components of the response to antitubercular therapy (Karnofsky Performance Score and symptom ratio). Third, almost all participants were not taking antiretroviral therapy at the time of the study. CRP is elevated in patients experiencing the immune reconstitution inflammatory syndrome after initiating antiretroviral therapy and the findings from this study should not be extrapolated to these patients [25] . Finally, participants in this study were adult, had a Karnofsky Score of ≥40; thus our findings cannot be extrapolated to hospitalized patients or children.
In conclusion our data indicate that CRP could have a role in sub-Saharan Africa in the evaluation of tuberculosis suspects who are sputum smear-negative. CRP has the potential to contribute to future algorithms for the primary care diagnosis of SNTB in high HIV prevalence settings in combination with clinical evaluation. The utility of point-of-care CRP in screening for SNTB should be evaluated in primary care settings.
Supporting Information
Figure S1
Receiver operating curves and sensitivity/specificity curves for confirmed TB vs. possible TB and not TB: comparison between the two CRP assays. A Initial 182 participants (confirmed TB n = 43) using the Olympus AU640 (normal range 0–8 mg/L); B Subsequent 182 participants (confirmed TB n = 92) using the Dade Dimension RXL (normal range 0–5 mg/L).
(TIF)
Figure S2
Receiver operating curves and sensitivity/specificity decision plots: intention to treat analysis (150 confirmed tuberculosis; 163 clinically diagnosed tuberculosis; 105 observed). A and B comparing participants with confirmed tuberculosis vs. those with no tuberculosis (n = 255); C and D combined confirmed and possible tuberculosis vs. those with no tuberculosis (n = 418); and E and F confirmed tuberculosis vs. those with possible tuberculosis and no tuberculosis (n = 418).
(TIF)
Table S1
Sensitivity and specificity for the comparison confirmed TB vs. possible TB and not TB in HIV seropositive participants (n = 200).
(DOC)
Table S2
Sensitivity and specificity for the comparison confirmed TB vs. possible TB and not TB in HIV seronegative participants (n = 39).
(DOC)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Non-alcoholic fatty liver disease (NAFLD) and its progressive form, non-alcoholic steatohepatitis (NASH) characterized by steatosis and necroinflammation, with or without centrilobular fibrosis, have emerged as prevalent problems in Western populations. The main risk factors for developing NAFLD and NASH are components of the metabolic syndrome; increased weight, insulin resistance, hypertension, and hyperlipidemia. Liver biopsy remains the gold standard for diagnosing and assessing the degree of injury in NASH [ 1 , 2 ]. Although exercise and weight loss, surgical bariatric procedures, and pharmacological interventions, including insulin sensitizing agents and Vitamin E, have shown promise in treating people with NAFLD or NASH, there is no approved therapy for these disorders. There is ongoing public health concern since it is estimated that up to 30% of the US population has NAFLD, and another one third has NASH.
Silymarin, an extract of milk thistle ( Silybum marianum ), is the botanical treatment most commonly used for liver disorders in the United States, owing to its purported hepatoprotective properties [ 3 ]. Through its antioxidant properties, silymarin may mitigate lipid peroxidation and the production of free radical injury, a suspected mechanism of liver injury in NASH. Studies evaluating the use of silymarin in this capacity have found it to be effective in scavenging hydroxyl radicals, preventing the release of TNF alfa, and restoring normal levels of superoxide dismutase, a precursor of glutathione [ 4 , 5 , 6 , 7 ].
Recently, Kheong and colleagues reported that Silymarin at a single dose was safe and well tolerated in a Malayasian NASH population, but did not result in a statistically significant reduction in the NAS compared with placebo.[ 8 ] In view of the limited data available on dosing and pharmacokinetics of silymarin, an initial dose-ranging study [ 9 ] was performed to identify adequate silymarin doses to be tested in proof of concept studies, including the current trial. Here, we aimed to confirm the safety and preliminarly assess the efficacy of silymarin in patients with biopsy confirmed NASH without cirrhosis.
Materials and methods
Trial design
The “ S il y marin in N ASH and C H epatitis (SyNCH)” study was a randomized, double-blind, placebo controlled phase II multicenter trial to evaluate the safety and explore the efficacy of 2 doses of a standardized form of silymarin (Legalon ® , Rottapharm|Madaus, Mylan) compared with placebo. Although the SyNCH study comprised two patient poulations, those with Hepatitis C [ 10 ] and NASH, the current study pertains only to the latter population. Diabetic and non-diabetic patients with NASH and without cirrhosis were randomized to 3 treatment groups for up to 48-50-weeks treatment duration. Enrollment began in May 2008, and was completed in August 2011, with follow-up completed in November 2012. The trial was initially funded as a cooperative agreement award between the National Center for Complementary and Alternative Medicine and the National Institutes for Diabetes, Digestive and Kidney Diseases. However, in May 2009, funding of the study was transferred to the sponsor, Rottapharm|Madaus, as an investigator-initiated clinical trial.
Participants
Patients over 18 years of age with AST or ALT > 40 IU/L within one year of screening and at least once during a 30-day screening period, and with suspected NAFLD were eligible for the study. At screening, patients were counseled to follow a healthy diet and lifestyle. Dietary recommendations included a decrease in saturated fats as well as total fats to <30% of total calories and macronutrient distribution of 45 to 55% carbohydrate, 25 to 35% fat and 15 to 20% protein. Patients were provided with dietary counseling to maintain glycemic control as well as to maintain a target weight/BMI that reflects no more than a +/- 10% change of body weight. Liver biopsy within 12 months of randomization confirming NASH was required for entry; the absence of cirrhosis and a NAFLD Activity Score (NAS) of 4 or greater on the baseline biopsy as read by a site pathologist qualified patients for the study. Patients meeting study entry criteria were stratified by the presence or absence of diabetes.
Patients were excluded if they had evidence for other chronic liver diseases or decompensation, a history of immunologically mediated liver disease or other severe medical illnesses, if they refused to adhere to limitations on alcohol consumption (average alcohol consumption of more than 1 drink per day or more than 2 drinks on any one day over the 30 days prior to the screening period), were diabetic with any change in anti-diabetic medication during the screening period, or had poor control of their diabetes as indicated by HbA1c > 8%. Secretagogues (sulfonylureas) and metformin were not permitted, given their purported impact on NAFLD, and anti-hyperlipidemics were permitted. In addition, patients with BMI > 45 kg/m 2 were excluded, and weight must have been shown to be stable, with no more than a 10% change between the baseline liver biopsy and enrollment. Patients were ineligible if they had used other milk thistle preparations for a period of 90 consecutive days or longer between biopsy and initial screening, or within 30 days prior to screening if the liver biopsy was performed during the screening period. Patients were also excluded if they had used other antioxidants such as vitamin E, vitamin C, glutathione, alpha-tocopherol, or non-prescribed complementary alternative medications (including dietary supplements, megadose vitamins, herbal preparations, and special teas) within 30 days prior to screening. Medications known to produce a NAFLD or NASH-like histological picture (eg; methotrexate) were not permitted.
Participants were recruited at 5 clinical sites in the United States. The study was approved by the institutional review boards at Thomas Jefferson University, Beth Israel Deaconness Medical Center, University of North Carolina-Chapel Hill, University of Pennsylvania, the Brooke Army Medical Center, and at the Data Coordinating Center (DCC, University of Pittsburgh). All patients provided written informed consent. An independent Data and Safey Monitoring Board (DSMB) approved the initial protocol and regularly reviewed the study progress.
Interventions and outcomes assessment
Participants were randomly assigned by the Data Coordinating Center at the University of PIttsburgh to 1 of 3 treatment groups: Legalon ® 420 mg or 700 mg, or placebo administered three times daily for up to 48–50 weeks. Legalon ® is a proprietary milk thistle seed extract standardized to a silymarin content of 140 mg per capsule. Treatment could be extended for up to 54 weeks following randomization to ensure that the post-treatment liver biopsy was performed while the participant was still taking study drug. After completing therapy, participants were monitored for an additional 12 weeks.
The doses for this study were selected based on the results of a phase I trial [ 9 ], which established safety and pharmacokinetics across a range of doses in hepatitis C and NAFLD patients. The doses, ranging from 140 to 700 mg given three times daily for 7 days, were safe and well tolerated [ 9 ]; the highest dose in the current study (700 mg three times daily) was selected as a balance between the need to achieve the highest systemic exposure and what was thought by the investigators to be a reasonable pill burden.
The primary outcome measure for efficacy was a reduction in the NAFLD Activity Score (NAS) by at least 2 points after the 48-week treatment period. The primary efficacy Intention-To-Treat (ITT) analysis was conducted based on interpretation of the baseline and end-of-treatment biopsies by a central pathologist (EB) who was masked with respect to timing and treatment group.
Secondary outcomes included reduction in the NAS by 1 point; improvement in the fibrosis stage; changes from baseline, normalization, and reduction by 50% in the serum ALT and AST; and decrease from baseline in HOMAr (determined through the formula: (Glucose mg/dL x 0.05551) x Insulin mcUI/mL 22.5) values. The primary outcome variable for safety was the occurrence of a dose-limiting toxicity during the treatment period. The toxicity rate for each dose group was calculated as the percent of subjects with an adverse event considered to be related to the study drug and resulted in a reduction or interruption of drug dose.
Following screening and randomization visits, participants were seen at weeks 2, 4, 12, 16, 24, 32, 40 and 48–54 throughout the treatment period, and followed for 12 additional weeks after therapy discontinuation. Liver biopsies were performed at the end of treatment, while the participant was still taking study medication.
Biopsies were scored by site pathologists according to the Nonalcoholic Steatohepatitis Clinical Research Network proposal for clinical trials for activity (the Nonalcoholic fatty liver disease Activity Score, NAS) and fibrosis stage.[ 2 ] NAS consists of scores for steatosis (0–3), ballooning (0–2) and lobular inflammation (0–3). Fibrosis stages are descriptive of location: zone 3 perisinusoidal (stages 1a, 1b) or periportal (1c); both zone 3 perisinusoidal and periportal (stage 2); bridging between any vascular structure (stage 3); cirrhosis (stage 4). Diagnosis of steatohepatitis (NASH diagnosis) was based on criteria published by Brunt et al. [ 11 ] The diagnosis consists of the presence of macrovesicular steatosis, hepatocyte ballooning and lobular inflammation. Fibrosis, scored separately, is not necessary for the diagnosis.
Adverse events were ascertained at each study visit, as were complete blood counts, serum biochemistries and liver tests, and urine pregnancy tests in female participants. Adherence to study medication was assessed using a summary of missed dose information obtained from patient diaries and dose counts.
Randomization scheme
Adaptive allocation was used to minimize the imbalance among treatment arms. Participants were allocated to treatment arm within strata defined by site and diabetes status, by means of a web-based system. Participants, investigators, clinical site staff and pathologists were masked to treatment assignment.
Statistical analyses
Sample size determination
Assuming a NAS reduction of at least 2 points in 15% of participants in the placebo arm (most likely to occur due to biopsy sampling error or misclassification or because of the lifestyle change) versus an average of 47.5% of participants taking silymarin (e.g. 40% and 55% in the lower and higher dose group, respectively), 26 participants in each treatment group would provide 80% power to reject the null hypothesis at α = 0.10 using a Chi-square test with 2 degrees of freedom. Per DSMB suggestion, the last observation carried forward method was used to impute the final outcome if a follow-up liver biopsy prior to the end of treatment biopsy was available, otherwise participants without a post baseline biopsy were considered treatment failures. Thus, no over-recruitment to account for loss to follow-up was necessary.
Statistical methods
Statistical analyses for the primary efficacy assessments were carried out on the Intention-To-Treat (ITT) population, defined as all the randomized patients. Notably, upon review of all liver biopsies by the central pathologist during the efficacy analysis, a proportion of pre-randomization liver biopsies did not meet entry histological criteria. Therefore, a supplementary analysis was performed with the subgroup of participants whose biopsies met entry histological criteria per the central pathologist.
Baseline characteristics across treatment groups are presented using frequencies and percents for categorical variables, using means and standard deviations for symmetric continuous variables, and using medians and percentiles for skewed continuous variables. By randomization scheme, they were supposed to be similar across groups and hence formal comparisons across treatment groups for these variables were performed only for verification purpose with a chi-square test or its exact version for categorical measures, and with parametric (F-test) or non-parametric (Kruskal-Wallis test) ANOVA for continuous measures.
The primary and secondary efficacy binary outcomes are reported using frequencies and percentages and compared across groups using the Chi-square test or its exact version, as appropriate. Missing NAS scores for participants who dropped out of the study were imputed using the last observation carried forward (LOCF) if a post-baseline liver biopsy was available at an earlier visit. Otherwise the drop out was considered to be a treatment failure (did not meet the endpoint). Subjects with baseline or end of treatment biopsy that could not be evaluated by the central pathologist were also considered as treatment failures.
For continuous secondary efficacy outcomes, changes from baseline were reported using means and standard deviations or medians and percentiles, as appropriate, and compared among treatment groups by means of a one-way ANalysis Of Variance (ANOVA) using an F-test or Kruskal-Wallis test, as appropriate. Toxicity rate and the incidence of AEs during the study were compared among treatment groups using a Chi-square test or its exact counterpart.
SAS version 9.3 (SAS Institute Inc., Cary, NC) was used for statistical analyses.
All authors had access to the study data and had reviewed and approved the final manuscript.
Results
A total of 116 patients were screened for enrollment: 78 were randomized; 26 to the Legalon ® 420 mg treatment arm, 27 to the Legalon ® 700 mg treatment arm and 25 to the placebo treatment arm. Among the 38 patients who were not randomized, the most common reasons were an ALT or AST not greater than 40 IU/L (14 patients), liver biopsy not demonstrating features consistent with NASH without cirrhosis as determined by the site pathologist (7 patients), and withdrawal of consent (7 patients). Fig 1 illustrates patient enrollment.
10.1371/journal.pone.0221683.g001
Fig 1
Displayed is patient enrollment.
Of the 116 patients assessed for eligibility, 38 failed screening, and 78 underwent randomization (25 to Placebo, 26 to 420 mg, and 27 to 700 mg); this group comprised the intention to treat study population. Twenty-nine of the 78 randomized patients actually met histological criteria for NASH, as determined by the study pathologist (BB). Specifically, 34 biopsies showed an NAS < 4 or no NASH; 1 showed NASH with cirrhosis; and 14 biopsies were either unavailable or not evaluable. Therefore, a subgroup analysis was conducted on the 29 patients, referred to as the intended target population.
The baseline demographic, clinical, laboratory and histological (NAS) characteristics of the three treatment groups were similar in the ITT population, as shown in Table 1 . Participants had median age of 48.3 years, median BMI of 34.1 kg/m 2 , were predominantly male (58%), of white race (95%), and non-Hispanic (83%).
10.1371/journal.pone.0221683.t001
Table 1 Baseline characteristics of the study population.
Legalon ® 420 mg (N = 26)
Legalon ® 700 mg (N = 27)
Placebo (N = 25)
Total (N = 78)
Demographics
Age, years (median)
47.3 (10.8)
48.2 (11.4)
49.5 (10.9)
48.3 (10.9)
Gender
Female
13 (50%)
9 (33%)
11 (44%)
33 (42%)
Male
13 (50%)
18 (67%)
14 (56%)
45 (58%)
Race (White)
25 (96%)
27 (100%)
22 (88%)
74 (95%)
Ethnicity (Hispanic)
5 (19%)
5 (18%)
3 (12%)
13 (17%)
Diabetes (stratum)
Yes
6 (23%)
8 (30%)
7 (28%)
21 (27%)
No
20 (77%)
19 (70%)
18 (72%)
57 (73%)
Laboratory examinations and metabolic factors
Platelets (x10 3 cells/mm 3 )
246 (66)
250 (60)
229 (51)
242 (59)
ALT (IU/L)
80(66,111)
61(51,94)
65(45,108)
70(53,101)
AST (IU/L)
57(43,71)
46(36,58)
51(37,62)
52(39,63)
Alkaline Phosphatase (IU/L)
64(56,84)
67(59,77)
82(61,96)
69(59,88)
Triglycerides (mg/dL)
130(119,173)
153(101,197)
153(114,179)
150(110,189)
Cholesterol (mg/dL)
191 (42)
175 (36)
174 (38)
180 (39)
Fasting glucose (mg/dL)
99(86,106)
98(87,105)
101(93, 127)
98(87,115)
HOMAr
4.9(3.4,7.4)
4.5(2.9,9.0)
5.5(3.1,9.1)
5.0(3.0,8.4)
BMI (kg/m 2 )
35.3 (4.8)
33.5 (4.3)
33.4 (4.8)
34.1 (4.7)
NAS
Site pathologist
4.9 (1.1)
4.8 (0.9)
4.6 (1.0)
4.8 (1.0)
Central pathologist §
4.4 (1.7)
4.4 (1.7)
4.4 (1.3)
4.4 (1.6)
Alcohol use
≥ 7 alcoholic beverages/day at least once in the past 12 months
2 (8)
2 (7)
3 (12)
7 (9)
Patient reported outcomes
SF-36
Physical component
46.8 (9.4)
43.4 (9.5)
48.8 (8.3)
46.3 (9.2)
Mental component
53.6 (4.9)
51.7 (8.5)
54.1 (5.6)
53.1 (6.6)
CES-D
16.4 (5.3)
16.8 (3.6)
15.3 (5.8)
16.2 (4.9)
CLDQ
5.4 (0.9)
5.2 (0.8)
5.4 (1.0)
5.3 (0.9)
Data are n (%) for categorical variables, mean (SD) for symmetric and median(25 th percentile, 75 th percentile) for skewed continuous variables.
§ NAS by central pathologist was missing in 3, 5 and 4 patients in the Legalon ® 420 mg, Legalon ® 700 mg and placebo groups, respectively.
CES-D: Center for Epidemiologic Studies Depression questionnaire. CLDQ: Chronic Liver Disease Questionnaire.
After 48–50 weeks of treatment, 4 participants of 27 (15%) in the 700 mg dose group, 5 of 26 (19%) in the 420 mg dose group, and 3 of 25 (12%) in the placebo group reached the primary endpoint of at least 2-point reduction in the NAS (p = 0.79; Table 2 ). Thus, there were no statistically significant differences among the treatment groups in the ITT analysis. Sixty two (80%) participants completed the study; 18/26 (69%) in the Legalon ® 420 mg dose group, 22/27 (82%) in the Legalon ® 700 mg dose group and 22/25 (88%) in the placebo group. Of the 16 participants who discontinued the study, the most common reasons for withdrawal were refusal to continue and loss to follow-up.
10.1371/journal.pone.0221683.t002
Table 2 Analysis of primary and secondary efficacy outcome measures.
Legalon ® 420 mg
Legalon ® 700 mg
Placebo
P values
ITT population
(N = 26)
(N = 27)
(N = 25)
Primary endpoint
≥2 NAS point reduction
5 (19%)
4 (15%)
3 (12%)
0.79
Secondary endpoints
≥1 NAS improvement
8 (31%)
7 (26%)
6 (24%)
0.85
ALT normalized °
2 (8%)
6 (25%)
1 (5%)
0.08
AST normalized °
4 (18%)
7 (37%)
6 (35%)
0.39
HOMAr decreased
14 (54%)
13 (48%)
11 (44%)
0.88
Fibrosis stage improved
3 (12%)
7 (26%)
7 (28%)
0.30
Data are n (%).
° Percentages calculated on patients with abnormal value (>40 IU/L) at baseline: Upper panel: ALT N = 25, 24 and 21 in the Legalon ® 420 mg, Legalon ® 700 mg and placebo groups, respectively; Upper panel: AST N = 22, 19 and 17 in the Legalon ® 420 mg, Legalon ® 700 mg and placebo groups, respectively; Lower panel: ALT N = 10, 8 and 10 in the Legalon ® 420 mg, Legalon ® 700 mg and placebo groups, respectively; Lower panel: AST N = 9, 7 and 8 in the Legalon ® 420 mg, Legalon ® 700 mg and placebo groups, respectively.
The analysis for secondary endpoints in the ITT population showed no statistically significant differences with respect to NAS improvement (at least 1 point), normalization of the ALT or AST, change in the HOMAr, and improvement in fibrosis stage ( Table 2 ). A relatively higher percentage of participants in the Legalon ® 420 mg (23%) and Legalon ® 700 mg (19%) dose groups showed an improvement in steatosis than in the placebo group (16%), but the difference was not statistically significant. No statistically significant differences among treatment groups were observed for the other NAS components ( Table 3 ).
10.1371/journal.pone.0221683.t003
Table 3 Hepatic histologic scores. Intention to Treat population.
Histologic Feature
Legalon ® 420 mg (N = 26) *
Legalon ® 700 mg (N = 27) °
Placebo (N = 25) #
Before Treatment
After Treatment
Before Treatment
After Treatment
Before Treatment
After Treatment
NAS
Patients with a reduction in score of ≥2 –no./total no. (%)
5/26 (19%)
4/27 (15%)
3/25 (12%)
Patients with any improvement in score–no./total no. (%)
8/26 (31%)
7/27 (26%)
6/25 (24%)
Steatosis
Score–no. of patients
0 (<5%)
0
0
1
1
1
0
1 (5–33%)
9
7
7
9
7
8
2 (>33–66%)
9
7
8
5
9
9
3 (>66%)
6
3
7
7
4
3
Patients with any improvement in score–no./total no. (%)
6/26 (23%)
5/27 (19%)
3/25 (12%)
Hepatocyte ballooning
Score–no. of patients
0 (None)
10
6
10
11
8
7
1 (Few)
5
5
6
3
6
7
2 (Many)
8
6
7
8
7
6
Patients with any improvement in score–no./total no. (%)
4/26 (15%)
5/27 (19%)
4/25 (16%)
Lobular inflammation
Score–no. of patients
0 (no foci)
1
2
1
1
0
1
1 (<2 foci per 200x field)
11
7
10
12
8
11
2 (2–4 foci per 200x field)
8
7
7
8
11
7
3 (>4 foci per 200x field)
3
1
4
1
2
1
Patients with any improvement in score–no./total no. (%)
5/26 (19%)
6/27 (22%)
5/25 (20%)
Fibrosis
Score–no. of patients
0 (None)
2
2
4
6
5
6
1 (Perisinusoidal or periportal)
12
5
8
5
9
8
2 (Perisinusoidal and portal/periportal)
3
2
5
5
2
3
3 (Bridging fibrosis)
5
5
5
3
4
3
4 (Cirrhosis)
1
2
1
3
0
0
Patients with any improvement in score–no./total no. (%)
3/26 (12%)
7/27 (26%)
7/25 (28%)
* The pre-treatment biopsy was considered not evaluable for 2 patients in regard to steatosis, and for 3 patients in regard to hepatocyte ballooning, lobular inflammation and fibrosis. The post-treatment biopsy was considered not evaluable for 1 patient in regard to steatosis, hepatocyte ballooning and lobular inflammation, and for 2 patients in regard to fibrosis, whereas for 8 patients the post-treatment biopsy was not available.
° The pre-treatment biopsy was considered to be not evaluable for 4 patients in regard to steatosis, hepatocyte ballooning and fibrosis and for 5 patients in regard to lobular inflammation. The post-treatment biopsy was not available for 5 patients.
# The pre-treatment biopsy was considered to be not evaluable for 1 patient in regard to steatosis, hepatocyte ballooning and lobular inflammation, and for 2 patients in regard to fibrosis, whereas for 3 patients the pre-treatment biopsy was not available to the central pathologist. The post-treatment biopsy was not available for 4 patients, whereas for 1 patient it was considered to be not evaluable.
As stated previously, upon review of the biopsies by the central pathologist, a large proportion of entry biopsies (63%) did not meet histological entry criteria, despite having been scored differently by site pathologists. Specifically, 34 of the 78 participants (43.6%) had biopsies showing NAS <4 or histological diagnosis criteria for NASH that were not met; 1 participant (1.3%) had cirrhosis; and 14 participants (17.9%) had pre-treatment biopsies that were either not available to the central pathologist or considered to be not evaluable due to insufficient tissue. Thus, the remaining sample comprised 10 in the 420 mg dose group, 9 in the 700 mg dose group, and 10 in the placebo arm. There were 4 participants of 9 (44.4%) in the 700 mg dose group, 3 of 10 (30%) in the 420 mg dose group, and 1 of 10 (10%) in the placebo group who reached the primary endpoint of at least 2-point reduction in the NAS (p = 0.27; S1 Table ). Analysis of the hepatic histologic score changes showed that more participants assigned to Legalon ® 420 mg and Legalon ® 700 mg compared to placebo had an improvement in steatosis and lobular inflammation, but the improvements were not statistically significant ( S2 Table ). As far as the analyses of the other secondary endpoints in this subgroup are concerned, no meaningful changes were observed among the three treatment groups in ALT, AST, or HOMAr ( S1 Table ).
During the study, a total of 89 Adverse Events (AEs) were reported by 44 (56%) participants ( Table 4 ). Gastrointestinal complaints were the most common. The incidence of AEs did not differ significantly in the three groups (p = 0.49). Four Serious Adverse Events occurred in 5% of patients, all considered not related to the study drug administration. Overall, only 2 patients experienced an adverse event that was rated as related to the study drug and resulted in a reduction or interruption of drug dose (both in the Legalon ® 420 mg dose group), leading to the following estimates of double-blind toxicity rates: 8% in the Legalon ® 420 mg dose group and 0% in Legalon ® 700 mg and placebo groups (p = 0.21).
10.1371/journal.pone.0221683.t004
Table 4 Adverse events by treatment arms.
AE Type
#Patients with AE [n(%)]
#AE
Legalon ® 420 mg (N = 26)
Legalon ® 700 mg (N = 27)
Placebo (N = 25)
Legalon ® 420 mg (N = 26)
Legalon ® 700 mg (N = 27)
Placebo (N = 25)
AEs
17 (65.4%)
15 (55.6%)
12 (48.0%)
28
28
33
Most common classes of AEs, by body system
Gastrointestinal
5 (19.2%)
4 (14.8%)
4 (16.0%)
6
8
4
Respiratory
2 (7.7%)
3 (11.1%)
2 (8.0%)
3
3
7
Musculoskeletal
2 (7.7%)
1 (3.7%)
5 (20.0%)
2
1
5
Headache
2 (7.7%)
2 (7.4%)
2 (8.0%)
2
2
2
Cardiac
3 (11.5%)
1 (3.7%)
2 (8.0%)
3
1
2
Other
11 (42.3%)
10 (37.0%)
8 (32.0%)
12
13
13
Discussion
Building upon the purported health benefits of milk thistle extract and ongoing interest in it as a therapeutic agent [ 12 ], this trial was designed to test whether a particular formulation of milk thistle could mitigate NASH related liver injury. In addition, the trial aimed to establish the safety of this milk thistle formulation over the range of doses used in this study, which were higher than customary. At all doses, we found that Legalon ® was safe and well tolerated, with no difference in adverse events among the treatment groups. This trial showed that more participants assigned to Legalon ® groups had an improvement in steatosis and lobular inflammation compared with placebo, but failed to show a statistically significant histological improvement of NASH.
No meaningful changes were observed among the three treatment groups in ALT, AST or HOMAr. No significant changes were observed in other efficacy assessments, which included change from baseline in liver fibrosis; proportion of participants with transaminases returning to normal range or with a reduction greater than 50%; and improvement of insulin resistance.
Kheong et al [ 8 ], in their randomized placebo controlled trial of Silymarin 700 mg given three times a day for 48 weeks to NASH patients, likewise did not demonstrate a statistically significant reduction in the NAS; ≥ 30% was targeted as the primary outcome. However, the treatment arm was associated with reductions in hepatic fibrosis by histology, and liver stiffness by transient elastography. The study was similar in design and duration of treatment as the current study, although only one dose was used. The majority of patients (89%) underwent post-treatment biopsy.
The current study excluded cirrhotic patients for several reasons, and thus the findings cannot be generalized to this population. Most importantly, histology of cirrhotics with suspected NASH often does not reflect the typical features of steatosis and steatohepatitis. Therefore, the diagnosis of NASH induced cirrhosis would have been based on circumstantial evidence, such as the history of insulin resistance or obesity. Such assumptions of disease causation without histological confirmation would have allowed for potential misclassification bias. Moreover, the primary endpoint of histological improvement could not have been assessed in this population. Furthermore, patients with cirrhosis demonstrate pharmacokinetics which differ from those without cirrhosis.
A limitation of the current study that is relevant to the primary endpoint and that may also limit generalizability is the large number of randomized patients ultimately found by the central pathologist to have liver biopsies that did not meet histologic inclusion criteria. This motivated an additional analysis of those participants for whom the liver biopsies met inclusion criteria, as judged by the central pathologist (EB). However, though this supplementary analysis showed a trend in the point estimates of patients reaching the primary endpoint consistent with the efficacy hypothesis defined in the protocol, the sample of the intended target population was too small to draw an adequately powered conclusion regarding efficacy. Importantly, the reduction in the NAS score observed in the intended target population of this study in the active and control arms are in line with those observed in the active and placebo arms and of a recently published trial.[ 13 ] This point notwithstanding, the fact remains that silymarin treatment in this study showed no improvement in the NAS in the ITT population. Furthermore, the large proportion of patients who failed to meet the histological entry criteria combined with a statistically non-significant improvement in fibrosis in the placebo group indicate the need for additional clinical trials.
In conclusion, Legalon ® at the higher than customary doses tested in this study is safe and well tolerated. Additional studies are warranted, ideally using improved methods to diagnose and grade NASH. At a minimum, future trials that rely on hiostological endpoints ought to make accommodations for optimizing liver biopsy samples.
Supporting information
S1 Table
Analysis of primary and secondary efficacy outcome measures in the patients who met histological inclusion criteria.
(DOC)
S2 Table
Hepatic histologic scores in patients who met histological inclusion criteria.
(DOC)
S1 File
Study protocol.
(PDF)
S2 File
Consort 2010 checklist.
(DOC)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Acting prosocially in a competitive intergroup context can be challenging. However, soccer is a highly competitive sport in which acts of kindness still occur. Despite substantial research on prosocial behavior (e.g., [ 1 , 2 ]), it is still a puzzle what motivates individuals to offer help in highly competitive contexts, such as a soccer game. Clearly, in soccer, the potential costs for offering help can change quickly and substantially depending on whether the player is in a goal scoring position (or midfield) or whether one’s own team is in a winning (or losing) position. Thus, focusing on helping in soccer allows us to better understand the circumstances under which prosocial behavior occurs in highly competitive intergroup settings.
One of the most significant prosocial acts during a soccer game is the decision to actually stop playing to help another player who fell on the ground and may have been injured. This is because deciding to offer help usually involves a trade-off between pursuing victory and assisting someone in need: There are typically few opportunities to take the lead in the game, and forgoing such opportunities for the sake of others can be too costly for the game’s outcome. But the play is regularly and intentionally interrupted by players, for example by kicking the ball out of the playing field or by offering a hand to a player who has been knocked to the ground.
The major purpose of the present research is to examine potential differences between male and female soccer players, as well as various differences in the game itself, on the decision to stop playing to check on a fellow player who is lying on the ground. Therefore, the key questions of the present research are: Do young female and male soccer players differ in inclinations to help? And does the inclination to help depend on the score of the game, such as whether one’s team is ahead or behind in score, and whether the game’s outcome is decided or uncertain? The latter variables are discussed in terms of “stakes” and how costly it is to offer help.
Although soccer has traditionally been a male-dominated sport at a global level (the USA being the well-known exception), female responses in such dilemma situations cannot be neglected, especially because female participation in soccer is rapidly increasing. Since 2012, the number of national academies for females has doubled whereas the number of registered female soccer players has exceeded one million [ 3 ]. Also, games at various levels (from local competitions to world championships) get more and more media attention. But despite the enormous growth in popularity of female soccer, we know little about female (versus male) responses in such dilemma situations during a soccer game.
Beyond the focus on sex differences in soccer, the present research extends past research in at least two theoretically meaningful ways. First, the present research focuses on prosocial behavior in a soccer context. This topic has also received little attention, even though there is increasing attention for research on prosociality in situations characterized by an intergroup conflict–the conflict between the interests of the own group and those of another group [ 4 , 5 ]. These intergroup conflicts pose interesting dilemmas in which parochial forms of cooperation can undermine collective interest. Specifically, helping members of the own group is sometimes conflicting with the broader, collective interests involving the interests of both groups. For example, encouraging unfair behavior (e.g., supporting a linesmen from the own club to favor the own team in signaling offside) might help the own team but it undermines the overall spirit of fairness in soccer–and to win and do so in a fair manner. Second, past research on cooperation and prosocial behavior has yielded a wealth of knowledge about the role of cognitive and affective processes, such as attitudes and social norms, religiosity, empathy, and evolved mechanisms such as kinship, reciprocity, and reputation [ 6 – 8 ].
The present research complements this (often lab-based) research by examining prosocial behavior in a real world context where attitudes and social norms are embedded in a long history of one of the most intense forms of civilized intergroup conflict (soccer), where empathy may still guide behavior, and where reciprocity and especially reputation may matter. But the intriguing aspect is that reputation and norms are not always clear–“should I stop playing when I see a player from the other team falling down?” In that sense, the intergroup context of soccer provides an ecologically meaningful context of intergroup conflict where a multifaceted package of motives, cognitions, and emotions might be activated, even under high levels of uncertainty and urgency.
High versus low stakes
In their classic Good Samaritan study, Darley and Batson [ 9 ] showed that people are less likely to go out of their way to help someone slumped by the side of the road when they are in a hurry: The inclination to stop to help this person was lower under time constraints as compared to situations in which there was no time pressure (see also [ 10 ]). This finding suggests that in many daily situations in which precious resources are limited or the stakes are high, prosocial behavior tends to decrease. Seeing someone lying on the ground is a rather uncommon event in real life, but very common in soccer: Often, players stumble, fall, and sometimes get severely injured. It also concerns decision-making under uncertainty, because one needs to quickly judge whether the player just falls with low risk of injury, or falls with a high risk of injury. Using a situation similar to the original Good Samaritan study, soccer provides a context where help is frequently needed and the dilemma whether to check on a person who fell on the ground is often encountered.
In line with Darley and Batson’s work [ 9 ], other studies show that low (rather than high) stakes can increase prosocial behavior [ 11 , 12 ]. For instance, people seem to work harder for charity than for themselves when financial incentives are low, but they prefer the opposite when the incentives are raised [ 12 ]. Thus, the inclination to benefit others instead of oneself appears to be greater when the decision to act prosocially involves relatively low costs (or potential costs) for oneself.
Although there is some research on prosocial behavior in soccer and sports in general (e.g., [ 13 , 14 ]), most of the studies to date have focused on the role of personality and individual differences (for a recent review, see [ 15 ]). However, we know very little about whether helping behavior during the game can vary according to characteristics of the game itself –especially the stakes (or how psychologically costly it is to offer help). Here we approach stakes dilemmas as follows: Stakes are high (i.e., it is psychologically costly to offer help) when the player is virtually in goal scoring position, one’s team is behind in score, or the game’s outcome is uncertain (the score difference is small). In contrast, stakes are low when the player is midfield, one’s team is ahead in score, or the game’s outcome seems to be decided (the score difference is large).
Based on the above evidence, we propose that soccer players will be more inclined to check on a player lying on the ground when the stakes are low rather than high. Furthermore, aiming to replicate previous research [ 16 – 18 ], we predict that soccer players will show greater helping intentions toward teammates than to opponents. Acting prosocially toward a teammate (rather than an opponent) appears to be more habitual as people have a strong tendency to favor the own group over other groups, a tendency that is likely to be more pronounced in a competitive context [ 19 ]. This can be attributed especially to “ingroup love” (i.e., positive affect toward teammates) including strong ingroup ties (i.e., perception of similarity and bonding with teammates), which, in turn, increase prosocial behavior toward teammates [ 16 ].
Sex differences in helping
Although some studies suggest that males and females behave differently during sports (e.g., [ 18 – 22 ]) it is not entirely clear how strong, meaningful, or robust sex differences are in prosocial behavior in the context of sports. For example, we know that as compared to males, female athletes tend to show a greater tendency to adhere to fair play norms by respecting the rules of the game and behaving cooperatively during competition [ 18 ]. As compared to males, females tend to show lower tolerance toward, and tendency to engage in, antisocial behaviors during competition, such as injurious behavior, intimidation of opponents and cheating behavior [ 14 , 18 , 20 ]. Thus, females appear to show a greater inclination to follow cooperative heuristics and to avoid antisocial behavior. However, we do not know whether there are also sex differences in helping behavior. More specifically, are female soccer players more inclined to check on a fellow player lying on the ground as compared to their male counterparts?
Whether to stop or continue playing may be guided specific social value orientations, which formally are described as preferences for maximizing own outcomes, a team’s outcomes, a relative advantage over the other team’s outcomes, and a concern with equality in outcomes [ 23 , 24 ]. Interestingly, research has uncovered interesting developmental differences in social value orientations. Specifically, research by Knight and Chao has demonstrated that girls and boys do not necessarily differ in terms of individualism, but that they differ in terms of egalitarianism and competitive orientation (sometimes also called rivalry or superiority, 23). Using various decision-making tasks, girls tend to be more egalitarian in orientation than boys: Girls favor more strongly equality so that the self and other are equally well-off (“fair share”). In contrast, boys are more prone to compete with others, seeking to obtain greater outcomes than others [ 23 , 25 ].
Complementary research suggests that girls tend to show greater prosociality in a broader sense, as expressed by empathic concern, nurturance, caring for others, and tending to the needs of other people [ 26 , 27 ]. This sex difference is mainly evident in close and long-term relationships (rather than interactions with strangers and contexts that involve danger) and tends to increase with age [ 28 – 31 ]. This can be due to the fact that, as compared to boys, girls to assign greater priority to self-transcendence values (universalism, benevolence), which emphasize concern for the welfare of others [ 32 ]. These general differences in social value orientation suggest that girls might be more prone to stop playing when a player falls, so as to do the same that one hopes another player would do for you. And perhaps even more importantly, if boys are generally more competitive in orientation, the desire to win may overshadow a concern with a player who falls on the ground.
While the scientific evidence would generally suggest that women are more likely to help than men, the question is whether such differences may also be uncovered on the soccer field. There are at least three reasons why such differences may be small or virtually absent. First, sex differences in helping and prosociality have been observed but tend to be small in magnitude [ 33 , 34 ]. Second, sex differences in helping among adolescents may be small because most people in those age groups tend to be individualistic rather than prosocial in orientation [ 34 , 35 ]. And third, most importantly, the soccer context is a “strong situation” (e.g., [ 36 ]) characterized by strong impulses and norms to compete [ 37 ]. It is plausible that the effects of context are so strong that they overshadow any differences between young men and women.
Present research
This research is one of the first attempts to examine differences between young men and women on the soccer field. We used scenario methodology that we first tested in a pilot study. Although there are limitations to such a methodology, it does provide initial insight into the potential differences between young men and women in helping behavior on the soccer field. Moreover, this methodology allows us to examine the effects of contextual variations on self-reported helping. To manipulate the size of stakes in the main study, the scenario described a hypothetical situation in which one’s own team is in a winning (or losing) position and the score difference between one’s own team and the opposing team is large (or small). Furthermore, the fallen player was portrayed to be a teammate (or an opponent).
We hypothesize that the size of the stakes and the player’s sex will affect helping behavior on the soccer field. More specifically, soccer players are expected to show greater inclination to help a fallen player in low (rather than high) stakes dilemmas and, hence, when one’s own team is in a winning (rather than losing) position, or when the score difference is large (rather than small). Furthermore, we expect female (rather than male) players to show greater inclinations to help a player who has fallen on the ground. Further, we predict that players will show greater helping intentions toward teammates rather than opponents.
It needs to be noted that we focused specifically on young soccer players (aged between 9 and 19 years) in amateur soccer. We chose this group because we speculated that stakes would vary more strongly in this age (young rather than older) and level (amateur versus professional). Furthermore, this convenience sample allowed us to access comparably high numbers of male and female soccer players. Given that the present study is one of the first studies examining sex effects in soccer, we advanced no formal hypothesis as to how sex may interact with either the stakes of the game or the teammate versus opponent difference. Rather, we wanted to explore whether the stakes of the game and the differences between helping a teammate versus an opponent would be more pronounced for men than for women, or vice versa.
Pilot study
Materials and methods
Ethics statement
The studies were reviewed and approved by the Scientific and Ethical Review Board (VCWE) of the Faculty of Behavioral and Movement Sciences, VU Amsterdam. Participants provided electronic consent prior to taking the online survey (for the pilot study) or written consent prior to taking the paper-and-pencil survey (for the main study).
Participants
One hundred and seven (107) young Dutch soccer players (52 female; M age = 14.27 years, SD = 2.35 years) completed an online survey. We recruited participants from local soccer clubs in different regions of the Netherlands. Initially, we contacted the directors of 13 clubs and asked them to distribute the survey among young soccer players. From the total of 13 clubs, four agreed to participate in this research and the trainers of the clubs sent the online survey link to the soccer players via email. Participation was voluntary but teams whose participation exceeded 50% were promised a small monetary reward (there was only one team that accomplished this goal and received 25 euros for drinks after a game).
Procedure, design, and measures
All participants filled out the online survey at home. After reading the informed consent form and agreeing to participate, the young soccer players read certain scenarios and answered questions related to helping during a soccer match. Next, they provided some demographic information, were debriefed and thanked. We employed a mixed design with sex as between-participants variable and scenarios as within-participants variable. We also controlled for age as a covariate.
We measured helping by examining the inclination to stop playing to check on another player who fell on the ground during a soccer match. The manipulation aimed to provide an initial test that helping might be affected by circumstances. In an effort to create an appropriate manipulation of stakes, the scenarios in this pilot study varied such that the position of the participant was either midfield or in goal scoring position. Helping when being midfield (as compared to goal scoring position) should be a low-stakes dilemma (as compared to high-stakes dilemma). Furthermore, we manipulated the team of the fallen player, such that the player was either a teammate or an opponent.
To cover a broad range of helping behavior, we assessed two complementary expressions of helping toward the fallen player: (a) the inclination to neglect the player and continue the game, and (b) the inclination to actually stop the game and check how the player was doing. Thus, using a 5-point scale (ranging from definitely not to definitely yes ), participants rated the extent to which they were inclined to ignore the fallen player and continue playing (reverse scored), as well as the extent to which they were inclined to stop playing to see how the fallen player was doing. For a more complete and accurate evaluation of helping, we used the combined average between the two items (all α s were between 0.84 and 0.91). Higher scores indicated higher levels of the inclination to help.
Results and discussion
We performed a 2 (Position: goal scoring, midfield) by 2 (Player: teammate, opponent) by 2 (Sex: male, female) mixed model ANOVA with Position and Player as within-participant factors, Sex as between-participants factor, and centered value of Age as covariate. Data revealed a significant main effect of Position, F (1, 104) = 71.202, p < .001, η p 2 = .406, such that participants were more inclined to help when they were midfield ( M = 3.27, SD = 0.97) rather than in a goal scoring position ( M = 2.57, SD = 0.92). Furthermore, there was a significant main effect of Player, F (1, 104) = 35.682, p < .001, η p 2 = .255, such that participants were more inclined to help teammates ( M = 3.16, SD = 0.94) than opponents ( M = 2.69, SD = 0.93). Against expectation, data revealed no significant main effect of Sex on helping ( p = .476). However, we found a significant interaction between sex and position, F (1, 104) = 6.867, p = .010, η p 2 = .062. Subsequent posthoc (LSD) pairwise comparisons showed that while female ( M = 3.22, SD = 0.98) and male ( M = 3.33, SD = 0.98) players did not differ in their willingness to help in a midfield position, F (1, 104) = 0.286, p = .594, females ( M = 2.74, SD = 0.92) were more inclined to offer help than males ( M = 2.41, SD = 0.92) in a goal scoring position, F (1, 104) = 3.551, p = .062, η p 2 = .033. None of the main effect of Age ( p = .840), or other two-way (all p -values > .089) or three-way interaction (all p -values > .252) was significant. Data and syntax are available as S1 Dataset .
Overall, the pilot study suggests that stakes matter in helping on the soccer field: Young soccer players indicated greater inclinations to help when the stakes were low (i.e., when being midfield rather than in goal scoring position). Furthermore, the inclination to help was greater when the fallen player was a teammate rather than an opponent. Thus, the pilot confirmed the validity of the method. Although marginally significant, females were more likely to help than males when they were in a goal scoring position (but not in a midfield position). Considering that effect sizes for sex differences in prosocial behavior tend to be small to moderate [ 31 ], having a small sample size might explain why we found no support for a general sex difference in helping behavior. In the main study, we aimed to address this issue by recruiting a larger sample.
Main study
Aiming to increase the statistical power, in the main study, we visited four different soccer clubs in the Netherlands, anticipating that we could recruit at least 200 participants to be able to detect even small effect sizes. Furthermore, in view of the positive finding of the pilot experiment, the main study aimed to refine the manipulation of stakes and improve its precision. Here, we used the same methodology but we focused on variations in stakes that are linked to the score in the present moment (ahead versus behind in the game, small versus large score difference). More specifically, to vary the size of stakes, instead of focusing on midfield (versus goal scoring) position, the new scenarios focused on winning (versus losing) position and on large (versus small) score difference. To strengthen the importance of the decision, we amended the scenarios so that the incident with the fallen player appeared to take place five minutes before the end of the game. We also addressed the limitations of the previous online sample by visiting the players and collecting data at their local clubs.
Materials and methods
Participants
Three hundred and seventy nine (379) young Dutch soccer players completed a paper-and-pencil survey. Of this initial sample, data of 13 participants were discarded due to incomplete responses. This yielded a final sample of 366 participants (157 female, 207 male, 2 unreported; M age = 14.39 years, SD = 2.24 years, 4 unreported). As with the pilot experiment, the young soccer players were recruited in four local soccer clubs in different regions of the Netherlands. Participation was voluntary and each participant received a sports drink as compensation for completing the survey.
Procedure, design, and measures
All participants filled out the paper-and-pencil survey at their local soccer club. After reading the informed consent form, the young soccer players read several scenarios and answered items assessing helping on the soccer field. To conclude they answered some demographic questions, received compensation, and were debriefed and thanked. We again used a mixed design with sex as between-participants variable and scenarios as within-participants variable, and included players’ age as covariate.
The measure of helping was identical to that used in the pilot study with two differences related to the size of stakes: (a) position now indicated either winning or losing position, and (b) we introduced score difference as a new parameter (small versus large). The game was told to last about another 5 minutes, and the score difference was either small (2–1) or large (5–1). Also, in addition to these scores, the difference in favor of the own team (winning) or other team (losing) was highlighted by the terms “small” (e.g., in Dutch, “je staat krap voor”) or “large” (e.g., in Dutch, “je staat dik voor”) using language that seems common among young players in the context of soccer games. Similar to the pilot study, higher scores indicated higher levels of inclination to help (all α s ranged from 0.77 to 0.91).
Results
We conducted a 2 (Position: winning, losing) by 2 (Score Difference: small, large) by 2 (Player: teammate, opponent) by 2 (Sex: male, female) mixed model ANOVA with Position, Score Difference and Player as within-participant factors, Sex as between-participants factor, and centered value of Age as covariate. The analysis yielded a significant main effect of Position, F (1, 358) = 75.304, p < .001, η p 2 = .174, suggesting that participants were more inclined to help in a winning ( M = 2.66, SD = 0.93) than in a losing position ( M = 2.34, SD = 0.89). Score Difference also had a main effect on helping, F (1, 358) = 256.028, p < .001, η p 2 = .417), in the sense that participants were more inclined to help when the score difference was large ( M = 2.86, SD = 1.03) rather than small ( M = 2.14, SD = 0.86). Furthermore, there was a significant main effect of Player, F (1, 358) = 72.981, p < .001, η p 2 = .169, such that participants were more inclined to help teammates ( M = 2.65, SD = 0.91) than opponents ( M = 2.35, SD = 0.90). This suggests that the size of stakes matter in the inclination to help.
We also found significant main effects of Age, F (1, 358) = 10.464, p = .001, η p 2 = .028, and Sex, F (1, 358) = 11.860, p = .001, η p 2 = .032. Younger players were more willing to help than the elder ones, and more importantly, females showed greater helping tendencies ( M = 2.65, SD = 0.83) than males ( M = 2.35, SD = 0.83). Although the effect size is small, this finding supports the hypothesis that women are more inclined than men to help others on the soccer field.
Furthermore, results yielded a significant two-way interaction between Position and Score Difference, F (1, 358) = 35.231, p < .001, η p 2 = .090. Subsequent posthoc (LSD) pairwise comparisons showed that compared to when the score difference was low ( M winning = 2.21, SD = 0.97; M losing = 2.07, SD = 0.91), when the score difference was high participants were more willing to help in winning position ( M = 3.11, SD = 1.16), F (1, 358) = 274.564, p < .001, η p 2 = .434, than in losing position ( M = 2.62, SD = 1.14), F (1, 358) = 96.627, p < .001, η p 2 = .213. Moreover, this two-way interaction was qualified by participants’ age, F (1, 358) = 13.513, p < .001, η p 2 = .036. Specifically, the interaction between Position and Score Difference was significant for younger (-1 SD ), F (1, 358) = 46.507, p < .001, η p 2 = .115, but not older (+1 SD ), F (1, 358) = 2.666, p = .103, participants. No other interaction effect was significant (all p -values > .079). Data and syntax are available as S1 Dataset .
Overall, the most novel finding from our main study concerns the sex differences in helping on the soccer field. Furthermore, this study replicates and extends findings from the pilot study by showing that the size of stakes (related to the team’s position in the game and the score difference) can affect the inclination to help. Also, the inclination to help was greater when the fallen player was a teammate rather than an opponent. And finally, helping was most likely when the younger players were in a comfortable position of winning with a large score difference.
Discussion
The present research sheds light on the circumstances under which helping behavior on the soccer field emerges. Both the main and the pilot study demonstrated that young soccer players were more inclined to stop playing to check on a fellow player when the stakes in the game were low rather than high. Perhaps younger (amateur) players do not take the game as seriously as older (amateur) players, and therefore are more likely to stop playing. Also, beginning players may be more easily distracted by unexpected events. Both tendencies may be stronger when they are in a comfortable position of winning by big numbers. Further, the main study revealed that helping was higher when one’s own team was in a winning rather than in a losing position, or the game’s outcome seemed decided rather than uncertain. Furthermore, helping was greater toward a teammate (rather than an opponent). Importantly, the main study revealed sex differences in helping: Females were more inclined to help than males.
Our results replicate and extend the classic findings of Darley and Batson [ 9 ] by showing that the size of stakes can affect helping intentions even in a competitive team sport environment. When the ultimate goal is to win the game, the decision whether or not to forgo what is in the best interest of one’s own team to help someone in need strongly depends on how costly this decision may be. From a sports science perspective, this finding suggests that helping intentions during competition are more flexible than previously assumed, because they are not solely affected by individual differences in variables such as autonomous motivation, prosociality, or moral disengagement (e.g., [ 13 , 15 , 38 ]), but also by important features of the situation, such as those linked to the stakes of the games. Thus, this research draws attention to the importance of factors related to the game itself in helping behavior during a sports competition. Clearly, more research is needed to replicate and extend these findings in different sports contexts, which may include other team sports or even comparisons between individual and team sports to illuminate whether the own team (the ingroup) might inhibit helping another player. The present research suggests that prosocial behavior on the playing field might vary substantially depending on continuous and dynamic changes in the game situation.
Broadly speaking, our findings are consistent with predictions from gain versus loss framing and prospect theory [ 39 , 40 ]; as with other choice dilemmas, here we find that losses loom larger than gains. This asymmetry in the importance of losses in relation to gains may further explain why soccer players were more inclined to offer help when their team was winning (instead of losing). Moreover, considering the certainty effect, it appears plausible that helping intentions increased when the degree of certainty (for winning the game, especially) was higher (i.e., the score difference was large rather than small).
Perhaps the most novel finding of the main study was a sex difference in helping. Extending prior research [ 15 ], our results indicate that female soccer players were more inclined to help as compared to their male counterparts. Although past studies have shown that females, more than males, tend to respect rules and avoid antisocial behavior in sports (e.g., [ 41 ]), here we provide support to the notion that there are sex differences in helping. In the introduction, we highlighted three general arguments why sex differences among young players in the context of soccer are likely to be small–past research has shown modest effect sizes, the young age, the strong situational context of soccer. In other words, although differences in empathy (and caring) between men and women, even as soccer players, are often stereotyped as being strong, but the scientific evidence is somewhat less strong.
How do we explain the sex difference? One possible explanation is empathy: As compared to males, females tend to express greater empathic concern and sensitivity to distress in others and this is evident in both sports and non-sports contexts (e.g., [ 41 – 43 ]). This concern for the well-being of fellow players could be even greater among females (as compared to males) because the risks for them appear to be higher: For instance, females appear less likely to engage in deceptive falling (diving) and more likely to sustain serious injuries during sports [ 44 , 45 ]. We should also acknowledge that the differences in empathy are somewhat overestimated in the present research, because our findings are based on scenario-methodology. Indeed, it is possible that this “explicit measurement” is to some degree affected by sex-related stereotypes, along with norms for how to behave, even on the soccer field [ 31 , 31 , 46 ].
As a second possibility, our findings may also be explained by the notion that, girls value equality and fairness more strongly than boys do, who tend to be more strongly orientated to rivalry and competition [ 23 , 25 ]. Thus, our finding provides support to the idea that there is a distinct female psychology that accounts for increased helping behavior at the expense of oneself; it is argued that such female psychology has evolved because it promotes fitness interests and is possibly enhanced by gender socialization [ 28 , 47 ]. Indeed, this reasoning is consistent with the view that the gender-roles of men and women are different, even for “strong” and specific situations such as competitive ingroup-outgroup games such as amateur soccer.
Needless to say, the present findings present some of the first evidence for sex differences in helping in sports. The results should thus be interpreted with caution. One possible explanation for the absence of main effects for sex in the pilot study is that the sample size was relatively small compared to that of the main study. Future replication studies with high statistical power are required to more firmly establish the link between sex and helping in soccer (and sports in general, team sports as well as individual sports).
It needs to be noted that the sex differences in helping behavior might be specifically observed in young soccer players but less so (or not at all) in professional players. Considering the heightened participation of females in soccer and the constantly increasing professionalism of female soccer [ 3 ], it is likely that professional female and male soccer players show comparable levels of helping as the stakes in top competitive teams are, by default, high. More research involving both amateur and professional players is required to generalize or identify boundary conditions for the present finding.
It is important to underline that the observed sex differences in helping were relatively small in magnitude. Furthermore, next to the sex differences, there were certain sex similarities that cannot be overlooked. For instance, male and female players responded similarly to the manipulation of stakes: Men and women were equally prone to show heightened help when being in a winning (versus losing) position. Also, both men and women tended to show heightened help when differences in score were large (versus small). This suggests that the two sexes are equally affected by the dynamic game situation, as they both tend to adjust their prosocial inclinations in response to the size of stakes. One potential exception to this “rule” is that men may be more likely than women to continue playing–and not stop to see what happened–when being closer to scoring. Perhaps at some critical and specific moments, when the stakes in soccer are very high (including personal stakes of scoring himself or herself), feelings of empathy may be more likely to be reduced in men than in women. This intriguing issue clearly deserves future research to provide insight into the robustness and generality across different situations on the field (e.g., other critical situations), types of team sport (e.g., volleyball, field hockey), and types of player (e.g., adult players).
Limitations and future directions
One limitation of the present research is that the findings are based on self-report measures. It is not argued that self-report intentions to help on the soccer field will always result in real-world helping behavior during the game. However, when taking the exploratory nature of this research into account, this work provides initial evidence that the size of stakes and the players’ sex can affect prosocial intentions during the game. Furthermore, the study design allowed us to a priori circumvent intervening variables that could potentially affect the results (e.g., inability to notice the fallen player). For instance, in high-stakes situations (e.g., when being in goal-scoring position), it is likely that players experience such elevated levels of adrenaline and excitement that they might not even notice the fallen player lying on the ground. The present studies allowed us to rule out such spurious relationships and arbitrary “noise” in the data. Nevertheless, future observational experiments could help confirm the present findings.
Second, the present research focused on a specific form of prosocial behavior during a soccer game and, therefore, the effects cannot yet be generalized to all forms of prosocial behavior in a sports competition. Whether to help a fallen player constitutes decision-making under uncertainty. It could be that a fall entails risk, but it is perhaps more likely that most falls are relatively free of any risk of injuries. Also, although females expressed greater inclinations to stop playing to help a fallen fellow player, it is likely that males express greater helping intentions in other types of prosocial dilemmas. This is because males, more than females, tend to engage in helping behavior that is heroic and chivalrous (rather than nurturant and caring, see [ 29 ]). Further research is required to evaluate the sex effect on acts of helping that are heroic and involve physical risks versus acts of helping that are nurturant in a sports competition (e.g., saving a fellow player’s life versus helping a fellow player get up).
Third, the two studies do not provide information on potential mechanisms underlying the effects of stakes and sex on helping. For example, although empathy is a likely explanation of the effects of sex [ 48 , 49 ], it is yet to be demonstrated whether the players were inclined to help because of heightened understanding of another person’s emotions and concern about their welfare. Furthermore, there could be alternative mechanisms driving the effects, such as expectations for rewards by one’s teammates or fear of sanctions. More specifically, helping a fellow player when the stakes are low might help obtain a positive social image and gain a reputation as a cooperator [ 8 , 50 ]. Contrariwise, helping another player when the stakes are high may be perceived as an act of weakness or even betrayal of one’s own team that could lead to experiencing sanctions by one’s teammates. We should also note that perhaps acts of helping are promoted (or undermined) when people feel and think in autonomous ways–in a manner independent of how own or other team players might evaluate such behavior [ 51 , 52 ]. But if the evaluations of the own team members guide helping (or not) on the soccer field, it is also likely that some processes are culture-specific. Growing evidence suggests that some cultures are more oriented to ingroup favoritism and collectivist mindsets [ 53 – 55 ]. Future research could look more deeply into possible explanations.
Conclusions
Acting prosocially during a competitive soccer game is a challenge. Yet, under the right circumstances, young soccer players are inclined to help a fellow player in distress at the expense of personal or team success. The present research showed that the stakes of the situation matter: When the stakes for personal and team success are low, the inclination to help increases; contrariwise, when the stakes for success and victory are high, the motivation to help tends to be lower. This finding suggests that in competitive situations like a soccer game, the cost of the prosocial act matters because “players appear to help when it doesn’t hurt.” Furthermore, we found that males and females respond differently to prosocial dilemmas on the soccer field: Female soccer players expressed greater helping intentions than their male counterparts. Being among the first to examine differences between men and women in soccer, the present study is, of course, in need of replication and the findings require further exploration to advance the literature on differences–and similarities–between men and women on the soccer field.
Supporting information
S1 Dataset
Dataset pilot study and dataset main study.
(ZIP)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
The processes through which transgender (trans) women recognize and express their gender identity are not well characterized in the literature. However, a growing body of research demonstrates the importance of reaching trans-specific developmental milestones (e.g., disclosing one’s transfeminine identity to others; expressing one’s transfeminine identity in public, changing one’s name/gender marker on government documents) in shaping the trajectories of the health and overall wellbeing of trans persons [ 1 – 3 ]. Indeed, reaching such milestones may be key components of gender affirmation , or the process of taking social, medical, and/or legal steps to achieve internal and external recognition of one’s gender identity [ 4 ]. Among trans women, pursuing gender affirmation may be critical to mitigating the impact of gender minority stress, or the pernicious, biopsychosocial sequalae of contending with anti-trans stigma and discrimination [ 5 ]. Despite recent data suggesting the importance of meeting trans-specific developmental milestones for trans women [ 2 ], few studies have explored how achieving these important milestones may play a protective role against the adverse mental health impact of stigma and structural violence. Although our understanding is limited, there is a growing body of literature that characterizes a significant association between adverse mental health outcomes and experienced violence, as well as the negative impact of discrimination on educational attainment among transgender populations [ 6 ].
Similarly, a dearth of research has explored potential associations between structural violence and the achievement of developmental milestones key to gender affirmation. Such an inquiry may be critical, however, given emerging evidence that reaching developmental milestones is associated with positive mental health outcomes, particularly for trans women. In a recent study, trans persons in the United States who changed their gender markers on government documents (i.e., passport and state ID) to reflect their gender identity (i.e., pursued legal gender affirmation) reported lower levels of negative mental health outcomes such as serious psychological distress, suicidal ideation, and suicide planning compared to those who did not change their gender markers [ 3 ]. In contrast, some research has found that experiencing developmental milestones during adolescence may subject trans persons to minority stress in addition to the societal and peer pressures that are common during this stage of life [ 7 ]. A similar study found that trans persons who experienced milestones earlier in life (i.e., during adolescence) were less likely to attain a four-year degree compared to those who experienced milestones during childhood or adulthood [ 1 ]. Associations between developmental milestones and health/social vulnerabilities among trans adolescents may occur as a result of lack of family support or limited access to gender-affirming services for adolescents. These findings suggest the complex nature and impact of experiencing trans-specific developmental milestones on trans persons’ quality of life, necessitating further research to understand how to affirm trans persons and ameliorate trans-related stigma during the course of gender identity development.
A recent paper examined developmental milestones in a diverse sample of young trans women aged 16–29 in the United States (US) [ 2 ]. The developmental milestones explored in this paper included internal processes such as developing an initial awareness of one’s transfeminine identity and transfeminine expression in private, as well as external or public processes such as disclosure of transfeminine identity to others, transfeminine expression in public, having first consensual sex with a partner as a trans woman, and initiation of hormones. Findings revealed that, on average, the earliest milestone achieved was initial awareness of transfeminine identity at approximately 10 years of age. The remaining milestones were typically experienced during adolescence, between ages 12 to 18. Although the timing of internal milestones tended to occur at similar ages across racial/ethnic groups, trans women of color tended to experience external milestones earlier than White trans women. Authors suggested that whereas earlier external milestones can promote positive mental and behavioral health in the presence of supportive social contexts, earlier development may pose risks for young trans women who do not have robust support and/or social resources. Additionally, reaching these milestones may differ by context/region due to cultural and structural factors that might facilitate or inhibit trans affirmation. As such, the current paper aims to expand research beyond the United States to examine patterns of milestone development in other global settings.
Unclear in the literature is the potential exacerbating impact of structural violence, or discrimination, violence, and stigma, on the developmental trajectories of trans women. A key aspect of this inquiry is the potential role of structural violence, defined as “avoidable impairment of fundamental human needs” that stops individuals and/or groups from reaching their full potential [ 8 ]. Structural violence is one of many factors that perpetuates the marginalization of trans women and, consequently, may be associated with achievement of trans-specific developmental milestones [ 9 ]. In the US, some of the structural factors linked to adverse health outcomes among trans women include lack of nondiscriminatory policies across institutional domains (e.g., healthcare, schools, employment), lack of health care insurance coverage for transgender-specific services like hormone therapy, and lack of provider training and education [ 5 , 10 – 13 ]. Similarly, a nascent body of literature has focused on trans women in the Philippines, where there is growing visibility and social recognition of this community as well as early findings documenting health and social vulnerabilities, including violence and discrimination among trans women [ 14 – 17 ]. For example, unique expressions of anti-trans structural violence among Filipinx (i.e., an inclusive term for describing non-binary genders in the Philippines) trans women were described in the context of healthcare settings, peer networks, educational settings, public retail establishments, as well as the Catholic Church. In a phenomenological study of Filipinx trans women [ 15 ], participants described being discriminated against and losing their jobs due to their feminine appearance, with employers citing company rules and policies on hair and dress codes. The instances described in these analyses demonstrate how structural violence and discriminatory policies not only discourages Filipinx trans women from exploring their gender identity and experiencing trans-specific developmental milestones, but also threatens the safety of those who do. While specific instances of structural violence have been described, there have not been any quantitative studies that describe how structural violence impacts trans women’s ability to experience important trans-specific developmental milestones in the Philippines or elsewhere.
The current study aims to fill gaps in knowledge about potential associations between structural violence and developmental milestones among trans women who are sexually active with men (trans-WSM) in the Philippines. Specifically, the study aims to explore such associations using data from a study with Filipinx trans-WSM. These analyses aimed to (a) describe the mean ages at which these trans-specific developmental milestones occur and (b) examine the associations between structural violence and mean ages at which at which Filipinx trans-WSM experience trans-specific developmental milestones. Identifying mean ages at which milestones occur holds potential to inform clinical outreach with young trans-WSM. Indeed, mapping age-estimates onto trans-specific milestone may reveal temporal intervention points to begin STI/HIV risk prevention, mitigate the impact of exposure to gender minority stress, and tailor programming to engender healthy development despite social vulnerabilities [ 18 ]. For Filipinx trans-WSM in particular, exploring how structural violence may interfere with health-promoting behaviors may hold implications for interventions to remove socioecological barriers across the lifespan.
Methods
Study design and participants
This analysis utilized data from a cross-sectional online survey, Project #ParaSaAtin, designed to characterize and examine access to HIV prevention and treatment services among Filipinx trans-WSM (n = 139). This analysis focused specifically on describing trans-specific developmental milestones and linkages to experiences of discrimination, stigma, and violence. The full study procedures have been described elsewhere [ 19 ]. Briefly, between June 2018 and May 2019, participants were recruited using venue-based sampling, study flyers, and snowball recruitment strategies via three local community-based organizations (CBOs) serving trans communities in the two highest HIV burden areas in the Philippines, which are Manila and Cebu metropolitan cities. Eligibility requirements included the following: (1) being 18 years old or above, (2) identify as a trans woman, (3) had condomless anal sex in the past year with a cisgender male partner, (4) currently living in Metro Manila or Cebu, and (5) demonstrated English and consent comprehension via a brief cognitive screening tool consisting of true/false questions about the consent form. All participants provided an electronic written informed consent and received a P300 ($5.58 USD) compensation for completing the survey. All of the study procedures were informed and deemed appropriate by the three CBOs involved as local study partners of this project. Study procedures were approved by the Brown University Human Research Protection Program Institutional Review Committee (IRB#: 1802001982). Electronic written informed consent was obtained from all individual participants included in the study.
Measure items
Participants completed a one-time, 20-25-minute survey that included measures on socio-demographics, trans-specific developmental milestones and hormone history for gender affirmation, as well as experiences of discrimination, stigma, and violence. Due to cross-sectional nature of the survey, exposures and outcomes of interest were measured simultaneously, providing no temporal relationships between measures.
Sociodemographic
We surveyed sociodemographic information including: (1) age (18–24 years old, 25–29 years old, 30–34 years old, and 35 years old or more), (2) current living location (Metro Manila or Cebu), (3) current employment status (employed or unemployed), and (4) education (high school or below, some college, or college and beyond).
Trans-specific developmental milestones, and hormone history
We used Restar and colleagues’ [ 2 ] indicators of trans-specific developmental milestones. Specifically, we asked participants at what age did they first experience the following milestones: (1) self-awareness of their transfeminine identity, (2) disclosure of their transfeminine identity to others; (3–4) private and public expression of transfeminine identity; and (5–6) initial sexual experience (i.e., oral, vaginal, anal sex) with a cisgender man before and after transfeminine identification. To assess hormone history, we asked participants if they have ever taken feminizing hormones for gender affirmation. Among those who reported not ever taking feminizing hormones, we asked whether they have ever desired to use it in the past (yes/no), and reasons for why they have not started hormones. Among those had ever taken feminizing hormones in the past, we asked about the locations where they get it from (e.g., clinic, private doctor, on the street, online, from a friend, pharmacy).
Structural discrimination, stigma, and violence
As a proxy to measuring structural discrimination, we utilized the discrimination portion of Davidson’s (2016) Gender Inequality Scale [ 20 ], which is a 4-item true/false subscale that asks participants whether they have experienced losing their job, been denied health services, been denied work promotion, or removed from direct contacts with patients/workmates/classmates because of their LGBT identification (Cronbach α = 0.87). Scores were summed and dichotomized at the median (low discrimination: 0 ≤ 1.20, and high discrimination: 1.21 > 4). To measure structural violence, we adapted Woulfe and Goodman’s Violence and Identity Abuse Scale [ 21 ]. This is a 7-item scale (Cronbach α = 0.98) that asks participants whether they have experienced violent reactions from others such as: “Having had a person call them pejorative names that have to do with my LGBTQ status / Having had a person threatened to tell their employer, family, or others about their gender identity / Having had a person used their gender identity against them.” Scores were summed and dichotomized at the median (low stigma: 0 ≤ 15.03, and high stigma: 15.03 > 35). We utilized Hill and colleagues’ Transphobia, Homophobia and Genderism Scale [ 22 ] to assess structural stigma. This 15-item, 5-point Likert scale measures perceptions of social beliefs towards trans women (Cronbach α = 0.94). Participants were asked to rate from 1 = strongly disagree to 5 = strongly agree on scale items such as: “People believe that it is morally wrong for a cisgender male person to present herself as a woman in public / People around me have teased a transgender woman before of her masculine appearance or behavior.” Scores were summed and dichotomized at the median (low violence: 0 ≤ 60.57, and high violence: 60.57 > 76).
Analysis strategy
We conducted descriptive analysis to describe patterns across our measures, and z-test procedures to identify mean age differences in trans-specific developmental milestones and hormone history by experiences of discrimination, stigma, and violence. We performed sensitivity analysis to determine the internal consistency of our scaled variables (i.e., discrimination, stigma, violence), providing Cronbach alphas, and dichotomized them at their median points. We utilized a two-tailed test (p<0.05) to determine statistical significance. All analyses were conducted using StataSE version 16.1 [ 23 ].
Results
Sample characteristics
Table 1 describes the sample sociodemographic characteristics and hormone history.
10.1371/journal.pone.0248248.t001
Table 1 Sociodemographic and hormone history(n = 139).
All n (%)
Demographics
Age (continuous, at the time of the survey)
18–24
45 (33.58)
25–29
54 (40.30)
30–34
20 (14.93)
35+
15 (11.19)
Study Site
Manila
110 (79.14)
Cebu
29 (20.86)
Currently Employed
Yes
80 (57.55)
No
59 (42.45)
Education
High School or below
55 (39.57)
Some College
23 (16.55)
College or beyond
61 (43.88)
Hormones
Ever taken hormones for gender affirmation
Yes
90 (64.75)
No
49 (35.25)
If Yes: which venues did you get hormones from ^ (n = 90)
A clinic / health center
27 (30.00)
A private doctor, private practice
12 (13.33)
On the street
7 (7.78)
Online
19 (21.11)
From a friend
90 (55.56)
Pharmacy
48 (53.33)
If No: Any desire to use hormones for gender affirmation in the past? (n = 49)
Yes
17 (34.69)
No
32 (65.31)
If No: Reasons for not starting hormones for gender affirmation (n = 49)
Not sure where to go to get hormones
21 (42.86)
I don’t have prescription for hormones
15 (30.61)
I can’t afford hormones
23 (46.94)
Note:
^ denotes check all that apply response option.
Most of the Filipinx trans-WSM in this study were between ages 25–29 years old (40.30%), reported currently residing in Manila (79.14%), were employed (57.55%), and had college-level education or beyond (43.88%).
The majority of the sample reported they had ever taken hormones for gender affirmation (64.75%). When asked about venues for hormone access, the most common sources of hormones included friends (55.56%), pharmacy (53.33%), a clinic or health center (30.00%), and online (21.11%). Few reported attaining their hormones from the streets (7.78%) or from a private practice (13.33%). Among those who reported they had never taken hormones (35.25%), about one-third (34.69%) reported desiring to use hormones for gender affirmation in the past. When asked about reasons for not starting hormones, common responses included not being able to afford them (46.94%), not sure where to go (42.86%), and not having a prescription (30.61%).
Developmental milestones and experiences of discrimination, stigma, and violence
Table 2 describes the developmental milestones and experiences of discrimination, stigma, and violence.
10.1371/journal.pone.0248248.t002
Table 2 Trans-specific development milestones and experiences of discrimination, stigma, and violence (n = 139).
All
Discrimination
Z statistic (p-value)
Violence
Z statistic (p-value)
Stigma
Z statistic (p-value)
Low
High
Low
High
Low
High
Trans-specific development milestones, Mean ages (SD)
1. Initial self-awareness of transfeminine identity
13.50 (6.31)
13.16
14.20
-5.62 (<0.001)
13.05
13.83
-4.42 (<0.001)
13.28
13.71
-2.47 (0.0135 )
2. Transfeminine identity disclosure to others
14.75 (5.28)
14.33
15.60
-6.82 (<0.001)
14.46
14.96
-2.81 (0.0048)
14.81
14.68
0.73 (0.4594)
3. Transfeminine expression in private
15.03 (5.99)
14.65
15.81
-6.22 (<0.001)
14.22
15.58
-7.33 (<0.001)
14.28
15.75
-8.35 (<0.001)
4. Transfeminine expression in public
16.77 (6.03)
16.50
17.32
-4.39 (<0.001)
16.45
17.01
-3.14 (0.0016)
16.25
17.28
-5.91 (<0.001)
5. First consensual oral/vaginal/anal sex with a cisgender male partner
15.15 (5.69)
14.82
15.87
-5.60 (<0.001)
14.81
15.41
-3.44 (0.0006)
14.74
15.55
-4.66 (<0.001)
6. First consensual oral/vaginal/anal sex with a cisgender male partner as a trans woman
16.11 (5.93)
15.61
17.16
-8.27 (<0.001)
15.41
16.64
-6.96 (<0.001)
15.49
16.69
-6.85 (<0.001)
7. Hormone integration (in age) (n = 90)
18.25 (3.86)
18.0
18.8
-3.60 (<0.001)
17.92
18.50
-2.71 (0.0067)
17.48
19.02
-7.27 (<0.001)
Note: outcomes dichotomized at median; SD = standard deviation.
On average, Filipinx trans-WSM in this sample reported initial awareness of their transfeminine identity at age 13.50 (SD = 6.31) years old. Filipinx trans-WSM reported initial disclosure of their transfeminine identification to another person within 1 year of their self-awareness (M = 14.75, SD = 5.28), and began private and public expression of their transfeminine identity within 2 to 4 years of their initial awareness (respectively, M = 15.03, SD = 5.99, and M = 16.77, SD = 6.03). The mean age for first consensual sex with a cisgender man was 15.15 (SD = 5.69) years old, and the mean age for first consensual sex as a trans woman with a cisgender man 16.11 (SD = 5.93) years old. Among those who reported ever taking hormones for gender affirmation (n = 90), hormone integration started on average by age 18.25 (SD = 3.86).
Mean age differences were observed for each developmental milestone by discrimination, stigma, and violence. Overall, participants who reported higher levels of discrimination, stigma, and violence also experienced a later age for nearly each milestone (i.e., initial self-awareness of transfeminine identity, transfeminine expression in private, transfeminine expression in public, first consensual oral/vaginal/anal sex with a cisgender male partner, first consensual oral/vaginal/anal sex with a cisgender male partner as a trans women, and hormone integration) (all p-values <0.05). Of note, the single exception to this pattern was the non-significant association between stigma and initial disclosure of transfeminine identification to another person.
Discussion
This study describes the rans-specific developmental milestones in a community-recruited sample of trans-WSM in Manila and Cebu, Philippines. Our findings expand previous research concentrated in the US [ 1 – 3 ], to an international setting and sample of trans-WSM in the Asia-Pacific; to our knowledge, this study provides the first account of this topic in this region. We found that on average, these milestones are experienced largely during adolescence, which supports previous research on the need for psychological, social, and medical resources to support trans adolescents and highlights the need for gender-affirming interventions targeting when these milestones occur [ 1 – 3 ]. As reported in the previous study of US trans women [ 2 ], we found that internal milestones (e.g., self-awareness of transfeminine identity) occurred earlier than external milestones (e.g., disclosure of transfeminine identity to another person, public expression of their transfeminine identity). Notably, initial self-awareness in this Filipinx sample occurred on average at 13.5 years of age, older than the average 9.9 years of age for the US sample. The majority of milestones occurred between the ages of 14 to 17, including disclosure of transfeminine identity and initiation of sexual activity. Use of hormones occurred later, started on average at approximately age 18. It is important to note that these ages are reported as averages, and that previous research in other settings have documented some of these milestones such as initial self-awareness to occur as young as 3 years old for some transgender children [ 24 ].
Importantly. we also found that experiences of structural discrimination, stigma and violence were significantly related to the timing of these milestones. In particular, Filipinx trans-WSM who reported higher rates of structural discrimination, stigma and violence also reported meeting the majority of milestones significantly later in life, with the exception of initial disclosure of transfeminine identification to another person. These findings urge for further programmatic support for adolescents who might be undergoing changes and stress related to trans-related identity development. Local community services and programming should encourage healthy developmental milestones for young Filipinx trans-WSM and must center and prioritize the psychological impact of violence, stigma, and discrimination. Gender-affirming sexual health services and programs mitigating the adverse impact of structural violence are needed to support trans adolescents in the Philippines. Perhaps more critically, it is critical for national laws along with institutional-level policies to adopt anti-transphobia and protect transgender people from discrimination, stigma, and violence across social institutions like healthcare, schools, and employment/labor. Currently, while there are anti-discrimination ordinances for transgender people at the local level, transgender populations continue to face legal challenges in the Philippines at the national level, leaving this population vulnerable to acts of discrimination on the basis of gender identity and expression [ 25 ]. As such, along with gender-affirming community services and programming, policies and laws that prevent gender-based anti-discrimination could support trans people and prevent delays in their trans-related identity development.
Previous research has reported that adolescents in the Philippines tend to initiate sexual intercourse at later ages compared with adolescents in other countries in the Global South, which might be due to pervasive cultural and religious norms in the Philippines that discourage premarital sex [ 26 ]. However, participants in this study reported age of sexual debut at an average 15 years old, commensurate with average age of sexual debut in a recent study with US trans women [ 2 ]. The average of first consensual sex among Filipinx trans-WSM in the current sample is younger compared with the national average (17 years old) in the Phillipines. Of note, it is possible the true mean age of first sex in this sample may be lower than reported, as this survey inquired only about consensual sex. Our findings point to the need for future research to understand dynamics of sexual consent among young Filipinx trans-WSM and their sexual partners, as well as the need for gender-affirming sexual health education programs for this population.
Perhaps most notable, participants who reported high (compared to low) levels of violence and stigma also reported older ages at which developmental milestones were met. There are multiple offer possible explanations for this observed association, which warrants further examination in subsequent research. One possible explanation could be that facing these negative experiences may lead trans-WSM to delay the process of important external gender-related discoveries and attainment of these milestones, i.e., pursuing social, legal, and medical gender affirmation. For example, experiences of discrimination in health care settings, including provider attitudes towards transgender people, likely prevent trans-WSM from seeking and/or initiating integration of hormone intervention [ 27 , 28 ], and exposure to social or familial reprimands for violating binary gender roles may lead to self-censorship with regard to gender identity exploration and development. Another possible explanation could be that trans-WSM who achieve gender identity milestones relatively early are likely to benefit from gender affirming resources that keep them relatively protected from structural discrimination, stigma, and violence, compared to those who undergo milestones later and do not gain access to gender-affirming resources [ 29 ]. Future longitudinal and qualitative work could delineate the directionality of these postulations, including the extent to which structural discrimination, stigma, and violence may inform these trans-specific developmental milestones, and vice versa. Lastly, it is notable that the only nonsignificant association in our findings is the association between stigma and transfeminine identity disclosure to others. Research on the psychological role of identity disclosure for trans people is mixed, with disclosure posing both positive rewards (i.e., receiving affirmation and support, building emotional connection) and potential adverse risks (i.e., increasing likelihood of interpersonal and structural violence) [ 30 ]. Disclosure has been described as a form of “stigma management” for some trans people, that is, some trans individuals who are visually conforming to societal norms of binary gender, often known as “passing privilege” or “living stealth,” may continue to choose to not disclose their gender identity to avoid stigma, violence, and discrimination [ 5 , 31 – 33 ]; our results add further support to the idea that intentional identity disclosure may be a resilience strategy to manage the impact of structural violence and stigma for some Filipinx trans-WSM as well.
It is noteworthy that while there are some developmental milestones that are independent of each other—for instance, trans-WSM choosing to not integrate hormone therapy as part of their trans-specific developmental goals [ 34 ]–a portion of women in this sample did express desiring to start hormone integration but experienced barriers to initiation. Specifically, among those who reported not having a history of hormone use, about a third (34.69%) reported desiring to use them for gender affirmation. Indicated barriers for initiating hormones in this sample included lack of information for available resources, not having access to a provider for hormone prescription, and financial cost; these barriers corroborate previous research documenting barriers to hormone use among trans women in other global settings [ 35 – 39 ]. Our findings, therefore, suggest that addressing access to hormones among young Filipinx trans-WSM, particularly those who are from low-income background and not fully connected to care but desire to start their initiation, must include providing educational information, referrals to providers, and eliminating cost related to hormone medications and clinical visits.
Limitations
This study has several limitations. First, the cross-sectional nature of the data did not allow us to establish temporality and directionality between our exposure and outcome variables. Second, the secondary data from a parent study that aimed to understand HIV-related behavioral risks of trans women who are sexually active with cisgender men, resulting in a limited sample that does not represent and omit a large proportion of trans women who have not yet experienced or have chosen to not engage in sexual activities with a cisgender male partner. Third, it is possible that trans-WSM in this sample may experience other gender beyond or besides their transfeminine identity at an earlier age, as such future constructions of measurements should further characterize these experiences and strive to situate or contextualize such experiences locally to illuminate the fluidity of gender in this population in the Philippines. Fifth, given our measurements were assessed at the individual-level to understand experiences related to structural factors, this study can be further strengthened by having multiple sources of data that are not reliant on self-report (e.g., evaluation of facility-level policies, analyzing institutional records on discrimination reports) to accurately examine structural factors. Lastly, as Filipinx trans-WSM represent a portion of a wide array of trans identities and gender diverse communities [ 40 – 42 ], future research is needed for a more comprehensive overview of trans-specific developmental milestones across the diverse range of trans identities in the Philippines.
Conclusion
This study is among the first to characterize trends in key developmental milestones among a sample of young trans-WSM in the Philippines. This study maps differences in the ages at which Filipinx trans-WSM living in Manila and Cebu affirmed their gender both internally and externally. Results are consistent with psychological literature outlining a temporal sequence of development among young trans and cisgender girls and women, including recent US data suggesting young trans women may first meet private/internal developmental milestones (e.g., developing a congruent, internal transfeminine identity) followed by a later-in-life achievement of external milestones (e.g., disclosing one’s trans identity to others) [ 2 ]. The achievement of such milestones has been shown to be potentially protective for the mental health of trans women, with studies demonstrating that taking steps to pursue both internal and external gender affirmation plays a mitigating role in the association between minority stressors and adverse health outcomes [ 43 ]. Results from the current study build on these findings with trans-WSM in the Philippines, specifically by demonstrating a positive association between gender minority stress (i.e., experiencing high levels of structural violence) and a later-in-life pursuit of gender affirmation through the achievement of these important milestones.
Filipinx trans-WSM in this sample tend to report an early-in-adolescence awareness of their gender identities, followed by a developmental process throughout their tends that includes steps toward identity disclosure and expression, hormone use (when accessible and desired), and sexual behaviors. Unique to trans-WSM’s experiences—both in the Philippines and globally—is the burden of contending with and managing structural anti-trans violence. Young trans-WSM are tasked not only with surviving adolescence—itself a profoundly difficult developmental rite of passage—but also with fostering and sustaining a positive, resilient sense of self despite a hostile environment. Indeed, structural violence has been shown to paradoxically stunt and accelerate sexual, gender, socioemotional, and cognitive development among all children and young adults. For young trans-WSM in the Philippines, data from this study suggest a significant association between the level of structural violence Filipinx trans-WSM experience and the age at which they achieve gender-related milestones. This key finding does not necessarily suggest that trans-WSM who experience higher levels of structural violence do not achieve milestones. More precisely, these data reveal that participants who experienced higher levels of structural violence achieved developmental milestones at slightly older ages: approximately one year later than those with lower levels of experiencing structural violence. Because data are not longitudinal, results do not necessarily indicate that violence caused a delay in development. These findings do, however, highlight a need for trauma-informed, strengths-based programming and institutional policies that measure and mitigate anti-trans violence, which is associated in this study with an approximately one-year delay in development. Finally, this study calls for an expansion in knowledge about the needs and strengths of trans communities beyond the US: gaps that can only be filled through earnest engagement with local community leaders already engaged in the struggle for gender health equity and trans and gender diverse liberation [ 5 , 30 ].
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Malaria is a devastating yet curable disease of global distribution, most prevalent in tropical and subtropical regions. According to the WHO report 2016, it remains one of the most deadly diseases in the world, with 200 million estimated cases and 400 thousand deaths per year. The causative agents of malaria are protozoan parasites of the genus Plasmodium .
When an infected female Anopheles mosquito takes a blood meal, it injects in the order of 100 Plasmodium sporozoites into the skin tissue [ 1 ]. From there sporozoites travel to the liver where they invade hepatocytes. When a sporozoite infects a liver cell, the host cell plasma membrane invaginates around the parasite, forming the parasitophorous vacuole membrane (PVM), in which liver stage schizogony takes place [ 2 ]. The PVM is the contact site between the parasite and its host. Despite its host cell origin, the PVM is quickly remodeled by the parasite and many Plasmodium -derived proteins can be found there [ 3 , 4 ]. The PVM has also been suggested as the target site for selective autophagy that occurs following sporozoite invasion [ 5 ].
In contrast to starvation-induced canonical autophagy, selective autophagy is nutrient insensitive and involves the selective degradation of proteins, organelles and pathogens. Ubiquitination is known to target proteins for degradation by the proteasome but can also label substrates for selective autophagy [ 6 ]. Such ubiquitin-conjugated or unfolded protein regions are recognised by a variety of autophagy receptors. They assemble through self-oligomerization and bind ubiquitin-like proteins (UBLs) of the LC3/GABARAP family. These UBLs are important for autophagosome formation, acting as a protein scaffold for the engagement of the autophagosome nucleation machinery and causing the expansion of the autophagosomal membrane [ 7 – 12 ]. The first selective autophagy receptor found in mammalian cells was nucleoporin 62 (p62), also called sequestosome 1 (SQSTM1) [ 13 , 14 ]. This protein is involved in the degradation of misfolded proteins, mitochondria, peroxisomes and pathogens [ 14 – 17 ]. Classically, p62 function depends on ubiquitination of the substrate. p62 contains an ubiquitin-binding domain (UBA), an LC3-interacting region (LIR) motif and a PB1 domain that mediates self-oligomerization. In selective autophagy, p62 is normally degraded together with its cargo. However, in autophagy-deficient cells, it is not degraded via macroautophagy and accumulates in the cytoplasm, therefore often being used to measure autophagic flux [ 11 , 18 ]. Another autophagy receptor is neighbor of BRCA1 gene 1 (NBR1), whose domain organization resembles that of p62. NBR1 is an important receptor in degradation of peroxisomes (pexophagy) [ 19 ]. Optineurin (OPTN) can act as a receptor for misfolded proteins in both a ubiquitin-dependent and -independent manner [ 20 ]. OPTN has an UBA and a LIR motif and is also involved in xenophagy and mitophagy [ 21 , 22 ]. Nuclear dot protein 52 kDA (NDP52) can also act as an autophagy receptor in xenophagy. In Salmonella infection, NDP52 labeling of bacteria-containing vacuoles is dependent initially on galectin 8 and then on ubiquitin. [ 23 ].
Whereas autophagy-dependent selective elimination is a well-known host cell reaction against bacteria after invasion, there are only very few reports in the literature about selective autophagy in cells infected by eukaryotic parasites. Successful elimination by selective autophagy has been reported for the apicomplexan parasite Toxoplasma gondii [ 24 , 25 ]. However, it has also been shown that T . gondii is capable of actively evading this autophagic destruction by activating EGFR, which inhibits LC3 accumulation around the parasite [ 26 ]. More recently, we investigated selective autophagy events in Plasmodium- infected hepatocytes and showed that the PVM of Plasmodium liver stage parasites is rapidly and heavily labeled by the host cell-derived autophagy marker protein LC3B, indicating that the host cell quickly recognises the invader [ 5 , 27 ]. Interestingly, this labeling is greatly reduced in later stages of normally developing parasites, suggesting that the parasite is able to escape from this host cell response in order to successfully establish infection and undergo replication [ 5 ]. In contrast, persistent LC3B-labeling is linked to parasite growth arrest and to elimination, indicating that the host cell can defend itself successfully using autophagy or a related mechanism. Importantly, in addition to LC3B, ubiquitin and the autophagy receptor p62 also accumulate around the parasite [ 5 ]. However, the mechanisms that allow different autophagy marker proteins to be recruited to the PVM remained unknown. It was also unclear whether other autophagy receptors are involved in the observed selective labeling of the PVM and these questions are the basis of the work presented here.
We used the rodent parasite Plasmodium berghei to infect wild type and LC3B-deficient HeLa cells generated using CRISPR/Cas9 technology [ 28 ]. In contrast to what has been shown for classical selective autophagy, we found that p62, NBR1 and ubiquitin recruitment to the PVM depends on the presence of LC3B.
Material and methods
Cell culture, treatment and infection of HeLa cells
Wild type HeLa cells (a gift from Robert Menard, Pasteur Institute, Paris), LC3B -/- and ATG5 -/- HeLa cells were cultured in Minimum Essential Medium with Earle’s salts (MEM EBS, 1-31F01-I, BioConcept, Allschwil, Switzerland), supplemented with 10% FCS (GE healthcare, Glattbrugg, Switzerland), 100 U/ml penicillin, 100 μg/ml streptomycin, and 2 mM L-glutamine (both BioConcept, BioConcept, Allschwil, Switzerland). Cells were cultured at 37°C and 5% CO 2 and split using Accutase (Innovative Cell Technologies, San Diego, California, USA) diluted 1:1 in PBS.
For starvation, cells were rinsed 3 times with PBS and subsequently incubated in Earle’s Balanced Salt Solution (EBSS, E2888, Sigma-Aldrich, Buchs, Switzerland) for 2 h before fixation. For simultaneous treatment with chloroquine and rapamycin, cells were grown in MEM EBS supplemented with 10 μM chloroquine (C6628, Sigma-Aldrich, Buchs, Switzerland) and 250 ng/ml Rapamycin (R-500, LC Laboratories, New Boston, Massachusetts, USA) for 4 h before cell lysates were prepared.
For infection of HeLa cells, salivary glands of P . berghei-infected Anopheles stephensi mosquitoes were isolated and disrupted to release sporozoites. Sporozoites were incubated with cells in the smallest possible volume of MEM EBS medium containing 25 μg/ml Amphotericin B (4-05F00-H BioConcept, Allschwil, Switzerland) for 2h. Subsequently, they were rinsed and incubated in the respective medium containing 2.5 μg/ml Amphotericin B. Parasites used in this study have a P . berghei ANKA background. Pb mCherry parasites are phenotypically wild-type and show cytosolic localisation of the fluorescent mCherry protein [ 29 ].
Transfection of HeLa cells
HeLa cells were harvested by Accutase treatment and 1 x 10 6 cells were pelleted by centrifugation at 1000 x g. Cells were resuspended in Nucleofector V Solution (VVCA-1003, Lonza, Switzerland) and transfected with 1 μg of plasmid DNA using program T-028 of the Nucleofector 2b transfection device (Lonza, Switzerland) and according to the manufacturer’s instructions. For IFA analysis, cells were seeded onto glass cover slides in 24 well plates (#0117530, 13 mm No. 1.5, Marienfeld GmbH, Lauda-Königshofen, Germany)
Plasmids
The various GFP expression vectors used are from Clontech. To generate the plasmid GFP-NBR1, the NBR1 cDNA was amplified by PCR using the forward primer CTGGTACCATGGAACCACAGGTTACTC and the reverse primer TGGGCCCTCAATAGCGTTGGCTGTA and plasmid pMXs-IP GFP-NBR1 (Addgene plasmid 38283, provided by Noboru Mizushima [ 30 ]) was used as a PCR template. The PCR product was subcloned into pJET1.2 (#K1232, Thermo Fisher Scientific, Reinach, Switzerland) via blunt end cloning and the NBR1 cDNA was verified by sequencing using the following primers: pJET1.2-fw CGACTCACTATAGGGAG ; pJET1.2-rev ATCGATTTTCCATGGCAG ; NBR1-915bp GCGAGCTGAGAAGAAACA . Finally the NBR1 cDNA was cloned into pEGFP-C1 using restriction enzymes KpnI and ApaI. To obtain the plasmid GFP-NDP52, the NDP52 cDNA was excised from plasmid pCR3-3x-Flag-NDP52 (a gift from Matias Faure, INSERM, Lyon) using the enzymes KpnI and ApaI. The NDP52-containing DNA fragment was cloned into pEGFP-C1 linearised with the same enzymes. For generating the GFP-OPTN plasmid, the OPTN cDNA was cut out from plasmid pOPTN-EGFP (Addgene plasmid 27052, provided by Beatrice Yue [ 31 ]) using the enzymes EcoRI and BamHI and ligated into pEGFP-C3 linearised with the same enzymes. The GFP-Gate16 plasmid is a gift from Zvulun Elazar, Weizmann Institute, Rehovot, Israel; pmRFP-ratLC3B (Addgene plasmid 21075, deposited by Tamotsu Yoshimori [ 32 ]), GFP-ubiquitin (Addgene plasmid 11928, deposited by Nico Dantuma [ 33 ]).
Generation of an LC3B knockout cell line
The CRISPR/Cas9 nickase system described by [ 34 ] was used to knock out LC3B in HeLa cells. CRISPR guide RNA pairs (gRNAs) were designed to target exon 1 of the LC3B gene. Cloning of the 2 gRNA oligonucleotides, GATCCCTGCACCATGCCGT and GGCGACGACGCGAGGGTCC , into the plasmid pX335-U6-Chimeric_BB-CBh-hSpCas9n(D10A), (Addgene plasmid 42335, supplied by Feng Zhang [ 35 ]) was performed following the protocol of the Zhang laboratory [ 34 ]. HeLa cells were transfected with the two pX335 plasmids, each encoding one gRNA sequence and in addition with the plasmid pcDNA3.1(-)-Puro-EGFP-C1 (this plasmid is a gift from Erich Nigg, Biozentrum, University of Basel, Switzerland) following the transfection protocol described above. Transfected cells were seeded into 2 wells of a 6 well plate and 24 hours after transfection, 1 μg/ml puromycin was added for 48 h to select for transfected cells. After puromycin selection, cells were cultured for 10 days in MEM, passaged when confluent and subsequently plated into 10 96-well plates to obtain clonal cell lines. Single colonies were expanded and a first screen, to assess for the absence of LC3B, was carried out by immunofluorescence staining. Cells of the clonal cell lines were seeded into glass-bottom 96-well plates and grown over night. The next day, cells were starved by incubation in EBSS for 2h to induce formation of autophagosomes, which can be visualised by fluorescence microscopy. Cells were then fixed and stained with α-LC3B antibodies. Putative positive clones that lacked an LC3B staining were further analysed by immunoblotting. Clones that showed no LC3B signal by Western blot analysis were further analyzed on the genomic DNA level. Genomic DNA was isolated from these cells using the QuickExtract ™ DNA extraction solution 1.0 (Epicentre, Madison, Wisconsin, USA) and the regions of interest were amplified using PCR (PCR primers are listed in S1 Table ). PCR fragments were cloned into pGEM-T-Easy (#1360, Promega, Madison, Wisconsin, USA) and at least 20 plasmids of each cell line were analyzed by sequencing using standard primers (T7: TAATACGACTCACTATAGG ; SP6: ATTTAGGTGACACTATAG ). Sequencing of the targeted genomic regions of knockout lines confirmed the presence of DNA alterations that lead to the knockout phenotype. Sequences of the different primers used for analyzing the three LC3B knockout cell lines and the detailed results of this analysis are summarised in S1 Table .
Protein lysates and western blotting
Cells were seeded into 24- or 6-well plates to reach confluency the next day. 24h later, after the appropriate treatment, cells in each well were rinsed with PBS and lysed for 30 min on ice with 50 μl or 200 μl ice-cold RIPA buffer (50mM Tris-HCl pH 7.0, 1% NP-40, 0.5% Na-deoxicholate, 150mM NaCl, 2mM Na-Fluoride, 0.1% SDS, Complete ™ Mini EDTA-free protease inhibitor cocktail (Sigma-Aldrich, Buchs, Switzerland)) and incubated on ice for 30 min with occasional shaking. Next, the lysate was centrifuged at 16,000 rcf / 15 min / 4°C. The supernatant was transferred into a fresh tube and mixed with Laemmli sample buffer. Proteins were denatured by incubating at 90°C for 5 min and then separated on 12% or 15% (for LC3B) SDS PAGE gels, followed by transfer to nitrocellulose membrane. 5% milk in TBST was used to block the membranes and for incubation with rabbit anti-LC3B (L7543, 1:1500, Sigma-Aldrich, Buchs, Switzerland), mouse anti-p62 (M162-3, 1:1000, MBL International, Woburn, Massachussetts, USA), chicken anti-GAPDH (AB2302, 1:5000, EMD Millipore, Darmstadt, Germany) and mouse anti-alpha-Tubulin (T9026, 1:1000, Sigma-Aldrich, Buchs, Switzerland). For secondary antibody incubation, anti-rabbit, anti-mouse IgG 800 CW IRDye and anti-chicken and anti-mouse IgG 680 LT IRDye (Li-Cor Biosciences, Lincoln, Nebraska, both 1:10,000) were diluted in 5% milk in TBST. A Li-Cor Odyssey Imaging system (Li-Cor Biosciences, Lincoln, Nebraska) was used for detection.
Indirect immunofluorescence analysis
For IFA analysis, cells were grown on glass cover slips (#0117530, Marienfeld GmbH, Lauda-Königshofen, Germany). After indicated time periods, cells were fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS; 137 mM NaCl, 2.7 mM KCl, 10 mM Na 2 HPO 4 , 1.8 mM KH 2 PO 4 , pH 7.4) for 10 min (all incubations at room temperature). Permeabilization was performed for five minutes in 0.1% Triton X-100 (T8787, Sigma-Aldrich, Buchs, Switzerland) in PBS or in 10 mg/ml digitonin (D141, Sigma-Aldrich, Buchs, Switzerland) in PBS for the anti-LC3B antibody. After washing with PBS, unspecific binding sites were blocked by incubation in 10% FCS/PBS for 10 min followed by incubation with primary antibody in 10% FCS-PBS for at least 1 hour. Primary antibodies used were mouse anti-LC3B (M152-3, 1:500, MBL International, Woburn, Massachusetts, USA), rabbit anti-UIS4 antiserum (provided by P. Sinnis, Baltimore, USA, 1:500), rabbit anti-GFP (SP3005P, 1:1000, Acris, Herford, Germany), mouse anti-GFP (11814460001, 1:1000, Roche Life Science, Rotkreuz, Switzerland), mouse anti-p62 (M162-3, 1:500, MBL International, Woburn, Massachusetts, USA), monoclonal rabbit anti-NBR1 (D2E6) (#9891, 1:1000, Cell Signaling Technology, Danvers, Massachusetts, USA), monoclonal mouse anti-Ubiquitin (FK2) (PW8810, 1:1000, Enzo Life Sciences Farmingdale, New York). Subsequently, cells were incubated with fluorescently labeled secondary antibodies in 10% FCS/PBS for at least 45 min using the following antibodies: anti-mouse Alexa488 (A11001, 1:1000, Invitrogen, Carlsbad, California, USA), anti-rabbit Alexa488 (A11008, 1:1000, Invitrogen, Carlsbad, California, USA), anti-mouse Alexa594 (A11032, 1:1000, Invitrogen, Carlsbad, California, USA), anti-rabbit Cy5 (Dianova, 1:1000, Hamburg, Deutschland) and anti-mouse Cy5 (Dianova, 1:1000, Hamburg, Deutschland). DNA was visualised by staining with 1 μg/ml DAPI (Sigma-Aldrich, Buchs, Switzerland) in PBS for 5 min. Cells were mounted on microscope slides with ProLong ® Gold Antifade Mountant (P36930, Thermo Fisher Scientific, Reinach, Switzerland) and analysed using a Leica TCS SP8 confocal microscope. For quantification of autophagy markers, a widefield Leica DM5500B epifluorescence microscope was used. Image processing was performed using Fiji software.
Quantification of autophagy markers
Association of autophagy receptors with P . berghei parasites
HeLa WT cells were transfected with plasmids expressing GFP-tagged versions of the autophagy receptors NBR1, NDP52, OPTN. Approximately 24 hours after transfection, cells were infected with Pb mCherry parasites and 6 hours post-infection, cells were fixed and stained with antibodies and monitored by fluorescence microscopy. For p62, non-transfected cells were infected and endogenous p62 was stained with antibodies. Receptor association with the P . berghei PVM was quantified visually. The slide was screened in the red channel for mCherry-expressing parasites. If the parasite PVM could also be clearly detected in the green channel (i.e. labeled with one of the autophagy receptors), this parasite was considered as positive. 60–100 parasites were counted for each group in each of two individual experiments.
Quantification of p62 association with P . berghei
HeLa WT and LC3B knockout cells were infected with Pb mCherry and 6 hours post-infection were fixed and stained with anti-p62 antibodies. Quantification was performed visually by counting the parasites that were stained strongly or weakly by the antibody. If the signal covered more than 30% of the area surrounding the parasite the signal was considered as strong. If less than 30% of the surrounding area was covered with p62, the signal was counted as weak or as having no association. Between 100 and 130 parasites were analyzed in each of two individual experiments.
For quantification of p62 association in the add-back experiment, HeLa WT and HeLa LC3B knockout cell lines were transfected with pmRFP-LC3B. Approximately 24 hours after transfection, cells were infected with Pb mCherry and 6 hours post-infection were fixed and stained with anti-p62 antibodies. Cells that were RFP-LC3B-positive and infected with Pb mCherry were monitored for p62 localization. Quantification was performed visually as described above. 60–120 parasites were analyzed in each of two individual experiments.
For calculating the Pearson’s correlation coefficient in the add-back experiment, cells were transfected and infected as described in the paragraph above. To visualise p62, fixed cells were stained with a mouse monoclonal anti-p62 antibody followed by a secondary anti-mouse antibody labeled with Alexa488. To visualise RFP-LC3B, we used a polyclonal rabbit anti-GFP (Acris) antiserum, which also stains RFP (but not mCherry) followed by a secondary anti-rabbit antibody labeled with Cy5. Confocal pictures of the green and far-red channels were taken and a Pearson’s correlation coefficient was calculated using the Coloc2 tool of the Fiji software.
Quantification of GFP-NBR1 association with P . berghei
HeLa WT and LC3B knockout cells were transfected with GFP-NBR1 and approximately 18 hours after transfection were infected with Pb mCherry. Cells were fixed 6 hours post-infection and stained with anti-NBR1 antibodies. Quantification was performed visually by counting the parasites that were stained strongly or weakly by the antibody. If the signal covered more than 30% of the area surrounding the parasite the signal was considered as strong. If less than 30% of the surrounding area was covered with GFP-NBR1 the signal was counted as weak. Between 84 and 111 parasites were analyzed in each of two individual experiments.
For quantification of GFP-NBR1 association in the add-back experiment, HeLa WT and HeLa LC3B knockout cell lines were transfected with pmRFP-LC3B and GFP-NBR1. Approximately 24 hours after transfection, cells were infected with Pb mCherry and 6 hours post-infection cells were fixed and stained with monoclonal anti-NBR1 and anti-LC3B antibodies. Cells that were RFP-LC3B-positive and GFP-NBR1-positive and infected with Pb mCherry were monitored for GFP-NBR1 localization. Quantification was performed visually as described above. 50 parasites were analyzed in each of two individual experiments.
For calculating the Pearson’s correlation coefficient in the add-back experiment, HeLa WT and HeLa LC3B knock out cell lines were transfected with pmRFP-LC3B and GFP-NBR1. Approximately 24 hours after transfection, cells were infected with WT P . berghei sporozoites and 6 hours post-infection, cells were fixed and stained with anti-NBR1 and anti-LC3B antibodies. Cells that were RFP-LC3B-positive and GFP-NBR1-positive and infected with P . berghei were analysed by confocal microscopy. Pearson’s correlation coefficient was calculated using Coloc2 module of the Fiji software.
Quantification of ubiquitin association with P . berghei
HeLa WT and LC3B knockout cells were treated as described for the p62 quantification experiments. Fixed cells were stained with anti-ubiquitin antibodies. Quantification was performed in the same way as for p62. Between 77 and 103 parasites were analyzed in each of two individual experiments.
For quantification of the ubiquitin association in the add-back experiment, cells were treated as described for p62 quantification. Fixed cells were stained with anti-ubiquitin antibodies. Quantification was performed in the same way as described for p62. Between 53 and 83 parasites were analyzed in each of two individual experiments.
For calculating the Pearson’s correlation coefficient in the add-back experiment, HeLa WT and HeLa LC3B knock out cell lines were transfected with pmRFP-LC3B. Approximately 24 hours after transfection, cells were infected with WT P . berghei sporozoites and 6 hours post-infection cells were fixed and stained with anti-ubiquitin and anti-LC3B monoclonal antibodies. Cells that were RFP-LC3B-positive and GFP-NBR1-positive and infected with P . berghei were analyzed by confocal microscopy. Pearson’s correlation coefficient was calculated using Coloc2 module of the Fiji software.
Statistical analysis
Two groups were compared using a 2-tailed, unpaired Student t-test. All statistical analyses were carried out using Prism 4.0c for Mac, GraphPad Software, San Diego, California, USA. P values of less than 0.05 were considered to indicate statistical significance. Pearson’s correlation coefficients were calculated with the Coloc 2 tool in Fiji by defining a region of interest around the P . berghei parasite.
Results
p62 and NBR1 are recruited to the PVM in Plasmodium -infected HeLa cells
When liver cells are infected with P . berghei , the parasites are immediately recognised and the PVM is labeled with the autophagy marker proteins LC3B, ubiquitin and p62 [ 5 ]. Approximately half of the invaded parasites are eliminated early in infection. Based on the PVM labeling, selective autophagy is obviously a possible means of elimination. The observed mechanism appears, however, to be different from classical selective autophagy, which is characterised by the generation of an additional membrane around the invading pathogen. To further analyze the selective PVM labeling we infected HeLa cells expressing GFP-tagged selective autophagy receptors, NBR1, NDP52 and OPTN in P . berghei -infected cells. Cells were then fixed 6 hours post-infection and stained with α-GFP or α-p62 antibodies to localize the different receptors and additionally with anti-UIS4 antibodies to allow visualization of the PVM. The localization of the different receptors was then analyzed by confocal microscopy. Fig 1 shows that endogenous p62 and GFP-NBR1 clearly associate with the PVM of the parasite. GFP-NDP52 shows some PVM localization but is primarily in the cytoplasm and OPTN-GFP has an entirely cytoplasmic localization. The control protein, GFP, localises to both the cytoplasm and nucleus. A quantitative assessment of this localization experiment ( Fig 1B and 1C ) revealed that in the majority of infected cells, the PVM was clearly labeled with p62 (77.9%) and with GFP-NBR1 (75.3%). In contrast, GFP-NDP52 association with the PVM was only found in 13.5% of infected cells and OPTN-GFP and GFP did not localise to the PVM in infected host cells.
10.1371/journal.pone.0183797.g001
Fig 1
Autophagy receptors p62 and NBR1 localise to the parasitophorous vacuole membrane of Plasmodium berghei .
(A) HeLa cells were infected with Plasmodium berghei . The parasitophorous vacuole membrane (PVM) of the parasite was stained with antibodies (UIS4, red). To visualise the autophagy receptors, cells were stained with antibodies (p62, green) or transfected 24 hours before infection with plasmids expressing GFP fusion proteins (GFP-NBR1, GFP-NDP52, OPTN-GFP, GFP, all shown in green). Infected cells were fixed 6 hours post-infection, stained with antibodies and analysed by confocal microscopy. DNA was stained with DAPI. Scale bar 10 μm (B) Numbers of receptor-labeled P . berghei parasites were determined by fluorescence microscopy. 60–100 parasites were counted in two separate experiments. Numbers of labeled parasites are expressed as percentages, error bars show standard deviations, p values were calculated using a t-test. (C) Pearson's correlation coefficients were calculated from at least 5 images, except for cells expressing GFP only. Standard deviations are depicted, p values were calculated using a t-test.
Generation of LC3B knock out cell lines
Having confirmed that several selective autophagy receptors are recruited to the PVM, we next explored the mechanism of how this occurs. It is well known that LC3 family proteins interact with autophagy receptors through their LIR domains [ 11 ]. To further analyze this process on the PVM, we constructed an LC3B knock out cell line in HeLa cells using CRISPR/Cas9 genome editing. To reduce potential off-target effects we used a double nick strategy [ 34 ] with two guide RNA (gRNA) sequences targeting the first exon of LC3B ( Fig 2A ). Introducing staggered nicks leads to a DNA double strand break that will be imperfectly repaired by the non-homologous end joining repair pathway, resulting in insertion/deletion mutations that will ultimately disrupt the gene function. The plasmids encoding the gRNAs and the Cas9 nickase were co-transfected together with a plasmid harboring a puromycin resistance cassette. Puromycin-selected and expanded clones were first tested by immuno fluorescence analysis using α-LC3B antibodies and then by immunoblot analysis. Three individual clones were found to be LC3B-deficient ( Fig 2B and 2C ). From these three clones the genomic DNA region was amplified by PCR, the PCR fragments were subcloned and at least 20 plasmids of each clone were sequenced. All three clones showed modifications at the predicted locus (see S1 Table ). For simplification, we present in the subsequent main figures the results from clone 25 only. The corresponding figures for clones 89 and 95 are available as supplementary data.
10.1371/journal.pone.0183797.g002
Fig 2
Generation of LC3B knockout HeLa cell lines.
(A) Genomic region of the LC3B gene. Exon 1 is shown in lowercase blue, the 5’-upstream region is printed in black capitals. Binding regions of the two gRNAs are highlighted in green and PAM sequences are shown in red. (B) Western blot of HeLa WT and three LC3B knock out cell lines to confirm the lack of LC3B protein in three clonal cell lines. GAPDH was used as a loading control. (C) IFA analysis of HeLa WT and three LC3B knock out cell lines. Cells were starved for 2 hours in EBSS, fixed, stained with anti-LC3B antibodies (green) and analysed by fluorescence microscopy. DNA was visualised with DAPI (blue). Scale bar 20 μm.
LC3B knockout does not block canonical autophagy
Before we started infection experiments with the newly generated LC3B-deficient clones, we wanted to know whether they are still proficient in executing canonical autophagy, including formation of autophagosomes and fusion with lysosomes. We therefore treated WT and LC3B -/- HeLa cells with rapamycin and chloroquine. Rapamycin is an inhibitor of mTOR and activates autophagy. Chloroquine inhibits the fusion of autophagosomes with lysosomes and therefore the autophagic flux. Proteins that are located in the autophagosomes, such as LC3B-II and p62, are thus expected to accumulate [ 36 ]. In WT HeLa cells, LC3B-II indeed accumulated after rapamycin and chloroquine treatment, confirming the inhibition of LC3B turnover ( Fig 3A and S1A Fig ), although no significant accumulation of p62 was observed. As a solid control for the inhibition of the autophagic flux, we used HeLa cells lacking the ATG5 protein [ 37 ]. These cells are not able to generate LC3B-II and show a clear accumulation of p62 even without rapamycin and chloroquine treatment. The short drug treatment of ATG5-deficient cells shows no additional effect because autophagy is always blocked in this cell line ( Fig 3A and S1A Fig ). For the LC3B-deficient cell lines, no difference could be detected in respect to the autophagic flux, suggesting that LC3B is not required for autophagy in HeLa cells. To confirm this result, we also investigated the formation of autophagosomes in WT and LC3B knockout cell lines using Gate16, another typical autophagosome marker protein, belonging to the LC3/GABARAP family of proteins. Gate16 is, similarly to LC3B, lipidated and incorporated into the autophagosomal membrane [ 38 ]. The different cell lines were transfected with a plasmid encoding GFP-Gate16 and again treated with rapamycin and chloroquine. To visualise autophagosomes, cells were fixed and stained with anti-GFP and anti-LC3B antibodies. In WT HeLa cells, GFP-Gate16-positive vesicles co-localise with LC3B-positive vesicles, confirming that they are true autophagosomes ( Fig 3B and S1B Fig , second top panel). Interestingly, LC3B-deficient HeLa cells are still fully proficient in forming autophagosomes, which are represented by GFP-Gate16 positive foci ( Fig 3B and S1B Fig , 3 lower panels) confirming the immunoblot analysis ( Fig 3A and S1A Fig ). The observation that LC3B-deficient cells are still fully capable of undergoing autophagy was expected since Nguyen and coworkers have recently shown that cells deficient in all ATG8 family proteins are still able to form autophagosomes [ 39 ].
10.1371/journal.pone.0183797.g003
Fig 3
LC3B knockout cells are able to undergo canonical autophagy.
(A) Representative western blot of non-infected HeLa WT, ATG5-knockout cells and one clonal LC3B-knockout cell line left untreated or treated simultaneously with 10 μM chloroquine and 250 ng/ml rapamycin for 4 hours. (B) HeLa WT and HeLa LC3B knockout cells ectopically expressing GFP-Gate16 were left untreated or treated with 10 μM chloroquine and 250 ng/ml rapamycin for 4 hours. Fixed cells were stained with anti-GFP antibodies to visualise Gate16 (green) or anti-LC3B antibodies (red). DNA was stained with DAPI (blue). White arrows in the LC3B panel indicate Gate16-transfected cells. Yellow arrowheads in the enlarged pictures indicate autophagic structures where Gate16 and LC3B colocalise. Scale bar 20 μm. Experiments using other clonal LC3B-knockout cells are shown in S1B Fig .
LC3B recruits p62 to the PVM in P . berghei -infected cells
To learn more about the kinetics of LC3B and p62 labeling of the PVM, we next infected HeLa cells lacking a functional LC3B gene and visualised p62 with anti-p62 antibodies 6 hours post infection. According to the existing literature on selective autophagy [ 11 , 40 ], we expected that the PVM would first be ubiquitinated, which would in turn attract p62 binding. The ubiquitin-bound p62 is then expected to recruit LC3B via a LIR domain. Contrary to this expectation, we found that in infected LC3B-deficient cells, p62 labeling of the PVM was strongly reduced compared to in parasite-infected WT HeLa cells ( Fig 4A and 4B and S2A and S2B Fig ). This implies that in P . berghei -infected cells, p62 association with the PVM mainly depends on LC3B. To confirm this observation, we also infected ATG5 -/- cells and monitored p62 localization. In infected ATG5 -/- cells, LC3B cannot be lipidated and is thus not integrated into the PVM [ 5 ]. In support of the previous result, p62 did not significantly associate with the PVM in ATG5 -/- cells. In a very small number of infected cells, some dot-like structures could be seen close to the parasite ( Fig 4A and 4B and S2A and S2B Fig ).
10.1371/journal.pone.0183797.g004
Fig 4
LC3B recruits p62 to the PVM.
(A) HeLa WT, HeLa LC3B- and ATG5-knockout cells were infected with Pb mCherry (red). 6 hours post-infection, cells were fixed and stained with α-p62 antibodies (green). All cell lines were transfected with RFP-LC3B (two lower panels) and infected 17 hours after transfection with P . berghei sporozoites expressing mCherry (red). p62 (green) was visualised using a specific monoclonal α-p62 antibody. DNA was labeled with DAPI (blue). Cells were analyzed by confocal microscopy. Scale bar 10 μm. (B) Numbers of p62-labeled P . berghei parasites in non-transfected and in RFP-LC3B-transfected cells were determined by fluorescence microscopy. 100–130 parasites were analyzed in non-transfected HeLa cells and 60–120 parasites were analyzed for RFP-LC3B-transfected HeLa cells. Two individual experiments were carried out. Labeled parasites are expressed as percentages. Standard deviations are depicted. (C) Pearson’s correlation coefficient of p62 and RFP-LC3B were calculated for six individual P . berghei parasites in HeLa WT and LC3B -/- cells transfected with RFP-LC3B. The mean values of 0.762 (HeLa WT) and 0.715 (LC3B -/- ) indicate a strong co-localization of p62 and RFP-LC3B. Depicted are standard deviations.
To finally prove that the observed lack of p62 recruitment to the PVM was indeed due to the lack of LC3B and was not an unspecific artifact of cell transfection and cloning, we transiently complemented all three LC3B-deficient HeLa cell lines with a plasmid encoding an RFP-LC3B fusion protein and infected them with P . berghei sporozoites. In all three RFP-LC3B-complemented LC3B -/- cell lines, the p62 localization to the PVM was comparable to in WT HeLa cells ( Fig 4A and S2A Fig , two lowest panels). In all three complemented LC3B -/- cell lines, more than 85% of the parasites showed a strong association with p62 ( Fig 4B and S2B Fig ). For one LC3B knock out cell line complemented with RFP-LC3B, the co-localization coefficient was calculated, confirming a strong association of RFP-LC3B and p62 at the PVM ( Fig 4C ). Indeed, in all RFP-LC3B-transfected cells harboring a parasite, we observed a very strong PVM labeling with p62, suggesting that indeed LC3B is responsible for the recruitment of p62 and not the other way around. Together, these results provide strong evidence that LC3B plays a decisive role in recruiting the autophagy receptor p62 to the PVM in P . berghei -infected host cells.
LC3B recruits GFP-NBR1 to the PVM in P . berghei -infected cells
As p62 and NBR1 are described as interacting with each other [ 41 ] and as we observed NBR1 at the PVM of P . berghei- infected cells ( Fig 1 ), we next analyzed NBR1 localization in LC3B- and ATG5-deficient cell lines. NBR1 was ectopically expressed as a GFP-NBR1 fusion protein. HeLa WT, LC3B -/- and ATG5 -/- cells were transfected with GFP-NBR1 and infected with Pb mCherry. Six hours post-infection, cells were fixed, stained with antibodies and GFP-NBR1 localization was qualitatively and quantitatively evaluated by microscopy. GFP-NBR1 clearly localised to the PVM of infected HeLa WT (72%) cells and to some extent also to the PVMs of parasite-infected LC3B -/- cells (53%) and ATG5 -/- cells (24%) ( Fig 5A and 5B ). This is in contrast to p62, which labels the parasite PVM in less than 25% of infected LC3B-knockout cells and in less than 1% of ATG5-knockout cells ( Fig 4B ). This indicates that GFP-NBR1 is able to interact with other proteins, most likely of the ATG8 family or with parasite proteins in the PVM. The distribution of GFP-NBR1 and p62 differs markedly. In both non-infected and infected cells GFP-NBR1 shows a more punctate pattern in contrast to p62, which is more evenly distributed (Figs 4A and 5A ). Also the localisation of GFP-NBR1 to the PVM appears more in clusters compared to p62, which is largely evenly distributed over the surface of the PVM. When RFP-LC3 and GFP-NBR1 are simultaneously transfected into HeLa WT and LC3B-deficient HeLa cells, at least 98% of the parasites show a GFP-NBR1-positive PVM with a strong co-localization of both molecules ( Fig 5C ). This result provides strong evidence that LC3B is indeed able to recruit GFP-NBR1 to the PVM in P . berghei -infected host cells. It now remains to be shown whether or not GFP-NBR1 is directly recruited by LC3B or through an interaction with p62.
10.1371/journal.pone.0183797.g005
Fig 5
LC3B recruits GFP-NBR1 to the PVM.
(A) HeLa WT, HeLa LC3B- and ATG5-knockout cells were infected with Pb mCherry (red). 6 hours post-infection, cells were fixed and stained with anti-NBR1 antibodies (green) as a control. Control and knockout cell lines were transfected with RFP-LC3B and GFP-NBR1 expression constructs (two lower panels) and infected 17 hours after transfection with Pb mCherry sporozoites (red). RFP-LC3B (red) and GFP-NBR1 (green) were visualised using mouse monoclonal anti-LC3B and rabbit monoclonal anti-NBR1 antibodies. DNA was labeled with DAPI (blue). Cells were analyzed by confocal microscopy. Scale bar 10 μm. (B) Numbers of GFP-NBR1-labeled P . berghei parasites in non-transfected and in RFP-LC3B-transfected cells were determined by fluorescence microscopy. 84–111 parasites were analysed in the non-transfected HeLa cells and 50 parasites were analysed for the RFP-LC3B-transfected HeLa cells. Two individual experiments were carried out. Labeled parasites are expressed as percentages. Standard deviations are depicted. (C) Pearson’s correlation coefficient of GFP-NBR1 and RFP-LC3B was calculated from at least five individual P . berghei parasites in HeLa WT and LC3B -/- cells transfected with RFP-LC3B. The mean values of 0.838 (HeLa WT) and 0.910 (LC3B -/- complemented with RFP-LC3) indicate a strong co-localization of p62 and RFP-LC3B. Depicted are standard deviations.
To exclude that we observed indirect binding of RFP-LC3B to GFP-NBR1 mediated through a RFP-GFP interaction, we carried out a control experiment. HeLa WT cells were transfected simultaneously with RFP-LC3B and GFP alone. Approximately 24 hours after transfection, these cells were infected with Pb mCherry. Cells were fixed 6 hours post-infection and analyzed by fluorescence microscopy. No interaction between RFP-LC3B and GFP was detected ( S3 Fig ) confirming that LC3B can indeed specifically recruit the receptor protein NBR1.
LC3B recruits ubiquitin to the PVM in P . berghei -infected cells
The PVM of P . berghei parasites is immediately labeled with the autophagy marker proteins LC3B, p62 and NBR1 after sporozoite infection. As our experiments strongly suggest that p62 and NBR1 labeling is dependent on L3CB, the observed mechanism is obviously different from classical selective autophagy. To determine whether in P . berghei -infected cells a completely inverted recruitment of autophagy proteins occurs, we next investigated the recruitment of ubiquitin to the PVM in the three different cell lines. GFP-ubiquitin was ectopically expressed in HeLa WT, LC3B- and ATG5-deficient cells and transfected cells were infected with Pb mCherry. Six hours post-infection, cells were stained with antibodies and the localization of ubiquitin was monitored visually by fluorescence microscopy.
Ubiquitin-labeled parasites could be detected in HeLa WT, LC3B -/- and in ATG5 -/- cells but labeling was strongest in HeLa WT cells ( Fig 6A and 6B ). In LC3B-deficient cells we observed less ubiquitin-labeled parasites than in WT cells. However, ubiquitin recruitment to the PVM labeling was weakest in ATG5-deficient cells. When HeLa WT and LC3B-deficient cells are transfected with RFP-LC3B, the parasite PVM shows a much stronger labeling with ubiquitin ( Fig 6A and 6B ). Transfection of ATG5-deficient cells with RFP-LC3B did not show this effect, confirming that the observed phenotype is highly specific ( Fig 6A and 6B ). These observations further support the notion that LC3B can recruit ubiquitin to the PVM of P . berghei -infected cells, probably via p62 and NBR1.
10.1371/journal.pone.0183797.g006
Fig 6
LC3B recruits ubiquitin to the PVM.
(A) HeLa WT, HeLa LC3B- and ATG5-knockout cells were infected with Pb mCherry (red). 6 hours post-infection, cells were fixed and stained with anti-ubiquitin antibodies (green) as a control. Control and knockout cell lines were transfected with RFP-LC3B (two lower panels) and infected 17 hours after transfection with P . berghei sporozoites expressing mCherry (red). Ubiquitin (green) was visualised using anti-ubiquitin antibodies. DNA was labeled with DAPI (blue). Cells were analyzed by confocal microscopy. Scale bar 10 μm. (B) Numbers of ubiquitin-labeled P . berghei parasites in non-transfected and in RFP-LC3B-transfected cells were determined by fluorescence microscopy. 77–103 parasites were analysed in the non-transfected HeLa cells and 53–83 parasites were analysed for the RFP-LC3B-transfected HeLa cells. Two individual experiments were carried out. Labeled parasites are expressed as percentages. Standard deviations are depicted. (C) Pearson’s correlation coefficients of ubiquitin and RFP-LC3B were calculated from seven individual P . berghei parasites in HeLa WT and LC3B -/- cells transfected with RFP-LC3B. Mean values are 0.4829 (HeLa WT) and 0.6871 (LC3B -/- ). Depicted are standard deviations.
Discussion
Selective autophagy regulates the degradation of specific cellular components and invading pathogens in autophagosomes. Classically, cargo to be degraded becomes ubiquitinated [ 40 ]. Subsequently, ubiquitin binding domain-containing (UBD-containing) receptors recognise the ubiquitinated cargo and mediate, via LIR domains, binding of LC3 family proteins [ 11 ]. This initiates the formation of an autophagosomal membrane around the cargo, which is finally degraded by fusion with lysosomes. In contrast to classical selective autophagy targeting intracellular cargo or pathogens, in P . berghei -infected cells, no formation of an autophagosomal membrane occurs. Plasmodium parasites reside in a vacuole that is directly targeted by autophagy marker proteins including autophagy receptors and ubiquitin [ 5 , 27 ]. Since no additional autophagosomal membrane is formed, it is safe to assume that receptor association with the PVM has a completely different function. A hint in this direction came from earlier work [ 5 ] showing that Plasmodium parasites try to avoid autophagic elimination by clearing the PVM of autophagy markers during parasite development [ 5 ]. This phenomenon was not only found for LC3 but also for ubiquitin and p62. Interestingly, at the same time, lysosome association with the PVM was strongly reduced suggesting that the parasite actively tries to avoid progression of autophagy by clearing the PVM of autophagy marker proteins. [ 42 ] It is tempting to speculate that the autophagy receptors are involved in PVM clearance but the deciphering of the molecular details of this event is beyond the scope of the present study and is a subject of current investigations.
Selective autophagy relies on an arsenal of receptors that target different cargo for degradation: p62 is known to recognise protein aggregates, bacteria, zymogen particles, midbodies and nucleic acids; NBR1 targets aggregated proteins, peroxisomes and midbodies for degradation; NDP52 labels bacteria, mitochondria and nucleic acids; OPTN is involved in degradation of protein aggregates, bacteria and mitochondria (reviewed in [ 43 ]). Autophagy receptors do not have an absolute specialization but often collaborate to target a cargo for degradation. p62 and NBR1 interact with each other in aggrephagy [ 41 ], pexophagy [ 19 ] and midbody ring degradation [ 44 ]. In xenophagy, a process that removes invading bacteria, p62 works together with NDP52 and OPTN [ 22 , 23 , 45 ]. NDP52 and OPTN locate on common subdomains on the ubiquitinated Salmonella bacteria containing vacuole, which raises the question whether these two proteins have similar roles in selective autophagy [ 22 ]. Ubiquitinated Salmonella bacteria are sensed by at least four autophagy receptors: p62, NDP52, OPTN and NBR1. With the exception of NBR1, they all appear to play a role in restricting growth of bacteria [ 17 , 22 , 23 , 45 ]. We have found that shortly after infection the PVM in the vast majority of P . berghei- infected cells is decorated with the autophagy receptor proteins p62 and NBR1 and to a much lesser extent with NDP52. The mechanism of their recruitment appears, however, to be completely different compared to that seen in bacterial infections.
We show that in contrast to our expectations, p62 and NBR1 binding to the PVM depends on LC3B. When we delete LC3B from HeLa cells, PVM labeling by p62 and NBR1 is diminished. Importantly, complementation of the LC3B knockout cell lines with RFP-LC3 fully restores PVM labeling by p62 and NBR1, confirming the central role of LC3B in this process in P . berghei -infected cells. Another important observation of this work is that depletion of LC3B does not have an obvious effect on autophagosome formation in general, confirming recent work in hexa knock out cells (HeLa cells knocked out for all LC3s and GABARAPs). In these cells, autophagosomes can still be formed and a certain level of autophagy is still possible [ 39 ]. We also did not observe an obvious effect on parasite growth in LC3B-negative HeLa cells (data not shown). It might well be that other members of the LC3 protein family partly compensate for the lack of LC3B. This is supported by the notion that in P . berghei -infected, LC3B-deficient HeLa cells, p62 and even more NBR1 association with the PVM was not completely abolished. However, when we deleted ATG5, which acts upstream in the autophagy cascade, the PVM of P . berghei -infected cells was never labeled with p62 but moderately labeled with NBR1. This indicates that ATG5-mediated lipidation and incorporation of LC3B or another member of the LC3 family into the PVM is necessary to recruit p62 and NBR1 and that NBR1 might be able to directly bind to other PVM proteins. Interestingly, ubiquitin recruitment to the PVM appears also to be dependent on LC3B presence in this membrane. This became most obvious when RFP-LC3B was co-transfected with GFP-ubiquitin. We reported earlier that in infected HepG2 cells a higher percentage of parasites exhibit ubiquitin labeling of the PVM [ 5 ]. It is possible that HepG2 cells express higher levels of LC3 than HeLa cells used in the present study. Together, our experiments show that in P . berghei -infected cells, the PVM labeling with autophagy proteins is reverted, with LC3B recruiting autophagy receptors and ubiquitin.
So far the role of receptor recruitment to the PVM has not been clear and one can only speculate on its function. For example, it might have a role in signaling pathways that promote survival of the host cell during parasite development. It has been shown that p62 can specifically interact with active RagC/D heterodimers, Raptor and TRAF6 and is able to activate the pro-survival factor mTORC1 on the lysosomal surface [ 46 , 47 ]. p62 can also activate the Keap1-Nrf2 pathway, which is a defense mechanism against oxidative and electrophilic stress [ 46 ]. Under normal conditions, Nrf2 is bound to the E3 ubiquitin ligase adaptor protein Keap1, which ubiquitinates Nrf2 and leads to its proteasomal degradation. When autophagic flux is compromised and p62 accumulates, Keap1 is sequestered by p62 and can no longer bind Nrf2, leading to increased Nrf2 signaling and down-regulation of oxidative stress. Since oxidative stress can be employed as a defense strategy against intracellular pathogens, activation of the Nrf2 pathway would support parasite survival. In fact, it has recently been shown that P . berghei -infected host cells do not show enhanced oxidative stress conditions [ 48 ]. It will now be interesting to investigate whether or not there is indeed a link between autophagy receptor signaling and oxidative stress regulation in P . berghei -infected hepatocytes.
Here we show that in the vast majority of P . berghei -infected cells, p62 and NBR1 are recruited to the PVM and that this clearly depends on the presence of LC3B. It will now be highly interesting but also very challenging to elucidate the function of this prominent interaction.
Supporting information
S1 Fig
LC3B knockout cells are able to undergo canonical autophagy.
(A) Representative western blot of non-infected HeLa WT, ATG5-knockout cells and two clonal LC3B-knockout cell lines left untreated or simultaneously treated with 10 μM chloroquine and 250 ng/ml rapamycin for 4 hours. (B) HeLa WT and HeLa LC3B knockout cells ectopically expressing GFP-Gate16 were left untreated or treated with 10 μM chloroquine and 250 ng/ml rapamycin for 4 hours. Fixed cells were stained with anti-GFP antibodies to visualise Gate16 (green) or anti-LC3B antibodies (red). DNA was stained with DAPI (blue). White arrows in the LC3B panel indicate Gate16-transfected cells. Yellow arrowheads in the enlarged pictures indicate autophagic structures where Gate16 and LC3B colocalise. Scale bar 20 μm.
(TIF)
S2 Fig
LC3B recruits p62 to the PVM.
(A) HeLa WT, HeLa LC3B- and ATG5-knockout cells were infected with Pbm Cherry (red). 6 hours post-infection, cells were fixed and stained with anti-p62 antibodies (green). All cell lines were transfected with RFP-LC3B (two lowest panels) and infected 17 hours after transfection with P . berghei sporozoites expressing mCherry (red). RFP-LC3B (red) and p62 (green) were visualised using antibodies. DNA was labeled with DAPI (blue). Cells were analysed by confocal microscopy. Scale bar 10 μm. (B) Numbers of p62-labeled P . berghei parasites in non-transfected and in RFP-LC3B-transfected cells were determined by fluorescence microscopy. 100–130 parasites were analysed in the non-transfected HeLa cells and 60–120 parasites were analysed for the RFP-LC3B-transfected HeLa cells. Two individual experiments were carried out. Labeled parasites are expressed as percentages. In the non-transfected cells, the two LC3B- and ATG5-knockout cell lines show significant less p62 associated with the parasite. In RFP-LC3B-transfected knockout cell lines, p62 association is not different to in RFP-LC3B-transfected WT cells. Standard Deviations are depicted.
(TIF)
S3 Fig
RFP-LC3B does not recruit GFP.
HeLa WT cells were simultaneously transfected with RFP-LC3B and GFP alone. Approximately 24 hours post transfection cells were infected with Pbm Cherry and 6 hours post infection cells were fixed and analysed by fluorescence microscopy. In the left panel RFP-LC3B and Pbm Cherry are shown. A white arrowhead points towards an LC3B-labeled parasite. The middle panel shows the GFP signal in greyscale. Scale bar 20 μm
(TIF)
S1 Table
Sequence analysis of three different LC3B -/- clonal cell lines.
Edited regions were amplified by PCR with the three different primer pairs indicated in the table. PCR products were cloned into a plasmid and sequence analysis of 24 individual plasmids was performed for each of the three LC3B knock out cell lines. 1 to 3 different alleles were found in each cell line.
(DOCX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Mental health is a state of well-being that enables people to achieve their potential, cope with daily stress, perform productive work and contribute positively to their communities [ 1 ]. It is a neglected essential component of health in low- and middle-income countries (LMICs). However, it is estimated that at least one billion people worldwide suffer from mental health problems (depression, bipolar disorder, anxiety disorders, posttraumatic stress disorder, schizophrenia, etc.) [ 2 , 3 ], which accounts for 32.4% of years lived with a disability and 15% of disability-adjusted life years (DALYs) [ 4 ].
Mental health problems constitute a huge global burden of disease, particularly in LMICs, most of which have health policies based on primary care services; however, there is a large treatment gap [ 5 ]. Studies show that 75–90% of people with mental health problems in these countries do not have access to the formal mental health care they need [ 6 – 8 ] for several reasons, of which the four main ones related to an insufficient annual national budget allocated to the mental health subsector, lack of integration of mental health provision into primary care in most LMICs, lack of services and stigmatization are developed below.
First, in LMICs, the share of the annual national health budget allocated to the mental health subsector remains below 1% [ 9 ]. Consequently, the availability of mental health services is very low. A study assessing the availability of services in five LMICs (India, Nepal, Ethiopia, South Africa and Uganda) showed that per every 100 000 inhabitants, there were 0.06 to 6.85 beds in inpatient mental health facilities and 0.04 to 2.70 psychiatric beds in general hospitals [ 10 ].
Second, in most LMICs, there is no offer of mental health at primary care level. However, the integration of mental health into primary care services could fill the gap in patient care. For example, in Guinea, where access to mental health care is considered extremely limited, with only five psychiatrists and 38 inpatient beds for twelve million inhabitants [ 11 ], the experience of the integration of mental health care in five health centers, all members of an NGO-driven mental health program, has proven beneficial. In primary health care (PHC) services, the proportion of patients with mental health problems seen at the general consultation increased from 2.7% to 3.9% between 2012 and 2017 [ 12 ]. In addition, this integration of mental health provision has improved the quality of care for patients treated at PHC centers, particularly in its patient-centered care dimension [ 13 ]. However, the fact that the mental health care is not integrated into PHC services in most LMICs makes the coverage of mental health services very low [ 8 , 10 ].
Third, due to a lack of formal mental health provision, patients resort to alternative solutions such as traditional medicine, herbal medicine and quackery [ 12 ]. Additionally, belief in the supernatural origin of diseases, including mental illnesses, leads patients to resort to fetishing, incantation and spiritual care practices [ 14 ]. Moreover, alternative approaches such as prayer, life coaching, yoga , tai chi and mindfulness meditation are sometimes recommended in hospitals for patients with psychic or psychosomatic disorders [ 12 , 14 , 15 ]. In addition to the therapeutic aspect, these alternative practices offer people a fundamental basis necessary for their psychosocial well-being [ 14 – 16 ].
Fourth, people with severe mental illness are often stigmatized within families, communities and even health services [ 17 ]. As a result, mental health remains poorly perceived by both the general public and health personnel, exposing people with severe mental disorders and psychosocial disabilities to discrimination and other forms of serious human rights violations.
Although mental health is one of the main urban health concerns around the world [ 18 , 19 ], in LMICs such as the Democratic Republic of the Congo (DRC), it is a low priority since the mental health subsector has several weaknesses, such as i) lack of funding for care and essential psychotropic drugs; ii) shortage of human resources for mental health; and iii) low availability of mental health data in the national health information system [ 20 ].
Based on 2020 World Health Organization (WHO) official estimations on the burden of mental disorders, in the DRC, the DALY (per 100 000 population) for mental and substance use disorders was 1 557.7 compared to other countries in the region such as Cameroon, Central African Republic, Congo and Gabon where the DALYs were respectively 347.7, 88.4, 85.7 and 33.7 [ 21 , 22 ], and the age-standardized suicide mortality rate (per 100 000 population) was 12.41 [ 21 ]. However, the country has only 6 public psychiatric hospitals and a dozen private mental health centers with 500 beds for nearly 90 million inhabitants, almost all of which are in located cities [ 23 ]. The coverage of mental health services remains very low (less than 5%); however, mental health provision is severely lacking in primary care services nationwide [ 24 ], and only 3% of PHC services have been able to integrate mental health care supply into their activity packages [ 24 ].
At the PHC level, no protocols for the care and assessment of mental health problems or official referral and counterreferral systems between PHC and mental health providers exist [ 24 ]. A 2006 report by the WHO together with the Congolese Ministry of Health shows that at most, 1 in 10 patients with a mental health condition attending PHC are referred to a mental health professional [ 23 ]. It is estimated that only 1% of PHC providers interact with mental health professionals once a year. At most, 5% of Congolese PHC providers interact with other psychosocial care providers, such as traditional healers and spiritual care workers [ 23 ]. These 5% of providers advise patients to consult traditional medicine and spiritual medicine because mental illness is considered to be of supernatural origin [ 25 ].
Regarding the burden of mental and psychosocial disorders and the important gap in the treatment of mental health problems around the DRC, the national mental health policy recommends integrating mental health care packages into the existing health services at the district level and assuring continuity of care [ 26 ]. Aligning with this national recommendation, the authorities of the health district (health zone, in the DRC) of Tshamilemba along with their development partners considered integrating mental health into the health services of this district based in the city of Lubumbashi in the Haut-Katanga province. It should be noted that for a total of 283 000 inhabitants, this health district does not currently have a dedicated psychiatric bed, nor does it have psychiatrists, psychologists, psychiatric nurses, or social workers trained in mental health.
To date, in the health district of Tshamilemba, no attempt has been made to integrate mental health into health services. Based on a study in the semirural district of Kinshasa by Mambanzi et al. [ 27 ], we could expect the gap in conventional mental health services to be filled by traditional medicine and houses of worship, despite the mistrust of intellectuals and biomedical practitioners and their lack of effective integration into the health care system.
To understand the mental health care situation in the health district of Tshamilemba, this study i) analyzed the existing demand for mental health care expressed by the populations throughout the assessment of the burden of mental health problems and care-seeking behavior, ii) described the available mental health care supply at the time of the study, and iii) critically examined the district’s operational response capacity to address mental health. The findings of this study can inform the policy design for integrating mental health into primary care in other urban health districts in the DRC and other LMICs.
Methods
Study context
The Congolese health system is organized into 3 levels: the central level, the provincial level and the operational level. The central level is responsible for defining policies, strategies, standards and guidelines. It supports provinces by providing advice, monitoring compliance and following up the implementation of the health development plan. The provincial level ensures the application of health policy and the translation of such policy and the related strategies and guidelines into operational instructions and technical sheets to facilitate their implementation in the form of actions in the health districts [ 28 ]. The operational level is responsible for implementing health policy, which is based on the PHC strategy [ 29 ].
This study was carried out at the operational level in the health district of Tshamilemba in urban areas of Lubumbashi. The 2020 consolidated operational action plan of the health district of Tshamilemba indicates that this district, which covers an area of about 42 km 2 and a population of 283 000 people, is divided into 13 health areas; counts 59 private clinics including for-profit (or commercial) and non-profit (e.g. confessional) facilities, and only one public primary care facility i.e., Tshamilemba PHC center, covering about 12 000 people. This clearly illustrates the huge (and problematic) extent of privatization of the first level of health care in the city of Lubumbashi in general, and in the health district of Tshamilemba in particular.
The Tshamilemba PHC center was established in 2012 and hosts training and research activities along with health services. It tests innovative experiments in the organization of PHC in the DRC. Because of its status as a learning and research site and the fact that it is the only public PHC service in the health district of Tshamilemba, this center was chosen as the site at which to implement a project for the integration of mental health care. In 2021, the analysis of activity reports of the Tshamilemba PHC center shows that 143 cases of tuberculosis, 433 cases of HIV/AIDS, 23 cases of diabetes and 34 cases of arterial hypertension were detected and managed. It is recognized that these chronic diseases can evolve into comorbidities with mental health problems. The analysis of activity reports also reveals an increase in the rate of utilization of general curative consultation from 0.26 new cases (NCs) per year in 2019 to 0.38 NCs per year in 2021.
Study design
A multimethod cross-sectional exploratory study was conducted to analyze the baseline mental health situation in urban areas of Lubumbashi, particularly in the health district of Tshamilemba. This cross-sectional descriptive and content analysis study provides useful information for the development of a strategy for integrating mental health in urban areas of the DRC.
Data sources and collection
Data were collected from three sources: (i) technical documents, (ii) key stakeholders on the mental health situation and (iii) residents of the health district of Tshamilemba. The technical documents (i.e., routine health information system canvas, consolidated district operational action plan, consultation registers, PHC centers’ activities reports) are from the health district of Tshamilemba. These documents were provided by the authorities upon request of the principal investigator (EMM). In addition, a soft copy of the document of health sector budget forecasts was shared by a senior official of the Ministry of Health. In total, 19 documents were selected that were directly related to mental health in the DRC and written in either French or English. These documents allowed us to gather quantitative and qualitative data ( Table 1 ). The documentary review was conducted by EMM and assisted by DCK (a local researcher from the health district of Tshamilemba) for two months, from December 8, 2020, to February 10, 2021. EMM together with DCK decided on the documents to be included according to whether they considered the documents relevant to the study depending on completeness, year of writing, and language of publication.
10.1371/journal.pone.0280089.t001
Table 1 Documents consulted and type of data collected.
Documents
Number
Year of production
Sources
Data extracted on:
Data type
Health Sector Budget Forecasts
1
2022
Ministry of Health
• Financial data of the mental health subsector
Quantitative
Consolidated District Operational Action Plan
1
2020
District management team of Tshamilemba
• Health demographics • Available resources
Qualitative and quantitative
Consultation registers
3
2020
District health services information system
• Patient’s health complaints at the medical consultation • Diagnostic presumptions
Qualitative and quantitative
Health centers activities reports
2
2020, 2021
• Use of services • Resources available
Quantitative
Routine Health Information System canvas
12
2021
• Epidemiological data on mental disorders • Availability of psychotropic drugs
Quantitative
Of all the documents reviewed, only the consultation registers of the Tshamilemba PHC center contained data likely to reveal the identity of the patients seen in general curative consultation. To ensure that this patient privacy data was kept anonymous and confidential, it was collected by a single local researcher (DCK), who is also a health care provider and the managing director of the same center. In the consultation registers, DCK extracted only the data useful for research (i.e., age, sex, complaints, diagnosis), which were transcribed into an Excel file prepared for this purpose. None of the data likely to reveal the identity of the patients (i.e., name, physical address) were extracted from the registers.
The quantitative data extracted from the reviewed documents were saved in an Excel file. The qualitative data were recorded in a notebook prepared for this purpose, copied into a soft file and saved on a laptop whose password was kept by the principal investigator.
Key stakeholders (including doctors, nurses, managing directors, district medical officers, community health workers, community leaders and health care users) in the mental health situation were invited to participate in focus group discussions (FGDs) [ 30 ]. These key stakeholders, constituting the convenience sample, had to meet the following criteria: i) live and/or work in the health district of Tshamilemba; ii) be at least 18 years old on the day of the survey; iii) agree to participate in the study; iv) be able to express themselves in French and/or Kiswahili; and v) declare that they have familiarity with the domain of mental health, even if only empirical.
Table 2 presents a summary of the characteristics of the participants in the FGDs. We organized five FGDs in the first quarter of 2021 in the health district of Tshamilemba. The groups were formed by taking into account cultural (living environments, proximity between actors, traditions), administrative (function and role of health professionals) and economic (allotted time, funds available, participant availability) imperatives [ 30 ]. Depending on who accepted the invitation to participate in the study, heterogeneous groups of 9 to 11 participants were formed according to their professional profiles. We formed heterogeneous groups because certain socioprofessional profiles (e.g., health care users) were underrepresented because they were reluctant to participate in face-to-face FGDs due to the emergence of COVID-19. To guarantee that everybody was able to raise his or her voice and to encourage the silent ones to speak, the moderator used two techniques: the round table and distribution of speech methods [ 30 ]. Moreover, to temper dominant speakers, the moderator kindly avoided allowing them to speak for a second time for the same question.
10.1371/journal.pone.0280089.t002
Table 2 Summary of characteristics of participants in the FGDs (N = 50).
#FGDs
Meeting places
Number of participants
Gender
Socioprofessional profile of participants (Number)
Years of experience (Min.–Max.) a
M
F
FGD1
Tshamilemba PHC center
11
7
4
Doctors (3) Nurses (6) Managing director (1) Pharmacy Assistant (1)
3–12
FGD2
Kabetha PHC center
10
4
6
Doctors (4) Nurses (6)
1–5
FGD3
District Central Office
9
6
3
District Medical Officer (1) Managing director (1) Nurses (5) Doctors (2)
4–11
FGD4
Community
11
8
3
Nurses (2) Community health workers (5) Community Leader (1) Health care users (3)
2–6
FGD5
Community
9
4
5
Nurses (2) Community health workers (2) Community Leaders (2) Health care users (3)
1–7
FGD, focus group discussion; PHC, primary health care; M, male; F, female; Min, minimum; Max, maximum.
a This data concerns only health professionals
To meet potential participants, an invitation accompanied by an information letter was sent to 90 persons seven days before the day scheduled for the meeting through the channel of community health workers based in the respective health areas. This letter included the objectives and representatives of the study, the FGD method, the confidentiality of the anonymity of the results, the overall description of the theme to be addressed and the practical arrangements for the meetings (place, date, etc.). Two days before the meetings, we tried to obtain verbal or telephone confirmation of those who would be present; 25 persons declined the invitation. The wish to avoid physical contact with other people during the COVID-19 pandemic and level of personal convenience were the two main reasons for declining. We developed the interview guide focused on three main themes: 1) demand for mental health, including the burden of mental health problems and care-seeking behavior, 2) mental health care supply available at the time of the study, and 3) the district’s operational response capacity to address mental health. This last theme was discussed mainly with health professionals and community health workers. The opinions of health care users were solicited to confirm (or not) the statements of professionals. The interview guide (see S1 File ) was developed in French. Once the questions were translated into the national language (Kiswahili), they were pretested in a neighboring health district to ensure that these questions could be understood by all participants.
After training the male and female investigators (MLM, MR, MTR) on the study protocol, interview tools, note taking, and observation and under the supervision of the principal investigator, the face-to-face discussions were organized and moderated by an independent researcher (public health physician, healthcare provider in a private clinic) (AKM) chosen for his expertise in conducting surveys. The FGDs were held at the PHC centers, at the central office of the health district and in the community. The interviews—which lasted an average of 1 h 45 minutes (±15 minutes)—were conducted during the day, between 8 a.m. and 5 p.m. Prior to study commencement, the interviewers introduced themselves to the participants in order to establish a relationship of trust. The discussions—in which only participants and researchers took part—were audio-recorded and the field notes made during and after FGDs by two research assistants (TKK, VIM) both doctors trained in health science and survey techniques, who originated from the city of Lubumbashi and have knowledge of the study context. During data collection, the main difficulty encountered was the restriction on gatherings of people imposed by the provincial government to limit the spread of COVID-19. Data saturation justified limiting the sample size to 50 participants.
The residents of the health district of Tshamilemba were the third source of data. A random sample in clusters was formed to carry out a household survey. The sample size was calculated for the two clusters at 769 subjects by applying the following formula [ 31 ]: n = Za 2 * c * ( pq )/( d 2 ), where: Za 2 = 1.96 2 ; c = 2 (number 2, means clusters factor used to minimize the effect of assumed heterogeneity of the study area); p = 0.5 and q = 0.5 (as the prevalence of mental health problems is not known in the study setting) and d 2 = 0.05 2 . A total of 591 participants responded, for a response rate of 76.9%. Participants had to meet the following criteria: i) live in the health district of Tshamilemba; ii) be at least 18 years old on the day of the survey; iii) agree to participate in the study; and iv) be able to express themselves in French and/or Kiswahili.
The registration of participants was performed by a door-to-door strategy conducted by community health workers.
The survey questionnaire (see S2 File ) we used, was previously pretested with a sample of 15 participants selected from a neighboring health district. This questionnaire included questions that aimed to explore the perceived and experienced psychosocial and mental health problems. Participants also gave their opinions on the possibilities offered by the local health system to respond to their requests for care.
Definitions of main mental health disorders categories used
In this study, we used the definitions presented by the National Mental Health Program of the DRC in its mental health module of the NHIS (National Health Information System). In this module, the Program refers to the WHO mental health Gap Action Program (mhGAP). The symptomatic approach has been adopted to clearly define these concepts, in order to facilitate their understanding by all non-specialists (e.g. community health workers, family members, primary care providers) participating in the study. These definitions are presented in Box 1 .
Box 1. Definitions of main mental health disorders
Anxiety disorders: in the past 6 months, regular (i.e. more often than not) experiencing disorders in which anxiety and concerns are associated with at least three of the following six symptoms: restlessness or feelings of excitation or nervousness, fatigue, difficulties in concentration, irritability, muscular tension, sleeping problems and restlessness. Stress disorders: patients experience fear when facing stressful events, feeling of helplessness, horror, fright, reliving traumatic memories (scenes, images); avoidance of traumatic memories and hypervigilance following adverse or traumatic situations. Depression: A disorder in which the patient experiences persistent sadness or depressed mood, fatigue, sleep problems, decreased energy and suicidal thoughts, anxiety, loss of interest or pleasure in normally enjoyable activities for at least 2 weeks. Substance use and related disorders: Disorders occurring in a person apparently under the influence of a psychoactive substance, manifested by, for example, lack of energy, agitation, inability to sit still, inarticulate speech; or signs indicative of substance use such as injection marks, skin infection, disordered appearance; requesting a prescription for sedative medication (sleeping pills, opioids); having financial difficulties or criminal problems; or difficulty with work, at household level, or in habitual social activities. Suicidal thoughts and attempts: Disorders in which patients experience extreme hopelessness and despair; with (past or present) thoughts, plans, or acts of self-mutilation, suicide attempts, and any other condition associated with chronic pain or extreme emotional distress. Neuroses: Disorders in which patients experience multiple physical complaints, without apparent medical explanation or physical cause, and that do not correspond to a known illness; experience of fear without apparent reason, nervousness, difficulties in living in community. Psychoses: Disorders in which patients exhibit incoherent or irrelevant speech; delusions (unreal ideas believed to be true); hallucinations (hearing voices and seeing things that do not exist); feelings of withdrawal, restlessness, disorganized and aggressive behavior, fixation on false beliefs not shared by others in the person’s culture, feeling that one’s thoughts are being stolen, or imposed; tendency to isolate oneself and neglect habitual tasks and responsibilities related to work, school, domestic or social activities, with lack of awareness that one has mental health problems.
Data management and analysis
After collection, the quantitative data compiled in the predefined data recording grid were entered into an Excel file and then processed by the principal researcher (EMM). Each of the two researchers chosen (EMM and DCK) independently assessed the scientific value of the information collected, which enabled them to obtain data deemed relevant. For these data, frequency measures (rate, absolute frequency, proportion, etc.) were computed. Clearly, in order to understand the extent of the burden of mental health problems and health care-seeking behaviors, we calculated the morbidity indicator (proportion of mental health cases), the relative frequency of mental health complaints reported by patients in PHC centers and health care utilization indicators.
To avoid the loss of the information collected and secure confidentiality, the qualitative data collected were stored on two computers chosen by the principal researcher. A unique digital folder was created and copied to each of the two computers and then password protected. The principal researcher was responsible for managing these data and analyzing them. The computer media (laptop and flash disk) containing the transcribed data, including the NVIVO file and the quantitative data, were kept in a locked cabinet to which only researchers (EMM and DCK) had access.
During transcription, the respondents’ identifiers were removed, and data were encrypted and managed anonymously by the researchers. The existing link between the code and the participants was removed, and the encryption key was saved in a separate file.
All verbatim transcripts were entered and stored in an NVIVO database for analysis. The quality and accuracy of the transcriptions were checked by the principal investigator (EMM) by randomly listening to a few recordings, and the transcriptions were then corrected if necessary.
Following the first transcriptions, clarifications were sought from the independent researcher (AKM) who had moderated the discussions, which allowed us to ensure that the questions were correctly asked and answered. The research assistants (TKK and VIM) separately read the transcripts and developed codes according to the three central themes identified in advance: 1) burden of mental health problems, 2) mental health care supply, and 3) district’s operational response capacities. The coding process was supervised by two members of the research team (EMM and DCK). A thematic analysis of the data was performed using NVIVO, and the key themes and subthemes from the interviews were used to structure the study findings. After data analysis, participants were invited to provide feedback on the findings.
For this research, to understand the complexity of the baseline mental health situation in the city of Lubumbashi, two triangulation methods were used, namely, the triangulation of sources, which consisted of analyzing the different documents (technical documents and documents of health sector budget forecasts) and key stakeholders (doctors, nurses, managing directors, community health workers, etc.), and the triangulation of data collection methods, which consisted of combining the two collection methods (documentary review and FGDs). The combination of triangulation methods applies when using the multiple methods approach to understanding complex situations [ 32 ]. The use of the aforementioned triangulation methods enabled us to reach, at the end of both the documentary review and the interviews held with the key stakeholders on the mental health situation in Lubumbashi, the ‘empirical’ saturation of the data.
Ethical considerations
The protocol for this study was approved by the medical ethics committee of the University of Lubumbashi (No. UNILU/CEM/034/2021) and the Institutional Review Board of the Institute of Tropical Medicine-Antwerp (IRB/AB/AC/022/1468/21). The study design was discussed with the district management team, which gave its permission, after a favorable opinion of the provincial health division of Haut-Katanga, to implement the project.
Both the interview guide and questionnaire we used were anonymous. The data extracted from the documents were already aggregated and anonymized. In the consultation registers, data likely to reveal the identity of the patients, such as names and physical address, were not collected. This made the collected data strictly anonymous for further data management steps (i.e., processing and analyses). Participation in the interviews was voluntary, and all targeted individuals freely consented to their participation.
Written informed consent was sought and obtained from each participant after reading the contents of the information letter. Before any recording and/or note-taking, the research assistants again requested permission from the participants.
Members of the research team (interview moderator, research assistants, coders and data analyst) who participated in the collection, coding and/or data analysis were trained on ethical aspects of research (respect for privacy, confidentiality, etc.), which they have undertaken to respect. Participants were informed that recordings and transcripts would be kept for a maximum of two years and then destroyed after the study results were published.
Results
Characteristics of household survey participants
Overall, 591 people participated in the household survey. The average age of the participants was 36.6 (SD ±11.7). Male participants were overrepresented (55.2%). Household heads represented 72.4% of the sample. A total of 249 (42.1%) participants did not have a paid job, and 294 (37.9%) worked in the liberal sector.
Demand for mental health care: Burden of mental health problems and care-seeking behavior
Respondents in the FGDs stated, in various ways, that mental health problems exist and represent a heavy social burden for the community and are a public health concern in the health district of Tshamilemba. Some participants stated during the FGDs that
‘We have family members with this type of disorder [namely , mental illnesses] ’ [FGD4].
They expressed the importance of mental health problems in the following terms:
‘The problem of mental disorders is there; it is emerging . It has become very serious , especially with the outbreak of the COVID-19 pandemic . The [seriously] mentally ill threaten the tranquility , and even the life of those around us; they are very numerous in our communities , and many of them are roaming around the city of Lubumbashi . Mental health conditions affect various categories; all our health areas are concerned…’ [FGD1].
In addition, health care providers recognized that they receive and treat people with mental health problems: ‘We have treated those patients with mental health problems in our health services’ [FGD1].
The care utilization indicators, resulting from the review of the selected technical documents, are presented in Table 3 below. From these results, we note that in 2020, of all general curative consultations carried out at the Tshamilemba primary care center, the proportion of patients with mental health problems out of all curative consultations was 5.3%. As well, the rate of utilization of curative mental health care at the general curative consultation was 14 new cases (NCs) per 1 000 inhabitants per year.
10.1371/journal.pone.0280089.t003
Table 3 Indicators of utilization of curative care at the Tshamilemba PHC center in the health district of Tshamilemba in 2020.
Indicators of utilization of the curative consultation in 2020
Statistics
Population of the health subdistrict of Tshamilemba (estimated in 2020)
11 975
Number of NCs at the curative consultation at the Tshamilemba PHC center in 2020 a
3 041
Rate of utilization of the general curative consultation at the Tshamilemba PHC center (NCs/inhab/year) 1
0.25
Number of NCs with mental health problems seen at the curative general consultation at the Tshamilemba PHC center
162
Proportion of cases consulting for a mental health problem b out of all recourse of general curative consultation 2
5.3%
Rate of utilization of curative mental health care estimated at the general curative consultation (NCs/inhab/year) 3
0.014
a Based on our count
b Excluding epilepsy
1 Numerator: Number of new cases (NCs) at the curative consultation at the Tshamilemba PHC center in 2020; Denominator: Population of the health subdistrict of Tshamilemba (estimated in 2020) (n = 11 975).
2 Numerator: Number of NCs with mental health problems seen at the curative general consultation at the Tshamilemba PHC center; Denominator: Total number of NCs at the curative consultation at the Tshamilemba PHC center in 2020 (n = 3 041).
3 Numerator: Number of NCs with mental health problems seen at the curative general consultation at the Tshamilemba PHC center; Denominator: Population of the health subdistrict of Tshamilemba (estimated in 2020) (n = 11 975).
Although it is a neurological condition, epilepsy is frequently encountered in PHC centers. Out of a total of 3 041 patients seen at the Tshamilemba PHC center in 2020, there were 40 cases of epilepsy, representing a proportion of 1.3%.
During the discussions, providers of one health service stated the following:
‘From a monthly average of 250 new cases seen in curative consultations , we receive up to 30 patients with mental health problems and/or complaints such as insomnia and drug addiction . ’ [FGD1].
The providers interviewed said that they noticed a gradual increase in mental health complaints at the general curative consultation during 2020. The documentary review confirmed this increase ( Fig 1 ). For example, in December 2020, out of a total of 312 patients admitted to the curative consultation at the Tshamilemba PHC center, 127 reported only mental complaints, 131 patients presented only physical complaints and 54 patients reported physical complaints associated with mental complaints.
10.1371/journal.pone.0280089.g001
Fig 1
Evolution of mental complaints at the general curative consultation at the Tshamilemba PHC center in 2020 (data source: Consultation registers).
The psychosocial consequences of mental health disorders are perceived in the living environments of participants. During the interviews, they mentioned some frequently observed consequences in the following terms:
‘Mentally ill people engage in self-aggressive acts , even suicide , and in heteroaggressive acts , even homicide . They disrupt public order , offend against public morals , and engage in begging . They are seen wandering around the city , even adopting a life on the streets (homelessness) . They destroy family or community property . They stigmatize and shame their families , who end up abandoning them and even wishing them death…’ [FGD2].
During the household survey, we explored the extent of the main mental health disorders felt/observed by respondents (n = 591). The corresponding results are shown in Table 4 .
10.1371/journal.pone.0280089.t004
Table 4 Distribution of mental health cases reported by respondents in the household survey (n = 591).
Mental health disorders reported by respondents
Statistics
n
Percentage
95%IC
Did not report mental health problems 1
297
50.3
[47.6–52.8]
Did report mental health problems 2
294
49.7
[47.1–52.3]
Anxiety disorders 3
12
4.0
[2.4–5.6]
Stress disorders 3
100
34.0
[30.2–37.8]
Depression 3
81
27.6
[24.0–31.2]
Substance use and related disorders 3
99
33.7
[29.9–37.5]
Suicide thoughts and attempts 3
19
6.5
[4.7–8.3]
Neuroses 3
55
18.7
[15.7–21.9]
Psychoses 3
7
2.4
[1.2–3.6]
1 Numerator: Number of respondents who did not reported a mental health problem (n = 297); Denominator: Total respondents (N = 591).
2 Numerator: Number of respondents who reported mental health problems according to the symptoms felt/observed (n = 294); Denominator: Total respondents (N = 591).
3 Numerator: Cases reported according to the symptoms felt/observed; Denominator: Total respondents (n = 294).
n = number of participants who responded
Of the 591 participants who responded to the questionnaire, 294 (49.7%) reported mental health problems either in themselves or in a family member. Of those who reported a mental health disorder (n = 294), the majority of participants and their family members were affected by stress disorders (including posttraumatic stress disorders) (34.0%; 95%CI: 30.2–37.8); substance use and related disorders (33.7%; 95%CI: 29.9–37.5); depression (27.6%; 95%CI: 24.0–31.2); and neuroses (18.7%; 95%CI: 15.7–21.9).
Mental health care supply available
Similar to many other urban health districts across the country, mental health has not yet been integrated into PHC in Lubumbashi. The mental health provision that exists in the city of Lubumbashi is offered in the following specialized mental health services: the Neuropsychiatric Center ‘ Docteur Joseph Guillain ’ of Lubumbashi (which has one neuropsychiatrist and one mental health nurse), the Neuropsychiatric Service of the Provincial Hospital Jason Sendwe (which has two neuropsychiatric interns) and the Neuropsychiatric Department of Lubumbashi University Clinics (which has one psychiatrist and one neurologist). These hospitals are private (not-for-profit) and public secondary and tertiary facilities, respectively, which can provide more complex treatments, including inpatient services during a mental health crisis. However, access to these facilities seems very limited due to the very high costs of the provided services. According to the activity reports consulted, in these health services, whose capacity for psychiatry services varies between 40 and 80 beds, 80% of the pathologies treated are mental disorders (depression disorders, mood disorders, manic episodes, delusional disorders, anxiety disorders, psychoactive substance abuse, puerperal psychosis, schizophrenia…) and 20% are neurological conditions (epilepsies, chronic headaches, pyramidal syndrome, somatoform disorders…). Nevertheless, the care packages offered in these hospitals have not been clearly described.
The minimum package of health services is offered in PHC centers of the health district of Tshamilemba. However, it is incomplete compared to the normative package, as it does not officially include mental health services. The content of this activity package is presented in Box 2 .
Box 2. Content of minimum package of activities currently offered in the PHC centers of the health district of Tshamilemba
The content of minimum package of activities currently offered in the health services of the health district of Tshamilemba consist of i) curative activities, including curative consultations, diagnostic services (ultrasound, medical laboratory), care for patients suffering from common diseases such as infectious diseases (malaria, typhoid fever…), trauma and violence, childbirth assistance and birth care, and management of chronic diseases (tuberculosis, HIV/AIDS, diabetes, arterial hypertension, etc.); general surgery activities; and referral and counterreferral activities, and ii) preventive activities relating to reproductive, maternal, new-born, child and adolescent health, vaccination, preschool consultation; health promotion activities such as family planning, behavior change communication, distribution of insecticide-treated mosquito nets; and rehabilitation activities such as the nutritional rehabilitation of children.
The interviewed health care providers said that they receive mental health cases for which they determine medical treatment. Epilepsy is the main condition and is treated with drugs such as diazepam and promethazine. Although it is classified as a neurological disorder, the providers consider epilepsy to be a mental disease. In addition, they have noticed that past attempts to treat patients with other mental health disorders (depression, psychoses…) by administering psychotropic drugs (amitriptyline, chlorpromazine) often failed, with patients returning to their family and community with unmet needs, showcasing that treatments for psychiatric illnesses are grossly lacking. Since then, they have decided to start referring them ex officio to provincial hospital Jason Sendwe, without offering any medical treatment. Due to the abovementioned limited financial accessibility, many patients resort to traditional healers and/or spiritual care interventions. The providers also expressed their fears regarding the mood instability and aggressiveness of some patients. From their responses, one could note the following:
‘We often wish to initiate the management of these cases , but we fear that it will be of poor quality … Currently , apart from epilepsy , which we treat well , we do not offer other mental health care for several technical , sociocultural and economic reasons . Patients who do not find their expectations met return home; some go to churches for prayer sessions , and others consult directly with traditional healers and diviners ’ [FGD2].
According to the participants, patients attending PHC services in the health district of Tshamilemba are sometimes not taken care of for several reasons, including the following:
‘Lack of integration of the mental health care package [containing curative activities such as the identification of cases , curative consultations , diagnosis and medical care , neuropsychiatric emergencies , basic psychosocial care , psychological consultation , psychological first aid , and referral/counterreferral; preventive activities such as aftercare follow-up of cases , and social reintegration; and promotional activities including home visits , mental health awareness , psycho-education , reporting , monitoring and supervision] into the minimum package of activities , which gives providers no ability to perform adequate acts of care , no mental health information and training for providers , and no supervision or technical support from the National Mental Health Program; providers’ poor perception of mental diseases; and the perception that mental health is a specific task of specialists’ [FGD3].
All participants (community participants and health care providers) expressed the need to see mental health care supply become available in primary care services because they believe that it is currently lacking. Participants in the study stated the following:
‘… we want psychiatric care to be integrated into health centers because when a family member is suffering from a mental illness , we often do not know where to take them . Sometimes we take them to a traditional healer or church for prayer . Either we give them medicine (paracetamol , diazepam , etc . ) ourselves at home without success… or sometimes we [family] do not do anything… We are afraid to go to the provincial hospital (neuropsychiatry department) because the care there is very expensive’ [FGD5].
Operational response capacity of the health district of Tshamilemba
The review of the national mental health system status report, consolidated district operational action plan and the activity reports of PHC centers showed that there is an absence of adequate and accessible community and hospital mental health services and specific community psychosocial support networks. It showed the absence of nonprofit organizations active in the field of mental health, as well as lack of associations of former patients or relatives of mental patients. The health areas have functioning health development committees.
Concerning mental health services, the qualitative data analysis showed that people with mental health conditions and/or psychosocial disabilities are not covered by formal health services. Community mental health care is not formally offered. Respondents mentioned that ‘access to essential psychotropic drugs is very limited for people who need them’ [FGD2].
Analyzing the domain of mental health in the PHC, participants stated the following:
‘Mental health is not integrated into the minimum package of activities of all the PHC services of the health district of Tshamilemba , and there are no formal links between PHC providers , psychiatric specialists and informal psychosocial care providers such as traditional healers’ [FGD5].
It is not currently possible to receive treatment for mental health conditions in primary care.
Participants of the health district management team declared the following: ‘ The Haut-Katanga provincial health division does not have operational mental health coordination led by a focal point who can oversee the integration of mental health activities in operational action plans . The primary health care supervisor of the health district of Tshamilemba does not currently include the mental health component in his activity package’ [FGD3]. They said there is a need to ‘strengthen the capacities of primary care providers on psychosocial and mental health care; supply health services with essential psychotropic drugs; harmonize diagnoses , case management and information tools; create links between primary care providers and mental health specialists; and provide health services with transport to escort patients to the psychiatric ward of provincial hospital Janson Sendwe’ [FGD3].
During the interviews, health care providers stated the following: ‘Our capacities for the identification and management of mental disorders , mental health care and the provision of psychosocial support are limited to what we acquired during our school or academic training’ [FGD1].
The district health services are financed mainly by subsidies from the Congolese state, their own income, support from partners and rare donations. The Congolese state pays salaries and bonuses; however, these are insufficient to cover the district’s daily needs. Analysis of the country’s 2022 finance bill indicates that the overall national budget was US$9.9 billion (US$1 = CDF 2 093). Yet, the health sector budget forecast documents we analyzed revealed that the share of the national budget allocated to the health sector in 2022 was US$225 million. Of this amount, the share specifically allocated to mental health care was US$39 000. However, this budget is used mainly for the payment of salaries and the functioning of the central level of the National Mental Health Program (NMHP), and it is considered very insufficient to carry out activities at the operational level. Respondents indicated that ‘patients [e . g ., epileptics] followed at the Tshamilemba PHC center pay for their care out-of-pocket since they are not covered by any social protection system . This constitutes a barrier to access to care for other mentally ill patients ’ [FGD1].
Regarding public awareness raising and linkages with other sectors, participants said, ‘We do not do mental health promotion in our district because we do not know what to tell people , and there are no nonprofit organizations active in the field of mental health in our district’ [FGD2].
Discussion
The results of this study, which aimed to analyze the existing demand and supply of mental health care in the urban health district of Tshamilemba and to examine the district’s operational response capacity to address mental health show that the situation of mental health is worrying and that response capacities are weak.
This study highlights that the burden of mental health problems is significant in this urban area. In the household survey, half of the respondents reported mental health problems either themselves or in their family. Documentary analysis showed that in 2020, 5.3% of all general consultations in PHC centers in Tshamilemba concerned mental health. According to the 2019 Global Burden of Disease study [ 33 ], the prevalence of mental disorders including substance use disorders in the DRC is estimated at 13.23% of the population, i.e. 11.42 million people. This prevalence is likely to be underestimated because except for the provincial health divisions of North Kivu and South Kivu which are currently connected to the District Health Information Software 2 (DHIS2), in the other provincial health divisions, mental health data is not routinely notified across primary care health services. Moreover, the epidemiological burden would probably be much heavier if population surveys were carried out, because the people in Congo generally do not consult for minor mental health conditions. There are currently no specific survey data available to determine the real extent of the situation. The present research precisely aims to filling this gap.
Surveyed participants stated that they and their family members were mostly affected by stress-related disorders, substance use and depression. There is however need for a more thorough investigation of the precise nature of the burden of mental health in the DRC in general, and in Tshamilemba district in particular.
The results indicated that only 1.4% of people with mental disorders had access to curative mental health care; thus sparking the debate on the very poor availability and accessibility of mental health care in the DRC. Based on this result, we can reasonably well assume that there is a significant mental health treatment gap in the Tshamilemba district, certainly when compared to the general figure of 10% of people with mental disorders accessing evidence-based treatments in LMICs overall [ 34 ].
Epilepsy is the most common neurological disease treated in PHC services in the health district of Tshamilemba. According to the standard classifications, epilepsy is a neurological disorder and not a mental health disorder [ 35 ]. However, in the Congolese context, it is considered a mental disease by health professionals and the community. This is a very interesting finding in itself. In addition, we believe that its perception by the population as a mental health disorder may mean that its successful clinical management could improve the confidence of care users in mental health services. This confidence would lead (potential) patients with mental health problems to accept and use the mental health care services offered. Furthermore, the findings pointed to the fact that epilepsy is currently not properly taken care of in the Tshamilemba district. Indeed, no effective anti-seizure medicines are available besides diazepam, which is used only for acute cases. The unmet need of epilepsy treatment is therefore considerable. The majority of people living with this neurological condition can unfortunately not access proper care. As in most LMICs where the treatment gap exceeds 75% [ 36 ], there is need to integrate epilepsy care in existing primary health care services.
Strengthening the technical capacity of health districts can reduce the burden of mental health problems [ 37 , 38 ]. This includes improving mental health management by expanding the minimum package of activities of PHC services. This would allow patients to receive care in an open setting, maintain a connection with their families and remain productive, thus helping to reduce the related stigma [ 38 ]. In this regard, there is a need to either integrate traditional medicine practices and spiritual care interventions into primary care services or to collaborate with providers of such unconventional care in mental health provision.
However, once integration is understood as a technique that promotes the provision of mental health care only in hospital settings and by formal care providers, it is difficult to achieve, especially for traditional healers, because of problems related to medical beliefs, philosophies and practices [ 39 ]. Therefore, making integration an opportunity for training and coreferral patients, encouraging collaborative knowledge sharing, and extending invitations to traditional healers to treat specific diseases in hospitals are possible avenues [ 39 ].
According to the beliefs of the inhabitants of Lubumbashi, traditional African medicine remains the main recourse for treating mentally ill people. Research indicated that in the DRC, the first contact of mentally ill people with a health care provider is with traditional healers and/or spiritual workers (priests, lay pastors or imams), and that only as last resort modern health professionals such as psychiatrists and psychologists are contacted [ 40 ]. Also in the Tshamilemba district, traditional healers are the main source of care for mental health problems; future research into the effectiveness of their services remains however to be done.
The study showed that beyond the willingness of decision-makers to integrate mental health into PHC services, there is a strong lack of the necessary resources. Currently, in the city of Lubumbashi, there is only one neuropsychiatrist for the nearly 2 700 000 people. There are a few medical doctors in this city who are currently specializing in neuropsychiatry at the university clinics of Lubumbashi, but the exact number is not known. This problem of a lack of specialized human resources is a national one, since the DRC has a total of 102 neuropsychiatrists [ 23 ] per 89 561 403 populations, i.e., 1 psychiatrist per 878 053 people, although on a global scale, there are 1.2 psychiatrists per 10 000 people, with an overrepresentation of psychiatrists in high-income countries [ 39 ]. It should be noted that in LMICs, the scarcity of specialized human resources and their inequitable distribution between urban and rural areas mean that very few people with mental illness receive the mental health care they need [ 41 ].
Researchers [ 42 ] suggest (i) training or building capacity and delegating tasks to lay primary care providers, other lay health workers (social workers, traditional healers, spiritual care workers, etc.) and community actors such as community leaders and community health workers, family and community members. This delegation of tasks should be done in stages. It should follow the correct strategy of providers and nonproviders of care, (ii) ensure that primary mental health care also includes brief psychotherapeutic interventions, (iii) promote recovery-oriented community interventions for people with chronic disabling mental disorders, (iv) conceptualize training as a continuous process of strengthening clinical skills through supervision, (v) involve community partners in psychosocial interventions, (vi) integrate changes toward primary mental health care into broader health policy reforms, and (vii) promote intersectoral approaches to address the social determinants of mental health. Most of these recommendations are feasible in the Congolese context since they have been successfully tested in the health district of Lubero in the eastern region of the DRC [ 24 ].
Mental health legislation often reflects a commitment by countries at the highest level to protect people with mental illness and psychosocial disabilities from human rights violations [ 23 , 37 ]. However, our results show that in the DRC, there is no mental health law. This means that alleged perpetrators of violations of the rights of psychiatric patients within families, communities and health services are not prosecuted. Thus, there is an urgent need for the country to accelerate the process of developing a mental health legislative framework that can open a pathway to the promotion of mental health and the rights of people with mental health problems and, by extension, equitable access to mental health care in a spirit of universal health coverage.
An important challenge remains to find a package for which all users of primary mental health care will be able to pay, given that 60–70% of Katanga’s population lives on US$0.80 per day [ 43 ].
The lack of specialized services can be filled by the provision of primary mental health care as the first line of defense. However, the relationship between formal health services and traditional medicine and spiritual care structures seems to be difficult at present due to different beliefs about the origin of diseases, especially mental illnesses. This raises the question of whether it would be appropriate to envisage collaboration between these nonconventional health structures, operating on their own, and the primary care services, or to simply encourage collaboration with their facilitators by reflecting on a model of coordination with the primary care services [ 44 ].
Researchers believe that the strategy of integrating mental health into primary care is effective and can improve the quality of care at PHC services [ 45 , 46 ]. However, such a strategy requires sufficient financial resources to establish and maintain a mental care offer in PHC services [ 38 ]. However, our findings showed that the state budget currently allocated to the mental health subsector, which is very insufficient, is mainly used only for the functioning of the national office of the NMHP. It is important to reflect on the optimal mechanisms for financing mental health to avoid those initiatives taken by external partners suffering from sustainability [ 47 ].
To date, access to medication for psychiatric disorders and epilepsy is very limited in the health district of Tshamilemba. According to a 2022 country analysis [ 48 ], the lack of availability of essential antipsychotic and neuroleptic drugs, combined with their prohibitive cost, contributes to further worsen the treatment adherence.
Our findings revealed that access to counseling and psychotherapy was virtually non-existing in Tshamilemba district. These results corroborate those of the 2022 country analysis which affirmed that the possibilities of counselling and psychotherapy in the DRC are extremely poor [ 48 ]. However, some local or international non-profit organizations provide some level of support services for victims to sexual violence. Additionally, mental health and psychosocial support is a growing concern of humanitarian organizations, especially in conflict and post-conflict areas.
Stigmatizing attitudes vis-à-vis mental health problems are unfortunately (still) quite dominant in the community. These attitudes are related to the prevailing belief that mental disorders are linked to supernatural causes [ 20 ] such as witchcraft, sorcery, magic, bad spirits, bewitchment, transgression of taboos… These beliefs are largely shared by family members of people with mental health problems. Primary care providers acknowledge their limitations on how to address mental health problems. This contributes to leading people facing mental health problems to seek help from traditional healers and/or spiritual workers [ 49 ].
Strengths and limitations
A first strength of our study is that, to the best of our knowledge, it is the first ever conducted in the city of Lubumbashi—and perhaps in the whole of the DRC—to analyze the demand and supply of mental health care, as well as the operational capacity of the health districts to address the huge unmet mental health needs of people. Our findings may inspire researchers, policy makers and health managers and constitute a basis for further reflection on the development of more evidence-based strategies to improve access to quality mental health care at the primary level.
A second strength of our study lies in its methodological set-up. The multimethod approach used, with triangulation of sources, methods and data, contributes to enhance the validity of our findings.
However, the levels of family and community participation in this study were low, which did not allow for an in-depth analysis of the role that these individuals play in the provision of mental health care. This is partly due to the context of the emergence of COVID-19, which is the period during which the current study took place. Additionally, missing from this study is the voice of traditional healers, although it is known that in the context of a lack of conventional medical care, these actors offer care. These aspects deserve to be considered in future studies.
Family and community members were a minority in the population included in the interviews conducted. Considering that these individuals may have little knowledge of the mental health issues being discussed, it is possible that they were influenced by the opinions expressed about health care providers. This was a limitation of the FGD variant (heterogeneous FGD) that we used. Indeed, soliciting perceptions from family and community members in the presence of health care providers could expose patients to a risk of response bias. This risk of bias was minimized by the style of moderation, which consisted of giving the floor to each participant and ensuring that everyone expressed themselves freely.
The classification of mental health disorders adopted herein almost follows the logic of the WHO’s mhGAP intervention guide, which is not exempt from the risk of bias. Furthermore, the reluctance and hesitation of participants during this period when physical contact restrictions were in place could have increased the risk of over- or underreporting [ 50 ]. To minimize these latter biases, we selected interviewers who were familiar with the study context and who are also well known by community members.
Conclusion
Our study established that the demand for mental health care in the Tshamilemba urban district in the city of Lubumbashi increased over time, but that, at the same time, the formal supply of mental health care remains poor. In Tshamilemba, traditional medicine remains the main source of mental health care provision. The current lack of evidence-based treatment results from a combination of poor operational capacity of primary care facilities, lack of resources, low priority of mental health care at policy level, and, last but not least, the reluctance of people with mental health problems to seek care given the high level of stigmatization of mental health issues.
The unmet mental health care needs are significant, and the responses to address them are grossly insufficient and inadequate. The need for urgent action in the area of mental health in this urban district, and undoubtedly throughout the city of Lubumbashi, is clear indeed. Among the 12 key shifts recommended by the WHO [ 51 ] to transform the mental health situation, three seem particularly relevant for our study setting: first, invest in universal health coverage policies that explicitly incorporate mental health care; second, raise community awareness for mental health problems and fight stigmatization and discrimination; and third, integrate a basic package of mental health care into existing primary care services. As Alegría et al. [ 52 ] point out rightly, such an ambition endeavor requires investing in the health workforce through task‐shifting, training and supervision of primary care providers, delivering care close to where people live rather than expecting them to travel long distances to access residential psychiatric care. Our findings showed that young people suffered less than adults from substance use disorders and depression. Yet we generally know that young age is not a protective factor against these disorders. There is a need to explore why this would be the case in this district of DRC.
Supporting information
S1 File
Focus group discussion-interview guide.
(DOCX)
S2 File
Household survey questionnaire.
(DOCX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Light affects nearly all biological activities such as finding food, avoiding predators, migration, and mate selection and is central to driving visual processes. Since the light available to an organism varies with behavior, habitat, and season, a myriad of photoreceptors have evolved, which are able to operate, often in the same eyes, over changes in illumination (irradiance) by more than a factor of 10 11 [ 1 , 2 ]. Coupled to this are changes in the spectral quality of light further dependent on elevation of the sun or moon in the sky both above and below the horizon. For example, at twilight with the sun below the horizon, a relative increase at blue and red wavelengths emerges as yellow wavelengths are selectively absorbed by atmospheric ozone, a phenomenon termed the “Chappuis Effect” [ 3 , 4 ]. Light in the sunlit shallow epipelagic (0 to 200 m) aquatic domain can be even more complex with even relatively small changes in an organisms’ vertical position resulting in very large changes in irradiance and spectral composition of light. For example, light intensity decreases by approximately 1.5 log units for every 100 m in clear ocean water, narrowing from full spectrum to the blue region [ 5 ].
Vision and light detection have evolved in parallel with the circadian clock, a self-sustained molecular machine reliant on positive and negative autoregulatory feedback loops and synchronized to external environmental cycles (most commonly light) via entrainment [ 6 , 7 ]. In marine organisms, other entrainment cues (e.g., tides, temperature, and salinity) are also often implicated [ 8 ]. Of particular significance is that the clock relies on the day–night cycle for temporally structuring both short-term processes (e.g., daily foraging) and long-term processes (e.g., gametogenesis) via photoperiodism [ 9 ]. Importantly, spectral cues characteristic of twilight can provide sufficient information to entrain circadian clocks [ 10 , 11 ]. For example, the blue dominant solar spectrum characteristic at twilight is consistent with the spectral absorption of cryptochromes known to be widely involved in circadian entrainment [ 12 – 14 ]. Light, therefore, plays multiple roles, among them vision and circadian clock entrainment, which, in combination, provide highly evolved and accurate spatial and temporal information for most behavioral and physiological processes.
Organisms inhabiting high latitudes experience a greater annual photoperiodic range than those at lower latitudes. At the poles, the sun is either permanently above or below the horizon for large parts of the year (Midnight Sun and Polar Night, respectively). During the Polar Night, traditional definitions of day and night and seasonal photoperiod become irrelevant since there are only “twilight” periods. These are defined by the sun’s elevation below the horizon at midday [ 4 ] and therefore considered hereafter as simply “midday twilight” ( Fig 1A ). Observations suggest biological relevance of midday twilight. Indirect acoustic measurements in Svalbard fjords during the Polar Night have revealed diel vertical migration (DVM) of zooplankton [ 15 – 17 ] and lunar vertical migrations (LVMs), which are also reflected across the Arctic Ocean as a whole [ 18 , 19 ]. However, direct observations on Arctic zooplankton endogenous visual sensitivity during the Polar Night are lacking [ 20 ], with the assumption that diel cycling of midday twilight is insufficient to entrain endogenous rhythms at this time of year [ 21 ]. With increased attention on poleward range shifts under climate change in marine organisms [ 22 ], it is now recognized that understanding photic barriers to such migrations is critical [ 23 , 24 ].
10.1371/journal.pbio.3001413.g001
Fig 1
Atmospheric light at solar noon across latitudes during the Polar Night.
(a) Northern Hemisphere of Earth, centered on the North Pole. Concentric circles represent the increasing extent of Polar Night as latitude increases; bands are based on the sun’s elevation relative to the horizon at solar noon on the winter solstice: civil twilight (0 to 6°), nautical twilight (6 to 12°), and astronomical twilight (12 to 18°) (modified from [ 78 ]). (b) Svalbard archipelago, with locations of light measurements shown as stars, and locations of krill collection for physiological experiments ( Fig 3 , S2 Fig ) shown as open circles. Basemap for panel (a) is from Natural Earth Data using the combined datasets for Physical Vectors with Land ( https://naturalearth.s3.amazonaws.com/10m_physical/ne_10m_land.zip ) and Minor Islands ( https://naturalearth.s3.amazonaws.com/10m_physical/ne_10m_minor_islands.zip ). Basemap for panel (b) is from Geonorge using S100 Map data ( https://kartkatalog.geonorge.no/metadata/s100-map-data/bd6050e8-7182-459b-9989-66c4ecbae874 ). (c) Hyperspectral irradiance measured at sunset for a midlatitude location (38.7 o N at 21:54 GMT, upper panel) and 3 high-latitude locations around Svalbard in January (lower panel). For high-latitude spectra, diagnostic spectral bands from aurora borealis (391, 557, and 630 nm) and the research vessel’s SVLs are annotated. All-sky images taken at the time of spectrum acquisition are shown at right. Inset is irradiance integrated as photosynthetically active radiation (400 to 700nm, E PAR ). Black, red, and green colors represent 3 stations of increasing latitude (measured at times: 11:35, 11:11, and 11:50 GMT, respectively), which correspond to stars (panel b). For data, see S1 Data . SVL, sodium vapor lamp.
In order to understand the role of light in vision and endogenous rhythms during Polar Night, we ask whether krill show rhythmic visual sensitivity during Arctic midday twilight. We used an in situ observational and experimental approach to study boreal/Arctic krill ( Thysanoessa inermis ), which dominate the Arctic macrozooplankton community and undergo extensive DVM [ 25 ]. The life cycle of krill is influenced by photoperiod, which determines annual patterns of metabolic activity, sexual maturity, and lipid utilization [ 26 ]. Furthermore, much like in Antarctic krill ( Euphausia superba ) [ 20 , 27 ], it is hypothesized that Arctic krill also possess a functional circadian clock. The aim of this paper is to (a) understand the components of atmospheric light which make up midday twilight during Arctic Polar Night; (b) observe the natural Arctic zooplankton community to characterize DVM at this time; (c) determine spontaneous changes in krill visual sensitivity under constant lab conditions; and (d) contextualize these endogenous changes in visual sensitivity with dim light acclimations. In order to provide a quantitative measure of Polar Night midday twilight, point sample hyperspectral light measurements were taken from a research vessel spanning a range of latitudes in the Barents Sea and Arctic Ocean. In addition, and to gain an understanding of daily changes in irradiance over time when the light cycle is driven by midday twilight, light data were gathered from a land-based light observatory [ 4 ]. Biological responses to changes in twilight cycles were determined acoustically near the light observatory. Finally, 3 types of electrophysiological visual sensitivity experiments were conducted on T . inermis : The first tested endogenous sensitivity rhythms over time in an individual, while the second compared visual responses in multiple krill across irradiance levels tested during subjective midday and midnight periods. Finally, the third experiment was used to contextualize endogenous visual sensitivity changes with those experienced under simulated ambient midday twilight intensities.
Results and discussion
Polar Night is a time of darkness punctuated by midday twilight
Using a spectroradiometer optimized for dim light detection [ 28 ], we quantified spectral irradiance of diffuse skylight at solar noon during the Polar Night from 76.5° N to 81° N under a range of weather conditions from a research vessel platform ( Fig 1 ). Broad peaks are present at blue and red wavelengths, with ozone absorption in the yellows (approximately 600 nm, Chappuis band) (e.g., [ 29 ]). Artificial light from the ship’s sodium vapor lighting is evident as peaks at approximately 590 nm ( Fig 1C ), but other peaks in this range represent ambient light (e.g., aurora borealis: 557 nm and 630 nm). Integrated diffuse skylight irradiance measured at the water’s surface (400 to 700 nm, E PAR ) ranged from 2.2 × 10 −6 to 2.2 × 10 −5 μmol photons m −2 s −1 ( Fig 1C , inset). These measurements demonstrate that atmospheric light at midday during Polar Night is, as expected, characteristic of twilight spectral composition at lower latitudes (e.g., [ 30 ]) ( Fig 1C , upper panel).
In order to capture the largest change in irradiance over the diel cycle, we measured a time series of diffuse skylight irradiance during Polar Night north of Rijpfjorden, Svalbard (80° 37.79N 22° 4.14E) over the midday period. During this time, the sun remained below the horizon reaching a maximum elevation of −11.69°, while the moon (approximately 89% illuminated) remained above the horizon but descended from 5.6° to 0.3° elevation ( S1 Fig ). A classic solar-driven diel light cycle (e.g., [ 31 ]) is damped at these latitudes during the extreme photic conditions of Polar Night. While a sun-dominated photoperiod following solar elevation remains evident ( S1A and S1B Fig ), moonlight and green/red light from the aurora borealis serve to extend what would be an otherwise much shorter solar-driven diel light cycle ( S1C and S1D Fig ).
Oscillating Polar Night light intensity is consistent with cyclic behaviors in marine zooplankton
Using atmospheric light data collected during the darkest part of Polar Night in Ny-Ålesund, Svalbard (78.9°N, 11.9°E) [ 4 ], we determined weekly periods (Tau, T) in light intensity (E PAR ) across the lunar cycle during the month of January 2018 ( Fig 2 ). These data reveal overt diel light intensity changes, onto which the monthly and daily lunar light cycles are superimposed. Lomb–Scargle period estimates for January were T = 24.8 hours during the first week (full moon) and T = 24.0 hours for all subsequent weeks. Increasing diel light intensities as a consequence of solar elevation at the end of January mask the rise and fall of the subsequent full moon. Regardless, light intensity remains cyclic with lunar and solar day periods, throughout midday twilight during the Polar Night.
10.1371/journal.pbio.3001413.g002
Fig 2
Cyclic atmospheric light and hydroacoustic patterns during Polar Night.
Solar elevation (degrees relative to horizon) at Ny-Ålesund, Svalbard (Kongsfjorden) during Polar Night, December 2017 to February 2018. Grayscale bar represents daily moon fullness (black = 0%, new moon; light gray = 100%, full moon). Atmospheric irradiance (E PAR ) at Ny-Ålesund is replotted from [ 4 ]. Red dashed vertical lines denote the month of January 2018, during which hydroacoustic observations were conducted with ADCPs. Acoustic MVBS (dB re 1 m −1 ) detection of zooplankton in the water column is plotted during January for Kongsfjorden and Rijpfjorden (Svalbard). Period analysis ( S1 Table ) shows significant diel rhythmicity at all depths and throughout January. Missing acoustic data between approximately 100 and 130 m is due to a “blind zone” of upward/downward facing ADCPs. For data, see S1 Data . We see evidence at the community level that the cyclic light we measured during January 2018 influences in situ migration behavior of marine zooplankton. We examined acoustic backscatter throughout a water column of approximately 200 m in Kongsfjorden and Rijpfjorden to monitor cyclic changes of zooplankton biomass over time and with depth (Fig 2, dashed red box). These biomass changes reflect both DVM and LVM of zooplankton [ 15 , 18 , 32 – 34 ], which is commonly triggered by the ambient light cycle [ 35 , 36 ]. Period analysis of acoustic backscatter revealed significant migrations at both stations in January when the maximum solar elevation at midday was only −6.2° ( S1 Table ). Significant periods in the circadian range for biomass movement were detected throughout the month of January in Kongsfjorden (57% of depth bins) and in Rijpfjorden (61% of depth bins). These data agree with previous observations (e.g., [ 15 ]) showing migrations that continue during the Polar Night in January are driven by solar and lunar cycles. While we do not have net samples from the acoustic mooring location to confirm the identities of the zooplankton migrators, previous net sampling coincident with acoustic surveys in these fjords during Polar Night show that krill ( Thysanoessa spp.) are the dominant migrators and contribute >90% of macrozooplankton biomass [ 37 , 38 ]. Our own net sampling at this time of year further confirms this observation. ADCP, acoustic doppler current profiler; MVBS, mean volume backscatter.
Krill express nocturnal endogenous rhythms in visual sensitivity
We subsequently tested T . inermis in January from 3 Svalbard fjord locations (Rijpfjorden, Kongsfjorden, and Isfjorden) ( Fig 1B ) for rhythmic visual sensitivity change by extracellular electroretinogram (ERG) recording (e.g., [ 28 , 39 ]). The choice of T . inermis was based on its dominance among Thysanoessa species within Svalbard fjords, including the fjord locations we sampled [ 40 , 41 ].
We observed a significant rhythm in visual sensitivity in ERG recordings from the eye of an individual tested by repeated stimulation with dim light flashes at 1°C ( S2A Fig ). ERG amplitude showed a period of 20.4 hours, with peaks in visual sensitivity in phase with low solar elevation (i.e., subjective night) ( S2B Fig ). Several studies support endogenous circadian rhythms in krill, including Antarctic krill E . superba [ 42 , 43 ] and northern krill Meganyctiphanes norvegica [ 44 ]. While these ERG time series data suggest that comparable endogenous rhythmic processes may occur in T . inermis , we adopted a different experimental approach to validate that conclusion. This was done to maximize replication during the limited ship time available in conducting these physiological experiments on live, freshly collected animals.
For these further experiments, we tested whether diel visual sensitivity rhythms were endogenous by determining the half-saturation (Log K) values from response irradiance ( V -log I ) curves for T . inermis eyes. The V -log I approach has been used in previous studies to show a difference in visual sensitivity over the diel cycle consistent with changes in ERG amplitude [ 39 , 45 – 47 ]. We conducted measurements around the times of subjective midday and midnight periods ( n = 8 individuals each) ( Fig 3A ). Freshly collected krill were held at 1°C in darkness for 24 hours before experiments began, with all diel environmental cycles removed to ensure that animals were in a free-running state. V -log I curves support endogenous control of visual sensitivity as Log K values were lower during subjective midnight than during subjective midday, but visual speed as response latency remained unchanged ( Fig 3B and 3C ). We can therefore assume that ambient light levels and extremely short duration midday twilight experienced by krill in situ during the Polar Night are sufficient to entrain a circadian rhythm of visual sensitivity. In further laboratory experiments ( n = 6 individuals), we reveal comparable visual sensitivity changes for krill under simulated midday twilight acclimation ( Fig 3A and 3B ). Visual speed increased with light acclimation, demonstrated as a decrease in response latency, which contrasts with results from our free-running experiments ( Fig 3C ). Thus, different mechanisms are likely involved in observed diel endogenous sensitivity changes as compared to the exogenous sensitivity changes observed here and for other polar crustaceans [ 48 – 50 ].
10.1371/journal.pbio.3001413.g003
Fig 3
Endogenous and exogenous change in krill visual sensitivity and speed with midday twilight.
(a) Response irradiance (V-log I ) curves for T . inermis , with experiments testing endogenous (circles, subjective) and exogenous (squares, acclimation) sensitivity change at 1°C. Endogenous experiments were conducted during subjective night (black circles, n = 8) and day (gray circles, n = 8). Exogenous experiments were conducted during the day with krill ( n = 6) tested first under dark acclimation (black squares) and then under simulated midday twilight acclimation (gray squares). Symbols represent means (± 1 SD) of V-log I models fit to normalized ERG data, with aggregate data fit by a V-log I model (solid lines). (b) Log irradiance at half-saturation for V-log I models (Log K) of each individual krill, with symbols as described for (a). Light acclimation experiments were paired Dark-Light, so points corresponding to the same individual are connected by a dashed line. Log K was lower during both subjective night ( p = 0.038, rank sum test) and dark acclimation ( p = 0.031, signed rank test) when compared to corresponding subjective day and light acclimation treatments in each experiment, respectively. (c) Speed of vision as determined by response latency (elapsed time between onset of the light flash and onset of the photoreceptor response at 50% V max ). Response latency was analyzed for all individuals plotted in (b) and plotted with the same symbol designation. Response latency did not differ between subjective night and day ( p = 0.833, rank sum test), but decreased with light acclimation ( p = 0.031, signed rank test). For data, see S2 Data .
Ecological benefits of extreme light sensitivity in krill
The light-mediated rhythmic behavioral and physiological processes we have demonstrated occur at extremely low intensities of solar/lunar illumination. The atmospheric irradiance values we measured (Figs 1C and 2 ) are conservative in representing the underwater light field as they do not account for refraction of light at the water surface (approximately 4% of incoming E PAR ) and subsequent light attenuation with depth [ 51 ]. To determine biologically relevant underwater light, we used our atmospheric measurements in radiative transfer models to quantify the underwater light krill would experience in the water column at 81° N during Polar Night sampling in January ( Fig 4 ). Intensities sufficient to evoke visual responses in krill extended to over 40 m below the surface, encompassing thresholds for photoentrainment in model crepuscular/nocturnal terrestrial organisms [ 52 – 55 ] and visual sensitivity thresholds in polar fish (i.e., 1% of V -Log I curve [ 56 ]). As photosensitivity may be nonvisual or extraocular [ 10 ], it is likely that light intensities sufficient for entrainment of cyclic processes, such as ERG rhythms in krill, are even less than that required for visual photoreception. The vertical distribution of T . inermis in Rijpfjorden during the morning (approximately 06:30 UTC, 07:30 local) suggests that krill are mostly at their daytime depths (consistent with Fig 2 ), which exceeds their visual threshold ( Fig 4 ).
10.1371/journal.pbio.3001413.g004
Fig 4
Modeled underwater light during Polar Night.
Underwater light intensity plotted as scalar irradiance (E o ; irradiance on a point from all directions weighted equally). Light values were derived from radiative transfer model results of diffuse skylight irradiance measured at Rijpfjorden, Svalbard (80° 37.79N 22° 4.14E) at solar noon on January 15, 2016, then propagated through the water column (bold black line showing irradiance as a function of depth). The dotted vertical line denotes the visual sensitivity threshold for T . inermis (1% of V max for subjective midnight V- log I ). Dashed vertical lines show thresholds for light-mediated processes in a range of other taxa: entrainment of circadian rhythms in fruit flies [ 53 ] and mice [ 55 ]; 1% of V max from V- log I for Antarctic fish Pagothenia [ 56 ]. Depth-stratified abundance of T . inermis at Rijpforden (06:30 on January 14, 2016) is plotted as bubbles of proportional size. For data, see S2 Data .
Endogenous clocks that underpin physiological and behavioral changes are primarily adaptive in their anticipatory capacity in environments where entrainment cues may, or may not, be sufficient [ 57 – 59 ]. Our ERG data show endogenous, rhythmic increases in visual sensitivity for krill during the subjective night and a reduction during the subjective day ( Fig 3 , S2 Fig ), consistent with the nighttime diel migrations of krill ( Fig 2 ). Circadian cycles of visual sensitivity mediated by retinal/interneuron processes and/or screening pigment migrations are common among terrestrial and aquatic animals (e.g., [ 60 – 62 ]). In a particularly well-studied marine organism, the horseshoe crab Limulus polyphemus , lateral compound eyes undergo changes in structure, gene expression, and rhabdom biochemistry to increase visual sensitivity at night (reviewed in [ 63 , 64 ]). This sensitivity increase is understood to be adaptive in nighttime mating at all states of the tide, day or night [ 65 ]. Equally for krill, this increased nighttime visual sensitivity could enhance the effectiveness of its bioluminescence. Krill perform counter-illumination by detecting downwelling light and matching the intensity with light emitted from photophores on their ventral surface [ 66 – 68 ]. This casts a light shadow, visually masking them from predators below [ 69 , 70 ]. Additionally, krill may feed by detecting bioluminescent flashes of prey [ 71 ]. Endogenous control of either would be adaptive when environmental light signals are weak or variable. Whether circadian rhythms in visual sensitivity persist during periods of constant light (i.e., Midnight Sun) remains an open question.
The most striking feature of meso- and macrozooplankton biology is arguably their vertical migrations. Typically, this behavior is understood to be a balance between predator avoidance at depth during the day and feeding near the surface at night under the cover of darkness (reviewed in [ 36 ]). Increased visual sensitivity at night may strengthen the swimming response to changes in light at a time when these animals are migrating ( Fig 2 ). Suppressed visual sensitivity during the day may, in turn, minimize responses to episodic changes in light (e.g., passing clouds) that would otherwise evoke vertical swimming, reducing metabolic costs and risk from visual predators. Even though the Polar Night presents little food from primary production, heterotrophic predators may still rely on migration to find their prey (e.g., [ 37 ]). With an endogenous rhythm in visual sensitivity, krill would anticipate the time of night through circadian entrainment and therefore return to the surface to feed omnivorously. This may explain the significant migrations we observed at depth (>100 m) (e.g., S1 Table and Fig 2 ). Previous molecular and behavioral studies have come to a similar conclusion for vertical migrators (e.g., [ 72 – 74 ]).
We conclude that midday twilight during high Arctic Polar Night has physiological and ecological relevance even at less than 1-fold change in diel light intensity, as compared to a 7-fold change at these locations during the equinoxes (e.g., [ 4 ]). It appears that in krill at least, circadian rhythms may be entrained by nonclassical midday twilight with required irradiance levels among the lowest for any organism to date. While mechanisms for this process are currently unknown, as are the potential role for other entrainment cues and pathways [ 74 – 77 ], these data provide a window into the likely flexibility of visual physiology in organisms from other dim light habitats.
Materials and methods
To analyze periodicity in light cycles during the darkest portion of the Polar Night (December/January), we used data from an atmospheric irradiance (E PAR ) time series measured near Ny-Ålesund, Svalbard (78.9°N, 11.9°E) in close proximity (<5 km) to where the Kongsfjorden acoustic data were collected ( Fig 2 ). These measurements were taken with an all-sky camera-based light sensor (Canon EOS 5D Mark III with 8 mm fisheye lens, Melville, New York, USA) calibrated for E PAR [ 4 , 79 ]. Period estimates were determined at weekly intervals for the irradiance time series using Lomb–Scargle periodograms within the Time Series Analysis (TSA) Serial Cosinor 6.3 package ( S1 Table ). Hyperspectral irradiance spectra of diffuse skylight during the winter Polar Night period were taken in January 2016 from the observation deck above the bridge of the R/V Helmer Hanssen using an Ocean Optics QE Pro spectroradiometer with Spectralon reflectance plate having a detection limit of approximately 4 × 10 −10 μmol photons m −2 s −1 (see [ 28 , 80 ] for detailed methods). Artificial light was minimized by extinguishing vessel lights on this part of the ship, but some work lights aft and below the observation deck remained illuminated for operational safety reasons; the spectral quality of these deck lights was measured, and its minimal influence readily differentiated from environmental sources in diffuse skylight spectra (see Fig 1C ). Point measurements were made at solar noon on January 10 and 12 at locations south and west of Svalbard (generally clear skies, some clouds on 76.5°N may have attenuated light) and on January 15 north of Svalbard over an extended 3-hour period centered on solar noon ( Fig 1 , S1 Fig ). The sun’s elevation relative to the horizon at solar noon ranged between −8.1° (i.e., nautical twilight) and −12.5° (i.e., astronomical twilight) (see [ 4 ] for further discussion of twilight definitions). Coincident with these hyperspectral irradiance measurements, an all-sky camera (as described above) adjacent to the spectroradiometer was used to capture aligned images of atmospheric conditions (e.g., clouds, sun/moon position, and aurora activity), with camera settings to a constant ISO of 12800 (light sensitivity), aperture (f) of 4.5, white balance manually set to “daylight,” and using the shutter speed as the only variable [ 28 , 79 ]. For comparison of the spectral composition of midday twilight, hyperspectral irradiance and an all-sky image were measured at twilight (after sunset) in January 2021 at a midlatitude location (Lewes, Delaware, USA; 38.7°N, 75.1°W).
Acoustics have been used in identifying vertical migration in zooplankton with calibrated echosounders [ 15 , 32 – 34 ]. In order to determine DVMs of the zooplankton community in the natural light environment, we used 2 RDI Workhorse 300 kHz acoustic doppler current profilers (ADCPs) moored in Kongsfjorden (78°58′N, 11°47′E) and Rijpforden (80°13′N, 22°26′E) in January 2018. All ADCP data were checked for quality using the RDI correlation index (a measure of signal to noise ratio) recorded at the instrument. Acoustic volume backscattering strength (S v ; dB re 1 m −1 ) was derived from echo intensity following the method described in [ 81 ] and later employed by [ 15 , 33 , 82 , 83 ] ( Fig 2 ).
We investigated visual sensitivity over the diel cycle in an individual krill using a shipboard assay ( S2 Fig ). Krill ( T . inermis ) were collected using a midwater trawl net from the upper 200 m in Kongsfjorden (78°57′N, 11°57′E) on January 20, 2016. Immediately upon recovery of the net, the cod end was transferred to a light-tight bucket and brought to a 3°C cold room aboard the vessel (ambient water temperature approximately 2°C). Once there, animals were sorted under dim red light, and a krill was immediately prepared for an experiment. In order to determine visual sensitivity over time, we measured the magnitude of extracellular ERGs in response to a standardized 50-ms flash of 488-nm light at 3.65 × 10 −9 photons cm −2 s −1 , given at 15-minute intervals with the animal otherwise in darkness. Equipment was as described in detail elsewhere [ 28 , 39 ]. Briefly, under dim red light (Schott RG630 longpass filter), T . inermis were restrained and submerged in a temperature-controlled seawater bath within a light-tight Faraday cage. An epoxy-insulated tungsten microelectrode (125-μm shank, FHC, Bowdoin, Maine, USA) was positioned subcorneally in the eye, and a reference electrode was placed in the seawater bath. Seawater bath temperature was maintained at 1°C as measured by a thermocouple positioned at the krill eye. The animal was not fed for the duration of the experiment. Periodicity in visual sensitivity was calculated using the Lomb–Scargle periodogram with results suggesting significant rhythmicity ( S2B Fig ). In order to maximize replication, we adopted the response irradiance ( V -log I ) experimental design described below focusing purely on sensitivity during subjective twilight day and twilight night periods.
To confirm the rhythmic visual sensitivity results and demonstrate that T . inermis collected at different latitudes possess endogenous rhythms entrained under twilight cycles, we determined V -log I curves by ERG recording ( Fig 3 ). Krill were collected by midwater trawl from Isfjorden-Karlskronadjupet (IsK; 78.32°N, 15.17°E on January 11, 2016 at 06:10 UTC), Rijpfjorden (80.30°N, 22.27°E on January 14, 2016 at 09:49 UTC), and Isfjorden-Trygghamna (IsT; 78.23°N, 13.83°E on January 17, 2016 at 16:37 UTC) ( Fig 1B ). Netted animals were treated as described above, only this time held in darkness for at least 24 hours without supplemental food until used in ERG experiments. Given this constant dark acclimation and removal of diel zeitgebers, any rhythmic activity present in the physiological experiments described below is considered to be endogenous [ 84 ]. We ceased testing krill 72 hours after collection. We quantified visual sensitivity by ERGs using V -log I curves during the subjective twilight day and twilight night periods, defined here as 09:00 to 15:00 and 21:00 to 03:00, respectively. Equipment and experimental details were as described above and in detail elsewhere [ 28 , 39 ]. The light stimulus was set to the wavelength of maximum spectral sensitivity for T . inermis (490 ± 7 nm, full width at half maximum) [ 28 ]. V -log I curves were generated for individual krill during the times of subjective twilight day ( n = 8) and twilight night ( n = 8) periods once ERG magnitude in response to a periodic dim test flash remained constant for a period of 1 hour. Different individuals were used for each replicate, with electrode position and depth in the eye kept constant across individuals to minimize variability in ERG response [ 39 ]. Peak-to-peak response heights of the ERG waveform ( V ) were measured over a range of 100-ms light intensity flashes (log irradiance; log I ) spanning several orders of magnitude. We modeled V -log I data to determine the log irradiance evoking 50% of the maximum response amplitude (Log K) [ 85 ]. We estimated speed of vision (temporal resolution) for flashes at Log K by determining the response latency as the elapsed time between onset of the light flash and onset of the photoreceptor response [ 39 ]. We tested for differences in Log K and response latency between subjective twilight day and twilight night by rank sum tests.
We supplemented these experiments on free-running krill with other data on exogenous effects of light on visual sensitivity in T . inermis . For these experiments, we used krill collected from Kongsfjorden (78.961°N, 11.895°E at 17:21 UTC on January 15, 2015) employing the general electrophysiological protocols described above. Following collection, krill were maintained in constant darkness at 2.6°C (± 0.1 SD) within a 200-L flow-through tank fed by pumped Kongsfjorden water. Within 2 weeks of collection, we tested visual sensitivity ( V -log I ) and speed (response latency) at 1°C under dark and light acclimation treatments. In a given experiment, an individual krill ( n = 6) was acclimated to darkness, defined by a constant ERG magnitude over a period of 1 hour in response to a dim test flash. After a V -log I curve was generated for the dark-acclimated krill, we light acclimated that individual to broadband blue light (Ocean Optics HL-2000 QTH lamp with Schott BG-18 broadband and 4.0 OD neutral density filters) yielding 5 × 10 −4 μmol photons m −2 s −1 measured at the position of the krill eye. After a constant ERG magnitude was observed over a period of 1 hour in response to a dim test flash, a second V -log I curve was determined for this individual, which remained in a constant state of light acclimation as judged by dim test flashes throughout the experiment. We modeled V -log I data and calculated response latency as described above. Because dark and light acclimation were conducted sequentially in the same individuals, we compared Log K and response latency values between paired dark and light acclimation treatments with signed rank tests.
In order to relate light sensitivity of krill in their habitat to that of other species studied for dim light rhythmic processes, we used the radiative transfer model HydroLight 5.2 to characterize the underwater light field during midday twilight ( Fig 4 ). Light input to the model was diffuse downwelling atmospheric spectral irradiance measured near Rijpfjorden during solar noon, with additional parameters and model details described in [ 28 , 86 ]. Model output was scalar spectral irradiance (E o ) from 400 to 700 nm at 10-nm resolution.
To assess krill distributions relative to modeled underwater light, we sampled krill in a single vertically stratified net tow in Rijpfjorden on January 14, 2016 at 06:33 UTC at the approximate location of krill sampling for ERG experiments using a Hydro-Bios Multinet (0.25 m 2 ) with mesh size 180 μm. Nets were lowered to 265 m and programmed to close 200 to 100 m, 100 to 50m, 50 to 20 m, 20 to 0 m. Cod end collections were fixed in 4% formaldehyde, and T . inermis abundances were later enumerated.
All work was carried out according to the Healthy, Safety and Environment guidelines of the local and national authorities for conducting fieldwork on Svalbard (see www.unis.no ), and the project was entered into the Research in Svalbard (RiS) database with project number 10624 ( https://www.researchinsvalbard.no/ ). For projects registered in the RiS database and carried out in compliance with the Kings Bay AS, no specific permissions are required for marine work. The work does not include protected or endangered species.
Supporting information
S1 Fig
Midday atmospheric light during Polar Night.
Spectral irradiance time series measured north of Rijpfjorden, Svalbard (80° 37.79N 22° 4.14E) on January 15, 2017 over midday period. (a) Solar (black) and lunar (red) altitude during measurements. Lunar phase was a waning gibbous moon, full on January 12. (b) Ratio of 492 nm (solar/lunar light at sensitivity maximum of krill) [ 1 ] and both the 557 nm (green) and 630 nm (red) aurora lines [ 2 ]. (c) Time series of E PAR (400 to 700 nm; upper panel) and 492 nm, 557 nm, and 630 nm light (lower panel). (d) Spectral irradiance at 3 time points during the time series shown in (c); with a running mean and 492 nm peaks (black lines), and aurora lines at 557 nm and 630 nm plotted green and red, respectively. For data, see S1 Data .
(DOCX)
S2 Fig
Rhythmic oscillations in krill visual sensitivity.
To test whether Arctic krill ( T . inermis ) showed rhythmic changes in visual sensitivity, and, in turn, warranted further experiments, we collected an individual krill from Kongsfjorden in January, and immediately prepared it for ERG recording. (a) ERG magnitude (red line = 1.75 hours running mean) is plotted in response to a 50-ms flash of 488-nm light at 3.65 × 10 9 photons cm −2 s −1 . Since this animal was in darkness, subjective solar elevation (negative degrees relative to horizon) is plotted for the collection location. Peaks in ERG response magnitude occurred during the time of subjective night. (b) Lomb–Scargle periodogram for ERG data in (a), resulted in a peak period at 20.4 hours. Dashed line represents significance at the ɑ = 0.05 level. For data, see S2 Data . ERG, electroretinogram.
(DOCX)
S1 Table
Period analysis of acoustic backscatter data.
Period (hours) estimates for Kongsfjorden and Rijpfjorden acoustic data (from Fig 2 ) are provided for discrete depths throughout the water column for January 2018. Periodicity was calculated by Lomb–Scargle periodogram. Gray shading indicates periods within the circadian range (20–28 hours), showing that circadian cycling is spread across the water column. The full moon occurred on January 2.
(DOCX)
S1 Data
Light and bioacoustic data for Figs 1 and 2 and S1 Fig .
(XLSX)
S2 Data
Electrophysiological and associated data for Figs 3 and 4 and S2 Fig .
(XLSX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Cellular response is a complex phenomenon that manifests both physically and chemically. While physical responses can most often be analyzed visually, chemical responses are difficult to characterize even with modern detection methods. Adding to the difficulty is the influence of timing when dealing with cellular response. When responding to a biochemically altering stimulus, such as naïve T-cell response to an antigen presented by a dendritic cell, a cell undergoes a series of biochemical pathway shifts that allow it to adapt to its new conditioned state. The cell’s temporal response, therefore, provides insight into the interspersed pathway shifts that occur throughout the various stages of stimulus exposure. Temporal response is especially relevant to toxicology, where, despite experiencing a toxin and undergoing numerous metabolic shifts, an organism may not show observable external symptoms until much later, when treatments may be less effective.
Nonetheless, time remains a largely underappreciated or neglected variable in most comprehensive cellular response experiments, not as much for its perceived lack of value as for the difficulty of precise temporal resolution and control of measurements. Biological measurements typically take one of two forms; either they sample a small number of targets very frequently (as in immunoassay or fluorescence measurements) or they sample a large number of targets over a broad time span ( e . g ., proteomic cataloging). The limitations of current methodological strategies and detection technologies hinder the combination of these two approaches. The most comprehensive analysis would track as many analytes as possible and sample these analytes as frequently as possible to capture the truly dynamic properties of metabolic responses. As the frequency of sample collection is increased, previously unidentified patterns in signal response may begin to emerge, according to the Nyquist-Shannon sampling theorem, which states that in order to properly characterize a pattern that has as its highest frequency Y, the sampling must occur at intervals less than 1/(2Y), thereby eliminating the possibility of aliasing, wherein undersampling of rapidly varying phenomena produces misleading features that appear to have a lower frequency.
Given the desire to study dynamic biological phenomena within the construct of robust microphysiological systems [ 1 ], a major logistical problem still remains: how does one sample a biological system multiple times over the course of an experiment without destroying, or even perturbing, that system? The exometabolome, or suite of biomolecules that are secreted or excreted from a cell into the surrounding matrix, represents a promising target. The exometabolome has been demonstrated as a viable indicator of internal cellular processes and can be easily assessed without any disturbance of the system from which it originates [ 2 ]. Changes in the transport of exometabolomic species provide information on the current state of the cell population, which can lead to a more thorough understanding of the particular cell biology and to the ability to control cell behavior [ 3 ]. Further promoting the benefit of metabolomics is the timescale of metabolic response to alterations in the environmental conditions. While alterations in the proteome occur over hours or days, metabolome changes occur within seconds or minutes.
The cell is constantly surveying the surrounding environment. Perceived changes in this microenvironment lead to alterations of intracellular and intercellular signaling, which in turn lead to shifts in gene regulation and modifications in protein and metabolite production. Depending on the signal received, the intracellular processes enacted may lead to the secretion of another signaling factor to extend the complex web of communication. These signaling factors are produced by a given cell for communication with 1) nearby cells of the same type (autocrine signaling), 2) adjoining cells (juxtacrine signaling), 3) nearby cells of a different type (paracrine signaling), and 4) distant cells (endocrine signaling). The exometabolome not only includes traditional signaling molecules and metabolites of nutrients, but also enzymatically produced metabolites of xenobiotics ( i . e ., drugs, toxins) introduced to cells and the nutrients or other factors in the culture media as modified through cellular uptake.
The Vanderbilt Institute for Integrative Biosystems Research and Education Multitrap Nanophysiometer (MTNP) is a polydimethylsiloxane (PDMS) microfluidic device that functions as a miniature bioreactor for unattached cells [ 4 – 10 ]. In contrast to traditional cell biology techniques that often require cellular populations on the order of >10 8 , the MTNP allows for studies on small populations of cells or simple organisms (<10 5 ). The MTNP also provides a solution to the volume challenge problem existing in traditional in-culture experiments, as the small volume of the device and continuous flow prevent the dilution of signaling factors [ 7 ]. The MTNP can be used for long-term optical measurements of the dynamic behavior of cells, including fluorescent labeling of cells to determine type and activation state and detect signaling dynamics and cell-cell interaction. This device provides a framework on which to study numerous cells, i . e ., T cells, beta cells, and breast cancer cells, as it traps both non-adherent and adherent cells with structural barriers instead of with chemical surface modification that may cause cells to be exposed to higher shear stresses resulting from direct contact with fluid flow. The MTNP is well suited to detect secreted molecules in cellular effluent, and it is also unique in its ability to provide a system for investigating the dynamic response of a cell population to a stimulus, possibly enabling challenge-response statistical analysis of cellular dynamics [ 11 ]. The constant perfusion design of the microfluidic bioreactor gives rise to a platform component capable of real-time alteration and control of the cellular microenvironment, in addition to providing an opportunity for monitoring the effluent output from the device.
The complexity of biological samples often overwhelms standard screening techniques seeking to discover biomarkers of disease or rapidly assess drug efficacy. This unique problem warrants an analytical technique capable of both rapid screening of samples and sufficient sensitivity to account for the many analytes in low abundance that typically are of interest in metabolomic and signaling experiments, which often either defy detection or appear collectively as chemical noise. The use of mass spectrometry (MS) as the leading detection method of proteomic [ 12 , 13 ], lipidomic [ 14 ], and glycomic [ 15 ] studies has led to many advances in elucidating the complexity of the cell.
The combination of constant microfluidic perfusion bioreactors with mass spectrometry has the potential to rapidly screen cell effluent for secreted species indicative of internal metabolic perturbations. This is potentially of great importance for analyzing the response of coupled organ-on-chip systems to drugs, toxins, and other agents [ 1 , 16 , 17 ]. Several studies have taken on the challenge of integrating these technologies to produce a powerful analytical platform. In an early experiment, Chan et al . verified that using PDMS bioreactor devices to transfer samples into the electrospray ionization mass spectrometer (ESI-MS) did not contaminate the samples [ 18 ]. On-chip ultrafiltration and analyte pre-concentration for high-throughput small molecule screening with ESI-MS were performed with the resulting detection sensitivity shown to increase by one to two orders of magnitude over off-chip screening strategies [ 19 ]. A significant impairment in coupling cellular bioreactor microfluidic devices with online-MS is in the suppression of signals of interest by salts present in biological samples. To overcome this challenge, some form of desalting is typically performed offline, using either a solid phase extraction (SPE) column and vacuum manifold or some form of liquid chromatography. These techniques desalt samples by providing a functionalized surface for which salts and analytes have differing affinities. For example, liquid chromatography (LC) typically uses C18 columns, which provide binding sites for non-polar molecules but lack sufficient interaction prospects for salts, thereby allowing for an aqueous rinse to clear the column of salts and a subsequent organic elution of analytes without the suppressing contributions from salts. These methods, while providing an efficient means of desalination, come at the expense of temporal resolution, and they are not designed for online analysis. One recent development in the online desalting of effluent from a microfluidic bioreactor is the work of Chen et al ., which incorporates a packed micro-solid phase extraction column [ 20 ].
We have previously demonstrated the ability to rapidly desalt a continuous sample stream using online SPE [ 21 , 22 ]. This work directly translates to the processing of continuous sample streams originating from an MTNP. Using online SPE for such applications provides a surface for which analytes of interest have affinities but unwanted salts do not. This allows for variable duration periods of pre-concentration of the analytes and even the potential for gradient elution of these analytes.
As an initial proof of concept experiment for this integrated platform, we selected cellular memory of drug exposure, specifically Jurkat T-cell memory of cocaine, for ease in identification of known drug metabolites and for its unique biological information. The choice of cell type was based upon cocaine’s classification as an immunosuppressive agent, by mechanisms of direct and/or indirect actions on immune cells. Findings from experiments tying cocaine to immune function suppression have been contradictory [ 23 – 27 ], most likely because of the complex biological systems under investigation and wide disparity of experimental procedures. A major hurdle in determining these mechanisms is the lack of an assay capable of tightly controlled environmental parameters, sufficient temporal resolution to avoid loss of transient changes, and multi-parameter sampling for unique evidence of interconnection of experimental variables. The platform demonstrated herein allows for such an assay to be conducted. Through the comparison of naïve Jurkat T cells and those with prior cocaine exposure on this platform, differences in cocaine metabolism are detected. Fig. 1 demonstrates the experimental concept. Exposure of a naïve cell to cocaine may lead to a certain dynamic exometabolomic profile that defines the state of the cell. However, when a cell with prior cocaine experience receives a subsequent dose, the dynamic exometabolomic profile may either be identical to that of the naïve cell under exposure or distinguish the cell as unique. With a near real-time readout of cellular metabolic events on our integrated microfluidic-solid phase extraction-ion mobility-mass spectrometry platform, it is possible to determine variations in dynamic exometabolomic profiles that will provide evidence of cellular memory of cocaine exposure.
10.1371/journal.pone.0117685.g001
Fig 1
Experimental scheme showing potential cell fates.
Upon exposure to cocaine in a microfluidic bioreactor, naïve and cocaine-experienced cells present different exometabolomic profiles, demonstrated here as color change. Our experiments were designed to determine whether cocaine-experienced cells went to a conditioned state A that was different from state B reached by naïve cells.
Methods
Microfluidic Bioreactor Design and Fabrication
The Vanderbilt Institute for Integrative Biosystems Research and Education Multitrap Nanophysiometer was previously designed in AutoCAD and converted to a chrome mask on glass (Advance Reproductions, North Andover, MA). A soft-lithographic master was produced through photolithography methods, which include spinning a negative photoresist, SU-8, onto silicon wafers, exposing them to UV light through the desired mask to crosslink the SU-8, then developing to remove non-crosslinked polymer [ 28 ]. Polydimethylsiloxane (PDMS) (Sylgard 184 Elastomer Kit, Dow Corning, Midland, MI) was then cast onto the silicon and SU-8 master, cured, and removed from the master. Inlet and outlet ports were punched and PDMS replicas were bonded to glass coverslips by O 2 plasma treatment (Harrick Plasma Cleaner, Ithaca, NY). PDMS surface modification was performed using alcohol deposition of 2-[methoxy(polyethyleneoxy) 6–9 propyl]trimethoxysilane immediately following plasma treatment and bonding (see Supporting Information , S1 File , S1 Fig. ).
Cell Culture and “In-Culture” Cocaine Exposure
Jurkat T cells (clone E6-1, TIB-152) (ATCC, Manassas, VA) were cultured in 90% RPMI 1640, 10% fetal bovine serum (FBS, heat inactivated) (Lonza, Allendale, NJ) at 37°C, 5% CO 2 according to the manufacturer’s instructions.
In-culture cocaine exposure experiments were performed as follows ( Fig. 2 ). Two populations of Jurkat T cells (passage 6, 2 million cells/mL, 500 μL, in biological triplicate) from the same culture flask were added to two separate microcentrifuge tubes. Both tubes were centrifuged at 200 × g for 2 minutes and supernatant was removed. Cells in Tube 1 were resuspended in 500 μL of RPMI 1640 and incubated for 270 minutes at 37°C, 5% CO 2 . Cells in Tube 2 were resuspended in 500 μL of cocaine in RPMI 1640 (60 μg/mL) and incubated for 216 minutes at 37°C, 5% CO 2 . The cells in Tube 2 were centrifuged and resuspended in RPMI 1640 for 54 minutes. At this point, the cells in Tube 1 are “cocaine naïve,” while those in Tube 2 are “cocaine experienced.”
10.1371/journal.pone.0117685.g002
Fig 2
Cocaine exposure scheme for both in-culture and online cell experiments.
The time course of cocaine administration to naïve (blue) and experienced (green) T-cell populations is shown. For the in-culture experiments, experienced samples 1E-4E and naïve samples 1N-4N were withdrawn for analysis at the times shown.
Both tubes were then centrifuged as above, with the supernatant being reserved as Samples 1E and 1N for protein precipitation and metabolomic analysis. The cell pellets were immediately resuspended in RPMI with cocaine at 60 μg/mL and incubated for another 54 minutes.
After this 54-minute cocaine exposure, the cells were centrifuged to obtain 2E and 2N, representing the results from the first common cocaine exposure for both naïve and experienced cells. The cells were resuspended in RPMI 1640 for another 54 minutes. Centrifugation provided samples 3E and 3N, representing recovery from cocaine exposure.
Next, the cells were resuspended with RPMI and cocaine (60 μg/mL) and incubated for 54 minutes. A final centrifugation provided samples 4E and 4N, representing the final cocaine exposure.
In order to investigate the degradation of cocaine over the time course of the experiment, samples of cocaine in RPMI but without cells were incubated in the same conditions as the cells and sampled at each 54-minute time point starting at time zero and ending at 10.5 hours.
Metabolomics Sample Preparation and Analysis
As each 54-minute time point was taken, supernatant was placed on ice to halt any continued metabolite modifications and immediately processed for protein removal with cold methanol. To each 300 μL supernatant sample, 900 μL of cold methanol was added, vortexed, allowed to sit for 10 minutes at 4°C and then centrifuged at maximum speed for 10 minutes in a Heraeus Fresco 21 temperature-controlled centrifuge (Thermo Scientific). Supernatant was transferred to a new microcentrifuge tube and kept at 4°C until all samples were collected and processed. All samples were then dried down in a Savant DNA 110 Speedvac overnight with low heat and reconstituted in 300 μL 5% methanol/95% water (0.01% formic acid).
Samples were placed in the sample tray of the nanoAcquity UPLC with autosampler (Waters, Milford, MA), which remains cooled to 4°C to prevent sample degradation. Ultraperformance liquid chromatography (UPLC) was performed on 1 μL samples loaded on a HSS C18, 1.8 μm particle size column with mobile phase A (0.1% formic acid in H 2 O) by ramping from 100% mobile phase A to 100% mobile phase B (0.1% formic acid in MeOH) over 11 minutes, then holding at 100% mobile phase B for 2 minutes. Ion mobility-mass spectrometry and MS e were then performed on the eluted analytes using an ion mobility-mass spectrometer (IM-MS, Synapt G2, Waters Corp., Milford, MA). Quality control samples were dispersed every 10 samples in the Waters MassLynx v4.1 software sample list among technical triplicates of each biological sample.
Online Cell Loading and Experimentation
Prior to experimentation, 500–1000 μL of cell suspension was removed from culture flasks. Cells at passage 6 were used for experiments. Cells were then gently pelleted and aspirated into polyether ether ketone (PEEK) tubing connected to pump-controlled syringes. The flow direction of the pump was reversed upon intubation of the MTNP and cells were collected into microfluidic traps for experimentation. Cell-loaded devices were then perfused with selected media components and brightfield images were collected by an inverted Nikon Eclipse Ti-e (Nikon Instruments, Melville, NY). Cells in the MTNP were maintained at 37°C and 5% CO 2 during experimentation.
For T-cell cocaine metabolism studies, populations of naïve T cells and experienced T cells (exposed to cocaine at 60 μg/mL in RPMI 1640 for 4 hours) were stimulated with either cocaine (60 μg/mL in RPMI 1640) or RPMI 1640. Cells were initially exposed to plain RPMI media for 54 minutes, followed by exposure to cocaine at 60 μg/mL in RPMI for 54 minutes. Both steps were repeated for a total of four steps. Experiments with naïve and experienced cells were performed on the same day, in series, to exclude any variation in cell population from day to day. The cocaine exposure scheme is the same as the in-culture exposure, shown in Fig. 2 .
Solid Phase Extraction Desalter
Columns were made of 360 μm OD/100 μm ID fused silica tubing and bomb-loaded in house with 3 μm, 300 Å, C18 phase Jupiter Bulk Packing (Phenomenex, Torrance, CA) using a PIP-500 Pressure Injection System (New Objective, Woburn, MA). Three 10-port Nanovolume UPLC Valves with 360 μm fittings, C72MH-4690ED (VICI Valco Instruments Co. Inc., Houston, TX) were used for the valve arrangement ( Fig. 3 ). The aqueous solvent and both organic solvent lines, running at 500 nL/min, were supplied with an Eksigent Nanoflow Metering System (AB SCIEX, Framingham, MA), which has four independent flow channels. The output lines from the two downstream valves were connected with a micro-T junction and fed directly into the mass spectrometer via a nanoelectrospray ionization (nESI) source. Cheminert 360 μm unions (with 100 μm bore) were used for all tubing-to-tubing connections (not shown). Only fittings for 360 μm OD tubing were used, as the more popular 1/16” fittings, which require sleeves to connect to smaller bore tubing, resulted in leakage at high backpressures. Other than the columns, which were made of fused silica, 360 μm/50 μm PEEK tubing was used. Prior to experimental use a 2 μg/mL solution of polyarginine was run through the system to bind all non-reversible interaction sites. Elution cycles were then run overnight to ensure that all reversibly bound material was removed before experimentation. Sample loops were added to the system (as shown in Fig. 3 ) to reduce backpressure buildup.
10.1371/journal.pone.0117685.g003
Fig 3
Solid phase extraction desalter.
Setup starting from initial sample effluent flow incorporates two sample loops, three valves, and two C18 columns. During (A), sample effluent fills sample loop 2 for 9 minutes, while the aqueous solvent flows through sample loop 1, over column 1, and to waste. The organic solvent flows over column 2 and to the mass spectrometer. (B) Upon switching of the valves, the sample effluent fills sample loop 1 for 9 minutes. Aqueous solvent forces the 1.5 μL head of aqueous solvent, the 4.5 μL of sample effluent, and an additional 2.1 μL of aqueous solvent over column 2 to equilibrate the column, load the sample, and rinse away the salts. Organic solvent runs through column 1 and to the detector. (C) The next valve switch again exchanges the sample loop filled by effluent, while column 1 is equilibrated, loaded, and rinsed. The analytes captured on column 2 are eluted by the organic solvent and sent to the detector. (D) When the valves switch again, the sample effluent fills the opposite loop, column 2 is equilibrated, loaded, and rinsed, and column 1 is eluted with organic solvents and those analytes are sent to the detector. The pattern repeats until the experiment is completed, with each cycle requiring 9 minutes.
Online Cell Effluent Desalting and Mass Spectrometry Analysis
All online cell effluent experiments were conducted using PDMS MTNP devices. Cellular effluent from the device was processed online prior to mass spectrometry analysis using the online SPE desalter in Fig. 3 . Cell effluents, driven by syringe pumps upstream of the bioreactor, were filtered with an inline filter (1 μm stainless steel frit, followed by a 0.5 μm polymer mesh, IDEX Health & Science, Oak Harbor, WA) and loaded into each sample loop. Sample loops, which were made of 360 μm OD/250 μm ID tubing, were 12.2 cm long, providing a sample loop volume of 6 μL. The continuous sample stream was diverted into each sample loop for exactly 9 minutes at 500 nL/min, thus filling the sample loop to 75% capacity. Because water was always flowing through these sample loops immediately prior to sample flow, a plug of 1.5 μL of water preceded each sample effluent plug. This plug served to quickly and roughly equilibrate the column with an aqueous solvent. Once the loop was filled to 75% with online sample effluent, the small water plug and sample effluent were passed over the column, using the aqueous solvent line to generate the necessary backpressure. Once the effluent had cleared the sample loop and had been entirely passed over the column, an additional 2.3 minutes or 2.1 μL of aqueous solvent (H 2 O with 0.1% formic acid) was run over the column to serve as the rinsing/purging step to remove residual salts. Following the salt purge, the column was eluted with organic solvent (90% methanol, 10% H 2 O, 0.1% formic acid). Each step of this process from the initial effluent flow through the SPE desalter is illustrated in Fig. 3 . Data were collected using MassLynx software (Waters Corp., Milford, MA) by Data Dependent Analysis method cycle files, where each column elution was collected as an individual file. Collecting each column elution as individual data files aided sample analysis, as explained later.
Data Processing and Multivariate Statistical Analysis
Resulting data sets from both online and in-culture experiments were processed using Waters MarkerLynx software along with Umetrics Extended Statistics software for multivariate statistical analysis (Waters Corp., Milford, MA). All spectra were corrected to sodiated HEPES buffer ([M+Na] + exact mass 261.0888) and centroided, and peaks were normalized to 10,000 counts per sample. Spectra from samples analyzed through UPLC were deisotoped and underwent chromatographic peak detection. Data from online experiments were processed by a combined scan range. An intensity threshold was set for all data at 1000. Principal component analysis with Pareto scaling was performed to verify initial sample grouping. Further statistical analysis with orthogonal partial least squares-discriminate analysis was performed to identify significant contributors for group separation. Significance in abundance of exometabolomic species was determined through a Welch’s unpaired t-test using conservative confidence levels less than 0.05.
Results
Platform Integration and Evaluation
Successful integration of the platform has been achieved, as shown in Fig. 4 . The microfluidic bioreactor (MTNP) is controlled upstream by Harvard Apparatus syringe pumps, viewed under the Nikon Eclipse Ti-e microscope for optional fluorescent and brightfield imaging acquisition, and outfitted with a stage incubator for control of temperature, gas, and humidity. The effluent exiting from the MTNP flows through two inline filters for catching cell debris (1 μm stainless steel frit, followed by a 0.5 μm polymer mesh, IDEX Health & Science, Oak Harbor, WA). Once through the filters, the effluent fills one of the two sample loops vented to open air to avoid high backpressures in series with the compliant microfluidic device. After the effluent undergoes salt removal by the solid phase extraction desalting system, it is directed into the nESI source and sprayed into the nESI-IM-MS.
10.1371/journal.pone.0117685.g004
Fig 4
Multitrap Nanophysiometer (MTNP)-solid phase extraction (SPE)-nanoelectrospray ionization (nESI)-ion mobility (IM)-mass spectrometry (MS) platform.
The platform includes Harvard Apparatus syringe pumps, a Nikon Eclipse Ti-e inverted fluorescence microscope with stage incubator, a solid phase extraction desalter with Eksigent and nanoAcquity pumps for solvent flow control, and a nanoelectrospray ionization source for continuous flow sample analysis with the Waters Synapt G2 ion mobility-mass spectrometer. Beyond the initial experimental setup, all components are fully automated and capable of running multi-hour experiments limited by the lifetime of cells in the MTNP.
Initial studies, which paired a previous arrangement of the online SPE design [ 22 ] with MTNPs, resulted in ruptured devices. When LC pumps were used to rinse and elute the SPE columns, pressure would build up behind the column. When the valve was switched so that the MTNP was directly in series with the column, this high backpressure (>200 psi), would cause massive flow reversal and induce the PDMS device to physically delaminate from the glass to which it was bonded. This was alleviated by altering the valve arrangement to incorporate pressure-eliminating sample loops. By cutting off the direct pathway between the columns and the microfluidics, the buildup of pressure that occurred during rinsing/eluting was no longer in series with the compliant microfluidic device and would instead be vented to the waste port, as shown in Fig. 3 . Although in extreme cases the inline filters downstream of the microfluidic device may become clogged with cells, thus interfering with effluent flows, this is a rare occurrence and can be prevented by using new filters for each experiment in addition to open-outlet cell loading (allowing cells to exit the device during loading before attaching to the downstream components).
The sample loop addition to the valve arrangement allowed for a two-step valve configuration, a simplification of the previous version. The new two-step arrangement generated a saw-toothed pattern of analyte elution as opposed to the rise-and-fall delta-function pattern observed with previous arrangements [ 22 ]. In this 180-minute experiment, an 18-minute-long cycle was used, producing an elution peak every 9 minutes. The sample loop volume was designed specifically to hold 9 minutes of sample effluent (4.5 μL at 500 nL/min flow rate). This cycle duration was determined to be optimal based on analyte concentrations. The file acquisition was set to account for the delay time from the device to the nESI source such that one column elution was captured per file while accounting for the roughly 12-second software delay between file acquisitions. Pump switching for control of the MTNP perfusion media was also timed with the SPE desalter valve switching.
Removal of salts in an online manner greatly increases signal-to-noise ratio. Yet with the low number of cells and high concentration of media components, detecting low concentrations of analytes is challenging. The signals are also affected by the dynamic range (∼10 5 ) of the mass spectrometer [ 29 ]. Presumably, post-desalting, and given cellular utilization of nutrients in the media, HEPES at 10 mM is the analyte of greatest concentration. The high concentration of this species limits the lower range of detection of exometabolomic contents to roughly five orders of magnitude lower than HEPES. This HEPES concentration can be reduced if necessary, but as its purpose is to buffer the media to maintain a physiological pH, there could perhaps eventually be a tradeoff. When considering the small volumes associated with this platform, HEPES at 10mM would equate to 4.5 x 10 –8 moles per filling of a sample loop. This would indicate that the dynamic range would allow for the detection of species as low as 10 –13 moles per filling of the sample loop, or roughly 100nM. While detection of low concentration species remains a possible challenge, the ongoing advances in mass spectrometry technology will increase the detection capabilities of this system.
PDMS Surface Passivation for Increased Signal-to-Noise Ratio
While insulin is not necessarily a prime target of these experiments, it serves as an example of the potential complications from non-specific adsorption. Though high sensitivity is characteristic of mass spectrometry, our system seeks to identify secreted molecules from roughly 10 5 cells. Detecting these low-level signals becomes a greater challenge when a portion of the signal is lost due to interactions with seemingly inert materials. Although the PDMS passivation schemes returned positive results (data provided in Supporting Information, S1 File , S2 Fig. ), some metabolite, peptide, or protein species, such as insulin, are particularly “sticky” to most polymers and glass. In testing the capabilities of the system, we have noticed drastic reduction in and even absence of signal from insulin standards over time, even at low temperatures. Additionally, insulin hysteresis in the combined platform has been discovered after as long as 4 days of continuous perfusion of the SPE desalter tubing and columns. While the columns may be a source of insulin retention, this particular hysteresis occurred with freshly made columns, pointing to an alternate source of contamination that resulted in memory effects. The remaining sources of contamination could be from insulin retention in the PEEK and/or fused silica tubing lines, the valve rotors, or even on the source block or cone of the mass spectrometer. Further efforts for overcoming or reducing memory effects could include passivation of all components of the system or avoiding use of certain materials known to interact more readily with biological materials. Although analyte interaction with materials, PDMS in particular, is unavoidable to some extent, surface passivation provides a means of vastly minimizing the effect.
Comparison of UPLC-ESI-IM-MS to MTNP-SPE-nESI-IM-MS
One of the major technical challenges of organs-on-chips is the small fluid volumes available for analysis [ 16 , 17 ]. A significant reduction in time and the avoidance of handling and storage of small fluid samples are among the benefits of continuous sample flow from the microfluidic bioreactor to the solid phase extraction desalter and into the nESI-IM-MS. This integrated platform allows for the setup (∼2–3 hours), execution (∼4–8 hours), and data collection (no additional time) in the course of a day. Traditional in-culture experiments with UPLC-ESI-IM-MS analysis require possibly less initial setup (∼1 hour), roughly the same execution time (∼4–8 hours), and significant additional sample processing time (∼15–20 hours including overnight sample evaporation) and data collection (∼50 hours for 120 samples with a 25-minute UPLC time per sample), for a total of about 4 days until data are ready for processing, compared to our platform’s essentially real-time capability with a 9-minute sampling interval and in-line sample processing.
This suggests a major advantage of the integrated platform compared to in-culture experiments: the ease of obtaining mass spectra at multiple time points. Our process is automated with constant media perfusion control timed with the switching of the SPE desalter valves as well as the data file collection timing. With the in-culture experiments, a lengthier process is required to collect a single time point. The cell suspension must be centrifuged for 2 minutes, the supernatant removed, and the cell pellet resuspended in new media. The length of time required for this media change is on the same order of the time points taken with the integrated platform. Repeated centrifuging and pellet aspiration can introduce unnecessary stress to the cells, which may affect the cellular exometabolome and lead to profiles resulting from both the media exposure and additional cell stress.
One downside compared to liquid chromatography data is the absence of retention time information. While the solid phase extraction desalter uses a column similar to those found in an LC system, any slight timing discrepancy from one file collection to the next prevents the use of any retention time data. Confounding this issue is the fact that processing data within the Waters Masslynx software without retention time data available prevents the option for the removal of isotopes from the data. While this is a hindrance in some respects (roughly 3 times more peaks, redundant data, etc.), it can prevent the accidental removal of isobaric peaks of interest from the study. Since our platform is adaptable, integrating a chromatographic separation could be easily accomplished in a number of ways. Gradient elution could be utilized to achieve this type of separation without any additional hardware or software. An additional column could be utilized as part of the solid phase extraction desalter setup for further chromatographic separation as well, thus achieving all the benefits of liquid chromatography if desirable for a specific experimental setup.
Cocaine Metabolism in Naïve and Experienced T Cells
Online Cellular Analysis
Two populations of Jurkat T cells were compared in this study: naïve T cells that had never been exposed to cocaine and experienced T cells that had been incubating in cocaine at 60 μg/mL in RPMI 1640 for 4 hours prior to online experimentation. As shown in Fig. 5 , a high degree of variance was observed based not only on what type of media was present in the bioreactor ( i . e ., plain RPMI media or cocaine RPMI media), but also whether the cells experienced cocaine earlier in the day ( i . e ., whether the cells were experienced or naïve to cocaine exposure). The major unique contributors to group separation between naïve and experienced exometabolomic profiles included m/z 283, m/z 187, m/z 399, m/z 157, and m/z 337 (all at elevated abundance in the experienced group compared to the naïve group). The metabolites contributing to group separation between type of media to which cells were exposed in the MTNP consisted of m/z 312, m/z 182, m/z 304, m/z 290, and m/z 150. Though the data analysis pipeline prevented removal of isotopes, any m/z present in the list of top contributors negated the inclusion of their respective isotopes from these lists.
10.1371/journal.pone.0117685.g005
Fig 5
Exometabolomic profiles of naïve and cocaine-experienced Jurkat T cells from online cellular analysis.
Walking principal component analyses of exometabolomic profiles of naïve (shades of yellow) and cocaine-experienced (shades of turquoise) Jurkat T cells were constructed following online cellular analysis. Each numerically labeled data point represents a 9-minute column elution, with six data points collected per step (all marked with the step number and connected in order of collection with the dotted line). In steps 1 and 3, both cell populations received plain RPMI. In steps 2 and 4, both received cocaine in RPMI at 60 μg/mL. As profiles switch from RPMI to cocaine exposure, the data points move towards the right and vice versa, with the exception of naïve cell step 4, which stays closer to step 1. Further analysis of the data reveals this inconsistency may be explained by the death of the cells. Data were grouped not only based on the experimental step, but also by the experience level of the cells, as the cells receiving a 4-hour pre-incubation in cocaine group separately from those that did not receive this dose.
Benzoylecgonine (BE) (m/z 290 as [M+H] + , m/z 312 as [M+Na] + ), a primary metabolite of cocaine, was identified as a contributing factor to the separation between the cocaine exposure steps and plain RPMI steps. This metabolite additionally contributed to the separation of populations of naïve cells and experienced cells. Analysis of this metabolite over the time course of the experiment revealed an expected increase during cocaine exposure steps, but also showed a significant increase from naïve to experienced cells with a very conservative p-value of 5 . 7 x 10 –4 ( Fig. 6 ). Expected levels of BE produced by degradation or metabolism of cocaine in the second cocaine exposure of the naïve cell population are notably absent. Further analysis of this apparent reduction in BE level during this exposure period revealed the likelihood of cellular death as a contributor to this result. Fragmentation spectra revealed fragment ions m/z 82, m/z 91, m/z 105, m/z 150, m/z 168, m/z 182 and m/z 272, as shown in Fig. 6 . Analysis of the remaining top three contributors to separation based on media revealed cocaine at m/z 304, anhydroecgonine methyl ester (AEME) at m/z 182 (produced from dehydration of ecgonine methyl ester (EME) rather than the pyrolysis of cocaine), and ecgonine aldehyde, the decomposition product of EME, at m/z 150. Cocaine metabolic pathways are described in Fig. 7 .
10.1371/journal.pone.0117685.g006
Fig 6
Benzoylecgonine (BE) time course and fragmentation data.
Top: BE time course data for experienced (blue) and naïve (purple) cells. While data were gathered sequentially, plots are overlaid to highlight the increased abundance of BE in experienced cells. The absence of the expected increase in BE corresponding to step 4 (the last step of cocaine exposure) suggests a decrease in cocaine metabolism, possibly due to cell death. Bottom: The fragmentation spectra of BE are shown with parent ion of m/z 290.
10.1371/journal.pone.0117685.g007
Fig 7
Metabolism of cocaine showing molecular weight for each metabolite.
The linkages in this network were adapted from Xia et al . [ 30 ]. Putatively identified species are shown in bold.
To verify that this increased BE abundance was not purely a result of non-enzymatic hydrolysis of cocaine to BE in aqueous solutions over the time course of the experiment, we omitted the Jurkat cells from an experiment conducted simultaneously with those for the naïve and experienced cells. The cell-free media, either with or without cocaine, depending upon interval in the protocol, was sampled and analyzed in the same manner as the media conditioned by the cells. In order to compare the experiment with cells to those without cells, we normalized the BE intensity to the cocaine intensity. On average, the percent of the total normalized BE created by non-enzymatic hydrolysis of cocaine was 34.1% in step 2 of the naïve cell experiment, while the corresponding percentage for step 2 of the experienced cell experiment was 34.5%. There may be other not-yet-identified mechanisms for the breakdown of both cocaine and BE, possibly involving processes shown in Fig. 7 . Hence the cells contribute to no more than ∼66% of the BE reported in Fig. 6 . The statistical significance of the differences between naïve and experienced BE production is not affected by this correction.
The time course of additional metabolites is provided in Fig. 8 , including cocaine metabolites anhydroecgonine (AHE) (m/z 168) and hydroxybenzoylecgonine (HOBE) (m/z 306) (shown in bold in Fig. 7 ), as well as several unknown metabolites (m/z 330, m/z 475, m/z 678). Some of these additional metabolites have higher abundance in the cocaine-experienced population while others have no overall change in abundance. Overall, BE, AHE, and m/z 645 show significant increases from naïve to experienced cell population ( p-values = 5 . 7 x 10 –4 , 1 . 12 x 10 –3 , and 1 . 60 x 10 –3 , respectively). While AHE is a typical product of AEME (the pyrolysis product of cocaine), some reports have shown that the metabolic pathway from cocaine into AEME and AHE could result from loss of water of EME or ecgonine [ 30 ]. While there is much evidence that AEME and AHE can form as a result of the analysis technique, this is typical of gas chromatography separations that require vaporization of compounds, thus risking alteration of thermolabile compounds such as cocaine and its metabolites [ 31 ]. Electrospray ionization is utilized when this type of compound is under investigation.
10.1371/journal.pone.0117685.g008
Fig 8
Additional metabolite time course data compared to benzoylecgonine (BE).
Experimental conditions for each group of cells are shown above the graph with solid black lines indicating exposure to cocaine media. Anhydroecgonine (AHE) and hydroxybenzoylecgonine (HOBE), two additional metabolites of cocaine, provide examples of both variation between naïve and experienced cell groups in the case of AHE and consistency between these two groups in the case of HOBE. The increases in BE and AHE from naïve to experienced groups are statistically significant with respective p values of 5 . 7 x 10 –4 and 1 . 12 x 10 –3 . Three unidentified metabolites (m/z 645, m/z 478, and m/z 330) that contribute to the separation between media exposure groups are also shown. m/z 330 and m/z 478 show no statistical significance between naïve and experienced cell groups while the increase in m/z 645 is statistically significant ( p = 1 . 6 x 10 –3 ) .
In-Culture Cellular Analysis
To compare this instrumentation platform, as well as the resulting biological data of naïve and cocaine-experienced T cells, with the current standard in mass spectrometry analysis of biological samples, we replicated the online experiment in culture using UPLC-ESI-IM-MS. Fig. 9 shows the principal component analysis plot demonstrating sufficient variance when comparing steps 1 and 3 (plain RPMI 1640 exposure) with steps 2 and 4 (cocaine exposure). In the online experiment, we are able to see separation between cocaine-experienced cells and naïve cells, a separation that is absent from the in-culture study, with the exception of the initial RPMI exposure of the naïve cell populations. While one major difference is the number of time points per step of media exposure, as discussed previously, replicating the 9-minute time point sampling of the online system would confound the length of time needed for media switching, as well as inflict unnecessary stress on the cells.
10.1371/journal.pone.0117685.g009
Fig 9
Exometabolomic profiles of naïve and cocaine-experienced Jurkat T cells from in-culture analysis.
Walking principal component analysis of exometabolomic profiles of naïve (shades of yellow) and cocaine-experienced (shades of turquoise) Jurkat T cells using in-culture analysis. Each numerically labeled data point represents the end point of a 54-minute exposure to the indicated media. As with the online experiment, during steps 1 and 3 both cell populations received plain RPMI, while during steps 2 and 4 both received cocaine in RPMI at 60 μg/mL. Variance between steps 1 and 3 and steps 2 and 4 results in unsupervised grouping according to media exposure type. Separation between naïve and experienced cell groups, particularly when under cocaine exposure, does not occur to the same degree, as yellow and turquoise points corresponding to steps 2 and 4 are heavily intermixed.
Conclusions
Platform Capabilities
We have described the integration, workflow, and proof of concept of a technology platform for near real-time detection of the dynamic cellular exometabolome based on the combination of a microfluidic bioreactor, an online SPE desalting arrangement, and mass spectrometry. A great strength of this platform is its adaptability to different cell types and experimental conditions. Microfluidic cell trapping devices can be customized to the size of any adherent or non-adherent cell type, and they provide a solution to the dilution issues found in traditional well plate experiments. In this work, we demonstrate cell trapping and experimentation on naïve and cocaine-experienced Jurkat T cells. While this work shows only one model system based around cocaine exposure, the environmental stimuli are limited only by the number of pumps one has available for providing variable perfusion conditions and the temperature change and gas exchange rates an incubator is capable of generating. Studies are under way to incorporate low-cost micropumps commensurate with a microfluidic platform, such as those reported by Darby et al . [ 32 ]. Thus, this platform is directly amenable to exposure of cells to other environmental drugs or toxins. The online SPE desalting arrangement allowed for sufficient desalting to permit a temporal resolution of 9 minutes. While this temporal resolution is not by any means a significant advance for cellular chemical detection methods, because it is dictated by the analyte abundance and detection capabilities of the time-of-flight mass analyzer, it is (depending on the mass analyzer) the best resolution possible for this biological system using mass spectrometry. Because it is trivial to scale the loading time in this arrangement based on detection power, we believe a technology platform of this general system will be of considerable utility to the biological community, particularly as mass analysis detectors improve in the coming years.
Cellular Memory of Cocaine Experience
Cocaine metabolism in naïve and experienced Jurkat T cells was investigated with this near real-time platform developed for the study of the cellular exometabolome. While it is well known that cocaine has an effect on immune cells, there has been no prior demonstration (though the idea has been suggested [ 33 ]) of even a short-term immune cell memory of prior cocaine exposure. With the advent of this innovative online platform, unique metabolic signatures ( Fig. 5 ) are obtained that are absent from the “in-culture” data ( Fig. 9 ) or perhaps lost due to the increased sample processing and UPLC analysis time. A concentration of cocaine higher than typically found in circulation was applied to cells to ensure that a cellular response was achieved for purposes of platform validation, not for modeling the in vivo conditions. Upregulation of cocaine metabolism into benzoylecgonine in experienced cells demonstrates one contributor to the unique exometabolomic profile resulting from previous cocaine experience. Anhydroecgonine, as well as unknown metabolites m/z 645 and m/z 478, are also upregulated in cell populations with prior cocaine exposure, leading to the possibility of indicators of immune cell memory of cocaine other than cocaine metabolites alone.
While there is a previously reported non-enzymatic degradation rate of cocaine into benzoylecgonine at physiological temperatures and pH [ 34 ], we were able to confirm a rate specific to this platform. Through comparison of the BE to cocaine ratios from naïve and experienced cell experiments, as well as the platform absent of cells, it is evident that the portion of BE abundance from non-enzymatic degradation does not entirely explain the significant increase in BE during the cocaine exposure steps in the experienced cells, indicating that the response is due to a unique exometabolomic profile of T cells with prior cocaine exposure. Further analysis of cellular memory of cocaine exposure, in particular at a range of concentrations, is warranted based upon these findings.
Supporting Information
S1 Fig
PDMS silanization scheme.
Hydrolysis of methoxy group from PEGn trimethoxysilane causes the formation of silanol groups (a). PDMS activation by O 2 plasma (b (top)), silane deposition (b (middle)), condensation of the silane into chains (b (bottom)), hydrogen bond formation between silanol and oxidized PDMS surface (c (left)), and covalent bond formation between silane and PDMS surface (c (right)) complete the silanization process.
(JPG)
S2 Fig
Reduction of nonspecific adsorption by PDMS silanization.
PDMS channels were perfused with FITC-insulin for 30 minutes (green), followed by 30 minutes of rinsing with Ringer’s solution (pink). The top three fluorescent images show the surface-bound insulin at 1 minute, 31 minute, and 60 minutes, with the bottom three images corresponding to the surface-modified channels. Silanized PDMS channels (blue line) retain (at maximum) only 25% of the total insulin retained by untreated PDMS (red line).
(JPG)
S1 File
PDMS Surface Modification for Reduction of Non-Specific Adsorption.
(PDF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
History
Our core resource laboratory maintains the three schistosome species that inflict the greatest human disease burden: Schistosoma mansoni , S. haematobium , and S. japonicum . Figure 1 shows the three species of snails (intermediate hosts) used for maintaining the three schistosome species above.
10.1371/journal.pntd.0000267.g001 Figure 1
The Three Species of Intermediate Hosts That Are Provided from The Schistosomiasis Center.
From left, Bulinus truncatus truncatus (host for S. haematobium ), Biomphalaria glabrata (host for S. mansoni ), and Oncomelania hupensis hupensis (intermediate host for the Chinese isolate of S. japonicum ; other subspecies of O. hupensis are also available). For size comparisons, the B. glabrata shown is approximately 5 mm in shell diameter.
National Institutes of Health–National Institute of Allergy and Infectious Diseases (NIH-NIAID) funding to support this effort began in 1967, with the first contract award going to the University of Michigan. These early years were especially important for developing optimal conditions for lab maintenance of the very fastidious Oncomelania sp. hosts for S. japonicum [1] – [4] . In subsequent years, the resource center was maintained at the University of Massachusetts–Lowell (1977–1995), and, since 1995, at the Biomedical Research Institute (BRI; Rockville, Maryland, United States of America). The original intent of the resource—to provide living schistosome life cycle stages to research investigators at no cost— still comprises the dominant portion of the effort. As will be discussed in more detail later, the work has been expanded in the past few years to keep up with the increasingly diversified interests of researchers.
Aside from shipping charges, living life cycle stages for the schistosome species and strains are provided free of charge as prepatent infections in the intermediate snail hosts, or as early infections in the definitive hosts (mouse or hamster). Due to space and funding considerations, the contract does not allow BRI to house the snails or mammals after infection, throughout the entire prepatent snail stage, or throughout the immature-to-mature stage in the definitive hosts. Therefore, the recipient labs must make some commitment to house the snails and/or mammals beginning about 1 week after exposure to the parasite.
In any given year, the resource provides life cycle stages and other materials to between 35 and 50 separate laboratories worldwide, with the majority in the US ( Figure 2 ). No distinction is made in Figure 2 regarding the proportion of snail versus mammal (mouse and/or hamster) shipments, but about two-thirds of total shipments are snails only. Approximately 39% of the recipient laboratories are located in graduate departments of universities, and 44% are in medical or veterinary schools. The remaining 17% are located in undergraduate institutions, where the investigators have active schistosomiasis research programs and/or use schistosomes as a teaching tool for biology and parasitology classes, taking advantage of the fact that schistosomes are good representative teaching models for helminth infections in general.
10.1371/journal.pntd.0000267.g002 Figure 2
The Geographic Location of Recipient Laboratories of Snails and/or Mammals Provided by the NIH-NIAID Schistosomiasis Resource Center.
Schistosome Species Provided
Schistosoma mansoni
Of the three schistosome species provided by this resource, S. mansoni has been, by far, the subject of most schistosomiasis research. It is the easiest of the three major schistosome species to maintain in the laboratory and, not surprisingly, it is first in terms of sheer numbers of infected snails and mammals requested. It is also the only species of the three provided that is endemic in several countries of both the Eastern and Western hemispheres. Of great historical importance in expanding schistosomiasis research was the work in Puerto Rico and Brazil with S. mansoni in the 1950s and 1960s, and Puerto Rican isolates of S. mansoni still are used widely as representative strains in the S. mansoni research community. Our two Puerto Rican isolates of S. mansoni (NMRI and PR-1) are maintained in the planorbid snail Biomphalaria glabrata , which is the major intermediate host for S. mansoni in the Western hemisphere. The most commonly used B. glabrata snail stock for maintenance of S. mansoni arose from an albino mutant stock from research by Newton [5] . Not only did it prove to be highly susceptible to S. mansoni , but it also allowed investigators using a dissecting microscope to examine intramolluscan development of the parasite, otherwise made difficult by the normal black pigmentation typically found in snails isolated from the field.
Schistosoma japonicum
Laboratory maintenance of S. japonicum is the most difficult of the three schistosome species provided by this resource. Workers experienced with this parasite's intermediate snail host ( Oncomelania sp.) have long been frustrated by the difficulty of coercing this species to grow vigorously in laboratory settings [1] . Its breeding characteristics, size, and cercarial shedding behavior in the laboratory are vastly different from that of its Biomphalaria and Bulinu s counterparts. S. japonicum –induced pathology in the mouse, on a worm pair comparison, is considerably more pronounced than that of S. mansoni infections, with only one or two S. japonicum worm pairs producing dramatic granulomatous disease in the laboratory mouse by 7 weeks post-cercarial exposure. Two subspecies of S. japonicum are provided; one a Philippine isolate, the other from China. These two subspecies are provided either in Oncomelania hupensis quadrasi (for the Philippine isolate) or in O. h. hupensis (for the Chinese isolate).
Schistosoma haematobium
Although S. haematobium is of wide-ranging clinical importance throughout much of Africa and the Middle East, significantly fewer requests have been made for this species from this resource than for S. mansoni or S. japonicum . This is probably because studying S. haematobium infections in the laboratory is complicated by the absence of a small laboratory animal model in which pathology resembles the human infection with this parasite. Through the years the standard mammalian laboratory host for S. haematobium has been the Syrian golden hamster. Although the parasite develops to maturity, the urogenital system of the hamster is not significantly involved in the disease. On the contrary, upon dissection, the involvement of the liver and intestinal tract resemble an infection with S. mansoni , and the majority of the adult worms can be found in the mesenteric venules, as with an S. mansoni infection. It is unfortunate for experimental studies that a more suitable small laboratory host has not been found that better approximates the infection in humans, since one of the more intriguing aspects of its infection in humans is its association with bladder cancer [6] , [7] . It is our belief that, if for no other reason, this alone should give investigators even more incentive to study this important parasite. The snail hosts for laboratory maintenance of this parasite, members of the genus Bulinus , are relatively easy to propagate in the laboratory, and their requirements for successful maintenance are much like those for Biomphalaria . This resource provides an Egyptian isolate of S. haematobium , maintained in Bulinus truncatus truncatus snails and in hamsters.
Impact of the Resource on the Schistosomiasis Research Community
It would be impossible to give an accurate number of publications in experimental schistosomiasis that have been made possible through the use of this resource over the 40-plus years of its existence. Certainly it would be in the thousands. At least two barometers, however, can give some idea of its impact on the schistosomiasis research community. Browsing through the past 10 years of presentations in experimental schistosomiasis from the annual meetings of The American Society of Tropical Medicine and Hygiene (ASTMH) shows that roughly 80% of those presentations were from laboratories that have relied upon material support from this resource. Another indication of its importance comes from PubMed searches. Over the last 3 years, approximately 90% of the experimental schistosomiasis papers are from laboratories relying on these materials.
Other Services
In recent years, additions have been made to the contract to meet other needs in schistosomiasis research.
Training
In the spring and fall of each year, our laboratory offers a training course in S. mansoni maintenance, allowing investigators to gain more experience with the various schistosome life stages and several of the procedures most common to schistosomiasis research laboratories. Our overriding concern is to help investigators avoid parasite-related problems that might lead to erroneous experimental outcomes—a common problem in many labs that handle these complex life cycles. An added benefit to the community is to establish standard, basic techniques to make each investigator's life cycles more productive. It is hoped that this effort will help investigators avoid, as much as possible, unproductive experiments due to some fundamental problem with the life cycle. Currently, each class is 2 days long and is limited to eight attendees per class. The attendees have ranged from experienced researchers and/or their students and technicians who maintain their own life cycles, to those who are new to the field of schistosomiasis research and lack prior hands-on experience with S. mansoni .
Cryopreservation of Schistosome Strains
Techniques have been developed to cryopreserve schistosomules as a means of long-term storage, foregoing the need for continually passing the parasite through snail and mammalian hosts [8] , [9] . Apart from the convenience of cryopreserving the parasites and restoring the strain only when needed, this technique can be of great practical benefit for studies such as those examining genetic changes in schistosomes as a result of several cycles of selection pressure under experimental conditions [10] . In addition, cryopreservation offers the possibility of storing large numbers of different field isolates for later comparative studies. Most of what is known about cryopreserving schistosomes comes from the work with S. mansoni , and with present techniques, approximately 2%–5% of the cryopreserved and thawed schistosomules can mature in mice. Although this is considerably lower than the typical 40%–50% maturation rate of skin-penetrating cercariae in mice, it has been our experience that if at least 20,000 cercariae are in the starting (pre-cryopreserved) population, recovery is usually sufficient to re-establish the strain once thawed [11] . BRI offers the service of cryopreserving individual S. mansoni strains, keeping them in liquid nitrogen for long-term storage, and re-establishing the strain in mice when requested.
The Schistosome Related Reagent Repository (SR3)
The Schistosome Related Reagent Repository (SR3) was launched in 2003 to serve as a central facility for the collection and long-term storage of schistosome and snail–host related reagents. This allows scientists: 1) improved access to parasite, snail, and related reagents; and 2) access to standardized reagents in a renewable form. This repository was developed to provide community-wide access to standardized and well-characterized materials that are generated and deposited by other investigators. At this time, the SR3 maintains a relatively small collection of S. mansoni cDNA libraries, and several B. glabrata –related molecular reagents (primarily cDNA libraries from various tissues). We anticipate that the repository will soon include a wider range of schistosome- and snail-related reagents such as: (a) genomic libraries, cDNA libraries, glycerol stocks and bacterial stabs of recombinant clones, recombinant plasmid DNA, oligonucleotide probes, and PCR primers; (b) antisera against schistosome products (storage only); and (c) reagents added on an as-need basis, such as high-density filters of gridded bacterial artificial chromosome, expressed sequence tag, and cosmid libraries.
The SR3 also continuously maintains, in culture, the B. glabrata embryonic ( Bge ) cell line, derived from the susceptible M-line snail [12] . This cell line is of great benefit for exploring molecular signaling events underlying the host–parasite interaction of S. mansoni . The cell line also has application for karyotyping and mapping the snail genome as part of the genome sequencing project. Hands-on training is available to help researchers with culturing this cell line and setting up co-cultures with sporocysts.
Conclusion
From its beginning, the NIH-NIAID-supported core schistosomiasis resource center has been a major driving force in the progress of schistosomiasis research. Not only has this been important in helping to decrease the health burden of schistosomiasis, but investigators are increasingly appreciating that a schistosome infection provides a good model system for studying a variety of inflammatory, allergic, and granulomatous diseases. For example, S. mansoni infection in mice provides an elegant model for helping unravel contributions of separate helper T cell populations (Th1 and Th2) in the development of asthma, allergic inflammation, and fibrosis [13] . Thus, laboratory studies of schistosomiasis are leading to advances in medicine that likely will have implications far beyond that of controlling the disease itself. Availability of this schistosome resource thus can serve as a foundation for individuals, not only in tropical medicine research, but in many diverse areas of basic medical research.
Box 1. Summary Points
For over 40 years, NIH-NIAID has supported a schistosome resource center to provide research material, free of charge, to principal investigators.
To improve efficiency, conditions have been developed to simultaneously house S. mansoni , S. haematobium , and S. japonicum .
The resource facility has been instrumental in creating a uniformity of source and standardization of study protocols.
The resource has been expanded in recent years to house schistosome-related molecular resources for genomic and proteomic research.
For capacity strengthening, training elements exist for established investigators and for those new to the field.
Web sites that can be accessed for information on provisions of the schistosome life cycle stages, schistosome-related molecular reagents, and other repository activities funded by NIH-NIAID are:
NIAID Schistosomiasis Resource Center, http://www.schisto-resource.org/
Schistosome Related Reagent Repository (SR3), http://www.afbr-bri.com/sr3/
NIAID Research and Development Contracts, http://www.niaid.nih.gov/contract/
Box 2. Testimonials from Scientists Receiving Materials from the Schistosomiasis Resource Center
As a relative newcomer to the field, this resource has been essential in helping us establish our research program. The training element and continued support have been invaluable for our technicians and graduate students. (Stephen Davies, USUHS, Bethesda, Maryland, United States of America)
Without this resource our research would have been impossible. The patient training, troubleshooting, and help during emergency needs are appreciated from the entire community of schistosomiasis researchers. (Miguel Stadecker, Tufts University, Boston, Massachusetts, United States of America)
This facility has been of immense help to our research on Schistosoma japonicum, and has provided extensive training in life cycle maintenance with this parasite. Serving such a large group of schistosome researchers, it is truly an invaluable resource and something that must be continued. (Don McManus, Queensland Institute of Medical Research, Brisbane, Queensland, Australia)
This is a highly valued resource to the entire blood fluke community. The consistency of mouse infections has always been highly uniform, such that we never have to be concerned with “batch-to-batch” variation in parasite yields or quality. The provision of supplemental numbers of snails during times of need has also been of great benefit and crucial to the success of our research program. (Tim Yoshino, University of Wisconsin-Madison, Wisconsin, United States of America)
My laboratory would not be able to perform studies on development of vaccines or immunoregulation without the life cycle support provided on this contract. Like clockwork, the snails are provided each month. A truly great resource. (Don Harn, Harvard University, Boston, Massachusetts, United States of America)
Box 3. Five Key Original Studies and Reviews Made Possible by the Materials/Stages from the NIAID Schistosomiasis Resource Center
1. Boros DL, Warren KS (1970) Delayed hypersensitivity granuloma formation and dermal reaction induced and elicited by a soluble factor isolated from Schistosoma mansoni eggs. J Exp Med 132: 488–507.
This study led to what we now know as the immune response to schistosome egg antigens being critical in driving the pathology of the disease.
2. Georgi JR (1982) Schistosoma mansoni : quantification of skin penetration and early migration by differential external radioassay and autoradiography. Parasitol 84: 263–281.
Application of this parasite tracking technique cleared up many of the issues of natural and vaccine-induced immunity in experimental animals.
3. Pearce EJ, MacDonald AS (2002) The immunobiology of schistosomiasis. Nature Rev Immunol 2: 499–511.
Comprehensive review of the immune response in schistosomiasis infections in the mammalian host. The complexities of the immune balance between the acute versus chronic phase responses are well described, and the role of Th1- and Th2-type responses in the search for vaccines is discussed.
4. Loker ES, Bayne CJ (2001) Molecular studies of the molluscan response to digenean infection. In: Beck G, Sugumaran M, Cooper EL, editors. Phylogenetic perspectives on the vertebrate immune system. New York: Plenum Publishers. pp. 209–222.
This review serves as groundwork for the burgeoning studies on the molecular phases of the intramolluscan development of digeneans, and in particular schistosomes.
5. McCutchan TF, Simpson AJG, Mullins JA, Sher A, Nash TE, Lewis F, Richards C (1984) Differentiation of schistosomes by species, strain, and sex by using cloned DNA markers. Proc Nat Acad Sci USA 81: 889–893.
Showed the feasibility of using molecular markers, in this case cloned ribosomal gene segments, for differentiation in population genetic studies for schistosomes.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Bacteria communicate via signal molecules either produced by the bacteria themselves, by the host organism or molecules in the environment. One such process of bacterial cell-cell signaling is quorum sensing (QS), that enables bacteria to sense and respond according to cell population density, and to regulate virulence gene expression [ 1 , 2 ]. Interference with QS may provide novel anti-virulence strategies to fight pathogenic bacteria [ 3 ]. The autoinducer-2 (AI-2) QS molecule is one of the most extensively studied, and AI-2 has been recognized as an intra- and inter-species communication signal. The AI-2 synthase, LuxS, is encoded by luxS homologues found in several different bacterial species [ 4 – 6 ]. The substrate of LuxS is S-Ribosylhomocysteine, which is cleaved to yield homocysteine and 4, 5-dihydroxy-2, 3- pentandione (DPD). DPD cyclizes spontaneously to form AI-2 [ 7 ].
In Escherichia coli , AI-2 binds the periplasmic receptor LsrB at the bacterial envelope [ 8 , 9 ]. This initiates uptake of AI-2 via the AI-2 transporter formed by two transmembrane proteins, LsrC and LsrD, and an ATPase, LsrA that provides energy to the AI-2 transport. Intracellular AI-2 is phosphorylated by a kinase, LsrK. The lsr operon is repressed by the lsr repressor; LsrR. Phosphorylated AI-2 inactivates LsrR resulting in transcription of lsr- regulated genes and subsequent increased uptake of AI-2. Thus, AI-2 regulates its own uptake.
In E . coli , AI-2 regulates the virulence factors involved in biofilm formation and motility [ 10 ]. Adhesion is a prerequisite for bacterial colonization of both abiotic and biotic surfaces [ 11 ]. Studies have shown increasing adhesion of pathogenic E . coli to epithelial cells induced by AI-2 signaling [ 12 , 13 ].
In addition to AI-2, other autoinducers such as AI-3 have been identified in E . coli [ 14 ]. AI-3 regulates gene expression through the two-component signaling system QseBC [ 14 ], with QseB being the regulator and QseC the sensor kinase [ 15 ]. Homologues of QseC are found in many important human and plant pathogens [ 16 , 17 ], suggesting an important evolutionary role. Prokaryotes and eukaryotes have coexisted for millions of years, and have consequently co-evolved to sense and respond to each other`s signaling molecules [ 18 ]. E . coli responds to hormones like epinephrine and norepinephrine through QseC located in the membrane [ 14 , 19 , 20 ]. QseC acts as an adrenergic receptor that activates virulence genes in response to inter-kingdom cross signaling [ 16 , 21 ]. Both epinephrine and norepinephrine have been shown to enhance growth and virulence in E . coli [ 22 ], and to increase motility and adhesion to HeLa cells by E . coli EHEC O157:H7 [ 23 ]. Because of the involvement of epinephrine and norepinephrine in bacterial signaling, we assessed possible interference of thiophenone with this signaling.
Several different chemical compounds have been identified as quorum sensing inhibitors (QSI) e.g. halogenated furanones [ 24 – 27 ]. Furanones were isolated from the red macroalgae Delisea pulchra and were discovered due to their capacity to inhibit bacterial growth and biofilm formation [ 28 ] by interference with AI-2 signaling [ 29 , 30 ]. Sulphur analogues of furanones, thiophenones [ 31 ], have been shown to effectively inhibit biofilm formation in various bacteria, including Staphylococcus epidermidis , E . coli and Vibrio harveyi [ 32 – 35 ], at non-toxic concentrations [ 33 ].
We have previously shown that biofilm formation and motility in E . coli O103:H2 are reduced by both furanone F202 [ 35 , 36 ], and its sulfur analogue thiophenone TF101 [ 35 ], with TF101 being the most efficacious [ 35 ]. Reduced motility by TF101 was explained by interference with the flagella synthesis, through reduced expression of flagella genes ( flhD ) [ 35 ], genes that are regulated by AI-2 [ 15 ].
This study aimed to elucidate the mechanisms of action of the quorum sensing inhibitor TF101 in E . coli O103:H2. The hypothesis was that TF101 interferes with virulence factors such as adhesion and biofilm formation regulated by AI-2, epinephrine or norepinephrine.
Materials and Methods
Thiophenone
Thiophenone TF101, ( Z )-5-(bromomethylene) thiophen-2 (5 H )-one ( Fig 1 ), was synthesized as reported previously [ 31 ]. Thiophenone was dissolved in 70% ethanol at 50 mM and stored at −20°C.
10.1371/journal.pone.0157334.g001
Fig 1
The chemical structure of (Z)-5-(bromomethylene)-thiophene-2(5H)-one (TF101).
AI-2
Synthetic DPD ((S)-4, 5-dihydroxy-2, 3-pentanedione, OMM Scientific Inc., TX, USA) was the source of AI-2 used.
Bacterial strains and culture media
Atypical enteropathogenic (aEPEC) Escherichia coli O103:H2 strain 2006-22-1153, isolated from sheep was used as a model organism in this study ( Table 1 ). The strain was verified and characterized at the National Reference Laboratory at the Norwegian Veterinary Institute. E . coli ABU83972 (OR: K5:H − , lsr − ), originally isolated from a young Swedish girl [ 37 ], was included in the biofilm assay. The strains were stored at -80°C in LB broth (Difco) supplemented with 15% glycerol, and recovered on LB agar plates (bacto-trypton 10 g/L, yeast extract 5 g/L, agar 15 g/L) at 37°C overnight. The bacterial cultures were transferred into LB broth and incubated with shaking at 37°C for 5 h to obtain working cultures. For biofilm experiments, E . coli O103:H2 was grown in LB without NaCl (bacto-trypton 10 g/L, yeast extract 5 g/L) [ 38 ], hereafter called LB b , whereas E . coli ABU83972 was grown in LB broth (Difco). The differences in the growth media used in the biofilm experiments are due to differences in the preferred growth conditions in the two strains.
10.1371/journal.pone.0157334.t001
Table 1 Bacterial strains used in this study.
E . coli strain
Characteristics
Phenotype
Reference
2006-22-1153 (1153)
Enteropathogenic E . coli O103:H2
Stx1 - , stx2 - , eae +
[ 52 ]
83972
ABU isolate (OR:K5:H − )
lsr -
[ 37 ]
Sample preparation, RNA isolation and qPCR
Overnight culture of E . coli O103:H2 was diluted in LB broth to OD 600 = 0.01, and incubated with shaking at 37°C. When the culture reached OD 600 = 0.5, 10 μM TF101, 10 μM AI-2, or 10 μM TF101 and 10 μM AI-2 in combination, was added. Bacteria in plain LB were included as control. The cultures were allowed to grow, and pellets were collected every hour by centrifugation (2000g x 4°C x 5min) and stored at– 80°C. Total RNA was isolated from harvested E . coli using the High Pure RNA isolation kit (Roche Applied Science, Mannheim, Germany) according to the manufacturer`s protocol. In addition to the DNase treatment included in the RNA isolation protocol, an additional DNase treatment was performed using Turbo DNase (Thermo Fisher Scientific Inc.). cDNA was synthetized using MMLV Reverse transcriptase 1 st -strand cDNA Synthesis Kit (Epicenter Biotechnologies) according to the manufacturer`s protocol. The primer pairs used are listed in Table 2 . Real time reactions were performed using the Thermo Scientific Maxima SYBR Green/ROX qPCR Master Mix (ThermoScentific), and real-time amplification was carried out using the Stratagene Mx3005 P Multiplex Quantitative PCR systems (Stratagene, La Jolla, CA). The gradient thermocycling program was set for 40 cycles at 95°C for 15 s, 59°C for 30 s, and 72°C for 30 s, with an initial cycle at 95°C for 10 min. The data were collected and analyzed by normalization against the housekeeping gene rpoA using the MxPro software.
10.1371/journal.pone.0157334.t002
Table 2 Primers used in this study.
ILW009_fimH
Forward
5`-atattgctgagtccacccgc-3`
ILW009_fimH
Reverse
5`-ttgcgtccaagtaccaccag-3`
ILW004_rpoA
Forward
5`-caaccattctggctgaacaa-3`
ILW004_rpoA
Reverse
5`-gcggacagtcaattccagat-3
ILW001_lsrB
Forward
5`-cggagtgccgctcttactac-3`
ILW001_lsrB
Reverse
5`-gtaacggtggggcttgagta-3`
ILW010_eae
Forward
5`-actgtggctcgatttgctga-3`
ILW010_eae
Reverse
5`-ctccgattcctctggtgacg-3`
Protein ligand interaction in silico
The protein ligand binding of AI-2 and TF101 in LsrB (LsrB from Salmonella typhimurium; PDB; http://www.rcsb.org/pdb/ ) [ 39 ] was predicted by using PyRx virtual screening tool (ver 0.9.2), followed by a visualization of the interaction using PyMol (ver 4.0).
Adherence to epithelial cells
The colorectal adenocarcinoma cell line Caco-2 was used as epithelial cells. The cells were grown in RPMI- 1640 (Sigma-Aldrich) with 2 mM L-glutamine (Sigma-Aldrich), 10% FBS and 1% Antibiotic Antimitotic Solution (Sigma-Aldrich), in 5% CO 2 at 37°C. Following trypsination, the cells were washed once in complete RPMI- 1640, and 1 mL seeded at a concentration of 4 x 10 5 cells per mL in 24 well plates (Nunc, Thermo Fisher Scientific) and grown to confluence.
E . coli O103:H2 incubated with aeration overnight in LB b medium was washed in PBS before centrifugation (5000g x 4°C x 5 min). The bacteria were re-suspended in RPMI-1640 without antibiotics and added to confluent Caco-2 cells in 24- well plates, to a multiplicity of infection (MOI) of 40:1. TF101 was added (10 μM final concentration) to assess the effect of TF101 on adhesion. Epinephrine (50 μM final concentration, Sigma-Aldrich, USA), norepinephrine (50 μM final concentration, Sigma-Aldrich, USA), or AI-2 (10 μM final concentration) was added in triplicate wells to assess their effect on adhesion. The plates were incubated for 4 h, 37°C. The cells were then washed twice with PBS to remove non-adherent E . coli and the Caco-2 cells were lysed with 0.1% Triton-X100. The lysates were diluted and plated on LB agar plates for CFU counts. Cells infected with bacteria without chemicals were included as negative control. To confirm that TF101 had no effect on receptors on the Caco-2 cells, the cells were exposed to TF101 prior to the adhesion assay in separate experiments. Fresh RPMI medium with TF101 (10 μM final concentration) was added to the cells. The cells were incubated at 37°C (5% CO 2 ) for 60 minutes. The medium with TF101 was removed; the cells washed twice with PBS to remove any remaining TF101, and fresh RPMI was added to each well. The bacteria were added to the cells and adhesion was quantified as described above.
Potential bacterial invasion was assessed using gentamicin protection assay as previously described [ 40 ], with some modifications. After the infection period described above, the Caco-2 cells with adherent E . coli were washed twice with PBS and fresh RPMI with gentamycin (50 mg/ L) was added to the wells to kill extracellular bacteria. The MIC of gentamycin was determined as 6 mg/L prior to the invasion experiment. RPMI without antibiotics was added to control wells. After 1 h incubation, the wells were washed four times with PBS to remove antibiotic residues. The cells were lysed by 0.1% Triton-X100. The lysates were serially diluted and plated on LB plates as above for CFU count.
Scanning electron microscopy (SEM) was used to visualize adherent E . coli on Caco-2 cells. Caco-2 cells were grown to confluence on polystyrene coverslips (Nunc Thermanox Coverslips, Thermo Scientific, Rochester, NY) with adherent E . coli as described above. The samples were fixed with 2.5% glutaraldehyde in 0.1 M Sørensen phosphate buffer and stored at 4°C until processed and examined by SEM (model XL 30 ESEM, Philips, Eindhoven) as described previously [ 30 ].
Cytotoxic effect of thiophenone
A possible cytotoxic effect of TF101 on Caco-2 cells was assessed by using the lactate dehydrogenase (LDH) release assay (CytoTox 96 Non-Radioactive Cytotoxicity Assay kit; Promega, Madison, WI). Caco-2 cells were cultured as described previously. A total of 50 000 cells/ mL were seeded in flat-bottom 24-well polystyrene microtiter plates (Nunc) and grown to confluence. The growth media were discarded and the Caco-2 cells were exposed to different concentrations of thiophenone TF101 (0 μM, 2.5 μM, 5 μM, 10 μM or 50 μM final concentration) dissolved in fresh RPMI. The cells were further incubated for 4 h, before the absorbance of the supernatant was measured (490 nm) according to the manufacturer`s protocol using the Synergy HT Multi-Detection Microplate Reader (Biotek).
Biofilm formation
The effect of TF101 on biofilm formation by lsrB proficient ( E . coli O103:H2, lsr + ) and non-proficient ( E . coli ABU83972, lsr −) strains was assessed and compared. Overnight cultures in LB medium were diluted 1:1000 in fresh LB b for E . coli O103:H2, and LB for E . coli ABU83972, incubated with aeration at 37°C for 5 h and diluted 1:200 in the respective fresh media (OD 600 = 0.02). To assess the effect on biofilm formation, TF101, AI-2, epinephrine or norepinephrine was added to the bacterial suspensions of E . coli O103:H2 at final concentrations of 10 μM, 10 μM, 50 μM or 50 μM, respectively. In order to investigate possible interference of TF101 with the different signaling systems, 10 μM TF101 was added simultaneously with 10 μM AI-2, and 50 μM of epinephrine or norepinephrine, in the biofilm assay.
The effect of TF101 and AI-2 on the ABU83972 ( lsr − strain) was tested by adding TF101 to the cultures at final concentrations of 5 μM, 10 μM or 50 μM, and AI-2 at a final concentration of 10 μM. Samples of 200 μL were added to flat-bottom, 96-well polystyrene microtiter plates (Nunc, Thermo Fisher Scientific). The plates were incubated statically for 48 h at 20°C for E . coli O103:H2 and overnight at 37°C for ABU83972, according to growth conditions required by the respective strains. Biofilm quantity was assessed after removing the planktonic cells by inverting the plates and washing the wells twice with 0.9% NaCl. Adherent cells were stained with 0.1% safranin solution for 30 min, followed by washing at least three times with 0.9% NaCl. The safranin stain was released with 30% acetic acid, and OD 530 nm was measured (Synergy HT Multi-Detection Microtiterplate Reader, Biotek, VT). The assay was performed in six parallels, and the experiment was repeated twice. The biofilm mass was calculated as % of control.
Statistical analysis
All experiments were performed as minimum two independent experiments with at least three parallels of each sample, using freshly prepared reagents. One-way ANOVA followed by Student- Newman- Keuls method was used for the comparisons in the adhesion test and the biofilm analysis involving the effect of TF101 and epinephrine/ nor-epinephrine, and the differences in gene expression. The effect of AI-2 on biofilm formation was determined using t-test. For all statistical analyses, the level of statistical significance was set at P < 0.05.
Results
Thiophenone interferes with AI-2 signaling
The possible interference of thiophenone with the AI-2 signaling system was investigated by measuring expression of the lsrB gene, encoding the AI-2 binding receptor, using quantitative real-time PCR (qPCR) with samples collected after 2 hours of exposure to AI-2, TF101 or AI-2 and TF101 in combination. Exposure of E . coli to 10 μM TF101, reduced expression of lsrB , while addition of AI-2 gave a significantly increased expression of lsrB ( P < 0.01). The increase in gene expression in response to AI-2, was attenuated by TF101 ( Fig 2a ).
10.1371/journal.pone.0157334.g002
Fig 2
The effect of TF101 or AI-2 on expression of lsrB , fimH and eae in E . coli O103:H2.
(a) TF101 significantly reduced expression of lsrB , while AI-2 gave a significant increase in expression of lsrB . TF101 and AI-2 added simultaneously attenuated the increase in lsrB expression following AI-2 stimulation (b) Expression of fimH and eae was significantly decreased following exposure to TF101, while AI-2 gave a significant increase in the expression of the same genes. (0 = control without exposure to TF101, AI-2 or both, respectively). The data are presented as mean values ± SD (n = 6). *Significantly different from control ( P < 0.05).
The binding affinity of TF101 to LsrB receptor was predicted using PyRx/Autodock Vina software. The program predicted binding of TF101 binding sites in the AI-2 binding pocket of the LsrB receptor from Salmonella ( Fig 3 ). Pairwise amino acid alignment of lsrB from E . coli and Salmonella are found in the supplementary material ( S1 Fig ). The predicted binding of AI-2 was estimated to be -7.1 kcal / mol , while the binding affinity of TF101 was -4.2 kcal / mol .
10.1371/journal.pone.0157334.g003
Fig 3
In silico interaction of TF101 and AI-2 with the LsrB receptor.
In silico analyses predicted possible protein ligand interaction of TF101 and the LsrB receptor, indicating that TF101 might act as a competitive antagonist for the AI-2 receptor.
Thiophenone TF101, AI-2, epinephrine and norepinephrine affect adherence to epithelial cells
Adhesion of E . coli to epithelial surfaces is the first step in colonization; we therefore investigated how thiophenone may interfere with AI-2 and inter-kingdom signaling molecules involved in E . coli adhesion. E . coli exposed to 10 μM TF101 showed a 2.6 fold reduction in adhesion, assessed by CFU compared to samples without TF101, while exposure to AI-2 increased adhesion 3.5 fold. Interestingly, TF101 attenuated the adhesion-enhancing effect of AI-2, suggesting that TF101 interacted with AI-2 signaling ( Fig 4a ). Scanning electron microscopy images confirmed the reduced adhesion of E . coli O103:H2 to Caco-2 cells, when exposed to TF101 compared to control ( Fig 4b ). The concentration of TF101 used did not affect planktonic growth in RPMI medium (data not shown). In addition, there was no difference in bacterial adhesion to Caco-2 cells when the cells were exposed to TF101 prior to the adhesion assay ( Fig 4c ). To investigate whether difference in adhesion could be attributed to a cytotoxic effect of TF101, cytotoxicity against Caco-2 cells was assessed by the LDH release assay. There was no difference in LDH release from Caco-2 cells exposed to TF101 at concentrations from 10 μM and lower, indicating no cytotoxic effect against Caco-2 cells at these concentrations ( Fig 4d ).
10.1371/journal.pone.0157334.g004
Fig 4
Adhesion of E . coli O103:H2 to Caco-2 cells.
(a) Mean ± SD number of bacteria (n = 12) attached to Caco-2 cells, relative to control (= 1 reference line). 10 μM TF101, 10 μM AI-2, 10 μM AI-2 + 10 μM TF101, 50 μM epinephrine (EPI), 50 μM epinephrine + 10 μM TF101, 50 μM norepinephrine (NE), or 50 μM norepinephrine + 10 μM TF101 was added during the adhesion assay. *Significantly different from control, P < 0.05. (b) SEM images of adherent E . coli O103:H2 on Caco-2 cells with or without addition of 10 μM TF101. Scalebar, 20 μm. (c) Adhesion of E . coli O103:H2 to Caco-2 cells pre-exposed and non-pre-exposed to TF101. The cells were exposed to 10 μM TF101 before the bacteria were added in order to test any possible effect of TF101 on the cells. The effect of pre-exposure of TF101 is presented relative to non-pre-exposed cells (indicated by the reference line = 1). The data are presented as mean values ± SD (n = 12). (d) Cytotoxic effect of TF101 on Caco-2 cells. Release of lactate from Caco-2 cells exposed to 0 μM, 2.5 μM, 5 μM, 10 μM and 50 μM T101. The data are presented as mean values ± SD (n = 6).
AI-2 has been shown to alter expression of genes encoding fimbria and flagella. We therefore investigated whether TF101 affected expression of the fimH and eae genes, encoding type 1 fimbria and intimin respectively. The fimH and eae genes showed significantly ( P < 0.05) reduced expression upon exposure to 10 μM TF101, while exposing E . coli to AI-2 led to increased expression ( P < 0.05) ( Fig 2b ).
To investigate whether TF101 also attenuated adhesion by other signaling molecules mediating adhesion, exposure to the inter-kingdom signaling molecules epinephrine and norepinephrine was performed. Epinephrine (50 μM), and norepinephrine (50 μM) increased adhesion to Caco-2 cells by 2.1 and 2.2 fold respectively, however TF101 did not attenuate the stimulatory effect on adhesion ( Fig 4a ).
The mean level of bacterial invasion in infected cells was close to nil (data not shown). The effect of TF101, AI-2, epinephrine or norepinephrine on invasion was therefore not investigated.
Effect of AI-2, TF101, epinephrine and norepinephrine on biofilm formation
TF101 reduced biofilm formation by E . coli O103:H2 significantly ( P < 0.05) at 10 μM, while AI-2 increased biofilm formations by 2.3 fold. By adding TF101 and AI-2 simultaneously, TF101 attenuated the enhancing effect of AI-2 on biofilm formation ( Fig 5a ). Epinephrine and norepinephrine increased biofilm formation by 2.5 and 2.7 fold, respectively, while TF101 diminished the stimulatory effect on biofilm formation ( Fig 5a ). Epinephrine and norepinephrine did not affect planktonic bacterial growth ( S2 Fig ). To further investigate the specificity of TF101 interference with AI-2 signaling, the lsr negative strain E . coli ABU83972 was used. Biofilm formation by this strain was unaffected by TF101 at 5 μm and 10 μm. However, 50 μm TF101 significantly enhanced biofilm formation in this lsr negative E . coli strain ( P < 0.05) ( Fig 5b ), while addition of AI-2 did not stimulate biofilm formation.
10.1371/journal.pone.0157334.g005
Fig 5
Biofilm formation.
(a) Relative effect of TF101, AI-2, epinephrine (EPI) or norepinephrine (NE) on biofilm formation by E . coli O103:H2. 10 μM TF101, 10 μM AI-2, 10 μM AI-2 + 10 μM TF101, 50 μM EPI, 50 μM EPI + 10 μM TF101, 50 μM NE, or 50 μM NE + 10 μM TF101 was added during the biofilm assay. The effect of the different chemicals is presented as relative to control value (reference line (= 1)). The data are presented as mean values ± SD (n = 10). *Significantly different from control ( P < 0.05). (b) Relative effect of TF101 or AI-2 on biofilm formation by E. coli ABU 83972 ( lsr - ). In E . coli ABU83972, neither TF101 nor AI-2 affected biofilm formation, except at 50 μM TF101 which gave a significant increase. The effect of TF101 or AI-2 on biofilm formation is presented as relative to control (= 1 reference line). The data are presented as mean values ± SD (n = 10). *Significantly different from control ( P < 0.05).
Discussion
In this study, we investigated how thiophenone TF101 interfered with AI-2 quorum sensing mediated regulation of virulence in E . coli O103:H2 (EPEC). We furthermore explored how the host-derived hormones, epinephrine and norepinephrine, affected adhesion and biofilm formation alone and in combination with thiophenone.
The exact mechanism of action of TF101 in E . coli is still not fully elucidated, but the present results give support to our hypothesis that TF101 interacts with the AI-2 controlled lsr operon and possibly competes for the LsrB receptor. The effect of TF101 on lsrB expression was studied to determine whether TF101 interferes with AI-2 signaling. The expression of lsrB was significantly reduced in response to TF101. The importance of AI-2 in the activation of lsr genes was confirmed by the upregulation of lsrB in presence of AI-2. Reduced expression of lsrB could consequently lead to a disruption in AI-2 internalization, resulting in down-regulation of AI-2 regulated genes involved in virulence. By exposing the bacteria to TF101 and AI-2 simultaneously, we showed that the upregulation of lsrB in response to AI-2 was diminished.
The in silico AI-2 and TF101 docking results predicted binding of TF101 in the AI-2 binding pocket of the LsrB receptor, suggesting that TF101 may act as a competitive antagonist for the AI-2 receptor in E . coli . AI-2 activates transcription of the lsr operon after phosphorylated AI-2 interacts with LsrR, which then relieves its repression of the lsr operon. The present results suggest that TF101 could act by distrupting AI-2 activity, and consequently inactivate the lsr operon and reduce AI-2 internalisation.
TF101 has previously been shown to reduce expression of flhD [ 35 ], the master regulator of flagella synthesis, which is regulated by AI-2 [ 29 , 41 ]. TF101 has also been shown to decrease the DNA- binding activity of the master regulator LuxR in V . harveyi [ 34 ]. To our knowledge, our study is the first to show that TF101 interfered with expression of a gene that is directly regulated by AI-2 signalling. This further supports the hypothesis that TF101 may function through interference with AI-2 mediated gene regulation. Consistent with this are the previous reports showing that TF101 interfered with AI-2 induced bioluminescence in the marine pathogen Vibrio harveyi , and that TF101 does not interfere with AI-2 synthesis [ 34 , 35 ]. Further to test whether TF101 interferes with AI-2 signaling, we exposed E . coli simultaneously to TF101 and AI-2 and assessed subsequent adhesion to Caco-2 cells and biofilm formation. We found that TF101 attenuated the enhancing effect of AI-2 on adhesion and biofilm formation, thus supporting the assumption that TF101 interfered with AI-2 signaling and its ability to activate genes involved in adhesion and biofilm formation. The in silico analysis predicted that TF101 migth bind to the LsrB receptor, and thus inhibit binding of AI-2. We thus propose that TF101 could function as a competitive antagonist, preventing AI-2 internalization and lsr operon activation.
Thiophenone TF101 reduces biofilm formation and motility in E . coli O103:H2 [ 35 ], possibly by interacting with AI-2 and lsr activity. This hypothesis was further explored by assessing the effect of TF101 on biofilm formation in E . coli ABU83972, a recognized good biofilm former [ 42 ] lacking the lsr operon [ 43 ]. While E . coli ABU83972 did not respond to TF101 at concentration expected to reduce biofilm formation, our results also showed that AI-2 did not stimulate biofilm formation in E . coli ABU 83972. Interestingly, 50 μM, a concentration that normally is toxic to bacteria, resulted in increased biofilm. With the lack of lsr genes, E . coli ABU 83972 appears to regulate biofilm formation in a non AI-2 dependent manner. Notably, E . coli O103:H2 and ABU83972 required different experimental conditions to form biofilm. Nevertheless, using two E . coli strains with different lsr status to study the effect of TF101, could give some clues of the mechanism of action of TF101 and its possible interference with AI-2 signaling in E . coli . As the results suggest that the lsr genes could be involved in the mechanism of action of TF101 in E . coli O103:H2, LsrB could be an attractive drug target.
Several studies have characterized different adhesion patterns of EPEC to epithelial cells [ 44 , 45 ]. According to the criteria stated in these reports we identify the adherence patterns of E . coli O103:H2 in this study, as IS (isolated bacteria) pattern; few isolated individual bacteria over the cells. ( Fig 4b ) [ 44 ]. The role of AI-2 mediated signaling in adhesion of E . coli to epithelial cells, has previously been explored only briefly. Bansal et al showed that AI-2 increased adhesion of E . coli O103:H2 EPEC strain to HeLa cells at concentrations of 100 and 500 μM [ 46 ]. However, in our study we show that an AI-2 concentration of 10 μM significantly increased adhesion to Caco-2 cells. Even though the cell line and media used in the two studies were different, they both indicate that AI-2 may play an important role in regulating adhesion to eukaryotic cells. We furthermore showed that the quorum sensing inhibitor (QSI) thiophenone TF101 significantly reduced the adhesion. To our knowledge, this is the first study to show the effect of a QSI on E . coli adhesion to epithelial cells. The transcriptional analysis showed decreased expression of fimH and eae , encoding the adhesion factors Type 1 fimbria and intimin respectively, in response to TF101, and increased expression in response to AI-2. Type 1 fimbria is a common adhesion factor found in both commensal- and pathogenic E . coli . The most important adhesion factor in EPEC is intimin, an outer membrane protein, encoded by the eae gene. Eae is found within the pathogenicity island LEE (locus of enterocyte effacement). Intimin is responsible for early bacterial adhesion to eukaryotic cells. Tir (translocated intimin receptor) acts as the receptor for intimin, and is translocated into the eukaryotic plasma membrane via Type III secretion system (T3SS). The LEE- encoded Type III secretion systems are key virulence factors of Gram negative enteric pathogens, and serve to inject bacterial proteins directly into host cells. Altogether, these LEE-encoded factors contribute to the characteristic attaching/effacing lesions (A/E lesions) in EPEC [ 47 ].
Our results indicate that decreased adhesion to Caco-2 cells following treatment with TF101 could be explained by a reduction in the production of adhesion factors such as intimin and Type 1 fimbriae. The increase in gene expression of the adhesion factors after stimulation with AI-2 also indicates that AI-2 is involved in the regulation of these virulence factors. Our results are thus consistent with other studies reporting that several genes involved in flagellar and fimbria biosynthesis are upregulated in response to AI-2 [ 46 , 48 ].
The cytotoxicity assay showed that TF101 at concentrations used in this study did not exert cytotoxic effects on Caco-2 cells. Our findings are in line with results from a previous study, showing that thiophenone TF101 at these concentrations did not affect human fibroblasts [ 33 ]. We furthermore tested whether the effect of TF101 on adhesion of E . coli to Caco-2 cells could be explained by altered surface properties of the Caco-2 cells. We exposed the cells to TF101 prior to the adhesion assay. There was no difference in adhesive capacity of E . coli O103:H2 EPEC between samples with pre-treated- versus non-pre-treated Caco-2 cells. Hence, the reduced adhesion related to TF101 could not be explained neither by cytotoxic effects of TF101 nor by alterations of the surface of the Caco-2 cells.
The mechanism of action of TF101 is still not completely revealed. In order to study whether TF101 interfered with other signaling pathways involved in regulation of virulence, we chose epinephrine and norepinephrine due to their association with enhanced bacterial growth, biofilm formation and adhesion [ 20 , 23 ]. This study confirmed that E . coli O103:H2 is able to sense epinephrine and norepinephrine, which could be the first step in a sequence of events leading to infection. Several studies have suggested that epinephrine and norepinephrine act as signaling molecules between the host and the bacteria [ 14 , 49 ]. Our results showed that epinephrine and norepinephrine increased adhesion to epithelial cells, and increased biofilm formation by E . coli O103:H2. This is in agreement with several other studies on the regulation of eukaryotic stress hormones in E . coli virulence and infection [ 16 , 23 , 50 ]. We added TF101 simultaneously with epinephrine or norepinephrine in order to test whether TF101 interfered with their role as signal molecules. From our adhesion assay we showed that TF101 did not interfere with host-bacteria interaction by interfering with epinephrine/norepinephrine. On the other hand, in the biofilm assay we observed that TF101 attenuated the biofilm-enhancing effect of epinephrine and norepinephrine. Biofilm formation and adhesion are two virulence factors regulated by different mechanisms in E . coli . It is still unclear how epinephrine and norepinephrine stimulate adhesion and biofilm formation. We could argue that the mechanisms regulating adhesion and biofilm are different, and TF101 might interfere with these mechanisms differently. None of these results give a strong indication that TF101 interfere with epinephrine/norepinephrine signaling, however it does suggest TF101 as an effective biofilm inhibitor able to attenuate the stimulating effect of epinephrine/norepinephrine and AI-2 signaling. Conversely, from our results, we cannot exclude the possibility that the effect of TF101 may not be specific to AI-2 signaling, and that other unknown mechanisms can be involved. This might be true especially for biofilm formation, but also for adhesion to epithelial cells.
Epinephrine and norepinephrine did not increase the growth rate of E . coli , a result that is in contrast to prior reports [ 22 , 51 ]. However, similar growth rates irrespective of epinephrine or norepinephrine addition emphasize that the increase in adhesion and biofilm formation cannot be explained by increased cell density due to increased bacterial growth. These results may be important in order to reveal the regulation of adhesion and colonization of E . coli O103:H2.
One of the challenges with using some bactericidal compounds for treating bacterial infections is lysis of the bacteria, and the concomitant release of toxins and pro-inflammatory mediators, which may lead to tissue destruction and treatment failure. This highlights the need for drugs that are effective without lysis of the bacteria. Thiophenone might be one such drug representing a non-bactericidal anti-virulence agent, hence; endotoxins and other products will not be released. Another important benefit of using non-bactericidal anti-virulence compounds like TF101 is that it does not exert a strong selective pressure for the development of resistance.
The pathogenicity of E . coli is a complex series of events including both bacterial quorum sensing molecules and a cross talk communication with the host. The present study show that TF101 interferes with E . coli O103:H2 virulence possibly by interfering with quorum sensing, however, future studies with additional pathotypes and other bacterial species are warranted. We propose that thiophenones represent promising anti-virulence agents in the fight against pathogenic bacteria.
Supporting Information
S1 Fig
Pairwise alignment of the amino acid sequence of lsrB from Salmonella enterica subsp.serovar enterica typhimurium and E . coli O103:H2 str.12009.
The yellow marks represent the amino acids in which AI-2 bind to in the binding pocket of the LsrB receptor.
(TIF)
S2 Fig
The effect of epinephrine and norepinephrine on planktonic growth.
No significant effect on planktonic growth was observed in response to epinephrine or norepinephrine.
(TIF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Structural revolution in the GPCR superfamily
G protein-coupled receptors (GPCRs) comprise the largest superfamily of integral membrane proteins, covering ∼3% of the human proteome. They mediate transmembrane (TM) signal transduction by allosterically facilitating information transfer across the cellular membrane in response to extracellular signals [ 1 , 2 ]. The GPCRs are pleiotropic proteins responsible for sensing a diverse set of extracellular signals ranging from photons and small molecules (neurotransmitters, metabolites, odorants, tastants) to large oligopeptides (chemokines, incretins), and converting them into one or more intracellular signaling cascades. This critical role of GPCRs in cellular signaling makes them therapeutic targets in a large number of diseases, either due to their direct role in the pathophysiology of a specific disease or due to their ability to modulate a set of signaling cascades implicated in a disease [ 3 ]. Currently, about 30–50% of all drugs and 20% of recently FDA approved drugs act through modulating GPCR functions [ 4 ].
Experimental structures are now available for more than 25 different GPCRs covering all four major GPCR phylogenetic classes. Several GPCRs have been crystallized in complex with ligands, and some have been crystallized in active conformations, capable of coupling to G proteins or arrestins. Furthermore, efforts to crystalize most human GPCR proteins are underway [ 5 ]. The human β 2 adrenergic receptor was crystallized in an active conformation in complex with the full heterotrimeric Gs protein, providing a snapshot of the conformational changes in both the receptor and the cognate G protein during GPCR activation [ 6 , 7 ]. Rhodopsin has been recently crystallized with arrestin [ 8 ] providing the first detailed snapshot of the receptor before and during internalization. These new structures inspire a full spectrum of mechanistic studies into the GPCR biology, and will guide the functional understanding and pharmacological targeting of these receptors [ 9 – 11 ].
The cellular membrane partitions a GPCR protein into 3 domains: extracellular, transmembrane, and intracellular [ 12 , 13 ]. The N-terminus and three extracellular loops (EC1, EC2, EC3) lie outside of the cell; seven transmembrane (TM) helices (TM1-7) span the membrane; and inside the cell, there are three intracellular loops (IC1, IC2, IC3) together with the C-terminus, which typically contains a shorter helix 8 resting parallel to the membrane. The N-terminus, intracellular loops, extracellular loops, and C-terminus can have very different lengths, not just across the major GPCR classes but also within a GPCR class [ 14 ]. The loop regions are flexible and display different conformations among the known GPCR crystal structures. However, the packing of the TM helices is remarkably well conserved ( Fig 1 ) even for proteins with very small sequence similarity (down to 20%, S2 Fig ).
10.1371/journal.pcbi.1004805.g001
Fig 1
TM domains of the available crystal structures.
Top: Two views of the 24 inactive crystal structures from classes A, B, C, and F (aligned to β 2 ) show the general GPCR fold of the transmembrane (TM) bundle. Class A in green, class B in blue (CRF1, GLR), class C in orange (MGLU1, MGLU5), class F in magenta (SMO). Bottom: Same views for only the 19 inactive class A structures showing the highly conserved class A TM fold. A detailed view of the conserved hydrogen bonding networks is shown in S1 Fig .
For class A GPCRs this is not a surprise, because several amino acids forming inter-helical hydrogen bonds are highly conserved. Indeed, the conserved residues directly correspond to each other in the structural alignment, as can be seen on S1 Fig . Within class A, the sequence similarity between the corresponding transmembrane regions for the receptors with experimentally determined structures is 35–100% ( S2 Fig ), which is high enough that a sequence alignment algorithm (such as Clustal Omega) typically aligns the transmembrane regions correctly. However, the sequence similarity between class A and other classes is so low, typically 20–30%, that sequence alignment algorithms generally fail.
At the time of this study, there were 5 non-class A GPCR structures available. Fig 1 shows that when these structures are aligned to class A using the GRoSS alignment (introduced below in the methods section), their transmembrane helices are in the same fold as that of class A GPCRs . A certain degree of similarity was expected for the intracellular face of the receptor, where they couple to the same set of G proteins and arrestins, but the similarity seems to be present in the full TM bundle. With this comparison of the structures of GPCRs from different classes, we are now able to examine the features that survived the evolutionary divergence in sequence potentially on the basis of their functional importance, enabling us to gain insights into the general features of the GPCR structural fold.
Transmembrane domains as a basis for modeling conformational diversity
Computational protein folding works best for small water soluble proteins and it has proved effective for up to about 100 residues [ 15 ]. The energy models used for soluble proteins have been extended to include approximations of the membrane environment [ 16 , 17 ]. However, the accuracy of these methods is not sufficient for drug design, partly due to the size of many membrane proteins. GPCRs are large proteins with more than 300 residues, making them difficult to model with standard protein modeling methods. Recent assessments [ 18 – 20 ] of the GPCR structure prediction community showed that approaches based on homology models remain the most successful [ 21 – 23 ].
Homology modeling requires accurate sequence alignment to existing structures, which fails completely for alignment between classes. Within each class, sequence alignments typically conserve key motifs, but they often lead to gaps in the transmembrane helices, which indicate serious problems. Recently, Isberg et al. [ 24 ] studied in detail the available crystal structures and obtained pairwise sequence alignments from the highly conserved 7 helix fold by aligning individual TM domains of two structures at a time. Their alignment leads to many single amino acid gaps, which correct for non-matching bends and turns of the TM helices that are often caused by mismatched Pro or Gly. Nevertheless, it is not possible to predict these gap positions with sequence information alone, without the knowledge of the correct structure. Therefore it is not possible to extend their alignment to other GPCRs. Our goal was to construct a sequence alignment of the TM domains of all human GPCRs , that would be suitable for 3D structure prediction. Thus we developed a method that avoids the gaps in the TM regions.
We can consider the common GPCR structural fold to be formed by alpha-helices, allowing us to think about the structural relationships in terms of rigid helices. Past studies [ 25 , 26 ] used the rigid transmembrane helices to compare the different GPCR receptors. The comparative analysis of available GPCR structures informs structure prediction methods (such as in [ 27 ]), where 3D rotations and tilts of the rigid helices are used to sample conformations of the protein in the membrane. Moving the whole helix at once can overcome large potential barriers and sample possible inactive and active states. Better definitions of the GPCR fold and its activation mechanism can help to reduce the number of coordinates needed for modeling.
Integrating sequence, structure, function
Several studies have analyzed inter-helical contacts [ 28 ], ionizable networks [ 29 ], or developed a phylogenetic analysis of class A GPCRs [ 30 ], but no one has yet reported a GPCR superfamily-wide sequence alignment that avoids gaps within the TM domains. The available GPCR structures in the PDB cover a wide sequence space, with some structures showing different ligands bound to the same receptor or the same ligand bound to different receptors, and some structures showing different conformations for the same receptor. The fact that the seven helix fold is conserved across divergent sequences can teach us about the nature of the GPCR fold and its features.
In this work, we first generate a structure-based alignment of the TM regions of the available 25 structures by maximizing the number of the corresponding inter-helical contacts. We use chemically nonspecific contacts instead of conserved residues, since, even though the amino acids are not conserved across this diverse superfamily, many inter-helical contacts are conserved across all GPCR classes as will be shown below. Our structural alignment is extended to include all human GPCRs by sequence alignments of small subgroups of similar receptors. The final GPCR sequence-structure alignment (GRoSS) includes the TM regions of all 817 human GPCRs. The GRoSS alignment is then used to generate a phylogenetic tree for all human GPCRs, which correctly distinguishes the different receptor classes, suggesting that the functional information about different GPCR classes, that are usually characterized by variable length N/C-termini and loop regions, is in fact encoded at least in part within the GPCR TM domains.
This observation that the TM domains and their relationships capture the similarities between the rather disparate set of GPCRs, has broad implications in terms of simplifying and enhancing the connection between receptor sequence and function. We identified 23 Conserved inter-HelIcal COntacts (CHICOs) that define the general GPCR structural fold. Furthermore, our structural comparisons of inactive and active conformations lead to the identification of 15 Native ACtivation “Hot-spOt” residues (NACHOs). We expect that mutations of the structurally important CHICO and NACHO residues will dramatically affect receptor function, and may be responsible for many diseases. Cross-checking these residues with available GPCR mutation databases (Uniprot [ 31 ], NAVA [ 32 ], TinyGRAP [ 33 ]), which contain disease/function association, has allowed us to identify several deleterious receptor mutations that are found at or adjacent to the “hot-spot” residues as shown later in the results section. Using the GRoSS alignment, we have mapped these important positions to all GPCRs leading us to suggest new testable hypotheses about molecular mechanisms behind natural or man-made mutations. We expect that this will create new rational drug discovery opportunities for efficacious therapeutics that minimize side-effects.
Methods
In this section we describe our procedure for sequence alignment of the TM domains of all 817 human GPCR proteins. The resulting alignment is available in S2 Table , and in fasta format in S1 File .
Human GPCR classification
After the Human Genome Project was completed, Fredriksson et al. [ 34 ] performed detailed phylogenetic analysis of GPCRs developing the GRAFS classification system [ 34 , 35 ], which identified 5 main groups of receptors: Glutamate, Rhodopsin, Adhesion, Frizzled/Taste2, and Secretin. The Rhodopsin family is the largest and was partitioned into four subgroups: α , β , γ , and δ . Olfactory receptors were attached to the δ branch. The bitter taste receptors (Taste receptors type 2, Taste2, or TAS2Rs) were grouped with the Frizzled family.
Currently, the list of receptors is maintained by the International Union of Basic and Clinical Pharmacology (IUPHAR) [ 36 ], which also keeps track of known endogenous ligands and signaling mechanisms. IUPHAR identifies 6 main classes A-F: Four of them, class A (rhodopsin-like receptors), class B (secretin and adhesion family), class C (glutamate receptors), and class F (frizzled receptors) contain human receptors. Classes D, E are part of the classification that includes other species, but contain no human orthologs. In contrast to GRAFS, secretin and adhesion groups have been merged since their sequences align well. The bitter taste receptors (Taste2, or TAS2Rs) and Vomeronasal receptors (VN1Rs) are beginning to be considered similar to class A, in contrast to the GRAFS classification, which is what we also find in the phylogenetic analysis presented below.
GPCR database
We extended the list of human GPCRs of Fredriksson et al. [ 34 ], with the proteins considered by the GPCR Network [ 5 ] and by IUPHAR [ 36 ]. Furthermore we added all sequences annotated as human GPCR proteins in the Uniprot database [ 31 ]. Altogether we collected 836 candidate sequences.
Most of these proteins have been assigned to a class by previous studies [ 34 , 36 ]. For sequences with unknown class, we first searched for related proteins by running BLAST, and then we aligned them against the candidate class with the program Clustal Omega [ 37 ]. If the sequence aligned without large gaps in the TM regions for any of the classes, we assigned the sequence to this class. The following 11 proteins might be GPCR proteins, but they do not align well to any of the classes, thus we ignore them in the further analysis (listed as Uniprot accession numbers, ACs):
P51810, Q5T9L3, Q5VW38, O60478, Q86V85, Q86W33, Q8N3F9, Q8NBN3, Q96K49, Q96N19, Q9NPR9.
The following 8 sequences are most likely pseudogenes, because they are missing one or more TM domains:
A6NFC9, Q32VQ0, Q8NGA4, Q8NGU1, Q8NGY7, Q8TDU5, Q96P88, Q99463.
We kept Q9P1P4 (TAAR3) which is a pseudogene in humans but functional in rodents, and we kept Q49SQ1 (GPR33), which may be functional in some people. A curious case is the protein GPR157 (Uniprot AC Q5UAW9), which is most similar to class B, but its TM1 has a gap in the alignment to class B. However, the TM1 aligns well to the class A TM1, so this protein appears to be a hybrid between these two classes.
Table 1 summarizes the count of all the sequences that we kept and S2 Table lists their Uniprot ACs. In total there are 817 candidate human GPCRs, 399 of which are non-olfactory. If the sequences in class A were present in the analysis of Fredriksson et al. [ 34 ], then we kept the subgroup labels α , β , γ , δ , otherwise we labeled them as A-other .
10.1371/journal.pcbi.1004805.t001
Table 1
Number of GPCR sequences by class.
The total number of candidate human GPCR sequences that were considered are listed. The full list of Uniprot ACs is in S2 Table .
88
A α
16
B
5
Vomeronasal
33
A β
22
C
25
Taste2
57
A γ
11
F
11
Other
58
A δ
33
Adhesion
8
Pseudogene
51
A-other
418
Olfactory
Available crystal structures
At the time of this study, crystal structures of 25 GPCRs were available (19 of these for human sequences): 4 in both active and inactive conformation [ 6 , 38 – 44 ], 1 in active only [ 45 ], and 20 in inactive conformation only [ 46 – 66 ]. When multiple crystal structures of the same protein were available, we used the one with the best resolution or the one with the best-defined transmembrane alpha helices. The PDB IDs of the structures used in analysis are listed in S1 Table .
Well-defined TM helices are a prerequisite for our analysis, however, different criteria have been used for annotating helices in different PDB files. Many of the transmembrane helices contain bends, and sometimes the helix termination is not well defined. We define the extent of each transmembrane helix as the residues positioned in the membrane (as placed by the Orientations of Proteins in Membranes (OPM) database [ 67 ]) extended until the end of the alpha helix by the DSSP secondary structure determination [ 68 ]. The helices were manually inspected and only a few manual corrections were needed. The final TM lengths used are displayed in S2 Table .
Alignment by minimizing RMSD is not unique
As suggested by Fig 1 , the transmembrane helical bundles of all the GPCR structures display the same structural fold (same relative position of all seven helices). The main basis for some published structural comparisons of classes A and B, C, F was an iterative structural alignment implemented by the program ICM-Pro [ 69 ] used for GLR, MGLU1, and SMO. Also, MGLU5 was aligned to class A with an iterative algorithm (SSM algorithm [ 70 ]) and CRF1 was aligned to class A manually. The iterative structural alignment algorithm, removes from the alignment all atoms that are too far in the previous rounds. This works relatively well, but the exact sequence pairing is not uniquely defined and depends on cutoff parameters. In many cases it leaves an ambiguity of ±4 residues (1 helical turn).
In order to remove this ambiguity and to determine which alignment would minimize the root mean squared deviation (RMSD) of the full TM bundle, we start from an approximate structural alignment, and try all nearby sequence alignments (±1 helical turn on each helix). For each sequence alignment, we first select the maximal overlapping lengths of the 7 TMs. Then we compute the C α RMSD by evaluating the least-squares superposition of the corresponding C α atoms. But minimizing only RMSD does not necessarily lead to an optimal alignment. For several cases in class A, we find TM alignments that have a lower RMSD than the alignment that conserves the correspondence of the same Ballesteros-Weinstein (BW) residue positions [ 12 ] (some BW residue positions in each TM domain are expected to be conserved at least within a GPCR class and hence align well). The reason for this is that the extracellular ends of the helices sometimes have significantly different tilts, making the most tilted helix dominate the RMSD measure. To avoid these issues with the RMSD measure, we instead look for an alignment, which maximizes the number of conserved inter-helical contacts. A contact is defined to be conserved if the residue pair in the contact has the same BW numbering in the two structures being compared. For example, if structure X has an inter-helical contact between residues 2.45–3.42 and structure Y also has an inter-helical contact between residues 2.45–3.42, then this contact is defined as conserved in the two structures. The BW numbering of residues in structure X and structure Y implicitly assumes that the sequence for structure X and the sequence for structure Y are aligned to superimpose residues with the same BW numbering. This measure accommodates the scenario commonly seen in GPCRs and described below, where residues may not be conserved but the corresponding structural contacts have been conserved during evolution.
Class A conserved contacts
In class A, the analysis of inter-helical interactions typically focuses on hydrogen bonds, since many of the hydrogen bonding residues are highly conserved. In order to compare receptors from different classes, which have poor sequence conservation, we compare inter-helical contacts ignoring their chemical nature. We use the definition of an inter-helical contact as in [ 28 ]: any two heavy atoms from different TMs that are closer than the sum of their van der Waals radii plus 0.6 Å. The inter-helical contacts, which are present in almost all class A structures, are shown in Fig 2 . This list is very similar to the contacts found by [ 28 ], but there are minor differences caused by using a different set of crystal structures.
10.1371/journal.pcbi.1004805.g002
Fig 2
Conserved inter-helical contacts.
Top left: Diagram of 40 conserved inter-helical contacts (CHICOs) present in at least 23 out of 24 studied class A structures. The contacts common to all classes are shown in purple, and contacts present only in class A in orange. Top right: List of these contacts in Ballesteros-Weinstein numbering scheme. Bottom: Extracellular view of the same contacts in the β 2 crystal structure. The contacts in the inner and outer half of the membrane are shows on the left and right respectively.
We briefly compare this list of chemically unspecific contacts to the conserved hydrogen bonds. Within class A, the focus is often on two conserved networks of hydrogen bonds (shown in S1 Fig ):
4-2-3 : W4.50 ↔ S/N/T2.45 ↔ S/N/T3.42
1-2-7 : N1.50 ↔ D2.50 ↔ N7.49.
Fig 2 shows that the network 4-2-3 is well conserved across classes. In particular, the contact 2.45–3.42 is present in all classes, and there are many conserved contacts in its immediate vicinity, such as 2.42–3.45, 2.42–3.46, and 2.46–3.42. The contact between TM 3 and 4 is also well conserved, as the highly conserved bulky residue W4.50 leans on A3.38.
The network 1-2-7 also has many conserved contacts, including N1.50-D2.50, but interactions with N7.49 are not conserved. Even in class A, N7.49 interacts with the other residues of this hydrogen bonding network indirectly through a water molecule. Still, the side chain packing in the regions where the helices are close together is important for structural stability. Conserved water-mediated interactions could be included in future analysis, after enough crystal structures resolve them. For class A it is clear that using BW numbering greatly facilitates sequence and structure comparisons. Next we will see how these ideas can be extended to other classes.
Sequence alignment across classes based on common contacts
To compare other GPCR classes with class A, we first find the inter-helical contacts of all the current crystal structures. Each structure has about 200 contacts. For any pair of structures and a given sequence alignment, we compute the number of conserved inter-helical contacts. We consider all possible TM sequence alignments that have no gaps (±10 residues from the starting sequence alignment). For class A, the number of common contacts between any two structures is maximized by the alignment that preserves the BW numbering.
To translate the BW numbering into classes B, C, and F, we start with a sequence alignment corresponding to an approximate initial structural alignment to class A. For each non-class-A structure, we then try all possible adjustments to BW n .50 residues (±10 residues, again with no gaps) on each helix and count the number of common contacts with each of the 24 class A structures. Table 2 shows the alignments with the highest cumulative number of the common inter-helical contacts.
10.1371/journal.pcbi.1004805.t002
Table 2
Selection of the alignment between class A and classes B, C, and F.
This table shows the selection process for assigning BW.50 residues to non class A proteins. Shifting BW.50 residue on each helix renumbers the relative BW numbers, effectively changing the labels of contacts observed in these proteins. Subsequently, the number of common contacts each structure shares with the class A structures changes for different BW residue assignments. The second rightmost column shows the cumulative number of contact occurrences among the 24 class A structures (including active conformations). The BW assignment with the highest number of contacts is selected (except for MGLU5, see text). The selected alignment is in bold.
Class
Protein
BW#
Common contacts
RMSD Å
1.50
2.50
3.50
4.50
5.50
6.50
7.50
B
CRF1
L134
F162
L213
W236
V279
L329
S360
2212
3.11
L134
F162
L213
W236
V279
I325
A363
2084
3.61
L134
F162
L213
W236
V279
I325
S360
2081
3.35
GLR
L156
F184
L249
W272
A314
V364
A397
1972
2.93
L156
F184
L249
G273
A314
V364
A397
1913
2.94
L156
F184
L249
M276
A314
V364
A397
1876
3.11
C
MGLU1
T607
I638
I682
I714
L763
A800
L827
2017
3.02
T607
I638
I682
S711
L763
A800
L827
2012
3.24
T607
I638
I682
S711
C767
A800
L827
1873
3.41
MGLU5
T594
I625
A669
F698
L750
A787
L814
1974
3.36
T594
I625
A669
I701
L750
A787
L814
1954
3.17
T594
I625
A669
F698
L750
I784
L814
1820
3.64
F
SMO
T245
F274
W339
W365
V411
I465
S533
2358
3.00
T245
F274
W339
W365
V411
C469
G529
2311
2.98
T245
F274
W339
W365
V411
C469
S533
2248
3.02
…
T245
F274
W339
W365
V411
S468
I530
1827
3.04
In class A, the BW n .50 ( n = 1 to 7 denotes the TM) residues correspond to the most conserved residues in each TM. After the projection of the BW numbering to the other classes, the n .50 residues are not necessarily the most conserved within each class. Moreover, they correspond neither to class B specific Wooten numbering [ 71 ], nor the class C specific Pin numbering [ 72 ], nor the class F specific numbering [ 65 ], which only define labels n .50 for residues that are the most conserved within the respective classes. In the context of Isberg et al. [ 24 ], our n .50 can be referred to as n .50 a and uses class A residues as a reference to number TM residues for GPCRs from all classes. This unified numbering scheme enables a view of structurally conserved or structurally similar positioned residues across all GPCR classes because conserved residues in individual GPCR classes have no structural similarity or positioning in the GPCR structural fold in cross-class comparisons.
In terms of absolute sequence alignment of TM domains, our Class B (CRF1, GLR) alignment agrees with the alignment suggested in [ 62 ], which was obtained by an iterative structural alignment. Similarly, our alignment for class C (MGLU1, MGLU5), agrees with the suggested alignment in [ 63 ]. For MGLU5, we chose the second highest scoring alignment to make the alignment consistent with the MGLU1. This choice was checked visually and the corresponding residues are in a more similar position in our selected alignment.
Our alignment for the SMO receptors is the only one that differs from the published alignment to class A presented with the crystal structure [ 65 ]. This published alignment corresponds to the last row of Table 2 , and it differs from our best alignment (fourth to last row of Table 2 ) in TM6 and TM7 by 3 residues each. The RMSD difference between this published alignment and our alignment is small, however, our alignment results in ∼500 more common structural contacts than the former alignment. Helix 7 of the SMO receptor does not have a proline residue, and so it is missing the kink that is typical for class A GPCRs. There are many inter-helical contacts in the extracellular part of the TM7, so that our chosen alignment gives a good spatial correspondence for the larger part of helix 7. We have presented a well-defined protocol for structural-sequence alignment. As more experimental structures become available, this protocol can refine the alignments where needed, which might occur especially for classes with a small number of known crystal structures.
This sequence alignment was used to generate the structural alignment shown in Fig 1 . The sequence similarities between the TM domains of the crystal structures are shown in S2 Fig , and the TM domain RMSDs are shown in S3 Fig . Sequence similarities correlate reasonably well with the RMSDs (structural differences). Note that we define percent similarity as the fraction of similar residues, where two residues are similar if their BLOSUM62 [ 73 ] (or GPCRtm [ 74 ]) matrix entry is positive.
GRoSS: Extension of alignment to all known GPCR sequences
We extend the structure-based alignment derived above to all human GPCRs by anchoring each subfamily to the correct crystal structure. As a guide for the quality of the sequence alignment, we check for the presence of any gaps in the transmembrane regions. The approximate positions of the TM regions are already annotated in the Uniprot database as predicted by the TMHMM program [ 75 ]. These predictions are quite noisy, and even for similar proteins that align well, they can differ by 5–8 residues and sometimes even misclassify a TM. However, for multiple sequences the overall trend clearly identifies the approximate TM location and allows us to judge the quality of the alignment of multiple sequences. If there are gaps in the TM regions, the alignment cannot be used to successfully create homology models.
First, we try to align directly all 817 sequences of the GPCR superfamily using a multiple sequence alignment program Clustal Omega [ 37 ]. However, the overall sequence conservation is very low, and the resulting alignment has many large gaps even within TM domains. Some highly conserved residues end up aligned incorrectly. In order to avoid this problem, we aligned class A sequences separately (705, including olfactory). Again the resulting alignment has large gaps even in the TM region. It seems that the large variability of the loop region is what confuses the alignment algorithm.
Fortunately, we find that sequences in individual subgroups can be aligned using Clustal Omega without large gaps in the TM regions. We take these individual subgroup alignments and fix them into a profile —a multiple sequence file for which aligned columns are kept fixed, and from which the hidden Markov model (HMM) is computed. We then align any two profile HMMs to see how similar are the two groups. A profile alignment of A α to each of the other class A groups (A β , A γ , A δ ) has no gaps in the TM regions, and also gives the correct alignment of the BW.50 residues.
The multiple sequence alignment of the group A-other showed gaps in the TM regions for several proteins (Uniprot ACs: Q96P67, Q8TDU6, Q16570, Q86SM8, Q9NS66, Q9NS67, P60893, Q86SM5), so a separate profile was created for these sequences. After the split, both profiles of the A-other proteins aligned separately to the A α profile without any gaps in the TM region, which anchors them to the class A alignment. Similarly, the profile of olfactory receptors (both tetrapod and fish-like) aligned to the A α profile without any gaps in the TM region, which anchors the olfactory profile to the class A alignment. The Vomeronasal and Taste2 groups were more problematic, and are discussed in the following section.
The profile of the adhesion class aligns well to secretin class (original class B). Class B is aligned to class A using the structural analysis described above. Aligning profiles of A α and B does not yield meaningful alignment, because the TM regions are offset and there are many gaps in the TM regions. Similarly, aligning classes A and C or A and F does not yield meaningful alignment, and again structural alignment was used for these cases.
Once the alignment is fixed, the TM lengths for new proteins can be predicted to be the average TM lengths from the available structures in the same class. For example, for sweet taste receptors (TAS1) the predicted TM length is the average TM length of GMR1 and GMR5; and for bitter taste receptors the average TM lengths of the 20 class A structures. These are meant to be the best initial guesses.
The complete listing of all the 817 human GPCR proteins is shown in S2 Table . The alignment of each helix is determined by the provided BW.50 residue. Expected TM range is also provided and it is estimated as the average TM region of the known crystal structures from the same class. For easy viewing, we provide the same alignment also in the fasta format together with the annotations of TM range and BW residues in Jalview [ 76 ] format in S1 File . The GRoSS alignment was also compared to alignments obtained by two other methods: HMM-HMM [ 77 ] and GPCRDB [ 24 , 78 ]. This comparison is described in S1 Text .
Bitter taste and vomeronasal receptors
Bitter taste (Taste2, TAS2R) and vomeronasal receptors are small groups of receptors that do not easily align to the profiles for classes A-F, and so their classification has not been unique. While IUPHAR assigns the bitter taste receptors into the class A, Singh et al. [ 79 ] points out the lack of conserved amino acids between the two.
The profile of the vomeronasal group aligns better with class A α compared to classes B and C, but there is still a gap of length 2 near the center of TM5. We remove the gap in such a way that the residue, which aligns with 5.50 stays fixed. To check that this is indeed the best alignment we explore small changes in the alignment by shifting individual TM by up to ±5 residues. In Fig 3 we see that for TMs 1 to 4, our current alignment gives the highest sequence similarity with A α , so the alignment of these TMs is correct. However, for TM5, the alignment shifted by -1 or +2 residues gives higher similarity with A α . Nevertheless, the similarity with groups A β , A γ , A δ , and B is the highest for our current alignment. We therefore keep the current choice.
10.1371/journal.pcbi.1004805.g003
Fig 3
Testing the robustness of the alignment of the Vomeronasal receptors with the other groups.
The table shows similarity between TMs averaged over all pairs of sequences formed from the two groups (red denotes high similarity, blue low similarity). For most TMs the optimal choices agree with the optimal alignment to A α (full table in S5 Fig ); all combinations are shown only for TM5. The same table but using the GPCRtm substitution matrix [ 74 ] instead of BLOSUM62 is shown in S7 Fig . GPCRtm was developed in particular for GPCR proteins, but in this case both matrices result in the same alignment.
We performed a similar analysis for the Taste2 receptors, for which adjustments were necessary. The profile alignment of Taste2 with A α has some gaps, but it is still the best alignment (i.e., it has the fewest gaps) compared to aligning to classes other than class A. TM3 has two gaps in the alignment: a gap of length 4 in the middle of TM3, and a gap of length 5 at the DRY motive. As the first iteration we kept the alignment fixed on residue 3.50, then we computed the similarity to other groups for ±5 residue shifts. The shift by +3 residues gives better similarity and so it was kept. See Fig 4 for the computed similarities after the shift has been made. All class A subclasses favor this new choice, as the highest similarity has offset 0. Class B would favor shift by 2 residues, but the similarity is less than 30%.
10.1371/journal.pcbi.1004805.g004
Fig 4
Testing the robustness of the alignment of the Taste2 receptors with the other groups.
The table shows similarity between TMs averaged over all pairs of sequences formed from the two groups (red denotes high similarity, blue low similarity). For most TMs the optimal choices agree with the optimal alignment to A α (full table in S6 Fig ) only TM6 shows a second possible alignment at offset +4. The same table but using the GPCRtm substitution matrix instead of BLOSUM62 is in S8 Fig . Again, both matrices result in the same alignment.
TM4 has low sequence similarity, and in particular the highly conserved Trp is not present in Taste2. Again as a starting point we kept the alignment at 4.50, but later had to adjust it by 4 residues. Fig 4 shows the similarity after this shift has been made. For TM4 the similarity is only slightly higher at the new best offset than at nearby offsets.
Taste2 TM6 showed the partially conserved motif IYFLS, with S being aligned to P6.50, which we kept as an initial try. This choice is kept in Fig 4 . However, we see that an offset of +4 residues, which corresponds to a one turn shift (the motif IYFLS aligns Ile with P6.50), also gives high similarity. Based solely on sequence similarity we cannot distinguish which alignment is better, and therefore both cases should be considered when building homology models and energy of the resulting structures should be used as a guide to select the best choice. These alignments will be revisited when the first experimental structure of one of the Taste2 receptors is determined.
Computation of the phylogenetic tree
We compute the similarity for each pair of sequences using the weights from the BLOSUM62 matrix (two residues are considered similar if their BLOSUM62 matrix entry is positive), and use the similarities as a distance metric to cluster the proteins. We used the unweighted pair-group clustering algorithm (implemented in Jalview [ 76 ]), which iteratively extends clusters by finding a non-member sequence with the lowest average dissimilarity over the cluster members. The phylogenetic tree constructed by this clustering algorithm was visualized using the Iterative Tree of Live toolkit [ 80 ].
List of natural variants
S3 Table lists all 2449 GPCR natural variants annotated by Uniprot. According to the GRoSS alignment 1289 of these lie in the TM regions and are listed here with the corresponding BW number. For each mutation we computed its distance to the closest NACHO (or CHICO respectively) residue on the same TM. Zero means this residue is the NACHO (CHICO) residue, in which case we also provide the multiplicity column counting to how many NACHO (CHICO) contacts this residue belongs to. We found 13 (23 including olfactory) mutations of residues on both lists, 48 (99 including olfactory) on the NACHO only list, and 161 (299 including olfactory) on the CHICO only list.
Molecular graphics
3D molecular views have been rendered using PyMOL [ 81 ].
Results and Discussion
Gaps in the alignment of TM regions
We constructed the GRoSS alignment in order to avoid gaps in the TM regions and to simplify preparation of homology models. With the BW residues correctly aligned, and without any gaps in the TM regions, we can use this alignment for direct generation of homology models of the TM helix bundle using essentially any structural template.
A general approach to preparing homology models is to create a new alignment for each target and available template, say using the HMM-HMM method [ 77 ]. However, HMM-HMM often produces false gaps in the TM regions. Recently, a sequence numbering for GPCR crystal structures was presented by Isberg et al. [ 24 ] (available at GPCRDB [ 78 ]) that used a structural alignment to identify gaps or bulges in TM regions, when comparing the same TM between any two crystal structures. The properly placed gaps, often improve structural alignment of helix kinks or loose turns. However, the best structural alignment also resulted in gaps in TM regions that can never be predicted by sequence alignment, such as HMM-HMM alone. Table A of S1 Text shows that the mismatches between the GPCRDB and HMM-HMM are common. The GPCRDB alignment is good for retrospective analysis of known structures, but cannot be used for predictions of unobserved gaps.
To quantify the differences between homology models based on the three difference alignments, we compare the RMSD, TM-score and number of common contacts in Fig A of S1 Text . Overall, these comparisons show that GRoSS performs similarly to GPCRDB for alignments within one class, and better for inter-class alignments. GRoSS performs better than HMM-HMM within a class, and significantly better between different classes.
Loop alignment
We omit loops from the GRoSS alignment, because in the GPCR protein superfamily loops are very diverse, especially the loops EC2 and IC3. EC2 is up to 171 residues long for some class A receptors, but it is shorter than 35 residues in class C, and shorter than 20 residues for all other receptors. IC3 is up to 223 residues long for some class A receptors, but it is shorter than 20 residues for all other receptors. There are likely important similarities among the loops across the GPCR classes. For example, on the intracellular side the receptors have to be sufficiently similar to accommodate G-proteins and arrestins. Furthermore, on the extracellular side, there is a highly conserved disulfide bond between TM3 and loop EC2 that is important for the assembly of the receptor in the membrane [ 82 ]. Thus it is possible that with more experimental GPCR structures, a more systematic understanding of the loop regions will emerge as well.
Sequence alignment from structural alignment
Fig 2 compares class A to the other classes for the alignment constructed by maximizing the number of common inter-helical contacts ( Table 2 ). The purple color in Fig 2 denotes the structural contacts common to all classes, and orange denotes contacts specific to class A. Only one contact, 6.51–7.39, is present in all of class A structures (active and inactive), but it is not in the structures of the other classes. Furthermore, the interactions of TMs 1–5 are more conserved across all classes, but the TM 6 and 7 contacts are more class A specific. It is possible that during the GPCR assembly the helices 1–5 form some intermediate partially folded state before helices 6 and 7 are fully present in the membrane. This might be the reason why the contacts between helices 1–5 are more similar across the classes.
Fig 5 shows the alignment of the TM3 regions for all the known crystal structures (other TMs are shown in Fig 6 ). We see that the DRY motif at positions 3.49–3.51 is highly conserved within the 20 class A sequences, and even when there are mutations only similar amino acids occur: ERY, DRF (however, there exist class A GPCRs without this motif, e.g. PTGDR has ECW [ 83 ]). In classes B, C, and F the DRY motif is not conserved at all.
10.1371/journal.pcbi.1004805.g005
Fig 5
TM 3 sequence alignment for the 25 crystal structures.
Other TMs are shown in Fig 6 . The sequences are taken from the selected PDB files. The TM helix residues are colored in the Zappos scheme, which captures the chemical nature of each residue (e.g. helix breakers, proline and glycine, are shown in purple). The loop residues are shown in grey. The BW n.50 residue (numbering displayed below the sequences) is the most conserved within the class A. The consensus sequence is most similar to class A, because most sequences are from this class. The largest differences are for the last 5 sequences, which belong to the classes B, C, and F. The figure was prepared using Jalview.
10.1371/journal.pcbi.1004805.g006
Fig 6
Sequence alignments for TMs 1,2,4–7 for the 25 crystal structures.
Same caption as Fig 5 , where TM3 is shown.
Proline residues often cause a helix kink and are commonly found in the TM domains of membrane proteins. They are structurally important for deciding which structures should be used as templates for modeling a new protein. In Fig 5 , prolines are highlighted in purple. For example, only MGLU5 has a proline in a central region of TM3, but in this case, the shape of TM3 is very similar to MGLU1, which does not have the corresponding proline.
The consensus sequence for TM3 mostly agrees with class A residues, because most of the crystal structures are from the class A. Interestingly the most conserved residue across all classes is Cys3.25, which forms a disulfide bond to the extracellular loop EC2. This bond is important for the stability of the protein, and shown to be critical for the GPCR assembly [ 82 ].
Fig 6 shows the consensus sequence and alignment of the remaining TM regions (1,2,4,5,6,7) for all experimental structures considered. Known conserved residues in these TMs for class A receptors are easily spotted. For TM1, residue 1.50 is a conserved polar residue for all GPCR classes except class B. For TM4, W4.50 residue is conserved across all classes except class C. For TM5, residue 5.60 is a positively charged residue for classes B, C, F, and most A α receptors. For TM6, residue W6.48 is conserved for all classes except A δ , B, and F. TM7 residue 7.45 is exceptional in being a conserved polar residue across all classes and also appears on our conserved contact residue list (see below).
The conserved inter-helical contacts of class A were the basis for the alignment between the GPCR classes. These contacts show interactions that should be considered first in analysis of the structure or function of these proteins. Fig 2 shows the contacts conserved only within class A, and a similar analysis is shown for classes B, C, and F in S9 Fig . However, since only one or two structures are available in classes B, C, and F, the resulting list is not averaged as it was for class A, and it will be refined as more crystal structures from these classes become available. Thus we cannot yet determine which residues are causing most of the systematic differences between the classes and which residues are critical within each particular class.
Common contacts between different classes define the GPCR structural fold
There are 40 inter-helical contacts common to class A GPCRs as shown in orange in Fig 2 . Out of these, 23 contacts (shown in purple) are present in the crystal structures from classes B, C, and F as well. These 23 conserved inter-helical contacts (CHICOs) formalize our initial insight that the TM bundles of all the different classes are similar and define the GPCR “structural fold”. As more structures become available, this structural fold will be refined.
Examining the inter-helical contacts that make up this GPCR structural fold, we find that TM6 is fully decoupled from both TM3 and TM5, whereas this was not the case for the fold that corresponds to only class A GPCRs (orange contacts in Fig 2 ). GPCRs across different classes couple to the same set of G proteins and arrestins, and now it is known from overwhelming structural and biophysical evidence that G proteins [ 6 ] and arrestins [ 8 ] couple to the GPCRs between TM3/TM6 or TM5/TM6 regions. The GPCRs have evolved to conserve these functionally important couplings with their intracellular signal transduction partners, but have not had the need to conserve contacts of TM6 with TM3 or TM5. This is consistent with the structural fold of GPCRs shown by purple contacts in Fig 2 .
Fig 2 seems to suggest that there are two separate conserved units: TMs 1–5 and TMs 6–7. It is possible that for receptors that have very long loop IC3 (these are only in class A), the TMs 1–5 need to stabilize in the membrane prior to assembly of the last two TMs.
Another important observation is that the specific conserved contact residues defining the GPCR structural fold are not conserved across the different GPCR classes. This tactic of nature to maintain a structural fold without conserving the residues is not uncommon, e.g., the MAT α 2 homeodomain-operator complex in yeast and drosophila has maintained the homeodomain-fold structure to interact with DNA, even though the species are separated by millions of years and have poor sequence homology in this domain [ 84 ].
Phylogenetic tree
The sequence similarities between the TM regions of the crystal structures are shown in S2 Fig (two residues are considered similar if their BLOSUM62 matrix entry is positive). The similarities are higher than 40% for proteins within the class A branches, from 34% to 54% across the class A branches, and 18–36% across the classes. Based on the GRoSS alignment, we computed the similarity for all the human GPCR proteins. The phylogenetic tree in Fig 7 graphically captures sequence similarity between all the proteins, which also indirectly corresponds to their structural differences (we compare these below). Even though evolutionary considerations were ignored when constructing this tree (for phylogenetic analysis see e.g. the Evolutionary Trace method [ 85 ]), this phylogenetic tree clearly contains evolutionary information, but it may miss the information encoded in the loops.
10.1371/journal.pcbi.1004805.g007
Fig 7
The phylogenetic tree based only on TM similarity using the GRoSS alignment (loops were ignored).
Color coding denotes the GPCR class. Proteins with known crystal structure are emphasized with a dot. The full resolution version of this figure is in S4 Fig .
The branches near the root of our tree are very sensible: First class C separates, then class B and adhesion proteins branch off, then class F, and finally class A comprising the rest of the tree. Evolutionarily, this is consistent with the recent most detailed analysis of 83 species [ 86 ], which showed that glutamate receptors (class C) and bacterial cAMP receptors are the oldest (>1400 MYA, million years ago), followed by class B and F (∼1275 MYA), and lastly rhodopsin-like receptors (class A)(∼1100 MYA). Except for several outliers, the first major branches to separate in class A are the sensory receptors: Vomeronasal, Taste2, and Olfactory. In the olfactory branch, the first split separates the fish-like receptors (families 51–56) from the tetrapod-like receptors (families 1–13). The subdivision of the rest of the class A does not follow the α − δ subclasses, but it is close. Near the leaves (i.e., for closely related proteins), the displayed tree might not provide the best classification, since our computation of similarity ignored loops. For related proteins, it may be advantageous to include similarity of the loops as well, since loops often interact with ligands, and therefore can determine receptor specificity. Indeed, it is somewhat surprising that using only the TM domain alignment of all human GPCRs, a phylogenetic tree can be constructed that correctly gets most evolutionary signatures of GPCRs. This suggests that the TM domains of GPCRs contain a good part of the signatures of the divergent evolution of GPCRs, including evolutionary separation of different classes, whose visible differences are usually seen in their soluble domains (N/C-termini and intracellular/extracellular loops).
Conserved inter-helical contacts involved in activation provide functionally important residues
We found that 40 inter-helical contacts are present in at least 23 out of 24 class A crystal structures (CHICOs in Fig 2 ). We infer that these residues are important for the interactions between the helices and that any changes to these residues may cause structural stability issues for the protein. Thus, naturally occurring mutations of the residues involved in the conserved contacts could be direct causes of physiological differences and/or diseases.
Comparing the common contacts among different proteins is not straightforward because many of the sequence differences appear random. Focusing on the difference between active and inactive conformations of the same protein makes the significance of the individual residues much clearer. There are 3 active-inactive crystal structure pairs available: RHO, β 2 AR, and M2 with accepted “fully active” conformations. The active structure of A2A is only partially active, and for NTS1act the inactive structure is not available.
The main signature of activation for rhodopsin is breaking the R3.50↔E6.30 salt-bridge and forming of the K5.66↔E6.30 salt-bridge. Instead of keeping track only of hydrogen bonds, our analysis of contacts allows us to determine more general changes during the activation. The changes in structural contacts upon activation are shown in Fig 8 along with the list of contact residues that change, referred to as native activation “hot-spot” residues (NACHOs).
10.1371/journal.pcbi.1004805.g008
Fig 8
Native activation “hot-spot” residues (NACHOs), which are contacts that change upon receptor activation.
The width of the green lines is proportional to the number of contacts common to all six structures (RHO, β 2 AR, M2, and their active structures). Blue shows the contacts present only in inactive structures, and not in inactive structures; while red shows the opposite. The upper diagrams show contacts in the extracellular half of the membrane. We see that there is no systematic change common to the class A receptors in the conformation of the extracellular half of the TMs. This is not obvious, because there are conformational changes accompanying ligand binding. All the systematic changes, which enable G protein binding, occur in the intracellular half of the TMs. The list only contains 15 different residues in 15 different contacts. Thus many of the residues switch partners upon activation.
An important observation is that the structural contacts in the extracellular half of the receptors do not change upon activation. All structural changes occur in the intracellular region, where the contacts get rewired upon the binding of the G protein. Most of the changes occur for TM 6, since the intracellular end of helix 6 undergoes the largest movement upon activation. However, TM 7 also shows a large number of systematic changes, as it breaks a contact with TM1 and creates new contacts with TM 2 and 3. The residues 3.43 and 3.46 occur in the list of conserved contacts in both active and inactive structures, therefore the conformational changes around these residues seem to be very important for the conformational changes during activation. The class A switching mechanism seems to rely critically on a small number of NACHO residues (15 residues). If any of these residues is mutated, the energy landscape of the active and inactive states might be modified, making the receptor likely to become either constitutively active or inactive, thereby altering or breaking its natural function.
Examples of mutations and natural variants modifying the function
It has been shown experimentally that single amino acid mutations can have a dramatic effect on GPCR activity. For example, the man-made mutation T3.46A makes the receptor CB1 fully inactive, while the mutations T3.46I and L3.43A make it constitutively active [ 87 , 88 ]. Both positions, 3.46 and 3.43, are on the NACHO list of residues critically involved in activation. These particular mutations were introduced by experimentalists, but the NACHOs ( Fig 8 ) are useful for judging the effect of natural variants as well. The positions of many single nucleotide polymorphisms (SNPs) are known from genetic studies, and by using the global GPCR alignment, we can determine the BW position of each SNP residue. Then each position can be directly compared against the list of activation hot-spots (and to the list of conserved contacts) to estimate the variant’s importance: whether the mutation causes some structural defects or whether it is likely to be benign. We scanned the Uniprot database [ 31 ] for naturally occurring mutations for all human GPCRs and converted the residue numbering to the BW scheme using our alignment. Table 3 provides several examples [ 87 – 98 ].
10.1371/journal.pcbi.1004805.t003
Table 3
Examples of natural variants and mutations that are associated with functional change or disease and which coincide with the NACHO residues.
Class
Protein
Uniprot
G-protein
Mutation
BW#
Activity Change
Disease Association
Reference
A-alpha
CB1
P21554
Gi/Go, Gs
T210A
3.46
Inactive
None
[ 87 ]
T210I
3.46
Highly constitutively active
None
[ 87 ]
L207A
3.43
Highly constitutively active
None
[ 88 ]
A-beta
V2R
P30518
Gs
R137C a
3.50
Constitutively active
NSIAD
[ 89 ]
R137L a
3.50
Constitutively active
NSIAD
[ 89 ]
A-gamma
CCR5
P51681
Gi/Go
R126N
3.50
Disables G-protein coupling
None
[ 90 ]
A-delta
FSHR
P23945
Gs
R573C a
6.36
Reduces AC stimulation
Ovarian dysgenesis 1
[ 91 ]
B
PTH1R
Q03431
Gs, Gq/G11
T410P a
6.37
Constitutively active
JMC
[ 92 ]
T410R a
6.37
Active (less than T410P)
JMC
[ 93 ]
H223R a
2.43
Constitutively active
JMC
[ 92 ]
C
CASR
P41180
Gi/Go, Gq/G11, G12/G13
F788C a
5.55
More active than wild type
Hypocalcemia
[ 94 ]
F806S a
6.36
No significant activating effect
Hypocalcemia
[ 95 , 96 ]
F788L a
5.55
More active than wild type
Hypocalcemia
[ 97 ]
F
FZD4
Q9ULV1
G12/G13
K436T a
6.36
Not known
Colorectal cancer
[ 98 ]
Predictions
A-alpha
DRD5
P21918
Gs
T297P a
6.36
Predicted change of function
Not known
Adhesion
GPR56
Q9Y653
Gq/G11, G12/G13
M493T a
3.43
Predicted change of function
Not known
a Natural variant.
For example, the natural variants R3.50C and R3.50L cause the vasopressin V2 receptor to be constitutively active. This causes “nephrogenic syndrome of inappropriate antidiuresis”, which presents itself as an inability to excrete a free water load, resulting in low sodium levels [ 89 ]. The mutations of R3.50 clearly interfere with arginine’s ability to form hydrogen bonds, and so they disrupt the activation mechanism.
Similarly the natural variant H2.43R in Parathyroid hormone receptor causes its constitutive activity. This mutation of class B receptor causes “Jansen metaphyseal chondrodysplasia”, which is characterized by short-limbed dwarfism [ 99 ]. Since the same G proteins couple to different GPCR classes, we can expect the same or similar structural signatures of activation in class B as in class A.
For both of the above examples, the mutations are known to cause constitutive activity. However, there are many observed natural variants, for which the effect is unknown. For example, we predict that the natural variant M3.43T of GPR56 will influence its activation, because the residue 3.43 has to switch contact residues during activation. This adhesion GPCR is involved in cell adhesion as well as in cell to cell interactions, and regulates the migration of neural precursor cells; thus the mutation likely has serious consequences. No databases of single nucleotide polymorphisms contain any functional information about this mutation (we checked Uniprot, and the GPCR specific TinyGRAP [ 33 ] and NAVA [ 32 ] databases), therefore this is a new prediction based on the analysis of the GPCR fold presented here. Another prediction can be made for the natural variant T6.36P of the D 5 dopamine receptor. This is a class A receptor and it influences the activity of adenylyl cyclase. Again, we predict that the natural variant T6.36P dramatically changes activation response of this receptor, either to be more constitutively active or less active.
We have illustrated the importance of the NACHO residues by finding disease associations that are caused by single mutations at these positions. The list of NACHO residues only contains 15 residues, which is about 5% of the transmembrane domain. Similarly, we hypothesize that mutations at the CHICO positions (that define the structural fold) can dramatically change the receptor function. There are many known natural variants whose effect has not been experimentally studied yet, and these criteria can be used to focus (experimental) attention on variants, which cause dramatic changes.
From Uniprot, we collected all 2449 GPCR natural variants, of which about half (1289) lie in the TM regions. These are listed in S3 Table together with their functional or disease associations, if available on Uniprot. Table 4 summarizes the limited disease-association data available for mutations in GPCRs. It shows that about half (∼53%) of the GPCR TM residue mutations have been found to be associated with diseases. This number jumps to about two-thirds for CHICO or NACHO residues (∼67% and ∼66% respectively) and almost all (12 out of 13 or ∼92%) for residues that appear on both CHICO and NACHO lists. This strongly suggests that NACHO and CHICO residues can help prioritize mutation sites to guide experimental validation of the structural and functional hypotheses presented by these specific residues.
10.1371/journal.pcbi.1004805.t004
Table 4
Summary of SNPs annotated on Uniprot. The complete list is in S3 Table .
Number of SNPs
With disease annotation
% with disease
All GPCRs
2489
694
27.9
All TM regions
1289
363
28.2
Excluding olfactory and unassigned
1463
635
43.4
TM
652
346
53.1
Non TM (Nterm+loops+Cterm)
811
289
35.6
CHICO only
161
105
65.2
CHICO
174
117
67.2
NACHO only
48
28
58.3
NACHO
61
40
65.6
Both CHICO and NACHO
13
12
92.3
There are still many SNPs in S3 Table that have unknown functional implications. We sort them with respect to a score capturing their relative position to the CHICO and NACHO residues: “distance to the closest NACHO + distance to CHICO - multiplicity of the closest NACHO - multiplicity of CHICO + Blosum62 of the mutation”. We hypothesize that the entries with the lowest score are very likely to cause dramatic changes in the receptor structure and function. The full list thus provides a large number of testable hypotheses about the molecular basis of disease-associated SNPs.
The CHICO and NACHO residues are results of a structural comparison, but the functional relevance of mutations is often obtained from phylogenetic considerations instead of structural ones. In S4 Table we compare these two approaches. We consider variations in sequences among a curated list of 77 P2Y12 orthologs [ 100 ], and among orthologs in an uncurated database for multiple proteins [ 101 ]. The CHICO and NACHO positions are more conserved than other TM residues in all GPCR classes among orthologs; and residues present on both lists are even more conserved. Thus both approaches are consistent, and should be combined to form more detailed insights.
Size of helix movement in available crystal structures and implications for homology modeling
By analysis of the inter-helical contacts we constructed the GRoSS alignment between all the GPCR proteins, from which new homology models can be derived. For structure prediction we would like to know how far the homology models are from the target structure. The variability of the TM bundle can be measured using the available crystal structures.
Fig 9 shows the observed move sizes, when the individual TM helices are treated as rigid bodies. Each pair of known structures was first aligned together, then each helix of the first protein was individually aligned to the corresponding helix of the second protein and the size of the move was measured. The center of mass translation was broken down into the direction along the helical axis and a direction perpendicular to it. The “tilt of axis” measures how much axis 1 had to be rotated to axis 2. And finally the “rotation around axis” measures the necessary rotation around the axis to map the corresponding atoms to each other.
10.1371/journal.pcbi.1004805.g009
Fig 9
Magnitude of the rigid body moves of the helices necessary to map one structure to another.
All TMs 1–7 from all available structure pairs were compared and each symbol denotes which TM is the data point from. The coordinate system is defined in the text. The maximal observed deviation is approximately proportional to the sequence dissimilarity of the two compared TMs, and it follows the same trend within class A (blue symbols) and across the GPCR superfamily (green symbols). The red symbols, which correspond to the active-inactive structure pairs, show rigid body moves caused by receptor activation. S10 Fig has an analogous plot of residual RMSD vs. similarity for each helix after the best rigid body transformation. RMSD shows a similar trend as the plots in this figure.
The maximal move sizes that need to be considered get smaller as the similarity of the TM sequence increases. If we are predicting a structure starting from a homology model with higher than 50% similarity, then we need only consider translating the helices up to 1.5 Å in any direction, tilting them up to 10°, and rotating around their axis by 40°. This is a very useful bound for refining homology models.
The same comparison can be applied to a single protein in multiple conformations. The red points in Fig 9 show the magnitude of rigid body moves undergone during activation for the 3 available pairs of active-inactive structures. Activation involves mainly the movement of TMs 5, 6, and 7. The computation of the move sizes ignores the bending of TM6 during activation, so it should be understood as an approximate description only.
Conclusion
A conceptual understanding of the molecular mechanisms behind biased signaling and functional selectivity is emerging around the conformational flexibility and dynamics of GPCRs. This is supported by an ever-increasing number of experimental and computational studies [ 102 – 107 ] that point to different ensembles of receptor conformations behind the pleiotropic signaling of GPCRs. These mechanistic studies have gotten a significant boost from the recent dramatic developments in GPCR structure determination methods.
We constructed the GRoSS sequence alignment of the transmembrane regions for all known human GPCRs. Although the inter-helical contact residue correspondence in the GRoSS alignment is in many cases approximate (intra-class) or non-existent (inter-class), several inter-helical contacts are highly conserved across the classes, which suggests their importance in the evolutionarily conserved GPCR fold. Our conserved contact analysis of the experimentally observed inactive and active conformations of Rhodopsin, Muscarinic M2, and adrenergic β 2 AR identified 15 residue positions in the TM regions that change molecular contacts upon activation. Targeted or natural mutations of these residues are known to cause dramatic changes in the receptor signaling. We recommend that these be the starting point for examining mechanisms for activation and for deleterious mutations. The GRoSS alignment of the TM domains also leads to a functional phylogenetic tree that captures many evolutionary signatures of GPCR evolution. This shows that the class-level differences in the GPCR superfamily are encoded in the TM domains even though the divergence in the loop domains is usually used to distinguish the classes from each other.
The GRoSS alignment is also a promising starting point for structure prediction, as there are no gaps present in the TM domains. For a protein in question one can build homology models based on any of the available templates by mutating the corresponding amino acids. While the increasing coverage of proteins by the crystal structures makes it easier to find a close template, the GRoSS alignment allows remote homology modeling, which can be particularly useful for modeling active states for all GPCR classes. GPCRs are too large (>300 residues) for exploring larger conformational changes only using molecular dynamics. Comparing the available structures with respect to the GRoSS alignment gives approximate bounds on the size of rigid body moves needed for the TM helices to reach the target structure.
The GRoSS alignment is unique in aligning all human GPCR sequences by maximizing the number of conserved inter-helical contacts. These conserved contacts provide a basis for defining a GPCR superfamily-wide structural fold, functionally conserved residue positions (even if residue type may not be conserved), and activation hot-spot residues (NACHOs).
Supporting Information
S1 Table
List of studied GPCR crystal structures.
When multiple structures are available, then the one with the highest resolution or the one with least deformed TM helices is used.
(PDF)
S2 Table
GRoSS sequence alignment for all 817 human GPCRs.
S1 File has this alignment in fasta format. Since there are no gaps in the TM domains, the alignment of each protein is uniquely determined by the BW.50 residues for each TM 1 through 7. We list also the expected range of the helical TM regions, which is estimated as the average TM region in the known crystal structures from the same class. In the discussion of the bitter taste receptors (TAS2Rs), we identified two possible alignments of TM6, but only the first one is presented in the following table. The second choice is to decrease the start, end, and BW50 residue of TM6 by 4.
(CSV)
S3 Table
GPCR natural variants annotated by Uniprot mapped to BW numbering and indicating their proximity to the NACHO and CHICO residues.
The mutations are ordered according to the following score: “distance to the closest NACHO + distance to CHICO - multiplicity of the closest NACHO - multiplicity of CHICO + Blosum62 of the mutation”.
(CSV)
S4 Table
Conservation of CHICO and NACHO residues among orthologs.
For orthologs of several proteins we computed average amino acids conservation over TM, and over CHICO/NACHO residues. The data shows that CHICO and NACHO positions are more conserved than other TM residues in all GPCR classes. Residues present on both lists are even more conserved. Two measures of conservation provided by Jalview are used: Consensus is the percentage of orthologs sharing the human amino acid; and Conservation is a qualitative measure counting the number of conserved chemical properties. For P2Y12, we used a curated list of 77 orthologs from [ 100 ]. For other proteins, we collected predicted orthologs from the MetaPhOrs database (release 201405 [ 101 ]), aligned them with Clustal Omega, and then removed sequences with gaps in the TM regions.
(PDF)
S1 Fig
Detailed view of conserved motifs in class A GPCRs.
The conserved residues in 24 different structures (including active) have very similar positions, which shows that the class A GPCR fold is highly conserved. The full TM bundle is shown in Fig 1 .
(PDF)
S2 Fig
Sequence similarity (%) of the TM bundles between crystal structures for the final sequence alignment.
Two residues are similar if their BLOSUM62 entry is positive.
(PDF)
S3 Fig
Backbone (atoms N, C α , C, O) RMSD of the TM bundles for the final sequence alignment.
For a given pair of structures, there may exist a different sequence alignment, which results in a lower RMSD than the listed one.
(PDF)
S4 Fig
High-resolution phylogenetic tree ( Fig 7 ) based on TM similarity only.
The pdf file is searchable for the UNIPROT accession numbers. Loops were ignored. Color coding denotes the GPCR class. Proteins with known crystal structure are emphasized with a dot.
(PDF)
S5 Fig
Testing the robustness of the alignment of the Vomeronasal receptors with the other groups.
This is an extended version of Fig 3 , same caption.
(PDF)
S6 Fig
Testing the robustness of the alignment of the Taste2 receptors with the other groups.
This is an extended version of Fig 4 , same caption.
(PDF)
S7 Fig
Testing the robustness of the alignment of the Vomeronasal receptors with the GPCRtm substitution matrix.
Same caption as in Fig 3 .
(PDF)
S8 Fig
Testing the robustness of the alignment of the Taste2 receptors with the GPCRtm substitution matrix.
Same caption as in Fig 4 .
(PDF)
S9 Fig
Diagram of interhelical contacts present in classes B, C, and F.
The width of the line connecting two TMs is proportional to the number of contacts present in all structures from the given class. The list in red font shows the contacts not present in any available structure from other classes.
(PDF)
S10 Fig
RMSD of helices after best rigid body move.
Same caption as Fig 9 .
(PDF)
S1 Text
Comparison of the GRoSS alignment to the HMM-HMM alignment [ 77 ] and to the GPCRDB alignment [ 24 , 78 ].
(DOCX)
S1 File
The GRoSS alignment in fasta format and annotation of the TM regions and BW residues in Jalview format.
The first 29 sequences are the actual sequences from the PDB files of used crystal structures; the rest of the sequences are from Uniprot. N-terminal, loops and C-terminal are not aligned. For interactive work it is useful to also highlight the TM regions and BW residues using the Jalview annotation gross-alignment.gff file.
(ZIP)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
A common modality in synthetic biology is to modulate the abundance of proteins for catalysis [ 1 ], computation [ 2 ], detection [ 3 ], and programmed therapeutics [ 4 ]. The vast majority of tools to modulate protein activity change mRNA and protein synthesis rates [ 5 ]. Such tools include transcriptional repressors [ 6 , 7 ], RNAi [ 8 ], riboregulators [ 9 ], and more recent tools using catalytically-inactivated Cas9, dCas9 [ 10 ].
Tools that change synthesis rates excel in applications where rapid transition to an ON state is required, specifically the increase in concentration of a desired protein, as transcription and translation occur on a time scale of minutes [ 11 , 12 ]. Turning OFF proteins (i.e. decreasing concentration) is indirect using synthesis tools. Protein degradation and dilution-by-growth reduce protein concentration when synthesis is low. In practice, most native proteins have long half-lives (over 20 hours) [ 13 – 15 ]; therefore, dilution is the primary mechanism to deplete these proteins. Dilution is directly related to growth rate; therefore, turning off the protein slows as growth slows [ 16 ]. This complicates knockdown of essential proteins because lower essential protein concentrations likely results in a slowed growth phenotype, where dilution and reducing the essential protein concentration is even further slowed.
Therefore to rapidly generate OFF states, degradation is much more desirable. Genetically-encoded targeted and inducible protein degradation is an attractive solution and has been demonstrated in a few strategies. A target protein is genetically fused to a destabilization tag. In prokaryotes, the tagged protein is stable, until degradation is induced by expression of a protease or adaptor protein, resulting in rapid decrease in protein concentration [ 17 – 19 ]. Several demonstrations of targeted, exogenous protein degradation have been applied recently for metabolic engineering [ 20 , 21 ], fundamental study of essential proteins [ 22 ] , enhancing recombinant protein production [ 23 ] , and protein-based circuits [ 24 ]. Most of these studies target essential proteins that otherwise could not be genetically knocked out.
Induced protein degradation strategies require that prior to induction the tag does not destabilize the protein (i), significantly affect native function (ii), or cause the tag to be inaccessible to the adaptor or protease (iii). Ideally, tagging can be done on the N- or C-terminus; however, all current tools for bacteria are limited to C-terminus [ 17 – 19 ]. This reduces the set of proteins able to be knocked down. For example, the mazE antitoxin is an essential protein [ 25 ] in Escherichia coli and requires a free C-terminus for native activity [ 26 ]. This protein may not be able to be tagged on its C-terminus and consequently may not be able to be studied using state-of-the-art induced protein degradation methods.
In response, we designed and characterized an N-degron tag (Ntag) for E . coli based loosely on the TEV-Induced Protein Instability/Inactivation (TIPI) system previously developed in Saccharomyces cerevisae [ 27 ] ( Fig 1A ). In this system, the target protein is N-terminally modified with a protection domain that hides an N-degron (an amino acid that destabilizes a protein when it is on the extreme N-terminus). When an encoded protease site is cleaved, the N-degron (non-methionine amino acid) is exposed. The protein is then degraded through the host N-end rule pathway. As we will discuss later, we do not use TEV for induction, but instead rely on expression of N-end rule pathway proteins, which in bacteria utilizes the ClpAP protease complex and the adaptor protein ClpS [ 28 – 30 ].
10.1371/journal.pone.0149746.g001
Fig 1
Targeted degradation is achievable using N-terminal fusions and elucidated with model-driven analysis.
( A ) A cleavable tag is fused N-terminally to the target protein. On cleavage, an N-degron is exposed leading to degradation of the target. ( B ) Balanced exponential phase growth experiments can be used to estimate parameters for a quantitative model ( C ) to calculate degradation rates. See S5 Fig for a detailed demonstration of calculations.
The degradome of prokaryotes is used to regulate growth rates, convoluting the effects of degradation and dilution-by-growth [ 31 , 32 ]. We utilize a model-driven framework to characterize the degradation rates in E . coli . Our analysis simultaneously accounted for changes in protein abundance related to targeted degradation, synthesis, and dilution in balanced, exponentially growing cultures ( Fig 1B and 1C ). For example, if degradation and dilution are increased, but not greater than synthesis, the protein abundance would still increase over time, masking the increased degradation.
In this study, we evaluated the genetic conditions that affect degradation rate toward optimizing the Ntag degradation system. We studied changes in degradation rate during balanced exponential growth. Growth and protein abundance were measured over time and fit to a differential equation. We confirm essential proteins (ClpP, ClpA, ClpS, and ClpX) in the degradation pathway support high degradation rates, and use these to demonstrate a dynamic induction of protein degradation via temporal availability of ClpP. We expect our Ntag to expand the applications of inducible protein degradation to metabolic engineering, protein circuit, and biological investigations.
Materials and Methods
Bacterial Strains and Media
All E . coli strains used in this study are described in Table A in S1 File . DH5α and Top10 strains were used for all plasmid construction. Derivatives of the MG1655 Z1 strain were used for all experiments [ 33 ]. The MG1655 Z1 strain constitutively expresses TetR and LacI so that synthesis of P Ltet-O and P trc promoters can be controlled through titration of doxycycline and IPTG respectively. Knockouts were introduced ( malE , sspB , clpASXP , aat ) to the Z1 strain using p1vir phage transduction from Keio library strains [ 25 ] (from Yale University E . coli Genetic Stock Center) and knockouts were verified by colony PCR. All kanamycin markers were excised using the pCP20 plasmid [ 34 ] using a previously described protocol [ 35 ].
Luria-Bertani (LB) media (10 g/L of tryptone, 5 g/L of yeast extract, 10 g/L of NaCl) with 0.4% glucose added to prevent carbon limitation [ 36 ] was used for all experiments. 40 ng/mL doxycycline was used to maintain expression of reporter proteins (Ntag-RFP, RFP, and Ntag-beta-galactosidase) and 17 mg/L chloramphenicol to maintain the plasmid. In experiments using pTrc-based plasmids, 100 mg/L of ampicillin was added to maintain the plasmid and 1 mM of IPTG was used to express the inserts.
Plasmid construction
All plasmids used in this study are listed and described in Table B in S1 File . All plasmids were constructed using Gibson Assembly techniques [ 37 ] and designed using the j5 software suite [ 38 ]. Construction oligonucleotides are listed in Table C in S1 File , and checking oligonucleotides are listed in Table D in S1 File . The plasmids which expressed the N-tagged proteins (called target plasmid from here on) were all pSC101-derived (specifically BioBricks pSB4C5). The Ntag sequence was codon optimized for E . coli and synthesized chemically (IDT, San Jose, CA). Details of the Ntag sequence and its motifs are provided in S1 File . Protein expression was driven by a P Ltet-O promoter and titratable using doxycycline ( S1 Fig ). A second plasmid bearing the proteins to modulate degradation (called the perturbation plasmid from here on) were all derived from pTrcHis2B. All plasmids were sequence verified and are available from AddGene (Deposit # 71910). All GenBank files are included in S2 File .
Culture Conditions
For all culture experiments, cells were freshly transformed with plasmids the previous day. Colonies were picked and cultured. Once concentration exceeded OD 600 1.0, cells were inoculated into 250 mL baffled shake flasks (working volume of 25 mL) at 1:1000 dilution. Shake flasks, once inoculated, were kept cold (12°C) until the next day and then cultivated at 37°C at 250 RPM. Sampling began once exponential phase was established (4–5 doublings, usually 2–5 hours from start of shaking) and measurements of protein activity and OD 600 occurred at least once an hour. OD was measured with UV-1800 Spectrophotometer (Shimadzu, Kyoto, Japan). Cultures that reached OD values beyond 1 were diluted into fresh shake flasks with prewarmed media of the same constituents to minimize transition disturbances and keep the cultures under balanced growth for continued measurements. All raw data for culture experiments are available in S3 and S4 Files .
Cleavage Assay
Cultures with relevant plasmids and inducers were grown to stationary phase (OD 5–10) in the presence of 100 ng/mL of anhydrotetracycline to maintain maximal synthesis of MBP-Ntag-RFP (compared to doxycycline). Cultured cells were then pelleted and resuspended in cold Tris-HCl buffer (50 mM, pH 7.5) with 1 mg/ml lysozyme. The samples were then incubated on ice for 10 minutes and frozen with dry ice/ethanol. Frozen cultures were then thawed on ice and centrifuged to separate protein from cellular debris at 17,000 g for 10 minutes in a precooled centrifuge. A Bradford assay (Biorad, Berkeley, CA) was used to quantify protein content in the lysate. Samples were diluted to 10 μg/mL with 0.1x Sample Buffer supplied with Wes Rabbit Kits (ProteinSimple, San Jose, CA). A polyclonal rabbit anti-FLAG primary antibody (#ab1162, Abcam, Cambridge, UK; Antibody Registry: AB_298215) and a monoclonal mouse MBP-HRP conjugated antibody (#e8038, New England Biolabs, Ipswitch, MA; Antibody Registry: AB_1559732) were used to visualize the bands in the Wes machine (ProteinSimple, San Jose, CA) using the standard instrument protocol. Both antibodies were used at a dilution of 1:50 (as recommended by the protocol) with the supplied Wes Antibody Diluent solution.
Fluorescence Assay
For RFP fluorescence measurements, 200 μL of culture was immediately chilled on ice and then measured (λ ex = 585 nm; λ em = 615 nm) on a Synergy H1 Plate Reader (Biotek, Winooski, VT) at a gain of 100. A non-RFP producing E . coli culture in LB media sample was used as the blank for the fluorescence measurement.
Beta-galactosidase assay
For beta-galactosidase optical density measurements, 0.5 mL of culture was placed into cold microcentrifuge tubes kept on ice. Cultures were pelleted at 5,000 g for 5 minutes using a precooled centrifuge. Cells were then resuspended to an OD of approximately 1 with PBS solution. 25 μL of concentrated cells were added to 100 μL permeabilization solution (100 mM Na 2 HPO 4 , 20 mM KCl, 2 mM MgSO 4 , 0.8 mg/mL CTAB (hexadecyltrimethylammonium bromide), 0.4 mg/mL sodium deoxycholate, 10 mg/mL bovine serum albumin, and 5.4 μL/mL beta-mercaptoethanol. Permeabilization solution was always made fresh on the day of assay from stock solutions. Standards were also prepared from purified E . coli beta-galactosidase (Sigma Aldrich, St. Louis, MO) in PBS and added to permeabilization solution in same manner as cells. After the last sample was collected for the day, all samples (cells and perm solution mix) were removed from room temperature and incubated at 30°C for at least 30 minutes. 90 μL of prewarmed (30°C) substrate solution (60 mM Na 2 HPO 4 , 40 mM NaH 2 PO 4 , 1 mg/mL o-nitrophenyl-β-D-Galactoside (ONPG), and 2.7 μL/mL β-mercaptoethanol) was dispensed into wells on a flat-bottom 96 well plate. 15 μL of each sample was added to designated wells, mixed for a few minutes using a platform shaker, and absorbance at 420 nm was read over one hour every five minutes in Synergy H1 plate reader. Plates were sealed with plastic film and kept warm (30°C) in plate reader during reading. Activity was measured as the change in OD over time, using the standards to calibrate to specific activity. All samples were measured in technical replicate.
Culture conditions for induction of ClpAP pathway proteins
For testing the effects of inducing ClpAP pathway proteins, cultures were first grown in the pre-induction condition in shake flasks (LB + 0.4% glucose, 40 ng/mL doxycycline, no IPTG), where the target protein (mCherry) was expressed. Upon OD reaching between 0.1 and 0.3, 1 mL of culture was sampled, centrifuged at 5,000 g for 5 minutes, and resuspended with warmed (37°C) PBS. Samples were centrifuged again in the same conditions and then inoculated into a fresh shake flask with new media. For “Repression Only” experiment, the plasmids pKS012 (Ntag-RFP) and pTrc (empty plasmid) were used and the post-induction media had no doxycycline but had IPTG to control for cellular effects from IPTG. For the “Degradation and Repression” condition, the pKS012 (Ntag-RFP) and pKS044 (ClpP expression) plasmids were used and the post-induction media had IPTG but no doxycycline. The fluorescence and OD at the zero time point was determined by extrapolating a negative control condition (no knockdown initiated) back to t = 0 h. All raw data for induction experiments are available in S6 File .
Model specifications and analysis
Change in protein abundance is related to synthesis, dilution, and degradation [ 39 , 40 ]:
d P d t = α − μ P − V m a x P K m + P
The synthesis rate, α, is dependent on physiological parameters related to cellular phenotype (e.g. growth rate, ribosomal availability, etc.) and to the plasmid construct (e.g. the protein, N-terminal region, RBS, promoter strength, plasmid copy number, mRNA stability, etc.) [ 41 , 42 ]. Therefore synthesis rates cannot be compared between different target protein constructs (i.e., RFP synthesis may be different from Ntag-RFP). Dilution is represented with μ P term. Our balanced-growth experiments typically operate at around 10,000 target proteins per cell, as determined from correlation of activity versus abundance ( S2 Fig ). K m values for ClpP-based degradation is around 600 proteins/cell[ 17 ]. We, accordingly, assume degradation to be zero order ( P >> K m ) and constant ( β) under our balanced-growth experiments. Rearrangement of the differential equation yields
d P d t + μ P = α − β
Thus, synthesis and degradation rates are equated to values that were measured. Specifically, protein and OD was measured over time. Rates (dP/dt and μ) could be calculated from the slopes at consecutive time points. P was the average protein between two time points. For a given biological replicate, at least three but no more than five time points were used to calculate α -β. Apparent degradation β was then calculated by subtraction from the no degradation control (Δ clpP or Δ clpP + pTrc). S5 Fig shows an example calculation of β from raw measurements.
For calculations and predictions of inducible time course dynamics, the full differential equation was used:
d P d t = α − μ P − V m a x P K m + P
Based on purified mCherry and fluorescence measurements ( S2 Fig ), α = 40,000 synthesized proteins/cell h (unrepressed protein synthesis), Vmax = β = 15000 approximately 1 Fluorescence/OD600 = 10 proteins/cell. Specific parameters are proteins/cell h, K m = 600 proteins/cell (approximately 1 μM), μ = 1 1/h, and P 0 = 40,000 proteins/cell. Parameters were changed based on the tested condition (e.g. for synthesis repression only, V max = 0 and α = 0). All numerical solutions were performed using the ode45 function on MATLAB R2014b (MathWorks, Natick, MA) running on a Macbook Pro running OS 10.9 (Apple, Cupertino, CA). All scripts are included in S2 File and all calculated data is included in S5 File .
Results and Discussion
Ntag is spontaneously cleaved in vivo
To characterize our system, we first evaluated cleavage of the Ntag protective domain and linker. To test cleavage in E . coli , we flanked the linker with two epitopes, MBP (N-terminal end of linker) and the FLAG tag (C-terminal end of linker / N-terminal end of mCherry). We denote the combined linker and FLAG tag region as the Ntag, which connected the upstream MBP to downstream RFP ( Fig 2A ). The cleaved product would be approximately 50 kD (MBP) and 40 kD (RFP). The full sequence of the Ntag is provided ( S1 File ).
10.1371/journal.pone.0149746.g002
Fig 2
The Ntag is natively cleaved in E . coli .
( A ) A protein was designed to test for cleavage of the Ntag consisting of an N-terminal maltose binding protein (MBP) and a C-terminal RFP. The Ntag is equivalent to the linker sequence in the original TIPI study but with a FLAG epitope on the C-terminal end [ 27 ] ( B and C ). Cleavage was found in all conditions where the MBP-Ntag-RFP was expressed using immunoblotting techniques regardless of the expression of the perturbation plasmid (Lanes 1–2 and 4–5). A negative control for cellular background produced no relevant fragments (Lanes 3 and 6). The intact construct (~90kD) was observed when blotting with both an Anti-FLAG ( B , Lanes 1–2) as well as Anti-MBP ( C , Lanes 4–5) primary antibody. Consistent cleavage products were observed with the relevant antibodies: ~40 kD band with Anti-FLAG (Lanes 1–2) and ~50 kD band with Anti-MBP (Lanes 4–5). Band image was produced using the Compass software via a lane view option. Dotted lines indicate where original gel image is cropped for lanes not used in analysis . An unaccounted band at 70 kD is seen when immunoblotting with the Anti-MBP antibody ( C , Lanes 4 and 5). This band is likely a C-terminally truncated MBP-Ntag-RFP construct. Sensitivity of the anti-MBP antibody is higher compared to anti-FLAG as suggested by the increased signal from the same 90 kD bands. Considering that the 70 kD bands only appears with anti-MBP, the abundance is likely at least an order of magnitude lower compared to the rest of the visualized products. Another spurious band is detected at around ~200 kD in Lane 2. This could be due to binding of the antibody to the lane standard. Protein Simple scientists have observed this phenomenon with many primary antibodies [ 43 ].
First, we tested for proper cleavage of the target protein MBP-Ntag-RFP at the Ntag. Specifically, we measure cleavage under the context of two plasmids because of planned inducibility/perturbations described later in our study. We cotransformed two plasmids, a “target” plasmid and a “perturbation” plasmid. The target plasmid had high expression of MBP-Ntag-RFP ( Fig 2B ). The perturbation plasmid constitutively expressed either no protein (empty vector) or GFP (which was not expected to cleave MBP-Ntag-RFP) (Lanes 1–2 and 4–5 respectively). GFP expression was confirmed for all GFP conditions by measuring fluorescence (data not shown). Samples were taken in late stationary phase, and protein target cleavage was assayed by fragment size using an automated western blotter (The Wes). To test for antibody background with cellular lysate, we used a plasmid expressing untagged RFP in place of the MBP-Ntag-RFP protein (Lanes 3 and 6).
The MBP-Ntag-RFP construct is cleaved endogenously as shown when GFP or no protein is induced, therefore cleavage is independent of coexpressed proteins. This is evident by the ~40 kD anti-FLAG band that corresponds to RFP ( Fig 2B , Lane 1–2 ) and a complementary ~50 kD anti-MBP band ( Fig 2C , Lanes 4–5 ). Genomic MBP ( malE gene) was knocked out to avoid background MBP staining ( Lanes 3 and 6 ). We further tested growth phase-dependent and protease-dependent cleavage and found that cleavage occurs throughout the growth curve ( S6 Fig ).
We also confirmed that cleavage occurs at a single location on the linker. Despite multiple species at the Ntag-RFP portion ( Fig 2B , Lane 1–2 ), only one band is seen for the MBP-Ntag portion ( Fig 2C & S6 Fig ). This observation suggests that cleavage occurs in one location and that the Ntag-RFP is processed proteolytically.
Degradation occurs on different proteins fused with the Ntag
Spontaneous cleavage of the Ntag suggests the protein may be degraded through the N-end Rule mechanism present in bacteria [ 30 ]. We cultured cells expressing Ntag fused to RFP (Ntag-RFP) under balanced exponential growth, synthesis, and degradation conditions as described in Methods and measured growth rate and protein concentration per cell. Using the measured values, we calculated the apparent degradation rate (β) as illustrated in S5 Fig . The N-end Rule pathway in E . coli is known to require the ClpP protease for degradation of tagged substrate [ 30 ]. We, therefore, expect that for a putatively unstable protein with an N-degron, degradation rate β would be insignificant in a Δ clpP strain. We tested our Ntag-RFP construct in the Δ clpP strain versus wild type (WT) strain (strains described in Table A in S1 File ). We see a significantly higher degradation rate in WT versus Δ clpP ( Fig 3A ). To verify that this is strictly a function of the Ntag, we show that with a stable mCherry construct (untagged RFP) that the degradation rate is very low and statistically indistinguishable from the Δ clpP strain with Ntag-RFP ( Fig 3B ). In these conditions, Δ clpP reduces growth rate by only 10%, indicating the deletion only has a minor effect on physiology ( S3 Fig ).
10.1371/journal.pone.0149746.g003
Fig 3
The Ntag confers ClpP-dependent degradation of RFP and beta-galactosidase.
Measuring the apparent degradation rate (β) for the Ntag-RFP construct was significantly higher in the wildtype versus Δ clpP strain ( A ). For an untagged RFP, there is no appreciable degradation evident by the lack of apparent degradation rate ( B ). Fusing Ntag to beta-galactosidase resulted in similar degradation as measured by a significant β value ( C ). For all calculations of the apparent degradation rate, the respective Δ clpP condition is used to calculate the synthesis rate (α). This methodology is illustrated in S5 Fig . Single * designates p < 0.05 and triple (***) designates p < 0.001. All p values are calculated between ΔclpP and WT strains for given constructs. Values are mean ± s.e.m. (n = 4 to 10).
To verify that this degradation is related to the general presence of the Ntag and not an artifact of RFP, we fused Ntag to beta-galactosidase as well. We again see appreciable degradation for the Ntag-beta galactosidase construct suggesting that this tag engenders generalizable degradation ( Fig 3C ). Note: Under these conditions (high glucose, no IPTG/lactose), endogenous beta-galactosidate activity is undetectable (data not shown). We estimate that degradation rates between the RFP (~6×10 6 residues•cell -1 •hr -1 ) and beta-galactosidase (~2×10 6 residues•cell -1 •hr -1 ) are consistent. Furthermore, we find that our calculated degradation rates compared well to established rates for native SsrA-tagged substrates, which are also ClpP dependent. For example, ArcA tagged with SsrA is approximately degraded at ~6.5×10 6 residues•cell -1 •hr -1 (full details in S1 File ) [ 17 , 44 ]. Raw values of RFP•OD -1 and OD over time are shown in S5 Fig to provide comparison with the model-derived calculations.
Ntag-protein degradation is dependent on N-end rule pathway components
N-terminal modification and degradation of our Ntag proteins by ClpP strongly suggests an N-end rule dependence mechanism. ClpA, ClpS, and ClpP are constituents of the E . coli N-end rule pathways ( Fig 4A ) [ 29 , 30 ]. Native substrates for the terminal N-end rule protease, ClpAP, include proteins modified by the Aat transferase to reveal N-degrons [ 30 , 32 ]. C-terminally-tagged proteins with the SsrA tag (AANDENYALAA) can also be degraded via the ClpAP protease [ 45 ]. Adaptor proteins ClpS and SspB are known to catalyze degradation by guiding substrate to protease complexes. Specifically, ClpS is known to facilitate degradation of proteins with N-degrons via ClpAP, and SspB is known to inhibit degradation of SsrA-tagged proteins via ClpAP but foster ClpXP-based degradation [ 28 , 29 , 46 ]. ClpXP is a structurally similar complex to ClpAP, and both use ClpP for protease activity; however, ClpXP has not been observed to degrade proteins with N-degrons [ 32 ].
10.1371/journal.pone.0149746.g004
Fig 4
Ntag fused proteins are degraded via the N-end Rule pathway.
The N-end Rule pathway requires an adaptor protein ClpS that promotes substrate toward the ClpAP protease complex ( A ). Native proteins are modified to have N-degrons via the Aat transferase enzyme or other mechanisms. These proteins with N-degrons are similarly facilitated by ClpS to the ClpAP complex for degradation as the Ntag fused target proteins. Another system tags native proteins with C-degrons, tags signaling degradation on the C-terminus. These proteins are preferentially degraded in the ClpXP complex by the SspB adaptor protein but can be degraded in the ClpAP complex. The apparent degradation rate (β) for the Ntag-RFP construct was measured across different knockouts of these degradation pathway members in comparison to the Δ clpP strain ( B ). The presence of N-end Rule members (ClpA, ClpS, ClpP) was required for degradation as indicated by no detectable degradation rate. The absence of ClpX reduced degradation to zero as well. Deletion of Aat resulted in, perhaps, enhanced degradation, and removal of the SspB adaptor protein reduced degradation but did not eliminate it. Single * designates p < 0.05 and triple (***) designates p < 0.001. Calculations of p-values are between Δ clpP and the given condition. Values are mean ± s.e.m. (n = 4 to 10).
To systematically determine which of the different pathway members in Ntag-based degradation are necessary, we measured apparent degradation rates for Ntag-RFP in strains with each pathway member described above knocked out ( Fig 4B ). In this study, growth rates changed by less than 20% and are accounted for in our analysis ( S3 Fig ). As expected, deletion of the other known N-end Rule components (Δ clpA and Δ clpS ) abolished degradation. Interestingly, we see that Δ clpX unfoldase protein eliminates RFP degradation. We surmise that blocking the ClpXP pathway may result in saturation of the ClpAP pathway. In the absence of ClpX, both proteins with N-degrons or C-degrons would be processed by ClpAP. Under this postulation, the Ntag protein would not only be competing with internal proteins with N-degrons but would now also with, for example, proteins with a SsrA tag. Previously this saturation has been shown to decrease the apparent degradation rate of specific proteins due to a queuing effect, and therefore may be a possible explanation for no degradation under Δ clpX [ 39 ].
The auxiliary, degradation-related proteins, Aat and SspB only had modest effects on protein degradation. Δ aat should reduce the number of native proteins targeted for degradation, allowing more degradation capacity to be available for Ntag-RFP. We see a slight increase in degradation rate above WT. Δ sspB appears to reduce the degradation rate slightly, presumably by mitigating the ClpXP pathway, forcing more proteins to use the ClpAP pathway for degradation.
Overexpressing some components of the N-end Rule pathway yields enhanced degradation
To improve the characteristic time of protein knockdown, protein degradation rates must be very high. We sought to identify the rate-limiting step from the N-end Rule pathway by overexpressing individual components and measuring β for Ntag-RFP ( Fig 5 ). In all conditions, we verify growth rates within 10% ( S4 Fig ). In this experiment, the perturbation plasmid overexpresses a protein of interest under balanced growth conditions. Overexpression of ClpS and ClpP increase degradation, while other components did not. We note that ClpA overexpression may in fact yield less degradation compared to the wild-type control (WT + pTrc); however, we could not distinguish the conditions statistically. It is interesting that ClpA overexpression did not enhance degradation. ClpP overexpression resulted in the most dramatic increase in protein degradation rate, and is therefore the primary limiting step. Previous work has shown that ClpP may be the limiting stoichiometric component for formation of the ClpAP complex [ 47 ]. ClpS overexpression also improved degradation. ClpS’ enhancement role cannot be due to pathway limitation because only one is needed per ClpAP complex [ 48 ]. ClpS, however, is an inhibitor of other ClpA substrates such as SsrA tagged proteins [ 49 ]. ClpS overexpression, therefore, likely decreases substrate competition for Ntagged proteins.
10.1371/journal.pone.0149746.g005
Fig 5
Ntag degradation is enhanced by ClpP and ClpS overexpression.
Overexpression of a perturbation protein was used to test for degradation effects on Ntag-RFP. Overall degradation is reduced with the target/perturbation plasmid system as indicated by the control conditions (Δ clpP + pTrc versus WT + pTrc). ClpA overexpression did not change degradation detectably. Overexpression of each ClpP and ClpS enhanced the degradation rate. Single * designates p < 0.05 and triple (***) designates p < 0.001. Calculations of p-values are between Δ clpP + pTrc and the given condition. Values are mean ± s.e.m. (n = 4 to 6).
We note that the control values of β are different than in the prior experiment, presumably due to introduction of the perturbation plasmid, IPTG addition, or antibiotics used to maintain the perturbation plasmid.
Protein degradation is minor during rapid growth, but important in slow growth
In the previous experiments, we measured degradation under a balanced growth state, however, the ultimate goal is to dynamically reduce protein for synthetic biology purposes. The two possible mechanisms are to repress synthesis (by TetR repression) or induce degradation. Because the presence or absence of ClpP results in large changes in degradation, the induction of ClpP should initiate the degradation of an Ntagged protein ( Fig 6A ). Furthermore, the presence or absence of ClpP was not significantly deleterious to fitness under balanced growth ( S3 and S4 Figs ). Prior to conducting this experiment, we modeled protein concentration dynamics with parameters derived from our experiments ( Fig 6B ). Under maximal growth conditions (growth rates ~1.1 h -1 ), dilution effects are significant; therefore, synthesis repression is a valuable means for protein knockdown. Combined with degradation, slightly more knockdown is achieved, but the effect is minor during rapid growth as suggested by our modeling.
10.1371/journal.pone.0149746.g006
Fig 6
Degradation rate is minor for protein knockdown in rapidly growing cells.
The target/perturbation plasmid system converts to an inducible, dynamic knockdown platform by temporally expressing ClpP within a clpP deletion strain ( A ). In high growth rate, knockdown is predicted to be primarily a result of dilution and degradation only slightly increases knockdown compared to synthesis repression alone ( B ). Experimental inducibility is tested using the target/perturbation system with Ntag-RFP as the target ( C ). β, degradation, increases post induction ( D ). Single * designates p < 0.05. All experimental values are mean ± s.e.m. (n = 6 to 8).
Consistent with the model, our measurements find that protein degradation only improves depletion slightly compared to repression alone ( Fig 6C ). However, the time required for knockdown is slower compared to our predictions ( Fig 6B ). This is likely due to a number of effects not considered in the model: (a) a delay associated with the required accumulation of ClpP or association with ClpA, (b) lag resulting from resuspending cells in fresh media at t = 0 and/or (c) leaky target protein expression due to residual doxycycline. A β estimate showed increased degradation post-induction ( Fig 6D ). The apparent β (~ 450 RFU•OD -1 •h -1 ) in this experiment is lower than the β in the previous balanced growth experiments (1000–1800 RFU•OD -1 •h -1 ). This lower rate may be due to lower concentration of the target protein (i.e., at the lower concentration, degradation may be a first order rate). Ntag-RFP concentration was lower in the induction experiments (100–3500 RFU•OD -1 ) compared to balanced growth where there was no repression (1000–20000 RFU•OD -1 ).
Looking forward, we expect induced protein degradation to be particularly valuable for slowed growth conditions where dilution effects are small. In general, we see the knockdowns efficiencies improve for combining degradation and synthesis repression versus simply synthesis repression as growth rates are lowered. Slow growth conditions are particularly pertinent when essential enzymes are the targets of protein degradation. The less of the essential enzyme, the slower the growth, as highlighted in this review [ 16 ].
In this study, we did not consider how degradation rate is affected by the sequence or structural features of the Ntag. Previous studies have found specific features of N-degrons that affect degradation rate. For example, acidic residues adjacent to the N-terminus (after cleavage) reduce degradation rates [ 50 ], while aliphatic residues increase the degradation rate [ 51 ]. A separate study was able to tune the degradation rate through modification of the ClpS adaptor protein [ 52 ]. These structural features could be used to rationally design an enhanced Ntag degradation system.
We have expanded the options for inducible protein degradation by establishing the Ntag system for facile, generalizable degradation of proteins by using a genetically encoded N-terminal modification. We implemented a modeling framework that focused on degradation rates, which allowed systematic analysis of the effects of protein degradation pathway components. We verify Ntag generalizability, N-end Rule dependence, enhancement conditions, and an inducible strategy. Our tool should be useful to a variety of metabolic engineering applications, protein-based circuits, and study of essential proteins.
Supporting Information
S1 Fig
Titrating doxycycline allows controllable synthesis of target protein.
Untagged RFP is driven by expression from a pl-tetO promoter. In MG1655 Z1 strain, constitutive expression of TetR represses synthesis of RFP. Synthesis can be controlled by adding varying levels of doxycycline, which is known to sequester TetR. Measurements were performed at a working volume of 200 μL in a 96 well plate running in Synergy H1 Plate Reader (Biotek, Winooski, VT).
(EPS)
S2 Fig
Purification of MBP-Ntag-RFP (from pKS011) allows for a rough correlation between fluorescence and protein number.
Method for protein purification can be found in S1 File .
(EPS)
S3 Fig
Growth rates for balanced growth one plasmid experiments.
Raw data available in S3 File .
(EPS)
S4 Fig
Growth rates for balanced growth two plasmid experiments.
Raw data available in S4 File .
(EPS)
S5 Fig
An overview of apparent degradation calculation.
Raw protein abundance/activity and biomass (OD) are measured during balanced growth (exponential phase) ( 1 ). From the data, dP/dt, μ, and P are calculated and used to calculate α-β over time ( 2 and 3 ). Average α-β is computed ( 4 ) and then apparent degradation (β) yields from the subtraction of the wild type condition from the control condition (in this case, Δ clpP ) ( 5 ). Note: this data is from one day of experiments and thus does not correspond to Fig 3 (n = 2). Raw data for this figure is available in S7 File .
(EPS)
S6 Fig
Time-course assay shows one cleavage event and cleavage occurring throughout the growth curve.
In this assay, MBP-Ntag-RFP was coexpressed with a second plasmid either the empty pTrc plasmid (P) or pJB028 (28). Cells were grown to midlog phase (OD = 0.6) at t0, at which point a sampling occurred ( Lanes 1 and 2 ). IPTG was then added to one of the conditions (28 + I) and sampled separately ( Lanes 5 and 8 ). Sampling occurred again 1 ( Lanes 3 , 4, and 5 ) and 3 ( Lanes 6, 7 , and 8 ) hours later. Intact MBP-Ntag-RFP is seen throughout with Anti-MBP antibody as indicated by band at around 85 kDa. Cleavage products seen throughout as indicated by 45 kD band on Anti-MBP blot and 40 kD bands on Anti-FLAG. Furthermore, a single cleavage event is indicated suggested by a singular band for the Ntag-MBP portion.
(EPS)
S1 File
Supplemental Information.
This document includes supplementary tables, molecular cloning information (sequences, oligonucleotides), methods for protein purification, and calculations of protein degradation rates.
(PDF)
S2 File
Plasmid GenBank files and MATLAB code.
GenBank files for all plasmids used in this study and MATLAB code for protein knockdown simulation.
(ZIP)
S3 File
Raw Data for Single Plasmid Experiments.
Excel file with all measurements of OD, fluorescence, beta-galactosidase activity, and calculations for apparent degradation rate.
(XLSX)
S4 File
Raw Data for Double Plasmid Experiments.
Excel file with all measurements of OD, fluorescence, and calculations for apparent degradation rate.
(XLSX)
S5 File
Knockdown simulation data.
Data used in Fig 6 from MATLAB simulation.
(CSV)
S6 File
Raw Data for Induction Experiments.
Excel file with all measurements of OD, fluorescence and calculations for the induction experiments.
(XLSX)
S7 File
Raw Data and Calculation for S5 Fig .
Excel file with all measurements of OD, fluorescence and calculations for S5 Fig calculations.
(XLSX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
According to Cancer Research UK ( https://www.cancerresearchuk.org/health-professional/cancer-statistics/statistics-by-cancer-type/lung-cancer ) there are approximately 50,000 new cases of lung cancer every year making lung cancer the third most common cancer in the UK. Furthermore, lung cancer is the most common cause of cancer death in the UK with 5- and 10-year survival as low as 10% and 5%, respectively (Cancer Research UK), which highlights the need to detect and correctly treat lung cancer as early as possible. This is particularly pertinent when imaging techniques identify indeterminate pulmonary nodules (IPNs) which are between 4mm and 20mm in size and carry a risk of malignancy of 10–70%. Often this is too low to justify a biopsy or other invasive procedure that carries a risk of morbidity and according to the British Thoracic Society Guidelines [ 1 ], an option for the clinician and patient is CT surveillance or ‘watchful waiting’. This entails repeat scanning at 3 months and 1 year to assess nodule volume doubling time (VDT). Patients with a VDT below 25% after 1 year are classed as negative and discharged, though some patients will not be discharged for up to 4 years follow up. Following the introduction of multidetector computed tomography (MDCT), the number of nodules detected, particularly those that are small, has increased dramatically with the prevalence of noncalcified lung nodules as 33% (range 17–53%) and 13% (range 2–24%), in screening and non-screening study populations, respectively [ 1 ].
Oncimmune's EarlyCDT–Lung is a simple ELISA blood test that measures seven lung cancer specific autoantibodies, and is used for the assessment of malignancy risk in patients with IPNs. Robust and easy to use, it can be run in any laboratory with standard laboratory equipment. The test has been marketed in the USA since 2012 and over 150,000 tests have been sold. A “kit” form of the test was CE marked in May 2017 for distribution to clinical laboratories outside the USA.
Using a decision analytic model, the objective of this study was to examine the cost-effectiveness of autoantibody test (AABT), EarlyCDT–Lung, in the diagnosis of lung cancer amongst patients with IPNs applied in the addition to CT surveillance, compared to CT surveillance alone as specified in the British Thoracic Society guidelines in which patients are offered surveillance through repeat CT scanning.
Methods
Developing the model structure
The comparison of the different testing strategies considered in this study are best represented using a modelling framework in which the various possible testing and treatment pathways can be compared. The patient group under examination in this study are 62 year old patients [ 2 ] with IPNs. Identified by imaging, these nodules are between 4mm and 20mm in size and carry a risk of malignancy of 10–65%. A model allows explicit representation of the impact of the accuracy of the tests, the costs incurred by the health care provider, and the impact on health-related quality of life (QoL) experienced by the patients that follow a particular diagnostic pathway.
A model which consists of a combination of a decision tree and Markov model was developed using TreeAge Pro 2001 software (TreeAge Software Inc., Williamstown, MA, USA). A life-time time horizon was adopted. Given that some events may occur on the patient pathways many years into the future, a Markov model approach was considered to be most appropriate.
Two different testing pathways were compared which describe alternative approaches to the testing and surveillance of these patients and their IPNs. The testing pathways are shown in Fig 1 :
Fig 1 shows the testing pathways for the two strategies considered in this analysis. These are defined as follows:
AABT+Surveillance –Patients receive the AABT (EarlyCDT-Lung), amongst those that test positive for malignancy, these are then given a biopsy followed by surgery for those that have confirmed malignancy identified through biopsy. Patients that are AABT test negative then follow the surveillance strategy below and are followed up with CT scans at 3 months, 12 months, and 24 months. Patients that are found to have a nodule that has grown at follow-up, are given a surgical biopsy. If the nodule is benign no further follow up is required; while patients that are found to have a malignant nodule receive surgery.
Surveillance –Patients receive a CT scan at 3 months, 12 months, and 24 months. Those that test positive for malignancy following a biopsy, are then given surgery for the confirmed malignancy. Patients that test negative continue on the surveillance pathway. Again, patients that are found to have a benign nodule that has grown at follow up are given surgical biopsy.
In all cases patients that have surgery are at risk of surgery related mortality and complications.
10.1371/journal.pone.0237492.g001
Fig 1
Testing pathways for the AABT and surveillance strategies.
Following the approach described by Gould, Sanders [ 2 ] Fig 2 shows the model structure describing the health states of patients and how these evolve over time.
10.1371/journal.pone.0237492.g002
Fig 2
Markov model (reproduced from Gould, Sanders [ 2 ]).
Patients enter the model undiagnosed either with an undiagnosed benign nodule, or an undiagnosed malignant nodule which may be at a local, regional, or distant stage. Patients that receive a true positive test then transition to their respective diagnosed state and receive surgery if the nodule is malignant. While those that are false negative (i.e. have a malignant nodule that is missed) remain undiagnosed but may progress to a more advanced disease state in the future. Patients that transition to the diagnosed benign state remain in this state under surveillance but may require surgical biopsy in the future if their nodule is subject to significant growth. Patients in the diagnosed local and regional states are at risk of recurrence or progression respectively for a period of 5 years after which they are assumed to be disease free. Patients in the diagnosed distant states are at constant risk of cancer related mortality for the rest of their lives. In all model states, patients are also subject to all-cause mortality.
Model assumptions
As part of the modelling framework and in order to conduct this analysis it is necessary to make some assumptions. These are described as follows:
Biopsy is 100% accurate
Compliance with surveillance is 100%
Nodules are diagnosed following a CT scan after doubling/progression
A CT scan is never performed after biopsy or surgery as imaging after an invasive diagnostic procedure is unusual [ 2 ]
Similarly, biopsy and observation are never performed after surgery [ 2 ]
Surgery will always be performed if biopsy confirms a malignancy [ 2 ]
If biopsy reveals a benign diagnosis, patients are subject to surveillance [ 2 ]
Surgical biopsy is always performed if benign nodule growth is observed during surveillance [ 2 ]
If no growth is observed after 24 months, then nodule is assumed to be benign and no further surveillance is conducted
CT scans performed during surveillance are 100% accurate at detecting growth in benign nodules
Patients in the diagnosed local and regional disease states are at risk of recurrence and progression for 5 years, after which they are considered to be free from cancer
Following a positive test, a patient is referred to a multi-disciplinary team (MDT)
Data requirements
The data required to parameterise the economic model were obtained through the extensive use of secondary sources. The parameters used in this model can be broadly categorized into prevalence of malignancy, accuracy parameters, transition probabilities between model states, resource use and costs, and utility values.
Prevalence
In the base case scenario, the prevalence of malignancy amongst patients presenting with IPNs was taken to be 9.5% (7/74) which is applied to both arms of the analysis. This is based on the logic as described by Edelsberg, Weycker [ 3 ] to interpret the data described in the study by Tanner, Porter [ 4 ], which is as follows:
In the study by Tanner, 74 patients were assigned to CT surveillance from an intermediate-risk group, amongst which 7 had lung cancer, yielding a lung cancer prevalence of 9.5 (7/74).
Test accuracy
Two alternative scenarios which describe the test accuracy of the AABT test are considered in this analysis, which are based on the availability of current test accuracy parameters for the AABT. The test accuracy parameters for the tests considered in this analysis are described Table 1 below:
The data from the 12 studies used in Gould, Sanders [ 2 ] to estimate the test accuracy values for a CT scan were re-analysed using the Metandi function in Stata so that the necessary information for the probabilistic sensitivity analysis could be obtained (see below). This led to slightly different sensitivity and specificity values compared to those reported by Gould, Sanders [ 2 ] (Gould values: Sensitivity = 0.965; specificity = 0.558).
10.1371/journal.pone.0237492.t001
Table 1 Test accuracy parameter values.
Test
Sensitivity
Specificity
Reference
AABT:
Scenario A
0.41
0.93
Taken from a ROC curve described in [ 5 ]
Scenario B
0.28
0.98
“”
CT Scan
0.923
0.723
[ 2 ] (taken from a meta-analysis of 12 studies)
Transition probabilities
Table 2 describes the probabilities and proportions that are applied to the economic model. In all cases, unless otherwise noted, the probabilities are monthly probabilities.
10.1371/journal.pone.0237492.t002
Table 2 Transition probabilities used in the model.
Parameter
Value
Reference
Proportion of malignant nodules that are initially:
Local Stage
0.875
[ 2 ]
Regional Stage
0.125 (0.078–0.165)
“”
Probability of detecting growth in a benign nodule during observation period: During first month
0.28 (0.13–0.29)
[ 2 ]
During each subsequent month
0.005 (0–0.01)
[ 2 ]
Natural Mortality
Varied by age
Office of National Statistics (2018)
Probability distant cancer related mortality
[ 2 ] (use+-50% as Gould)
0–12 months
0.1255 (+-50% Gould)
13–24 months
0.0670 (+-50% Gould)
25–36 months
0.0589 (+-50% Gould)
37–48 months
0.0150 (+-50% Gould)
Mortality Undiagnosed malignant nodule
0.02688 (+-50% Gould)
Based on a life expectancy of 36.7 months for patient with untreated 2cm nodule that doubled every 5.24 months [ 2 ]
Probability of progression from undiagnosed local to regional, and from undiagnosed regional to distant cancer
0.19224 (0.18887–0.21005; 95% CI)
[ 2 ]
Probability mortality due to biopsy
29/31960
[ 6 ]
Probability guided needle biopsy complications
230/31960
“”
Probability mortality due to recurrence
Calculated from data. See text [ 2 ]
0–12 months
0.0106
(use +-50% see Gould)
13–24 months
0.0100
25–36 months
0.0090
37–48 months
0.0114
Probability mortality due regional cancer
“”
0–12 months
0.0340
(use +-50% see Gould)
13–24 months
0.0296
25–36 months
0.0225
37–48 months
0.0155
Surgery related mortality for malignant nodule
0.042 (0.017–0.053)
[ 2 ]
Surgery related mortality for benign nodule
0.005 (0.002–0.016)
“”
Surgery complications for malignant nodules
0.084 (0.048–0.11)
[ 2 ]
Surgery complications for benign nodules
0.065 (0.033–0.13)
“”
Proportion of patients that receive radiotherapy with surgery
5/35
[ 7 ]
Proportion of patients that receive chemotherapy with surgery
11/35
“”
Calculation of cancer related mortality rates
Mortality rates for patients in the local, regional, and distant cancer states post-surgery were calculated by fitting a model by maximum likelihood to data survival curves for patients with pathologically staged lung cancer (T1N0M0), pathologically staged regional lung cancer (any T N1–3 M0), and distant lung cancer (any T any N M1) from the linked Medicare claims–Surveillance, Epidemiology and End Results (SEER) tumour registry, US [ 2 ].
Local cancer-related mortality was derived from survival data for 1,207 Medicare beneficiaries with surgically treated, T1N0M0 non–small-cell lung cancer. The regional cancer-related mortality was derived from survival data for 1954 Medicare beneficiaries with pathologically staged regional lung cancer. The distant cancer-related mortality was estimated from 10 835 Medicare beneficiaries with distant-stage non-small-cell lung cancer. All data here are from the SEER tumour registry, 1990–1993, as described in Gould, Sanders [ 2 ] ( Fig 3 ).
10.1371/journal.pone.0237492.g003
Fig 3
Survival curves for patients with pathologically staged local lung cancer (T1N0M0), pathologically staged regional lung cancer (any T N1-3 M0) and distant lung cancer (any T any N M1) from the linked Medicare claims–Surveillance, Epidemiology and End Results (SEER) tumour registry, USA.
(Reproduced from Gould, Sanders [ 2 ]).
Calculation of cancer progression rates
To calculate the progression rates amongst patients with undiagnosed malignant nodules, the observed doubling times in the figure above were used. A model (probability of progression = 1-exp(-rate*t)) to obtain the monthly probability of progression was fit to the data ( Fig 4 ) using maximum likelihood. The resulting model output showing 1 –probability of progression over time is also shown in Fig 4 .
10.1371/journal.pone.0237492.g004
Fig 4
The frequency plot of the observed doubling times for 67 pulmonary nodules and mass lesions from the veterans administration–armed forces cooperative study on asymptomatic pulmonary nodules (described in Gould, Sanders [ 2 ]).
Dashed line shows model output to estimate monthly progression rate.
Costs and resource use
Table 3 describes the costs applied in the economic model. All costs are in pounds (£) sterling for the 2016/17 price year. NHS Reference costs were used to attribute costs to resource use.
10.1371/journal.pone.0237492.t003
Table 3 Resource use unit costs.
Parameter
Value
Reference (NHS Reference Costs 2016/17 unless otherwise noted)
AABT
£70
Oncimmune
CT Scan (single area, no contrast)
£85.56
RD20A - Computerised Tomography Scan of One Area, without Contrast, 19 years and over
MDT (Multidisciplinary team)
£111.99
CMDT_Oth
Guided Needle biopsy
£948.92
DZ71Z Minor Thoracic Procedures
Surgery no complications
£7,713.03
DZ02K Complex Thoracic Procedures, 19 years and over, with CC Score 0–2
Surgery + complications
£10,177.74
DZ02H Complex Thoracic Procedures, 19 years and over, with CC Score 6+
Surgery: biopsy no complications
£3,091.08
DZ63C Major Thoracic Procedures, 19 years and over, with CC Score 0–2
Surgery: biopsy with complications
£6,733.76
DZ63A Major Thoracic Procedures, 19 years and over, with CC Score 6+
Radiotherapy
£3,252
[ 7 ], inflated from 2011/12 prices (£3,039)
Chemotherapy
£4,155.15
[ 7 ], inflated from 2011/12 prices (£3,883)
Utility values
The utility values used in the analysis to inform the quality adjusted life year (QALY) are described in Table 4 below.
10.1371/journal.pone.0237492.t004
Table 4 Utility values.
Parameter
Value
Reference
Age 55–64
0.810
[ 8 ] EQ-5D index value population norms for the UK–England, using country specific Time-trade off (TTO) values
Age 65–74
0.773
“”
Age 75+
0.703
“”
Serious adverse event due to biopsy
-0.2
[ 6 ]
Local
0.71
n = 33 (stage IA and IB) at 12 months [ 9 ]
Regional
0.65
n = 12 (stage I and Stage II) at 12 months “”
Distant
0.62
n = 4 (stage IV) at 12 months “”
Analysis
This model-based economic evaluation utilizes the primary outcome of the cost per QALY, where one QALY is defined as one year lived in perfect health. A time step of 1 month was applied in the Markov model with a life-time time horizon. This time horizon was chosen to allow the full impact of the interventions that may occur many years in the future to be included. Half cycle correction was incorporated into the analysis. Discounting was applied at 3.5% for costs and outcomes as recommended by NICE [ 10 ], with the analysis conducted from the health-care provider perspective The results are presented using the incremental cost-effectiveness ratio (ICER) which is defined as the difference in the costs of the two strategies divided by the difference in their outcomes, and net-monetary benefit (NMB) which is defined for each strategy as:
N M B = Q A L Y s g a i n e d x w i l l i n g n e s s t o p a y ( W T P ) f o r a Q A L Y – C o s t o f t h e i n t e r v e n t i o n .
Where the WTP for the QALY is taken to be £20,000, which is at the lower end of the £20,000 to £30,000 acceptance threshold as recommended by NICE [ 10 ].
Sensitivity analysis
This analysis contains a number of important uncertainties that must be examined. These were examined through one-way and probabilistic sensitivity analyses.
Given this is an early economic evaluation the optimum price of the AABT test is examined. The price of the AABT test was varied to show the point at which the price of the test leads to an ICER of £20,000/QALY which is at the low end of the threshold for acceptance of an intervention as given by NICE [ 10 ].
Probabilistic sensitivity analysis (PSA) was implemented by using Beta distributions where data made this possible, using the method of moments to obtain the Alpha and Beta parameters in each case [ 11 ]. Where a range was described for parameter uncertainty, then the standard error for the Beta distribution was estimated as follows:
S E = ( U ‐ L ) / ( 2 × 1.96 )
Where U and L are the upper and lower limits of the range respectively [ 12 ].
In order to apply probabilistic sensitivity analysis (PSA) to the CT scan test accuracy values the following equation was used
l o g i t ( s e n s i t i v i t y ) = Λ e − β 2 − e − β l o g i t ( s p e c i f i c i t y )
From the metandi output in Stata (see Test Accuracy Section above), specificity was 0.7234 (se = 0.0276), Λ was found to be 3.156 (se = 0.2296) and β = -0.5362433. The resulting sensitivity was estimated by sampling from a beta distribution for the specificity and sampling from a normal distribution for Λ. Beta was kept constant.
Expected value of perfect information
The expected value of perfect information (EVPI) is based on the probability of a decision maker making the wrong decision about which testing strategy to choose and the impact of making that wrong decision. It provides insights into the maximum value of conducting further research to resolve the uncertainty in the parameter values. Expected value of perfect parameter information (EVPPI) takes this idea forward and shows the maximum value of resolving the uncertainty in specific parameters or specific groups of parameters. Both methods are useful in providing insights into the direction of future research. The methods used here to estimate the EVPI and EVPPI are those described by Strong, Oakley [ 13 ].
Results
The results here are presented for two scenarios based on alternative estimates for the test accuracy of the AABT test (Scenario A–Sensitivity 0.41 Specificity 0.93; Scenario B–Sensitivity 0.28 Specificity 0.98).
Scenario A
At a price of £70, the cost-effectiveness results for AABT vs Surveillance for Scenario A showing the average cost and QALYs gained per patient are shown in Table 5 below:
It can be seen that when the price for AABT = £70, and adopting the test accuracy parameters as described for Scenario A, AABT+Surveillance is more costly and more effective in terms of QALYs gained than surveillance alone. Given that the ICER is well under £20,000, AABT+Surveillance can certainly be regarded as cost-effective.
10.1371/journal.pone.0237492.t005
Table 5 Cost-effectiveness of AABT vs. surveillance for testing Scenario A, where the price of AABT = £70.
Scenario A:
Total Cost
Inc . Cost
QALYs Gained
Inc . QALYs
ICER (Cost/QALY)
Surveillance
£2,261
10.6850
AABT+Surveillance
£2,410
£149
10.7465
0.0614
£2,417
The results when the uncertainty in the parameter values is considered in the analysis for Scenario A are shown in Fig 5 above. It can be seen that AABT+Surveillance is always more costly than surveillance alone and almost always (99.4%) more effective in terms of QALYs gained. The cost-effectiveness acceptability curve (CEAC) shows that AABT+Surveillance is more likely to be cost-effective at a WTP for the QALY of £2,000 and above. At a WTP of £20,000/QALY AABT is approximately 99% likely to be cost-effective.
10.1371/journal.pone.0237492.g005
Fig 5
Probabilistic sensitivity analysis results for Scenario A for 1,000 model runs showing the cost-effectiveness plane and the cost-effectiveness acceptability curve for Scenario A.
The net benefit for the AABT+Surveillance and Surveillance alone scenarios with variation in the price of the AABT test for Scenario A are shown in Fig 6 .
10.1371/journal.pone.0237492.g006
Fig 6
Net monetary benefit at a WTP = £20,000 / QALY for the AABT+Surveillance and surveillance strategies with variation in the price of the AABT test for Scenario A.
It can be seen that the price of the AABT can be up to £1,150.37 and still have greater NMB than compared to Surveillance alone (WTP = £20,000). Above this price, surveillance alone becomes more cost-effective.
Assuming an annual incidence of 50,000 patients presenting with IPNs, a discount rate of 3.5% and a 10-year time horizon after which this technology will be superseded, the expected value of information for Scenario A is shown in the figure below.
It can be seen in Fig 7 that at a WTP for the QALY of £20,000, the EVPI for Scenario A is approximately £1,000,000.
10.1371/journal.pone.0237492.g007
Fig 7
Expected value of information for Scenario A.
Assuming an incidence of 50,000 new patients presenting with IPNs, a discount rate of 3.5% and a 10-year time horizon.
In terms of expected value of perfect parameter information. The EVPPI for different groups of parameters for Scenario A are shown in Fig 8 below.
10.1371/journal.pone.0237492.g008
Fig 8
Expected value of perfect parameter information for Scenario A.
It can be seen ( Fig 8 ) that there is some value in resolving the uncertainty in the disease related mortality rates, initial patient characteristics (i.e. prevalence and proportion of patients with local and regional disease) and the utility values. The most value of resolving this uncertainty over 10 years is achieved by targeting the mortality rates (approximately £600,000), while the values for the other parameters is actually very low being approximately £80,000 for the initial patient characteristics and £25,000 for the utility values.
Scenario B
Similar to Scenario A, when the price for AABT = £70, and adopting the test accuracy parameters as described for Scenario B, AABT+Surveillance is more costly and more effective in terms of QALYs gained than Surveillance alone. Again, given the low ICER value, AABT can certainly be regarded as cost-effective ( Table 6 ).
10.1371/journal.pone.0237492.t006
Table 6 Cost-effectiveness of AABT+Surveillance vs. surveillance alone for testing Scenario B, where the price of AABT = £70.
Scenario B:
Total Cost
Inc . Cost
QALYs Gained
Inc . QALYs
ICER (Cost/QALY)
Surveillance
£2,261
10.6850
AABT+Surveillance
£2,358
£97
10.7308
0.0457
£2,121
It can be seen from the results of the PSA for Scenario B ( Fig 9 ) that AABT+Surveillance is always more costly than surveillance alone and always more effective in terms of QALYs gained. The cost-effectiveness acceptability curve (CEAC) shows that AABT+Surveillance is more likely to be cost-effective at a WTP for the QALY of £3,000 and above. At a WTP of £20,000/QALY AABT is more than 98% likely to be cost-effective.
10.1371/journal.pone.0237492.g009
Fig 9
Probabilistic sensitivity analysis results for Scenario B for 1,000 model runs showing the cost-effectiveness plane and the cost-effectiveness acceptability curve for Scenario B.
The net benefit for the AABT+Surveillance and Surveillance scenarios with variation in the price of the AABT test for Scenario B are shown in Fig 10 .
10.1371/journal.pone.0237492.g010
Fig 10
Net monetary benefit at a WTP = £20,000 / QALY for the AABT+Surveillance and surveillance alone strategies with variation in the price of the AABT test for Scenario B.
In the case of the test accuracy parameters in scenario B the AABT test can be priced up to £887.28, and be more cost-effective than surveillance alone (WTP = £20,000/QALY).
It can be seen in Fig 11 that at a WTP for the QALY of £20,000 the EVPI for Scenario B is approximately £100,000. In the case of Scenario B, there was found to be no value in resolving the uncertainty in any of the parameter groups (not shown).
10.1371/journal.pone.0237492.g011
Fig 11
Expected value of information for Scenario B.
Assuming an incidence of 50,000 new patients presenting with IPNs, a discount rate of 3.5% and a 10-year time horizon.
Analysis of Scenario A and Scenario B
Given that both Scenarios A and B show that AABT+Surveillance is cost effective compared to surveillance alone, it is important to establish whether the extra QALYs gained from Scenario A compared to Scenario B are worth paying for.
As shown in Table 7 , the ICER for AABT+Surveillance Scenario A compared to Scenario B is £3,277 which is well below the NICE acceptance threshold of £20,000. Thus, it can be concluded that Scenario A has the most cost-effective test accuracy parameters and as such these should be adopted.
10.1371/journal.pone.0237492.t007
Table 7 Cost-effectiveness results showing a comparison between Surveillance and AABT for Scenarios A and B.
Total Cost
Inc . Cost
QALYs Gained
Inc . QALYs
ICER (Cost/QALY)
Surveillance
£2,261
10.6850
AABT+Surveillance Scenario B
£2,358
£97
10.7308
0.0457
£2,121.43
AABT+Surveillance Scenario A
£2,410
£52
10.7465
0.0157
£3,277.41
The CEAC for the 3 scenarios is shown below:
It can be seen from the CEAC shown in Fig 12 that up to a WTP for a QALY of approximately £2,000, Surveillance alone is most likely to be the most cost-effective scenario, and then from WTP of approximately £3,000 upwards AABT+Surveillance Scenario A is most likely to be cost-effective. At a WTP for a QALY of £20,000 Scenario A is approximately 90% likely to be the most cost-effective option, with this probability increasing with increased WTP values.
10.1371/journal.pone.0237492.g012
Fig 12
Cost-effectiveness acceptability curve comparing surveillance, AABT+Surveillance Scenario A, and AABT+Surveillance Scenario B.
Discussion
Using a decision modelling framework consisting of a decision tree and Markov model the analysis here has examined the cost-effectiveness of the AABT test in addition to CT surveillance compared to the current practice of CT surveillance alone for patients with indeterminate pulmonary nodules (IPNs), as recommended in the British Thoracic Society guidelines. Two alternative pairs of test accuracy parameters were considered for the AABT test. Scenario A (sensitivity 41% specificity 93%) with its higher sensitivity and lower specificity compared to Scenario B (sensitivity 28% Specificity 98%). This analysis took a baseline price of the AABT test to be £70, but also investigated the maximum price the test could be set at, and still remain cost-effective at these values.
Based on the results of this analysis at baseline, it is quite clear that the use of the AABT test in addition to CT surveillance is cost-effective compared to CT surveillance alone. And while at £70 per test, the test was never found to be cost saving, based on the NICE threshold of acceptance (i.e. less than £20,000 / QALY) the extra effect in terms of QALYs was always found to be worth paying for. In terms of which test accuracy parameters should be adopted, again the results here are clear, with Scenario A (Sensitivity 41% Specificity 93%) being the preferred option. The probabilistic sensitivity analysis supports the main conclusions and indeed provide reassurance that these results are robust to realistic variations in the input parameters. Thus, it can be concluded that the extra sensitivity of Scenario A compared to Scenario B (41% vs. 28%) at the expense of some specificity (93% vs. 98%) leads to improved patient outcomes that are worth paying for.
The results here also demonstrate that at £70, the AABT test is significantly under-priced and could be priced at between approximately £900 to £1,170 (depending on the Scenario) and still be cost-effective based on the NICE acceptance threshold for the QALY. Although it is acknowledged that while this would still be under the NICE acceptance threshold, the increased budget impact would obviously make this much less attractive to decision makers.
“The intuition behind the results here is that by adding the additional AABT diagnostic test to current surveillance, there is a positive trade off between the patient benefits associated with the detection of true positives and the negative impact of a slight increase in the number of false positive test results due to imperfect specificity that will lead to a further biopsy. However, given the very high specificity of the AABT test and the relatively high prevalence of malignancy (approx. 10%) , there is very little downside to a patient receiving this test, apart from the cost of the test. If a patient is found to require surgery as a result of a true positive test, then this clearly has a positive impact on patient outcomes, while a false negative test simply leads the patient to the surveillance test pathway, which is what the patient would have received in the absence of the AABT test. Even though it could be argued that the sensitivity is still relatively low (41% is the highest in Scenario A), this is still sufficient to have a positive impact on patient outcomes.”
The conclusions drawn from this analysis are supported in the study by Edelsberg, Weycker [ 3 ] which also found the AABT test in addition to CT surveillance to be cost-effective compared to surveillance alone. However, our analysis differs from the approach taken by these authors in that we use a Markov model which allows for patients to be followed over their life-time. While Edelsberg, Weycker [ 3 ] do attempt to draw conclusions over longer time horizons, the Markov model is regarded as the most appropriate model design for chronic diseases.
This analysis has a number of limitations that have to be acknowledged. The cost of palliative care for patients that die of lung cancer have not been incorporated into this analysis. However, given that patient outcomes are improved in the AABT plus surveillance scenario, their inclusion would cause the AABT to appear even more cost-effective than has been presented in these results. Rather than doing an extensive systematic review to identify the best available evidence to populate the model, this study has made extensive use of the parameters, data and model structure from the study by [ 2 ]. While this should be regarded as a limitation, the uncertainty in the parameter values used has have been subject to extensive sensitivity analysis and this has shown that the conclusions drawn from this analysis are robust to realistic variation in the parameter values.
Conclusion
The results here have demonstrated that the use of the EarlyCDT–Lung AABT test in addition to CT surveillance will have a positive impact on patient outcomes and is a cost-effective approach to the management of patients with IPNs, with all the results well under the NICE threshold for acceptance. The conclusions drawn from this analysis are also very robust to realistic variation in the parameters used in this model.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Large-scale quantitative proteomic analysis acquired under different conditions has been used to gain deeper insight into protein function and regulation [ 1 , 2 ]. One widely used approach consists of comparing the level of expression of a given protein between different conditions and to determine whether or not the difference between the various groups is meaningful based on statistical analysis [ 3 ]. The following step, which consists of assigning a biological function context to the proteomics data or identifying key molecular targets, remains a challenging task. Correlation within gene expression (i.e. co-expression analysis) has been used to extract biologically meaningful information from different data sets [ 4 , 5 ], but has rarely been used on proteomics data with the exception of the work of Gibbs et al 2013 [ 6 ].
Here, we have used different topologically-based strategies to divide the main list of identified proteins into different modules by first using a Weighted Gene Co-expression Network Analysis (WGCNA) developed by the Horvath group [ 7 , 8 ]. These modules were, in turn, separated and broken down into clusters and sub-clusters using MCODE [ 9 ] and hierarchical clustering was applied to the protein expression patterns. As these approaches rely solely on expression profiles without priory functional knowledge, we then employed several knowledge-based tools to both verify and assign biological relevance to the observed sub-clusters of data. We compared the protein-protein interaction networks generated de novo using WGCNA against predicted networks for the same subset of proteins using STRING [ 10 , 11 ] which clearly shows a significant overlap between the WGCNA analysis of the proteomics data and STRING.
In this study, we present a protein co-expression analysis of the dataset for glioblastoma multiforme previously acquired and published by Deighton et al . [ 12 ]. Our new findings support these previous observations. In addition to the previous findings from this study, we have highlighted three major modules of co-expressed proteins that are associated with specific membrane structures; the mitochondria, the endoplasmic reticulum (ER), and vesicle membranes. We show that within these modules, we can generate protein networks, which are similar to protein interaction networks predicted by data-mining from the literature without using an immunoprecipitation approach or native gel separation.
In addition to a major disruption of the Electron Transfer Chain (ETC) observed in the tumour samples, we show that the proteins composing each of the main ETC complexes (Complex I to IV) are mostly co-expressed but that each of the complexes are affected differently. In the ER, the unfolded protein response as well as the oxidative stress pathway are up-regulated. Furthermore, two isoforms of Pyruvate kinase (PKM) (isoform M1 and isoform M2) were differentially co-expressed with a high PKM2/PKM1 ratio supporting aerobic glycolysis (a hallmark feature of cancer) at the expense of oxidative phosphorylation (most likely inefficient due to the disruption of the ETC). While the M2 isoform seems poorly co-expressed with other proteins, the M1 isoform is part of a more defined network which is involved in ion transport, cellular response to insulin stimulus, glutamate secretion as well as syntaxin binding. In this study, we show that the use of a weighted protein co-expression analysis provides a level of information about protein interaction networks which is not possible to obtain using a standard data analysis approaches.
Methods
The data used were the quantitative proteomics data from a glioblastoma multiforme study conducted by Deighton et al . [ 12 ]. All protein identities are publically available through PRIDE ( http://www.ebi.ac.uk/pride ) PRD000620 and the label-free quantitation output presented in S1 Table . In that study, 6 controls and 6 tumour samples were used. A mitochondrial extraction was performed, the samples were trypsinised, followed by a shotgun proteomics analysis. The quantitative analysis was performed using Progenesis (Non Linear Dynamics, UK). The MS data for this present study were searched against a human RefSeq database (34 284 sequences) using Mascot (version 2.4.1), Matrix Sciences), with a significance threshold p < 0.05 in addition to peptide ion score cut-off of 20. Each analysed protein needed at least 2 identified peptides. Conversion from RefSeq to gene symbol was performed using the biological DataBase network (bioDBnet) [ 13 ].
Label-free intensity data were ArcsinH transformed prior to analysis as log transform of 0 is not ideal. A simple trait matrix was built as follows; the parameter “state” was a single number defined as “1” for disease and “0” for control. The R WGCNA package [ 7 ] was used to perform the analysis of the data set. The Topological Overlap Matrix (TOM) was created using a cut height of 0.25 and a minimum module size of 30. The analysis produced five modules, identified with different colours (‘brown’, ‘turquoise’, ‘blue’, ‘grey’ and ‘yellow’) with the ‘grey’ module containing all proteins that were not sorted to any of the other modules shown in Table 1 . A hard threshold approach was used for comparison purposes where:
a ij = corr(prot i , prot j ) β The correlation a ij between the ArcsinH intensity of the protein prot i and prot j is measured. The factor β is a thresholding parameter, for hard thresholding, we used a β of 1 only for validation purposes (for FDR evaluation by comparing the same dataset against a randomised one). For the remainder of the study, we used a β of 10, justified from S1 Fig which corresponds to the lowest value showing a good scale-free topology.
10.1371/journal.pone.0161828.t001
Table 1 GO term enrichment assignments for the five main clusters.
Module
GO
GO Term
Description
P-value
FDR
Category
q-value
Blue
Process
GO:0022900
electron transport chain
1.04E-26
2.98E-23
Function
GO:0008137
NADH dehydrogenase (ubiquinone) activity
3.74E-15
4.85E-12
Component
GO:0044455
mitochondrial membrane part
2.93E-25
2.10E-22
Brown
Process
GO:0006397
mRNA processing
4.28E-08
2.44E-04
Function
GO:0003676
nucleic acid binding
7.16E-11
9.29E-08
Component
GO:0005783
endoplasmic reticulum
1.06E-06
7.62E-04
Turquoise
Process
GO:0007268
synaptic transmission
4.89E-08
2.79E-04
Function
GO:0030276
clathrin binding
5.95E-05
3.86E-02
Component
GO:0097458
neuron part
4.80E-14
3.45E-11
Yellow
Process
NO ENRICHMENT FOUND
Function
NO ENRICHMENT FOUND
Component
GO:0005739
mitochondrion
4.37E-10
3.14E-07
Grey
Process
NO ENRICHMENT FOUND
Function
NO ENRICHMENT FOUND
Component
NO ENRICHMENT FOUND
Selected groups of proteins were exported to the online Gene Ontology enRIchment anaLysis and visuaLizAtion tool (GOrilla) [ 14 ] with the gene names of each individual module used as a target set and those of the remaining modules used as a background set for the enrichment. The network data were exported to Cytoscape v3.2.1 [ 15 ], with the corresponding WGCNA function [ 8 ] where they were visualised. “Hub” clusters were defined using MCODE v1.4.1 [ 9 ]. Default suggested parameters were used.
Protein interaction networks for each module were generated from co-expression similarity using WGCNA. The same set of proteins was then clustered into a network using STRING v10 using specific confidence parameters presented in table [ 10 , 11 ]. The two generated networks were then compared using “Network Analysis Tool” (NeAT) [ 16 ] with default parameters, randomisation was based on the Erdos-Renyi method. The WGCNA networks were used as the ‘Query’ networks and those from STRING as the ‘Reference’ networks. For both network types, different cut-off points were tested and are described in Table 2 .
10.1371/journal.pone.0161828.t002
Table 2 Different cut-off combinations for comparing between networks generated using WGCNA and STRING prediction.
WGCNA
STRING
Module name 1
Cut-off 2
Cut-off 3
P-value 4
Jaccard 5
Brown
0.3
0.4
1.40E-32
0.0871
0.7
5.50E-20
0.081
0.2
0.4
6.10E-63
0.1029
0.7
4.00E-53
0.0693
0.1
0.4
3.50E-47
0.0681
0.7
1.40E-40
0.0377
Turquoise
0.3
0.4
1.20E-12
0.048
0.7
2.10E-11
0.0289
0.2
0.4
1.70E-15
0.0455
0.7
1.60E-13
0.024
0.1
0.4
2.30E-14
0.0427
0.7
4.00E-11
0.0206
Blue
0.3
0.4
0.00E+00
0.2209
0.7
0.00E+00
0.1767
0.2
0.4
6.00E-289
0.1631
0.7
2.40E-221
0.1137
0.1
0.4
1.20E-154
0.1206
0.7
5.30E-141
0.078
1) Different Modules extracted using WGCNA.
2) Threshold values for pair-wise protein co-expression (Pearson correlation^10).
3) Confidence score cut off to generate protein-protein network in STRING ( http://string-db.org/ ).
4) p-value calculated for the overlap of the 2 protein networks (from WGCNA and from STRING) using NeAT.
5) Jaccard similarity coefficient: size of the intersection divided by the size of the union of the sample sets
Hub proteins were associated with proteins having a number of interactions which was two-fold greater than the standard deviation above the average number of interactions found in a specific module (i.e. z-score above 2). A hierarchical clustering of protein intensity was applied on the largest clusters generated from MCODE for each of the module networks using R version 3.1 GPLOT package and Ward’s method. Sub-clusters were then generated and their nature analysed using ToppCluster [ 17 ] for comparative analysis. Additionally, the clusters were analysed using the database of differentially expressed proteins in human cancers, dbDEPC 2.0 [ 18 ]. Fig 1 illustrates the overall data analysis platform used for this work and the number of proteins associated to each of the modules.
10.1371/journal.pone.0161828.g001
Fig 1
Workflow illustrating the analysis performed.
On the left, the step-wise protein list fragmentation is illustrated. On the right, the different bioinformatics tools used are described. The number of proteins per group are presented and coloured using the WGCNA colour coding. The names associated to the sub-clusters are illustrated on the right side of each sub-Clusters.
Results
Defining the different modules
It has been previously described that peptide and protein interaction networks possess a scale-free network topology similar to those found in gene co-expression networks [ 6 , 19 ]. The glioblastoma dataset from Deighton et al . [ 12 ] is composed of a set of 799 proteins identified with at least two peptides (presented in S1 Table ). As shown in S1 Fig , a power of β = 10 has been extracted from the original data and used for further analysis.
The overall data analysis strategy used in this work is presented in Fig 1 . A series of data analysis tools, based on either network topology characteristics or literature knowledge was used to cluster groups of proteins. One possible concern with the use of the Deighton et al (10) dataset for network analysis is its rather small size (a 2 group comparison with 6 replicates only) where normally datasets of at least 25–30 samples are commonly used for co-expression analysis (4). We have estimated the false discovery rate (FDR) by using a permutation approach as described elsewhere [ 20 , 21 ]. Hierarchical clustering of the pair wise correlation coefficient was evaluated first using a thresholding parameter of β = 1, shown in Fig 2A each row and column are represents proteins, the colour purple is associated with clusters having a high correlation coefficient, and white is associated to a high anti-correlation coefficient. Protein intensities were also randomly permutated and the same clustering method was used once again (shown in Fig 2B ). As expected, a significant decrease in the level of correlation is observed. The plot of the distribution of the correlation coefficients for both datasets (the direct dataset and the randomised one) is shown in Fig 2C . The randomised dataset is centred around 0 (blue) while the normal dataset (turquoise) exhibits two distributions roughly centred on 0.5 and -0.5 associate to either correlated or anti-correlated pairs or proteins, respectively. We have evaluated the FDR for different correlation coefficients and a FDR of 5% was calculated for a correlation coefficient of 0.754 and above and values of -0.728 and less for meaningful anti-correlation ( Fig 2D ). A similar calculation was performed using a β = 10 ( Fig 2E ), a FDR of 5% was found for value an a ij of 0.0993 and above ( Fig 2F ).
10.1371/journal.pone.0161828.g002
Fig 2
Evaluation of the confidence in the protein pair-wise measured correlation coefficient.
Fig 2A hierarchical clustering of the protein pair-wise correlation coefficient for a β = 1; in Fig 2B, correlation coefficient evaluated after intensity randomisation for a β = 1. Fig 2C Distribution of the correlation coefficient from direct correlation in turquoise (extracted from Fig 2A) or after intensity randomisation in blue (extracted from Fig 2B). Fig 2D is the ratio false positive hits versus measurements obtained in the dataset. A correlation coefficient of 0.754 and above indicates positive correlation while -0.728 and less for negative correlation corresponds to a ratio of false positive below 5%. In Fig 2E the same measurements as in 2C in the case of a soft threshold β = 10. In Fig 2F, a false positive rate equal to or below 5% corresponds to a value of 0.0993 and above.
The initial analysis was based on the WGCNA package for R [ 7 , 8 ]. Fig 3 shows the Topological Overlap Matrix (TOM) plot applied to the dataset from Deighton et al . [ 12 ]. Each row and column represents proteins. The colours red and yellow indicate the high and low weighted correlation values, respectively, and are assigned by the TOM-based dissimilarity between each protein co-expression level. Each of the five modules, represented by the coloured bar on the top and left of the matrix (blue, yellow, brown, turquoise, and grey), are associated to set of proteins sharing a high value for co-expression level. Only 18 proteins did not cluster into a module and were allocated to the grey module. The turquoise module is the largest one, containing 272 proteins, followed by the blue module with 256 proteins, the brown module with 177 proteins, and the yellow module with 76 proteins (illustrated in Figs 1 and 3 ).
10.1371/journal.pone.0161828.g003
Fig 3
Clustering of the proteomic label-free analysis of glioblastoma multiforme.
The data shows five major clusters. The clustering heatmap was created using a soft thresholding of β = 10 on the entire proteomics dataset. Data clustering and module membership generation are described in Materials and Methods. The scale ranges from yellow to red, with yellow demonstrating low topological overlap and red representing high topological overlap.
Similar to gene expression patterns, proteins within a given module are co-expressed with higher correlation than with proteins from different modules. We then asked if those proteins which are part of the same cluster share some similarities in terms of biological function. To address this, we performed a Gene Ontology (GO) enrichment analysis for each module. The proteins from each cluster were analysed using Gorilla.
Table 1 shows the results of the GO enrichment analysis. Both, the yellow and the grey modules do not show any major functional enrichment. On the other hand, the three other modules clearly show significant GO terminology enrichment. They represent primarily three different components, the blue module being associated with the mitochondrial membrane part (q-value of 2.10e-22), the brown module being associated with the ER (q-value of 7.62e-4) and the turquoise module being associated with the neuronal part/membrane vesicles (q-value of 3.45e-11), suggesting that the proteins from a given cellular location have similar function and display a higher degree of co-expression.
The five main protein modules were then correlated to phenotype data (trait matrix) to highlight possible trends. This step was performed in order to identify possible links between the clusters of proteins and higher-level information. The phenotype information is presented in S2 Table . Fig 4 is a heatmap showing the correlation between the five modules and three different traits. The traits used for correlation were general traits (age and gender of patient taken from Deighton et al . [ 12 ] and state (extracted from S2 Table ). The numbers within the heatmap squares show Pearson correlation coefficients quantifying the correlation between the modules and the phenotype traits. The numbers in brackets are the respective p-values and corrected p-values, respectively.
10.1371/journal.pone.0161828.g004
Fig 4
Data-trait correlation between the first principal component (Eigengene) of each module (y-axis) and the clinical traits (x-axis).
All positive correlations are shown in red and the negative correlations are shown in blue. The correlation coefficients between cells are shown and p-values are displayed within brackets below the correlation coefficient itself. The modules with the lowest and highest significant p-values are the brown, the turquoise, and the blue module.
In the two modules having the lowest functional information (yellow and grey), no significant correlations were found. No significant correlation was found with Age and Gender for any of the modules. The brown module (ER) is anti-correlated to both the blue (mitochondrial membrane) and turquoise modules (membrane vesicles). The brown module shows a high level of correlation with the state (i.e. control = 0, tumour = 1) (r = 0.93). These results indicate that the proteins within the brown module are mostly up-regulated in glioblastoma tumour samples. The turquoise module shows strong anti-correlation with state whereas the blue module shows a similar, but less pronounced anti-correlation with the state.
Validating the protein networks generated from WGCNA
High values of co-expression between two proteins may be predictive of protein- protein interactions. In order to assess the validity of the interactions generated with the presented analysis, the networks that were generated using WGCNA were compared with networks of known interactions obtained from STRING for the same protein dataset. The statistical comparison between the two different approaches was performed using the Network Analysis Tool NeAT (see Materials and Methods section for detailed description). Table 2 shows the results of the comparison between WGCNA and STRING outputs for different threshold values. The Jaccard coefficient was used to determine the similarity between two sample sets.
A combination of different parameter thresholds for both the WGCNA analysis and STRING was tested in order to optimise the best overlap of the two independent approaches to predict protein-protein interactions (illustrated in Table 2 ). The parameter threshold for WGCNA is the minimal threshold Pearson’s correlation coefficient of co-expressed paired proteins. The STRING score is defined as the confidence in the interaction between two protein nodes. Different combinations of parameters have been used and their effect on network overlap (Jaccard coefficient) and prediction quality (shown by p-value) is shown in Table 2 . In addition, we have evaluated the similarity between the STRING output and a randomised pairing (using the same node but having randomised the same number of edges predicted by WGCNA for a given cut-off). The chosen combination of cut-off was based on several factors including minimal p-value and Jaccard score as shown in Table 2 and the highest difference in p-value obtained between the WGCNA and a randomised similar data set against STRING.
The best threshold combination appears to be 0.3 (associated to a FDR of 0.5% and less) for WGCNA and 0.4 for the confidence score generated by STRING (expressed as 0.3/0.4 pair in the text), which gives the higher Jaccard value for the blue and turquoise modules. On the other hand, for the brown module the optimal threshold combination seems to be 0.2 for WGCNA(which is associated to a FDR of 1% and less) and 0.4 for the confidence score in STRING (0.2/0.4 pair). However, more pronounced differences between WGCNA prediction and a random dataset were observed with a cut-off of 0.2/0.4 for the brown and turquoise modules. We have calculated a p-value of 6e-289 for the blue module against STRING, whereas a randomised dataset under the same condition had a p-value of 1.2e-13. In the brown module, we measured a p-value of 6.1e-63, whilst a random dataset generated a p-value of 9.1e-9. In the turquoise module we observed 1.2e-12 whilst a randomised dataset generated a p-value of 3.3e-7. Although low p-values were observed with randomised datasets using STRING, they were largely different from what was predicted with the real dataset. Those highly significant p-values for the random datasets are a consequence of a high ratio of number of edges versus nodes. The higher this ratio, the less of an effect the edge position randomisation has on the predicted network.
In general, the number of interactions predicted by WGCNA was significantly higher compared to what has been reported in STRING. The resulting outcomes are densely interconnected protein networks. In order to reduce the dimensions of the three main large modules identified and extract more subtle information regarding their nature, we used other topological based tools. MCODE [ 9 ], a tool that identifies highly interconnected nodes within a complex network, was used to isolate smaller groups of proteins (which will be referred to as a “cluster”) within each of the three main modules (blue, brown and turquoise module) and identify key highly connected proteins (i.e. Hubs). For each module, a major dense cluster was identified and several minor clusters were also generated (Figs 5 , 6 and 7 ).
10.1371/journal.pone.0161828.g005
Fig 5
Visualisation of the brown module using a network generated in Cytoscape.
The global network is shown in Fig 5A. The main large cluster identified by the MCODE application, being coloured in green and the most interconnected (‘hub’) proteins shown in purple. This main cluster extracted from the brown module is shown in Fig 5B and other secondary clusters identified by MCODE are also shown (Fig 5C, 5E and 5F). In Fig 5B, 5C, 5E and 5F, proteins highlighted in red are associated with a defined GO term assigned by GOrilla. The main cluster as shown in Fig 5B was analysed using a hierarchical clustering approach based on protein intensity across the tumour (Tu) and the control (Ct) samples and is shown in Fig 5D. Five sub-clusters were identified and further analysed using Toppcluster.
10.1371/journal.pone.0161828.g006
Fig 6
Visualisation of the blue module using a network generated in Cytoscape.
The global network is shown in Fig 6A with the main large cluster, identified by the MCODE application, being coloured in green. The main cluster extracted from this module is shown in Fig 6B and other secondary clusters identified by MCODE are also shown in Fig 6D and 6E. In Fig 6B, 6D and 6E, proteins highlighted in red are parts of a defined GO term according to GOrilla. The main cluster as shown in Fig 6B was analysed using a hierarchical clustering approach based on protein intensity across the tumour (Tu) and the control (Ct) samples and is shown in Fig 6C. Four sub-clusters were identified. Distribution of the proteins from the five complexes across the different sub-clusters is shown in Fig 6F.
10.1371/journal.pone.0161828.g007
Fig 7
Visualisation of the turquoise module using a network generated in Cytoscape.
The global network is shown in Fig 7A which contains a large (Fig 7B) and a small (Fig 7C) network. In Fig 7A, the main large cluster, identified by the MCODE application, is coloured in green and the most interconnected (‘hub’) proteins are visualized in purple. The main cluster extracted from this module is shown in Fig 7B and a smaller secondary cluster identified by MCODE is shown in Fig 7D. In Fig 7B, proteins highlighted in red are part of a described biological function according to GOrilla, no specific enrichment has been found for cluster 7D. The main cluster as shown in Fig 7B was analysed using a hierarchical clustering approach based on protein intensity across the tumour (Tu) and the control (Ct) samples and is shown in Fig 7C.
Partitioning the modules; cluster and sub-clusteranalysis
Figs 5 – 7 show network representations of each module. Figs 5A , 6A and 7A are the global networks for the brown, blue and turquoise modules, respectively. Coloured in purple are the highly connected hub proteins for each module (the blue module has no identified protein hub). In the global network, proteins that belong to the first cluster generated by MCODE (as presented in Figs 5B , 6B and 7B , respectively) are coloured in green in Figs 5A , 6A and 7A .
The three main clusters in Figs 5B , 6B and 7B were still densely interconnected, with an overlap of 59 out of 177 proteins for the brown module (33%), 129 out of 256 proteins for the blue module (50%) and 94 out of 272 proteins for the turquoise module (35%). As those main clusters represent an important part of each module, they are mostly an enriched version in terms of function and protein localisation of each module. In addition to those main clusters, several smaller clusters were as well identified and are described below.
Some smaller clusters for the three modules (see Figs 5C, 5E, 5F , 6D and 6E ) showed significant GO term enrichment based on GOrilla (node coloured in red). The main terms describing the most significant enrichment varied in most cases, but were mainly found to be well described by cellular component and biological process GO terms.
In order to identify subtle variation within each main cluster, we applied hierarchical clustering on the protein intensity for each of the main clusters (i.e. the large clusters in Figs 5B , 6B and 7B ), which highlighted some possible sub-clustering. These heatmaps are shown in Fig 5D for the brown module, Fig 6C for the blue module and Fig 7C for the turquoise module. Each of those clusters and sub-clusters were analysed using the comparative tool Toppcluster.
Description of the ER (Brown) Clusters and Sub-clusters
From the initial 177 proteins composing this module, 154 proteins had at least one WGCNA co-expression parameter above 0.2 the threshold value we used for that clusters/sub-clusters. From those 154 proteins, 87 proteins are sub-grouped into four clusters. The 67 proteins that did not associate with any cluster were also not assigned any major biological function. A group of seven proteins were identified as “hub” proteins, i.e. proteins which are highly interconnected (shown in purple in Fig 5A and 5B ). Those proteins are CAT, PDIA6, CALU, SCP2, TMX1, MYH9, and VIM, of which PDIA6 and TMX1 are involved in disulphide isomerase activity and SCP2, CAT and VIM in peroxisome signalling.
In Fig 5B , the proteins highlighted in red are associated with cell compartment GO terminology “ER”. In Fig 5E , the term used to describe the proteins in red was the GO Cellular Component “cell cortex part” (ANK1, SLC2A1, SLC4A1, SPTA1, SPTB) primarily involved with cytoskeletal protein binding, while the subgroup ANK1, SPTA1, SPTB is also related to biological processes associated with the tetrapyrrole and porphyrin-containing compound biosynthetic process. RHD is the only protein not associated with the cell cortex part, but is linked to the plasma membrane along with the other proteins in this cluster.
For Fig 5C , the “cell cortex” and “cortical cytoskeleton” are over-represented in the cellular compartment GO terminology (EZR, FLNA, MAPRE1), the proteins EZR, FLNA, PFN1, TLN1 are involved in maintenance of protein location while the large group of proteins containing EEF2, EIF4A1, EZR, FLNA, HSPB1, KPNB1, MAPRE1, PFN1, RPL4, SERPINH1 share the molecular function “poly(A) RNA binding”. The term used to describe the proteins in red was the GO Function term “nucleic acid binding”. One can notice that two proteins, were not characterised by the prevalent GO term (i.e., protein TAGLN2 and TLN1 both in blue in Fig 5C ). These two proteins are associated with actin binding. However, TAGLN2 is a poorly characterised protein without a determined function.
Fig 5F shows proteins involved in poly(A) RNA binding (APEX1, FUS, HMGB2, HNRNPA2B1, HNRNPA3, PARP1), and most of the proteins found in this cluster are primarily located in the nucleoplasm (Cellular compartment). They are: APEX1, FUS, H2AFY, HMGB2, HNRNPA2B1, HNRNPA3 and PARP1. The term used to describe the proteins in red was the GO Function “DNA binding.”
The main brown cluster illustrated in Fig 5B contains proteins enriched in the ER part, with proteins involved in ER stress. Some specific domain enrichments were found, such as Thioredoxin-like fold (EEF1G, P4HB, PDIA3, PDIA4, PDIA6, PRDX4, TMX1) and ER targets (CALR, HSP90B1, HSPA5, P4HB, PDIA4, PDIA6, PRKCSH). The three main pathways represented in these data are:
mRNA processing (HNRNPA1, HNRNPH1, HNRNPK, HNRNPU, NONO, PTBP1, SFPQ, TMED10)
Protein processing in ER CALR, CANX, CKAP4, DDOST, HSP90B1, HSPA5, P4HB, PDIA3, PDIA4, PDIA6, PRKCSH, RPN1, STT3A
Calnexin/calreticulin cycle (CALR, CANX, PDIA3, PRKCSH)
The main cluster shown in Fig 5B was separated into five sub-clusters, Br1a to Br1e, of 31,6,8,2 and 12 proteins, respectively (shown in Fig 5D ). Mainly the two sub-clusters Br1a and Br1e generate functional information. The ER parts are found in the Br1a and Br1e sub-cluster with CALU, CKAP4, DDOST, PDIA4, PRKCSH, RPN1, STT3A, TMED10, TMX1 for Br1a.
Proteins associated with mRNA processing were found in the sub-cluster Br1a (HNRNPA1, HNRNPH1, HNRNPK, HNRNPU, NONO, PTBP1, SFPQ) and nucleoplasm (HNRNPA1, HNRNPH1, HNRNPK, HNRNPU, LMNB1, NONO, PTBP1, SFPQ, XRCC5, XRCC6). Interestingly, the pair XRCC5 and XRCC6 were identified, which play a major role in the non-homologous end joining (NHEJ) pathway [ 22 ].
Proteins associated to the cytoplasmic membrane-bound vesicles are unique to the Br1e sub-cluster (CALR, CANX, HSP90B1, HSPA5, P4HB, PDIA3 and PPIB). In addition, unique proteins associated to calcium ion binding such as ANXA1, CALR, CANX, HSP90B1, HSPA5 are found in the sub-cluster Br1e which also contains unique proteins involved in response to ER (CALR, HSP90B1, HSPA5, P4HB, PDIA3). Proteins found in this last subgroup (CALR, HSP90B1 and especially HSPA5) are well known to be involved in the activation of signalling protein activity and unfolded protein response (UPR).
Description of the Mitochondrial (Blue) Clusters and Sub-clusters
From the initial 256 proteins composing this module, 214 proteins have a WGCNA co-expression parameter above 0.2. From those 214 proteins, a group of 147 proteins can be sub-divided into three clusters. The 67 proteins that are not part of any major cluster are not significantly co-expressed, however, they did share some biological function such as fibrinogen Complex FGA, FGB, FGG, FN1 and are parts of the Integrin signalling linked to the MAP kinase pathway by recruiting Grb2 to the FADK1/SRC activation complex. In contrast to the majority of protein in this module, this small subset of proteins is up-regulated in the tumour samples.
Fig 6A shows the main module Blue while Fig 6B is associated to the main blue cluster generated by MCODE. The main cluster in Fig 6B is largely composed of proteins involved in the ‘Electron Transport Chain’. The small cluster ( Fig 6E ) (HPX, ORM1, SERPINA1, TF) is associated to the cellular component “extracellular space”. The larger network ( Fig 6D ) has no significant functional enrichment according to GOrilla, although STRING significantly associates (p-value of 2.059e-5) all of its proteins to the extracellular region, except for GLS, CKMT1B and GDAP1L1. This module also contains three proteins having a thioredoxin fold domain (GDAP1L1, PRDX1, and PRDX6). The cluster in Fig 6D contains several proteins which have been associated with a variety of different cancer types including breast cancer which are WDR1, PRDX1, PRDX6 [ 23 ], and HSP90AB1 [ 24 ], hepatocellular carcinoma HSP90AB1, PRDX1, PRDX6 [ 25 ], gastric cancer WDR1, HSPB90AB1[ 26 ], cervical cancer PRDX1, HSP90AB1 [ 27 ], thyroid cancer HSP90AB1, PRDX6 [ 28 ], prostate cancer HSP90AB [ 29 ], and colorectal cancer WDR1 [ 30 ].
The main blue cluster from Fig 6B can be separated into four sub-clusters (shown in Fig 6C ), Bl1a to Bl1d, consisting of 52, 33, 16, and 28 proteins, respectively. The smallest cluster (Bl1c) had little biological information deduced. The blue cluster 1 ( Fig 6B ) is mainly composed of proteins associated to the mitochondrial respiratory chain which, in turn, comprises Complexes I to V. These different complexes are co-expressed slightly differently and are therefore distributed across the four sub-clusters ( Fig 6F ). The Complex I proteins are mainly found in sub-clusters Bl1a (18 proteins out of 26 identified in this study), proteins from Complex III are mainly found in Bl1b (5 out of 7 proteins identified in this study), Complex IV is found across Bl1b and Bl1d while Complex V is distributed between sub-clusters Bl1a and Bl1d. Only two proteins from Complex II were identified (SDHA and SDHB) that were not found to be part of the same sub-cluster.
Description of the Neuronal (Turquoise) Clusters and Sub-clusters
From the initial 272 proteins composing this module, 198 proteins have a WGCNA co-expression parameter above 0.3. From those 198 proteins, a group of 118 proteins is involved in two clusters. The 80 proteins not part of any major clusters although not significantly co-expressed shared some biological function such as fatty acid beta oxidation (ACAA2, ACADS, ACADVL, DECR1, ECI2, HADHA, HADHB), gluconeogenesis (ALDOA, ENO2, PGAM1, SLC25A1, SLC25A13), and glucose metabolism (ALDOA, ENO2, PGAM1, PKM1, SLC25A1, SLC25A13).
The main turquoise module ( Fig 7A ) generated both a large and a small network while using a threshold of 0.3 for the WGCNA coefficient. The large main network ( Fig 7A ) is composed of proteins involved in different “membrane vesicles” structures whilst the small network is mostly related to the myelin sheath (CNP, MBP, PLP1, SIRT2). The overall module containing the “neuronal part” is associated with proteins assigned the terms endocytic vesicles and cytoplasmic membrane-bounded vesicles with some ATPase and GTPase activity; furthermore, a subgroup of proteins is associated to glial cell differentiation (CNP, GAP43, MBP, PLP1, TPPP).
A group of proteins which are highly connected (i.e. Hub proteins; VSNL1, YWHAG, ATP6V1E1, ATP6V0A1, GNAZ, SYT1, DNM1, ATP6V1A, STXBP1) were identified. Three ATPase H+ transporting lysosomal units were found to be quite interconnected and are involved in several different functions (e.g. ATP hydrolysis coupled proton transport and ferric ion transport). Three proteins combined with SYT1, DNM1 and STXBP1 are part of the synaptic vesicle cycle.
In Fig 7C , the main cluster in 7B has been divided into five sub-clusters of 24, 23, 9, 31, and 7 proteins, respectively, (Tu1a to Tu1e). According to Toppcluster, mainly three sub-clusters show biological enrichments which are Tu1a, Tu1b and Tu1d. The sub-cluster Tu1a is rich in proteins involved in ion/cation transport (ANK2, ATP1B1, ATP6V0A1, ATP6V1A, ATP6V1B2, CAMK2A, CNTN1, NSF, SNAP25, STX1A, STX1B, SYT1, THY1, and YWHAZ). In addition, proteins from cluster Tu1a have molecular functions associated to SNARE binding (NSF, SNAP25, STX1A, STX1B, and SYT1).
A group of three proteins from Tu1a is involved in regulation of mitochondrial membrane permeability (CAMK2A, YWHAG, and YWHAZ). The sub-cluster Tu1b is rich in proteins involved in pathways associated to coated vesicle membrane and clathrin-coated vesicle (AP2A1, AP2M1, DNAJC5, SNAP91, and VAMP2), while Tu1d is mainly composed of proteins involved in synaptic vesicle endocytosis and synaptic vesicle recycling (AMPH, RAB3A, SH3GL2, SNCA, SYNJ1, and SYP).
The data presented in Fig 7D showed no strong functional enrichment after being analysed by GOrilla, although according to Toppcluster and STRING the following proteins are associated to protein targeting to ER as biological process: RPL18, RPL7A, RPLP0, and RPN2. Additionally, STRING identified several proteins as parts of membrane-bound vesicle from this cluster (CAMKV, CAMK2G, PALM, RPLP0, RAP2A, RPL7A, GSTK1, TUBB2A, AP2A2, DPYSL2, FTH1, PFKP, DPP6, and AK2).
Pyruvate kinase isoforms co-expression network
The two isoforms of PKM (PKM1 and PKM2) were identified ( Fig 8 ). While PKM1 was found to be down-regulated and associated to the turquoise module, the PKM2 isoform was up-regulated and associated to the brown module. The direct co-expressed proteins for each pyruvate kinase protein isoform are illustrated in Fig 8 . Twenty-nine proteins were found to be co-expressed with PKM1 while only three showed co-expression with PKM2 in this study.
10.1371/journal.pone.0161828.g008
Fig 8
Pyruvate kinase isoform M1 (Left) and M2 (Right) and their respective co-expression networks (direct interactors only). Nodes in dark blue are the 2 PKM proteins, in pink are proteins defined as HUB proteins from the turquoise Module. The proteins in green are proteins associated to the larger sub-cluster presented in Fig 7 .
Regarding PKM1, the largest group of proteins exhibiting direct co-expression are those related to the synaptic vesicle cycle (KEGG Pathway): ATP6V0A1, ATP6V0D1, ATP6V1A, ATP6V1B2, ATP6V1E1, CPLX2, DNM1, NSF, RAB3A, SNAP25, STXBP1, SYT1, and VAMP2.
Several other metabolite-associated groups of proteins were identified, such as proteins related to cellular response to insulin stimulus (YWHAG, VAMP2, ATP6V0A1, ATP6V0D1, ATP6V1A, ATP6V1B2, ATP6V1E1, and GOT1), glutamate secretion (VAMP2, SYT1, SNAP25, STXBP1, and RAB3A), and to syntaxin binding (CPLX2 NAPB NSF SNAP25 STXBP1, and VAMP2).Some of the highly correlated expression profile proteins with PKM1 include guanine nucleotide binding protein (GNAO1) and syntaxin binding protein 1 (STXBP1) for which no known direct interaction has been reported yet. In a similar manner, the protein cell adhesion molecule 3, CADM3 involve in the calcium-independent cell-cell adhesion molecules is as well highly correlated with PKM1 but no known relation between CADM3 and PKM1 has been previously reported. Both cases merit to be explored by studying the role of both PKM in specific tissues, in this cases, PKM1 role in the synaptic vesicle. For the PKM2 cluster, no significant term enrichment was found.
Discussion
In the present study, we used a combination of different analytical methods to characterise protein co-expression measured from a quantitative proteomics analysis. The main method (WGCNA) was applied to the Deighton et al dataset [ 12 ] and allowed for the subgrouping of all proteins into five main modules, of these modules, three are associated with membrane-based organelles. The soft threshold power used in this study (β = 10) is in the same order of magnitude as that used in previous work [ 31 ]. This approach has resulted in identification and sub-grouping of proteins by their distinct features into three different cellular locations.
Although the dataset is of a modest size (a total of 12 experiments), we have shown that it was possible to extract valid and meaningful information. We evaluated the FDR for different correlation coefficient thresholds using the same dataset but with the position of each intensity for a given protein being randomised. The threshold values selected to generate the different networks in Figs 5 – 7 have a FDR of between 0.5% and 1% which is quite conservative. The 3 major modules clearly show significant enrichments thus supporting the validity of the approach even on small datasets. The networks generated using WGCNA have been compared to the knowledge-based method STRING and shows the overlap between the two independent methods to be significant. Thus, we have shown that the use of WGCNA to generate protein networks de novo without the need for an immunoprecipitation-based approach. These networks could not have been generated with the initial type of analysis used in Deighton et al [ 12 ] The over-represented GO term to describe functional enrichment was mainly the cellular component with the blue module’s proteins being significantly localised in the mitochondria. The brown module was enriched in ER proteins and the turquoise module enriched in various types of vesicle membranes. In addition, these abundance of the proteins in these modules correlated to traits which included the relative increase, or decrease of expression in cancer tissue ( Fig 4 ). The brown module (ER) proteins correlate with proteins that are up-regulated in tumour samples, while the blue module (mitochondrial membrane part) and turquoise module (membrane vesicles) both correlate with proteins that are down- regulated in tumour samples.
The original proteomic analysis reported by Deighton et al . [ 12 ] was based on a mitochondrial fraction enrichment. However, in this current study we have identified several proteins from other membrane-based organelles such as the ER and vesicular membranes. Although these membrane fractions share similar physical properties to the mitochondrial fractions and could introduce complexity to the samples, their identified interaction networks reveal the broader of the many effects of glioblastoma multiforme. In addition, the concomitant enrichment of ER in the mitochondrial fraction might be a result of those two organelles being interconnected through mitochondria-associated membranes (MAM) [ 32 ] a finding that may provide a deeper understanding of intra-cellular organelle coordination during tumorigenesis.
Despite the use of a soft threshold β = 10 to generate the different networks, these networks were significantly denser than what was predicted by STRING. Although the overlap between STRING and WGCNA was found to range between 5 and 22%, the calculated p-values clearly support that the observed networks were not simply due to chance (p-values between 1.2e-12 to 6e-289).
One observation, also reported in Deighton et al . [ 12 ] is that the electron transfer chain (ETC) is significantly down-regulated in cancer cells (part of the blue module). Proteins from the major complexes of the ETC were identified in this study and were found to be mostly down-regulated. This observation was supported by electron microscopy showing that the inner membrane of the mitochondria is severely disrupted [ 12 ]. However, in this manuscript we have found that the different complexes were marginally co-expressed in different sub-clusters especially for Complex I (70% of Complex I proteins were found in sub-cluster Bl1a) and Complex III (70% of Complex III proteins found in sub-cluster Bl1b) which suggests that these two complexes are not affected in the same way, with Complex I proteins being slightly more down-regulated than the proteins from Complex III. A similar observation on the different effects on complexes of the ETC has been made on mitochondrial fractions isolated from a transgenic mouse model [ 33 ]. A disruption of the electron transfer chain and oxidative phosphorylation could potentially lead to elevated ROS generation [ 34 ]. Several proteins involved in the oxidative damage response were also found to be up-regulated such as catalase, superoxide dismutase 2, peroxiredoxin 1, 4 and 6.
Several key proteins involved in the “ER stress response” or the “unfolded protein response” (UPR) were found to be up-regulated. The disruption of the ETC and the up-regulation of several proteins involved in oxidative stress support a link with cellular events such as protein oxidation and protein folding. Oxidative stress and ROS generation are important components of the ER stress response. The major enzymatic components of ROS production during UPR induction are protein disulfide isomerase (PDIA4 was found up-regulated in this study); ER proteins involved in stress response were found significantly co-expressed (CALR, HSP90B1, HSPA5, P4HB, and PDIA3) specifically in the sub-cluster Br1e. Most of these proteins were also found up-regulated during oxygen and glucose deprivation for 18h [ 35 ] which supports an integrated cellular survival response. Furthermore, mitochondrial HSP90 has been reported to play an important role in controlling core metabolic processes by stabilising Complex II of the ETC and allowing cellular respiration to continue under compromised conditions, contributing to tumorigenesis [ 36 ].
Cells under normal conditions have a basal level of ROS, which is intrinsic to signalling mechanisms. However, an increase of ROS levels is observed upon exposure to specific stress such as cytotoxic reagents, irradiation, and environmental pollutants and during some specific enzymatic reactions such as: mitochondrial respiratory chain reactions, activity of glucose oxidase, amino acid oxidase, xanthine oxidase, and NADP/NADPH oxidase). Triggering of the unfolded protein response (UPR) consequential to the exposure to oxidative stress is most likely a mechanism to preserve both cell function and survival. On the other hand, continuous oxidative stress and protein misfolding induce apoptotic pathways and play crucial roles in the pathogenesis of multiple human diseases including diabetes, atherosclerosis, and neurodegenerative diseases.
HSPA5 (also known as GRP78, Bip) is a chaperone protein whose expression is significantly enhanced under various conditions including glucose deprivation, oxidative stress, treatment with Ca2+ ionophores, and hypoxia [ 37 ]. Higher levels of HSPA5 are essential for sustaining cell viability under specific kinds of stress. The up-regulation of stress proteins in tumour cells has been shown to inhibit programmed cell death and to contribute to drug resistance [ 37 ]. Therefore, HSPA5 has some potential as a novel therapeutic target for both anti-tumor and anti-angiogenesis activity [ 38 ].
Similar to the blue module, the turquoise module is mainly composed of proteins which are down-regulated under tumour-forming conditions and are mainly enriched in “vesicle membrane” fractions. The main cluster in Fig 7B contains most of the proteins having known biological functions. Surprisingly, the YWHAZ protein was found to be down-regulated in our study, whilst Nishimura et al. [ 39 ] observed that YWHAZ-overexpression plays a major role in tumour cell proliferation. One of the highly interconnected protein members of the hub proteins was YWHAG, which is a 14-3-3 adapter protein involved in the regulation of a broad spectrum of signalling pathways. YWHAG binds to a large number of partners, usually by recognition of a phosphoserine or phosphothreonine motif. Binding generally results in the modulation of the activity of the binding partner by protein kinase C inhibitor activity. A protein kinase C (PRKG) was also found co-expressed in the turquoise module. Both, 14-3-3 protein YWHAG and YWHAZ in combination with CAMK2A were found in the same sub-cluster Tu1a and are involved in the regulation of mitochondrial membrane permeability.
The soft threshold method used in this study (β = 10) significantly reduced the importance of module interconnection. However, a few interesting proteins were identified in the ER which are more strongly co-expressed with proteins in the mitochondria including PDIA6 and HSPA5/GRP78 which are known to play a crucial role on apoptosis inhibition [ 38 , 40 ]. Regarding the blue module, a few proteins were found to be highly co-expressed with other proteins outside the module suggesting a co-ordination role far beyond their immediate environment. One of the identified proteins is CKMT1B, which was also found to be down-regulated in squamous cell carcinomas and in clinical samples [ 41 ].
A component of the turquoise module is the isoform 1 of pyruvate kinase (PKM1), which was highly co-expressed with more proteins than its counterpart PKM2 from the brown module ( Fig 8 ). It has often been described in the literature that the PKM protein expression switches from PKM1 to the PKM2 isoform during tumourigenesis [ 42 , 43 ]. We observed a change in isoform ratio where the PKM1 isoform is down-regulated with a ratio tumour/control = 0.26 associated to the turquoise module. While the PKM2 isoform is up-regulated (ratio tumour/control = 2.14 and clustered in the brown module). The observed changes in this current study, although meaningful, do not support a complete shift from one isoform to the other one as described by Bluemlein et al .[ 44 ]. The two isoforms of PKM are differentially expressed (M1 and M2) with the different co-expression network proteins of each isoform supporting an increase in aerobic glycoysis at the expense of oxidative phosphorylation (rendered inefficient due to the disruption of the ETC).
Decreasing the PKM2/PKM1 ratio has recently been described as a therapeutic strategy in patients with glioblastoma multiforme [ 45 ]. As shown in Fig 8 , co-expression of PKM2 was limited to only three other proteins (ANXA5, PFN1, and RPS11). Conversely, PKM1 was found co-expressed with more than 30 other proteins from the turquoise module which are mostly involved in ion transport, cellular response to insulin stimulus, glutamate secretion as well as syntaxin binding; a common theme among these proteins is related to the synaptic vesicle cycle with 12 out of the 32 proteins being directly involved in this pathway. Although a broad range of functions is associated to the different proteins co-expressed with PKM1, our findings support that pyruvate kinases are possibly bound to synaptic vesicles with substrates that may be supporting vesicular glutamate uptake [ 46 ]. In addition, several of the highly PKM1 co-expressed proteins reported in this study were newly identified. Guanine nucleotide binding protein (GNAO1) and syntaxin binding protein 1 (STXBP1) and the protein cell adhesion molecule 3, CADM3 involved in the calcium-independent cell-cell adhesion molecules has not been previously reported and merit to be explored by more tissue targeted analysis of both PKM.
It is intriguing that the 2 PKM isoforms show expression patterns which are not co-localised; PKM2 found mostly co-expressed with proteins from the mitochondrial fraction while PKM1 found co-expressed with proteins related with vesicular membrane. In summary, protein co-expression analysis of the mitochondrial protein fraction revealed novel protein networks with several intrinsically linked functions and uncovered functional modulesHere we have shown and validated with several different strategies that a weighted protein co-expression analysis complements more conventional approaches based on differentially expressed proteins from different groups and can serve as a valuable method for revealing new trends and information clustering which are impossible to capture otherwise.
Supporting Information
S1 Fig
Soft threshold parameters and resulting scale-free topology.
(DOC)
S1 Table
Complete proteomics dataset used in this study (proteins identified with at least 2+ peptides).
(XLS)
S2 Table
Table of the trait matrix used for the correlation with the modules.
(XLS)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Chronic lung function decline, punctuated by recurrent airway exacerbations is characteristic of CF pulmonary disease and contributes substantially to patient morbidity and mortality [1] . Several culture-independent studies examining bacterial diversity in a variety of human niches have demonstrated that shifts in bacterial community composition underlie states of host health or disease [2] , [3] , [4] , [5] , [6] , [7] , [8] . The advent of high-throughput technologies to examine microbial community composition permits high-resolution profiling of organisms that are beneficial or detrimental to the health of the human super-organism [9] . Recent studies focusing on CF have demonstrated that bacterial assemblages, that include a number of pathogens, exist in both adult and pediatric airways [8] , [10] , [11] , [12] , [13] , [14] , [15] , [16] . These assemblages undoubtedly contribute to the long-term lung function decline experienced by CF patients, but the depth of bacterial diversity and the relationship between airway microbiota and chronic pulmonary disease observed in patients from birth through adulthood, is currently unknown.
To perform an in-depth characterization of age-based bacterial community composition of CF airways, independent deep-throat swab and expectorated sputum samples from 51 pediatric and adult patients respectively, were collected at the pediatric and adult CF clinics at University of California, San Francisco. Analysis was performed using the 16S rRNA PhyloChip, a high-density, phylogenetic microarray, capable of identifying approximately 8,500 bacterial taxa in parallel [17] , [18] , [19] .
Materials and Methods
Ethics Statement
The Committee on Human Research at University of California San Francisco (UCSF) approved all study protocols, and all patients or surrogates provided written, informed consent.
Sample collection
Airway samples from clinically stable adult and pediatric CF patients, defined as having no change in pulmonary function for ≥2 months prior to sample collection were used for this study. Adult sputum samples (n = 37) were collected following routine spirometry at the UCSF Adult Cystic Fibrosis Centre clinic. Samples were mixed with approximately twice the volume of RNAlater (Ambion) prior to nucleic acid extraction within 4 hours of collection. Pediatric deep-throat swab samples (n = 26) were collected at the UCSF Pediatric Cystic Fibrosis Centre clinic from all patients (aged up to 18 years), submerged in a sterile tube containing 3 ml of RNALater prior to nucleic extraction within 4 hours of collection. Deep-throat swabs were used as an alternative to sputum since the majority of pediatric patients, particularly the younger individuals are unable to produce sputum. The primary dataset for this study comprised single, independent samples from 51 patients. To determine the temporal changes in bacterial diversity within patients of different age groups, a subset of 13 of these patients from whom we had also collected a subsequent stable sample were analyzed. Patient demographics are presented in Table S4 .
Sample processing
Optimized DNA extraction protocols developed in-house using CF sputum samples and representative Gram positive and Gram negative bacteria were used for this study. Adult samples were centrifuged for 10 minutes at 12,000 g and RNALater removed prior to total genomic DNA extraction from 300 µl of sputum (containing sputum plugs) using the Wizard Genomic DNA extraction kit according to the manufacturer's instructions (Promega, CA). Deep-throat swabs were added to lysis buffer from the AllPrep DNA/RNA extraction kit (Qiagen, CA) in a lysing matrix B tube (Qbiogene, CA) and lysed by bead-beating using a FastPrep system (Qbiogene, CA) for 30 seconds at 5.5 m sec −1 . Supernatant was then transferred to the DNA column of the AllPrep kit and nucleic acids extracted according to the manufacturer's instructions. Universal primers 27F ( 5′-AGAGTTTGAT CCTGGCTCAG-3′ ) and 1492R ( 5′-GGTTACCTTGTTACGACTT-3′ ) [20] were used to amplify the 16S rRNA gene using 12 PCR reactions per sample run across a gradient of annealing temperatures (48–58°C) to maximize diversity recovered. PCR reactions contained 0.02 Uµl −1 Takara Ex Taq DNA Polymerase (Takara Bio Inc Japan), 1× Takara buffer, 0.8 mM Takara dNTP mixture, 0.4 mg ml −1 bovine serum albumin (BSA) and 1.0 µM of each primer. PCR conditions were 1 cycle of 3 min at 95°C followed by 25 cycles of 95°C for 30 s, the gradient annealing temperature for 30 s, 72°C for 2 min and a final extension at 72°C for 10 min. A total of 100 ng of extracted DNA from adult sputum or pediatric deep-throat swab samples was used per PCR reaction. Amplified products from all 12 annealing temperatures were pooled, gel-purified and processed for PhyloChip analysis as previously reported [17] , except that 250 ng of each amplicon was hybridized. As a negative control, nucleic acid was extracted from sterile swabs and assayed for 16S rRNA product as described above. No PCR product was detected.
Microarray analysis
A two-step normalization procedure was adopted to correct the fluorescence intensities for amplicon target quantification variation and in microarray technical variation. For analysis of bacterial community composition we used the PhyloChip, a 16S rRNA custom microarray developed at Lawrence Berkeley National Laboratory and synthesized by Affymetrix Inc. (Santa Clara, CA, USA). The array has 506,944 probes arranged in 712 rows and columns representing approximately 8,500 bacterial taxa, with the capability of detecting bacteria comprising at least 0.01% of a population [21] . Each chip has additional probes that serve as the following controls: (1) prokaryotic and eukaryotic metabolic genes (added prior to fragmentation) to control for variations in fragmentation, biotinylation, hybridization, washing, staining, and scanning and (2) pre-labeled oligonucleotides added to the hybridization mix to account for variation in hybridization, washing, staining, and scanning. In addition, to minimize cross-hybridization, at least eleven probe pairs (positive match and mis-match oligonucleotides) are used to interrogate each taxon at at least two discriminatory loci on the 16S rRNA gene. Combined PCRs and control amplicons were fragmented to 50–200 bp using DNase I (0.02 U mg −1 DNA; Invitrogen, USA) and One-Phor-All buffer (NJ, USA). Biotin labeling was performed using terminal deoxynucleotide transferase and ddUTP as per the manufacturer's instructions. The reactions were denatured at 99°C for 5 min and hybridized to the PhyloChip for 16 hours at 48°C at 60 rpm. The arrays were subsequently washed, stained and scanned as in [22] . Scanning of the arrays was performed using a GeneArray Scanner (Affymetrix, CA, USA) and the intensity of all the probes was treated as previously reported [17] , [19] . Positive probe pairs met two criteria: (1) the fluorescence of the perfectly matched probe was at least 1.3 times greater than the intensity of the control (mismatch probe); (2) the value of the difference between perfectly matched probe and mismatch probe intensities was at least 130 times greater than the squared noise value. The value of the positive fraction (pf) was calculated for each probe set as the number of positive probe pairs divided by the total number of probe pairs in a probe set.
Data sets were conservatively filtered, with taxa determined as present if pf ≥0.9 (90% of probes in a probe set for an individual taxon) were positive. Changes in probe-set fluorescence intensity are equivalent to changes in taxon relative abundance between samples. Fluorescence intensity for every taxon determined to be present in at least one sample was log transformed prior to analysis using packages in the R statistical environment [23] . FastUnifrac (Hamady et al , in review [24] ) with branch length weighting by taxon relative abundance (normalized log-transformed fluorescence intensity) was used to generate a phylogenetic distance matrix from the PhyloChip data.
Quantitative PCR Validation
Quantitative PCR (Q-PCR) was performed on a subset of samples with sufficient remaining DNA to validate the presence and relative abundance of Pseudomonas spp., Staphylococcus spp., and Haemophilus spp. using primers designed on the Greengenes 16S rRNA gene alignment [22] using Arb [25] ( Pseudomonas : 9056aF 5′-CCGCATACGTCCTGAGGGA GAAAGT-3′ , 9056aR 5′-TCTCAGACCAGTTACGGATCGTCGC-3′ ; Staphylococcus : SaurF2 5′-AACCCTTAAGCTTAGTTGCCATC-3′ , SaurR2 5′-TTGACCTCGCGGTTTC GCTG-3′ ; Haemophilus : HinF 5′-AATGGCGTATACAGAGGGAAG-3′ , HinR 5′- CAATCCGGACTTAGACGTACT-3′ ). A total of 20 ng of DNA per reaction was used in triplicate, 25 µl Q-PCR reactions at an annealing temperature of 56°C with the Quantitect SYBR Green QPCR kit (Qiagen, MD) according to the manufacturer's instructions. Pearson's correlation analysis of inverse cycle threshold values against array fluorescence intensities was used to confirm relative abundance of the three organisms and concordance between the two independent molecular methods.
In addition to Q-PCR validation, to confirm that specific members of the community were truly present, a pair of primers designed to detect Veillonella parvula (based on PhyloChip probes; VDF5′- CGTAATCAA CCTGCCCTTCAGAGG -3′ , VDR 5′-TTTCTGGCTTCC GAAGAAGAGGAAC-3′ ) was used to amplify a PCR product which was then submitted for bi-directional sequencing. Sequence identity was assigned by comparing reads to GenBank using NCBI's BLAST ( http://blast.ncbi.nlm.nih.gov/ ).
Data analysis
From the FastUnifrac distance matrix the amount of variance that each clinical variable e.g. sample collection method, age etc. contributed to the dataset was calculated using the adonis function of the vegan package in R [26] . Nearest-taxon index (NTI) and Net-relatedness index (NRI) [27] , [28] measures of the phylogenetic-relatedness of communities, were calculated using the picante package in R [29] . NRI, is a standardized measure of the mean pairwise phylogenetic distance of taxa in a sample, relative to a phylogeny of an appropriate species pool, and quantifies overall clustering of taxa on a tree. NTI, is a standardized measure of the phylogenetic distance to the nearest taxon for each taxon in the sample and quantifies the extent of terminal clustering, independent of deep level clustering. A neighbor-joining with nearest-neighbor interchange phylogenetic tree of representative 16S rRNA sequences downloaded, pre-aligned from the Greengenes database [22] was constructed using FastTree [30] . FastTree is a neighbor-joining with nearest neighbor interchange phylogenetic inference method using alignment profiles rather than distance matrices. The topology of an initial neighbor-joining tree is improved by rounds of nearest neighbor interchanges, which recompute the profiles of each internal node. Splits in the tree that result from this process are then tested by local bootstrapping to see how well they are supported. This approach has the advantage over alternative methods, of speed whilst approximating the accuracy of character-based methods such as maximum-likelihood, Bayesian analysis, or parsimony. This was used, together with taxon richness, to calculate the mean phylogenetic distance (MPD) and mean nearest phylogenetic taxon distance (MNTD) using the phylogeny shuffle null model for each sample. MPD and MNTD values were used, as previously described [27] , to calculate NRI and NTI values respectively for each sample. Inverse Simpson's diversity index [31] , Pielou's evenness [32] and community richness were calculated using the vegan package in R. To accurately reflect community richness for this calculation, individual taxa were deemed to have an abundance of 0 if they did not meet the pf ≥0.9 criterion. Temporal change in community diversity was calculated by subtracting the calculated diversity index of the second sample from the initial sample collected from individual patients and normalized by dividing the result by the number of days between sample collections to provide the change in diversity over time.
Linear regression analysis was performed using calculated metrics of community composition and patient variables e.g. age, CFTR mutation etc. in Statplus (Analystsoft, Vancouver, Canada). Correlation analysis of each individual taxon abundance against patient age or CFTR mutation was performed using the multtest package [33] available as part of the Bioconductor suite of analysis programs [34] . P-values were adjusted for false discovery using the Benjamini-Hochberg procedure [35] or q-value method [36] where applicable. A phylogenetic tree of taxa that significantly positively or negatively correlated with age was constructed using FastTree as described above. Samples were assigned to age bins of 5 years (all bins contained ≥3 samples) and taxon fluorescence values in each bin were averaged and scaled relative to the highest value in the overall dataset. The tree illustrating these changes was imported to the Interactive Tree of Life ( http://itol.embl.de/ ; [37] ) and annotated. Taxa that were identified by Bellerophon [38] running on the Greengenes database as chimeric or those taxa represented by sequences of less than 600bp were not reported.
Results
Relationship between Pulmonary Function and Airway Bacterial Diversity
Sample type (deep throat swab or sputum) explained only 7% of community variance (p = 0.001), with age group explaining 33% (p = 0.001), suggesting that the method of sample collection did not substantially impact community compositional differences, although differences in the site of sample collection likely explains this low level of variance observed. Between 78 and 1012 taxa were detected in the samples and 1,837 different taxa (defined as species or strains that share≥97% 16S rRNA sequence identity) were detected in the entire cohort, which represents substantially greater bacterial richness than previously reported using other culture-independent methods [10] , [11] . This is likely due to the ability of the PhyloChip to detect members that represent as little as 0.01% of the community in parallel with highly abundant taxa [18] . This aspect of array analysis is particularly advantageous when analyzing CF airway samples that tend to be dominated by a small number of highly abundant species [10] . However due to the potential for cross-hybridization at the taxon-level, the majority of our subsequent analyses was performed at higher levels of classification. These taxa belonged to 43 phyla with the majority belonging to the phyla Actinobacteria, Bacteroidetes, Firmicutes and Proteobacteria with smaller numbers belonging to the Acidobacteria, Planctomycetes and Spirochaetes (a complete list of all taxa detected is provided in Supplementary Table S1 ). Known CF pathogens, routinely cultured from CF sputum, were detected in the samples. S. aureus could be detected in 65% of samples (61% of adult samples, 72% of pediatric) and P. aeruginosa in 73% of samples (91% of adult samples, 39% of pediatric). A subset of 14 adult CF samples was used to examine the abundance of three organisms by Q-PCR. Correlation analysis of Q-PCR data with PhyloChip fluorescence intensity demonstrated good concordance for P. aeruginosa (otu_9056; r = 0.76; p≤0.001), S. aureus (otu_3258; r = 0.86; p≤0.0001) and H. influenzae (otu_8555; r = 0.56; p = 0.002). Although multiple safeguards are in place to minimize cross-hybridization, the PhyloChip, like all other culture-independent profiling technologies is susceptible to false positives. Therefore, targeted PCR amplification using primers designed on probes for OTU 940, whose representative species is Veillonella parvula was used to amplify a PCR product which was then sequenced. Sequence analysis demonstrated that the sequence exhibited 99% homology with Veillonella parvula across 631 nt.
To determine whether a linear relationship existed between patient age and pulmonary function (forced expiration volume in one second [FEV 1 % predicted]), we tested for correlation between these variables, for samples where FEV 1 measurements were available (n = 44). Despite the absence of standardized airway function testing for the very young pediatric patients (age ≤5, n = 7) the data confirmed that older patients CF patients exhibited significantly lower pulmonary function (r = −0.48; p<0.0003; Fig. 1A ). To determine whether age-related pulmonary function was associated with aspects of the airway bacterial community structure or composition, we tested for correlations between patient age (n = 51) and taxonomic richness (number of bacterial taxa detected), community evenness (relative distribution of taxa within communities), taxonomic diversity (metric based on both richness and evenness) and phylogenetic relatedness; Net Relatedness Index (NRI) and Nearest Taxon Index (NTI), two indices that measure the degree of phylogenetic clustering or dispersion within communities [39] . These analyses revealed that older CF patients possessed airway bacterial communities that were less rich (r = −0.35, p<0.001), less even (r = −0.38, p<0.0006) and had significantly lower bacterial diversity (r = −0.33, p<0.004; Table 1 ). In addition, bacteria in the airway communities of older patients were more related to each other (NTI, r = 0.47, p<0.0001; NRI, r = 0.35, p<0.0001; Table 1 ; Fig. 1B and C ) with the NTI and NRI indicating significant phylogenetic clustering of the taxa in these samples at both the terminal branches (taxon level) and at higher phylogenetic levels (family, class etc.). These data demonstrate that compared to younger CF patients, older individuals exhibit reduced airway bacterial diversity and assemblages of more closely-related organisms which occurs in parallel with age-related pulmonary function decline, implicating this structural shift in community composition in the pathophysiology of late stage CF airway disease. This has not previously been demonstrated in an age-stratified cohort spanning neonates to adults.
10.1371/journal.pone.0011044.g001 Figure 1
Patient age is related to pulmonary function and aspects of community relatedness.
Boxplots with age bins containing three or more samples are displayed; Relationships between patient age and (A) pulmonary function (B) Net Relatedness Index and (C) Nearest Taxon Index are illustrated. As pulmonary function declines, increasing phylogenetic relatedness at both the deeper levels and terminal branches of the phylogenetic tree respectively is evident.
10.1371/journal.pone.0011044.t001 Table 1
Relationship between CF patient age and metrics of bacterial community composition.
Analysis
All clinically stable
Homo-ΔF508
Hetero-ΔF508
Non-ΔF508
Age vs Richness
−0.35 a p b <0.001
0.53 p<0.02
0.007 p<0.98
0.36 p<0.25
Age vs Evenness
−0.38 p<0.0006
0.08 p<0.75
0.32 p<0.12
0.35 p<0.27
Age vs Diversity
−0.33 p<0.004
0.53 p<0.02
0.0005 p<0.99
0.35 p<0.27
Age vs NRI
0.47 p<0.0001
0.37 p<0.13
0.63 p<0.005
0.62 p<0.03
Age vs NTI
0.52 p<0.0001
0.56 p<0.01
0.26 p<0.31
0.74 p<0.006
a R values. b P values <0.05 are considered significant.
Airway Bacterial Diversity and CFTR Genotype
Our cohort consisted of a variety of CFTR genotypes, permitting the opportunity to examine the relationship between mutation and community composition. Patients were classified in 3 groups: homozygous-ΔF508 (n = 19), heterozygous-ΔF508 (n = 19) or non-ΔF508 mutation (n = 13). To determine which aspects of community composition were most related to mutation-specific loss of pulmonary function, regression analyses against community metrics and bacterial taxon abundance were performed. Significant negative correlations between patient age and bacterial diversity, richness and NTI ( Table 1 ) were identified for the homozygous-ΔF508 group. Compared with younger patients, community restructuring in older homozygous-ΔF508 patients was primarily due to loss of multiple members of the Mycobacteriaceae, Staphylococcaceae, Enterobacteriaceae, Acholeplasmataceae, Pasteurellaceae and Chlamydiaceae; older individuals were primarily associated with higher abundance of members of the Pseudomonadaceae (Supplementary Table S2 ). Compared to younger heterozygous-ΔF508 and non-ΔF508 patients, older individuals with these genotypes did not exhibit significant reductions in taxonomic richness, evenness or diversity” ( Table 1 ); however, for both groups, these patients did exhibit increased NTI and NRI of their airway communities ( Table 1 ). Similar to the homozygous patients, the older heterozygous-ΔF508 patients also exhibited increased abundance of members of the Pseudomonadaceae and Xanthomonadaceae; in addition, they also exhibited significant increases in the Moraxellaceae, and Sphingomonadaceae. In contrast to ΔF508 patients, non-ΔF508 subjects did not exhibit an age-based positive correlation with the Pseudomonadaceae or with any other bacterial family. Instead younger patients were strongly associated with multiple members of the Enterobacteriaceae, Campylobacteraceae and Helicobacteraceae (Supplementary Table S2 ). These families appear to be specifically associated with the non-ΔF508 group, as they did not feature prominently in heterozygous-ΔF508 or homozygous-ΔF508 airways.
Temporal Diversity Changes
The temporal change in bacterial diversity of a subset of 13 patients ranging from 9 months to 43 years for whom a second, subsequent clinically stable airway sample was available was examined. All pediatric patients examined up to 11 years of age (n = 10) exhibited a net increase in bacterial diversity over time. In contrast a net decrease in community diversity was observed for older patients (n = 3; Fig. S1 ), suggesting that CF patients exhibit an initial expansion of airway diversity (colonization and diversification period) over the first decade of life, after which a progressive decline in diversity and development of increasingly phylogenetically-related consortia occurs (establishment of competitively dominant species). However, while provocative, given the small number of patients and samples analyzed for this section of the study, this data should be interpreted cautiously. An expanded study with substantially more temporal samples coupled with extensive medical histories for these patients is necessary to confirm that this observation holds true in larger populations over more protracted time-frames and if so, to determine the factors that drive this phenomenon e.g. frequency of antimicrobial administration.
These shifts in community structure and diversity observed in older CF patients may contribute to the phenomenon of age-specific airway pathogen abundance in CF patients [40] , [41] . To examine these relationships in more detail, correlations between the abundance of all known CF airway pathogens and patient age were determined to identify CF pulmonary disease pathogens associated with pediatric and adult CF patients respectively. Of the known CF pathogens, only Haemophilus influenzae (p = 0.02, r = −0.31), Stenotrophomonas maltophilia (p = 0.02, r = 0.31) and Pseudomonas aeruginosa (p = 0.001, r = 0.42) exhibited significant correlations with age ( Table 2 ). P. aeruginosa and S. maltophilia abundance was greatest in older CF patients ( Fig. 2 ), who exhibited lower bacterial diversity. In contrast, H. influenzae exhibited the opposite trend with a peak in abundance in younger patients when community diversity was greatest ( Fig. 2 ), observations, which are supported by previous culture-based reports [41] , [42] , [43] .
10.1371/journal.pone.0011044.g002 Figure 2
Phylogenetic tree displaying relationship between patient age and taxon abundance.
Taxa exhibiting a significant increase (red) or decrease (blue) in relative abundance with increasing CF patient age are illustrated. Scale bar indicates 0.01 nucleotide substitutions per base.
10.1371/journal.pone.0011044.t002 Table 2
Correlation between pathogen abundance and increasing CF patient age.
Pathogen
R
p-value
Haemophilus influenzae
−0.31
0.02
Pseudomonas aeruginosa
0.42
0.001
Stenotrophomonas maltophilia (Xanthomonas axonopodis)
0.31
0.02
Shaded pathogens exhibit a significant correlation.
An advantage of the array is the ability to detect taxa not normally isolated by conventional culture. Thus the PhyloChip data was analyzed to identify other bacterial species that exhibited age-related changes in abundance, in an attempt to more broadly define those organisms associated with pediatric and adult CF airways. Following false discovery adjustment, a total of 113 taxa exhibited a significant correlation with age; 45 were positive correlations, 68 negative (a complete list is provided in Supplementary Table S3 ). Of the 45 taxa significantly positively correlated with age, almost half (22 taxa) were members of the Pseudomonadaceae ( Fig. 2 ). This confirmed our previous observation that older CF patients possess airway communities that are less diverse and more phylogenetically related. Whether this phenomenon is a result of the Pseudomonadaceae actively defining a niche with less diversity or is due to their exploitation of an existing low diversity niche is unknown.
Many of the bacteria identified in this analysis have not yet been phenotyped, nor their potential for pathogenesis assessed. However, a number of known pathogens were more prevalent in younger CF patients, including members of the Mycobacteriaceae [44] and obligate intracellular members of the Chlamydiaceae [45] , [46] and the Mycoplasmataceae [47] (Supplementary Table S3 ). In contrast, known or potential pathogens associated with older CF patients included members of the Burkholderiaceae and Thermoactinomycetaceae [48] , [49] . Age-based relationships with anaerobic species were also examined since their relevance to CF airway disease is a subject of much recent research and discussion [50] , [51] . Of the strict anaerobic bacteria that had a significant correlation with age, the majority exhibited a negative relationship and were detected in higher abundance in younger patients (Supplementary Table S3 ). However, a very small number of Clostridia and sulfate-reducing species exhibited an increase in relative abundance in older CF patients.
Discussion
In this study, CF patient airway colonization was examined using a culture-independent phylogenetic microarray and samples from a cohort of patients defined as clinically stable (no change in pulmonary function for ≥2 months prior to sample collection), aged between 9 months and 72 years ( Table S4 ). For the purpose of this study it was important to consider only clinically stable time points to determine the longer-term evolution of the CF airway microbiota and avoid the pronounced short-term impact of pulmonary exacerbation and antimicrobial therapy. Deep throat swabs were used to sample pediatric patients in this study, this was to ensure inclusion of neonatal and younger pediatric patients who do not produce sputum. Expectorated sputum samples (following routine spirometry) were collected for adult patients. Despite differences in sampling between the age groups, the mode of sampling did not seem to drive the bacterial community composition. One caveat of airway studies is the necessity to collect the samples through the oral cavity and hence the potential for sample contamination. Both deep throat swab and sputum samples are routinely used for clinical laboratory culture to guide antimicrobial treatment regimes for CF patients. In addition, Rogers and colleagues have previously demonstrated that distinct bacterial communities exists in oral and sputum CF samples [14] , also supporting the hypothesis that oral contamination minimally impacts the assessment of airway samples from CF patients.
A total of 158 bacterial families were detected in these samples, the majority of taxa detected in these families have not been previously associated with cystic fibrosis. There are several potential explanations for this including that the PhyloChip is a highly sensitive molecular assay, capable of detecting organisms that are present at levels of only 0.01% of the total community. Validation by sequence analysis of Veillonella parvula confirms its reported presence in this niche by the array, suggesting that multiple organisms may co-habit this niche. Indeed, other culture-independent T-RFLP-based studies (estimated by the authors as having a sensitivity of 1% of the population), clone library sequencing and temporal temperature gradient gel electrophoresis (TTGE) have also detected a number of species that had not previously been associated with CF airways [10] , [11] , [15] , [16] . Even with the limitations of culture-based methods, previous studies have also demonstrated the presence of uncharacteristic species in CF airways using this approach [52] . More recently a sequence-based study of a CF patient airway sample revealed the presence of multiple unusual species not previously reported in this niche such as Dolosigranulosum pigrum, Kocuria rosea, Granulicatella spp. and Bergeyella spp. amongst others [53] . All of the reported species detected in this sequence-based study were also detected by PhyloChip in several of the patients in our study. This in addition to our sequence-based validation, collectively suggest that a multitude of bacterial species do indeed exist in CF airway samples and underscore the need to perform truly deep sequencing to overcome the issue of the dominant species present and detect the “rare biosphere” present.
The impact of sequence depth on interpretation of results was recently demonstrated by Qin et al [54] , in a large-scale sequence-based human microbiome study of 124 European individuals. Increasing the sequence depth coverage for samples from two individuals (from ∼4 Gb to >8.5 Gb) resulted in an increase in the number of strains common to these two individuals by 25% [54] . This demonstrates that human host microbiota are intrinsically diverse and that our current view of these assemblages is only curtailed by the limited depth of sequencing that has been used to interrogate them. Though we recognize that the array-based technology used in this study is potentially subject to cross-hybridization at the taxon-level, it nonetheless provides a standardized tool for high-resolution profiling of samples, permitting detection of low abundance species in dominated communities and relative changes in community composition that can be related to clinical measurements and features of the disease.
In natural systems, colonizing populations are observed to develop progressively, typically leading to increased biomass, productivity and diversity [55] , [56] , [57] , [58] . This has recently been exemplified in the human gastrointestinal tract; initial colonization by pioneer aerobic species is followed by facultative anaerobes and finally strict anaerobes, concomitant with an increase in bacterial biomass and productivity characteristic of the stable adult microbiota [59] . Though this is a cross-sectional study, data presented here provides insights into the progression of CF airway colonization. Evidence for initial diversification in younger patients, decreases in diversity in older patients and the presence of specialized communities of phylogenetically related species ( Fig. 3 ) associated with poor pulmonary function in these older patient suggest that CF airway microbiota may also follow the rules of community assembly previously reported at other host niches. This form of colonization involving an initial rapid rise in species diversity as successive invasions occur, followed by species replacement as the community develops is common, and has previously been reported for other ecosystems and at higher trophic levels [60] , [61] , [62] . Once established, severe antimicrobial-based perturbation of microbial communities has been shown to lead to long-term changes in community composition and a loss of diversity in animal models [63] . Subsets of older CF patients exhibit diminished improvement in lung function in response to antimicrobial administration compared with younger CF patients [64] , who also typically exhibit better pulmonary function. This suggests that the more phylogenetically-related microbial assemblages detected in older CF patients may be more resistant to these antimicrobials and contribute to poorer airway function, while the more diverse pediatric microbial assemblage is more sensitive to antimicrobial perturbation and linked to improved pulmonary function. More recently a study has demonstrated a “like begets like” phenomenon, in that colonization of a specific niche by particular keystone species results in “invasion” of that community by other phylogenetically related species [65] . It appears from our data that a similar scenario exists in older CF patients whose airway microbiota exhibits phylogenetically related members and is largely composed of Pseudomonadaceae. That multiple members of the Pseudomonadaceae may co-exist in CF airways is not unprecedented, Harris and colleagues have previously demonstrated with relatively shallow sequencing depth, the presence of other Pseudomonas species other than P. aeruginosa, in 50% of their CF patient airways [10] , suggesting that multiple members of the Pseudomonadaecaeae may co-exist in this niche.
10.1371/journal.pone.0011044.g003 Figure 3
Bacterial community structure and composition associated with CF patient age.
Compared to pediatric communities, adult CF patient airway communities exhibit lower bacterial diversity and are more uneven. Increased diversity in younger airways is correlated with a large number of known pathogenic families. Loss of diversity in older airways is strongly correlated with loss of pulmonary function and emergence of competitively dominant species such as members of the Pseudomonadacece, Burkholderiaceae and Xanthomonadaceae. For simplicity, the initial increase in diversity exhibited by younger CF patients is not illustrated.
It seems counter-intuitive that a greater diversity of organisms (in younger CF patients) would be associated with better lung function, however several studies have demonstrated that dramatic changes in community structure through loss of diversity as reported here, are increasingly being associated with chronic inflammatory diseases, pathogen outgrowth and poor clinical outcome [3] , [66] , [67] , [68] , [69] . It appears therefore, that bacterial community structure and composition represents an important factor in defining the functionality of the microbial assemblage and host health status. However analysis of CF cohorts is difficult and confounded by the fact that a number of the older patients in this study have less severe disease, which may be associated with their microbial community composition or different treatment regimens. In addition, there is also the possibility that the findings reported here are due to other factors that track with age such as antibiotic use, chest physical therapy, adherence, nutrition or other factors. Nonetheless, this study demonstrates that distinct microbial assemblages are associated with CF patient age and that community composition is correlated with CF genotype and aspects of pulmonary disease in this patient population.
This study also provided information on the relationship between CFTR mutation and the airway microbiota. It has previously been reported that CFTR mutation changes the airway microenvironment [70] , [71] , [72] , which would presumably influence the microbial community that establishes in this niche. A strong relationship between ΔF508 CFTR mutation and absence of multiple members of the Mycobacteriaceae, Staphylococcaceae, Enterobacteriaceae amongst others with a concomitant rise in abundance of members of the Pseudomonadaceae was identified in older CF patients with this genotype. This suggests that mutations rendering the CFTR non-functional are associated with a loss of airway bacterial (including pathogen) diversity and outgrowth of a small group of phylogenetically-related species as these patients age. Whether this is directly related to the severity of the mutation and the creation of a distinct niche due to lack of functional CFTR or due to the treatments necessary to manage these patients (or a combination of both) is unclear. However, patients with heterozygous ΔF508 or non-ΔF508 mutations also exhibited distinct pathogen profiles. Older heterozygous ΔF508CF patient airways were associated with Moraxellaceaea and Sphingobacteriaceae, two bacterial families that have recently been associated with chronic obstructive pulmonary disease [73] and invariant Natural Killer T cell induction in asthmatic mice [74] . While non-ΔF508 patients were associated with a relatively less severe pathogen profile. This study, albeit small, demonstrates that homozygous-ΔF508 mutation is associated with the most substantial change in airway community structure and phylogeny in older patients. Furthermore, particular CFTR mutations, which are known to influence the airway environment [70] , [71] , [72] , are associated with distinct pathogen profiles, a finding that may explain the range of severity in pulmonary symptoms commonly observed with various CFTR genotypes, and has implications for patient-tailored care [75] .
It is important to note that detailed functional analyses of longitudinal CF airway samples are necessary to comprehensively determine the impact of treatment on the airway microbiota. Given our data, it appears that older CF patients possess a stable core of pathogenic organisms which presumably are selected for over time due to repeated antimicrobial pressure. With the increasing lifespan of CF patients due to successful disease management, understanding the long-term impact of antimicrobials on the CF microbiota may lead to further improvements in treatment strategies and life expectancy. Novel approaches that involve manipulation of microbial consortia, rather than destruction of the community structure may offer an alternative therapeutic approach for patient-tailored management of chronic airway disease. This concept is gaining increasing support [8] , [63] and there is evidence for its efficacy, although the mechanism underlying these benefits remains unclear. A recent pilot study of pediatric CF patients supplemented with a probiotic Lactobacillus species or placebo demonstrated that in addition to reduced gastrointestinal inflammation, patients who received the probiotic species exhibited a significant reduction in hospitalizations for pulmonary exacerbation [76] . Certainly with the advent of newborn screening programs and sophisticated culture-independent tools to comprehensively monitor patient airways, the opportunity to intervene at a very early age and alter the course of airway microbial colonization to improve patient outcome is unprecedented.
The CF microbiota is a complex community, but the use of molecular ecological approaches permits the progression of CF airway disease to be comprehensively explored. This work provides the foundation for an improved understanding of polymicrobial CF airway colonization, the relationship between community structure, composition and lung function in CF patients and the influence of CFTR mutation on the airway microbiota. Antibiotic administration, the acquisition of organisms from the internal and external environments, and changes associated with patient age (increased lung surface area, hormonal changes) represent key influences on this ecosystem. Understanding the mechanistic relationship between community dynamics, pathogen abundance and behavior, and the host immune response is crucial to further extending the lifespan of this patient population.
Supporting Information
Table S1
Taxa identified by 16S rRNA PhyloChip in the airways of CF patients.
(2.39 MB DOC)
Table S2
Correlation of taxon abundance and CFTR mutation.
(0.23 MB XLS)
Table S3
Correlation of taxon abundance with patient age.
(0.05 MB XLS)
Table S4
Patient demographics.
(0.09 MB DOC)
Figure S1
Change in bacterial community diversity over time. Change in diversity, normalized to length of time between sample collection points was calculated for CF patients ranging in age from 9 months to 43 years old and illustrates an initial net increase in diversity (per day) in younger patients in comparison with a decrease in diversity in older patients.
(0.18 MB TIF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Alterations in the gut microbiome (dysbiosis) have been reported in human colonic neoplasia [ 1 – 6 ]. However it remains unclear as to whether dysbiosis represents a response to tumorigenesis or whether it precedes tumor formation. One of the most prominent genetic mutations associated with the pathogenesis of sporadic and hereditary colorectal cancers (CRC) lies in the tumor suppressing adenomatous polyposis coli (APC) gene [ 7 – 13 ]. A germ-line mutation of the APC gene causes familial adenomatous polyposis (FAP), which results in the development of multiple colorectal adenomas at an early age that unequivocally lead to CRC if no surgical interventions are taken. APC mutations also represent an early event in the adenoma-carcinoma sequence and are present in about 70–80% of sporadic human colorectal adenomas and carcinomas.
The multiple intestinal neoplasia (Min) mouse model of FAP carries a truncation mutation at codon 850 of the Apc gene [ 14 ]. Studies comparing the number of intestinal polyps in germ-free and conventionally raised C57Bl/6 APC Min/+ mice suggest that the gut microbiome may promote development of intestinal neoplasia [ 15 , 16 ]. One study reported decreased incidence of polyps in only the mid small intestinal segment, however a subsequent study reported a significant reduction of intestinal adenomas in both the small and large intestine of germ-free mice compared with conventionally raised mice. Antibiotic treatment of C57BL/6 APC Min/+ MSH2 -/- mice, which carry both the APC mutation and an HNPCC DNA mismatch repair mutation, reduced the number of polyps in both the small and large intestine [ 17 ].
We hypothesize that mutation of the APC gene results in alterations in host-microbiota interactions prior to tumor formation. To test this hypothesis, gut microbial composition was compared between 6 week-old C57Bl/6 APC Min/+ , prior to the development of detectable neoplasia [ 18 ], and congenic WT mice.
Materials and Methods
Animal Type and Housing
All of the mice were acclimated for two weeks in order to reduce stress from traveling. Carbon dioxide was used during euthanasia of the mice. This study was approved by the Institutional Animal Care and Use Committee (#202449) and Division of Laboratory Animal Resources at Stony Brook University. Three shipments of 10 four-week-old female C57BL/6J APC Min/+ and 10 four-week-old female C57BL/6J WT mice were received from The Jackson Laboratory (Bar Harbor, ME) between June 2012 and May 2013. APC Min/+ mice and WT mice were housed separately in groups of three to four in specific pathogen free (SPF) cages for two weeks prior to euthanization. All the experiments strictly followed guidelines from the Institutional Animal Care and Use Committee and Division of Laboratory Animal Resources at Stony Brook University.
Tissue and Luminal Content Sample Collection
All of the mice were euthanized at 6 weeks of age using carbon dioxide. Immediately after sacrifice, the gastrointestinal tract was divided along its cephalocaudal axis as previously described [ 19 ]. The segments analyzed included the ileum, cecum, proximal colon, and distal colon. Each small intestinal segment was washed in sterile phosphate buffered saline to remove the luminal content. A 1.0–1.5-cm section was obtained from the proximal ends of duodenum, jejunum, distal ends of ileum, proximal colon, and distal colon, and placed into RNAlater solution (Life Technologies, Grand Island, NY, USA) for RNA/DNA studies. The cecum was placed in its entirety in RNAlater. Three pellets of distal colonic luminal content (formed stool) were collected from the distal colon and stored in RNAlater. All the samples were kept at room temperature overnight and then stored in -80°C. In the first cohort, the remainder of each intestinal segment was processed into “swiss rolls” and fixed in 10% buffered formalin for histological analyses. In the second and third cohort, the remainder of each intestinal segment was stained with 0.25% methylene blue and inspected under a dissecting microscope for adenomas and aberrant crypt foci [ 20 ]. Stools were collected from nine 12–14 week APC Min\+ female mice with intestinal neoplasia [ 18 ] and six WT female mice, and placed into RNAlater.
DNA and RNA Extraction of Intestinal Tissue and Luminal Content Samples
Total RNA and DNA, (host and associated bacterial mixed community) were extracted from the duodenum, jejunum, ileum, cecal pouch, proximal colon and distal colon tissues using TRI Reagent (Sigma, St. Louis, MO) according to the manufacturer’s recommendations. For distal colonic luminal content samples, DNA was extracted using the UltraClean Fecal Kit (Mo BIO Laboratories, Inc., Carlsbad, CA).
Quantitative PCR (qPCR) for targeted bacterial clades
QPCR assays were performed using established primers for Bacteroides–Prevotella–Porphyromonas [ 21 ], Lachnospiraceae [ 21 ], and total bacteria [ 22 ] on all tissue and distal colonic luminal content samples as previously described [ 23 , 24 ]. The relative abundance of taxa within the Bacteroidetes phylum measured by ΔCt = Ct (threshold cycle) total bacteria - Ct Bacteroides–Prevotella–Porphyromonas . The relative abundance of taxa within the Lachnospiraceae clade (i.e., Clostridia Group Xi’an clade) was measured by ΔCt = Ct (threshold cycle) total bacteria - Ct Lachnospiraceae as previously described [ 23 , 24 ]. All assays were carried out in triplicate. Plasmid quantification standards were prepared from representative clones of the target organisms. The Mann-Whitney test using GraphPad Prism 5 (La Jolla, California) was performed to compare APC Min/+ and WT ΔCt values. The Bonferroni correction was made to correct for multiple comparison, thus significance required p< 0.025.
16S rRNA Amplicon Library Construction and Illumina V1V2 Sequencing Analysis
Bacterial profiles were determined by broad-range amplification and sequence analysis of 16S rRNA genes following our previously described methods [ 25 , 26 ]. In brief, amplicons were generated using primers that target approximately 300 bp. of the V1V2 variable region of the 16S rRNA gene (primers 8F and 338R, modified by addition of Illumina adapter and dual index sequences). PCR products were normalized using a SequalPrep kit (Invitrogen, Carlsbad, CA), pooled, lyophilized, purified and concentrated using a DNA Clean and Concentrator Kit (Zymo, Irvine, CA). Pooled amplicons was quantified using Qubit Fluorometer 2.0 (Invitrogen, Carlsbad, CA). The pool was diluted to 4nM and denatured with 0.2 N NaOH at room temperature. The denatured DNA was diluted to 15pM and spiked with 25% of the Illumina PhiX control DNA prior to loading the sequencer. Illumina paired-end sequencing was performed on the Miseq platform with version v2.3.0.8 of the Miseq Control Software and version v2.3.32 of MiSeq Reporter, using a 600 cycle version 3 reagent kit.
Paired-end sequences were sorted by sample via barcodes in the paired reads with a python script [ 25 ]. Sorted paired end sequence data were deposited in the NCBI Short Read Archive under BioProject Accession Number: PRJNA270112 ( www.ncbi.nlm.nih.gov/bioproject/PRJNA270112 )PRJNA270112. The sorted paired reads were assembled using phrap [ 27 , 28 ] and paired reads that did not assemble were discarded. Assembled sequence ends were trimmed over a moving window of 5 nucleotides until average quality met or exceeded 20. Trimmed contigs with more than 1 ambiguity or shorter than 200 nt were discarded. Potential chimeras identified with Uchime (usearch6.0.203_i86linux32) [ 29 ] using the Schloss [ 30 ] Silva reference sequences were removed from subsequent analyses. Assembled sequences were aligned and classified with SINA (1.2.11) using the 418,497 bacterial sequences in Silva 115NR99 as reference configured to yield the Silva taxonomy [ 31 , 32 ]. Operational taxonomic units (OTUs) were produced by clustering sequences with identical taxonomic assignments. OTU counts were normalized between samples by dividing sequence counts by the total number of sequences generated per sample. Phylum-level and family-level OTU tables were generated by collapsing lower level OTUs into higher-level categories. OTUs with a relative abundance <0.0001 and with a prevalence <0.01 were culled and relative abundances then transformed using the square root function [ 33 ]. The software package Explicet (v2.9.4, www.explicet.org ) was used to display OTU data and estimate alpha diversity indices (i.e., S Chao1 , Shannon complexity [H], and Shannon Evenness [H/H max ]) through 1000 replicate samplings of rarefied datasets [ 34 ].
Comparisons of individual phyla and families passing the initial filtering step, were conducted as follows. Because of the commonly observed over-dispersion in microbiome count data [ 35 ], the effects of APC Min/+ genotype on individual OTU abundances were examined using the negative binomial (NB) regression model as follows:
l o g ( μ i j k ) = ( β i 0 ) k + β 1 k g e n o t y p e i j + ( l o g t o t a l c o u n t ) i j ( β i 0 ) k = b 0 k + b i k I { s h i p m e n t i j = i } Y i j k ~ N B ( μ i j k , ϕ k )
Y ijk denotes the OTU k’s observed count for mouse j in shipment i, μ ijk is the mean parameter for OTU k’s count distribution of mouse j in shipment I and ϕ k is the dispersion parameter. Shipment refers to three different deliveries of mice (mice 1–20, mice 21–40 and mice 41–60), which has a zero-mean random coefficient. Coefficients b 0 and β 1 are fixed constant representing grand mean and APC genotype respectively. The log total sequence count for each mouse is considered as an offset. In addition, the skewness of the distribution of low abundance OTUs causes a large proportion of zero counts. Therefore, a zero-inflated version of the negative binomial (NB) model is also fitted to OTUs with zero counts in additional to the NB model:
l o g ( μ i j k ) = ( β i 0 ) k + β 1 k g e n o t y p e i j + ( l o g t o t a l c o u n t ) i j ( β i 0 ) k = b 0 k + b i k I { s h i p m e n t i j = i } Y i j k ~ N B ( μ i j k , ϕ k ) ( Y i j k | X i j k = 0 ) = 0 , ( Y ijk |X ijk = 1 ) = Y ijk X i j k ~ B e r n o u l l i ( π )
The NB and zero-inflated NB models are chosen based on AIC criterion [ 36 ]. The p values for the genotype effects on each OTU were then adjusted by the Benjamini-Hochberg [ 37 ] procedure to calculate the FDR. Significance was set as FDR<0.05. In addition, a 10-fold cross-validation was performed to validate the significant OTUs that were identified. The square root transformation was then applied to the relative abundances to correct for the skewness and to reduce the coefficient of variation. Comparisons of overall microbial composition between APC Min/+ mice versus the wild type mice were subsequently conducted using the permutation Hotelling T2 test with 10,000 permutations using the R package ‘Hotelling’ [ 38 ]. Principle coordinate analysis (PCoA) was conducted at the lowest taxonomic level (genus) using the wcmdscale function implemented by the vegan R package [ 39 ] and using Morisita-Horn dissimilarity scores.
Histological Analysis of the Intestinal Sections
Histological analysis was carried out in a subset of 10 APC Min/+ mice and 10 wild type mice (1 st cohort) by constructing “Swiss rolls” of intestinal segments. These segments were stained with hematoxylin and eosin and scored for adenomas and inflammation by a pathologist who was blinded with respect to the genotype of the mice (N.O.), as previously described [ 40 ]. Inspection of intestinal segments stained with 0.25%methylene blue was carried out on the second and third cohorts of mice with the aid of a Zeiss dissecting scope for detection of adenomas and aberrant crypt foci [ 20 ].
Comparison of Mouse Proximal Colonic mRNA Expression in APCMin/+ and Wild type Mice
IL-1β mRNA expression relative to actin mRNA, was measured in cecal, proximal colonic, and distal colonic intestinal tissue RNA samples in all 30 APC Min/+ and 30 WT mice as previously described [ 8 ]. RNA extracted from the proximal colon of 9 week-old WT mice treated with 3% DSS in water for 7 days was used as a positive control for the assay. The IL-1β ΔCt values (ΔCt = Ct actin -Ct IL1β ) were compared between APC Min/+ and WT groups using the Mann-Whitney test. Significance was set at a threshold of p <0.05.
Aliquots (1 μg) of proximal colon RNA samples from 6 APC were subjected to paired-ends 100 bp Illumina sequencing. The RNA-Seq libraries were prepared and sequenced at the New York Genome Center. Between 81 and 314 million reads were generated for each of the RNA samples. The RNA-Seq data were deposited in NCBI's Gene Expression Omnibus database with accession number GSE67634. The short reads were aligned to the GRCm38 genome ( http://useast.ensembl.org/Mus_musculus/Info/Annotation ) using STAR (Spliced Transcripts Alignment to a Reference) [ 41 ], and then converted to raw gene counts using featureCounts [ 42 ]. The edgeR package [ 43 ] was used to identify differentially expressed (FDR<0.05) genes (DEGs) between the APC Min/+ and wild type mice, using additional cutoff of 2-fold differential expression between groups. Hierarchical clustering based on the reads per kilobase of exon per million mapped reads (RPKM) value of the 130 DEGs was carried out by using 1-r dissimilarity measurement and Ward linkage, and the cluster number (n = 7) was chosen based on inspection of the coefficient of determination (R2) plot as previously described [ 44 ]. Second, a negative binomial (NB) regression model was fit with gene clusters as following:
l o g ( μ i j k ) = β i 0 + β 1 k g e n o t y p e i j + Σ h α h x i j h + ( l o g t o t a l c o u n t ) i j ( β i 0 ) k = b 0 k + b i k I { s h i p m e n t i j = i } Y i j ~ N B ( μ i j , ϕ k ) .
x ijh is gene cluster h’s expressions centroids (medians) of mouse j in shipment i. μ ijk is the mean parameter for phylum k’s count distribution of mouse j in shipment i and ϕ k is the dispersion parameter. “Shipment” is as defined above. b 0 and β 1 are fixed coefficients representing grand mean and APC genotype respectively. The log total count of each mice is considered as an offset. Best subset model selection was conducted to choose the model with lowest AIC. All models were fitted with R package: glmmADMB [ 45 ].
Results
The relative abundance of Bacteroidetes spp is increased in APC Min/+ mice colonic mucosa and luminal content prior to the development of intestinal neoplasias.
In preliminary targeted qPCR studies, we observed that the relative abundance of Bacteroidetes spp . in fecal DNA was higher in 12–14 week-old APC Min/+ female mice compared to age-matched WT female mice (ΔCt = -2.4 vs. ΔCt = -5.1, p = 0.0004). To test the hypothesis that the increase in the relative abundance of Bacteroidetes spp . preceded polyposis, we compared the relative abundance of this clade in the ileal, cecal, proximal colonic, distal colonic mucosa and the distal colonic luminal content in 6 week-old mice. Because gender effects have been previously reported on the number and location of polyps, we restricted our analysis to female mice [ 46 ]. Previous studies conducted on 6 week APC Min/+ female mice had established the absence of detectable neoplasias at that age [ 18 ]. The absence of intestinal adenomas and aberrant crypt foci was confirmed in the mice included in this study by microscopic inspection of the intestinal segments. The mean histological scores for inflammation were 0.1 and 0 (p = 0.37) for APC Min/+ and WT colons, respectively (n = 10 in each category). We observed a significant increase in the relative abundance of Bacteroidetes spp . in proximal colonic, distal colonic and distal luminal contents between the APC Min/+ and WT-control mice, but no significant difference in the ileal or cecal mucosal samples ( Table 1 ). In contrast we observed no significant difference is Lachnospiriceae spp ., a prominent group of Firmicutes, except in the luminal content of the distal colon, which exhibited significantly higher loads in WT mice (p = 0.003).
10.1371/journal.pone.0127985.t001
Table 1 QPCR comparison of the relative abundances of the Bacteroidetes phylum and the Lachnospiriceae clade within the Firmicutes phylum in 6 week old APC Min/+ and WT mice.
Bacteroidetes
APC Min/+ ΔCt Median (range)
Wild type ΔCt Median (range)
P-value
Ileal mucosa
-2.5 (-6.4, 0.5)
-3.0 (-7.5, 0.4)
0.65
Cecal mucosa
-4.0 (-8.0, -1.0)
-4.3 (-6.2, -3.1)
0.10
Proximal colonic mucosa
-3.4 (-6.9, -1.6)
-5.0 (-6.4, -2.2)
0.0008
Distal colonic mucosa
-3.1 (-7.1, -1.6)
-3.9 (-7.9, -1.7)
0.005
Distal colonic luminal content
-1.4 (-2.6, -0.1)
-2.3 (-5.3, -1.0)
<0.0001
Lachnospiriciae
APC ΔCt Median (range)
Wild type ΔCt Median (range)
P-value
Ileal mucosa
-4.2 (-12.1, -0.9)
-3.9 (-11.6, -2.4)
0.89
Cecal mucosa
-2.9 (-8.6, -1.3)
-2.4 (-6.1, -0.7)
0.06
Proximal colonic mucosa
-2.2 (-9.2, -0.3)
-1.9 (-11.4, -0.3)
0.15
Distal colonic mucosa
-3.6 (-8.3, -1.5)
-2.8 (-7.4, -0.5)
0.06
Distal colonic luminal content
-4.8 (-9.0, -3.4)
-3.9 (-10.4, -1.6)
0.003
The qPCR assays were conducted using established primers as described in Methods. The median and range of ΔCt values (~Log 2 relative abundance of targeted taxa) are listed for.the ileal, cecal, proximal colonic, distal colonic mucosal samples and the distal colonic luminal samples collected from 30 6 week old APC Min/+ and 30 wild type mice. The p-values were carried out using the Mann-Whitney U test. The Bonferroni correction was made to the p-value, so that significance was set at p<0.025.
16S rRNA sequence analysis of proximal colonic mucosal samples from APC Min/+ and WT mice
Illumina sequencing of the 16S rRNA gene V1V2 hypervariable region was carried out for the proximal colonic DNA samples. A total of 13,248,412 high-quality sequences were generated (average sequence length: 317 nt; average sample size: 220,807 sequences/sample; minimum: 9,381 sequences; maximum: 411,636 sequences, exclusive of negative controls which were near zero). The median Good’s coverage score was ≥ 99.9987% at the rarefaction point of 9,381 sequences, indicating deep sequence coverage of the intestinal microbiome.
The 16S rRNA sequencing results confirmed the targeted qPCR results in demonstrating a significant increase in the relative abundance of taxa within the Bacteroidetes phylum (FDR = 0.0009) in the APC Min/+ mice ( Fig 1A and Table 2 ). Within the Bacteroidetes phylum, the predominant family was S24-7 , whose relative abundance was also significantly increased in APC Min/+ mice ( Fig 1A and Table 3 , FDR = 0.0015). 16S rRNA sequencing detected a significant reduction in the relative abundance of taxa within the Tenericutes phylum (FDR< 0.0001), as well as a significant reduction in the relative abundance of the Anaeroplasmataceae family (FDR < 0.0001), which was the most prevalent family in the Tenericutes phylum. The results were also confirmed in all 10 cross-validations. The relative abundance of the Cyanobacteria phylum and the Chloroplast family, which is the most prevalent family in this phylum), was decreased in APC Min/+ mice (FDR = 0.047). However this observation was confirmed in only 4 out of 10 cross-validations.
10.1371/journal.pone.0127985.g001
Fig 1
Comparison of phyla and families between wildtype (WT) and APC Min/+ (APC) mice.
A. The mean relative abundances of phyla (left panel) and families (right panel) as inferred from the 16S rRNA sequence analysis. Only phyla and families with relative abundances >0.5% are shown. The Hotelling T2 test was used to compare the overall microbial composition, with p-values noted above each barchart.
10.1371/journal.pone.0127985.t002
Table 2 16S rRNA Sequence comparison of the relative abundances of phyla in the proximal colonic mucosa of 6 week old APC Min/+ and WT mice.
Phyla
Mean Relative Abundance
Test type, P-value and FDR
APC
WT
Test Type
P-value
FDR
Firmicutes
0.65247
0.69756
NB regression
0.10125
0.19302
Bacteroidetes
0.30827
0.21810
NB regression
0.00022
0.00089
Tenericutes
0.029
0.07425
NB regression
0.00001
0.00008
Bacteria
0.00535
0.00672
NB regression
0.16889
0.19302
Proteobacteria
0.00432
0.00276
NB regression
0.15898
0.19302
Actinobacteria
0.00055
0.00047
NB regression
0.31061
0.31061
Cyanobacteria
0.00004
0.00012
NB regression
0.01753
0.04675
Verrucomicrobia
0
0.00002
NB regression
0.16798
0.19302
Seven phyla remained after preprocessing filtering (maximum relative abundance < 0.0001, prevalence <0.01). Significant differences were detected in the bolded phyla , with the threshold set as FDR <0.05.
10.1371/journal.pone.0127985.t003
Table 3 16S rRNA Sequence comparison of the relative abundances of families in the proximal colonic mucosa of 6 week old APCMin/+ and WT mice.
Families
Mean Relative Abundance
Test type, P-value and FDR
APC
WT
Test Type
P value
FDR
Firmicutes phylum
Lachnospiriceae
0.50691
0.52577
NB regression
0.4327
0.5024
Ruminococcaceae
0.08136
0.09148
NB regression
0.1694
0.2072
Unassigned Clostridiales
0.03145
0.03665
NB regression
0.0589
0.0727
vadinBB60
0.02152
0.02471
NB regression
0.2689
0.3203
Erysipelotrichaceae
0.00369
0.00622
NB regression
0.0020
0.0027
Lactobacillaceae
0.00288
0.00402
NB regression
0.0399
0.0497
Peptococcaceae
0.00237
0.00650
NB regression
0.0000
0.0000
Unassigned Firmicutes
0.00118
0.00121
NB regression
0.9504
0.9717
Family XIII Incerta Sedis
0.00081
0.00076
NB regression
0.6219
0.6762
Peptostreptococcaceae
0.00010
0.00012
NB regression
0.0001
0.0001
Clostridiaceae
0.00006
0.00007
NB regression
0.9443
0.9717
Paenibacillaceae
0.00005
0.00000
NB regression
0.0000
0.0001
Bacillaceae
0.00004
0.00003
NB regression
0.7316
0.7831
Staphylococcaceae
0.00003
0.00001
NB regression
0.0157
0.0203
Thermoactinomycetaceae
0.00000
0.00001
zero-inflated NB regression
0.9933
0.9975
Unassigned Bacilli
0.00001
0.00000
zero-inflated NB regression
0.0000
0.0000
Bacteroidetes phylum
S24-7
0.30782
0.21769
NB regression
0.0002
0.0015
Unassigned Bacteroidales
0.00016
0.00007
NB regression
0.0000
0.0000
Bacteriodaceae
0.00014
0.00018
NB regression
0.5226
0.5818
Rickenellaceae
0.00009
0.00008
NB regression
0.5008
0.5670
Prevotellaceae
0.00003
0.00003
NB regression
0.7769
0.8251
Porphyromonadaceae
0.00002
0.00004
NB regression
0.4825
0.5508
Unassigned Bacteroidetes
0.00001
0.00000
NB regression
0.6390
0.6893
Ternicutes phylum
Anaeroplasmataceae
0.02664
0.07133
NB regression
0.0000
0.0000
RF9
0.00236
0.00293
NB regression
0.2202
0.2647
Proteobacteria phylum
Enterobacteriaceae
0.00319
0.00039
NB regression
0.0002
0.0003
Phyllobacteriaceae
0.00046
0.00144
NB regression
0.9975
0.9975
Bradyrhizobiaceae
0.00038
0.00065
NB regression
0.8941
0.9280
Sphingomonadaceae
0.00008
0.00011
NB regression
0.3601
0.4253
Methylobacteriaceae
0.00010
0.00007
NB regression
0.8034
0.8467
Ricketsiella/mitochondria
0.00001
0.00004
NB regression
0.0338
0.0425
Burkholderiaceae
0.00001
0.00001
NB regression
0.5053
0.5674
Alcaligenaceae
0.00001
0.00001
NB regression
0.5501
0.6029
Moraxellaceae
0.00001
0.00001
NB regression
0.0336
0.0425
Pseudomonadaceae
0.00001
0.00001
NB regression
0.9914
0.9975
Desulfovibrionaceae
0.00001
0.00000
NB regression
0.0120
0.0156
Actinobacteria phylum
Coriobacteriaceae
0.00050
0.00042
NB regression
0.3842
0.4499
Proprionibacteriaceae
0.00003
0.00004
NB regression
0.4533
0.5218
Corynebacteriaceae
0.00000
0.00001
NB regression
0.5266
0.5818
Cyanobacteria phylum
Chloroplast
0.00004
0.00012
NB regression
0.0213
0.0273
Verrucomicrobia phylum
Verrucomicrobiaceae
0.00000
0.00002
zero-inflated NB regression
0.8836
0.9240
Other Phyla
Other Bacteria
0.00535
0.00672
NB regression
0.2041
0.2475
Forty-two families remained after preprocessing filtering (maximum relative abundance < 0.0001, prevalence <0.01). The 42 families are listed within their respective phyla in order of their relative abundance, with the remaining families combined in Other categories. The phyla are underlined in the table. Significant differences were detected in the bolded families , with the threshold set as FDR <0.05.
Although 16S rRNA sequence analysis did not detect a significant difference in the overall relative abundance of the Firmicutes phylum, it also confirmed the lack of any change in the relative abundance of taxa within Lachnospiriciae clade, thus confirming the qPCR data (see Table 1 ). Furthermore decreases as well as increased were observed in some of the individual families within the Firmicutes phylum ( Table 3 ). Similarly, although 16S rRNA sequence analysis did not detect a significant difference in the overall relative abundance of the Proteobacteria phylum, increases as well as decreases were observed in some of the individual families within the Proteobacteria phylum.
Overall, the proximal colonic microbiomes differed between APC Min/+ and WT mice at both the phylum (p<0.0001) and family (p <0.0001) levels, as assessed by a 10000-permutation Hotelling T2 test of the 7 phyla and 42 families that passed the initial filtering step ( Fig 1A ). The proximal colonic microbiomes of APC Min/+ mice were also characterized by significantly lower OTU complexity (Shannon H; p = 0.009) and evenness (Shannon H/Hmax; p = 0.006) compared with WT mice ( Fig 1B ); OTU richness (S Chao1 ) was comparable in the two groups. Finally, principle coordinates analysis (PCoA) demonstrated significant associations of principle component axes 1 and 2 with APC genotype (see Fig 2 ).
10.1371/journal.pone.0127985.g002
Fig 2
Principle Coordinate Analysis.
PCoA was conducted at the family taxonomic level using a pairwise dissimilarity matrix calculated using the Morisita-Horn beta-diversity index. Each triangle is representative of a single APC Min/+ (APC) mouse and each square is representative of a single WT mouse, plotted along the first two principal component axes (left panel). PC1 and PC2 accounted for 61% and 5.6%, respectively, of total variance. The middle and right panels display the PC scores along axes 1 and 2, respectively. Differences in scores between genotypes were assessed by student t-test.
Differentially expressed genes in 6 week-old APC Min/+ mice
We reasoned that differences in colonic microbial composition associated with the APC Min/+ genotype must be linked to alterations in mouse colon gene expression, despite the lack of histological evidence of neoplasias. To examine how the APC mutation could alter the host colon gene expression, we conducted parallel RNA-sequence analysis on 6 APC Min/+ mice and 6 WT mice, (sampling all three cohorts). A total of 130 host genes (fold change > 2 fold, FDR <0.05) were selected using edgeR [ 43 ]. The DEGs were grouped into seven clusters as described in Methods (see Table 4 ), with 106 upregulated genes distributed among three clusters and the 24 downregulated genes distributed among four clusters.
10.1371/journal.pone.0127985.t004
Table 4 Hierarchical clustering of DEGs in APC Min/+ proximal colonic RNA transcripts vs. control RNA samples.
Gene name
Log 2 FC
FDR
Gene name
Log2 FC
FDR
Cluster 1
Up ↑
Cluster 4
Down↓
Hoxd13
7.95
5.00E-04
Rmrp
-3.9
4.00E-04
Tgm3
9.07
6.00E-04
Gm22513
-3.97
8.00E-04
Atp12a
6.4
8.00E-04
Gm26035
-5.49
8.00E-04
Ly6g
9.39
8.00E-04
Igkv5-48
-3.23
0.0022
Sval1
7.44
8.00E-04
Gm24146
-4.69
0.0023
Fut9
6.44
0.001
Igkv12-98
-5.86
0.0023
Gm8540
8.63
0.0018
Metazoa_SRP
-3.23
0.0034
AI854703
4.65
0.0022
Rn7sk
-3.04
0.0036
B3gnt7
5.27
0.0023
Igkv3-5
-5.88
0.0044
Cpn2
4.92
0.0023
Rpph1
-2.62
0.0073
Evx1
8.64
0.0023
Igkv3-1
-5.52
0.0224
Gm15053
8.81
0.0023
Vaultrc5
-2.57
0.0267
Itih2
5.16
0.0023
Ighv1-7
-2.54
0.0352
Mptx1
5.36
0.0023
Klk1b22
-2.09
0.0352
Myh2
3.28
0.0023
Olfr424
-4.87
0.0352
Thbs4
4.23
0.0023
Gm22179
-3.19
0.0353
Casp14
10.53
0.0029
Igkv9-124
-3.76
0.0445
Gm16341
5.86
0.0032
Cluster 5
Down↓
Best4-ps
6.75
0.0034
Ighv5-15
-6.07
5.00E-04
Ctse
4.5
0.0034
Igkv2-109
-5.08
0.0066
Fxyd4
9.83
0.0034
Cluster 6
Down↓
Gm2539
10.01
0.0034
Ighv9-4
-3.72
0.0176
Sval3
7.01
0.0034
Igkv9-120
-3.07
0.0432
Klk15
9.86
0.0035
Ighv1-80
-3.7
0.0456
Tmprss13
7.92
0.0035
Cluster 7
Down↓
Hoxb13
10.06
0.0039
Igkv14-100
-3.87
0.0388
Hoxa13
3.92
0.0044
Sel1l2
-7.57
0.0444
Mptx2
5.49
0.0044
Insl5
4.45
0.0045
Pla2g4f
3.83
0.0045
Slc28a3
4.12
0.0046
Vsig1
3.39
0.0046
Rims4
3.76
0.0047
Vtcn1
5.41
0.0049
Anxa8
3.61
0.0053
Pdzd7
3.17
0.0053
4930552P12Rik
3.66
0.0059
HOXA11-AS1_5
4.62
0.0059
Gpr83
4.27
0.006
B3gnt5
2.98
0.0064
Gjb5
3.34
0.0066
Tnip3
2.94
0.0066
Eno3
2.53
0.007
Gjb4
3.84
0.0074
Nxpe4
3.51
0.0074
Spink3
3.63
0.0083
Trpv3
3.11
0.0091
Hoxa11os
3.58
0.0098
2310079G19Rik
8.03
0.01
Gm11535
7.68
0.01
Cyp2f2
3.9
0.0112
HOXA11-AS1_4
3.93
0.0112
Muc1
3.11
0.0112
St8sia5
7.08
0.0126
HOXB13-AS1_2
7.9
0.015
Cyp2a12
6.98
0.0156
Ttr
2.01
0.0172
Nccrp1
3.85
0.0173
Evx2
7.43
0.0187
Slc46a1
2.38
0.0187
Csta
4.42
0.0188
Gm16556
3.57
0.0191
Sycn
2.76
0.0191
Slc15a1
3.24
0.0204
Grin2b
3.81
0.022
Cyp2d12
3.51
0.0223
Gm11830
3.09
0.0236
HOTTIP_2
7.46
0.0236
HOXB13-AS1_1
6.11
0.0236
Psg17
3.46
0.0236
Cela1
2.1
0.0258
Iqch
3.58
0.0258
Foxq1
2.17
0.0259
Ms4a10
1.48
0.0259
Nt5c1a
3.4
0.026
Il18
1.6
0.0264
Myo16
2.71
0.0264
Hoxd12
5.36
0.0271
Brinp2
2.7
0.0291
Ankdd1b
2.84
0.0292
Defb45
4.97
0.0292
Gp6
3.93
0.0328
Ggh
2.36
0.0336
Gm15401
1.69
0.0336
A930011G23Rik
1.78
0.035
Rdh16
1.75
0.035
Hrg
4.05
0.0369
1700042G15Rik
5.12
0.0411
Wnt8b
3.96
0.0413
Ctgf
1.44
0.0432
Slc36a1
2.23
0.0432
Nt5e
1.75
0.0434
Sh3d21
1.87
0.0461
Slc16a12
1.93
0.0466
Gm16557
5.58
0.0469
Ano4
2.8
0.0477
Pla2g5
1.87
0.0477
Gpr137b
2.36
0.0483
Cd207
6.32
0.0493
Gm17384
6.35
0.0494
Cluster 2
Up ↑
Gm10800
10
0.0023
Gm10801
9.72
0.0034
Gm21738
9.54
0.0035
Gm26870
7.7
0.0191
Gm10718
7.06
0.0236
Cluster 3
Up↑
Apon
2.18
0.0419
One hundred thirty DEGs were selected by edgeR analysis of RNA sequence data (see Methods ) and grouped into seven clusters by hierarchical clustering. Shown on the left are the three upregulated clusters (1–3), and shown on the right are the four dowregulated clusters (4–7).
NB regression (see Methods ) selected four (1, 4, 6, 7) out of seven gene clusters (see Table 5 ) significantly (p-value < 0.05) in addition to APC genotype, which were positively associated with the relative abundance of Bacteroidetes. Among those four clusters, cluster 1 (coefficient 0.051) is composed of 100 out of 106 upregulated genes. The downregulated gene clusters 4,6,7 covers 22 out of 24 downregulated genes. While APC genotype had a dominant effect on the relative abundance of Bacteroidetes, detection of additional associations with mouse colonic gene expression, suggest that alterations in host colonic gene expression play a role in influencing mucosal associated microbial composition.
10.1371/journal.pone.0127985.t005
Table 5 Association between gene cluster expression (centroid medians), APC genotype and the relative abundance of Bacteroidetes in the proximal colonic mucosa of 6 week old APC Min/+ and WT mice.
Bacteroidetes
Coefficient
p-value
APC genotype
0.915
8.36E-21
cluster1
0.051
0.000632
cluster4
0.031
7.17E-10
cluster6
0.006
0.000556
cluster7
0.029
0.000158
The cluster medians of seven geneclusters along with APC genotype were used in the following model as described in Methods . The significant effects are bolded , with the threshold set as p-value <0.05. Regression coefficients are also reported as index of effect size.
Because elevated IL-1β levels have previously been reported in 18–25 week old APC Min/+ compared with WT mice, RT-PCR assays were conducted on proximal colonic RNA samples in all of the mice as previously described [ 15 ]. No significant difference was observed in the ΔCt IL1β-actin values between APC Min/+ and WT mice (-13.5 vs. -16.2, p = 0.485). These values were both very low compared to that measured in DSS treated mice (-4.9), indicating that IL-1β was not highly expressed in the colons of either mouse group in our study.
Discussion
Alterations in the gut microbiome (dysbiosis) have been reported in human colonic neoplasia and in mouse models [ 1 – 6 , 47 ]. This study demonstrates that alterations in the gut microbiome, characterized by an increased relative abundance of Bacteroidetes spp. observed in association with intestinal neoplasias, actually precedes the development of microscopically detectable intestinal neoplasias in 6 week old APC Min/+ mice. Increased loads of Bacteroidetes spp . have been reported in another colitis-associated mouse model of colon cancer [ 47 ], and in some but not all studies of human colorectal neoplasia [ 1 – 6 ]. 16S rRNA sequence analysis revealed that the increased relative abundance of Bacteroidetes spp . corresponded primarily to an increased relative abundance of taxa within the uncultured family S24-7. Similar increases in S24-7 have also been reported in conventionally raised C57BL6 mice that were fed a high fat diet [ 48 ]. This association with intestinal dysbiosis is of interest, because increased dietary fat has previously been associated with increased number and /or size in both WT and APC Min/+ mice [ 49 , 50 ].
The relative abundance of the phylum Tenericutes observed in this study is higher than reported by some studies of C57Bl/6 mice [ 51 ], but similar to another study using C57Bl/6 mice purchased from the same vendor [ 52 ]. In this study, the APC Min/+ and WT mice were housed in separate cages, which could influence the reported microbial compositions [ 53 , 54 ], possibly related to coprophagic behavior.
Alterations in gene expression have been previously reported in normal appearing mucosa of APC mutant mice after the development of intestinal polyposis [ 55 ]. We report differential expression of genes (DEG) in 6 week APC Min/+ mice prior to the detection of intestinal polyposis. In order to integrate host colonic gene expression with the microbial taxonomic data, we reduced the gene expression input variables by first selecting DEGs, reasoning that these genes would be most likely to be involved in disrupted colonic microbial interactions in the mutant mice. Variable dimensionality was further reduced by clustering the 130 DEGs into seven groups. The detection of significant associations between host colonic gene expression and the relative abundance of microbial taxa, after taking into consideration APC genotype, support the concept that host colonic microbial cross talk influences mucosal associated microbial composition. Cluster 4, which included downregulated genes encoding immunoglobulins and non-coding functional RNAs, demonstrated a significant linear relationship with the relative abundance of Bacteroidetes after controlling for APC genotype. The inverse correlation between immunoglobulin gene expression and the relative abundance of Bacteroidetes spp . is intriguing in light of previous reports linking an increased relative abundance of Bacteroidetes spp . with a reduction of immunoglobulin coated bacteria in humans [ 56 , 57 ]. The observation that some of the non-coding RNAs in this cluster may be located in the mitochondria, is intriguing in light of the observation that mutated APC proteins in contrast to WT APC proteins are detected in mitochondria [ 58 ]
In summary, our results support the concept that APC haplo-insufficiency of the host colonic epithelial cell alters colonic microbial interactions prior to polyposis. It is thus conceivable that such microbiome changes contribute to the pathogenesis of colon cancer. An important corollary to such a notion would be that the colonic microbiome represents an important (and druggable) target for the prevention of colon cancer. Indeed, interventions directed at the microbiome (germ free and antibiotic treatment) have been reported to modulate tumor formation in mouse models of colon cancer [ 17 ]. However, it remains to be determined whether interventions directed at ameliorating dysbiosis in APC Min/+ mice, such as through probiotic, prebiotic or antibiotic interventions, could reduce tumor formation.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: | Figure 9A does not appear. Please view the entire Figure 9 here:
Download corrected item |
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Nonsense-mediated decay (NMD) is a molecular mechanism whereby potentially defective messenger RNAs (mRNAs) are degraded. The term “nonsense” refers to the type of mutation ( i . e . nonsense mutation, or a mutation that results in generation of a stop codon) that induces this mechanism. According to the “scanning model” of protein translation, the translation machinery scans an mRNA from the translation start site downwards until it encounters a stop codon and decouples from the mRNA [ 1 – 4 ]. However, if the translation machinery detects a premature translation-termination codon (PTC), it starts to recruit the NMD machinery, which then serves to degrade the “problematic” mRNA to avoid yielding a truncated peptide [ 1 ].
NMD has been conventionally regarded as an important mechanism for mRNA quality control. NMD-targeted transcripts (NMDTs) could result from point mutations, insertions/deletions, or alternative splicing events that give rise to a PTC [ 5 , 6 ]. NMD is observed in all investigated organisms, from bacteria to mammals [ 1 , 7 , 8 ]. This mechanism is involved in gene regulation and a wide spectrum of biological processes [ 9 – 11 ]. Importantly, NMD has been associated with human diseases [ 12 – 17 ]. For instance, Ullrich disease, an autosomal recessive congenital muscular dystrophy, has been found to be regulated by NMD factors [ 18 , 19 ]. Despite the biomedical importance of NMD, the evolutionary and regulatory origins of NMDTs have been less explored [ 20 , 21 ].
One major regulatory source of NMDT is alterations in transcript structure [ 22 – 25 ]. This is because inclusion/exclusion of coding exons or changes in exon boundaries may result in frameshift events, which in turn can generate PTCs. Of note, transcript structural alterations per se may be influenced by other mechanisms such as splice site mutations or structural variations in the genome. As mentioned above, NMD is mainly a translation-dependent mechanism. When the translation machinery halts at the first stop codon, and the stop codon is located more than 50–55 nucleotides (NTs) upstream of the last exon-exon junction, the NMD machinery will be engaged to initiate degradation of the mRNA [ 23 , 26 ]. This stop codon is defined as a PTC. Notably, however, exceptions to this rule have been reported. A transcript may be degraded even when the PTC is located within 50 NTs from the last exon junction ( e . g . T cell receptor β-transcript), or be resistant to degradation when the PTC is far upstream (e.g. PTCs within β-globin exon 1) [ 2 ].
Evolutionary and regulatory studies of NMD require correct classification of NMD events. For example, it has been reported that the conservation level of many exon-inclusion-caused, but not exon-exclusion-caused NMD events have emerged and been conserved in placental mammals [ 25 ]. Meanwhile, intron-retention-caused NMD events have been reported to regulate gene expression in retinitis pigmentosa and Taybi-Linder syndrome [ 27 ]. Despite the importance of NMD classification, there have been no publicly available tools to serve this purpose.
Here we report NMD Classifier, a tool for systematic classification of NMD events. NMDTs have been suggested to emerge during the evolution of vertebrates because of changes in splicing patterns [ 25 ] or point mutations [ 5 , 28 ]. NMDTs are also observed to result from single nucleotide polymorphisms in the human population [ 29 ]. Theoretically, an evolutionary/regulatory event that involves the smallest number of changes is the most likely to occur. We thus develop the NMD Classifier on the assumption of “minimal evolution/regulation”. We hypothesize that an NMDT has resulted from an evolutionary or regulatory event that alters the reading frame of a non-NMDT ( i . e . a “normal” coding transcript). By comparing an NMDT against its most similar coding transcript isoform, we could identify the transcript structure-altering event that has led to the NMD event. Our simulation results indicate that NMD Classifier yields highly accurate results in the identification of NMD-causing changes in transcript structure. This tool will be useful for future NMD-related studies, and is available at https://sourceforge.net/projects/transcriptome-analysis/files/NMD_Classifier.tar.gz
Result
Overview of NMD classifier
The analysis flow of NMD Classifier is shown in Fig 1 . The analysis starts with input data, which are either user-generated transcript assembly annotations (in GTF format) or annotation files from Ensembl (in GTF format) or NCBI (in GFF format). For user-generated transcript assembly, NMD Classifier detects NMDTs according to the 50-NT rule (see the next section) before analyzing transcript structural changes. For Ensembl/NCBI annotation files, the NMDT detection step is optional. NMD Classifier by default skips the detection step, and takes the annotated NMDTs for classifications. Next, for each NMDT, NMD Classifier identifies the best matching coding transcript isoform (“best partner”, see Methods ), which supposedly is the splicing isoform most similar to the interested NMDT. Each NMDT is then compared against its best partner. Each exon from an NMDT is “grouped” with an exon (or exons) from its best partner if the corresponding genomic regions of these exons overlapped with each other by at least one nucleotide. NMD Classifier then scans for frameshift events starting from the first exon group (the one that contains the translation start codon). If an upstream frameshift event is “rescued” by a downstream event (or events), the search for frameshift re-initiates downstream of the rescue event and continues until the last PTC is detected. Except in complex NMD events, the first transcript structure-altering event that results in the non-rescued frameshift event is considered as the cause of NMD, and is classified according to the splicing type of the specific event. In cases where no NMD-causing events are identified between an NMDT and its best partner, the NMD event is classified as “UTR alteration”. If an NMD event is caused by multiple types of transcript structural changes, it is classified as a “complex event” (Methods).
10.1371/journal.pone.0174798.g001
Fig 1
The analysis flow of NMD Classifier.
Detection of NMDTs
Ensembl and NCBI annotation files include NMDT annotations. However, if the user performs de novo transcriptome assembly, we must first correctly detect NMDTs before we can classify them. To this end, we have tested several commonly recognized NMDT detection criteria and used the Ensembl NMDT annotations (version 75) as a gold standard. Ensembl-annotated NMDTs are mostly supported by experimental evidence. The strength of supporting evidence is shown as the “Transcript Support Level (TSL)”. Approximately 56% of the Ensembl-annotated NMDTs have at least one supporting EST (TSL 1–3; S1 Table ). The tested NMD-detection criteria include a PTC located 50 NTs or 55 NTs upstream of the last exon junction, the presence of an upstream open reading frame (uORF), and inclusion of a long (> 650 NTs or > 2000 NTs) 3’untranslated region (3’UTR) ( Fig 2A ) [ 30 ]. We applied these NMDT detection criteria to all of the transcripts of coding genes annotated in Ensembl V75 (including 21,037 coding genes that encoded 42,637 coding transcripts and 9,357 NMDTs). An annotated coding transcript erroneously detected as an NMDT was defined as a false positive, while an annotated NMDT not detected was considered as a false negative. Fig 2A also shows that the 55-NT rule yielded a 1.83% false positive rate and a 2.42% false negative rate. The corresponding rates of the 50-NT rule were 1.92% and 1.57%, respectively. Both of the 55-NT and 50-NT rule yielded an overall accuracy of 97.8%. In comparison, detection based on the presence of uORFs or 3’UTR length yielded unacceptably high false positive and negative rates.
10.1371/journal.pone.0174798.g002
Fig 2
(a) The distribution of MCC values across different distances between a PTC and the last exon-exon junction. A positive (negative) distance indicates that the PTC is located upstream (downstream) of the last exon-exon junction; (b) False positive and false negative rates of different NMD prediction rule.
To more precisely determine which PTC-last exon junction distance was the best for detecting NMDTs, we calculated the Mathews Coefficient of Correlation (MCC)[ 31 ] for different distances. Interestingly, the largest MCC value occurred at 51 NT ( Fig 2B ), which was very close to 50 NT. According to these results, we selected the 50-NT rule to detect NMDTs for de novo assembled transcripts, and integrated it into NMD Classifier.
Evaluation of the accuracy of NMD classifier
To evaluate the accuracy of NMD Classifier, we conducted a simulation study based on annotated human transcripts (Ensembl V75). We generated artificial transcript structure-altering events by randomly inserting or deleting an exon from a coding transcript, or changing the 5’ or 3’ boundary (or both) of an exon. Specifically, we randomly selected one coding transcript. Then a coding exon of this transcript was randomly selected and removed (random deletion). For random insertion, an intron between two coding exons was randomly selected from a transcript. Part of the intronic sequence was then “turned into” a coding exon. The length of this artificial exon followed the length distribution of real coding exons. A similar approach was applied to generate random boundary changes—an exon length and a target coding exon were randomly selected. If the random length was larger (or smaller) than that of the target exon, the target exon was extended (or abridged) at 5’, 3’, or both ends with equal probabilities. Only one transcript structure-altering event was generated in each transcript. Five thousand artificial transcripts were created in each simulation. NMD Classifier was then tested on the mock transcriptome. A total of one thousand simulations were conducted. Fig 3A shows that NMD Classifier could correctly identify an average of 99.3% of the NMD-causing transcript structural changes. Note that whether an artificial transcript represented an NMDT was determined by the 50-NT rule. Our results demonstrate that NMD Classifier was highly accurate in identifying NMD-causing transcript structural changes.
10.1371/journal.pone.0174798.g003
Fig 3
(a) Distribution the accuracy of NMD Classifier in 1,000 simulation experiments. (b) The numbers of NMDTs identified in the transcriptomes of paired normal-tumor tissues from lung adenocarcinoma; the numbers in the parentheses indicate the percentages of NMDTs that are annotated by Ensembl; (c) The distribution of relative expression level (D value) of NMDTs between tumor and normal tissue.
On potential concern in the above simulation study is that the “best partner” transcripts may have specific features not considered in the simulation. To address this issue, we compared the lengths and expression levels between the “best partner” transcripts and the other transcript isoforms from the same genes. Indeed, the “best partner” transcripts of NMDTs tended to be longer and less expressed ( S1A and S1B Fig ). However, the accuracies of NMD Classifier stayed at 98~99% regardless of length and expression level of the best partner transcript ( S1C Fig ).
Of note, here we only conducted simulations of single-exon transcript structural alterations (insertion/deletion or extension/shortening). In reality, multiple-exon alteration events might occur. However, the simulation of multiple-exon events is far more complex than single-exon simulations. Furthermore, in complex transcript structure-altering events, it is difficult to clarify the evolutionary/regulatory path leading to the emergence of an NMDT. Therefore, at this moment we may not be able to correctly assess the accuracy of NMD Classifier in detecting complex NMD-causing events. Nevertheless, complex transcript structural changes appear to be infrequent in NMDTs ( Table 1 ), and thus may be less important.
10.1371/journal.pone.0174798.t001
Table 1 An exemplar classification of NMD events in paired normal-tumor tissues from a lung adenocarcinoma patient.
Sample ID
ERR164502
ERR318893
Tissue Type
Normal
Tumor
NMD_ex
2346
2328
NMD_in
3531
3452
multi_NMD_ex
414
399
multi_NMD_in
28
27
A5SS
1320
1341
A3SS
152
141
A5SS+A3SS
834
779
NMD IR
87
89
nNMD IR
40
38
UTR_Diff
573
543
UTR_Diff_CDSdiff_NoFrameDiff
39
34
Complex
33
36
Total
9397
9207
NMD_ex: exclusion of an exon; NMD_in: inclusion of an exon; multi_NMD_ex: exclusion of multiple exons; multi_NMD_in: inclusion of multiple exons; A5SS: the changes occurred at 5’ splicing site; A3SS: the changes occurred at 3’ splicing site; A5SS+A3SS: the changes occurred at both 5’ and 3’ (A3SS) splicing site; NMD_IR: intron retention that occurred in NMDT; nNMD_IR: intron retention that occurred in NMDT’s best partner; UTR_Diff: the NMDT and its best partner has identical coding sequences but different untranslated regions; UTR_Diff_CDSdiff_NoFrameDiff: the NMDT and its best partner has different coding sequences but no frame shift, the NMD may be caused by differences in untranslated regions; Complex: multiple types of transcript structure-altering events are involved.
Application of NMD classifier to real data
We applied NMD Classifier to the transcriptomes of paired tumor-normal tissues from one lung adenocarcinoma patient. (Methods) [ 32 ]. Table 1 shows that each transcriptome included more than 9,000 NMDTs, with exon inclusion/exclusion-caused NMDTs (NMD_in and NMD_ex) representing the largest groups (~3,500 and ~2,300 NMDTs, respectively). The next largest group of NMDTs resulted from exon boundary changes (A3SS–alternative 3’splice site, A5SS–alternative 5’ splice site, and A3SS+A5SS), which represented ~2,300 NMDTs. Together NMD_in/ex and exon boundary changes accounted for ~87% of all of the NMDTs. The observation that NMD_in constituted the largest group was consistent with the result of a previous study [ 25 ]. About 6.6% (617 in 9,397) and 5.7% (525 in 9,207) of the NMDTs in normal and tumor tissue, respectively, were not annotated by Ensembl. This result indicated that most of the identified NMDTs in these transcriptomes have been previously annotated. However, caution should be taken because this might have resulted from insufficient RNA-sequencing depth [ 25 ] or the assembly approach adopted here (reference-based de novo assembly; see Methods ). Meanwhile, the vast majority (92.8–94.7%) of the identified NMDTs were shared between tumor and normal tissue ( Fig 3B ). Nevertheless, hundreds of NMDTs were observed in tumor or normal tissue only, indicating that disease state-specific NMD events were present for lung adenocarcinoma. This observation is intriguing considering that the two transcriptomes were derived from the same organ of the same individual. Of note, even if tumor and normal tissue shared the same NMDTs, these NMDTs might have different expression levels. To illustrate this phenomenon, we defined a “D value” to measure the relative expression level of an NMDT between tumor and normal tissue (see Methods ). D falls between -1 and +1, which indicate an NMDT is expressed exclusively in normal and tumor tissue, respectively. Fig 3C shows that ~2,000 NMDTs had their D values fall between -0.8 and -1.0, and another ~2,000 with D values between +0.8 and +1.0. Of note, a D value close to +1 indicated that the expression of the interested NMDT was close to zero in normal tissue, but its expression in the paired tumor tissue might not be high because D was an index of “relative expression”. Note that the large numbers of NMDTs at both D-value extremes might have resulted from very low expression levels of the relevant NMDTs. We thus screened out NMDTs with < 0.5 FPKM expression level in both of the samples. Indeed, the numbers of NMDTs at both extremes decreased ( S2 Fig ). The biomedical implications of these NMD events are worth further explorations.
Discussion
In this study, we develop a convenient tool for identification and classification of NMD-causing transcript structural changes. Of note, these transcript structural alterations result from changes in RNA splicing pattern (exon inclusion/exclusion, exon boundary changes, intron retention…etc). What NMD Classifier does not address is mutation-caused generation of stop codon, which may also be a PTC in an NMDT. The detection of mutation-caused PTC requires another analysis flow, and is not included in NMD Classifier. However, if a mutation leads to a change in splicing pattern (such as a mutation at a splice site), NMD Classifier could detect this change in de novo assembled transcripts given an adequate number of exon junction reads.
We demonstrated by a simulation study that NMD Classifier could identity nearly 100% of the structural changes that led to NMD events. The small number of events not detected by NMD Classifier resulted from “erroneous selection” of best partner. Recall that NMD Classifier relies on pair-wise comparison between an NMDT and a best-matching coding transcript isoform from the same gene. We found that occasionally an annotated “coding transcript” selected as a best partner could be an NMDT according to the 50-NT rule. This inconsistency between 50-NT rule and annotation undermined the accuracy of NMD Classifier. Fortunately the number of such events was fairly small (<1% of all the analyzed cases). This observation suggests that selection of best partner is crucial for the accuracy of NMD Classifier. In case of de novo assembly, some of the assembled transcript structures may be less reliable. The qualities of transcript assembly can be examined by using RSEM-EVAL [ 33 ]. For the two transcriptomes examined in this study (ERR164502 and ERR318893), the de novo Cufflinks-assembled NMDTs actually had higher RSEM-EVAL scores than Ensembl-annotated NMDTs ( S3 Fig ). However, users are encouraged to examine the assembly qualities of their transcriptomes before applying NMD Classifier.
NMD Classifier can be applied to evolutionary and regulatory studies. For example, NMD Classifier can be used to identify NMD-causing events in one interested species. The evolutionary trajectory of these events can then be studied by using comparative approaches [ 25 ]. For regulatory and disease-oriented studies, one could compare the patterns and activities of NMD in different conditions such as diseased vs. normal tissues or the same tissue at different developmental stages. Such comparisons may lead to discoveries of the roles that NMD plays in important biological functions.
Methods
Preprocessing of transcriptome data
NMD Classifier takes a GTF or GFF annotation file as input to identify and classify NMDTs. These annotation files can be downloaded from Ensmebl (GTF) or NCBI (GFF). The user can also generate his/her own annotation file in GTF format from RNA-sequencing raw data. Firstly, the RNA-sequencing data (in FASTA or FASTQ format) should be mapped to the corresponding genome by using a sequencing read-mapping tool (e.g. TopHat or STAR) [ 34 , 35 ]. The mapping output file (in BAM or SAM format) then can be submitted to an assembly tool (e.g. Cufflinks or Trinity) [ 32 , 36 ] to yield a GTF file. A GTF file contains transcript structure information and genomic coordinates, which can be analyzed directed by NMD Classifier. In the current analysis, two transcriptomes (ERR164502 and ERR318893) were mapped to the human genome by using STAR [ 34 ](version 2.4.2) with default parameters. The mapping results were input to Cufflinks for reference-based de novo assembly and estimation of expression level. The example files are included in the downloadable NMD Classifier package.
Analysis procedure
If the user uses standard annotation files downloaded from Ensembl or NCBI, NMDT annotations have been included in these files. However, if the user performs de novo assembly of transcripts by using tools such as Cufflinks, whether a transcript contains a PTC is unknown. In this case, NMD Classifier must predict the translation start site and the first in-frame stop codon. Since the transcriptional orientation was given by the assembler, NMD Classifier determined the coding region of a de novo assembled transcript by using three-frame conceptual translation. The reading frame that yielded the longest coding region was considered as the “correct” frame, and this longest coding region was defined as the main coding sequence of the de novo assembled transcript. The locations of the translation termination site and the last exon-exon junction could then be determined, and the distance in-between could be evaluated according to the 50-NT rule.
NMD Classifier classifies NMD-causing events based on the minimal evolution/regulation hypothesis. Particularly, it is hypothesized that an NMDT resulted from transcript structural alteration(s) of a coding transcript isoform, and that a minimal alteration was more likely to occur than a major one. Except in complex NMD events, the first structure-altering event that resulted in the non-rescued frameshift was considered as the cause of NMD. Therefore, for each NMDT, we selected the most similar coding transcript isoform (best partner) for comparison. Specifically, the coding transcript isoforms that shared translational start site with the interested NMDT were identified. Among these coding transcripts, the one that shared the largest proportion of nucleotides with the NMDT was chosen as the best partner of the interested NMDT.
Based on pair-wise comparisons between NMDTs and their best partners, NMD-causing events were classified into five groups: (i) boundary alterations: the changes occurred at either 5’ (A5SS) or 3’ (A3SS) splicing site, or both (A5SS-A3SS) ( Fig 4A ); (ii) cassette exons: inclusion (NMD_in) or exclusion (NMD_ex) of an exon or multiple exons (multi_NMD_in, multi_NMD_ex) resulted in NMD ( Fig 4B ); (iii) intron retention: intron retention that occurred in NMDT (NMD_IR) or in its best partner (nNMD_IR) ( Fig 4C ); (iv) UTR alterations: the NMDT and its best partner had identical coding sequences but different UTRs; and (v) complex events: the difference between an NMDT and its best partner included multiple types of transcript structural changes. An exemplar complex event comprises an exon inclusion in the NMDT and an intron retention event in the best partner.
10.1371/journal.pone.0174798.g004
Fig 4
Alternative splicing events that result in NMDTs.
(a) Changes at exon boundaries; (b) inclusion or exclusion of one or more coding exons; (c) Intron retention. Changes in untranslated regions and complex NMD events are not shown here. The orange boxes indicate the exon groups identified to be the cause of NMD events. NMDT: NMD transcript; nNMDT: non-NMD transcript; A5SS/A3SS: alternative 5’/3’ splice site; A5SS-A3SS: exon boundary changes at both 5’ and 3’ ends; NMD_in/ NMD_ex: inclusion/exclusion of an exon causes the NMD event; multi_NMD_in/multi_NMD_ex) inclusion/exclusion of multiple exons causes the NMD event; NMD_IR/ nNMD_IR: intron retention in the NMDT/non-NMDT causes the NMD event; CDS: coding sequence; UTR: untranslated region.
Specifically, for each NMDT and its best partner, the genomic coordinates of each exon (retrieved from the GTF file) were compared. Exons whose corresponding genomic regions overlapped with each other by at least one nucleotide were grouped together as an “exon group”. For example, if an exon of an NMDT was located at genomic coordinates 100–500, whereas an exon of its best partner was located at 300–700, these two exons were considered as an exon group that spanned coordinates 100–700. In another example, if an exon of an NMDT was located at 2000–2500, and two exons of its best partner were located at 1800–2200 and 2300–2700, respectively. The resulting exon group would span coordinates 1800–2700. In each exon group upstream of the PTC, the difference in reading frame between NMDT and its best partner was calculated and summed up from 5’ to 3’. Except in complex events, the first (most upstream) frameshift event was considered as an NMD-causing event ( Fig 5A ) unless the frameshift was “rescued” by a second, downstream event. If a “rescue” event occurred, the first frameshift event downstream of the “rescue” event was regarded as the NMD-causing event ( Fig 5B ).
10.1371/journal.pone.0174798.g005
Fig 5
Examples of how NMD Classifier identifies NMD-causing event by calculating differences in reading frame between an NMDT and its best partner.
(a) Insertion of a single exon [exon group (ii)] causes a one-base frameshift and therefore an NMD event; (b) A one-base frameshift occurs at exon group (ii) but is offset downstream at group (iii). NMD Classifier thus continues to scan for the next frameshift event, which occurs at group (iv). This latter frameshift is maintained throughout to the stop codon, and is identified as the NMD-causing event.
Test transcriptome data source and data processing
Two transcriptomes (ERR164502 and ERR318893) from lung adenocarcinoma were downloaded from the Gene Expression Omnibus (GEO) database [ 37 ] with accession number GSE40419. The transcriptomes were mapped to the human genome (Hg19, Ensembl V75) by using STAR [ 34 ]. The transcripts were assembled with the reference-based de novo assembly function of Cufflinks [ 32 ], and then analyzed by using NMD Classifier.
To analyze the relative expression level of NMDTs in tumor and normal tissues, we defined the “D” value as follows:
D = { 0 if the FPKM values in normal and tumor tissue are both 0 ; E T − E N E T + E N otherwise
(1)
Where E T and E N , respectively, indicated the expression level (in FPKM) of the interested NMDT in tumor and normal tissue.
Supporting information
S1 Fig
Distributions of relative transcript length (left panel) and relative expression level (right panel) of best partners and the other transcripts in ERR164502 (a) and ERR318893 (b). The accuracies of NMD Classifier in the simulation study (c) across different relative transcript lengths (left panel) and relative expression levels (right panel).
(TIFF)
S2 Fig
Distribution of D values for transcripts with ≧ 0.5 FPKM expression level in at least one sample.
(TIFF)
S3 Fig
Distributions of RSEM-EVAL score (a) and expression level (b) of Ensembl-annotated NMDTs and de novo assembled NMDTs in two test samples (left panel: ERR164502; right panel: ERR318893).
(TIFF)
S1 Table
Transcript support levels of NMDTs
(XLSX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Ingested food, through various chemical and mechanical signaling pathways, induces peristaltic reflexes in the gut. Due to this motility, cells present in intestinal villi and colonic mucosa are responsive to both chemosensory and mechanosensory signaling [ 1 ]. Mucosal subepithelial PDGRFα + cells (MuPαC, aka subepithelial fibroblasts or fibroblast-like cells), located in the basement membrane under the epithelial layer of the colon, participate in the creation of contractile cellular networks via gap junctions [ 1 ]. These cells form subepithelial reticular intertwined networks around the crypts [ 2 ]. The networks enclose the lamina propria, in which MuPαC are in close proximity to neural and capillary networks, as well as myofibroblasts, epithelium, and immune cells [ 1 ].
MuPαC are closely associated with, but distinct from, myofibroblasts that express both α-smooth muscle actin (aka α-SMA: Acta2 ) and smooth muscle myosin ( Myh11 ) [ 3 ]. Both MuPαC and myofibroblasts in the lamina propria are mesenchymal cells that predominantly originate from the visceral mesoderm [ 4 ]. Together, MuPαC and myofibroblasts play a role in acute and chronic epithelial injury, fibrosis, chronic inflammatory diseases, and colitis-associated cancer [ 2 ].
Previously, our group has reported that primary MuPαC isolated from colonic mucosa express Toll-like receptor genes, purinergic receptor genes, 5-hydroxytryptamine (5-HT) 4 receptor gene, and hedgehog signaling genes [ 3 ]. However, a comprehensive resource that encompasses genome-wide transcriptomic analysis within these cells still has yet to be developed.
We have previously isolated GFP-labeled PDGRFα cells from the jejunal and colonic muscularis of Pdgfra-eGFP mice [ 5 ], and characterized genome-scale gene expression data from these cells. With this trove of data, our group constructed a Smooth Muscle Genome Browser [ 6 ] linked to the bioinformatics data repository found at the University of California, Santa Cruz (UCSC) genome database [ 7 ]. For this study, we utilized a similar strategy to isolate MuPαC from Pdgfra-eGFP mice and then sequenced the transcriptomes of these cells, as well as whole mucosal tissue from the murine colon. This information was incorporated into the previously mentioned UCSC Smooth Muscle Genome Browser. Through analysis of the obtained transcriptome, we were able to identify several new cell-selective markers for MuPαC including the metalloendopeptidase ADAM-Like Decysin 1 ( Adamdec1 ), fibronectin 1 ( Fin1 ), and collagen type VI alpha 4 ( Col6a4 ). We also identified several gene categories expressed in MuPαC including those encoding for growth factors, transcription factors, receptors, gap junction proteins, extracellular proteins, cytokines, peptidases, kinases, and phosphatases that are characteristic of MuPαC cellular identity and function. The MuPαC transcriptome we have added to the UCSC Smooth Muscle Genome Browser will serve as a resource that provides vital information about possible cellular structure, variously expressed transcript isoforms, and further insights into the regulation of all genes expressed in these cells.
Methods and materials
Animal and tissue preparation
Pdgfra eGFP/+ mice were obtained from Jackson Laboratory [ 8 ]. Mice were housed 4 per cage, maintained on a 12–12 hour light-dark cycle, and given access to food and water. All experiments were performed using 4–8 week old male and female Pdgfra eGFP/+ mice. The animal protocol was approved by the Institutional Animal Care and Use Committee at the University of Nevada-Reno Animal Resources. All experiments were performed in accordance with institutional guidelines and regulations.
Flow cytometry and fluorescence-activated cell sorting (FACS)
Cells were dispersed from the colonic mucosa of Pdgfra eGFP/+ mice using an enzyme digestion buffer comprised of 4mg/ml collagenase type 2, 8mg/ml trypsin inhibitor, 8mg/ml bovine serum albumin and 0.125mg/ml papain and incubated at 37°C for 30 min. GFP + PDGRFα cells were sorted from dispersed cells using FACS [ 5 ]. Isolated GFP + PDGRFα high cells (as differenctiated PDGFRα cells) from Pdgfra eGFP/+ mice (15 males and 15 females) were lysed and these cell lysates were pooled together with all other lysate samples. This pooled lysate from thirty Pdgfra eGFP/+ mice was used as one collective sample in the isolation of total RNAs.
Isolation of total RNAs
Total RNA was isolated from the colonic mucosa of mice using the mirVana miRNA isolation kit (Life Technologies, Carlsbad, CA) according to the manufacturer’s instructions. The quality of total RNAs was analyzed using a NanoDrop 2000 Spectrometer (Thermo Fisher Scientific, Waltham, MA) and a 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA).
Construction of RNA-seq libraries and next-generation sequencing
Two RNA-seq libraries were generated and sequenced via Illumina HiSeq 2000 (Illumina, San Diego, CA) following the vendor’s instruction at LC Sciences (Houston, TX) as previously described [ 5 ].
Bioinformatics data analysis
Paired-end sequencing reads were processed and analyzed as previously described [ 9 ]. A cutoff of FPKM = 0.025 generated equal false positive and false negative ratios of reliability. The expression level of transcripts with a FPKM value of less than 0.025 were considered to be 0.
Real time polymerase chain reaction
cDNA libraries were made using reverse transcription of the total RNAs isolated from FACS-purified MuPαC (n = 6) from colonic mucosa and smooth muscle PDGFRα cells (SMPαC: n = 6) [ 5 ], Interstitial cells of Cajal (ICC: n = 6) [ 10 ] and smoothe muscle cells (SMC: n = 6) [ 11 ] from colonic muscularis as well as colonic mucosa (n = 5) and smooth muscle, (n = 6) as previously described [ 5 , 10 , 11 ] [n = 5–6 mice (3 males and 2/3 females) per cell and tissue type]. Reverse-transcription polymerase chain reaction (RT-PCR) and quantitative PCR (qPCR) analyses of cDNA were performed as previously described [ 5 ]. All primer sequences used can be found in S1 Table .
Confocal microscopy and immunohistochemical analysis
Frozen murine tissues were fixed in cold acetone and 4% PFA before 8 μm cryosections were cut using a cryostat. Cryosections were then placed onto slides coated with Vectabond. anti-PDGFR-alpha (goat, 1:100, R&D system, MN), anti-Fibronectin (FN1) (rabbit, 1:100, abcam, MA), anti-Collagen VI (COL6) (rabbit, 1:200, abcam, MA), anti-PLAU (rabbit, 1:100, abcam, MA), anti-PROCR (Rabbit, 1:50, Bioss antibodies, MA), anti-BMP7 (Rabbit, 1:100, AVIVA system biology, CA), anti-SEMA3F (Rabbit, 1:50, Bioss antibodies, MA), and anti-PCSK6 (Goat, 1:100, antibodies-online, GA) were the primary antibodies used. Primary and secondary antibodies were diluted in 4% skim milk/1x TBS/0.1% Triton-X114. Each slide was washed twice with 1x TBS and treated with Fluoroshield mounting medium with DAPI (Abcam, ab104139) after incubation with the secondary antibodies. An Olympus FV1000 confocal laser scanning microscope (Olympus, Tokyo, Japan) was used to capture the immunohistochemically stained images and these images were analyzed through Fluoview FV10-ASW 3.1 Viewer software (Olympus, Tokyo, Japan).
Statistical analysis
qPCR data obtained in the present study was compared using a one-way analysis of variance (ANOVA) in order to determine whether the differences were statistically significant. Measured variables were expressed as the mean ± standard errors of the mean (SEM). The differences in mean values between the two groups (MuPaC and SMPaC) were evaluated and considered significantly different when * P < 0.05 and ** P < 0 . 01 .
Results
Identification and isolation of mucosal subepithelial PDGFRα + cells
Mucosal subepithelial PDGFRα + cells (MuPαC) were identified through eGFP expression within the subepithelial region of colonic mucosa in Pdgfra eGFP/+ mice [ 8 ] ( Fig 1A ). The identity of MuPαC was confirmed through immunohistochemical labeling with anti-PDGFRA antibodies that coincided with endogenous nuclear eGFP, as seen in previous work [ 3 ]. The PDGFRA protein is mainly localized in the plasma membrane of MuPαC, while eGFP is exclusively found within the nucleus of the cells as the eGFP gene is fused with a human histone H2B type 2-E gene in Pdgfra eGFP/+ mice [ 8 ]. MuPαC are in close proximity to each other under the epithelial barrier ( Fig 1A ). MuPαC are concentrated at the subepithelial area of the cryptic plateau within plicae in contrast to the lower number seen in the cryptic base and axis. Primary MuPαC from colonic mucosa were further analyzed through the use of flow cytometry. Our group previously reported two distinct populations of eGFP + MuPαC within the murine colonic mucosa: cells with brighter eGFP fluorescence (report higher expression of PDGFRα: P1) and cells with dimmer eGFP fluorescence (report lower expression of PDGFRα: P2) [ 3 ]. The P1 and P2 cells within eGFP + MuPαC, identified by fluorescence-activated cell sorting (FACS), were 6.3% and 17.8%, respectively, of the total events ( Fig 1B ), which was consistent with our previous cell sorting data [ 3 ]. Since P1 cells express Pdgfra at a higher level than P2 cells [ 3 ], we identified and isolated only the brighter eGFP + MuPαC population (MuPαC, P1) for RNA-seq. We sorted MuPαC from 30 mice (15 from each sex), extracted total mRNAs from each mouse’s isolated colonic mucosa, and pooled these mRNA samples together. This pooling process was also carried out on unsorted cells (colonic mucosal tissue).
10.1371/journal.pone.0261743.g001
Fig 1
Identification of colonic mucosal subepithelial PDGFRα + (MuPαC) cells and analysis of their transcriptome.
A : PDGFRα + cells in the colonic mucosa identified with Pdgfra-eGFP and through PDGFRA antibody. Pdgfra-eGFP mice express the H2B-eGFP (nuclear eGFP) fusion gene from the Pdgfra locus. L, lumen; SE, subepithelium; E, epithelium; LP, lamina propria. B : Primary eGFP + PDGFRα + populations (bright, P1, MuPαC, and dim, P2, MuPαC) from colonic mucosa identified (circled) through flow cytometry. C : Chart showing the number of genes identified in colonic mucosal tissue (Mu) and MuPαC cells in the colonic mucosa through RNA-seq. D : Comparison of expression levels of genes in Mu and MuPαC. E : Gene ontologies reported in Mu and MuPαC. The gene ontology (GO: function, process, and component) for Mu-/ MuPαC-specific genes was analyzed, and key GO terms were compared using normalized expression (FPKM) percentile. Blue and orange bars indicate a percentage of the gene number and an expressed amount of the gene, in each GO term category, to a total gene number or expressed amount respectively.
Transcriptomic analysis of mucosal subepithelial PDGFRα + cells
To identify the genes expressed within MuPαC, we obtained and analyzed the transcriptomes of isolated mucosal tissue (Mu) and MuPαC. The transcriptomes consisted of 15,933 (Mu) and 15,777 (MuPαC) known genes ( Fig 1C and Table 1 ). We obtained 169 and 154 million reads, of which 91% and 92% were mapped to the genome in Mu and MuPαC, respectively. We found 52,113 and 51,282 unique gene isoforms in Mu and MuPαC, respectively. Complete lists of all isoforms identified in this study along with tracking IDs, gene ID/names, chromosome location, isoform length, and expression levels in both Mu and MuPαC can be found in Table 1 . MuPαC expressed an average of 3 isoforms per gene that were produced from alternative transcription start sites, and/or alternative splicing sites (NCBI GEO GSM1388414 and GSM1388415, S2 Table ). Most genes (15,777) were expressed in both Mu and MuPαC; however, we found 156 genes that were expressed in Mu that were not found to be expressed in MuPαC ( Fig 1C ). A complete list of the genes expressed in Mu and MuPαC with their combined isoform expression levels and numbers of splice variants can be found in S3 Table . Although most genes are expressed in both Mu and MuPαC, the overall expression profiles of both samples were not very similar (correlation coefficient = 0.54) ( Fig 1D ). To further investigate cellular identity and function from our transcriptome data, we employed gene ontology (GO) analysis of genes abundantly expressed in Mu and MuPαC. Key GO terms and numbers of genes found to be associated with each term obtained from both samples were similar. The most highly expressed genes in the Mu population are involved in ion binding. In contrast, genes coding for extracellular proteins were the most highly expressed category in MuPαC ( Fig 1E ). This suggests that MuPαC may have an important role in extracellular function.
10.1371/journal.pone.0261743.t001
Table 1 Summary of transcriptomes obtained from colonic mucosal tissue (Mu) and subepithelial PDGFRα + cells (MuPαC).
Sample
Total read
Mapped read
Known gene
Total isoform
Average isoform
Mu
168,835,236
153,208,218
15,933
52,113
3.3
MuPαC
154,151,508
141,592,530
15,777
51,282
3.3
Identification of genes exclusively expressed in mucosal subepithelial PDGFRα cells
We have previously obtained and analyzed the transcriptomes of colonic smooth muscle tissue (SM) as well as three cell types that reside within gastrointestinal tissue: smooth muscle cells (SMC) [ 9 ], interstitial cells of Cajal (ICC) [ 10 ], and smooth muscle PDGFRα + cells (SMPαC) [ 5 ]. To identify genes selectively expressed in mucosal subepithelial PDGFRα + cells (MuPαC), we analyzed and compared the transcriptomes of MuPαC and Mu to the existing transcriptomes of SM, SMC, ICC, and SMPαC. We identified 76 genes that are highly, and selectively, expressed within colonic MuPαC when compared to Mu, SM, and the previously mentioned cell types (SMC, ICC, and SMPαC) ( S4 Table ). The thirty most selectively enriched gene expression signatures in MuPαC are shown in Fig 2A . The top three most highly enriched genes in MuPαC are Col3a1 , Adamdec1 and Col1a2 . Adamdec1 , Fn1 , and Col6a4 also show selective expression in Mu compared to SM ( Fig 2B ). Additionally, the top three MuPαC-enriched genes in comparison to Mu are Procr , Col6a4 and Bmp7 ( Fig 2C ). Lastly, the top three MuPαC-enriched genes vs SMPαC are Adamdec1 and Fn1 , and Plau ( Fig 2D ). Taken together, the most selective genes in MuPαC at mRNA levels include Adamdec1 , Fn1 , and Col6a4 . To validate the cell-restricted expression of genes expressed in MuPαC, we selected 8 genes ( Adamdec1 , Fn1 , Col6a4 , Plau , Procr , Bmp7 , Sema3f , and Pcsk6 ) and performed immunohistochemistry on murine colonic tissue in order to label the protein product of each previously listed gene (Figs 3A and 3B and S1 ). This screening identified ADAMDEC1, FN1, and COL6A4 as being selectively expressed in MuPαC. In a separate parallel study, we found that ADAMDEC1 is not only a selective marker for MuPαC but also a biomarker induced by colonic mucosal inflammation (in review) [ 12 ]. However, PLAU, PROCR, BMP7, SEMA3F, and PCSK6 expression were not able to efficiently label MuPαC ( S1 Fig ). Restricted localization of FN1 and COL6A isoforms in MuPαC isolated from colonic mucosa is shown in Fig 3 . FN1 was more prominently found in MuPαC compared to SMPαC ( Fig 3A ). In addition, the FN1 protein was abundantly detected in mesothelial cells in the serosal layer. FN1 abundantly colocalized in subepithelial PDGFRα + cells under the epithelial cells in cryptic plateaus (vertical sections) and cryptic bases or axes (horizontal sections), where epithelial stem/progenitor cells are located. Another marker, COL6A4, was labeled with the anti-collagen 6 (COL6A) antibody due to the isoform specific anitbody (anti-COL6A4) being unavailable. COL6A showed about equal signal strength in MuPαC as FN1. However, there are five collagen type 6 genes, Col6a1-5 , expressed in the colonic mucosa of mice ( S3 Table ). This results in the labeling of all proteins that are translated from the collagen type 6 genes when using the COL6A antibody. Our transcriptome data show that MuPαC have medium to high expression of Col6a1-4 , while these cells have very low expression of Col6a5 ( S3 Table ). Thus, the signal in MuPαC is likely mostly from COL6A1-4. Additionally, this antibody would likely also label SMPαC as these cells have abundant expression of three of the COL6A isoforms ( S4 Table ).
10.1371/journal.pone.0261743.g002
Fig 2
Identification of genes expressed in isolated colonic MuPαC.
A : A heat map of genes expressed in colonic MuPαC compared to colonic mucosa (Mu), colonic smooth muscle (SM), interstitial cells of Cajal (ICC), smooth muscle cells (SMC), and smooth muscle PDGFRα + cells (SMPαC). Col3a1 and Adamdec1 are highly expressed in colonic MuPαC. B : Colonic Mu-specific genes compared to colonic SM. C : MuPαC-specific genes compared to Mu. D : Colonic MuPαC-specific genes compared to colonic SMPαC.
10.1371/journal.pone.0261743.g003
Fig 3
Selective expression of FN1 and COL6A4 in colonic MuPαC.
A and B : Restricted expression of FN1 and COL6A protein within colonic subepithelial PDGFRA + cells. Anti-FN1 and anti-COL6A antibodies were used. Vertical and horizontal cross-sections (CS) images are indicated. Scale bars are 50 μm. C : Expression of Adamdec1 , Fn1 , and collagen type 6 isoforms ( Col6a1 - 5 ) in Mu, isolated MuPαC, SM, and isolated SMPαC examined by RT-PCR. D : Quantitative analysis of Adamdec1 , Fn1 , and Col6a1 - 5 mRNA expression in Mu (n = 5), isolated MuPαC (n = 5), SM (n = 5), and isolated SMPαC (n = 6) measured by qPCR. ** p ≤ 0.01, MuPαC versus SMPαC. Gapdh was used as an endogenous control.
However, according to the transcriptome data, Col6a4 is expressed at very low levels in SM and SMPαC ( Fig 2A and S4 Table ). Similar to FN1, COL6A isoforms were also found in subepithelial PDGFRα + cells at the cryptic plateaus and axes or bases ( Fig 3B ). To further validate expression of Fn1 and Col6a4 in MuPαC, we examined their cell-restricted expression in colonic MuPαC and SMPαC by RT-PCR. Consistent with the transcriptome data, colonic mucosa tissue and isolated MuPαC detected varying transcript levels of Adamdec1 , Fn1 , Col6a1 , Col6a2 , Col6a3 , and Col6a4 and very low expression of Col6a5 across all samples ( Fig 3C ). In addition, colonic SM tissue and isolated SMPαC showed very low transcript levels of Adamdec1 , Fn1 , and Col6a4 and abundant transcript levels of Col6a1 , Col6a2 , and Col6a3 . In terms of differences in expression levels between MuPαC and SMPαC, we observed Adamdec1 , Col6a4 , and Fn1 had significantly higher expression in MuPαC than SMPαC through qPCR analysis ( Fig 3D ). These findings were consistent with our transcriptome data ( S4 Table ). Taken together, the immunohistochemical, RT-PCR and qPCR data show that Adamdec1 , Fn1 , and Col6a4 are likely selective markers for MuPαC.
Identification of growth factors, transcription factors, cell signaling genes, receptors and receptor binding proteins expressed in mucosal subepithelial PDGFRα cells
MuPαC expressed 52 growth factors ( S5 Table ). The thirty most predominantly expressed growth factors in MuPαC, compared to Mu, are shown in Fig 4A . Cxcl12 and Ogn appeared to be the most highly expressed, while Bmp7 and Bmp5 were the most specific to MuPαC ( Fig 4B ). All ten of the most predominantly expressed growth factors in MuPαC were also expressed in SMPαC with Gpi1 being the only growth factor of these ten that is more highly expressed in SMPαC than levels seen in MuPαC ( Fig 4B ). Bmp5 and Bmp7 had very low expression in SMPαC but high expression in MuPαC ( Fig 4C ), suggesting that these two growth factors may be required for the growth of MuPαC.
10.1371/journal.pone.0261743.g004
Fig 4
Identification of the growth factors and transcription factors predominantly expressed in colonic MuPαC.
A : A heat map of the growth factors enriched in MuPαC compared to mucosal tissue (Mu) and smooth muscle PDGFRα + cells (SMPαC). B : MuPαC-specific growth factors compared to Mu. C : MuPαC-specific growth factors compared to SMPαC. Sorted by MuPαC/Mu, cut off 10 fold MuPαC in MuPαC/Mu; sorted by MuPαC/SMPαC, cut off 10 fold MuPαC and 0 fold SMPαC in MuPαC/SMPαC. D : A heat map of the transcription factors enriched in MuPαC compared to Mu and SMPαC. E : MuPαC-specific transcription factors compared to Mu. F : MuPαC-specific transcription factors compared to SMPαC. Sorted by MuPαC/Mu, cut off 10 fold MuPαC in MuPαC/Mu; sorted by MuPαC/SMPαC, cut off 10 fold MuPαC and 0 fold SMPαC in MuPαC/SMPαC.
In addition, MuPαC expressed 134 transcription factors ( S6 Table ). Fos and Jun were the most highly expressed transcription factors in MuPαC ( Fig 4D ), with these two genes also having highly expressed in SMPαC ( Fig 4D ). Tbx2 and Foxf2 appeared to be the most specific to MuPαC over Mu ( Fig 4E ), while Tbx2 and Foxf2 were the most specific to MuPαC over SMPαC ( Fig 4F ).
MuPαC also expressed 133 genes involved with cell signaling ( S7 Table ). The thirty most predominantly expressed cell signaling genes in MuPαC are shown in Fig 5A . Each one of these cell signaling genes was also found to be expressed in SMPαC, albeit at differing levels of expression. Interestingly, two Wnt singling genes ( Wnt5a and Wnt4 ) were specifically expressed in MuPαC when compared to either Mu ( Fig 5B ) or SMPαC ( Fig 5C ). Wnt4 and Wif1 were more highly specific to MuPαC as compared to levels found in SMPαC ( Fig 5C ).
10.1371/journal.pone.0261743.g005
Fig 5
Identification of the cell signaling genes, receptors and receptor binding proteins predominantly expressed in colonic MuPαC.
A : A heat map of the cell signaling genes enriched in MuPαC compared to mucosal tissue (Mu) and smooth muscle PDGFRα + cells (SMPαC). B : MuPαC-specific cell signaling genes compared to Mu. C : MuPαC-specific cell signaling genes compared to SMPαC. Sorted by MuPαC/Mu, cut off 10 fold MuPαC in MuPαC/Mu; sorted by MuPαC/SMPαC, cut off 10 fold MuPαC and 0 fold SMPαC in MuPαC/SMPαC. D : A heat map of the receptor binding proteins enriched in MuPαC compared to Mu and SMPαC. E : MuPαC-specific receptor binding proteins compared to Mu. F : MuPαC-specific receptor binding proteins compared to SMPαC. Sorted by MuPαC/Mu, cut off 10 fold MuPαC in MuPαC/Mu; sorted by MuPαC/SMPαC, cut off 10 fold MuPαC and 0 fold SMPαC in MuPαC/SMPαC.
Finally, MuPαC expressed 203 receptor and receptor binding protein genes ( S8 Table ). Gnas was the most highly expressed and Pdgfra was the most specifically expressed gene in MuPαC ( Fig 5D and 5E ). These two genes were also highly expressed in SMPαC ( Fig 5D ). Nlrp6 and Agt were the most specific genes to MuPαC over SMPαC ( Fig 5F ). However, Nlrp6 was also highly expressed in Mu, while Agt was expressed at much lower levels in Mu when compared to MuPαC levels ( S8 Table ). This suggests that Nlrp6 may be expressed in other mucosal cells, while Agt was predominantly expressed in MuPαC.
Identification of predominantly expressed genes related to gap junctions and extracellular activity in mucosal subepithelial PDGFRα cells
MuPαC expressed 18 genes related to gap junctions ( S9 Table ). Of these genes, Gja1 was the most highly and specifically expressed in MuPαC over both Mu and SMPαC ( Fig 6A–6C ). In SMPαC, Des was the most highly and specifically expressed ( S9 Table ). However, MuPαC specifically expressed Gjb1, Gjb3, and Gja1 over SMPαC ( Fig 6C ), suggesting that these gap junction proteins have a unique role in MuPαC.
10.1371/journal.pone.0261743.g006
Fig 6
Identification of the gap junction and extracellular proteins predominantly expressed in colonic MuPαC.
A : A heat map of the gap junction proteins enriched in MuPαC compared to mucosal tissue (Mu) and smooth muscle PDGFRα + cells (SMPαC). B : MuPαC-specific gap junction proteins compared to Mu. C : MuPαC-specific gap junction proteins compared to SMPαC. D : A heat map of the extracellular proteins enriched in MuPαC compared to Mu and SMPαC. E : MuPαC-specific extracellular proteins compared to Mu. F : MuPαC-specific extracellular proteins compared to SMPαC. Sorted by MuPαC/Mu, cut off 10 fold MuPαC in MuPαC/Mu; sorted by MuPαC/SMPαC, cut off 10 fold MuPαC and 0 fold SMPαC in MuPαC/SMPαC.
MuPαC expressed 600 genes related to extracellular activity ( S10 Table ). Col3a1 and Adamdec1 were the most highly expressed extracellular activity genes in MuPαC ( Fig 6D ) with Col3a1 being the most highly expressed in SMPαC, and Adamdec1 being minimally expressed in SMPαC ( Fig 6D ). Many extracellular proteins, including Masp1 , Bmp7 , and Penk , are preferentially expressed in MuPαC over Mu ( Fig 6E ). Additionally, Adamdec1 was the most specifically expressed gene in MuPαC over SMPαC ( Fig 6F ). Many extracellular activity genes are preferentially expressed in MuPαC over SMPαC: 44 genes are more than 100 fold enriched in MuPαC compared to SMPαC ( S10 Table ).
Identification of cytokine, peptidase, protein kinase, and phosphatase genes found in mucosal subepithelial PDGFRα cells
MuPαC expressed 77 genes encoding for cytokines ( S11 Table ). The thirty most highly expressed genes encoding for cytokines within MuPαC are shown in Fig 7A . Cxcl12 is the most highly expressed in MuPαC and SMPαC ( Fig 7A ). The genes most specific to MuPαC as compared to Mu are Bmp7 and Wnt5a (Figs 7B and S1B and S1D ). Twelve cytokine genes are preferentially expressed in MuPαC over SMPαC, with Cxcl9 and Fam3b being the most specific ( Fig 7C and S11 Table ).
10.1371/journal.pone.0261743.g007
Fig 7
Identification of the cytokines and peptidases predominantly expressed in colonic MuPαC.
A : A heat map of the cytokines enriched in MuPαC compared to mucosal tissue (Mu) and smooth muscle PDGFRα + cells (SMPαC). Sorted by MuPαC/Mu, cut off 10 fold MuPαC in MuPaC/Mu; sorted by MuPαC/SMPαC cut off 10 fold MuPαC and 1 fold SMPαC in MuPαC/SMPαC. B : Cytokines enriched in MuPαC compared to colonic SMPαC. C : MuPαC-specific cytokines compared to SMPαC. D : A heat map of the peptidases enriched in MuPαC compared to Mu and SMPαC. E : MuPαC-specific peptidases compared to Mu. F : MuPαC-specific peptidases compared to SMPαC. Sorted by MuPαC/Mu, cut off 10 fold MuPαC in MuPaC/Mu; sorted by MuPαC/ SMPαC, cut off 10 fold MuPαC and 1 fold SMPαC in MuPαC/SMPαC.
MuPαC expressed 283 peptidase genes ( S12 Table ). The thirty most highly expressed peptidases in MuPαC are shown in Fig 7D . Adamdec1 (previously categorized as an extracellular gene in Fig 6D ) is the most highly expressed peptidase in MuPαC, with a negligible expression level in SMPαC ( Fig 7D and 7F ). Interestingly many peptidase genes are preferentially expressed in MuPαC over SMPαC: 11 genes have an over 100 fold enrichment in MuPαC ( S12 Table ).
Finally, MuPαC expressed 354 protein kinase genes ( S13 Table ) and 105 phosphatase genes ( S14 Table ). The thirty most highly expressed protein kinases in MuPαC are shown in Fig 8A . Axl and Pdgfra (also previously categorized as a receptor in Fig 5D ) were the most highly expressed kinases in MuPαC ( Fig 8A ). As expected, Pdgfra was also highly expressed in SMPαC, being the most specific to both PαC ( Fig 8A and 8B ). The two most specific genes to MuPαC as compared to SMPαC are Rps6ka1 and Vegfa ( Fig 8C ). Rps6ka1 was also highly expressed in Mu, but Vegfa was expressed at a much lower level in Mu, suggesting that Rps6ka1 may be expressed in other mucosal cells; however, Vegfa is predominantly expressed in MuPαC ( Fig 8A and S13 Table ). The thirty most highly expressed phosphatase genes in MuPαC are shown in Fig 8D . Each one of these phosphatase genes were also abundantly expressed in SMPαC ( Fig 8D ). Dusp10 and Ptpn13 were the most specifically expressed in MuPαC over Mu and SMPαC, respectively ( Fig 8E and 8F ).
10.1371/journal.pone.0261743.g008
Fig 8
Identification of the protein kinases and phosphatases predominantly expressed in colonic MuPαC.
A : A heat map of the protein kinases enriched in MuPαC compared to mucosal tissue (Mu) and smooth muscle PDGFRα + cells (SMPαC). B : MuPαC-specific protein kinases compared to Mu. C : MuPαC-specific protein kinases compared to SMPαC. Sorted by MuPαC/Mu, cut off 10 fold MuPαC in MuPαC/Mu; sorted by MuPαC/SMPαC, cut off 10 fold MuPαC and 1 fold SMPαC in MuPαC/SMPαC. D : A heat map of the phosphatases enriched in MuPαC compared to Mu and SMPαC. E : MuPαC-specific phosphatases compared to Mu. F : MuPαC-specific protein phosphatases compared to SMPαC. Sorted by MuPαC/Mu, cut off 10 fold MuPαC in MuPαC/Mu; sorted by MuPαC/SMPαC, cut off 10 fold MuPαC and 1 fold SMPαC in MuPαC/SMPαC.
Validation of MuPaC-selective genes
As shown in Figs 2 and 4 – 8 , the 26 MuPαC-selective genes ( Col3a1 , Adamdec1 , Bmp7 , Bmp5 , Ogn , Foxf2 , Tbx3 , Tbx2 , Wnt4 , Wnt5a , Pdgfra , Nlrp6 , Agt , Gja1 , Gjb1 , Gjb3 , Dmp1 , Cxcl9 . Fam3b , Masp1 , Penk1 , Axl , Rps8ka1 , Vegfa , Dusp10 and Ptpn13 ) were indentified by the transcriptome analyses. To valiate the RNA-seq profiles of these genes, we quantified expression levels of each gene in isolated MuPαC, SMPαC, ICC, SMC, colonic Mu and SM tissue using qPCR analysis. Expression levels of the 24 genes were significantly higher in MuPαC than the other cell and tissue types, suggesting these genes are indeed MuPαC-selective ( S2 and S3 Figs). The other two genes, Pdgfra and Penk1 , were also more highly expressed in both MuPαC and SMPαC than SMC and ICC, implying they are MuPαC- and SMPαC-selective. These qPCR data confirmed the expression profiles of the 26 MuPαC-selective genes identified by the transcriptome analyses.
Addition to UCSC Smooth Muscle Genome Browser
Using data obtained from our previous smooth muscle transcriptome studies, we built a smooth muscle genome browser utilizing transcriptomes from jejunal and colonic SMC [ 9 ], ICC [ 10 ], and SMPαC [ 5 ] using the UCSC genome browser (UCSC Smooth Muscle Genome Browser; SMGB) [ 7 ]. We have now updated the browser with the colonic MuPαC and Mu transcriptome data found in this study. The SMGB now contains the transcriptomic data from colonic SM, SMC, ICC, SMPαC, Mu, and MuPαC along with jejunal SM, SMC, ICC, and SMPαC [ 6 ]. This SMGB (found at: https://med.unr.edu/physio/transcriptome ) provides not only the genomic map of each splice variant (promoter region, exons, and introns) for all genes expressed in MuPαC, SMPαC, SMC and ICC, but it also allows for analysis of our transcriptome data using the gene expression and regulation data from ENCODE [ 13 ] that is available in the database. For example, the genomic structure of Fn1 , identified as a new marker for MuPαC in this study, is shown in Fig 9A . The Fn1 gene consists of 46 exons, which are transcribed into 11 different variants in MuPαC (PaC Mu Colon) ( Fig 9A ). Alternative transcription start sites can be found at E1 (V1), E9 (V2), and E18 (V3). There are also 6 exons (E25, E33, E34, E40, E42, and E44), that are alternatively spliced. Expression levels for each variant can be found in Fig 9B with results showing that V1 (TCONS_00006336: 8,020 bp) is the most highly expressed variant followed by V2 (TCONS_00005374: 6,479 bp), and V3 (TCONS_00001747: 4,954 bp). Fn1 is expressed at a high level in colonic MuPαC, confirming a low level in the whole tissue Mu, but it is not expressed in colonic SMC, ICC, and SMPαC ( Fig 9C ), suggesting a mucosa-specific expression of Fn1 . Through further structural exploration of Fn1 , a CpG island was found around the promoter, first exon and intron where RNA polymerase 2 was previously found to bind in embryonic fibroblasts ( Fig 9A ). A region hypersensitive to DNase 1 in NIH3T3 and adult fibroblasts is also found within the same area. These data suggest that Fn1 is expressed in both embryonic and adult fibroblasts, as well as NIH3T3 cells (fibroblast cell line). In addition, there are two c-Jun binding sites in CH12 cells (lymphoma cell line), found at intron 1 (I-1) and intron 41 (I-41) within the gene ( Fig 9A ). The two corresponding regions of the c-Jun binging site, 361 bp (chr1: 71698290–71698650) within I-1 and 398 bp (chr1:71640784–71641181) within I-41, were located on the SMGB, and the two DNA sequences were analyzed for the presence of c-Fos and c-Jun binding sites in the transcriptional regulatory element search database, “PROMO” [ 14 ]. This search identified three binding sequences, CATAGTCAT , ATGACGTCAT , and CCAAGTCAG at I-1 of Fn1 gene and one TTGACTCTT at I-41, for both c-Fos and c-Jun ( Fig 9D ), suggesting the two transcription factors with the highest expression, FOS and JUN ( Fig 4D ), may transcriptionally regulate Fn1 via these binding sites. Among 11 Fn1 variants, V1 (TCONS_00006336), V2 (TCONS_00005374) and V3 (TCONS_00001748) are three major transcripts expressed in MuPαC ( Fig 9B ). Finally, Fn1 gene expression was examined in the colonic tissues and cells in the SMGB. The gene expression was restricted to MuPαC, which was confirmed by the expression in Mu and little to no expression in SMC, ICC, SMPαC, and SM ( Fig 9C ).
10.1371/journal.pone.0261743.g009
Fig 9
Genomic structure and expression data of Fn1 analyzed using the UCSC Smooth Muscle Genome Browser.
A : A genomic map view of Fn1 variants expressed in MuPαC. Three alternative initial exons (V1-3) are circled and six alternative exons are boxed in blue. A CpG island denoted by a green box. Red arrows indicate either binding sites of Polymerase 2 (Pol2) in mouse embryonic fibroblasts (MEF), a heterodimer of FOS and JUN (c-Jun) constituting transcription factor AP1 in CH12, or DNase I hypersensitive sites in NIH 3T3 and fibroblasts. Custom Tracks have view options (hide, dense, squish, pack, full: full is selected in the image) for the transcriptome data of MuPαC (eGFP + -PaC Mu Colon). B : Expression (FPKM) levels of Fn1 transcriptional variants in MuPαC whose structure is shown in A. The three most highly expressed variants (V1-3) are marked. C : Expression (FPKM) levels of total Fn1 mRNAs in colonic Mu, MuPαC, SM, ICC, SMC, and SMPαC. D : c-Fos and c-Jun binding DNA sequence within two peaks (I-1 and I-41) of c-Jun biding sites in A. Three binding sites of c-Fos and c-Jun binding, CATAGTCAT , ATGACGTCAT , and CCAAGTCAG are found at the peak of intron 1 (I-1) from PROMO while one binding site TTGACTCTT is found at intron 41 (I-41).
Discussion
In this study, we analyzed the transcriptome obtained from colonic MuPαC and identified signatures of genes including three new MuPαC-specific markers, Adamdec1 , Fn1 , and Col6a4 . Furthermore, we added the transcriptomic data to our Smooth Muscle Genome Browser [ 9 ] that already contains transcriptomic data from colonic and jejunal SMC [ 9 ], ICC [ 10 ], and SMPαC [ 5 ]. The browser offers a comprehensive reference for genes expressed not only in colonic MuPαC and Mu, but also colonic and jejunal SMPαC, SMC, and ICC as well as SM.
MuPαC were identified in the colonic mucosa as a unique cellular population that is distinct from subepithelial myofibroblasts [ 3 ]. MuPαC and subepithelial myofibroblasts are located in the same anatomical regions and are closely associated underneath epithelial cells [ 3 , 15 , 16 ]. Several markers including PDGFRA, ACTA2, MYH11, DES, and VIM can distinguish the two populations: subepithelial MuPαC (PDGFRA + , DES - , ACTA2 low , MYH11 low , and VIM low ) and subepithelial myofibroblasts (PDGFRA - , DES + , ACTA2 high , MYH11 high , and VIM high ) [ 3 , 15 ]. However, these markers still have overlap between the two cell types at varying levels [ 3 ]. Our transcriptome data from colonic MuPαC show a moderate to moderately high expression of of Acta2 (FPKM: 376) and Myh11 transcripts (FPKM: 30) in MuPαC ( S3 Table ). The Acta2 and Myh11 gene expression detected in our MuPαC transcriptome data is unlikely due to SMC contamination due to the observation that Des is not, or negligibly, detected in the SMC (FPKM: 2) ( S3 Table ) agreeing with previous findings [ 3 ]. This suggests that subepithelial myofibroblasts may be a sub-population of MuPαC.
MuPαC have at least two subpopulations, PDGFRA high (P1: near the apical area of the lamina propria) and PDGFRA low (P2: around the cryptic nadir) (Figs 1A and 2B ). PDGFRA high cells are expressed in telocytes/SEMFs (subepithelial myofibroblasts) in the villus and have a role in cell-to-cell communication [ 16 , 17 ]. Foxl1 , Pdgfra , Gli1 , CD34 , and Cspg4 are ascribed as molecular markers [ 17 , 18 ]; however, our transcriptome data from colonic MuPαC show high expression of Pdgfra (FPKM: 228), while other telocytes/SEMFs markers had low expression [ Foxl1 (FPKM: 32), Gli1 (FPKM: 44), CD34 (FPKM: 20) and Cspg4 (FPKM: 14)] in MuPαC ( S4 Table ). This suggests that telocytes/SEMFs may be a sub-population of MuPαC. PDGFRA low cells may also express smooth muscle genes. In fact, SMC and PαC are derived from the same mesenchymal precursor cells [ 19 ]. Recently, Roulis et al ., reported the identities of the mesenchymal cell population, which also expresses the PDGFRα + cell marker. Mesenchymal cells have four different fibroblast populations, all populations express Pdgfra [ 20 ]. In addition, PαC transdifferentiate into SMC in embryonic smooth muscle cells [ 19 , 21 ], while SMC have the ability to become PDGFRA low cells in response to intestinal injury and under cell culture conditions [ 11 ]. These phenotypic overlaps make it hard to identify definitive markers for MuPαC over subepithelial myofibroblasts. The three newly identified MuPαC markers, Col3a1 , Adamdec1 and Col1a2 , are more highly expressed in MuPαC than SMPαC ( Fig 2A ), suggesting they are better markers for MuPαC than PDGFRA alone. Further studies should explore if these new markers can distinguish MuPαC over subepithelial myofibroblasts.
Through transcriptomic analysis, we compared genes of interest between 1) MuPαC and Mu or 2) MuPαC and SMPαC throughout this manuscript. We were able to identify the top thirty genes that are enriched and selectively expressed in MuPαC over Mu. Next, we identified the top thirty genes that are enriched and selectively expressed in MuPαC over SMPαC. With limited space, we discussed only the most or second most expressed genes in each functional gene category. As it pertains to growth factors, we found Cxcl12 and Ogn are the most highly expressed genes in MuPαC. Cxcl12 encodes for the C-X-C Motif Chemokine Ligand 12 which functions as a ligand for the G-protein couple receptor 4 (CXCL12). This ligand regulates embryogenesis [ 22 ], stem cell homeostasis [ 23 ], immune surveillance [ 22 ], tissue regeneration [ 22 ], inflammation [ 24 ], and tumorigenesis [ 25 ]. Another growth factor found in MuPαC, Ogn , encodes osteoglycin which also regulates fibrosis [ 26 ], immune response [ 27 ], inflammation [ 28 ], and colon cancer [ 29 ]. A family of growth factors, the Bmp genes ( Bmp 7, 5, 3, 4, 1), are within the top ten genes of growth factors that are highly expressed in MuPαC. BMPs (bone morphogenetic proteins) belong to the transforming growth factor-β (TGFβ) superfamily. BMP7 is mostly expression in tumor including colon cancer, and it is regulated of cell proliferation [ 30 ]. Recent studies have shown that BMPs play an important role in regulating the immune response to infection, inflammation [ 31 ], and cancer [ 30 , 32 ]. In regard to the transcription factors expressed in MuPαC, we found Fos and Jun to be the most highly expressed genes in MuPαC. FOS and JUN form a heterodimer, forming the transcription factor AP1 (Activator Protein 1) that regulates the expression of genes involved in cell proliferation, differentiation, and apoptosis [ 33 ]. JUN is essential for fibroblast proliferation [ 34 ], and TGFβ stimulated cell proliferation via FOS [ 35 ], suggesting that AP1 could regulate the proliferation of MuPαC via BMPs.
The most abundantly expressed gene group in MuPαC are those related to extracellular activity ( Fig 1E ). Not only are they highly expressed, but they are also the largest gene group (600 genes) represented in MuPαC ( S10 Table ). Collagen types, 3, 1, and 6 are among the thirty most highly expressed genes in MuPαC ( Fig 6D ). Collagen is amongst the most abundant protein made in mammals, representing 25–30% of all proteins produced. Twenty collagen isoform genes, including type 1, 3, 4, 5, 6, 8, 12, 14, 15, 16, 18, 20, 24, and 27 are expressed in MuPαC. Most of these isoforms are also abundantly expressed in SMPαC with Col6a4 being negligibly expressed in S SMPαC, thus, we have identified Col6a4 as a MuPαC-specific marker ( Fig 2 ). The transcriptome data from isolated cells confirm that this gene is expressed in MuPαC, but insignificantly expressed in SMPαC ( Fig 3 ). The mucosa specific expression is consistent with the gene expression level in the transcriptome data found in both mucosa (Mu) and smooth muscle tissue (SM). Unfortunately, a COL6A4-specific antibody is not currently available. Therefore, we used an anti-collagen type 6 antibody which detects all COL6A isoforms, COL6A1-6. The immunohistochemical data in Fig 3 shows the protein is abundantly found within mucosa restricted to MuPαC as well as smooth muscle tissue mainly in SMPαC. The transcriptome data show that Col6a1 , Col6a2 , and Col6a3 had more expression in both MuPαC and SMPαC compared to SMC and ICC ( S4 Table ), suggesting that COL6A1-3 are mainly expressed in colonic PαC (SMPαC and MuPαC). The Col6a4 gene encodes for COL6A4 protein in mice, but is only a pseudogene in humans, which limits direct human application of this gene. Nevertheless, we demonstrated that Col6a4 gene products (mRNAs and protein) in mice can be used for a selective marker for MuPαC. In addition, the gene locus may be a useful target to generate MuPαC-restricted mouse lines.
In total, we were able to identify 15,777 genes expressed into 51,282 unique transcripts within isolated MuPαC. This valuable gene expression data was added to our “Smooth Muscle Genome Browser” [ 6 ] that contains the transcriptome data from colonic and jejunal SMC [ 9 ], ICC [ 10 ], and SMPαC [ 5 ]. This browser provides comprehensive genetic information in designated cell populations in both the colon and jejunum. In addition, through the browser, users can access the gene expression and regulation data (ENCODE) in the genome database [ 13 ] that allows for the study of genetic and epigenetic regulation of genes expressed within specific cell populations. However, there are some limitations to using the Smooth Muscle Genome Browser. For example, it cannot display an expression level and cDNA of an individual gene or transcriptional variant expressed in these specific cell populations. To rectify this issue, we built another browser: “Smooth Muscle Transcriptome Browser” [ 6 ]. This additional browser offers genetic references and expression profiles (expression levels, cDNAs, and encoded protein) of all transcripts expressed in individual cell populations and their associated tissues. Both browsers are available online, hosted by the University of Nevada, Reno at https://med.unr.edu/physio/transcriptome . These two browsers provide genome-wide genetic references and expression levels that bring advanced levels of insight into genetic structure, expression profile, and the isoforms of each gene expressed in intestinal cell groups and muscularis and mucosal tissue. We anticipate that these two browsers will greatly improve studies on GI smooth muscle biology and physiology.
In summary, we have analyzed the transcriptome of colonic MuPαC and identified signature genes including new selective markers relevant to cell identity and functionality. This transcriptome data was added to our Smooth Muscle Genome Browser and Smooth Muscle Transcriptome Browser that both offer vital genetic references for PDGFRα + cells that can aid further functional studies in intestinal diseases and physiology.
Supporting information
S1 File
(DOCX)
S1 Fig
Non-selective expression of PLUA, PROCR, BMP7, SEMA3F and PCSK6 in colonic MuPαC.
Vertical and horizontal cross-sections (CS) images are indicated. Scale bars are 50 μm.
(TIF)
S2 Fig
Validation of expression levels of MuPαC-selective genes.
( a-f ) Expression levels of MuPαC-selective genes in MuPaC, SMPαC, ICC, SMC, colonic Mu and SM tissue measured by qPCR. A : Col3a1 and Adamdec1 in Fig 2 . B and C : Bmp7 , Bmp5 , Ogm , Foxf2 , Tbx3 and Tbx2 in Fig 4 . D and E : Wnt4 , Wnt5a , Pdgfra , Nlrp6 and Agt in Fig 5 . F : Gja1 , Gjb1 and Gjb3 in Fig 6 . n = 5–6 per groups. * p ≤ 0.05 and ** p ≤ 0.01, versus MuPαC.
(TIF)
S3 Fig
Validation of expression levels of MuPαC-selective genes.
A - D : Expression levels of MuPαC-selective genes in MuPαC, SMPαC, ICC, SMC, colonic Mu and SM tissue measured by qPCR. A and B : Dmp1 , Cxc9 , Fam3b , Masp1 and Penk1 in Fig 7 . C and D : Axl , Rps6ka1 , Vegfa , Dusp10 and Ptpn13 in Fig 8 . n = 5–6 per groups. * p ≤ 0.05 and ** p ≤ 0.01, versus MuPαC.
(TIF)
S4 Fig
(PDF)
S1 Table
Oligonucleotides used in this study.
(XLSX)
S2 Table
List of transcriptional variants expressed in colonic Mu and MuPαC.
(XLSX)
S3 Table
List of genes expressed in colonic Mu and MuPαC.
(XLSX)
S4 Table
List of genes highly and selectively expressed in colonic Mu and MuPαC.
(XLSX)
S5 Table
List of growth factors expressed in colonic Mu and MuPαC.
(XLSX)
S6 Table
List of transcription factors expressed in colonic Mu and MuPαC.
(XLSX)
S7 Table
List of cell signaling genes expressed in colonic Mu and MuPαC.
(XLSX)
S8 Table
List of receptors and receptor binding proteins expressed in colonic Mu and MuPαC.
(XLSX)
S9 Table
List of gap junction proteins expressed in colonic Mu and MuPαC.
(XLSX)
S10 Table
List of extracellular proteins expressed in colonic Mu and MuPαC.
(XLSX)
S11 Table
List of cytokines expressed in colonic Mu and MuPαC.
(XLSX)
S12 Table
List of peptidase expressed in colonic Mu and MuPαC.
(XLSX)
S13 Table
List of protein kinase activity expressed in colonic Mu and MuPαC.
(XLSX)
S14 Table
List of phosphatases expressed in colonic Mu and MuPαC.
(XLSX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Lung cancer is the leading cause of malignancy-related mortality in both developed and developing countries, and non-small cell lung cancer (NSCLC) accounted for approximately 80% of all lung cancer deaths [ 1 , 2 ]. Despite the tremendous development of technique, lung cancer remains dissatisfactory improvement in survival during the past decades [ 3 ]. To ameliorate the survival of NSCLC patients, specific and sensitive prognostic factors for identifying cancer risk which can provide a more appropriate estimation of individual outcomes and allow the optimization of patient stratification are desired in clinical trials.
Existing researches have certified that systemic inflammation has a close relationship with malignancies [ 4 – 7 ]. Besides, patients’ nutritional status is also closely linked to cancer mortality, with one third of deaths being caused by malnutrition rather than the cancer itself [ 8 ]. Thereafter, the predictive value for malignancies of combination of both systemic inflammation and nutritional status were increasingly investigated, such as Glasgow prognostic score (GPS) which defined as the combination of C-reactive protein (CRP) and albumin [ 9 ], Inflammatory-Nutritional Index (INI) which defined as ratio of albumin to CRP [ 10 ]. However, such kind of indexes were really scarce and dubious, and they were not tested routinely in clinical practice, especially in NSCLC.
As for systemic inflammation, cancer can induce local or systemic inflammation, which is mediated by activation of transcription factors and production of major inflammatory cytokines [ 11 ]. On the other hand, cancer-related inflammation can influence cell proliferation, cell survival, angiogenesis, tumor cell migration, invasion and metastasis of adaptive immunity [ 11 ]. Admittedly, neutrophil which is the most common and indispensable component of inflammation plays very important role in inflammatory tumor microenvironment [ 12 ]. Some studies have revealed derangement in the full blood count such as neutrophilia, was known poor independent prognostic factor for many solid tumors [ 13 – 15 ]. In addition, CRP which is the most widely accepted proxy for systemic inflammation was also identified as a prognostic factor in both advanced and resectable NSCLC [ 16 , 17 ]. However, CRP is not commonly used in clinic because of low sensitivity and unconventional detection. Nutritional status of cancer patients, commonly reflected by serum albumin, is also a determinant of survival in many kinds of cancer. Hypoalbuminemia, an objective parameter of malnutrition [ 18 ], has been reported as a negative prognostic factor for survival in advanced NSCLC [ 19 ], and other malignancies [ 20 , 21 ]. Nevertheless, the predictive efficiency of combination of pretreatment albumin level which reflects nutritional status and pretreatment neutrophil level which reflects systemic inflammation has not been reported before.
To find an index which can roundly and systematically reflect patients’ condition, we put albumin and neutrophil together to form a new index named albumin and neutrophil combined prognostic grade (ANPG). The aim of this study was to investigate and validate our hypothesis that ANPG might be a new, convenient and more powerful predictive factor for NSCLC patients.
Patients and Methods
Patient collection
A series of 1033 consecutive patients who underwent surgical treatment for primary NSCLC between January 2006 and September 2011 at the Department of Oncology of Shandong Provincial Hospital affiliated to Shandong University were retrospectively identified from the original electronic databases. Patient collection based on the following inclusion criteria: (1) patients who were pathologically diagnosed with NSCLC; (2) patients who underwent completely tumor resection; (3) patients who had complete serum albumin and neutrophil records before treatment within 1 week and (4) patients who had complete follow-up data. Patients were excluded who matched any of the following criteria: (1) patients who had ongoing non-cancer related inflammation, immunity disease or end-stage liver disease within 1 week; (2) patients who underwent previous chemotherapy and/or radiation therapy; (3) patients whose data were incomplete; (4) patients with previous or concomitant other cancers.
This research was approved by the Ethical Committee of Shandong Provincial Hospital affiliated to Shandong University and written informed consent was obtained by participants for their clinical records to be used in this study.
Clinical and follow-up data collection
Clinical characteristics including gender, age, histological subtype, degree of tumor differentiation, pathological TNM stage (pTNM), postoperative radiotherapy and/or chemotherapy, pretreatment albumin and neutrophil count were recorded for all patients. After surgical treatment, all patients were regularly followed-up by phone interview. The pTNM was carried out according to the 8th edition of the TNM classification [ 22 ]. The degree of tumor differentiation was obtained from pathological report.
Progression free survival (PFS) was calculated from the time when firstly definite diagnosis to progression or death of any cause. Overall survival (OS) was measured from the date on which the first time definite diagnosis was made or the date of surgery until the date of death for any cause, the date of loss to follow-up or the date on which the patient was last known to be alive.
Allocation of ANPG
Blood laboratory data, especially for albumin and neutrophil, was obtained before commencement of treatment. Three binary classification methods of pretreatment albumin and neutrophil were designed and the different cut-off points were as follows. (1) According to receiver operating characteristic (ROC) curve, the optimal cut-off values of pretreatment albumin and neutrophil were 42.55g/L and 2.895×10 9 /L not merely for OS but also for PFS, respectively. (2) The quartile values of pretreatment albumin and neutrophil were 43.80g/L and 3.070×10 9 /L, respectively. (3) The clinical reference values of pretreatment albumin and neutrophil were 35.00g/L and 7.000×10 9 /L, respectively.
The ANPG was calculated into 3 grades according to dichotomization of pretreatment albumin and neutrophil. Grade1 = elevated albumin and low neutrophil; grade2 = low albumin and low neutrophil, as well as elevated albumin and elevated neutrophil; grade3 = low albumin and elevated neutrophil.
Statistical analyses
Statistical analyses were calculated by SPSS (version 20.0) software program (SPSS Inc., Chicago, IL, USA). Descriptive statistics were utilized to describe the characteristics of the study cohort. The optimal cut-off points were found out from ROC curve which is now widely recognized as the best approach for measuring the quality of diagnostic system. To ensure the best accuracy and diagnostic effect, the optimal cut-off value was located at the point on which the maximum absolute value [sensitivity-(1-specificity)] was calculated out. Kaplan-Meier (K-M) method was performed to determine the significance of variables for OS, PFS, and the log-rank test was utilized to examine the significant differences of survival distributions between different levels of albumin, neutrophil and ANPG. The Cox proportional hazards regression models were used for confirming the independent predictors of OS and PFS and multivariate Cox analyses were performed in a step-forward logistic regression approach. A two tailed p -value≤0.05 was considered to indicate a statistically significant difference.
Results
Patients’ characteristics
According to the inclusion criteria, a total of 1033 patients from the original files were finally retrospectively enrolled in our study. All samples were primary NSCLC patients who were pathologically diagnosed. Of these, the mean age was 59.18 years (range 20 to 83 years), and there were 745 (72.1%) patients ≤65years and 288 (27.9%) patients >65years, with 741 (71.7%) males and 292 (28.3%) females. Mean albumin was 41.3 (range from 25.2 to 48.3) g/L and mean neutrophil count was 4.28 (range from 1.01 to 12.32) ×10 9 /L. According to ROC curve, the number of patients with low and high level pretreatment albumin was 614 (59.4%) and 419 (40.6%), respectively; low and high level pretreatment neutrophil was 205 (19.8%) and 828 (80.2%), respectively; grade1, grade2 and grade3 of ANPG was 93 (9.0%), 438 (42.4%) and 502 (48.6%), respectively. According to quartile values, 782 (75.7%) were low albumin level and 251 (24.3%) were high; 259 (25.1%) were low neutrophil level and 774 (74.9%) were high; 75 (7.3%) achieved grade1, 360 (34.8%) achieved grade2 and 598 (57.9%) achieved grade3 for ANPG. According to clinical reference values, only 61 (5.9%) patients were low albumin level; only 68 (6.6%) patients were high neutrophil level and only 7 (0.7%) patients achieved grade3 for ANPG. All of the main baseline characteristics are detailed in Table 1 .
10.1371/journal.pone.0144663.t001
Table 1 Baseline characteristics of all 1033 NSCLC patients.
Characteristic
Data (%)
No. of patients
1033 (100)
According to the optimal cut-off values
Pretreatment albumin level (mean±sd, g/L)
41.3±3.67
Low
614 (59.4)
High
419 (40.6)
Pretreatment neutrophil level (mean±sd, 10 9 /L)
4.28±1.66
Low
205 (19.8)
High
828 (80.2)
ANPG
Grade1
93 (9.0)
Grade2
438 (42.4)
Grade3
502 (48.6)
According to quartile values
Pretreatment albumin level
Low
782 (75.7)
High
251 (24.3)
Pretreatment neutrophil level
Low
259 (25.1)
High
774 (74.9)
ANPG
Grade1
75 (7.3)
Grade2
360 (34.8)
Grade3
598 (57.9)
According to clinical reference values
Pretreatment albumin level
Low
61 (5.9)
High
972 (94.1)
Pretreatment neutrophil level
Low
965 (93.4)
High
68 (6.6)
ANPG
Grade1
911 (88.2)
Grade2
115 (11.1)
Grade3
7 (0.7)
Age (mean±sd, years)
59.18±9.69
≤65years
745 (72.1)
>65years
288 (27.9)
Gender
Male
741 (71.7)
Female
292 (28.3)
Postoperative radio-chemotherapy
Yes
325 (31.5)
No
708 (68.5)
pTNM
Stage I,II
721 (69.8)
Stage III
312 (30.2)
Differentiation
Well or moderate
716 (69.3)
Poor or undifferentiated
317 (30.7)
Histological subtype
Adenocarcinoma
531 (51.4)
Squamous cell
428 (41.4)
Other
74 (7.2)
OS (mean±sd, month)
44.29±23.63
PFS (mean±sd, month)
36.47±26.72
Abbreviations: OS,overall survival; PFS, progression free survival; HR, hazard risk; CI, confidence interval; ANPG, pretreatment albumin and neuprophil combined prognostic grade; pTNM, pathological TNM stage; sd, standard deviation.
Survival analyses in pretreatment albumin, neutrophil and ANPG
The median PFS of all patients was 31 months [mean ± sd (standard deviation), 36.47±26.72] and the median OS was 45 months (mean ± sd, 44.29±23.63) during all patients’ follow-up period. 449 (43.5%) patients died and 563 (54.5%) patients made progression. K-M curves of pretreatment albumin, neutrophil levels and ANPG for OS and PFS according to optimal cut-off values, quartile values and clinical reference values were shown in Fig 1 , Fig 2 , Fig 3 , respectively. Patients with high pretreatment albumin had a significantly better OS and PFS than low group ( p <0.05) while it lost its significance for clinical cut-off determination. Moreover, OS and PFS rate for patients with high pretreatment neutrophil were worse than low group (all p <0.05). There was also a significant different survival in different ANPG grades both for OS and PFS (all p <0.05).
10.1371/journal.pone.0144663.g001
Fig 1
Kaplan-Meier curves for optimal cut-off values.
The OS according to different pretreatment albumin levels, neutrophil levels and ANPG grades is shown in A, B, C, respectively. The PFS according to different pretreatment albumin levels, neutrophil levels and ANPG grades is shown in D, E, F, respectively.
10.1371/journal.pone.0144663.g002
Fig 2
Kaplan-Meier curves for quartile values.
The OS according to different pretreatment albumin levels, neutrophil levels and ANPG grades is shown in A, B, C, respectively. The PFS according to different pretreatment albumin levels, neutrophil levels and ANPG grades is shown in D, E, F, respectively.
10.1371/journal.pone.0144663.g003
Fig 3
Kaplan-Meier curves for clinical reference values.
The OS according to different pretreatment albumin levels, neutrophil levels and ANPG grades is shown in A, B, C, respectively. The PFS according to different pretreatment albumin levels, neutrophil levels and ANPG grades is shown in D, E, F, respectively.
Univariate survival analyses
All the results of univariate survival analyses were shown in Table 2 .
10.1371/journal.pone.0144663.t002
Table 2 Univariate analyses.
OS
PFS
HR (95% CI)
p value
HR (95% CI)
p value
According to the optimal cut-off values
Albumin
High level
1
1
Low level
1.699 (1.392–2.074)
<0.001
1.433 (1.205–1.704)
<0.001
Neutrophil
Low level
1
1
High level
1.714 (1.313–2.238)
<0.001
1.511 (1.206–1.893)
<0.001
ANPG
Grade 1
1
1
Grade 2
1.830 (1.162–2.882)
0.009
1.582 (1.103–2.270)
0.013
Grade 3
2.987 (1.914–4.660)
<0.001
2.206 (1.548–3.145)
<0.001
According to quartile values
Albumin
High level
1
1
Low level
1.478 (1.171–1.864)
0.001
1.329 (1.088–1.625)
0.005
Neutrophil
Low level
1
1
High level
1.384 (1.103–1.736)
0.005
1.341 (1.099–1.636)
0.004
ANPG
Grade 1
1
1
Grade 2
1.409 (0.899–2.209)
0.135
1.287 (0.884–1.873)
0.188
Grade 3
1.980 (1.283–3.056)
0.002
1.709 (1.189–2.455)
0.004
According to clinical reference values
Albumin
High level
1
1
Low level
1.356 (0.946–1.943)
0.097
1.204 (0.855–1.696)
0.287
Neutrophil
Low level
1
1
High level
1.489 (1.068–2.077)
0.019
1.470 (1.084–1.993)
0.013
ANPG
Grade 1
1
1
Grade 2
1.414 (1.081–1.849)
0.011
1.362 (1.063–1.744)
0.015
Grade 3
1.911 (0.713–5.119)
0.198
1.399 (0.523–3.743)
0.504
Postoperative radio-chemotherapy
No
1
1
Yes
1.535 (1.271–1.854)
<0.001
1.476 (1.245–1.749)
<0.001
Gender
Female
1
1
Male
1.281 (1.035–1.586)
0.023
1.215 (1.007–1.466)
0.042
Age
≤65years
1
1
>65years
1.360 (1.118–1.655)
0.002
1.195 (0.999–1.430)
0.051
pTNM
Stage I,II
1
1
Stage III
3.148 (2.611–3.796)
<0.001
2.360 (1.994–2.794)
<0.001
Differentiation
well or moderate
1
1
poor or undifferentiated
1.420 (1.171–1.721)
<0.001
1.319 (1.108–1.571)
0.002
Histological subtype
Adenocarcinoma
1
1
Squamous cell
1.248 (1.031–1.512)
0.023
1.054 (0.888–1.252)
0.546
Other
1.025 (0.695–1.511)
0.903
0.861 (0.606–1.223)
0.403
Abbreviations: OS,overall survival; PFS, progression free survival; HR, hazard risk; CI, confidence interval; ANPG, pretreatment albumin and neuprophil combined prognostic grade; pTNM, pathological TNM stage.
As for ROC cut-off determination, low pretreatment albumin level [hazard ratio (HR) = 1.699, 95% confidence interval (CI): 1.392–2.074, p <0.001], high pretreatment neutrophil level (HR = 1.714, 95%CI: 1.313–2.238, p <0.001), high ANPG (HR = 1.830, 95%CI: 1.162–2.882, p = 0.009, grade2/grade1; HR = 2.987, 95%CI: 1.914–4.660, p <0.001, grade3/grade1) were associated with worse OS. As for quartile cut-off determination, low albumin level (HR = 1.478, 95%CI: 1.171–1.864, p = 0.001), high neutrophil level (HR = 1.384, 95%CI: 1.103–1.736, p = 0.005), high ANPG (HR = 1.409, 95%CI: 0.899–2.209, p = 0.135, grade2/grade1; HR = 1.980, 95%CI: 1.283–3.056, p = 0.002, grade3/grade1) were also proved to be poor outcome factors for OS. As for clinical cut-off determination, only high neutrophil level (HR = 1.489, 95%CI: 1.068–2.077, p = 0.019) and grade2 of ANPG (HR = 1.414, 95%CI: 1.081–1.849, p = 0.011, grade2/grade1) retained significance for OS. In the analyses about PFS, the HRs for ANPG were also generally higher than pretreatment albumin and neutrophil levels according ROC and quartile cut-off determinations, implying more important prognostic value. Other identified prognostic factors for OS and PFS including postoperative radio-chemotherapy, gender, age, pTNM, degree of tumor differentiation (all p <0.05). Of note, age>65years was only a nearly univariate prognostic predictor for PFS ( p = 0.051) and histological subtype only significantly predicted for OS (HR = 1.248, 95%CI: 1.031–1.512, p = 0.023, squamous cell/adenocarcinoma).
Multivariate survival analyses
The multivariate Cox proportional regression in which variables were tested in a step-forward logistic regression approach was performed to examine independent factors for OS and PFS. Pretreatment albumin level, pretreatment neutrophil level and ANPG were respectively brought into the model with all other significant factors in univariate survival analyses. Results of the three multivariate survival analyses were successively shown in Table 3 .
10.1371/journal.pone.0144663.t003
Table 3 Multivariate analyses.
According to the optimal cut-off values
According to quartile values
OS
PFS
OS
PFS
HR (95% CI)
p value
HR (95% CI)
p value
HR (95% CI)
p value
HR (95% CI)
p value
Albumin
High level
1
1
1
1
Low level
1.645 (1.347–2.009)
<0.001
1.417 (1.191–1.687)
<0.001
1.325 (1.048–1.675)
0.019
1.259 (1.028–1.541)
0.026
Age
≤65years
1
1
1
1
>65years
1.601 (1.310–1.958)
<0.001
1.366 (1.111–1.605)
0.002
1.605 (1.311–1.963)
<0.001
1.353 (1.125–1.626)
0.001
pTNM
Stage I,II
1
1
1
1
Stage III
3.260 (2.680–3.966)
<0.001
2.367 (1.984–2.823)
<0.001
3.403 (2.812–4.119)
<0.001
2.350 (1.970–2.803)
<0.001
Postoperative radio-chemotherapy
No
1
1
*
1
Yes
1.229 (1.011–1.493)
0.038
1.249 (1.048–1.488)
0.013
*
1.245 (1.045–1.484)
0.014
Gender
Female
*
*
1
*
Male
*
*
1.278 (1.031–1.583)
0.025
*
Neutrophil
Low level
1
1
1
1
High level
1.639 (1.255–2.141)
<0.001
1.439 (1.147–1.804)
0.002
1.349 (1.076–1.692)
0.01
1.297 (1.062–1.583)
0.011
Age
≤65years
1
1
1
1
>65years
1.639 (1.324–2.001)
<0.001
1.373 (1.143–1.648)
0.001
1.665 (1.364–2.033)
<0.001
1.386 (1.154–1.663)
<0.001
pTNM
Stage I,II
1
1
1
1
Stage III
3.390 (2.802–4.101)
<0.001
2.365 (1.983–2.821)
<0.001
3.403 (2.813–4.117)
<0.001
2.366 (1.985–2.822)
<0.001
Postoperative radio-chemotherapy
No
*
1
*
1
Yes
*
1.213 (1.017–1.445)
0.031
*
1.225 (1.028–1.459)
0.023
ANPG
Grade 1
1
1
1
1
Grade 2
1.771 (1.124–2.792)
0.014
1.493 (1.039–2.145)
0.03
1.538 (0.980–2.413)
0.061
1.323 (0.908–1.928)
0.145
Grade 3
2.790 (1.786–4.359)
<0.001
2.065 (1.445–2.950)
<0.001
1.922 (1.244–2.969)
0.003
1.638 (1.139–2.357)
0.008
Age
≤65years
1
1
1
1
>65years
1.570 (1.284–1.919)
<0.001
1.318 (1.097–1.584)
0.003
1.165 (1.321–1.974)
<0.001
1.346 (1.120–1.617)
0.002
pTNM
Stage I,II
1
1
1
1
Stage III
3.385 (2.797–4.095)
<0.001
2.364 (1.982–2.820)
<0.001
3.381 (2.793–4.094)
<0.001
2.349 (1.969–2.803)
<0.001
Postoperative radio-chemotherapy
No
*
1
*
1
Yes
*
1.219 (1.022–1.453)
0.027
*
1.227 (1.030–1.462)
0.022
Abbreviations: OS,overall survival; PFS, progression free survival; HR, hazard risk; CI, confidence interval; ANPG, pretreatment albumin and neuprophil combined prognostic grade; pTNM, pathological TNM stage.
* Not in the final step of multivariate analyses.
As for ROC cut-off determination, pretreatment albumin level, pretreatment neutrophil level, ANPG were all significantly independent prognostic factors ( p <0.05) and ANPG [HR = 1.771(grade2/grade1), HR = 2.790(grade3/grade1) for OS; HR = 1.493(grade2/grade1), HR = 2.065(grade3/grade1) for PFS] presented higher HR than albumin (HR = 1.645 for OS; HR = 1.417 for PFS) and neutrophil (HR = 1.639 for OS; HR = 1.439 for PFS). Age and pTNM were also independently prognostic for OS and PFS (all p <0.05). Noteworthily, postoperative radio-chemotherapy was also an independent predictor but it had no significant impact on OS when it was analyzed with neutrophil or ANPG.
As for quartile cut-off determination, pretreatment albumin level, pretreatment neutrophil level, ANPG (grade3/grade1) were also significantly independent prognostic factors ( p <0.05). ANPG [HR = 1.538(grade2/grade1), HR = 1.922(grade3/grade1) for OS; HR = 1.323(grade2/grade1), HR = 1.638(grade3/grade1) for PFS] still presented higher HR than albumin (HR = 1.325 for OS; HR = 1.259 for PFS) and neutrophil (HR = 1.349 for OS; HR = 1.297 for PFS). Age and pTNM were also independently prognostic for OS and PFS (all p <0.05). Besides, postoperative radio-chemotherapy only had significantly independent impact on PFS instead of OS. Gender was also proved to be an independent predictor for OS when it was analyzed with pretreatment albumin level ( p = 0.025).
Pretreatment albumin level, pretreatment neutrophil level and ANPG were not included in the final step of multivariate analyses according to clinical cut-off determination (not shown in Table 3 ).
Discussion
Since high mortality and dissatisfactory improvement of lung cancer, the major prognostic factors have been searched all the time and numerous clinical indexes were identified to be significantly related to lung cancer survival. It has been reported systemic inflammation and nutritional status were closely related to NSCLC [ 7 , 23 ]. The connection between inflammation and survival of NSCLC dates back to early of 21 century [ 24 ]. After decade years, mounting reports have provided solid evidence to prove prognostic value of systemic inflammation and nutritional status which can be easily quantified and reflected by peripheral neutrophil and serum albumin [ 13 , 19 ]. In our study, we firstly took albumin and neutrophil together to evaluate whether the combination of them could present a better predictive value for NSCLC patients’ survival. Strikingly, we found ANPG not only was a strong independent predictor but also had a higher sensitivity than either of them.
Neutrophil, an important component of inflammation, is the first type of immune cell that responds to the site of infection and attack invaders directly. Actually, neutrophil guards its conventional positive character as a defender by killing not only invading pathogens but also malignancies. However, the inflammatory cells and cytokines found in tumors are more likely to contribute to tumor growth, progression, and immunosuppression than they are to mount an effective host antitumor response [ 25 ]. In 1986, Shoenfeld and colleagues found that high peripheral blood leukocyte count indicated worse prognosis in patients with non-hematologic neoplasms [ 26 ]. Thereafter, an increasing number of studies demonstrated neutrophil was related to poor outcome in multiple tumors [ 27 ], also in NSCLC [ 13 ]. These early reports were consistent with our findings in this study that high pretreatment neutrophil level was associated with worse survival in NSCLC patients. Although exact mechanism for this remains unclear, the reason may be that neutrophils can be recruited by kinds of chemoattractant mediators into tumor microenvironment then become pro-tumor N2 phenotype tumor-infiltrating neutrophil (N2-TIN) which can improve tumor progression [ 12 , 28 ]. Albumin, which is produced by liver, is usually regarded as an index of malnutrition and cachexia when decreased. Evidence suggested that inflammation could suppress albumin synthesis [ 29 ] and the progressive decrease of albumin was a consequence of systemic inflammation [ 30 ]. As another inflammatory index, hypoalbuminemia was also reported to be associated with poor survival in NSCLC [ 19 ]. We speculated the predict value of low pretreatment albumin might be due to dystrophic, lack of immunity and weak status of body. Moreover, chemotherapeutics may remain high residue and high toxicity in the blood stream because of lacking of albumin to bind to the drugs and this may also contribute to cancer mortality. However, both of pretreatment albumin and neutrophil levels have not been commonly used in clinic because of their low predict efficiency. The volatility of albumin also limits its application in clinic. Therefore, we gave a hypothesis here that taking pretreatment albumin and neutrophil together, ANPG would be a more powerful and feasible predictor for NSCLC patients. Additionally, other researches referring to this combined prognosis score is really scarce, especially for NSCLC.
In this study, we firstly explored the prognostic value of the combination of serum albumin and neutrophil in NSCLC patients. Our results suggested that when adopting ROC and quartile cut-off determinations for dichotomy of albumin and neutrophil, patients with low pretreatment albumin, elevated pretreatment neutrophil, high ANPG grade had worse OS and PFS in certain extent. Although all three of them were independently related to poor survival outcomes, ANPG offered higher HRs than the two other indexes, which supported our hypothesis that ANPG presented more powerful prognostic value in univariate and multivariate analyses.
When using clinical cut-off determination, pretreatment albumin lost its significance on predicting outcome for NSCLC, and ANPG was also disqualified for independent prognostic factor in multivariate analyses. This might be due to serious maldistribution of dichotomy and tiny sample size of low albumin level, high neutrophil level and grade3 (ANPG). Besides, for patients initially diagnosed as primary lung cancer, most haematoglgical parameters of them had not changed too much and were still in normal clinical reference range. Therefore, the clinical reference cut-off values may not be suitable for dichotomy of these patients. Even so, K-M curve and survival data indicated that there still existed worse prognostic trend for patients with low pretreatment level, high pretreatment level and high ANPG grade ( Fig 3 ).
Noteworthily, haematological tests are one of the routine kinds of tests carried out in cancer patients, and the ANPG will be easily obtained in clinic. Compared with existing factors, our study might provide a new, highly reproducible, easily obtainable, low cost and most of all more powerful index for predicting outcome and making therapeutic decisions. Even so, the authentic value of ANPG should be confirmed in clinic. Moreover, what we have discovered did not clarify precise mechanism underlying the relationship between this combined index and NSCLC prognosis. Further studies are required to address this question.
Conclusion
Our study firstly established a new index named ANPG for predicting outcome of primary NSCLC after surgical treatment by putting pretreatment albumin and neutrophil together. We not only demonstrated ANPG was an significantly independent prognostic factor for NSCLC patients and patients with high ANPG seemed to have more chance to live in poor prognosis, but also found ANPG outperformed better than either albumin or neutrophil for predicting outcome and making therapeutic decisions for NSCLC. However, the potential mechanisms and performance for clinical practice should be validated in further prospective studies.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
According to American Cancer Society, 241,740 will be diagnosed with prostate cancer (CaP) and 28,170 CaP patients were projected to die in the year 2012 in USA alone [1] . The unsatisfactory outcome of overall management (treatment strategies and prognosis monitoring) for CaP disease could be associated to the lack of a reliable prognostic serum-biomarker. Although widely used, several important caveats have been reported in serum-PSA as a prognostic biomarker [2] . For example, in some CaP cases, serum-PSA is (a) detected little if any, (b) lacks adequate sensitivity, and (c) fails to discriminate potentially significant cancers from insignificant ones [2] – [4] . PSA does not reflect cancer biology and a high risk of mistaken results [5] – [6] . Further, discrepancies in PSA as a diagnostic marker among different racial groups such as Caucasians and African-American have confounded the management of this cancer [6] – [7] . Therefore a great need persists for the development of improved serologic biomarkers in CaP, which is reliable for prognosis and diagnosis in Caucasian and African-American patients.
There is increasing evidence that polycomb group (PcG) proteins play a crucial role in cancer development and disease recurrence [8] . B-cell-specific Moloney murine leukemia virus integration site 1 (BMI1) is a well-known marker used in stem cell biology [8] – [9] . BMI1 which has an ubiquitous pattern of expression in almost all tissues is frequently upregulated in various types of human cancers [8] – [10] . We recently reviewed significance of BMI1 in the emergence of chemoresistance in various types of cancers including CaP [8] . The current study is the first clinical evidence showing that BMI1 is a secretory protein that has tremendous potential to be developed as a serum-biomarker for CaP and its prognosis in both Caucasian and African-American population. We suggest that serum-BMI1 as a biomarker would perform better than PSA. Further, BMI1 could be used as a dual biomarker in serum as well as biopsy.
Materials and Methods
Prostate tissues and Serum samples from human CaP patients
Prostatic tissues surgically harvested from human CaP patients and matching paraffin blocks were procured from Cooperative Human Tissue Network Midwestern Division, The Ohio State University (Columbus, OH). Serum samples of human CaP patients were procured from serum bank (BioServe, Beltsville, MD). Additional paraffin-embedded sections of human prostate tissues of 70 patients with normal and adenocarcinoma were obtained from the ISU Abxis Co. Ltd., (Seoul, South Korea).
Cell Lines
Cell lines originated from both Caucasian and African American mans were used in our study. Normal and immortalized prostate epithelial cell line (RWPE1), CaP cell lines (LNCaP, C42b, PC3, Du145, VCaP and PCa-2b), prostatic stromal myofibroblasts (WPMY1), colon normal epithelial cells (FHC) and colon cancer cell lines (SW480, HCT116 and HT29) and human pancreatic carcinoma cell lines PANC1 and AsPC1 were obtained from ATCC (Manassas, VA). Normal pancreatic ductal epithelia cells, premalignant Kras mutant E6E7-Ras and malignant Kras mutant E6E7-Kras-st cells were obtained from D. Paul M. Campbell (H. Lee Moffitt Cancer Center, Tampa, FL) [11] . BPH-1 cells were procured from Dr. Simon Hayward (Vanderbilt University, Nashville, TN) who developed them as described [12] . Establishment and characterization of RC77N/E, RC77T/E and E006 cells was described earlier [13] – [14] . Cells were grown in appropriate media supplemented with 10% FBS (ATCC, Manassas, VA) and 1% Penicillin-Streptomycin (Invitrogen, Carlsbad, CA) under standard cell culture conditions of 5% CO 2 in an incubator at 37°C.
Cell Selection
(a) Caucasian Cells: RWPE1 (normal), BPH-1 (non-malignant hyperplasia) and, LNCaP, C4-2B, PC-3, Du145 and VCaP representing Caucasian prostate cancer. WPMYI1 stromal fibroblasts were also used. (b) African American Cells: RC77N/E (normal), and RC77T/E, PCa-2B, E006 representing African American prostate cancer.
Antibody, Plasmids and siRNA
Monoclonal anti-BMI1 antibody was procured from Millipore (Temecula, CA). pbabe-BMI1 plasmid (BMI1-overexpressing) was a kind gift from Dr. Chi V. Dang (The John Hopkins University, Baltimore, MD). BMI1-siRNAs were commercially purchased from Dharmacon (Lafayette, CO).
Immunohistochemistry
Immunohistochemical staining was performed as described earlier [15] – [16] . Briefly, paraffin sections (to be evaluated for BMI1) were pretreated with citrate buffer (pH 6) for 10 min in a microwave for antigen retrieval. Sections were incubated with primary antibody (anti-BMI1) at a dilution of 1∶50 for 12 h at 4°C. Slides were then incubated for 2 h at room temperature with appropriate HRP-conjugated secondary antibody. Slides were developed in 3, 3′-diaminobenzidene (DAB kit, Invitrogen, Carlsbad, CA) and counter stained with hematoxylin. The stained slides were dehydrated and mounted in permount solution under cover slips.
Western blot Analysis
Immunoblots analysis was performed as described earlier [16] – [17] . Briefly, cell lysates were prepared in cold lysis buffer [(0.05 mmol/L Tris-HCl, 0.15 mmol/L NaCl, 1 mole/L EGTA, 1 mol/L EDTA, 20 mmol/L NaF, 100 mmol/L Na 3 VO4, 0.5% NP-40, 1% Triton X-100, 1 mol/L phenyl methylsulfonyl flouride (pH 7.4)] with protease Inhibitor Cocktail (Roche, Indianapolis, IN). The lysate was collected and stored at −80°C. The protein content in the lysates was measured by BCA protein assay (Pierce, Rockford, IL), as per the vendor's protocol. For Western blot analysis, 40 µg protein was resolved in 10% SDS-PAGE gels, transferred onto PVDF membranes (Millipore, Bedford, MA) and subsequently incubated in blocking buffer (5% nonfat dry milk/1% Tween 20; in 20 mmol/L TBS, pH 7.6) for 2 hours. The blots were incubated with BMI1 primary antibody, washed and incubated with HRP-conjugated secondary antibody (Sigma, Saint Louise, MO). The blots were detected with chemiluminescence (ECL kit, Amersham Biosciences, Piscataway, NJ). Equal loading of protein was confirmed by stripping the blots and re-probing with β-actin (Sigma, St. Louis, MO). Densitometry measurements of the scanned bands were performed as described earlier [16] .
Detection of Protein in cell culture media
Cells were allowed to grow up to 80% confluence in complete media. At 80% confluent level, media was discarded and cells were washed with PBS twice. After washing, cells were added with serum-free media. Cells were cultured in serum-free media for 24 h. After 24 h, media was collected and analyzed for BMI1 secretory protein by using Immuno-Slot-blot assay. The Slot-blot assay was performed as per the manufacturers' protocol (Whatman, Florham Park, NJ). Briefly, Slot-blot apparatus was assembled using Whatman filter paper and a pre-wetted nitrocellulose membrane. Next, the apparatus was connected to a vacuum pump. Slots were filled with samples (media/serum) and then drawn by vacuum (unused slots were filled with PBS). The membranes were then blocked for 2 h in blocking buffer (5% nonfat dry milk). The blots were incubated with BMI1 primary antibody, washed and incubated with HRP-conjugated secondary antibody (Sigma, Saint Louise, MO). The blots were detected with chemiluminescence (ECL kit, Amersham Biosciences).
Removal of Albumin from serum samples
Albumin was removed from human serum samples by using Albumin Removal Kit (Pierce, Rockford, IL) as per vendor's protocol. Samples containing 1000 µg of total protein were loaded onto a single removal disc, where each disc is reported to have a binding capacity of >2 mg of albumin.
Estimation of PSA protein levels by ELISA
This was performed by using human PSA-specific ELISA (Anogen, Ontario, Canada) as per vendor's protocol.
Quantification of secretory BMI1 protein in culture media
This was performed by using a BMI1-specific ELISA (Antibodies-online Inc., Atlanta, GA). Recombinant BMI1 protein was used to serve as standard for this assay.
BMI1-siRNA and pbabe-BMI1 (BMI1-expressing plasmid) transfection to validate that BMI1 is indeed secreted by CaP cells
Transfections were performed by using Lipofectamine (Invitrogen, Carlsbad, CA) as per vendor's protocol. For this reason, first intracellular BMI1 from Caucasian CaP (LNCaP and Du145) and African American CaP (E006) epithelial cells was determined.
Under 1 st approach
BMI1 was knocked down by shRNA in Caucasian and African American cells. 12 h after transfection, cells were grown in complete media for 12 h. After 24 h post-transfection, media was discarded and cells were grown in serum-free media for 24 h. After 24 h, serum-free media from BMI1-knockdown cells was collected and secreted-BMI1 levels were measured by ELISA.
Under 2 nd approach
prostate cancer cells representing Caucasian and African American disease were transfected with BMI1-overexpressing plasmid. 12 h after transfection, cells were grown in complete media for 12 h. After 24 h post-transfection, media was discarded and cells were grown in serum-free media for 24 h. After 24 h, serum-free media from BMI1-overexpressing cells was collected and secreted-BMI1 levels were measured by ELISA.
Androgen treatment of cells
For this reason, Caucasian and African American prostate epithelial cells were treated with androgen analogue (R1881) for 12 h. After 12 h, media was discarded and cells were grown in further 12 h. After 24 h, cells were harvested to be evaluated for intracellular BMI1 expression by western blot analysis.
Statistical analyses
Graphical summaries of the distribution of staining intensity were made using scatter plots and box plots. Simple linear egression and correlation methods were use to evaluate associations between BMI1, PSA and CaP rank (1 = normal, 2 = Stage II, 3 = Stage III, 4 = Stage IV). To correct for skewness, BMI1 and PSA were analyzed on a log (base2) scale. A p-value of <0.05 was considered to be statistically significant.
Results
Bmi1 protein levels in prostatic tissues increases with progressive stages of disease in transgenic TRAMP mouse models
Glinsky et al. [18] previously showed that Bmi1 protein is elevated in the prostatic tissues of TRAMP mice, an autochthonous mouse model of CaP development, we investigated if a progressive increase in the levels of Bmi1 in prostatic tissues could be detected during progressive age of CaP. For this purpose we used prostatic tissue samples collected at different ages of TRAMP transgenic mice. As shown in Fig 1A ; Bmi1 protein was observed to be detectable in all ages of TRAMP mice. In general, the staining was stronger in prostatic epithelial cells from older mice than in prostatic epithelial cells from younger mice. Smooth muscle cells have much stronger staining than fibroblast cells. The staining pattern of Bmi1 protein was compared in age 17 weeks to 45 weeks old prostatic specimens ( Fig. 1A ). These data showed increased expression levels of Bmi1 protein in prostate of older aged mice ( Fig. 1A ). There was an intense staining at apical region of epithelial cells. Stromal regions were observed to have a positive staining ( Fig. 1A ).
10.1371/journal.pone.0052993.g001
Figure 1
BMI1 protein levels in prostatic tumor tissues of humans and TRAMP transgenic mice.
( A ) Photomicrographs represent immunostaining of BMI1 in prostatic tissues of transgenic TRAMP mice. Arrows indicate staining for BMI1. Magnification ×40. ( Bi ) Photomicrographs show BMI1-positive neoplastic and non-neoplastic regions of prostatic specimens of CaP patients as assessed by immunostaining. Arrows indicate staining for BMI1. Magnification ×40. ( Bii ) Box plots for BMI1 protein based on score pertain to immunostaining pattern in normal and CaP specimens in stromal region.*, P<0.05; black bar in box, median values.
BMI1 protein expression in prostatic tissue specimens of CaP patients
Notably, some epithelial cells of transgenic mouse prostate epithelial cells showed dense apical staining suggesting that Bmi1 could be a secretory protein. We next identified the expression of BMI1 in human CaP specimens by immunohistochemical analysis and determined its expression levels in stromal regions of 70 pair-matched specimens of normal and CaP representing all tumor stages. The intensity of immunoperoxidase staining for BMI1 was scored as 0 (negative), 1 (weak), 2 (moderate) and 3 (strong). Immunostains showed staining in both non-neoplastic and neoplastic stroma. In general, the staining was stronger in neoplastic stroma than in non-neoplastic stroma. The epithelial cells also showed positive staining for the antibody ( Fig. 1Bi ). Smooth muscle cells have much stronger staining (3) than fibroblast cells (0–1+). The staining pattern of BMI1 protein was compared in stage II–IV CaP specimens ( Fig. 1Bi ). These data showed increased expression levels of BMI1 protein in high grade tumor in human CaP ( Fig. 1Bi ). The box plots of the data for BMI1 protein expression in stroma exhibited a wide inter-specimen variation in cancer specimens, compared with normal tissues and revealed a significant difference in the level of protein between normal and CaP tissues (p<0.05, Fig. 1Bii ). The average score for the staining intensity of BMI1 in stroma of normal tissues was 0.81±0.07 (n = 70), and was significantly lower than high-grade stage II (1.8±0.08; n = 36), stage III (2.26±0.10 n = 28) and stage IV (2.8±0.11; n = 6) cancer specimens ( Fig. 1Bii ; p<0.05). A similar pattern of staining in pair-matched CaP specimens was observed in the epithelial of the prostatic specimens. Taken together, these data show that expression of BMI1 increases with increasing stage of CaP.
BMI1 expression in normal and neoplastic prostatic cells representing CaP disease in Caucasian men
As an attempt towards identifying the expression of BMI1 in CaP progression, we first measured protein expression levels by immunoblot analysis in several human Caucasian CaP cell lines, LNCaP, Du145 and PC3 and compared them to NHPE (normal primary prostate epithelial cell) and RWPE1 (representing normal immortalized prostatic epithelial cells), respectively. Among the CaP cell lines used, LNCaP is androgen-dependent whereas Du145 and PC3 are androgen-independent. The choice of these cells was based on the fact that 80% CaP patients present with androgen-dependent disease at the time of diagnosis which later transforms into more aggressive, androgen-independent disease [19] . As shown in Figure 2(Ai–ii ), all CaP cell lines exhibited a higher expression of BMI1 protein than in normal prostate epithelial cells. When the protein expression of BMI1 was compared, based on the densitometric analysis of the immunoblots, highly aggressive PC3 cells and Du145 exhibited higher expression than in LNCaP cells ( Fig. 2Aii ). Interestingly, we also detected BMI1 expression in the prostate stromal cells (WPMY1) ( Fig. 2Ai–ii ). Interestingly, BMI1 expression was found to be very low in prostate epithelial cells representing benign prostatic hyperplasia (BPH) condition (data not shown).
10.1371/journal.pone.0052993.g002
Figure 2
BMI1 protein levels (in both intracellular and secretory forms) correlate to the aggressiveness of tumor cell type representing Caucasian and African American CaP disease.
( Ai ) Figure represents the level of BMI1 protein in normal and CaP cells of Caucasian origin as assessed by immunoblot analysis. Equal loading of protein was confirmed by reprobing immunoblot for β-actin. The blot shown here are representative of three samples. ( Aii ) Histogram showing the densitometry analysis of immunoblots of BMI1. *, P<0.05; black bar in gray box, median values. ( Bi ) Figure represents the level of BMI1 protein in normal and CaP cells of African American origin as assessed by immunoblot analysis. Equal loading of protein was confirmed by reprobing immunoblot for β-actin. The blots shown here are representative of three samples. ( Bii ) Histogram showing the densitometry analysis of immunoblots of BMI1. *, P<0.05; black bar in gray box, median values. ( C ) Figure represents the detection of BMI1 in conditional culture medium of different cells as assessed by Slot-blot analysis. The blots data shown here are representative of three samples. ( D ) Detection of secreted BMI1 protein in conditioned culture medium of cells. Each bar in the histogram represents mean ± SE of 3 independent experiments, *represents P<0.05.
BMI1 expression in normal and neoplastic prostatic cells representing CaP disease in Africa- American men
Age-adjusted data from SEER study showed that African-American men have a 60% higher incidence and 125% higher mortality rates from CaP than Caucasian men [1] , [13] . Race and family history are the two most widely accepted risk factors for this disease [1] . Understanding the underlying biological mechanisms responsible for CaP progression will eventually lead to the development of more effective therapeutic strategies. We determined the levels of BMI1 expression in a cell-based in vitro model representing different phenotypes of CaP disease in African-American men. These include RC77N/E (representing normal prostatic epithelial cells in African-American men), RC77T/E (representing androgen-dependent tumorigenic prostatic epithelial cells), E006 (representing androgen-dependent non-tumorigenic prostatic epithelial cells) and PCa-2b (representing CRPC phenotype; however retain androgen responsiveness) [13] – [14] . As shown in Figure 2(Bi–ii ), all CaP cell lines RC77T/E, PCa2b and E006 exhibited a higher expression of BMI1 protein than in normal cells RC77N/E. These data ( Fig. 2A–B ) suggest a possibility that expression of intracellular BMI1 protein may be correlated with the secretory BMI1 levels in human tissues and may play a role in aggressiveness of human CaP.
BMI1 is a secretory protein: Detection in serum-free media from CaP cell cultures
The presence of BMI1 in the apical region of prostate epithelial cells and stromal region prompted us to hypothesize that it could be a secretory protein in nature. To test our hypothesis we asked if BMI1 is secreted by tumor cells under culture conditions. In order to detect BMI1 protein in culture media of CaP cells, we employed Slot-blot technique. Cells at a confluency level of 80% were allowed to grow in fresh serum-free media for 24 h. As evident from Fig. 2C , serum free media (harvested from CaP cells culture) tested positive for BMI1 protein. Notably, media collected from cultures of epithelial cells representative of normal and BPH condition exhibited very low BMI1 protein ( Fig. 2C ).
Quantification of secretory BMI1 in culture media of cells representing CaP in Caucasian and African-American men
By employing a human specific BMI1-ELISA technique, we were able to detect and quantify BMI1 protein secreted by cells representing CaP in Caucasian and African-American men ( Fig. 2D ). We determined secreted BMI1protein levels (a) in the culture media of normal, BPH1, and (b) in the culture media of tumor cells representing various cancer types. BMI1 was detected in the culture media of normal prostate cells (RWPE1; 0.45 ng/ml media) and interestingly the levels of BMI1 were not elevated in BPH1 cells ( Fig. 2D ). As compared to normal RWPE1 cells, CaP cells exhibited increased secretory BMI1 protein levels in media ( Fig. 2D ). LNCaP, C42b, PC3 and Du145 cells were observed to secrete BMI1 protein in a range of 1.3–3.4 ng/ml of media ( Fig. 2D ). It is noteworthy that BMI1 secreted protein was observed in the serum-free culture media of all types of CaP cell lines representing from normal RWPE1 to lesser aggressive LNCaP to castration-resistant prostate cancer (CRPC) cells C42b through highly aggressive Du145 and PC3 cells. This finding corroborates with the data obtained CaP patients representing progressive stages of disease who were analyzed for serum-BMI1 protein levels. Notably, media collected from the cultures of epithelial cells E006 (derived from African American CaP patient) exhibited significantly high BMI1 protein ( Fig. 2D ). On the contrary, the culture media of prostate stromal cells (WPMY1), normal colon epithelial cells (FHC) and normal pancreatic ductal epithelial cells (PDE) did not exhibit any secreted BMI1 levels (data not shown). Interestingly, secreted BMI1 levels were not to be observed in all types of pancreatic (Kras-mutant PDE, E6E7-Ras and E6E7-Ras-st) and colon carcinoma cell lines (SW480, HCT116), but only in highly aggressive pancreatic cell lines AsPC1 (at very low levels; data not shown) and colon HT29 cells (data not shown). The ELISA data of secretory BMI1 conforms to our observations in immunhistochemical analysis of CaP tissue specimens where we observed an increased stromal staining for BMI1 protein. This would be the first report showing BMI1 as a secretory protein from tumor cells.
Secreted BMI1 in the culture media is directly related to intracellular BMI1 of tumor cells
Since BMI1 was observed to secrete in the culture media, we sought to determine if this secretion is related to intracellular BMI1. We employed a two-way approach where BMI1 was either knocked-down or overexpressed in CaP cells. After 24 h post transfection, BMI1-suppressed and BMI1-overexpressed cells were cultured in the serum-free media for further 24 h. Next, serum free media from transfected cell cultures were harvested and analyzed for BMI protein by employing an ELISA. BMI1 protein levels were observed to be highly reduced in BMI1-knocked-down cells and increased in BMI1-overexpressed cells ( Fig. 3A–F ; p<0.05). These data show that BMI1-silenced tumor cells significantly secrete low levels of BMI1 protein and BMI1-overexpressed CaP cells secreted significantly high levels of BMI1 protein in the culture media, the data suggest that intracellular BMI1 is directly correlated with the secretory BMI1 protein levels ( Fig. 3A–F ; p<0.05). We speculate that increase in the intracellular BMI1 levels in CaP cells amounts to its subsequent release by epithelial cells into the extracellular space and causes a spike in the secretory BMI1 protein levels.
10.1371/journal.pone.0052993.g003
Figure 3
Secretory BMI1 is correlated with its intracellular levels in prostatic tumor cells and is independent of androgen.
( A–F ) Figure represents the effect of (A– C ) BMI1-silencing and ( D–F ) BM11-overexpression on the level of secreted BMI1 protein in conditional media of different cells as assessed by ELISA assay. Equal loading of protein was confirmed by reprobing immunoblots for β-actin. Each bar in the histogram represents mean ± SE of 3 independent experiments, *represents P<0.05. ( Gi ) Figure represents the level of BMI1 protein in androgen (R1881) treated and non-treated CaP cells as assessed by immunoblot analysis. Equal loading of protein was confirmed by reprobing immunoblot for β-actin. ( Gii ) Histogram showing the densitometry analysis of immunoblots of BMI1. *, P<0.05; black bar in gray box, median values.
BMI1 expression in cells representing CaP in Caucasian and African-American men is independent of influence of androgen
The differences between races in androgen concentrations and sensitivity are considered as important factors for the racial disparities in CaP [20] . However, androgen concentrations do not always correlate to PSA in cancer patients and sometimes mislead the outcome [21] . We next asked if the BMI1 levels in humans CaP disease has a correlation with presence or absence of androgen. For this purpose we selected VCaP (representing androgen-independent CRPC phenotype in Caucasian population), E006 (representing androgen-dependent non-tumorigenic prostatic epithelial cells from African-American population) and PCa-2b (androgen responsiveness CRPC cells from African-American population). Androgen treatment (R1881; 1 nM) of VCaP, E006, and PCa-2b cells did not cause significant change in the levels of BMI1 protein ( Fig. 3Gi–ii ; p<0.05) thus suggesting that BMI1 expression is independent of androgen status. This data is significant because aggressive CaP in both Caucasian and African-American is often Androgen independent [6] , [19] .
Detection of BMI1 protein in blood of human CaP patients
Since, BMI1 protein was observed to be secreted by human prostatic epithelial cells in vitro. We next asked if BMI1 could be detected in the serum of CaP patients. By employing Slot-blot analysis, we determined the levels of BMI1 protein in albumin-free cleared sera, prepared from human blood (randomly selected from normal and CaP patients). As evident from the Fig. 4A , BMI1 protein was detected in the serum of CaP patients.
10.1371/journal.pone.0052993.g004
Figure 4
Measurement of serum-BMI1 protein levels in human CaP patients.
( A ) Figure represents the detection of BMI1 in human serum as assessed by Slot-blot analyses. The blot data shown here are representative of three samples. ( B ) Plot of BMI1 (ng/ml, log-2) versus CaP group rank (n = 58). Horizontal line is the group mean. ( C ) Plot of PSA (ng/ml, log-2) versus CaP group rank (n = 58). Horizontal line is the group mean. Each group (Fig. F & G) represented as 1 = Normal, 2 = Stage II, 3 = Stage III, and 4 = Stage IV CaP. ( D ) Figure represents the correlation between serum-PSA (ng/ml, log-2) and serum-BMI1 (ng/ml, log-2) (Spearman r = 0.58, p<0.001) in 58 men. Line is from simple linear regression.
Serum-BMI1 protein levels increase progressively with CaP development in human patients
We next asked if serum-BMI1 protein levels bear translational relevance as a potential biomarker for staging and development of CaP disease in humans. For this purpose we investigated if serum-BMI1 protein levels exhibit a significant difference with respect to different stages of CaP. We determined serum-BMI1 levels in a cohort of 58 human subjects representing normal disease free condition, and different CaP stages, viz., normal (n = 10), Stage II CaP (n = 16), Stage III CaP (n = 15), and Stage IV CaP (n = 17). The average serum-BMI1 protein levels in normal human subjects (n = 10) were estimated to be approximately 1.72±0.30 ng/ml of serum ( Table 1 ). The serum BMI1 level for each patient is provided in Table 2 . Serum-BMI1 levels were lower in normal human subjects than in CaP patients. BMI1 protein levels in human CaP patients was 3.91±0.60 ng/ml in stage II CaP, 8.55±1.95 ng/ml in stage III CaP and 10.84±2.44 ng/ml in stage IV CaP ( Table 1 ). These data showed that mean serum-BMI1 protein levels were progressively increased with increasing stage of CaP disease in humans (r = 0.72, p<0.001, Fig. 4B ). These data suggest that serum-BMI1 protein levels possess a translational potential to be developed as a novel serum-biomarker for CaP disease however further studies in a large cohort of patients are warranted.
10.1371/journal.pone.0052993.t001
Table 1
Serum-BMI1 protein levels in human prostate cancer patients.
ng/ml serum
Stage
Number of Human Subjects
PSA (mean ± SE)
BMI1 (mean ± SE)
Average GS
Normal
10
2.60±0.54
1.72±0.30
None
Stage II
16
12.82±9.67
3.91±0.60 *
6.0±0.092
Stage III
15
16.77±3.91 *
8.55±1.95 *
6.9±0.12
Stage IV
17
38.04±12.15 *
10.84±2.44 *
7.8±0.40
GS represents Gleason score;
*
Represents p<0.05.
10.1371/journal.pone.0052993.t002
Table 2
Comparative analysis of serum-PSA and serum-BMI1 in prostate cancer patients vis-à-vis Gleason score.
Therapy
ng/ml serum
S.N.
Age
Stages
GS
DA
MS
CT
RT
HT
Serum collected during
CaP Type
PSA
BMI1
1
78
-
-
-
-
-
-
-
-
-
1.18
0.81
2
57
-
-
-
-
-
-
-
-
2.8
3.2
3
66
-
-
-
-
-
-
-
-
-
4.43
2.32
4
70
-
-
-
-
-
-
-
-
3.63
3.09
5
65
-
-
-
-
-
-
-
-
-
2.01
0.97
6
80
-
-
-
-
-
-
-
-
2.02
0.97
7
78
-
-
-
-
-
-
-
-
-
0.29
0.94
8
57
-
-
-
-
-
-
-
-
2.45
1.02
9
66
-
-
-
-
-
-
-
-
-
2.72
2.37
10
70
-
-
-
-
-
-
-
-
4.51
1.46
11
73
II
3+3
71
N
N
Y
N
Remission
AC
10.38
3.83
12
73
II
3+3
51
N
N
Y
Y
Treatment
AC
11.27
4.39
13
83
II
3+3
77
N
Y
N
Y
Treatment
AC
2.51
10.96
14
74
II
3+3
73
N
N
Y
N
Treatment
AC
13.51
4.24
15
70
II
3+3
57
N
N
Y
Y
Remission
AC
1.48
3.72
16
82
II
3+3
79
N
N
Y
Y
Treatment
AC
13.39
4.83
17
72
II
3+3
71
N
N
N
Y
Remission
AC
0.57
2.16
18
72
II
3+2
51
N
N
Y
N
Treatment
AC
6.62
3.04
19
81
II
3+3
77
N
Y
Y
Y
Treatment
AC
10.56
2.24
20
57
II
3+3
57
N
N
N
N
Detection
AC
1.91
7.80
21
60
II
3+3
60
N
N
N
N
Detection
AC
4.62
2.37
22
64
II
3+3
60
N
N
N
Y
Treatment
AC
3.75
2.61
23
75
II
4+3
75
N
N
N
N
Detection
AC
2.72
1.37
24
47
II
3+3
47
N
N
N
N
Detection
AC
14.45
2.89
25
58
II
3+3
58
N
Y
N
Y
Treatment
AC
19.91
2.35
26
63
II
3+3
63
N
N
N
N
Detection
AC
25.33
3.67
27
72
III
3+4
66
N
N
Y
Y
Remission
AC
13.04
5.26
28
78
III
4+3
77
N
N
N
Y
Treatment
AC
35.87
17.52
29
67
III
3+3
62
N
N
Y
N
Remission
AC
7.02
15.49
30
73
III
3+4
67
N
N
Y
Y
Remission
AC
3.30
13.23
31
77
III
4+4
73
N
N
N
Y
Treatment
AC
20.06
15.30
32
69
III
4+3
64
N
N
Y
Y
Remission
AC
23.16
9.39
33
70
III
3+4
66
N
N
Y
Y
Remission
AC
21.83
10.01
34
73
III
3+3
77
N
N
N
Y
Treatment
AC
33.35
6.79
35
65
III
4+3
62
N
N
Y
N
Remission
AC
12.29
7.95
36
68
III
3+4
66
N
N
N
Y
Treatment
AC
6.45
4.56
37
70
III
3+4
69
N
N
N
Y
Treatment
AC
7.70
4.37
38
60
III
3+4
60
N
N
Y
N
Treatment
AC
12.72
3.86
39
54
III
3+4
54
N
N
N
N
Detection
AC
23.18
7.51
40
65
III
4+3
65
N
N
N
N
Detection
AC
13.54
3.83
41
57
III
4+3
57
N
Y
N
Y
Treatment
AC
18.08
3.13
42
69
IV
3+3
66
N
N
N
Y
Remission
AC
35.96
14.02
43
85
IV
4+4
80
N
N
N
Y
Treatment
AC
57.67
17.69
44
52
IV
9+0
51
Li, L
Y
Y
Y
Treatment
AC
32.07
12.21
45
68
IV
3+4
64
N
N
Y
Y
Remission
AC
57.64
14.29
46
80
IV
4+4
76
N
N
Y
Y
Treatment
AC
74.57
16.34
47
57
IV
6+4
53
Li
Y
N
Y
Treatment
AC
67.97
5.59
48
76
IV
3+3
65
N
Y
N
Y
Treatment
AC
89.70
4.33
49
74
IV
2+1
74
B
N
Y
Y
Treatment
AC
25.91
6.71
50
73
IV
6+2
62
B, L
N
Y
Y
Treatment
AC
17.27
13.23
51
70
IV
4+5
70
N
N
N
N
Treatment
AC
71.32
24.89
52
47
IV
4+4
46
N
Y
N
Y
Treatment
AC
6.78
14.09
53
60
IV
4+5
60
N
N
N
N
Detection
AC
0.25
3.50
54
59
IV
3+5
59
N
N
N
N
Detection
AC
11.0
6.58
55
72
IV
4+5
72
N
N
N
N
Detection
AC
13.16
4.79
56
63
IV
5+4
63
N
N
N
N
Detection
AC
4.02
5.39
57
75
IV
4+3
73
N
N
Y
N
Treatment
AC
10.35
6.30
58
75
IV
4+4
64
N
N
N
Y
Treatment
AC
71.16
14.41
N represents NO; Y represents Yes; AC represents adenocarcinoma; CaP represents prostate cancer; GS represents Gleason score; DA represents diagnosis age; MS represents metastatic site; CT represents chemotherapy; RT represents radiation therapy, HT represents hormonal therapy; L represents lung; Li represents Liver ; B represents bone; Bold and italic represent values in patients with low PSA and high BMI1 levels.
Serum-BMI1protein levels were correlated with serum PSA levels
Next we investigated if an increase in BMI1 during CaP developments has a correlation with PSA levels in these patients (n = 58). In this Cohort of CaP patients, association of PSA with progression of CaP was also observed (r = 0.57, p<0.001, Fig. 4C ). The serum PSA level for each patient is provided in Table 2 . BMI1 was modestly correlated with PSA (r = 0.58, p<0.001, Fig. 4D ). BMI1 remained significantly (p<0.001) when adjusting for PSA in a regression model predicting cancer stage group.
Discussion
A prognostic biomarker provides evidence about a patient's eventual outcomes from a disease independent of a given therapy, whereas a predictive-biomarker estimates the likelihood of response/benefit to a specific therapy in a specific context [22] . PSA still remains the marker of choice for CaP diagnosis, prognosis, and active surveillance. However, PSA has several limitations [23] – [26] . For example, sipuleucel-T is known to improve survival without having an impact on early PSA levels [27] . PSA progression during CRPC therapy is reported to be prognostic for overall survival but likewise is not a surrogate for overall survival [22] . Some CaP types such as neuroendocrine tumors, produce little if any PSA and decreased secretion of PSA in patients suffering from ductal CaP has also been reported [23] , [25] . In these cases, PSA alterations do not correlate well with clinical benefit [23] , [25] . There is an unmet need to identify a robust and reliable biomarker which can detect disease progression in patients in whom PSA is not a reliable indicator. Thus, the development of biomarker(s) that can correlate with disease stage along the course of tumor progression is important for intervention and treatment of disease, especially chemoresistant CaP.
In the current study, we provide evidence that BMI1 secretory protein has high potential to be developed as a reliable serum-biomarker for human CaP. We provide compelling evidence that BMI1 protein is (i) secreted by tumor cells in greater amounts proportionate to tumor stage and grade, (ii) detectable in blood of human CaP patients in an order of increasing tumor/Gleason Score grade and, (iii) detectable in some CaP patients which exhibit very low levels. Further, we showed a good correlation (r = 0.58) between secretory-PSA and secretory-BMI1 in the serum of human CaP patients. Thus, expression of PSA, along with the detection of BMI1 in serum and biopsy tissue samples, may offer a new approach for CaP diagnosis, prognosis, and active surveillance. We suggest that serum-BMI1 could bring under surveillance some cases in which PSA levels do not correlated with disease progression.
African-American men exhibit the worst prognosis of CaP disease which could be due to several reasons [20] , [28] – [29] . It is being suggested that absence of a reliable predictive biomarker for African American CaP is one of the contributory factors for the failure of prognosis in African-American CaP patients. Clinical studies suggested significant differences and in the levels of PSA of between Caucasian and African-American CaP patients [20] . PSA is androgen-dependent and its expression is regulated by androgen receptor [30] . The difference in androgen concentrations between African-American and Caucasian is considered as important factor for the racial disparities in CaP prognosis [20] . It has been reported that androgen receptor expression is 81% higher in African-American CaP patients that in Caucasian and high androgen receptor stimulation has been considered as one of the reasons for CaP development at a younger age with rapid progress in African-American men [29] . Changes in PSA may be informative in patients treat with anti-androgen therapy. However, changes in serum-PSA do not always predict the action of therapy or the disease condition [27] , [31] – [33] . Furthermore, in neuroendocrine or small cell prostate cancer, very little or no PSA is produced, and therefore PSA changes do not correlate with disease status [33] . Thomson et al. [34] reported that CaP can be detected in approximately 15% of men with normal or low levels of total PSA level. This data questions the validity of PSA as a global serum-biomarker for men. Our study in this context is significant as we provide evidence that BMI1 could be a reliable predictive secretory biomarker for both the races especially African-American CaP. This is evident from our data we show that E006 cell (derived from African American CaP patient) does not express PSA [14] , however secrete BMI1 in culture media ( Fig. 2D ). Notably, E006 cell line also expressed intracellular BMI1 ( Fig. 2B ). This data suggest that BMI1 could be used as a biomarker for even those cases in African-American men who exhibit very low PSA levels but develop CaP disease. This corroborates to our data in Caucasian men, where we were able to detect BMI1 in patients which exhibited very low PSA levels ( Table 2 ). Furthermore, BMI was found to independent of androgen and thus, it may be very useful as a prognostic biomarker in patients with both early as well as advanced prostate cancer.
Thus, analysis of BMI1 in tissue biopsies and serum analysis may serve as a prognostic biomarker in CaP and may ultimately lead to monitoring therapeutic response during CaP treatment protocols. We suggest that BMI1 stands out as a promising molecule to be developed as an ideal serum-biomarker for prognosis of CaP in humans. We suggest that this study has high translational potential however, warrants further investigation in a big cohort of human patients. It is imperative that BMI1 as a biomarker be studied rigorously in parallel with drug development (which is underway in our laboratory), given the potential to maximize benefit and management of CaP disease in Caucasian as well as African-American patients, that in turn will minimize the harms and costs to society.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Summary Points
Enthusiasm and hope are increasing around point-of-care (POC) diagnostics for diseases of global health importance.
The mere availability of rapid or simple tests does not automatically ensure their adoption or scale-up. A range of barriers prevent the successful use of POC testing—economic, regulatory, and policy-related, as well as user/provider perceptions and cultural barriers.
Technology as such does not define a POC test. Rather, it is the successful use at the POC that defines a diagnostic process as POC testing. Thus, the focus must be on POC testing programs, rather than POC technologies.
We discuss a framework that envisions POC testing as a spectrum of technologies (simplest to more sophisticated), users (lay persons to highly trained), and settings (homes, communities, clinics, peripheral laboratories, and hospitals).
A deeper appreciation of this diversity in target product profiles, and likely barriers in each setting, might help test developers and public health managers to identify the most impactful product and delivery model.
The Promise of Point-of-Care Testing
Point-of-care (POC) tests have the potential to improve the management of infectious diseases, especially in resource-limited settings where health care infrastructure is weak, and access to quality and timely medical care is a challenge [1] , [2] . These tests offer rapid results, allowing for timely initiation of appropriate therapy, and/or facilitation of linkages to care and referral. Most importantly, POC tests can be simple enough to be used at the primary care level and in remote settings with no laboratory infrastructure. POC tests can potentially empower patients to self-test in the privacy of their homes, especially for stigmatized diseases such as HIV [3] . In fact, home-based, over-the-counter HIV testing was approved in July 2012 by the Food and Drug Administration in the United States [4] .
Several agencies, notably, the Bill & Melinda Gates Foundation and Grand Challenges Canada, have recently announced grants for the development of new POC diagnostics for global health. Several million dollars are being invested in this area, and substantial enthusiasm and hope are increasing around POC diagnostics. Efforts are also underway to engage diagnostic and biotech companies in emerging economies such as India and China in developing new and affordable diagnostics for TB and HIV [5] . Furthermore, donors such as UNITAID clearly value the importance of good diagnostics and are actively supporting projects on TB, HIV, and malaria diagnostics [6] .
In this context, various stakeholders need deeper insights into the challenges for use and scale-up of POC testing, and a framework for thinking about the diversity of product profiles involved in POC testing, including where and how POC testing can be implemented successfully, what barriers need to be overcome, and what characteristics are necessary for impact.
Diversity of Definitions and Target Product Profiles within POC Testing
According to one textbook, point-of-care testing (POCT) can be defined as the “provision of a test when the result will be used to make a decision and to take appropriate action, which will lead to an improved health outcome” [7] . Another definition is: “patient specimens assayed at or near the patient with the assumption that test results will be available instantly or in a very short timeframe to assist caregivers with immediate diagnosis and/or clinical intervention” [8] . However, there are dozens of definitions of POCT and no accepted universal definition [9] . Regardless of the exact definition, we believe that the most critical elements of POCT are rapid turn-around and communication of results to guide clinical decisions and completion of testing and follow-up action in the same clinical encounter [10] – [12] .
Rapid turn-around of results is critical for the test results to impact clinical management (e.g., triage, referral, treatment decisions, decision to discharge, etc.). Indeed, without a clear link to a treatment or counseling plan, test results, even if rapid, are unlikely to have an impact [13] . “Rapid” can range from within seconds, to minutes, to a few hours (“while the patient waits”). At the least, results “on the same day” can still help disposition of clients with a clear plan (e.g., initiation of anti-tuberculosis or anti-retroviral therapy). Convenience to patients and care providers mainly derives from the fact that the POC diagnostic process is completed “in the same clinical encounter,” as compared to conventional testing where clients/patients may not come back for testing or go far away (or wait long) for testing. One of the biggest concerns about conventional laboratory-based testing is the long turn-around times and delays, and the resultant loss of patients from the testing and treatment pathway. This is, for example, a well-recognized concern with conventional sputum smear microscopy for tuberculosis (TB) [14] , and laboratory-based CD4 and viral load testing for HIV [15] .
We suggest that the technology as such does not define a POC test nor determine its use at the POC. Rather, it is the successful use at the POC that defines a diagnostic process as POC testing. So, it may be best to think of POC testing programs, rather than POC tests. It is how the tests are deployed or implemented in a health system that defines a POC testing program. For example, one could implement a rapid diagnostic test (RDT) or dipstick in a reference laboratory, and that will not be a POCT program. Indeed, laboratories in resource-limited countries often use RDTs, but results are often delivered after days. On the other hand, one could implement a molecular test in an out-patient clinic and successfully allow POC usage. The Xpert MTB/RIF test based on GeneXpert technology (Cepheid Inc) is one such technology that can potentially be implemented in TB clinic settings and peripheral laboratories [12] .
Thus, systems for rapid reporting of test results to care providers, and a mechanism to link test results to appropriate counseling and treatment are as important as the technology itself. If systems for reporting the results and follow-up care are not in place, then POC testing is unlikely to have an impact on clinical or public health outcomes [13] . Also, POCT programs require viable business models to ensure that they are sustainable in the real world and will actually get used. This means POC testing must fit within real-world workflow patterns and economic/incentive structures.
It is widely believed that POC tests should be equipment free, simple RDTs (that is, those that meet the “ASSURED” criteria: affordable, sensitive, specific, user friendly, rapid and robust, equipment-free, and delivered [2] ) and always be done outside of laboratories and hospitals by non-laboratorians. Such criteria were probably necessary when RDTs were introduced, but in today's context, they impose artificial restrictions on the concept of POCT. POC testing done at the point of clinical contact is preferable but not required, so long as a system is in place for rapid reporting of results that can inform clinical decisions. For example, testing at a peripheral laboratory attached to (or near) a clinic or hospital can still allow for POCT. Such POC testing can be done by hospital staff in emergency rooms, operating rooms, intensive care units, and labour wards, without waiting for results to come from laboratories, and this means POCT can also occur in hospital settings [16] . Indeed, there are successful examples of POC testing by non-laboratorians in hospital-based settings such as emergency rooms (e.g., rapid influenza testing [17] ) and labour wards (e.g., rapid HIV testing to reduce mother-to-child transmission [18] ).
Also, restricting POCT to really cheap and equipment-free tests (e.g., RDTs—also called “first-generation POC tests”) imposes barriers for use of newer technologies such as cartridge-based POC nucleic acid amplification tests (NAATs; “second-generation POC tests”) and hand-held devices such as mobile phone-based technologies (“third-generation POCTs”) [19] . These newer POCTs may not be as cheap and equipment free as RDTs and dipsticks, but may prove to be very impactful and cost-effective in the right context [20] .
Thus, we propose that it is easier to think of POC testing as a spectrum of technologies (simplest to more sophisticated), users (lay persons to highly trained), and settings (homes to hospitals). This diversity of target product profiles (TPPs) within POCT is illustrated in Figure 1 . This framework shows that POCT can be done in at least five distinct settings: homes (TPP1), communities (TPP2), clinics (TPP3), peripheral laboratories (TPP4), and hospitals (TPP5). Unique barriers may operate at each level, and prevent the adoption and use of POCTs. As shown in the schematic, there are several examples of POC tests in each of these settings. The relative importance of these settings may vary by country, and also within a country, where there may be important differences in public versus private sectors, and rural versus urban areas.
10.1371/journal.pmed.1001306.g001
Figure 1
Diversity of target product profiles, users, and settings within the spectrum of POC testing.
HBV, hepatitis B virus; HCV, hepatitis C virus; UTI, urinary tract infection; MRSA, methicillin-resistant staphylococcus aureus; C. diff, clostridium difficile; RDT, rapid diagnostic test; Strep A, group A streptococcus.
In the framework that we propose ( Figure 1 ), the type of device does not define a POC test. As mentioned, POC tests can range from the simplest dipsticks to sophisticated automated molecular tests, portable analysers, and imaging systems. The same lateral flow assay, for example, could be used across all TPPs. Hence, the device does not automatically define the TPP, although some types of devices will immediately rule out some TPPs or users. For example, conventional ELISA cannot be performed by lower level health workers or even doctors. Microscopy is another technology that requires a trained user and quality assurance mechanism—this restricts the technology to laboratories and hospitals. In general, even the simplest molecular tests require basic infrastructure such as power supply and temperature control, and are therefore unlikely to be used in TPP1–TPP3 in resource-limited countries.
Also, the end-user of the test does not automatically define a POC test. The same device (e.g., lateral flow assay), can be performed by several users across the TPPs—from untrained (lay) people, to community health workers, to nurses, to doctors, and laboratory technicians. Rapid oral-fluid–based HIV tests are a good example of such a test [9] , because it now spans the entire spectrum—from in-home testing to hospital-based testing. So, the actual user does not immediately identify the TPP, although targeting the end-user helps narrow down the type of TPP needed (e.g., lay person or lower level health worker necessarily means the simplest type of device and the most robust design). For example, in-home testing for HIV demands the simplest type of device, along the lines of a pregnancy test, and may also require telephonic counseling and support services [4] , [21] , [22] .
Depending on the end-user and the actual setting, the purpose of POC testing may vary—from triage and referral, to diagnosis, treatment, and monitoring. Some POC test users may not be empowered to prescribe medicines, while others can use POCT results for treatment. This has implications for test developers. A test that is intended for triage and referral can have different accuracy (e.g., lower specificity), compared to a test that is used to make treatment decisions [23] .
The Need to Understand Barriers for POC Testing
The best POCT will not have any impact unless it is widely used and followed-up with appropriate treatment interventions [24] . The mere availability of rapid or simple tests does not automatically ensure their adoption or scale-up [24] . This observation is evident from the global experience with rapid HIV tests and malaria RDTs [5] , [25] – [28] . In India, for example, simple RDTs are available for a variety of diseases (e.g., HIV, malaria, dengue, syphilis, hepatitis), are quite inexpensive (US$1 per test or less), and some meet all the ASSURED criteria. However, these RDTs are often not used in community or clinic settings to make clinical decisions (with the possible exception of pregnancy tests and possibly HIV and malaria RDTs). It appears that very little POCT occurs in homes, communities, or clinics (TPP1–TPP3). Testing predominantly takes place in laboratories and hospitals (TPP4 and TPP5). In fact, small, stand-alone laboratories are the biggest consumers of RDTs.
What are the most important barriers to widespread use of POCT at lower levels of the health care delivery system, where we hope POC testing will reduce diagnostic delays and interrupt transmission? On the basis of our observations and the published literature [2] , [11] , [26] – [30] , we believe there are a variety of barriers to successful use of POCT—from economic, regulatory, and policy-related barriers to user/provider perceptions and cultural barriers. Table 1 provides illustrative examples of these barriers. Some barriers are generic, while others are restricted to a specific TPP or setting. Barriers for POC testing may depend on the country, and may also differ across public versus private, and urban versus rural settings. In addition, some barriers may be disease-specific, while others will apply to all types of tests. For example, stigma and confidentiality may be important barriers for HIV testing [26] , [28] , [31] , while they may be less relevant with malaria or dengue testing.
10.1371/journal.pmed.1001306.t001 Table 1
Barriers to adoption and scale-up of POC technologies.
Barrier for POCT
Example
Economic
It may be more expensive to place test instruments at the POC, as compared to laboratories. Some POCTs may be priced at a level that is unaffordable in many countries. Private care providers may receive incentives from laboratories for each test that they order; this means they can earn more by sending their patients to labs rather than do any POC testing.
Policy-related
Existing guidelines and policy documents may not provide clear recommendations on how to include POC tests in algorithms that are in place. Lack of a strong evidence-base on POCTs can result in weak evidence and uncertain policy recommendations.
Regulatory
Poor regulation of diagnostics may result in easy availability of suboptimal and poor quality rapid tests on the market; this makes it challenging to scale up validated POCTs.
Laboratory capacity
Some POCTs may require peripheral labs with sufficient capacity to run them (e.g., nucleic acid amplification tests). Poor laboratory capacity poses a barrier for scale-up of such technologies.
Infrastructure
Clinics and primary care centers often lack infrastructure such as constant power supply, refrigerators, storage space, waste disposal units, phlebotomy supplies, and temperature control; this makes it hard to implement some types of POCTs.
Quality control and quality assurance
Even simple POC tests require quality assurance and training before they can be performed. Primary care providers may not have the expertise or training to do them with quality assurance.
Work-flow balance
Staff shortages and high workload may reduce uptake of POCT. Health care providers are overburdened with a high volume of patients, and work-flow and time constraints do not permit easy use of POC tests.
Training
Unqualified and informal care providers may lack the knowledge and training needed to implement even simple RDTs. Erroneous results then erode the health system's faith in POCT. Lack of continuous, ongoing proficiency training can result in diminishing performance of POCT programs.
Supply chain
Supply chain deficiencies can lead to suboptimal or poor quality POC tests, which, in turn, may discredit POCT.
Infection risk
Health providers may be unwilling to do tests that may expose health care workers to the risk of infection.
Administrative/operational
It is not easy for health providers to seek reimbursement from insurance providers and third-party payers when POC tests are used in community or home settings.
Technical/medical
Doctors and front-line care providers in some settings may prefer clinical diagnosis and empiric treatment over diagnostic certainty. Widespread empiric treatment of common diseases reduces the felt need for any testing, POCT or otherwise.
Awareness
Health workers and care providers may not be aware of the various tests that are now available for POC use. Thus, they may still refer their patients to laboratories for testing.
Health system-related
Laboratory professionals in hospitals and larger health care facilities are opposed about any testing that is done outside of lab settings. They fear this will impact their own business, and they also worry about relinquishing control over testing.
Fit with user needs
Available rapid tests are often single disease focused when primary care providers are more worried about syndromes of unknown etiology (e.g., febrile illness, chronic cough). So, available tests may not quite meet user needs.
Cultural/societal
Perceived lack of confidentiality and stigma may reduce acceptance of POC testing in the community (e.g., HIV rapid tests).
Why Do We Need to Understand the Diagnostic Ecosystem in Countries?
It is particularly important to look beyond the technology, and understand current diagnostic practices and the health systems within which POC testing has to get scaled up. At the country level, POC diagnostics ultimately need to be integrated within health systems, supported by financing (who will pay and how much?), incentives (do various stakeholders benefit from the economics?), training and information and communications technology (ICT). These other factors (“the business model”) may be as important as the POCT itself and need to be taken into account when developing tests [32] .
Indeed, the best POCT without a good business model is unlikely to get scaled up, while, paradoxically, inaccurate tests can become popular because of economic reasons, as illustrated by the apparent market success of inaccurate TB serological tests in many developing countries [33] , [34] . In India, we have shown that although serological TB tests are inaccurate, various players along the value chain profit from their use, and this sustains a market for these tests [34] .
ICT, when combined with POC, can help expand care to lower tiers of the health care delivery system, all the way to home-based, self-testing [30] . Thus, the rapid expansion of mobile telephony makes telephonic counseling and rapid reporting of results (to patients as well as to public health programs) feasible in many settings. Diagnostic devices linked with mobile phones can also allow for automatic data capture, external quality assurance, and proficiency testing. Smart phones can also provide decision support to health workers on what follow-up action is necessary after testing. Indeed, new TPPs and business models are now feasible, thanks to the rapid expansion of ICT. Interestingly, the mobile phone itself is becoming a POC testing device, and this is a vibrant area for incentive prizes [35] .
In India, the diagnostic ecosystem is worrisome with systematic market failures throughout the value chain for diagnostics—private doctors receiving payments or incentives for tests ordered, over-reliance on useless tests, and under-use of good diagnostics [33] , [34] , [36] – [38] . There is little quality assurance for laboratories in India and private labs offer tests of doubtful value. Laboratories in the public health sector suffer from poor infrastructure and limited capacity, while dealing with massive volumes. The regulatory framework for in vitro diagnostics in India is weak and most diagnostics do not undergo rigorous validation before approval [34] . As a result a large number of inaccurate and ineffective products can be found on the market; many of these are imported, but not approved by the US Food and Drug Administration (FDA) or other such credible regulatory bodies outside of India [34] . In fact, the Government of India has recently banned TB serological antibody tests, which are widely used in the private sector [34] .
It is within this context that we need to understand the barriers for use of POC diagnostics in India. Based on our work in India on TB and HIV diagnostics [3] , [18] , [29] , [33] , [34] , [36] , [37] , [39] – [43] , we have identified several potential barriers to implementation of POC tests by primary care providers ( Table 2 ). Additional barriers may operate at the community level in India. Much of the community-level health care in India is done by village health nurses, auxiliary nurse midwives, and Accredited Social Health Activist workers, under the National Rural Health Mission. These workers are generally not well trained or empowered to adequately use POC tests or prescribe drugs on the basis of test results (with some exceptions). In fact, the medical lobby in India vigorously prevents any move to empower health workers to prescribe drugs. Furthermore, these community health workers are heavily burdened with paperwork associated with maternal and child health-related programs and often do not have the time to conduct any testing. Furthermore, the incidence of many diseases is quite low at the community level, and this might reduce the motivation and resources for community-based testing.
10.1371/journal.pmed.1001306.t002 Table 2
Barriers for use of point-of-care tests in India.
Category
Potential Reasons Why POC Tests Are Not Being Used at the POC
Technical, administrative, and operational
Widespread empiricism in clinical practice: Doctors and front-line care providers in India generally prefer clinical diagnosis and empiric treatment (e.g., with broad spectrum antibiotics) over diagnostic certainty.
Work-flow, time constraints, and patient volume: Care providers are overburdened with a high volume of patients, and because average consultation times last for just a few minutes, the work-flow and time constraints do not permit POC testing. In the time it takes to do the POC test and read it, providers could see several waiting patients and that is more important for their popularity and reputation.
Expertise and self-confidence: Care providers in India have a diversity of backgrounds—ranging from MBBS-trained medical doctors, to unqualified providers, and those with training in alternative medicine. Thus, the ability of these providers to do any clinical testing is highly variable. Unqualified practitioners and those with non-medical backgrounds may lack the skills (or the confidence) to perform and interpret tests (even if they are simple to use). Even MBBS doctors may not want to take on the responsibility of doing and interpreting test results.
Infrastructure and support staff: A large number of care providers in India practice medicine from small, single-room clinics, with very little space for even a small side laboratory. They often practice alone with no support staff (e.g., nurse). This makes it difficult to implement even simple lateral flow tests. Clinics often lack basic equipment such as refrigerators, storage space, waste disposal units, phlebotomy supplies, temperature logs, etc.
Quality assurance training: Care providers may not have the expertise nor training to do testing with quality assurance (e.g., run positive/negative controls).
Fit with user needs: Care providers manage several commonly encountered infections (e.g., TB, malaria, dengue, and influenza), and it is not feasible for them to remember all the standard operating procedures of these tests, nor is it possible for them to perform several rapid tests on the same patients to work through their differential diagnoses.
Tests for multiple infections: Available rapid tests are often single disease focused (e.g., malaria RDTs) when primary care providers are more worried about syndromes of unknown etiology (e.g., febrile illness, chronic cough, diarrhea). In the absence of multiplexed POC tests for a panel of related infections, they find it hard to do multiple POC tests at the POC.
Investments at the primary care provider level: At the level of the single primary care provider, there is insufficient volume of each disease to make it worth their while to stock multiple RDTs. Also, care providers do not like to make any capital investments, especially in small centers.
Reputation of POCTs: Easy availability of suboptimal and poor quality rapid tests in the market may have resulted in lack of faith in them—care providers consider rapid tests to be inferior to conventional tests.
Health system-related
Awareness of POCTs: Doctors and care providers may not be aware of the various POC tests that are now available for POC use.
Professional exclusivity: Laboratory professionals in hospitals and larger health care facilities are opposed to any testing that is done outside of lab settings. They fear this will impact their own business, and they also worry about relinquishing control over testing. Laboratory professions do not believe that doctors or health care workers can do POC tests with quality assurance.
Monitoring and tracking outside of labs: Health care delivery systems struggle to audit and monitor the use of POC tests in settings outside of the laboratory. In India, medical records are poorly managed in most settings.
Impact on hospital attached labs: Hospitals and medical establishments often have their own laboratories, and to generate business for their labs, they prefer that doctors refer patients to these laboratories, instead of doing testing by themselves.
Investment in post-test counselling: Health care systems may be unwilling to invest in counselors who can provide post-test counseling services in clinic settings.
Weak capacity to absorb POC tests and insufficient treatment availability: Implementation of POCT might increase workload and demand for treatment services. Limited resources (e.g., drugs) and human resources might diminish enthusiasm for POCT.
Economic
Clinical relationship and exchange of money: The usual practice in India is for patients to pay someone else (e.g., a secretary or nurse) rather than the doctor directly. If the doctor were to perform and read the POC test, it will be awkward for them to take money from patients (unless support staff are involved).
Referral fees and incentives: Private care providers often receive referral fees (i.e., incentives) from laboratories for each test that they order; this in turn greatly limits their financial incentive to directly conduct tests.
Patients ' willingness to pay: While patients may be willing to pay for tests done at a laboratory, they may be unwilling to directly pay the doctor for testing costs, beyond the consultation fees that they are already paying.
Affordability of POCTs: Poor patients may not be able to afford any testing. They may instead prefer a prescription for drugs (which may be less expensive and more easily available compared to tests).
Traceability for insurance reimbursements: It is not easy for health providers to seek reimbursement from insurance providers and third-party payers when POC tests are used in office, community or home settings, with no invoice getting generated for each test.
Bypass of medical consultation and diagnostic process: Patients seeking quick relief from symptoms may directly buy over-the-counter antibiotics from pharmacies, and thereby completely bypass the medical consultation process.
Mark-up on POCTs: Laboratories that offer POC tests may demand a high price for these tests, making them unaffordable for poor patients.
In South Africa, discussions with TB/HIV experts suggest that much of the POC testing currently occurs in laboratories and hospitals (TPP4 and TPP5). For example, the scale-up of Xpert MTB/RIF in South Africa is happening via the National Health Laboratory Service (NHLS) network of laboratories. At the clinic level (TPP3), tests for infections like HIV, syphilis, and malaria are widely used. At the community level (TPP2), HIV rapid tests are the most widely used POC tests. HIV self-testing is known to happen at home in South Africa (TPP1), especially among health care workers who avoid conventional voluntary counseling and testing (VCT) to protect their confidentiality.
South African experts identified several potential barriers for POCT in their setting. For example, laboratory professions are concerned about widespread use of POC tests for many reasons: (1) NHLS cannot control or provide oversight to any testing that is done outside of the NHLS lab network; (2) it is not clear which agency will accept ownership of a POC testing program in South Africa—who will conduct training, quality assurance, maintenance; (3) it is unclear who will provide overall management of a decentralized POC testing program at the level of communities and clinics. Cost can also be a major barrier for POCT—a recent study from South Africa suggests that placing the Xpert MTB/RIF test at the POC will be substantially more expensive than placing the instruments in the NHLS laboratories [44] .
Overcoming Barriers to POCT Programs
POC testing holds a lot of promise, but it is important to understand the complexity and diversity of POCT, and identify the biggest barriers to successful implementation of POCT programs. This step is critical for development and scale-up of POCTs, because it will allow test developers and public health programs to target the TPP that is most likely to succeed and has the most impact.
The framework we have proposed may have utility in shaping many of the ongoing efforts to develop and deploy POC tests for global health. Firstly, test developers and manufacturers need to understand the real-world context (e.g., conditions, settings, users, resources) within which tests need to get scaled up. Only then can TPPs by test developers match the TPPs required by public health programs. Indeed, technologies may need to be designed in resource-limited settings (“frugal or reverse innovation”), from the ground up, to ensure that they are robust, field-tested in a variety of conditions, have built-in capacity for reporting/notification, and appropriately priced. Secondly, donors and funding agencies must consider the downstream implications of the health technologies that they are funding, and ensure that product development initiatives are simultaneously coordinated with pricing and delivery mechanisms, supported by innovative business models for scale-up. Lastly, health care managers must invest in POCT programs, rather than merely purchase rapid tests, and ensure the mechanisms are put in place for quality assurance, reporting of results, notification of cases, and initiation of action on the results of the tests. Only then will the true benefits of POC testing be realized.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: | Introduction Genes may vary in evolvability for many reasons, including physical susceptibility to mutagenesis. Here we show that a class of genes with distinctive physical features—heat-shock genes—is remarkably prone to mutagenesis by insertion of a specific transposable element (TE), the P element of Drosophila . TEs are mobile, repetitive DNA sequences and a structurally dynamic component of genomes [ 1 ]. TEs can cause gene and chromosome evolution in numerous ways, including insertional mutagenesis, retroposition, conveyance of regulatory elements to novel sites, and service as pivotal sites for ectopic recombination, and thus chromosomal rearrangements and gene duplication. For such evolution to occur, however, TEs must first insert into chromatin, which in turn requires that the target site be accessible to the transpositional machinery [ 2 ]. Indeed, insertion of Drosophila P elements, among the best-studied of TEs [ 3 ], into specific sites is associated with features of local chromatin architecture such as DNase I hypersensitivity, location in 5′-flanking sequence, presence of pre-existing TEs, and physical structure, but only weakly with insertion sites' nucleotide sequence (e.g., [ 4 – 6 ]). These features vary widely and frequently throughout genomes [ 7 ], which is consistent with the irregular, but repeated, occurrence of TEs. Entire classes of genes also vary in TE frequency—and hence potentially evolvability via transposition—in laboratory studies [ 8 ], but for natural populations, neither the mechanistic basis for this variation nor its relevance for evolvability is clear. In such experimental work with Drosophila, heat-shock genes (e.g., the major heat-shock gene Hsp70 ) stand out as a class receiving numerous TE insertions [ 8 – 10 ]. (By “gene,” we intend both the transcribed sequence and associated non-transcribed regulatory sequence.) This distinction is not unexpected from two perspectives. First, the local chromatin architecture of heat-shock proximal promoters is peculiar, incorporating constitutively decondensed chromatin and nucleosome-free regions [ 11 , 12 ], and constitutive engagement of the transcriptional machinery. In addition to the 5′ location of these promoters, such features should predispose these regions to TE insertion (see above). Second, TEs segregate at high frequency in natural populations in the 5′-flanking regions of the five genomic copies of Hsp70 [ 13 – 17 ]. This finding is remarkable given that TEs typically are at low allelic frequency in the Drosophila genome, presumably because they are deleterious [ 18 – 20 ]. The Hsp70 intragenic TEs are seemingly adaptive, exhibiting repeatable demographic variation in allelic frequency along natural thermal gradients and beneficial impacts on Hsp70 expression and components of fitness [ 14 – 17 , 21 ]. Nonetheless, TEs constitute 22% of the Drosophila genome [ 22 ] and are numerous (more than 6,000 elements) [ 23 ]. Thus TE insertions in heat-shock genes could simply be a manifestation of general patterns because TEs are common in the Drosophila genome, rather than indicative of a specific insertion susceptibility and/or adaptive role. To distinguish between these possibilities, we carried out an unbiased screen with both negative and positive controls. Our working hypotheses were as follows: First, because TE insertion can be mutagenic, naturally occurring transposition into Hsp70 genes could simply reflect that these are multicopy genes [ 24 ] and functionally redundant, and thus permit insertional mutagenesis of one to two copies. If so, then TEs occurring in proximal promoter regions should be restricted to multicopy genes like Hsp70 and not widespread in the “heat-shock genome” of natural populations, typically comprising single-copy genes. Second, if as a class, heat-shock genes are especially susceptible to TE insertion in their proximal promoter region, the “heat-shock genome” of natural Drosophila populations should harbor numerous TEs in this region. Accordingly, we screened for TEs in the proximal promoters of 18 heat-shock genes other than Hsp70 . This set of genes represents the prototypical heat-shock genes and cognates in Drosophila melanogaster other than Hsp70 (Gene Set I in Table 1 ). 10.1371/journal.pgen.0020165.t001 Table 1 Genes Other Than Hsp70 Screened for Transposable Element Insertions in 5′-Flanking Sequence Third, if heat-shock genes' peculiar chromatin architecture and its correlates (see above) predispose the heat-shock genome to TE insertion, then the proximal promoter regions of other genes in the Drosophila genome sharing some or all of these features should likewise harbor numerous TEs in natural populations. Accordingly, we screened for TEs in 18 non–heat-shock genes resembling heat-shock genes in relevant features (Gene Set II in Table 1 ). Finally, if heat-shock genes' chromatin architecture and its correlates predispose the heat-shock genome to TE insertion, then in natural populations, genes dissimilar to heat-shock genes should less frequently harbor TEs in their proximal promoter regions. Accordingly, we screened for TEs in the proximal promoters of a “negative control” set of 18 such genes (Gene Set III in Table 1 ). Relevant to all working hypotheses is that a TE in a gene will signify both that the TE has successfully inserted and that the TE has not (yet) been eliminated. Remobilization of TEs, their mutagenesis, and negative selection may all affect TEs' presence at a specific site. Such screens pose a substantial analytical challenge. Only a single D. melanogaster genome has presently been sequenced, and that for an isogenized laboratory strain [ 25 ]. Although the sequenced genome is typical of wild D. melanogaster with respect to many TEs, it is intentionally dissimilar with respect to others [ 26 ].Moreover, an isogenized strain obviously cannot represent variability present in natural populations. Furthermore, most attempts to characterize the “transposome” of natural Drosophila populations, whether experimentally or in silico, are sequence-based; i.e., they rely on the distinctive canonical sequences of the various TEs for TE recognition and subsequent identification of the gene (or intergenic region) in which TEs have inserted. These methods range from genomic Southern blots to TE-specific PCR to TE display to bioinformatics searches. Our objective, by contrast, is to ascertain how often specified gene regions contain TEs. Given that each region to be screened might contain one of more than 120 different TE families in Drosophila [ 23 , 26 ], a sequence-based screen specific for each possible element in numerous genes and populations would be prohibitively laborious. Furthermore, our region of interest (proximal promoter) is non-coding, which may frustrate simple PCR-based screens when highly variable. For these reasons we have exploited universal fast walking (UFW) [ 27 , 28 ], a method that can report TEs, not by their sequence, but by the size polymorphisms they create. Here we demonstrate, by applying this technique to a screen of 48 natural Drosophila populations from around the world ( Figure 1 ), that heat-shock genes as a class are a distinctive and repeatable natural target for TE insertion, as is predictable from the distinctive characteristic features of these promoters. Remarkably, of the many active TE families that might target heat-shock genes, the vast majority of the naturally occurring TEs that we discovered are P elements, notorious for their recent invasion of the D. melanogaster genome [ 29 – 31 ]. Accordingly, we conclude that the proximal promoters of heat-shock genes in Drosophila are especially conducive to transposition of P elements in nature, which creates significant variation upon which evolutionary processes may act. Furthermore, dissimilarities between frequencies of naturally occurring and experimental P element transpositions into the various classes of promoters imply that weakened purifying selection and/or positive selection may contribute to the persistence of P elements in natural populations—a suggestion that invites future testing. 10.1371/journal.pgen.0020165.g001 Figure 1 Geographic Origins of D. melanogaster Populations Screened in This Study Screens revealed zero to 14 P elements per population (indicated by the number of squares), distinctive by insertion location, in the proximal promoter regions of genes examined ( Table 1 ). Colors of squares correspond to gene set (see Introduction). Inset: Percentages of distinctive P elements discovered in Hsp70 genes and each of the three gene sets screened. A total of 161 P element insertions (the ten P elements in the coding sequence and the five non– P element insertions are not included in the figure). These tallies potentially under-report the actual number of P elements; see Results . F06 (Celera) is the strain whose genome has been sequenced [ 25 ] and is the reference strain for the present study. Populations F18, F50, and F52 (in light gray text) were removed from the analysis after screens failed for multiple genes and primer sets. Results Summary Findings and Methodology The UFW screen revealed numerous differences in amplicon size between the reference strain (Celera, F06) and the natural populations (see exemplary gel images in [ 28 ]). These polymorphisms were characterized by sequencing and/or TE-specific PCR. A total of 97% were insertions of P elements into the proximal promoters of the surveyed genes, with the balance jockey and gypsy elements ( Figure 2 ). In fact, 19 (35%) of all investigated promoters ( N genes = 55) had at least one P element insertion in the proximal promoter region in one or more of the populations investigated ( N populations = 48; Figures 1 and 3 ). Many of these insertions are into identical sites in different populations ( Figures 2 , 4 , and 5 ). Most (42 of 48) populations had a P element insertion in at least one investigated gene. 10.1371/journal.pgen.0020165.g002 Figure 2 Locations of TEs Integrating into the Proximal Promoters of Hsp70 Genes Six nearly identical Hsp70 genes are present in the sequenced Drosophila genome, but only five copies in natural populations. The locations of selected promoter elements and sites are indicated for all copies. (A) Previously discovered TEs and experimental transpositions relative to the conserved Hsp70 sequence. a, Jockey element in Hsp70Ba [ 16 ]; b, c, and d, P elements in Hsp70Ba [ 14 , 15 , 21 ]. An S element is present between the oppositely oriented paralogs Hsp70Aa and Hsp70Bb [ 13 ], and is represented twice, corresponding to its location relative to each paralog, as are the HMS Beagle (e) [ 16 ] and “ 56H8 ” (f) [ 88 ] elements inserted within it. Triangles below the line indicate transgene insertion sites (FlyBase; http://flybase.bio.indiana.edu ). (B) and (C) Bottom: newly discovered TEs, with the natural population in which they were discovered (F01–F54, exclusive of F06) indicated for each. (B) TEs other than P elements. Four are Gypsy elements that have integrated into the S element in specific populations, the fifth is a Gypsy that has inserted into a Gypsy, and the sixth is a Jockey that has inserted into a P element. The Gypsy s are arbitrarily plotted relative to Hsp70Ab and Hsp70Aa, respectively. (C) Natural P elements in Hsp70 . The arrows indicating the number of independent EPgy2 insertion sites recently described by Shilova et al. [ 10 ]. Except for the Gypsy s, TEs were not mapped to a specific Hsp70 gene. Insertion sites localized within the Hsp70 region were all established by sequencing. For population codes, see Figure 1 . 10.1371/journal.pgen.0020165.g003 Figure 3 Number of Natural P Element Insertions (161 Total) Distinctive by Population and Location into the “Proximal Promoter Region” of Each of the Screened Genes ( Table 1 ) Genes without any such insertions are not represented in the main figure. These tallies and estimates are conservative in three ways: (1) P elements inserting within 1,000 bp of the transcription start site of Bsg25D and CG6396 are included although they actually insert into neighboring genes (see Figure 4 ); (2) The tally for Hsp70 excludes non– P elements and those previously discovered ( Figure 2 ), and divides the remaining total (44, light gray bar in background) by five, the Hsp70 copy number for natural populations [ 17 ]; and (3) Re-screening of a subset of insertions implies an underestimation of the tally at the 161 P insertion sites (see Results and Figure 7 ). Inset: frequencies of genes in each Gene Set (I, including Hsp70, II, and III) in which 0, 1, or >1 P elements had inserted. 10.1371/journal.pgen.0020165.g004 Figure 4 Locations of P Elements Integrating into the Proximal Promoters of Heat-Shock Genes Other than Hsp70 (Gene Set I) Data are plotted as in Figure 2 except as follows: F-numbers in columns refer to natural populations with transposons integrating at identical sites. Primer sets used in the screens amplified regions 3′ to transcription start site of different length; P elements discovered upstream of the initiator are plotted (pale), but not included in comparative analyses (i.e., in Hsp22 in population F31, Hsp68 in F05, and in Hsrω in F14). In Hsrω , numerous P elements were discovered in one region (box); a randomly chosen subset of these were localized (by sequencing) within that region (see enlargement). Several putative deletions were also discovered, and are plotted. For population codes see Figure 1 . Table S1 provides additional information about Gene Set I and relevant sources for the organization of promoter regions. These findings prompt two methodological concerns, which are unwarranted. First, insertions common to multiple populations could be shared by descent, and hence tallying them as independent would overestimate insertion events; we exclude this possibility below. Second, the high proportion of P elements could stem from oversensitivity of the UFW screen to P elements. To address this concern, we re-screened a subset of genes known to contain both P elements and other TEs (the Hsp70 genes in each natural population) with a technique known to detect all these TEs. The re-screening used a reliable PCR with one primer complementary to Hsp70 and the other to each of the six TEs common in the Celera strain (roo, 1360, 297, Jockey, I, and Gypsy) [ 26 ]. This procedure re-detected each Jockey and Gypsy previously implicated by UFW in Hsp70, but revealed no additional TEs in any of the other genes. Two genes from each gene set (Gene Sets I–III) were likewise re-screened for four populations (F04, F40, F53, and F54), with the result entirely consistent with the previous screens. These results affirm that the method is universal, detects TEs when present, and does not favor P elements. Abundance of Transposon Insertions in the Three Gene Sets The UFW- and TE-specific PCR screens together initially detected 177 TE insertion sites (containing 171 P elements; Figures 2 , 4 , and 5 ), one Jockey inserted in a P element, and five Gypsy (one a Gypsy inserted into a Gypsy; Figure 2 ) in the 55 genes surveyed ( Table 1 ) in the 48 natural Drosophila populations ( Figure 1 ). Because the screens intentionally focused on the proximal promoter region of each surveyed gene, 167 of these insertion sites (containing 161 P elements, the Jockey, and the Gypsy s) were in this region, often near the TATA box or initiator of the associated gene (e.g., Hsp22 and Hsp27 ). The screen also included more 3′ regions for some (but not all) genes, and detected ten P element insertion sites in coding sequence (e.g., for Hsp70, Hsp22, Hsp68, Hsp83, Hsrω, su(s), act5C, and elf; Figures 2 , 4 , and 5 ) or in nearby genes (e.g., nearby bsg25D and CG6296; Figure 5 ). Because these regions were surveyed only in those genes in which the UFW primers encompassed them, P elements in these regions were excluded from the following comparative analyses. 10.1371/journal.pgen.0020165.g005 Figure 5 Locations of P Elements Integrating into the Proximal Promoters of Non–Heat-Shock Genes Resembling Heat-Shock Genes in Relevant Features of Their Proximal Promoters (Gene Set II), and in Genes Dissimilar to Heat-Shock Genes (Gene Set III) Data are plotted as in Figure 2 except as follows: Primer sets used in the screens amplified regions 3′ to transcription start site of different length; P elements discovered upstream of the initiator are plotted (pale), but not included in comparative analyses (i.e., in su(s) in population F17, in Act5C in F03 and F31, and in Elf in F43 and F54). Note that in Gene Set III, the two P elements discovered are not clearly associated with their focal genes, integrating into or just upstream of genes neighboring the focal genes. For population codes, see Figure 1 . Table 1 provides additional information about the gene sets. The data support three of our a priori expectations: (1) that novel TE insertions into Hsp70 genes should be readily discoverable ( Figure 2 ), (2) that TE insertions should be numerous in Hsp genes other than Hsp70 ( Figure 3 ), and (3) that TE insertions should be rare in genes dissimilar to Hsp genes ( Figure 3 ). A total of 29 natural populations (with at least 44 distinct insertions) harbored P element insertions into at least one of the five Hsp70 gene copies ( Figure 2 ), with 13 into the Hsp70Aa gene, four into Hsp70Ab, and the balance not localized to a specific Hsp70 gene. Hsp70 genes also harbored all of the non– P transposable elements detected by the screen (one Jockey and five Gypsy s). P element insertions were also numerous (in 30 natural populations with 37 distinct insertions) for Hsrω ( Figure 4 ), a single-copy sequence encoding a heat-inducible mRNA. Other heat-shock genes with insertions in more than five natural populations include Hsp22, Hsp23, Hsp26, Hsp27, Hsp68, and Hsp83 ( Figure 3 ). Excluding Hsp70, which presents an expanded target due to its multiple copies, one or more P elements were present in the proximal promoter of the first gene set (Gene Set I) in 88 samples (9.8%, 50 populations × 18 genes = 900 samples). In fact, 94% (152) of all 161 P insertions discovered in proximal promoters were located in heat-shock genes ( Hsp70 and Gene Set I; Figure 1 , inset; Figure 4 ). By contrast, in Gene Set III, selected for dissimilarity to Hsp promoters (see Introduction and Table S1 ), only 0.2% of the samples (50 populations × 18 genes = 900 samples) included a transposon (one insertion in one population each for CG6295 and CG14030; Figure 5 ). These transposons, furthermore, seem not to have inserted into the proximal promoter of their genes, but may have inserted into neighboring genes. The data do not support our last expectation, however, that TE insertions should be common in proximal promoter regions of non- Hsp genes similar to those of Hsp genes (Gene Set II). Only 0.8% of samples (50 populations × 18 genes = 900 samples) included a TE. Although ordinarily such “natural experiments” do not permit replication, the FlyBase Database ( http://flybase.bio.indiana.edu/ ) records anthropogenic insertions of natural and synthetic transposons. For example, the Berkeley Drosophila Gene Disruption Project (BDGDP) has undertaken genome-wide P element mutagenesis of laboratory stocks. Indeed, analysis of these data for the three gene sets (Gene Sets I–III) recapitulates the outcome of natural mutagenesis ( Figure 3 ). As of 2004, 157 experimental transpositions into the same genes that we screened in the natural populations are on record, of which 69% were into Gene Set I ( Hsp genes), 24% were into Gene Set II ( Hsp -like non- Hsp genes), and 7% were into Gene Set III (negative control set with Hsp -dissimilar genes; Figure 6 ). Thus most genes from Gene Set I (14 out of 18; 78%) and II (13 out of 18, 72%) had at least one insertion, whereas only six out of 18 (33%) of Gene Set III had insertions ( Figure 6 ). As in natural mutagenesis, Hsrω is distinctive, receiving more than twice as many P element insertions as any other gene in the three sets. The two datasets are highly concordant when all genes surveyed are ranked according to number of transposon insertions in the proximal promoter for (1) the 48 natural populations and (2) the synthetic transposon mutant strains (Spearman rank correlation test; p < 0.001). In many cases this similarity extends to the specific insertion sites themselves ( Figures 4 and 5 ). 10.1371/journal.pgen.0020165.g006 Figure 6 Frequencies of Experimental P Element Insertions Reported by FlyBase Database into the Proximal Promoter Regions of Each of the Genes Screened in Natural Populations in the Present Study Data are plotted as in Figure 3 . Note that insertions in the different Hsp70 copies are not combined as in Figure 3 . The FlyBase database ( http://flybase.bio.indiana.edu/transposons/ ) terms all tallied P element insertions as “transgene insertions.” 10.1371/journal.pgen.0020165.g007 Figure 7 Distinctive P Elements Revealed by Re-screening a Random Sample of P Element Insertion Sites in Natural Populations for Four Genes, Hsp23, Hsp27, Hsrω, and Hsp70 The P element insertion sites were selected from Gene Set I. A plus sign (+) indicates successful PCR amplification with one primer complementary to the focal gene and another complementary to a unique sequence in the P element (top), and thus reports the size and orientation of the P element; a minus sign (−) indicates no amplification. Table S7 provides sequences of these primers. At each insertion site in a population, one to six distinctive P elements segregated; these are designated a–f. For Hsp23, Hsp27, and Hsrω, nine insertion sites shared by two or more natural populations (indicated by boxes) and 17 unique insertion sites were re-screened. Amplicons that share a symbol (filled square [█], filled triangle [▴], filled circle •], etc.) occurred at the same integration site in different populations and were indistinguishable by size or orientation. For Hsp70, a five-copy gene in natural populations [ 17 ], the specific gene of insertion was not determined; thus, each distinctive amplicon (a–f) could represent insertion(s) at the same site in one to five of the Hsp70 genes. For population codes see Figure 1 . ORF, open reading frame. According to the Fisher exact test, the natural P element insertions were more numerous in Gene Set I than in Gene Set II ( p = 0.018) and in Gene Set III ( p = 0.002). By contrast, experimental P element insertion in Gene Sets I and II were not significantly different ( p = 1.0), but more numerous in each than in Gene Set III ( p = 0.004 and p = 0.010, respectively). Characteristics of P Elements within Promoters The insertion sites of P elements in the Hsp promoters were themselves highly clustered, with up to eight populations putatively receiving different insertions at the same site ( Figure 2 ). The elements also varied in orientation relative to the associated coding sequence, with no apparent orientation preference ( Figure 7 ). Re-screening Establishes that P Elements Are More Numerous than the UFW Screen Indicates Of the transposon inserts discovered, 47 were clearly unique (i.e., no two populations shared the same transposon at the same location in the same gene). By contrast, many different populations exhibited identical insertion sites, with up to eight populations showing P element insertions at the same position (this, for example, in the proximal promoter of Hsp70; Figure 2 ). As cited in the methodological concerns (above), these could represent insertion(s) of a P element into a common ancestral population and its vertical transmission into daughter populations, rather than independent multiple insertion events at the same site, and hence overestimate P element insertions. Alternatively, at any insertion site detected by UFW, multiple but similarly sized P elements might have inserted and presently segregate in any natural population so that UFW screening might underestimate TE insertions. To re-examine our above estimate of TE abundance in natural populations, 43 insertion sites (17 unique, with nine sites shared in two to six populations) in three different genes (Hsp23, Hsp27, and Hsrω) were chosen for re-analysis. We used a PCR-based technique that reports both the size and orientation of the P element ( Figure 7 ). In the 17 unique sites are 30 P elements distinguishable by size and orientation. In the nine shared sites are 30 P elements distinguishable by size and orientation. Of these, five are present in two populations, and one is present in four populations ( Figure 7 ). In all but one of these instances, the shared P element is singular in a first population and one of two to three forms segregating in a second population. In the remaining case, in Hsp27 in populations F51 and F54, two P elements segregate at the same insertion site in both populations. A similar re-screening of the Hsp70 genes at a “single” insertion site in each of five populations ( Figure 7 ) detected one to six distinctive P element variants at each site (15 total). Each was localized relative to a sequence shared by the five Hsp70 genes in natural populations, but not to a specific Hsp70 . Thus, each P element could be present at the same insertion site in one to five Hsp70 copies, with the tally of 15 under-representing the actual number of insertions. In summary, excluding the re-screened P elements in Hsp70, 60 distinctive P element variants were found at the 43 sites re-analyzed, suggesting that the UFW screen undercounted distinctive P elements by nearly 30%. Corrected for this undercount, all 117 P element insertion sites in the three gene sets likely harbor 163 distinctive transposable elements in the natural populations, or 225 with those in the Hsp70 s included. These distinctive P elements may represent distinctive insertion events, distinctive evolution after common insertion events, or both. To estimate a lower bound for the frequency of insertion events, distinctive P elements at the same 43 insertion sites were re-tallied based on orientation only (and not size). Any two P elements with opposite orientation likely arose from independent insertion events rather than from evolution after a common insertion (but more than two independent insertions cannot be distinguished). On this basis, 43 distinctive P elements are distinguishable at the 43 sites. In other words, the P element tally based on UFW likely does not overestimate the number of independent insertion events, although it may underestimate this number. Allelic Frequencies of the P Element Insertions Individual P elements varied in both nucleotide sequence and allelic frequencies in populations. A prior study [ 14 ] suggested that transposition into the proximal promoters of Hsp genes can be selectively advantageous because of its impact on Hsp gene expression. Moreover, although deleterious TEs might be inactivated if not purged from populations [ 32 , 33 ], adaptive TEs might be maintained at high frequencies or modified. Surveys of several randomly selected populations ( Table 2 ; Figure 7 ) are consistent with the simultaneous modification and maintenance of TEs. All P elements discovered whose size was determined were less than full length. The allelic frequencies of P elements at each site surveyed ( Table 2 ) varied considerably among populations, ranging from very low (e.g., in F40 for Hsp22 ) to high (e.g., in F04 for Hsp27 ) or even fixation (e.g., in F51 for Hsrω ). Populations also differed in the number of different variants of the P element inserted in a particular region of a gene or the number of insertion sites ( Table 2 ), and frequency and number of insertion sites are not correlated. Population F40, for example, harbors three different P elements at two different sites in Hsrω . The allelic frequencies, however, of these insertions are low (6%), whereas that for a single P element in Hsp27 in population F04 is much higher (85%; Table 2 ; Figure 4 ). 10.1371/journal.pgen.0020165.t002 Table 2 Frequency of P Element Insertions in the Promoter Region of Nine Heat-Shock Genes in Different Populations In six (12%) natural populations (F09, F10, F11, F16, F41, and F42; Figure 1 ), no P elements had inserted in the proximal promoter regions of the genes under investigation according to UFW. Re-screening these populations with PCR revealed no P elements; P elements either are absent in these populations or too distant from one another to support PCR amplification. Interestingly, all six populations are geographically adjacent ( Figure 1 ). Only about half (ten out of 21) of the African populations harbored two or more P element insertions, in contrast to 78% of the populations outside Africa. The population with the most numerous insertions ( n = 14, F40 from Marrakech), however, is African. Although specific mobile element insertions vary in frequency along geoclimatic gradients [ 14 – 17 , 21 ], in our data, the number of insertions bears no apparent relationship to latitude or climate. Discussion We have hypothesized that heat-shock genes as a class are distinctively evolvable because TEs integrate into their proximal promoter regions at unusually high rates, creating unique regulatory variation on which evolutionary processes such as natural selection can act. Here we test a major component of this hypothesis, that heat-shock proximal promoter regions are especially susceptible to the integration of TEs. TEs and other repetitive elements constitute more than 20% of the D. melanogaster genome [ 22 ], comprising 6,013 specific elements in more than 120 families [ 23 , 26 ]. Although these numbers are less than for other multicellular eukaryotes (e.g., ~45% in humans) [ 34 ], they nonetheless establish that TEs are numerous and diverse in D. melanogaster . TEs in Drosophila, moreover, are active, accounting for more than 50% of spontaneous mutation [ 35 ] (versus 0.2% in humans [ 36 ]). Accordingly, the discovery of more than 170 additional TEs in natural Drosophila populations is, in itself, unremarkable. What is striking is the predilection of these TEs for insertion in the proximal promoters of one class of genes, their persistence after insertion, and the fact that almost all are P elements. Methodological Issues Deducing this susceptibility from compilations of insertion sites is prone to bias unless (1) all TEs are detected, and (2) similarly sized regions are compared. With respect to (1), UFW is inclusive of all insertions, even currently undescribed TEs, because it is sensitive to size and not sequence [ 28 ]. UFW can be problematic, however, if deletion exactly counterbalances the insertion of novel elements and/or if its PCR steps favor amplification of small products. With respect to (2), we scrutinized equally sized regions of the three gene sets. One prospective weakness of this approach is that no two selected genes are the same and thus may differ as targets in ways not relevant to the main hypothesis. For example, TEs might be rare in those sets with many genes in regions of high recombination, which are thought to disfavor the persistence of TEs [ 37 ]. Genes that are located in regions of high recombination [ 38 ] are equally numerous in the three gene sets ( Table 3 ; Fisher exact test, p = 0.7). In fact, those genes in Gene Set I most numerous in P element insertions are in highly recombining region in most cases. At any rate, to guard against other unforeseen confounding factors, we surveyed 18 genes in each set on the assumption that similarities would manifest themselves if robust. Another prospective weakness is that any given natural population may be unrepresentative of entire species. To compensate, we surveyed 48 populations (and a reference strain). Thus, although no screen is free of biases, the results provide a reasonably unbiased assessment of naturally occurring TE insertions in the proximal promoters of three contrasting sets of genes. 10.1371/journal.pgen.0020165.t003 Table 3 P Element Insertions into Genes in Regions of High and Low Recombination Are Heat-Shock Promoters Conducive to P Element Insertion, and Why? The abundance of P elements in Drosophila heat-shock promoters may arise from the species' distinctive (but common) ecological niche. Like many organisms, D. melanogaster often undergoes hyperthermia in nature (e.g., [ 39 , 40 ]). Drosophila infests necrotic fruit, wherein eggs, larvae, and pupae are prone to heat stress. At the biochemical level, this hyperthermia is deleterious to proteins and membranes, and for the former may initiate a cytotoxic cascade of denaturation and aggregation of proteins [ 41 ]. At the organismal level, this heat stress compromises development, reproduction, and survival. A primary and important response to heat stress is therefore the expression of heat-inducible molecular chaperones, which can deter protein aggregation, target damaged proteins for degradation, help non-native proteins refold in the cell, and/or remove proteins from aggregates for refolding or degradation [ 42 ]. Given these essential roles, heat-inducible molecular chaperones are poised for rapid and massive accumulation upon heat shock [ 41 ]—hence their original name, “heat-shock protein.” Indeed, several distinctive features of the heat-shock genes of complex eukaryotes appear to facilitate heat-induced expression: constitutively decondensed chromatin, nucleosomes positioned outside the proximal promoter, a pre-assembled (but paused) polymerase apparatus, a pre-expressed, inactive, but readily activatable transcription factor (HSF), and the absence of introns needing splicing (e.g., [ 12 , 43 – 48 ]). Each feature can be viewed as an elimination or minimization of a time-consuming step in gene expression, and is thus an appropriate component of an emergency response to rapid and unpredictable thermal damage. As explained in the Introduction, these same features give TEs accessibility to chromatin, which could facilitate insertion [ 2 ]. Because Hsp genes share some of these features and represent extreme manifestations of others, Lerman et al. [ 15 ] suggested that proximal promoters of Hsp genes in general were natural “hotspots” for TE integration. Although this suggestion was consistent with the discovery of naturally occurring TEs in Hsp70 genes' promoters [ 15 ], (1) these TEs were few, (2) with few exceptions [ 49 ], naturally occurring TEs had not been discovered in other Hsp genes, and (3) TEs were not known to be comparatively rare in the proximal promoter regions of non–heat-shock genes. Three independent lines of evidence now establish that insertions of one TE, P elements, are common, not only in Hsp70 promoters, but also in other (single copy) heat-shock promoters: The present screen of natural populations documents (1) numerous P elements inserted into the proximal promoters of Hsp70 genes ( Figure 2 ), (2) numerous P elements inserted into the proximal promoters of Hsp genes other than Hsp70 ( Figure 3 ), and (3) few or no P elements inserted in the proximal promoters of Hsp -dissimilar genes ( Figure 4 ). That P elements have inserted in non-transcribed sequence is not unexpected [ 50 ], but our comparison is of identical regions in each gene set. If transposition into a region is solely a function of that region's representation in the euchromatin, then each gene set should accumulate equal numbers of P elements. The UFW and P element screens, by contrast, detect 152 (94.4%) in the heat-shock genes, seven (4.4%) in Gene Set II, and only two (1.2%) in Gene Set III ( Figure 1 , inset; Figure 3 ). As explained, the two in Gene Set III actually reside in nearby genes ( Figure 5 ) and could therefore be excluded (although we have not done so). The accompanying survey of transgene insertion experiments ( Figure 6 ), mainly from a genomic mutagenesis scheme that relied on P elements [ 8 ], has largely the same outcome (in number and position), at least for heat-shock genes ( Hsp70 and Gene Set I) and the negative gene set (Gene Set III). Shilova et al. [ 10 ] mobilized P transposon constructs (EPgy2) adjacent to the Hsp70A gene cluster and documented numerous new insertions into Hsp70 via local transposition ( Figure 2 C). They recovered 46 independent insertions of which, remarkably, 50% were into the same two nucleotides, −96 and −97, which is in correspondence with our results ( Figure 2 C). These two nucleotides harbor 12 (27%) of the 44 P elements we discovered in Hsp70 genes in natural populations, and represent a remarkable conjunction of DNase I hypersensitivity, adjacency to GAGA factor binding sites, absence of nucleosomes, and insertion site preference [ 10 ]. This largest number of P element insertions, both natural and transgenic, is into Hsrω, a highly conserved single-copy gene that encodes a heat-inducible mRNA, but no protein [ 51 ]. Like many non-coding RNAs, Hsrω transcripts play diverse regulatory roles [ 51 ], including regulation of distribution of RNA binding proteins [ 52 ] and the suppression of non- Hsp gene expression upon heat shock [ 53 ], and suppression of polyglutamine neurotoxicity [ 54 ]. In 30 (62.5%) of the 48 natural populations surveyed, a P element has integrated in Hsrω, with 70% distinctive integration events into nucleotides −89 to −161 upstream of the transcription start site. Why this region is so susceptible to P element integration is not apparent, but these insertions are not into any major regulatory element [ 55 ], and most of this region is not conserved within D. melanogaster . Two additional aspects favoring successful P element integration in the germline are expression of the host gene, and the occurrence of this expression in germ cells prior to meiosis, an embryonic process in Drosophila [ 9 , 56 , 57 ]. Gene Set III, essentially negative controls, was selected for restriction to narrow developmental windows and low levels of expression ( Table S1 ). Genes in this set also exhibit neither constitutive chromatin decondensation, nor positioned nucleosomes, nor regulation via polymerase pausing; i.e., none of the attributes hypothesized to favor TE integration in Hsp promoters. Accordingly, the near absence of TEs discovered in this set and their complete absence from the proximal-promoter region is consistent with the above conclusions. The paucity of transposons discovered in Gene Set II, the “ Hsp -like” genes, is more difficult to reconcile, however. One possible explanation is that the “true” Hsp s are prone to massive environmentally induced expression during the embryonic period in which germline transformation is possible, whereas the Hsp -like genes are not. As we discuss below, an alternative, but non-exclusive, explanation is that natural selection more effectively eliminates TEs from Hsp -like genes than from Hsp genes. Why Have P Elements Persisted in the Promoters of Hsp Genes in Nature? A common view is that TE insertions into genes are generally deleterious because mutations in general are usually deleterious and are therefore eliminated from the genome by purifying selection. Evidence includes the deleterious phenotypes that commonly result when TEs integrate into genes (for P elements, reviewed by [ 58 ]), the comparative rarity of TE insertions into coding or regulatory regions of genes, the low frequencies of intragenic TEs segregating in natural populations [ 18 , 19 ], and the rapid evolution of suppression of transposition in natural populations (e.g., trans -acting RNAs, DNA methylation, and specific antisense RNA). For Drosophila, instances in which intragenic TEs are beneficial are so rare that accounts of such instances are newsworthy [ 59 – 62 ]. Why, then, are P elements in the Hsp promoters of naturally occurring Drosophila so numerous? One explanation is that P elements are distinctive transposable elements in several respects. First, unlike other TEs (for example, the Ty elements of Saccharomyces, which insert in specific sites of specific genes [ 2 ]), the information content of P insertion sites is relatively low [ 63 ]. P element insertions thus are general and robust reporters of exposed chromatin. Second, P elements have invaded the D. melanogaster genome relatively recently. Biogeographic patterns of P element occurrence in D. melanogaster and the recent near-disappearance of D. melanogaster populations without P elements together suggest that P elements invaded the D. melanogaster genome within the last century [ 29 – 31 ]. Since then, P elements spread widely and effectively throughout the D. melanogaster genome, likely due to self-regulation and hybrid dysgenesis [ 29 , 64 ]. TEs typically undergo excision or degeneration with time. Indeed, whereas full-length P elements encode a transposase and are thus autonomous, every P element discovered in the UFW screen whose size we have determined is less than full length. Thus, the occurrence of P elements in the Hsp proximal promoters could be transient, representing a successful genomic invasion that has not yet been purged. Although this may be so, populations we have surveyed average two to three P elements in their Hsp promoters. Given that natural populations typically carry no more than 50 P elements distributed throughout their more than 13,000 genes [ 65 ], their abundance in Hsp promoters is remarkably high. Importantly, in natural Drosophila populations TEs in Hsp promoters may be advantageous and therefore persist via positive selection (e.g., [ 21 ]). Although Hsp s typically encode proteins that can function as molecular chaperones, these proteins have numerous other functions including intra- and extra-cellular signaling (small Hsp s, Hsp60, and Hsp70 ) [ 66 ], regulation of cell cycle and apoptosis [ 67 , 68 ], and maturation and regulation of nuclear receptors, among others [ 66 , 69 ]. Hsp90 (encoded by Hsp83 in Drosophila ) alone interacts with numerous (>200) client proteins in the cell [ 70 ]. Thus, whereas massive Hsp accumulation often may be advantageous for temperature tolerance and mitigation of thermal damage, in the absence of heat stress, stringent regulation of Hsp levels may be essential. Indeed, unbridled expression of Hsp genes in the absence of heat stress is often deleterious (reviewed by [ 15 ]). The typical phenotype of TE insertions into Hsp70 promoters is decreased gene expression [ 14 ], and natural P insertions have similar phenotype in Hsp26 in population F32 (unpublished data). Accordingly, selection may favor the retention of TEs in Hsp promoters because these elements reduce expression of a situationally harmful protein. Natural and experimental selection can either increase or decrease Hsp expression depending on which outcome a given thermal environment favors (reviewed by [ 15 ]). P element insertion may thus simply be an opportunistic way to suppress Hsp activity in the wild. Intragenic TEs segregating in natural Drosophila populations are typically at very low frequencies, consistent with their usually deleterious phenotypes (see above). On the other hand, intragenic TEs considered advantageous, by contrast, are typically at much higher frequencies, if not fixed, in natural populations [ 13 , 20 , 71 ]. In our survey, although frequencies of P elements at specific sites varied considerably in natural populations, these sometimes were much higher than those invoked as evidence for deleterious or neutral phenotypes [ 18 , 19 ]. Ten of the 16 P element insertions we arbitrarily chose for examination in detail were at population frequencies ≥15%, seven at ≥35%, and four at ≥75% ( Table 2 ). This outcome resembles that for prior studies of TEs in Hsp70 promoters in natural populations, in which frequencies were high and varied inversely with Hsp70 protein levels [ 14 – 17 , 21 ]. A testable prediction is that experimental evolution in contrasting thermal regimes should be capable of altering P element frequencies in Hsp genes according to the phenotypes of P element insertions; i.e., increasing P element allelic frequency when beneficial and decreasing frequency when deleterious. The same prediction ought to be applicable to other instances in which TE-derived sequences modify a host function and have been assimilated by the host genome [ 1 , 20 , 49 , 72 , 73 ]. Interestingly, in the transgene lines, laboratory strains that are intentionally isolated from natural selection, the Hsp and Hsp -like genes included in our survey received transposon insertions at similar frequencies ( Figure 6 , inset) and positions. In nature, by contrast, the proximal promoters of the Hsp -like genes (Gene Set II) were relatively depauperate P elements. This contrast further implicates positive or balancing selection for maintenance of P elements in the Hsp genes (Gene Set I and Hsp70 ). In other words, in nature, selection may routinely purge P elements from the proximal promoters of non- Hsp genes and/or favor insertions in Hsp genes. It suggests that accessibility of chromatin to P element integration may be necessary, but not sufficient, to generate stable P element insertions in the proximal promoters of these genes in nature because of negative or purifying selection. Heat-shock genes are a component of an ancient, but effective, response to acute thermal stress in the natural environment, and include features that facilitate rapid and massive gene expression (e.g., [ 40 , 42 , 74 ]). These same features, however, may facilitate the integration of P elements into Hsp promoters, which in turn affect gene expression. These intragenic P elements thereafter may segregate in natural populations and are a form of genetic variation upon which natural selection and other evolutionary processes may act. The present study and another (unpublished data) establish that this scenario is not specific for a single multicopy gene (Hsp70) in a small number of natural populations as previously described, but generally applicable to Drosophila populations worldwide and to the entire heat-shock genome. From this perspective, intragenic TEs in natural Drosophila populations are less newsworthy exceptions [ 59 – 62 ] than a normal expectation. The near-exclusive involvement of P elements and the frequencies of these elements establish, moreover, that the TEs in Hsp promoters are not the remnant of an ancient event, but a manifestation of active and ongoing microevolution in natural populations. Heat-shock genes have been posited to make a special contribution to evolvability [ 75 ]. The capacity of their products—molecular chaperones—for conformation-specific recognition of diverse client proteins has led to their involvement in diverse regulatory processes and their assignment to diverse structural roles. Importantly, molecular chaperones may transiently suppress conformational mutations in client proteins, thereby protecting such variation from selection under routine conditions, but exposing it during stress [ 76 ]. TEs have likewise been posited to make a special contribution to evolvability [ 1 ]. They are the most widespread and effective of natural mutagens, can redistribute genetic material throughout the genome, and form the recombinational substrate for much gene duplication, retroposition, and creation of hybrid genes. Here we show that these two components of evolvability intersect: heat-shock genes as a class are distinctively prone to the integration of at least one TE, the P element, and that, accordingly, the expression of heat-shock genes may evolve distinctively from the rest of the genome. Materials and Methods Drosophila strains. The Drosophila stocks used in this survey were derived from wild-caught flies and were maintained as isofemale lines with the exception of F32, F51, and F52. Flies were from 51 different worldwide locations ( Figure 1 ), plus a reference strain. Most lines were obtained from Dr. Jean David (Centre Nationale de la Recherche Scientifique, France), and have been the subject of previous investigations [ 77 – 80 ]. In addition, strains F01 (14021-0231.21), F02 (14021-0231.22), F03 (14021-0231.23), F04 (14021-0231.25), and F05 (14021-0231.26) were obtained from the Drosophila Species Stock Center, Tucson, Arizona, whose reference numbers are in parentheses. The reference strain (F06: y 1 ; cn 1 bw 1 sp 1 ), “the Celera strain,” was the strain whose genome has been sequenced [ 25 ]. This strain is free of P elements [ 26 , 80 ]. Strain F32 was provided by Dr. Michael Rose (University of California, Irvine) and is one of the “base” or control strains used in his studies of laboratory evolution [ 81 ]. Strains F51 and F52 were from the north- and south-facing slopes, respectively, of “Evolution Canyon” (Lower Nahal Oren, Israel) [ 82 ]. Additional strains and origins include: F49, Dr. Arne Mooers (Simon Fraser University, Canada); F48 and F50, Jennifer Shirriffs (La Trobe University, Australia); and F53 and F54, Dr. Masayoshi Watada (Ehime University, Japan) [ 83 , 84 ]. All live flies were reared on a yeast, cornmeal, molasses, and agar medium at 25 °C. DNA isolation. Bulk samples of genomic DNA were extracted from 2 × 50 individual adults for each population. Flies were fresh or preserved in 70% or 100% ethanol. Ethanol-preserved files were air-dried and washed in 500-μl phosphate-buffered saline solution (PBS) for 2 min prior to DNA isolation. The washing buffer was removed and another 180-μl PBS was added for grinding. Total DNA was extracted according to [ 28 ]. Gene sets. Genes ( Tables 1 and S1 ) were selected and grouped into sets for analysis a priori according to the following criteria: (1) the nearly identical Hsp70 genes [ 17 ]. These were included for re-analysis because of the prior discovery of several TEs in their 5′-flanking regions (see Introduction). These genes are arranged in two clusters ( Hsp70A at 87A7, comprising Hsp70Aa and Hsp70Ab, and Hsp70B at 87C1, comprising Hsp70Ba, Hsp70Bb, Hsp70Bbb (if present), and Hsp70Bc ); (2) other heat-shock genes (Gene Set I) [ 74 , 85 ]. Although all genes included in this group of genes increase in expression upon heat shock or other proteotoxic stresses, most (unlike Hsp70 ) [ 86 ] are expressed constitutively. Also, although several subsets of these genes share similarities in sequence, they are not multicopy genes in the same sense as the Hsp70 s, which encode proteins of identical sequence [ 17 ]; (3) genes resembling heat-shock genes in regulation of expression, chromatin configuration, associated promoter elements, etc. (see Introduction) (Gene Set II). These were identified from literature reports ( Table S1 ); and (4) Genes dissimilar to heat-shock genes or with no known features similar to those of heat-shock genes (Gene Set III). These were initially selected from the data of Arbeitman et al. [ 87 ], available online (http://flygenome.yale.edu), according to least expression throughout the Drosophila life cycle. All else equal, genes with limited embryonic expression were preferred. Initially selected genes were discarded if a literature search disclosed characteristics that might qualify them for inclusion in Gene Set II. In most cases, genes included in this set had not been studied in detail when the set was compiled, or had hypothetical functions according to sequence homology with better-studied genes ( Table S1 ). Transposable element screening. We executed two screens with differing characteristics. The first, a UFW screen reported size polymorphism in amplicons between the reference strain Celera (F06) and natural populations. We modified UFW as originally described [ 27 ] to minimize false positives, and included an additional nested PCR step [ 28 ]. A total of 5 μl of genomic DNA (100 ng) extracted in bulk from 2 × 50 adult flies was used for each reaction. Vista ( http://genome.lbl.gov/vista/index.shtml ) and Primer 3 ( http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi ) were used to design six to seven UFW primers for all 55 genes binding either to a highly conserved region in the CDS (coding sequence) of the gene or to a neighboring gene ( Tables S2 , S3 , and S4 ). Because of the extraordinary conservation of Hsp70 coding sequence [ 17 ], UFW primers specific to each of the five to six Hsp70 genes were impossible to design. Therefore the UFW screen of Hsp70 indicated amplicon size polymorphism in one or more of the genes. In brief, UFW is a non-restrictional, non-ligational genome walking method that uses primer complementary to a known sequence (in our case, coding sequence), random primers, PCR, and nuclease digestion in combination to amplify unknown sequence flanking the known sequence. Insertions (e.g., TEs) or deletions present in the amplified unknown region are detectable because they increase or decrease the size of the amplicon. TEs were subsequently identified by sequencing and localized via the PCR screen described below. More detailed descriptions of the assay, guidelines for primer design, and exemplary gel images are in Myrick and Gelbart [ 27 ] or Walser et al. [ 28 ]. About 15% of all UFW reactions initially failed or produced only modest amplification even after the nested PCR step, and were therefore repeated with a different set of primers. Populations F18, F50, and F52 were removed from the dataset after the reaction failed for multiple genes and primer sets. Presumptive positives (defined as extra bands in the UFW footprint compared to the reference strain) were cloned and then sequenced for identification (at least one per gene). Topo TA Cloning Kit (Invitrogen, Carlsbad, California, United States) for amplicons fewer than two kilobases and Topo XL PCR Cloning Kit (Invitrogen) for larger fragments were used. FlyBase query and Repeat Masker software (with the latest release of the RepBase database update; http://www.girinst.org ) were used to assess the similarity of insertions with known sequence. When the vast majority of insertions (98%) that UFW screening revealed were P elements, a second PCR-based screen was designed to confirm the UFW results and discover further insertions. In this second screen, one primer site was chosen in the conserved region of the associated gene, and the other in the 31-bp (base pair) inverted terminal repeat region of the P element. The P -specific PCR screen also included a further population from Israel (F51) and two populations from Japan (F53 and F54). Although primers for the second screen were gene specific, all positive samples (samples amplified in the second screen) were re-screened with another primer specific to the conserved region of the associated gene. PCR reactions with only one primer, the P element–specific primer, served as a control for inadvertent amplification of sequence between multiple P element insertions. Additionally, we sequenced 75% of all positive samples, including all fragments bigger then 450 nt. For amplicons smaller then 450 nt, a −6FAM–labeled P element–specific primer was used, and the PCR products were sized on an ABI 3730 DNA sequencer with LIZ-500 as internal size standard using Genemapper 3.0 (Applied Biosystems, Foster City, California, United States). Control screening. To confirm that the discovery of P elements was not due to some inadvertent bias of the UFW screen, we performed additional PCR-based screens for six TEs common in the Celera strain ( roo, 1360, 297, Jockey- family, I elements, and Gypsy -family elements) [ 26 ] and still active in the genome. The same PCR-based approach as for P element screening was used, with one primer specific for a conserved region of the associated gene and the other for the TE being screened. Because these TEs lack inverted terminal repeats, two element-specific primers were used to establish TE orientation. TE-specific primers were designed from alignments of TE sequences deposited in the National Center for Biotechnology Information (NCBI) GenBank database ( Table S5 ). For all populations, the promoter region of Hsp70 was screened for these six TEs. Furthermore, two genes from each set of genes were likewise re-screened for four populations (F04, F40, F53, and F54). Screening for deletions. A final PCR-based screen used primers ( Table S6 ) complementary to conserved regions in a gene of interest and a neighboring gene, and thus amplified the entire intergenic/intervening region. Because PCR preferentially amplifies smaller sequences in size polymorphisms, this screen was expected to detect primarily deletions and small insertions rather than larger TEs at a low frequency. Frequency of insertion. For selected genes and five populations (F04, F05, F12, F40, and F51) the frequency of P element insertions was estimated by analyzing 35–48 individual flies per population ( Figure 1 ). DNA was purified from single flies according to the manufacturer's recommendations (Puregene DNA Purification Kit; Gentra System, Minneapolis, Minnesota, United States). Distinguishing P element insertion events. P elements discovered at the same site in multiple populations might have been inherited from a single ancestral population or have inserted independently. Also, any P element discovered at a site might either be singular or represent multiple elements segregating in a population. To distinguish among these alternatives, 43 P element insertions (at different sites with populations sharing zero to six P transposons at each site) from three different genes (Hsp23, Hsp27, and Hsrω) were re-screened with a PCR-based technique that reports both the size and orientation of the P element ( Figure 7 ; Table S7 ). This screen exploited the tendency of the internal regions of P elements (but not their termini) to truncate during evolution [ 3 ], but in highly variable fashion. From an alignment of full-length and truncated P element sequences deposited in the NCBI GenBank database, six forward and seven reverse primer sites were chosen. The corresponding amplicons (indicated by “+” in Figure 7 ) are indicative of the size and orientation of the corresponding P element. Transgenic insertion sites. The flanking region of all experimental P element transpositions thus far reported by the FlyBase Database ( http://flybase.bio.indiana.edu ) were used to characterize insertion site for the genes of the different gene sets (Gene Sets I–III). In addition, we also included locations of EPgy2 element insertions for Hsp70 recently described by Shilova et al. [ 10 ] for comparison with the naturally occurring P elements. Spearman rank correlation was used to test for a monotonic relationship between the natural and the transgene insertion sites for the different genes. The Fisher exact test for count data was used to compare the number of gene with element inserts in the three gene sets. Other polymorphisms. The UFW screen also detected several insertions/deletions, presumably deletions by parsimony ( Figure 4 ). In population F40, a 1,381-nt deletion is in the 5′ region of Hsp27 . The deletion occurs upstream of the TATA box and removes all five heat-shock elements (HSE). Another deletion of 565 nt occurs in Hsp68 in population F21. This removes all four HSEs, the TATA box, and the initiator. Population F01 exhibited a pair of deletions in Hsrω, in which 17 nt and 5 nt of the 5′ and 3′ regions flanking the TATA box were absent. Screens of individual Drosophila ( n = 50) suggest that these deletions segregate in populations at very low frequencies (<2%). Supporting Information Table S1 Supplementary Information on Genes in the Three Gene Sets (219 KB PDF) Table S2 UFW Primers for Gene Set I (36 KB PDF) Table S3 UFW Primers for Gene Set II (44 KB PDF) Table S4 UFW Primers for Gene Set III (31 KB PDF) Table S5 Transposable Element–Specific Primers for Additional Screens (80 KB PDF) Table S6 Primers for Intergenic Regions (58 KB PDF) Table S7 P Element–Specific Primers for Orientation and Length Determination (299 KB PDF) |
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
In recent years, there has been a growth in the number of museum-based research studies that revisit and reanalyze archaeological legacy collections [ 1 – 12 ]. These projects have investigated a wide range of issues, including colonialism, environmental studies, gender, human subsistence, museum curation practices, and several other topics. Researchers are increasingly revisiting legacy collections and conducting new excavations at extant sites that apply modern excavation and sampling standards to revise site-specific and regional chronologies and earlier interpretations [ 9 , 13 – 18 ]. In certain instances, archaeological sites are no longer accessible for new excavations due to site destruction, modification, permitting processes, heritage conservation practices, or concerns raised by stakeholders. In others, unanalyzed and understudied museum collections exist for the site(s) and do not warrant further excavations on sensitive, finite, and nonrenewable cultural resources. In these circumstances, museum collections offer an exceptional opportunity to contribute important new information regarding archaeological sites that can confirm, revise, and refine previously reported chronologies and interpretations for specific sites or broader archaeological regions.
Significant advances in radiocarbon dating have occurred since the inception of the method facilitated by accelerator mass spectrometry (AMS), increasing the accuracy and precision of radiocarbon dating measurement. Furthermore, advancements in sample preparation and pretreatment, smaller sample sizes required for dating, standardization of laboratory protocols, compound-specific analyses, refinement of calibration curves, improved statistical analyses, and a deeper understanding of reservoir effects continue to advance the method and its application. These developments have resulted in a critical reappraisal of previously reported archaeological chronologies, often through the application of site and region-specific chronometric hygiene assessments to ensure that radiocarbon samples are reliable for chronology building. Numerous studies have shown that various preceding radiocarbon dating projects frequently lack chronometric hygiene assessments and suffer from other biases. These biases commonly result from sampling long-lived rather than short-lived organisms, selecting mixed samples rather than single entities, lack of proper sample pretreatment procedures, estimated rather than measured δ 13 C and δ 15 N isotopic values, and dating samples of ambiguous cultural association [ 19 – 26 ].
Along the northern Oregon Coast, recent AMS radiocarbon dating of cervid bones at the Par-Tee site (34CLT20) by Sanchez and colleagues [ 9 ] significantly revised the Par-Tee chronology through the application of site-specific chronometric hygiene assessments to previously reported radiocarbon dates from the 1960s and 1970s and Bayesian statistical modeling. Sanchez and colleagues [ 9 ] found that radiocarbon measurements from the 1960s and 1970s at the Par-Tee site derive from composite or bulk samples of unidentified charcoal and shell and bone. Many samples were not appropriately pretreated to remove potential contaminants, and often, samples were not corrected for δ 13 C isotopic fractionation [ 19 , 27 – 29 ]. Instead, δ 13 C isotopic ratios were estimated rather than measured, making these older dates problematic for building chronologies. These biases in the radiocarbon data are significant considering recent analyses of the Par-Tee site museum assemblage investigating ancient fishing practices, potential whaling events, and the use of cetaceans, sea mammals, and terrestrial mammals more broadly [ 30 – 35 ]. The lack of accurate radiometric measurements for the Par-Tee site places these studies in chronological limbo resulting in significant uncertainties regarding how the timing of the human activities identified at Par-Tee interdigitates with the Palmrose site and other sites and practices regionally.
Previous radiocarbon dating of a nearby archaeological site known as Palmrose (35CLT47) with evidence of an early plank house suggests the Palmrose site was inhabited millennia before Par-Tee. Because the majority of Palmrose radiocarbon samples were obtained by the same researchers, analyzed using the same methods, and samples processed by the same laboratory—the Smithsonian Institution Radiocarbon Laboratory (SI)—as Par-Tee, there are significant questions about their hygiene and the chronology’s reliability.
In this paper, I present the results of recent AMS radiocarbon dating and Bayesian analysis for the Palmrose site, a large village site that produced a sizeable and diverse material culture record that includes formal tools, faunal remains, and early evidence of fully- to semi-sedentary lifeways along the Oregon Coast. In this study, I primarily selected culturally modified cervid remains for radiocarbon dating, including cut marked elk ( Cervus ) and deer ( Odocoileus sp.) bones that exhibit evidence of human processing. One exception is a single elk premolar/molar fragment from an excavation level that lacked other diagnostic postcranial cervid specimens and direct evidence of human processing. Previous research suggests that elk and deer dominate the Palmrose terrestrial mammal assemblage [ 30 , 31 ] and represent the primary raw material in the bone and antler tool assemblage [ 9 , 34 , 36 ]. The Bayesian analysis of the new AMS radiocarbon dates for the Palmrose site sequence will assist forthcoming museum-based studies through the construction of a refined site chronology with relevance to broader regional chronological frameworks and provides an important context for enhancing the interpretation of existing collections and increasing their broader value to the scientific community [ 37 – 39 ].
Background
Large-scale archaeological excavations along the northern Oregon Coast were conducted by George Phebus, a collections assistant in the Department of Anthropology, Smithsonian Institution, and avocational archaeologist Robert Drucker from 1967 to 1977 [ 40 , 41 ]. Together Phebus and Drucker excavated three significant sites in Seaside, Oregon, specifically the Palmrose, Par-Tee, and Avenue Q (35CLT13) sites ( Fig 1 ). Excavations resulted in the recovery of a diverse range of material culture currently curated by two institutions, including the National Museum of Natural History (NMNH), Smithsonian Institution, and the Museum of Natural and Cultural History (MNCH), University of Oregon [ 9 ]. However, many formal artifacts from the sites remain in possession of private collectors who participated in the initial excavations [ 42 ]. Phebus and Drucker note that Par-Tee and Palmrose each measured over 65 m in length with deposits of at least 1.4 m but up to 3.0 m in depth. The Avenue Q site lies beneath residential structures, yards, and roads but yielded stratified and undisturbed deposits.
10.1371/journal.pone.0255223.g001
Fig 1
Overview of the northern Oregon Coast and the location of the Palmrose (35CLT47), Par-Tee (35CLT20), Avenue Q (35CLT13) sites.
Based on evidence from field notes curated by the MNCH and the National Anthropological Archives, Smithsonian Institution, Phebus and Drucker excavated at least 227 5 x 5 ft wide excavation units at the Palmrose site, 256 5 x 5 ft units at Par-Tee, and a single 5 x 5 ft test unit at Avenue Q. All excavation units were dug in 1 ft arbitrary levels—each assigned a numeric number from top to bottom—with the recovery of materials from excavated sediments screened over ¼ in. sieves. According to Phebus and Drucker’s estimates, excavations at Palmrose and Par-Tee may have totaled ~1415 m 3 .
Palmrose site excavations 1967−1988
Among the Seaside sites, Palmrose provides compelling evidence for the presence of a rectangular plank house, the earliest reported and known example of an ancient plank house along the Oregon Coast [ 40 – 43 ]. According to Phebus and Drucker’s field notes and reports, the site’s western portion was significantly impacted by looting activities and contained largely unstratified deposits [ 40 , 41 ]. However, the eastern portion of the site appeared to be mostly intact with stratified deposits. Phebus and Drucker’s [ 40 , 41 ] reports and field notes suggest they encountered a rectangular house feature, most likely a plank house, possibly measuring 6 m in width and 12 m in length, on the eastern portion of the site where they focused the majority of their excavation efforts ( Fig 2 ). According to these records, the house included multiple superimposed sand-lined hearth features. Radiocarbon dates obtained in the 1960s and 1970s, primarily on unidentified charcoal, suggest the house was inhabited for millennia with at least three house rebuilding events [ 40 , 41 ]. Phebus and Drucker’s interpretations of these dates suggest the site was inhabited in three significant episodes, with the earliest occurring from ~700−600 cal BC, an intermediate occupation from ~300−200 cal BC, and the terminal occupation from ~cal AD 200−300 [ 40 – 43 ].
10.1371/journal.pone.0255223.g002
Fig 2
Grid map of the Palmrose site depicting excavation units sampled for radiocarbon dating (colored units) and the MNCH profiles from 1988.
Grey units represent the extent of Phebus and Drucker’s excavations. Adapted from Connolly (42) with permission from the Museum of Natural and Cultural History, original copyright 1992.
In the summer of 1988, subsequent testing of the Palmrose site was conducted by MNCH archaeologists to establish the boundaries of a highway right-of-way for a proposed alteration to the local highway [ 42 ]. The MNCH field crews were able to relocate Phebus and Drucker’s former excavation units while confirming and establishing the site boundaries through the placement of seventeen 20 cm diameter auger probes, two 50 cm square test units, and one l x 1 m square test unit. The fieldwork confirmed earlier reports of extensive disturbance to the western section of the site. Given the disturbance level on the western portion of the site, MNCH archaeologists abandoned that section’s excavations. Subsequently, they focused their efforts on the site’s eastern segment, opening three vertical profiles of undisturbed midden deposits from Phebus and Drucker’s excavation units ( Fig 2 ) [ 42 ].
Connolly [ 42 ] summarized that two profiles designated North and South both measured two meters in length and revealed stratified and intact deposits related to the plank house occupation (see Fig 2 ). However, Connolly terminated the South Profile’s excavation due to the presence of human remains in the basal deposits. Next, a 50 cm x 50 cm column sample designated as Unit A was excavated into the North Profile of Phebus and Drucker’s excavation block and northeast of the plank house. Lastly, a third profile designated as the East Profile measured 11 m in length and exposed a cross-section at the plank house feature’s eastern edge. All midden constituents were recovered using 1/8 in. mesh screens. The 1988 excavations confirm several factors originally reported by Phebus and Drucker [ 40 , 41 ]. First, the house appears to have a well-defined bench along the north wall. Second, a central fire hearth provides evidence for a series of four superimposed sand-lined hearths from subsequent occupations, each marked "Sand or Ash/Sand" in field notes, reports, and profiles. Third, evidence suggests infilling occurred after site abandonment following each occupation [ 42 ].
Palmrose radiocarbon dating 1967−1988
Previous research suggests habitation of Palmrose occurred between 2340 cal BC to cal AD 640 [ 40 – 42 ]. The bulk of the Palmrose site’s available radiocarbon dates derive from assays on charcoal samples submitted by Phebus and Drucker and processed by the Smithsonian Institution Radiocarbon Laboratory in the 1960s and 1970s. Table 1 presents the 19 radiocarbon assays. Before including or excluding these previously reported dates from Bayesian modeling, I applied the following chronometric hygiene assessments, which were previously used by Sanchez and colleagues [ 9 ] at the Par-Tee site, to evaluate each sample’s reliability: 1) are the samples derived from identified or unidentified charcoal, and do they represent bulk samples or individual specimens; 2) are samples from long-lived or short-lived organisms; 3) were sample pretreatment procedures conducted to remove potential contaminants; 4) were samples corrected for δ 13 C isotopic fractionation; 5) are samples accurately dating the event of interest or stated otherwise is there ambiguity regarding the association of the sample with cultural remains, deposits, and events of interest.
10.1371/journal.pone.0255223.t001
Table 1 Previously reported radiocarbon ( 14 C) dates for the Palmrose site. Lab numbers beginning with Smithsonian Institution Radiocarbon Laboratory (SI) represent Phebus and Drucker samples, while Connolly submitted samples to Beta Analytic Inc. (Beta).
14 C Lab Number
Provenience
Material
Pretreatment
Conventional 14 C Age BP
cal BC/AD (95.4% CI)
SI_612
NWA2-4
Charcoal
---
1760 ± 50
AD 200−420
SI-613
NWA2-5
Charcoal
---
1650 ± 100
AD 210−640
SI-614
NWA6-6
Charcoal
---
1640 ± 100
AD 220−640
SI-582
NWA2-2
Charcoal
NaoH, HCl
2410 ± 110
800−200 BC
SI-582R
NWA2-2
Charcoal
NaoH, HCl
2610 ± 90
980−420 BC
SI-583
NWA2-6
Charcoal
NaoH, HCl
2260 ± 100
750−40 BC
SI-584
NWA6-7
Charcoal
NaoH, HCl
2620 ± 90
990−420 BC
SI-584R
NWA6-7
Charcoal
NaoH, HCl
3840± 150
2860−1880 BC
SI-585
NWA6-8
Charcoal
NaoH, HCl
2180 ± 80
400−10 BC
SI-586
NWA10-10
Charcoal
NaoH, HCl
2180± 100
420−70 BC
SI-2385
NE2C-6
Charcoal
NaoH, HCl
2495 ± 65
790−410 BC
SI-2386
NE2B-7
Charcoal
NaoH, HCl
2475 ± 65
780−410 BC
SI-2387
NE1D-5
Charcoal
NaoH, HCl
2490 ± 65
790−410 BC
SI-2388
NE1D-6
Charcoal
NaoH, HCl
2380 ± 65
770−260 BC
SI-3229
SE3B-3
Charcoal
NaoH, HCl
1765 ± 65
AD 120−420
SI-3230
SE3B-5
Charcoal
NaoH, HCl
1840 ± 65
AD 20−370
SI-3231
SE3B-7
Charcoal
NaoH, HCl
1830 ± 70
AD 30−410
SI-3232
SE3B-9
Charcoal
NaoH, HCl
2135 ± 65
380 BC−AD 10
SI-3233
SE3B-10
Charcoal
NaoH, HCl
2565 ± 70
890−420 BC
Beta-28848
Unit A-6
Charcoal
NaoH, HCl
1760 ± 60
AD 130−420
Beta-28849
Unit A-9
Charcoal
NaoH, HCl
2270 ± 100
750−40 BC
Beta-28852
Unit D-4
Charcoal
NaoH, HCl
3650 ± 100
2340−1740 BC
Beta-28853
Unit F-18
Charcoal
NaoH, HCl
2060 ± 100
370 BC−AD 210
Chronometric hygiene assessments have been applied in archaeological studies to assess the reliability of radiocarbon dates for various regions throughout the world [ 19 – 26 ]. While the criteria applied in chronometric hygiene assessments vary between regions due to differences in preservations biases, excavations practices, radiocarbon sample selection, and freshwater and marine reservoir effects, these assessments are applied to ensure that reported radiocarbon dates reflect the cultural phenomena of interest and to identify which samples should be included in analyses or excluded. Therefore, the chronometric hygiene criteria applied in this study seek to mitigate potential biases from historically reported dates that primarily derive from unidentified charcoal samples and composite charcoal samples [ 9 , 42 ].
The chronometric hygiene assessments developed by Sanchez and colleagues [ 9 ] and applied in this study to evaluate the reliability of the 19 previously reported radiocarbon dates by Phebus and Drucker reveal numerous issues. First, the charcoal samples submitted by Phebus and Drucker represent large composite samples of wood, often combining separate entities in one sample. These findings are consistent with their use of composite samples at the Par-Tee site [ 9 ]. As Ashmore demonstrates [ 19 ], composite samples of wood are unreliable due to the combination of separate entities, resulting in the dating of multiple events rather than more discrete cultural activities. Second, in Pacific Northwest rainforests, long-lived trees and drift logs were a common fuel source, so dates of multiple unidentified charcoal fragments are likely significantly offset by in-built age and/or the old wood effect [ 9 , 36 , 44 , 45 ]. Third, samples that lack stable carbon isotope measurements are prone to inaccuracies [ 28 ].
Based on archival records, δ 13 C isotopic values for all Smithsonian Institution radiocarbon samples from Palmrose were estimated rather than measured, raising uncertainties about correcting these dates. Fourth, several of these dates have large standard deviations (≥100 years) that result in large calibration ranges limiting their potential to provide the chronological data required to define and constrain the cultural events of interest. Fifth, the laboratory reanalyzed two samples submitted by Phebus and Drucker, and in each instance, discrepancies exist between the dates reported. For instance, SI-584 and SI-584R have conventional radiocarbon ages of 2620 ± 90 and 3840 ± 150, Table 1 . When calibrated at 2-sigma in OxCal 4.4 using the IntCal20 calibration curve [ 46 , 47 ], the dates span from 990−420 cal BC and 2860−1880 cal BC. To a lesser degree, SI-582 and SI 582R have conventional radiocarbon ages of 2410 ± 110 and 2610 ± 90, respectively, Table 1 . When calibrated at 2-sigma, the dates span from 800−200 cal BC and 980−420 cal BC. For all these reasons, in re-examining the potential age range for Palmrose site human occupation and applying the chronometric hygiene assessments developed for this study, I exclude all dates previously reported by Phebus and Drucker [ 20 , 23 – 26 , 48 , 49 ].
In addition to the dates compiled by Phebus and Drucker, four additional radiocarbon dates for the site were collected by MNCH staff and submitted to Beta Analytic Inc. following the 1988 field project. Similar to the dates reported by Phebus and Drucker, the radiocarbon dates reported by Connolly [ 42 ] were not corrected for δ 13 C isotopic fractionation, and the majority—three out of the four—is either derived from ambiguous contexts or lack sufficient data reporting to assess their cultural association fully. For example, sample Beta-28852 is derived from below the shell midden deposits of the Palmrose site beneath two clay lenses in humic loam [ 42 ]. The charcoal sample was derived from a charcoal-rich sandy loam near a whale bone fragment. However, given the lack of a well-defined association with cultural materials based on the lack of stone tools, shell midden, or other cultural items, it is unclear if the charcoal-rich sandy loam and whale bone represent natural background materials or if they were deposited through human agency. Given the ambiguous cultural association, the date is excluded. Of the remaining three samples submitted by Connolly, the provenience for two (Beta-28848 and Beta-28849) are not adequately reported. Beta-28853 represents charcoal from deposits Connolly interpreted as a sand-line hearth. Nevertheless, the samples analyzed represent unidentified charcoal, which may derive from long-lived organisms and includes multiple entities [ 19 ]. Applying the chronometric hygiene standards developed for this study and applied to other sites in the region [ 9 ], given the ambiguous and unreported cultural context for most of the samples, the use of unidentified charcoal, possibly from long-lived organisms, and lack of δ 13 C isotopic fractionation measurements, I excluded the four dates reported by Connolly [ 42 ] from chronological modeling.
Methods and materials
The MNCH curates the Palmrose collections and materials sampled in this study. Twelve culturally modified elk and deer bone samples from four excavation units, including NE1K, NE4B, SE3C, and SE5F, were sampled in this analysis ( Fig 2 ). The only specimen that lacked diagnostic cultural modification is sample 1593–5, an elk premolar/molar selected from an excavation level where other diagnostic postcranial deer and elk elements were not identified. These four units were selected as previous research by Phebus and Drucker [ 40 , 41 ] and Connolly [ 42 ] places two of the four units, SE3C and SE5F, within the rectangular plank house feature. Units NE1K and NE4B lie to the north of the plank house feature but within possibly intact and stratified midden deposits. However, it is important to note that a portion of NE4B was impacted by looters affecting the integrity of the southern portion of the unit. Nonetheless, the selection of units from within and outside the plank house provides the opportunity to accurately date the overall Palmrose occupation and duration of the plank house habitation.
I selected three samples per unit, each from distinct arbitrary excavation levels and different strata within each excavation unit. In general, I selected samples from the basal, intermediate, and upper deposits of the unit to measure the site occupation’s extent. Where possible, I attempted to select specimens that did not crosscut strata noted by the original excavators. However, given the complex stratigraphy of the site, that was not always possible. I used a Dremel TM drill to remove at least one gram of bone. Samples were sent to the W.M. Keck Carbon Cycle AMS Laboratory, University of California, Irvine (UCIAMS) for AMS radiocarbon dating.
AMS methods
At UCIAMS, bone collagen was extracted and purified using the modified Longin method with ultrafiltration [ 50 , 51 ]. Samples (200–400 mg) were demineralized for 24−36 h in 0.5 N HCl at 5°C, followed by a brief (< 1 h) alkali bath in 0.1 N NaOH at room temperature to remove humates. The pseudomorph was rinsed to neutrality in multiple changes of 18.2 MΩ H 2 O, and then gelatinized for 10 h at 60°C in 0.01 N HCl. Gelatin solution was pipetted into precleaned Centriprep® 30 ultrafilters (retaining >30 kDa molecular weight gelatin) and centrifuged three times for 20 min, diluted with 18.2 MΩ H 2 O, and centrifuged three more times for 20 min to desalt the solution. More detailed ultrafilter cleaning methods are described by McClure and colleagues [ 52 ]. Ultrafiltered collagen was lyophilized and weighed to determine the percent yield as a first evaluation of the degree of bone collagen preservation. All δ 13 C and δ 15 N values were measured to a precision of <0.1‰ and <0.2‰, respectively, on aliquots of ultrafiltered collagen, using a Fisons NA1500NC elemental analyzer/Finnigan Delta Plus isotope ratio mass spectrometer. Sample quality was evaluated by % crude gelatin yield, %C, %N, and C:N ratios before AMS radiocarbon dating. C:N ratios for the samples ranged from 3.2 to 3.4, indicating good collagen preservation and within the threshold advocated for by DeNiro (2.9−3.6) and van Klinken (3.1−3.5) [ 53 , 54 ]. Given the initial collagen yield of 0.9% for UCIAMS 229652, the sample was reanalyzed as sample UCIAMS 229653 with a collagen yield of 2.0%. However, both dates are included in this study. Radiocarbon samples (~2.5 mg) were combusted for 3 hours at 900°C in vacuum sealed quartz tubes with CuO wire and Ag wire. Sample CO 2 was reduced to graphite at 550°C using H2 and a Fe catalyst, with reaction water drawn off with Mg(ClO 4 ) 2 [ 55 ]. Graphite samples were pressed into targets in Al cathodes and loaded on the target wheel for AMS analysis. Radiocarbon ages were corrected for mass-dependent fractionation with measured δ13C values on the AMS [ 28 ] and compared with samples of 14 C free whale bone and mammoth bone.
Bayesian statistical modeling
The construction and modeling of archaeological chronologies through Bayesian approaches incorporates prior information about the archaeological site(s) and regional cultural histories, emphasizing the context, provenience, relative dating, and stratigraphic relationships of samples [ 13 , 27 , 38 , 56 , 57 ]. Given that the primary goal in the current research is to provide a reliable and precise chronological model for the occupation of the rectangular plank house structure and the Palmrose site generally, half the samples in the present study are derived from excavation units within the plank house feature. The remaining samples derive from north of the house feature in sediments interpreted as stratigraphically intact by Phebus and Drucker [ 40 , 41 ]. Therefore, the prior knowledge used in chronological models’ construction includes archaeological context, stratigraphic, and sedimentary data derived from archival field notes and previously published reports [ 40 – 42 ].
In this analysis, radiocarbon dates were calibrated using the IntCal20 Northern Hemisphere calibration curve and Bayesian models developed and tested in OxCal 4.4 [ 46 , 47 ]. Bayesian modeling allows researchers to statistically test potential chronological events providing probabilities for terminus post quem , terminus ante quem , chronological sequence, phase(s), and their chronological span [ 46 , 56 ]. As noted by Bronk Ramsey [ 46 ], a vital consideration of any chronological model is the recognition that stratigraphic information may not necessarily reflect chronological order; therefore, individual agreement indices and three other indices, model agreement, overall agreement, and convergence, are crucial.
OxCal chronological modeling calculates an individual agreement (A) index for each dated item or sample and an index for the model ( A model ), which is a measure of the agreement between the model and the observed data [ 46 ]. An overall agreement ( A overall ) index for the model is also determined, calculated from the individual agreement indices [ 46 , 56 ]. Individual sample indices, model indices, and overall indices can have a 100% value but can be higher and might fall as low as 60% to 0%. As Bronk Ramsey [ 46 ] notes, model agreement indices should not fall below 60%. If the model agreement index falls below 60% (analogous to 0.05 significance level in a Χ 2 test), the radiocarbon results or the models are problematic [ 56 ]. Therefore, these various agreement indices allow researchers to test unreliable models, dates, or identify intrusive dates [ 46 , 56 ].
The combination of Bayesian analysis and chronological modeling of 12 new AMS radiocarbon dates for the Palmrose site and the plank house feature, along with field notes and provenience information, provides an excellent opportunity to create a revised and precise chronology for the Palmrose site. These new data have the potential to change our understanding of site chronology, the development of semi- to fully-sedentary lifeways on the northern Oregon Coast, and alter regional chronologies broadly [ 6 , 9 ].
In the models’ construction, I assumed that all deposits were in undisturbed stratigraphic order, based on information in the existing field notes. To test this assumption and the stratigraphic integrity of the site and radiocarbon samples, I initially created simple calibration models and sequences for individual excavation units, applying priors from stratigraphic levels within each unit, before constructing more intricate chronological modeling following Sanchez and colleagues [ 9 ]. Radiocarbon dates were placed in a sequence in OxCal with boundary start and end dates calculated.
Results
Eleven of the 12 samples produced sufficient collagen yield. However, the analysis of sample 1593–9 from unit SE5F level 5 resulted in zero collagen yield and was not processed further. As previously noted, sample 1593–4 has duplicate dates resulting in 12 new AMS dates for the site. The conventional radiocarbon ages for the 11 samples range from 2135 ± 20 to 1785 ± 20 ( Table 2 ). Based on unmodeled calibration for the 12 dates, the site may have been inhabited from 345−55 cal BC to cal AD 225−340. To test the general stratigraphic integrity of the samples, I calibrated each unit through Bayesian methods. I created sequences for each unit by organizing samples based on excavation levels and included start and end boundaries to test for and identify radiocarbon reversals before merging all dates in a broader chronological model integrating additional stratigraphic data. All radiocarbon ranges presented below represent 95.4% probability.
10.1371/journal.pone.0255223.t002
Table 2 Conventional and calibrated AMS 14 C dates on cervid bone from the Palmrose site. Context designations derived from Phebus and Drucker field notes.
Sample ID
Taxon
Context
Element
UCIAMS #
δ 13 C (‰, VPDB)
δ 15 N (‰, Atm N2)
C/N
Provenience(Unit-Level)
Conventional 14 C Age BP
cal BC/AD (95.4% CI)
1593–1
Odocoileus sp .
Basal ash lens/crushed shell above the subsoil
Calcaneus
229649
-22.6
2.5
3.2
NE4B- 8
2135 ± 20
345−55 BC
1593–2
Cervus elaphus
Terminal crushed shell and humus deposits
Phalanx
229650
-21.1
2.9
3.3
NE4B-2
1845 ± 20
AD 125−240
1593–3
Odocoileus sp .
Intermediate deposits with crushed shell, humus, rock
Astragalus
229651
-24.2
3.1
3.2
NE4B-6
1930 ± 20
AD 25−205
1593–4
Cervus elaphus
Terminal crushed shell and humus deposits
Astragalus
229652
-21.7
3.9
3.3
NE1K-2
1810 ± 20
AD 205−330
1593–4 (Dup.)
Cervus elaphus
Terminal crushed shell and humus deposits
Astragalus
229653
-21.7
4.2
3.4
NE1K-2
1785 ± 20
AD 225−340
1593–5
Cervus elaphus
Intermediate crushed shell and humus deposits
Lower/Upper Premolar
229654
-21.2
5.2
3.2
NE1K-4
2125 ± 20
340−50 BC
1593–6
Cervus elaphus
Lower crushed shell deposits
Astragalus
229655
-25.1
3.7
3.3
NE1K-6
2100 ± 20
175−45 BC
1593–7
Odocoileus sp .
Basal ashy, crushed shell deposits
Astragalus
229656
-25.0
2.2
3.3
SE5F-10
2095 ± 20
170−45 BC
1593–8
Cervus elaphus
Terminal shell and humus deposits
Astragalus
229657
-21.3
3.2
3.3
SE5F-7
1815 ± 20
AD 170−330
1593–9
Cervus elaphus
Surface deposits
Astragalus
---
-25.3
3.6
---
SE5F-5
---
---
1593–10
Cervus elaphus
Lower crushed shell deposits
Calcaneus
229658
-24.9
4.2
3.3
SE3C-9
2035 ± 20
100 BC−AD 55
1593–11
Cervus elaphus
Ashy sand/crushed shell deposits
Astragalus
229659
-22.0
3.5
3.2
SE3C-7
1885 ± 20
AD 80−220
1593–12
Odocoileus sp .
Upper crushed shell and humus deposits
Astragalus
229660
-22.6
2.5
3.3
SE3C-3
1840 ± 20
AD 125−245
Chronology building first iteration: Excavation unit stratigraphic models
NE1K
The cervid remains from excavation unit NE1K, outside of the plank house structure, include three samples from levels 6, 4, and 2 ( Fig 3 ). Level 6 appears to represent shell midden deposits that overlie basal components of the occupation. Level 4 is an intermediate level that crosscuts three stratigraphic differences, including shell midden, crushed shell, and humus deposits. Level 2 represents terminal shell midden deposits overlaid by humus.
10.1371/journal.pone.0255223.g003
Fig 3
Palmrose unit NE1K east wall profile.
Level provenience of radiocarbon samples noted. Adapted from Palmrose excavation notes.
The conventional radiocarbon age for the specimens ranges from 2100 ± 20 BP to 1785 ± 20. In this modeling stage, samples from level 2, including two duplicates dates from the same sample, are combined using the R_Combine command. The agreement indices for the model are A model = 90.6 and A overall = 91.7 within the tolerance suggested by Bronk Ramsey [ 46 ]. No statistically significant stratigraphic reversals are present in the unit. The units modeled sequence suggests a possible start of occupation between 1125−65 cal BC and ending around cal AD 220−1145 with modeled radiocarbon dates from the basal and upper components of the midden spanning 180−60 cal BC to cal AD 215−325 , Table 3 .
10.1371/journal.pone.0255223.t003
Table 3 Radiocarbon dates for unit NE1K, including modeled sequence, 95% probability ranges, and boundaries.
UCIAMS #
Sample ID
Level
Conv. 14 C age (BP)
Modeled 95 . 4% CI (BC/AD)
Boundary
End of occupation
---
AD 220 − 1145
229653
1593–4
2 (Dup.)
1785 ± 20
---
229652
1593–4
2
1810 ± 20
---
R_Combine
1593–4
2
---
AD 215 − 325
229654
1593–5
4
2125 ± 20
155 − 50 BC
229655
1593–6
6
2100 ± 20
180 − 60 BC
Boundary
Start of occupation
---
1125 − 65 BC
NE4B
Radiocarbon samples from excavation unit NE4B are not within the plank house structure. The unit was partially disturbed by looters on its southern portion. Three samples from the unit were selected for analysis from levels 8, 6, and 2 ( Fig 4 ). Level 8 represents midden deposits near the basal component of the occupation, which crosscuts at least two stratigraphic differences noted by the field crew, including ashy deposits and shell midden, above the previously noted rocky subsoil. Level 6 crosscuts several stratigraphic differences within the unit, including crushed shell midden, humus, and rock deposits. Lastly, level 2 represents terminal shell midden deposits overlain by humus.
10.1371/journal.pone.0255223.g004
Fig 4
Palmrose unit NE4B east wall profile.
Level provenience of radiocarbon samples noted.
The conventional radiocarbon ages for the specimens range from 2135 ± 20 to 1845 ± 20. The agreement indices for the model are A model = 94.9 and A overall = 95.6 within the tolerance suggested by Bronk Ramsey [ 46 ]. No statistically significant stratigraphic reversals are present in the unit. The units modeled sequence suggests a possible start of occupation at 1130−55 cal BC and ending around cal AD 130−1105 with modeled radiocarbon dates from the basal and upper components of the midden spanning 340−50 cal BC to cal AD 125−240 , Table 4 .
10.1371/journal.pone.0255223.t004
Table 4 Radiocarbon dates for unit NE4B, including modeled sequence, 95% probability ranges, and boundaries.
UCIAMS #
Sample ID
Level
Conv. 14 C age (BP)
Modeled 95 . 4% CI (BC/AD)
Boundary
End of occupation
---
AD 130 − 1105
229650
1593–2
2
1845 ± 20
AD 125 − 240
229651
1593–3
6
1930 ± 20
AD 20 − 165
229649
1593–1
8
2135 ± 20
340 − 50 BC
Boundary
Start of occupation
---
1130 − 55 BC
SE3C
Samples from excavation unit SE3C are derived from deposits associated with and in the plank house feature, which offers the potential to approximate the extent of plank house occupation, potential rebuilding episodes, and site occupation. Three samples from the unit were selected for analysis from levels 9, 7, and 3 ( Fig 5 ). Level 9 represents shell midden deposits directly above the ashy sand stratigraphy identified by Phebus and Drucker and by OSMA archaeologists, which has been interpreted as the initial house building episode. Level 7 crosscuts a second ashy sand deposit and a shell midden deposit that overlies the basal ashy sand. Level 3 appears to represent shell midden deposits near the termination of the midden formation and occupation.
10.1371/journal.pone.0255223.g005
Fig 5
Palmrose unit SE3C east wall profile.
Level provenience of radiocarbon samples noted.
The conventional radiocarbon ages for the specimens range from 2035 ± 20 to 1840 ± 20. The agreement indices for the model are A model = 102 and A overall = 101.3 within the tolerance suggested by Bronk Ramsey [ 46 ]. No statistically significant stratigraphic reversals are present in the unit. The modeled sequence suggests a possible start of occupation around 1060 cal BC−cal AD 60 and ending around cal AD 130−1070 with modeled radiocarbon dates from the middens basal and upper components of the midden spanning 95 cal BC−cal AD 60 to cal AD 130−245 , Table 5 .
10.1371/journal.pone.0255223.t005
Table 5 Radiocarbon dates for unit SE3C, including modeled sequence, 95% probability ranges, and boundaries.
UCIAMS #
Sample ID
Level
Conv. 14 C age (BP)
Modeled 95 . 4% CI (BC/AD)
Boundary
End of occupation
---
AD 130 − 1070
229660
1593–12
3
1840 ± 20
AD 130 − 245
229659
1593–11
7
1885 ± 20
AD 80 − 215
229658
1593–10
9
2035 ± 20
95 BC − AD 60
Boundary
Start of occupation
---
1060 BC − AD 60
SE5F
Lastly, unit SE5F is also within the plank house structure. Three samples were submitted for radiocarbon dating from levels 10, 7, and 5 ( Fig 6 ). However, sample 1593–12 from level 5 lacked sufficient collagen preservation for radiocarbon dating—level 5 represented surface materials from the midden. Level 10 appears to be associated with ashy shell midden deposits above the subsoil, potentially indicative of initial site occupation. Level 7 represents terminal midden deposits above the middle ashy sand deposit found in SE3C, demonstrating consistency between the house occupation’s midden deposits.
10.1371/journal.pone.0255223.g006
Fig 6
Palmrose unit SE5F east wall profile.
Level provenience of radiocarbon samples noted.
The conventional radiocarbon ages for the specimens range from 2095 ± 20 to 1815 ± 20. The agreement indices for the model are A model = 100.4 and A overall = 100.4 within the tolerance suggested by Bronk Ramsey [ 46 ]. No statistically significant stratigraphic reversals are present in the unit. The units modeled sequence based on the available data suggests a possible start of occupation at 1150−55 cal BC and ending around cal AD 210−1200 with modeled radiocarbon dates from the middens basal and upper components spanning 170 cal BC−cal AD 55 to cal AD 130−325 , Table 6 .
10.1371/journal.pone.0255223.t006
Table 6 Radiocarbon dates for unit SE5F, including modeled sequence, 95% probability ranges, and boundaries.
UCIAMS #
Sample ID
Level
Conv. 14 C age (BP)
Modeled 95 . 4% CI (BC/AD)
Boundary
End of occupation
---
AD 210−1200
---
1593–9
5
---
---
229657
1593–8
7
1815 ± 20
AD 130−325
229656
1593–7
10
2095 ± 20
170 BC−AD 55
Boundary
Start of occupation
---
1150−55 BC
Chronology building second iteration: Palmrose occupation sequence
Based on the overall agreement between the sequences from the individual units, I constructed a chronological sequence model for the Palmrose site by analyzing the individual unit profiles for the four units. Overall, while there are differences across the four units, a general trend occurs across all four, which informed initial sample selection and the model’s construction. Each unit’s basal component includes sand or ashy sand deposits overlain by shell midden, except for unit NE1K. Therefore, the first boundary I included in the model was the ashy sand/ash and rock lens, which I termed Phase A ( Fig 7 ). These deposits include level 8 from NE4B and level 10 from SE5F, both within the house feature and interpreted as indicative of the house’s initial occupation. Next, I termed level 9 from unit SE3C as Phase B as the level generally corresponds with and includes shell midden that does not contain components of the ashy sand level, which it overlies. These stratigraphic components of the profiles correspond with shell midden deposits often noted as loose midden or crushed shell midden in the field notes.
10.1371/journal.pone.0255223.g007
Fig 7
Unit profiles with phase designations derived from stratigraphic data from archived Phebus and Drucker field notes.
Phase C comprises stratigraphic level 6 in units NE1K and NE4B, both of which appear to represent midden deposits above the basal deposits or the ashy sand but not associated with the intermediate sand lens, especially the small sand lens present in NE4B. Phase D is based on a single date from SE3C level 7, which crosscuts the intermediate sand lens’s upper deposits and the overlying shell midden. I interpret these deposits to represent the second house construction episode or, at a minimum, a reestablishment of the house floor through the addition of new sand. Lastly, Phase E represents shell midden overlying the second intermediate sand lens until midden formation ends.
The second iteration of chronological models informed by the stratigraphic variation fails due to two stratigraphic reversals within the model resulting in a model agreement index of A model = 0 ( Fig 8 ). Therefore, the second iteration model results informed the treatment of samples in the creation of the third iteration of modeling. Specifically, I excluded two significant outliers found in model two. Both outliers in the model derive from NE1K levels 6 and 4. While the exact cause of the stratigraphic reversals is unknown, given the complex stratigraphy for the site, evidence for multiple rebuilding episodes, and significant looting, it is not surprising to discover discontinuities in the site stratigraphy and radiocarbon reversals. As noted by Bayliss and colleagues [ 56 ] and Bronk Ramsey [ 46 ], such findings from model construction are critical a priori information that can and should be used in later iterations of model building.
10.1371/journal.pone.0255223.g008
Fig 8
Results of the second iteration of the Palmrose chronological modeling.
Chronology building third iteration: Palmrose occupation sequence excluding outliers
With the exclusion of the outliers from NE1K, the constructed Bayesian model of the Palmrose sequence based on my interpretation of the stratigraphy of the unit profiles suggests that the site and plank house’s primary occupation may have spanned from 580−55 cal BC to cal AD 210−300 , based on modeled start and end calculations in the sequence ( Fig 9 and Table 7 ). The agreement indices for the model are A model = 124.9 and A overall = 127.6. Based on the two radiocarbon assays—level 8 from NE4B and level 10 from SE5F, both within the house feature—Phase A, the basal sandy ash lens overlying the subsoil may have been occupied from 195−50 cal BC (Figs 7 and 9 ).
10.1371/journal.pone.0255223.g009
Fig 9
Results of the third iteration of the Palmrose chronological modeling.
10.1371/journal.pone.0255223.t007
Table 7 The third iteration of the Palmrose chronological modeling, including modeled sequence, 95% probability ranges, and boundaries. Model indices: Amodel = 124.9 and Aoverall = 127.6.
Name
Modeled 95 . 4% CI (BC/AD)
Agreement
Convergence
Difference Span
855 to 290 cal yr
97.5
Boundary End
AD 210−300
99.5
R_Combine Duplicate
AD 210−255
87
99.7
R_Date 229657-SE5F-L7
AD 205−250
129.7
99.8
R_Date 229650-NE4B-L2
AD 195−245
132.3
99.8
R_Date 229660-SE3C-L3
AD 200−245
137.1
99.7
Phase E
Boundary Loose Shell and Above
AD 160−240
99.6
R_Date 229659-SE3C-L7
AD 120−215
104.4
100
Phase D
Boundary Ashy Sand and Above
AD 65−210
99.9
R_Date 229651-NE4B-L6
AD 25−155
106.7
99.9
Phase C
Boundary Loose and Mixed Shell
50 BC−AD 125
99.9
R_Date 229658-SE3C-L9
95 BC−AD 25
104
99.9
Phase B
Boundary Shell Above Ashy Sand
150−1 BC
99.9
R_Date 229656-SE5F-L10
165−50 BC
103.3
99.8
R_Date 229649-NE4B-L8
195−50 BC
84.8
99.7
Phase A
Boundary Basal Ashy Sand Matrix
270−50 BC
99.5
Boundary Start
580−55 BC
97
Sequence
Phase B spans from 95 cal BC−cal AD 25 and is derived from a single date from unit SE3C level 9. It represents shell midden that does not contain components of the basal ashy sand level that it overlies or the sand and ash lens above. Phase C is represented by a single date from NE4B level 6 and spans from cal AD 25−cal AD 155 and appears to represent midden deposits above the basal ashy sand but not associated with the intermediate sand lens, especially the small sand lens present in NE4B.
Phase D spans from cal AD 120−215 and is derived from a single date from unit SE3C level 7. Based on stratigraphic data, this sample crosscuts the intermediate sand lens’s upper deposits and the overlying shell midden. I interpret these deposits to represent the second house construction episode or, at a minimum, a reestablishment of the house floor through the addition of new sand.
Lastly, Phase E represents shell midden above the intermediate sand lens and spans from cal AD 200−255 . These data are derived from five radiocarbon dates from SE3C level 3, NE4B level 2, SE5F level 7, and NE1K level 2. Therefore, the Bayesian models I constructed in this study based on my interpretations of the site stratigraphy, previous interpretations of the site, and the Bayesian modeling results may indicate three occupation phases and two house rebuilding episodes. The first occupation occurred sometime between 195 cal BC−cal AD 25 during Phase A and B. The second building episode spanned cal AD 25−215 sometime between Phase C and D. The terminal occupation occurred sometime between cal AD 200−255 during Phase E.
Chronology building fourth iteration: Palmrose house occupation sequence
As previously mentioned, one of the primary goals of the present study is to define the duration of the Palmrose plank house occupation. Given that two excavation units directly correlate to the house feature, I created a fourth chronological model using these units solely. Like the third chronological model, the fourth model relies heavily on the excavation unit profiles and stratigraphic data reported by Phebus and Drucker [ 40 , 41 ]. For example, the basal component of unit SE3C level 10 and 9 and unit SE5F level 11 and 10 include ashy, sandy, rock, clay deposits, and dense shell midden that overlie the rocky subsoil of the Palmrose site.
Unit SE3C levels 8 and 7 and SE5F levels 9 and 8 indicate a change in stratigraphy with a second intermediate ashy sand, crushed shell, and humus lens that I have interpreted as a second house occupation and reestablishment of an interior floor. Ancient floor zones often include ash and charcoal [ 58 ]. Hearths were often sand lined in Northwest Coast style plank houses [ 58 ]. Lastly, levels 6–2 in unit SE3C and levels ~7–6 in unit SE5F have been interpreted as indicative of the plank house’s terminal occupation. Based on these observances, the fourth chronological sequence may provide a refined model for the house occupation.
The constructed Bayesian model results for the sequence suggest the maximum range of the plank house occupation may have spanned from 1360−10 cal BC to cal AD 170−430 , based on modeled start and end calculations in the sequence ( Table 8 , Fig 10 ). The agreement indices for the model are A model = 119.4 and A overall = 118.4.
10.1371/journal.pone.0255223.g010
Fig 10
Results of the fourth iteration of the Palmrose chronological modeling.
10.1371/journal.pone.0255223.t008
Table 8 The fourth iteration of the Palmrose chronological modeling, including modeled sequence, 95% probability ranges, and boundaries. Model indices: Amodel = 119.4 and Aoverall = 118.4.
Name
Modeled 95 . 4% CI (BC/AD)
Agreement
Convergence
Difference Span
1670 to 250 cal yr
97.9
Boundary End
AD 170−430
99.4
R_Date 229657-SE5F-L7
AD 170−320
115.9
99.8
R_Date 229660-SE3C-L3
AD 170−250
128.1
99.9
Phase Terminal House Occupation
Boundary Above Ashy Sand
AD 130−240
99.9
R_Date 229659-SE3C-L7
AD 80−220
101.1
99.9
Phase Second House Occupation
Boundary Ashy Sand
60 BC−AD 200
99.7
R_Date 229658-SE3C-L9
100 BC−AD 30
96.9
99.9
R_Date 229656-SE5F-L10
160−1 BC
100.2
99.8
Phase Initial House Construction
Boundary Basal Ashy Sand Matrix
420−10 BC
99.6
Boundary Start
1360−10 BC
98.4
Sequence
The Bayesian modeling suggests that three plank house occupation periods may have occurred. The first suggests an occupation from cal BC 160−cal AD 30 and is associated with midden deposits above the subsoil, including ashy sand, rock, clay, and shell midden in units SE3C and SE5F ( Fig 11 ). The second occupation represented by a single date from unit SE3C and derived from the second intermediate sand lens or the shell midden overlying the sand deposit suggests that the second occupation period may have spanned from cal AD 80− cal AD 220 ( Fig 11 ). Lastly, based on two radiocarbon dates—one from SE5F level 7 that overlies the sand ash lens and another date from SE3C level 3—the plank house’s terminal occupation likely spans from cal AD 170−cal AD 320 ( Fig 11 ).
10.1371/journal.pone.0255223.g011
Fig 11
Unit SE3C and SE5F profiles with house occupation designations derived from reported stratigraphic data by Phebus and Drucker (40, 41).
Discussion
Radiocarbon dating by Phebus and Drucker [ 40 , 41 ] and Connolly [ 42 ] suggest that the Palmrose site was inhabited from 2340 cal BC to cal AD 640. Interpretations of the dates by Phebus and Drucker [ 40 , 41 ] and Connolly [ 42 ] suggest that the plank house may have been inhabited in three episodes, with the earliest occurring from 700−600 cal BC, intermediate occupation from 300−200 cal BC, and the terminal occupation around cal AD 200−300.
In this study, the third iteration model suggests start and end boundaries from 580−55 cal BC to cal AD 210−300 . The fourth iteration model start and end boundaries range from 1360−10 cal BC to cal AD 170−430 . The third model iterations indicate the possibility of three occupation episodes, dated between 195 cal BC−cal AD 25 , cal AD 25−215 , and cal AD 200−255 . The fourth model suggests the three occupations of the house may have occurred from cal BC 160−cal AD 30 , cal AD 80−cal AD 220 , and cal AD 170 −cal AD 320 . These results are in sharp contrast to previous reports. The reasons for that are that the models presented in this study would constrain the maximum range of the site and plank house occupation to 1360−10 cal BC to cal AD 170−430 (fourth model iteration), based on modeled start and end calculations, rather than 2340 cal BC to cal AD 640 as suggested by Phebus and Drucker [ 40 , 41 ] and Connolly [ 42 ]. The models in this study indicate that the three occupations of the plank house likely occurred in a much-constrained period and likely indicate a continuous occupation of the site. These new data affect regional chronologies and interpretations of human subsistence, occupation, and human-animal relationships across time and space discussed further below.
Reconsidering the Seaside regional chronology: Implications for human-environmental relationships and subsistence practices
The long-standing regional chronology for the Seaside area was primarily comprised of radiocarbon dates from Palmrose, Avenue Q, and Par-Tee, with the majority of radiocarbon assays derived from Phebus and Drucker’s work. These data suggested the Palmrose site was inhabited from 2340 cal BC to cal AD 640, Avenue Q from 1925 cal BC to cal AD 995, and Par-Tee from 350 cal BC to cal AD 1150 [ 40 – 42 ]. Therefore, the previous Seaside regional chronology suggested that the Seaside area’s initial occupation began with the Palmrose occupation, followed by Avenue Q. It was long thought that Palmrose and Avenue Q were both occupied contemporaneously. Lastly, it was believed that the initial occupation of Par-Tee overlapped for a limited time with Palmrose and a more extended period with Avenue Q.
The Palmrose site economy has been interpreted as more terrestrially, marine, and riverine focused, while marine taxa dominate Par-Tee [ 30 , 31 ]. These interpretations are derived from extensive faunal museum collections. For example, Colten [ 30 , 31 ] suggests Palmrose has more bones of migratory marine mammals, such as northern fur seals ( Callorhinus ursinus ) and Steller sea lions ( Eumetopias jubata ), than Par-Tee. The Par-Tee marine mammal assemblage has many more bones of sea otters ( Enhydra lutris ) and harbor seals ( Phoca vitulina ) than Palmrose. In terms of birds, Par-Tee has many more pelagic bird species, such as albatross (Diomedeidae), shearwaters ( Puffinus sp.), and murres ( Uria aalge ), than Palmrose. In contrast, Palmrose has the remains of more coastal and estuary birds, such as cormorants (Phalacrocoracidae), ducks and geese (Anatidae), and grebes (Podicipedidae) Colten [ 30 , 31 ].
Sanchez and colleagues [ 59 ] recently conducted an ichthyofaunal analysis of the Par-Tee collection and compared their findings to previously reported data from the Palmrose and Avenue Q sites. As previously mentioned, the Palmrose faunal assemblage was recovered with 1/8 in. mesh sieves. The Palmrose site is dominated by salmon ( Oncorhynchus sp.), representing 67% of the site assemblage. Therefore, it appears that the fishery’s focus was directed toward the acquisition of salmon supplemented by other fishes. Avenue Q was also recovered with 1/8 in. mesh sieves, with the fishery divided across multiple species including greenlings (Hexagrammidae), surfperches (Embiotocidae), skates (Rajidae), and hakes (Merlucciidae), among others, and suggest more variability and diversity in fishing practices, as no single fish organism dominates the assemblage as evidenced at Palmrose. Therefore, the Avenue Q fishery likely represents a broad-based fishery.
At Par-Tee, Phebus and Drucker recovered the faunal assemblage using 1/4 in. mesh sieves. It appeared to be a broad-based hook and line fishery focused on large fishes such as sturgeon ( Acipenser sp.) and large predatory fishes such as lingcod ( Ophiodon elongatus ), rockfish ( Sebastes sp.), and cabezon ( Scorpaenichthys marmoratus ) with limited evidence for salmon fishing. The inclusion of the fauna from bulk sediment samples hint at the possibility that mass-capture techniques were practiced targeting herrings (Clupeidae), Pacific tomcod ( Microgadus proximus ), smelts (Osmeridae), and Northern anchovy ( Engraulis mordax ) [ 59 ].
Previous research regarding the potential for cetacean hunting at Par-Tee is also significant. Losey and Yang [ 34 ] suggested the possibility that opportunistic whaling for humpback whales ( Megaptera novaeangliae ) occurred at the site. Radiocarbon dating by Sanchez and colleagues [ 36 ] suggested that the potential whaling event occurred around cal AD 430−550. Analysis of the Par-Tee and Palmrose marine mammal assemblage by Colten [ 30 , 31 ] suggested that cetacean remains differ between the sites. Both sites had significant numbers of harbor porpoises ( Phocoena phocoena ). The Palmrose site had the remains of many bottlenose dolphins ( Tursiops truncata ), while Par-Tee has larger cetacean bones, notably those of Minke whale ( Balaenoptera acutorostrata ) and humpback. Subsequent analysis of the larger cetaceans by Wellman and colleagues [ 35 ] suggests the use of stranded whales may have been more common than opportunistic whaling. Analysis of small cetaceans by Loiselle [ 32 ] suggests that Par-Tee residents were more frequently hunting rather than scavenging the small cetaceans, predominantly harbor porpoise, Dall’s porpoise ( Phocoenoides dalli ), bottlenose dolphin ( Tursiops truncatus ), and Pacific white-sided dolphin ( Lagenorhynchus obliquidens ).
Consideration of the variation between the Palmrose and Par-Tee assemblages and, to a lesser extent Avenue Q, reveal several interpretations to explain these differences. First, the variation may result from the chronological separation of the sites [ 30 , 31 ]. Second, there is the possibility of environmental variation in the Seaside area due to the potential infilling of an ancient bay in the sites’ vicinity [ 60 ]. Third, a cultural explanation has been suggested offering the possibility that different ethnic or tribal groups were living in close proximity, possibly reflective of historical patterns of Tillamook and Clatsop indigenous communities residing in Seaside at the time of European colonization [ 31 ]. Fourth, the variation may be driven by economic differences between the sites given the presence of a plank house at Palmrose and the lack of unambiguous residential structures at Par-Tee, especially as plank houses have been interpreted as the primary economic production and storage centers [ 60 – 62 ]. Fifth, the potential for differences in seasonal occupations of the sites [ 60 ].
The recent radiocarbon dating of the Palmrose and Par-Tee sites offers insights into the feasibility of these various possibilities. First, the interpretation that temporal differences between the Palmrose and Par-Tee sites may explain these differences is unlikely based on the refined chronology. Rather than the Palmrose and Par-Tee site occupations ranging from 2340 cal BC to cal AD 640 and 350 cal BC to cal AD 1150, the new chronological models suggest the maximum extent of the Palmrose occupation occurred from 1360−10 cal BC to cal AD 170−430 (fourth iteration model), but could be as constrained as 580−55 cal BC to cal AD 210−300 (third iteration model). The occupation of Par-Tee ranged from cal AD ~100−800 . Therefore, the chronological difference between the two sites changes significantly.
Regarding environmental variation between the two sites, the revised chronology suggests previous chronological research related to the timing of the infilling of an ancient bay near Seaside needs to be reconsidered [ 60 ]. As the radiocarbon dates reported by Connolly [ 60 ] and Phebus and Drucker [ 40 , 41 ] provided the basis for the analysis of molluscan remains by Connolly [ 60 ] and the subsequent interpretations of shifts in estuarine shellfish to open coast species, the findings of Sanchez and colleagues [ 9 ] and the present study strongly suggest the presently reported timing of the bay infilling should be reconsidered and reinvestigated, due to the inclusion of radiocarbon samples which do not adhere to chronometric hygiene standards as applied in this study. However, this study’s findings suggest the timing of the bay infilling occurred much more recently than previously believed. The present study cannot offer further support or refute interpretations regarding potential ethnic, seasonal, or economic variation between the sites or the use of different habitats by site inhabitants. However, the faunal data summarized does suggest differences in economic activities between the Palmrose and Par-Tee sites.
Conclusions
AMS radiocarbon dating and Bayesian modeling for the Palmrose and Par-Tee sites significantly alter site-specific and regional chronological models altering interpretations regarding human economic and environmental variation across space and time. The study suggests the Palmrose site was inhabited much more recently than previously believed and indicates the antiquity of fully- to semi-sedentary communities along the Oregon Coast needs to be reconsidered. In addition, the revised Palmrose chronology, along with the Par-Tee site chronology, suggests the sites overlapped in their occupations. These data possibly constrain the potential infilling of the former bay near Seaside and affect interpretations of the material record differences between the sites. These findings are consistent with recent Bayesian analyses and chronological studies of previously reported radiocarbon dates applying chronometric hygiene assessments developed for northern Oregon Coast sites [ 1 – 12 ] and support these previous studies’ findings. It demonstrates how AMS dating of museum collections can increase the scientific value of these collections while contributing information to chronologically situate forthcoming and future analyses of the Palmrose and Par-Tee collections. The results of this study suggest the Avenue Q assemblage would benefit from advanced chronological studies while also advocating for the use of short-lived or unambiguous samples in future radiocarbon dating of Oregon Coast sites.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: | Introduction Microarray gene expression profiling has become a common method for gaining insight into molecular disease mechanisms that are involved in host-pathogen interaction, and the outcome of the infection process, in terms of development of disease. The increasing public availability of microarray data allows for combining data in a meta-analysis, to identify common clusters of genes induced upon infection. Since most innate immune responses, especially those to pathogen-associated molecular patterns, are evolutionary conserved, it is likely that such common responses can be found. Indeed, it has been shown that under controlled in vitro conditions macrophages respond to a broad range of bacteria with a shared gene expression pattern [1] and similar findings have been described for dendritic cells [2] and peripheral blood mononuclear cells [3] . The meta-analysis of in vitro data by Jenner and Young [4] revealed a common infection cluster across a larger number of pathogens and studies, that included several genes for which this role was not recognized in the underlying studies. However, whether these findings in in vitro infection models are representative for what happens in vivo is unknown. Our laboratory recently published data on in vivo lung infection responses to respiratory syncytial virus (RSV) [5] and Bordetella pertussis [6] , [7] . Comparing these data sets showed several similarities as well as differences in the genes involved, although the kinetics of the responses was completely different [5] – [7] . To make additional comparisons and identify a common set of upregulated genes in different inflammatory diseases of the lung, we collected additional data for acute lung inflammation models from literature studies and other studies at our institute. As interpreting pair wise comparison between models is hampered by the large data size for each study, our goal was to use the data in a meta-analysis. Because the number of different pathogens or other exposures in each group is small compared to e.g. toxicogenomics experiments, it is not possible to determine pathogen-specific responses with sufficient certainty. The dataset is, however, suitable to detect responses common to all – or to a substantial number of – pathogens, or to reflect the “acute phase response” in the lung. Therefore, our aim was to employ a meta-analysis to identify gene expression changes in in vivo lung inflammatory models that are common to all, or subsets of, inducers of lung inflammatory lesions and pathogens. Results We combined microarray data from 12 studies [5] – [17] to compile a table containing gene expression data for 4551 genes upon 45 treatment conditions causing pulmonary pathology. These include 4 exposures to chemical (i.e. air pollution) sources, namely ozone and particulate matter (PM); 12 to bacterial infection models (including LPS, a mimic of infection); 19 to viral infections; 5 to parasitic infections; and 5 to allergic asthma models. More details on these exposures are given in Table 1 . Hierarchical clustering on the data showed that there was one cluster showing a similar response pattern across the rodent and macaque pulmonary inflammation models. This cluster contains 383 genes and is available as supplementary data at the journal's website (Supporting Dataset S1 ). Please note that, because gene symbols for rodents and macaques differ in which typecase is used, we will refer to genes in uppercase throughout this paper and supplementary data to connect to conventions for writing human gene symbols. 10.1371/journal.pone.0002596.t001 Table 1 Data and studies used in the meta-analysis; including labels used in Figure 1 . Source data Species Exposure Time points, labels Chemical Kooter, 2007 mouse ozone A: 12 h Kooter, 2005 rat Particulate matter B: 2–6 h, C: 15–21 h, D: 24–40 h Bacterial Banus, 2007 mouse (two strains : C3H and HcB28) Bordetella pertussis A: C3H 1 d, B: C3H 3 d, C: C3H 5 d, D: HcB28 1 d, E: HcB28 3 d, F: HcB28 5 d Lewis, 2008 mouse LPS aerosol G: 4 h Mycoplasma pulmonis H: 7 d Pseudomonas aeruginosa I: 2 h Rosseau, 2007 mouse Streptococcus pneumoniae J: 1 d, K: 2 d, L: 4 d Viral Janssen, 2007 mouse Respiratory Syncytial Virus A: 1 d, B: 3 d Kash, 2004 mouse Influenza (three strains: 1918, NC, and WSN) C: 1918 1 d, D: 1918 3 d, E: NC 1 d, F: NC 3 d, G: WSN 1 d, H: WSN 3 d Rosseau, 2007 mouse Influenza I: 1 d, J: 2 d, K: 4 d Baskin, 2004 macaque Influenza A: 4 d, B: 7 d Kobasa, 2007 macaque Influenza (two strains: K173 and 1918) C: K173 3 d, D: K173 6 d, E: K173 8 d, F: 1918 3 d, G: 1918 6 d, H: 1918 8 d Parasitic Lewis, 2008 mouse Nippostrongylus brasiliensis A: 5 d Reece, 2006 mouse Nippostrongylus brasiliensis B: 2 d, C: 3 d, D: 4 d, E: 8 d Allergic asthma Kuperman, 2005 mouse OVA A: 1 d Lewis, 2008 mouse OVA B: 1 d Aspergillus extract C: 4 h Zimmermann, 2003 mouse OVA D: 18 h Aspergillus extract E: 18 h Initial inspection showed that among the 383 genes there were a considerable number of immune and inflammation-related genes. Gene Ontology (GO) term enrichment analysis by the DAVID resource website ( http://david.abcc.ncifcrf.gov/ ) [18] , [19] showed that 100 of the 383 genes are annotated with the GO-term “immune response” (Fisher Exact p-value = 4.1E-43) and 40 genes are annotated with the term “inflammatory response” (Fisher Exact p-value = 1.1E-29) (Supporting Dataset S1 ). Also, the list shows substantial overlap with genes previously identified in in vitro experiments on host-pathogen responses. This includes 120 genes that were previously found by Jenner and Young [4] as being involved in the host-pathogen response; 82 genes described by Huang et al. [2] as dendritic cell common responsive genes; 42 genes described by Nau et al. [1] as being induced as part of the macrophage activation program; and 30 genes described by Boldrick et al. [3] as common responsive genes in peripheral blood mononuclear cells (Supporting Dataset S1 ). The largest overlap between the in vivo and in vitro responses is found for inflammatory cytokines and chemokines. Within the common up-regulated set of 383 genes, several immunological processes are represented by a substantial number of genes. These can be summarized in order of decreasing overlap with the in vitro studies. The functional class of genes that is most prominently up-regulated consists of cytokines and chemokines. Among these genes are interleukins ( IL1A , IL1B , IL5 , IL6 , IL12B , IL13 , IL15 ), CC-chemokines ( CCL2 ( MCP1 ), CCL4 ( MIP-1B ), CCL5 ( RANTES ), CCL7 ( MCP3 ), CCL8 ( MCP2 ), CCL11 (eotaxin), CCL17 ( TARC ), CCL19 ( MIP-3B ), CCL20 ( MIP-3A ), CCL22 ), CXC-chemokines ( CXCL1 ( GRO1 ), CXCL2 ( MIP-2A ), CXCL5 , CXCL9 ( MIG ), CXCL10 ( IP-10 ), CXCL13 ) and other cytokines such as CSF1 ( M-CSF ), CSF3 ( G-CSF ), TNF , SPP1 (osteopontin), AREG (amphiregulin). These genes not only show a consistent up-regulation in in vivo pulmonary inflammation studies, but also a clear overlap with genes found to be induced in vitro [1] – [4] (Supporting Dataset S1 ), corroborating the pivotal role these genes play in response to a wide range of agents. A second major group among the up-regulated genes are interferon-induced genes. These include guanylate binding proteins 1 and 2 ( GBP1 and GBP2 ), myxovirus resistance 1 and 2 ( MX1 and MX2 ), chemokines CXCL9 and CXCL10 , indoleamine-pyrrole 2,3 dioxygenase ( INDO ), and tryptophanyl-tRNA synthetase ( WARS ). In addition, these are several functionally less well characterized interferon-induced genes such as IFI27 , IFI30 , IFI35 , IFIT1 , IFIT2 , and IFITM3 . This set of genes contains both IFNα/β and IFNγ regulated genes. Other markers that are consistent with activation of the IFNγ response are several immunoproteasome components ( PSMB8 , PSMB9 , PSMB10 , PSME2 ). Increased expression of interferon-induced genes is mainly found for bacterial and viral infection models, illustrating the role of interferon in the innate immunity system's first line of defense against both viral and bacterial pathogens. Most of the genes in this class are also induced in in vitro models (Supporting Dataset S1 ). Although not as strongly induced as the previous classes, several genes involved in immunological signaling pathways were found to be consistently up-regulated. These belong to pathways such as interferon signaling ( IFNAR2 , IRF1 , IRF4 , IRF7 , ISGF3G , JAK2 , STAT1 , STAT2 ), NF-κB signaling ( NFKB2 , NFKBIB , NFKBIE , IKBKE , REL , RELB , TNFAIP3 ), AP-1 signaling ( JUNB , FOS , FOSL1 , FOSL2 ), MAPK signaling ( MAPK13 , MYC ), and TLR signaling ( TLR2 , CD14 , MYD88 , PIK3CD , PIK3CG , and TBK1 ). These pathways are interconnected, and several of the genes mentioned play a role in more than one signaling process. These interconnections between the various signaling routes makes the immunological signaling as a whole more robust and may also explain why the common in vivo response extends across all these pathways and even several dozens of other transcription factors that are not directly connected to these pathways. It is interesting to note that although the majority of the common up-regulated signaling genes activate the immune response, several inhibitory genes are also found for the NF-κB signaling pathway, namely NFKBIB , NFKBIE , TNFAIP3 . The expression patterns for these three genes are similar to other signal transduction genes. This indicates that the induction of these inhibitory genes is an equally important aspect of the response, as these genes will help restore the host cell to its normal state when the inflammatory stimulus is no longer present, thus keeping the system in check. For the immunological signaling genes, the overlap with in vitro studies is less pronounced than for cytokines and chemokines and interferon-induced genes. In addition to these three groups of immune-related genes, there are several smaller gene classes involved in the immune or infection response. These include known inflammation markers (such as S100A8 , S100A9 , LCN2 , SAA2 ), genes involved in the complement cascade ( C1QA , C1QB , C3 , C3AR1 , C5R1 ), cytotoxicity ( GZMA , GZMB ), or tissue remodeling ( CHI3L1 , MMP8 , MMP9 , MMP12 , MMP15 ). Others play a role in immune cell adhesion such as VCAM1 , integrins ( ITGAM , ITGAX , ITGB2 ), or selectins ( SELE , SELL , SELP ). Finally, some genes are not exclusive to the immune system but connect this to other cellular functions such as genes involved in leukotriene and prostaglandin metabolism ( ALOX5AP , ALOX12 , ALOX15 , PTGS2 , PTGES ), nitric oxide metabolism ( NOS2 ( iNOS ) and ARG2 ), or protection against oxidative stress ( SOD2 , HMOX1 ). Taken together, these processes play various roles in the immune or inflammatory response. The finding that a broad range of immune-related processes are induced across the different exposure categories shows that the in vivo pulmonary inflammatory response to various pathogens or exposures proceeds through – at least partly – shared molecular mechanisms. Apart from genes involved in immune or infection-related processes, we also found a substantial number of genes involved in cell division or proliferation. These include CCNA2 , CCNB1 , CCNF , CDC2A , CDC6 , CDC20 , CDKN1A ( P21 ), CDKN2D , AURKB , BUB1B , MKI67 , and UBE2C . For these genes, the strongest induction was observed for parasitic (helminth) and protein sensitization (allergic asthma) models. Surprisingly, there was practically no overlap between these genes and the genes mentioned in in vitro studies, suggesting a mechanism specific for the in vivo response. To further characterize gene expression response patterns, the full 4551-gene dendrogram was reduced to the most significant branches. Using the GeneMaths option Cluster Plot (see Methods section for details) provided a recommended branch cutoff level of 88% cluster similarity. Clipping the dendrogram at this level resulted in a total number of fifteen branches, ranging in size from 23 to 1713 genes. Most of these show no apparent induction or down-regulation upon exposure or show only a (moderate) change within a single study (data not shown). Five branches – which together form the common pulmonary inflammation response cluster of 383 genes – show an induction pattern that is consistent across multiple studies and/or exposures ( Fig. 1 , subset A–E). These five subsets display different levels and/or patterns of induction ( Fig. 1 , subset A–E). 10.1371/journal.pone.0002596.g001 Figure 1 Gene sets with common expression responses. Fragment of the hierarchical clustering dendrogram containing the abridged clusters for five common up- (A–E) and one down-regulated (Z) gene cluster. Expression changes compared to control levels are indicated by the color bar and corresponding ratio: green represents down-regulation, yellow no difference, red up-regulation. The heat map color gradient within each block indicates the variation within that block. Full details on the exposures used and the corresponding labels can be found in Table 1 and Supporting Table S1 . Among the five subsets group C, containing 23 genes, shows the strongest and most common response to lung inflammation in rodents, even extending to particulate matter. In primate influenza models this subset is also up-regulated, although not as strong as in rodent models. GO-term enrichment showed that this is the only subset clearly enriched for inflammatory response genes (Supporting Dataset S1 ). This is also reflected in the regulated genes ( Table 2 ), which include several known genes involved in inflammatory response such as the TLR4 co-receptor CD14 ; inflammatory cytokines such as IL1B and IL6 ; and chemokines like CCL2 , CCL4 , CXCL9 , and CXCL10 . Additionally, several well-known inflammation markers were present, including S100A8 and S100A9 (Calgranulin A and B), LCN2 (lipocalin 2 or NGAL ), and serum amyloid A2 ( SAA2 ). Pathway analysis by MetaCore (GeneGo, San Diego, CA) showed a high rank for the pathway “Toll-like receptor (TLR) ligands and common TLR signaling pathway leading to cell proinflammatory response”, which is consistent with up-regulation of proinflammatory cytokines and chemokines. 10.1371/journal.pone.0002596.t002 Table 2 Subset C of the common lung inflammation response, containing the 23 genes with the strongest and most common response across the various pathogens and exposures. Symbol Gene Name CCL2 Chemokine (C-C motif) ligand 2 CCL4 Chemokine (C-C motif) ligand 4 CCL7 Chemokine (C-C motif) ligand 7 CD14 CD14 antigen CXCL1 Chemokine (C-X-C motif) ligand 1 CXCL2 Chemokine (C-X-C motif) ligand 2 CXCL5 Chemokine (C-X-C motif) ligand 5 CXCL9 Chemokine (C-X-C motif) ligand 9 CXCL10 Chemokine (C-X-C motif) ligand 10 FCER1G Fc receptor, IgE, high affinity I, gamma polypeptide FCGR3 Fc receptor, IgG, low affinity III GBP2 Guanylate binding protein 2, interferon-inducible IFIT1 Interferon-induced protein with tetratricopeptide repeats 1 IL1B Interleukin 1, beta IL1RN Interleukin 1 receptor antagonist IL6 Interleukin 6 (interferon, beta 2) LCN2 Lipocalin 2 S100A8 S100 calcium binding protein A8 (calgranulin A) S100A9 S100 calcium binding protein A9 (calgranulin B) SAA2 Serum amyloid A2 TGFBI Transforming Growth Factor, beta-Induced, 68 kDa TGTP T-cell specific GTPase TIMP1 TIMP metallopeptidase inhibitor 1 A second subset showing a general response to pulmonary inflammation is group A. This subset is particularly induced upon bacteriological and viral infections in mice. Functional and pathway analysis showed that the genes in this subset are especially involved in interferon signaling, with the three highest-ranking MetaCore pathways being “Antiviral actions of Interferons”, “IFN alpha/beta signaling pathway”, and “Antigen presentation by MHC class I”. The last of these pathways is also mediated by interferon gamma through the formation of immunoproteasomes and the synthesis of the proteasome activator PA28 [20] . Among the 50 genes in this cluster are genes involved in interferon signaling such as STAT1 and STAT2 , and more than 15 interferon-induced proteins such as myxovirus resistance 1 ( MX1 ), tryptophanyl-tRNA synthetase ( WARS ), and indoleamine-pyrrole 2,3 dioxygenase ( INDO ) (Supporting Dataset S1 ). Subset D, containing 29 genes, shows mainly a gene expression response in parasitic and asthma models in mice. Both of these models are associated with T helper 2 (Th2) responses [15] , [17] . DAVID and MetaCore revealed that this is the only subset that is not enriched for immunological genes. Instead, it is enriched for cell cycle related genes, especially those involved in cell division such as cyclins A2, B1, F, and antigen MKI67 . Finally, there are two subsets with a general, although less pronounced, response. The larger of these is subset B, containing 182 genes. MetaCore indicated that this subset is enriched for several cytokine and chemokine signaling pathways, such as NF-AT, NF-κB, and MAPK signaling. This is reflected by the presence of genes such as IL5 , IL13 , NFKB2 , NFKBIB , and MAPK13 in this cluster. The smaller of the two subsets is cluster E, containing 99 genes. DAVID and MetaCore analysis indicated that this subset contains a broad range of immunological genes. High ranking MetaCore pathways were “Classic complement pathway”, “Lectin Induced complement pathway”, and “Alternative complement pathway”, based on complement and integrin genes such as C1QA , C1QB , C3 , C3AR1 , ITGAM , ITGAX , and ITGB2 . Other immunological genes in this subset include CCL8 , AREG , and NFKBIE . Unlike the other subsets, which are induced in rodents and (albeit to a lesser extent) in macaques, this subset is not induced in macaques but only in the rodent models. In addition to the common set of up-regulated genes, we also identified a cluster that displayed a general down-regulation. This cluster (which we will refer to as subset Z) contained 157 genes. DAVID and MetaCore showed this subset was enriched for development-related terms such as cell differentiation, organ development, blood vessel development, and growth factor activity. Some of the genes involved in these processes include BDNF , BMP4 , FGF1 , FIGF , IGFBP3 , TNNI3 , and WNT3A . Other processes that were overrepresented in this subset are muscle contraction and several metabolic processes. Only three immunological genes ( CD81 , PLUNC , EFNB1 ) were found among this subset. A complete listing of this subset is available as supplementary data at the journal's website (Supporting Dataset S2 ). Discussion Combining gene expression data from multiple studies creates the possibility to compare effects and look for common or specific responses. In this study, we focused on in vivo acute lung inflammation models. We included allergic asthma models and exposures to air pollutants, as these also cause pulmonary inflammation and therefore provide gene expression data to which the nature and the extent of infection responses can be compared. When data from different studies are combined, it should be kept in mind that not al studies are equally comparable, as there are differences between inflammation models as well as between species, time points, as well as other practical details on how the study was performed. However, combining studies also results in a larger data set, which allows for an analysis to reveal additional information that would not be apparent in the original studies used. When more microarray data on pulmonary inflammation models will become available in the future, it can therefore be expected that the number of identifiable common and specific responsive genes and pathways will increase. When different studies employ different methods in analyzing raw data this can cause unwanted (study-dependent) differences on the normalized data. For this reason we used the same normalization procedure on all raw two-color array data. Downloaded Affymetrix data were already normalized according to standard methods. Also, as the included studies used several kinds of microarrays, the initially collected data contain a large number of genes (mostly ESTs) for which only data from one or two studies are available. Therefore, to reduce the influence of missing data on the analysis, we also applied a filtering on the set of included genes, as described in the Methods section. The criteria were chosen so that a sufficiently large number of genes was included and small adjustments to the criteria had only a minor effect on the resulting clustering (data not shown). The data used contained information for 45 compared exposures that could be grouped into five main categories, namely chemical (air pollution related), bacterial, viral, parasitic, and allergic asthma models. These data led us to identify a common cluster of 383 genes with a similar in vivo response pattern characterizing acute lung inflammation. Of these 383 genes, 120 were previously identified as belonging to an in vitro common infection response [4] . Within this cluster there were subsets associated with more specific functional roles such as the response to bacterial and viral infection (subset A), cytokine and chemokine signaling (subset B), general inflammatory response (subset C), and the response to parasites and allergic asthma models (subset D). A closer analysis on these subsets could enable us to identify new genes that are of mechanistic importance and suitable as biomarkers to evaluate infection with unknown pathogens. When we compare gene expression profiles for the various exposures, it becomes apparent that the differences within each treatment category were smaller than those between categories. This indicates that in addition to a general common infection response there are additional, more dedicated, responses for categories of pathogens or treatments. An example of this is the difference we observed between responses to parasitic versus bacterial or viral infection. Although further elucidation of such responses would require data from a larger number of infection studies, the five subsets identified could serve as an initial starting point to see which processes are associated with a shared infection response to all or some categories of exposure. Among the chemical exposures, the most pronounced response was observed for subset C. This subset is involved in a general inflammatory response to pathogens, allergic asthma, and even air pollutants. An inflammation-related gene expression response to air pollutants corresponds to the finding that both particulate matter (PM) and ozone cause lung inflammatory and cytokine responses [11] , [12] . The response to ozone in the several subsets is both up- and down-regulated. This can be explained by the finding that in addition to an inflammatory response, ozone also causes suppression of immune responsive genes [12] , [21] . It has been suggested that the inflammatory response to PM is caused by the presence of bacterial endotoxins such as LPS in particulate matter [22] – [24] . Indeed, the response to PM correlates best with the bacterial infection models (including a mimic of infection, LPS), as is visualized in Fig. 1 . Our data underline that the response to LPS may be an important element of the response to PM. Bacteria induce gene expression in several subsets, and the response to LPS aerosol matches the other responses to bacterial infection in these subsets. Even though LPS aerosol is not an actual infection, it mimics exposure to a pathogen and accordingly induces an inflammatory response. Interestingly, this response is not strictly LPS-specific as the bacterial infections used include not only gram negative ( Bordetella , Pseudomonas ) bacteria, but also the LPS-lacking Mycoplasma and Streptococcus . This indicates that the response to bacterial infection is not only dependent on LPS signaling via TLR4 but signaling through other Toll-like receptors also plays an important role. A shared response in non-immunological genes was observed for parasitic (helminth) and protein sensitization (allergic asthma) models. Considering that both induce a Th2 response, this is not surprising. However, subset D, which shows the strongest response upon these exposures, did not include any Th2-associated cytokines, suggesting that the major expression changes take place downstream of Th2 cytokine signaling. The shared response between helminth infection and allergic asthma involves increased expression of cell cycle-related genes. In these models there is apparently a more rapid turnover or proliferation of cells than in other models. This can be explained by an increased lung epithelial renewal or proliferation, or alternatively by an increase in proliferation for immune cells involved. Although the latter possibility is feasible, the former matches known clinical pathology for asthma models. Asthma is associated with hyperplasia of the mucin-secreting goblet cells [25] and in the studies used this is also described for asthma as well as helminth infection [14] , [15] . This is in agreement with the finding that these genes are not found to be induced in in vitro models using isolated immune cells. As pulmonary inflammation often involves leukocyte infiltration, it raises the question whether the common responses occur primarily in sessile lung cells or can be attributed to infiltrating immune cells. Although the common upregulated cluster contains some markers associated with monocytes and macrophages ( e.g. CD14 , CD68 ) or lymphocytes ( e.g. CD72 , CD80 ), these do not show a parallel expression pattern, as would be expected if the gene expression responses are caused by cellular influx. In addition, the common cluster does not include several other markers associated with these types of immune cells, nor those associated with granulocyte lineages, even though several of these are present in the initial 4551 gene set used ( e.g. CD36 , CD19 , CD3E , CD7 , CD4 , CD8A , MPO ). For these markers that are not part of the common cluster, the expression changes to the controls are much weaker than those in the common cluster and the responses are also much less consistent across the various models (data not shown). Finally, two of the studies include time points where gene expression responses are at their maximum before a detectable cellular influx is found by pathological analysis, namely the response one day after RSV infection [5] and the early response (2–6 h) after PM exposure [11] . Therefore, it can be concluded that the common infection response can be predominantly attributed to gene expression changes in sessile pulmonary cells rather than to leukocyte influx. Besides a common up-regulated cluster, we also identified a cluster characterized by a common down-regulation in response to inflammation. This cluster contained a large number of genes involved in growth and development, both of which are important processes for continuous renewal of lung epithelium. As can be seen from Fig. 1 , the extent of the down-regulation in this cluster is moderate compared to the effects in the common up-regulated subsets. This suggests that the effects found are not a direct targeted response, but rather a secondary effect that reflects tissue damage or that the induction of an inflammatory response goes at the expense of normally active processes in lung tissue. The finding that the degree of down-regulation is strongest in groups with a stronger up-regulation for subsets A–D and especially E corroborates this assumption. In comparison with the transcriptional changes found in mice, the effects in primates are generally weaker. More specifically, the inflammatory response-related subset C, which shows the strongest response in mice, shows considerably less up-regulation in macaques. The degree of induction for the interferon signaling-related subset A is also reduced, albeit to a lesser extent. The other subsets show a more moderate response to influenza in both mice and macaques and for these subsets the difference in response is comparatively small, although it is interesting to note that the induction of subset E is virtually absent in macaques. A minor difference in effect is also found for subset Z, which is down-regulated in mice as well as macaques. These differences in expression response can not be ascribed to mere study differences as the effect was reproducible in two studies carried out with different species, namely cynomolgus macaques ( Macaca fascicularis ) in the study by Kobasa et al. [10] and pigtailed macaques ( Macaca nemestrina ) in the Baskin et al. [8] study. It can not be ascertained whether these differences are characteristic for either influenza or macaques, as microarray data for primate infections with other pathogens were not available. However, it is likely that differences in expression response between mice and macaques can be explained by different disease characteristics between these species. First, differences in effect are most distinct for the two subsets (A and C) where the mouse response upon influenza is strongest, which suggest an association between these responses and disease severity. Second, the influenza studies in mice [9] , [16] both reported more severe lung inflammation and pathology than those in macaques [8] , [10] . In macaques, the most severe pathology was observed in those infected with the 1918 virus strain. Of the three influenza strains used in the macaque studies, this particular strain causes a gene expression response that compares best to the common mouse response. This corroborates our assumption that the different expression response between rodents and macaques reflects the extent in which lung tissue is infected and/or the virus multiplies. In conclusion, our study shows that there is a shared in vivo expression response to different inducers of lung inflammation. This response comprises several processes involved in host defense and inflammation, and the extent of the response represents the degree of lung inflammation. Our meta-analysis shows considerable overlap with findings from in vitro studies (Supporting Dataset S1 ), especially in cytokines, chemokines, and interferon-induced genes. Some of the differences can be attributed to complex cell-cell interactions, that are absent from in vitro systems, such as the induction of the cell division-related subset D. However, as additional microarray data will allow for a more powerful meta-analysis that reveals more common genes for both the in vivo and in vitro response, genes that do not overlap between the common in vivo and in vitro response will not always be specific for either response and merely represent the developing knowledge in this field. In the future, additional microarray data from rodents, primates, and perhaps other mammals will contribute to a further understanding of the common in vivo response and, ultimately, identification of disease mechanisms that are unique to specific agents or pathogens. Methods We searched Pubmed, GEO ( www.ncbi.nlm.nih.gov/geo/ ), and ArrayExpress ( www.ebi.ac.uk/arrayexpress/ ) for gene expression profiling studies related to acute lung inflammation. If corresponding microarray data were available, they were downloaded from websites indicated by the authors. Data were included in the meta-analysis if they met the following conditions: (a) complete microarray raw or normalized data are available; (b) a suitable uninfected or mock infection control is included in the study; (c) time points are at most eight days after infection (to exclude chronic effects). Furthermore, we excluded experiments with transgenic pathogens or hosts focused on specific research questions, as these typically show inflated responses that are not representative of normal disease. Based on these criteria, we included 45 treatment conditions from 12 experiments [5] – [17] . Of these studies, 4 were carried out in our laboratory and 8 were from the literature. Note that we count the data from the two related articles by Banus et al. as one study. Full details of the studies are given in Table 1 . Affymetrix probe sets identifiers were converted to gene symbols using probe set annotation data downloaded from the Affymetrix website ( www.affymetrix.com ). When necessary, gene symbols in two-color or Affymetrix data files were adjusted to remove tags such as “predicted” and converted to uppercase symbols for further handling. All further calculations were carried out in R [26] or Microsoft Excel. To minimize the influence of data handling procedures, we normalized all raw two-color data with the same algorithm [5] , [27] . This consisted of a four-step approach of (1) natural log transformation, (2) quantile normalization of all scans, (3) taking the sample/reference ln-ratio, and (4) averaging replicate spot data. To remove negative values and inflated ratios, MAS4 normalized Affymetrix data were cut off at a lower value of 100, based on the findings of Grundschober et al. [28] . MAS5 normalized Affymetrix data were used without adjustment. Affymetrix data were ln-transformed and values for replicate gene symbols were averaged. Finally, for all data sets, the average ln-ratio for treatment to control was calculated per gene. Treatment ratio data for the various studies were merged into one table. To minimize the impact of missing data and non-regulated genes on further analysis, we restricted the initial table of 39312 genes to 4551 genes that were measured in at least 6 out of 12 studies, 30 out of 45 treatment conditions, and had a ratio exceeding ±1.5 in at least one condition. Hierarchical clustering on these 4551 genes was performed in GeneMaths (Applied Maths, St-Martens-Latem, Belgium), using Euclidian distance and Ward linkage. The GeneMaths option Cluster Plot was used to plot the logarithm of the cluster size versus the cluster similarity. This method typically results in a graph with data points that lie closely along a curve that drops off sharply at a level where the number of branches is sufficient to show enough detail, but increasing the number of branches does not result in much additional information. This level can be considered as a recommended cluster similarity cutoff. The resulting value of 88% cluster similarity was used to abridge dendrogram branches above this similarity value. The resulting dendrogram was used to identify branches or clusters with up- or down-regulation that cover multiple studies or pathogens and therefore indicate a common response. Functional annotation and Gene Ontology (GO) term enrichment analysis were performed with the DAVID website ( http://david.abcc.ncifcrf.gov/ ) [18] , [19] . For GO-term enrichment, the functional annotation clustering option was used, with default settings. Functional annotation terms were considered enriched for an Enrichment Score larger than 3, which corresponds to a geometric average p-value of 0.001. MetaCore (GeneGo, San Diego, CA) was used for additional pathway enrichment and visualization. Supporting Information Dataset S1 Genes involved in the common up-regulated infection response (0.25 MB DOC) Dataset S2 Genes involved in the common down-regulated infection response (0.12 MB XLS) Table S1 Full information on the studies and treatments included (0.00 MB PDF) |
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Since its discovery, much of the research of the type 1 transmembrane protein β-amyloid precursor protein (APP) has focused on its proteolytic components, particularly the β-amyloid (Aβ) peptide that accumulates in Alzheimer’s disease. However, full length APP is yet to be attributed a conclusive function. It has been described to have roles in transcriptional signaling, synapse formation, ion transport, neuroprotection and neuroplasticity [1] . Recently we have added to this growing functional list by reporting that both the full-length membrane bound and the cleaved soluble extracellular form of sAPPα, but not other family members amyloid precursor-like protein (APLP) 1 and 2, facilitate the efflux of iron from APP-expressing cells such as neurons [2] .
As an integral cofactor in many metabolic processes, iron must be closely regulated for the wellbeing of any cell, particularly where oxygen consumption is high such as in the neuron. The ability for iron to undergo redox-cycling is harnessed by some enzymes for catalysis [3] , however under aerobic conditions iron may also catalyze the production of reactive oxygen species (ROS) through the Haber-Weiss [4] and Fenton [5] reactions. Unregulated hydroxyl radical (OH•) and ROS production is damaging to the cell [6] and have been associated with aging and disease, particularly in neurodegenerative diseases such as Alzheimer’s disease, Parkinson’s disease and aceruloplasminemia, where iron accumulates in affected tissue [7] .
As both iron deficiency or excess may compromise cell viability, homeostasis is tightly controlled with cell entry, storage and exit [8] . Import of iron was previously considered to be solely through either divalent metal transporter 1 (DMT1) [9] or by transferrin import through interaction with the Transferrin Receptor (TfR) [10] . However, other import mechanisms have now been described including ZIP14 [11] , indicating that uptake of iron into the cell may not be as simple as previously thought. Currently, there is only one known iron export pore protein, ferroportin (FPN), which is believed to traffic Fe 2+ from the cytoplasm to the plasma membrane surface. While a variety of mechanisms are thought to facilitate the release of iron from the exofacial surface of FPN, multicopper ferroxidases such as ceruloplasmin (CP), hephaestin (Heph) and the bacterial ferroxidase Fet3, were previously considered the only facilitators of intracellular iron efflux. This was mediated through their ability to secure Fe 2+ from stabilized FPN on the cell surface and promote Fe 2+ oxidation for Fe 3+ loading into iron-transporting proteins such as transferrin (TF) [12] – [14] . We concluded that APP might also fulfill an analogous function for iron release [2] .
We found that the major proportion of APP in human and mouse post-mortem brain tissue samples is complexed to FPN, and that APP knockout mice markedly accumulate iron in several organs, including the brain [2] . Several reports have since corroborated the impact of APP expression on cellular iron levels [15] – [18] . This is likely to be the mechanism by which sAPPα is neuroprotective against glutamate excitotoxicity, and a point mutation within the REXXE motif within APP, a site common within other iron homeostatic proteins, negates this neuroprotection [2] . A peptide fragment of APP containing this motif was recently reported to interact with FPN and stabilize it on the surface of human brain microvascular endothelial cells [19] .
We also reported that APP could catalyze the oxidation of Fe 2+ through a mechanism we thought analogous to ferritin, a ferroxidase that does not have a multicopper active site [2] . Concerns have since arisen about the validity of this chemistry, and other groups were unable to show similar activity when using regions within APP containing the ferritin-homologous REXXE motif that we originally suggested was required for this activity [19] – [21] . Here we re-examine the mechanism for APP-promoted iron export, and whether APP enzymatically catalyzes the oxidation of Fe 2+ . Based on established assays used extensively to monitor iron oxidation, we developed a more reliable assay system [22] that enable studies in an environment that takes into account the physiological levels of phosphate (0.80–1.45 mM [23] , [24] ), and transferrin (25.5–45.0 µM [25] ). We confirm that APP stabilizes surface FPN, and that it is indeed unable to catalyze ferroxidation, which therefore appears to be irrelevant to its ability to facilitate cellular iron export.
Materials and Methods
Reagents
Reagents were all analytical grade and were purchased from Sigma (Australia), unless otherwise stated. Purified human ceruloplasmin was purchased from Vital Products (USA). Recombinant full-length human APP with a C-terminal fused Fc region of human IgG was purchased from Sino Biological Inc. (China). FPN was detected by antibodies, gifted by Prof. Tracy Rouault, raised to epitopes in either an extracellular (MAP23; α/α165–181) or intracellular (MAP24; α/α240–258) region of the protein.
Recombinant protein preparation
Recombinant human sAPP695α was expressed in the methylotrophic yeast Pichia pastoris strain GS115 and purified from culture media by FPLC (BioRad) as previously described [26] . Chromatographic purification was by anion exchange using Q Sepharose (1.6×25 cm column, GE Healthcare) followed by hydrophobic exchange with phenyl Sepharose (1.6×25 cm column, GE Healthcare). APP eluted in phenyl Sepharose buffer B (50 mM Na 2 HPO 4 , pH 7) was then concentrated using Amicon Ultra-15 Centrifugal Filter Units (30 kDa, Merck Millipore, Australia) and stored at –80°C as 20 µM stocks. Elimination of anionic buffer in original sAPPα, APP, CP and BSA stocks was carried out by buffer-exchange with 50 mM HEPES, 150 mM NaCl, pH 7.2 (HBS) using a Superdex 200 10/GL filtration column (GE Healthcare).
Direct and indirect enzymatic measurement of Fe 2+ oxidation
Three previously established procedures that either directly or indirectly spectrophotometrically measure catalyzed oxidation of Fe 2+ to Fe 3+ have recently been combined and adapted to a plate assay [22] . Assaying the consumption of Fe 2+ (“Ferrous Loss Assay”) was adapted from Erel [27] and measured the colorimetric change at absorbance 590 nm when the chromagen 3-(2-pyridyl)-5,6-difurylsulfonic acid-1,2,4-triazine sodium salt (Ferene S) is bound to ferrous ions compared to ferric ions. Fe 3+ production was directly measured by the absorbance (310 nm) of Fe 3+ (“Ferric Gain Assay”) as previously described by Minotti and Ikeda-Saito [28] . Finally, the physiological loading of apo-TF (apo-TF) with Fe 3+ , which induces an absorbance increase at 460 nm, was assayed using a procedure adapted from that originally reported by Johnson et al. [29] and used in the original investigation on apparent ferroxidation promoted by APP [2] . All of these reactions were performed at 24 °C so that the reactions would proceed more slowly to enable measurement with standard spectrophotometers [22] .
To a 200 µL reaction, the protein of interest (500 or 250 nM), or media, was added to HBS (50 mM HEPES, 150 mM NaCl, pH = 7.2) ± apo-TF (50 µM). Ferrous ammonium sulfate ((NH 4 ) 2 Fe(SO 4 ) 2 ·6H 2 O; 100 µM) was added to start the reaction. Absorbance readings (310 nM and 460 nM) were kinetically monitored with the Flexstation 3 (Molecular Devices) or PowerWave HT (BioTek) microplate spectrophotometer over 20 min at 30 s intervals with continual agitation and kept at 24°C. Ferene S (500 µM) was added at the end of 20 min and absorbance (590 nm) read. Reactions were blanked against HBS only and all dilutions are shown as final concentrations.
Preparation of mouse primary neuronal cultures
All animal procedures were approved by the Howard Florey Institute animal ethics committee (#12-078; Melbourne, Australia), and were carried out in accordance with the Australian code of practice for the care and use of animals for scientific purposes as described by the National Health and Medical Research Council of Australia. Pregnant mice were deeply anesthetized with isofluorane and then sacrificed by cervical dislocation, without suffering. Primary neurons cultures from the cortices of wild-type embryos were prepared from embryonic day 14 or 15 mice, as previously described [30] . Cortices were removed, dissected free of meninges, and dissociated in 0.025% trypsin. Cortical cells were plated onto poly-L-lysine (50 µg/ml)-coated 6-well plates (Nunc) at a density of 600,000 cells/cm 2 in DMEM supplemented with 10% fetal bovine serum (FBS; PAA Laboratories), 5% horse serum and 10 µg/ml gentamycin sulfate. The neurons were allowed to adhere for 2–3 h before the plating medium was replaced with Neurobasal-supplemented medium (serum free and with B27 minus antioxidants, 500 µM glutaMAX and 10 µg/ml gentamycin sulfate). On the day of the iron experiments the medium was replaced with fresh Neurobasal-supplemented medium and for all further experimentation the medium was serum-free.
Surface Biotinylation assay
Parental HEK293T cells, which constitutively express APP [31] but not CP [2] , [14] , were maintained in Opti-MEM media with 10% FBS. 48 hours after plating, cells were treated with serum-free Opti-MEM media (Life Technologies). Where required, HEK293T or wild-type mouse primary neurons cells were incubated with ferrous ammonium sulfate (FAS; 50 µM) for 3 h before washing with their respective serum-free media and then replaced with serum-free media ±1 µM sAPP695α or CP for 30 min, prior to surface biotinylation using the Pierce cell surface protein isolation kit (Thermo Scientific).
The procedure for cell surface biotinylation and isolation was carried out as per the manufacturer’s instructions and at 4°C to minimize protein internalization. In brief, cells were labeled with Sulfo-NHS-SS-Biotin in PBS for 30 min. Unreacted biotin was quenched by the Quenching solution provided, before cells were harvested by scraping and lysed using Lysis buffer+Phosphatase Inhibitor I and II (1∶1000). Cell lysates were clarified by centrifugation at 10,000 × g for 2 min and a sample of the supernatant was taken to measure protein levels in the ‘total homogenate’ fraction. To precipitate biotinylated surface proteins from each cell lysate, 200 µg of total protein was incubated with NeutrAvidin Agarose for 1 hour. Non-bound ‘intracellular’ proteins were separated by centrifugation (1,000 × g for 1 min) before agarose was washed three times, and bound proteins eluted with XT sample buffer (Bio-Rad) containing 50 mM DTT, followed by analysis by western blotting.
From each experimental condition total homogenate lysate starting material (10 µg protein), as well as samples of both non-biotinylated (‘intracellular’) and biotinylated (surface) eluates from the 200 µg of lysate separated by NeutrAvidin Agarose were separated on 4–20% PAGE (Bis-Tris, Invitrogen) and transferred to PVDF membrane using a Transblot (Bio-Rad). Primary antibodies used were rabbit anti-FPN (1∶1000, MAP23 recognizing residues 165–181 or MAP24 recognizing residues 240–258 [32] , kind gift from Tracey Rouault), mouse anti-APP (1∶1000, WO2 recognizing residues 672–699 [33] or 22C11 recognizing residues 66–81 [34] , in house), and rabbit anti-Cp (1∶5000, Dako; #Q0121). The load control was either mouse anti-β-actin (1∶5000, Sigma) or mouse anti-Na/K ATPase (1∶5000, Sigma). Proteins were visualized with ECL (Amersham) and a LAS-3000 Imaging suite, and analyzed using Multi Gauge (Fuji). Densitometry using Image J (v1.48k, NIH) was performed in triplicate on 3 separate experiments. All quantitation was standardized against β-actin and Na/K ATPase levels and secondary antibody alone was used to eliminate the risk of non-specific binding to proteins on the PVDF membrane.
Fluorescence-activating cell sorting analysis
Parental N2a neuroblastoma cells, were maintained in DMEM media (Lonza) containing 10% FBS until required. For siRNA transfections, cells were washed once with OptiMEM and incubated for 30 min at 37°C in fresh OptiMEM. During this incubation, APP (SMARTpool: ON-TARGETplus Human APP siRNA; Fisher Scientific Ltd) and scrambled (ONTARGETplus Nontargeting Pool; Fisher Scientific Ltd) siRNA were prepared at a final concentration of 50 nM in siRNA buffer (Thermo Scientific) and OptiMEM to a volume of 1 mL. In turn, this was added to 1 mL of OptiMEM containing 20 µL of Dharmafect 1 (Fisher Scientific Ltd), prepared separately. After 15 min incubation, 8 mL of DMEM containing 10% FBS was added to the 2 mL mixture. OptiMEM media was removed from cells and an appropriate volume of siRNA was added. Cells were incubated with the siRNA mixture for 36 hours at 37°C, 5% CO 2 prior to fluorescence-activated cell sorting (FACS) analysis.
Prior to the experiment, cells were pre-incubated with ferric ammonium citrate (50 µM) in OptiMEM for 6 hours in order to generate sufficient surface FPN expression to be detected on the subsequent FACS, to test whether this would be suppressed by RNAi of endogenous APP or increased by the subsequent 30 min addition of 1 µM sAPP695α in OptiMEM without iron. RNAi under baseline conditions (no additional iron) was not feasible because FPN immunoreactivity was just above the limit of detection, and an experimental treatment that might suppress FPN surface expression would be difficult to exhibit clearly. FACS preparation involved washing and collection of cells in PBS (without Ca and Mg) (Lonza) at room temperature. After cells were pelleted by centrifugation at 2000 rpm for 5 min the remaining procedure was performed at 4°C. Cells, re-suspended in ice-cold FACS buffer (PBS, 2.5 mM EDTA pH 8.0), were incubated with a primary antibody against N-terminal APP (1∶50; Abcam; ab15272) or FPN (1∶50; BioScience Life Sciences) for 30 min. FPN and APP antibodies used were raised to epitopes on extracellular domains and intensity of fluorescence was compared to cells stained with secondary only to minimize the detection of non-specific binding. Cells were washed with FACS buffer before lightly fixing with 1% paraformaldehyde (Alfa Aesar) for 4 min. Cells were then re-suspended in FACS buffer containing AlexaFluor 488 goat anti-rabbit IgG secondary antibody (1∶200; Life Technologies) for 30 min in the dark. Cells were further washed with FACS buffer before incubating with DAPI (Cell Signaling Technology) to differentiate dead cells. Cells were sorted by forward and side scatter on a BD-LSR-Fortessa (BD Biosciences) with a 488 nM blue laser according to fluorescence at 530±30 nM. Minimums of 10,000 cells were recorded in each experiment, having gated the cell population to ensure that only live cells were monitored. Experiments were carried out in duplicate on 3 separate occasions and data were analyzed using BD FACS DiVa 6 and FlowJo 7.6.4 software.
Confocal Microscopy
A coverslip was placed at the bottom of each well of a 24-well plate, on which wild-type mouse primary neurons were incubated at 50% confluency for 14 days in neuroblastoma media. Where required, cells were incubated with FAS (50 µM) for 3 h before the medium was replaced with fresh neurobasal-supplemented medium ±1 µM sAPP695α or CP for 30 min, prior to immunofluorescence staining. After each experimental condition, cells were washed with cold PBS and then fixed with 4% paraformaldehyde for 4 min at room temperature. Non-permeabilized cells were rinsed again in cold PBS before incubation with blocking buffer (5% (v/v) BSA) for 1 h at 4°C. Cells were then incubated with the appropriate primary antibodies diluted in blocking buffer for a further 2 h at 4°C. Primary antibodies used were rabbit anti-FPN (1∶100, MAP23) and mouse anti-APP (1∶50, 22C11). Cells were then incubated with the fluorescently conjugated secondary antibodies for 1 h in blocking buffer after further washes in PBS. Alexa Fluor 488 Goat anti-Rabbit IgG and Alexa Fluor 488 Goat anti-Mouse IgG (Millipore) were used in combination at 1∶500. Cells were washed a final time before counterstaining with DAPI (1∶1000) and mounted onto slides using fluoromount G mounting medium. A Leica SP8 confocal microscope was used to collect z-stacks of neurons, deconvoluted using the Huygens deconvolution software (Scientific Volume Imaging), and each stack compiled using Image J software.
Statistical Analysis
Statistical analysis was performed with Microsoft Excel 2011 and GraphPad Prism v5.0 software. Primarily, analysis was carried out with a 2-tailed t-test with the level of significance set at P = 0.05.
Results
Evaluating the in vitro ferroxidase activity of APP
We revisited the in vitro experiments testing sAPPα ferroxidase activity previously presented as part of our primary publication [2] . In a physiologically relevant environment the enzymatic activity of sAPPα was again compared to the established ferroxidase, CP. A triplex assay we have recently developed from existing assays [22] was utilized to kinetically appraise Fe 2+ oxidation and the loading of Fe 3+ into TF simultaneously at a physiological pH and with minimal auto-oxidation interference.
For reevaluating the enzymatic activity of APP, human sAPPα was expressed and purified from yeast as per Henry et al’ s original purification procedure [26] . This recombinant protein was comparable to our original report [2] and to others (e.g. [35] , [36] ). In addition, a further source of human recombinant full-length APP expressed in bacteria was commercially obtained to corroborate findings using our in-house purified sAPPα.
Within a physiologically relevant buffer (HBS, pH 7.2) the abilities of sAPPα and CP to convert Fe 2+ to Fe 3+ were measured in the absence and presence of TF. In all conditions the sAPPα (250 nM) preparation was able to decrease Fe 2+ and increase Fe 3+ levels compared to control (buffer alone, or with 250 nM BSA) ( Fig. 1A–D ). Fe 2+ oxidation proceeded significantly faster in the presence of CP compared to the presence of sAPPα ( Fig. 1A–D , Table 1 ), and the addition of TF to the assay increased the kinetics of both oxidation readouts ( Fig. 1B & D , Table 1 ). Upon measuring Fe 3+ incorporation by apo-TF, sAPPα and CP were observed to promote comparable generation of holo-TF by the end of the reaction (20 min), similar to that previously reported [2] but kinetically their initial rates of reaction differed (i.e. slope of curve in Fig. 1F , Table 1 ). Similar results were obtained from the commercially-obtained full-length APP (data not shown).
10.1371/journal.pone.0114174.g001 Figure 1
Original preparation of recombinant sAPPα has apparent ferroxidase activity.
Using a triplex assay that simultaneously measures kinetically the loss of Fe 2+ ( A & B ), conversion of Fe 3+ ( C & D ) and physiologically-relevant loading of Fe 3+ into transferrin ( E &F ) [22] , the apparent oxidase activity of sAPPα (500 nM) (▪) was compared to CP (500 nM) (•), BSA (500 nM) ( ) and buffer only ( ) with ( B, D & F ) and without ( A, C & E ) the presence of TF. While activity of the originally prepared recombinant sAPPα was not kinetically comparable to CP in all conditions, activity was greater than BSA used as a control and buffer alone. Assays were run at 26°C in HBS, pH = 7.2+ FAS (100 µM) ± apo-TF (50 µM). Individual data points were mean ± S.E. of 2 experiments, performed in duplicate.
10.1371/journal.pone.0114174.t001 Table 1
Initial rates of reaction for originally purified sAPPα and CP for all 3 outputs in the triplex assay with the absence or presence of TF.
Without TF
With TF
sAPPα
CP
sAPPα
CP
Ferrous loss Fe 2+ (µM)/min
4.90
10.84
7.01
14.24
Ferric gain Fe 3+ (µM)/min
4.84
14.83
14.89
25.09
TF loading Fe 3+ (µM)/min
–
–
20.99
28.11
We went further to investigate whether a cofactor co-purified with sAPPα was required for the apparent ferroxidase activity. We determined that during the last step of chromatographic purification of sAPPα from yeast, the phenyl Sepharose elution buffer of HPO 4 2− (50 mM) yielded a final concentration in the ferroxidation assay of 0.5 mM HPO 4 2− . Prior reports have indicated that polyanions such as phosphate and bicarbonate can promote TF loading [22] [37] – [39] . We had not previously appreciated this 1% contaminant (after dilution with HBS) as significant. We therefore repeated the experiment after removing the residual anions from recombinant sAPPα by size exclusion chromatography with the buffer completely exchanged to HBS, pH 7.2. After this exchange, sAPPα activity was abolished for all outputs of the triplex assay ( Fig. 2 ). Further examination indicated that 0.5 mM HPO 4 2− alone was enough to oxidize Fe 2+ as well as incorporate Fe 3+ into TF at a rate comparable to the original sAPPα preparation ( Fig. 2 ). This mimicked ferroxidase activity. Again, similar results were identified with the commercial full length APP (data not shown), which is delivered lyophilized in PBS.
10.1371/journal.pone.0114174.g002 Figure 2
Apparent APP ferroxidase activity derives from the presence of contaminating polyanions.
Recombinant sAPPα was originally eluted from phenyl Sepharose with a buffer (50 mM Na 2 HPO 4 , pH = 7.4) that lead to a final assay concentration of HPO 4 2− at 0.5 mM. Buffer exchanging the sAPPα preparation with HBS prior to its use in the assay (▴) eliminated the presence of this trace polyanion and subsequently ablated activity in all measurements of the triplex assay ( A–C ). The original iron oxidation rate of the sAPPα (250 nM) preparation (•) was comparable to Na 2 HPO 4 alone ( ), and reintroduction of Na 2 HPO 4 with sAPPα (▪) by further buffer exchange produced similar activity to that observed with the original sAPPα preparation. Thus, contaminating HPO 4 2− mimicked ferroxidase activity. Individual data points were mean ± S.E. of 2 experiments, performed in duplicate.
While the apparent ability of sAPPα and APP to oxidize Fe 2+ originated from the presence of contaminating HPO 4 2− ( Figs. 3A & C ), this anion was not required for CP enzymatic activity and had no effect on enhancing the kinetics of the reaction as measured by Fe 3+ production ( Fig. 3B ).
10.1371/journal.pone.0114174.g003 Figure 3
Contribution of phosphate to iron oxidation from preparations of sAPPα, CP or BSA.
Within the original triplex assay conditions (HBS, pH = 7.2, 26°C), FeSO 4 (100 µM) alone had minimal capability to produce Fe 3+ within 10 min. The presence of Na 2 HPO 4 (0.5 mM) markedly increased the production of Fe 3+ over the same period. As previously determined in Fig. 2 , sAPPα (250 nM) was unable to facilitate Fe 3+ production unless in the presence of Na 2 HPO 4 ( A ), similar to BSA ( C ), whereas CP (250 nM) was unaffected and demonstrated greater formation of Fe 3+ than Na 2 HPO 4 alone ( B ). Ferric ion production was measured as previously described [22] , [28] , [37] . Individual data points were mean ± S.E. of 2 experiments, performed in duplicate.
APP increases the stability of Ferroportin on the cell surface
Having excluded ferroxidase activity as being of relevance to the mechanism by which APP facilitates the efflux of intracellular iron [2] , [17] , we investigated APP interaction with the iron exporter FPN. Our original findings of APP interaction with FPN [2] were recently confirmed using the REXXE-containing peptide fragment of APP, with both this peptide, and exogenous sAPPα, stabilizing surface levels of FPN and promoting iron efflux in human brain microvascular endothelial cells [19] .
Immortalized HEK293T and primary neuronal cultures both utilize APP to promote iron efflux [2] , [17] . We tested the effects of a brief exposure to exogenous sAPPα added to the media (1 µM for 30 min) to surface FPN expression in these cells as well as neuroblastoma cells. The cells were pre-incubated with 50 µM iron to achieve detectable starting levels of FPN. Indeed, brief sAPPα treatment increased FPN detected by cell surface biotinylation and FACS ( Fig. 4 ). A similar increase in FPN expression on the cell surface was also observed following incubation with CP (1 µM for 30 min) ( Fig. 4A&B ), despite this enzyme not being endogenously expressed in either of the cell lines used [2] , [14] . The increase of surface FPN induced by CP did not increase cell surface APP. As total FPN expression was unchanged by either sAPPα or CP treatment, the increased cell surface FPN is consistent with stabilization of surface FPN rather than increased FPN production.
10.1371/journal.pone.0114174.g004 Figure 4
Extracellular sAPPα or CP increases cellular expression of surface ferroportin.
FPN location was examined in ( A ) HEK293T and ( B - D ) primary murine neuronal cultures, preincubated with iron (50 µM, 3 h) followed with CP or sAPPα (1 µM, 30 min). Both cell lines have been previously shown to utilize APP to promote iron efflux, and do not express CP [2] , [14] . Surface proteins on ( A ) HEK293T cells, and ( B ) primary neurons, were biotinylated to identify changes to endogenous FPN and APP expression on the cell surface, as well as exogenously attached sAPPα or CP. Surface levels of FPN were significantly increased in the presence of CP or sAPPα, despite total levels of FPN remaining unchanged. The graphs show the distribution of FPN when normalized against the β-actin content of the intracellular+surface fractions, and adjusted for protein load. Similar results for FPN distribution were achieved even without adjusting for β-actin (not shown). ( C ) Fluorescence-activated cell sorting of non-permeabilized N2a neuroblastoma cultures preincubated with iron (50 µM, 6 h) confirms an increase in surface expression of FPN, quantified in ( D ), after a 30 min incubation with sAPPα (1 µM) in OptiMEM. ( E ) Deconvoluted confocal microscopy shows overlap of endogenous APP and FPN at the surface of non-permeabilized primary neurons preincubated with iron (50 µM, 3 h), as well as ( F ) increased FPN on the neuronal surface following further treatment with CP or sAPPα (1 µM, 30 min). Endogenous surface FPN was below detection limits in neurons that were not treated with FAS (not shown). Data in ( A ), ( B ) & ( D ) are means ± S.E. of 3 experiments, performed in duplicate. * p<0.05, ** p<0.01 and *** p<0.001 analyzed treatment vs control, by two-tailed t tests. ( C ) is a representative histograms of>10,000 live cells normalized to the control (treated with secondary antibody only) mean signal, set at 10 2 . ( E ) & ( F ) are representative images of a neuron from 2 experiments, performed in duplicate. Scale bar = 10 µm.
Histological assessment of the proximity of APP and FPN on the primary neuronal cell surface was determined by deconvoluted confocal microscopy. Upon the addition of iron, endogenous levels of APP and FPN were colocalized on the surface of non-permeabilized primary neurons ( Fig. 4E ). Similar to cell surface biotinylation studies, the addition of either sAPPα or CP greatly increased the expression of FPN on the cell surface ( Fig. 4F ).
The interaction between endogenous APP and FPN was also analysed by FACS of neuroblastoma cells. Treatment with iron (50 µM) induced increased surface levels of FPN, as expected [40] , [41] , and a concomitant increase in cell surface levels of APP ( Fig. 5A–C ). In this system, longer duration of iron (50 µM) treatment (for 6 hours rather than just 3 hours treatment for the cell culture data in Figure 4 ) was optimized to achieve a clear separation of flow cytometry peaks, ± Fe, for FPN, so that RNAi of APP could be tested. When endogenous APP expression was suppressed by siRNA (as confirmed by negligible APP expression after treatment, data not shown), FPN levels on the cell surface were markedly suppressed ( Fig. 5D–E ) despite the increase in intracellular labile iron that APP suppression can induce [2] .
10.1371/journal.pone.0114174.g005 Figure 5
Presence of neuronal surface APP and FPN correlate in response to levels of intracellular iron and the expression of the congruent protein.
Fluorescence-activated cell sorting of non-permeabilized N2a neuroblastoma cultures preincubated with iron (50 µM, 6 h) showed an increase in surface expression of ( A ) endogenous APP and ( B ) FPN, quantified in ( C ). ( D ) Despite the previously reported accumulation in intraneuronal iron induced by APP siRNA [2] , endogenous surface FPN expression was decreased. ( E ) Surface FPN was exaggerated in the presence of iron (50 µM, 6 h), but not with APP siRNA. ( A ), ( B ) & ( D ) are representative histograms of>10,000 live cells normalized to the control (treated with secondary antibody only) mean signal, set at 10 2 . Quantitation using FlowJo software in ( C ) & ( E ) represent the means ± S.E. of 3 experiments, performed in duplicate. *** p<0.001 in ( C ) analyzed control vs. iron, in ( E ) Scrambled (SCR) vs. APP siRNA, by two-tailed t tests.
Culture media alone promotes iron oxidation
We had previously reported that iron efflux from neuroblastoma and primary neurons was promoted by APP and sAPPα in the absence of a ferroxidase enzyme [2] . However, McCarthy et al [19] recently reported that while sAPPα promotes iron export, cells require the presence of an exofacial ferroxidase (Fet3). The conditions of these two sets of experiments are noted to differ, as summarized in Table 2 ; namely, different cells, media, sAPPα concentration and presence of extracellular TF. To reconcile some of these observations we examined the oxidation of Fe 2+ achieved by the various culture media alone, since if there is sufficient oxidation of Fe 2+ , this could account for iron efflux through Fe 3+ release from FPN. All media tested contained select proprietary formulations, whose full contents are not published, therefore, there was no way of predicting what amount of oxidation would occur without measuring it empirically.
10.1371/journal.pone.0114174.t002 Table 2
Comparison of media conditions relevant for iron oxidation in cell types used to study APP facilitated iron efflux.
Cell Type
hBMVEC [19]
HEK293T [2] & SH-SY-5Y [17]
Primary Neurons [2] , [17]
Media
RPMI
OptiMEM
Neurobasal
Inorganic salts
Sodium Phosphate
∼5 mM
∼1 mM
∼1 mM
Sodium Bicarbonate
∼25 mM
∼25 mM
∼25 mM
Sodium Chloride
∼100 mM
∼115 mM
∼50 mM
HEPES
+ (unknown)
+ (unknown)
+ (∼10 mM)
Supplements
Transferrin
−
+ (unknown)
+ (unknown)
Indeed, all three commercial media utilized by the two studies were able to induce markedly greater Fe 2+ oxidation than HBS pH 7.2 ( Fig. 6A&B ). While this could account for our observation that APP and sAPPα promotes iron efflux without the presence of a ferroxidase, McCarthy et al [19] reported that the ferroxidase (Fet3) promoted efflux above the levels achieved by sAPPα alone. McCarthy et al [19] did not supplement with extracellular TF in their media, whereas we did ( Table 2 ), so we examined the impact of TF on the oxidation induced by each media. Indeed, the presence of TF augmented oxidation by each media ( Fig. 6A ), and the loading of Fe 3+ into TF was comparable in all media ( Fig. 6C ), demonstrating that TF dominates the reaction, and can limit the amount of oxidation that is achieved by the polyanionic environment. Therefore, the presence of TF in the cell culture media is predicted to be a dominating influence on the efflux of iron, and may be the reason why McCarthy et al [19] only saw sAPPα promoting iron efflux in the presence of a ferroxidase (but the absence of TF), whereas we [2] observe sAPPα and APP promoting iron efflux in the absence of a ferroxidase (but the presence of TF).
10.1371/journal.pone.0114174.g006 Figure 6
Apparent ferroxidase activity derived from culture medium.
Media of interest was tested by the triplex assay to measure loss of Fe 2+ ( A ), conversion of Fe 3+ ( B ) and loading of Fe 3+ into TF ( C ). Without apo-TF, a one in five dilution of OptiMEM, neurobasal (NB) and RPMI-1640 (RPMI) media oxidized Fe 2+ to Fe 3+ ( A & B ). The addition of apo-TF (50 µM) promoted Fe 2+ oxidation by all media ( A ), and had comparable ability to load Fe 3+ onto TF ( C ). Assays were run at 26°C in HBS, pH = 7.2+ FAS (100 µM) ± apo-TF (50 µM). Individual data points were mean ± S.E. of 2 experiments, performed in duplicate.
Discussion
Using a recently reported adaptation of existing assays for measuring Fe 2+ oxidation in physiologically relevant buffers [22] , we found that our previously-reported oxidase activity of APP actually originated from contaminant phosphate that co-eluted during protein purification. Despite this, our data confirm the previous findings [2] , [17] , [19] that APP decreases intraneuronal iron by stabilizing FPN on the neuronal cell surface. McCarthy et al [19] recently reported that the presence of an extracellular ferroxidase (Fet3) was necessary for sAPPα-assisted iron export in endothelial cells. However, our previous data indicate that the facilitatory role of APP in effluxing neuronal iron did not require ferroxidase activity or an endogenous multicopper ferroxidase, as neither were detectably present in the original neuronal culture studies [2] . However, similar to previous iron efflux studies [12] both TF and polyanions such as phosphates and bicarbonates were present in the media used (OptiMEM and Neurobasal), and therefore may have promoted Fe 3+ loading into TF [22] , [42] as well as the chemical oxidation of Fe 2+ released from the exofacial surface of FPN. It remains to be determined whether an exogenous ferroxidase will augment the neuronal iron efflux that is promoted by sAPPα in the same manner as shown with endothelial cells. Nonetheless, rapid oxidation of Fe 2+ is evident in the presence of apo-TF and phosphate [43] and we have determined that APP and sAPPα can promote Fe 3+ efflux in a suitable polyanionic/TF environment without the presence of an extracellular ferroxidase [2] . This use of a non-transferrin bound iron efflux route with RPMI media in the absence of TF (as in [19] ), may explain the variable outcomes obtained between studies.
Here we have attempted to appraise the involvement of neuronal APP in the 3 known criteria of iron export; namely, stabilization of ferroportin on the cell surface (by APP), iron oxidation (by the media), and loading of TF (by polyanions). Taken together, we hypothesize that under conditions requiring rapid iron efflux, or within an environment where Fe 2+ oxidation by polyanions is limited, a ferroxidase enzyme may be needed to assist iron efflux. Such a circumstance would be where pH and the anion gap are altered, such as within a hypoxic environment and hypophosphatemia, or when apo-TF and/or APP/sAPPα are in limited supply, such as with anemia of chronic inflammation and Alzheimer’s disease respectively [44] – [47] . Despite CP not being endogenously expressed in neurons, its recruitment from other cell types may still promote FPN persistence on the neuronal cell surface ( Fig. 4 ) and release of Fe 3+ from FPN to facilitate iron efflux [19] . Our previous evidence illustrating an increased interaction between FPN and CP in brains of APP knockout mice [2] supports this hypothesis.
We conclude that despite no current evidence of endogenous ferroxidase expression in a neuron, APP in a physiological environment still facilitates the efflux of iron in the presence of TF. We hypothesize that by APP stabilizing FPN on the cell surface, the anionic interstitial environment oxidizes Fe 2+ whereupon Fe 3+ is incorporated into TF. We propose that this process for iron efflux is predominant where a ferroxidase is not present, but can be superseded in conditions of stress where a ferroxidase is used to increase the efflux of iron with minimal ROS production. It is important to consider that, while neurons have high metabolic activity and therefore generate more ROS, the iron elevation in brain only causes neurodegeneration with age [48] , [49] . This phenotype is accelerated when either APP or CP is genetically ablated in mice [2] , [50] and may be a contributory pathogenic mechanism in human diseases associated with each protein, such as Alzheimer’s disease [51] , [52] , Parkinson’s disease [53] or aceruloplasminemia [54] , [55] .
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Macrophages are key cells that recognise, ingest and destroy foreign microorganisms and their products as part of the innate immune system. Macrophages have an additional role as antigen-presenting cells, and so are central to the optimal functioning of both the innate and acquired immune response [1] . Pathogen-associated molecular patterns (PAMPs) such as bacterial endotoxin (LPS) and CpG DNA and other toll-like receptor (TLR) ligands, induce the release of proinflammatory products such as cytokines and chemokines, thereby enhancing pathogen clearance [2] . Ligation of surface receptors commonly activates protein phosphorylation cascades, which are mediated by the selective activation of protein tyrosine kinases (PTKs). Such responses are usually transient and can be negatively regulated by protein tyrosine phosphatases (PTPs) [3] – [5] , and perturbation of the balance between PTK and PTP activity may result in a failure of inflammation to resolve or dysregulation of cell proliferation, which can lead to life-threatening chronic inflammatory diseases or malignancy [6] – [11] . An example of this is the constitutive tyrosine phosphorylation of the PI3 kinase/Akt pathway due to the reduction in the activity of SHP-1 in the allelic moth-eaten viable (Me v /Me v ) mouse, resulting in severe autoimmunity [12] .
PTPRJ (CD148, DEP-1, PTPη, Byp or PTPβ-like tyrosine phosphatase) is a type-III receptor PTP containing a single cytoplasmic phosphatase domain and an extracellular domain containing eight to twelve FNIII repeats, depending on species [13] . PTPRJ is found in a wide range of cell types [14] , [15] and evidence for a tumour suppressor role has been indicated due to its reduced expression in some malignant tumours, its regulation by cell density, and the reversion of the transformed phenotype when PTPRJ function is restored [16] – [21] . PTPRJ has several substrates, including PDGF β-receptor [22] , hepatocyte growth factor (HGF) receptor [23] , vascular endothelial growth factor (VEGF) receptor-2 [24] and the p85 subunit of PI3-kinase [25] .
Within human and mouse tissues, macrophages exhibit the highest expression of Ptprj [26] – [30] , and in this cell type Ptprj expression is up-regulated by LPS and down-regulated by CSF-1 [30] . Knockout of Ptprj indicates a positive role in monocyte activation as it dephosphorylates the negative regulatory tyrosine in src family kinases in a manner similar to CD45 [31] , [32] . Little is known about the molecular mechanism underlying the regulation of Ptprj expression. Although there is little data on regulation of transcription, the 5′ end of the mRNA is thought to attenuate translation [33] .
Recently, we characterized differential expression of Ptprj in normal and cancerous human breast tissue and in the developing mouse mammary gland [21] . Microarray analysis of the Ptprj gene locus identified seven probes that targeted long noncoding RNA species originating from the first intron [21] , [34] . Although previously regarded as “junk”, it is now becoming increasingly understood that ncRNAs play a vital role in regulating and co-ordinating the developmental complexity of eukaryotic organisms [35] . The differential and developmental specificity of long ncRNA (lncRNA) expression, in combination with the widespread conservation of their promoters, splice sites and primary sequence [36] , suggests that they are generally functional [37] – [39] . The molecular mechanisms of lncRNA functions are diverse and cannot be easily generalised, and unlike protein-coding genes, their function cannot currently be predicted from their primary sequence [40] . Previous functional studies of lncRNAs reveal that they can act by influencing target gene expression at specific genomic loci, either by directly interacting with chromatin regulatory proteins or by modulating the activity of their interacting partners [41] – [43] . While lncRNAs play important roles during normal cellular development and differentiation [44] , lncRNAs are also associated with several diseases, including heart disease, Alzheimer’s disease, psoriasis, and cancer [45] . In addition to our observation of multiple lncRNAs originating from the Ptprj locus, the potential for RNA regulation of Ptprj was also recently highlighted by a report showing that PTPRJ is negatively regulated by the short RNA microRNA328 [46] .
Here we investigate the expression of Ptprj and an antisense lncRNA originating from the Ptprj locus in macrophages. We show that Ptprj is highly expressed in macrophage-enriched tissues, that it is upregulated in response to various toll-like receptor ligands and downregulated by CSF-1. We further show that the expression of Ptprj-as-1 , an lncRNA that is transcribed antisense to the coding sequence, is regulated by proinflammatory factors in a manner similar to the PTPRJ gene. Finally we characterise the promoter region of human and mouse Ptprj and identify putative transcription factor binding sites. An understanding of the biology of phosphatases such as Ptprj in macrophage-specific signalling cascades may enable the identification of key endogenous regulators of inflammation and therapeutic targets for inflammatory diseases.
Materials and Methods
Ethics Statement
All animal handing and treatment protocols were approved by the University of Queensland Animal Ethics Committee, certificate 570-08. Mice were monitored daily for any adverse reactions and euthanized by exposure to CO 2 .
Animal Handling
Bone marrow-derived macrophages (BMMs) were derived from cells obtained from the femurs of carbon dioxide euthanized C57BL/6 mice in accordance with University of Queensland ethics guidelines. Thioglycollate-elicited peritoneal macrophages (TEPMs) were obtained by injecting 6–8 week-old male mice with 1 mL of 10% w/v thioglycollate broth and recovering the peritoneal exudate by peritoneal cavity lavage with 10 mL of phosphate-buffered saline (PBS) 5 days post injection as described previously [47] .
Cell Culture
RPMI 1640 medium (Invitrogen Life Technologies) supplemented with 10% FCS (JRH Biosciences), 20 U/mL penicillin (Invitrogen Life Technologies), 20 µg/mL streptomycin (Invitrogen Life Technologies), and 2 mM L-glutamine (Invitrogen Life Technologies) (complete medium) was used for culture of RAW 264.7 cells and bone marrow-derived macrophages (BMMs). Briefly, bone marrow cells were cultured for 7 days in complete medium in the presence of 10,000 U/mL of recombinant human CSF-1 (Chiron Corporation, Emeryville, CA, USA) on bacteriological plastic plates. RAW 264.7 [48] cells were maintained on bacteriological plates (Sterilin, Staffordshire, UK) in complete media containing 5% heat-inactivated fetal bovine serum for a maximum of four weeks in culture. RAW 264.7 cells were passaged with an 18-gauge hypodermic needle and syringe. NIH3T3 and HEK 293 cell lines were maintained in D-MEM (Dulbecco/Vogt Modified Eagle’s Minimal Essential Medium, GIBCO) supplemented with 2 mM L-glutamine (Glutamax-1, Invitrogen), 10% heat-inactivated fetal bovine serum (JRH Biosciences, Lenexa, KS, USA), 20 U/mL penicillin and 20 µg/mL streptomycin (Invitrogen). The cells were cultured in T75 filter cap flasks (Nunc). All primary cells were maintained in a 37°C incubator containing 5% CO 2 . Lipopolysaccharide (from Salmonella Minnesota, Sigma Aldrich, St. Louis, MO, USA) was used at a final concentration of 10 ng/mL. Recombinant human colony stimulating factor-1 (a gift from Chiron, Emeryville, CA, USA) was used at a final concentration of 10 4 U/mL. Phorbol 12-myristate 13-acetate (PMA) (Sigma Chemical Co.) was used at a final concentration of 10 −7 M. Where shown, cells were treated with 10 µg/mL of anti-CD148 antibody or hamster isotype control antibody (Serotec).
Quantitative Real Time PCR
RNA extraction from mammalian cells or tissues was performed using the RNeasy mini kit (Qiagen, Valencia, CA, USA) as per the manufacturer’s instructions. RNA was quantified by spectrophotometry at 260 and 280 wavelengths. RNA purity was ensured by an A 260 /A 280 of at least 1.8. Genomic DNA contamination was removed from RNA preparations using DNase (Ambion, Austin, TX, USA) and cDNA was synthesised from 2–5 µg of total RNA using Superscript III (Invitrogen, Carlsbad, CA, USA), using oligo-dT primers (Geneworks, Adelaide, Australia). cDNA was quantitated using SYBR Green (Applied Biosystems, Foster City, CA, USA) in 20 µL reactions in a 96 well plate. Each cDNA was quantified in experimental triplicate. No amplification controls (a minus-reverse transcriptase control and a minus sample control) were included in each reaction plate to ensure the absence of contaminating genomic DNA. Data was collected and analysed using the ABI Prism software. Gene expression was determined relative to Hprt (hypoxanthine-guanine phosphoribosyl transferase) mRNA using the comparative threshold method as previously described [21] . Calculations were performed in Microsoft Excel following equations provided by Applied Biosystems. Unless otherwise stated, error bars in figures indicate the standard deviation of duplicate cDNA quantitation in the same thermal cycle run. Expression profiles were typically quantitated in at least two separate experiments (as stated in the figure legends) using completely independent preparations of cells and RNA extracts. Primers (f, forward; r, reverse) used were as follows: mouse csf1r f: CCAGAGCCCCCACAGATAA , r: AGCTTGCTGTCTCCACGTTTG ; human csf1r f: CCTTCAGGAGCAGGCCCAAG , r: CCTTGCTCGCAGCAGGTCAG ; mouse Ptprj f: CAGTACAGTGAATGGGAGCACTGAC , r: GTCCGTATTCTGCCACTCCAACT ; human Ptprj f: AGTACACACGGCCCAGCAAT , r: GAGGCGTCATCAAAGTTCTGC ; mouse Ptprj-as1 f: CCATCTCCCATT GTCCAAAC , r: TGATTGAAGGACAGCTGGAA mouse Hprt f: GCAGTACAGCCCCAAAATGG , r: AACAAAGTCTGGCCTGTATCCAA ; human Hprt f: TCAGGCAGTATAATCCAAAGATGGT , r: AGTCTGGCTTATATCCAACACTTCG .
Antibodies
Monoclonal hamster anti-mouse CD148 antibody was generated as described [14] and affinity purified with Protein G sepharose. This antibody was a gift from Prof. Arthur Weiss (Howard Hughes Medical Institute, Rosalind Russell Medical Research Centre for Arthritis, University of California-San Francisco, San Francisco). Rat anti-mouse Mac-2 monoclonal antibody (un-purified culture supernatant from clone TIB-166 hybridoma, ATCC) was kindly provided by Dr Andrew Cook, University of Melbourne. Antibodies to the following proteins were purchased from commercial sources: F4/80 monoclonal antibody from Serotec, goat anti-hamster IgG biotin conjugated (as a secondary against hamster anti-mouse CD148 antibody) from Pierce and goat anti-rat IgG biotin conjugated (as a secondary against rat anti-mouse F4/80 and Mac-2 antibodies) from Serotec.
Immunoblotting
Total protein extracts were prepared using the SDS-boiling method [26] . Cells were washed twice with ice cold PBS and lysed on the culture plate with 500 µL boiling lysis buffer (66 mM Tris-HCl pH 7.4, 2% SDS) per 1×10 6 cells. Extracts were homogenised by repeated passage through a 26-gauge needle and further boiled for 5 minutes. Protein concentration of total lysates was determined using the BCA assay kit (Pierce, Rockford, IL, USA). Extracts were resolved by SDS-PAGE (4–12%), transferred to Imobilon-P (Millipore, North Ryde, NSW, Australia), blocked and probed with anti-CD148 antibody or antiphosphoprotein antibodies in the presence of phosphatase inhibitors. Blots were washed, probed with HRP-labelled secondary antibodies (Cell Signaling Technology) or Streptavidin-HRP-labelled (Pierce) tertiary antibody in case of CD148 blots, and detected using enhanced chemiluminescence (ECL) reagents (Amersham Pharmacia Biotech, Piscataway, NJ, USA). Membranes were stripped with 66 mM Tris-Cl (pH 6.7)/2% SDS/100 mM 2-mercapto-ethanol and reprobed with total Akt antibodies (Cell Signaling Technology).
Immunohistochemistry
Immunohistochemistry was performed using an immunoperoxidase technique with diaminobenzidine (DAB) as the chromogen as described previously [21] , [49] . Expression of CD148, F4/80 and Mac-2 were examined in serial sections. Briefly, sections were deparaffinized and rehydrated followed by antigen retrieval. For CD148, microwave antigen retrieval was performed in citrate buffer pH 6.0 for 2 min and allowed to cool overnight. For F4/80 Carezyme Trypsin digestion kit (Biocare Medica) was used as per the manufacturer’s instructions. Sections were washed in Tris buffered saline (TBS) and endogenous peroxidase activity was blocked by incubating the sections in 3% H 2 O 2 (diluted in TBS) for 30 min. Sections were incubated for 60 min in serum block [10% fetal calf serum (FCS) plus 10% normal serum (species of secondary antibody) in TBS] and then treated with the primary antibody for 60 min. Sections were subsequently incubated for 30 min with a biotinylated F(ab′)2 fragment of goat anti-rat or rabbit anti-hamster immunoglobulin (DakoCytomation), followed by horseradish peroxidase (HRP)-conjugated streptavidin (DakoCytomation) and developed with DAB chromogen (DakoCytomation) according to the manufacturer’s specifications. The specificity of the staining was confirmed by using matched isotype control antibodies. All sections were counterstained with Mayer’s haematoxylin. Slides were allowed to dry on the bench before mounting using permanent mounting media (Cytoseal, Stephens Scientific). All incubations were carried out at 25°C sections were washed between each step with TBS. Slides were examined and photographed using a transmitted light microscope (Olympus BX-51, with DP-70 camera).
Statistical Analysis
One-way analysis of variance (ANOVA) was used to determine any significance differences between the samples. In addition significance differences between individual time points for each group were calculated using a two-tailed independent Student’s t-test. Data sets yielding a p value greater than 0.05 were regarded as not statistically different.
Results
Ptprj is Preferentially Expressed in Macrophage-enriched Tissues and Cell Types
To characterize the expression of Ptprj in various mouse tissues, quantitative real-time PCR analysis was performed. Ptprj expression was highest in bone marrow derived macrophages ( Figure 1A ). Tissue distribution revealed that Ptprj was high in tissues with a significant macrophage content. In particular, expression of Ptprj was highest in the spleen. However, other macrophage-rich tissues such as lung, liver, kidney, intestine, thymus, ovary, muscle and testis showed detectable expression levels of Ptprj . Minimal expression of Ptprj was detected in brain, heart and placenta. The expression of Ptprj in a range of different cell types was also examined. Ptprj mRNA was expressed at relatively elevated levels in inflammatory (TEPMs) and primary macrophages (BMMs) and at lower levels in the macrophage-like cell line RAW 264.7, differentiated RAW/C4 osteoclast-like cells and myeloid precursor cell line M1 ( Figure 1B ). Ptprj mRNA levels were low in pre-B lymphoid cell line WEHI-231, embryonic fibroblasts and Ptprj mRNA was virtually undetectable in the fibroblast cell line NIH 3T3.
10.1371/journal.pone.0068306.g001 Figure 1
Ptprj mRNA expression in murine tissues.
(A). qPCR of Ptprj from RNA extracted from mouse tissues. (B) qPCR of Ptprj from RNA from pre-B lymphoid cell line WEHI-231, osteoclast-like cell line (RAW 264.7, C4), TEPM, macrophage-like cell line (RAW 264.7), BMM, myeloid cell line M1 and fibroblasts (NIH3T3 and mouse embryonic). (C). Immunohistochemistry of cell-specific expression of PTPRJ in mouse spleen sections. Sections were immunostained for CD148 (A1, B1) or F4/80 (A2, B2) and with CD148 (C1) and F4/80 (C2) isotype control antibodies. All sections were counterstained with haematoxylin. RP, red pulp; WP, white pulp. Original magnification: x100 (A), x200 (B, C). Bar, 100µm.
As Ptprj expression was highest in spleen compared to other tissues ( Figure 1A ), immunohistochemistry was performed to identify the cell types within spleen that express the protein product of Ptprj , CD148. CD148 expression pattern in the spleen was restricted to the red pulp region, and was indistinguishable from the expression pattern of F4/80, a mononuclear phagocyte marker ( Figure 1C ). This confirmed the macrophage lineage–specific expression of CD148 in the mouse spleen and was consistent with the quantitative real-time PCR data ( Figure 1A,B ).
Regulation of Ptprj mRNA and Protein in Response to Inflammatory Stimuli in Mouse Macrophages
Quantitative real-time PCR analysis of BMMs stimulated with LPS showed regulated expression of Ptprj . In the absence of CSF-1, LPS induced Ptprj mRNA expression approximately six-fold peaking at 4 hours. However, the presence of CSF-1 caused a marked reduction in both the basal and LPS-induced expression levels of the Ptprj transcript ( Figure 2A ). As expected, the expression pattern of c-fms , a classical marker for mature, differentiated macrophages was suppressed by CSF-1 ( Figure 2B ). To investigate other murine macrophage populations, the expression profile of Ptprj was examined in the macrophage-like RAW 264.7 cell line and in thioglycollate-elicited peritoneal macrophages (TEPM). Although, the pattern of Ptprj induction in response to LPS in RAW 264.7 cells was identical to that of BMMs ( Figure 2C ), the fold induction was less compared to BMMs. Because RAW 264.7 cells are deficient in cell surface expression of CSF-1R [50] , Ptprj mRNA expression in these cells was examined only in the absence of CSF-1. As with BMMs, in TEPM Ptprj was strongly induced by LPS and this was repressed by CSF-1 ( Figure 2D ). Pre-treatment of macrophages with IFNγ (priming) results in a more rapid and heightened response to LPS and other TLR agonists [48] , [51] . Unlike LPS and CpG DNA, IFNγ alone had no effect on induction of Ptprj in BMMs (data not shown). Conversely, priming of BMMs with IFNγ repressed LPS-mediated induction of Ptprj mRNA expression at early time points ( Figure 3E ).
10.1371/journal.pone.0068306.g002 Figure 2
Regulation of Ptprj expression in mouse macrophages by proinflammatory stimuli.
A–C: Regulation of Ptprj expression by CSF-1 and LPS in mouse bone marrow-derived macrophages (BMMs). BMMs were maintained overnight in the presence or absence of CSF-1 (1×10 4 U/mL) before treatment with LPS (10 ng/mL) (A, B). RAW 264.7 cells were maintained overnight in the absence of CSF-1 before treatment with LPS (10 ng/mL) (C). Ptprj (A, C) and c-fms (B) expression profiles were assessed by quantitative real-time PCR. Profiles are representative of two independent experiments. D: Regulation of Ptprj expression by CSF-1 and LPS in mouse thioglycollate-elicited peritoneal macrophages (TEPMs). TEPMs were maintained overnight in the presence or absence of CSF-1 (1×10 4 U/mL) before treatment with LPS (10 ng/mL). Ptprj expression profile was assessed by quantitative real-time PCR. E: IFNγ treatment of bone marrow derived macrophages suppresses the LPS mediated induction of ptprj . BMMs were maintained overnight in the presence of CSF-1 (1×10 4 U/mL) and presence or absence of IFNγ (500 pg/mL) before treatment with LPS (10 ng/mL). RNA was extracted at each time point and used for the synthesis of cDNA. Ptprj expression profile was assessed by quantitative real-time PCR. Datapoints (+/− SD) represent the average of triplicate samples each from triplicate independent experiments. Significance values were determined by one-way analysis of variance (ANOVA). *denotes p<0.05; **denotes p<0.005; n = 3. F: Regulation of PTPRJ protein in response to LPS, CpG DNA and CSF-1. BMMs were maintained overnight in the presence or absence of CSF-1 (1×10 4 U/mL) before treatment with LPS (10 ng/mL) [top panel] or CpG DNA (0.1 µM) [bottom panel]. Protein lysates were separated by SDS-PAGE, transferred to PVDF membranes and immunoblotted for PTPRJ. The membrane was then stripped, and reprobed for total Akt as a loading control. Profiles are representative of two independent experiments.
10.1371/journal.pone.0068306.g003 Figure 3
Ptprj expression in response to LPS in human mononuclear phagocytic cells.
THP-1 cells were maintained for 24 hours in the presence or absence of PMA (10 −7 M) to induce differentiation, before treatment with LPS (10 ng/mL) (A, B). Human dendritic cells were treated with LPS (10 ng/mL) over a time course (C, D). Ptprj (A, C) and c-fms (B, D) expression profiles were assessed by quantitative real-time PCR. Datapoints (+/− SD) represent the average of triplicate samples. Significance values were determined by one-way analysis of variance (ANOVA). *denotes p<0.05; **denotes p<0.005; n = 3. *denotes p<0.05; **denotes p<0.005; n = 3.
To determine whether Ptprj was regulated at the protein level, western blot analyses of BMM cell lysates from LPS and CpG time courses were performed ( Figure 2F ). CSF-1 increased the basal level of CD148 protein (0 h lanes), however CD148 protein levels increased with time after treatment with LPS in both replete and CSF-1-starved cells. Although CpG DNA showed a similar basal pattern of induction of CD148 by CSF-1, an increased level of CD148 protein with time after stimulation was observed. Thus both qPCR ( Figure 2D ) and western blotting ( Figure 2F ) showed an increase in CD148 mRNA and protein respectively, albeit with a different timecourse.
Regulation of PTPRJ Gene Expression in Human Mononuclear Phagocytic Cells
The expression profile of PTPRJ in response to LPS was also examined in the human monocytic cell line THP-1 and in human dendritic cells. Unlike mature differentiated macrophages such as RAW 264.7 cells or BMMs, THP-1 cells are premonocytes that are committed to the monocytic lineage, but are non-adherent and lack many macrophage-specific cell surface markers. Therefore, for the induction of terminal differentiation to macrophage-like cells, THP-1 cells were cultured in the presence of PMA. PMA mimics some actions of CSF-1, but activates distinct signalling pathways [52] . In the absence of PMA, LPS stimulation had no effect on PTPRJ expression ( Figure 3A ). However, presence of PMA led to a dramatic induction of PTPRJ in response to LPS. Even in the absence of LPS, PMA induced basal PTPRJ expression approximately ten-fold, whereas in the presence of LPS, the induction was even stronger. PMA strongly induced c-fms , consistent with the view that PMA imparts a macrophage-like phenotype to these monocytic cells, however LPS treatment suppressed c-fms expression in PMA-differentiated cells ( Figure 3B ).
CD148 has been recognised as an accessory molecule present on the surface of peripheral blood dendritic cells [53] . Although overall fold induction was lower than in mouse macrophages, LPS induced PTPRJ expression in human peripheral blood dendritic cells (PBDCs) ( Figure 3C ). As with PMA-differentiated THP_ cells, c-fms was down-regulated by LPS in human dendritic cells ( Figure 3D ).
Expression of Long Noncoding RNA Ptprj-as1 during Macrophage Activation
Expression of Ptprj-as1 in murine tissues
In a previous study, we examined microarray expression profiling data of mammary epithelial cells derived from pregnant, lactating and involuting mice [34] . The microarray that this dataset is based upon features probes targeting ∼29,550 mRNAs and 8,693 long ncRNAs. Within this dataset, we identified seven probes that targeted long ncRNAs that originated from the Ptprj locus [21] . All of these long ncRNAs occurred within the first intron of Ptprj ; two were on the antisense strand and the remaining five were on the sense strand ( Figure 4A ). On the basis of its high expression in monocytic cells, cytoplasmic localization, and spliced character, we selected the lncRNA that we term Ptprj-as1 (GenBank Accession ID AK016880) for further examination. As the other lncRNAs were expressed at low levels or unexpressed in monocytic cells, and they were derived from unspliced transcripts, they were not pursued for further characterization as such low-expressed single-exon transcripts are challenging to study because they cannot be easily distinguished from genomic DNA in expression studies.
10.1371/journal.pone.0068306.g004 Figure 4
Characterisation of Ptprj-as1.
A–B: Ptprj-as1 maps to the reverse strand within the boundaries of the murine Ptprj gene. Comparison of the mouse Ptprj (A) and human PTPRJ (B) loci. Protein coding transcript isoforms of Ptprj/PTPRJ are shown in red and long noncoding transcripts are shown in blue. Arrows indicate the direction of transcription. The human microRNA miR-3161 is shown in green. Position of PCR primers used for qRT-PCR for mouse Ptprj-as1 are indicated. C-D: Mapping (C) and expression (D) of a splice variant of murine Ptprj-as1 in brain, kidney and testis. E: Predicted secondary structure of Ptprj-as1 splice variant.
Ptprj-as1 is a spliced 1,006 nt lncRNA that is transcribed antisense to Ptprj-as1 and is expressed at levels similar to Ptprj . Interestingly, Ptprj-as1 is antisense to the 5′UTR of a short isoform of Ptprj that lacks the canonical first exon, raising the possibility that Ptprj-as1 may be co-expressed with the short isoform of Ptprj-as1 . Ptprj-as1 was identified as a long noncoding RNA of unknown function that is transcribed from the reverse strand of the Ptprj gene ( Figure 4A ). Note that the PTPRJ-AS1 in human is in a different location to Ptprj-as1 in mouse and should not be considered as an orthologue ( Figure 4B ). RT-PCR analysis of the murine transcript in different tissues confirmed the existence of tissue-specific splice variants ( Figure 4C ). Spleen, brain, and testis revealed two isoforms of Ptprjas-1 , one; as reported previously [21] and a second that contains an additional 80 base exon ( Figure 4D ). The examination of the secondary structure of the additional exon showed a stable stem loop structure ( Figure 4E ).
Expression of Ptprj-as1 in murine macrophages
Ptprj-as1 was readily detectable by qRT-PCR ( Figure 5A ). Preliminary screens of murine tissues demonstrated that whilst Ptprj-as1 was expressed in many tissues, the transcript showed comparatively elevated levels in lung, brain, kidney and spleen; all tissues that have a sizeable macrophage component ( Figure 5A ). In unstimulated cells, LPS activation of murine bone marrow-derived macrophages (BMMs) induced a transient increase of Ptprj mRNA, peaking at a 4 fold increase 8 hours post addition of LPS. In parallel with Ptprj , Ptprj-as1 expression was also transiently induced by LPS activation of BMMs ( Figure 5B ). Furthermore, Pam3Cys induced a very similar expression pattern of Ptprj and Ptprj-as1 ( Figure 5C ). The comparable expression trends of Ptprj and Ptprj-as1 during mBMM activation is consistent with the developing hypothesis that the Ptprj and Ptprj-as1 are regulated by the activity of the same promoter.
10.1371/journal.pone.0068306.g005 Figure 5
Expression of Ptprj-as1 in murine tissues and in response to TRL ligands.
A: Expression of Ptprj-as1 in murine tissues. B, C: Ptprj and Ptprj-as1 mRNA expression in BMMs in response to LPS (B) or Pam3Cys (C). mRNA expression was quantified by qRT-PCR and expressed as fold change compared with untreated (0h). Plots represent mean fold change +/− SD; n = 3. Significance values were determined by one-way analysis of variance (ANOVA). *denotes p<0.05; **denotes p<0.005; n = 3. *denotes p<0.05; **denotes p<0.005; n = 3.
Bioinformatic Identification of Regulatory Sequences in the Ptprj Gene
To further investigate the potential mechanisms that direct the expression of Ptprj , bioinformatic analysis to identify putative promoter regions was performed.
Identification of Ptprj putative promoter by CAGE analysis
CAGE (cap analysis of gene expression) analysis identified the dominant start sites for transcription of the mouse and human Ptprj genes, which in turn allowed functional alignment of the promoters. Ptprj appears to have only one promoter, and both the mouse and human Ptprj promoters lack a TATA box and are rich in GC (data not shown). Whilst the majority of the macrophage-specific promoters also lack a TATA box [54] – [56] the Ptprj promoter differs from typical macrophage promoters such as c-fms [57] , [58] in that these generally lack GC-rich elements. Analysis of the Ptprj-as1 promoter did not reveal the presence of either a TATA-box or GC-rich element, suggesting that it may be regulated independently to Ptprj.
Identification of Putative Transcription Factor Binding Sites
The mouse and human sequences corresponding to ECR1 were aligned using ClustalW alignment. This sequence was further examined for consensus transcription factor binding sites conserved in sequence and position between mice and humans using the RVista browser and TRANSFAC ( Figure 6 ). This analysis revealed the presence of multiple transcription factor binding sites that are conserved in both sequence and position between the mouse and human putative promoters. These include many transcription factor binding sites that are commonly functional in myeloid promoters including PU.1, SP1, myeloid zinc finger protein (MZF1), nuclear factor kappa B (NF-κB) or p50, octamer factor binding sequences (OCT) and acute myelogenous leukemia (AML1). The presence of a CpG island upstream of the transcription start site might permit epigenetic control of transcriptional activity of the Ptprj promoter [58] , [59] . One important observation of the assignment of TSS by CAGE is that the Ptprj mRNA is longer than previously appreciated. As a consequence, there may be up to four potential initiation codons or AUGs (uAUGs) in the 5′ untranslated region (UTR) upstream of the putative start site of translation of CD148.
10.1371/journal.pone.0068306.g006 Figure 6
Clustal W alignment of the mouse and human Ptprj putative promoters.
Consensus transcription factor binding sites conserved in sequence and position between mouse and human were predicted using RVista browser and TRANSFAC. Only the myeloid-specific transcription factors have been shown in this figure. Evolutionary conserved region (ECR1) lies within the blue brackets. Blue line marks the region of transcription start site (TSS) cluster represented in Figure 6 . mTSS refers to the transcription start site for mouse ptprj and hTSS to the transcription start site for human Ptprj predicted by CAGE analysis. The four uAUG codons are highlighted within grey boxes and the stop codons within red boxes. Asterisk (*) represents the translation start site. The boxes represent the conserved sequences between mouse and human. N represents any nucleotide.
Discussion
PTPRJ has been widely accepted as an epithelial cell expressed gene and has been proposed to be a tumor suppressor in this cell type. However, we show here that Ptprj is highly expressed in macrophages and its expression is regulated in response to CSF-1 and proinflammatory stimuli. Ptprj expression was highest in spleen, lung, ovary, liver, kidney, intestine, thymus and testis; tissues that are rich in resident macrophage population. Indeed, approximately 18% of the cells in the spleen (the highest expressing tissue for Ptprj ) are macrophages [60] . Others have shown CD148 is expressed in macrophages and related tissues of human origin [61] . The macrophage-enriched expression of CD148 was further confirmed by the strong correlation of the expression pattern of CD148 and F4/80 surface markers on the splenic macrophages in the red pulp. qRT_PCR showed a transient increase in CD148 mRNA after stimulation with proinflammatory factors such as LPS, and that both basal and stimulated expression is suppressed by CSF-1. Western blotting showed a general sustained increase in CD148 protein after LPS stimulation but no detectable suppression by CSF-1. Although the reason for this discrepancy is unknown, lack of correlation between RNA and protein may be related to the half-life of the protein within the cell. For example, if CD148 has a long half-life (as many membrane/cytoskeletal proteins do), then a short-term down-regulation in mRNA levels would not immediately translate into a down-regulation of protein. On the contrary, protein with a long half-life would continue to appear up-regulated due to the accumulation of the protein in the cell before degradation.
Whilst there have been a number of studies investigating the role of Ptprj in both epithelial and haematopoietic cells, its exact role in cell function has not been defined. Overall, Ptprj appears to be a negative regulator of cell proliferation in epithelial cells and fibroblasts, by down-regulating receptor tyrosine kinase activity [16] , [17] , [20] . In contrast, Ptprj plays a positive role in the activation of B cells and monocytes by dephosphorylating the negative regulatory carboxyterminal tyrosine of src family kinases (SFK), thereby activating these enzymes and potentiating a signalling cascade. Ptprj expression has been reported to be up-regulated in epithelial cells as they reach confluence and is down-regulated in some tumours [16] , [17] . Although loss of heterogeity has been reported for colon, lung, breast and thyroid cancers [62] , we have not detected any major down-regulation of PTPRJ protein or mRNA in a large cohort of breast cancers [21] .
Clearly, the regulation of Ptprj is of profound importance in a number of cell types, however the regulation of PTPs in general and Ptprj in particular is poorly understood [11] . Ptprj translation has been shown to be regulated by the use of an alternative start site at the 5′ end of the gene [33] . There is growing evidence that microRNAs (miRNAs) can target the expression of both oncogenes and tumour suppressor genes and recently miRNA-328 has been reported to decrease Ptprj expression in epithelial cells [46] . In contrast to the relatively few miRNAs that have been identified, it appears that there are at least as many long noncoding RNAs (lncRNAs) as coding RNAs in the human genome [45] , [63] , and it is likely that many of these can regulate gene expression [64] . We have identified a lncRNA, Ptprj-as1 , that is transcribed off the Ptprj gene locus and is co-regulated with Ptprj coding transcripts in murine monocytes. Although we have not investigated any direct roles of Ptprj-as1 in regulating Ptprj , its tissue-specific expression and co-incident location with Ptrpj in the genome, raise the possibility that it may have some role in the expression or splicing of Ptprj . Previous studies have revealed that antisense long ncRNAs can directly affect the expression or alternate splicing of nearby genes. For example, expression of a long ncRNAs antisense to the tumor-suppressor gene p15 results in silencing of the p15 gene by inducing heterochromatin formation [65] . In another example, differential expression of the lncRNA LUST , which originates antisense to an intronic region of RBM5 , regulates the expression of RBM5 splice variants [66] . It also noteworthy that Ptpre , a related tyrosine phosphatase-encoding gene, also features a contextually analogous antisense noncoding RNA, Ptpreas [67] . Finally, although examination of the corresponding Ptprj loci in the human genome revealed a number of lncRNAs, because the human PTPRJ loci differs considerably in terms of both splice variants, exon number, primary sequence, and promoter initiation, it is not possible to identify orthologous lncRNAs. This in itself is not surprising, as lncRNAs seldom reveal identifiable primary sequence conservation, and as an intrinsic feature of their regulatory function as RNAs are thought generally to show great plasticity in their sequence constraints compared to protein-coding genes [37] .
CAGE analyses revealed the presence of clustered transcription start sites in both mouse and human Ptprj promoters, including conserved multiple upstream AUG (uAUG) codons in the 5′ UTR (data not shown). These uAUGs are common features of mRNAs that encode regulatory proteins. These upstream open reading frames (uORFs) encode for upstream peptides (uPEPs), which are highly evolutionarily conserved and are important for regulating translation of the main CDS [68] , [69] . It is conceivable that one of these AUGs is the preferred start codon, instead of the one (marked with asterisk in Figure 6 ) recognised by the publicly available sequence databases. As shown in Figure 6 , there is a conserved stop codon in the 5′ UTR and another in the mouse putative promoter. In addition, translational analysis initiating from the fourth uAUG encodes for a 52 amino acid long peptide in case of mouse Ptprj .
As there was a similar pattern of Ptprj gene regulation across a range of mouse and human macrophages, a combination of bioinformatic data mining and functional analysis was used to dissect the potential mechanisms underlying the transcriptional regulation of Ptprj . NF-κB binding sites were identified in the Ptprj promoter ( Figure 6 ). NF-κB has a key role in inflammation as it regulates apoptosis, cell-cycle progression, proliferation and cell differentiation [70] , [71] . The responsiveness of both mouse and human Ptprj genes to LPS and CpG DNA could therefore be attributed to the presence of an NF-κB binding site in the Ptprj putative promoter.
The Ptprj promoter is a GC-rich TATA-less promoter, which is the major promoter class in mammals and is a more broad and evolvable category of promoters compared to the TATA box-containing promoters [59] . A number of multiple transcription factor binding sites that are conserved in both sequence and position between the mouse and human putative promoters were identified. The presence of PU.1, SP1, MZF1, NF-κB, AML1 and OCT binding sites in the Ptprj promoter could explain the macrophage-specific expression of Ptprj . Many myeloid promoters have a functional PU.1 binding site upstream of the transcription start site. PU.1 binds to and recruits TATA binding protein (TBP), the primary component of the basal transcription factor TFIID in promoters lacking a TATA box [55] , [72] . The expression of PU.1 has been shown to be important for macrophage differentiation and the expression of a number of molecules that mediate some of the immunological actions of macrophages [73] . Amongst other transcription factors that are involved in myeloid gene regulation and can interact with TBP are SP1 and OCT [55] . MZF1 expression is restricted to myeloid cell lines and is implicated in the development of cells of the myeloid lineage. The presence of a conserved MZF1 binding site in the Ptprj promoter is consistent with its expression in myeloid cells. The RNA binding zinc finger proteins EWS and FUS/TLS bind to the consensus binding sites for MZF1, implicating their significance in the assembly of the basal transcriptional complex [74] . Taken together, the bioinformatic analyses of the Ptprj putative promoter revealed that the Ptprj gene is highly regulated in macrophages and the Ptprj promoter resembles a macrophage-specific promoter.
IFNγ augments the immune response to LPS and other TLR agonists, thereby orchestrating distinct cellular signalling pathways through transcriptional control over a large number of genes [50] . In contrast, IFNγ priming of BMMs repressed Ptprj induction in response to LPS at early time points and had no effect at later time points ( Figure 2E ). Thus, Ptprj differs from many other macrophage-specific genes, which are regulated by both LPS and IFNγ. Many LPS-inducible genes including Mpeg-1 , Itm2b and Ramp2 , are repressed by CSF-1, but LPS-inducible in the presence of CSF-1 [75] . However, LPS-induction of Ptprj expression was more pronounced in the absence of CSF-1. In this respect, Ptprj regulation is similar to that of Tlr9 , as absence of CSF-1 causes an elevation in Tlr9 mRNA expression, but differs in response to IFNγ. In both mouse macrophages and human monocytes or dendritic cells, LPS induced CD148 expression whereas it repressed c-fms expression. Thus the increases in Ptprj gene expression are unlikely to be dependent on CSF-1 signalling, as LPS treatment not only suppresses expression of the CSF-1 receptor (C-fms) but also induces cleavage of the surface receptor making cells refractive to CSF-1 [75] . Dendritic cells are a heterogeneous population of cells with a number of different lineage relationships. C-fms is downregulated upon differentiation of monocytes into myeloid-derived dendritic cells and is also expressed at lower levels in other debdritic cell polulations [76] . Whilst expression is low in these cells, clearly, c-fms is functional since absence of CSF-1 results in >50% reduction in dendritic cell numbers [76] , and CSF-1 can also regulate the number of tissue and blood elicited dendritic cells [77] – [79] . Thus ligation of CSF-1 to its receptor c-fms, provides both survival and differentiation signals for dendritic cells [80] .
CD148 expression is up-regulated in chronic inflammatory diseases, such as Crohn’s disease and Cogan’s syndrome [61] , [81] . The possible function of CD148 in inflammation could be similar to that of CD45, which plays an important role in lymphocyte development and function and has a crucial role in inflammation and cancer [82] . CD45 enhances T-cell and BCR signalling by dephosphorylating the auto-inhibitory phospho-tyrosine residues on Src family tyrosine kinases (SFKs) [83] . Indeed, recent studies of double knockout mice have revealed that CD45 and CD148 have overlapping functions, and can activate src family kinases in monocytes [84] ; therefore activation or enhancement of CD148 expression could lead to modulation of the inflammatory response. The current results indicate that Ptprj is highly regulated by inflammatory stimuli in vitro and in vivo , however, the precise mechanism involved in the regulation of inflammation by Ptprj remains to be elucidated.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
What is the minimum sampling effort needed to adequately document plant community richness and composition? This question forms a fulcrum upon which multiple branches of plant science have revolved for decades [ 1 – 4 ]. We still do not have a universal solution, because the answer depends both on the characteristics of the sampled community and the goals of the researcher [ 5 ]. Further, there are several dimensions of biodiversity including richness, abundance, and evenness, each of which can manifest differently across temporal and spatial scales, as well as study systems [ 6 , 7 ]. Given the complexity, and increasing urgency, of describing Earth’s biodiversity, it is necessary to continue refining our sampling approaches.
Combining knowledge and research tools from multiple disciplines is one way to better describe complex systems [ 8 – 10 ]. One instance where collaboration remains elusive is between floristic botanists and plant ecologists. While these groups flank each other on the spectrum of biodiversity scientists, they are often siloed, in part because of their different disciplinary histories and research goals [ 11 ]. Here we explore the unique research lenses and sampling approaches that floristic botanists and plant ecologists use to describe plant community diversity. We then illustrate how these different approaches are complementary in describing diversity using research conducted along an urban greenway. We close by discussing the circumstances under which collaboration is likely to be most beneficial in this time of rapid biodiversity loss.
Understanding plant community diversity requires documenting both pattern and process. We must first know which species exist, and where (i.e., taxonomic and biogeographic patterns), before we can determine why they exist there (i.e., ecological and evolutionary processes). Botanical specimens, which are routinely collected by floristic botanists, form the backbone of what we know about plant taxonomy and biogeography, from species discovery to the generation of exhaustive and meticulously vouchered local and regional floras [ 12 ] ( Fig 1 ). While such data can in aggregate be used to test hypotheses about what shapes biodiversity, the primary data (i.e., physical specimens and species lists) are not designed to uncover site-level ecological processes. In contrast, ecological data are collected for the express purpose of answering a question or testing a hypothesis [ 13 ]. This often requires statistical design and analysis, including the use of replicated transects or plots, which are distributed in a manner that reduces biases in the resulting data [ 5 , 14 , 15 ]. Unavoidably, this requirement drastically shrinks the area that plant ecologists can sample [ 16 ], which has led to a multitude of papers concerned with determining “how much sampling is enough?” (with “enough” often approximated by the asymptote on a species accumulation or rarefaction curve) [ 17 – 20 ].
10.1371/journal.pone.0244982.g001
Fig 1
Conceptual diagram illustrating the different research foci, sampling approaches, and research outputs of floristic botanists and plant ecologists, placed within the context of the species-rank abundance curve.
This curve typifies many plant communities in which a few species are common, while many species are rare. Plant ecology research often focuses on common species that drive ecological processes. Describing these processes requires transects or plots that allow for statistical hypothesis-testing to answer specific questions. This approach necessarily shrinks the area that can be sampled, causing species far out on the tail of rarity to be missed. Conversely, floristic botany focuses on species discovery, taxonomy, systematics, and the building of floras, which do not rely on statistically designed field sampling. This frees floristic botanists to search exhaustively for unique species in a study area, moving them farther out on the tail of rarity for a better estimate of a site’s “true” species pool. Each approach yields unique outputs. For example, the ecological approach may quantify contemporary relationships between plant relative abundances and their environment (top left of figure), while the floristic approach generates specimens, phylogenies, or floras with species distribution maps (top right of figure). The research outputs of each field could be more mutually informative (cycling arrows), e.g., if a collected specimen was linked to a plot-based estimate of that species relative abundance at the collection location.
Floristic botanists, quite powerfully, are not tied to plots. They are free to make informed decisions about how much area must be covered to adequately sample a plant community of interest, which can often be completely traversed using an opportunistic sampling approach. Thus, botanical collectors can cover a considerably larger area than ecologists, thereby finding species further out on the tail of rarity ( Fig 1 ) and generating a better estimate of the species pool in a given study area. Further, because botanists collect high-veracity specimens that are in flower or fruit, they usually have the structures needed for a species-level identification. Conversely, as part of their effort to reduce sampling biases, ecologists must identify everything captured in a plot regardless of phenological stage, potentially reducing the veracity of the identification or constraining it to the genus level (especially for graminoids). Still, ecologists typically capture the most abundant species in a community, with such species often driving ecological processes including nutrient cycling and response to disturbances like fire and grazing [ 21 ]. Importantly, only by using plot- or transect-based approaches can researchers move beyond documenting plant presence (richness) to quantitatively relating plant relative abundances to environmental variation.
Here we illustrate that the sampling approaches of botanists and ecologists are complementary, with the former providing critical context about the community species pool, and the latter layering on quantitative data for a subset of abundant species. We demonstrate the utility of implementing both approaches with a case study from an urban greenway in Colorado, USA. First, we compare species richness estimates generated using opportunistic sampling to those generated using ecological (line-point intercept) transects, assuming that the intensive floristic search will best approximate the “true” community species pool (sensu [ 28 ]). We then use rarefaction and species richness estimation to determine how closely the transect-based species-pool estimate reflects the larger species pool. This opportunistic versus bounded comparison of species pools is unusual (although not unique; see [ 5 , 22 ] for examples with mosses and lichens), with most vascular plant studies instead using plots of increasing size to estimate species pools [ 23 ]. Second, we explore whether the composition of the species pools captured by each approach leads to different ecological interpretations of the greenway’s flora. To this end, we compare the taxonomic coverage of each approach, as well as how species are distributed in relationship to functional group, biogeographic origin, floristic quality (based on Coefficients of Conservatism), and Wetland Indicator Status. Third, we integrate abundance estimates from the ecological transects into our interpretation of the greenway’s ecological condition to show that abundance is a desirable, if not necessary, complement to richness in any setting with applied conservation or management goals. Finally, we highlight areas of opportunity for collaboration between floristic botanists and plant ecologists.
Methods
Study system and field sampling
We sampled plant communities along the High Line Canal greenway, a 66-mile recreational trail that passes through 11 municipalities in the Denver-Metro area of Colorado. The trail runs alongside a 71-mile earthen canal (owned by Denver Water), which was excavated in the late 1800s to support agriculture and human settlement in what was historically native plains and foothills shrubland vegetation (see S1 Fig for a map of the greenway in relationship to EPA Ecoregions). The greenway thus represents a human-created waterway that is highly managed, yet supports a species pool that contains native flora (see [ 24 ] for habitat descriptions).
The greenway’s length (with the Canal’s inception located at 39.48362, -105.11293) is demarcated by mile markers that we used to generate a random subset of 45 locations at which the botanical and ecological field crews could synchronize their data collection. As is typical for collections-based floristic surveys, the botanical crew sampled exhaustively from early spring (7 May) through late summer (28 September) to capture early-, middle-, and late-season species. Starting at each of the 45 mile markers, the crew walked in the Canal’s downstream direction, searching the greenway for newly encountered species to collect and accession to the Kathryn Kalmbach Herbarium (KHD) at Denver Botanic Gardens [ 25 ]. Permission to collect plant specimens was provided by Denver Water. Most mile marker locations were sampled once during the inventory, but a few were revisited if they occurred in a vegetation type that would not be re-encountered later in the season at the other mile markers (e.g., mile markers zero and one at the inception of the Canal were the only locations in the foothills shrubland Ecoregion; S1 Fig ).
We used a staggered sampling design in which 5 mile markers spanning the southwestern to northeastern extent of the Canal were sampled every other week from May to September. The floristic survey was carried out over 57 days, comprising 850 search-hours and an estimated distance covered of 42 miles (calculated from our daily starting and stopping waypoints logged with a GPS unit; S2 Fig ). The botanical crew consisted of two botanists trained in the local flora and one to two additional non-botanists who assisted with specimen collection. All members of the crew searched for species within an ~50 to 75-foot-wide viewshed moving from the bed of the Canal, up the Canal bank, across the greenway trail, and over to the property line that marked the end of Denver Water’s ownership ( S3 Fig ). High-veracity (with identifying structures) herbarium specimens were accessioned for every species encountered during the floristic survey (numbering 1570 specimens, including duplicates; collections data available; 26). Identifications were made using [ 27 – 29 ].
The ecological sampling was carried out over 10 days, from May 22, 2018 to June 6, 2018, to capture a snapshot of plant communities around peak biomass. This method of deploying a concerted sampling effort over a short time period is common in ecological sampling, because it is often of interest to detect treatment differences that could be obscured by confounding time lags between sample dates (as opposed to the floristic botany goal of exhaustively delineating a species pool over time). At each of the 45 miles markers, we laid a 12 m × 2 m transect, the length of which captured habitat variation across the greenway corridor ( S3 Fig ). We used the line-point intercept method [ 30 ] to make field observations of plant species presence every 0.25 m along the 12 m transect (as well as bare ground, plant litter, and rocks, which we do not report herein) [ 31 ]. In the associated data set [ 32 ], the “first hit” was used to generate the reported percent cover estimates (number of hits per species per total number of hits), while the “second hit” was used to add species to our presence list. We also searched each of the two, one-meter-wide belt transects for additional species that were not encountered along the line-point transect. Voucher specimens were collected for the species encountered during the observational ecological sampling (collected outside the transects so as not to influence long-term sampling). However, given the short time period of the ecological sampling, not all specimens had flowers or fruits, and therefore were not of sufficient quality to be curated. All specimens were kept during the field season and subsequent analyses to facilitate identifications, but only higher quality specimens were accessioned to the herbarium [collections data available; 32]. Please note that one example specimen exists for potentially hundreds of field observations (i.e., each time a species was encountered along the transects).
We chose the line-point method as the most appropriate for our system, with its narrow and steep canal bank that could not accommodate other plot designs. Additionally, our aim with the ecological transects was to estimate not only species presence, but also composition. For questions about composition, the line-point method is highly repeatable across individuals and rapidly deployed, thereby maximizing sampling replicates across many locations in a single season [ 30 ]. Such transects will capture fewer species than other methods (e.g., Modified-Whitaker plots); however, any bounded sampling approach will cover considerably less area than can be achieved with opportunistic sampling based in the floristic tradition of using the habitat itself as the sampling unit [ 5 ]. Importantly, we note that it was not our goal to equalize the temporal or spatial scales of the two sampling approaches (which in our experience is not often done in practice), but rather to sample in a manner that is broadly consistent with collections-based versus ecological disciplines. The particulars of our comparison, such as the sampling window and the use of transects rather than any number of plot types, contextualize the results.
Community species pools and ecological metrics
We first estimated the species pools captured by the collections-based floristic and quantitative ecological sampling approaches, and then compared the pools in ecologically meaningful ways. Species pools are hierarchical and scale dependent, having been variously defined, but they are typically partitioned from larger scale (regional pools), to mid-scale (local pools), to smaller scale (actual or community pools). Here, we define a community species pool according to [ 33 ] as “the set of species present in a target community,” with our target community being the urban greenway. We consider the species found during the intensive floristic sampling as a best estimate of the greenway’s “true” community species pool and expect that the ecological transects will capture a subset of this larger pool. (We acknowledge that even the intensive floristic sampling will not capture the true pool, but the goal is to employ realistic sampling schemes used in botanical floristics and plant ecology to see how they compare).
After delineating the species pools, we chose several ecologically informative metrics to compare them: species distributions among families; plant functional group based on longevity and growth form; floristic quality (based on Coefficients of Conservatism or C values) [ 34 ]; Wetland Indicator Status [ 35 ]; and biogeographic origin (assigned as native or introduced to Colorado according to [ 27 ], and if introduced, whether it is cultivated). Plant functional groups are extensively used to aggregate large numbers of species into a few classes that are expected to respond similarly to changes in their environment, or to similarly affect their environment [ 36 – 38 ]. Floristic Quality Analysis is often used in the conservation realm to assess an area’s conservation value [ 34 ]. Sites with high floristic quality are relatively pristine, having departed little from the disturbance regime that existed prior to European settlement. Related to this, species in a community can be ranked on a scale of zero to 10 according to their “conservatism,” or their fidelity to habitats that are more (or less) degraded by human use. Species that can only persist in undegraded, native habitats are assigned high C values, while ruderal species, which can withstand substantial degradation, are assigned low scores (rankings below adapted from [ 39 ]).
0–3: Introduced species (always = 0), plus native species that occur in moderately to highly degraded sites (1–3)
4–6: Native species that show some affinity to natural areas and are often dominant or are present across a wide range of habitats and environments
7–8: Native species associated mostly with natural areas, but that can sometimes persist in degraded habitat
9–10: Native species that tolerate very little or no habitat degradation
Wetland Indicator Status ranks species according to their dependence on saturated soils, or wetland conditions. This metric is meaningful in our study system because the Canal represents a novel waterway in an otherwise semi-arid landscape. The status rankings are as follows. 1) Obligate Wetland: almost always a hydrophyte, rarely found in uplands; 2) Facultative Wetland: usually a hydrophyte but occasionally found in uplands; 3) Facultative: commonly occurs as either a hydrophyte or non-hydrophyte; 4) Facultative Upland: occasionally a hydrophyte, but usually occurs in uplands; 5) Upland: rarely a hydrophyte, almost always found in uplands [ 40 ]. Our study area occurs on the interface of the Western Mountain Valleys and Coasts and Great Plains regions, which can occasionally have different wetland indicators assigned for the same species. If there was a discrepancy between regions, we chose the more hydrophilic option to generate a conservative list in terms of species reliance upon water.
Data analysis
We calculated the expected number of species in our pooled samples using species rarefaction. The rarefaction curve was produced from 1000 random resamples drawn without replacement from the pool of species in the transects. We extrapolated out to 1.5X the original sample size (68 transects), as extrapolation past doubling or tripling of the reference sample size is not recommended due to increased uncertainty [ 41 ]. Asymptotic species richness was estimated using the Chao2 estimator in EstimateS version 9.1.0 [ 42 ] using the incidence of each species within each sampling transect.
We used Pearson’s Chi-square tests of independence ([ 43 ]; chisq.test in R v. 3.6.2) to explore whether the floristic and ecological sampling approaches generated different distributions of species in relationship to taxonomic coverage, functional groups, C values, Wetland Indicator Status, and biogeographic origin. For each data set (floristic and ecological) we summed species frequencies from the 45 mile markers to alleviate low cell counts within mile markers. When the sampling approaches generated significantly different ( P ≤ 0.05) distributions of species among groups, we used the adjusted standardized residuals to assess which groups contributed to the disparity (with residuals exceeding an absolute value of approximately 2 considered important; [ 44 ]). We could not assign functional groups to nine species that were only identified to the level of genus. We also could not assign C values to a subset of species that did not have them available (n = 18 species or 4% of all collections and six species or 5% from transects). The same was true for Wetland Indicator Status, in particular for cultivated species, which are not assigned this type of indictor (n = 76 species, or 17% of all collections, and 10 species, or 8% from transects).
For the purpose of assessing species composition along the greenway, we calculated abundance using percent cover from the ecological transects. Abundance estimates were calculated by taking the number of hits per species divided by the total number of hits sampled over the entire Canal [raw data available; 31].
Results
We found 452 species using the opportunistic sampling approach used in floristic botany [collections data available; 45] and 126 species using ecological transects (see S1 Table for full species list). Species richness modeled from the transect data underestimated the floristic estimate by 41% (mean = 184; 95% CI lower bound = 151; 95% CI upper bound = 253; Fig 2 ). This marked underestimate manifested despite using the Chao2 estimator, which statistically accounts for the fact that uncommon species will likely be missed. The species rarefaction curve showed that a 51% increase in our sampling effort, from 45 to 68 transects, would only capture 31% of the species pool observed during the opportunistic sampling (mean = 141; 95% CI lower bound = 125; 95% CI upper bound = 157; Fig 2 ). While neither the species rarefaction curve nor the Chao2 estimator reached a definitive asymptote (although they were distinctly leveling off), our transect sampling effort was based on what we could reasonably achieve with the available financial and personnel resources. We expect that other researchers are similarly constrained in most applied situations.
10.1371/journal.pone.0244982.g002
Fig 2
Species richness estimated using transect-based rarefaction (black line shown with 95% CI in red shading) based on 45 sampled transects and extrapolated up to 68 transects (values after the vertical black line).
An asymptotic richness estimate was calculated using the Chao2 estimator (grey line with 95% CI in green shading).
The taxonomic coverage of the ecological versus botanical sampling became less complete moving hierarchically from family, to genus, to species (with transects capturing 50% of families [39 versus 78], 36% of genera [103 versus 289], and 28% of species [126 versus 452]; see S4 Fig for species distributions among families). The distribution of species among functional groups was statistically similar for the two sampling approaches ( Fig 3A ; χ 2 = 5.6, df = 7, P = 0.59), although perennial grasses were weakly over-represented (by 6%) along transects, reflecting the pattern found for family distributions (see Poaceae, S4 Fig ). The floristic quality of the greenway, based on the distribution of species among C values ( Fig 3B ; χ 2 = 9.9, df = 10, P = 0.45), did not differ significantly between the two sampling approaches. Wetland Indicator Status significantly differed ( Fig 3C ; χ 2 = 10.6, df = 4, P = 0.03), with transects underestimating obligate wetland species by 5% (adjusted residuals = -2.05) and upland species by 10% (adjusted residuals = 2.0) relative to the larger species pool. In terms of biogeographic origin, the transects over-estimated (although non-significantly) the proportion of introduced species (55%) relative to the opportunistic sampling (46%; χ 2 = 2.75, df = 1, P = 0.10). The composition of introduced species captured by the two approaches differed in an important respect, with the floristic approach capturing a substantial number of uncommon, non-native garden cultivars (79 cultivars out of 208 introduced species, or 38%) that the transects missed (four cultivars out of 68 introduced species, or 6%).
10.1371/journal.pone.0244982.g003
Fig 3
A-C. Distribution of species captured by the floristic botany (collections) and ecological (transects) sampling approaches in relationship to functional groups, floristic quality (based on Coefficients of Conservatism or C values), and Wetland Indicator Status. The distribution of species among functional groups and C values did not significantly differ, while the distribution among wetland indicator status differed. Functional groups: PF = perennial forb; W = woody (and perennial); AF = annual forb; PG = perennial grass; VFG = variable forbs and grasses (annual to short-lived perennials; includes only two species of grasses); AG = annual grass; BF = biennial forb. Wetland Indicator Status: OBL = obligate wetland; FACW = facultative wetland; FAC = facultative; FACU = facultative upland; UPL = upland. See text for statistics and definitions of C values and wetland indicators.
In terms of abundance, the greenway revealed a typical species-rank abundance curve in which a few species dominated, while a long tail of uncommon species contributed to species richness ( Fig 4 ). The three most abundant species, which comprised ~50% of plant cover during our sampling window, were three non-native grasses: Bromus inermis L. (smooth brome), Bromus tectorum L. (cheatgrass), and Agropyron cristatum (L.) Gaertn. (crested wheatgrass).
10.1371/journal.pone.0244982.g004
Fig 4
Species relative abundances (percent cover) estimated using the ecological transects showing that only three species (the introduced grasses Bromus inermis L., Bromus tectorum L., and Agropyron cristatum (L.) Gaertn.) made up nearly half of the greenway’s percent cover during the 2018 sampling window.
Compare with the conceptual diagram of the species-rank abundance curve in Fig 1 .
Discussion
Here we illustrate that sampling approaches from floristic botany and plant ecology capture complementary dimensions of biodiversity. As hypothesized, opportunistic sampling generated a markedly more robust (nearly four times larger) empirical estimate of the greenway’s community species pool than did the transects. Yet, the transects revealed that only three species of introduced grasses comprised 50% of plant cover at the time of sampling. Generating this critical indicator of the greenway’s ecological condition necessarily traded off with a fuller assessment of the species pool. This long-acknowledged trade-off is typically addressed by using rarefaction and accumulation curves and species richness estimators based on transect or plot data [ 46 ]. Our findings suggest that for sampling designs with limited areal coverage, such tools may not adequately capture the long tail of uncommon species that contribute disproportionately to species richness [ 13 , 47 , 48 ]. This issue was raised by Heilmann-Clausen and Læssøe [ 49 ], who clarified that species accumulation curves and species richness estimators address “how many species will be recorded if [a particular] sampling regime is followed in perpetuity or extended to cover all available habitat,” rather than telling the size of the species pool in the system. Related, Newmaster et al. (2005) illustrated that rarefaction and Chao estimates of common forest moss species reached an asymptote at small sample sizes (25 plots), while estimates of richness for rare species never leveled off. Thus, the very transect- or plot-based studies that often rely on estimator tools to determine the completeness of sampling likely produce underestimates of the true species pool [ 50 , 51 ]. Of course, many ecological questions can be rigorously answered without exhaustive documentation the species pool. Still, it is worthwhile to address these interpretive nuances when presenting results based on species estimators.
The question then becomes whether and when the transect-based sampling constraint, and any attendant underestimate of the species pool, affects how the ecology of the sampled area is interpreted (at least in terms of species richness). To assess this, we grouped species into ecologically meaningful categories to reduce complexity and uncover general patterns that are independent of species identity per se [ 52 , 53 ]. We found that based on functional groups and floristic quality (i.e., C -values), the ecological condition of the greenway appears similar between the two species pools. This comparable delineation of functional groups is especially desirable given that functional group identity and diversity are routinely used to gauge community response to disturbances such as fire and biological invasions [ 54 , 55 ]. In our system, functional traits such as longevity and woodiness (and attendant traits like rooting depth) likely shape soil and hydrologic conditions along the Canal banks. In terms of floristic quality, both sampling approaches captured the bell-shaped distribution of native species and similarly indicate that few conservative ( C -value of 7 or more) species remain along this highly modified urban corridor. These findings corroborate previous work showing that floristic quality performs well when using transects or plots, because mean C -values are less dependent on the area sampled than is species richness [ 34 ].
While transects robustly captured patterns associated with functional groups and floristic quality, differences between the sampling approaches arose for Wetland Indicator Status and the presence of uncommon cultivated species. In particular, patchily distributed wetland areas and the obligate wetland species they harbor went largely undetected by the transects. This is a non-trivial miss in a Canal system where episodic drought and various management strategies strongly affect the hydrologic regime and thus persistence of sensitive wetland areas. Transects also failed to detect the nascent incursion of garden plants from adjacent private properties onto the Canal banks, potentially hindering Early Detection and Rapid Response (EDRR) management interventions [ 56 , 57 ]. These examples illustrate that without knowing the more complete species pool, the ability to thoroughly uncover sensitive ecological conditions along the greenway would be hampered. Still, if only species richness from the floristic inventory were used to assess the greenway, the high abundance of non-native grasses would go unreported. Without these abundance data, it would be impossible to assess costs associated with potential control or revegetation efforts, as well as to relate these dominant grasses to ecological processes of interest. For example, we are currently asking how the implementation of green stormwater infrastructure will affect vegetation along the Canal banks, and in turn, how in situ vegetation will affect stormwater infiltration, retention, and removal of pollutants. We are now positioned to take a two-pronged approach to this question by exploring how the most abundant species might shape stormwater dynamics, while also integrating information about the identity and location of uncommon species likely to be particularly responsive to hydrologic changes (e.g., obligate wetland species) and disturbance from infrastructure installations (e.g., establishing individuals of ruderal non-natives).
Broadening out from our greenway example, when should floristic botanists and ecologists develop on-the-ground collaborations to better describe contemporary biodiversity? We suggest that any question about the maintenance of community-level species diversity would benefit from a paired approach, as it is ultimately the interplay of local (competition, predation, microenvironmental variation) and regional (immigration and extinction) processes that shape biodiversity [ 58 – 60 ]. To better integrate across spatial scales, it would be powerful to link quantitative data from bounded sampling to floristically based best estimates of the community species pool that functions as the backdrop for species immigration into embedded transects or plots. (While many terrestrial plant studies sample hierarchically across plot sizes to infer species pool sizes, they fall short of breaking free of plots to approach a comprehensive site- or habitat-level survey; [ 23 ]). Moreover, pairing floristic and ecological approaches addresses the call for increased metric complementarity in assessing biodiversity [ 61 ]. In particular, the historically heavy reliance upon species richness as a sole indicator of ecosystem health or biodiversity change has proven insufficient, as it fails to capture changes in other key phenomena such as species turnover and changes in species relative abundances [ 61 , 62 ]. Re-imagining floristic surveys of species richness as integral components of hypothesis-driven ecological work that uses other metrics can lead to new insights.
For example, to achieve metric complementarity in a restoration context, a floristic inventory could be used to assess the feasibility of passive restoration (which depends on the community species pool; [ 63 , 64 ]), while paired ecological sampling could quantify the degree of habitat degradation and monitor effectiveness of restoration efforts. Invasive species management would also benefit from a combined sampling approach, where initial arrivals of rare introduced species to an area are captured during unbounded floristic sampling bouts, while the spread and population biology of already established populations are monitored using transects or plots. Indeed, integrating researchers versed in alpha taxonomy is critical in invasion biology, as the resolution of taxonomically challenging groups, including those that hybridize, is critical to proper ecological interpretation [ 10 ]. A further boon of integrating floristic botany into settings where the species pool is of interest is excellent temporal sampling of early- to late-season bloomers, which can be missed using the “peak biomass” approach typical of ecological sampling [ 65 ]. Moreover, validation of transect- or plot-based data with vouchered specimens is tantamount to institutional knowledge that can be readily accessed by all researchers who carve out projects from a particular locale. The long-standing specimen and data curation practices used in natural history collections have achieved a level of standardization and data-sharing not yet realized in the ecological realm. However, improvements are being made in this arena, such as the application of the event-based Darwin Core data standard when publishing ecological data, as we have done herein [ 26 , 45 ].
To most informatively link plot- and site-scale diversity data, best practices would be to pair the replicable plot data with equally replicable floristic sampling of the community species pool [e.g., 22,62,63]. Collections-based floristic botany has not fully adopted standardized field sampling practices across individual collectors [ 66 ], which can limit the use of collections data for analyzing hypothesis-driven questions about community-level change in diversity [ 67 ]. The value of specimens and species lists generated using the opportunistic sampling approach could be increased by reporting the spatial extent of the surveyed area [ 68 ], as well as the intended goal of the collection event (e.g., an exhaustive inventory [implying species absence] versus targeted sampling based on an investigator’s taxa of interest). Such reporting practices can be achieved within the Ecological Metadata Language standard contained within a data package [ 69 ] and would provide additional context for downstream uses of aggregated data (e.g., species pool estimates derived from geo-referenced specimen databases) [ 70 ]. However, it must be considered that standardizing collections data (or at least reporting accurate areal coverage of survey sites) requires some degree of bounding that is both time-consuming and at odds with maximizing the number of species encountered [ 68 ]. Thus, it may only be worth bounding collections-based sampling when the specific question calls for it (e.g., quantifying immigration into experimental plots from the surrounding species pool).
Our experience is that physical collections are not often considered by ecological principal investigators as essential to their field protocols (despite substantial movement in this direction by, e.g., the National Ecological Observatory Network). This is partly because it is no small task to integrate the disparate training, project planning, data curation, and analyses implemented in floristic botany and ecology [ 71 – 73 ]. Thus, cross-disciplinary partnerships across non-profit, governmental, and academic institutions are key. We suggest that ecologists reach out to their campus or regional herbaria, connect with curators and collections managers, and dedicate a line item in their research budgets for vouchered floristic (or faunistic) surveys of their study sites (see [ 74 ] for discussion of under-funding in collections-based research). We similarly suggest that curators and collections managers build relationships with ecology principal investigators and members of their labs, sharing their skills as integral assets to be included a priori into the proper design of biodiversity-focused ecological fieldwork (i.e., it is imperative to quash the “end-user mode” attitude that views botanists as simply providing identification services to those who use their keys and field guides [ 75 , 76 ]). For example, gaps in biodiversity data, including both species discovery and ecological monitoring, are high in tropical relative to temperate ecosystems [ 77 , 78 ], and would thus be best addressed by teams of floristic botanists and ecologists. As large-scale digitization of collections data has revealed, there are myriad, previously unimaged ways that natural history collections can inform ecological questions [ 79 ]. We believe our understanding of biodiversity 100 years from now can only benefit from thoughtful co-exploration of today’s ecosystems by floristic botanists and ecologists.
Supporting information
S1 Fig
Map of High Line Canal greenway in relationship to US EPA level IV Ecoregions.
The “Character Zones” overlaid on the Ecoregions represent large-scale variation from the southwest to the northeast of the greenway, characterized by a transition from foothills to plains habitat, which is in turn overlaid by different degrees of land use intensity. These habitat and land use factors shape the “character” of the greenway, as the viewshed changes in relationship to topography, type and density of vegetation, and the type and density of surrounding development.
(TIF)
S2 Fig
Map of greenway extent covered during the opportunistic floristic sampling as measured by daily starting and stopping waypoints logged with a GPS unit.
The “Character Zones” overlaid on the satellite imagery represent large-scale variation from the southwest to the northeast of the greenway, characterized by a transition from foothills to plains habitat, which is in turn overlaid by different degrees of land use intensity. These habitat and land use factors shape the “character” of the greenway, as the viewshed changes in relationship to topography, type and density of vegetation, and the type and density of surrounding development.
(TIF)
S3 Fig
Schematic of ecological transect orientation in relationship to High Line Canal greenway.
(TIF)
S4 Fig
Species distributions among the most common families observed along the High Line Canal greenway in Colorado, USA, using floristic botany (opportunistic) and ecological (transect-based) sampling approaches.
(TIF)
S1 Table
List of species found using the floristic collections-based versus ecological transect-based sampling approaches.
(XLSX)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Interleukin-2 (IL-2) has long been recognized for its role in promoting T cell proliferation in vitro , yet only recently has appreciation grown for its paradoxical role in maintaining peripheral tolerance in vivo . One key means by which IL-2 appears to promote tolerance involves its influence over CD4 + CD25 + T cells, often termed regulatory T cells (Treg). In 1995, Sakaguchi and colleagues identified the α-chain of the IL-2 receptor (CD25) as a putative marker of functional Treg [1] . The high-affinity, trimeric IL-2 receptor complex consists of three major sub-units; the α-chain (designated CD25), the β-chain (CD122), and the common cytokine receptor γ-chain (CD132) [2] . These subunits, as well as other fragments of the molecule, appear to play important roles not only in immune physiology, but as markers of immune activity.
The concept that the expression level and affinity of cell surface receptors dictates cellular fate was introduced as the quantal theory by Kendall A Smith [3] . In activated T cells, signaling through the IL-2 receptor results in increased high-affinity receptor expression and eventual activation induced cell death (AICD) through up-regulation of Fas and FasL [4] , [5] . It has been known since 1985 that production of the soluble form of CD25 (sCD25) associates with T cell activation in vitro [6] . In healthy individuals, sCD25 is present in serum at approximately 2 ng/ml, whereas increased serum sCD25 levels have been associated with lymphocyte activation in infection as well as in hematologic malignancies [7] , [8] . More recently, polymorphisms in the CD25 gene have been associated with autoimmune diseases such as type 1 diabetes, multiple sclerosis, and Graves' disease; with measurable differences in serum levels of the soluble form of the corresponding molecule (i.e., sCD25) [9] – [12] . These data point to, but do not specifically identify, a functional role for sCD25 separate from its value as a biomarker of activation in serum. Indeed, several key questions regarding sCD25 remain unanswered. These include those asking what biological actions result in the formation of sCD25, and which role (if any) does sCD25 play in modulating immune responsiveness?
While an alternative splice variant cannot be ruled out as contributing to the production of sCD25 [13] , [14] , it is more commonly thought that sCD25 results from a proteolytic cleavage event driven by a family of enzymes collectively referred to as “sheddases”. To date, at least three enzyme candidates have been reported to possess the capacity for cleaving CD25 including the endogenous enzymes elastase and matrix metalloproteinase-9 (MMP-9), as well as the environmental house dust mite allergen DerP1 [15] – [17] . While these enzymes have each been shown to be sufficient to elicit cleavage, the formation of sCD25 is often still observed in their absence or in the presence of enzymatic inhibitors. Thus, there may be additional unidentified-enzymes or overlapping enzymes responsible for sCD25 formation.
It has been suggested that sCD25 itself may bind IL-2 in vivo , preventing it from forming complexes with α2-macroglobulin and thus preserving its bioavailability for T cells [18] . The affinity of the monomeric soluble receptor for IL-2 is relatively low (K d = 30 nM) [19] . This is comparable to the affinity of the monomeric membrane α-chain for IL-2 (K d = 10 −8 M), followed by that of the intermediate βγ complex (K d = 10 −9 M) and finally the high-affinity αβγ-trimeric complex (K d = 10 −11 M) [20] . An antagonistic role for sCD25 has also been suggested [21] . More recent studies have depicted sCD25 as an early inhibitor of IL-2 signaling in T cells that also enhances their proliferation in short and long term in vitro culture [12] . If this molecule is indeed participating in IL-2 signaling as suggested, then it may be possible to apply the quantal theory to this soluble receptor. sCD25 could then be expected to play multiple roles depending on the local environment in regards to IL-2 concentration, cellular activation state, and the amount of sCD25 present.
Methods
Objectives
We hypothesized that the control of membrane and soluble forms of CD25 on CD4 + T cells is critical for regulating immunity and tolerance, and that the presence of sCD25 and other factors in serum may be influencing in vitro assays. Herein, we describe efforts addressing these notions as well as our attempts to shed light on the aforementioned knowledge voids regarding sCD25 production and function. We sought to investigate the production of sCD25 during in vitro suppression assays as well as during PBMC activation. In addition, we explored the role of sCD25 during suppression and proliferation via addition of recombinant protein to in vitro assays for this activity. These efforts attempted to avoid interference from sCD25 present in supplemented serum by using serum free media. Taken together, we believe this effort provides valuable insights into the role of CD25 stability and sCD25 production on the process of immune regulation.
Participants
Peripheral blood was obtained from normal healthy control subjects (11M/6F; median age 30.4, range 21.1 to 45.4 years) without a recent infection, allergic episode, or having a known autoimmune disorder.
Description of Procedures or Investigations undertaken
Staining and Flow Cytometry
Cells were aliquoted (0.5–1×10 6 in 100 µl per tube) along with each appropriate antibody, including anti-CD3 (clone HIT3a), anti-CD25 (M-A251), anti-CD4 (SK3), anti-CD127 (clone hIL-7R-M21), or isotype controls mouse IgG1 (X40), mouse IgG1 (MOPC-21), and mouse IgG2b [27] – [35] (BD Biosciences). FOXP3 staining was conducted with the anti-human FOXP3 (clone 206D) staining kit according to manufacturer recommendations (Biolegend). Flow cytometric analysis was conducted using FCS Express (De Novo Software) or FlowJo software (version 7.2.2, TreeStar).
Cell Purification and Fluorescence-Activated Cell Sorting
Cells used in functional suppression assays were separated into a T cell depleted accessory cell population (irradiated 3300 rads) and functional CD4 + CD25 + Treg and CD4 + CD25 − Teff cells. The untouched accessory cell population was produced by incubating an aliquot of blood (5 ml) with a T cell depletion antibody cocktail (RosetteSep; StemCell), followed by density gradient centrifugation (Cellgro) with Ficoll-hypaque solution (Amersham/GE Healthcare). CD4 + T cells were pre-purified from the remaining blood by negative selection using a CD4 + T cell enrichment cocktail (Stem Cell). Briefly, 50 µl of cocktail per 1 ml of whole blood was incubated for 20 min at 23°C, subjected to density gradient centrifugation, washed twice in PBS (Ca ++ and Mg ++ free) containing 0.1% BSA (Sigma-Aldrich), then this “untouched” CD4 + population underwent an additional high-speed FACS sorting procedure to yield CD4 + CD25 Hi Treg cells and CD4 + CD25 − Teff cells. For specified assays, the CD127 marker was also used during sorting to yield CD4 + CD127 −/lo CD25 + T cells, as these cells have been reported to contain a highly enriched FOXP3 + population [22] . Sorted Treg and Tconv cell populations were analyzed for purity following FACS with anti-human FOXP3 ( Figure S1 ). PBMC for later proliferation experiments were isolated by density gradient centrifugation over Ficoll-Hypaque.
Isolation of Purified Treg and Teff Cell Populations by FACS
Pre-enriched CD4 + T cells were stained by the addition of anti-CD3 (clone HIT3a), anti-CD25 (M-A251), and anti-CD4 (SK3) antibodies at 5 µl/10 6 cells. PBS containing 2% human AB serum was added for a final staining volume of 10 8 CD4 + T cells/ml (30 min at 4°C), washed in PBS/0.1% BSA wash buffer, centrifuged (300 x g), and resuspended in wash buffer (3 ml) prior to high-speed cell sorting on a BD FACSVantage Cell Sorter. Sort gates for CD4 + CD25 Hi T cells were set to optimize the yield and purity of FOXP3 expressing T cells within the CD4 + CD25 + T cell fraction [23] . Where indicated, anti-CD127 ((clone hIL-7R-M21), BD Biosciences) was added to discriminate Treg from Teff cells. Purified CD4 + CD25 HI and CD4 + CD25 − T cell fractions were collected in sterile tissue culture tubes (BD Biosciences) containing cold (4°C) AIM V SFM (Invitrogen).
Cell Culture
Suppression assays were performed in AIM V complete SFM (Invitrogen) with the addition of freshly isolated autologous serum (heat inactivated at 56°C for 30 min), where indicated. PBMC proliferation assays took place in either RPMI 1640 (CellGro) supplemented with 5% human AB serum (Cellgro), or AIM V, XVIVO 15, or CTL Test serum free media. All assays were performed in U-bottom 96-well plates (Costar) and maintained at 37°C in 5% CO 2 .
Suppression Assay Co-culture System
CD4 + CD25 Hi Treg cells were added in decreasing ratios (1∶0, 1∶1, ½∶1, and 0∶1) to a constant number of CD4 + CD25 − Teff cells (5×10 3 cells/well). A combination of 5 µg/ml soluble anti-CD3 (clone HIT3a) and 2.5 µg/ml soluble anti-CD28 (clone CD28.2; eBioscience,) provided proliferation stimulus for a 120 h culture period. 5×10 4 irradiated (3300 rads) T cell depleted accessory cells were also added to each well in a total volume of 200 µl. One µCi of 3 H-Thymidine (Amersham Biosciences) was added at 96 h for a final 16 h of culture to assess proliferation. Supernatants (20 µl/well) from triplicate wells were collected at 48 h and 96 h to assess sCD25 and cytokine production. Suppression was calculated by the reduction of 3 H-thymidine incorporation using the following equation: Percent suppression = (1 - (mean CPM Treg+Teff)/(mean CPM Teff) ×100%). For FACS based suppression assays, proliferation was calculated by division index (DI), with suppression calculated as previously described [24] . For T cell subset proliferation assays, suppression assays were performed as above but with co-cultures containing either normal or irradiated Treg and Teff cell populations (3300 rads), or both.
Fluorescent Cell Labeling
To track proliferation of Treg and Teff cells individually, each cell population was labeled with either the “red” dye PKH26 (Sigma, final labeling concentration 2 µM) or “green” dye CFSE (Invitrogen, 1 µM), for Treg and Teff cells, respectively [24] . For cell labeling experiments, functional CD4 + CD127 lo/− CD25 + Treg and CD4 + CD127 + Teff cell populations were obtained by FACS. Suppression experiments were conducted as described for 3 H-thymidine based suppression assays, in the following combinations (Treg to Teff cell ratios 1∶0, 1∶1, 1/2∶1, 1/4∶1, 1/8∶1, 1/16∶1, 1/32∶1 and 0∶1; where 1 = 5×10 4 cells). In addition, 5×10 5 T cell depleted and irradiated (3300 rads) accessory cells were added. Cultures were run in triplicate and pooled prior to FACS analysis. At harvest, cells were washed in PBS containing 0.2% BSA, stained with CD4-APC (BD Biosciences), and analyzed by FACS for loss of cellular fluorescence in both proliferating CD4 + populations.
PBMC Proliferation Assays
Freshly isolated PBMC were plated at 10 5 cells per well. Cells were left unstimulated or treated with PHA-L (Sigma-Aldrich) at 5 ug/ml, or α-CD3 and α-CD28 coated beads (Miltenyi) at a bead∶cell ratio of 1∶2. In addition, recombinant sCD25 was added at 2 and 20 ng/ml to duplicate wells for each condition. One µCi of 3 H-Thymidine (Amersham Biosciences, Piscataway, NJ) was added at 96 h for a final 16 h of culture to assess proliferation. Supernatants were collected at 96 hours to measure sCD25 and nuclear matrix protein 41/7. Stimulation indices were determined by dividing the mean CPM of stimulated cultures over relevant unstimulated controls.
Supernatant Cytokine, sCD25 and Nuclear Matrix Protein Detection
Cytokines were assessed from cell culture supernatants in a multiplex format utilizing the Luminex 100 xMAP System (Austin, TX), with a multiplexed kit (Beadlyte® Human 22-plex Multi-Cytokine Detection System 4; Upstate Biotechnology). Samples were diluted 1∶2 in AIM V medium prior to analysis. sCD25 levels were determined by ELISA according to manufacturer instructions (BD Biosciences). Samples were diluted (when necessary) in PBS containing 10% FBS (pH 7.0). To identify a normal range of sCD25 in human serum, 60 samples were randomly selected from a healthy control population of adolescents and adults, whose origins were as previously described [23] . Soluble nuclear matrix protein 41/7 was measured by ELISA (Calbiochem, San Diego, CA) according to manufacturer instructions.
Reagents
Recombinant sCD25 was obtained from R&D Systems (Minneapolis, MN) for suppression assays, and from Spring Biosciences for PBMC proliferation assays. Recombinant MMP-2, MMP-9 and the selective MMP2/9 inhibitor were acquired through Calbiochem (San Diego, CA).
Ethics
Institutional Review Board (IRB) approved informed consents were obtained in accordance with approved protocols.
Statistical Methods
Statistical analyses utilized GraphPad Prism 4.00 software (GraphPad, San Diego, CA), with Student's paired t tests or ANOVA with the Bonferroni post hoc correction for multiple comparisons along with Spearman's correlation analyses. For all tests, p <0.05 was deemed significant.
Results
Treg Cells Require Serum for Optimal Suppression of T Cell Proliferation
To investigate sCD25 production during the in vitro suppression assay, peripheral blood CD4 + CD25 − conventional T cells (Tconv) and CD4 + CD25 Hi Treg were isolated by FACS. The isolated Treg population demonstrated modestly lower CD4 expression and CD25 Hi surface expression, which enriches for cells expressing FOXP3 ( Figure S1 ). To analyze the production of de novo sCD25 produced during the suppression assay from the levels already present in healthy control human serum (2039.0±899.8 pg/ml), assays were conducted under serum-free media (SFM) conditions or in SFM supplemented with specified concentrations of autologous serum.
Surprisingly, serum absence markedly abrogated the ability of Tregs to suppress proliferation as observed in 3 H-thymidine-based suppression assays ( Figure 1A ). This is in contrast to the standard suppressive response; an assay whose outcomes could be fully recovered by adding small (i.e., 1.0%) amounts of serum ( Figure 1B ). Proliferation was significantly reduced in SFM compared to those containing serum ( Figure 1C ). The reduced capacity of Treg to suppress under serum free conditions could be a result of altered cellular survival. To examine this possibility, we measured a marker of cellular apoptosis and necrosis during the suppression assay in the presence and absence of serum. Nuclear matrix protein (NMP) levels in culture supernatant were as follows in SFM: ((mean +/− SEM) 1∶0 Treg to Teff = 57.7+/−18.8, 1∶1 = 279.3+/−46, 0∶1 = 254.3+/−35) and in 1% autologous serum: ((mean +/− SEM) 1∶0 = 39.6+/−10.7, 1∶1 = 193.7+/−16.9, 0∶1 = 185.3+/−51.5). Teff cell death did not change in coculture with Treg, suggesting that the altered suppression patterns observed in SFM were unlikely a result of significant changes in cell death. Moreover, Tregs alone did not exhibit significantly increased death in SFM alone. In fact, under SFM conditions, Treg and Teff co-cultures often exhibited increased proliferation over Teff cells alone ( Figure 1D ; % suppression at a 1∶1 Treg to Teff ratio; mean + SEM, −197.8+270.6 for SFM vs. 70.4+17.3 with 1.0% serum; p = 0.04; Student's paired T test). This response was also observed when cells were cultured at a ratio of ½∶1 Treg to Teff cells (−103.1±91.5 vs. 34.8±18.12, for SFM and 1.0% serum, respectively; p = 0.01).
10.1371/journal.pone.0007980.g001 Figure 1
CD4 + CD25 + Treg cells require serum for suppression of T cell proliferation.
Suppression assays were conducted utilizing FACS sorted cells run in parallel under either (A) SFM or (B) SFM supplemented with 1.0% autologous serum (n = 7 healthy controls). Treg cells were plated alone (5×10 3 cells/well), and in decreasing ratios (1∶1, ½∶1, and 0∶1) to a constant number of Teff cells. Cells were stimulated with soluble anti-CD3 (5.0 µg/ml) and anti-CD28 (2.5 µg/ml) in the presence of a ten-fold excess (5×10 4 ) of irradiated accessory cells. (C) Graph indicates data plotted from panels (A) and (B) to highlight the differences in proliferation between serum free media (open bars) and 1% serum (closed bars). Cells proliferated significantly less in SFM at Treg to Teff ratios of ½∶1 and 0∶1 (p<0.001 for both conditions). (D). Graph indicates the percent suppression calculated under serum-free conditions ( open squares ) and supplemented with 1.0% serum ( closed squares ) at a ratio of 1∶1-Treg to Teff cells (left data points) and at a ratio of ½∶1 Treg∶Teff cells (right points). Under serum free conditions, Treg fail to suppress proliferation and often lead to increased proliferation in the co-culture (% suppression = mean + SEM, −197.8+270.6 for SFM vs. 70.4+17.3 with 1.0% serum; * p = 0.04 at a ratio 1∶1 Treg to Teff. This trend continued at a ratio of ½∶1 Treg to Teff cell (−103.1+91.5 vs. 34.8+18.12, for SFM and 1.0% serum, respectively; ** p = 0.01). (* p <0.05, ** p <0.01, and *** p <0.001).
The degree of in vitro suppression was dependent on both Treg and Teff cell subsets. We noted a dramatically reduced proliferation of Teff cells in SFM versus those in 1.0% serum ( Figure 1A and 1B , 0∶1 condition; mean CPM = 22,840+16,456 for SFM vs. 252,755+24,207 with 1.0% serum, p = 0.0006), whereas Treg proliferation was negligible irrespective of the presence of serum when Treg were cultured alone.
Treg Cells Exhibit Reduced Suppressive Properties and Proliferate when Cultured with Teff Cells under Serum Free Medium Conditions
To assess the individual contributions of Treg and Teff to proliferation in co-culture, these cells were FACS sorted ( Figure 2A ) and then subjected to specific dyes (i.e., the red dye PKH26 for Treg, green dye CFSE for Teff) ( Figure 2B , left plot ). In vitro suppression assays were performed and proliferation tracked by the loss of cellular fluorescence in each responding CD4 + T cell population ( Figure 2B–2D ).
10.1371/journal.pone.0007980.g002 Figure 2
Contribution of Treg and Teff cells to increased proliferation in suppression assay co-cultures.
Peripheral blood CD4 + T cells ((A), left plot) were FACS sorted based on surface expression of CD25 (y-axis) and CD127 (x-axis) to yield CD4 + CD127 lo/− CD25 + T cells ((A), middle plot) and CD4 + CD127 + CD25 − Teff cells ((A), right plot). Following isolation, Teff cells were labeled with CFSE ((B), left plot, y-axis) and Treg cells labeled with PKH26 (( B ), left plot, x-axis). Proliferation of each gated CD4 + T cell population was then assessed following activation conditions utilized in the suppression assay. ( B ) Left plot indicates fluorescence levels of unstimulated cells (depicted in red for Treg and green for Teff), and following 96 h of in vitro activation ((B), right plots). Proliferation of Treg and Teff cells are shown at a ¼∶1 Treg to Teff cell ratio and plotted under conditions of SFM (dashed lines) or in the presence of 5% serum (solid lines). (C) For all ratios of Treg to Teff cells, Treg maintain increased suppressive potential and anergic properties in the presence of serum as measured by division index (DI). (D) In vitro suppression calculated from DI indicated increased suppression by Treg in the presence of serum. Data plotted represent one example of five independent experiments. (E) A standard suppression assay was set up under SFM conditions with either standard or irradiated (3300 rads) Treg and Teff cells as indicated. This analysis demonstrated a synergistic response at a 1∶1 Treg to Teff ratio, with near maximal proliferation following addition of irradiated Teff cells. Statistical significance indicated as * p <0.05, ** p <0.01, and *** p <0.001.
In this system, CD4 + CD127 −/lo CD25 + T cells become anergic in the presence of serum ( Figure 2B , 1 /4∶1 Treg to Teff cell ratio; dashed green histograms representing culture in SFM and solid overlaid histogram representing culture in 5.0% serum). Proliferation of PKH26 labeled Treg and CFSE labeled Teff cells were tracked according to division index (DI) and plotted for all ratios of Treg to Teff cells during the in vitro suppression assay ( Figure 2C ).
In an effort to more accurately measure the contribution of each cell population to the increased proliferation, we also established a suppression assay in SFM with either normal or irradiated Treg and Teff cell populations ( Figure 2E ). This analysis indicated that increased proliferation within the CD4 + CD25 + T cell fraction is predominantly responsible for the synergistic response observed in the co-culture as similar proliferation levels were detected in the presence of normal or irradiated Teff cells (33120.0±4866.0 vs. 28828.33±1382.0 CPM, p = NS).
Production of sCD25 During Suppression Assay
Treg and conventional T cells have classically been selected based upon their expression of the membrane-bound form of CD25 under baseline conditions [1] . In spite of this, little is known about what happens to soluble and membrane CD25 during the in vitro suppression assay. We collected and analyzed supernatants from each condition during the suppression assay in order to determine sCD25 production at 48 and 96 h time points ( Figure 3 ). Overall, this analysis indicated that Treg cells, which were selected based upon high membrane CD25 expression levels, maintain CD25 in a membrane-bound form following polyclonal activation in vitro ( Figure 3 , 1 ∶0 conditions). Teff cells, on the other hand, are known to upregulate CD25 in response to activation, but then subsequently “shed” it into the tissue culture medium ( Figure 3D , 0∶1 condition).
10.1371/journal.pone.0007980.g003 Figure 3
Production of sCD25 during the in vitro suppression assay.
Supernatants from triplicate wells were pooled and analyzed for the production of sCD25 by ELISA (n = 7 control subjects). Under conditions of both serum-free and 1.0% serum, the highest levels of sCD25 were detected in the 1∶1 Treg to Teff co-culture condition at the early 48 h time point, but were only significantly higher in media containing 1% serum ((A) and (B)). At the 96 h time point ((C) and (D)), the pattern of sCD25 production more closely resemble the responses observed in the proliferation assay assessed by uptake of 3 H-thymidine. (* p <0.05, ** p <0.01, and *** p <0.001).
This type of analysis highlights the key influence of culture conditions and timing on the production of sCD25. At an early time point of 48 h, the highest levels of sCD25 are uniformly observed in the 1∶1 Treg to Teff cell co-culture under SFM and serum supplemented conditions ( Figure 3A,B ). At a later time point of 96 h, the detection of sCD25 more closely resembled the proliferation data observed in the suppression assays ( Figure 3C,D ). In this assay, the absence of serum resulted in detection of the highest levels of sCD25 in the 1∶1 Treg to Teff co-culture condition ( Figure 3C ; mean + SEM, 3880.9+4756.5 pg/ml vs. 1043.3+863.8 for Teff cells alone). In line with the proliferation results in the presence of serum, the highest levels of sCD25 were also detected from Teff alone wells ( Figure 3D , 0∶1 condition; 5369.5+2575.6 pg/ml). A modest level of suppression in sCD25 levels was observed from the co-culture conditions at a ratio of 1∶1 and 1/2∶1 Treg to Teff cells (4311.1+2029.0 and 5245.3+2666.1, respectively). Finally, in line with the anergic and suppressive properties of Treg, the lowest levels of sCD25 were detected from Treg cultures alone ( Figure 3D , 682.2+367.1 pg/ml).
The Production of sCD25 Correlates with Cellular Proliferation in the Suppression Assay
Considering the similar patterns of cellular proliferation and sCD25 production during the in vitro suppression assay, we sought to determine the relationship between proliferation and sCD25 production. Under both serum-free and serum-supplemented conditions, the production of sCD25 at 96 hours correlated with the levels of cellular proliferation observed under all ratios of Treg to Teff cells ( Figure 4A and 4B ; r = 0.67, p = 0.0002 for SFM and r = 0.76, p <0.0001 for 1.0% serum).
10.1371/journal.pone.0007980.g004 Figure 4
Levels of cellular proliferation correlate with the production of sCD25.
Levels of sCD25 produced during the in vitro suppression assay were plotted versus the amount of proliferation detected at the 96 h time point. Under both (A) SFM and (B) 1.0% serum conditions, for all ratios of Treg to Teff cells (1∶0, 1∶1, ½∶1, and 0∶1, n = 7), the levels of sCD25 detected (x-axis) correlate with cellular proliferation as assessed by the incorporation of 3 H-Thymidine (y-axis). Isolated Treg and Teff cells were then assessed for their capacity to proliferate and produce sCD25 in an autocrine fashion over a 5 day time period. Purified CD4 + CD25 + (open bars) and CD4 + CD25 − T cells (closed bars) were cultured under expansion conditions utilizing beads coated with anti-CD3 and anti-CD28 as well as exogenous human recombinant IL-2 (300 U/ml) under serum-free conditions. Shown are (C) proliferation and (D) sCD25 production at 48, 72, 96, and 120 h time points by each indicated cell population. The data plotted represent the mean CPM and pooled sCD25 levels of replicate cultures ( n = 6 individuals) with situations identifying significance between Treg and Teff cells indicated (* p <0.05, ** p <0.01, and *** p <0.001).
In order to determine the cellular source of sCD25, freshly isolated CD4 + CD25 − and CD4 + CD25 + T cells were stimulated individually in the presence of anti-CD3 and anti-CD28 coated beads and exogenous IL-2. This analysis revealed that sCD25 can be generated in an autocrine fashion from both Treg and Teff cells following activation in the absence of any other cell types ( Figure 4 ). We found it intriguing that the culture conditions which abrogate the anergic and suppressive properties of Treg (i.e., by provision of cross-linking anti-CD3 and anti-CD28 co-stimulation, as well as exogenous IL-2) represent the same conditions that elicit the production of sCD25 from purified Treg. It should once again be noted, however, that both proliferation and sCD25 production were lower in Treg cells during expansion cultures when compared to Teff cells ( Figure 4C and 4D ). This is particularly apparent at the later time points of 96 and 120 h, when Treg likely have consumed exogenous IL-2 ( Figure 4D ). We would also note that production of sCD25 appears specific for the growth factor present (herein IL-2), as production of sCD127 (the IL-7R alpha-chain) was detected near background levels following robust in vitro expansion (data not shown).
Finally, given a previous report suggesting MMP-2 and MMP-9 could elicit CD25 cleavage [15] , we analyzed the capacity of these two proteases to modify Treg and Teff cell responses. Consistent with that report, we noted that addition of MMP-2 or MMP-9 in vitro reduced Treg ( Figure 5A , p <0.05 vs. vehicle) and to an even greater degree, Teff cell proliferation under anti-CD3/anti-CD28 stimulation ( Figure 5B , p <0.05 vs. vehicle). Interestingly, addition of a selective MMP-2/9 chemical inhibitor trended towards an increase in Teff proliferation ( Figure 5B , albeit p = NS), whereas Treg proliferation was significantly reduced by addition of this inhibitor ( Figure 5A , p <0.01). Consistent with this notion, addition of the MMP-2/9 inhibitor to the standard in vitro suppression assay increased the proliferative capacity of Teff, while augmenting the suppressive function of Treg ( Figure 5C ).
10.1371/journal.pone.0007980.g005 Figure 5
Alteration of MMP-2 and MMP-9 activity alters Treg and Teff cell proliferation and in vitro suppression.
FACS isolated Treg (A) and Teff (B) were activated with anti-CD3 and anti-CD28 coated microbeads in the presence of vehicle control or the selective MMP-2/9 chemical inhibitor II (IC 50 = 250 nM). Addition of active recombinant MMP-2 and MMP-9 (1 µg/ml each) led to reduced proliferation of Treg (A) and significantly reduced proliferation of Teff (B) cells. The inhibitor reduced proliferation of Tregs (p<0.01) while having no effect on Teffs. (C) The addition of MMP-2/9 inhibitor to the in vitro suppression assay resulted in increased Teff proliferation while at the same time, augmenting Treg-mediated suppression. Percent (%) suppression at 1∶1 (Treg/Teff) and ½∶1 ratios are noted above the bars, as a comparison to vehicle (0∶1) or MMP-2/9 inhibitor (0∶1) CPM. Graphs show one representative experiment of three individual experiments. (* p <0.05, ** p <0.01, and *** p <0.001).
Treg Maintain the Capacity to Suppress Teff Cell Cytokine Production in Serum Free Media Conditions
In addition to suppressing proliferation, one of the hallmark phenotypes of Treg is their capacity to suppress cytokine production by Teff cells [25] . In light of the findings of deficient suppression of proliferation by Treg in SFM conditions, we questioned whether Treg were also deficient in their capacity to suppress cytokine production by Teff cells. To test this notion, supernatants were collected and analyzed for the production of a panel of 22 cytokines by multiplex analysis (complete cytokine results are contained in Table S1 ).
Surprisingly, Treg maintained their capacity to suppress Teff cell cytokine production under both SFM and serum supplemented conditions for a majority of cytokines analyzed. However, some cytokines did exhibit increased production in co-culture under SFM conditions, paralleling the proliferation results. In this case, it is of interest to note that these responses tended to include cytokines classically associated with a T R 1 or T H 2-associated phenotype (e.g., IL-10 and IL-5). It is also worth noting that despite the highest levels of proliferation in the Treg and Teff co-culture condition (1∶1 in SFM), the highest levels of IL-2 were still detected from Teff cells alone (mean ± SEM, 11.7±10.2 pg/ml for the 1∶1 Treg to Teff cell condition vs. 129.6+155.5 for Teff only). On the other hand, despite dramatically higher Teff cell proliferation in the presence of serum ( Figure 1C , 0∶1 conditions), little detectable IL-2 was observed at the 96 h time point in supernatants.
Exogenous sCD25 Does Not Affect Suppression, but Alters Proliferation of Teff Cells
We sought to determine the effect of sCD25 supplemented at “serum” concentrations on Treg function in the suppression assay ( Figure 6A ). The addition of sCD25 did not affect suppression, but did result in decreased proliferation in Teff cells alone (0 ng/ml = 5865±2410 CPM vs. 2 ng/ml = 1656±143.9, p = NS). These results conflicted with what has been reported in the literature [12] .
10.1371/journal.pone.0007980.g006 Figure 6
Addition of soluble CD25 differentially effects proliferation of Teff cells and PBMC.
A suppression assay was set up under serum-free media ((A), open bars), or in the presence of a human recombinant sCD25 protein ((A), closed bars). Addition of sCD25 did not significantly restore Teff cell responses or the anergic properties of Treg in serum-free conditions ( p = NS for all Treg to Teff cell ratios, 1∶0, 1∶1, ½∶1, and 0∶1). In the 0∶1 condition, sCD25 did appear to reduce proliferation. sCD25 was added to PBMC cultures (n = 10) in the presence of increasing concentrations of sCD25, with anti-CD3/28 microbeads ((B), open bars) or PHA ((B), hatched bars) stimulation. Proliferation increased in some culture conditions and decreased in others. Statistically significant increases were seen by ANOVA under 3/28 stimulation in RPMI ( p = 0.05), AIMV ( p = 0.001), and XVIVO ( p = 0.001) media, and under PHA stimulation in XVIVO medium ( p = 0.01) when comparing 0 and 20 ng of sCD25 added. Comparisons among the same stimulation showed that cells cultured in RPMI and AIMV media had significant differences in proliferation regardless of the amount of sCD25 added, with both PHA and CD3/28 stimulation. Under PHA stimulation, CTL cultures proliferated less than either XVIVO or AIMV cultures. To determine if this differential proliferation was due to altered cell death, we measured the amount of nuclear matrix proteins (NMPs) released into the supernatant by dying cells (C). Addition of sCD25 did not result in a dose-responsive increase in cell death. The outcome varied depending on the culture media and stimulus. Plotting the stimulation index versus the production of NMP showed a negative correlation between proliferation and cell death under CD3/28 stimulation ((D), bottom panel). * p <0.05, ** p <0.01, and *** p <0.001.
Exogenous sCD25 has Variable Effects on PBMC Proliferation and is Dependent on the Culture Milieu
Our previous data reflecting the influence of serum presence along with the trend seen in ( Figure 6A ) led us to investigate the effects of multiple culture conditions and stimuli on production and function of sCD25. To reflect native immune conditions rather than isolated effects, we used total PBMC (n = 10) and direct TCR stimulus (anti-CD3 and anti-CD28 coated beads) as well as a mitogenic T cell stimulus (PHA-L). We chose four media formulations to compare the results between these assays - standard conditions with RPMI 1640 with 5% human AB serum, as well as three different serum free media - AIMV, XVIVO, and CTL Test media. sCD25 was added at 0, 2, and 20 ng/ml to approximate the levels present in serum and those that may persist at areas of inflammation. With direct TCR stimulus ( Figure 6B , hatched bars), the high dose of sCD25 significantly increased proliferation as compared to the control by ANOVA in RPMI (0 ng/ml = 70.85±6.894 vs. 20 ng/ml = 112.2±6.55, p <0.05), AIMV (0 ng/ml = 144.7±8.695 vs. 20 ng/ml = 206.6±19.33, p <0.001) and XVIVO media (0 ng/ml = 114.4±6.453 vs. 20 ng/ml = 174.4±16.33, p <0.01). Under mitogenic stimulus ( Fig. 6B , open bars), only cultures in XVIVO medium exhibited increased proliferation (0 ng/ml = 164.1±13.39 vs. 20 ng/ml = 231.8±26.55, p <0.001) while showing a decreasing trend in CTL serum free medium ( Figure 6B ). These comparisons indicate that substrate as well as matrix constituents in media lead to significantly different proliferation.
Exogenous sCD25 May Increase Survival in PBMC Cultures
To determine if this proliferation difference was due to differential cell death in these cultures, nuclear matrix protein 41/7 (NMP 41/7) was measured in supernatant by ELISA ( Figure 6C ). Increasing concentrations of sCD25 alone had no affect on cell death by ANOVA ( p = NS for all comparisons). Comparisons following stimulation in the various culture medium revealed that cell death could be altered by both factors: stimulus (AIMV: 0 ng/ml = 738.1±47.9 (with PHA) vs. 301.1±49.33 (3/28), p <0.01; 2 ng/ml = 832.2±151.8 (PHA) vs. 313.6±42.36 (3/28), p <0.01; 20 ng/ml = 824.9±136.7 (PHA) vs. 329±53.41 (3/28); ANOVA) and media formulation only with mitogenic stimulus (PHA) (AIMV = 798.4±30.24 vs. RPMI = 315.2±8.172 and vs. XVIVO = 378.2±11.64, both p <0.05; ANOVA). The stimulation index correlated with cell death only with CD3/28 stimulation ( Figure 6D , r = −0.3371, p = 0.0334). Interestingly, in comparisons between media formulations, there were both positive and negative correlations that were statistically significant. Cells cultured in AIMV and XVIVO media showed decreased proliferation with increased cell death, whereas PHA stimulation in CTL medium resulted in increased proliferation with increased death (data not shown). This suggests that sCD25 at high concentrations in local microenvironments may protect T cells from AICD, thus allowing increased viability and proliferation.
Production of sCD25 by PBMC Correlates with Cell Death
The differing proliferation results in PBMC cultures as compared to isolated T cell cultures led us to investigate the production of sCD25 as a function of PBMC proliferation ( Figure 7 ). sCD25 levels measured in cultures with no exogenous sCD25 ( Figure 7A ) varied in response to different stimuli in both RPMI and AIMV medium (RPMI: PHA = 18952±2872 vs. 3/28 = 7457±916.8, p<0.001; AIMV: PHA = 21482±2114 vs. 3/28 = 10957±893.8, p <0.001). Comparison to proliferation showed that the correlation seen earlier was absent in PBMC ( Figure 7B , p = NS by linear regression and Spearman correlation). Instead, production of sCD25 correlated with cell death ( Figure 7C , PHA: r = 0.7182, p <0.001, CD3/28: r = 0.3882, p = 0.0133).
10.1371/journal.pone.0007980.g007 Figure 7
sCD25 production in the absence of exogenous recombinant protein correlates with cell death, not proliferation, of PBMC.
To determine if the variable results seen previously were present in other areas, we measured the production of sCD25 in the 0 ng/ml condition with anti-CD3/28 microbeads ((A), open bars) or PHA ((A), hatched bars) stimulation. sCD25 production varied dependent on the stimulus provided in RPMI and AIMV media conditions (RPMI: (PHA) 18952±2872, (3/28) 7457±916.8, p<0.001; AIMV: 21482±2114, (3/28) 10957±893.8, p<0.001), but did not significantly differ between media formulations. In (B), sCD25 production did not correlate with proliferation as seen previously. Rather, sCD25 was generated at levels correlating with cell death as measured by NMP production, (C) (PHA: r = 0.7182, p <0.001, CD3/28: r = 0.3882, p = 0.0133). ***p<0.001.
Discussion
In this report, we describe a critical facet influencing T lymphocyte function; that being the differential control of membrane and soluble forms of CD25 on T cell subsets. We observed that CD25, which has commonly been used to select functional Treg, remains stable on the surface of anergic and suppressive Treg during the in vitro suppression assay. On the other hand, CD25 is upregulated on recently activated Teff cells, but is then subsequently released into the culture medium. These findings imply that Treg may obtain quantitatively and qualitatively greater levels of IL-2 signaling compared to Teff cells as a result of increased membrane CD25 stability and affinity for IL-2. These data also suggest that IL-2 production by Teff cells is required to drive continued cell cycle progression and sCD25 production by both cell populations.
In an effort to isolate the production of sCD25 from the levels added through serum supplementation, we conducted these experiments in a complete SFM formulation. To our surprise, the absence of serum impaired the in vitro suppression of proliferation assessed by standard 3 H-thymidine-based assays but not Teff cell cytokine production. In terms of developing an explanation for this phenomenon, we proposed three non-mutually exclusive explanations. First, Tregs may be more sensitive to a lack of serum components during co-culture. Second, Tregs may be contaminated with non-regulatory T cell populations that are relatively serum independent. Third, Tregs lose their anergic and suppressive properties under serum free conditions. In testing these three potential explanations we noted that the degree of in vitro suppression simultaneously takes into account the responses of Treg and Teff cells. Here, we observed functional deficiencies in both populations under SFM conditions, with a marked hyporesponsiveness of Teff cells and a concomitant failure by Treg to remain anergic and suppress proliferation. Despite deficient responses of both populations during in vitro suppression, cell tracking dye studies and markers of apoptosis/necrosis did not indicate significant differences in cell death due to the absence of serum. The addition of increased sorting stringency by inclusion of CD127 did restore some of the suppressive capacity of Tregs under serum free conditions, implying some contribution by non-Tregs when only CD4 and CD25 were used to sort. However, cell dilution dye assays tracking proliferation of Tregs and Teff cells during suppression indicated increased cell cycling from the entire population of Tregs in SFM making outgrowth of a small subset during the culture an unlikely explanation for the increased proliferation observed. In sum, the in vitro suppression assay is highly sensitive to potential artifacts due to media components when assessing proliferation alone and may provide only limited insight into the potential in vivo function of Tregs.
In terms of which component(s) within serum are required for full T cell activation as well as suppression by Treg, several candidates are the focus of our ongoing investigations. One such candidate, TGF-β, serves a wide variety of pleiotropic functions, with a role for TGF-β in immune regulation forming the subject of much recent debate. TGF-β is a major cytokine product of Treg and has also been reported to exist in an active form on the surface of these cells [26] . Furthermore, TGF-β can elicit the conversion of CD4 + CD25 − T cells into FOXP3 + T cells [27] . The function of Treg has been reported to be influenced by both the presence and activation state of TGF-β (i.e., in serum, TGF-β often exists in an inactive latent form (TGF-LAP)). In a related concept, the balance of serum proteases and protease inhibitors represent additional components which may serve dual roles by influencing the cleavage of CD25, as well as through their capacity to modify the activation state of TGF-β [15] , [28] . To the existing proteases reported to cleave CD25 [15] , [16] , here we add our findings that addition of MMP-9 or MMP-2 has effects on both Treg and Tconv cell responses. In addition, our data suggest that chemical inhibition of specific proteases may also influence CD25 stability and immune responsiveness. Controlling the proteolytic release of CD25 from the surface of Treg or Teff cells may provide novel therapeutic targets to alter IL-2 receptor signaling.
These findings raise the intriguing question of why CD25 appears to be more stable on the surface of Treg compared to Teff cells following stimulation and how this observation influences the downstream function of these cells. The IL-2 signaling axis can elicit proliferation of T cells, but also contains at least two mechanisms for controlling T cell expansion in response to antigen. IL-2/CD25 interactions reinforce the metabolic fitness of Treg, leading to dominant immune suppression [29] . In Teff cells, excess IL-2 signaling can lead to AICD. We would speculate that the CD25 shedding from freshly activated Teff cells observed in these studies may be necessary to prevent these processes, allowing for continued cycling as well as maintaining activation requirements in daughter cell populations.
If a stable IL-2 signal reinforces FOXP3 expression and Treg function, one might expect situations that interrupt IL-2 signaling to interfere with Treg activity. An example of this situation was reported by Tang and colleagues who observed that intra-islet Treg cells expressed reduced amounts of CD25 and Bcl-2, which led to a functional imbalance of Treg and Teff cell populations leading to the development of type 1 diabetes in non-obese diabetic mice [30] . In addition, multiple reports in inflammatory autoimmune diseases and studies of human T-cell leukemia virus type-I (HTLV-1) associated lymphoma report elevated serum levels of sCD25 and defective suppression by Treg [31] – [33] . Therefore, it appears consistent through multiple disease processes that disturbances in the IL-2 and CD25 signaling axis are often paralleled by defective immunoregulation by Treg. Some of the factors which may influence CD25 stability, and subsequent immune reactivity, are modeled in Figure S2 .
As mentioned previously, at least two major hypotheses on the formation of sCD25 exist. Either Treg express a membrane-stable form of CD25, or Treg cultures fail to elicit the factors that lead to the proteolytic cleavage of CD25. Tregs are capable of producing sCD25 upon activation and IL-2 exposure suggesting that production of IL-2 by Teff cells may account for their continued sCD25 production and proliferation. These observations taken during the in vitro suppression assay appear to indicate that not only are adaptive cytokine levels important for T cell proliferation and function but in addition, the stability of cytokine receptors warrants further consideration.
The influence of serum seen in suppression assays indicated that a more thorough investigation of culture conditions was required to accurately compare inter-laboratory data. Therefore, PBMC responses were measured under TCR or mitogenic stimulation under various culture conditions. What we have described is that the presence or absence of serum as well as the components of the culture media influenced the data collected in terms of sCD25 production, proliferation and cell death. Overall, the addition of exogenous sCD25 increased proliferation in three out of four culture conditions. The results of these assays are difficult to interpret, in that slight modifications in the in vitro environment have an effect on multiple outcome measures. This may provide an explanation for the variation between laboratory results, but likely does not complete the story. Indeed, the presence of sCD25 appears to have differential effects on pre-activated naive T cells versus whole PBMC.
These studies highlight the importance of the soluble and membrane associated forms of CD25 in terms of influencing cell proliferation, survival, and suppression via Treg. The extensive variability between investigations as well as various culture conditions highlights the need for increased standardization during in vitro assays. This is also contingent upon a more complete understanding of the in vivo microenvironments controlling immune responses. In terms of future implications, these studies highlight important facets to consider in diseases associated with abnormal levels of sCD25 as well as defective regulation by Treg.
Limitations
The role of sCD25 remains obscure because of the variations in culture conditions, and ultimately will have to be resolved in vivo by knock-in of only the soluble form into mice. The observation that sCD25 may promote survival is an intriguing possibility, one that might have a basis in the stoichiometric competition for other receptor subunits in the common gamma chain cytokine family, where cleaved CD25 allows for a more homeostatic cytokine (i.e., IL-7, IL-15) signaling. We would argue that the relatively abundant amount of sCD25 in serum would not simply be waste and this is the subject of ongoing investigation.
What we can infer from these experiments though is that sCD25 production on the whole is more likely associated with activation induced cell death than proliferation per se. In light of this, the multiple proteases linked with apoptotic cell death may be involved in releasing CD25 from the cell surface. Chief among these candidates are caspases, which we assume would have access to CD25 upon internalization of the IL-2 receptor. This is particularly pertinent considering recent studies linking disease-associated polymorphisms in CD25 with increased susceptibility to type 1 diabetes and multiple sclerosis [9] , [11] . In the case of type 1 diabetes, we reported susceptible alleles in type 1 diabetes were linked with lower serum levels of sCD25 [9] . In addition, in the non-obese mouse model of type 1 diabetes, allelic variation of IL-2 was associated with reduced IL-2 expression correlating with impaired Treg activity [34] . Thus, defects within the IL-2/CD25 signaling axis may constitute a conserved immune/AICD defect predisposing to autoimmunity.
This study also draws attention to the necessity for standardization of in vitro assays to allow for proper interpretation and comparison of results. Proprietary media formulations in combination with batch and vendor variation in serum supplements hamper in vitro comparisons. A recent publication regarding difficulties in in vitro polarization of Th17 cells due to media composition shows a problem that is understood but not fully explored – that immune cell activity in vivo is highly dependent on the location [35] . Thus, it may be necessary to define in vitro conditions that are more truly reflective of the many environments where immune cells perform their functions.
Supporting Information
Table S1
(0.07 MB PDF)
Figure S1
Isolation of Treg and T conventional cells from human peripheral blood by FACS. Representative plots showing (A) pre-sort expression profiles of CD4 (x-axis) and CD25 (y-axis, left plots) or CD25 (y-axis) and CD127 (x-axis, middle plots) following CD4 negative selection. Plots represent post-sort purity of sorted cells indicating expression of CD4, CD25, CD127, and FOXP3 for (B) T conventional cells and (C) Tregs. Post-sort FOXP3 analysis was assessed immediately following FACS isolation (right plots, black overlays represent isotype control and blue overlays indicate FOXP3 staining, respectively).
(0.77 MB TIF)
Figure S2
Membrane CD25 stability influences immune responsiveness. A hypothetical model is shown depicting how alterations in membrane-bound and soluble forms of CD25 (sCD25) may influence the activities of Treg and conventional T (Tconv) cell subsets. The immunological microenvironment provides costimulatory and cytokine signals through APC which influence CD25 stability on Treg and Tconv cells. Both intrinsic signals, through TCR signal strength and costimulation, as well as extrinsic inflammatory signals such as TOLL ligand induced proinflammatory cytokines may skew T cell phenotypes through the production of proteases and subsequent cleavage of membrane-bound CD25. Loss of membrane CD25 and IL-2 signaling in Treg would result in reduced phosphorylation of STAT5 and downstream FOXP3 expression. Conversely, in Tconv cell populations IL-2 signal strength along with costimulatory and cytokine differentiation signals would influence the lineage fate of responding T cell populations. Loss of CD25/IL-2 high affinity signaling from Teff cell populations would result in lower IL-2 affinity and/or dependence on signaling from other common γ-chain receptors for survival (such as IL-7, IL-15, and IL-21). Conversely, Tconv cells activated under tolerogenic conditions would lead to increased CD25 membrane stability and subsequent FOXP3 expression attenuating effector responses (Th17, Th1, and Th2) and reinforcing an induced Treg (iTreg) phenotype. The signal strength as measured by both affinity and avidity for IL-2 (related to the alpha beta gamma combinations) is important in driving proliferation but also AICD.
(0.53 MB TIF)
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Protein unfolding induced by chemical denaturants such as urea and guanidine hydrochloride (GdnHCl) is a common approach to study protein folding in vitro [1] . Meanwhile, it has been shown that low concentrations of GdnHCl can cause protein stabilization by eliminating the strains in protein caused by the electrostatic interactions of charged groups on its surface [2] , [3] . Our results also prove it. At low concentration of GdnHCl (around 0.1 M) we have recorded small changes in fluorescence spectrum position and tryptophan fluorescence anisotropy for creatine kinase [4] , carbonic anhydrase II (CA II) [5] and increase in chromophore fluorescence intensity of EGFP, DsRed1 and their mutant forms (0.5 M of GdnHCl) [6] .
In this work, we have shown that along with the stabilizing effect on protein structure, small amounts of GdnHCl can also cause protein aggregation. The account of this effect clarifies why the transition of some proteins (e.g. α-lactalbumine and CA II) into intermediate state like molten globule is accompanied by blue shift of fluorescence spectrum, increase in anisotropy of intrinsic fluorescence, and increase in fluorescence of hydrophobic dye 1-anilinonaphthalene-8-sulfonic acid (ANS) when denaturation is caused by GdnHCl but not by urea [5] , [7] , [8] . The increase in ANS fluorescence in the narrow range of GdnHCl concentration for urease is also connected with protein aggregation [9] , [10] . For the first time, protein aggregation in the solution of low concentration of GdnHCl we have revealed for actin. In this case the effect is especially pronounced because in aggregation are involved large supramolecular complexes of inactivated actin [11] , [12] and it is accompanied not only by the increase in ANS fluorescence intensity, but by the increase in light scattering also.
Results and Discussion
Aggregation and changes of inactivated actin surface characteristics induced by GdnHCl
The study of actin unfolding – refolding showed that actin denaturation results in the formation of the so-called inactivated actin (I), which represents an ordered aggregate (supramolecular monodisperse complex of 14–16 monomers of partially unfolded actin molecules) with hydrophobic clusters on the surface [11] , [12] . The fluorescence intensity of ANS (5⋅10 −5 M) in the presence of inactivated actin (0.15 mg/ml) is around 20 times greater than in the presence of native actin at the same concentration. This can be explained not only by the existence of hydrophobic clusters on the surface of inactivated actin, but also by the appearance of “hydrophobic pockets” between actin molecules in partially folded state forming inactivated actin. Inactivated actin characteristics are independent of the way of its formation [11] , [13] . We decided to examine the characteristics of inactivated actin in solutions of different concentrations of urea and GdnHCl. As it can be expected, the intensity of ANS fluorescence weakly depends on urea concentration up to the concentration when supramolecular complexes are destroyed. However, the dependence of ANS fluorescence on the concentration of GdnHCl was found to be the curve with maximum in the narrow range of small concentrations of denaturant ( Figure 1 ). Furthermore, in the same range of GdnHCl concentrations maximum of light scattering (or even precipitation at high protein concentration) was observed. This means that in this narrow range of GdnHCl concentrations inactivated actin forms large aggregates, and that ANS molecules affinity to these aggregates is very high. ANS incorporates into the hydrophobic pockets between the molecules forming aggregates that result in the dramatic increase in its fluorescence intensity.
10.1371/journal.pone.0015035.g001
Figure 1
Actin aggregation induced by GdnHCl in low concentrations.
The dependence of ANS fluorescence intensity ( A ) and of light scattering ( B ) of the solutions of inactivated actin (red) and initially native actin after 10 min and 24 h of incubation in a solution of denaturants at the appropriate concentration (blue and green) on GdnHCl (closed symbols) and urea (open symbols) concentration. Insert in panel A . Scheme of actin denaturation and aggregation (N, U * and I are native, essential unfolded and inactivated actin [11] , I ag is aggregates of inactivated actin, the details are given in the text). Insert in panel B . The dependence of the total macromolecule charge of actin on the pH of a solution calculated on the basis of protein amino acid content [24] . The protein concentration was 0.15 mg/ml, ANS concentration was 5⋅10 −5 M.
We explain protein aggregation by the interactions between the GdnHCl cations (GuH+) and the side chain C = O group of the glutamic acids and glutamine, aspartic acid and asparagine amino acid residues of the molecule. The possibility of such interactions has been shown earlier [14] , [15] . In actin, the number of negatively-charged groups from glutamic and aspartic acids (OD2 - 22 groups and OE2 - 28 groups) is greater than that of the positively-charged groups from lysine (NZ – 18 groups), arginine (NH1 – 18 groups) and histidine (NE2 – 9 groups). Therefore, the actin molecule is negatively charged (pI 5.07, see Figure 1A , Insert) at a neutral pH. With an increase in the number of GuH+ ions bound to inactivated actin, the number of positively-charged groups increases, and at some concentration of GdnHCl (0.2–0.3 M), the initially negatively-charged molecules become neutral, which leads to their aggregation. Upon the further increase in GdnHCl concentration, the number of positively-charged groups on the surface of the protein molecules will exceed the number of negatively-charged groups. Therefore, the conditions will no longer be favorable for aggregation. This is the reason for the abrupt decrease in light scattering intensity. The less abrupt decrease in the intensity of ANS fluorescence in comparison with light scattering with the increase in GdnHCl concentration can be explained by the higher affinity of negatively-charged ANS molecules with inactivated actin when it is positively charged, though aggregates are already destroyed.
We have ascertained that native proteins with a pI values at acidic pH, and native actin in particular, do not aggregate at low concentrations of GdnHCl. Due to complex process of actin denaturation and the dependence of the transitions rates upon GdnHCl concentration [11] , [12] (see also, Figure 1A , Insert) maximum of ANS fluorescence intensity shifts to lower concentration of GdnHCl with the increase in incubation time. Thus after 24 h of incubation maximum of light scattering and intensity of ANS fluorescence intensity were recorded practically at the same concentrations of GdnHCl as for inactivated actin.
Hydrophobic interactions apparently play a significant role in both inactivated actin formation and in the formation of inactivated actin aggregates in the presence of low concentrations of GdnHCl. As mentioned above, due to existence of hydrophobic pockets in inactivated actin ANS fluorescence intensity in the presence of inactivated actin is 20 times greater than in the presence of native actin. Inactivated actin already has hydrophobic clusters on its surface, but molecules of inactivated actin do not “stick together” because of the negative charges on their surfaces, which prevent this process. At low concentrations of GdnHCl the aggregation of inactivated actin leads to the significant increase in the number of hydrophobic pockets and consequently to the increase in the number of bound ANS molecules, that is recorded by increase in ANS fluorescence. Protein aggregation in the solution of low concentration of GdnHCl is especially pronounced for actin, because in this case large supramolecular complexes of inactivated actin [11] , [12] are involved in aggregation.
Intermediate states in the pathway of CA II unfolding induced by urea and GdnHCl: molten globule and aggregates of molten globules
Taking into account the aggregating effect of GdnHCl it became clear why the transition of some proteins into intermediate state like molten globule is accompanied by blue shift of fluorescence spectrum, increase in anisotropy of intrinsic fluorescence, and increase in fluorescence of ANS when protein denaturation is caused by GdnHCl but not by urea [5] , [7] , [8] , [9] , [10] . This exactly was observed for CA II [5] . Now it is clear that CA II unfolding is two stage processes both induced by GdnHCl and urea. The appearance of intermediate state of CA II in the pathway of unfolding induced by urea agrees with the results of several works which are reviewed in [16] . The increase of ANS fluorescence intensity is recorded only in GdnHCl solution ( Figure 2 ), because the dye interacts with aggregates of CA II intermediate state. Aggregates of CA II intermediate state are comparative small therefore their formation is seen by the increase of intrinsic fluorescence polarization ( Figure 2 ), but not light scattering.
10.1371/journal.pone.0015035.g002
Figure 2
Denaturation of CA II induced by GdnHCl (red symbols) and urea (blue symbols).
Panels A and B represent the changes in anisotropy of intrinsic fluorescence and ANS fluorescence intensity, respectively. The protein concentration was 0.15 mg/ml, ANS concentration was 5⋅10 −5 M, pH 7.5.
In conclusion, this work proposes new view on three important points of protein folding: (i) strong chemical denaturant GdnHCl in narrow range of small concentrations, in contrast to urea, can cause aggregation of some proteins in molten globule state; (ii) hydrophobic dye ANS binds with the aggregates of proteins in the molten globule state rather than with the hydrophobic clusters on the surface of a protein in the molten globule state, as was commonly accepted [1] , [17] ; (iii) for some proteins, what was previously believed to be the molten globule state in the pathway of protein denaturation by GdnHCl in reality represents the aggregates of protein molecules in this state.
Materials and Methods
Rabbit skeletal muscle actin was purified by one or two cycles of polymerization-depolymerization [18] . The native state of actin was checked by its fluorescence spectrum position characterized by parameter A = ( I 320 / I 365 ) 297 , where I 320 and I 365 are fluorescence intensities at λ em = 320 and 365 nm, respectively (λ ex = 297 nm). Actin samples had a parameter A value ≥2.53, which corresponds to an inactivated actin content of <4% [19] . The molar extinction coefficient for actin was taken as E 280 = 1.09 (mg/ml) −1 cm −1 [20] . The final actin concentration varied from 0.1 to 0.44 mg/ml. CA II from bovine erythrocyte was purchased from Sigma (USA) and used without further purification. The extinction coefficient for CA II was taken as E 280 = 1.87 (mg/ml) −1 cm −1 [7] . The protein concentration varied between 0.05 and 0.5 mg/ml. GdnHCl (Nacalai Tesque, Japan), urea, and ANS (Sigma, USA) were used without additional purification. The concentration of GdnHCl and urea were determined by the refractive index [21] with an Abbe refractometer (LOMO, Russia). The extinction coefficient of ANS was taken as ε 350 = 5000 M −1 cm −1 [22] . The concentrations of proteins and ANS were determined using a spectrophotometer U-3900H (Hitachi, Japan).
Fluorescence experiments were carried out using a Cary Eclipse spectrofluorimeter (Varian, Australia) and a homemade spectrofluorimeter for registration of fluorescence polarization [23] . Protein intrinsic fluorescence was excited at the long-wavelength edge of absorption spectrum (λ ex = 297 nm) where the contribution of tyrosine residues to the bulk protein fluorescence is negligible. The position and form of the fluorescence spectra were characterized by parameter A [19] , [23] . The values of parameter A and of the fluorescence spectra were corrected by the instrument spectral sensitivity. The fluorescence intensity of the hydrophobic dye ANS was detected at 480 nm (λ ex = 365 nm). The anisotropy of tryptophan fluorescence was calculated by the equation , where and are the vertical and horizontal components of the fluorescence intensity excited by vertically polarized light and G is the relation of vertical and horizontal components of fluorescence intensity excited by horizontally polarized light ( ) [23] . The intensity of light scattering was detected by the spectrofluorimeter when λ reg = λ ex . In the majority of experiments the light scattering was recorded at 297 or 350 nm. The choice of the wavelength did not influence the result.
|
Vytvor v slovenčine zhrnutie totho článku. Laicky vysvetli všetky základné pojmy v článku. Ak používaš úvodzovky, nezabudni, že v slovenčine sa prvé úvodzovky píšu dole. Využi 8000 až 8100 tokenov. Tu je text článku: |
Introduction
Tobacco smoking harms nearly every organ in the body and is a leading cause of preventable morbidity and premature mortality [ 1 ]. Individuals who smoke lose at least a decade of life expectancy compared to those who have never smoked [ 2 ]; major causes of this excess mortality include cancer as well as vascular and respiratory disease [ 1 , 2 ]. Quitting smoking greatly reduces the risk of developing smoking-related diseases including cardiovascular disease, several types of cancer, and chronic obstructive pulmonary disease (COPD) [ 3 ]. The extent of risk reduction, and timeline over which it occurs, varies between diseases. Risk of cardiovascular events is reduced substantially within five years, and after 15 years of quitting the risk is close to that of individuals who have never smoked [ 4 , 5 ]. Approximately 10–15 years after smoking cessation, lung cancer risk decreases to half that of those who continue to smoke and continues to decline as time since cessation increases [ 3 ]. While smoking cessation can prevent the development of COPD and attenuate disease progression [ 6 ], lost lung function already present at the time of smoking cessation is not fully recovered [ 7 ]. In addition to cessation itself, reducing the number of cigarettes per day may be associated with some health benefits among those who continue to smoke. A meta-analysis of people who smoke heavily (followed from 5 to 40 years) found that those who reduced their smoking lowered their lung cancer risk, but not their risk of all cancers or all smoking-related cancers [ 8 ]. The risk of cardiovascular disease was also lowered among those who reduced from heavy to light smoking [ 9 ], but was not lowered by a 50% reduction in cigarettes per day [ 8 , 10 , 11 ].
Although smoking cessation has the greatest impact on reducing the health risks associated with smoking, it is important to also study the health benefits associated with receiving smoking cessation treatments in real-world settings [ 12 ], and not just smoking cessation. This is because, in part, with only 3–5% of untreated quit attempts achieving abstinence [ 13 ], interventions that result in 5–10% abstinence may be considered effective [ 14 ], and most patients presenting for treatment will not achieve sustained abstinence. Furthermore, the focus on abrupt abstinence is too narrow to encompass the breadth of treatment trajectories and outcomes that occur. Individuals seeking treatment may relapse and transition between smoking and non-smoking repeatedly, as many make several failed attempts to quit before they finally succeed [ 15 – 17 ]. In addition, some individuals presenting for treatment will set a goal to reduce, but not stop, smoking, or will adopt a reduce-to-quit strategy, whereby they gradually reduce cigarette consumption as a cessation strategy [ 18 – 21 ]. An awareness of the health outcomes associated with smoking cessation treatment can help guide healthcare decision makers and clinicians treating these patients, such as deciding whether and when to provide screening or preventive treatment for particular conditions.
Aside from studies conducted to detect adverse events associated with cessation medications (nicotine replacement therapy (NRT), varenicline, bupropion) [ 22 , 23 ], few studies have examined incident health outcomes following smoking cessation treatment. Where published, existing studies have examined cause-specific hospitalizations or re-hospitalizations. For example, among Massachusetts Medicaid enrolees, use of a smoking cessation pharmacotherapy insurance benefit was associated with a significant decrease in hospitalization for acute myocardial infarction and other acute coronary heart disease diagnoses, but no significant change was observed in hospitalizations for respiratory diagnoses [ 24 ]. In another study, re-hospitalization and mortality outcomes were examined among people who smoke and initiated smoking cessation treatment while in hospital; these patients experienced significantly lower rates of smoking-related readmissions compared to patients who smoke and received usual care [ 25 ]. The Lung Health Study compared the health outcomes of people with prevalent COPD who either received cessation treatment, consisting of NRT and behavioural support, or usual care [ 26 ]; after 5 years, there was no significant difference in lung cancer or hospitalizations for respiratory disease between the usual care and smoking cessation treatment groups, although there was greater smoking cessation and a slower decline in lung function among those who received smoking cessation treatment.
This dearth of research examining the incidence of chronic disease following smoking cessation treatment indicates a gap in the literature. Data collected during routine interactions with the healthcare system provides an opportunity to address this gap. Such data, from actual patients who have received smoking cessation treatment in typical clinical settings and conditions, can allow us to assess whether the healthcare outcomes we hope to achieve are ultimately being realized in the real world. Therefore, the objective of the current study was to compare the incidence of cancer, COPD, diabetes, hypertension, and major cardiovascular events during a 5-year follow-up period for people who smoke and had accessed a smoking cessation treatment program, versus people who smoke and had not accessed the program, in Ontario, Canada.
Materials and methods
Study design
We conducted a retrospective matched cohort study to compare the risk of developing chronic disease among individuals who had enrolled in the Smoking Treatment for Ontario Patients (STOP) program against matched control Ontarians who smoke but had not accessed the program. Incident chronic disease was determined through linkage with health administrative data. This study was approved by the Research Ethics Board of the Centre for Addiction and Mental Health (#110–2019) and adheres to the Reporting of Studies Conducted Using Observational Routinely-Collected Health Data (RECORD) guidelines (see S1 Checklist ) [ 27 ].
Smoking cessation treatment
The STOP program delivers smoking cessation treatment to patients at partnering healthcare organizations across the province of Ontario. Prior to receiving treatment, patients provide written informed consent. Patients are eligible to receive up to 26 weeks of NRT within a 12-month period, and behavioural counselling delivered by healthcare practitioners trained in smoking cessation interventions. Although treatment is tailored to individual need and can vary, a majority of patients receive a combination of transdermal patch plus a single form of short-acting NRT (e.g., gum, inhaler) [ 28 ]. Further description of the STOP program is provided elsewhere [ 29 ].
Matched cohort creation
In a parent study, we derived a treatment cohort who sought smoking cessation treatment via the STOP program and a matched control cohort who smoked but had not accessed the program (detailed selection criteria for participants that made up these cohorts can be found in the parent study or S1 Fig ) [ 29 ]. The parent study treatment cohort consisted of patients who had enrolled in the STOP program between 1 July 2011 and 31 December 2012. The matched control cohort was formed using the 2007/2008, 2009/2010, and 2011/2012 cycles of the Canadian Community Health Survey (CCHS), a cross-sectional population-based survey that collects self-reported health-related data including smoking behaviours; detailed survey methodology is reported elsewhere [ 30 ]. If a CCHS respondent appeared in multiple cycles, only data from their most recent survey was retained in order to better align with the enrolment timeframe for the treatment cohort. Index date was the date of enrolment in the STOP program for the treatment cohort and the date of CCHS survey completion for the control cohort. Each treated individual was matched to one control individual using a combination of hard matching (sex and age ± 2 years at index date) and propensity-score matching (using a greedy algorithm without replacement with a caliper width of 0.2 standard deviations of the estimated propensity score logit). The propensity score estimated the probability of treatment for each individual. This was done using multivariable logistic regression with the following baseline variables: age at index, education, household income, number of cigarettes smoked per day, age first started smoking, comorbidity burden (determined by Aggregated Diagnostic Groups [ 31 ]) and the rate of emergency department visits and hospitalizations in the two years prior to index date. Further details of the matching process used in the parent study are described elsewhere [ 29 ].
For the current study, we selected participants from the population of matched treatment-control cohorts that were derived previously. We identified five matched sub-cohorts from the parent study (one matched sub-cohort for each of the five chronic disease outcomes). These sub-cohorts were at risk of: (i) cancer, (ii) COPD, (iii) diabetes, (iv) hypertension, or (v) a major cardiovascular event. Individuals were deemed to be “at risk of” each outcome if there was no known record of having experienced the outcome. For each sub-cohort, individuals who had already experienced the disease outcome at index were excluded because they were no longer at risk of incident disease after index; we also excluded the individual with whom they had been matched in the parent study to preserve the 1:1 matched design. Thus, for each chronic disease, we retained previously matched treatment-control pairs who were both at risk for developing the disease outcome after index. For a flow chart illustrating derivation of the at-risk matched sub-cohorts, see S1 Fig .
Outcomes
The primary outcomes were incidence of cancer, COPD, diabetes, hypertension, and major cardiovascular events (i.e., acute myocardial infarction, stroke, percutaneous coronary intervention, coronary artery bypass graft, or death from ischaemic heart disease or cerebrovascular disease) from index date until 31 December 2017, loss of Ontario Health Insurance Plan eligibility (OHIP; this plan provides jurisdiction-wide health coverage), or death, whichever occurred first.
Data sources
Data from the STOP program and the CCHS were previously linked to health administrative data sets. All datasets were linked using unique encoded identifiers and analyzed at ICES [ 32 ]. ICES is an independent, non-profit research institute whose legal status under Ontario’s health information privacy law allows it to collect and analyze healthcare and demographic data, without consent, for health system evaluation and improvement.
STOP baseline assessment or CCHS survey
Descriptive characteristics pertaining to smoking history were self-reported in the STOP baseline assessment questionnaire or the CCHS: smoking frequency, number of cigarettes smoked per day, and age when first tried smoking. Duration of smoking was calculated by subtracting age when first tried smoking from age at index.
ICES datasets
Derivation of at-risk sub-cohorts, and ascertainment of chronic disease outcomes, were achieved using several datasets. These datasets were checked at index to identify people at risk of each health outcome, and over the follow-up period to ascertain if, and when, a health outcome had occurred. Diabetes, COPD, and hypertension were identified based on existing chronic condition datasets derived by ICES: the Ontario Diabetes Dataset [ 33 ], Ontario Chronic Obstructive Pulmonary Disease database [ 34 ], and Ontario Hypertension Dataset [ 35 ]. ICES applies validated algorithms to the healthcare administrative data of all Ontarians on a recurring basis to identify patients with these medical conditions. Cancer diagnoses were identified from the Ontario Cancer Registry [ 36 ], which collects data on Ontario residents newly diagnosed with cancer (except for basal cell carcinoma and squamous cell carcinoma of the skin) or who have died of cancer. Major cardiovascular events were defined as hospital admissions or emergency department (ED) visits for acute myocardial infarction, stroke, percutaneous coronary intervention, coronary artery bypass graft, or death from ischaemic heart disease or cerebrovascular disease [ 37 ]. These events were ascertained from the Canadian Institute for Health Information Discharge Abstract Database (DAD), National Ambulatory Care Reporting System (NACRS), and Office of the Registrar General Vital Statistics–Deaths database.
Additional datasets were used to identify baseline characteristics. Age at index, sex and postal code were obtained from the Registered Persons Database. Immigration category was obtained from the Immigration, Refugee and Citizenship Canada Permanent Resident database [ 38 ]. Postal code was linked to census data to obtain neighbourhood-level socioeconomic indicators (household income, employment and educational attainment quintiles) and the Rurality Index of Ontario (RIO 2008) [ 39 ] scores to assess rurality of residence. In addition to the chronic disease datasets used to define the at-risk cohorts and ascertain outcomes, the following datasets were used to assess comorbidities at baseline: asthma (Ontario Asthma Dataset [ 40 ]), congestive heart failure (Congestive Heart Failure Dataset [ 41 ]) and myocardial infarction (Ontario Myocardial Infarction Dataset [ 42 ]) at index. Comorbidity burden was determined using The John Hopkins ACG ® System (Version 10) Aggregated Diagnosis Groups (ADGs); scores were calculated using a two-year lookback window from index date and categorized into four groups (0, 5, 6–9, 10+), with higher scores indicating greater comorbidity burden [ 31 ]. Healthcare utilization in the two years prior to index date was ascertained using data on outpatient physician visits (for any reason) from the OHIP database, hospitalizations from the DAD and Ontario Mental Health Reporting System, and emergency department visits from the NACRS. Occurrence and date of any deaths were identified using the Registered Persons Database. Causes of death were ascertained from the Office of the Registrar General Vital Statistics–Deaths database. Further details about baseline measures and data sources are provided elsewhere [ 29 , 32 ].
Statistical analyses
Baseline characteristics for each of the matched sub-cohorts were described using frequencies and percentages for categorical measures and mean and standard deviation for continuous measures. Standardized mean differences (SMD) were computed to examine balance in the distributions of baseline characteristics between the treatment and control groups; an SMD > 0.1 was considered a meaningful imbalance.
Within each matched sub-cohort, the risk of chronic disease was estimated for treatment and control groups using the cumulative incidence function approach, where death (occurring prior to chronic disease) was treated as a competing event. Individuals whose observation terminated due to study end were considered right-censored at that time. Under this approach, the estimated 5-year risk of chronic disease was reported for treatment and control groups. Gray’s test was used to determine if the risk of chronic disease over time statistically differed between treatment and control groups.
All analyses were stratified by sex. As sex was a hard-matched variable, all stratified analyses could be done without breaking matched pairs. A p value less than 0.05 was considered statistically significant. There were no missing data. Analyses were conducted using SAS Enterprise Guide version 7.12 software.
Results
Cancer
The treatment and control groups in the matched sub-cohort at risk for cancer were well-balanced at index on all sociodemographic characteristics, including age, sex, education, employment, rurality/neighbourhood income and migrant status. Some imbalances remained on smoking characteristics, prevalent comorbidities and healthcare utilization in the two years prior to index. There was a higher proportion of those who smoked daily and a lower proportion who smoked occasionally in the treatment versus control group (both sexes). In addition, the number of outpatient visits and the proportion of those having at least one ED visit were also higher in the treatment group versus the matched control group (males only). The above was true for all five at-risk matched sub-cohorts.
The following additional imbalances were found between the treatment and control groups in the matched sub-cohort at risk for cancer. Three prevalent health conditions were more common in the treatment group versus control group: COPD (females: 29.5% vs. 19.8%; males: 26.5% vs. 17.3%), diabetes (males: 16.2% vs. 11.4%) and asthma (females: 24.3% vs. 19.0%). The proportion having made an outpatient visit two years prior to index was also higher in the treated versus controls (females: 96.6% vs. 93.7%; males: 93.9% vs. 83.7%). Baseline characteristics of the treatment and control groups in the matched sub-cohort at risk for cancer, stratified by sex, are described in S1 Table .
See Table 1 for the estimated incidence of cancer within 5 years of follow-up in the treatment and control groups, accounting for right-censoring and treating death as a competing risk. The cumulative incidence of cancer did not differ significantly between treatment and control groups, among both sexes (females: Gray’s test p = 0.84, see Fig 1A ; males: Gray’s test p = 0.84, see Fig 1B ).
10.1371/journal.pone.0288759.g001
Fig 1
Cumulative incidence of cancer among female (A) and male (B) smoking cessation treatment patients versus matched controls.
The follow-up period begins the day after enrolment in smoking cessation treatment (treatment cohort) or survey completion (control cohort) in 2011/2012 and ends December 31, 2017 (or date of death if it occurred first). Shaded areas indicate the 95% CI. Number of individuals at risk at each time point is presented below the x axis.
10.1371/journal.pone.0288759.t001
Table 1 Estimated risk of chronic disease among treated and matched control groups at 5 years post index, accounting for death as a competing event.
Females
Males
N
Cumulative incidence of disease outcome at 5 years (95% CI)
N
Cumulative incidence of disease outcome at 5 years (95% CI)
Cancer
Treatment
4,832
4.9% (4.3%–5.6%)
4,302
5.7% (5.0%–6.5%)
Control
4,832
5.2% (4.5%–5.8%)
4,302
5.2% (4.5%–5.9%)
COPD
Treatment
3,024
11.0% (9.9%–12.2%)
2,881
12.3% (11.0%–13.6%)
Control
3,024
10.7% (9.6%–11.8%)
2,881
9.1% (8.1%–10.2%)
Diabetes
Treatment
4,074
5.8% (5.1%–6.5%)
3,469
6.8% (6.0%–7.8%)
Control
4,074
4.3% (3.7%–5.0%)
3,469
4.7% (4.0%–5.4%)
Hypertension
Treatment
3,000
8.9% (7.9%–10.0%)
2,577
10.4% (9.1%–11.6%)
Control
3,000
7.9% (7.0%–8.9%)
2,577
10.0% (8.9%–11.2%)
Major CV events
Treatment
5,007
4.5% (3.9%–5.1%)
4,272
6.9% (6.2%–7.8%)
Control
5,007
4.5% (3.9%–5.1%)
4,272
6.4% (5.6%–7.1%)
Abbreviations: COPD = chronic obstructive pulmonary disease; CV = cardiovascular; IQR = interquartile range; SD = standard deviation.
COPD
In addition to the common imbalances described above (first paragraph of Results), there were further imbalances between the treatment and control groups in the matched sub-cohort at risk of COPD. The treatment group smoked more cigarettes per day than the control group (females: 15.48 vs. 14.42). Compared to the control group, the treatment group had a higher prevalence of diabetes at index (males: 12.6% vs. 8.4%) and a higher proportion had at least one outpatient visit (males: 92.1% vs. 80.8%). There was also a higher proportion with 0–5 ADG comorbidities in the treatment versus control group (females: 51.2% vs. 44.7%). Thus, findings suggest somewhat higher comorbidity and healthcare utilization among treated males versus control males at risk of COPD, but lower overall comorbidity burden among treated females versus control females. Baseline characteristics of the matched sub-cohort at risk of COPD, stratified by sex, are described in S2 Table .
See Table 1 for the estimated incidence of COPD within 5 years of follow-up in the treatment and control groups. While the cumulative incidence of COPD did not differ significantly between treated and control females (Gray’s test p = 0.63, see Fig 2A ), it was significantly higher over time for treated males (Gray’s test p < 0.001, see Fig 2B ).
10.1371/journal.pone.0288759.g002
Fig 2
Cumulative incidence of COPD among female (A) and male (B) smoking cessation treatment patients versus matched controls.
The follow-up period begins the day after enrolment in smoking cessation treatment (treatment cohort) or survey completion (control cohort) in 2011/2012 and ends December 31, 2017 (or date of death if it occurred first). Shaded areas indicate the 95% CI. Number of individuals at risk at each time point is presented below the x axis.
Diabetes
In addition to the common imbalances described above (first paragraph of Results), there were further imbalances between the treatment and control groups in the matched sub-cohort at risk of diabetes. Two prevalent health conditions were more common in the treatment group: COPD (females: 27.3% vs. 17.6%; males: 24.5% vs. 15.0%) and asthma (females: 23.3% vs. 18.0%). The proportion having at least one outpatient visit in the two years prior to index was higher in the treatment versus control group (females: 96.3% vs. 93.5%; males: 93.2% vs. 82.1%), as was the proportion who had been hospitalized (males:13.5% vs. 10.1%). Baseline characteristics of the matched sub-cohort at risk of diabetes, stratified by sex, are described in S3 Table .
See Table 1 for the estimated incidence of diabetes within 5 years of follow-up in the treatment and control groups. The cumulative incidence of diabetes was significantly higher over time for treated versus control females (Gray’s test p = 0.004, see Fig 3A ) and males (Gray’s test p < 0.001, see Fig 3B ).
10.1371/journal.pone.0288759.g003
Fig 3
Cumulative incidence of diabetes among female (A) and male (B) smoking cessation treatment patients versus matched controls.
The follow-up period begins the day after enrolment in smoking cessation treatment (treatment cohort) or survey completion (control cohort) in 2011/2012 and ends December 31, 2017 (or date of death if it occurred first). Shaded areas indicate the 95% CI. Number of individuals at risk at each time point is presented below the x axis.
Hypertension
In addition to the common imbalances described above (first paragraph of Results), there were further imbalances between the treatment and control groups in the matched sub-cohort at risk of hypertension. Three prevalent health conditions were more common in the treatment versus control groups: COPD (females: 19.4% vs. 11.6%; males: 17.9% vs. 9.9%), diabetes (females: 7.8% vs. 5.0%; males: 9.2% vs. 5.5%) and asthma (females: 24.7% vs. 19.1%). A lower proportion had 0–5 ADG comorbidities (males: 67.8% vs. 73.7%), and a higher proportion had 10+ ADG comorbidities (males: 7.4% vs. 4.9%), in the treatment group, indicating an overall higher comorbidity burden among treated versus control males. The proportion having at least one outpatient visit in two years prior to index was higher in the treatment group of both sexes (females: 96.0% vs. 92.4%; males: 91.7% vs. 79.0%). Baseline characteristics of the matched sub-cohort at risk of hypertension, stratified by sex, are described in S4 Table .
See Table 1 for the estimated incidence of hypertension within 5 years of follow-up in the treatment and control groups. There was no significant difference in the cumulative incidence of hypertension (females: Gray’s test p = 0.30, see Fig 4A ; males: Gray’s test p = 0.81, see Fig 4B ).
10.1371/journal.pone.0288759.g004
Fig 4
Cumulative incidence of hypertension among female (A) and male (B) smoking cessation treatment patients versus matched controls.
The follow-up period begins the day after enrolment in smoking cessation treatment (treatment cohort) or survey completion (control cohort) in 2011/2012 and ends December 31, 2017 (or date of death if it occurred first). Shaded areas indicate the 95% CI. Number of individuals at risk at each time point is presented below the x axis.
Major cardiovascular events
In addition to the common imbalances described above (first paragraph of Results), there were further imbalances between treatment and control groups in the matched sub-cohort at risk of major cardiovascular events. Three prevalent health conditions were more common in the treatment versus control groups: COPD (females: 30.2% vs. 20.3%; males: 26.2% vs. 17.4%), diabetes (males: 15.7% vs. 10.8%) and asthma (females: 24.0% vs. 18.8%). The proportion having made at least one outpatient visit in two years prior to index was higher among the treated versus control group (females: 96.6% vs. 93.9%; males: 93.8% vs. 83.8%). Baseline characteristics of the matched sub-cohort at risk of major cardiovascular events, stratified by sex, are described in S5 Table .
See Table 1 for the estimated incidence of major cardiovascular events within 5 years of follow-up in the treatment and control groups. There was no significant difference in the cumulative incidence of major cardiovascular events (females: Gray’s test p = 0.82, see Fig 5A ; males: Gray’s test p = 0.56, see Fig 5B ).
10.1371/journal.pone.0288759.g005
Fig 5
Cumulative incidence of major cardiovascular events among female (A) and male (B) smoking cessation treatment patients versus matched controls.
The follow-up period begins the day after enrolment in smoking cessation treatment (treatment cohort) or survey completion (control cohort) in 2011/2012 and ends December 31, 2017 (or date of death if it occurred first). Shaded areas indicate the 95% CI. Number of individuals at risk at each time point is presented below the x axis.
Discussion
In this retrospective matched cohort study, we observed a higher incidence of certain chronic diseases within five years follow-up among Ontarians who smoked and who had enrolled in a smoking cessation treatment program compared to Ontarians who smoke but had not enrolled in the program. Specifically, there was a higher incidence of diabetes among the treatment group versus control group of both sexes, and a higher incidence of COPD among male treatment versus control groups. No difference was observed between treatment and control groups in the incidence of cancer, hypertension, or major cardiovascular events during follow-up.
Increased incidence of diabetes in the group of Ontarians seeking smoking cessation treatment was an expected finding. While cigarette smoking itself is a risk factor for incident diabetes [ 43 ], quitting smoking is associated with an increased short-term risk of incident type 2 diabetes compared with continued smoking [ 44 , 45 ]. This risk was reported to be greatest within 3 years of quitting in one study [ 44 ], and between 5 and 7 years after quitting in another study [ 45 ], before decreasing over time until there was no excess risk. A reduction in cigarettes smoked per day over 3 years was also associated with increased insulin and glucose levels in males, but not females, in one study [ 46 ]. Various analyses suggest this increased risk for diabetes is at least partially explained by weight gain [ 44 – 46 ]. Quit status could not be incorporated into our analyses due to the cross-sectional design of the CCHS and resulting lack of longitudinal smoking status for the control sub-cohorts. However, given the previously reported proportions who quit in the STOP program (27%) [ 47 ] versus past-year quit attempt success in the general population (12% in 2017 [ 48 ]), the proportion of recent quitters was likely higher in the treatment sub-cohort. This may have driven up the number of incident cases of diabetes in the treatment group relative to the control group. The greater proportion of daily (versus occasional) smoking in the treatment sub-cohort at baseline may also have been associated with a higher incidence of diabetes during follow-up, given the increased risk of diabetes among heavier versus lighter smokers [ 43 ]. Regardless of the underlying mechanisms, this finding has important implications for the health care system. It is important that health care providers delivering smoking cessation treatment are aware of an elevated incidence of diabetes in this population and combine smoking cessation interventions with strategies for prevention and early detection of type 2 diabetes [ 44 ]. This may include regular monitoring of glucose levels during and following a quit attempt, and recommending treatments associated with less weight gain, particularly for individuals with a higher risk of diabetes. A recent network meta-analysis found several pharmacologic treatments for smoking cessation minimized weight gain, with nicotine patches combined with fluoxetine being associated with the least weight gain [ 49 ]. Notably, the increased short-term risk of type 2 diabetes does not negate the beneficial impact of smoking cessation on cardiovascular or all-cause mortality [ 45 ].
We also found an increased incidence of COPD in the treatment group compared to the control group, although for males only. At index, treated males had higher recent healthcare utilization and prevalent diabetes compared to control males, whereas treated females appeared to have less comorbidity burden than control females. Greater incidence of COPD in treatment versus control group males, but not females, may have been due to these underlying differences in health factors at index, but differential diagnosis of COPD may also have occurred. A review of several studies, including one conducted in Ontario, suggests that only 20% to 30% of individuals with evidence of persistent airflow limitation on spirometry have been diagnosed with COPD [ 50 ], and under diagnosis is more common among males [ 50 ]. Thus, cases of COPD that were undiagnosed may have been more common among treatment versus control group males, and as a result, more diagnoses may have occurred in this group during follow-up. The greater proportion of diabetes among treatment versus control group males supports this possibility given that diabetes is more common in patients with COPD [ 51 ]; however, shared underlying mechanisms may also put one at risk for both diabetes and COPD [ 52 ]. We are unaware of any prior research or mechanism by which smoking cessation treatment, per se, would increase the incidence of COPD, and thus suggest this finding is likely due to extenuating factors. The greater prevalence of COPD at both index (in the other four chronic disease sub-cohorts) and at follow-up in our sample suggest that primary care patients seeking smoking cessation treatment are an at-risk group who should be targeted for COPD case finding. Moreover, a case-finding approach can increase detection of COPD among all patients who smoke, not just those seeking smoking cessation treatment (e.g., see algorithm by Diab et al. [ 50 ]). The National Lung Health Education Program Consensus Statement recommends that primary care providers perform spirometry in patients 40 years of age or older who are current or former smokers and have chronic cough, excessive sputum production, wheezing, or shortness of breath out of proportion to age or activity performed [ 53 ]. Similar recommendations have been made by others [ 50 ].
In contrast to diabetes and COPD, incidence of the three other conditions (cancer, major cardiovascular events, and hypertension) did not differ between the treatment and control groups over follow-up. Smoking cessation reduces risk of cancer, but smoking cessation treatment may not have a similar association. However, incidence of cancer is inversely related to time since cessation [ 54 – 58 ], thus five years may not have been long enough to find a difference in overall cancer incidence following smoking cessation treatment with a 27% cessation rate. Examining overall cancer incidence may have also masked changes in select types of cancer, particularly those most highly linked with smoking. For example, a recent meta-analysis found that a reduction in smoking was associated with lower incidence of lung cancer but did not lower all-cancer or smoking-related cancer risk [ 8 ]. Thus, future studies examining the association between smoking cessation treatment and incidence of cancer may wish to examine incidence of select types of cancer as well as all cancers (if feasible to do so). Although smoking cessation is associated with lower risk of cardiovascular disease within five years compared to continued smoking [ 5 ], the present study suggests major cardiovascular events are not reduced within five years of smoking cessation treatment. However, COPD, which is associated with higher incidence of both cardiovascular disease and hypertension [ 51 ], was higher in the treatment groups relative to the control groups and could have masked any potential reduction in both these disease outcomes following smoking cessation treatment.
Ultimately, imbalances between the treatment and control groups in the matched sub-cohorts prevent us from being able to make firm conclusions regarding the impact of smoking cessation treatment on incident chronic diseases.
Producing real-world evidence is important so that simulation-based modeling exercises are not the only source of population-based information available. However, real-world evidence relies on the availability of real-world data; we were able to leverage circumstances that made such an investigation possible: 1) access to data from a large sample of treatment-seeking people, 2) a large sample of respondents to a population-based survey to serve as controls, 3) both previously linked to routinely collected healthcare administrative data. However, low overall smoking prevalence in the population of Ontario meant that the number of survey respondents who smoked was also relatively low. For this reason, we combined multiple cycles of the CCHS, a common approach used to increase the number of potential controls. STOP, being a treatment program, tends to serve those who smoke more heavily overall (and relatively few of those who smoke lightly or non-daily), but the CCHS, being a population-based survey, has a broader distribution of smoking behaviors. To achieve a more appropriate comparison, we 1:1 matched on age and sex, and propensity score-matched with a caliper width of 0.2 standard deviations of the estimated propensity score logit. As described above, imbalances remained on some smoking measures and some prevalent comorbidities. Decreasing the caliper width would have resulted in pairs that were more tightly matched but with a resulting decrease in the number of matched pairs (i.e., greater internal validity but decreased external validity). As is, we obtained matches for 76.8% of the STOP participants whose data was linked. Thus, we chose an approach that balanced internal and external validity.
The current study highlights the challenge in using existing real-world data to obtain a control group of people who smoke with which to compare those who smoke and seek treatment. To use population-based surveys, as we have done, those surveys would need to be extraordinarily large and, likely, would need to oversample those who smoke more heavily and/or who have made quit attempts. Alternatively, context-specific surveys of people who smoke could be used if the survey data was then linked to existing healthcare administrative data. Linkage of survey data from large numbers of people has obvious cost implications and cross-sectional data capture only a snapshot in time. But using survey data to identify controls is only necessary because smoking status, though central to obtaining a fulsome patient history, isn’t easily obtained from existing healthcare datasets. Indeed, many important patient characteristics, like smoking behaviours, are captured only in free text fields as case notes in special sections of electronic medical records; the unstructured nature of these data present an obstacle for use in producing real world evidence [ 59 ]. Annually, at least 80% of Ontarians access the publicly funded healthcare system and as a result an administrative record is generated [ 60 ]. If smoking behaviours and history (e.g., smoking status, cigarettes/day, age first smoked), were routinely captured in a usable format, we would have had over 2 million (roughly 19% of Ontario’s population in 2011/2012 [ 61 , 62 ]) Ontarians as potential controls. Further, and though outside the scope of the current study, other traditional health behaviours that are difficult to measure using existing administrative data (i.e., alcohol consumption, poor diet, physical inactivity, lack of sleep) are also poorly ascertained yet important determinants of health and healthcare service use. Improving capture of these types of factors would be a significant leap forward in the ability to use real world data to produce real world evidence.
As discussed above, the main limitation of the current study is that individuals were not randomized to groups and despite matching, the treatment and control group in the matched sub-cohorts were not balanced on several baseline variables. Most importantly, there was imbalance on conditions known to be associated with disease outcomes. On the other hand, a strength of the current study was that data from the treatment group was collected from patients receiving smoking cessation treatment under real-world circumstances with their primary care provider, enhancing the generalizability of our findings to real-world treatment in healthcare settings. Further, as some individuals in the control group likely sought treatment (e.g., over the counter NRT or other prescription medications such as varenicline), the current study is not a comparison of any versus no smoking cessation treatment, but speaks more specifically to the outcomes associated with delivering a smoking cessation treatment program providing NRT with behavioural support in primary care, in a context where other treatments are available. Lastly, our study speaks to incident chronic disease within a five-year follow-up period and cannot rule out the possibility that differences presented after that timeframe. Therefore, future studies should extend the observation period.
Conclusions
In summary, further research able to adequately control for baseline differences between those who do and do not seek smoking cessation treatment is needed using real-world data to establish whether there is a differential change in incidence of chronic disease in the short and long term. Although there was imbalance between the treatment and control groups, our study shows those seeking smoking cessation treatment have higher levels of comorbidity than population-based controls, both prior to and following treatment. Health care providers should be aware of higher incidence of diabetes and COPD among those seeking smoking cessation treatment and conduct appropriate screening for these conditions.
Supporting information
S1 Checklist
RECORD checklist.
(DOCX)
S1 Fig
Derivation of at-risk matched treatment and control cohorts.
(DOCX)
S1 Table
Baseline characteristics of matched treatment and control females and males, at risk for cancer.
(DOCX)
S2 Table
Baseline characteristics of matched treatment and control females and males, at risk for chronic obstructive pulmonary disease.
(DOCX)
S3 Table
Baseline characteristics of matched treatment and control females and males, at risk for diabetes.
(DOCX)
S4 Table
Baseline characteristics of matched treatment and control females and males, at risk for hypertension.
(DOCX)
S5 Table
Baseline characteristics of matched treatment and control females and males, at risk for major cardiovascular events.
(DOCX)
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.