text
stringlengths 65
113k
|
|---|
A large body of evidence indicates that pigeons use olfactory cues to navigate over unfamiliar areas with a differential contribution of the left and right hemispheres. In particular, the right nostril/olfactory bulb (OB) and left piriform cortex (Cpi) have been demonstrated to be crucially involved in navigation. In this study we analysed behaviour-induced activation of the olfactory system, indicated by the expression of the immediate early gene ZENK, under different homing conditions. One experimental group was released from an unfamiliar site, the second group was transported to the unfamiliar site and back to the loft, and the third group was released in front of the loft. To evaluate the differential contribution of the left and/or right olfactory input, the nostrils of the pigeons were either occluded unilaterally or not. Released pigeons revealed the highest ZENK cell density in the OB and Cpi, indicating that the olfactory system is activated during navigation from an unfamiliar site. The groups with no plug showed the highest ZENK cell density, supporting the activation of the olfactory system probably being due to sensory input. Moreover, both Cpis seem to contribute differently to the navigation process. Only occlusion of the right OB resulted in a decreased ZENK cell expression in the Cpi, whereas occlusion of the left nostril had no effect. This is the first study to reveal neuronal activation patterns in the olfactory system during homing. Our data show that lateralized processing of olfactory cues is indeed involved in navigation over unfamiliar areas. |
The supraoptic nuclei of the hypothalamus display a remarkable anatomical plasticity during lactation, parturition and chronic dehydration, conditions associated with massive neurohypophysial hormone secretion. This structural remodeling is characterized by a pronounced reduction of the astrocytic coverage of oxytocin neurons, resulting in an increase in the number and extent of directly juxtaposed neuronal surfaces. Although the exact role played by such an anatomical remodeling in the physiology of the hypothalamo-neurohypophysial system is still unknown, several findings obtained over the last decade indicate that synaptic and extrasynaptic transmissions are impacted by these structural changes. We review these data and try to extrapolate how such changes at the cellular level might affect the overall activity of the system. One repercussion of the retraction of glial processes is the accumulation of glutamate in the extracellular space. This build-up of glutamate causes an increased activation of pre-synaptic metabotropic glutamate receptors, which are negatively coupled to neurotransmitter release, and a switch in the mode of action of pre-synaptic kainate receptors that control GABA release. Finally, the range of action of substances released from astrocytes and acting on adjacent magnocellular neurons is also affected during the anatomical remodeling. It thus appears that the structural plasticity of the hypothalamic magnocellular nuclei strongly affects neuron-glial interactions and, as a consequence, induces significant changes in synaptic and extrasynaptic transmission. |
This review focuses on the plasticity of the regulation of a particular neuroendocrine transducer cell, the melanotrope cell in the pituitary pars intermedia of the amphibian Xenopus laevis. This cell type is a suitable model to study the relationship between various external regulatory inputs and the secretion of an adaptive endocrine message, in this case the release of α-melanophore-stimulating hormone, which activates skin melanophores to darken when the animal is placed on a dark background. Information about the environmental conditions is processed by various brain centres, in the hypothalamus and elsewhere, that eventually control the activity of the melanotrope cell regarding hormone production and secretion. The review discusses the roles of these hypothalamic and extrahypothalamic nuclei, their neurochemical messengers acting on the melanotrope, and the external stimuli they mediate to control melanotrope cell functioning. |
It is important to determine the mechanisms controlling the number of neurons in the nervous system. Previously, we reported that neuronal activity plays a central role in controlling neuron number in the neonatal hippocampus of rodents. Neuronal survival requires sustained activation of the serine-threonine kinase Akt, which is initiated by neurotrophins and continued for several hours by neuronal activity and integrin signaling. Here, we focus on the CA3 region to show that neuronal apoptosis requires p53. As in wild-type animals, neuronal death occurs in the first postnatal week and ends by postnatal day (P)10 in p53(-/-) mice. During this period, the CA3 region of p53(-/-) mice contains significantly lower numbers of apoptotic cells, and at the end of the death period, it contains more neurons than the wild type. At P10, the p53(-/-) CA3 region contains a novel subpopulation of neurons with small soma size. These neurons show normal levels of tropomyosin receptor kinase receptor activation, but lower levels of activated Akt than the neurons with somata of normal size. These results suggest that p53 is the key downstream regulator of the novel survival-signaling pathway that regulates the number of CA3 neurons in the first 10 days of postnatal life. |
Involvement of fronto-parietal structures within the right hemisphere in bodily self recognition has gained convergent support from behavioural, neuropsychological and neuroimaging studies. Increases in corticospinal excitability via transcranial magnetic stimulation (TMS) also testify to right hemisphere self-related processing. However, evidence for self-dependent modulations of motor excitability is limited to the processing of face-related information that, by definition, conveys someone's identity. Here we tested the hypothesis that vision of one's own hand, as compared with vision of somebody else's hand, would also engage specific self-hand processing in the right hemisphere. Healthy participants were submitted to a classic TMS paradigm to assess changes in corticospinal excitability of the right (Experiment 1) and left (Experiment 2) motor cortex, while viewing pictures of a (contralateral) still hand, which could either be their own (Self) or not (Other). As a control for body selectivity, subjects were also presented with pictures of a hand-related, but non-corporeal object, i.e. a mobile phone, which could similarly be their own or not. Results showed a selective right hemisphere increase in corticospinal excitability with self-hand and self-phone stimuli with respect to Other stimuli. Such a Self vs. Other modulation of primary motor cortex appeared at 600 ms and was maintained at 900 ms, but was not present at earlier timings (100 and 300 ms) and was completely absent following stimulation of the left hemisphere. A similar pattern observed for self-hand and self-phone stimuli suggests that owned hands and objects may undergo similar self-processing, possibly via a different cortical network from that responsible for self-face processing. |
The hypothesis of communication through coherence proposes that coherent or synchronous oscillations in connected neural systems can promote communication. It has been applied mainly to how oscillations interact in connected networks. We tested by simulations whether information transmission about an external stimulus from one network to a second network is influenced by gamma oscillations, by whether the oscillations are coherent, and by their phase. Gamma oscillations were induced by increasing the relative conductance of AMPA to NMDA excitatory synapses. It was found that small associative connection strengths between the networks were sufficient to produce information transmission (measured by Shannon mutual information) such that the second attractor network took the correct decision based on the state of the first network. Although gamma oscillations were present in both networks, the synaptic connections sufficient for perfect information transmission about the stimulus presented to the network (100% correct, 1 bit of information) were insufficiently strong to produce coherence, or phase-locking, between the two networks; this only occurred when the synaptic strengths between the networks were increased by > 10 ×. Further, the phase of the oscillations between the networks did not influence the information transmission or its speed at these connection strengths. Moreover, information transmission was as good when the AMPA-to-NMDA ratio was reduced to its normal value, and oscillations were not present. Similar results were found when the second network was not an attractor decision-making network. Thus information transmission can occur before synapses have been made sufficiently strong to produce coherence. |
The matrix metalloproteinase (MMP) enzyme family contributes to the regulation of a variety of brain extracellular matrix molecules. In order to assess their role in synaptic plasticity following traumatic brain injury (TBI), we compared expression of stromelysin-1 (MMP-3) protein and mRNA in two rodent models of TBI exhibiting different levels of recovery: adaptive synaptic plasticity following central fluid percussion injury and maladaptive synaptic plasticity generated by combined TBI and bilateral entorhinal cortical lesion (TBI + BEC). We sampled the hippocampus at 7 days postinjury, targeting a selectively vulnerable brain region and a survival interval exhibiting rapid synaptogenesis. We report elevated expression of hippocampal MMP-3 mRNA and protein after TBI. MMP-3 immunohistochemical staining showed increased protein levels relative to sham-injured controls, primarily localized to cell bodies within the deafferented dendritic laminae. Injury-related differences in MMP-3 protein were also observed. TBI alone elevated MMP-3 immunobinding over the stratum lacunosum moleculare (SLM), inner molecular layer and hilus, while TBI + BEC generated more robust increases in MMP-3 reactivity within the deafferented SLM and dentate molecular layer (DML). Double labeling with GFAP confirmed the presence of MMP-3 within reactive astrocytes induced by each injury model. Semi-quantitative RT-PCR revealed that MMP-3 mRNA also increased after each injury, however, the combined insult induced a much greater elevation than fluid percussion alone: 1.9-fold vs. 79%, respectively. In the TBI + BEC model, MMP-3 up-regulation was spatio-temporally correlated with increased enzyme activity, an effect which was attenuated with the neuroprotective compound MK-801. These results show that distinct pathological conditions elicited by TBI can differentially affect MMP-3 expression during reactive synaptic plasticity. Notably, these effects are both transcriptional and translational and are correlated with functionally active enzyme. |
Bone marrow (BM) is a rich source of stem cells and may represent a valid alternative to neural or embryonic cells in replacing autologous damaged tissues for neurodegenerative diseases. The purpose of the present study is to identify human adult BM progenitor cells capable of neuro-glial differentiation and to develop effective protocols of trans-differentiation to surmount the hematopoietic commitment in vitro. Heterogeneous cell populations such as whole BM, low-density mononuclear and mesenchymal stem (MSCs), and several immunomagnetically separated cell populations were investigated. Among them, MSCs and CD90+ cells were demonstrated to express neuro-glial transcripts before any treatment. Several culture conditions with the addition of stem cell or astroblast conditioned media, different concentrations of serum, growth factors, and supplements, used alone or in combinations, were demonstrated to alter the cellular morphology in some cell subpopulations. In particular, MSCs and CD90+ cells acquired astrocytic and neuron-like morphologies in specific culture conditions. They expressed several neuro-glial specific markers by RT-PCR and glial fibrillary acid protein by immunocytochemistry after co-culture with astroblasts, both in the absence or presence of cell contact. In addition, floating neurosphere-like clones have been observed when CD90+ cells were grown in neural specific media. In conclusion, among the large variety of human adult BM cell populations analyzed, we demonstrated the in vitro neuro-glial potential of both the MSC and CD90+ subset of cells. Moreover, unidentified soluble factors provided by the conditioned media and cellular contacts in co-culture systems were effective in inducing the neuro-glial phenotype, further supporting the adult BM neural differentiative capability. |
To better understand the particular vulnerability of mesencephalic dopaminergic neurons to toxins or gene mutations causing parkinsonism, we have taken advantage of a primary cell culture system in which these neurons die selectively. Antimitotic agents, such as cytosine arabinoside or cAMP, prevent the death of the neurons by arresting astrocyte proliferation. To identify factors implicated in either the death of the dopaminergic neurons or in the neuroprotective effect of cAMP, we constructed cDNA libraries enriched by subtractive hybridization and suppressive PCR in transcripts that are preferentially expressed in either control or cAMP-treated cultures. Differentially expressed transcripts were identified by hybridization of the enriched cDNAs with a commercially available cDNA expression array. The proteoglycan receptors syndecan-3 and the receptor protein tyrosine phosphatase zeta/beta were found among the transcripts preferentially expressed under control conditions, and their ligand, the cytokine pleiotrophin, was highly represented in the cDNA libraries for both conditions. Since pleiotrophin is expressed during embryonic and perinatal neural development and following lesions in the adult brain, we investigated its role in our cell culture model. Pleiotrophin was not responsible for the death of dopaminergic neurons under control conditions, or for their survival in cAMP-treated cultures. It was, however, implicated in the initial and cAMP-dependent enhancement of the differentiation of the dopaminergic neurons in our cultures. In addition, our experiments have provided evidence for a cAMP-dependent regulatory pathway leading to protease activation, and the identification of pleiotrophin as a target of this pathway. |
It has recently been reported that adult hematopoietic stem cells can differentiate into neural cells, opening new frontiers in therapy for neurodegenerative diseases. In this study, adult human hematopoietic stem cells (HSCs) were isolated via magnetic bead sorting, using a specific CD34 antibody and cultured with human astrocyte culture conditioned medium (ACM). In order to evaluate their differentiation into neurons and/or astrocytes, ACM-treated cultures were probed for the expression of several neural markers. We observed morphological modifications and, after 20 days of treatment, cell morphology displayed extending processes. Immunocytochemistry, Western blotting and RT-PCR showed the expression of neuronal markers such as neurofilaments, neuron specific enolase (NSE) and NeuN in ACM-treated HSCs cultured in poly-L-lysine-coated dishes. On the contrary, when the same ACM-treated cells were grown on a plastic substrate, they expressed high levels of glial fibrillary acidic protein (GFAP), with only weak expression of neuronal markers. Nestin, a neural progenitor cell marker, was present in treated cells, regardless of the substrate. These results demonstrate that astrocytes can generate a suitable microenvironment for inducing HSCs to differentiate into neural cells. Therefore, adult bone marrow may represent a readily accessible source of cells for treating neurodegenerative diseases. |
The detrimental effects of traumatic brain injury (TBI) on brain tissue integrity involve progressive axonal damage, necrotic cell loss, and both acute and delayed apoptotic neuronal death due to activation of caspases. Post-injury accumulation of amyloid precursor protein (APP) and its toxic metabolite amyloid-beta peptide (Abeta) has been implicated in apoptosis as well as in increasing the risk for developing Alzheimer's disease (AD) after TBI. Activated caspases proteolyze APP and are associated with increased Abeta production after neuronal injury. Conversely, Abeta and related APP/Abeta fragments stimulate caspase activation, creating a potential vicious cycle of secondary injury after TBI. Blockade of caspase activation after brain injury suppresses apoptosis and improves neurological outcome, but it is not known whether such intervention also prevents increases in Abeta levels in vivo. The present study examined the effect of caspase inhibition on post-injury levels of soluble Abeta, APP, activated caspase-3, and caspase-cleaved APP in the hippocampus of nontransgenic mice expressing human Abeta, subjected to controlled cortical injury (CCI). CCI produced brain tissue damage with cell loss and elevated levels of activated caspase-3, Abeta(1-42) and Abeta(1-40), APP, and caspase-cleaved APP fragments in hippocampal neurons and axons. Post-CCI intervention with intracerebroventricular injection of 100 nM Boc-Asp(OMe)-CH(2)F (BAF, a pan-caspase inhibitor) significantly reduced caspase-3 activation and improved histological outcome, suppressed increases in Abeta and caspase-cleaved APP, but showed no significant effect on overall APP levels in the hippocampus after CCI. These data demonstrate that after TBI, caspase inhibition can suppress elevations in Abeta. The extent to which Abeta suppression contributes to improved outcome following inhibition of caspases after TBI is unclear, but such intervention may be a valuable therapeutic strategy for preventing the long-term evolution of Abeta-mediated pathology in TBI patients who are at risk for developing AD later in life. |
The wobbler mouse is one of the most useful models of motoneuron degeneration, characterized by selective motoneuronal death in the cervical spinal cord. We carried out two parallel studies in wobbler mice, comparing the anti-glutamatergic drug riluzole and the AMPA receptor antagonist RPR119990. Mice were treated with 40 mg/kg/day of riluzole or with 3 mg/kg/day of RPR119990 from the 4th to the 12th week of age. Here, we show that chronic treatment with riluzole improves motor behavior, prevents biceps muscle atrophy and decreases the amount of motoneuron loss in treated wobbler mice. Chronic treatment with the AMPA antagonist RPR119990 is ineffective in improving motor impairment, in reducing motoneuronal loss and muscular atrophy in treated mice. These results, together with the unchanged immunostaining for the AMPA receptor subunit GluR2 in wobbler mice, suggest that AMPA receptor-mediated injury is unlikely to be involved in neurodegeneration in wobbler disease, and that the protective effect of riluzole in wobbler mice seems to be independent of its anti-glutamatergic activity, as suggested in other models of neurodegeneration. Immunostaining of cervical spinal cord sections shows that in riluzole-treated wobbler mice BDNF expression is significantly increased in motoneurons with no changes in the high-affinity receptor Trk-B. Our data confirm that riluzole has beneficial effects in wobbler mice, and suggest that these effects could be associated to the increased levels of the neurotrophic and neuroprotective factor BDNF. |
Hepatocyte growth factor (HGF), mainly produced and acting in the periphery, attenuates cerebral ischemia-induced cell death and thus shows therapeutic potential in CNS regeneration. Accordingly, we tested its ability to permeate the blood-brain barrier (BBB). HGF was stable in the circulating blood of adult mice for up to 20 min, as HPLC showed intact (125)I-HGF in both serum and brain homogenate. Multiple time regression analysis revealed a rapid blood-to-brain influx rate of 0.38 +/- 0.07 microl/g min, faster than might be expected for a protein of this size. Although excess unlabeled HGF failed to inhibit of the influx of (125)I-HGF in mice, the use of a higher dose of unlabeled HGF in cellular uptake studies showed the presence of saturable endocytosis. Furthermore, capillary depletion studies showed that about 32% of the HGF present in brain entered the parenchymal compartment in contrast to the 11% entrapped in endothelial cells 10 min after intravenous bolus injection. The amount of HGF that crossed the BBB in intact form was substantial and could be physiologically important in the CNS. |
Previous studies of patients with focal cerebellar damage underscored the importance of the cerebellum for balance control. These studies were restricted to postural control in the pitch plane, and focused mainly on leg muscle responses. Here, we examined the effect of degenerative cerebellar lesions on postural control in multiple directions, and studied how such lesions affect intersegmental coordination of the legs, trunk and arms. We formulated two main questions. (a) Do patients with cerebellar ataxia predominantly have balance problems in the sagittal or frontal planes? (b) Is instability in cerebellar ataxia associated with increased joint motion or with reduced joint motion? We selected nine patients with autosomal dominant spinocerebellar ataxia (SCA)--three with pure ataxia and six with mild extra-cerebellar features--and 12 matched controls. Upright standing subjects received support surface rotations (7.5 degrees at 60 degrees /s) that were randomly delivered in eight different directions of pitch or roll. We used full body kinematics to determine displacements of the center of mass (COM) and of individual body segments. We also collected surface EMG from 10 leg, trunk and arm muscles. Primary variables of interest were COM displacement and trunk control (angles and muscle responses). Secondary analyses focused on angles and muscle responses of the legs and arms. COM analysis demonstrated that SCA patients had greatest instability following backward and laterally directed perturbations. Major factors in causing this instability were, first, a marked reduction of stimulus-induced knee flexion and, second, excessive "hypermetric" motion of the pelvis (in roll) and trunk (in pitch). Muscle responses of SCA patients were characterized by increased late balance correcting activity. Responses of patients with pure ataxia were comparable to those of patients with mild extra-cerebellar features. A main underlying cause of postural instability in SCA patients appears to be "locking" of the knees, which may reflect compensation (by reducing interaction between body links) or reduced vestibulocerebellar control over leg muscles. The observed pathophysiology is very different from that seen in other patient populations. |
Thiadiazolidinones (TDZDs) are small molecules that inhibit glycogen synthase kinase 3-beta (GSK3-beta) activity in a non competitive manner to ATP. NP00111, a new TDZD, besides causing inhibition of GSK-3beta, has also shown to be an agonist of PPARgamma . Since phosphorylation and consequent inhibition of GSK-3beta by PI-3K/Akt and agonism of PPARgamma have shown to afford neuroprotection in several in vitro and in vivo models, we have studied the potential neuroprotective effect of NP00111 in an "in vitro" model of ischemia-reperfusion. NP00111, at the concentration of 10 microM, significantly protected adult rat hippocampal slices subjected to oxygen and glucose deprivation (OGD) for 1 h followed by 3 h re-oxygenation, measured as lactic dehydrogenase (LDH) released to the extracellular media. The protective effects of NP00111 were more pronounced during the re-oxygenation period in comparison to the OGD period. Other GSK-3beta inhibitors like lithium or AR-A014418 did not afford protection in this model. However, the PPARgamma agonist rosiglitazone was protective at 3 microM. Protection afforded by NP00111 and rosiglitazone were prevented by the PPARgamma antagonist GW9662, suggesting that both NP00111 and rosiglitazone were preventing cell death caused by oxygen-glucose deprivation via activation of PPARgamma. NP00111 increased by two fold phosphorylation of ERK1/2 and its protective effects were lost when the hippocampal slices were co-incubated with the mitogen-activated protein kinase (MAPK) inhibitor PD98059. In conclusion, the novel TDZD NP00111 was protective against OGD in rat hippocampal slices by a mechanism related to phosphorylation of ERK1/2 via activation of PPARgamma. |
Embryonic spinal cord motor neurons (MNs) can be maintained in vitro for weeks with a cocktail of trophic factors and muscle-derived factors under serum-containing conditions. Here we investigated the beneficial effects of muscle-derived factors in the form of muscle-conditioned medium (MCM) on the survival and neurite outgrowth of adult rat spinal cord MNs under serum-free conditions. Ventral horn dissociated cell cultures from the cervical enlargement were maintained in the presence of one or more of the following factors: brain-derived neurotrophic factor (BDNF), glial cell-derived neurotrophic factor (GDNF), a cell permeant cyclic adenosine-3',5'-monophosphate (cAMP) analog and MCM. The cell cultures were immunostained with several antibodies recognizing a general neuronal marker the microtubule-associated protein 2 (MAP2) and either one or more motor neuronal markers: the non-phosphorylated neurofilament heavy isoform (SMI32), the transcription factors HB9 and Islet-1 and the choline acetyl transferase. We found that treatment with MCM together with the cAMP analog was sufficient to promote selective survival and neurite outgrowth of adult spinal cord MNs. These conditions can be used to maintain adult spinal cord MNs in dissociated cultures for several weeks and may have therapeutic potential following spinal cord injury or motor neuropathies. More studies are necessary to evaluate how MCM and the cAMP analog act in synergy to promote the survival and neurite outgrowth of adult MNs. |
Chronic CNS infection by several families of viruses can produce deficits in prefrontal cortex (PFC) and striatal function. Cannabinoid drugs have been long known for their anti-inflammatory properties and their ability to modulate adult neuro and gliogenesis. Therefore, we explored the effects of systemic administration of the cannabinoid agonist WIN55,212-2(WIN) on prefrontal cortex (PFC) and striatal cytogenesis in a viral model of CNS injury and inflammation based on Borna Disease (BD) virus encephalitis. Active BrdU(+) progenitor populations were significantly decreased 1 week after BrdU labeling in BD rats [p<0.001 compared to uninfected (NL) controls] while less than 5% of BrdU(+) cells colabeled for BDV protein. Systemic WIN (1mg/kg i.p. twice daily×7 days) increased the survival of BrdU(+) cells in striatum (p<0.001) and PFC of BD rats, with differential regulation of labeled oligodendroglia precursors vs microglia/macrophages. WIN increased the percentage of BrdU(+) oligodendrocyte precursor cells and decreased BrdU(+) ED-1-labeled phagocytic cells, without producing pro- or antiviral effects. BDV infection decreased the levels of the endocannabinoid anandamide (AEA) in striatum (p<0.05 compared to NL rats), whereas 2-AG levels were unchanged. Our findings indicate that: 1) viral infection is accompanied by alterations of AEA transmission in the striatum, but new cell protection by WIN appears independent of its effect on endocannabinoid levels; and 2) chronic WIN treatment alters the gliogenic cascades associated with CNS injury, promoting oligodendrocyte survival. Limiting reactive gliogenesis and macrophage activity in favor of oliogodendroglia development has significance for demyelinating diseases. Moreover, the ability of cannabinoids to promote the development of biologically supportive or symbiotic oligodendroglia may generalize to other microglia-driven neurodegenerative syndromes including NeuroAIDS and diseases of aging. |
Nerve growth factor (NGF) has been previously shown to support neuron survival and direct neurite outgrowth in vitro, and to enhance axonal regeneration in vivo. However, a systematic analysis of NGF dose and dose duration on behavioral recovery following peripheral nerve injury in rodents has not been previously investigated. Here, we show that NGF promotes a bell shaped dose-response, with an optimal threshold effect occurring at 800 pg/μl. High dose NGF inhibited regeneration. However, this effect could be reversed through functional blockade of p75 receptors, thus implicating these receptors as mediators of the inhibitory response. Longer term evaluation showed that animals administered NGF at 80 ng/day for 3 weeks had greater sensorimotor recovery compared to all other treatment groups. These animals made significantly fewer errors during skilled locomotion, and displayed both increased vertical and fore-aft ground reaction forces during flat surface locomotion. Furthermore, terminal electrophysiological and myological assessments (EMG, wet gastrocnemius muscle weights) corroborated the behavioral data. Overall, these data support the hypothesis that both appropriate dose and duration of NGF are important determinants of behavioral recovery following nerve injury in the rat. |
Neuropeptide Y (NPY) is widely expressed throughout the CNS and exerts a number of important physiological functions as well as playing a role in pathological conditions such as obesity, anxiety, epilepsy, chronic pain and neurodegenerative disorders. In this review, we highlight some of the recent advances in our understanding of NPY biology and how this may help explain not only its role in health and disease, but also its possible use therapeutically. |
Stroke is a devastating neurological disease with no satisfactory therapies to preserve long-term neurological function, perhaps due to the sole emphasis on neuronal survival in most preclinical studies. Recent studies have revealed the importance of protecting multiple cell types in the injured brain, such as oligodendrocytes and components of the neurovascular unit, before long-lasting recovery of function can be achieved. For example, revascularization in the ischemic penumbra is critical to provide various neurotrophic factors that enhance the survival and activity of neurons and other progenitor cells, such as oligodendrocyte precursor cells. In the present study, we hypothesized that chronic dietary supplementation with fish oil promotes post-stroke angiogenesis, neurogenesis, and oligodendrogenesis, thereby leading to long-term functional improvements. Mice received dietary supplementation with n-3 PUFA-enriched fish oil for three months before and up to one month after stroke. As expected, dietary n-3 PUFAs significantly increased levels of n-3 PUFAs in the brain and improved long-term behavioral outcomes after cerebral ischemia. n-3 PUFAs also robustly improved revascularization and angiogenesis and boosted the survival of NeuN/BrdU labeled newborn neurons up to 35days after stroke injury. Furthermore, these pro-neurogenic effects were accompanied by robust oligodendrogenesis. Thus, this is the first study to demonstrate that chronic dietary intake of n-3 PUFAs is an effective prophylactic measure not only to protect against ischemic injury for the long term but also to actively promote neurovascular restorative dynamics and brain repair. |
Fragile X Syndrome (FXS) is the most common heritable form of intellectual impairment as well as the leading monogenetic cause of autism. In addition to its canonical definition as a neurodevelopmental disease, recent findings in the clinic suggest that FXS is a systemic disorder that is characterized by a variety of heterogeneous phenotypes. Efforts to study FXS pathogenesis have been aided by the development and characterization of animal models of the disease. Research efforts in Drosophila melanogaster have revealed key insights into the mechanistic underpinnings of FXS. While much remains unknown, it is increasingly apparent that FXS involves a myriad of spatially and temporally specific alterations in cellular function. Consequently, the literature is filled with numerous discordant findings. Researchers and clinicians alike must be cognizant of this dissonance, as it will likely be important for the design of preclinical studies to assess the efficacy of therapeutic strategies to improve the lives of FXS patients. |
The mechanisms by which sepsis triggers intensive care unit acquired weakness (ICUAW) remain unclear. We previously identified difficulty with motor unit recruitment in patients as a novel contributor to ICUAW. To study the mechanism underlying poor recruitment of motor units we used the rat cecal ligation and puncture model of sepsis. We identified striking dysfunction of alpha motor neurons during repetitive firing. Firing was more erratic, and often intermittent. Our data raised the possibility that reduced excitability of motor neurons was a significant contributor to weakness induced by sepsis. In this study we quantified the contribution of reduced motor neuron excitability and compared its magnitude to the contributions of myopathy, neuropathy and failure of neuromuscular transmission. We injected constant depolarizing current pulses (5s) into the soma of alpha motor neurons in the lumbosacral spinal cord of anesthetized rats to trigger repetitive firing. In response to constant depolarization, motor neurons in untreated control rats fired at steady and continuous firing rates and generated smooth and sustained tetanic motor unit force as expected. In contrast, following induction of sepsis, motor neurons were often unable to sustain firing throughout the 5s current injection such that force production was reduced. Even when firing, motor neurons from septic rats fired erratically and discontinuously, leading to irregular production of motor unit force. Both fast and slow type motor neurons had similar disruption of excitability. We followed rats after recovery from sepsis to determine the time course of resolution of the defect in motor neuron excitability. By one week, rats appeared to have recovered from sepsis as they had no piloerection and appeared to be in no distress. The defects in motor neuron repetitive firing were still striking at 2weeks and, although improved, were present at one month. We infer that rats suffered from weakness due to reduced motor neuron excitability for weeks after resolution of sepsis. To assess whether additional contributions from myopathy, neuropathy and defects in neuromuscular transmission contributed to the reduction in force generation, we measured whole-muscle force production in response to electrical stimulation of the muscle nerve. We found no abnormality in force generation that would suggest the presence of myopathy, neuropathy or defective neuromuscular transmission. These data suggest disruption of repetitive firing of motor neurons is an important contributor to weakness induced by sepsis in rats and raise the possibility that reduced motor neuron excitability contributes to disability that persists after resolution of sepsis. |
Parkinson's disease (PD) is a neurodegenerative disorder characterized by loss of dopaminergic neurons in the substantia nigra compacta (SNpc) and the only risk factor is aging. We showed that in 6-hydroxydopamine (6-OHDA)-model of PD there is a reduction in the neuronal profile within the brainstem ventral respiratory column with a decrease in the hypercapnic ventilatory response. Here we tested the involvement of orexin cells from the lateral hypothalamus/perifornical area (LH/PeF) on breathing in a 6-OHDA PD model. In this model of PD, there is a reduction in the total number of orexinergic neurons and in the number of orexinergic neurons that project to the RTN, without changing the number of CO<sub>2</sub>-activated orexinergic neurons during the dark phase. The ventilation at rest and in response to hypercapnia (7% CO<sub>2</sub>) was assessed in animals that received 6-OHDA or vehicle injections into the striatum and saporin anti-Orexin-B or IgG saporin into the LH/PeF during the sleep and awake states. The experiments showed a reduction of respiratory frequency (f<sub>R</sub>) at rest during the light phase in PD animals only during sleep. During the dark phase, there was an impaired f<sub>R</sub> response to hypercapnia in PD animals with depletion of orexinergic neurons in awake and sleeping rats. In conclusion, the degeneration of orexinergic neurons in this model of PD can be related to impaired chemoreceptor function in the dark phase. |
Metabotropic glutamate receptors (mGluRs) are G-protein-coupled excitatory amino acid (glutamate) receptors and are abundantly expressed in basal ganglia nuclei. We used behavioral, regional glucose uptake metabolic mapping, and FOS protein expression to examine the effects of stimulating striatal and subthalamic mGluRs in rats. Stimulation of striatal Group I mGluRs produced behavioral effects mediated by polysynaptic activation of subthalamic neurons. Stimulation of subthalamic Group II mGluRs produced similar effects. Excessive activity of subthalamic neurons is a key feature of parkinsonism. mGluR Group I or Group II antagonists may prove to be useful for symptomatic treatment of parkinsonism. Stimulation of Group III mGluRs produced behavioral effects in only 6-hydroxydopamine-lesioned animals. Regional glucose uptake metabolic mapping and FOS expression studies suggested that striatal dopamine denervation produced increased sensitivity of Group III mGluRs. Agents active at Group III mGluRs may also be useful for treatment of parkinsonism. |
Aging is the main risk factor for neurodegenerative diseases. In aging, microglia undergoes phenotypic changes compatible with their activation. Glial activation can lead to neuroinflammation, which is increasingly accepted as part of the pathogenesis of neurodegenerative diseases, including Alzheimer's disease (AD). We hypothesize that in aging, aberrant microglia activation leads to a deleterious environment and neurodegeneration. In aged mice, microglia exhibit an increased expression of cytokines and an exacerbated inflammatory response to pathological changes. Whereas LPS increases nitric oxide (NO) secretion in microglia from young mice, induction of reactive oxygen species (ROS) predominates in older mice. Furthermore, there is accumulation of DNA oxidative damage in mitochondria of microglia during aging, and also an increased intracellular ROS production. Increased ROS activates the redox-sensitive nuclear factor kappa B, which promotes more neuroinflammation, and can be translated in functional deficits, such as cognitive impairment. Mitochondria-derived ROS and cathepsin B, are also necessary for the microglial cell production of interleukin-1β, a key inflammatory cytokine. Interestingly, whereas the regulatory cytokine TGFβ1 is also increased in the aged brain, neuroinflammation persists. Assessing this apparent contradiction, we have reported that TGFβ1 induction and activation of Smad3 signaling after inflammatory stimulation are reduced in adult mice. Other protective functions, such as phagocytosis, although observed in aged animals, become not inducible by inflammatory stimuli and TGFβ1. Here, we discuss data suggesting that mitochondrial and endolysosomal dysfunction could at least partially mediate age-associated microglial cell changes, and, together with the impairment of the TGFβ1-Smad3 pathway, could result in the reduction of protective activation and the facilitation of cytotoxic activation of microglia, resulting in the promotion of neurodegenerative diseases. |
Understanding speech in the presence of background sound can be challenging for older adults. Speech comprehension in noise appears to depend on working memory and executive-control processes (e.g., Heald and Nusbaum, 2014), and their augmentation through training may have rehabilitative potential for age-related hearing loss. We examined the efficacy of adaptive working-memory training (Cogmed; Klingberg et al., 2002) in 24 older adults, assessing generalization to other working-memory tasks (near-transfer) and to other cognitive domains (far-transfer) using a cognitive test battery, including the Reading Span test, sensitive to working memory (e.g., Daneman and Carpenter, 1980). We also assessed far transfer to speech-in-noise performance, including a closed-set sentence task (Kidd et al., 2008). To examine the effect of cognitive training on benefit obtained from semantic context, we also assessed transfer to open-set sentences; half were semantically coherent (high-context) and half were semantically anomalous (low-context). Subjects completed 25 sessions (0.5-1 h each; 5 sessions/week) of both adaptive working memory training and placebo training over 10 weeks in a crossover design. Subjects' scores on the adaptive working-memory training tasks improved as a result of training. However, training did not transfer to other working memory tasks, nor to tasks recruiting other cognitive domains. We did not observe any training-related improvement in speech-in-noise performance. Measures of working memory correlated with the intelligibility of low-context, but not high-context, sentences, suggesting that sentence context may reduce the load on working memory. The Reading Span test significantly correlated only with a test of visual episodic memory, suggesting that the Reading Span test is not a pure-test of working memory, as is commonly assumed. |
The biology of brain microvascular pericytes is an active area of research and discovery, as their interaction with the endothelium is critical for multiple aspects of cerebrovascular function. There is growing evidence that pericyte loss or dysfunction is involved in the pathogenesis of Alzheimer's disease, vascular dementia, ischemic stroke and brain injury. However, strategies to mitigate or compensate for this loss remain limited. In this review, we highlight a novel finding that pericytes in the adult brain are structurally dynamic <i>in vivo</i>, and actively compensate for loss of endothelial coverage by extending their far-reaching processes to maintain contact with regions of exposed endothelium. Structural remodeling of pericytes may present an opportunity to foster pericyte-endothelial communication in the adult brain and should be explored as a potential means to counteract pericyte loss in dementia and cerebrovascular disease. We discuss the pathophysiological consequences of pericyte loss on capillary function, and the biochemical pathways that may control pericyte remodeling. We also offer guidance for observing pericytes <i>in vivo</i>, such that pericyte structural remodeling can be more broadly studied in mouse models of cerebrovascular disease. |
<b>Objective</b>: The aim of this study was to analyze quantitative sleep changes and their implication on subjective cognitive decline (SCD). Objective sleep patterns were investigated by an actigraph and recorded at the baseline and 2-year after in order to examine specific sleep alterations in SCD. <b>Background</b>: Sleep disorders are very common among average elderly adults and an altered sleep pattern is known to be a risk factor for future development of mild cognitive impairment (MCI) and dementia. Recent studies have shown how sleep is objectively altered in average senior adults with SCD, without any other significant change in cognition and behavior or brain structure. Considering that both SCD and disrupted sleep are risk factors for future MCI and dementia, with sleep only as a modifiable risk factor, further research is required to deeply investigate the interaction between sleep and SCD. <b>Methods</b>: Among 70 community-dwelling elderly individuals who had been enrolled at baseline, 35 (64.6 ± 5.6 years, 15 M/20 F) underwent a complete neuropsychological battery and 1-week wrist actigraphy recording 2 years later during the follow-up stage. Individuals were divided into two groups according to their SCD Questionnaire (SCD-Q) score. Sleep hours, sleep efficiency and onset latency, napping and time awake after sleep onset (WASO) were collected. All individuals underwent structural magnetic resonance imaging (MRI) examination to exclude brain disorders. Data collection was performed at baseline and after 2 years at the follow-up phase. <b>Results</b>: A significantly different night sleep time between the two groups was observed: SCD showed a lower total sleep time (TST) than non-SCD subjects. Moreover, a total time spent in bed (TIB) was significantly lower in SCD subjects over 2 years of observation. <b>Conclusions</b>: Objective changes over time of the sleep pattern, specifically TIB and TST, are present in SCD individuals. The results of the study show that sleep alterations are common in SCD and underline the clinical importance of screening in order to assess sleep alterations as well as improve sleep in average adults with SCD complaints. |
<b>Introduction</b>: Tactile sensitivity is impaired in older adults, which contributes to the loss of manual dexterity and mobility function. The reliability of classical psychophysical tests, such as two-point gap discrimination, has been questioned. Here we tested a new method to determine tactile acuity during dynamic touch, which is more functional than static touch. The aim was to validate a method providing a high level of discrimination of tactile acuity in the elderly. <b>Methods</b>: We tested the ability of subjects to evaluate the distance between bands printed on poly-methyl-methacrylate (PMMA) sheets. Pairs of sheets were compared in two groups of participants aged from 60 to 74 years; the test group was required to apply a cosmetic foam with an active ingredient on both their hands twice a day for 1 month, the control group had an identical task but used the same cosmetic foam without any active ingredient. The tests were run in a double-blind, placebo-controlled study. <b>Results</b>: The tactile discrimination threshold decreased by 83 μm after 1 month of cosmetic application in the group using the active ingredient, while it was unchanged in the control group. <b>Discussion</b>: The test presented here provided highly accurate results and should be useful to determine tactile performance. It allows the monitoring of tactile rehabilitation and/or skin treatments used to restore tactile acuity in the elderly. |
Middle cerebral artery occlusion (MCAO) induces ischemia characterized by a densely ischemic focus, and a less densely ischemic penumbral zone in which neurons and astrocytes display age-dependent dynamic variations in spontaneous Ca<sup>2+</sup> activities. However, it is unknown whether penumbral nerve cells respond to sensory stimulation early after stroke onset, which is critical for understanding stimulation-induced stroke therapy. In this study, we investigated the ischemic penumbra's capacity to respond to somatosensory input. We examined adult (3- to 4-month-old) and old (18- to 24-month-old) male mice at 2-4 h after MCAO, using two-photon microscopy to record somatosensory stimulation-induced neuronal and astrocytic Ca<sup>2+</sup> signals in the ischemic penumbra. In both adult and old mice, MCAO abolished spontaneous and stimulation-induced electrical activity in the penumbra, and strongly reduced stimulation-induced Ca<sup>2+</sup> responses in neuronal somas (35-82%) and neuropil (92-100%) in the penumbra. In comparison, after stroke, stimulation-induced astrocytic Ca<sup>2+</sup> responses in the penumbra were only moderately reduced (by 54-62%) in adult mice, and were even better preserved (reduced by 31-38%) in old mice. Our results suggest that somatosensory stimulation evokes astrocytic Ca<sup>2+</sup> activity in the ischemic penumbra. We hypothesize that the relatively preserved excitability of astrocytes, most prominent in aged mice, may modulate protection from ischemic infarcts during early somatosensory activation of an ischemic cortical area. Future neuroprotective efforts in stroke may target spontaneous or stimulation-induced activity of astrocytes in the ischemic penumbra. |
Hyperphosphorylation of the microtubule-associated protein tau and its resultant aggregation into neurofibrillary tangles (NFT) is a pathological characteristic of neurodegenerative disorders known as tauopathies. Tau is a neuronal protein involved in the stabilization of microtubule structures of the axon and the aberrant phosphorylation of tau is associated with several neurotoxic effects. The discovery of tau pathology and aggregates in the cortex of Temporal lobe epilepsy (TLE) patients has focused interest on hyperphosphorylation of tau as a potential mechanism contributing to increased states of hyperexcitability and cognitive decline. Previous studies using animal models of status epilepticus and tissue from patients with TLE have shown increased tau phosphorylation in the brain following acute seizures and during epilepsy, with tau phosphorylation correlating with cognitive deficits in patients. Suggesting a functional role of tau during epilepsy, studies in tau-deficient and tau-overexpressing mice have demonstrated a causal role of tau during seizure generation. Previous studies, analyzing the impact of seizures on tau hyperphosphorylation, have mainly used animal models of acute seizures. These models, however, do not replicate all aspects of chronic epilepsy. In this study, we investigated the effects of acute seizures (status epilepticus) and chronic epilepsy upon the expression and phosphorylation of tau using the intra-amygdala kainic acid (KA)-induced status epilepticus mouse model. Status epilepticus resulted in an immediate increase in total tau levels in the hippocampus, in particular, the dentate gyrus, and phosphorylation of the AT8 epitope (Ser202, Thr205), with phosphorylated tau mainly localizing to the mossy fibers of the dentate gyrus. During epilepsy, abnormal phosphorylation of tau was detected again at the AT8 epitope with lower total tau levels in the CA3 and CA1 subfields of the hippocampus. Chronic epilepsy in mice also resulted in a strong localization of AT8 phospho-tau to microglia, indicating a distinct pattern of tau hyperphosphorylation during chronic epilepsy compared to status epilepticus. Our results reaffirm previous observations of tau phosphorylation post-status epilepticus, but also elaborate on tau alterations in epileptic mice which more faithfully mimic TLE. Our results confirm seizures affect tau hyperphosphorylation, however, suggest epitope-specific phosphorylation of tau and differences in cell-specific localization according to disease progression. |
<b>Background:</b> Repetitive sit-to-stand (rSTS) is a fatigue perturbation model to examine the age-effects on adaptability in posture and gait, yet the age-effects on muscle activation during rSTS <i>per se</i> are unclear. We examined the effects of age and exhaustive rSTS on muscle activation magnitude, onset, and duration during ascent and descent phases of the STS task. <b>Methods:</b> Healthy older (<i>n</i> = 12) and younger (<i>n</i> = 11) adults performed rSTS, at a controlled frequency dictated by a metronome (2 s for cycle), to failure or for 30 min. We assessed muscle activation magnitude, onset, and duration of plantar flexors, dorsiflexors, knee flexors, knee extensors, and hip stabilizers during the initial and late stages of rSTS. Before and after rSTS, we measured maximal voluntary isometric knee extension force, and rate of perceived exertion, which was also recorded during rSTS task. <b>Results:</b> Older vs. younger adults generated 35% lower maximum voluntary isometric knee extension force. During the initial stage of rSTS, older vs. younger adults activated the dorsiflexor 60% higher, all 5 muscle groups 37% longer, and the hip stabilizers 80% earlier. Older vs. younger adults completed 467 fewer STS trials and, at failure, their rate of perceived exertion was ~17 of 20 on the Borg scale. At the end of the rSTS, maximum voluntary isometric knee extension force decreased 16% similarly in older and younger, as well as the similar age groups decline in activation of the dorsiflexor and knee extensor muscles (all <i>p</i> < 0.05). <b>Conclusion:</b> By performing 467 fewer STS trials, older adults minimized the potential effects of fatigability on muscle activation, voluntary force, and motor function. Such a sparing effect may explain the minimal changes in gait after rSTS reported in previous studies, suggesting a limited scope of this perturbation model to probe age-effects on muscle adaptation in functional tasks. |
<b>Background:</b> The intricate relationship between type 2 diabetes mellitus (T2DM) and Alzheimer's disease (AD) suggests that insulin is involved in modulating AD-related proteins. Alpha-lipoic acid (ALA) can improve insulin resistance (IR) in diabetic rats. However, the role of ALA in alleviating the cognitive decline of T2DM is not yet clear. This study examined the ameliorative effect of ALA on cognitive impairment, cerebral IR, and synaptic plasticity abnormalities in high-fat diet (HFD) plus streptozotocin (STZ) induced diabetic rats. <b>Methods:</b> The HFD/STZ-induced T2DM male Wistar rats were orally administered with ALA (50, 100, or 200 mg/kg BW) once a day for 13 weeks. Abilities of cognition were measured with a passive avoidance test and Morris water maze. Specimens of blood and brain were collected for biochemical analysis after the rats were sacrificed. Western blotting was used to determine protein expressions in the hippocampus and cortex in the insulin signaling pathways, long-term potentiation (LTP), and synaptic plasticity-related protein expressions. <b>Results:</b> Alpha-lipoic acid improved hyperinsulinemia and the higher levels of free fatty acids of the T2DM rats. Behavioral experiments showed that the administration of ALA improved cognitive impairment in HFD/STZ-induced T2DM rats. ALA ameliorated insulin-related pathway proteins [phosphoinositide 3-kinase (PI3K), phospho-protein kinase B (pAkt)/Akt, and insulin-degrading enzyme (IDE)] and the LTP pathway, as well as synaptic plasticity proteins (calmodulin-dependent protein kinase II, cyclic AMP response element-binding protein, and postsynaptic density protein-95) of the cerebral cortex or hippocampus in HFD/STZ-induced T2DM rats. <b>Conclusion:</b> Our findings suggested that ALA may ameliorate cognition impairment <i>via</i> alleviating cerebral IR improvement and cerebral synaptic plasticity in diabetic rats. |
Age-related deterioration of balance control is widely regarded as an important phenomenon influencing quality of life and longevity, such that a more comprehensive understanding of the neural mechanisms underlying this process is warranted. Specifically, previous studies have reported that older adults typically show higher neural activity during balancing as compared to younger counterparts, but the implications of this finding on balance performance remain largely unclear. Using functional near-infrared spectroscopy (fNIRS), differences in the cortical control of balance between healthy younger (<i>n</i> = 27) and older (<i>n</i> = 35) adults were explored. More specifically, the association between cortical functional activity and balance performance across and within age groups was investigated. To this end, we measured hemodynamic responses (i.e., changes in oxygenated and deoxygenated hemoglobin) while participants balanced on an unstable device. As criterion variables for brain-behavior-correlations, we also assessed postural sway while standing on a free-swinging platform and while balancing on wobble boards with different levels of difficulty. We found that older compared to younger participants had higher activity in prefrontal and lower activity in postcentral regions. Subsequent robust regression analyses revealed that lower prefrontal brain activity was related to improved balance performance across age groups, indicating that higher activity of the prefrontal cortex during balancing reflects neural inefficiency. We also present evidence supporting that age serves as a moderator in the relationship between brain activity and balance, i.e., cortical hemodynamics generally appears to be a more important predictor of balance performance in the older than in the younger. Strikingly, we found that age differences in balance performance are mediated by balancing-induced activation of the superior frontal gyrus, thus suggesting that differential activation of this region reflects a mechanism involved in the aging process of the neural control of balance. Our study suggests that differences in functional brain activity between age groups are not a mere by-product of aging, but instead of direct behavioral relevance for balance performance. Potential implications of these findings in terms of early detection of fall-prone individuals and intervention strategies targeting balance and healthy aging are discussed. |
Evidence exists for age-related decline in face cognition ability. However, the extents to which attentional demand and flexibility to adapt viewing strategies contribute to age-related decline in face cognition tests is poorly understood. Here, we studied holistic face perception in older (age range 65–78 years, mean age 69.9) and young adults (age range 20–32 years, mean age 23.1) using the complete design for a sequential study-test composite face task (Richler et al., ). Attentional demand was varied using trials that required participants to attend to both face halves and to redirect attention to one face half during the test (high attentional demand), and trials that allowed participants to keep a pre-adjusted focus (low attentional demand). We also varied viewing time and provided trial-by-trial feedback or no feedback. We observed strong composite effects, which were larger for the elderly in all conditions, independent of viewing time. Composite effects were smaller for low attentional demand, and larger for high attentional demand. No age-related differences were found in this respect. Feedback also reduced the composite effects in both age groups. Young adults could benefit from feedback in conditions with low and high attentional demands. Older adults performed better with feedback only in trials with low attentional demand. When attentional demand was high, older adults could no longer use the feedback signal, and performed worse with feedback than without. These findings suggest that older adults tend to use a global focus for faces, albeit piecemeal analysis is required for the task, and have difficulties adapting their viewing strategies when task demands are high. These results are consistent with the idea that elderly rely more on holistic strategies as a means to reduce perceptual and cognitive load when processing resources are limited (Konar et al., ).
## 1. Introduction
As it is found for other cognitive abilities, face cognition performance also undergoes age-related decline (Bartlett et al., ; Crook and Larrabee, ; Searcy et al., ; Pfutze et al., ; Chaby et al., ; Hildebrandt et al., ; Germine et al., ). Clearly, the well-documented decline in memory function with age may, at least partly, underlie the decline observed in face recognition tests (Fulton and Bartlett, ), and loss in general perceptual functions (Sekular and Sekular, ; Lott et al., ), speed limitations (Salthouse, , ), and top-down suppression and attentional control (Gazzaley et al., , , ) may contribute to these effects. Because there are multiple sources of age-related decline it is hard to judge whether impaired performance of the elderly is due to a decline in face-specific mechanisms or to impairment in general cognitive functioning, which is necessarily involved in face cognition tests.
Recent cross-sectional studies (Wilhelm et al., ; Hildebrandt et al., ) revealed that face cognition ability is predicted by non-facial general ability in memory function, speed, and object cognition with about 50% explained variance. The degree of predictability proved to be relatively stable across young, middle, and late adulthood, indicating no age-related dedifferentiation of face and non-face cognition (Hildebrandt et al., ). These findings suggest that the special status of face cognition, as a set of distinct abilities, is preserved in late adulthood.
A key feature that characterizes face cognition as a distinct and highly developed ability is its holistic nature. Processing of face parts is highly sensitive for the facial context such that a change of parts usually changes the overall appearance of a face. Striking demonstrations are the “part-to-whole effect” (Tanaka and Farah, ; Tanaka and Sengco, ) and the “composite effect” (Young et al., ). The part-to-whole effect shows that facial features are more easily identified when they appear in their natural face contexts. The composite face effect shows that upper and lower face halves interact perceptually, and cannot be judged independently. When two composite faces are shown that combine upper and lower face halves from different persons, observers have difficulty matching the identity of only the upper or lower halves (see example in Figure 3 ). Meanwhile, the composite effect is frequently used to assess holistic face processing (for an overview, see Rossion, ).
Aging studies have addressed whether age-related decline exists in the ability to apply holistic viewing strategies for faces. Corresponding to Wilhelm et al. ( ) and Hildebrandt et al. ( ), recent studies have corroborated that the integrative nature of face processing is not affected by aging. A study with young and older adults (mean age 68.6) found age-related decline in face identification was reported, but, no decline in the composite effect (Konar et al., ). Further, the composite effect predicted face identification performance to the same degree in both age groups, which indicates that the association of holistic face perception and face cognition is maintained at mature ages.
Meinhardt-Injac et al. ( ) studied how external features modulate the perception of internal features in young and older adults (mean age 70.4), and tested the accuracy of sequentially matching faces by attending either feature class. They found about equally strong holistic effects in both age groups. However, older adults performed better with external features, while the accuracy of assessing the inner face details declined. Daniel and Bentin ( ) recorded the face specific N170 potential and the P300 component to assess global, configural and featural face-processing strategies in younger and older adults (mean age 77.1). They found that older adults processed faces by relying on global features, which shows deficits in tasks that require local configural cues. Taken together, present evidence suggests that the elderly do not suffer from deficits in the ability to process faces holistically, but may have difficulties attending diagnostic facial cues. These findings point to the role of attentional capabilities.
Ample evidence exists that attentional selection of older adults is impaired in tasks with simultaneous presentation of target and non-target stimuli (Quigley et al., ; Schmitz et al., ). Further, serious age-related deficits have been reported for tasks that require one to change attentional focus during a trial (Georgiou-Karistianis et al., ). However, in tests of holistic face processing, researchers have assessed how unattended facial features affect the judgments of attended facial features. Holistic face processing is concluded from the failure to selectively attend to face parts (Richler et al., ). Therefore, the sensitivity of holistic face perception should be controlled in regard to the variation of attentional task demands. If higher attentional demands yield stronger holistic effects for older adults, then their preference for global and holistic viewing strategies would, at least partly, be due to age-related decline in attentional mechanisms (see Discussion).
The effect of attentional demand can be examined in a sequential study-test composite face task by varying the temporal position of the cue that indicates which of both face halves, the upper or lower, have to be attended (see Figure ). If the cue comes with the study image (Figure , upper panel) the observer can try to attend just the cued half and maintain the attentional focus throughout the trial. If the cue comes after the study image (Figure , lower panel) she/he must attend to both halves, and switch attention toward the target test half within the trial.
Examples of a single trial for the 1st cue (upper row) and the 2nd cue (lower row) condition . The white brackets informed the observer whether upper or lower face halves were to be compared. The upper row shows a same trial in congruent condition with upper target half and early cue, the lower row a different trial in incongruent condition with lower target half and late cue.
The effects of the unattended face halves are expected to be stronger in the late cue condition because the whole study face is attended. Both conditions differ in two respects, which are relevant for comparison among age groups. First, the late cue condition requires one to change the attentional focus from the whole study face toward only one half during the test. If age-related differences in attentional control and reallocation of resources modulate performance, the effect of cue position should be different in both age groups. Second, varying the temporal cue position alters not only attentional requirements, but also memory demands. In the late cue condition, features from the upper and lower halves must be encoded and held in memory until the test. If working memory load is crucial for performance in the composite face task, a differential effect of cue position in both age groups should also exist because working memory capacity differs strongly among young and old adults (Brockmole and Logie, ). Hence, varying the temporal cue position can reveal whether age-related differences in coping with increased task demands in the composite face task.
A further aspect is the role of cognitive control that may be used to regulate the influence of the unattended on the attended facial features. Meinhardt-Injac et al. ( ) observed that young adults could enhance accuracy in judging the identities of internal features in the presence of incongruent external features by about 10% when trial-by-trial feedback about correctness is provided. This result was stable for exposure durations between 200 and 650 ms, indicating that young adults are able to replace holistic features by piecemeal face processing if they have sufficient temporal resources and the opportunity to adjust their viewing strategy with the help of feedback. For older adults, the role of feedback for optimizing the face viewing strategy, to date, has not been addressed.
The focus of the present study was threefold. First, we remeasured the composite effect for young and older adults because the current state of evidence for maintenance of holistic face perception at mature ages is not yet settled. In recent studies (Boutet and Faubert, ; Konar et al., ), the composite face effect was examined by comparing aligned and misaligned composite face arrangements, as in the seminal study on the composite effect (Young et al., ). However, in the last years, there was progress in the methodological development of the composite face paradigm, leading to a fully balanced and complete design (Gauthier and Bukach, ; Cheung et al., ). We decided to use this new design because of its methodological advantages (see Methods) to first add results on aging effects with the complete design to the literature. Second, we varied task demands, allowing the observer to select the attentional focus in advance and maintain it throughout the trial, or to force her/him to reallocate attentional resources during a brief time interval. Comparing across age should dismantle age-related capabilities and limitations in coping with higher task demands. Third, we provided trial-by-trial feedback, or not, to reveal whether older adults are able to use higher-level cognitive control to learn and refine their viewing strategies in the same way as young adults.
## 2. Materials and methods
### 2.1. Experimental outline
We used a variety of the sequential composite face tasks (Richler et al., ). In the experimental trials, subjects first fixated on the screen center and then saw a composite study face. The image remained on the screen for 800 ms. After masking with a carefully designed mask pattern (see below), another blank screen interval followed, and then the composite test face was presented for one of three possible presentation times chosen at random. Subjects then decided by button press whether the study and test agreed or disagreed in the face halves that were being attended (upper or lower). In the first cue condition, a large white bracket marking the face half to be attended was shown with the study image. In the second cue condition, the bracket appeared after the study image, together with its subsequent mask (see Figure ).
Cue position conditions were run in separate experimental blocks because pilot measurements showed that the task was too hard for the elderly if the target cue position was varied randomly interleaved. Each experimental block was run with acoustical trial-by-trial feedback about correctness and without. Three exposure durations were chosen for the test image, one brief timing precluding saccades and serial scans (50 ms), an intermediate timing (233 ms), and a relaxed timing (633 ms) to allow for detailed image scrutiny.
### 2.2. Experimental design
We employed the “complete design” (CD) of the composite face paradigm (Gauthier and Bukach, ; Cheung et al., ). In contrast to a former variety (called the “partial design,” PD, by Cheung and colleagues) congruent and incongruent face half pairings are fully balanced in the CD, and performance in terms of accuracy as well as holistic effects are calculated from both response categories in order to avoid confounds with a possible preference (bias) toward either response category. The design is illustrated in Figure . Same-trials and different-trials are realized in the congruent and the incongruent variety. In congruent trials, the non-attended halves agree when the attended halves agree (same-trial), and disagree when the attended ones disagree (different-trial). This means that attended and non-attended halves are congruent with respect to the correct decision. In incongruent trials, however, the unattended halves disagree when the attended halves agree (same-trial), and agree when the attended ones disagree (different-trial). Hence, attended and unattended halves are incongruent with regard to the correct decision. Holistic effects are operationally defined as congruency effects , reflecting the performance difference achieved in congruent and incongruent trials (see Performance Measures) .
Overview of the trial types used in the complete design with upper and lower face halves as the targets . Equal halves of a composite face pair are marked by same letters and same levels of gray.
### 2.3. Stimuli
Photographs of 20 male models were used for stimulus construction (see Figure for examples). These were frontal view shots of the whole face, captured in a professional photo studio under controlled lighting conditions. The original images were edited with Adobe Photoshop CS4 to generate the set of stimuli used in the experiment. Photographs were initially converted to 8 bit grayscale pictures and superimposed with an elliptical frame mask to obliterate all external facial features such as hair, ears, or chin line. The elliptical cutouts were then split horizontally at the bridge of the nose, thus yielding 20 upper and 20 lower face halves. Each upper half was recombined with three lower halves to constitute the final set of 60 compound faces. The cutline between the face halves was concealed with a white bar of 5 pixels thickness. It was warranted that any upper face part was never recombined with the lower half of the same original face. In addition, each of the twenty lower and upper halves appeared exactly three times in the final set of stimuli. Stimulus size was 250 × 350 pixels (width × height), which corresponded to 10 × 12.5 cm of the screen. For each face stimulus a corresponding mask was constructed by sampling randomly ordered 5 × 5 pixel blocks from the face image. Masks subtended 350 × 450 pixels (width × height), and covered the whole region where two subsequent face stimuli were displayed.
Stimulus example for upper face half comparison in incongruent trials (mid row of Figure ) . The left composite face pair shows same upper halves combined with different lower halves, the right one shows different upper halves combined with same lower halves.
### 2.4. Subjects
Overall, 46 young adults and 40 senior subjects participated in the present study. The two samples were halved, one group participated in the experiment with feedback, the other without feedback. All participants had normal or corrected to normal vision and reported normal neurological and psychiatric status. Senior subjects lived independent lives and were paid for participation.
The mini-mental state examination (MMSE; Folstein et al., ) was used to evaluate mental status. Young adult subjects were undergraduate students, 20% were male and 80% female. The mean age of the student group was 23.1 (range 20–32). These participants were given course credit points for participation, or received payment. Senior subjects were assigned to the feedback and no-feedback groups in a pseudo-random procedure with the constraint to keep the age structure of the groups equivalent. Feedback group: 20 subjects (11 female; mean age = 69.7; range 65–78 years), and No feedback group: 20 subjects (14 female; mean age = 70.1; range 65–77 years). All subjects were naive with respect to the purpose of the experiment. The study was conducted in accordance with the Declaration of Helsinki. In detail, subjects participated voluntarily and gave written informed consent to their participation. In addition, participants were informed that they were free to stop the experiment at any time without negative consequences. The data were analyzed anonymously.
### 2.5. Apparatus
The experiment was executed with Inquisit runtime units. Stimuli were displayed on NEC Spectra View 2040 TFT displays in 1280 × 1024 resolution at a refresh rate of 60 Hz. Screen mean luminance L was 100 cd/m at a michelson contrast of ( L − L )/( L + L ) = 0.98, therefore the background was practically dark (about 1.4 cd/m , measured with a Cambridge Research Systems ColorCAL colorimeter). No gamma correction was used. The room was darkened so that the ambient illumination approximately matched the illumination on the screen. Stimuli were viewed binocularly at a distance of 70 cm. Subjects used a distance marker but no chin rest throughout the experiment. Stimuli were viewed at 70 cm viewing distance. Subjects responded via an external key-pad, and wore light headphones for acoustical feedback in the feedback condition.
### 2.6. Preparation and preliminary measurements
Preliminary measurements were taken with four senior subjects to assure that the task could, in principle, be executed by the elderly, and to determine the proper exposure durations for the test stimuli. Several exposure durations were probed to find a relaxed timing that allowed for maximum performance of senior subjects under the experimental conditions with the lowest attentional and perceptual demands (i.e., for the target cue with the study image, providing feedback, and for congruent trials). It turned out that senior subjects could respond to these trials with about 90% correctness at test stimulus exposure durations of half a second and longer.
Enlarging exposure duration to about a second did not increase accuracy any further. Note that 90% correct corresponded to only three errors out of 32 replications. We then presented incongruent and congruent trials mixed in random order, which did not lead to a stronger decline in accuracy for the congruent trials when exposure durations of well beyond 500 ms were used. We decided to use 633 ms (36 frames of the monitor at 60 Hz refresh rate) as the largest exposure duration.
### 2.7. Procedure
Subjects were informed that face pairs could differ in the cued halves, but also in non-cued halves, and face halve comparison was to be done for just the cued halves. They were also instructed to compare the face halves as accurately as possible, without speed pressure for the response. The temporal order of events in a trial sequence was: fixation mark (750 ms), blank (300 ms), study face stimulus (800 ms), mask (400 ms), blank (800 ms), test face stimulus (50, 233, or 633 ms), mask (400 ms), and blank frame until response (see Figure ).
In the 1st cue condition a rectangular bracket marking the target face half was shown simultaneously with the study face, and remained until the test face was masked. In the 2nd cue condition the cue presentation began with the mask of the study face. Stimulus position jittered randomly within a region of ±50 pixels around the center of the screen to preclude image region matching strategies between two subsequent stimulus presentations.
Young adults were made familiar with the task by going through randomly selected probe trials to ensure that the instruction was understood and could be put into practice. Senior subjects were carefully prepared for the experiment. First, the researcher explained the sequential composite face task using paper print examples of the stimulus pairings. To ensure that subjects understood the composite face task with incongruent face halve pairings, the experimenter displayed paper prints of 10 stimulus pairs, and asked participants to name the five pairs showing objects with the same upper (lower) halves and five showing different upper (lower) halves. Subjects were given as much time as needed to label the 10 pairs. If errors occurred, the experimenter adverted to the wrongly labeled pairs and drew attention to just the halves to be compared. The first minutes at the computer were spent on just congruent trials presented with the longest viewing time (633 ms), which all subjects could do with good accuracy. They then saw probe trials of the experiment with congruent and incongruent trials for about 8 min. After the preparation phase, the experimental blocks started.
Each subject went through 2 (cue position) × 2 (congruency) × 3 (duration) = 12 conditions. Each condition was measured with 16 same- and 16 different- trials. Eight of these 16 replications were done with upper half, and 8 with lower half as the target, resulting in 384 trials. These were subdivided into a block of 192 trials where the target cue came at the first position and a block of 192 trials where the cue came at the second position. Going through a block took about 20 min. Interleaved by a brief pause, the two blocks were administered on a single day, one with 1st cue, and one with 2nd cue, in random order across subjects.
### 2.8. Performance measures
Accuracy was measured in terms of the proportion of correct judgments, P . The rates were calculated from the frequencies of correct “same” [ h ] and correct “different” [ h ] judgments, i.e., P = ( h + h )/( n + n ). With n = n = 16 replications per trial, each proportion correct datum rested on n = 32 trials. Congruency effects were calculated as the difference
Originally, Cheung et al. ( ) referred to the d ′ measure as a bias-free measure. We used the proportion correct measure, because, as d ′, proportion correct also derives from the performance achieved for both response alternatives. However, it avoids hypothetical assumptions about sensory mapping of face stimuli, and the distribution of the corresponding sensory states. Further, it reflects task difficulty on a direct and intuitive scale.
Moreover, a direct and intuitive measure of response bias can be defined by referring to the relative frequencies for the errors of both kinds (Meinhardt-Injac et al., ). For the same/different experiment the “same” response category is commonly defined as the target category (e.g., Richler et al., ). Accordingly, hit-rate (Hit) was defined as the rate of correctly identifying same target halves and correct rejection rate (CR) was defined as the rate of correctly identifying different target halves. False alarm rate (FA) and the rate of misses (Miss) were defined as being the complementary rates to CR and Hit, respectively. We measured response bias in terms of the error proportion, Q , which indicates which of both errors is more likely:
If Q = 0.5, then both kinds of errors are made with the same frequency. A ratio of Q > 0.5 indicates a tendency to say “different” while Q < 0.5 indicates a preference toward “same” responses. The Q - measure has the advantage that it easy to interpret. For example, a value of Q = 0.7 means that 70% of all errors are wrong “different” responses and 30% are wrong “same” responses .
### 2.9. Data analysis
The proportion correct data and the Q - measure were analyzed with ANOVA, having feedback (FB) and age group (Age) as grouping factors and cue position (Cuepos), congruency (Congru) and exposure duration (Time) as repeated measurement factors. We do not report ANOVA results for the CE measure, since the results for the difference measure are already included in the results for all interactions involving congruency at the original P data.
## 3. Results
Figure shows the mean proportion of correct responses as a function of exposure duration for all experimental conditions. Generally, both younger and older adults reached good accuracy levels above 90% correct at intermediate (233 ms) and large (633 ms) viewing times for congruent trials. For incongruent trials, performance did not come close to these levels, and even declined. Hence, a large congruency effect was found in all experimental conditions, which became obvious by the space between the black and gray curves.
Mean proportion correct rates as a function of exposure duration for the two age groups with feedback (upper panels) and without (lower panels), and target face half cue given at study image (1st cue, left panels) and before test image (2nd cue, right panels) . Data for the congruent trials are shown as open black circles, gray symbols indicate data for incongruent trials. Error bars indicate 95% confidence limits of the means.
Data analysis using ANOVA revealed main effects of exposure duration [ F = 69.08, p < 0.001], congruency [ F = 191.10, p < 0.001], cue position [ F = 89.65, p < 0.001], and age group [ F = 64.04, p < 0.001], but no main effect of feedback [ F = 3.02 × 10 , p = 0.986]. We explored these effects further explored by analyzing first and higher order interactions.
### 3.1. Effects of feedback and cue position
There was no main effect of feedback, and no interaction of feedback with age [ F = 0.62, p = 0.434]. Hence, feedback did not change the general level of performance in both age groups. However, feedback substantially modified the effect of congruency [congruency × feedback, F = 10.18, p < 0.002, see below], and the effect of cue position [cue position × feedback, F = 4.66, p < 0.04]. However, the latter was further moderated by age group [cue position × feedback × age group, F = 6.28, p < 0.02]. Figure illustrates this interaction.
Cell means plot for the proportion correct measure, illustrating the cue position × feedback × age group interaction . Data for the feedback condition are shown as open black circles, gray filled circles indicate data for the no feedback condition. Error bars indicate 95% confidence limits of the means.
For young adults the effects of feedback were the same in both cue positions. For older adults, performance in the 2nd cue condition was disproportionately worse with feedback. This finding was confirmed by pairwise Fisher LSD post-hoc tests. Older adults did not perform significantly different in both feedback conditions at the 1st cue position (Δ P = 0.023, p = 0.373), but significantly worse with feedback at the 2nd cue position (Δ P = 0.052, p < 0.04). Exploring the role of trial type showed the same performance for both feedback conditions in incongruent trials (Δ P = 0.019, p = 0.582), but worse performance with feedback in congruent trials (Δ P = 0.085, p < 0.02; see also right panels of Figure ). At the 1st cue position older adults performed better with feedback than without in incongruent trials (Δ P = 0.064, p < 0.05), and the same in both feedback conditions in congruent trials (Δ P = 0.018, p = 0.328). This finding indicates a paradox effect of feedback in the old age group for the condition with high attentional demand. For young adults, the same results scheme for the effects of feedback was found for the 1st cue and the 2nd cue position. These participants performed better with feedback in incongruent trials (1st cue: Δ P = 0.041, p < 0.05; 2nd cue: Δ P = 0.054, p < 0.04) and the same with and without feedback in congruent trials (1st cue: Δ P = 0.015, p < 0.40; 2nd cue: Δ P = 0.02, p = 0.516).
Figure illustrates that cue position modified performance strongly, which led to significantly lower performance in the 2nd cue condition. The effect of cue position was not modulated by age [cue position × age group, F = 2.71, p = 0.104], however, it was by age and feedback (see above). Table shows the effects of cue position for both age groups and feedback conditions and their effect sizes (Cohen's d effect size measure). The data show that cue position had large effects of comparable sizes in both age groups in the no feedback condition. Adding feedback did not affect much for young adults, but more than doubled the effect for older adults, both in the accuracy measure, and in effect size.
Effects of cue position for the two age groups and feedback conditions .
The table shows the accuracy difference for 1st and 2nd cue position, F - value, significance level, and Cohen's d measure.
### 3.2. Congruency effects
Variation of the congruency relation (congruent/incongruent) among face halves strongly modulated performance. With respect to age group we found larger congruency effects for older adults [congruency × age group, F = 5.34, p < 0.02]. Comparing across age for congruent and incongruent trials separately with Fisher LSD post-hoc tests showed that young adults were better than older adults particularly in incongruent trials (congruent: Δ P = 0.096, p < 0.001; incongruent: Δ P = 0.146, p < 0.001). This finding indicates age-related differences in the ability to suppress incongruent facial context.
Feedback strongly modified the effect of congruency [congruency × feedback, F = 10.18, p < 0.002]; the congruency effect was strongly attenuated when feedback was provided, which is readily seen when the space between the black and the gray curves shown in Figure is compared among the upper and the lower data panels. Further, cue position strongly modulated the congruency effect [cue position × congruency, F = 13.48, p < 0.001], which reflects larger congruency effects for the 2nd than for the 1st cue position.
Interestingly, no higher interactions were found with age group, indicating that the congruency effect was modulated by feedback and cue position in the same way for younger and older adults [congruency × feedback × age group, F = 0.05, p = 0.819; cue position × congruency × age group, F = 0.03, p = 0.853]. Table lists the congruency effects of both age groups, for both cue positions and feedback conditions. The data reflect that older adults had consistently larger congruency effects than did the young adults, in the order of magnitude of 5% (see last column). The table also shows that the modulating effects of feedback and cue position on the congruency effect were the same in both age groups, and in the range of 4–6% (cue position), and 5–8% (feedback), respectively.
Congruency effects for the two target cue positions and the two feedback conditions, for both age groups .
The table shows mean congruency effects, the difference of the mean congruency effects for both cue positions, both feedback conditions, and the difference of congruency effects across age group for the same conditions.
### 3.3. Effects of exposure duration
The effect of exposure duration was different in the two age groups [exposure duration × age group, F = 9.14, p < 0.001], with smoothly rising performance across viewing times for young adults, while performance showed stronger improvement with viewing time for older adults. There were no time-related effects of feedback, cue position, or congruency, which indicates that all these effects were relatively constant across exposure duration. There was time related effect that concerned the congruency effect at the two cue positions [cue position × congruency × exposure duration, F = 3.45, p < 0.04]. This effect reflected that congruency effects tended to decline with increasing viewing time when the cue came at the 1st position, while congruency effects tended to increase with exposure duration when the cue came at the 2nd position. No age-related differences were indicated for this effect by statistical testing [cue position × congruency × exposure duration × age group, F = 0.29, p = 0.746].
### 3.4. Response bias
Figure shows the data for the Q - measure. ANOVA revealed main effects of age group [Δ Q = 0.08, F = 12.05, p < 0.001], congruency [Δ Q = 0.09, F = 101.6, p < 0.001], and feedback [Δ Q = 0.07, F = 9.55, p < 0.001], but no effects of cue position [ F = 1.38, p = 0.24] and exposure duration [ F = 0.36, p = 0.70]. Young adults tended to prefer the “different” response category [ Q = 0.53, CI = [0.50, 0.56]], while older adults preferred “same” responses [ Q = 0.45, CI = [0.42, 0.48]]. The Q - measure was consistently larger in incongruent trias, compared to congruent trials [ Q ( IC ) = 0.54, CI = [0.51, 0.56], Q ( CC ) = 0.45, CI = [0.42, 0.47]], and also consistently larger in the no feedback condition, compared to the feedback condition [ Q ( NoFB ) = 0.53, CI = [0.49, 0.56], Q ( FB ) = 0.45, CI = [0.42, 0.48]]. There was a significant interaction of congruency and feedback [ F = 4.87, p < 0.03], which indicated that the difference in the Q - measure for congruent and incongruent trials was stronger in the no feedback condition, compared to the feedback condition (see Figure ). Young adults showed a strong response bias toward “different” responses in incongruent trials when there was no feedback (see Figure ). The bias vanished when feedback was provided. Older adults did not prefer “different” responses in any experimental condition.
Error proportion, Q , for assessing response bias . Data for congruent trials are shown as open black circles, gray filled circles indicate data for incongruent trials. Values of Q > 0.5 indicate a bias toward “different” responses, and values of Q < 0.5 toward “same” responses. Error bars indicate 95% confidence limits of the means.
## 4. Discussion
We studied holistic face perception with the complete design of the composite face paradigm, to explore the particular role of attentional demand, feedback, and viewing time, and to compare these factors across younger and older adults. Younger adults could do the study-test composite face task at brief timings (50 ms exposure duration) and at good performance levels. Older adults started at lower levels for the shortest timing, but well above chance, and reached good performance of about 90% accuracy at relaxed viewing times (633 ms).
We obtained strong congruency effects in all experimental conditions, which were consistently larger for older adults. Age-related differences were particularly pronounced for incongruent trials. However, the modulation of congruency effects by feedback and attentional demand was highly similar in both age groups. Generally, congruency effects were strongest when subjects were forced to change their attentional focus within a trial, and when no feedback was provided. A strong interaction of attentional demand, feedback, and age group was observed. Young adults could exploit trial-by-trial feedback to improve performance in incongruent trials with high and low attentional demand. Older adults could do so only in trials with low attentional demand. When participants were forced to reallocate attentional resources within a trial, performance was worse with feedback than without.
Analysis of response bias revealed a tendency of older adults toward preferring “same” responses, while young adults were slightly biased toward “different” responses. Feedback led toward more frequent “same” responses in both age groups.
### 4.1. No age-related decline in congruency effects
One aim of this study was to re-examine age-related changes in the congruency effect as an important hallmark of perceptual integration in face perception. We obtained consistently larger congruency effects for the elderly in all experimental conditions. Further, the strong performance difference of congruent and incongruent trials was observed in both age groups at brief timings of 50 ms, and remained for more relaxed timings. Hence, no indication was found of age-related decline in the general capabilities to view faces holistically. In line with recent results (Daniel and Bentin, ; Konar et al., ; Meinhardt-Injac et al., ), our results suggest the elderly prefer global and holistic viewing strategies, albeit part-based viewing strategies are more effective for task success.
### 4.2. Effects related to task demands
Face half comparisons are more difficult in the 2nd cue condition, since a late cue enforces fast reallocation of resources (Greenwood and Parasuraman, ; Georgiou-Karistianis et al., ). A second reason for higher task difficulty in the late cue condition is enhanced demand for encoding and fast retrieval from working memory. When the cue comes with the study image observers can encode only the face half of interest and compare it to the target test half, while trying to ignore the non-target half. When the cue comes late it is not possible to proceed this way, and the observers must encode information of both halves at study.
Both attentional control and working memory are known to be affected by aging. Several studies have shown that the elderly operate much worse than young adults in tasks that require attentional switch (Lincourt et al., ; Greenwood and Parasuraman, ; Vanneste and Pouthas, ; Georgiou-Karistianis et al., ). Using Navon-like stimuli and task (Navon, ), Georgiou-Karistianis and colleagues showed that older adults exhibited a similar global precedence effect as young adults, but they performed worse when a switch from global to local or from local to global was required. In contrast, young adults exhibited only moderate or no switching costs. Age-related decline in working memory is a well-established finding that is substantiated by many studies (for a review, see Rajah and D'Esposito, ). Both the decline in working memory function and loss of attentional control can be understood within the framework of the frontal lobe hypothesis of aging (West, ), because divided attention, attentional and executive control, and working and episodic memory were found to be mediated by frontal brain areas (Goldman-Rakic, ; Cabeza et al., ; Fink et al., ; Rajah and D'Esposito, ; Prakash et al., ). From these results it can be expected that the combined effects of higher attentional demands and stronger working memory requirements in the late cue condition should disproportionately affect the performance of older adults. Interestingly, our results do not support a disproportionate age-related decline of performance in the 2nd cue condition.
As outlined in the Results section (see Figure and Table ) the effect of cue position was the same in both age groups, as long as there was no feedback. The effect of cue position was larger for older adults only in the feedback condition, for specific reasons (see below). Hence, the increase of task demands in the 2nd cue condition compared to the 1st cue condition affected performance of young and older adults to the same degrees. This finding indicates that younger and older adults handled increased task demands equally well. In view of the fact that cue position modulated task difficulty strongly, this finding is at odds with expectation from the known aging effects on working memory function and attentional control.
We also found that the effect of cue position on the congruency effect was not different for young and older adults (see Congruency Effects). Increased task demands strengthened the influence of the unattended face halves, in the same way for both age groups. The surprising fact that both performance and congruency effects of older adults were not disproportionately affected by the much higher task demands in the late cue condition points to a potential benefit of holistic encoding, which might have been used as a strategy. Holistic encoding spares the costs of divided attention to lower and upper halves at study, which precludes the effects of restricted capabilities in divided attention to become effective (Greenwood and Parasuraman, ). However, the encoding advantage is at the costs of having to recall the diagnostic features of just one half from a holistic representation, which results in stronger interference among target the half and incongruent non-target half. Accordingly, an increase of contextual interference for 2nd cue trials should result, which was indeed observed.
The composite face task was generally much more difficult for the elderly, as indicated by the strong main effect of age. One likely reason why face comparisons were more difficult for older adults is the use of elliptical frames that leave only the inner face parts and mask global face shape and further external features. Meinhardt-Injac et al. ( ) used full and intact face stimuli, and had subjects attend to either the internal or external features. They found that older adults were nearly as good as young adults in comparing external features, but were much worse when internal features were the focus. This finding indicates that global face shape is a relevant face identity cue for the elderly (see below).
### 4.3. Response bias
A considerable advantage of the CD compared to the PD is that the CD is fully balanced with respect to congruency relation and the number of same and different face halves (Richler et al., ). Thus, the CD avoids that response bias is induced due to methodological artifacts. Analysis of response preferences can therefore reveal true age-related differences, as well as influence of experimental conditions on decision behavior. In this study we found evidence for different response behavior in both age groups, and modulatory influence of feedback and congruency relation, but no influence of task demands and exposure duration. Young adults strongly preferred the “different” response in incongruent trials when there was no feedback. The bias toward “different” responses was found in several studies using the CD (Cheung et al., ; Richler et al., ; Gao et al., ), and might indicate that the difference of the wholes and the unattended parts bias the observer toward responding “different,” albeit the attended parts are same (Gao et al., ). Interestingly, trial-by-trial feedback canceled this effect. With the help of feedback young observers noticed that they relied on the wrong features, and they could revise their decisional strategy. This is in line with the observation that feedback helped to improve young adults' performance in incongruent trials. Older adults, in contrast, did not show a “different” bias in any experimental condition. While they responded “different” more often in incongruent trials, compared to congruent trials, they stayed generally biased toward “same” responses. With feedback the overall preference toward “same” responses even increased. The general “same” bias might indicate that elderly tend to overlook local diagnostic features that are crucial for facial comparisons. This is supported by earlier and recent findings which show that older adults tend to more likely identify new faces as previously seen ones (Bartlett et al., ; Fulton and Bartlett, ; Lee et al., ). In a recent aging study of Konar et al. ( ) no response bias was found for young and older adults. However, the authors used the PD and concluded holistic processing from the difference achieved with aligned and misaligned presentation. This might account for differences of their results and the findings of this study.
### 4.4. The paradox effect of feedback in the older adults group
Perceptual learning studies have found that feedback enables observers to revise and to optimize their viewing strategies (Herzog and Fahle, , ). Face perception studies have also found that young adults identify diagnostic facial features and regulate the influence of irrelevant context with the help of feedback (Meinhardt-Injac et al., ). The results obtained here show that feedback had exactly this effect for young and older adults, as long as task demands were moderate. In the late cue condition young adults were still able to benefit from feedback, particularly in the incongruent trials. In contrast, the performance of older adults was not better with feedback in incongruent trials, while performance in the easier congruent trials declined (see Results). Seemingly, older adults were confused by the feedback signal in the late cue condition, and failed to establish a correlation of strategy revision and success. At the same time, the lower performance levels of older adults indicate that they experienced high task difficulty (see Figure ). This finding corresponds to an interaction of task difficulty and learning observed in perceptual learning (Ahissar and Hochstein, , ). When task difficulty is high, learning usually does not occur, even when external markers are provided. Subjects need some easy trial instances to initiate learning (“eureka effect,” see Ahissar and Hochstein, ). Hence, the inability to benefit from feedback in the condition with the highest task demands may indicate an interaction of learning and task difficulty for the elderly. This effect should not be over-estimated, as it is observed for the first time in the context of the composite face task. However, it would be interesting to see whether the effect is also obtained with non-face objects because older adults do not seem to apply global viewing strategies (Meinhardt-Injac et al., ). As the stronger congruency effects for older adults indicate, it is adherence to global viewing strategies that is in conflict with feedback. The difficulties of elderly to replace a global viewing strategy with a more effective piecemeal strategy when task demands are high is in line with recent claims that older adults use holistic processing as a strategy to reduce perceptual and cognitive load (Dror et al., ; Konar et al., ).
### 4.5. How do elderly look at faces?
Looking at the composite effects for the elderly (see Table ) shows that the influence of unattended face halves in the feedback condition is still as great as for young adults in the no feedback condition. Therefore, the general level of contextual influence remains high for older adults, even in conditions that are optimal for setting up a piecemeal viewing strategy.
The large global-contextual influence for older adults indicates that age-related decline in face perception does not concern mechanisms of perceptual integration. Rather, the elderly suffer from deficits when analytical processing of faces and control of facial context is required. Further evidence that face-specific processing is intact in older adults comes from the face inversion effect (FIE, Yin, ). Comparing across the life span, Germine et al. ( ) reported that the FIE in a face recognition task gradually increases up to ages 62 years, indicating that the experience dependent advantage of upright face processing is not lost in mature ages. Murray et al. ( ) found that elderly were much more vulnerable to face rotation than were young adults, which indicates that they strongly rely on configural information of facial features. Similar findings were reported by Creighton et al. (submitted). For older adults accuracy, response latency, and intensity rating for facial expressions of anger, happiness, fear and sadness were notably impaired when faces were turned upside down. Inversion effects for young adults were much smaller (fear, sadness) or even absent (anger, happiness).
Comparing the FIE for horizontal (eye distance) and vertical (eye-mouth distance) relational face manipulations across age, Chaby et al. ( ) observed that the FIE for vertical-relational manipulations was preserved in the elderly, while the FIE for horizontal-relational manipulations was lost. However, the overall accuracy level was lower than for young adults in detecting vertical relational changes. Obermeyer and colleagues obtained similar findings concerning age-related decline in face recognition with images that contained only horizontal spatial frequency information (Obermeyer et al., ). They also found a strong FIE of more than one d ′ unit in both age groups for this type of image manipulation. The strong FIE for vertical-relational manipulations, together with the loss of the FIE for horizontal-relational manipulations is diagnostic of the facial cues preferred by older adults. Eye distance (horizontal) is a local-relational feature judged relatively independent of facial context (Leder et al., ). In contrast, eye height (vertical) is defined in terms of its distance to the mouth, forehead and face outline, and is a global, long-range relational feature (Sekunova and Barton, ; Meinhardt-Injac et al., ). Chaby et al. ( ) reported a strong age-related decline in assessing local-configural facial features, while global-configural features could still be assessed. This finding is in-line with Daniel and Bentin ( ), who recorded the face specific N170 potential and the P300 component to reveal global, configural and featural face-processing strategies. Daniel and Bentin ( ) found that older adults relied on distal global information, and tended to process faces merely at the basic level of categorization until identification was required. Moreover, the elderly did not apply configural information by default, and showed deficits in subordinate categorization (gender classification based on internal features), which strongly relies on local-configural cues. Recent results from sequential same/different tasks with whole or just part-based agreement in external and internal features showed that the elderly rely more on global shape information than do young adults, and they experience deficits in judging inner face details (Meinhardt-Injac et al., ). Also the finding of a global bias toward “same” responses indicates that elderly have difficulties to focus the diagnostic features when they compare faces. These results, together with the findings of a less flexible handling of viewing strategies show that the elderly generally process faces holistically, but suffer from losses in assessing local-configural features, particularly when maintenance of attentional focus is impeded by the complexity of the visual task.
## Author contributions
All authors contributed equally to the conceptualization of the study. Bozana Meinhardt-Injac set up the basic design. Malte Persike conducted the experiments and data preparation. Günter Meinhardt contributed data analysis and interpretation. All authors were involved in writing, preparation of the manuscript and final approval. All authors agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are investigated and resolved appropriately.
### Conflict of interest statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
It is common wisdom that practice makes perfect; but why do some adults learn better than others? Here, we investigate individuals’ cognitive and social profiles to test which variables account for variability in learning ability across the lifespan. In particular, we focused on visual learning using tasks that test the ability to inhibit distractors and select task-relevant features. We tested the ability of young and older adults to improve through training in the discrimination of visual global forms embedded in a cluttered background. Further, we used a battery of cognitive tasks and psycho-social measures to examine which of these variables predict training-induced improvement in perceptual tasks and may account for individual variability in learning ability. Using partial least squares regression modeling, we show that visual learning is influenced by cognitive (i.e., cognitive inhibition, attention) and social (strategic and deep learning) factors rather than an individual’s age alone. Further, our results show that independent of age, strong learners rely on cognitive factors such as attention, while weaker learners use more general cognitive strategies. Our findings suggest an important role for higher-cognitive circuits involving executive functions that contribute to our ability to improve in perceptual tasks after training across the lifespan.
## Introduction
Successful interactions in dynamic environments are known to benefit from past experience. But why do some adults learn better than others? Despite the general consensus that practice makes us “perfect”, the striking variability in learning ability among individuals remains largely unexplained (Ackerman, ; Saarinen and Levi, ; Withagen and Van Wermeskerken, ). Previous behavioral, neurophysiology and neuroimaging studies (for reviews, see e.g., Gilbert et al., ; Fine and Jacobs, ; Kourtzi and DiCarlo, ) have advanced our understanding of the learning mechanisms that facilitate behavioral improvements through training; yet the socio-cognitive factors that underlie individual variability in learning ability remain largely unknown.
In this study, we sought to understand the roles of cognitive and social capacities that may underlie individual variability in our ability to improve in perceptual tasks through training (cf. Hutchens et al., ). Training is shown to facilitate performance in a wide range of perceptual skills in both young (Fine and Jacobs, ; Sagi, ) and older adults (Ball and Sekuler, ; Richards et al., ; Andersen et al., ; Bower and Andersen, ). For instance, recent studies show that training enhances performance on a wide range of tasks, including brightness discrimination (Ratcliff et al., ), acuity (Fahle, ), texture discrimination (Andersen et al., ), motion direction discrimination (Ball and Sekuler, ; Bower and Andersen, ; Bower et al., ) and global form perception tasks (Kuai and Kourtzi, ).
To understand how learning improves our ability to recognize objects, we trained young and older participants on a global form discrimination task that entails extracting task-relevant information from distracting background noise similar to identifying a friend in the crowd or a familiar object in a cluttered scene. In particular, we used parametric manipulations of Glass patterns that comprise oriented dot dipoles (Figure ). For these stimuli, small local changes to dot patterns have a predictable influence on the perception of global forms (concentric vs. radial patterns). Further, adding background noise (i.e., randomly oriented dipoles) increases the difficulty of the task and worsens our ability to discriminate between global patterns. Our previous work (Kuai and Kourtzi, ) has shown that training on this task improves global form discrimination in both young and older adults. However, tolerance to external noise varies across individuals, especially in older age, suggesting that visual selection processes may impose limits to perceptual learning and result in individual variability. Thus, we predict that variability in perceptual learning tasks may relate to cognitive (i.e., attentional, memory) skills that facilitate extracting relevant information while suppressing distracting patterns.
Example stimuli . Examples of Glass pattern stimuli (stimulus parameters are adjusted for demonstration purposes). The top panel shows Glass patterns stimuli with different spiral angles from radial (0°) to concentric (90°). The bottom panel shows radial Glass patterns at different levels of signal-to-noise-ratio (SNR) from 0.43 to 9.
To this end, we developed a battery of cognitive tests and theory-grounded individual differences measures, assessing the extent to which cognitive abilities combine with individual strategies to determine learning performance. Next, we sought to relate cognitive and social profiles from participants of all ages to individual learning ability in the context of a visual discrimination task. We asked whether an individual’s cognitive and social skills profile predicts training-induced improvements in perceptual tasks and tested for abilities that mediate learning independent of chronological age. While there is no single guiding theory that explains learning-dependent improvement in nonverbal tasks, previous studies have linked accuracy of learning to variables such as individual’s social perceptions (including cognitive style; Sternberg and Zhang, ), motivation (Dweck, ; Lau and Roeser, ), self-confidence and self-esteem (Lamont et al., ). Thus, we tested the hypothesis that these social factors may interact with cognitive processes that support our ability to extract relevant information and facilitate behavioral improvement in perceptual tasks through training in young and older adults.
## Materials and Methods
### Participants
Sixty participants, 30 young adults (11 male and 19 female ranging in age from 19 to 36 years old, M = 21.43), and 30 older adults (17 male and 13 female, ranging in age from 65 to 90 years, M = 73.60) took part in the study. All participants had normal to corrected vision and underwent the following visual tests: Visual acuity (Bailey-Lovie near and far acuity tests; Bailey and Lovie, ), contrast sensitivity (Pelli-Robson Contrast Sensitivity test; Pelli et al., ), and color blindness. None of the participants had been exposed previously to the task. Older participants also completed the Mini Mental exam (Folstein et al., ), and all scored within the normal range (25–30). This study was approved by the University of Birmingham Ethics Committee.
#### Recruitment
Two strategies were used to guide recruitment. Young participants were recruited from university’s research participation scheme; whereas, older participants were recruited from the university’s database for research into aging (which is drawn from university alumni and therefore indicates a secondary education), or by adverts placed in local publications. Most participants reported having some educational background (16 were University alumni, 6 reported no educational achievement). Comparison of means indicates that there are no significant differences between older participants scores on study variables due to avenue of recruitment. Further, older participants in our study were in frequent contact with research groups in the university and were regularly asked to participate in various studies. Thus, they were likely to have had exposure to the same types of equipment and questionnaires as the younger adults in the study. Importantly, our experiments were conducted by well-trained researchers that explained in detail and repeatedly the tasks and instructions and monitored closely progress in the experiment, ensuring that participants were familiar with the computer equipment and understood well all aspects of the task and instructions in questionnaires. Finally, all young and older participants were paid for their participation.
### Visual Learning Task
We used Glass patterns (Glass, ) as stimuli that are generated by pairing randomly positioned dots to form dipoles (dot pairs). Our stimuli comprised 600 white dot pairs (dipoles) displayed within a square aperture (7.5 × 7.5°) on a black background. Each dipole consisted of two 0.0375° dots with 0.26° separation between them. By placing dipoles tangentially and orthogonally to the circumference of a circle centered on the fixation dot we created concentric and radial patterns respectively. We also generated intermediate Glass patterns between these two pattern types by parametrically varying the spiral angle between 0° (radial) to 90° (concentric). The spiral angle was defined as the angle between the dot dipole orientation and the radius from the center of the dipole to the center of the stimulus aperture. Further, we manipulated signal-to-noise ratio (SNR; i.e., the ratio of signal dipoles to noise dipoles: randomly positioned and oriented dipoles) and presented stimuli at 30%, 45%, 60%, 75%, 90%, and 100% signal corresponding to SNR of 0.43, 0.82, 1.5, 3, 9 and ∞, respectively. We set the lowest SNR at the detection threshold of Glass patterns in noise (29.8 ± 1.59% signal) as indicated by our pilot and previous studies (Kuai and Kourtzi, ). A new pattern was generated for each stimulus presented in a trial, resulting in stimuli that were locally jittered in their position. These parameters were chosen based on pilot psychophysical studies and in accordance with previous work (Wilson and Wilkinson, ; Kuai and Kourtzi, ) showing that coherent form patterns are reliably perceived for these parameters. All participants were familiarized with the task and stimuli during a short practice session (100 trials). Participants took part in one pre- and one post-training session without error feedback and four to five training sessions with feedback. Feedback was delivered by an auditory signal (beep) when the participants responded incorrectly in a trial. On each trial, a stimulus pattern was presented for 200 ms followed by a 500 ms blank screen. Participants were instructed to report whether the pattern was radial or concentric. We measured participants’ performance using a 3-down-1-up staircase method resulting in 79.4% convergence level. That is, task difficulty increased following correct response in three trials, while it decreased following one incorrect response. Task difficulty was manipulated by changing the spiral angle (i.e., spiral angles closer to 0° or 90° corresponding to radial and concentric patterns were easier to discriminate than spiral angles closer to 45°). We measured spiral angle thresholds by averaging the last two-third reversals in each staircase. In the pre- and post-training sessions, we measured participants’ performance using three to five staircases with nine or ten up-down reversals at each SNR. In each training session, participants received training on 1200–2000 trials with feedback.
### Individual Differences Measures
Previous research exploring individual variability in aging has typically used measures of control beliefs and self-reported measures of personal approaches as a means for detecting individual variability. However, these measures are more indirect and general in their focus. Here, we develop a more direct measure of the influence of individual differences and motivational factors on learning. In particular, we investigated learning styles (Biggs, ; Evans et al., ; Sadler-Smith et al., ) within the context of the applied social cognition framework drawn from work exploring intrinsic motivation (Dweck, ) and the influence of personal self-esteem (Rosenberg et al., ). Additionally, evidence from the field of social psychology suggests that style of thinking (Sternberg’s Theory of Mental Self Government; Sternberg, , ; Sternberg and Grigorenko, ) or learning approach (J B. Biggs’ “Theory of Learning Approaches”; Biggs, ), influences learning outcome (Gully and Chen, ). These social constructs including: self-esteem (Entwistle and Ramsden, ), and intrinsic motivation (Dweck, ; Lau and Roeser, ) have been argued to affect individuals’ ability and confidence to learn new information. While research has sought to explore the role of individual differences in learning ability, it has largely overlooked the role of these social constructs on learning ability across the lifespan (cf. Hutchens et al., ). The individual difference measures consisted of the following items: Deep vs. surface learning style, strategic approach, achievement motivation, and self-esteem.
To test the ecological validity of individual difference questionnaires, we conducted pilot trials with 60 (30 young and 30 older) participants. Participants were asked to rate the questions included in the individual difference measures for clarity. Items rated highly by most participants as clear and comprehensible were included in the questionnaires used. These measures were then administered to the 60 participants in our study. In order for the individual differences measures to be consistent with the scoring of the cognitive variables, three of the scales (strategic approach, AMT, and self-esteem) were re-coded; i.e., a low score indicates a high rating on the scale. The individual difference measures consisted of the following scales:
#### Deep-Surface Learning Style
Participants were presented with five statements, taken from Tait et al.’s ( ) Approaches and Study Skills Inventory for Students (ASSIST; Tait et al.’s, ). These included, “I look at evidence carefully to reach my own conclusions;” and “What I have learned frequently seems unrelated to other bits and pieces” (reverse coded). All items were scored 1 = disagree completely through to 5 = agree completely . The mean of the 5-items was taken as a measure of learning style with high scores indicating deep learning style, while low scores indicating a “non-deep” surface learning style (∀ = 0.78).
#### Strategic Approach
The strategic approach scale comprised three subscales, also adapted from the ASSIST scale (Tait et al.’s, ). These subscales consisted of: (1) a strategic approach scale , with items such as: “I pay careful attention to any advice I am given, and try to improve my understanding”; (2) an effort scale , with items including, “I generally keep working hard even when things aren’t going all that well”; and (3) an organized scale , with items like, “I carefully prioritize my time to make sure I can fit everything in.” All items were scored 1 = disagree completely through to 5 = agree completely . After recoding, the mean of all of the items across the three subscales was taken as a measure of a “strategic learning approach”, with low scores indicating a strategic learning approach towards learning (∀ = 0.75).
#### Achievement Motivation
The achievement motivation scale consisted of 6-items, such as: “I work to cultivate people who will be helpful to me in the future.” All items were scored 1 = disagree completely through to 5 = agree completely . After recoding, the mean of the items was taken as a measure of achievement motivation, with low scores indicating high achievement motivation (∀ = 0.69).
#### Self-Esteem
Self-esteem was measured using a 10-item scale (Rosenberg, ) including items like, “On the whole, I am satisfied with myself.” All items were scored 1 = disagree completely through to 5 = agree completely . After recoding, the mean of the items was taken as a measure of self-esteem, with low scores indicating positive self esteem (∀ = 0.78).
### Cognitive Tasks
We used a battery of cognitive tasks to measure a range of abilities that have been suggested to affect learning ability. In particular, previous work provides evidence for the importance of working memory (WM) capacity (Law, ), attention span (Hambrick and Engle, ), processing speed (Salthouse and Ferrer-Caja, ; Mayes et al., ) and cognitive inhibition (St Clair-Thompson and Gathercole, ) in learning ability in young adults. Further, work in older adults suggests that individual differences in cognitive decline may drive age related differences in perceptual learning ability (Hultsch et al., ). Here, we tested the young and older participants on the following tasks: WM, cognitive inhibition, selective and divided attention (DA) and multiple object tracking.
#### Memory: Working Memory Task
The WM task was designed based on the sequential WM task used by Luck and Vogel (Luck and Vogel, ). Colored dots were displayed on a gray background for 500 ms, followed by a 1000 ms delay. After the delay, the dot display re-appeared with one of the dots highlighted by a white square. Participants reported whether the highlighted dot had remained the same color on the second presentation. An initial display of two dots was used. By using a two down one up staircase and a step size of 1 we manipulated the number of dots in the display, resulting in 70.7% performance. For example; each time the participant had two responses correct in a row an additional dot would be added to the next trial’s display, while for every incorrect answer, one dot was removed from the display for the next trial. WM thresholds (i.e., number of dots in the display) were calculated by averaging the last two-third reversals in each staircase. For each trial, each dot was randomly assigned a color, and one dot was randomly chosen as the target. Each dot had a radius of 12 pixels and dots were displayed in random locations within a 10 × 10 grid (jittered +/− 10 pixels). Each run consisted of 10 staircase reversals, participants completed 3 runs, after which we computed the average threshold as their WM score. In this task, a higher score (greater number of items in display) denotes better performance.
#### Inhibition: Stop-It Task
We used the Stop-It task developed by Verbruggen et al. (Verbruggen et al., ), which measures response inhibition based on the stop-signal paradigm (Lappin and Eriksen, ). Participants were asked to respond to the “go signal” (a white square or a circle presented in the center of a black background, displayed for 250 ms) by pressing a right or left response key to indicate the shape’s identity. The “go signal” remained on the screen until the participant responded, or for a maximum of 1,250 ms. “Go signals” were separated by a white fixation cross, displayed for 2,000 ms. On 25% of the trials a “stop signal” (750 Hz auditory tone, presented for 75 ms) was presented after the “go signal” had been displayed, instructing participants to inhibit their response for that trial. This delay (SSD: Stop Signal Delay) varied across trials. It was initially set at 250 ms and adjusted continuously using a staircase tracking procedure: When inhibition was successful, SSD increased by 50 ms; when inhibition was unsuccessful, SSD decreased by 50 ms. The task comprised of 3 blocks which consisted of 64 trials each. We used the latency of the stop process (SSRT), as initiated by the stop signal (see Verbruggen et al for full details), as our measure of cognitive inhibition. Poor inhibition is indicated by a slow SSRT; that is lower scores (faster SSRT) indicate good performance on this task.
#### Attention: Multiple Object Tracking Task
Multiple object tracking tasks measure human attention span and short term memory. We designed a task similar to that used by Sekuler et al. ( ). For each trial, a display of 10 stationary, blue and red dots (radius of 8 pixels each) appeared on a gray background for 1000 ms. After this initial fixation period the target red dots turned blue and all dots moved around the display for 5000 ms. Once the dots were stationary, a number (1–10) appeared in each blue dot. Participants were asked to indicate where the red dots were in the display by entering their corresponding numbers. This task consisted of 80 trials (20 per condition comprising 2, 3, 4 and 5 red target dots). Dots were assigned to random positions within a 10 × 10 grid (with a jitter of +/− 10 pixels), ensuring all dots were at least 16 pixels apart. Target locations and the angle at which each dot should move were randomly assigned for each trial. Dots had a maximum velocity of 2 pixels per frame and the Euclidean distance between two disks was always less than the diameter of a disk. If dots collided in the display, motion direction was altered. Task performance was measured by plotting the percentage of correctly tracked dots for each condition and calculating the slope of the fitted performance across conditions. Thus, a lower score (shallower slope) indicates better performance in this task.
#### Attention: Useful Field of View
Useful Field of View (Visual Awareness Inc.) is a task that assesses three attentional processes: processing speed, DA and selective attention (SA). This version of the task is explained in full by Edwards et al. ( , ), who also report a test-retest reliability of 0.74. Each trial consisted of 4 stages: (1) a fixation bounding box (1 s duration), (2) the test stimuli (variable duration; see below), (3) a white noise visual mask to control for after images (1 s duration), and (4) the response screen (displayed until a response is made). Participants responded using the mouse. The first test, “processing speed”, required participants to identify a centrally presented stimulus. This stimulus (a silhouette of a 2 cm × 1.5 cm of a car or a truck) was presented on a black background inside a 3 cm × 3 cm white bounding box. Participants were asked to indicate whether the central stimulus comprised a car or truck by mouse click. The second task, “divided attention”, required participants to identify the central stimulus (car vs. truck), and also identify the location of a simultaneously presented peripheral stimulus (2 cm × 1.5 cm silhouette of a car). This peripheral stimulus was fixed at 11 cm from the central stimulus at one of 8 radial locations. The third task “selective attention” followed the same procedure as “divided attention” but the target stimuli were presented in the context of distractors (47 triangles of the same size and luminance as the targets). Participants were instructed to ignore the triangles, and indicate whether the central stimulus comprised a car or a truck, as we all the location of the peripheral target. Using a double staircase method the duration of the display within each task varied between 16.7 ms and 500 ms. This allowed us to establish the minimal display duration at which the participant could correctly perform each of the three tests 75% of the time. This means that a lower score (shorter duration) indicates better performance. Further, this manipulation allowed for the tasks to be adjusted for difficulty across age groups appropriately.
### Data Analysis: Partial Least Squares Regression Modeling
PLS regression is a component based multivariate statistical technique that allows predicting single or multiple response variables Y (i.e., threshold reduction) from highly correlated or collinear multiple explanatory variables X (i.e., cognitive abilities and individual differences variables, respectively; Wold, ). In contrast to principal components regression, the goal of PLS regression is not to form only components that capture most of the information in X, but components that are also predictive of Y. As such, the algorithm reduces the dimensions of X through a weighted linear combination of X variables to form orthogonal components that are correlated to the dependent variable. The analysis shows how much of the variation in Y and is accounted for by each additional component obtained from X. For Y (threshold reduction) cumulative variance can be interpreted in the same way as unadjusted R-square. Adjusted R-square shows the adjusted version of the cumulative Y variance.
## Results
To provide a sensitive and controlled measure of perceptual learning, we asked young and older participants to discriminate global visual forms (radial vs. concentric) defined by simple patterns of dots (Glass patterns). We manipulated participants’ ability to perceive these global patterns by varying: (a) the amount of background noise (i.e., randomly placed dots), and (b) the similarity between global forms, using linear morphing between concentric and radial patterns. To quantify the effect of learning, we used the following index:
where Th(session) is the mean shape discrimination threshold for each session.
Performance on this form discrimination task improved through training in both young and older adults (i.e., a reduction in the signal needed for 79.4% threshold performance was similar across age groups) with overall better performance for young than older participants (Figure ). In particular, a mixed design ANOVA, showed a significant main effect of session ( F = 147.82, p = 0.001) and age ( F = 14.84, p = 0.002), but no significant interaction between age and session ( F = 2.62, p = 0.11). Interestingly, we observed strong individual variability in performance for both young and older adults (Figure ).
Behavioral improvement in visual discrimination task. (A) Normalized ( z -score) thresholds (deg of spiral angle at 79.4% threshold performance) across training sessions for young (circles) and older (crosses) participants. (B) Normalized ( z -score) threshold reduction (i.e., difference in thresholds between post- and pre-training) for young and older participants. Box plots show individual variability in learning performance: in threshold performance ( z -scores) ranged from −1.25 to 2.02 for younger adults and from −2.87 to 0.74 for older adults. The upper and lower error bars display the minimum and maximum data values and the central boxes (“bowties”) represent the interquartile range (25th to 75th percentiles). The notch of the “bowtie” represents the median.
### Individual Differences
To investigate the sources of this individual variability in learning improvement, we used the battery of cognitive tests (multiple object tracking: MOT, DA, SA, WM, and cognitive inhibition) and theory-grounded individual differences measures (learning style: deep vs. surface, strategic approach, self-esteem, and achievement motivation). Consistent with previous studies (Hedden and Gabrieli, ), these measurements showed that older adults differ in cognitive abilities and social profile from young adults (Table ). Specifically, older adults had significantly lower performance in: cognitive inhibition ( t = −2.454, p = 0.019), divided attention ( t = −3.242, p = 0.003), selective attention ( t = −6.288, p = <0.001) and WM ( t = 4.046, p = <0.001) tasks compared to young adults. Further, older adults were more likely to engage in deep learning ( t = −2.715, p = 0.009), and rely on achievement motivation as a drive for learning ( t = −4.291, p = <0.001) than young adults (See Table ).
Performance in cognitive and individual differences measures .
Mean scores and standard errors for each cognitive and individual difference measures for young and older participants. All variables, other than working memory, are coded so that lower scores indicate higher performance on cognitive tasks and a higher rating on individual differences measures. Working memory is coded so that higher scores indicate high performance. The learning style (surface/deep) variable is coded so that a low score (1) indicates surface learning while a high score (5) indicates deep learning. Asterisks indicate a significant difference between the two age groups. *p < 0.05, **p < 0.01, ***p < 0.001 .
We then sought to relate the measured cognitive and social abilities to individual learning ability across age groups, using a partial least squares (PLS) regression model. This procedure allows us to test the predictive utility of a model even in smaller datasets (30 young vs. 30 older adults), that otherwise could be subjected only to correlational analysis. Our results show that a combination of cognitive abilities (i.e., performance in cognitive inhibition, MOT, DA, WM tasks), and individual differences variables (i.e., extent to which one engages in deep vs. surface learning, motivational impetus, and higher self-esteem), account for 60% of the variance in performance threshold reduction independent of age (Table ). Excluding age from the PLS model showed similar results; that is, cognitive and social variables alone accounted for a significant portion (i.e., 58%) of the explained variance in performance. Further, we found that learning is best predicted by the ability to inhibit irrelevant information (cognitive inhibition), and select targets (MOT). That is, PLS values are higher for these cognitive variables when age is included or excluded from the model (Figure ). Removing these cognitive variables from the model reduces substantially the explained variance (i.e., 46%).
Variance predicted by the PLS model .
Adjusted R values, and loadings of each of the cognitive and individual differences measures onto each of the components in the PLS model .
The role of cognitive and social profiles in learning variability . Outcome of the PLS model showing predictive utility of each variable in the PLS model when (A) age is included or (B) excluded from the model. Predictive utility values indicate the relative importance of each variable in predicting learning performance (i.e., threshold reduction). Variables included in the model comprise: (i) cognitive abilities measures (black bars): cognitive inhibition, multiple object tracking ability (MOT), Age, divided attention (DA), selective attention (SA), working memory (WM); (ii) individual differences measures (white bars): achievement motivation (AMT), self-esteem, learning style and strategic approach.
### Comparing Strong and Weaker Learners
To provide an alternative approach to the study of individual variability, we compared strong and weaker learners independent of age using partial correlations. To assign participants to these performance groups we first normalized threshold reduction scores within each age group. Participants with a normalized score (z-score) of above 0 were classified as strong learners while those with a score below 0 were classified as weaker learners. We correlated cognitive variables and individual difference measures with threshold reduction, while controlling for pre-training performance in the visual form discrimination task and age. This analysis revealed that different cognitive abilities predict individual learning ability in strong vs. weaker learners (Figure ), despite similar performance in these measures between groups (Table ). Learning improvement (i.e., higher threshold reduction) correlated with higher SA scores ( r = −0.499, p = 0.011) for strong learners, while with cognitive inhibition ( r = −0.522, p = 0.006), working memory ( r = 0.411, p = 0.037) and divided attention ( r = −0.479, p = 0.021) scores for weaker learners. These results were supported by power calculations indicating that we have 80%–87% power to detect correlations of 0.48 for the sample sizes included in this study.
Comparing profiles for strong and weaker learners . Results of partial correlations ( r values) for “strong” and “weaker” learners. Threshold reduction is correlated with cognitive and individual differences measures, while controlling for pre-training performance in the visual discrimination task and age. Note that for graphical representation purposes, the signs of any negatively coded variables have been reversed, indicating that increased scores in cognitive and individual difference measures correlate with increase in threshold reduction.
Cognitive and individual differences measures for strong and weaker learners .
Mean performance and standard errors for cognitive variables and individual difference measures. Results for strong and weaker learners are displayed outside vs. inside brackets, respectively. Partial correlations of cognitive and social variable scores with threshold reduction are reported separately for strong and weaker learners. All variables with the exception of working memory are coded so that a low score indicates strong performance; that is, negative correlations signal that high scores/ratings on these variables correlate with increased threshold reduction. For working memory, high performance correlates with increased threshold reduction, as shown by the positive correlation. For learning style, a negative correlation indicates that surface style correlates with better threshold reduction, while a positive correlation indicates that deep learning style correlates with better threshold reduction .
### Modulatory Effects of Age
Our findings so far suggest that both cognitive and individual difference variables play a role in determining learning ability across age. However, this does not rule out the possibility that age may modulate the relationship between cognitive and individual difference variables. To test this hypothesis, we conducted additional analysis using threshold reduction as dependent variables in the PLS model. Our results showed that age modulates the importance of cognitive (i.e., cognitive inhibition, attention span as assessed by divided and SA) and social (i.e., learning style, strategic approach, motivational impetus and self-esteem) variables in predicting learning ability (Table ), suggesting that some variables become more important in predicting threshold reduction at older age. Specifically, the pattern emerging is that age directly impacts cognitive inhibition, learning style and esteem. Further, older adults drew on a different style to guide their learning of novel information as compared to that of younger participants: they are motivated to look for more detail and a greater overall understanding of the task. Additionally, higher personal self-esteem is significantly more important for higher learning improvement in older participants. That is, older adults rely more on deep or strategic learning strategies than young adults, as self-esteem mostly and to a lesser extent cognitive abilities decrease with age.
(A) Adjusted R-square values for PLS Regression without Age as an independent measure. (B) PLS model with threshold reduction as dependent variable and age as a moderator.
Unstandardized estimates for cognitive and individual difference measures (left column) and the interaction between these independent variables and age (right column). Increased values (from left to right column) indicate the moderating effect of age on the independent variables. For learning style, a low score indicates surface learning and a high score indicates deep learning; thus a negative sign indicates that strong “surface learning style” predicts strong learning performance. Note: *p < 0.05, **p < 0.01 ***p < 0.001 .
## Discussion
Our findings provide evidence that an individual’s cognitive and social skills profile rather than age per se predicts the ability to improve in perceptual judgments through training. Testing for age related differences alone may obscure the role of these variables in measuring individuals’ learning ability. In contrast to previous studies focusing on age differences (Ball and Sekuler, ; Richards et al., ; Andersen et al., ; Bower and Andersen, ), we demonstrate that attentional capacity, learning style and intrinsic motivation are critical for improving in perceptual tasks through training. Interestingly, our results show that strong learners are better able to select the most appropriate cognitive strategy (i.e., SA to targets) to improve at the task in hand (i.e., visual form discrimination in noise), while weaker learners rely on more general cognitive strategies.
Our work focuses on learning as a result of training on perceptual tasks. Previous studies have suggested that aging may result in reduced efficiency (Bennett et al., ), increased internal noise (Bennett et al., ) or reduced tolerance to external noise (Bower and Andersen, ) affecting performance in perceptual tasks. However, learning in young adults has been suggested to enhance performance efficiency (Gold et al., ), improve exclusion of external noise and reduce internal noise (Dosher and Lu, ). Extending beyond these previous studies, we show that perceptual learning is influenced by executive functions (i.e., the ability to inhibit distractors and select task-relevant features) and social attitudes (i.e., strategic or deep learning). These findings are consistent with previous studies (Kuai and Kourtzi, ) showing that visual form learning in aging is limited by visual selection processes rather than fine feature processing. Further, the ability to suppress irrelevant background information has been shown to deteriorate with age (Betts et al., , ) possibly due to weakening of inhibitory processes (Leventhal et al., ; Hua et al., ) or attentional functions in aging (Ball et al., ; Kane et al., ). Interestingly, in our previous work (Mayhew and Kourtzi, ) we have shown that visual shape learning engages primarily parietal regions in older adults, suggesting a stronger role of attentionally-guided learning that enhances the perceptual salience of behaviorally relevant targets in cluttered scenes (Gottlieb et al., ; Corbetta and Shulman, ; Roelfsema and van Ooyen, ; Mevorach et al., ).
Our findings have potential implications for understanding compensatory brain mechanisms that may support individual ability for learning in older age. Understanding the socio-cognitive profile of individuals and how it influences learning ability is critical for determining the brain mechanisms that underlie individual variability and may support better learning in some older adults than others. For example, our findings suggest that older participants find it more difficult to inhibit irrelevant details. It is possible that older adults may attempt to compensate for this change in cognitive capacity by drawing on strategic learning, or the use of deep learning strategies focusing on more thorough understanding of new information. In future work, it would be interesting to test whether varying the training task recruits different socio-cognitive variables as best predictors of learning ability. It is possible that cognitive inhibition and attention are critical when detecting targets from noise and discriminating highly similar stimuli. However, other cognitive variables (e.g., WM) may be more important in the context of associative or probabilistic learning tasks. Further, learning in other domains, such as verbal or motor learning, may be influenced by a different set of socio-cognitive abilities. It may also be interesting to enrich the individual difference measures using a measure of control beliefs (Hutchens et al., ), as control beliefs may offer further insights into the factors that learners perceive as beneficial for their general learning ability.
Overall, our findings have practical relevance for the optimization of training programs targeting cognitive abilities and social attitudes, which are critical for improvement in perceptual tasks but more importantly for generalizing learning to real-life situations. Future research would investigate why older people may adopt different strategies to maximize learning; and how readily they may adopt alternate strategies for learning, if they diverge from the ones that they may feel comfortable employing.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
The research on staging of pre-symptomatic and prodromal phase of neurological disorders, e.g., Alzheimer's disease (AD), is essential for prevention of dementia. New strategies for AD staging with a focus on early detection, are demanded to optimize potential efficacy of disease-modifying therapies that can halt or slow the disease progression. Recently, neuroimaging are increasingly used as additional research-based markers to detect AD onset and predict conversion of MCI and normal control (NC) to AD. Researchers have proposed a variety of neuroimaging biomarkers to characterize the patterns of the pathology of AD and MCI, and suggested that multi-view neuroimaging biomarkers could lead to better performance than single-view biomarkers in AD staging. However, it is still unclear what leads to such synergy and how to preserve or maximize. In an attempt to answer these questions, we proposed a cross-view pattern analysis framework for investigating the synergy between different neuroimaging biomarkers. We quantitatively analyzed nine types of biomarkers derived from FDG-PET and T1-MRI, and evaluated their performance in a task of classifying AD, MCI, and NC subjects obtained from the ADNI baseline cohort. The experiment results showed that these biomarkers could depict the pathology of AD from different perspectives, and output distinct patterns that are significantly associated with the disease progression. Most importantly, we found that these features could be separated into clusters, each depicting a particular aspect; and the inter-cluster features could always achieve better performance than the intra-cluster features in AD staging.
## 1. Introduction
Alzheimers disease (AD) is the most common neurodegenerative disorder among aging people, which accounts for nearly 70% of all dementia cases. The symptoms of cognitive impairment develop gradually over years, and eventually lead to death (Kalaria et al., ). Currently, there is no cure for AD. The early signs of AD include a noticeable and measurable decline in memory, language, thinking, and other cognitive abilities. Patients with these symptoms are usually diagnosed as the Mild Cognitive Impairment (MCI). MCI does not notably interfere with daily activities, but those with MCI have a higher risk of later progressing to AD or other forms of dementia (Dubois and Albert, ; Jicha et al., ; Nettiksimmons et al., ). Many medical interventions may only be effective in the early course of the disease (Bond et al., ). Therefore, accurate staging of the disease, especially the detection of MCI, could help the physicians to identify the subjects at higher risk of developing dementia and allow the patients to receive early medical interventions before irreversible brain damages are formed.
Numerous biochemical and genetic biomarkers, e.g., increased cerebrospinal fluid (CSF) tau, phosphorylated tau and ubiquitin levels, low CSF Amyloid-β ( Aβ ) concentration, and apolipoprotein E (ApoE) ϵ4 allele, have been proposed to detect AD onset and predict conversion of MCI and normal control (NC) to AD with high specificity and sensitivity (Trojanowski et al., ; Kandimalla et al., , , ; Andreasson et al., ). Recently, neuroimaging biomarkers have been increasingly used as additional markers for assessing the likelihood of such detection and prediction, since they can detect the changes in brain structure (e.g., atrophy) and function (e.g., hypometabolism, amyloid plaque, and neurofibrillary tangles formation) before the cognitive impairment symptoms appear (Perrin et al., ; Davatzikos et al., ; Ewers et al., , ; Hinrichs et al., ; Singh et al., ; Jacobs et al., ). Several large multi-modal neuroimaging data repositories, such as the Alzheimers Disease Neuroimaging Initiatives (ADNI) (Jack et al., ; Jagust et al., ) and Australian Imaging, Biomarker and Lifestyle Flagship Study of Aging (AIBL) (Sona et al., ), have been founded to facilitate the neuroimaging research in AD and MCI.
A variety of quantitative measures can be extracted from the neuroimaging data as biomarkers in the evaluation of AD and MCI patients, such as hippocampal volume loss (Schuff et al., ), ventricular boundary shift integral (Freeborough and Fox, ) extracted from structural MRI, and z-score (Minoshima et al., ) and t-map (Cai et al., ) extracted from FDG-PET. We refer to same type of features as a “view.” The terms, “view” and “modality,” are often used interchangeably in the computer vision community, but not in the medical imaging community. A modality, in medical imaging domain, usually means the image acquisition technique or scanning protocol, such as Magnetic Resonance Imaging (MRI), Positron Emission Tomography (PET), Computed Tomography (CT), Ultrasound, Single Photon Emission Computed Tomography (SPECT), functional MRI (fMRI), and Diffusion Tensor Imaging (DTI). However, a view means a specific type of measure extracted from a modality. Therefore, a modality may contain multiple views, but a view pertains to one modality.
MRI and PET are the two most widely used neuroimaging modalities in visualizing AD and MCI brains (Liu et al., ). A diversity of biomarkers extracted from the PET and/or MRI data have been proposed in previous studies. Liu et al. recently gave a comprehensive review of these biomarkers (Liu et al., ). The most well-known MRI biomarkers include the regional gray matter volume (GMV), e.g., hippocampus and ventricles (Klöppel et al., ; Heckemann et al., ), cortical thickness (Fischl and Dale, ; Dickerson et al., ; Frisoni et al., ), local gyrification index (Schaer et al., ), curvedness and shape index (Awate et al., ; Cash et al., ). PET-derived biomarkers are generally pertaining to the radioactive tracers. Amyloid-binding compounds, i.e., F-BAY94-9172, C-SB-13, C-BF-227, F-AV-45, and C-Pittsburgh compound B ( C-PiB), have been used for imaging amyloid plaques in AD (Carpenter et al., ; Perrin et al., ; Thompson et al., ; Ni et al., ), whereas 2-[ F]fluoro-2-deoxy-D-glucose (FDG) has mainly been used to depict glucose metabolism (Minoshima et al., ; Cai et al., ). Various static and kinetic biomarkers can be extracted from the PET data, i.e., the standard uptake value (SUV) (Clark et al., ; Landau et al., ), cerebral metabolic rate of glucose consumption (CMRGlc) (Sokoloff et al., ; Cai et al., ), mean index (Batty et al., ), z-scores (Minoshima et al., ), hypo-metabolic convergence index (HCI) / amyloid convergence index (ACI) (Chen et al., ), tissue time activity curve (TTAC) (Cai et al., ). In our previous studies, we proposed the convexity ratio and solidity ratio (Liu et al., ) to detect the brain atrophy with MRI and the Difference-of-Gaussian (DoG) features (Cai et al., ) to detect the hypo-metabolism with FDG-PET, respectively. All of these biomarkers have been proved to have great potentials of differentiating the AD and MCI patients from normal controls (NC), and they also have demonstrated different strengths in characterize the disease pathology, e.g., PET views, such as SUV and CMRGlc in FDG-PET, are effective in detecting the functional anomalies in the brain, whereas MRI views, such as GMV and cortical thickness, are more sensitive to the brain morphological changes (Fan et al., ; Desikan et al., ; Risacher et al., ).
Researchers have carried out many studies on fusing these multi-view features. As pointed out by Atrey et al. ( ) and Zhang et al. ( ), current multi-view fusion methods could be roughly categorized into two groups, i.e., feature fusion and decision fusion. The feature fusion methods create a new feature space for the multi-view features and subsequently train a single model to classify the patients. Feature selection is a special feature fusion algorithm, that selects the most discriminant features based on certain selection criteria, such as t -test (Heckemann et al., ), Lasso (Zhu et al., ), or Elastic Net (EN) (Shen et al., ). The advanced feature fusion methods include multi-view spectral embedding, which embed the multi-view feature spaces into a unified space based on manifold learning (Park, ; Liu et al., ; Che et al., ), the multi-kernel support vector machine (MK-SVM) that combines the feature spaces with kernel tricks (Hinrichs et al., , ; Zhang et al., ), and deep learning methods that extract highly abstract features with a multi-layered neural network (Liu et al., , ). The decision fusion methods train different models for different views, and subsequently aggregate the predictions of the all classifiers to make the final decision. Decision fusion, as compared to feature fusion, requires repeatedly training the classifiers and tuning their weighting parameters. In our recent study (Liu et al., ), we proposed the Multifold Bayesian Kernelization (MBK) method to synthesize the multi-view biomarkers. MBK could construct a set of non-linear kernels to obtain the classification probabilities for individual views, and then infer their weighting parameters by minimizing the diagnostic errors and kernelization errors based on a Bayesian framework.
The aforementioned studies show that multi-view biomarkers could achieve better performance than single biomarkers, and imply that the multi-view biomarkers could create the synergy in the classification of AD and MCI (Hinrichs et al., ; Zhang et al., ; Singh et al., ; Liu et al., , ; Jacobs et al., ). However, researchers do not yet understand the cause of such synergy, and there is a lack of the methods for quantitatively analyzing the synergy between individual biomarkers. Therefore, this study differs from the other multi-view studies in that our interest is to investigate the synergy between the multi-view biomarkers instead of solely improving the staging performance.
We propose a cross-view pattern analysis framework to investigate the synergy between the multi-view biomarkers. With this framework, we found that the biomarkers derived from MRI and PET could be separated into four clusters, each having a unique strength in detecting certain pathological changes in AD and MCI. We evaluated these biomarkers in a task of classifying the AD, MCI, and NC subjects obtained from the ADNI baseline cohort, and found the inter-cluster combination could always achieve the best performance compared to the intra-cluster combination. This study does not require the ethical approval since it is purely based on the analysis of the medical imaging data with no involvement of the patients, and the permission has been obtained to use the ADNI datasets.
The reminder of this paper is organized as follows. In Section 2, we first described the ADNI datasets, the pre-processing steps and the multi-modal features used in this study, and then elaborated the single-view and cross-view pattern analysis methods as well as the classification and evaluation methods. The pattern analysis and classification results were shown in Section 3, followed by the discussion on our findings in Section 4. Finally we concluded in Section 5.
## 2. Materials and methods
Figure illustrates the work-flow of our analysis. We first acquired the raw MRI and PET datasets from the ADNI baseline cohort, then registered the brain volumes to a template and segmented them into a set of 3D regions of interest (ROI). Totally nine views of biomarkers were extracted from each ROI. Single-view and cross-view pattern analyses were carried out on these views based on their pathology patterns in terms of the brain atrophy and hypo-metabolism. Finally, we evaluated the single-view biomarkers and their combinations in the classification of AD, MCI, and normal control (NC) subjects using the MK-SVM algorithm.
The work-flow of the cross-view pattern analysis . It is a five-step pipeline, which takes brain T1-MRI and FDG-PET images as inputs and generates the classification results as the outputs.
### 2.1. Datasets
Data used in the preparation of this article were obtained from the Alzheimers Disease Neuroimaging Initiative (ADNI) database ( ). The ADNI was launched in 2003 as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early Alzheimer's disease (AD).
The ADNI datasets consist of a variety of imaging and non-imaging biomarkers, such as MRI, PET, cerebrospinal fluid (CSF) measures, genetic biomarkers, and clinical assessments. Our focus of this study was to investigate the neuroimaging biomarkers, therefore, we selected all of 369 subjects who had both the T1-weighted MRI volume scanned on a 1.5 Tesla MR scanner and the FDG-PET volume from the ADNI baseline cohort. After pre-processing, we visually checked the images and excluded those with intolerable distortions, that resulted in a downsized database of N ( N = 331) subjects. These patients were divided into three groups according to their baseline diagnoses, including 85 AD, 169 MCI, and 77 NC subjects.
### 2.2. Pre-processing
In Step (a), as indicated in Figure , we pre-processed all these 3D MRI and PET volumes using the following protocols. We retrieved the MRI and PET volumes from ADNI database (Jack et al., ; Jagust et al., ). The PET data have a common isotropic voxel size of 1.5 mm and a full width at half maximum resolution of 8 mm. We then removed the non-brain tissue from MRI images using FSL BET (Smith, ). To enable the subsequent joint analysis of PET and MRI, i.e., feature extraction and pattern analysis, we then linearly registered the PET image to the MRI image of the same subject using FSL FLIRT (Jenkinson et al., ).
ROI-based features, as compared to voxel-based features, had lower dimensions and would avoid the curse-of-dimensionality. In addition, brain ROI features, such as hippocampal and ventricular volumes, have shown promising potential in characterizing AD and MCI. Therefore, we chose to use ROI-based features instead of voxel-based features in this analysis. The MRI data in ADNI baseline cohort have been labeled with K ( K = 83) brain ROIs through the multi-atlas propagation with enhanced registration (MAPER) approach (Heckemann et al., , ). These MAPER-generated labelmaps were then used to extract the ROI features in the next step. A complete list of these ROIs can be found in the previous papers (Heckemann et al., ; Liu et al., ).
### 2.3. Feature extraction
As shown in Figure - Step(b), totally M ( M = 9) views of biomarkers were investigated in this study, including four biomarkers extracted from the T1-weighted MRI data: Gray Matter Volume (GMV), Local Gyrification Index (LGI), Convexity Ratio (CNV), and Solidity Ratio (SLD); and five biomarkers extracted from FDG-PET data: Mean Index (M-IDX), Fuzzy Index (F-IDX) and three Difference-Of-Gaussian features (DoG-M, DoG-C, DoG-Z). Since the features were all ROI-based, each feature element had two attributes, the location in the brain and the feature value. These two attributes together formed a signature neurodegeneration pattern of each view.
#### 2.3.1. Gray matter volume
GMV, is the most commonly used MRI biomarker in AD characterization in laboratories, since the GMV is closely related to the cortical neuronal loss as well as synaptic loss due to the disease (Carison et al., ). In this study, we extracted the GMV features from all K ROIs except for the ventricles, central structures, cerebellum and brainstem (whole volumes were used for these ROIs). We further normalized the GMV features by the intracranial volume as measured on the same source image to eliminate the impact of linear scaling in segmentation.
#### 2.3.2. Local gyrification index
LGI, is a metric that quantifies the ratio of the cortex buried within the sulcal folds to the outer visible cortex (Schaer et al., ). A normal healthy cortex with extensive folding usually has a larger LGI, whereas a degenerative cortex with limited folding has small LGI. The LGI features are usually computed in circular 3D ROIs in each hemisphere. In order to match the other views of features in this study, we computed the LGI features in the K pre-defined ROIs instead, i.e., the intersection of the 3D circular ROIs and pial surface were replaced by the outer surface of the pre-defined ROIs. For the non-cortical regions, the surface areas were used as the LGI features.
#### 2.3.3. Convexity ratio
CNV, also aims to capture the cortical folding features. CNV differs from LGI in that it is not limited to the cortex surfaces. It is defined as the area of the convex hull surface divided by that of the ROI surface (Liu et al., ). Similar to LGI, a normal healthy brain usually has a larger CNV, and a degenerative brain has low CNV.
#### 2.3.4. Solidity ratio
SLD, quantifies the fullness of the ROI in the convex hull. It is defined as the ratio of volume of the ROI to that of the convex hull. SLD describes the extend to which the shape is convex or concave. Compared to the normal healthy brains, the degenerative brains with atrophy usually have a shrinking shape, which leads to a lower SLD value. SLD and CNV are usually used together to enhance the GMV features due to large inter-subject brain volume variations (Liu et al., ).
#### 2.3.5. Mean index
M-IDX, is defined as the mean activity levels of the ROIs (Batty et al., ). It is a simple and effective feature in capturing the brain metabolism activity levels and has been widely used in AD and MCI characterization. In particular, M-IDX is very sensitive to the brain hypo-metabolism and has better performance in early detection of MCI than many complex feature descriptors, such as 3D Gabor Filters, Gray Level Co-occurrence Matrix, and Discrete Curvelet (Liu et al., ). To eliminate the intensity variations during acquisition or parameter estimation, we further normalized the M-IDX features with the average cerebellum metabolism rate.
#### 2.3.6. Fuzzy index
F-IDX, evaluates the consistency of the metabolism activity levels, or the fuzziness, of the ROIs. It is defined as the standard deviation divided by the mean value of the ROI voxels. F-IDX is particularly useful for characterizing the ROIs that are partially hypo-metabolic. The voxels in these ROIs have less consistent activity levels, thus lead to higher F-IDX. On the contrary, the normal ROIs are expected to have more consistent activity levels and smaller F-IDX values.
#### 2.3.7. Difference-of-Gaussian mean
DoG-M, quantifies the degeneration levels of the hypo-metabolic regions (lesions) at different spatial scales estimated by the Difference-of-Gaussian (DoG) descriptor. It is defined as the mean metabolism rate of the lesion area within the segmented ROI (Cai et al., ). Different from M-IDX, DoG-M considers the activity level of the lesions only. The mean metabolism rate of all lesion areas across the brain is first computed, and further normalized by the mean metabolism rate of the cerebellum to remove the bias of global intensity variation. It is originally called the lesion mean index. To avoid the ambiguity with M-IDX, we referred to it as DoG-Mean (DoG-M) in the rest of this paper.
#### 2.3.8. Difference-of-Gaussian contrast
DoG-C, quantifies the contrast between the lesions and non-lesion parts. Since there are large variations of the metabolism rates in different ROIs, DoG-C offsets this effect by focusing on the contrast instead of the actual activity level of the ROI. It is originally called the lesion contrast index and defined as the ratio of the mean metabolism rate of the lesions to that of the non-lesion parts and further corrected using the variances of both parts in the ROIs, where the lesions are also approximated by the DoG descriptor.
#### 2.3.9. Difference-of-Gaussian Z-score
DoG-Z, similar to the conventional Z-score (Minoshima et al., ), quantifies the proportion of the abnormal voxels in the ROIs. However, conventional Z-score requires voxel-wise registration which will involve registration error, instead we used DoG operator to estimate the hypo-metabolism lesions in this study. DoG-Z is a good indicator to approximate the progress of the disease. Late-stage patients usually have higher DoG-Z values than the early stage patients.
### 2.4. Single-view pattern analysis
In single-view pattern analysis, as shown in Figure - Step(c), we analyzed the pathology patterns of the nine individual views extracted from the imaging data.
For each view, we performed ANOVA on the three disorder groups, AD, MCI, and NC, against the null hypothesis that all groups were simply random samples of the same population. Given a view, P , the p -values of ANOVA, P = { P (1), P (2), …, P ( K )}, showed the discriminating power of the ROIs in this view. To make it comparable to other views' patterns, we transformed the p -values to non-negative valued weights, which were positively correlated to the ROI discriminating power, as Equation (1):
where σ is the bandwidth parameter which controls how quickly P ′( i ) falls off with the P ( i ). If P ( i ) is small, then P ′( i ) is close to 1; and if P ′( i ) is greater than σ, then P ′( i ) will plummet to 0. In this study, we set the bandwidth σ as 0.05.
In order to quantify the differences between the patterns in the following analysis, we further normalized the P ′ as Equation (2):
The normalized weights, P ″(1), P ″(2), …, P ″( K ), together formed a distinct pathology pattern of the view.
There are three types of ROIs in terms of their consistency across different views. The first type of ROIs is the disease-spared ROIs, which are not affected by the disease and have low discriminating power across most of the views, e.g., cerebellum is believed to be spared by AD and always used to calibrate the PET metabolism rates. The second type of ROIs is the disease-affected ROIs. Hippocampus, for instance, has been widely used as an effective biomarker for characterizing AD and MCI. The third type of ROIs is the view-specific ROIs, which have varying p -values across different views. These ROIs show the different effects of the disease on the brain, and potentially lead to the synergy or interference between different views.
### 2.5. Cross-view pattern analysis
Figure - Step(d) shows the cross-view pattern analysis. The goal of this step was to compare the patterns and quantitatively analyze the variability among them.
We first paired up these M views, which lead to M × ( M − 1)/2 pairs of views. In this study, there were 9 single views and 36 pairs. We then quantitatively analyzed each pair based on their patterns. Assuming P ″ and Q ″ represent the patterns of two views, we computed their affinity, A ( P, Q ), as Equation (3):
where D ( P || Q ) is the Kullback-Leibler (KL) divergence of Q from P , and D ( Q || P ) is the KL divergence of P from Q . A ( P, Q ) = 0 if P ″ = Q ″. Note that KL divergence is non-symmetric measure of difference between P and Q , and cannot be used as a distance metric as it does not satisfy the symmetry condition. Therefore, we actually measured the affinity between two views based on their mutual divergence.
The affinity value of all pairs formed the affinity matrix A . To see how the views were related to each other, we further computed the clustering of them based on the symmetric normalized Laplacian matrix ( L ) of A (Ng et al., ), as Equation (4):
where I is the unit matrix, D is defined as the diagonal matrix whose ( i, i )-element was the sum of A 's i th row. If we consider the patterns to be the points in a K -dimensional space, then the top- k eigenvectors of L could be stacked in columns to form a new k -dimensional space for the patterns, therefore it allowed us to observe the embedding of the views in a low dimensional space. In this study, we set k to 2 and displayed the views as points in a 2-dimensional space.
### 2.6. Classification and evaluation
The last step of our work-flow was to evaluate the performance of these 9 single views and 36 pairs of views in the task of staging of the disease progression, i.e., classifying the AD, MCI and NC subjects, as illustrated in Figure - Step(e). The goal of this step is to see how the single-view biomarkers interact with each other and find out what biomarkers have more effective synergy than others.
Since the datasets used in this study were highly skewed that MCI subjects accounted for a large percentage (over 50%) of the entire population, we designed three classifiers instead of one classifier in order to reduce the data bias onto classification and achieving more accurate staging. The first classifier was a binary SVM aiming to distinguish NC subjects from the AD and MCI patients. We kept the NC subjects predicted by the first classifier and sent other subjects to the second classifier. The second classifier was also a binary SVM, which classified the subjects into AD or non-AD patients. The predicted AD patients of the second classifier were retained and the rest of the patients were sent to the third classifier. The third classifier was a multi-class SVM, which classified AD, MCI, and NC subjects in one setting. The Radial-Basis-Function SVM (RBF-SVM) was used for the single views, whereas the Multi-Kernel SVM (MK-SVM) was used for the combinations of the views. Both the RBF-SVM and MK-SVM were implemented using LIBSVM library (Chang and Lin, ).
The 5-fold cross-validation paradigm was adopted in performance evaluation. Specifically, we divided the datasets into 5 equal-sized subsets, and each subset was used for testing in turn while other subsets were used for training the model. While training, the three classifiers were trained together and the hyper-parameters were optimized using the random search optimization algorithm (Bergstra and Bengio, ). Totally six performance metrics were used in this study, including three precision metrics for AD, MCI, and NC respectively, and the overall accuracy, specificity and sensitivity. Note that when computing the specificity and sensitivity, NC was considered as the negative class, and both MCI and AD were considered as the positive class. The corresponding standard deviations from cross-validation were also reported with the performance metrics.
## 3. Results
### 3.1. Single-view pattern analysis result
Figure shows the back-projection of the MRI single-view patterns onto the ICBM_152 brain template (Mazziotta et al., ), which is also labeled using the MAPER approach. The color bar indicates the p -values of the ROIs in each view. Note that the ventricles and corpus callosum are not displayed here. Based on these patterns, we found that a large proportion of the brain was spared by the disease, such as the insula, brain stem, corpus callosum, and parts of the frontal lobe, parietal lobe and subcortical regions. The disease-affected regions include the repeatedly reported ventricles, middle and inferior temporal lobe and limbic gyrus. We also observed a strong agreement across most views on parts of the occipital lobe (lateral part, lingual, and cuneus) and frontal lobe (superior part), which were less investigated in previous studies. GMV further detected the hippocampus, parahippocampal and ambient gyrus, and amygdala. CNV detected two particular ROIs, the cerebellum and the thalamus, although these two structures were usually considered spared by AD. SLD also has two signature ROIs in the parietal lobe, including the superior and post-central parts.
Back projection of the normalized weights of the ROIs for four MRI views onto the ICBM_152 template using 3D Slicer (Fedorov et al., ) .
Figure shows the back-projection of the PET single-view patterns onto the ICBM_152 brain template. In addition to the temporal lobe and limbic gyrus that were detected by MRI views, the PET patterns also included more frontal (subgenual, orbital, inferior, middle, and superior parts) and parietal areas (post-central and superior parts). These regions are believed affected at the later course of AD and MCI, after the hippocampus, entorhinal cortex, temporal regions and posterior cingulate (Fan et al., ). This indicated that frontal and parietal lobe were essential in staging AD and MCI, and we may more effectively detect functional changes rather than structural changes in these regions. Compared to MRI views, they were less sensitive to pathological changes in the occipital lobe, where only the cuneus was detected by the DoG-M and DoG-Z. The patterns of M-IDX and DoG-M were larger than the other views, both covering the inferiolateral parietal area.
Back projection of the normalized weights of the ROIs for five PET views onto the ICBM_152 template using 3D Slicer .
To summarize, we found that parts of the brain were disease-spared regions verified by both PET and MRI views. MRI views were capable of capturing the brain structural changes on temporal lobe, limbic gyrus, the ventricles, and part of the occipital lobe, which were usually shaped in the late course of the disease. The PET views, on the other hand, reflected the metabolic activities of the brain and were able to detect the early functional anomalies, therefore they tended to involve more ROIs in their patterns than the MRI views, especially in the frontal and parietal areas. In addition, some ROIs could only be detected by certain views, and led to distinct patterns. The differences of these patterns indicated that the disease had different effects on the brain and no single-view biomarkers were able to capture all the pathological changes.
### 3.2. Cross-view pattern analysis result
Table shows the KL divergence ( D ( Col || Row )) of the row item ( Row ) from the column item ( Col ) for these nine views. PET views had a low mean KL divergence of 16.6, which was close to that of MRI views 17.8. However, the mean KL divergence of PET views from MRI views ( D ( MRI || PET ) = 47.37) was markedly higher than that of MRI views from PET views ( D ( PET || MRI ) = 20.87). These results indicated that the views in the same modality usually look more similar than those in different modality. A typical example to show inter-modal and intra-modal differences was the GMV, which had limited divergence from other MRI views (LGI:6.83; CNV:2.31; SLD:3.99), but large divergence from PET views (M-IDX:48.09; F-IDX:31.13; DoG-M:49.95; DoG-C:39.28; DoG-Z:33.53). In addition, the MRI views always gain more information from the PET views than otherwise, e.g., D ( M - IDX || CNV ) = 51.80 is much greater than D ( CNV || M - IDX ) = 4.01; D ( DoG - M || LGI ) = 50.15 is also greater compared to D ( LGI || DoG - M ) = 9.21. The only exception was the pair of CNV and DoG-Z, both having high divergence from each other. As for the individual views, the divergence had a very wide range from the minimum D ( M - IDX || DoG - M ) = 0.35 to the maximum D ( SLD || DoG - C ) = 61.47.
The cross-view Kullback-Leibler divergence between different views .
To see how individual views related to each other, Figure displays their clustering results in a 2D space using the cross-view pattern analysis method described in Section 2.5. The blue color indicates the MRI views, the red color indicates the PET views, and the distance between two views in this coordinate system is proportional to their mutual divergence. We noticed that the MRI and PET views were clearly separated. More importantly, these views also formed clusters within the same modality. There were two clusters for the MRI views and two clusters for the PET views. The first sub-cluster (C1) for MRI included CNV, GMV, and LGI. All of these three features had strong correlation with the brain cortical atrophy, such as the loss of cortical neurons, the changes of cortical foldings. The second cluster (C2) for MRI had one isolated view only, SLD. Different from other MRI views, SLD focused on the shape changes of the brain caused by the disease. The third cluster (C3) contained three PET views, F-IDX, DoG-C, and DoG-Z. These views were effective in evaluating the consistency of the activity levels within a ROI, particularly when the ROI was partially hypo-metabolic. The M-IDX and DoG-M formed the fourth cluster (C4). These two views both were sensitive to the metabolic activity changes of the brain, which were important in the early detection of the AD and MCI.
The clustering results for the nine views in the 2D space . The structural and functional features are substantially separated when considering the first two eigenmodes.
### 3.3. Single-view classification performance
The classification performances of the individual views are summarized in Table . The best result of each performance metric is highlighted in bold-face. In general, PET views tended to have better performance than MRI views, especially on NC precision, MCI precision and the overall specificity. The only exception was DoG-Z, which had lower NC precision, MCI precision, accuracy and sensitivity compared to the MRI views. In addition, the sensitivity was always higher than the specificity across all the views, with an average difference of 30.44%. This was because we considered both AD and MCI as the positive class when computing the sensitivity and specificity. We argued that sensitivity was more important than specificity in this classification task, because the strong ability to detect the positive class (AD and MCI) would avoid treating the patients as normal subjects.
The classification performance of single-view biomarkers .
The bold value is the highest value in each column.
It was very clear that no single view could win all. F-IDX was the best view with the highest NC precision (53.30%), MCI precision (64.06%), overall accuracy (56.49%), and specificity (63.67%). GMV achieved the highest AD precision (67.64%), and DoG-M had the highest sensitivity (82.27%). One interesting discovery about the top three views (F-IDX, GMV, and DoG-M) was that they were from three distinct clusters (C3, C1, and C4), as described in Section 3.2. This fact implied that different views had different strengths in classification, and such strengths might be related to the clusters they belonged to.
### 3.4. Cross-view classification performance
Table shows the classification performance of 36 pairs of biomarkers in the same classification task. The best result of each performance metric is highlighted in bold-face. We separated the pairs into three groups according to their modalities, including six intra-MRI pairs, 10 intra-PET pairs, and 20 inter-PET&MRI pairs.
The classification performance of the 36 combinations of the multimodal neuroimaging biomarkers .
The bold value is the highest value in each column.
Most of the biomarker pairs could outperform the single biomarkers with marked improvements. Similar to the single-view biomarkers, none of the pairs could be leading in all aspects. The pair of CNV and DoG-M achieved the best precision for NC at 65.94%, which was 12.64% higher than the best single-view performance. They also had the highest sensitivity of 93.69%, increased the best single-view sensitivity by 11.42%. The pair of GMV and F-IDX performed best on MCI classification with a precision of 70.89%, improved the best single-view MCI precision by 6.83%. GMV also had the best performance on AD classification when paired with F-IDX, and their AD precision was 80.56%, 12.92% higher than the best single-view AD precision. F-IDX had the best single-view accuracy of 56.49%, and it further improved the accuracy to 67.37% when combined with DoG-M. The highest specificity was 69.92%, achieved by CNV and DoG-C with an increase of 6.25% compared to the best single-view specificity.
We noticed that the inter-MRI&PET pairs usually gave better results than the intra-PET and intra-MRI pairs. As detailed above, the best results were always obtained from the inter-MRI&PET pairs, except for the overall accuracy, which was achieved by two PET views. However, when the views were separated into different clusters as in Figure , we found the two views in the best pairs were always from different clusters with no exception, i.e., C1 (CNV) and C4 (DoG-M) achieved best NC precision and overall sensitivity; C1 (GMV) and C4 (DoG-M) had best MCI precision; C1 (GMV) and C3 (F-IDX) led in AD precision; C3 (F-IDX) and C4 (DoG-M) attained highest overall accuracy; and finally C1 (CNV) and C3 (DoG-C) achieved best specificity. C2 (SLD) was the only cluster that made no contribution to any of the best results.
In summary, there were two clear tendencies based on the cross-view results. First, the biomarker pairs could achieve much better results than the single-view biomarkers. Second, the best performance was always achieved by the views from different clusters.
## 4. Discussions
The mutual divergence was an effective measure to quantize the variability of the biomarkers. In this study, we identified four clusters of the biomarkers based on their mutual divergence, and the best joint performance in classification was always achieved by the combination of views from different clusters. However, it was not clear whether mutual divergence could be used as a general performance predictor for any two biomarkers. Therefore, we further asked this question, what could we expect from the biomarkers when combining them in classification.
To answer this question, we first looked at the correlation between the joint performance of the biomarker pair and performance of individual biomarkers. We used the E , E , and E to represent the joint performance, the higher performance and lower performance of the biomarkers. Table shows the Pearson's Correlation Coefficients (ρ) and the corresponding p -values with a significant value of 0.05. It was very clear that the multi-view classification performance was strongly correlated to the performance of individual views, and largely affected by the view with higher performance.
Pearson's correlation coefficient of the performance and the divergence .
We further examined the correlation between the joint performance and the mutual divergence D , as well as the higher KL divergence D and lower KL divergence D between the two views. A large D and a large D mean that the two views have dramatically different patterns, such as DoG-Z and CNV. A large D and a small D mean one pattern covers the other, such as M-IDX and CNV. If the D and D are both small, then the patterns are very similar, such as DoG-M and M-IDX. As shown in Table , the mutual divergence D did not show a correlation with the joint performance E , except that it had a weak anticorrelation with the accuracy (ρ = −0.1, p_value = 0.048). The higher KL divergence D had a weak correlation with the NC precision (ρ = 0.11, p_value = 0.040), whereas the lower KL divergence D , showed a decreasing linear relationship with the NC precision (ρ = −0.12, p_value = 0.027), accuracy (ρ = −0.17, p_value ~ = 0.001), and sensitivity (ρ = 0.11, p_value = 0.041). In other words, an increase of D or a decrease of D might lead to better classification performance. Large divergence does not necessary lead to better performance, as the views might not only create synergy, but also cause the interference to each other.
Therefore, our answer to the above mentioned question is that the multi-view performance is primarily determined by the performance of individual views. If one view's pattern covers the other, they tend to perform better than those with highly different or similar patterns.
There are also limitations of the datasets and the classifiers of this study. The datasets used in this study consisted of 331 subjects, but these subjects were not evenly distributed in each group, i.e., MCI patients accounted for over 50% of the entire population. The large disparity of number of patients in individual groups might have an impact on the SVM classifier. To offset such impact, we designed a cascade of three classifiers instead of only one classifier to increase the chance of classifying NC and AD. However, there might be a high correlation between classifiers, which might result in redundancy and consequent reduced performance. Such design is rather ad-hoc and would not be necessary for the future datasets with evenly distributed patients in each group. In this study, we adopted a design of 3-class classification (AD / MCI / NC) with a focus on the staging of the disease. However, such design poses great challenges to interpret our detected patterns of ROIs, since we don't see which regions are significant for AD or MCI. In addition, MCI is essentially a heterogeneous group and a substantial number of MCI subjects had primary non-AD pathologies, such as vascular dementia (VD) and frontotemporal dementia (FTD), as suggested by a recent ADNI study (Nettiksimmons et al., ). Therefore, it will be particularly useful to further distinguish the MCI subjects, including stable MCI patients not converting to other pathologies (ncMCI), and MCI converters who convert to AD (cMCI) or other pathologies. Totally nine views of neuroimaging biomarkers were investigated in this study. All of the biomarkers were based on the same template with 83 pre-defined ROIs, thus their patterns can be compared to each other. However, the multi-modal biomarkers might not always be ROI-based, such as the voxel-based features and the non-imaging biomarkers. In addition, certain biomarkers were able to bring additional information than the ROI-based features. For instance, the popular connectome (Wang et al., ) derived from DTI could not only capture the features of the ROIs, but also quantize the correlation between them. Currently, our cross-view analysis framework could quantitatively analyze and predict the synergy between two biomarkers. However, it is still very challenging to predict the synergy of more than two biomarkers.
## 5. Conclusions and future work
In this study, we presented a cross-view pattern analysis framework to quantitatively analyze the synergy between the multi-modal biomarkers derived from T1-MRI and FDG-PET, and predict their performance in AD and MCI classification. Several important conclusions can be draw based on the preliminary experiment results. Firstly, the single-view biomarkers had distinct pathology patterns, and no single-view biomarkers were able to capture all the pathological changes. Secondly, the MRI and PET views could be clearly separated, and the views in the same modality could also form different clusters, each depicting a certain type of pathological changes. Thirdly, the different views had different strength in classification, and the clusters could provide a good reference of their strength. Fourthly, the combination of biomarkers could achieve much better results than the single-view biomarkers, and the inter-cluster biomarkers always gave the best results. Last but not least, the multi-view classification performance was primarily determined by the performance of individual views, but we could use the divergence to estimate the trade-off between the interference and synergy and predict the performance.
For the future work, we would include more subjects into our datasets and refine the current design of classifiers to convey more meaningful findings on AD / NC; MCI / NC; cMCI / ncMCI; AD / cMCI / ncMCI / NC. Current framework could only test two views at a time. We will extend this framework to accommodate multiple (greater than 2) views using the multivariate methods. Another future direction is that we will employ this cross-view pattern analysis framework to investigate cross-ROI synergies, since many ROIs have been repeatedly reported in previous multi-modal neuroimaging studies, i.e., the pattern of AD pathology start mainly in the hippocampus and entorhinal cortex, and subsequently spreads throughtout most of the temporal lobe and the posterior cingulate, finally reaches the parietal, prefrontal and orbitofrontal regions (Fan et al., ; Desikan et al., ; Risacher et al., ). It would also be interesting to incorporate other non-imaging features, such as ApoE genotype (Pastor and Goate, ) or CSF concentrations of Aβ (Motter et al., ) and tau (Vandermeeren et al., ). We will investigate their single-view and cross-view patterns.
## Author contributions
All authors listed, have made substantial, direct and intellectual contribution to the work, and approved it for publication.
### Conflict of interest statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
The explosive growth of mobile technologies combined with the rapid rise of aging populations fearful of their risk for Alzheimer's disease has led to a number of marketed products aimed at enhancing cognitive health. However, an increasing number of product claims that are not substantiated has led regulatory agencies, such as the Federal Trade Commission (FTC), to issue warnings or penalties against some companies. Therefore, it is likely that a number of computerized cognitive training (CCT) companies will conduct clinical trials to prove their efficacy to gain Food and Drug Administration (FDA) clearance as a medical software/device. This raises a number of issues such as optimal trial design for establishing efficacy. The type of control condition is unique issue for CCT, given the variety of non-specific known to produce beneficial effects on cognition that are difficult to isolate from the content of the program. These include participant expectancy, engagement, motivation, novelty, and therapist interaction. We herein discuss the types of non-specific factors, desirable qualities of an active control condition, and the nuances that exist between previously used control conditions within the context of CCT for mild cognitive impairment.
## Expectancy
One nonspecific factor is expectancy, which refers to the participant's anticipation of positive or negative treatment effects. Expectancies come in two major forms: outcome expectancy, which is the belief that the treatment itself will result in a particular outcome, and response expectancy, which is the participant's subjective response to the treatment. Expectancies of both kinds impact outcomes across modalities, including psychotherapy (Goossens et al., ; Smeets et al., ; Weinberger, ), pharmacological interventions (Rutherford et al., ), and neurosurgery (Freeman et al., ). Long known to produce clinically significant outcomes, the belief that one will symptomatically improve is met with high response rates. For example, a meta-analysis of randomized placebo-controlled trials of antidepressants found the placebo effect to be responsible for the majority of observed change, in depression symptoms, with a response rate of 50% for those receiving medications compared to 30% for groups assigned to the placebo condition (Walsh et al., ).
Further evidence for the contribution of expectancy to treatment effectiveness can be found in the different response rates across different trial designs. Greater response rates have been observed in comparator trials between two drugs than in placebo-control trials in late-life depression (Sneed et al., ). This discrepancy may be a consequence of participants in comparator trials knowing they will receive treatment, while participants in placebo-controlled trials are hoping they are assigned to the treatment arm (and hence have lower expectations). This finding has been extended in middle-aged and adolescent depression (Rutherford et al., ) as well as in schizophrenia (Rutherford et al., ). Although these groups are diverse in terms of presenting ailment, consistent among them is that degree of expected improvement coincides with magnitude of actual improvement.
Rutherford et al. ( ) elaborated on a model of expectancy effects driving the differences in response rates between types of trials in which the orbitofrontal cortex (OFC), rostral anterior cingulate cortex (rACC), and nucleus accumbens (NAC) generate and maintain expectancies (Rutherford et al., ). The OFC and rACC are active during anticipation of pain and unpleasant experiences (Petrovic et al., ). Placebos lower activation of these regions which coincides with reduced severity of experienced pain (Wager et al., ). This analgesic effect may be partially explained by heightened opioid activity in the OFC and rACC that follows placebo administration and correlates with reported pain alleviation (Wager et al., ). Activation of the NAC, on the other hand, occurs in anticipation of reward. Greater activation of the NAC occurs with expectancy of analgesia and is correlated with placebo-induced pain reduction (Scott et al., ). Taken together, heightened expectancy of relief attenuates appraisal of negative experiences to a clinically significant degree.
In addition to altering perception of negative symptoms, raising expectancy increases participant's belief of self-efficacy in their functioning, making them more confident in their own capability to perform tasks. Judgments of high self-efficacy are associated with greater exerted effort in the face of challenges, as well as lengthened persistence (Bandura, ). A systematic review of placebo responses found self-efficacy and locus of control to be significant predictors of symptom improvement (Horing et al., ). The impact of response expectancy on cognition is particularly important in trials of CCT, given the use of performance-based outcome measures. In trials of CCT, participants are told that improvement in their cognitive functions is possible as a benefit from their involvement in the study. This creates an expectation of better performance following training, which is known to create positive cognitive outcomes. Indeed, participants who enrolled in a trial of CCT via a flyer suggesting CCT will improve working memory and fluid intelligence had greater post-training performance than participants who enrolled via a non-suggestive flyer, despite participating in the same program (Foroughi et al., ). Similarly, in a trial of healthy adults randomly assigned to a placebo pill or no-pill condition, those who took the placebo pill described as a “cognitive enhancer” had greater performance on tests of attention and delayed recall, though no effect was found in five secondary outcome measures (Oken et al., ). Given the potential effect of expectancy on cognitive performance, and the wide use of neuropsychological tests as primary outcomes in CCT trials, treatment, and comparison conditions must be balanced in terms of anticipated improvement.
## Engagement
How well a CCT program captivates and sustains attention will affect the capability of subjects to participate and focus during training. How engaging a task is depends on numerous factors, including usability, focused attention, positive affect, esthetics, endurability, richness, and control (Wiebe et al., ). These characteristics can be independent of the main program content. For example, a 2-back task necessitates working memory ability by nature of the design alone. By adding auditory feedback, sound effects, colorful animations, score tracking, and countdown timers, the task becomes more attractive to the participant. Such design elements create a nonspecific cognitive load, are not unique to CCT, yet make the CCT condition more interesting than the comparison group. When comparing an engaging CCT task to either waitlist or uninteresting control activities, it becomes impossible to isolate the central treatment effect.
## Motivation
Whereas engagement may be understood as emotional or attentional investment during a task, motivation is a global personal orientation to the activity (Wiebe et al., ). A participant may believe that cognitive improvement is contingent upon active involvement in their training. Numerous cognitive training platforms track then share performance data with participants. Following conclusion of a module, participants are shown their score, typically alongside a report of previous attempts. Congratulatory messages for surpassing earlier records are common. This positive feedback and potential for goal-setting behavior creates an environment where the participant is motivated to perform well.
Participant's beliefs about the malleability of their intelligence will affect their performance as well. In a study evaluating performance on tests of general knowledge, participants who held the believe that intelligence is malleable corrected more errors in their responses during retesting than participants who believed that intelligence is fixed (Mangels et al., ). When given negative feedback following errors, participants who believed in fixed intelligence demonstrated reduced memory-encoding activity in the temporal lobe. This suggests that one's attitude toward learning will affect the amount of effort placed in training tasks. This is particularly important for older adults with MCI, whose motivation toward the task may be increased by the prospect of preventing further cognitive decline.
## Novelty
The novelty effect is the tendency for improved performance due to interest in a fresh experience, rather than the content. A review of educational research found that when not controlled, novelty effects create an increase in test scores. On average, performance increases due to novelty by 50% of a standard deviation for the first 4 weeks, and by 20% of a standard deviation after 8 weeks (Clark and Sugrue, ). Novel experiences may also simply be remembered better than familiar ones. On tasks of explicit recognition, subjects show higher accuracy in identifying novel words from a studied list than familiar words viewed multiple times (Tulving and Kroll, ; Habib et al., ). Given that CCT consists of tasks not encountered in everyday life, and prior familiarity with training protocols is grounds for excluding participants, there is ample opportunity for sheer novelty of the task to constitute a considerable portion of any measured effects.
## Active controls
To address these issues, studies have employed active controls, where the comparison condition completes an activity designed to account for non-specific factors. This type of control condition aims to determine whether the benefits of such mental exercise are unique to CCT or can be obtained by any stimulating activity. Commonly used active controls include crossword puzzles, word searches, newspapers, and questionnaires. Active controls alleviate ethical issues of placebo-controlled treatments, as both the treatment and comparison condition may expect improvement. Although active controls are methodologically superior to waitlist conditions, differences in effect sizes between active control conditions and passive control conditions are not consistently observed (Karbach and Verhaeghen, ; Au et al., ), and the nature of the active control task may not account for all non-specific effects. Given the diversity of CCT paradigms and quantity of non-specific factors, researchers have argued that there is no universally applicable active control condition for all design cases (Boot et al., ). Instead, the control condition must be reasonably related to the tasks of the CCT condition such that participants expect improvement in the same cognitive domain regardless of their group assignment.
Table contains a list of studies of CCT in patients with mild cognitive impairment (MCI). Most studies use waitlist control conditions (Rozzini et al., ; Finn and McDonald, ) or control conditions that do not account for engagement and motivation in the task (Galante et al., ; Talassi et al., ; Gagnon and Belleville, ; Herrera et al., ; Carretti et al., ; Gaitan et al., ). In such designs, the treatment condition is at an unfair advantage because patients assigned to CCT have a greater chance of improvement simply because the tasks are engaging and motivating whereas those assigned to waitlist control lose interest and motivation. Even when active control conditions have been used in previous studies of CCT in MCI, they are not consistently computerized (Talassi et al., ; Herrera et al., ; Carretti et al., ; Gaitan et al., ) and consist of CCT tasks of invariable complexity (Gagnon and Belleville, ; Gaitan et al., ). Comparing CCT that scales in difficulty with participant performance to CCT that remains at a fixed difficulty does not determine whether effects are specific to CCT or can be obtained with any engaging computerized game. Further, trials of scaling difficulty fail to take into account the novelty of experience they introduce, beyond just harder problems.
Studies of CCT for MCI .
The CCT conditions themselves vary considerably, both by targeted cognitive domains and whether they are administered alone (Galante et al., ; Barnes et al., ; Finn and McDonald, ; Rosen et al., ; Gagnon and Belleville, ; Herrera et al., ; Carretti et al., ; Gooding et al., ), or as part of a wider intervention (Rozzini et al., ; Talassi et al., ; Gaitan et al., ). If CCT is administered alongside additional therapies, the magnitude of participant's expectancy, engagement, and motivation may be greater than if they were treated with CCT alone. Such nuances in design will impact the magnitude of the observed differences between the training and control groups, and the measured effect sizes across trials. One needs to consider the nonspecific factors present in both the CCT and control conditions in order to accurately interpret the results.
## Model control conditions
An adequate comparison condition must be matched for engagement, motivation, training time, computer interface, and novelty of stimuli. Preferably, the control group should account for nonspecific factors without introducing neurocognitive demands. If it does create specific effects, their impact on cognitive functioning should be known before comparing it to a CCT group. By using a CCT group, an active control group, and a passive control group, the relative contributions of each nonspecific factor can be estimated (Greenwood and Parasuraman, ). There is no “one size fits all” control condition. Instead, the selection of a control condition must be made in view of the content of the CCT platform and the specific research question.
An example of a well-balanced active control condition is a CCT program targeting domains of cognition that are different from those being targeted in the CCT program of the intervention condition. Should one group find greater improvement in cognitive domains of interest that transfer to untrained domains and quality of life, this can be taken as evidence that the content of CCT matters more than the non-specific factors. Participants in the active control group must expect the same types of cognitive benefits as those in the training group, principally by the control group and training group having identical descriptions of anticipated effects.
Another active control condition is one that incorporates non-adaptive versions of CCT. These programs use the same types of tasks as those in the experimental training condition, though they do not scale in difficulty with participant performance. When compared with adaptive versions of the same procedure, the tasks are balanced on participant expectancy and motivation, and practice effects can be ruled out. However, scaling difficulty introduces novelty, both in terms of activities encountered and strategies necessary for completing the task. Further, participants who complete the same task at unchanging difficulty may become less engaged as they reach their peak performance early. Controlling for practice effects is particularly important in older adults with cognitive impairment, given their propensity toward greater practice effects than their cognitively-stable peers (Suchy et al., ).
Remedial skills training has also been used as an active control condition. In these groups, participants discuss top-down strategies aimed at compensating with cognitive deficits. Such designs make it difficult to disentangle the plasticity of representations (knowledge, skills) from plasticity of processes (cognitive ability, brain function, brain structure, Fissler et al., ). It is possible that improvement following CCT is due to familiarity with neuropsychological tests and not due to improvement in underlying cognitive capacities. For example, learning to chunk numbers may improve digit span score, but may reflect one's ability to apply knowledge of cognitive strategies rather than one's actual attentional capacity. Comparisons between cognitive remediation and CCT are useful for balancing expectancy, motivation, and novelty, but offer little insight into the precise mechanism of any observed cognitive improvement.
## Looking forward
Future studies would benefit from inclusion of measures to evaluate non-specific factors as covariates. For example, the User Engagement Scale measures aspects of engagement, usability, and satisfaction on a 5-point Likert scale and comprises both negative (“I felt annoyed when on this site,” “the game was confusing”) and positive (“I really had fun,” “It was really worthwhile”) items (Wiebe et al., ). The Immersive Experience Questionnaire provides ratings of temporal distortion, challenge, emotional involvement, enjoyment, and attentional involvement in the task (Jennett et al., ). Expectancy can be evaluated using the Credibility and Expectancy Scale (Devilly and Borkovec, ). The CES asks participants to report how logical the therapy seems, how successful they expect the treatment to be, and whether they would recommend the treatment to a friend. Two factors, credibility, and expectancy, have been found within the CES, with expectancy ratings successfully predicting treatment outcome in a randomized controlled trial of cognitive therapies for generalized anxiety disorders.
Until CCT can be found to improve cognitive and everyday functioning after accounting for each non-specific factor, its future as a treatment remains uncertain. What is more certain is the usefulness of being generally mentally active. Indeed, high levels of mental activity may be associated not only with higher cognitive performance, but reduced risk of dementia. A systematic review of 22 population studies found mental exercises may reduce overall incident dementia risk by 46% (Valenzuela and Sachdev, ). Whether CCT is the same as any other form of mental activity or represents a unique method for augmenting cognitive processes shall be a continuing topic of investigation.
## Author contributions
JM: Manuscript conceptualization and design, writing of article. DD: Manuscript conceptualization and design, critical revision. PD: Manuscript conceptualization and design, critical revision. JS: Manuscript conceptualization and design, writing of article, critical revision. All authors have contributed to and have approved the final manuscript.
### Conflict of interest statement
PD has received grants and advisory/speaking fees from several pharmaceutical and technology companies including antidepressant manufacturers. He owns stock in several companies whose products are not discussed here. DD has received fees for scientific advisory boards from AbbVie, Lundbeck, and Intracellular Therapeutics. The other authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
By 2050, the aging population is predicted to expand by over 100%. Considering this rapid growth, and the additional strain it will place on healthcare resources because of age-related impairments, it is vital that researchers gain a deeper understanding of the cellular interactions that occur with normal aging. A variety of mammalian cell types have been shown to become compromised with age, each with a unique potential to contribute to disease formation in the aging body. Astrocytes represent the largest group of glial cells and are responsible for a variety of essential functions in the healthy central nervous system (CNS). Like other cell types, aging can cause a loss of normal function in astrocytes which reduces their ability to properly maintain a healthy CNS environment, negatively alters their interactions with neighboring cells, and contribute to the heightened inflammatory state characteristic of aging. The goal of this review article is to consolidate the knowledge and research to date regarding the role of astrocytes in aging. In specific, this review article will focus on the morphology and molecular profile of aged astrocytes, the consequence of astrocyte dysfunction on homeostatic functions during aging, and the role of astrocytes in age-related neurodegenerative diseases.
## Introduction
Aging is yet to be pinpointed to one specific factor or cell, but the current consensus is that aging is a multitude of factors (genetic, biological, environmental) that interplay together amongst the various cell types to contribute to physiological and cognitive changes characteristic of the aging process. Aging of the human brain is mainly observed as structural and physiological alterations that are most notably accompanied by cognitive decline. In terms of structure, imaging studies have shown that the aged brain displays a loss of dendritic spines (Pannese, ), decreases in oligodendrocyte (OL) number (Pelvig et al., ; Fabricius et al., ), stem cell loss in the hypothalamus (Zhang et al., ), and decreased brain volume across multiple neural regions (Ge et al., ; Resnick et al., ; Scahill et al., ; Raz et al., ; Freeman et al., ; Driscoll et al., ; Fjell et al., ; Li et al., ; Mu et al., ). The decrease in brain volume has been found to vary by sex, with males having more brain atrophy than females (Xu et al., ). Further, the reduced brain volume was hypothesized to be due to a loss of neurons (Brody, ; Coleman and Flood, ; Sturrock, ). However, a number of studies have shown that neuronal count does not change with age (Cragg, ; Terry et al., ; Freeman et al., ) and as such, the search for what was responsible (e.g., glia) for the structural changes became an active area of research.
With respect to physiological changes that occur with aging in the central nervous system (CNS), researchers observed an overall increase in gene expression in the brain (Berchtold et al., ; Boisvert et al., ). Although these genetic changes varied across brain regions and were sex dependent, with males showing more gene changes than females (Berchtold et al., ), there were some consistent changes. For example, across all brain regions, microglia and endothelial cells increased their gene expression levels, whereas astrocytes and OLs, but not neurons, shifted their expression pattern (Boisvert et al., ). Of note, these changes in gene expression in microglia, endothelial cells and astrocytes were of an inflammatory profile, which was intriguing given the well-known development of a mild inflammatory state with age (Howcroft et al., ; Morrisette-Thomas et al., ; Rea et al., ). Further, because the shift in gene expression pattern occurred in glial cells but not neurons, it implied that neurons are not a good indicator of age; instead, it has been proposed that glial specific genes may be a better predictor of age (Soreq et al., ). This review article will focus on what is known thus far about changes that occur to astrocytes during aging, the consequence of astrocyte dysfunction during aging on other resident CNS cells, and the role of astrocytes in age-related neurodegenerative diseases.
## Astrocytes
Astrocytes were first observed in the brain by Camillo Golgi in 1871. They are the largest cell population in the CNS with a ratio of astrocytes to neurons in the human cortex averaging 1.4:1 (Nedergaard et al., ). Astrocytes also represent the most heterogeneous group of glial cells and tile the entire CNS in a non-overlapping manner (Sofroniew and Vinters, ). Once considered to be no more than “brain glue” that provides structural support through extracellular matrix formation, astrocytes have now become known for key roles in a variety of complex and essential functions. A few of these functions include synapse development, neurotransmitter homeostasis, glycogen storage and blood brain barrier (BBB) maintenance (Sofroniew and Vinters, ). Due to their multi-faceted roles, it is not surprising that astrocytes are involved in aging.
### Phenotype and Molecular Profile of Aged Astrocytes
Astrocytes represent the most heterogeneous group of glial cells in terms of molecular, structural and physiological profiles and their classification relies on both morphological and molecular criteria. These glia are morphologically diverse, but are typically stellate-shaped cells with radial processes (Oberheim et al., ). With age in humans, there is an alteration in astrocyte phenotype from long and slender processes in young subjects to short and stubby processes in older individuals (Kanaan et al., ; Cerbai et al., ; Jyothi et al., ). This age-related morphological change has also been shown to occur in aged rodents (Castiglioni et al., ; Amenta et al., ) and primates (Kanaan et al., ; Robillard et al., ). With respect to numbers, there appears to be region-dependent differences in their density where reductions were noted in the retina of aged rodents (Mansour et al., ), while no changes were observed in the hippocampus (Lindsey et al., ; Jinno, ), and an increase seen in the human cortex (Hansen et al., ) and hypothalamus (Wang et al., ).
With respect to molecular criteria, glial fibrillary acidic protein (GFAP) has been used as a classical marker for astrocyte identification; although it is important to note that GFAP is not detectable in all astrocytes (Sofroniew and Vinters, ). Reports on aging astrocytes have shown that GFAP expression increases (Nichols et al., ; Kohama et al., ; Yoshida et al., ; Rozovsky et al., ; Wu et al., ; Clarke et al., ) and since an increase in GFAP expression is a common feature of reactive/activated astrocytes (Zamanian et al., ; Sofroniew, ; Liddelow et al., ), these findings suggest that astrocytes become reactive with age. More evidence in support of this notion was provided when it was found that astrocyte reactivity involves the differential expression of over 1,000 genes, some of which have been collectively attributed to “A1” reactive astrocytes—astrocytes of a more inflammatory state (Zamanian et al., ). Some of these “A1” reactive genes (Zamanian et al., ), such as complement system factors, antigen presentation molecules, secretory factors, peptidase inhibition and cholesterol synthesis, were recently identified as additional molecular profiles of aged astrocytes (Boisvert et al., ; Clarke et al., ). In addition, there is evidence that aged astrocytes undergo epigenetic changes and may alter BBB function and circadian rhythm.
### Aging Astrocytes and the Complement System
The complement system is part of the innate immune system, and aids in regulating inflammation as well as resistance to infection (Markiewski and Lambris, ). The system consists of approximately 30 soluble factors that are widely expressed in neurons and glia in the postnatal brain (Stevens et al., ; Hammad et al., ). C1q and C3 are major complement proteins that allow for cells to be tagged and targeted for phagocytosis (Stevens et al., ). During development, postnatal neurons express C1q at synapses which appears to target unwanted synapses for elimination (Stevens et al., ). In the mature brain, the complement system is dampened, however studies have shown that during CNS disease the system becomes enhanced by various cells, including astrocytes, and can lead to pathology via synapse, dendritic spine and neuronal loss, as well as changes in neuronal morphology (Stevens et al., ; Stephan et al., ; Zamanian et al., ; Lian et al., , ). An increase in the genes belonging to components C3 and C4B is seen in astrocytes during aging (Boisvert et al., ; Clarke et al., ). C4B is a protein involved in both the classical and lectin pathway of the complement system. It is also a substrate that allows C3 convertase to cleave C3 into C3a and C3b, which direct both of these pathways toward opsonization and cell lysis. C3 is required for both these pathways, however it can also function on its own through the alternative pathway where it is involved in inflammation, cell migration, activation and cell lysis. This suggests that astrocytes are able to functionally partake in all three pathways of the complement system.
The activation of complement by aged astrocytes may contribute to cognitive decline. More specifically, because overall neuronal loss was not seen in aging humans, it has been suggested that changes in neuronal structure may instead be the culprit responsible for age-associated cognitive impairment (Berchtold et al., ). Since the current consensus is that memories are formed when neurons fire to one another, and recollection of that memory strengthens the associated synapses (Mayford et al., ), it is feasible that loss of a strengthened connection (synapse) between two neurons via complement targeting by reactive aged astrocytes could contribute to the memory deficits characteristic of elderly people (Choi et al., ).
### Antigen Presentation and Aging Astrocytes
The aging astrocyte also shows an upregulation in genes comprising major histocompatibility complex I (MHC I; Orre et al., ; Boisvert et al., ; Clarke et al., ), which is in agreement with other literature showing that there is an overall age-related increase in brain MHC I (Mangold et al., ). Interestingly, H2-K1 and H2-D1, both of which are upregulated in aging astrocytes and form part of MHC I, are associated with increased expression of pro-inflammatory genes (Mangold et al., ), implying that astrocytes could contribute to the low level inflammatory state that is characteristic of aging. That is, although contentious as to whether astrocytes present antigens in vivo , an increase in MHC I in astrocytes would suggest that these glia have an increased propensity to present antigens. More specifically, even though these glia are not professional antigen presenting cells (APCs) by definition, and they present antigens weaker than professional APCs, if only a subset of astrocytes presented antigens, the amount of overall antigen presentation would be large based on the sheer number of this glial cell in the CNS. Further support for the antigen presenting ability of aged astrocytes in vivo is their significant expression of various factors involved in phagocytic pathways, including Pros1, Mfge8, Megf10, Lrp1 (Clarke et al., ); Mertk expression was not significantly increased, but was still expressed by aged astrocytes (Clarke et al., ). These results suggest that astrocytes possess the tools needed to tag ( Mfge8, Pros1, C3b ), phagocytose ( Mertk, MegF10, Lrp1 ) and present ( H2-K1, H2-D1 ) antigens. However, Megf10, Mertk and Lrp1 expression does not necessarily signify negative consequences such as inflammation, because astrocytes may utilize these receptors for debris clearance. That is, astrocytes were found to possess lipid inclusions which contained myelin in the aging human optic nerve suggesting that astrocytes may be trying to clear myelin debris during aging (Nag and Wadhwa, ). Furthermore, apart from its immunological role, MHC I has been shown to be increased significantly with age in cognitively-intact individuals, but decreased in cognitively impaired individuals (Lazarczyk et al., ). This suggests that MHC I expression may have a positive role in brain aging by trying to preserve cognitive function. Thus, the antigen presentation capability of astrocytes with age may be both advantageous and detrimental depending upon the context. Further work needs to be done to elucidate exactly how these phagocytic factors in aged astrocytes are contributing to aging and whether they are detrimental, protective, or both.
### Aged Astrocytes and Secretory Molecules
Another characteristic feature of aged astrocytes is an increase in production of cytokines such as CXCL10/inducible protein-10 (IP-10; Clarke et al., ). CXCL10 serves as a chemoattractant for peripheral immune cells, and aids in T cell adhesion to endothelial cells (Sorensen et al., ). Interestingly, the receptor for CXCL10 is CXCR3, a microglia marker, suggesting that astrocytes and microglia may be communicating with each other during aging. This is entirely plausible as recent literature shows that microglia are able to induce reactivity in astrocytes (Liddelow et al., ; Rothhammer et al., ).
CXCL5 is another cytokine whose expression is augmented in aged cerebellar astrocytes (Boisvert et al., ). Like CXCL10, CXCL5 has chemotactic capabilities and has been shown to be involved in neutrophil recruitment , and like astrocytes, neutrophils display aging deficits including an increase in their inflammatory properties (Adrover et al., ). Astrocytes have been shown to have both a direct (cell to cell) and indirect effect on polymorphonuclear neutrophils (Xie et al., ). That is, astrocytes are able to attenuate pro-inflammatory cytokine expression in neutrophils through direct contact, while an indirect interaction has been attributed to enhanced neutrophil necrosis (Xie et al., ). Therefore, it is plausible that aged astrocytes and aged neutrophils could synergistically elicit a heightened inflammatory response.
In addition to chemotactic and survival effects on immune cells via cytokine release, aged cortical astrocytes may negatively impact CNS health because of a decline in production of certain metabolic and trophic factors such as ATP and neurotrophins. For instance, a decline in ATP release may contribute to cognitive decline and impaired synaptic plasticity because ATP is involved in regulating neuronal activity through synaptic and tonic inhibition (Lalo et al., ). In addition, neuronal survival and neurogenesis may be decreased due to a reduction in the secretion of neurotrophins like vascular endothelial growth factor (VEGF), fibroblast growth factor 2 (FGF2; Bernal and Peterson, ) and brain derived neurotrophic factor (BDNF) in aged astrocytes (Bellaver et al., ).
### Oxidative Stress and Aging Astrocytes
Oxidative stress is a key component to aging and age-related diseases, including neurodegenerative diseases. The accumulation of oxidative damage, specifically to lipids, DNA, and proteins (Dizdaroglu et al., ; Cooke et al., ; Evans et al., ; Swain and Subba Rao, ), is proposed to result in age-associated functional losses in a process referred to as the free radical theory of aging or oxidative stress theory of aging (Beckman and Ames, ; Liguori et al., ). Indeed, increased reactive oxygen species (ROS) which cause damage to lipids, proteins, and DNA is frequently observed in aged tissue (Kudin et al., ; Brawek et al., ; Lukiw et al., ).
Middle-aged astrocytes have been shown to accumulate ROS and display an overload in Ca (Ishii et al., ). Moreover, the over-abundance of Ca was associated with an increase in JNK/SAPK activation, which is a member of the stress-activated MAPK signaling pathway, and which has been linked to cell death signaling (Ishii et al., ). Of further note, astrocytes that have been exposed to oxidative stress factors during aging begin to undergo oxidative stress themselves (Lei et al., ; Lu et al., ; Bellaver et al., ), and in age-associated neurodegenerative diseases such as multiple sclerosis (MS), astrocytes in active lesions have been shown to increase their accumulation of oxidized lipids and proteins in their cytoplasm (van Horssen et al., ) and oxidized DNA in their nuclei (Haider et al., ). Accumulation of oxidative stress suggests that these glia can no longer support neurons (Lin et al., ) which could lead to decreased neuronal function and damage as seen in the disease. However, whether this damaging mechanism really does occur in MS is unknown because astrocytes in active MS white matter (WM) lesions also augment expression of mitochondrial antioxidants (Nijland et al., ; Licht-Mayer et al., ; Lassmann and van Horssen, ). More specifically, enhanced expression of peroxisome proliferator-activated receptor-gamma coactivator 1-alpha (PGC-1α) was observed in reactive astrocytes, and astrocytes that over-express PGC-1α not only secrete less pro-inflammatory cytokines and chemokines, but they protect neuronal cells against oxidative attack greater than those co-cultured with control astrocytes (Nijland et al., ). Further, Nuclear factor erythroid 2-related factor 2 and its stabilizer and positive regulator, DJ1, is another antioxidant that is increased in reactive astrocytes in both active and chronic active MS lesions (van Horssen et al., ), and found to be protective against ROS in co-cultured neurons (Shih et al., ; Kraft et al., ; Calkins et al., ). Altogether, these findings imply that astrocytes are actively trying to combat oxidative damage both within themselves and neighboring cells, but their protective measures are insufficient to combat the overwhelming oxidative damage occurring with age and during age-associated neurodegenerative diseases.
### Aged Astrocytes and Peptidase Inhibition
Serine proteases control many key elements of the immune system including granzymes which activate apoptotic pathways, complement system proteins that mediate inflammation and phagocytosis, and production of cytokines and chemokines (Safavi and Rostami, ). These proteases have been linked to aging. For example, an increase in serine protease HTRA1 in mouse retinal pigment epithelium may contribute to age-related macular degeneration due to degeneration of the epithelium (Jones et al., ). Further, a significant increase in the serine proteases plasmin, trypsin and elastase was found in the blood of aged rats and appear to be involved in extracellular matrix degradation (Paczek et al., ). In aged astrocytes, serine protease inhibitor called Serpina3n is significantly upregulated (Boisvert et al., ; Clarke et al., ) which may be an attempt by aged astrocytes to combat the aging effects of serine proteases.
### Cholesterol Synthesis
Due to the BBB, the CNS does not uptake cholesterol from the blood stream, and instead has to locally synthesize the majority of cholesterol that it needs. Astrocytes are believed to play a major role in brain cholesterol synthesis because of their expression of sterol regulatory element-binding protein 2 (SREBP2; Ferris et al., ). SREBP2 activates the transcription of enzymes, including HMG-CoA reductase ( Hmgcr ), that are needed for cholesterol synthesis and cholesterol uptake receptors, including the LDL receptor (Madison, ). SREBP2 presence in astrocytes is critical for CNS function since its knockout in astrocytes in mice results in impaired brain development and function and reduced neurite outgrowth (Ferris et al., ). In the aged murine brain, the rate limiting cholesterol synthesis enzyme Hmgcr is downregulated, while the receptors for cholesterol transport are increased in astrocytes (Boisvert et al., ). This suggests that there is an overall dysregulation of the cholesterol synthesis pathway in astrocytes, and since many cells, including neurons, rely on cholesterol from astrocytes (Zhang et al., ), this dysregulation may lead to metabolic disruption in these nerve cells.
### Epigenetics of Aging Astrocytes
Studies that have looked at overall methylation changes during aging have shown that both hyper- and hypo-methylation occur with age at comparable rates (Maegawa et al., ; Issa, ). In addition, histone modification has also been shown to occur in aging cells (Wood et al., ; Liu et al., ). DNA methylation is an important process in the development of astrocytes since demethylation of astrocyte-specific genes such as GFAP, S100β and Aqp4 in neural stem cells (NSCs) promotes the switch from neurogenesis to astrogenesis (Namihira et al., , ; Hatada et al., ; Takouda et al., ). In mature astrocytes, global DNA methylation patterns have been shown to occur in psychiatric disorders (Nagy et al., ) and alcohol use (Miguel-Hidalgo, ), however our understanding of epigenetic changes in astrocytes during aging is limited. Chrisholm and colleagues noted that H3K4 specific methyltransferase activity was lower in astrocytes from middle aged vs. young female rats following ischemia (Chisholm et al., ). As such, they proposed that future therapies may be able to target epigenetic modifications to provide neuroprotection against aging in astrocytes (Chisholm et al., ). Indeed, it has been suggested that inhibitors of histone deacetylases (HDACs) may be neuroprotective via effects on glial cells (Staszewski and Prinz, ). Some evidence in support of this presumption was seen in astrocyte cultures where it was shown that HDAC inhibition increased neurotrophic cytokines (Chen et al., ) and mitigated changes in Parkinson’s and Alzheimer’s disease (PD and AD; Nuutinen et al., ), although no changes were seen in reducing astrocyte activation (Xuan et al., ).
Environmental factors can also induce epigenetic changes. In a recent study that assessed how environment impacted young vs. old astrocytes, it was reported that aged mice from a standard environment could not distinguish between stationary and displaced objects (Diniz et al., ). Upon further examination of the astrocytes in these animals, the authors found two distinct morphological phenotypes and speculated that aging and environment reduce the complexity of astrocytes (Diniz et al., ). This agrees with previous literature showing morphological changes between young and old astrocytes (Kanaan et al., ; Cerbai et al., ; Jyothi et al., ). Furthermore, disturbances in astrocyte mitochondrial function have been shown to occur with age and these changes can also be seen when astrocytes are exposed to environmental neurotoxicants such as 3-chloropropanediol (Cavanagh et al., ), 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP; Sundar Boyalla et al., ) and manganese (Zheng et al., ).
### Blood Brain Barrier (BBB) and Aged Astrocytes
The BBB consists of endothelial cells, astrocytic endfeet processes, pericytes, smooth muscle cells, neurons and perivascular microglia (Abbott and Friedman, ; Banks, ; Osipova et al., ). Considering that the BBB is comprised of such a variety of cell types, it is evident that disturbances in any one of these cells could result in changes to BBB integrity. The BBB is a compact complex held together by tight junctions which limits the diffusion of molecules and migration of cells from freely entering and exiting the brain. Transport of substances and cells that are required for CNS homeostasis occurs mainly by transcellular routes and acquisition of small molecules requires expression of specific transporters in order to pass across the BBB (Goodall et al., ).
As mentioned, astrocytes are an integral part of the BBB as they are involved in its development, maintenance and regulation. For example, astrocytes secrete Sonic Hedgehog (Shh) while BBB-endothelial cells bear the receptor for Shh (Alvarez et al., ). Blocking the Hedgehog pathway in astrocytes was found to increase barrier permeability while a lack of Shh led to compromisation of the barrier phenotype (Alvarez et al., ). Astrocytes also secrete a variety of factors that influence the function and permeability of the BBB. For instance, inflammatory factors such as nitric oxide, IL-6, TNF-α and matrix metalloproteinases, which astrocytes are known to secrete when they acquire an inflammatory state, have been shown to impair vasculature control and function (Zhang, ). A common pathway that is involved in the production of these inflammatory factors includes the NF-κB pathway, which has been shown to have a direct effect on the tight junction proteins occludin, ZO-1 and claudin-5 (Lamberti et al., ; Stamatovic et al., ; Aveleira et al., ). It is also important to note however, that astrocytes can also be protective in BBB function. This includes the NF-κB pathway, which has been shown to promote barrier function by inducing the expression of Shh (Kasperczyk et al., ).
BBB changes have been documented in neurodegenerative diseases, which are typically associated with age progression (Kirk et al., ; Zlokovic, ). However, changes in BBB permeability with age are conflicting as some groups have shown an increase in permeability with age (Mooradian and McCuskey, ; Hosokawa and Ueno, ; Kirk et al., ; Hafezi-Moghadam et al., ; Farrall and Wardlaw, ; Popescu et al., ; Goodall et al., ) while others have shown no alteration (Wadhwani et al., ; Vorbrodt and Dobrogowska, ; Banks et al., ; Mackic et al., ). Due to this dichotomy, further research needs to be done to clarify the changes that are occurring with age, and specifically how the aged astrocyte contributes to these changes. Based on the changes that we know occur in aging astrocytes we can speculate that if there is age-dependent BBB alterations, aged astrocytes could contribute to any phenotype and permeability impairment because of their inflammatory secretions.
#### Chronodisruption and Aged Astrocytes
An interesting observation about aged astrocytes is their association with chronodisruption. Circadian rhythm generates a time clock to a 24 h cycle that influences our behavior and physiology and adapts us to light/dark cycles of the earth (Kondratova and Kondratov, ; Hood and Amir, ). Our circadian rhythm is generated in the suprachiasmatic nucleus (SCN) of the hypothalamus via a transcriptional/translational feedback loop (Golombek and Rosenstein, ; Duhart et al., ). Briefly, the loop begins when transcription factors, CLOCK and BMAL1 heterodimerize and initiate transcription of Per and Cry genes. PER and CRY proteins form a complex in the cytoplasm and block transcription of CLOCK/BMAL1, which in turn inhibits transcription of Per and Cry . PER/CRY complex is also targeted for degradation via the proteasomal pathway, which initiates the transcription of CLOCK/BMAL1 again. This feedback loop forms the core oscillator mechanism (Lowrey and Takahashi, ; Duhart et al., ). The SCN is composed of a variety of classes of neurons and astrocytes (Abrahamson and Moore, ). Recent work has demonstrated that SCN neurons are metabolically active during daytime while SCN astrocytes are active during night and that these astrocytes are able to suppress the activity of SCN neurons through extracellular glutamate (Brancaccio et al., ).
Disruption of circadian rhythm has been suggested to be carcinogenic to humans (Straif et al., ) and has been well documented to be altered during aging and neurodegenerative diseases of aging (Kondratova and Kondratov, ; Mattis and Sehgal, ). Oxidative stress has not only been proposed as a mechanism of aging, and is indeed a common feature amongst aging neurodegenerative diseases, but it is speculated to contribute to changes in circadian rhythm with age (Grimm et al., ; Kondratova and Kondratov, ; Krishnan et al., ). As discussed above, oxidative stress negatively impacts astrocytes by reducing their ability to support neurons. Because the neuron-astrocyte relationship is integral in the SCN, disruption of proteins in the SCN could perturb this collaboration. Indeed and for example, BMAL1 deletion in SCN neurons can cause partial astrocyte activation, while BMAL1 deletion in SCN astrocytes induced astrogliosis and astrocyte dysfunction (Lananna et al., ). In addition, circadian disruption can induce GFAP expression (Lananna et al., ), while aging (Wyse and Coogan, ) and inflammation (Curtis et al., ) have been shown to suppress BMAL1 levels, thus potentially influencing astrocyte activation state as well (Lananna et al., ). However, loss of BMAL1 in astrocytes does not disrupt circadian rhythm (Tso et al., ). Instead, it was found that deletion of CK1ε, an enzyme that controls phosphorylation of PER/CRY in SCN astrocytes, resulted in an increase in behavioral period in mice (Tso et al., ). These results suggest a feedback loop involving astrocytes and neurons in circadian rhythm impairment. That is, the reciprocal relationship between astrocyte and neurons in the SCN suggests that a disruption in astrocytes would in turn affect neurons resulting in disruption of the feedback loop.
#### Summary
The majority of changes in aged astrocytes may be fueling the pro-inflammatory processes that occur with aging—albeit there are instances (e.g., serine protease inhibition) where aged astrocytes are trying to combat detrimental mechanisms (Figure ). Since the majority of these alterations are of the pro-inflammatory nature, it appears that the anti-inflammatory processes within astrocytes and other CNS cells are inadequate to quell the inflammatory properties, thus contributing to a heightened inflammatory state as we age. It is likely also that the enhanced expression of pro-inflammatory changes in aged astrocytes has effects beyond these glia by influencing the function of other CNS cells, the subject of the next section.
Schematic representation that summarize the phenotype change of astrocytes as they age (A) and the molecular profile of aged astrocytes (B) .
## Aged Astrocytes and Other CNS Cells
### Microglia
It was a long-held dogma that glial cells only provided support to neurons, however recent research suggests a much more complex and active interplay between glial cells themselves and their relationship with neurons. While more work is needed to definitively elucidate the relationship between different glial cells during aging, there are some insights into what is occurring amongst these cells during senescence. In a model of aging neuroinflammation in the hippocampus, astrocyte branches were found to be actively bisecting the cell body of neurons while microglia contained neuronal debris (Cerbai et al., ), suggesting that astrocytes and microglia may both be involved in eliminating neurons during aging. Whether these are joint efforts or each cell performing their function autonomously is however unclear. Recent work from the Barres lab demonstrated that activated microglia are able to influence astrocyte activation through their secretory profile (e.g., C1q, TNF-α and IL-1α; Liddelow et al., ) implying that microglia could influence astrocyte activation during inflammation. In the context of aging, microglia have been shown to increase their reactivity (Hickman et al., ; Norden and Godbout, ; Grabert et al., ), which would suggest that they would augment their secretory profile. There is some support for this idea since it has been shown that the increase in reactive astrocyte genes associated with aging was significantly reduced in mice lacking C1q, TNF-α and IL-1α (Clarke et al., ). Of no surprise, this feedback loop between microglia and astrocytes is also present when trying to resolve inflammation, of which becomes disrupted in the aging brain. More specifically, activated microglia secrete the anti-inflammatory cytokine IL-10. Astrocytes express the IL-10 receptor (IL-10R) and upon activation of this receptor, secrete TGF-β, which in turn decreases microglial activation (Norden et al., ). However, aged astrocytes do not increase their expression of IL-10R or secretion of TGF-β, which would release the brakes needed to suppress astrocyte and microglial activation during aging (Norden et al., ). Therefore, astrocytes and microglia function both autonomously and non-autonomously in aging and age-related neurodegenerative disorders to contribute to neuronal loss and inflammaging of the brain (Franceschi et al., ).
### Neurons
Astrocytes support neuronal function in many ways including ion homeostasis, metabolic support and transmitter homeostasis (Sofroniew and Vinters, ). As mentioned previously, neuronal numbers do not change with age (Cragg, ; Terry et al., ; Freeman et al., ), but their function and morphology change with age (Pannese, ). This led researchers to explore whether age-related changes in astrocytes could impact neuronal function. Indeed, astrocytes gain an “A1” phenotype with age, and previous work with this phenotype found that “A1” astrocytes secrete a neurotoxin that kills neurons (Liddelow et al., ). In addition, aging astrocytes increase expression of Sparc (Boisvert et al., ), which blocks synapse formation and decreases α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (Nobuta et al., ; Allen, ) thereby potentially inhibiting synapse function of neurons. As mentioned previously also, upregulation of the complement system factors, C3 and C4B in aged astrocytes, suggests that these glia have the potential to eliminate neuronal synapses (Boisvert et al., ; Clarke et al., ). Finally, disruption in lipid metabolism (see “Cholesterol Synthesis” section) could lead to impaired synapse formation and function of neurons (van Deijk et al., ). However, it should be noted that the majority of astrocyte homeostatic functions involved in neuronal support are unaltered with age, including genes involved in ion, glutamate and lactate regulation (Boisvert et al., ). This suggests that astrocytes are still capable of supporting certain neuronal functions with age and thus the balance between the various beneficial and detrimental effects will determine the health of a neuron. For instance, although the majority of astrocyte support of neuronal cells is not altered, the changes that occur in every aged cell (neuron, microglia, OL) that influence neuronal function may eventually add up and tip the balance towards an impaired neuron.
### Oligodendrocytes
In regard to OLs, there is some evidence that astrocyte dysfunction with age could negatively impact the function of these cells. OLs synthesize myelin and wrap axons that are seen as WM in the brain and spinal cord. Aging results in a 30% decrease in WM and a 45% decrease in myelinated fiber length from the age of 20–80 years old (Bartzokis et al., ; Marner et al., ). While myelin production still occurs as we age, aging results in thinner myelin sheaths and shorter internodes (Marner et al., ) that can be traced back to OL dysfunction. The hypothesis is that due to oxidative stress, OLs suffer from DNA damage that accumulates during aging to result in WM changes (Tse and Herrup, ). In addition to OLs, precursors for these glia may also contribute to the WM changes with age. In the mature CNS, a pool of progenitor cells known as OL progenitor cells (OPCs) reside to serve as replacements for OLs. While the OPC pool remains constant throughout age (Boda et al., ) there is an inherent defect in their differentiation with senescence. Their differentiation time is doubled (Zhu et al., ) and there is an impairment in their recruitment to replace OLs (Doucette et al., ).
Astrocytes are known to support OL function including proliferation, differentiation and myelination. While OLs synthesize cholesterol to produce myelin they also rely on cholesterol synthesis from astrocytes for their myelin production; however, as discussed above, cholesterol synthesis is impaired in aging astrocytes. As a consequence, OLs may not receive the necessary amount of cholesterol during aging, thus resulting in reduced myelin production characteristic of the aging brain. Another process of OLs that aged astrocytes may disrupt is differentiation. Before OLs can produce myelin they need to undergo differentiation from their OPC state. FGF2, an important factor in OPC differentiation (Bögler et al., ), is decreased in aged astrocytes (Bernal and Peterson, ), suggesting that astrocytes may hinder the differentiation process.
In contrast to the detrimental effects aging astrocytes may have on OLs and OPCs, they may also be potentially beneficial. Astrocytes increase their expression of CXCL10, which is involved in the migration of OPCs (Omari et al., ), and an increase in the inflammatory state of aging astrocytes coincides with an increase in STAT3 (Monteiro de Castro et al., ), that may be in involved in crosstalk with microglia and OPCs to prevent impairment of OPC maturation (Nobuta et al., ). It is possible however, while these mechanisms are integral in recruiting OPCs to sites that need OL repopulation and differentiation, the astrocyte-contributing defects in differentiation and reduced myelin production may overcome these beneficial properties of aged astrocytes on OL functions.
### Summary
The interactions of aged astrocytes with other CNS cell populations suggest that they can propagate inflammation and directly affect the health and function of these other cells (Figure ). Since all of these cells communicate with each other, a direct effect of astrocyte interactions with any one of these cell populations can create a feedback loop that results in the dysfunction of multiple cell types and ultimately functional impairment as seen in age-related diseases.
Schematic representation of the relationship between aged astrocytes and other CNS cells during senescence.
## Aged Astrocytes and Neurodegenerative Diseases
In 2015, the aged population (aged 65 and over) represented 8.5% (or 617.1 million) of the total population (He et al., ). By 2050, it is predicted that the aging population will rise to over 16.7% or 1.6 billion (He et al., ). That is a 150% expansion of the aged population (He et al., ). With age comes an increased risk for age-related diseases, which includes neurodegenerative diseases. Aging and neurodegeneration share many features including gliosis, loss of myelination, cognitive decline, loss of working memory and astrocyte dysfunction. AD, PD and amyotrophic lateral sclerosis (ALS) are three common age-related diseases known to have astrocyte involvement. There are many in-depth reviews on AD, PD and ALS and thus we will limit our discussion to what is known about the role of astrocytes in these diseases.
### AD and Astrocytes
AD affects more than 35 million people worldwide and is the most frequent form of dementia in the aged population, accounting for 50%–56% of cases (Querfurth and LaFerla, ). Aging does not mean individuals will progress to AD, however, AD and aging share many common features including oxidative stress, mitochondrial dysfunction, inflammation, proteotoxicity and altered gene expression (Chakrabarti and Mohanakumar, ). These common features can also be found in AD astrocytes. For example, neurons in AD have been shown to increase insulin-like growth factor 1 (IGF-1) via astrocyte interactions, which can lead to increased beta-amyloid formation by neurons (Costantini et al., ). Beta-amyloid, in turn, is capable of stimulating astrocytes and causing activation of NF-κB pathways (considered an important mediating agent in neuroinflammation in AD; Akama and Van Eldik, ; Wang et al., ; Shi et al., ) and consequently production of pro-inflammatory cytokines such as IL-1β, IL-6, iNOS and TNF-α (Bales et al., ; Akama and Van Eldik, ; Hou et al., ). Other evidence of astrocyte involvement in AD pathogenesis is a decline in antioxidant defense, which could result in the accumulation of oxidative damage in astrocytes leading to neurodegeneration (Hung et al., ). For example, oxidative stress can result in activation of astrocytes (Andersen, ) and activation of NADPH oxidase in astrocytes, which has been shown to lead to beta-amyloid-induced neuronal death through mitochondrial dysfunction (Abramov et al., ).
### Astrocytes and PD
PD is the second most common age-related disease after AD (Reeve et al., ). Age is the strongest risk factor for PD, with prevalence of the disease increasing more than 400 times as one ages (Rodriguez et al., ). Although volume reduction can be seen throughout the aging brain, the substantia nigra dopaminergic neurons are particularly affected in PD displaying neuronal loss at a rate of 4.7% to 9.8% per decade (Fearnley and Lees, ; Ma et al., ; Rudow et al., ). It is incompletely known why dopaminergic neurons are targeted, but it has been suggested that astrocytes may be involved. Astrocytes and neurons share a very close relationship due to the fact that astrocytes are part of the tripartite synapse, provide structural and metabolic support, buffer neurotransmitters, and help regulate synaptic transmission (Sofroniew and Vinters, ). Because these functions become compromised in the aged astrocyte, it is likely that neuronal health would be directly impacted. Furthermore, genes that have been identified as part of astrocyte biology are also causative in the development of PD; these include PARK7, SNCA, PLA2G6, ATP13A2, LRRK2, GBA, PINK1 and PARK2 (Zhang et al., ; Booth et al., ). These genes are involved in a variety of functions in astrocytes including inflammatory responses, cholesterol synthesis, and mitochondrial dysfunction which are also known to be dysfunctional in PD.
### ALS and Astrocytes
ALS is a degenerative disease of motor neurons in the brain, brain stem and spinal cord, that leads to paralysis and death (Rowland and Shneider, ). Although ALS is not considered a prototypical aging disease, the median age of onset occurs at 70.8 years of age in European populations (Marin et al., ) and has an incidence curve similar to that of PD (Logroscino et al., ). In addition, recent genetic advances have revealed that pathways involved in the development of the disease are also modulated during the aging process. These include autophagy, inflammation and cellular maintenance (Niccoli et al., ). The precise mechanisms that target motor neurons in ALS remain elusive. However, it has been proposed that during aging in rodents a loss of astrocytic support to motor neurons may result in loss of both motor neurons and their function (Das and Svendsen, ; Das et al., ). This loss of support is further accelerated in a rodent model of ALS (Das and Svendsen, ). Furthermore, astrocytes generated from post-mortem brain tissue from both familial and sporadic ALS are toxic to motor neurons and a knockdown of the superoxide dismutase 1 gene in astrocytes, a known genetic risk factor in ALS, attenuates this toxicity (Haidet-Phillips et al., ). In addition, treatment of motor neuron/astrocyte co-cultures with an inhibitor for necrotic cell death prevented loss of motor neurons (Re et al., ). It appears that the NF-κB signaling pathway in astrocytes may be participating in the astrocyte-motor neuron death (Haidet-Phillips et al., ).
### Summary
The general inflammatory state of astrocytes gives us a good prediction of different processes that may become dysfunctional during disease. By exploring how this inflammatory state is manipulated or altered in various age-related neurodegenerative disorders, we can gain better insight into potential disease mechanisms and pathology.
## Conclusion
In conclusion, a growing number of studies have shown that astrocytes play a more central role than previously appreciated in aging and age-related diseases. These glia become more pro-inflammatory and contribute to the general low-grade inflammation that is characteristic of the aging brain. Further, because of their active interactions and secretory products, aged astrocytes could negatively afflict other CNS cells. Further work is needed to unravel how these glia interact with other CNS cells at the cellular and molecular level since this will have implications when developing therapeutics for age-related diseases.
## Author Contributions
SO and AP conceived and wrote the manuscript.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Mild cognitive impairment (MCI) is a clinical state with a high risk of conversion to Alzheimer's Disease (AD). Since there is no effective treatment for AD, it is extremely important to diagnose MCI as early as possible, as this makes it possible to delay its progression toward AD. However, it's challenging to identify early MCI (EMCI) because there are only mild changes in the brain structures of patients compared with a normal control (NC). To extract remarkable features for these mild changes, in this paper, a multi-modality diagnosis approach based on deep learning is presented. Firstly, we propose to use structure MRI and diffusion tensor imaging (DTI) images as the multi-modality data to identify EMCI. Then, a convolutional neural network based on transfer learning technique is developed to extract features of the multi-modality data, where an L1-norm is introduced to reduce the feature dimensionality and retrieve essential features for the identification. At last, the classifier produces 94.2% accuracy for EMCI vs. NC on an ADNI dataset. Experimental results show that multi-modality data can provide more useful information to distinguish EMCI from NC compared with single modality data, and the proposed method can improve classification performance, which is beneficial to early intervention of AD. In addition, it is found that DTI image can act as an important biomarker for EMCI from the point of view of a clinical diagnosis.
## 1. Introduction
Alzheimer's disease (AD) is an irreversible degenerative brain disease and the most common cause of dementia (Wilson et al., ), which often happens to people aged over 65 years. In Alzheimer's disease, neurons in parts of the brain involved in cognitive function are eventually damaged or destroyed. In 2018, the estimated number of AD patients was 5.7 million in America, and this number will continue to grow to 13.8 million by 2050 (Alzheimer's Association, ). There is no effective prevention or treatment against AD at present. Early Mild Cognitive Impairment (EMCI) is the stage between age-related cognitive decline and AD or other types of dementia (Schneider, ). Early intervention against EMCI will possibly delay the progression of EMCI toward AD.
People with MCI exhibit mild but measurable changes in thinking abilities. A systematic review found that an average of 32% individuals with MCI developed into AD within 5 years, which shows that MCI patients have a high risk of conversion to AD (Alzheimer's Association, ). Early diagnosis of MCI is therefore of great importance to early intervention in the preclinical state of AD (Lim, ; Wen and Li, ), and this has received extensive attention from researchers in the recent decades (Jiang et al., ). However, identifying EMCI is a challenging clinical problem due to the mild changes between EMCI and NC.
Many studies utilize machine learning methods to complete computer-aided diagnosis for EMCI in which various neural imaging techniques, such as structure Magnetic Resonance Imaging (sMRI), functional MRI (fMRI), and diffusion MRI (dMRI), work as data sources. Based on dMRI [specifically, Diffusion Weighted Imaging (DWI)], Prasad et al. ( ) calculated a 68 × 68 connectivity matrix and a set of network measures from 68 cortical areas as the input of support vector machine (SVM), and a classification accuracy of 59.2% was achieved for EMCI vs. NC. Using sMRI data, Raeper et al. ( ) proposed a cooperative correlational and discriminative ensemble learning framework. Each individual brain was represented by a set of shallow convolutional brain multiplexes (SCBMs) used to train an ensemble of canonical correlation analysis (CCA)-SVM and linear discriminative analysis (LDA)-based classifiers, achieving an accuracy of 80.95%. Kang and Suk ( ) separated the fMRI signals into a true source signal and a noise component by means of a stochastic variational Bayesian inference and then calculated source correlations of the inferred source signals as input features of a linear SVM, achieving an accuracy of 74.45%. Chen et al. ( ) and Jiao et al. ( ) also utilized fMRI data, they constructed low-order and high-order functional networks to train SVM and achieved an accuracy of 88.14 and 91.13%, respectively. It worth noting that Jiao et al. performed least the absolute shrinkage and selection operator (LASSO) feature selection algorithm.
The above methods based on tradition machine learning commonly need complex feature engineering to extract region of interest (ROI)-based or voxel-wise features used for a classification task. The validity of the extracted features thus largely depends on image preprocessing steps, such as segmentation and registration, as well as prior hypotheses. Recently, researchers have shown an increasing interest in the convolutional neural network (CNN) in medical image classification field (Islam and Zhang, ; Yue et al., ). The CNN can alleviate the above problems by automatically extracting the most discriminating disease-related features from voxel values of complex high-dimensional image data in end-to-end modes, which is conducive to avoiding errors caused by feature engineering and retains the subtle differences between EMCI and NC.
Some studies have tried to use the CNN to extract latent features of neuroimaging data for EMCI classification. Kam et al. ( ) proposed a novel 3DCNN framework to extract deeply embedded features from both static and dynamic brain functional networks of fMRI data for EMCI classification, and they reported an accuracy of 76.07%. However, time-consuming, multi-channel, and multi-model training did not result in higher classification accuracy. Yue et al. ( ) utilized a 2DCNN to acquire the most useful features of the gray matter of sMRI. This deep learning method achieved high accuracy for EMCI vs. AD and LMCI vs. EMCI, but the classification task excluded EMCI vs. NC. Puranik et al. ( ) employed a 2DCNN model with transfer learning technique to classify AD, EMCI, and NC and obtained an accuracy of 98.41%. However, the inputs of the CNN are the 2D slices of fMRI images, which means that the classification task is not based on subject-level, deviating clinical needs. Apart from one paper, the above methods did not process the problem of binary classification of EMCI vs. NC (Kam et al., ).
On account of the similar brain structure and brain functions between EMCI and NC, it is very challenging to distinguish EMCI from NC. The diagnostic accuracy of EMCI in all the above studies is much lower than that of AD or MCI due to the subtle differences between EMCI and NC. On the other hand, with single modality data it is difficult to extract enough features to classify EMCI from NC (Qi et al., ; Cabrera-León et al., ). For example, sMRI data cannot catch the mild changes of brain structure. Since multi-modality data can provide more useful auxiliary information, it seems to be more promising to extract the most discriminative features to perform the classification task.
Fortunately, multi-modality-based diagnosis has attracted extensive attention and become a hot area of research within medical image analysis. Some studies have successfully applied multi-modality neuroimaging analysis to the diagnosis of AD, as can be seen in Baiying et al. ( ), Khvostikov et al. ( ), and Cheng and Liu ( ). For EMCI classification using multi-modality data, Forouzannezhad et al. ( ) made a preliminary attempt. They combined the features extracted from cortical region and subcortical region of sMRI and PET images to train a deep neural network (DNN) and achieved an accuracy of 84%. The authors also trained a SVM classifier utilizing the same data in paper (Forouzannezhad et al., ) and reported an accuracy of 81.1% (Forouzannezhad et al., ).
In this paper, we further develop a multi-modality diagnosis method for EMCI. Specifically, we fused sMRI and DTI data with the multi-modality fusion strategy and then combined the CNN model and SVM classifier to identify EMCI. To the best of our knowledge, we are the first to utilize DTI data to train CNN for EMCI diagnosis. DTI data has been proven to be a useful diagnostic marker for distinguishing EMCI from NC, especially its measures, namely fractional anisotropy (FA) and mean diffusivity (MD). Moreover, DTI data can reflect brain microstructure changes by quantifying the integrity of white matter (Nowrangi et al., ; Marizzoni et al., ; Brueggen et al., ; Gyula, ), which makes it promising for identifying the subtle differences of EMCI compared with NC.
## 2. Materials and Methods
### 2.1. Data Acquisition
sMRI and DTI data used in this study were obtained from Alzheimer's Disease Neuroimaging Initiative (ADNI) project, which was launched in 2003 as a public-private partnership. The goals of the ADNI study are to identify biomarkers for clinical use and detect AD at the early stage (EMCI, MCI, and LMCI) using biomarkers. Multiple biomarkers, including sMRI, fMRI, DTI, and related neuropsychological assessments, are combined in an effort to detect the progression of EMCI and early AD.
In this study, we selected 120 subjects, including 70 EMCI subjects and 50 age-matched Normal Controls (NCs) from the ADNIGO and ADNI2 database. For each subject, there is a T1-weighted sMRI image, a FA-DTI image, and an MD-DTI image in an NIfTI file format. The selected subjects coming from the basebline/screening visit have passed strict inclusion criteria. shows the detailed demographic information of subjects used in this study, where Mini Mental State Examination (MMSE) is a mental test quantifying the cognitive function. As an auxiliary diagnostic index, the lower MMSE score indicates poor cognitive ability.
Corresponding statistical information of subjects.
All raw data were acquired by 3T GE medical system scanners at multiple sites with a rigorous quality control to reduce site effect. The raw 3D T1-weighted sMRI scans were collected with the following imaging parameters: 256 × 256 × 196 voxels per volume, 1.0 × 1.0 × 1.2 mm every voxel size, inversion time = 400 ms, and flip angle = 11°. The raw DTI data of each subject is formed of 41 diffusion images with b = 1,000 s / mm and 5 T2-weighted b0 images. Each DTI slice was acquired with the following imaging parameters: 256 × 256 pixels per slice, 1.37 × 1.37 every pixel size, slice thickness = 2.70 mm , and flip angle=90°. More information about the scan parameters of images can be searched on the ADNI website ( ).
### 2.2. Preprocessing
All downloaded data have been preprocessed through a series of standard preprocessing procedures. Using the FMRIB software library (FSL) and FreeSurfer software, the T1-weighted sMRI images were preprocessed with skull-stripping, intensity normalization, and registration with a standard template Colin27 having the same coordinate system as MNI152. For DTI data, FSL was used to perform the preprocessing steps, including skull-stripping, eddy current correction, head motion correction, diffusion tensors estimation generating FA and MD maps, and registration with Colin27. Consequently, for each subject, we acquired three preprocessed and aligned 3D images in NIfTI file format from ADNI, which are sMRI, FA-DTI, and MD-DTI respectively. Three images have the same resolution of 110 × 110 × 110 voxels.
In order to reduce the interference of useless information and improve the computational efficiency, we investigated 2D slices of the above three kinds of images. For each subject, 32 slices with indexes 37 to 68 were selected from each kind of image. Then, the multi-modality fusion strategy were adopted: the slices with same index were merged into an RGB slice and saved as a JPEG image format, as shown in . Consequently, each subject has 32 RGB slices which can be seen in . It is worth noting that some researches have reported that the temporal lobe may make an essential contribution during the early stage of MCI (Bi et al., ; Cui et al., ; Zhang et al., ), and the 32 slices, including the whole temporal lobe and other memory related regions, such as the hippocampus and callosum, were thus selected.
The FA, MD, and sMRI slice with the same index on the left are merged into an RGB slice on the right.
All RGB slices were used to build the following two multi-modality dataset. A slice-level dataset consisting of 2,240 (32 × 70) EMCI RGB slices and 1,600 (32 × 50) NC RGB slices was built for training CNN model. In addition, the slice-level dataset was split into train set and validation set with the ratio of 8:2. Another subject-level dataset was built for SVM classification, which consists of 120 folders, each of which contained all RGB slices of one subject.
### 2.3. Proposed System Framework
The framework of the proposed approach is shown in , where the VGG16 (Simonyan and Zisserman, ) network structure is exploited to perform feature extraction and an SVM classifier is used for feature classification. The dataset is split into five-folds, where four-folds (96 subjects) are used for training and one-fold (24 subjects) for testing. Firstly, the VGG16 network is trained with a slice-level training set using a transfer learning technique, and the optimal VGG16 model is then saved in terms of the lowest loss value. Secondly, all the slice features of each subject in the subject-level training set are extracted by the optimal VGG16 model. Thirdly, feature selection is performed by a LASSO algorithm to reduce the feature dimension and redundant information of training set, and then the outputs are used to train the SVM classifier to distinguish EMCI from NC. Finally, the features of the subject-level test set are extracted by optimal VGG16 model trained by training set, and the extracted features are further selected by LAASO model fitted by training set. After feature selection, the predicted labels are acquired through SVM classification.
Total framework of proposed method, where the VGG16 model consists of five convolution blocks and a fully connected layer. For one slice, the output of VGG16 model is a matrix of 1 × 256. For each subject with 32 slices, 32 feature matrixes of 1 × 256 are acquired from VGG16, and they are then concatenated as a feature representation with the dimension of 1 × 8192. A total of 120 feature representations of 120 subjects are finally obtained, where 96 subjects used for training VGG16 network make up training set for feature selection and classification and 24 subjects make up test set for prediction.
#### 2.3.1. Convolutional Neural Network and Transfer Learning
The VGG16 network used for feature extraction in this study is one of the classical CNN. With the rapid development of deep learning, the CNN has achieved great success in large-scale image recognition in recent years. The CNN is mainly composed of a convolution layer, pooling layer, and fully connected layer (Yue et al., ). The convolution calculation is performed in the convolution layer with some convolution kernels, which can learn various features of the input images. The pooling layer can reduce the feature dimension and reduce vast network parameters and training time. The fully connected layer converges all learned features to produce the classification score of input data. In this study, due to the small dataset, transfer learning technique was used to train the VGG16 network to realize adaptation from source domain to target domain, which is to utilize the pre-trained weights to initialize the network whose structure is the same as pre-trained model trained with a larger dataset, and this has been applied to medical image classification in many studies (Pan and Qiang, ; Tajbakhsh et al., ; Hon and Khan, ; Karri et al., ; Kermany et al., ).
In this study, the pre-trained weights trained by nature image dataset Imagenet with 1,000 categories were transferred into VGG16 network. Due to the difference of categories number and image attributes, a fine-tuning strategy was taken. Firstly, the pre-trained weights of the first four convolution blocks were frozen, the fully connected layer was replaced, and then the pre-trained weights in the fifth convolution block shown in the orange part of and the initial weights of the new fully connected layer were continually updated until the model converged.
#### 2.3.2. LASSO
The feature matrix extracted from the VGG16 model possibly contains a lot of irrelevant or redundant features for EMCI diagnosis. To remove these features and reduce feature dimension, the feature selection algorithm of least absolute shrinkage and selection operator (LASSO) was adopted to select a small set of crucial features related to EMCI disease. LASSO is performed through minimizing the penalized objective function with L1 regularization which tends to give zero weight to irrelevant features so that the important features can be saved (Jiao et al., ). The objective function of LASSO is defined as follows:
where is a feature matrix. N is the number of subjects and d is the number of features. is a set of corresponding class labels of subjects. θ represents a regression coefficient and λ is the regularization parameter to balance the complexity of the model.
#### 2.3.3. SVM
Support vector machine is most suitable classifier to deal with high-dimensional small dataset, which seeks a maximum margin hyper-plane to separate EMCI from NC. Given a training set with input data and corresponding binary class labels y ∈{−1, +1}, the output of primal SVM is presented as follows:
Here, φ( x ) is a non-linear function, mapping the input space to higher dimensional feature space, which makes the input data linearly separable in the hyperplane. b is a bias term. The optimization objective function is defined as follows (Suykens, ):
subject to:
ξ is a slack variable, indicating the tolerance of misclassification. w is the weight applied for input data x . c is a tuning parameter which must be a positive real constant.
### 2.4. Implementation
VGG16 network is trained based on Keras with a single GPU (i.e., NVIDIA GTX TITAN 12GB). The network is optimized by Root Mean Square Propagation (RMSProp) with a learning rate of 10 . The weights update is performed in mini-batches of 32 samples per batch and stops after 50 epochs.
## 3. Results
### 3.1. Experiment Setup
In this study, several experiments are designed to validate the effectiveness of the proposed method in this paper. Specifically, we want to know whether multi-modality diagnosis can effectively improve classification performance than single modality. Firstly, slice-level dataset and subject-level datasets are built for FA, MD, and sMRI data. Except for modality fusion, these three single-modality datasets are built in the same way as a multi-modality dataset, as described in section 2. Then, the above three datasets and multi-modality dataset are used for EMCI diagnosis using the proposed method. Finally, the performance evaluation results of multi-modality are compared with that of single modality.
### 3.2. Feature Extraction
Feature extracted from VGG16 model is abstract in deepest layer, but we can visualize the output of the lower layers. shows the output of the first Maxpooling layer of VGG16, where different filters (or convolution kernels) learn different features from various aspects. Some filters learn the brain shape, and others learn the interior structure of the brain. In deeper layers, such as the seventh convolution (CONV), the features become more and more localized, as shown in . It illustrates that although brain image is very different from nature image, the first few frozen layers can extract many generic features, such as side, angle, color, etc. In addition, the deeper fine-tuned layer can extract high-level target-specific features used for distinguishing different categories of images.
The feature activation maps of an RGB slice in the VGG16 model were visualized. (A) A total of 64 feature maps of 55 × 55 pixels from the first Maxpooling layer. (B) A total of 256 feature maps of 27 × 27 pixels from the seventh CONV layer.
### 3.3. Feature Selection
Feature selection can reduce the effect of redundant features through adjusting the regularization parameter α. For one of the experiments in five-fold cross validation, the variation curve of classification accuracy changing with α is shown in . As we can see, the value of α will affect the accuracy to some extent because the complexity of the model and the quantity of selected features rely on the value of α.
The variation curve of classification accuracy changing with α. Some values of α will arouse the problem of overfitting in a classification task, and such values were discarded in experiment.
As shown in , the classification accuracy of multi-modality data improves significantly comparing with single modality. In order to further validate the effectiveness of multi-modality data, the selected features are visualized in form of cluster figures, as shown in . Single modality data can roughly distinguish EMCI from NC, especially sMRI data. The divisibility of multi-modality data is much better than that of any single modality data. It is worth noting that as shown in , the features of several EMCI subjects are similar to those of NC. The possible reason is that the differences between these EMCI subjects and NC subjects are especially subtle.
Manifold visualization of different neuroimaging feature, by t-SNE projection (Laurens and Hinton, ). (A) FA features; (B) MD features; (C) sMRI features; and (D) the features fusing FA, MD, and sMRI.
### 3.4. Feature Classification and Performance Evaluation
Before feature classification, data distribution of the selected features was linearly transformed to a normal distribution with unit standard deviation and zero mean. Then, the normalized features were fed into an optimal SVM classifier selected through exhaustive search. In order to alleviate the problem of data imbalance, class weights were imposed during SVM classification. The test accuracy (ACC), sensitivity (SEN), specificity (SPE), and area under the receiver operating characteristic curve (AUC) were finally acquired with five-fold cross-validation. We took the mean of each metric to evaluate classification performance quantitatively, and the classification results comparison between single modality and multi-modality method are shown in .
Classification performance comparison between the single-modality and multi-modality method for distinguishing EMCI from NC.
As shown in , the classification performance of multi-modality method is superior to that of single modality method. An average classification accuracy of 94.2%, a sensitivity of 97.3%, a specificity of 92.9%, and an AUC of 95.3% have been achieved to distinguish EMCI from NC with the multi-modality method, while the classification accuracy of FA, MD, and MRI data were 67.5, 71.7, and 73.3%, respectively. It is obvious that the method of combining the features from FA, MD, and sMRI data can enhance the classification accuracy significantly. These results indicate that it is difficult for single modality to represent all the attributes of EMCI because different modalities reflect pathological changes in different forms. For example, sMRI modality can show the change of macrostructure of brain while DTI modality can reflect the abnormity of microstructure of brain through random motion of water molecules affected by sensitive gradient field. The results also illustrate that DTI data is useful for distinguishing EMCI and NC to some extent. What's more, sMRI combined with DTI data can capture more disparate differences between EMCI and NC, which can improve classification performance greatly. This suggests that it is hihgly necessary to fuse data from different modalities for neuroimaging analysis.
### 3.5. Comparison With Other Methods
To our best knowledge, we are the first to fuse sMRI modality and DTI modality for distinguishing EMCI from NC. provides an assessment of the proposed approach in comparison to related studies using the same metrics, where it can be clearly seen that the proposed method yielded the best results in all metrics. It is worth noting that the neuroimaging data of these competitive studies all comes from the ADNI website.
Classification performance comparison with pervious researches for distinguishing EMCI from NC.
On the one hand, as described in section 1, paper (Forouzannezhad et al., ; Raeper et al., ) both utilized a traditional machine learning method to classify EMCI and NC. We further introduced the CNN to perform feature extraction for multi-modality fusion data. As shown in , the proposed method using the sMRI modality alone outperformed paper (Forouzannezhad et al., ), achieving an accuracy of 73.3%, 12.2% higher than paper (Forouzannezhad et al., ), and the classification accuracy of using diffusion MRI alone shown in is much higher than that of paper (Prasad et al., ), which shows the effectiveness of our method. It is worth noting that the preprocessing procedures of diffusion MRI in paper (Prasad et al., ) are the same as ours completely, which are performed by the group led by Pual M. Thompson. The above results illuminates that the CNN used in this study plays a significant role, which can efficiently extract different levels of features and reduce the error resulting from incomplete prior hypothesis. For example, despite the different pathology between EMCI and AD, many studies still assume that EMCI lesions are based on AD. Fortunately, CNN can ignore this difference by using full images instead of ROI as input and capture more features in a larger range. In other words, the CNN can extract the most discriminative features no matter what the pathology is. On the other hand, Raeper et al. acquired higher accuracy using sMRI data in paper (Raeper et al., ) than the proposed method; a possible reason is that transfer learning just can acquire a better performance when the target data are RGB images. We used the same three sMRI grayscale images to connect into a pseudo RGB image as the input of the CNN, and the relatively lower accuracy illustrates that the CNN utilizing transfer learning needs richer data to fit, such as the proposed multi-modality data.
Compared with the only method using multi-modality diagnosis, the proposed method using sMRI and DTI modality achieved an accuracy of 94.2%, which is 10.2% higher than that of paper (Forouzannezhad et al., ), which used sMRI and PET modality. The right choice of multi-modality neuroimaging data and the introduction of CNN are the main reasons for acquiring high accuracy in the proposed multi-modality diagnosis. As we can see in , in the proposed method, the classification accuracy of using sMRI modality alone is 20.9% lower than that of fusing sMRI and DTI modality. It illustrates that DTI data can offer some extra complementary information for sMRI modality so as to improve classification performance significantly.
## 4. Discussion
Alzheimer's disease is an irreversible neurodegenerative disease, and identifying EMCI accurately contributes to early intervention of EMCI due to its high conversion rate to AD. However, studies of EMCI diagnosis suffer from some serious limitations, such as complex feature engineering and low accuracy. To solve these problems, we proposed a CNN-based multi-modality diagnosis method to distinguish EMCI from NC efficiently. Using the special multi-modality fusion method, we achieved a high classification accuracy of 94.2%, which is superior to other state-of-the-art methods. Three factors are identified as being potentially important. Firstly, DTI data is an effective supplement to sMRI, and it acts as an significant biomarker of EMCI since DTI data describes the changes of brain microstructure. Secondly, the transfer learning technique greatly improves the learning ability of small medical datasets and helps excavate more high-level target-domain features. Thirdly, it is very necessary to perform L1-norm feature selection by LASSO algorithm so as to remove redundant features in the high-dimensional features generating from multi-modality method. The contributions of different experimental steps were shown in . As we can see, LASSO and the transfer learning technique are both key steps and make almost an equal contribution to the proposed method.
Classification performance comparison with different experiment setups, where without LASSO and without fine-tuning , representing transfer learning technique and LASSO algorithm, respectively, were not performed in the experiment.
Complete experiment is the complete process of the proposed framework .
In addition, compared with other multi-modality fusion methods (Cheng and Liu, ; Khvostikov et al., ), the training time and computing resources involved are reduced because the number of models is reduced using the proposed multi-modality fusion strategy. Other studies generally train multiple models to extract the features of different modalities and then fuse these extracted features. In the proposed method, only one model is trained for extracting the features of multi-modalities because the multi-modality data are input into the network through multiple channels.
In the past few years, several pioneering studies only focused on sMRI data to detect EMCI (Raeper et al., ; Yue et al., ; Taheri and Naima, ; Wee et al., ), and they have utilized morphological features and demographic factors to perform feature selection after the sMRI image is divided into 45 subcortical regions or 68 cortical regions. Interestingly, Wee et al. ( ) constructed cortical thickness graphs using sMRI data and input them into the popular graph CNN. sMRI is one of the common neuroimaging tool for disease diagnosis; however, there are many studies illustrating that multi-modality data are more effective than single-modality data for EMCI classification (Amoroso et al., ; Cheng et al., ; Forouzannezhad et al., ; Hao et al., ; Lei et al., ), and these studies have shown that different neuroimaging data may provide complementary information that is beneficial to diagnose EMCI. In addition, more and more researchers have turned their attention away from structural changes of the brain to functional change. They use fMRI data to construct a brain functional network of the EMCI and NC groups, and it has been found that the temporal lobe is the discriminating disease-related region. It is worth noting that in this study, we intentionally selected the 2D slices including the temporal lobe.
In this study, we applied transfer learning technique to train the CNN, which can alleviate the problems caused by a small dataset. Due to the small sample size of medical images, deep-learning-based diagnosis methods suffer many limitations. In paper (Puranik et al., ; Taheri and Naima, ), the classification accuracy is achieved by using 2D slices of neuroimaging data as input of the CNN, which is based on slice-level classification. In order to acquire the subject-level classification accuracy, we integrate all slice features of each subject.
In summary, multi-modality fusion diagnosis using the CNN is an effective medical image analysis method, and good classification performance can be achieved by selecting the most suitable neuroimaging modality according to the pathological characteristics of disease. Although it has been proven that DTI data is an effective imaging-biomarker for MCI diagnosis (Nowrangi et al., ; Marizzoni et al., ; Brueggen et al., ; Gyula, ), there are few studies to report. The high classification accuracy obtained in this paper again proves that DTI image can act as a remarkable biomarker for EMCI from the point of view of clinical diagnosis. The present study also posed several limitations. First, a larger sample size and other stages of cognitive impairment should be further considered to verify the stability and generalization ability of the proposed method. Secondly, the transfer learning technique used in this study relies on nature images. Although the fine-tuning method is adopted, a source domain model trained by brain images may fit the target domain better. These limitations will guide us to further enhance the robustness of the proposed method in future works.
## Data Availability Statement
Publicly available datasets were analyzed in this study. This data can be found here: .
## Author Contributions
All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Background: Although exercise is known to have a neuroprotective effect in aging, the mediators underlying the exercise-cognition association remain poorly understood. In this paper we aimed to study the molecular, brain, and behavioral changes related to physical activity and their potential role as mediators.
Methods: We obtained demographic, physical activity outcomes [sportive physical activity and cardiorespiratory fitness (CRF)], plasma biomarkers (TNF-α, ICAM-1, HGF, SDF1-α, and BDNF), structural-MRI (brain volume areas), psychological and sleep health (mood, depressive and distress symptoms, and sleep quality), and multi-domain cognitive data from 115 adults aged 50–70 years. We conducted linear regression models and mediation analyses stratifying results by sex in a final sample of 104 individuals [65 women (age = 56.75 ± 4.96) and 39 men (age = 58.59 ± 5.86)].
Results: Women engaging in greater amounts of exercising showed lower TNF-α levels and greater dorsolateral prefrontal cortex and temporal lobe volumes. Men engaging in greater amounts of exercise showed greater temporal lobe volumes. CRF levels were not related to any of the analyzed outcomes in women but in men higher CRF was associated with lower TNF-α, HGF and ventricle volumes, greater volume of temporal and parietal lobes and fewer depressive symptoms and better mood. In men, reduced TNF-α and HGF levels mediated brain and cognitive CRF-related benefits.
Conclusion: Our results show that exercise is a promising approach for influencing inflammation and brain volume and also contributes to ongoing discussions about the physiological mediators for the association between CRF and cognition in men.
## Introduction
“Let's move your body; let's rock your cells, brain and self!” Epidemiological studies have indicated that an active lifestyle including exercise positively impacts several major hallmarks of aging and has neuroprotective benefits (Garatachea et al., ). Exercise, a subtype of physical activity (PA) applied in a regular manner in order to improve physical fitness, is related to reduced risk of dementia and better cognitive health in clinical and non-clinical populations, with greatest effects for measures of executive function (Barha et al., ; Northey et al., ). Those benefits have been described when using a measure of self-reported exercise habits and when assessing physiological correlates of habitual PA such as cardiorespiratory fitness (CRF). CRF is the ability of the cardiovascular system to supply oxygen to the organism during sustained PA. Previous literature has identified potential mechanisms of the exercise-cognition association at multiple levels: molecular, brain, and behavioral. In this paper we will discuss them as Level 1, 2, and 3 mechanisms, respectively, based on Stillman et al. ( ).
At the molecular level, Level 1, aging is related to altered levels of inflammatory, oxidative stress, metabolic and neuronal, and cell growth markers (Garatachea et al., ). Exercise reduces levels of systemic inflammation modulating markers such as C-reactive protein (CRP), Interleukin 6 (IL-6), Interleukin 1 (IL-1), and Tumor Necrosis Factor alpha (TNF-α) (Sallam and Laher, ). However, Woods et al. ( ) indicated that the evidence for a relationship between CRF and TNF-α levels in humans was still insufficient for making definitive conclusions. Studies with mice have demonstrated that exercise also produces an increased shear stress on endothelial cells that modulates other markers of the inflammatory process such as Intercellular Adhesion Molecule 1(ICAM-1), Nuclear factor-κB (NF-κB), Mitogen-activated protein kinase (MAPK), and Cyclooxygenases (COX-2) (Sallam and Laher, ). Shear stress and mechanical load related to exercise induce changes in other anabolic and metabolic growth factors that affect muscles and bones such as Hepatocyte growth factor (HGF), Vascular endothelial grow factor (VEGF), and Insulin-like growth factor-1 (IGF1). HGF, which is commonly related to obesity and insulin resistance and is capable of modulating the inflammatory response, promotes angiogenesis, and neuroprotection in the brain (Kiliaan et al., ). Exercise-induced VEGF is also related to angiogenesis and improved circulation in humans. It upregulates the chemokine Stromal cell-derived factor 1 (SDF1-α), also known CXCL12 (Stimpson et al., ), which promotes both endothelial progenitor cells and the endothelial nitric oxide synthase enzyme in mice (Gertz et al., ). Exercise also influences brain derived neurotrophic factor (BDNF), a neuronal growth factor involved in neurogenesis. A systematic review (Huang et al., ) reported a negative association between long-term regular PA and CRF with peripheral BDNF in observational studies. Those results may reflect a more efficient uptake mechanism of circulating BDNF into the brain in active subjects (Currie et al., ).
At a more macroscopic level, Level 2, reviews and systematic reviews reported beneficial effects of exercise on brain volume in healthy older adults, specifically for areas more related to normal brain aging (Erickson et al., ; Sexton et al., ; Stillman et al., ). Cross-sectional studies showed that higher amounts of PA were associated with greater total brain volume (Benedict et al., ; Spartano et al., ) and gray matter (GM) volume in the frontal lobe (Erickson et al., ; Flöel et al., ; Bugg and Head, ; Eyme et al., ), hippocampus (Erickson et al., ; Yamamoto et al., ; Raichlen et al., ), cingulate cortex (Flöel et al., ), precuneus (Benedict et al., ; Eyme et al., ), and nucleus accumbens (Yamamoto et al., ). CRF levels have also been positively correlated with overall gray matter (Raichlen et al., ), multiple areas of the frontal lobe (Gordon et al., ; Weinstein et al., ; Wittfeld et al., ), medial-temporal lobe (Gordon et al., ; Wittfeld et al., ), hippocampus (Erickson et al., ; Szabo et al., ; Wittfeld et al., ), and cingulate cortex (Wittfeld et al., ) volume in healthy older adults. Findings related to white matter (WM) indicated that engaging more frequently in PA could increase WM global volume (Gow et al., ; Benedict et al., ; Arnardottir et al., ). Only a few studies have examined the relationship between CRF and WM volume and did not find significant results (Burns et al., ; Gordon et al., ; Honea et al., ).
At a behavioral level, Level 3, aging is commonly associated with increased sedentarism (Copeland et al., ), changes in sleeping patterns (Li et al., ), and more psychological problems such as depression or anxiety symptoms (World Health Organization (WHO), ). For older adults, exercising is related with better physical and mental health (Bertheussen et al., ), quality of life (Fox et al., ), and well-being (Lee and Hung, ; Black et al., ). Higher levels of PA (Strawbridge et al., ) and CRF (Sui et al., ; Willis et al., ) are also protective for prevalent and incident depression. PA is also associated with better sleep quality (Kline et al., ; Tan et al., ) and efficiency (Kline et al., ; Wilckens et al., ) and total sleep time (Murray et al., ). However, the relationship between sleep patterns and CRF in healthy older adults is not well-established.
Conceiving mechanisms at multiple levels might be helpful to better understand the relationship between exercise and cognition. As the evidence suggests, exercise might initiate a molecular cascade that promotes macroscopic changes in the brain and/or behaviors that in turn enhance cognition. For example, PA has been related to reduced inflammatory profile and greater brain volume (Stillman et al., ). However, there is a poor understanding of the mediating role of these variables in the PA-cognition relationship and the multiple pathways by which microscopic and macroscopic biomarkers might influence each other to promote cognition are in current research. Identifying these pathways by which the benefits of exercise are realized is a challenge not only due to the multi-level mechanisms but also because of the role of factors that may moderate the association such as sex (Barha et al., ). Previous evidence suggests that both women and men might positively benefit cognition from exercise, although the CRF-cognition relationship was only significant in men (Castells-Sánchez et al., ). Moreover, there are sex differences in the mediators of this relationship described in the ongoing research. For example, greater PA was related to reduced TNF-α only in men (Elosua et al., ), while daily walking has been associated with greater hippocampal (Varma et al., ) and dorsolateral prefrontal volume (Barha et al., ) and larger surfaces of the subiculum (Varma et al., ) in women, but not in men. Therefore, addressing the role of these mechanisms stratified by sex might be interesting to better understand exercise as a personalized approach to enhance cognitive health.
To our knowledge, previous research focused on single levels of analyses when examining and describing possible mechanisms of exercise on cognition (Stillman et al., ). In this paper, we first aim to study the relationship between exercise and CRF with key markers at molecular, brain volume, and behavioral levels of analysis and further stratifying the results by sex because of established sex-related differences in many of these measures. Secondly, we aim to study their potential mediating role of each of these markers in the exercise-cognition relationship and perform exploratory analyses to address the potential pathways by which these markers might influence each other.
## Materials and Methods
This is a cross-sectional study based on Projecte Moviment (Castells-Sánchez et al., ) and is drawn from our previously published results (Castells-Sánchez et al., ). The study was carried out by the University of Barcelona in collaboration with Institut Universitari d'Investigació en Atenció Primària Jordi Gol and Hospital Germans Trias i Pujol. It was approved by the responsible ethics committees following the Declaration of Helsinki.
### Participants
One hundred and fifteen community dwelling healthy late-middle-aged adults were recruited from the Barcelona metropolitan area using multiple strategies (lists of volunteers from previous studies, advertisements in local media, presentations in local community organizations, etc.). They were 50–70 years old, were not cognitively impaired [Mini Mental State Examination, MMSE ≥ 24 (Blesa et al., ), and Montreal Cognitive Assessment 5-min, MoCA 5-min ≥ 6 (Wong et al., )], had competency in Catalan or Spanish and had adequate sensory and motor skills. Participants were excluded from the study if they had a neurological diagnosis, psychiatric disease or Geriatric Depression Scale score >9 (Martínez et al., ), a history of drug abuse and alcoholism, consumed psychopharmacological drugs, history of chemotherapy, and had any contraindication to magnetic resonance imaging (MRI). As previously published (Castells-Sánchez et al., ), the sample consisted of Projecte Moviment baseline low-active participants (<2 h/week over the last 6 months) and 20 additional participants with a higher physical activity profile (≥5 h/week moderate PA or 2.5 h/week intense PA) in order to enlarge the interval of the PA and CRF levels and increase the sample size for a cross-sectional analysis. All participants were recruited, selected, and assessed following the same protocol during the same period of time. They were screened by phone and an on-site interview and signed an informed consent prior to the assessment.
### Assessments and Outcomes
All participants underwent a multimodal assessment organized into three appointments in 2 weeks in the following order: (1) Medical assessment and blood extraction (30 min), (2) Cognitive psychological health and physical activity assessment (2, 5 h), (3) MRI protocol (45 min). In order to control the effects of acute exercise, participants were advised not to exercise 8 h before all appointments. Assessments were conducted in clinical facilities: medical and cognitive assessments were carried out in two Primary Health Care Centers and MRI scans were performed in the Hospital Germans Trias i Pujol.
#### Physical Activity and Cardiorespiratory Fitness
Self-reported PA was evaluated by the Validated Spanish short version of Minnesota Leisure Time Physical Activity Questionnaire (VREM) (Ruiz et al., ). Participants reported frequency and duration of the following activities during the last month: sportive walking (walking in order to exercise), sport/dancing, gardening, climbing stairs, shopping, walking, and cleaning the house. We transformed hours per month expended in each category into units of metabolic equivalent of tasks (METs). This allowed us to estimate energy expenditure in Sportive PA (S-PA) adding up the METs spent in sportive walking and sportive/dancing activities.
We obtained estimated CRF applying the Rockport 1-Mile Walking Test which is a less invasive and valid method commonly used in healthy elderly population. Participants walked one mile on a treadmill (Technogym®, Italy) adjusting their speed in order to be as fast as possible without running. We registered time to complete the mile, heart rate, and average speed during the test once they finished. We estimated maximal aerobic capacity (VO max) using the linear regression developed by Kline et al. ( ).
#### Biomarkers
Blood extraction was performed between 8:00 and 9:00 a.m. following an overnight fast by nurses in the Primary Health Care Centers. All participants were instructed to not exercise 8 h before the blood test. Blood samples were obtained from the antecubital vein and collected in EDTA tubes for plasma analyses. Tubes were immediately transferred to the IGTP-HUGTP Biobank integrated in the Spanish National Biobanks Network of Instituto de Salud Carlos II (PT13/0010/0009) and Tumor Bank Network of Catalonia, and they were processed following standard operating procedures with the appropriate approval of the Ethical and Scientific Committees. Plasma aliquots were stored at −80°C.
As Level 1 outcomes, we obtained peripheral BDNF levels using an ELISA kit (Human Free BDNF Quantikine ELISA Kit; R&D Systems, Minnesota, USA). The rest of the molecular markers were selected according to the Projecte Moviment trial (Castells-Sánchez et al., ). TNF-α, ICAM-1, HGF, and SDF1-α levels were analyzed quantitatively using the corresponding ELISA immunoassay method (Human TNF-α Quantikine HS ELISA, Human ICAM-1/CD54 Allele-specific Quantikine ELISA Kit, Human HGF Quantikine ELISA Kit, Human CXCL12/SDF-1 alpha Quantikine ELISA Kit; R&D Systems, Minnesota, USA).
#### Neuroimaging
Structural MRI data was collected in a 3T Siemens Magnetom Verio Symo MR B17 (Siemens 243 Healthineers, Erlangen, Germany) for all participants (Castells-Sánchez et al., ). We acquired T1-weighted multi-planar reformat sequences (acquisition time: 5:26 min, voxel: 0.9 x 0.9 x 0.9 mm, TR/TE/TI: 1900/2.73/900 ms, flip angle: 9°, slices: 192; thickness: 0.9 mm) and an expert neuroradiologist visually checked them for artifacts or clinical brain conditions. All participants received a clinical report of the MRI.
Brain images were analyzed using MRICloud ( ) (Mori et al., ), an online cloud-computing platform that includes a function to calculate gray and white matter brain volumes (Wu et al., ). It performs a fully automated parcellation of 287 volumes based on multiple atlases and fuses different algorithms (transformation algorithm, Large Deformation Diffeomorphic Metric Mapping (LDDMM), and the atlas label-fusion algorithm) (Christensen et al., ; Oishi et al., ; Wang et al., ) with a local search algorithm (Coupé et al., ). For our sample, we used atlas library version 10A, which includes 30 atlases from cognitively-normal individuals and individuals with cognitive impairment or dementia. For Level 2 outcomes, we selected relevant areas based on previous literature (Erickson et al., , ; Verstynen et al., ) to perform analyses: ventricles, total WM and GM of the frontal lobe, dorsolateral prefrontal cortex, cingulate cortex, parietal lobe, precuneus, temporal lobe, and hippocampus. In accordance with other volumetric studies, we used the ANCOVA method to regress brain volumes on outcomes of interest. All volumes were normalized by head size by including intracranial volume (ICV) as a covariate. We calculated ICV summing volume of the brain tissue (Left Hemisphere + Right Hemisphere + Brainstem + Cerebellum) and CSF (Ventricles + Sulci).
#### Psychological Health and Daily Activity
Psychological health and daily activity of participants were assessed using self-reported questionnaires and the raw scores of each test were used as Level 3 outcomes. We applied GDS-15 (Martínez et al., ) for depressive symptoms, the Modified Version of Visual Analog Mood Scale (VAMS, Stern et al., ) for mood states and the Short Informant Questionnaire in Routine Evaluation-Outcome Measure (CORE-OM, Trujillo et al., ) to assess psychological distress in the domains of subjective well-being, problems/symptoms, general functioning, and risk. We also administered the Pittsburgh Sleep Quality Index (PSQI, Rico and Fernández, ) to evaluate the quality and patterns of sleep and the Short Informant Questionnaire on Cognitive Decline in the Elderly (S-IQCODE, Morales et al., ) as a measure of subjective cognitive performance in daily activities. We used the raw scores of each test as Level 3 outcomes.
#### Cognition and Demographic Data
We assessed cognitive function using an extensive neuropsychological battery which included standard tests selected for their psychometric qualities and high relevance in the area of study. These tests provided measures of multiple cognitive functions grouped following a theoretically-driven approach (Strauss and Spreen, ; Lezak et al., ) into five domains: (1) Executive (Inhibition, Flexibility, Fluency, Working Memory); (2) Visuospatial Function, (3) Language, (4) Memory (Verbal Memory, Visual Memory), (5) Attention-Speed (Attention, Speed). Extended details of the assessment are provided in the . We obtained age, sex, and years of education, height and weight to calculate BMI, diagnoses of hypertension and diabetes as cardiovascular health variables and current medication for cardiovascular risk factors such as dyslipidemia, hypertension, and diabetes.
### Statistical Analyses
We used IBM SPSS Statistics for Windows, Version 24.0, for statistical analyses. Linear regression models were performed to examine the associations between S-PA and CRF with molecular (Level 1), brain volume (Level 2), and psychological (Level 3) outcomes. We stratified analyses by sex and included age and years of education as covariates. We also introduced BMI as a covariate for Level 1 and 2 analyses (Colbert et al., ; Bourassa and Sbarra, ) and ICV when using Level 2 outcomes. Since molecular outcomes in serum could be influenced by current cardiovascular risk factors medications, this variable was included as a dichotomous covariate in complementary analyses for Level 1 outcomes.
We analyzed the potential mediating role of these outcomes in the relationship between S-PA and CRF with cognition in women and men separately (see Model 1 in ). Then, we performed exploratory mediation analyses addressing the potential pathways by which outcomes at Level 1 and 2 might influence markers at Level 2 and 3 (see Model 2, 3, and 4 in ). We applied mediation analyses using the PROCESS Macro ( ). These analyses were computed with bias-corrected bootstrap or Monte Carlo 95% confidence intervals (CIs) based on 5,000 bootstrap samples. Significance of mediation was indicated if the confidence intervals (CIs) in Path AB did not overlap with 0 (Hayes, ).
Mediating models.
## Results
### Participants
We recruited and assessed 115 healthy adults. The final sample consisted of 104 participants (age = 57.44 ± 5.36; 63% female; years of education = 13.35 ± 5.29; MMSE = 28.22 ± 1.45; BMI = 27.52 ± 4.91) from which we could obtain a valid measure of CRF. There were no significant differences in the demographic data between the 65 women and 39 men included in the sample except in years of education that is used as a covariate (see ). Despite sex differences in S-PA and CRF levels, the correlations between S-PA and CRF in women ( r = 0.551, p < 0.001) and men ( r = 0.631, p < 0.001) were significant and comparable (see ). There were no significant differences in age, years of education and MMSE scores between participants of Projecte Moviment and the 20 additional participants (see , ).
Demographic and physical activity variables.
MMSE, Mini-Mental State Examination (Blesa et al., ); S-PA, Sportive Physical Activity; CRF, Cardiorespiratory fitness; METs, Metabolic equivalent .
### Associations Between Physical Activity, CRF, and Mechanisms at Level 1, 2, and 3
#### Level 1: Physical Activity, CRF, and Molecular Biomarkers
Linear regression models examining the relationship between S-PA and CRF with Level 1 biomarkers in women and men separately are in . Engaging in greater amounts of S-PA was significantly associated with reduced levels of TNF-α in women. There were not any significant associations between S-PA and molecular markers in men. In contrast, higher CRF levels were associated with lower TNF-α and HGF levels in men. There were not significant relationship between CRF and molecular markers in women. Extended details for each model are included in . Besides, similar results were obtained for these same regression linear models when accounting for the potential influence of cardiovascular risk factors medications as a covariate (see ).
Linear regression models in women and men: relationship between physical activity variables and molecular biomarkers.
S-PA, Sportive Physical Activity; CRF, Cardiorespiratory Fitness; β, standardized beta .
Covariates: age, years of education, BMI .
S-PA is measured in METs units and CRF in ml/kg*min .
#### Level 2: Physical Activity, CRF, and Brain Volumes
Linear regression models examining the relationship between S-PA and CRF with brain volumes, Level 2 outcomes, in women and men separately are described in . Engaging in greater amounts of S-PA were positively correlated with the volume of the dorsolateral prefrontal cortex in women and with temporal lobe volume in both women and men. There was also a relationship between S-PA and precuneus volume in men. Greater levels of CRF were associated with larger precuneus and temporal lobes and with smaller ventricles in men. There was also a correlation between CRF and frontal and parietal lobe volumes in men. In women, there were no significant relationships between CRF and brain volumes. Extended details for each model are included in .
Linear regression models in women and men: relationship between physical activity variables and brain volumes.
S-PA, Sportive Physical Activity; CRF, Cardiorespiratory Fitness; β, standardized beta .
Covariates: age, years of education, ICV, BMI .
S-PA is measured in METs units and CRF in ml/kg*min .
#### Level 3: Physical Activity, CRF, and Psychological Health and Daily Activity
Linear regression models examining the relationship between S-PA and CRF with behavioral outcomes at Level 3 in women and men separately are in . There were no significant associations between S-PA and any of the scores in women and men. However, when analyzing the subscales, we found that in men, higher levels of S-PA were related to better subjective sleep efficiency (β = 0.41, p = 0.008) and fewer sleep disturbances (β = −0.43, p = 0.008) in the Pittsburgh Sleep Quality Index and higher scores of well-being (β = −0.31, p = 0.065) in the corresponding subscale of the CORE-OM test. In addition, in men, higher CRF was significantly associated with fewer depressive symptoms measured by the GDS. There was also a moderate negative relationship between levels of CRF and VAMS scores in men. Extended details for each model are included in .
Linear regression models in women and men: relationship between physical activity variables and behavior outcomes.
S-PA, Sportive Physical Activity; CRF, Cardiorespiratory Fitness; β, standardized beta; GDS, Geriatric Depression Scale (Martínez et al., ); VAMS, Visual Analog Mood Scale (Stern et al., ); S-IQCODE, Short Informant Questionnaire on Cognitive Decline in the Elderly (Morales et al., ); PSQI, Pittsburgh Sleep Quality Index (Rico and Fernández, ); CORE-OM, Short Informant Questionnaire in Routine Evaluation-Outcome Measure (Trujillo et al., ) .
Covariates: age, years of education .
S-PA is measured in METs units and CRF in ml/kg*min .
### Mediating Effects
#### Model 1
Model 1 (see ) tested the mediating effect of each molecular, brain volume and behavioral outcome in the relationship between S-PA and CRF with the assessed cognitive functions in women and men. In the relationship between S-PA and cognitive domains, none of the outcomes at Level 1, two or three showed significant indirect effects in the mediation analyses in women or men. When CRF was the predictor, we found statistically significant indirect effects for HGF in the CRF-executive function (Path AB = β = 0.29, SE = 0.17, 95% CI: 0.02, 0.70) and in the CRF—working memory (Path AB = β = 0.25, SE = 0.15, 95% CI: 0.02, 0.59) relationships in men. Indirect effects were also significant when TNF-α was the mediator in the association between CRF and inhibition (Path AB = β = 0.31, SE = 0.19, 95% CI: 0.04, 0.77) only in men. There were no significant indirect effects for any outcome in the CRF-cognition relationship in women (see ).
Mediation results of Models 1, 2, 3, and 4 in women and men.
#### Model 2
Model 2 (see ) examined the mediating effects of molecular outcomes in the relationship between S-PA and CRF with brain volumes in women and men. In the association between S-PA and brain volumes, none of the molecular outcomes showed significant mediating effect in women or men. In the mediation analyses with CRF as a predictor, we found statistically significant indirect effects for TNF-α in the relationship between CRF and cingulate cortex volume (Path AB = β =-0.26, SE = 0.14, 95% CI: −0.58, −0.03) only in men. We did not find any significant mediating effects for any molecular outcome in the CRF-brain volume association in women (see ).
#### Model 3
Model 3 (see ) assessed the mediating effects of molecular outcomes in the relationship between S-PA and CRF with psychological health and daily activity. In the relationship between S-PA and behavioral outcomes, none of the molecular outcomes showed significant indirect effects in women or men. When CRF was the predictor, we found statistically significant indirect effects for HGF in the relationship between CRF and the subscale of subjective sleep quality of the Pittsburgh Sleep Quality Index (Path AB = β = −0.21, SE = 0.13, 95% CI: −0.53, −0.03) in men. There were no significant mediating effects for any molecular outcome in the CRF-behavioral relationship in women (see ).
#### Model 4
Model 4 (see ) tested the mediating effects of brain volume outcomes in the relationship between S-PA and CRF with psychological health and daily activity. In the association between S-PA and behavioral outcomes, none of the brain volume outcomes showed significant mediating effects in women or men. In the mediation analyses with CRF as a predictor, we found statistically significant indirect effects of precuneus volume in the relationship between CRF and a subscale of subjective sleep quality from the Pittsburgh Sleep Quality Index (Path AB = β = −0.26, SE = 0.17, 95% CI: −0.65, −0.00) in men. We did not find any significant mediating effects for any brain volume outcome in the CRF-behavioral association in women (see ).
## Discussion
To our knowledge, we are the first to describe the relationship between physical activity outcomes and three different levels of potential mechanisms–molecular, brain volume and behavioral–in healthy late-middle-aged women and men. Moreover, we analyzed these potential mediators in the relationship between exercise and CRF with cognition and between themselves stratifying results by sex.
At the molecular level, Level 1, our results support previous evidence (Sallam and Laher, ) suggesting that physical activity might be related to reduced levels of inflammation. However, despite that self-reported exercise and CRF measures were highly correlated, we found sex differences when examining these outcomes. In particular, we found that reduced TNF-α levels were related to greater energy expenditure in sportive physical activity only in women and to higher CRF levels only in men. Men with higher CRF also showed reduced levels of HGF, independently of BMI. Both biomarkers are related to pro-inflammatory processes, obesity, and insulin resistance and, therefore, reduced levels might induce a neuroprotective effect (Kiliaan et al., ). Nevertheless, we can not discard that these results could be related to other processes besides inflammation given the pleiotropic nature of these markers. Interestingly, when analyzing the mediating role of molecular outcomes in the physical activity-cognition association, we found that the reduced levels of these inflammation-related markers might be an important mediator of the brain and cognitive CRF-related benefits. In men, HGF was a significant mediator in the association between CRF and executive function, working memory, and sleep quality and TNF-α was a significant mediator in the relationship between CRF and inhibition. Interestingly, those men with higher CRF also showed that TNF-α was a significant mediator of the cingulate cortex volume, which is functionally involved in executive function processes such as inhibition (Huang et al., ). This fact highlights the potential role of reduced neuroinflammation as a result of exercise and enhanced CRF to promote greater executive functions.
Our findings at brain volume level, Level 2, are consistent with previous findings that physical activity has been consistently associated with the maintenance of brain volume and with less atrophy across the lifespan (Erickson et al., ). One of our relevant results is that frequent exercise was related with greater temporal lobe volume in women and men, which is a key structure for memory and it deteriorates with aging. Moreover, consistent with Barha et al. ( ), women that expended more energy in sportive physical activities had greater dorsolateral prefrontal cortex volume, which is involved in supporting executive functions, especially working memory. In other studies greater CRF levels have been linked to greater brain volumes (Raichlen et al., ; Wittfeld et al., ). Interestingly, in our study, higher CRF levels were associated with less ventricular volume and higher volumes of temporal and parietal lobes, specifically of the precuneus, in men but not in women. Despite previously published evidence for the mediating effects of brain volume in the relationship between physical activity outcomes and cognition, such as the mediating role of the hippocampus in the association between CRF and spatial memory (Erickson et al., ), we found that physical activity and fitness were associated with larger volumes, but that this was unrelated to cognition.
Previous literature about Level 3 outcomes reported that active lifestyles such as exercise are also related with mood (Fox et al., ; Willis et al., ) and sleep patterns (Kline et al., ) in older adults. Curiously, we found similar results, but this was significant only in men. Men performing more exercise reported higher sleep efficiency and fewer sleep disturbances and those with higher levels of CRF presented fewer depressive symptoms and better general mood. In mediation analyses, and in accordance with previous papers relating diminished precuneus with insomnia (Grau-Rivera et al., ) and sleep restriction (Long et al., ), we found that greater volume of the precuneus in men mediated the higher CRF-better subjective sleep quality association. Surprisingly, in our sample, physical activity was not related with these behavioral outcomes in women. Based on previous papers, we hypothesize that mood (Wharton et al., ) and sleep (Baker et al., ) patterns in women could be influenced by other parameters, such as hormonal changes.
Our results are consistent with previous findings about exercise on health outcomes as we described above. Nevertheless, we not only add support to the multi-level benefits of regular exercise but also highlight the sex differences in the role of each physical activity outcome as we stratified results by sex. CRF was an outcome highly related to cognition and to the molecular, brain, and psychological cascade of changes in men but not in women. However, benefits of exercise at the molecular level including reduced TNF-α levels and brain volumes –dorsolateral prefrontal cortex and temporal lobe, were also observed in women but were linked to the amount of self-reported energy expended in sportive activities during last month and not to CRF. These results could be related to sex differences in the musculoskeletal and cardiovascular systems and their adaptations in response to exercise (Barha and Liu-Ambrose, ; Ansdell et al., ). Current literature suggests that women experience less metabolic stress for the same amount of exercise and, in turn, lesser adaptative response (e.g., less increase in CRF levels). Moreover, evidence suggests that the integrative response to exercise might be mediated by an oestrogenic effect in females. Therefore, differences in sex hormones that act as neurosteroids and interact with molecular growth factors, brain structures and cognition in a different manner for women and men could explain part of this variability (Ansdell et al., ). Besides, this is in accordance with recent bibliography stating sex differences in the association of exercise with brain and cognitive outcomes (Lindwall et al., ; Varma et al., , ; Barha et al., ; Dimech et al., ). Another potential explanation could be a dose-effect bias related to the described differences in the amount of physical activity performed by women and men. Although, this fact could be a source of variance, sex differences in the distribution of SPA and CRF observed in our sample are consistent with previously published literature reporting that men are more physically active (Al-Mallah et al., ). Additionally, there is evidence reporting no significant differences between PA levels but showing sex differences in CRF levels and significant associations between CRF and functional brain outcomes only in men (Dimech et al., ). Therefore, future studies should address the influence of dose by sex in sex-balanced samples.
It must be acknowledged that our results are based on a cross-sectional design which allows us to describe statistical relationships but not causal conclusions about the neuroprotective effects of exercise. Moreover, further studies should address these aims in samples with different physical activity profiles, age groups, sex-balanced groups, and a non-estimated measure of CRF. Including Heart Rate Reserve to estimate intensity/exertion during the test could provide further information about the physical activity status of participants and inform about its role in relation with physiological mechanisms. We stress the need to stratify results by sex and study the role of the hormonal profile in the molecular cascade that might explain the benefits of exercise to brain health.
## Conclusion
From a clinical perspective, we showed that exercise might be related to reduced levels of inflammatory markers and increased brain volume in areas commonly deteriorating in aging. Regular exercise is also associated to better psychological and sleep health in men.
Our results contribute to the field of research adding evidence about the mediating effects of molecular biomarkers capable to modulate the immune system in the relationship between CRF and cognition in men. Moreover, our results suggest sex differences in the association between physical activity outcomes and molecular, brain and psychological outcomes. This might be related to sex differences in the musculoskeletal and cardiovascular adaptations after exercise as well as sex differences in the hormonal profile. Future studies should address the interaction of the hormonal profile with the molecular and brain measures commonly studied in cross-sectional designs and RCTs.
## Data Availability Statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics Statement
The studies involving human participants were reviewed and approved by Bioethics Commission of the University of Barcelona (IRB00003099) and Clinical Research Ethics Committee of IDIAP Jordi Gol (P16/181). The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
AC-S and FR-C participated in the study concept and design, acquisition, analyses, and interpretation of data as well as in the elaboration of the manuscript. RD-A contributed processing the neuroimaging data. NL-V collaborated in the acquisition of the data. RD-A, NL-V, AKS, PT-M, GP, PM-A, AH-T, SD, MV, and KE critically reviewed the content of the article. MM conceptualized the study, contributed to the study design and the implementation as Principal Investigator and supervised all procedures, and the elaboration of the manuscript. All authors contributed to the article and approved the submitted version.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
In Alzheimer's disease (AD), amyloid β deposition-induced hippocampal synaptic dysfunction generally begins prior to neuronal degeneration and memory impairment. Lycium barbarum extracts (LBE) have been demonstrated to be neuroprotective in various animal models of neurodegeneration. In this study, we aimed to investigate the effects of LBE on the synapse loss in AD through the avenue of the retina in a triple transgenic mouse model of AD (3xTg-AD). We fed 3xTg-AD mice with low (200 mg/kg) or high (2 g/kg) dose hydrophilic LBE daily for 2 months from the starting age of 4- or 6-month-old. For those started at 6 month age, at 1 month (though not 2 months) after starting treatment, mice given high dose LBE showed a significant increase of a wave and b wave in scotopic ERG. After 2 months of treatment with high dose LBE, calpain-2, calpain-5, and the oxidative RNA marker 8-OHG were downregulated, and presynaptic densities in the inner plexiform layer but not the outer plexiform layer of the retina were significantly increased, suggesting the presynaptic structure of retina was preserved. Our results indicate that LBE feeding may preserve synapse stability in the retina of 3xTg-AD mice, probably by decreasing both oxidative stress and intracellular calcium influx. Thus, LBE might have potential as a neuroprotectant for AD through synapse preservation.
## Introduction
The number of people with dementia is about 55 million worldwide in 2021, and is expected to increase to about 140 million by 2050 (World Health Organization., ). Alzheimer's disease (AD) is the most common type of dementia (60–70% of cases). It is an evolving humanitarian challenge, with growing numbers due to the aging of the population. In the United States alone, the estimated total healthcare cost for the treatment of AD in 2020 was estimated at $305 billion, and is estimated to increase to more than $1 trillion by 2050 (Wong, ). Early detection and diagnosis of AD, leading to early initiation of current AD therapies, is associated with improved quality of life and economic outcomes. This is despite the modest effectiveness of current AD therapies, therefore finding an improved and affordable treatment would be very important for society.
Transgenic mouse models harboring mutated human genes associated with familial forms of AD are advanced preclinical tools in the study of mechanisms underlying AD. The triple transgenic mouse model of AD (3xTg-AD) incorporates a Swedish amyloid precursor protein (APP) mutation, a human mutant presenilin 1 (PSEN1) gene PS1 (M146V) knock-in, and a tau (P301L) transgene (Oddo et al., ). These mice develop intracellular amyloid beta (Aβ), Aβ plaques and neurofibrillary tangle (NFT) like pathology in a progressive and age-dependent manner similar to human AD (Oddo et al., ; Edwards et al., ). AD disproportionately affects women in both disease prevalence and rate of symptom progression (reviewed by Fisher et al., ). 3xTg-AD mice also display faster progression in females than in males. Starting at 6 months of age, female 3xTg-AD mice exhibit greater cognitive deficits than males. This disparity is still evident at 9 months of age, when female mice have significantly higher stress responses (Clinton et al., ). Thus, it is crucial to specify sex and age to examine early-stage AD mechanisms and/or novel therapeutic interventions.
In the early stage of AD, memory impairment and loss of hippocampal synapses show up prior to frank neuronal degeneration (Selkoe, ; Chen et al., ). In early stages of familial AD, neuronal RNA oxidation occurs (Nunomura et al., , ; Nunomura and Perry, ). Increased oxidative stress (Xu et al., ), uncontrolled hyperactivation of a family of cysteine proteases called calpains (Mahaman et al., ), and synaptic dysfunction (Jackson et al., ) may contribute to the disease process. The abnormal activation of calpain could induce synaptic dysfunction (Chen et al., ). Calpain inhibition not only prevented Aβ (1-42) induced Ca influx and neuronal death in primary cortical neuron culture (Lee et al., ), but also decreased the AD-like pathology and cognitive decline in aged 3xTg-AD mice (Medeiros et al., ). Mounting evidence indicates that therapeutic approaches aiming to protect against these in the early stage of AD might stop or reverse disease progression (Jackson et al., ).
Emerging evidence suggests that visual performance is impaired early in AD (Van Wijngaarden et al., ; reviewed by Shah et al., ). Visual impairments reported in AD patients include nerve fiber layer thinning, degeneration of retinal ganglion cells (RGC), and changes to vascular parameters (Hart et al., ). Moreover, Aβ deposits were observed in multiple layers of the retina, and were significantly associated with brain Aβ burden. The retina (and its related vasculature and optic nerve) shares the same embryological origin as well as some anatomical and physiological properties as the brain (Patton et al., ). In transgenic AD animals, retinal abnormalities demonstrated by deterioration in visual function or reactive gliosis were reported (Chidlow et al., ; Chiquita et al., ; Zhang et al., ). Any neuroprotective agents benefitting the visual system might be useful for AD treatment.
Lycium barbarum (LB) is named Gouqizi in Chinese, and wolfberry in English. It has been used medicinally for more than 2500 years for its abilities to delay aging and improve visual acuity. LB has been regularly consumed by a vast number of people, with little to no side effects, making it a good candidate for an anti-aging agent (Gao et al., ). The neuroprotective effect of LB was first demonstrated in animal models of retinal degenerative diseases such as glaucoma, age-related macular degeneration (ARMD) (Chiu et al., ). It appears to protect the visual system through four primary processes: neuroprotection, blood-retinal barrier stabilization, antioxidation, and modulation of retinal immune function via the retinal microglial cells and Müller cells (reviewed by Manthey et al., ). LB can inhibit two key pro-apoptotic signaling pathways (JNK and PKR) in Aβ peptide neurotoxicity (Chang et al., ; Suen et al., ; Yu et al., , ). The neuroprotective effects of LB were further demonstrated in the preservation of cognitive functions and decrease in Aβ deposition in APPswe/PS1-del9 Tg-AD mice (Zhang et al., ; Zhou et al., ).
Studies using extracts revealed the beneficial effects of LB. The major LB constituents demonstrating anti-aging properties include LB polysaccharides, carotenoids (zeaxanthin and β-carotene), betaine, flavonoids and vitamins (Gao et al., ). Our previous study using hydrophilic extract from LB (LBE) demonstrated the enhancement of retinal light response in the retina of 5xFAD mice (Zhang et al., ). Further study in cell culture using the IMG microglial cell line showed that LBE promoted activation of microglia with anti-inflammatory character (M2 polarization) and reduced oligomeric Aβ induced inflammatory reactions in microglia (Sun et al., ). LBE also had antioxidant effects under H O stimulation of primary mixed glial cells (Zheng et al., ). Therefore, if LBE is applied early in AD, disease progression might be significantly delayed.
In the current study, we aimed to explore the effects of LBE feeding on the retinal changes in young female 3xTg AD mice, and to unveil the underlying mechanisms. The evaluation was focused on retinal function changes and RNA oxidation, calpain activation and synaptic proteins in the retina. Because much of retinal function and pathology can be observed non-invasively in living animals and humans, the ability to evaluate treatment effects in the retina would be an invaluable and low cost translational research platform to fulfill an unmet need of healthy aging.
## Materials and Methods
### Animals
3xTg-AD mice [B6;129-Psen1tm1Mpm Tg (APPSwe, tauP301L)1Lfa/J] (Oddo et al., ) were purchased from the Jackson Laboratory (stock No. 004807, Bar Harbor, ME, USA). Since the mice are homozygous for mutations in the PS1, APP, and tau genes, we maintained the colony by breeding homozygous 3xTg-AD mice to each other. Non-transgenic C57BL/6J mice were obtained from the Laboratory Animal Unit of the University of Hong Kong and used as wild-type (WT) controls. To reduce sexual dimorphism, especially in the early stage of AD (from 4 to 8 months), only female mice were used in the study. All animals were maintained in a temperature-controlled room with a 12-h light/dark cycle throughout the observation period. All animal procedures were performed according to the ARRIVE guidelines and were approved by the Committee on the Use of Live Animals in Teaching and Research of the University of Hong Kong. All efforts were taken to minimize the number of animals used and their suffering.
### LBE Preparation, Animal Feeding and Grouping
LBE was provided by Eu Yan Sang (HK) Ltd. LB from NingXia Huizu Autonomous Region, the People's Republic of China, was used. The extraction procedure was the same as in our previous report (Sun et al., ). Briefly, 2.5 kg of LB was washed and soaked in 40°C ultrapure water for 15 minutes. After 1 h boiling, the filtered drug residues were boiled again. The extracts from two boils were combined and concentrated to 1.25 kg. The final LBE was weighed and diluted in ultrapure water (w/v) to produce a stock solution (100 g/L).
3xTg-AD mice were orally fed with a low dose (200 mg/kg body weight) or high dose (2 g/kg) of LBE daily for 2 months with distilled water feeding as vehicle control, and age-matched C57BL/6J mice with the same treatments were regarded as WT controls. For each age stage (4 and 6 months old), five groups were included: water treated WT mice, high dose LBE treated WT mice, water treated AD mice, low dose LBE treated AD mice, and high dose LBE treated AD mice. Each group consisted of six mice.
### Flash Electroretinography
Retinal function was evaluated by an electroretinography (ERG) system (Espion E2 Electrophysiology System, Diagnosys LLC, USA) according to the standard protocol of the International Society for Clinical Electrophysiology of Vision (ISCEV). Mice were anesthetized with intraperitoneal injection of a mixture of ketamine (0.1 mg/g) and xylazine (10 μg/g). The eyes were administered 1% Mydriacyl (Alcon, Fort Worth, USA) to dilate the pupils and 0.5% Alcaine (Alcon, Fort Worth, USA) to reduce cornea sensitivity. ERG signals under a scotopic flash intensity of 3.0 cd·s/m and photopic flash intensity of 22.8 cd·s/m were recorded. Amplitude and latency of ERG signals were filtered and analyzed by Axon pCLAMP 10 (Molecular Devices Corp., Sunnyvale, CA, USA).
### Histology and Morphometric Analysis of Retinal Sections
Mice were sacrificed by an overdose of anesthesia with an intraperitoneal injection of Dorminal (0.1–0.15 mg/g). The eyeballs were harvested, fixed, processed and paraffin-embedded for sectioning. Five micrometer thick retinal cross sections with intact optic nerves were deparaffinized and stained with hematoxylin and eosin. Images were captured with a light microscope (Eclipse80i, Nikon) under 40x magnification. Morphometric analysis was carried out as in our previous report (Chan et al., ). The thickness of the inner retinal layer (IRL) was measured from inner limiting membrane to the outermost point of the inner nuclear layer (INL), and the outer nuclear layer (ONL), as well as the number of RGC both in the central or peripheral retina, were calculated using Fiji software (NIH, MD, USA).
### Immunohistochemical Detection in Retinal Sections
Retinal sections were rehydrated and boiled in 95°C citric acid buffer (10 mM, pH 6.0) for 15 min for antigen retrieval. Sections were then incubated in 10% normal goat serum with 1% bovine serum albumin (Sigma-Aldrich, St. Louis, MO, USA) and 0.1% Triton X-100 (Sigma-Aldrich, St. Louis, MO, USA) in PBS (pH 7.2–7.4) for 1 h at room temperature to block nonspecific binding. Following blocking, sections were incubated with either of the following primary antibodies overnight at 4°C: anti-goat Brn3a (1:500, Santa Cruz, Dallas, USA); anti-mouse PKC-α (1:500, Millipore, Burlington, Massachusetts, USA); anti-rabbit GFAP (1:500, Abcam, Cambridge, UK); anti-rabbit Iba-1(1:500, WAKO, Chou-ku, Osaka, Japan); anti-rabbit synaptophysin (1:500, Abcam); anti-rabbit PSD-95 (1:500, Abcam); anti-rabbit 8-OHG (1:500, Abcam); anti-rabbit calpain-2 (1:200, Thermo Fisher Scientific, Waltham, Massachusetts, USA); or anti-rabbit calpain-5 (1:200, Thermo Fisher Scientific). After washing with PBS, the sections were incubated with Alexa-568 or 488 fluorescent-conjugated goat IgG secondary antibody (1:500; Thermo Fisher Scientific) for 1 h at room temperature. The slides were then washed in PBS and counterstained with 4',6-Diamidino-2-phenylindole (DAPI) (1:1000). Images were captured using a ZEISS LSM 800 Confocal microscope (Carl Zeiss Microscopy GmbH, Germany).
### Statistical Analysis
Statistical analysis was performed by t -test or one-way ANOVA followed by post-hoc analysis using SPSS software for Windows (version 19.0; SPSS, Inc., IL, USA). Data were reported as mean ± SD. A value of P < 0.05 was considered significant for statistical analysis.
## Results
### Age Dependent Retinal Functional Impairment and Accumulation of Intracellular Amyloid in the 3xTg-AD Mice
Retinal function reflected by ERG parameters demonstrated age dependent changes from 4 to 14 months ( ). Compared with WT control, the scotopic retinal response in a wave started to decline significantly at age 6 months, followed by a rapid decline between 6 and 8 months ( ). Significant decline in b wave could be detected between 6 to 8 months and 9 to 14 months in 3xTg-AD mice. Under light adaptation, the changes in b wave were not significant when compared to WT control. There was significant PhNR reduction at 12 months of age ( , p < 0.05). Deterioration of retinal function in ERG might be caused by decreased neuronal activity or connectivity or by loss of retinal neurons. At the age of 8 months, morphometric comparison of the thickness of IRL and ONL, and number of RGC in H&E-stained retinal sections ( ), did not show any significant changes between WT and 3xTg-AD mice. Further immunohistochemical detection and counting of Brn-3a positive RGCs, PKC-α labeled retinal bipolar cells and GFAP labeled astrocytes and active Müller cells also confirmed that there was no significant change in the cell densities at the age of 8 months ( ).
Flash-ERG and amyloid beta (Aβ) deposition in retinas of non-Tg control mice and 3xTg-AD mice at different ages. The scotopic retinal response (A) of 3xTg-AD mice started to decline at 6 months of age, followed by a rapid decrease from 6 to 8 months of age (A,C) . * p < 0.05. Further significant decline in b wave changes can be detected between 6–8 months and 9–14 months. Significant reduction in a wave started at 6 months when compared with non-Tg controls. PhNR detected in the photopic response revealed a significant reduction at 12 months when compared with non-Tg controls (B) . * p < 0.05; NS, no significant change. Typical ERG wave forms are shown in C. Intracellular Aβ deposition of 4G8 (D) and Aβ1-42 (E) were detected in the RGCL and INL of retinas of 3xTg-AD mice at 8 months only (arrows). No positive signal could be detected in the non-Tg controls or 6 month old 3xTg-AD retinas. Scale bar: 20 μm.
Aβ expression in the retinas of 3xTg-AD mice was detected by two amyloid antibodies, anti-Aβ1-42 and anti-Aβ17-24 (4G8) at the ages of 6 and 8 months, when significant functional changes are found in AD retina. No extracelluar Aβ was detected at either time point. Accumulation of intraneuronal Aβ was found in the ganglion cell layer (GCL) and inner nuclear layer (INL) of the retinas of 3xTg-AD mice at 8 months of age ( ). Increased Aβ expression was predominately located in the soma of neurons. Aβ staining was not observed in the retinas of age-matched WT control mice.
### LBE Feeding Preserved Retinal Function of 3xTg-AD Mice
The 3xTg-AD mice begin to show impairment on the Morris water maze and inhibitory avoidance, coincident with the onset of intracellular Aβ in the brain at age 4 months. Extracellular Aβ is detectable in the cortex at 6 months of age (Billings et al., ). The ages at which feeding was started in this study were chosen as the ages when intracellular (4th month) and extracellular (6th month) Aβ become detectable in the brain. Our previous study using the early onset 5xFAD mouse model found that the retinal function in dark conditions (scotopic response) was enhanced by LBE at 20 g/kg oral feeding (Zhang et al., ). In the current study of 3xTg-AD mice, lower doses were chosen because less Aβ accumulation was reported in these mice than in the 5xFAD model. The ERG test was performed at 1 or 2 months after feeding started. As shown in , no significant changes in scotopic or photopic retinal function were induced by LBE feeding when the feeding started at age 4 months. When the feeding started at age 6 months, 1 month of 2 g/kg LBE gavage significantly increased both a-wave and b-wave amplitude in scotopic ERG compared to the no-feeding 3xTg-AD mice. The same tendency of a-wave and b-wave preservation was observed after 2 months of 2 g/kg LBE feeding, but without statistical significance ( ). No significant changes in photopic b-wave or PhNR were observed in LBE-fed 3xTg-AD mice ( ). At 8 months of age, morphometric analysis on the neuronal cell number and Müller cell activation also remained unchanged in 3xTg-AD mice with LBE feeding ( , ).
Effect on retinal function of LBE oral feeding started at different ages. LBE feeding started at 4 months or 6 months of age in both non-Tg and 3xTg-AD mice. For two months, two doses [low (200 mg/kg) or high (2 g/kg)] of LBE were given daily, while water served as the control and non-feeding as the age-matched AD control. LBE feeding from age 4 months showed a trend toward better scotopic retinal response for both a-wave and b-wave in 3xTg-AD mice (A) . When started at 6 months of age, high dose LBE feeding of 3xTg-AD mice for 1 month showed significantly higher retinal response to scotopic light intensity (3.0 cd·s/m ) compared to 3xTg-AD mice given only water (C) . Photopic b-wave and PhNR, indicating the function of the inner nuclear layer and RGCs, showed no differences among all the groups (B,D) . * p < 0.05, ** p < 0.01; NS, no significant change. Typical ERG wave forms of various groups at 1 month and 2 months after the start of feeding (E) .
### LBE Feeding Decreased RNA Oxidation and Activation of Calpain-2, Calpain-5 in the Retinas of 3xTg-AD Mice
The guanosine oxidation product 8-hydroxyguanosine (8-OHG) is one of the most abundant and best characterized biomarkers for RNA oxidative lesions (Feyzi et al., ). The amount of cellular 8-OHG is a sensitive measurement of oxidative stress and is an RNA damage biomarker (Xu et al., ). At 8 months of age, there was an obvious trend of increased 8-OHG in the retinas of water-fed 3xTg-AD mice compared to the age matched WT controls ( , first lane). Comparing with water-fed mice, LBE oral feeding did not change the 8-OHG level in WT mice, while reducing its level in the 3xTg-AD mice, especially with the high dose (2 g/kg) ( ).
Two months LBE oral feeding downregulated expression of 8-OHG, calpain-2, and calpain-5 in 3xTg-AD mice at the age of 8 months. (A) Representative images of 8-OHG, calpain-2, and calpain-5 stained retinal sections of 3xTg-AD mice and age matched non-Tg control mice fed with water, low dose LBE or high dose LBE. Semi-quantification analysis showed increased 8-OHG (B) , calpain-2 [ (C) , * P < 0.05; ** P < 0.01], and calpain-5 [ (D) , * P < 0.05; ** P < 0.01] in the retinas of 3xTg-AD mice, which could be suppressed by LBE feeding.
With an increase in age, increased oxidative stress and Aβ deposition, the normally tightly controlled calpain activation regulatory system becomes impaired. Increased calpain activation is involved in the pathogenesis of AD (Mahaman et al., ). In the retinas of 3xTg-AD mice, the expression of both calpain-2 ( , second lane) and calpain-5 ( , third lane) showed a tendency toward increase compared to WT mice at 8 months of age. While there is no effect of LBE feeding comparing with water feeding in the WT mice, LBE feeding at both doses significantly decreased the expression of calpain-2 and calpain-5 in the retinas of 3xTg-AD mice ( ).
### LBE Feeding Preserved Impaired Pre-synaptic Densities in Retinas of 3xTg-AD Mice
Synaptic loss is a common pathological change in AD patients and Tg-AD models. Both pre- and post-synaptic proteins were detected in the retinas. While there were no obvious changes in the immunoreactivity of a post-synaptic protein marker (PSD-95) ( ), a pre-synaptic protein marker (synaptophysin) was significantly decreased in the retinal inner plexiform layer (IPL) of 8 month old 3xTg-AD mice ( ). The reduced synaptophysin expression level was restored by LBE feeding, with this restoration reaching significant levels at high dose ( ). Pre-synaptic protein expression in the OPL layer showed a similar but non-significant tendency. Furthermore, we explored post-synaptic changes with PSD-95 staining, and no significant changes were observed in 3xTg-AD mice compared with wild type mice ( ).
LBE oral feeding significantly preserved presynaptic terminals in 3xTg-AD mice at 8 months of age. (A) Representative images of retinal sections stained for the presynaptic marker, synaptophysin, in 8 month old 3xTg-AD mice and age matched non-Tg control mice fed with water, low dose LBE or high dose LBE. Semi-quantification analysis revealed decreased presynaptic densities in the IPL but not the OPL in 8 month old 3xTg-AD mice, and rescue of the decrease by high dose LBE feeding [ (B) , * P < 0.05, ** p < 0.01]. Representative images of retinal sections stained for the postsynaptic marker, PSD-95, in 8 month old 3xTg-AD mice and age matched wild type mice fed with water, low dose LBE or high dose LBE (C) . No significant changes were observed in postsynaptic densities among all the groups (D) .
## Discussion
Retinal Aβ aggregation has been reported in various AD transgenic mouse models (reviewed by Chiu et al., ). In the current study, we used the retina as a window to study early changes in female 3xTg-AD mice. At the 8th month of age, the retinal function detected by flash ERG revealed a significant decrease of scotopic b-wave in the 3xTg-AD mice compared to WT mice. Morphometric analysis showed no significant loss of neurons in the 3xTg-AD retina. Intracellular Aβ, oxidative RNA marker 8-OHG, and calpain-2 and -5 were increased in the retinal neurons in the RGCL and the INL. The pre-synaptic protein, synaptophysin, was significantly decreased in the IPL of 3xTg-AD mice retinas. Oral feeding of LBE (2 g/kg) starting at 6 months of age decreased the activation of calpain-2 and -5 and restored the synaptophysin expression after 2 months of feeding.
As an extension of the central nervous system, the retina is considered a valuable tool for the study of CNS disorders (London et al., ). Accumulating evidence has revealed retinal abnormalities in AD patients, including retinal function impairment and structural changes, such as optic nerve damage, loss of ganglion cells, and retinal thinning (Chiquita et al., ). These facts provided support for the idea to use the retina as a window into the AD nervous system, to help overcome the difficulty in accessing the brain and the high cost of current modalities, including PET and CSF biomarkers (Liao et al., ; Ngolab et al., ). More importantly, retinal Aβ plaques tended to show up in the retina prior to brain plaques (Liao et al., ). In 3xTg-AD mice, scattered extracellular Aβ deposition in the caudal hippocampus existed at the age of 6 months (Belfiore et al., ). In this study, intracellular Aβ immunoreactivity (4G8 or Aβ1-42) was detected in the retina at 8 months of age, together with RNA oxidation, calpain activation and reduced synaptophysin expression in the inner retina. These changes might be the reason that significant scotopic retinal function reduction in both a and b wave in the 3xTg-AD mice was detected. Morphometric analysis of the thickness change in the inner and outer retina in H&E retinal sections can be translated to the clinical detection of retinal structure using optical coherence tomography (OCT). Retinal function detected by ERG might be used as an early marker for AD diagnosis. There was no significant change in the retinal thickness using classical histopathology analysis. Moreover, no significant neuronal loss in the retina could be confirmed using immunohistochemical detection of specific cell markers such as Brn-3a and PKC. Loss of synapses in the affected brain regions correlates best with cognitive impairment in AD patients that has been considered as an early mechanism that precedes neuronal loss (Chen et al., ). The decrease of retinal function can be explained by the reduction of synaptic protein, especially in the inner retina in the early stage of AD.
A cytoplasmic oxidative RNA nucleoside, 8-OHG, is markedly increased within the hippocampus and temporal neocortex in AD patients at an early stage of the disease (Nunomura et al., ; Nunomura and Perry, ). Synaptic dysfunction is closely associated with oxidative stress in AD (Kamat et al., ). LBE has been proved to be protective against oxidative damage in various diseases (Gao et al., ), and to modulate glial cell function in vitro (Zheng et al., ). Two months of 2 g/kg LBE oral feeding starting at 6 months proved the effectiveness of antioxidation and preservation of synaptic function in the retina of 3xTg-AD mice. Although the preserved retinal function can only be significantly detected at the 1 month but not the two-month time point in the LBE-fed 3xTg-AD mice, an increase in animal numbers might reveal a significant effect at 2 months.
In neurodegenerative diseases including AD, abnormal activation of calpains favors Aβ accumulation and tau hyperphosphorylation in neurons and is associated with synaptic dysfunction (Trinchese et al., ; Medeiros et al., ; Diepenbroek et al., ). Upregulated calpain-2 and -5 were detected in the retinas of 8 month-old 3xTg-AD mice. They were down regulated by 2 months of LBE feeding, which also stabilized pre-synaptic protein expression and improved retinal function.
Gliosis is an activation of glial cells in response to central nervous system injury. In the brain tissue of 3xTg-AD mice, increased densities of active astrocytes and microglia were observed at 7 months (Caruso et al., ). Activation of astrocytes and Müller cells was also identified in the retinas of 3xTg-AD mice at 8 months of age (Edwards et al., ). However, in our study, immunochemical staining of GFAP and Iba-1 did not detect significant changes in astrocytes and microglial cell numbers, respectively, in either the brain or the retina of 8 month old 3xTg-AD mice. These findings may be explained by the fact that neuronal cell damage and extracellular Aβ deposition may be the primary inducers of glial activation in AD, but neither were present in the retinas of the 8 months old 3xTg-AD mice in the current study. Still, a possibility exists that activation of glial cells might have occurred in our mice but may have been too subtle to detect. Oligomeric Aβ induces M1 activation (pro-inflammatory) of the IMG cell line, but pretreatment with LBE shifts the cells to M2 activation (anti-inflammatory) (Sun et al., ). Therefore, besides the direct LBE effect on retinal neurons, modulation of retinal glial function cannot be ruled out.
A limitation of this study is that we did not identify which molecular component was responsible for the neuroprotectant activity we examined. However, we used a very simple extraction procedure to make it easy for the general population to use the same method. Therefore, it is possible that, even without pinpointing one chemical for further investigation, this LBE can be quickly and affordably adopted by more patients or even the entire aging population with the ultimate goal of reducing illness and economic burden.
## Conclusion
Our study demonstrated a protective effect of LBE feeding on synaptic function in the retinas of 3xTg-AD mice. This may be due to the ability of LBE feeding to reduce intracellular RNA oxidation and calpain-2 and -5 hyperactivation, and thus to stabilize synapses. LBE acted through multiple pathways to stabilize retinal function in 3xTg-AD mice. Daily LBE supplementation might be beneficial for neuronal function and survival even during aging in the presence of excess Aβ.
## Data Availability Statement
The original contributions presented in the study are included in the article/ . Further inquiries can be directed to the corresponding authors.
## Ethics Statement
The animal study was reviewed and approved by Committee on the Use of Live Animals in Teaching and Research of the University of Hong Kong.
## Author Contributions
KC, KS, and JL: conceptualization. JL, LB, SY, YL, and GX: methodology. JL and GX: data analysis. KC and KS: project administration. JL and LB: writing original draft. KC, LB, and RC: manuscript review and editing. All authors read and approved the final manuscript.
## Funding
This project was supported by Health and Medical Research Fund (HMRF, Project No: 14151281) and Midstream Research Program for Universities (MRP, Project No: MRP-092-17X) in Hong Kong to KC.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
|
Biologically speaking, normal aging is a spontaneous and inevitable process of organisms over time. It is a complex natural phenomenon that manifests itself in the form of degenerative changes in structures and the decline of functions, with diminished adaptability and resistance. Brain aging is one of the most critical biological processes that affect the physiological balance between health and disease. Age-related brain dysfunction is a severe health problem that contributes to the current aging society, and so far, there is no good way to slow down aging. Mesenchymal stem cells (MSCs) have inflammation-inhibiting and proliferation-promoting functions. At the same time, their secreted exosomes inherit the regulatory and therapeutic procedures of MSCs with small diameters, allowing high-dose injections and improved therapeutic efficiency. This manuscript describes how MSCs and their derived exosomes promote brain neurogenesis and thereby delay aging by improving brain inflammation.
## Introduction
Aging is a topic that has fascinated scientists and philosophers throughout history ( ). As the global population is aging, the prevention of brain aging is a common problem. Aging is associated with a progressive decrease in the effectiveness of mechanisms that maintain homeostasis of the body and its organs and tissues, which leads to an increased risk of various pathologies and death ( ). At present, China has entered an aging society, and it is expected that in 2050, 30% of the total population will be over 60 years old, The problem of aging is becoming increasingly severe. It is common in science to think of human aging as a set of characteristics that change over time and to refer to someone as “older” or “younger” ( ). Brain aging is a significant cause of most neurodegenerative diseases and is often irreversible and lacks an effective treatment, leading to a dramatic decline in quality of life ( ). As with other organ systems, brain function gradually declines during the aging, mainly in learning and memory functions ( ). Cognitive dysfunction can be described as an imbalance in the structural and functional organization of the brain at all three levels: the molecular/cellular level, the local circuit level, and the large-scale network level. Each of these levels interacts dynamically with the others and exhibits the characteristics of an open complex system ( ). Cognitive function is complex and may also be affected by diet, and adequate nutrition is effective in preventing cognitive decline ( ). Therefore, since the development of medicine, scientists have been working to explain the phenomenon of cognitive decline in the elderly.
Some studies point out that age-related cognitive decline is characterized by a considerable reduction or even death of neurons in the brain ( ; ). In the hippocampus (and perhaps in other brain areas), neuronal death can partially compensated by neuronal generation. However, neuronal production is significantly impaired with age ( ). In the adult mammalian hippocampus, new neurons are derived from the stem and progenitor cell divisions, a process known as adult neurogenesis ( ). Neurogenesis occurs throughout life in the ventricular-subventricular zone (V-SVZ) of the lateral ventricles and the subgranular zone (SGZ) of the hippocampal dentate gyrus (DG) ( ). Neurogenesis plays a critical role in neuroplasticity, brain homeostasis, and central nervous system (CNS) maintenance. It is essential to maintaining cognitive function and repairing damaged brain cells affected by aging and brain disease ( ). Adult hippocampal neurogenesis directly impacts cognitive function since hippocampal formation is closely linked to the storage and processing of memory ( ; ). In recent years, evidence has accumulated that neurogenesis can restore a more youthful state during aging. In addition, increased adult neurogenesis contributes to a variety of human diseases, including cognitive impairment and neurodegenerative diseases ( ). The appearance of neurodegenerative diseases (including Alzheimer’s and Parkinson’s) increases exponentially with age ( ), so aging is considered to be the most critical risk factor for almost all neurodegenerative diseases ( ).
Neurogenic inflammation is triggered by neural activation, resulting in neuropeptide release, rapid plasma extravasation and edema, leading to conditions such as headaches. Neuroinflammation is a local inflammation of the peripheral nervous system (PNS) and CNS ( ). Neuroinflammation has been shown to alter neurogenesis in adults. Various inflammatory components, such as immune cells, cytokines, or chemokines, regulate neural stem cells’ survival, proliferation, and maturation ( ). During normal brain aging, increased inflammatory activity is caused by the activation of glial cells ( ). It has been shown that mesenchymal stem cells (MSCs) can stimulate neurogenesis and angiogenesis and delay neuronal cell death ( ). At the same time, their secreted exosomes are smaller in size and cause less immune response in the body, which is a hot topic of current research ( ).
In the aging brain, the number of neuronal cells is significantly reduced, the levels of inflammatory factors IL-1β, IL-6, and TNF-α are increased leading to neuroinflammation, and the levels of reactive oxygen species (ROS) are increased causing oxidative stress in the brain. After treatment with exosomes secreted by mesenchymal stem cell-extracellular vesicle (MSC-EV), the number of neuronal cells increased, the levels of inflammatory factors IL-1β, IL-6, and TNF-α decreased, and the levels of ROS decreased thereby reducing oxidative stress in the brain.
## Mechanisms and manifestations of brain aging
Cellular senescence is an important factor in tissue deterioration and the accumulation of senescent cells is considered a hallmark of and a pathological cause of aging ( ). Among the organelles most closely related to senescence is the nucleus ( ), mitochondria ( ), and lysosomes ( ; ). The core is mainly involved in the cell cycle, telomere, and epigenomic changes ( ). A new study finds that age-related epigenetic changes can be reversed by interventions ( ); Mitochondria are mainly involved in oxidative stress due to the increase of reactive oxygen species (ROS) and mutations in mitochondrial DNA (mtDNA), inflammation, and apoptosis ( ), which are important factors that induce the onset of aging; In lysosomes, it was found that lysosomes and lysosome-related organelles play an important role in the regulation of aging and longevity ( ), which is mainly associated with autophagy ( ); in the cytoplasmic matrix and extracellular, etc., are primarily involved in signaling pathways related to inflammation and fibrosis ( ), such as who stated that cardiac fibrosis is usually one of the hallmarks of cardiac aging. These signaling pathways release inflammatory factors and chemokines that contribute to the deterioration of the senescent cells’ microenvironment, which transmits aging signals and affects the transformation of surrounding healthy cells into senescent cells.
The brain is the most complex and vital human organ ( ; ; ), consuming more energy than any other tissue in proportion to its size. Microstructural degeneration of the gray and white matter in the human brain during aging leads to tissue softening and tissue atrophy ( ). The rate of brain atrophy during aging can predict whether someone will develop cognitive impairment and dementia, and analysis of cross-sectional histological sections suggests that atrophy is the combined result of dendritic regression and neuronal death ( ). Some scholars have used magnetic resonance imaging (MRI) to find that the frontal, parietal, and temporal lobes decrease with age ( ) while the frontal, parietal, and temporal lobes control language, memory, auditory ( ), motor ( ), and attention functions of the human brain. Initially, these aging mechanisms occur mainly at the cellular level due to slowed metabolic activity and ischemia, such as inflammation, mitochondrial dysfunction ( ), oxidative stress ( ), and calcium dysregulation ( ), but then gradually manifest themselves in tissue and eventually organ-level changes in brain shape ( ). In addition, some environmental factors can affect the rate of structural changes in the brain during aging. For example, adequate aerobic exercise increases hippocampal volume, effectively improving memory function ( ); overweight and obesity can lead to hippocampal atrophy and affect brain health ( ).
## Mesenchymal stem cell and exosome properties
Mesenchymal stem cells, officially named 29 years ago, represent a class of cells in the human and mammalian bone marrow (BM) and periosteum that can be isolated and expanded in culture while maintaining their ability to be induced to form a variety of different cells in vitro ( ). MSCs have a solid proliferative capacity ( ) and can self-renew and differentiate into tissue-specific cells (e.g., osteoblasts, chondrocytes, and adipocytes), and therefore have great potential in regenerative medicine ( ; ). In addition to its pluripotency, MSC has immunomodulatory properties and has been investigated as a potential treatment for various immune diseases ( ). MSCs influence most immune effector cells through direct contact with immune cells and local micro-environmental factors. According to studies, the immunomodulatory effects of MSC are mainly delivered through cytokines secreted by MSC. However, apoptotic and metabolically inactivated MSCs have recently shown immunomodulatory potential, with regulatory T cells and monocytes playing a pivotal role ( ). Secondly, since MSCs do not express significant histocompatibility complexes and immunostimulatory molecules, they are not detected by immune surveillance and do not cause graft rejection after transplantation, which is a significant breakthrough point for regenerative medicine ( ). Various animal models (myocardial infarction mice, burned mice, and diabetic mice) and clinical trials have shown that MSCs show good results in repairing damaged tissues ( ; ; ). MSCs have the homing ability, which means they can migrate to the site of injury and secrete some growth factors, cytokines, and chemokines that are beneficial for tissue repair ( ; ). Many experimental studies in ischemic stroke have shown that MSCs are able to modulate immune responses and play a neuroprotective role by stimulating neurogenesis, oligodendrogenesis, astrogliogenesis, and angiogenesis. MSCs may also have the ability to replace damaged cells, but paracrine factors released directly into the environment or via extracellular vesicles (EVs) appear to play the most significant role ( ).
Exosomes are nanoscale vesicles (30–150 nm in diameter) secreted by most cells ( ). They are surrounded by a lipid bilayer and carry a variety of biomolecules, including proteins, lipids, metabolites, RNA, and DNA. When exosomes are taken up by other cells, these exosomes are transferred and affect the phenotype of the recipient cells. Exosomes play a crucial role in bioactive molecule transport, immune response, antigen presentation, protein regulation, cellular homeostasis, and extracellular matrix remodeling ( ). Thus, exosomes are considered to be an essential mediator of intercellular communication ( ). MSC-derived exosomes contain cytokines, growth factors, lipids, and messenger RNA (mRNA) and regulate microRNAs (miRNAs) function ( ). Exosomes mainly act on the organism in a paracrine manner ( ), and it has been established that the mode of action of the therapeutic effect of stem cells is mainly paracrine mediated by stem cell secretory factors ( ), so it is presumed that exosomes primarily act during stem cell therapy. Exosomes have a relative therapeutic effect on a variety of diseases. In a mouse model of acute kidney injury (AKI), MSC-derived exosomes (MSC-Exo) accumulated mainly in inflamed kidneys, whereas in a brain hemorrhage model, MSC-Exo was detected in the injured brain ( ). Two proteins commonly found in exosomes, CD81 and tumor susceptibility gene 101 (TSG101), have been confirmed by Western blot ( ; ). Exosome therapy has shown similar therapeutic effects to direct MSC transplantation without causing multiple adverse outcomes. The complexity of the integrated function of its contents improves the therapeutic effect of MSC-Exo ( ). Multiple studies have found that after intravenous administration of exosomes, they are predominantly distributed in vascular-rich organs and organs associated with the reticuloendothelial system, such as the liver, lungs, spleen and kidneys ( ). In addition, found that DIR (lipophilic, near-infrared fluorescent anthocyanine dye)-labeled exosomes could be detected in mouse brain after intravenous injection of exosomes by near-infrared fluorescence (NIRF).
## Therapeutic effects of mesenchymal stem cells and exosomes on brain aging
### Mesenchymal stem cells repair neuronal cells in the hippocampus to slow brain aging
Aging is a natural process; the most obvious outward manifestation accompanying brain aging is a decline in cognitive function. The structure to consider for cognitive decline in the brain is necessarily the hippocampus-a brain region known to play an essential role in learning and memory consolidation as well as in affective behavior and emotion regulation and whose functional and structural plasticity [e.g., neurogenesis ( )] occurs in adulthood. Neurobiological changes seen in the aging hippocampus, including increased oxidative stress and neuroinflammation, altered intracellular signaling and gene expression, and reduced neurogenesis and synaptic plasticity, are thought to be associated with age-related declines in cognitive function ( ). In animal experiments, the Morris water maze is usually used to test the learning memory ability of mice. In contrast, after intracerebroventricular injection of MSC of human BM origin, the aged MSC-treated group showed significant improvements in spatial memory accuracy and prolonged persistence in single- and three-hole target areas as demonstrated in the Morris water maze compared with the aged control group. MSC treatment increased the number of neuroblasts in the hippocampal DG, decreased the number of reactive microglia, and restored presynaptic protein levels compared to older controls. And after MSC transplantation, MSCs mainly migrated to the DG, CA1, and CA3 regions of the hippocampus. Cognitive deficits are associated with altered levels of several neurological factors, such as brain-derived neurotrophic factor (BDNF), nerve growth factor (NGF), and glial cell-derived neurotrophic factor (GDNF) ( ). In contrast, human MSCs express a variety of neuromodulators that promote neuronal survival and neurogenesis ( ). We can see this result in the hematoxylin–eosin (HE) pathology and Nissl staining of the MSC-treated and senescent control groups. This experiment concluded that intracerebroventricular injection of human bone marrow-derived mesenchymal stem cells (hBM-MSCs) was effective in improving spatial memory in aged rats and that the treatment improved some of the functional and morphological brain characteristics that are typically altered in aging rats ( ).
### Mesenchymal stem cells slow brain aging by promoting angiogenesis
discussed that the neuroprotective effect was mainly attributed to soluble factors secreted by stem cells. Furthermore, it has been shown that stem cells can form vascular structures and secrete pro-angiogenic factors in vitro , positively influencing the growth of blood vessels in vitro and in vivo ( ). For example, cerebral ischemia is the most common disease in the elderly ( ). Age is the main unmodifiable risk factor for cerebral ischemia. When located in the inflammatory microenvironment in vivo , MSC can release a variety of angiogenic and neurotrophic factors as well as anti-inflammatory molecules; in addition, MSC appears to have an excellent homing ability when administered by systemic routes. Several studies have shown that bone marrow-derived stem cell transplantation in the peripheral circulation improves neurological function and reduces infarct volume. Cellular therapies using MSCs can enhance endogenous repair mechanisms in the damaged brain by supporting the processes of neoangiogenesis, neurogenesis, and neural reorganization. The mechanism by which MSCs improve infarcted brain tissue appears to be more related to the ability of MSCs to release neuroprotective factors (a paracrine mechanism) than to their ability to replace ( ).
### Mesenchymal stem cells and its secreted exosomes slow brain aging by suppressing the expression of pro-inflammatory factors
Aging is characterized by developing a persistent pro-inflammatory response ( ), and the aging brain is also susceptible to inflammation ( ). Yet, inhibiting of pro-inflammatory factor expression alleviates cognitive impairment in the brain ( ). Microglia are resident immune cells of the CNS and play a key role in maintaining brain homeostasis ( ). In the aging brain and neurodegeneration, microglia lose their homeostatic molecular signature and can promote increased production of pro-inflammatory cytokines ( ). Studies have demonstrated that reactive microglia infiltrate the hippocampus in aging rats and cause it to exhibit an inflammatory state ( ). MSC can maintain the resting phenotype of microglia or control microglia activation by producing multiple factors ( ), thereby controlling the inflammatory response and delaying brain aging. Studies have shown that exosomes obtained from MSC secretion also have anti-inflammatory effects. It regulates the brain infiltration of leukocytes and thus protects the nerves ( ). Inflammatory responses have long been associated with neurodegenerative processes. And TNF-α, IL-1β, and IL-6, cytokines that support inflammation, are significantly increased in the aging brain. found that the pro-inflammatory regulators TNF-α, IL-1β, and IL-6 were significantly reduced in the brains of aging mice after MSC-Exo treatment. In the Morris water maze test, the MSC-Exo-treated group significantly reduced escape latency from 3 days after acquisition training, and learning and memory abilities were significantly improved in mice treated with MSC-Exo. Thus, MSC-derived exosomes can rescue memory deficits by modulating the inflammatory response ( ). Data suggest that MSC-derived exosomes can enter microglia and inhibit their activation to shift them back to function, thereby suppressing inflammation and promoting recovery of brain function ( ).
### Mesenchymal stem cells and its secreted exosomes slow brain aging by promoting neurogenesis
Recent studies have shown the neurogenesis in the adult brain ( ). In addition, the treatment of MSCs has been shown to stimulate neurogenesis in the rat brain and is proposed to be implemented in a number of neurodegenerative diseases ( ; ). Some scientists have also used exosomes secreted by MSCs for treatment and found that exosomes promote neurogenesis in the subventricular zone (SVZ) and DG of the hippocampus and reduce cognitive impairment associated with Parkinson’s disease, stroke, and traumatic brain injury ( ; ). From the preceding, it is known that MSC-derived exosomes disrupt the polarization of M1 microglia and trigger their transition to the M2 phenotype, thereby significantly reducing inflammation. In addition, exosome treatment is neuroprotective against oxidative stress and also expands neuronal nerve density ( ). Several studies provide evidence that exosomes interact with neurogenic ecotopes through miRNA transfer to neural precursor cells, triggering neural remodeling events, neurogenesis, angiogenesis, and synaptogenesis ( ). Some scientists have also modified exosomes with miRNAs. For example, found that elevated miR-26a enhanced axonal growth in hippocampal neurons and axonal regeneration in the PNS, and then they overexpressed miR-26a in exosomes and found that it could activate the mammalian target of rapamycin (mTOR) pathway to enhance axonal growth and renewal in the nervous system, thus promoting neurogenesis ( ).
Application of mesenchymal stem cell (MSC) and exosome in brain aging.
## Discussion
Based on the data collected, it is known that brain aging subsequently increases the incidence of neurodegenerative diseases, which seriously affect the quality of life of the elderly. Scientists have been seeking better ways to slow down aging, and MSCs and their derived exosome therapies are emerging promising strategies for treating various diseases. In recent years, there has been increasing research interest in exosomes, their innate ability to transport genetic material, protect it from cellular degeneration, and deliver it to recipient cells in a highly selective manner, suggesting that MSC-derived exosomes are an ideal delivery system for small molecules and a means of gene therapy for cancer treatment and potentially regenerative drugs. In addition, encouraging preclinical data suggest that MSC-derived exosome therapies may be superior to cell-based therapies in terms of safety and versatility. Today, the preparation of MSCs has become proficient and the extraction of exosomes is being refined. However, the technology for purification of exosomes after extraction still needs to be addressed. Furthermore, the progression of exosomes on tumors has been widely reported in the last decade or so. For example, MSC-derived exosomes from bone marrow (BM MSCs) stimulate the hedgehog signaling pathway in osteosarcoma and gastric cancer cell lines, thereby promoting tumor growth ( ). Therefore, the translation of therapies from the laboratory to the clinic requires a clear understanding of component characterization, immune response, etc., in order to optimize their clinical application.
## Author contributions
JJ and XW contributed to the conception and design. XZ wrote the manuscript and figure. XH collected the data and designed the figure. LT and ZZ performed literature search and provided valuable comments. All authors contributed to the article and approved the final manuscript.
|
Vascular alterations often overlap with neurodegeneration, resulting in mixed forms of dementia (MD) that are hard to differentiate from Alzheimer’s Disease (AD). The 26 bp intergenic polymorphism of VAMP2 , a key component of SNARE complex, as well as its mRNA and protein levels are associated with neurological diseases. We evaluated ApoE4 and VAMP2 26 bp Ins/Del genotype distribution in 177 AD, 132 MD, 115 Mild Cognitive Impairment (MCI) and 250 individuals without cognitive decline (CT), as well as VAMP2 gene expression in a subset of 73 AD, 122 MD, 103 MCI and 140 CT. Forty-two MCI evolved to AD (22 MCI-AD) or MD (20 MCI-MD) over time. VAMP2 mRNA was higher in MD compared to AD ( p = 0.0013) and CT ( p = 0.0017), and in MCI-MD compared to MCI-AD ( p < 0.001) after correcting for age, gender, MMSE and ApoE4 +/− covariates ( p = 0.004). A higher VAMP2 expression was observed in subjects carrying the minor allele Del compared to those carrying the Ins/Ins genotype ( p = 0.012). Finally, Del/Del genotype was more frequently carried by MD/MCI-MD compared to CT ( p = 0.036). These results suggest that VAMP2 mRNA expression can discriminate mixed form of dementia from AD, possibly being a biomarker of AD evolution in MCI patients.
## Introduction
Dementia is a common condition of the elderly characterized by multiple cognitive impairments leading to high disability. This clinical condition results from heterogeneous neurodegenerative and/or cerebrovascular diseases characterized by selective neuronal loss and by intra- or extracellular deposition of proteins ( ) and/or by blood vessel blockage leading to dead tissue or bleeding area in the brain ( ). Neuropathological classification depends on the anatomical distribution of neuronal loss and cellular distribution of specific aggregated proteins ( ).
Between the cognitive changes of normal aging and those associated with dementia there is an intermediate stage called Mild Cognitive Impairment (MCI), which is characterized by memory deficits that do not impair global cognitive functions or daily lifestyle and activities ( ). Neuropathological studies showed that synaptic loss is already evident in MCI ( ) and MCI subjects exhibit loss of pre- and postsynaptic proteins ( ). Unfortunately, MCI is often the initial step toward development of different kinds of dementia, including Alzheimer’s Disease (AD), Vascular dementia (VAD), Dementia with Lewy Body (DLB) and Frontotemporal Dementia (FTD) ( ).
In older people vascular alteration often overlap with neurodegenerative pathology ( ; ), making the diagnosis process hard. For this reason, it would be of great importance to find out biomarkers able to predict how MCI condition can develop on the different kind of dementias. The release of neurotransmitters in the synaptic cleft, through the fusion of synaptic vesicles with the presynaptic membrane, is the basis of neuronal communication ( ). To this regard the N -ethylmaleimide-sensitive factor attachment protein receptor (SNARE) complex, a fusion machinery that provides the energy required for the synaptic vesicles fusion with the presynaptic membrane, plays a crucial role ( ) and defects in this machinery are associated with a range of neurological disorders associated to dementia ( ; ; ). The SNARE complex includes three specific proteins: the Syntaxin1a (STX1a), the synaptosomal-associated protein of 25 kDa (SNAP-25) and the vesicle associated membrane protein 2 (VAMP2). VAMP2 plays a fundamental role in the SNARE complex formation and functionality. The VAMP2 gene is located on chromosome 17p13.1 and consists of 6 exons that encode for the SNARE complex key protein required for membrane fusion. A VAMP2 gene 26 bp Ins/Del polymorphism, located in a intergenic region, suspected to be involved in gene expression, distant 2 kb from the 3′ flanking region of VAMP2 was associated with neurological disease like Attention-deficit hyperactivity disorder and with neurodegenerative disease like Multiple Sclerosis ( ; ; ). Notably, a reduction of VAMP2 expression in the hippocampus and the entorhinal cortex was found to be associated with cognitive decline ( ) and AD brains have a reduced level of VAMP2, STX1a and SNAP-25 compared to control brains ( ).
Considering the crucial role of VAMP2 in the SNARE complex assembly, we evaluated whether VAMP2 could be a novel peripheral biomarker that can distinguish among different forms of cognitive impairment in old persons. For this purpose, VAMP2 gene variant 26 bp Ins/Del and VAMP2 mRNA expression in human peripheral blood mononuclear cells (PBMCs) were analyzed in patients with a diagnosis of either MCI, AD, or mixed dementia (MD).
## Materials and Methods
### Patients and Controls
The cohort consisted of 674 community-dwelling adults aged 63–100 years enrolled at Geriatric Unit of the Fondazione IRCCS Ca’ Granda Ospedale Maggiore Policlinico, Milan, Italy. All participants were admitted to the Geriatric Unit for the investigation of a suspect cognitive decline and their data recorded in Registro di Raccolta Dati della Unità di Geriatria (REGE). One hundred and seventy-seven individuals suffered from AD and 132 from MD. One hundred and fifteen had a diagnosis of MCI; 42 out of them evolved in either AD ( N = 22: MCI-AD) or MD ( N = 20: MCI-MD) over 2 years of the study period. Two hundred and fifty sex-matched subjects without cognitive decline (CT) were enrolled in the study as well.
Information about medical history, physical examination, and neurocognitive assessment (i.e., Mini Mental State Examination-MMSE, Clock Drawing Test-CDT), neuropsychological tests-NPS (i.e., Trail Making Test, verbal fluency test, digit span forward and backward tests, verbal learning tests, Token Test, Rey’s figure copy and delayed recall, Raven Colored Progressive Matrices), were recorded for all participants. The diagnosis of AD was made according to the criteria by , while MCI patients met the criteria outlined by . Individuals showing the presence of both neurodegenerative and vascular findings were diagnosed as MD ( ). CT were subjects with an MMSE score ≥ 27, the neuropsychological (NPS) tests negative for dementia and no neurological or psychiatric disorders.
The study was conducted in accordance with the declaration of Helsinki and the research protocol was approved by the Ethical Committee of the IRCCS Fondazione Ca’ Granda Ospedale Maggiore Policlinico (protocol number: 00223248). Informed consent to participate in the study has been signed by all participants.
### VAMP2 and ApoE Genotyping
Genomic DNA was extracted using a previously described salting-out method ( ) and its concentration and purity were determined by the means of spectrophotometric analysis. The VAMP2 gene 26 bp Ins/Del polymorphism was genotyped by polymerase chain reaction (PCR) with a VeritiPro Thermal Cycler (Applied Biosystems by Life Technologies, Foster City, CA, United States), using VAMP2 F-5′-ACAAAGTGCGCCTTATACGC-3′ and VAMP2 R-5′-GATTTTCCTTGACGACACTC-3′ primers as previously described ( ). Amplicons (10 μL) were detected by electrophoresis on a 3% agarose gel. Finally, the ApoE genotype was determined as previously described ( ). In brief, a 244 bp ApoE fragment was amplified by PCR (step 1: 96 °C for 5 min; step 2 for 30 cycles: 95 °C for 1 min, 60 °C for 1.10 min, and 70 °C for 2 min; step 3: 70 °C for 10 min) with the primer pairs: 5′-GATCAAGCTTCCAATCACAGGCAGGAAG-3′ and 5′-GATCCGGCCGCACACGTCCTCCATG-3′. The amplified fragment was digested by using the Hha I enzyme and the products were visualized on agarose gel.
### VAMP2 Gene Expression
Peripheral blood mononuclear cells (PBMCs) were isolated by density gradient (Lympholyte-H, Cedarlane, Canada). Total RNA was extracted from PBMCs using the Chomczynski and Sacchi’s modified method ( ). Two micrograms of total RNA were reverse-transcribed using the SuperScript VILOTM cDNA Synthesis Kit (Invitrogen by Thermo Fisher Scientific, Massachusetts, United States). Quantitative PCRs were performed in the OpenArray system QuantStudio 12K Flex Real-Time PCR System (Applied Biosystems by Thermo Fisher Scientific, Massachusetts, United States). For the VAMP2 gene expression analysis, qPCR reactions based on TaqMan probes Hs00360269_m1 (Thermo Fisher Scientific) were performed using high-performance OpenArray chip. Three genes have been selected as endogenous according to their stable expression in human cells ( GAPDH , ACTB and 18S ) and included into the OpenArray chip. One hundred and twenty ng of every cDNA sample (1.2 μL of each) were mixed with 1.3 μL of PCR-grade water and 2.5 μL of TaqMan™ OpenArray Real-Time PCR Master Mix (Applied Biosystems by Thermo Fisher Scientific, Massachusetts, United States). Samples were loaded in duplicate into Open-Array plates. For gene expression analysis, Ct values were obtained using the Thermo Fisher ConnectTM (Thermofisher) online application, and the Relative Quantification (RQ) software.
### Statistical Analysis
Pearson’s chi-square test was performed to compare categorical variables, to exclude any deviation of SNP genotype distribution from Hardy–Weinberg equilibrium and to evaluate case-control differences of SNP distribution. The distribution of demographic and clinical parameters was evaluated by Kolmogorov-Smirnov test to assess possible deviations from the Gaussian model; parametric data were reported as mean ± standard deviation, whereas non-parametric data were shown as median and interquartile range (IQR: 25th–75th percentile). Regarding the continuous values, parametric data were analyzed using the Analysis of Variance (ANOVA) and Student’s t -test, whereas non-parametric data were analyzed using the Kruskal–Wallis test and the Mann–Whitney test. A binomial logistic regression model (with forward stepwise selection) was computed, considering pathology status as response variable (MCI-AD vs. MCI-MD) and VAMP2 expression as explanatory variable and inserting age, gender, MMSE and ApoE4 positivity as covariates. p -values corresponding to < 0.05 were described as statistically significant in the text. Corrected p -values (p ) are reported for statistical analysis corrected for covariates. Data of VAMP2 mRNA expression was used in Receiver operating characteristic (ROC) analysis to draw the ROC curve and to calculate the area under the curve (AUC). Statistical analyses were accomplished using commercial software (IBM SPSS Statistic 26.0, IBM Inc., Chicago, IL, United States).
## Results
### Characteristics of the Cohort
Demographic and clinical characteristics of the cohort are summarized in . Sex distribution was similar in all the groups examined. Age was different among the four groups ( p < 0.001) with CT and MD subjects being older than MCI and AD patients ( p = 0.03 and p < 0.001 for both, respectively). As expected, the MMSE score was different among the four groups ( p < 0.001) with AD and MD patients having a significant lower MMSE score compared to CT and MCI ( p < 0.001 for all the comparisons), as well as for the MCI compared to CT ( p < 0.001). The distribution of the ApoE4 variant showed a significant difference among the four groups ( p < 0.001), with the ApoE4 allele being more frequent in AD compared to CT, MCI and MD patients (AD vs. CT: p < 0.001; OR: 3.76; 95% CI: 2.43–5.89; AD vs. MCI: p < 0.001; OR: 2.57; 95% CI: 1.54–4.33; AD vs. MD: p < 0.001; OR: 2.20; 95% CI: 1.34–3.64). Similarly, the ApoE4 variant was significantly more frequent in MD patients compared to CT ( p = 0.043; OR: 1.71; 95% CI: 1.02–2.86). Forty-two of the 115 MCI with known clinical evolution in dementia were categorized according to disease progression: 22 evolved to AD (MCI-AD) and 20 to MD (MCI-MD). Demographic and clinical characteristics of these two sub-groups are summarized in . Sex distribution, MMSE and ApoE4 distribution were comparable between the groups. MCI-MD were older than the MCI-AD patients ( p = 0.021) ( ).
Demographic and clinical characteristics of the cohort.
Demographic and clinical characteristics of the MCI evolved in dementia.
### VAMP2 mRNA Expression According to VAMP2 26 bp Ins/Del Polymorphism
The VAMP2 mRNA expression level was evaluated according to the VAMP2 26 bp Ins/Del variant genotypes and minor allele Del (Ins/Del + Del/Del) categorization. The VAMP2 mRNA expression level were obtained from PBMCs of 435 subjects (139 CT and 296 cognitively impaired subjects with VAMP2 26 bp polymorphism genotyping available data) out of the 674 enrolled in the cohort. For the remaining 239 excluded subjects the PBMCs sample was exhausted in our repository or the VAMP2 26 bp polymorphism genotyping was missing.
VAMP2 mRNA expression was significantly different among the Ins/Ins (n: 314; median: 0.67: IQR: 0.50–0.84), Ins/Del (n: 109; 0.70: 0.55–0.93) and Del/Del (n: 12; 0.80: 0.53–1.07) genotypes ( p = 0.043). In particular, VAMP2 expression was significantly increased in Ins/Del compared to Ins/Ins genotype ( p = 0.017) ( ); VAMP2 mRNA expression was also consistently increased in Del/Del compared to Ins/Del and Ins/Ins carriers; these differences approached but did not reach statistical significance. VAMP2 expression was evaluated next according to VAMP2 minor allele Del categorization. Results showed the presence of a significant difference between Ins/Ins (n: 314; 0.67: 0.50–0.84) and minor allele Del (n: 121; 0.70: 0.55–0.94) ( p = 0.012) ( ).
VAMP2 mRNA expression according to VAMP2 26 bp Ins/Del polymorphism (A) and to VAMP2 26 bp minor allele Del (Ins/Del + Del/Del) (B) .
### VAMP2 mRNA Expression Level According to Cognitive Impairment
The VAMP2 mRNA expression level obtained from available PBMCs of 438 subjects out of the 674 enrolled were next evaluated according to cognitive impairment (73 AD, 122 MD, 140 CT and 103 MCI of which 18 MCI-MD and 19 MCI-AD). For the remaining 236 subjects (104 AD, 10 MD, 110 CT and 12 MCI of which 2 MCI-MD and 3 MCI-AD) the PBMCs sample was exhausted in our repository.
VAMP2 mRNA expression was significantly different when AD (n: 73; 0.63: 0.50–0.73), MD (n: 122; 0.71: 0.55–0.95), MCI (n: 103; 0.75: 0.56–0.94) and CT (n: 140; 0.63: 0.47–0.81) individuals were compared ( p < 0.001) ( ). In particular, VAMP2 expression was significantly increased in MCI compared to AD ( p < 0.001) and CT ( p = 0.001) and in MD compared to AD patients ( p = 0.0013) and CT ( p = 0.0017) ( ). To note, no difference in VAMP2 mRNA expression was found in CT compared to AD patients ( p = 0.64), or between MCI and MD patients ( p = 0.78).
VAMP2 expression according to diseases (A) and according to the MCI evolution (B) . Global significance: p < 0.001 (A) . AD, Alzheimer’s Disease; MD, mixed dementia; CT, subject without cognitive decline; MCI, Mild Cognitive Impairment; MCI-AD, MCI individuals evolved to AD; MCI-MD, MCI individuals evolved to mixed dementia.
Considering MCI subjects with known clinical evolution, at recruitment, when all these individuals had a diagnosis of MCI, VAMP2 expression was significantly augmented in MCI that converted to MD (n: 18) compared to those converted to AD (n: 19) (0.87: 0.74–1.00 vs. 0.67: 0.42–0.74, p = 0.001) ( ). Binomial logistic regression analysis confirmed that VAMP2 gene expression ( p = 0.004) was significantly increased in MCI-MD compared to MCI-AD after correction for age, gender, MMSE and ApoE4 positivity/negativity.
### Receiver Operating Characteristic Curve Analysis
Receiver operating characteristic curve analysis was employed to evaluate the ability of VAMP2 gene expression to discriminate between clinically different forms of dementia. Eighteen MCI-MD, 19 MCI-AD and 140 CT subjects were considered in the analyses. The total area under the curve (AUC) of MCI-AD vs. CT was 55% (95% CI: 0.47–0.63) and of MCI-MD vs. CT was 77% (95% CI: 0.70–0.83; sensitivity: 77.8%, 95% CI: 52.4–93.6; specificity: 65.7%, 95% CI: 57.2–73.5) ( ). Finally, the AUC of MCI-MD vs. MCI-AD was 82% (95% CI: 0.66–0.93; sensitivity: 77.8%, 95% CI: 52.4–93.6; specificity: 73.7%, 95% CI: 48.8–90.9) ( ).
The ROCs curve of VAMP2 gene expression for MCI-AD vs. CT, MCI-MD vs. CT and MCI-MD vs. MCI-AD evaluation. ROC, receiver operating characteristic; AUC, area under the curve.
### VAMP2 26 bp Ins/Del Polymorphism Distribution in Alzheimer’s Disease, Mixed Forms of Dementia and Subject Without Cognitive Decline
Molecular genotyping of VAMP2 26 bp Ins/Del polymorphism was performed in all the subjects; distribution analysis was done grouping AD patients together with MCI subjects which developed AD (AD/MCI-AD) and MD patients together with MCI subjects that developed MD (MD/MCI-MD). The 73 MCI with unknown clinical evolution were excluded from this analysis.
VAMP2 26 bp Ins/Del polymorphism distribution was in Hardy–Weinberg equilibrium in all the groups ( p > 0.05), but VAMP2 26 bp Ins/Del polymorphism genotypes distribution was significantly different among the three analyzed groups ( p = 0.032; d.f.: 4; χ = 10.56). Thus, results showed that the Del/Del homozygote genotype was more frequently present in MD/MCI-MD (4.6%) compared to AD/MCI-AD and CT (1 and 0.8%, respectively) ( ). In particular, a significant different genotype distribution was seen in MD/MCI-MD compared to CT ( p = 0.036; d.f.: 2; χ = 6.65) and a trend of significance in MD/MCI-MD compared to AD/MCI-AD ( p = 0.096; d.f.: 2; χ = 4.67). Finally, although VAMP2 26 bp Ins/Del allelic distribution analysis did not show the presence of significant differences between the three analyzed groups ( p = 0.13; d.f.: 2; χ = 4.06), the Del allele was more frequent in MD/MCI-MD compared to CT showing a trend of significance ( p = 0.05; OR: 1.51; 95% CI: 0.99–2.28).
Genotypes and allelic distribution of VAMP2 26 bp Ins/Del polymorphism in AD, MD patients and CT subjects.
## Discussion
In the present paper we studied VAMP2 gene expression and VAMP2 26 bp polymorphism distribution in a cohort of elderly persons with different forms of cognitive impairment and dementia. Results showed that VAMP2 mRNA is more expressed in PBMCs from MD patients compared to AD; notably, higher levels of VAMP2 expression seems to occur even in the first stages of disease and, in particular, in those MCI patients who converted to MD compared to those who converted to AD. These results were also reinforced by the ROC analysis, which confirmed the specific discriminatory power of VAMP2 expression level between MCI-MD and MCI-AD.
Additionally, results indicated that VAMP2 mRNA expression is influenced by VAMP2 genotype. Thus, subjects carrying the VAMP2 26 bp minor allele Del (Ins/Del + Del/Del) were characterized by higher VAMP2 mRNA expression compared to those carrying the Ins/Ins genotype. How the 26 bp Ins/Del variant could influence the VAMP2 expression is still unknown. It is important to note that several studies showed that Ins/Del intergenic variants could lead to higher gene transcription by influencing the affinity of the transcription machinery for the transcription binding site ( ; ; ). Interestingly, the genotype distribution showed that the VAMP2 26 bp Ins/Del polymorphism associates with MD and with MCI-MD. In particular, the frequency of individuals carrying the VAMP2 26 bp Del/Del genotype was significantly increased in MD patients compared to AD and CT subjects. This higher frequency of VAMP2 26 bp Del among the MD patients may account for the higher VAMP2 mRNA expression seen in mixed form of dementia. Although being preliminary, these results strongly suggest that VAMP2 mRNA expression is a possible early indicator able to distinguish MCI evolution to MD or AD as reinforced by the ROC evaluation and reinforce the hypothesis that VAMP2 is a crucial gene in mixed form of dementia, but not in AD ( ).
VAMP2 is a vesicle SNARE protein that plays a crucial role in neural communication and plasticity together with the target SNARE proteins STX1a and SNAP-25, through the formation of the SNARE complex. Impairments in SNARE complex assembly, changes in SNARE protein expression, and genetic variants of SNARE protein are widely observed in human as well as in mouse models of neurological diseases, and these alterations might be related to disease pathology and progression ( ; ; ). Defects in this machinery can result in a range of neurological disorders associated to dementia ( ; ). In particular, a reduction in the SNARE complex efficiency led to a lower neurotransmitters release inducing neurodegeneration ( ). In particular, a reduction in presynaptic proteins was found in the temporal cortex of VAD patients, even if this reduction was not nearly as important as what has been described in AD ( ). Dysfunction in SNARE complex was also indicated as the initial trigger of accumulation of α-Synuclein ( ) in DLB, and a recent study showed the physical interaction of α-Synuclein with VAMP2 to promote SNARE assembly ( ) in this condition. Finally, a reduction of Synaptophysin, a presynaptic protein that binds VAMP2 and plays a role in SNARE complex assembly and vesicle fusion, was observed to be present in FTD patients ( ), suggesting a fundamental role of synaptic decline in dementia ( ). VAMP2 involvement, both at mRNA and protein levels, in the pathogenesis of VAD and in different forms of well-defined dementia, indeed suggest a possible role for this gene in the vascular alteration seen in the mixed form of dementia, that will need to be further investigated. Moreover, our data need further studies confirming levels of VAMP2 in central nervous system (brain tissue or liquor). Anyway, PBMCs may serve as a peripheral laboratory to find molecular signatures that could aid in differential diagnosis with other forms of dementia and in monitoring of disease progression ( ). The number of patients we analyzed is relatively small, in particular regarding the MCI subjects with a clinically defined evolution, and as a consequence these findings will need to be confirmed in larger cohorts of patients. Longitudinal studies and analyses focusing on the impact of the 26 bp Ins/Del variant on VAMP2 gene and protein expression will also shed more light on the role of VAMP2 in the pathogenesis of dementia.
These caveats notwithstanding, the identification of easy-to-collect biomarker such as VAMP2 able to discriminate between different forms of mixed form of dementia could be useful to allow early therapeutic and/or rehabilitative intervention, with beneficial efforts for the patients, caregivers and the public health system.
## Data Availability Statement
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author/s.
## Ethics Statement
The studies involving human participants were reviewed and approved by Ethical Committee of the IRCCS Fondazione Ca’ Granda Ospedale Maggiore Policlinico (protocol number: 00223248). The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
AC, FG, and BA conceived and designed the project. AC, EF, and PR assisted with biological samples and data collection. AC and EF performed the experiments and data analysis. MC supervised the project. AC and FG wrote the manuscript. All authors read and approved the final manuscript.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
|
Tau is a microtubule-associated protein known to bind and promote assembly of microtubules in neurons under physiological conditions. However, under pathological conditions, aggregation of hyperphosphorylated tau causes neuronal toxicity, neurodegeneration, and resulting tauopathies like Alzheimer’s disease (AD). Clinically, patients with tauopathies present with either dementia, movement disorders, or a combination of both. The deposition of hyperphosphorylated tau in the brain is also associated with epilepsy and network hyperexcitability in a variety of neurological diseases. Furthermore, pharmacological and genetic targeting of tau-based mechanisms can have anti-seizure effects. Suppressing tau phosphorylation decreases seizure activity in acquired epilepsy models while reducing or ablating tau attenuates network hyperexcitability in both Alzheimer’s and epilepsy models. However, it remains unclear whether tauopathy and epilepsy comorbidities are mediated by convergent mechanisms occurring upstream of epileptogenesis and tau aggregation, by feedforward mechanisms between the two, or simply by coincident processes. In this review, we investigate the relationship between tauopathies and seizure disorders, including temporal lobe epilepsy (TLE), post-traumatic epilepsy (PTE), autism spectrum disorder (ASD), Dravet syndrome, Nodding syndrome, Niemann-Pick type C disease (NPC), Lafora disease, focal cortical dysplasia, and tuberous sclerosis complex. We also explore potential mechanisms implicating the role of tau kinases and phosphatases as well as the mammalian target of rapamycin (mTOR) in the promotion of co-pathology. Understanding the role of these co-pathologies could lead to new insights and therapies targeting both epileptogenic mechanisms and cognitive decline.
## Introduction
Tau is a microtubule-associated protein encoded in humans by the microtubule-associated protein tau gene, MAPT , on chromosome 17 ( ). In the brain, tau is most abundant in neurons, including neuronal axons, somatodendritic compartments, and nuclei, but it is also present in glia and, to a lesser degree, extracellularly ( ; ). The functions of tau in the brain are multifaceted, but its most well-characterized role is in microtubule binding and assembly. Tau is natively unfolded and highly soluble, thus exhibiting little tendency for aggregation. However, under pathological conditions, the hyperphosphorylation of tau reduces its affinity for tubulin and is thought to drive abnormal aggregations of phosphorylated tau (p-tau), such as neuropil threads or neurofibrillary tangles (NFTs), resulting in tauopathies ( ; ).
Endogenous tau is also implicated in neuronal activity ( ), though this role of tau is less well understood. Neuronal excitation, in turn, also regulates tau by promoting extracellular release and phosphorylation. Rapid and persisting increases in extracellular tau following in vivo ( ) or in vitro ( ) neuronal stimulation suggest that tau amplification is associated with pathological neuronal activation. Given that seizure and chronic epilepsy animal models result in prolonged tau phosphorylation ( ; ; ), a growing body of research is examining the role of pathological tau in epilepsy and mechanisms underlying epilepsy and tauopathy comorbidities.
In Alzheimer’s disease (AD), which is the most common tauopathy, an estimated 60% of patients have seizures and subclinical epileptic activity ( , ; ; ). Seizures are more common in AD and dementia with Lewy bodies than in primary tauopathies, such as frontotemporal dementia and progressive supranuclear palsy ( ). However, the possibility of seizures and hyperexcitability in primary tauopathies should not be ruled out, as they occur more frequently in these diseases than in the general population ( ; ). Myoclonus, a sign of network hyperexcitability, is observed in a subset of patients with corticobasal degeneration ( ), and epileptic activity is present in the FTDP-17 animal model of frontotemporal dementia with parkinsonism ( ).
Furthermore, tau pathology is repeatedly found in human epilepsy ( ). In a post-mortem series of 138 refractory epilepsy cases of diverse causes, Braak staging of NFTs in the age group 40–65 years revealed increased Braak stages III/IV compared with data from an age-matched series of non-epilepsy cases ( ). Abnormally high total tau and p-tau levels were also detected in cerebrospinal fluid of status epilepticus patients, with increased total tau correlating with greater risk of developing chronic epilepsy ( ). Given that neurodegenerative conditions characterized by hyperphosphorylated tau aggregations exhibit increased rates of epilepsy, epilepsies are being re-conceptualized within a tauopathy context ( ; ; ).
As such, the present review seeks to explore the following three main questions: (1) Does tau play a role in mediating network hyperexcitability and seizure activity across different epilepsy disorders? (2) Do comorbid tauopathies and epilepsies stem from independent or common mechanisms? (3) How do tauopathy and epilepsy comorbidities contribute to disease-related cognitive impairment? In light of evidence indicating tau-mediated epileptic activity and dysregulation of tau-related cell signaling pathways across seizure disorders, we propose a potential overarching mechanism ( ) whereby endogenous tau helps enable network hyperexcitability, which triggers homeostatic responses aimed, in part, to disable tau activity by phosphorylation. Resulting tau hyperphosphorylation and aggregation may in turn further contribute to cognitive impairments seen in some seizure disorders.
Cascade of events in development of seizures and tau pathology. Endogenous tau has an enabling function in the development of seizure activity following disease onset or traumatic insult. Network hyperexcitability in turn leads to cognitive decline and the activation of mechanisms involving mTOR and tau kinases and phosphatases, resulting in abnormal phosphorylation of tau. Overactivation of these cell signaling pathways increases susceptibility to pathological tau hyperphosphorylation and aggregation, which may also contribute to epilepsy-associated cognitive decline. GSK-3β, glycogen synthase kinase-3β; CDK5, cyclin-dependent kinase 5; PP2A, protein phosphatase 2A; mTOR, mammalian target of rapamycin; p-tau, abnormally phosphorylated tau. Created with .
## Role of Tau in Epileptic Activity
Across various animal models that exhibit epilepsy, reducing tau levels reduces network hyperexcitability, seizure severity, and latency to seizure stages ( ; ; ; ; ; ). Endogenous levels of tau also positively correlate with chemically induced seizure susceptibility in wild-type mice ( ), suggesting a role of endogenous tau in mediating epileptic activity.
Though not a primary epilepsy, autism spectrum disorder (ASD) is a disease with links to both seizures and tau. Several studies reveal a strong relationship between epilepsy in individuals with autism and autism in those with epilepsy ( ; ; ), with the prevalence of epilepsy in ASD doubling in adolescence (26%) compared to childhood (12%) ( ; ; ). Likely explanations supporting the conjugation of the two conditions include an imbalance in neuronal excitation/inhibition ( ; ; ). In the Cntnap2 mouse model of autism with focal epilepsy, global tau knockdown prevents epileptic activity in addition to other autistic-like behaviors ( ), indicating an epileptogenic role of tau in ASD.
Genetic ablation of tau, even by 50% by inactivation of a single Mapt allele, also reduces epileptic activity, high mortality rates, and cognitive deficits in the Scn1a mouse model of Dravet syndrome, a severe and intractable childhood epilepsy that is caused by mutations in the SCN1A gene and can lead to autism ( ; ; ; ; ). This effect also occurs in the Scn1a model following postnatal injection of tau-targeting antisense oligonucleotides ( ), suggesting that antisense oligonucleotides may be a promising treatment avenue for children with Dravet syndrome. further found that selective genetic ablation of tau in hippocampal excitatory neurons but not in inhibitory neurons mediates the neuroprotective effects of tau reduction in the Scn1a model. The authors propose that the suppression of epileptic activity by tau reduction may therefore result from a lower hypersynchrony of excitatory neuronal activity rather than greater inhibitory regulation.
In some models, pathological tau, rather than endogenous tau, can contribute to epileptic activity. Temporal lobe epilepsy (TLE) is one of the most prevalent forms of focal epilepsy ( ; ), and in the electrical amygdala kindling rodent model of TLE, tau-knockout mice do not differ from wild-type mice in seizure outcome following repeated kindling ( ). However, kindling produces longer epileptic discharge durations and accelerated seizure progression in rTg4510 transgenic mice, which overexpress P301L tau in forebrain and develop increased p-tau and NFTs ( ). These findings suggest that an increase in p-tau and tau aggregation promotes kindling-induced epileptogenesis.
Taken together, findings across different seizure disease models reveal a significant function of endogenous tau, as well as pathological tau, in the mediation of epileptic activity. Given tau’s role in modulating neuronal activity under normal physiological conditions ( ; ), endogenous tau likely contributes to network hyperexcitability across primary and secondary epilepsies. In humans, tau mRNA expression and protein levels in the brain can vary greatly ( ). And though exact reasons for individual differences in endogenous tau levels remain unknown, high levels may consequently predict a person’s susceptibility to epileptic activity. Elevated tau measurements in cerebrospinal fluid have in fact been shown to correlate with seizure type and duration in patients with epilepsy ( ). Higher levels of endogenous tau alone may not cause seizures, but it is possible that this may predispose an individual to seizure development upon pathogenesis.
## Presence of P-Tau Pathology in Seizure Disorders and Links to Epileptic Activity
While levels of endogenous or total tau differ across examinations of patients with epilepsy, increased levels of p-tau in the brain are found across many seizure disorders ( ). These include patients with TLE, Dravet syndrome, Nodding syndrome ( ; ; ), Niemann-Pick type C disease (NPC) ( ; ; ; ), focal cortical dysplasia IIB (FCDIIb) ( ; ), and tuberous sclerosis complex (TSC) ( ; ), as well as animal models of post-traumatic epilepsy (PTE) ( ; ), Lafora disease ( Epm2a ) ( ; ; ), and ASD ( ). It should be noted that tau aggregation in the brain is associated with older age and is generally uncommon in healthy young adults ( ; ). However, the presence of tau pathology in childhood or adolescent-onset epilepsies, such as Dravet syndrome, Nodding syndrome, Lafora disease, NPC, TSC, and ASD, and in relatively younger TLE patient cohorts ( ; ; ) suggest a causal link between seizure activity and p-tau accumulation.
In animal models of TLE, tau hyperphosphorylation is observed in relevant brain regions, including the amygdala, hippocampus, and cortex, following chemical and electrical amygdala kindling ( ; ; ). And in humans with chronic epilepsy, elevated p-tau is present in post-mortem ( ) as well as surgically resected tissue ( ; ; ; ; ; ). For instance, analysis of resected temporal lobe tissue by found pathological tau phosphorylation in the form of neuropil threads, NFTs, and pre-tangles in 31 of 33 TLE patients between 50 and 65 years of age. Interestingly, observations of subpial bands formed by cortical p-tau depositions have been made across separate studies ( ; ; ), providing evidence for a novel pattern of tau pathology in TLE that may result from seizure-induced reorganization of temporal lobe networks ( ).
Tau hyperphosphorylation is also consistently present in the initial and long-term secondary mechanisms initiated by traumatic brain injury (TBI) ( ; ; ; ; ), a leading cause of morbidity and mortality worldwide ( ). For instance, sustaining even a single TBI can result in progressive NFT formation that is more extensive and severe than what is expected with normal aging ( ; ). Furthermore, it is estimated that over 50% of severe TBI cases will result in seizures or PTE ( ), and animal models reveal increased p-tau levels in the brain associated with TBI-induced epileptic activity ( ; ). In a recent study, presented a novel model of transgenic zebrafish expressing a fluorescent tau biosensor where TBIs by blast-like pressure waves induced progressive tauopathies. Tau aggregation positively correlated with TBI severity and the presence of seizure-like clonic shaking. Furthermore, tau aggregation following TBI administration was prevented by the anti-convulsant ezogabine and exacerbated by kainate treatment, demonstrating a role of seizure activity in mediating tauopathy development ( ).
A mechanism by which epileptogenesis gives rise to tau hyperphosphorylation may underlie the high incidence of tauopathy and epilepsy co-pathology that is found in diseases such as AD and dementia with Lewy bodies ( , ; ). Given that endogenous tau plays a role in regulating neuronal activity, disruption in the homeostatic balance of tau modifications may mediate seizure comorbidities observed in these diseases, and epileptic activity may in turn help drive tau hyperphosophorylation. In classical tauopathies where overt epilepsy infrequently occurs, tau hyperphosphorylation can arise from a variety of different causes ( ). However, it is also possible that epileptic activity in these tauopathies is clinically underrecognized due to being non-motor or subclinical in nature, as suggested by the detection of subclinical epileptic activity in over 40% of AD patients during overnight electroencephalography and 1-h magnetoencephalogram recordings ( ). More studies involving extended periods of neurophysiological monitoring are therefore required to investigate the presence of epileptic activity and its potential contribution to tau hyperphosphorylation in primary tauopathies.
It should also be noted that tau pathology is not universally found in connection with epileptic activity. For instance, 31% of the post-mortem refractory epilepsy cases studied by were classified as Braak Stage 0, and analysis of surgically resected tissue from 56 TLE patients by found p-tau-positive neurons in only two samples. The absence of pathological tau deposition in these cases indicates that epileptogenesis does not always lead to tauopathies. However, the factors that determine the subsequent development of tau pathology in some cases of aberrant network excitability but not others remain unclear. It is possible that the formation of pathological tau deposits is linked to specific seizure disorders or that mechanisms mediating tau hyperphosphorylation are overactivated in cases of more severe epilepsy.
## Dysregulation of Cell Signaling Activity Upstream of Tau Phosphorylation
### Kinases
Given that the balance of tau phosphorylation states is regulated by enzymatic activity, investigations into the impairment of tau kinases and phosphatases in seizure disorders reveal links between epileptic activity and tau hyperphosphorylation. Investigations into novel pharmacological interventions targeting tau hyperphosphorylation in epilepsy have therefore concentrated on inhibiting and enhancing related phosphorylation and dephosphorylation mechanisms, respectively ( ; ).
One relevant kinase responsible for tau phosphorylation is glycogen synthase kinase-3β (GSK-3β) ( ). Upregulation of GSK-3β is found in surgically resected tissue samples from patients with intractable epilepsy ( ; ), and GSK-3β overactivation co-occurs with increased p-tau levels in mesial TLE patients ( ). Inhibition of GSK-3β may have dual benefits given that GSK-3β inhibition reduces tau hyperphosphorylation and NFT formation in tau-overexpressing transgenic mice ( ; ; ) and produces anticonvulsant effects against pentylenetetrazol-induced seizures in zebrafish larvae ( ). However, the observation of sustained increases in p-tau levels following kainic acid administration being accompanied by only transient increases in GSK-3β activity ( ) and the lack of effect on hippocampal p-tau by GSK-3β inhibitor pretreatment in the intra-amygdala kainic acid-induced status epilepticus mouse model ( ) indicate that GSK-3β is not the only kinase responsible for tau phosphorylation following epileptic activity.
Another protein kinase highly implicated in tau phosphorylation is cyclin-dependent kinase 5 (CDK5). Dysregulation of CDK5 signaling can contribute to neurodegeneration, excitotoxicity, and tau hyperphosphorylation ( ). As is seen with GSK-3β, CDK5 overactivation is present in resected tissue from refractory epilepsy patients ( ; ), and dysplastic cortical neurons in FCD patients express CDK5 aggregations ( ). Furthermore, progressive activation of CDK5 co-occurs with increasing tau phosphorylation in rodent seizure models ( ; ), indicating significant mediation of seizure-associated tau hyperphosphorylation by CDK5. For example, in the genetic mouse model of NPC, increased activation of CDK5 and its activator, p25, coincides spatially and temporally with tau pathology, and CDK5 inhibition by roscovitine and olomoucine prevents cytoskeletal protein phosphorylation ( ; , ).
Both GSK-3β and CDK5 play a role in neuronal excitability through involvement in GABAergic and glutamatergic neurotransmission, and inhibiting their activity can affect network activity through various mechanisms ( ; ; ; ). Therefore, inhibiting these kinases should be approached with caution. For instance, genetic ablation of the CDK5 activator, p35, increases susceptibility to spontaneous seizures in mice ( ). Tau levels were not measured in this study, but it is possible that the absence of activated CDK5 in this genetic model results in higher levels of dephosphorylated tau that contribute to neuronal hyperexcitability. These considerations highlight the complexity of kinase regulation in the setting of normal activity and hyperexcitable states. Targeting tau rather than upstream kinases may therefore be a more viable intervention option for seizure disorders ( ).
### Phosphatases
In addition to kinase activity, tau hyperphosphorylation associated with seizure activity may also be due to a lack of tau dephosphorylation by tau phosphatases. Following kainic acid administration in mice, biphasic changes in p-tau levels occur, where decreased phosphorylation is first observed within the first 6-h period followed by a gradual 3–5-fold increase until a 48-h endpoint. This progression is accompanied by a corresponding increase and then decrease in the activation of protein phosphatase 2A (PP2A) ( ), which is estimated to account for 70% of human brain tau dephosphorylation ( ). It is possible that phosphatase activity is initially triggered to offset elevated tau phosphorylation caused by upregulated kinase activity following an epileptic event. For instance, GSK-3β upregulation causes PP2A activation ( ). However, long-lasting phosphatase downregulation ultimately occurs, as evidenced by decreased PP2A activity paired with increased p-tau levels observed in epileptogenic brain regions following post-kainic acid status epilepticus, amygdala kindling, and fluid percussion injury in rats ( ).
Similarly, abnormal p-tau in the form of NFTs are also observed in the Epm2a mouse model ( ), which replicates many of the features of Lafora disease caused by EPM2A mutations, including laforin deficiency, neuronal degeneration, spontaneous epileptic activity, and the development of Lafora bodies ( ). Laforin is another tau phosphatase ( ), though further research is required to investigate connections between hyperexcitability states and laforin downregulation in other seizure disorders. At least in the Epm2a model, pathological tau levels are also associated with increased GSK-3β activation ( ), suggesting that tau hyperphosphorylation is not mediated by the absence of laforin alone in Lafora disease.
Interestingly, the lack of phosphatase activity may also contribute to epileptic activity. Sodium selenate is a specific agonist for PP2A expressing the regulatory B subunit, an essential subunit for tau dephosphorylation by PP2A ( ; ), and shows promise as a potential antiepileptic treatment option. Sodium selenate treatment attenuates seizure activity and tau hyperphosphorylation and accumulation following administration of pentylenetetrazol or kainic acid as well as in the TLE model of amygdala kindling and the fluid percussion injury model of PTE ( ; ). The antiepileptic effects of sodium selenate persist following drug washout in animal TBI models ( ), highlighting a potential disease-modifying effect of PP2A upregulation by sodium selenate during epileptogenesis when applied in early PTE disease stage.
The mechanisms through which tau dephosphorylation by phosphatase function alleviates epileptic activity remain unclear. Dephosphorylated tau at sufficient levels may be favorable in chronic epileptic states, or phosphatases may participate in independent signaling pathways that abate neuronal hyperexcitability. Regardless, as was proposed with tau kinases, long-term phosphatase inactivation may serve as a homeostatic response aimed at maintaining higher levels of phosphorylated tau and preventing endogenous tau from enabling network hyperexcitability ( ). Taken together, the discussed findings indicate that seizures give rise to disruptions in the intricate balance of tau kinase and phosphatase activity and that the combined effects of kinase upregulation and phosphatase downregulation contribute to progressive tau hyperphosphorylation and accumulation in seizure disorders.
### Mammalian Target of Rapamycin Pathways
The mammalian target of rapamycin (mTOR) is a highly conserved protein kinase that is implicated in a wide array of cellular and metabolic functions, including cell survival, growth, proliferation, migration, and differentiation ( ; ). Activation of mTOR is also a proposed driver of tau pathology given the involvement of tau-related kinases both upstream and downstream of mTOR signaling ( ) and the contribution of mTOR-mediated autophagy dysfunction to tau hyperphosphorylation ( ; ). For instance, the downstream targets of mTOR activation include signaling cascades involving 4EBP1, S6K1, and CDK5, all of which result in tau phosphorylation ( ).
Simplified diagram of activated kinase signaling cascades in epilepsy. Epileptic activity leads to the activation of tau kinases, GSK-3β and CDK5, as well as mTOR. Dashed-line arrows indicate indirect activation of mTOR by GSK-3β through the mTOR complex 1 and of CDK5 by mTOR through amyloid-β aggregation and calpain activation. The downstream targets of mTOR activation involve the activation of additional tau kinases, p70S6K1 and eIF4E. GSK-3β, glycogen synthase kinase-3β; mTOR, mammalian target of rapamycin; CDK5, cyclin-dependent kinase 5; S6K1, ribosomal protein S6 kinase 1; 4EBP1, 4E binding protein 1; p70S6K1, phosphorylated S6K1; eIF4E, eukaryotic translation initiation factor 4E; p-tau, abnormally phosphorylated tau. Created with .
Furthermore, mTOR hyperactivation accompanies epileptic activity across different seizure models including animal models of TLE, PTE, FCDII, Dravet syndrome, ASD, and TSC ( ; ; ; ; ; ; ; ; ). mTOR therefore likely contributes to tau and seizure co-pathology, warranting further pharmaceutical consideration of mTOR inhibition by rapamycin. Rapamycin treatment inhibits both tau hyperphosphorylation ( ; ; ) and the development of status epilepticus and chronic epilepsy in models of pharmacological seizure induction ( ; ), TLE ( ), and PTE ( ; ). Given that active GSK-3β also activates mTOR ( ), tau hyperphosphorylation resulting from seizure-associated GSK-3β upregulation may be further exacerbated by GSK-3β-mediated mTOR activation ( ; ).
While the dysregulation of mTOR signaling pathways may manifest differentially across seizure disorders, the many connections between tau and mTOR highlight the significance of maintaining an optimal ratio of dephosphorylated and phosphorylated tau through balanced kinase/phosphatase regulation. Endogenous tau enables mTOR activation through a disinhibition mechanism whereby tau inhibits phosphatase and tensin homolog deleted chromosome 10 (PTEN), which normally inhibits mTOR ( ; ). In genetic mouse models of ASD and Dravet syndrome, tau ablation prevents epilepsy and normalizes mTOR overactivation ( ; ; ), suggesting that reducing tau may be beneficial in these diseases via PTEN disinhibition. mTOR also functions as a negative regulator of autophagy ( ). Therefore, mTOR hyperactivity could prevent clearance of both normal and pathological tau ( ). Furthermore, reduced autophagy resulting from mTOR overactivation is implicated in elevated endogenous tau levels in the TSC2 mouse model of TSC ( ). In seizure disorders such as TSC and NPC that are characterized by autophagy dysregulation ( ; ), increase in normal tau levels may in turn contribute to both seizure activity and PTEN inhibition, creating a feedback loop of mTOR overactivation that results in further hyperphosphorylation of tau ( ).
Interplay of endogenous tau, mTOR, and autophagy in epilepsy. (A) Under normal physiological conditions (blue), endogenous tau positively regulates mTOR activity via PTEN inhibition, and mTOR in turn negatively regulates autophagy mechanisms that contribute to the clearance of tau and p-tau. (B) In hyperexcitability states found in pathophysiological conditions such as tuberous sclerosis complex and Niemann-Pick type C disease (red), overactivation of mTOR due to increased PTEN inhibition causes excess inhibition of autophagy, resulting in reduced clearance of tau species. Elevated levels of normal tau in turn exacerbate epileptic activity and mTOR disinhibition. PTEN, phosphatase and tensin homolog deleted chromosome 10; mTOR, mammalian target of rapamycin; p-tau, phosphorylated tau. Created with .
## Tau-Associated Cognitive Decline in Epilepsy Disorders
Cognitive impairment is a common comorbidity of seizure disorders. Though cognitive deficits can independently occur in seizure disorders as a direct result of disease etiology, such as trauma, epileptic activity contributes to, and exacerbates cognitive decline ( ). However, there are also investigations into a potential role of tau pathology in epilepsy-associated cognitive decline. While individuals with dementia have higher rates of epilepsy, seizures are experienced more frequently in tauopathy-associated dementias like AD than in other dementias ( ). Cognitive decline is also accelerated in patients who have both AD and seizures compared to those with only AD ( ), suggesting that pathological tau and seizures can synergistically worsen cognitive outcomes.
Correlations between tau pathology and cognition are in fact observed in epilepsy. Post-mortem analysis of 138 refractory epilepsy cases revealed that 77% of patients with Braak staging III or higher exhibited progressive cognitive decline ( ), and increased total and p-tau levels measured in surgically resected tissue from TLE patients are inversely correlated with cognitive scores ( ; ; ). In younger TLE patients, an association between post-operative naming decline and subtle tau hyperphosphorylation localized to only the subiculum and dentate gyrus suggest that tau-associated pathological changes in relevant brain regions over time may underlie progressive cognitive impairment seen in TLE ( ). Furthermore, the neuroprotective effects of tau ablation against not only seizures but also cognitive deficits in animal models of ASD and Dravet syndrome ( ; ) provide evidence for a role of tau in mediating cognitive impairment in these diseases as well. Therefore, tau pathology present in seizure disorders may exacerbate cognitive decline resulting from epileptic states, with seizure-driven tau hyperphosphorylation further compounding this effect with disease progression.
## Conclusion
As presented in this review, a mounting body of literature has elucidated connections between tau pathology and epilepsy disorders of diverse etiologies. The antiseizure effect of tau ablation that can be reproduced in a variety of seizure models indicates a significant mediating role of endogenous tau in epileptogenesis. Given findings of upregulated kinase and downregulated phosphatase activity across different seizure disorders, we propose that epileptic activity can trigger homeostatic responses whereby enzymatic pathways disable endogenous tau by increased phosphorylation to stabilize aberrant network hyperexcitability. Subsequent hyperphosphorylation and accumulation of tau results from overactivation of such mechanisms, especially with recurring epileptic activity, and continuous epileptic states. Furthermore, growing evidence indicates a potential contribution of tau hyperphosphorylation to progressive cognitive decline in seizure disorders. Though the exact degrees to which tau involvement in seizures and cognitive decline are mediated by convergent or divergent mechanisms in distinct diseases remains unclear, the overlapping of tau-related cell signaling pathways and prevalence of tau hyperphosphorylation found throughout different types of epilepsies ( ) warrant continuing efforts into understanding epilepsies from a tauopathy perspective. Greater focus on tau in epileptic pathophysiology may yield advances in diagnostic and prognostic tools and novel therapeutic approaches targeting tau and tau-associated pathways.
Characterization of seizure disorders and their links to tau pathology and tau-associated mechanisms.
## Author Contributions
KV and KH developed the manuscript concept. KH prepared the figure and table. All authors contributed to the manuscript and approved the final submitted version.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
|
## Background
Diabetes mellitus (DM) has been reported to be associated with perioperative stroke, but the effects of preoperative hyperglycemia on the risk of perioperative stroke in diabetic patients undergoing non-cardiovascular surgery remain unclear. This study investigated the association between preoperative hyperglycemia and the risk of perioperative ischemic stroke in type 2 diabetic patients undergoing non-cardiovascular surgery.
## Methods
This retrospective cohort study screened 27,002 patients with type 2 DM undergoing non-cardiovascular surgery with general anesthesia between January 2008 and August 2019 at The First Medical Center of Chinese People’s Liberation Army (PLA) General Hospital. The exposure of interest was preoperative hyperglycemia, defined as a fasting plasma glucose (FPG) ≥ 7 mmol/L. The outcome of interest was a new diagnosis of perioperative ischemic stroke within 30 days after surgery. Residual confounding was minimized by controlling for observable patient and intraoperative factors. Logistic regression was conducted in the total and propensity score matched cohorts. In addition, we stratified patients into six subgroups to investigate whether the association between preoperative hyperglycemia and perioperative ischemic stroke differs in these subgroups.
## Results
The overall incidence of perioperative ischemic stroke was 0.53% ( n = 144) in the current cohort. The odds of perioperative ischemic stroke were significantly increased for patients with preoperative hyperglycemia after adjusting for patient- related variables (OR: 1.95; 95% CI: 1.39–2.75; p < 0.001), surgery-related variables (OR: 2.1; 95% CI: 1.51–2.94; p < 0.001), and all confounding variables (OR: 1.78; 95% CI: 1.26–2.53; p < 0.001). The risk of perioperative stroke was significantly increased in patients with preoperative hyperglycemia (OR: 2.51; 95% CI: 1.66–3.9; p < 0.001) in the propensity score matched cohort. Preoperative hyperglycemia was associated with the outcome for all the subgroups except for patients undergoing neurosurgery.
## Conclusion
Preoperative hyperglycemia is associated with an elevated risk of perioperative stroke in patients with type 2 DM undergoing non-cardiovascular surgery. The effect could be eliminated for patients undergoing neurosurgery, during which specific risk factors should be considered.
## Introduction
Perioperative stroke is a severe neurological complication after surgery and is associated with considerable morbidity and mortality rates ( ; ). The perioperative stroke incidence is approximately 0.1–1.9% in non-cardiac surgeries ( ; ) and can reach 9.7% in complicated cardiac surgeries ( ). Despite the low incidence, perioperative stroke seriously affects the prognosis of surgical patients, and can impose extra burden on families and society.
Type 2 diabetes mellitus (DM) is characterized by persistent insulin resistance and hyperglycemia. Type 2 DM has been reported to be causally associated with ischemic stroke ( ), as hyperglycemia affects arterial remodeling ( ; ) and increases arterial stiffness ( ). Hyperglycemia exerts considerable influence on diabetic microvascular pathology and is associated with an elevated risk of vascular disease ( ). Preoperative hyperglycemia is also associated with several poor clinical outcomes in a variety of surgical courses ( ; ; ; ; ) and has proven an independent predictor of perioperative stroke in patients undergoing carotid endarterectomy ( ).
In a multicenter, international prospective cohort study, a casual glucose level above 7.92 mmol/L before surgery was most likely to develop postoperative myocardial injury in diabetic patients undergoing non-cardiac surgery ( ). To the best of our knowledge, a paucity of research to date has assessed the impact of preoperative hyperglycemia on perioperative stroke in non-cardiovascular surgical patients with type 2 DM. However, it is unclear whether preoperative hyperglycemia in diabetic patients is associated with an elevated risk of perioperative stroke compared to those with relatively normoglycemic status.
The goal of this retrospective cohort study is to assess the association between preoperative hyperglycemia and perioperative stroke in patients with type 2 DM undergoing non-cardiovascular surgery. We hypothesized that preoperative hyperglycemia fasting plasma glucose [(FPG) ≥ 7 mmol/L] in type 2 diabetic patients is associated with an elevated risk of perioperative stroke compared to those with normoglycemic status (FPG < 7 mmol/L).
## Materials and methods
### Study design and study population
The Chinese People’s Liberation Army (PLA) General Hospital has a central computerized database, and there is a digital record of the demographics, diagnoses, laboratory results, and other clinical data for all inpatients. This cohort study was approved by the Medical Ethics Committee of The First Medical Center of Chinese PLA General Hospital (reference number: S2021-493-01), and the requirement for informed content was exempted. The current research adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines ( ). We electronically retrieved from the database all surgical inpatients between January 2008 and August 2019. Inclusion criteria were patients aged 18 years old or older, who underwent non-cardiovascular surgery, and received general anesthesia with a surgery length of more than 1 h. A total of 251,793 adult patients met the inclusion criteria. Exclusion criteria were ASA classification of IV or above, no diabetes diagnosis, type 1 DM, intraoperative hemorrhage (transfusion of > 4 U of packed red blood cells or whole blood) and missing data for any confounders. Type 2 diabetes mellitus was identified through the discharge diagnosis of the index procedure. For patients with multiple procedures within the study period, the first surgery was used as an index procedure. In total, 27,002 participants comprised the study cohort. The patient flow diagram for this study is shown in .
Patient flow diagram. ASA, American Society of Anesthesiologists; PSM, propensity score matching.
### Preoperative hyperglycemia and perioperative stroke
Preoperative glucose level was determined by the fasting plasma glucose (FPG) level at the time of preoperative evaluation, which was tested by the central laboratory. If there were multiple measurements before surgery, the value closest to the date of surgery was used, and only the results from the central laboratory were used. An FPG of ≥ 7 mmol/L is often used to diagnose diabetes, and an increase in the prevalence and incidence of diabetic retinopathy begins approximately at an FPG of 7 mmol/L ( ). Additionally, a preoperative FPG ≥ 7 mmol/L was considered to provide useful information in the perioperative setting ( ). Considering these points, preoperative hyperglycemia was defined as FPG ≥ 7 mmol/L. Perioperative stroke was defined as new-onset brain infarction during hospital stay. Diagnoses of stroke were confirmed by a combination of neuroimaging and clinical evidence of cerebrovascular ischemia within 30 days after surgery, identified through ICD9/ICD10 diagnosis codes ( ).
### Confounding variables
The following covariates were included as potential confounders in our models: age, American Society of Anesthesiologists (ASA) classification, hypertension, coronary heart disease, heart failure, peripheral vascular disease, previous ischemic stroke, preoperative platelet, preoperative albumin, preoperative insulin, and preoperative anticoagulants were defined as patient related variables; while surgical category, surgery length, emergency surgery, intraoperative blood product usage, and intraoperative vasoactive drugs were defined as surgery-related variables.
### Statistical analysis
Continuous variables were summarized as the median and interquartile range (IQR), and categorical variables were summarized as the number and percentage of patients.
We used multivariable logistic regression for the aforementioned covariates in the analyses to determine whether preoperative hyperglycemia was independently associated with an elevated risk of perioperative ischemic stroke in diabetic patients. Ischemic stroke was modeled as the dependent variable and hyperglycemia was modeled as independent variable. Four models were built to analyze this association. Model 1 was a univariable model for crude analysis. Model 2 was a multivariable model including patient-related variables. Model 3 was a multivariable model including surgery-related variables, and model 4 included all variables. We also used the propensity score matching (PSM) method to further validate the association between preoperative hyperglycemia and perioperative stroke. In PSM, patients in the two groups were matched by propensity score (PS) at a 1:1 ratio with a caliper of 0.05.
To further investigate whether the association between preoperative hyperglycemia and perioperative stroke differs among selected patient subgroups, we also conducted multivariable logistic regression stratified by several key variables (age, sex, previous ischemic stroke, preoperative insulin, neurosurgery, and surgery length). Statistical analysis was performed using GraphPad Prism (version 9.0) and R (version 4.0), along with the MatchIt, rms, MASS, cobalt, and car packages. With all statistical tests being two-sided, a p -value of < 0.05 was considered statistically significant. Odds ratios with 95% confidence intervals (CI) were reported for all models.
## Results
### Characteristics of study cohort
A total of 251,793 adult patients met the inclusion criteria during the study period, and the study cohort consisted of 27,002 patients after application of exclusion criteria. Of these, 9,572 (35.4%) patients were defined as having preoperative hyperglycemia (FPG ≥ 7 mmol/L), 8,746 (32.4%) were aged ≥ 65 years old, 15,630 (57.9%) were males, 1,459 (5.4%) had a history of ischemic stroke, 13,822 (51.1%) were under insulin medication, 2,319 (8.6%) received neurosurgery, and 10,431 (38.6%) underwent a surgery length exceeding 3 h. Descriptive statistics comparing patients in the hyperglycemic group to those in the normoglycemic group in the total and PS matched cohorts are shown in .
Patient characteristics in total and propensity score matched cohorts.
### Association between preoperative hyperglycemia and perioperative stroke
Perioperative stroke significantly increased for patients in the hyperglycemic group compared to patients in the normoglycemic group (OR: 2.1; 95% CI: 1.51–2.92; p < 0.001). The odds of perioperative stroke for patients in the hyperglycemic group significantly increased in the logistic regression analysis after adjusting for patient-related variables (OR: 1.95; 95% CI: 1.39–2.75; p < 0.001), or surgery-related variables (OR: 2.1; 95% CI: 1.51–2.94; p < 0.001), or all confounding variables (OR: 1.78; 95% CI: 1.26–2.53; p < 0.001) ( ). The complete data of the univariate and multivariate models are detailed in .
Odds ratio for preoperative FPG ≥ 7 mmol/L for risk of stroke in the total and propensity score matched cohorts.
### Propensity score analysis
Variables including age, sex, ASA status, hypertension, preoperative hemoglobin, preoperative total bilirubin, preoperative oral hypoglycemics, preoperative insulin, surgical category, and emergency surgery were matched in PSM. We obtained 9,132 pairs after PSM, with a standardized mean difference (SMD) of less than 0.10 for all variables ( ). The distribution of propensity scores in the hyperglycemic and normoglycemic groups is graphically illustrated before and after matching ( ). Perioperative stroke occurred in 75 (0.82%) patients in the hyperglycemic group and 30 (0.33%) patients in the normoglycemic group in the PS matched cohort, and the risk of perioperative stroke was significantly increased in patients with preoperative hyperglycemia (OR: 2.51; 95% CI: 1.66–3.9; p < 0.001) ( ). Complete data for the PS matched cohort are detailed in .
Propensity score distribution before and after matching. FPG, fasting plasma glucose.
### Subgroup analysis
The overall incidence of perioperative stroke was 0.53% ( n = 144). The incidence was 0.8% ( n = 77) in the hyperglycemic group and 0.38% ( n = 67) in the normoglycemic group. We evaluated the effects of preoperative hyperglycemia on perioperative stroke in subgroups of patients stratified by age, sex, previous ischemic stroke, preoperative insulin medication, surgical category, and surgery length ( ). The OR of perioperative stroke were significant in spite of age (aged ≥ 65 years old: OR: 1.76; 95% CI: 1.1–2.79; p = 0.017; aged < 65 years old: OR: 1.88; 95% CI: 1.11–3.19; p = 0.02), sex (male: OR: 2.07; 95% CI: 1.28–3.39; p = 0.003; female: OR: 1.67; 95% CI: 1–2.79; p = 0.048), previous ischemic stroke (with previous ischemic stroke: OR: 2.11; 95% CI: 1.24–3.62; p = 0.006; without previous ischemic stroke: OR: 1.65; 95% CI: 1.04–2.6; p = 0.03), preoperative insulin medication (receiving insulin medication: OR: 1.55; 95% CI: 1.01–2.4; p = 0.047; not receiving insulin medication: OR: 2.25; 95% CI: 1.26–3.99; p = 0.006) and surgery length (exceeding 3 h: OR: 1.87; 95% CI: 1.16–3.04; p = 0.01; within 3 h: OR: 1.73; 95% CI: 1.05–2.86; p = 0.03). The OR of perioperative stroke was significant for patients undergoing non-neurosurgical procedures (OR: 2.08; 95% CI: 1.39–3.13; p < 0.001), but not significant for patients undergoing neurosurgery (OR: 1.04; 95% CI: 0.5–2.11; p = 0.92).
Effects of preoperative glucose on perioperative stroke risk. CI, confidence interval; FPG, fasting plasma glucose; OR, odds ratio.
## Discussion
Poor perioperative glycemic control increases postoperative morbidity and mortality in a variety of surgical cohorts ( ; ; ; ; ). In this cohort of non-cardiovascular surgical patients with type 2 DM, we found that preoperative hyperglycemia was associated with an elevated risk of perioperative stroke. Our findings will inform the role of preoperative glycemic status in optimal preoperative stroke risk assessments for diabetic patients undergoing non-cardiovascular surgery.
An overall incidence of perioperative stroke of 0.25% was recently reported in a large cohort of non-cardiac surgical patients ( ). In the current cohort of 27,002 non-cardiovascular surgical patients with type 2 DM, perioperative stroke occurred in 144 (0.53%) patients. Using an FPG ≥ 7 mmol/L as primary exposure (preoperative hyperglycemia), perioperative stroke occurred in 67 (0.38%) patients with normoglycemia, and in 77 (0.8%) patients with hyperglycemia. Our results are consistent with previous researches that found that the risk of stroke ranges from approximately 0.1–1.9% in non-cardiac surgeries depending on risk factors ( ), and that diabetic patients undergoing surgical procedures have higher rates of perioperative stroke than non-diabetic patients ( ). Our study demonstrated that preoperative hyperglycemia elevated the known increased risk of perioperative stroke in type 2 diabetic patients undergoing non-cardiovascular surgeries, regardless of age, sex, history of stroke, preoperative insulin medication or surgery length.
There are several possible explanations for the findings of the current cohort study. First, preoperative FPG < 7 mmol/L represents relatively tighter glycemic control, which means reduced risk factors for cardiovascular events due to improved endothelial cell function and decreased inflammatory mediators ( ; ). Second, diabetic patients could suffer aggravated insulin resistance resulting from preoperative stress and starvation, which is described as “stress hyperglycemia,” irrespective of satisfying glucose control on ordinary days. A higher level of preoperative FPG could indicate a harsher state of insulin resistance, which is also associated with poor outcomes for surgical patients ( ). Finally, since poor preoperative glucose control is independently associated with postoperative hyperglycemia ( ; ), and postoperative hyperglycemia leads to a higher rate of adverse events after surgery ( ), adverse effects of preoperative FPG ≥ 7 mmol/L could be worsened in surgical patients with type 2 DM.
We also evaluated the association between preoperative hyperglycemia and perioperative stroke across various subgroups. In our cohort of type 2 diabetic patients, the normoglycemic group showed a higher incidence of neurosurgery compared with the hyperglycemic group (9.6% vs. 6.7%). An intriguing finding in our cohort is that the association of preoperative hyperglycemia with perioperative stroke only existed in the non-neurosurgical subgroup. Presumably this is related to the fact that neurosurgical patients are exposed to very specific risks as brain tissues are very vulnerable to neurosurgical maneuvers (e.g., surgical brain injury induced by direct incisions, electrocauterization, and retraction) ( ). It is worthwhile to note that owing to the risk of postoperative intracranial hemorrhage, anticoagulant medication was often deferred or discontinued after neurosurgeries, which also increased the incidence of thrombosis ( ). Intraoperative mechanical insults combined with postoperative hypercoagulability predisposed neurosurgical patients to a higher risk of perioperative stroke, so that the association between preoperative hyperglycemia and perioperative stroke may become weak due to these specific risk factors. Our results indicated that patients in the hyperglycemic group were more likely to accept insulin medication than those in the normoglycemic group (63.9% vs. 44.2%). Considering that patients under insulin medication could have greater diabetes severity (resulting in more severe hyperglycemia and putting them at a higher risk of perioperative stroke), a subgroup analysis stratified by preoperative insulin medication was conducted. It turned out that preoperative hyperglycemia, regardless of whether the patients received preoperative insulin medication, was associated with an elevated risk of perioperative ischemic stroke (OR: 1.55; 95% CI: 1.01–2.4; p = 0.047; OR: 2.25; 95% CI: 1.26–3.99; p = 0.006; respectively).
The effects of preoperative hyperglycemia on the risks of postoperative complications in cardiovascular surgeries have been widely studied ( ; ; ; ; ). Preoperative hyperglycemia was also associated with elevated postoperative infections and prolonged hospital length of stay after non-cardiac surgeries ( ; ). The effects of preoperative glucose level on postoperative cardiovascular events during non-cardiac procedures have also received growing attention in recent years. demonstrated that preoperative casual glucose level exceeding 7.92 mmol/L was predictive of myocardial injury after non-cardiac surgery. also reported that preoperative hyperglycemia, but not glycosylated hemoglobin levels, was associated with myocardial injury after non-cardiac surgery. As preoperative glucose level plays an important role in postoperative outcomes, satisfying preoperative glycemic control could reduce the aforementioned risks ( ; ; ). However, few studies have investigated the relationship between preoperative hyperglycemia and the risk of perioperative stroke after non-cardiovascular surgeries. The current study demonstrated that preoperative FPG ≥ 7 mmol/L increased the risk of perioperative stroke in non-cardiovascular surgical patients with type 2 DM.
Hyperglycemia is considered an independent predictor of adverse events after surgery, however, it is also a modifiable factor. As preoperative glucose control is a foundation of preoperative preparation for patients with DM, concerning our outcomes, preoperative FPG level is particularly important for non-cardiovascular surgical patients with type 2 DM, considering the risk of perioperative stroke. This study demonstrated that diabetic patients with preoperative hyperglycemia were more likely to suffer from perioperative stroke relative to those with normoglycemic status. Preoperative FPG ≥ 7 mmol/L could provide additional value when assessing the risk of perioperative stroke patients with type 2 DM undergoing non-cardiovascular surgeries.
There are some limitations to our findings that must be considered. In the current cohort, all the stroke cases were diagnosed postoperatively, meaning we were unable to discriminate intraoperative stroke from postoperative stroke, which are very different entities. Eleven (7.6%, data not shown) stroke cases were diagnosed 1 week after surgery, and it is difficult to identify the long-term impact of preoperative hyperglycemia on perioperative stroke risk. Nevertheless, we found a strong association between preoperative hyperglycemia and perioperative stroke. Since glycosylated hemoglobin was not a routine measurement, only 3,700 (13.7%, data not shown) subjects in the cohort had glycosylated hemoglobin results. We were unable to analyze the effect of glycosylated hemoglobin due to the significant missing data. In this respect, we highlighted the effect of short-term rather than long-term glycemic status. We conducted correlation analysis of FPG and glycosylated hemoglobin for the 3,700 participants, and we found a relatively strong correlation between preoperative FPG and glycosylated hemoglobin ( r = 0.402, p < 0.001; ). Further studies are needed to verify the effects of preoperative glycosylated hemoglobin on the risk of perioperative stroke. In addition, this is a cohort from a single institution, and a larger, multicenter cohort study is required to validate our conclusions. Our findings highlight the need for further studies to establish the ideal level of glycemic control for diabetic patients before non-cardiovascular surgeries, and for randomized clinical trials to determine whether improving glycemic control reduces the risk of perioperative stroke for diabetic patients undergoing non-cardiovascular surgeries.
## Conclusion
Preoperative hyperglycemia is associated with an elevated risk of perioperative stroke in type 2 diabetic patients undergoing non-cardiovascular surgery. The effect could be eliminated for patients undergoing neurosurgery, during which specific risk factors should be taken into consideration. As preoperative fasting plasma glucose ≥ 7 mmol/L predisposes type 2 diabetic patients to perioperative stroke, clinicians should take note of fasting plasma glucose level during preoperative evaluation.
## Data availability statement
The original contributions presented in this study are included in the article/ , further inquiries can be directed to the corresponding author/s.
## Ethics statement
The studies involving human participants were reviewed and approved by the Medical Ethics Committee of The First Medical Center of Chinese PLA General Hospital. The ethics committee waived the requirement of written informed consent for participation.
## Author contributions
SL wrote the manuscript with contributions from all authors. WM, YM, and SL designed the study. LS, MS, FZ, HY, JL, ML, and YS were responsible for data extraction and acquisition. YM, SL, and BW designed and conducted the statistical analyses. All authors critically reviewed the report and approved the final version.
|
An individual's genetic background affects their emotional behavior and response to stress. Although studies have been conducted to identify genetic predictors for emotional behavior or stress response, it remains unknown how prior stress history alters the interaction between an individual's genome and their emotional behavior. Therefore, the purpose of this study is to identify chromosomal regions that affect emotional behavior and are sensitive to stress exposure. We utilized the BXD behavioral genetics mouse model to identify chromosomal regions that predict fear learning and emotional behavior following exposure to a control or chronic stress environment. 62 BXD recombinant inbred strains and C57BL/6 and DBA/2 parental strains underwent behavioral testing including a classical fear conditioning paradigm and the elevated plus maze. Distinct quantitative trait loci (QTLs) were identified for emotional learning, anxiety and locomotion in control and chronic stress populations. Candidate genes, including those with already known functions in learning and stress were found to reside within the identified QTLs. Our data suggest that chronic stress history reveals novel genetic predictors of emotional behavior. |
Alteration in social behavior is one of the most debilitating symptoms of major depression, a stress related mental illness. Social behavior is modulated by the reward system, and gamma oscillations in the nucleus accumbens (NAc) seem to be associated with reward processing. In this scenario, the role of gamma oscillations in depression remains unknown. We hypothesized that gamma oscillations in the rat NAc are sensitive to the effects of social distress. One group of male <i>Sprague-Dawley</i> rats were exposed to chronic social defeat stress (CSDS) while the other group was left undisturbed (control group). Afterward, a microelectrode array was implanted in the NAc of all animals. Local field potential (LFP) activity was acquired using a wireless recording system. Each implanted rat was placed in an open field chamber for a non-social interaction condition, followed by introducing another unfamiliar rat, creating a social interaction condition, where the implanted rat interacted freely and continuously with the unfamiliar conspecific in a natural-like manner (see Supplementary Videos). We found that the high-gamma band power in the NAc of non-stressed rats was higher during the social interaction compared to a non-social interaction condition. Conversely, we did not find significant differences at this level in the stressed rats when comparing the social interaction- and non-social interaction condition. These findings suggest that high-gamma oscillations in the NAc are involved in social behavior. Furthermore, alterations at this level could be an electrophysiological signature of the effect of chronic social stress on reward processing. |
The perinatal window is a critical developmental time when abnormal gestational stimuli may alter the development of the stress system that, in turn, influences behavioral and physiological responses in the newborns. Individual differences in stress reactivity are also determined by variations in maternal care, resulting from environmental manipulations. Despite glucocorticoids are the primary programming factor for the offspring's stress response, therapeutic corticosteroids are commonly used during late gestation to prevent preterm negative outcomes, exposing the offspring to potentially aberrant stress reactivity later in life. Thus, in this study, we investigated the consequences of one daily s.c. injection of corticosterone (25 mg/kg), from gestational day (GD) 14-16, and its interaction with offspring early handling, consisting in a brief 15-min maternal separation until weaning, on: (i) maternal behavior; and (ii) behavioral reactivity, emotional state and depressive-like behavior in the adolescent offspring. Corticosterone plasma levels, under non-shock- and shock-induced conditions, were also assessed. Our results show that gestational exposure to corticosterone was associated with diminished maternal care, impaired behavioral reactivity, increased emotional state and depressive-like behavior in the offspring, associated with an aberrant corticosterone response. The early handling procedure, which resulted in increased maternal care, was able to counteract the detrimental effects induced by gestational corticosterone exposure both in the behavioral- and neurochemical parameters examined. These findings highlight the potentially detrimental consequences of targeting the stress system during pregnancy as a vulnerability factor for the occurrence of emotional and affective distress in the adolescent offspring. Maternal extra-care proves to be a protective strategy that confers resiliency and restores homeostasis. |
Together with genetic factors, early-life experience governs the expression and function of stress-related genes throughout life. This, in turn, contributes to either resilience or vulnerability to depression and to aging-related cognitive decline. In humans and animal models, both the quality and quantity of early-life maternal care has been shown to be a predominant signal triggering bi-directional and enduring changes in expression profiles of genes including glucocorticoids and corticotropin releasing factor (CRH; hypothalamic and hippocampal), associated with the development of resilient or vulnerable phenotypes. However, many crucial questions remain unresolved. For examples, how is the maternal-derived signal transmitted to specific neuronal populations where enduring (likely epigenetic) regulation of gene expression takes place? What is the nature of this information? In other words, how do neurons know to 'turn on' epigenetic machinery? What are the direct functional consequences of altered gene expression? This review describes the voyage of recurrent bursts of sensory input from the mother ('mother's love') to CRH-expressing hypothalamic neurons that govern the magnitude of the response to stress. In addition, the acute and enduring effects of both nurturing and fragmented maternal care on the structure, cellular signaling and function of specific hippocampal and hypothalamic neurons are discussed. The evolving understanding of the processes initiated by the early life experience of 'mother's love' suggest novel molecular targets for prevention and therapy of stress-related affective and cognitive disorders. |
Norepinephrine (NE) is thought to play a key role in fear and anxiety, but its role in amygdala-dependent Pavlovian fear conditioning, a major model for understanding the neural basis of fear, is poorly understood. The lateral nucleus of the amygdala (LA) is a critical brain region for fear learning and regulating the effects of stress on memory. To understand better the cellular mechanisms of NE and its adrenergic receptors in the LA, we used antibodies directed against dopamine beta-hydroxylase (DβH), the synthetic enzyme for NE, or against two different isoforms of the beta-adrenergic receptors (βARs), one that predominately recognizes neurons (βAR 248) and the other astrocytes (βAR 404), to characterize the microenvironments of DβH and βAR. By electron microscopy, most DβH terminals did not make synapses, but when they did, they formed both asymmetric and symmetric synapses. By light microscopy, βARs were present in both neurons and astrocytes. Confocal microscopy revealed that both excitatory and inhibitory neurons express βAR248. By electron microscopy, βAR 248 was present in neuronal cell bodies, dendritic shafts and spines, and some axon terminals and astrocytes. When in dendrites and spines, βAR 248 was frequently concentrated along plasma membranes and at post-synaptic densities of asymmetric (excitatory) synapses. βAR 404 was expressed predominately in astrocytic cell bodies and processes. These astrocytic processes were frequently interposed between unlabeled terminals or ensheathed asymmetric synapses. Our findings provide a morphological basis for understanding ways in which NE may modulate transmission by acting via synaptic or non-synaptic mechanisms in the LA.
## Introduction
Norepinephrine (NE) has long been implicated in fear and anxiety (Gray, ; Redmond, ; Aston-Jones and Bloom, ; Aston-Jones et al., , ; Sullivan et al., ), and is known to play a role in learning and memory (Ferry et al., ; Bailey et al., ; McGaugh, ). However, the contribution of NE to Pavlovian fear conditioning, a leading model for understanding the neural basis of fear and anxiety and learning and memory, is poorly understood. Recent studies suggest that NE contributes to the acquisition but not the consolidation of fear conditioning (Lee et al., ; Debiec and LeDoux, ; Grillon et al., ; Murchison et al., ; Bush et al., ). The lateral nucleus of the amygdala (LA) is a critical brain region for fear learning (LeDoux, , ; Rodrigues et al., ; Maren, ; Lang and Davis, ) and blocking beta-adrenergic receptors (βARs) within the LA disrupts the acquisition of fear conditioning (Bush et al., ). Previous studies have shown that the amygdala is extensively innervated by noradrenergic fibers, primarily originating from the locus coerulus (Moore and Card, ; Fallon and Ciofi, ; Asan, , ). Several studies have examined the ultrastructural relations of NE terminals in the basolateral complex or basolateral nucleus (BLA) (Asan, , ; Li et al., , ). However, the BLA contains two major subdivisions, the LA and the basal nucleus (B), each with distinct connections (Pitkänen et al., ), and the LA, but not the B, is necessary for fear conditioning (Amorapanth et al., ; Nader et al., ). Moreover, the results of these ultrastructural NE studies differed: one study showed that within the basal complex, NE terminals rarely formed synapses (Asan, ) while another showed a greater proportion of NE terminals forming synapses, including with other terminals (Li et al., ). Electrophysiological studies have shown that βARs play a role in synaptic transmission (Gean et al., ; Huang et al., , ; Ferry et al., ; Buffalari and Grace, ) and LTP (Johnson et al., ) in the amygdala. While anatomical studies have characterized the cellular and subcellular distribution of βARs in other brain regions (Aoki et al., ; Aoki, , ; Aoki and Pickel, ; Milner et al., ), little is known about the cellular and subcellular distribution of βARs within the amygdala. Using commercial antibodies directed against the β and β receptor subtypes, one confocal microscopic study found that the β and β receptor subtypes are widely distributed in basal amygdala neurons but not in astrocytes (Qu et al., ). This group also found that while both receptor subtypes were seen in the membranes and cytoplasm of cell bodies, the β receptor subtype, but not the β , was localized to the nucleus. To understand better the cellular mechanisms of NE's contributions to fear learning, we examined the anatomical organization of NE terminals and βARs in the LA. In this study, we employed immunoelectron microscopy to determine whether terminals immunoreactive for dopamine beta-hydroxylase (DβH), the synthetic enzyme for NE, form synaptic junctions in the LA and if so, examine these synapses and identify the post-synaptic targets on NE terminals. To determine the cellular and subcellular distributions of βARs in the LA, we used previously characterized antibodies directed against two different isoforms of βARs: βAR 248, an antibody that predominately recognizes neurons, and βAR404, which primarily detects astrocytes.
## Materials and Methods
Male Sprague–Dawley (Hilltop Lab Animals, Inc; Scottdale, PA, USA) rats weighing 300–400 g ( n = 8) were used for studies. All procedures used were approved by the Animal Use and Care Committee of New York University, and conform to the guidelines of the National Institutes of Health on Care and Use of Experimental Animals in Research.
### Tissue fixation
Naïve animals were anaesthetized with chloral hydrate (25%; 1–1.5 g/kg BW) and transcardially perfused with 25–30 ml of heparinized 0.9% saline followed by either: 50 ml of 3.0% acrolein mixed into 4% paraformaldehyde (PFA), followed by 450 ml of 4% PFA dissolved in 0.1 M phosphate buffer (PB, pH 7.4; n = 4), or 500 ml 0.1% glutaraldehyde/4% PFA ( n = 4). The brains were removed from the skull, blocked, and post-fixed in 4% PFA for 30 min. Blocks containing the amygdala were cut on a Vibratome and 40-μm coronal sections were collected. Tissue sections were treated with 1% sodium borohydride in PB for 30 min and rinsed with phosphate-buffered saline (0.01 M PBS, pH. 7.4).
### Immunocytochemical labeling
Following rinses in PBS, tissue sections were preincubated in PBS containing 1% (w:v) bovine serum albumin (BSA) for 30 min. The sections were incubated overnight at room temperature with either a mouse monoclonal antibody directed against DβH (1:500; Millipore, Temecula, CA, USA) or rabbit polyclonal antiserum directed against the one of the β-adrenergic receptors. The following day, the tissue was rinsed in PBS, incubated for 30 min in a solution containing a 1:200 dilution of either goat anti-mouse for DβH or goat anti-rabbit biotinylated IgG for βAR (1:200; Vector Labs, Burlingame, CA, USA), rinsed, and incubated for 30 min in ABC (Vector Labs) solution. The reaction product was then visualized by incubation in 0.022 % 3-3′-diaminobenzidine (DAB; Sigma, St. Louis, MO, USA) and 0.003% hydrogen peroxide. All primary and secondary antisera incubations included 1% BSA and diluents containing PBS. Triton-X was added to the DβH primary antibody solution (0.2% for light microscopy, 0.05% for electron microscopy). All incubations were performed at room temperature with continuous agitation. For dual-label fluorescent experiments, tissue was incubated overnight in βAR 248 as described above, rinsed, and incubated with goat anti-rabbit IgG conjugated to Alexa 594 (1:200; Invitrogen, Carlsbad, CA, USA) for 1 h. The tissue was then rinsed, preincubated in 1% BSA and incubated overnight in either antibodies made in mouse either to GABA (1:2000; ICN Biochemicals, Costa Mesa, CA, USA) or CAMKII (1:200; Upstate, Lake Placid, NY, USA). The following day, the tissue was rinsed, and incubated in goat anti-mouse conjugated to Alexa 488 (1:200). Tissue sections for light and confocal microscopy were mounted on slides coated with gelatin. Tissue sections designated for LM analysis were dehydrated in a graded series of alcohol, defatted in xylene and coverslippped with Permount (Fisher Scientific). Fluorescent tissue was mounted and coverslipped with Prolong Gold (Invitrogen). The final preparations were examined on either a Nikon FXA and photographed with a Coolsnap digital camera (Roper Scientific, Trenton, NJ, USA) or a Zeiss LSM 310 and Leica TCS SP2 confocal microscope.
### Electron microscopic processing
Tissue sections designated for EM were processed as previously described (Farb and LeDoux, ). In brief, tissue sections containing the amygdala were incubated in 1% osmium tetraoxide/PB, dehydrated in a graded series of alcohols, stained en bloc in uranyl acetate, further dehydrated in acetone and subsequently flat-embedded in EMbed. Portions of the tissue containing the amygdala were cut and glued (Super Glue; Rancho Cucamonga, CA, USA) onto Beem capsules and placed at 60°C for 10 min. Photographs of the amygdala were taken and ultrathin sections (85 nm) were cut from the dorsolateral division of the LA (Figure A). Ultrathin sections were collected on 8–12 nickel grids and the tissue was examined on a JEOL 1200EX electron microscope. Photographs were taken using a Hammamatsu digital camera (AMT; Danvers, MA, USA). Electron micrographs were collected from the dorsolateral amygdala of four animals with the best morphological preservation. For each brain, ultrathin sections from at least two vibratome sections containing the AL were examined. Labeled terminals were identified by the presence of peroxidase reaction product within processes and were distinguished from preterminal axons by the presence of vesicles. Immunoreactive terminals without distinct membrane boundaries or whose peroxidase reaction product was too dense to distinguish between it and the post-synaptic density were not included in the analysis. Immunoreactive terminals were characterized as either forming or not forming synaptic contacts by the presence of a post-synaptic membrane specialization, intercleft filaments, and widened (10–20 nm) parallel spacing of plasma membranes (Peters et al., ). Labeled terminals with thickened post-synaptic densities and widened synaptic clefts were classified as asymmetric while terminals with thin post-synaptic densities and narrower synaptic clefts were identified as symmetric. Appositions were characterized by close membrane associations not separated by astrocytic processes, the lack of conventional synaptic clefts, intercleft material or dense specializations. Dendritic shafts were arbitrarily characterized as large (i.e. proximal) if their diameter was greater than 0.5 μm, or small (i.e., distal) if their diameter was less than 0.5 μm. Dendritic spines were smaller than dendrites and lacked mitochondria.
Light micrographs showing dopamine beta-hydroxylase (DβH) and beta-adrenergic receptors (βARs) in the amygdala . (A) Dark-field micrograph shows DβH in the amygdala. Trapezoid corresponds to area examined by electron microscopy. (B) Higher-magnification dark-field micrograph illustrates thin, varicose DβH axons. Arrowheads point to varicosities. (C,D) Low-magnification bright-field micrographs show βAR 248 (C) , and βAR 404 (D) are distributed homogenously throughout the amygdala. (E) Higher magnification shows that βAR 248 intensely labels somata and some proximal dendrites (arrowheads). The nuclei of some labeled cells are seen in some cells (arrows) but are difficult to distinguish in others. (F) Higher-power Nomarski optics show labeled astrocytic cell bodies (arrowheads) and their radiating processes (arrows). Also shown is an astrocytic process (asterisks) surrounding a blood vessel (BV). Scale bar = 100 μm in A – D , 50 μm in B, E , and F . BLA, basolateral amygdala; Ce, central amygdala; LA, lateral amygdala.
### Antibody specificity
In this study, we used mouse monoclonal antibodies directed against DβH, CAMKII, and GABA that have been characterized in previously published reports (Li et al., ; McDonald et al., ; Balcita-Pedicino and Rinaman, ; Schiltz and Sawchenko, ; Howorth et al., ). The pattern of DβH-immunoreactivity we describe is consistent with previously published reports in the amygdala complex (Fallon et al., ; Fallon and Ciofi, ; Roder and Ciriello, ; Asan, , ; Li et al., , ). Earlier studies have shown that within the LA, DβH is a specific marker for NE and does not label the other catecholaminergic biosynthetic enzymes, phenyl-methyl transferase (PNMT), the marker for adrenergic axons, or tyrosine-hydroxylase (TH), the marker for dopaminergic axons (Fallon et al., ; Fallon and Ciofi, ; Asan, ; Roder and Ciriello, ). Specifically, these studies have shown that PNMT immunoreactivity in the LA is scarce or nearly absent while TH-ir differs markedly and is non-overlapping. We also used rabbit antisera directed against β AR that was generated by using synthetic peptides corresponding to amino acids 248–256 (βAR 248) and 404–418 (βAR 404) of hamster lung β ARs (Dixon et al., ). The βAR 404 antiserum recognizes both the β - and β -subtypes (Strader et al., , ). The antiserum to βAR 248 (1:1K) was directed against the third cytoplasmic loop while the antiserum for βAR 404 (1:1K) was directed against the C-terminus of the receptor. The specificity of βAR 404 antisera has been previously characterized using Western blot (Strader et al., ) and immunoprecipitation of radiolabeled βAR (Strader et al., ). Preadsorption controls to the synthetic peptides using the exact correspondence to the antigens that were used to generate the βAR 248 and 404 antisera were also performed (Aoki, ). Antisera against βAR 248 and βAR 404 were generously provided by Dr. C.D Strader of Merck Sharp and Dohme Research Laboratories. Immunoreactivity was absent from tissue in which the primary antisera was omitted from the incubation solutions or the secondary antibody was mismatched to the primary antibody, e.g., anti-mouse IgG instead of the anti-rabbit IgG, and the tissue was reacted as described above.
## Results
### Light microscopy
#### The LA contains a dense plexus of dopamine beta-hydroxylase fibers
The pattern of DβH-immunoreactivity (-ir) in the amygdala has been described previously (Moore and Card, ; Fallon and Ciofi, ; Asan, , ; Li et al., , ). In brief, in both glutaraldehyde and acrolein-fixed tissue, DβH fibers appeared as a dense plexus and were distributed throughout the LA (Figure A). DβH fibers were fine and varicose and coursed through the amygdala in both dorsal–ventral and medial–lateral directions (Figure B).
#### Beta-adrenergic receptor immunoreactivity occurs throughout the LA
Immunoreactivity for both β AR antisera was distributed throughout the LA. By light microscopy, numerous cells were labeled. (Figures C,D). βAR 248 densely labeled neuronal perikarya and the proximal portions of their dendrites. In some LA cells, the reaction product rimmed the cytoplasm and the nucleus was well delineated whereas in other cells, the reaction product obscured the nuclei (Figure E). βAR 404-ir was observed in small cell bodies that appeared astrocytic: many labeled processes radiated from small perikarya (Figure F). Some labeled processes followed the contours of blood vessels.
### Confocal microscopy reveals βars are localized to both excitatory and inhibitory cells
To determine whether βAR248 was localized to specific cells types, we dually labeled tissue for βAR248 and CAMKII, a marker for excitatory, pyramidal-like cells in the LA (McDonald et al., ) and examined the tissue by confocal microscopy. We also dually labeled tissue for βAR 248 and GABA to establish whether GABAergic cells contain βARs. By confocal microscopy, βARs were localized to both LA excitatory and inhibitory cells (Figures A,B).
Confocal micrographs show that βAR 248 is localized to both excitatory and inhibitory LA cells . (A) Cells immunopositive for CAMKII (green) are also immunoreactive for βAR 248 (red). Arrows point to dually labeled cells while the asterisk indicates a cell singly labeled for βAR 248. (B) GABAergic cells (green) also contain βAR 248 (red). Arrows point to cells dually labeled for βAR 248 and GABA; asterisks indicate βAR 248 singly labeled cells.
### Electron microscopy
#### Most DβH terminals do not form synapses within single sections
Most of our EM analysis was performed on tissue fixed with acrolein since both the ultrastructure and membrane preservation were superior to tissue fixed with low levels of glutaraldehyde. Four hundred and ten DβH-labeled terminals were analyzed from tissue taken from the four animals with the best morphology. Analysis was performed on three animals perfused with acrolein and one animal perfused with glutaraldehyde. Ultrathin sections were collected from 3–4 vibratome sections from each animal for a total of 14 samples. DβH-labeled terminals were unmyelinated and varied in size from 0.4–1.5 μm. DβH terminals contained small, clear vesicles, though many terminals also contained 1–5 dense-core vesicles (Figures A–F). DβH terminals frequently contained mitochondria and some DβH-labeled axons appeared to follow the contours of blood vessels (Figure A). Frequently, the reaction product filled the axoplasm and obscured the morphological features of the terminal. Those terminals whose membranes were not intact due to the use of detergent were not included in the analysis. The vast majority of DβH terminals did not form synapses in a single plane of section (282/410 or 69%) (Figures A,E). About half the DβH terminals (223/410, or 54%) were directly apposed to unlabeled terminals (Figures B,C,E,F). In some instances (9/410 or 2%) possible axo–axonic contacts were observed: the plasma membranes of DβH terminals showed close and parallel alignment with the plasma membrane of unlabeled terminals and some intercleft density was present (Figure A). When DβH terminals did form synapses, most formed symmetric synapses (90/128 or 70%) (Figure A) and the vast majority of these (75/90, 83%) occurred on dendrites (Figures B,D) though some symmetric synapses were made on spines (13/90, 14%) and two occurred on somata (2/90, 2%). DβH symmetric synapses on spines usually occurred on the spine neck. When DβH terminals formed junctions on spines, almost one-quarter (9/37 or 24%) of these spines received another synapse from an unlabeled terminal. Most of these second synapses were asymmetric (8/9 or 89%) and were formed on the spine head. Almost one-third of synapse-forming DβH terminals made asymmetric synapses (38/128 or 30%). The majority of asymmetric synapses occurred on spines (24/38, or 63%) ( Figures 3F and 5B ), though many were formed on dendrites (14/38 or 37%) ( Figures 3C and 5B ).
Electron micrographs show DβH-terminals in LA . (A) A DβH – terminal (DBH) apposes a dendritic spine (sp) and an unlabeled terminal (ut) forming an asymmetric synapse (asterisks) onto a spine (sp). (B) A DβH-terminal forms a symmetric synapse (arrows) with a dendrite (d). (C) A DβH-terminal forms a synapse (arrowheads) onto a dendrite (d) that also receives a synapse (arrows) from an unlabeled terminal (ut). Glial processes (g and asterisk) are also shown. (D) A DBH-terminal forms a symmetric synapse (arrows) onto a dendritic (d) whose spine (sp) receives a synapse (arrowheads) from an unlabeled terminal (ut). Also shown is a glial process (g). (E) A DβH-terminal is apposed to an unlabeled terminal (ut ) that forms a symmetric synapse (arrows) on a dendrite (d). An unlabeled terminal (ut ) forms an asymmetric synapse (arrowheads) on the dendrite's spine (sp). Unlabeled glial processes (g and *) are also shown. (F) A DβH-terminal (DBH ) forms a synapse (arrows) onto a dendritic shaft (d ), whose spine (sp) receives an asymmetric synapse (arrowheads) from a second DβH-terminal (DBH ). DβH apposes unlabeled terminals (ut ) forming asymmetric synapses (arrowheads) with a spine (sp) and a dendrite (d ). Scale bars = 0.500 μm.
DβH-profiles and glial processes immunoreactive for βAR 404 are associated with blood vessels . (A) A DβH-axon (LAx) follows the contours of a blood vessel. The plasma membrane of the DβH-terminal (DβH) shows parallel alignment with the plasma membrane (small arrows) of an unlabeled terminal (ut). The basal lamina (asterisks) and an endothelial cell (End) separate the DβH-terminal from the blood vessel (BV). (B) A βAR 404-labeled glial process (LG) encircles a blood vessel (BV), which is bounded by an endothelial cell (end) and the basal lamina (asterisks). The βAR 404 astrocyte apposes unlabeled terminals (ut), one of which forms an asymmetric synapse (arrowheads) onto a dendritic spine (sp). Scale bars = 0.500 μm.
Graphs show synaptic targets and specializations made by DβH terminals that form synapses . (A) The vast majority of DBH symmetric synapses occurred on dendrites but a small proportion were made onto spines and somata. (B) Most DBH asymmetric synapses occurred on spines though some were formed on dendrites.
### βar immunoreactivity
#### βAR 248 is frequently localized to post-synaptic densities
By electron microscopy, βAR 248-ir was localized to neuronal perikarya, large and small dendritic shafts, dendritic spines, some axon terminals, and astrocytic processes (Figures A–C). Within dendritic shafts, reaction product rimmed the microtubules and the mitochondrial membranes, and in both shafts and spines was frequently concentrated along the plasma membranes and at post-synaptic densities (Figures A–C). βAR 248-ir axon terminals forming asymmetric synapses were occasionally observed (Figure B). βAR 248-ir was seen in perikarya with the morphological features of both inhibitory, e.g., invaginated nuclei and abundant cytoplasm, and excitatory cells, e.g., large nuclei and a thin rim of cytoplasm (Ribak and Seress, ; Farb et al., ). The subcellular distribution of βAR 248 immunoreactivity was consistent with previously published studies (Aoki, , ).
Both LA neurons and glia are immunoreactive (ir) for βAR . (A) Unlabeled terminals (ut) form asymmetric synapses onto βAR 248-ir spines (LSp). The spinous portion (LSp) of a dendrite (d) is βAR 248-ir. (B) An unlabeled terminal (ut) synapses (arrowheads) onto a βAR 248-ir spine (LSp). βAR 248-labeled terminals (LT ) synapse (arrowheads) onto unlabeled spines (usp). (C) The βAR 248 reaction product in a dendrite (LD) is concentrated at the synapse (arrowheads) formed by an unlabeled terminal (ut) but also rims the microtubules (mt) and mitochondria (m). Also shown are a βAR 248-ir terminal (LT) and a small labeled dendrite (LD). (D) βAR 404-ir glial processes (LG, asterisks) are interposed between two unlabeled terminals (ut ). Also shown are an unlabeled terminal (ut ) synapsing (arrowheads) onto a βAR 404 ir spine (LSp) and a βAR 404-ir terminal (LT) synapsing (arrowheads) onto an unlabeled spine (usp). (E) A βAR 404-ir glial process (LG) surrounds a terminal (ut) that synapses (arrowheads) onto a spine (usp). Asterisks denote βAR 404-ir along the plasma membrane. (F) A βAR 404 ir glial process (LG, asterisks) surrounds an unlabeled terminal (ut) that forms an asymmetric synapse (arrowheads) onto a βAR 404-ir spine (LSp). Scale bars = 0.500 μm.
#### βAR 404 is predominantly localized to glial processes
Ultrastructural examination revealed that βAR 404 was predominantly localized to glial perikarya and processes but some neuronal processes were also immunoreactive (Figures D–F). Glial perikarya were distinguished from neuronal perikarya by the presence of filamentous organelles or glycogen granules whereas glial processes were recognized by their irregular contours and scarcity of organelles. When βAR 404 immunoreactivity occurred in large glial processes, the immunoperoxidase product rimmed the glial vesicles and mitochondria but was frequently concentrated along the plasma membranes (Figure E). Labeled glial processes frequently ensheathed or directly apposed unlabeled terminals forming asymmetric terminals (Figures D–F). Often, small labeled glial processes were interposed between unlabeled axon terminals (Figure D). Some axon terminals and dendritic shafts and spines were also immunoreactive for βAR 404 (Figures D,F). Large glial processes intensely immunoreactive for βAR 404 were sometimes apposed to the basal lamina and endothelial cells that bounded blood vessels (Figure B).
## Discussion
The present study used immunocytochemistry to identify and characterize: (1) terminals that contain norepinephrine, and (2) the cellular and subcellular distribution of βARs in the LA. The results show that most DβH terminals within the LA do not form synaptic junctions, but when they do, most synapses occur on dendritic shafts and a small proportion are formed on dendritic spines. While the majority of DβH synapses are symmetric, asymmetric synapses are also formed and most of these occur on spines. βARs are localized to both neurons and glial cells in the LA, and within neurons, βARs are localized to both excitatory and inhibitory cells and are frequently concentrated at the PSDs of dendritic shafts and spines. These results provide the morphological basis for understanding the role that NE and βARs play in modulating synaptic transmission within the LA.
### Methodological considerations of βar antisera
The βAR 404 antibody we used in this study was generated by using synthetic peptides directed against the amino acid sequences of hamster lung β ARs (Dixon et al., ) but recognizes both the β - and β -subtypes (Strader et al., , ). This antibody has been extensively characterized using Western blot (Strader et al., ), immunoprecipitation of radiolabeled βAR (Strader et al., ) and preadsorption to the synthetic peptides using the exact correspondence to the antigens that were used to generate the antisera (Aoki, ). Immunolabeling with the antibody directed against the third intracellular loop, βAR 248, was consistent with previous studies, in which a monoclonal antibody directed the third intracellular loop was used (Aoki et al., ; Aoki, ). Though the results from several autoradiographic (Palacios and Kuhar, ; Minneman et al., ; Rainbow et al., ; Johnson et al., ) and in situ (Asanuma et al., ; Abraham et al., ) studies report different levels of β - and β -ARs across various brain regions, each of these studies demonstrated the presence of either β - or β -ARs in the amygdala complex and adjacent areas. Though the pattern of βAR immunolabeling we observe is homogenous compared to the distinct patterns reported in the autoradiographic and in situ studies, the antibodies we used likely recognize receptors in the perikaryal cytoplasm undergoing sequestration, desensitization, degradation or synthesis, as well as those that are ligand-binding (Strader et al., , ; Zemcik and Strader, ). Additionally, the antibodies we used recognize both the β - and the β -subunits, though they were directed against the β subunit, and recognize a greater population of cells, compared to those identified by in situ . It is thus not unexpected that the distribution of βARs identified by immunolabeling differs from patterns seen by other methods.
### Synaptic characterization of dβh within the la
Our results show that most DβH terminals within single planes of section do not form synapses but instead form non-junctional appositions with dendrites or unlabeled terminals. Though these findings are consistent with previous studies showing the non-junctional nature of DβH terminals in the amygdala (Asan, , ) and other brain regions (Descarries et al., ; Seguela et al., ; Aoki et al., ), it is likely that we underestimated the degree to which DβH terminals form synapses. Several factors might account for our failure to detect these synapses: the use of detergent to permeabilize membranes and improve penetration of the DβH antibody, dense DAB reaction product that obscures the morphological features of labeled terminals, and the thin, small size of synapses that may be overlooked without serial section examination. Though we followed some non-junctional DβH terminals for 2–5 sections, this series was too small to establish whether these DβH terminals ultimately formed synapses. Our results, showing that approximately 30% of DβH terminals in the LA form synaptic junctions, is higher than what has been reported for the BLA (Asan, ) and may represent regional differences within the amygdala complex or methodological differences attributable to our use of a stronger fixative that results in better preservation of membrane ultrastructure, enabling detection of a greater number of synapses. Our study showing that DβH terminals form both symmetric and asymmetric synapses, is consistent with previous studies done in basal amygdala (Asan, ; Li et al., ) and other brain regions (Seguela et al., ; Aoki et al., ). Though Li et al. ( ) reported that the proportion of symmetric and asymmetric synapses was almost equivalent in the BLA, our results showing that DβH terminals form approximately twice as many symmetric synapses than asymmetric synapses, are similar to what Asan ( ) found in BLA. Though asymmetric junctions have been correlated with glutamatergic transmission and symmetric synapses with GABAergic transmission (Peters et al., ), the distinction in catecholaminergic systems is less clear and may instead reflect the target dendrite and the synaptic machinery present on that target. For example, though DβH terminals were more likely to form symmetric rather than asymmetric synapses in LA, most asymmetric synapses were formed on dendritic spines, which receive most of the glutamatergic synapses in the LA (Farb et al., ). Though serial section analysis shows that just 7% of dendritic spines in LA receive more than one synapse, and only 2% of LA spines receive both an asymmetric and a symmetric synapse (Ostroff et al., ), we found that when DβH terminals synapse on spines, approximately 25% of these spines also receive synapses from other unlabeled terminals and most of these junctions are asymmetric. This observation and the prevalence of βAR labeling at the PSDs of asymmetric synapses in dendritic spines suggest that NE may modulate glutamatergic transmission at LA spines. These results are consistent with electrophysiological findings showing that NE modulates glutamatergic neurotransmission in amygdala (Huang et al., , , ) and activation of βARs enhances synaptic transmission in amygdala (Gean et al., ; Huang et al., , ), hippocampus (Raman et al., ), and prefrontal cortex (Ji et al., ). Preliminary data from our lab indicate that LA dendritic spines and shafts that are immunoreactive for βARs receive synaptic contacts from axon terminals originating either from the acoustic thalamus or cortex (unpublished observations), pathways known to be glutamatergic (Farb et al., ; Farb and LeDoux, , ). Thus, NE may modulate synaptic transmission of these sensory pathways to the LA either by its convergence onto the same dendritic shafts and spines as cortical or thalamic axons or via activation of βARs on these dendritic processes.
### βars are present in both excitatory and inhibitory cells in the la
Our confocal and EM findings, showing that the neuronal form of the βAR is localized to both excitatory and inhibitory LA cells, are consistent with anatomical and pharmacological studies showing that βARs are present on pyramidal and GABAergic cells in hippocampus (Milner et al., ; Hillman et al., ; Cox et al., ). Within amygdala, in vitro electrophysiological studies have shown that the activation of βARs on pyramidal cells results in enhancement of excitatory transmission by NE whereas blockade of βARs by propranolol reduces excitatory transmission (Gean et al., ; Huang et al., , ; Ferry et al., ; Buffalari and Grace, ) and blocks late LTP (Johnson et al., ) on these cells. While anatomical and in vitro studies within the amygdala have shown a relationship between NE and GABA, they have yet to demonstrate whether this association is attributable to the activation of βARs on GABA cells in the LA. For example, GABAergic cells in BLA receive synaptic contacts from DβH terminals (Li et al., ), indicating that NE may directly modulate GABAergic transmission. Additionally, application of NE in LA slices, suppresses feed-forward GABAergic inhibition of projection neurons (Tully et al., ). In vivo, when the βAR antagonist propranolol is administered i.p., the memory-enhancing effects of the GABA antagonist bicuculline are blocked, while clenbuterol, the βAR agonist, blocks the memory-impairing effects of the GABA agonist muscimol (Introini-Collison et al., ). Our data showing that βARs are present on GABA cells in the LA provides a framework for understanding these physiological and behavioral findings.
### βars at non-synaptic sites
The large proportion of non-junctional appositions formed by DβH terminals in amygdala may reflect NE release via volume transmission (Descarries et al., ; Agnati et al., ) or non-synaptic mechanisms. Consistent with these ideas are dual-label ultrastructural studies in cortex and hippocampus showing that dendrites that are immunoreactive for βAR were near catecholaminergic axons but rarely in direct contact with them, though astrocytic processes were (Aoki et al., ; Aoki, ; Aoki and Pickel, ; Milner et al., ). Our findings, showing extensive βAR immunoreactivity of glial processes, support the idea that NE might act indirectly through astrocytic processes and are consistent with previous ultrastructural studies (Aoki et al., , Aoki , ; Milner et al., ) and in vivo and in vitro binding studies from various brain areas showing βARs expression or binding in astrocytes (Burgess and McCarthy, ; Lerea and McCarthy, ; Stone and John, ). Activation of astrocytic βARs may modulate glutamatergic transmission at excitatory synapses via close appositions or glial ensheathment of these synapses (Shao and McCarthy, ). Additionally, astrocytic βARs may modulate gap junction permeability, release glucose for energy metabolism, or play a role in cytoskeletal rearrangements that accompany neuronal plasticity (for reviews, see Gibbs et al., ; Giaume et al., ).
## Conclusions
Together, results from this study suggest that norepinephrine-containing terminals in the LA may engage in non-synaptic transmission in the LA. The presence of β ARs in both excitatory and inhibitory neurons suggests that NE has a prolific role in the modulation of synaptic transmission in LA. Further, the prevalence of βARs in glial cells adds a further dimension to the role of NE in modulating synaptic transmission in LA since glial cells may play a role in regulating excitatory transmission. These data provide an anatomical foundation for interpretation of physiological and behavioral studies of the role of NE in the amygdala.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Tailoring treatments to the specific needs and biology of individual patients—personalized medicine—requires delineation of reliable predictors of response. Unfortunately, these have been slow to emerge, especially in neuropsychiatric disorders. We have recently described a real-time functional magnetic resonance imaging (rt-fMRI) neurofeedback protocol that can reduce contamination-related anxiety, a prominent symptom of many cases of obsessive-compulsive disorder (OCD). Individual response to this intervention is variable. Here we used patterns of brain functional connectivity, as measured by baseline resting-state fMRI (rs-fMRI), to predict improvements in contamination anxiety after neurofeedback training. Activity of a region of the orbitofrontal cortex (OFC) and anterior prefrontal cortex, Brodmann area (BA) 10, associated with contamination anxiety in each subject was measured in real time and presented as a neurofeedback signal, permitting subjects to learn to modulate this target brain region. We have previously reported both enhanced OFC/BA 10 control and improved anxiety in a group of subclinically anxious subjects after neurofeedback. Five individuals with contamination-related OCD who underwent the same protocol also showed improved clinical symptomatology. In both groups, these behavioral improvements were strongly correlated with baseline whole-brain connectivity in the OFC/BA 10, computed from rs-fMRI collected several days prior to neurofeedback training. These pilot data suggest that rs-fMRI can be used to identify individuals likely to benefit from rt-fMRI neurofeedback training to control contamination anxiety.
## Introduction
Dysregulation of anxiety is a core component of many neuropsychiatric conditions. Obsessive-compulsive disorder (OCD) is characterized by intrusive obsessions, which are often associated with anxiety, and with repetitive compulsions that seek to control that anxiety (Jenike, ). One common presentation of OCD is characterized by extreme contamination anxiety, often triggered by thoughts or images of, or contact with, potential contaminates such as dirt, body secretions, or mold (Bloch et al., ). Improving control of contamination anxiety is a key step in improving the quality of life for many individuals with OCD.
Treatments for contamination anxiety and for OCD exist, but none are universally effective (Franklin and Foa, ). For individuals who do not respond to standard behavioral and pharmacological treatments, interventions that more directly modulate the specific brain regions whose dysfunction is implicated in the disorder may be of benefit. In extreme cases, this modulation is sometimes done through invasive procedures such as deep brain stimulation (Greenberg et al., ). Targeted brain modulation using neurofeedback via real-time functional magnetic resonance (rt-fMRI) may prove to be an alternative (Scheinost et al., ).
rt-fMRI neurofeedback involves monitoring the blood oxygenation level dependent (BOLD) signal, a measure of brain activity, and providing immediate feedback to the subject showing them how specific brain activity patterns are changing over time. This form of feedback can facilitate learned control over brain activity and associated behaviors (Sulzer et al., ; Ruiz et al., ). Neurofeedback training has shown promise as a potential treatment in several clinical disorders including addiction (Hanlon et al., ; Li et al., ), tinnitus (Haller et al., ), stroke (Sitaram et al., ), depression (Linden et al., ; Young et al., ), Parkinson’s Disease (Subramanian et al., ), and schizophrenia (Ruiz et al., , ). rt-fMRI neurofeedback can produce changes in brain function (Hampson et al., ; Harmelech et al., ) and related behaviors (Shibata et al., ). In individuals with significant but subclinical contamination anxiety, we have shown that neurofeedback of activity in the orbitofrontal cortex (OFC) and anterior prefrontal cortex Brodmann area (BA) 10 can reorganize functional brain networks associated with anxiety and reduce the anxiety produced by contamination-related stimuli (Scheinost et al., ).
Clinically, a trial of an intervention that ultimately proves ineffective carries substantial cost, in both time, resources, and ongoing patient suffering. How best to match an individual to an intervention is therefore a crucially important question. Predictors of response that can help with treatment selection in neuropsychiatric conditions such as OCD would be of enormous clinical value but have been slow to emerge.
Here, we ask whether resting-state fMRI (rs-fMRI) can predict response to neurofeedback training and, thus, potentially guide treatment selection in the future. Previous research suggests that imaging-based biomarkers can be used to predict performance with a brain-computer interface (Halder et al., ). rs-fMRI, in particular, provides a great opportunity for identifying biomarkers to aid clinical decisions, given that it can be collected in clinical populations without requiring any task performance and yet provides a wealth of information about brain function (Constable et al., ; Lee et al., ). To investigate whether brain connectivity at rest can predict reduction in contamination anxiety induced during a neurofeedback intervention, we correlated a voxel-wise measure of functional connectivity, computed from rs-fMRI collected prior to neurofeedback training, with behavioral response to the neurofeedback intervention in a cohort of healthy subjects with subclinical contamination anxiety (Scheinost et al., ). We then examined whether a similar relationship existed in a small cohort of patients.
## Methods
Data were from studies performed at Yale University School of Medicine, New Haven, CT. All protocols were reviewed and approved by Human Research Protection Program at Yale University. Written informed consent was obtained. All scans were obtained and analyzed at Yale University.
### Subjects
Two cohorts of subjects were used in this study. The first cohort has been described previously (Scheinost et al., ) and consisted of 10 subjects without any clinical diagnosis of OCD, but with high levels of contamination anxiety. Only the 10 subjects who received true neurofeedback in our previous study—not the 10 who received sham neurofeedback in the control condition—are included in the present analysis. The second cohort consisted of five OCD patients with moderate symptom severity (Table ) and prominent contamination obsessions.
Clinical characteristics and symptom improvement in five OCD patients who underwent rt-fMRI biofeedback .
MDD—major depressive disorder. Panic D/O—panic disorder, with agoraphobia. GAD—generalized anxiety disorder. SUD—substance use disorder (in remission). BDD—body dysmorphic disorder. * taken occasionally, as needed. rs-fMRI data collected prior to neurofeedback and used in connectivity analysis; see Figure .
### Neurofeedback training
Healthy subjects and OCD patients received neurofeedback training following a previously detailed protocol (Hampson et al., ). Of the five OCD patients, the first two underwent only a single neurofeedback session, without pre- or post-neurofeedback resting-state scans. All other individuals participated in four separate MRI scanning sessions, spaced several days apart. In the first session, rs-fMRI data were collected and a functional localizer was used to identify the target area of the OFC/BA 10 region to be used for neurofeedback. The second and third sessions involved rt-fMRI neurofeedback training based on the target OFC/BA 10 region. A final session (not of relevance to this work) involved collecting post-intervention rs-fMRI data. The rs-fMRI data were always collected before any other functional scans in a given session to avoid possible effects of previous task on the rs-fMRI data.
The overlap of the target area for feedback for all 15 subjects is shown in Figure . Overlap was calculated by (1) smoothing the target region of each individual with a 6 mm Gaussian smoothing kernel to account for differences in functional anatomy and registration errors; (2) warping the target regions to a common reference; and (3) averaging across subjects the likelihood of a voxel being included in the target region.
Overlap of target regions for neurofeedback. All subjects received neurofeedback from a region in the OFC/BA 10 (Hampson et al., ). These target regions were determined on individual basis from a functional localizer task allowing for differences in individual functional anatomy. The percent overlap of all these target regions is shown on a template brain using Radiological convention (left is on the right for axial slices). Warmer colors indicate that the voxel was included in a greater number of individual target regions.
### Behavioral measures
Behavioral measures of control over contamination anxiety (for the first cohort) and clinical measures (for the second cohort) were collected before and after neurofeedback training. The pre-intervention assessment data was collected before the first neurofeedback session (second overall imaging session), immediately prior to the start of neurofeedback training. The post-intervention assessment data was collected several days after the completion of neurofeedback training, either in a separate session with no imaging (for the first two OCD patients) or in conjunction with the fourth fMRI session (for all other patients). Finally, midpoint assessment data was collected in between the first and second neurofeedback sessions for the subjects who received two sessions of neurofeedback.
For the healthy subjects, with subclinical contamination anxiety we used a behavioral measure designed to assess the subjects’ ability to control their anxiety. Subjects were instructed to try to control their anxiety while viewing 25 contamination-related images and to indicate their experienced anxiety for each image on a 1–5 scale. A rating of one indicated the least anxiety and a rating of five indicated the most anxiety. The ratings for the 25 contamination-related images were then averaged yielding a single measure of anxiety. Different sets of images were used before and after the intervention, but the sets were designed to induce similar levels of contamination related anxiety and piloted to verify that they were balanced in this respect (Hampson et al., ).
For the patients, we administered a modified version of the Yale–Brown Obsessive Compulsive Scale (Y-BOCS), in which they were instructed to report on their symptoms over the last 3 days, rather than over the past week as in the traditional Y-BOCS (Goodman et al., , ). The Y-BOCS ranges from 0–40, with higher scores representing more severe symptoms, and measures the frequency, intrusiveness, and distress associated with obsessions and compulsions. Scores in the mid-twenties, as these patients had (Table ), correspond to moderate to severe disease.
For both groups, change in behavior measures were calculated as score prior to neurofeedback minus score after neurofeedback, such that a positive change indicates an improvement in anxiety.
### Imaging parameters
All imaging was done on a 1.5-T Siemens Sonata scanner (Siemens Medical Systems, Erlangen, Germany). A sequence designed to optimize signal in the OFC was used for all functional data collection (repetition time = 2000 ms, echo time = 30 ms, flip angle = 80, bandwidth = 2604, 200 mm field of view for 3.1 mm isotropic voxels, 31 axial-oblique slices covering almost the whole cerebrum and most of the cerebellum). Two 5 min resting data runs were collected.
### Resting-state connectivity
Images were preprocessed using a previously detailed pipeline (Hampson et al., ). All images were slice time and motion corrected using SPM. Unless otherwise specified, all further analysis was performed using BioImage Suite (Joshi et al., ). Several covariates of no interest were regressed from the data including linear and quadratic drift, six rigid-body motion parameters, mean cerebrospinal fluid (CSF) signal, mean white-matter signal, and mean global signal. The data were low-pass filtered via temporal smoothing with a 0 mean unit variance Gaussian filter (approximate cutoff frequency = 0.12 Hz). Finally, a gray matter mask was applied to the preprocessed data so that only voxels in the gray matter were used in subsequent calculations. After preprocessing, all resting-state runs were concatenated and the connectivity for each voxel was then calculated in each subject’s individual brain space.
The gray and white matter and CSF masks were defined on a template brain (Holmes et al., ), and warped to individual subject space using a series of transformations, described below. The gray matter mask was dilated to ensure full coverage of the gray matter after warping into individual subject space. Regions that were not included in all subjects’ data (for e.g., the bottom of the cerebellum) were excluded from analysis. Likewise, the white matter and CSF masks were eroded to ensure only pure white matter or CSF signal were regressed from the data.
Global functional connectivity of each voxel was measured from rs-fMRI data using the network theory measure degree (Bullmore and Sporns, ) as previously described (Martuzzi et al., ). The BOLD time course for each voxel was correlated with every other voxel in the gray matter. Two voxels were considered connected if correlation of their timecourses was greater than r = 0.25; the degree of each voxel was defined as the number of such connections. The process was repeated for every voxel in the gray matter. Each subject’s degree map was normalized by subtracting the mean across all voxels and dividing by the standard deviation across all voxels. This normalization has been shown to reduce the impact of confounds related to motion (Yan et al., ).
To facilitate comparisons of imaging data, all degree maps were spatially smoothed with a 6 mm Gaussian filter and then warped to a common template space through the concatenation of a series of linear and non-linear registrations, as previously described (Scheinost et al., ). All transformations were computed using the intensity-based registration algorithms in BioImage Suite (Papademetris et al., ).
### Evaluating the relationship between response to intervention and rs-fMRI data
To identify which brain regions predicted response to neurofeedback training, we related the rs-fMRI data acquired before any neurofeedback training with changes in the behavioral measure of control over contamination anxiety (for the healthy subjects) and changes in clinical severity (for the patients). For the healthy subjects, we performed a data-driven, whole-brain analysis by correlating the change in control of anxiety with the degree maps in a voxel-wise manner. Significance was assessed at a p < 0.05 level after correcting for multiple comparisons across the gray matter via AFNI’s AlphaSim program. From this voxel-wise analysis, we defined a region of interest (ROI) that showed significant effects in the healthy subjects to explore whether this finding translated to the smaller cohort of OCD patients. For the three OCD patients on whom pre-neurofeedback rs-fMRI was collected, degree averaged over all voxels in this ROI was extracted and related to changes in Y-BOCS scores.
## Results
### Imaging predictors of behavior in subclinically anxious subjects
As reported previously (Scheinost et al., ), healthy subjects with subclinical contamination anxiety showed a significant ( p < 0.05) increase in control over anxiety after neurofeedback training. Whole-brain connectivity analysis revealed a single significant cluster ( p < 0.05 corrected; MNI coordinate of peak voxel: 0, 66, −4, max t -value = 5.84, cluster size = 5857 mm ) in which degree prior to neurofeedback training was significantly correlated with improved control over anxiety (Figure ). This cluster was located in the OFC/BA 10 target region. Subjects with the highest connectivity in this region prior to neurofeedback training exhibited the most improvement in post-treatment anxiety. A scatterplot of the average connectivity change in this region vs. the change in control of anxiety is shown in Figure . As the choice of threshold used to consider whether two voxels are connected can impact connectivity results (Scheinost et al., ), we repeated this analysis over a range of thresholds (0.10 < r < 0.65). This produced no qualitative change in the findings. Additionally, as motion has been shown to confound functional connectivity results, average frame to frame displacement was calculated for each group (Van Dijk et al., ). Motion was not correlated with improved control of anxiety ( r = 0.18, p > 0.60) and adding motion as a covariate in the group analysis did not change the presented results.
Correlation of improved control over anxiety and rs-fMRI. (A) Subjects with the highest connectivity in the OFC/BA 10 (MNI coordinate of peak voxel: 0, 66, −4) prior to neurofeedback training had the largest improvement in control over anxiety over the course of the intervention. Results shown using Radiological convention at p < 0.05 level, corrected for multiple comparisons. (B) Scatterplot showing improved control over anxiety and pre-neurofeedback rs-fMRI.
### Clinical improvement after neurofeedback in subjects with OCD
Five patients with moderate-to-severe OCD and prominent contamination symptoms underwent one or two sessions of neurofeedback (Table ). All five tolerated the procedure well and exhibited reduced symptoms, as evaluated by the Y-BOCS several days after the last neurofeedback session. Average symptom improvement was 20%. Of the OCD patients, the three with the greatest symptom improvements also had a co-diagnosis or a history of major depressive disorder (MDD). Demographic and clinical details are given in Table .
### Imaging predictors of clinical improvement
Next, we tested whether a similar relationship between connectivity and behavioral improvements would be found in OCD patients. Pre-neurofeedback rs-fMRI was not measured on the first two subjects; this analysis was therefore performed only on the three subjects who underwent the full two-session neurofeedback protocol. To maximize power in this very limited dataset, we used the OFC/BA 10 region defined in the first cohort as an a priori ROI, Average degree in this ROI prior to neurofeedback training was related to clinical improvement for the three patients. Consistent with the pattern seen in the healthy subjects, a strong linear relationship was observed ( r = 0.99). Thus, in both groups, increased connectivity in the OFC/BA 10 measured from rs-fMRI data collected prior to neurofeedback training was associated with greater behavioral improvements.
## Discussion
Advances in understanding individual differences motivate a new approach to health care in which treatment is tailored to the specific needs and biology of an individual patient. This “personalized medicine” approach has been endorsed by the National Institute of Mental Health, NIMH, but its adoption depends critically on our ability to identify which patients are likely to respond to which interventions. rs-fMRI holds great promise as a tool for providing this information. It is easy to collect, does not require patients to perform any difficult tasks, and yet is a rich source of potentially clinically relevant information about brain function (Constable et al., ; Lee et al., ).
In a pilot study, we demonstrate, for the first time, that rs-fMRI can be a useful tool to predict response to neurofeedback training via rt-fMRI. After receiving two sessions of neurofeedback training, healthy subjects showed improved control over anxiety and OCD patients showed a reduction in OCD symptom severity. For both groups, these behavioral improvements were strongly correlated with the pre-intervention level of whole-brain connectivity in the anterior prefrontal cortex.
The resting state functional connectivity analysis used in this study was unbiased by a priori expectations regarding regions of interest. Therefore, it is striking that the region that emerged from our whole-brain analysis as most relevant for predicting improvements in contamination anxiety was in our target area of the OFC/BA 10. Taken together with a large body of data highlighting the importance of the OFC and anterior prefrontal cortex in obsessive-compulsive symptoms (Swedo et al., ; Chamberlain et al., ; Menzies et al., ; Harrison et al., ; Sakai et al., ; Anticevic et al., ; Beucke et al., ), this gives us confidence that we are targeting a biologically relevant brain area.
Notably, OFC/BA 10 connectivity predicted the response to the intervention in both healthy subjects and OCD patients, suggesting a shared neurobiological mechanism for improved control over contamination anxiety across groups. It is possible that the phenomenon of contamination anxiety is a dimensional construct, differing in a quantitative rather than a qualitative sense in patients when compared to healthy individuals. Supporting this view are previous reports of OFC/BA 10 activations to contamination related imagery in both healthy subjects and OCD patients (Mataix-Cols et al., , ).
To the extent that neurobiology of a phenomenon is shared across patients and healthy subjects, interventions developed in the healthy group are likely to translate into the patient population. In this particular intervention, based on our preliminary patient data, translational potential appears high. A variety of other applications of rt-fMRI neurofeedback trainings have been developed in healthy populations (Hampson et al., ; Shibata et al., ; Chiew et al., ; Garrison et al., ). It will be interesting to see how well the findings from these studies translate into clinical populations. If the dimensional approach implicit in NIMHs Research Domain Criteria is an accurate description of pathological brain dysfunction, many of these studies may successfully translate into the respective patient groups.
An important consideration for predictive validity is the reliability of rs-fMRI. Overall, graph theory measures have been shown to be reliable (Telesford et al., ; Braun et al., ) and, in particular, voxel-wise degree has shown good test-retest reproducibility across different sites and scanners (Tomasi and Volkow, ). While generally reliable, a variety of factors can reduce the predictive power of rs-fMRI. Medications and other drugs such as caffeine can alter connectivity patterns (Rack-Gomer et al., ; Martuzzi et al., ). Sleep also changes connectivity patterns (Tagliazucchi et al., ) which can be an issue if subjects are falling asleep and not reporting it. Finally, factors related to subject comfort such as hunger may reduce data quality and prediction accuracy due to motion artifacts and effects on subject compliance. The degree to which all these variables are controlled is likely to affect the power of future studies to identify clinically relevant biomarkers that predict treatment response.
The major limitation of this pilot study is the small number of subjects, particularly in the patient group, in which we only had three subjects with resting data. Although the finding in the healthy subject group is statistically significant, the finding in the patient group must be considered preliminary. However, the tight correspondence between connectivity and intervention response in our modest clinical sample, and its similarity to the relationship seen in healthy subjects, are promising. Future studies are needed to rigorously examine whether this biomarker is an effective predictor of response in the clinical group. A large study that can examine possible modulating variables would be particularly valuable. For example, the data in our small sample suggest that patients with a current co-diagnosis or a history of MDD show the greatest improvement in clinical symptoms, but we were unable to investigate this given our limited data in the patient group. A study with the power to test that possibility could yield interesting insights.
## Conclusion
These pilot data provide evidence that rs-fMRI connectivity can be used to identify individuals likely to benefit from rt-fMRI neurofeedback interventions for training control over contamination anxiety. Specifically, we have identified a biomarker that may be useful in developing personalized treatment programs in patients with OCD. More generally, these findings illustrate the potential utility of rs-fMRI data for identifying biomarkers of treatment response and thereby facilitating a personalized medicine approach to treating mental illness.
## Conflict of interest statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
This study aimed to investigate the effects of a long-term resistance exercise intervention on executive functions in healthy elderly males, and to further understand the potential neurophysiological mechanisms mediating the changes. The study assessed forty-eight healthy elderly males randomly assigned to exercise ( n = 24) or control ( n = 24) groups. The assessment included neuropsychological and neuroelectric measures during a variant of the oddball task paradigm, as well as growth hormone (GH), insulin-like growth factor-1 (IGF-1), and homocysteine levels at baseline and after either a 12 month intervention of resistance exercise training or control period. The results showed that the control group had a significantly lower accuracy rate and smaller P3a and P3b amplitudes in the oddball condition after 12 months. The exercise group exhibited improved reaction times (RTs), sustained P3a and P3b amplitudes, increased levels of serum IGF-1, and decreased levels of serum homocysteine. The changes in IGF-1 levels were significantly correlated with the changes in RT and P3b amplitude of the oddball condition in the exercise group. In conclusion, significantly enhanced serum IGF-1 levels after 12 months of resistance exercise were inversely correlated with neurocognitive decline in the elderly. These findings suggest that regular resistance exercise might be a promising strategy to attenuate the trajectory of cognitive aging in healthy elderly individuals, possibly mediated by IGF-1.
## Introduction
While life expectancy has been increasing in developed countries, one risk associated with a rapid growth in aged populations is that the number of people suffering from cognitive impairments and dementia could increase, due to age-related deteriorations in an array of cognitive processes involving central executive functions, attention, and short- and long-term memory (Anderson and McConnell, ). This can be attributed to gradual declines in physical activity levels (Kimura et al., ), the structural integrity of the brain (e.g., frontal, parietal, and temporal lobes) (Jernigan et al., ; Raz et al., ), and the secretion of growth factors (e.g., growth hormone (GH) and insulin-like growth factor-1 (IGF-1)) from the neurochemical system (Sonntag et al., ), which inevitably occur with aging. Cognitive impairment is closely related to decreases in both living-independence and general health. Therefore, determining how to counteract neurocognitive decline in order to reduce the costs associated with geriatric care is becoming an important issue for many public health systems around the world.
Recently, researchers have focused on the role of the GH/IGF-1 axis in attenuating age-related neurocognitive decline. Elderly individuals experience a fall in the ability to secrete GH, and a subsequent decrease in its secondary mediator (i.e., IGF-1), which is primarily produced in the liver (90%) but also by other cell types in the brain and vasculature (Sonntag et al., ). Both substances cross the blood-brain barrier and bind to receptors in the central nervous system to stimulate the growth of glial cells, myelination, and neurons: therefore, age-related decreases in serum GH and IGF-1 levels might contribute to cognitive decline in the elderly (Sonntag et al., ). A number of studies have attempted to understand the relationship between neuropsychological test performance and serum GH and IGF-1 levels in the elderly. Although some argue that there is no relationship between serum IGF-1 and GH concentrations and specific neurocognitive functions (e.g., GH vs. event-related potential (ERP) N2b component; IGF-1 vs. delayed recall and Brus reading) (Papadakis et al., ; Aleman et al., ; Quik et al., ), there is growing evidence for a significant association between neuropsychological measures (e.g., GH vs. reaction times (RTs) in selective attention and short-term memory; IGF-1 vs. speed of information processing) and serum IGF-1 and GH levels in elderly populations (Rollero et al., ; Aleman et al., ; Kalmijn et al., ; Dik et al., ; Quik et al., ). These cross-sectional studies, however, do not provide sufficient evidence for effective interventions to prevent or reduce the risk of cognitive decline in the elderly. The inconclusive findings of such studies regarding the relationship between serum GH and IGF-1 levels and age-related neurocognitive decline in healthy elderly individuals mean that more research needs to be carried out that uses long-term exercise interventions to clarify the role of growth factors in this phenomenon.
Exercise is a lifestyle factor crucial to the prevention or delayed onset of mild cognitive impairment in later life (Smith et al., ). The potential mechanisms for this effect include exercise-induced increases in the levels of nerve growth factors, such as GH and IGF-1, as these play central roles in the health of neurons in the brain. Previous studies investigating the effect of resistance training on cognition mostly discussed the potential mechanisms using GH, IGF-1, and homocysteine, possibly due to the fact that these biomarker secretions are exercise-sensitive, since they are characterized by different physiological and metabolic demands (Cassilhas et al., , ; Seo et al., ). However, several experimental studies suggest that basal levels of serum IGF-1 (Singh et al., ; Cassilhas et al., , ) and GH (Seo et al., ) in the elderly are only affected by long-term resistance training at moderate and high intensities.
A growing number of studies strongly support the beneficial effects of resistance exercise on various aspects of cognitive performance, such as on executive control, memory, attention, and mini mental state examination (MMSE) score (Perrig-Chiello et al., ; Ozkaya et al., ; Cassilhas et al., ; Liu-Ambrose et al., ). In contrast, Kimura et al. ( ) and Venturelli et al. ( ) failed to replicate these results, reporting that 3 months of resistance training did not significantly improve cognitive performance (e.g., on MMSE and task switching) in healthy elderly populations. These inconsistent results can be attributed to differences in the forms of resistance exercise that were prescribed, and the neuropsychological measures that were assessed. The kinds of resistance exercise that successfully enhanced cognition in elderly subjects included moderate to high intensity training performed two or three times per week. Additionally, different cognitive functions require different training periods: for example, short-term (e.g., 8 weeks) resistance training can only benefit short-term memory (Perrig-Chiello et al., ), whereas long-term (e.g., 12 months) resistance training is better with regard to enhancing long-term memory and executive functions (Liu-Ambrose et al., ).
Executive functions are susceptible to senescence (Alvarez and Emory, ), as the neural networks involved in these are subject to age-related atrophy (West et al., ; O’Connell et al., ). The oddball task is a cognitive test extensively used by previous studies to examine the effects of aging on executive functions involving information processing. The performance of ERPs during the oddball task is a sensitive index of the changes in neural activity related to cognition that are associated with aging, with the P300s (e.g., P3a and P3b) being the most sensitive biomarkers for normal aging and age-related pathology among the neuroelectric markers elicited during this task (Pontifex et al., ; West et al., ; O’Connell et al., ). Since neuronal loss in the cerebral cortex and the cerebral white matter commences during the third decade of life (Jernigan et al., ), this causes a linear decrease in the amplitude of the potentials, coupled with a marked anterior shift in the topographical orientation of the P300 components observed in the elderly when performing the oddball task (West et al., ; Richardson et al., ; O’Connell et al., ). Fortunately, executive functions are more strongly affected by physical activity or exercise than other aspects of cognitive functioning (Colcombe and Kramer, ). A number of studies have demonstrated that active lifestyles in the elderly can enhance performance on the oddball task. For example, fitter elderly subjects showed shorter RTs and larger P3b amplitudes compared with less-fit age-matched counterparts (Pontifex et al., ). Similarly, physically active elderly subjects with habitual moderate exercise demonstrated better performance on the oddball task, such as faster RTs and larger P3 amplitudes (McDowell et al., ; Hatta et al., ). Exercise intervention could thus elicit training effects on the performance of such cognitive tasks in the elderly.
Previous studies on exercise and brain aging have intensively examined neuropsychological (i.e., behavioral and psychomotor) measures. However, thus far, research has not yet considered the effects of long-term resistance exercise on neuroelectric performance, nor has it explored the potential neurophysiological mechanisms impacting performance of the visual oddball task in healthy elderly individuals. Therefore, the aims of this study were as follows: (1) to investigate the effects of a 12 month resistance exercise intervention on executive functions, with respect to age-related effects on neuropsychological and neuroelectric performance elicited during the visual oddball task; and (2) to further explore whether changes in basal levels of serum GH, IGF-1, and homocysteine relate to the effects of long-term resistance exercise on executive functions. Based on the above review of research on cognition and neurobiology, the current study examined the following hypotheses: (1) long-term resistance exercise can attenuate or minimize certain aspects of the progression of cognitive degeneration in healthy elderly individuals when performing an oddball task designed to elicit P3a and P3b components; and (2) exercise-induced changes in growth factors can act as potential mechanisms for this effect. We believe the combination of neuropsychological and neuroelectric measures, with biochemical measures obtained during long-term exercise intervention, can provide more compelling evidence and deeper insights into the nature of the effects of exercise on attenuation or prevention of the cognitive decline associated with aging.
## Methods
### Subjects
This study recruited forty-eight elderly males (aged 65–79 years, mean 71.40 ± 3.79) from senior community centers in rural areas of Taiwan, with these subjects being more sedentary and less educated than their city-dwelling counterparts, and thus presumably more responsive to exercise training. We only recruited male subjects because the effects of exercise on endocrinological responses may be gender-dependent (Baker et al., ). Subjects were recruited with the use of print advertisements and underwent screening by a standardized telephone interview. They then underwent a medical examination, including heart rate and blood pressure measurements, electrocardiography, routine laboratory testing, a standardized neurological examination, and a structured interview on their previous medical histories. The medical exam ascertained whether subjects were free of a history of brain injury, liver and kidney dysfunction, severe medical conditions affecting the dopaminergic system, neurological disorders, and psychiatric illnesses, such as depressive symptoms [defined by scores above 13 on the Beck Depression Inventory, second edition (BDI-II)], dementia, and mild cognitive impairment (defined by scores below 26 on the MMSE) (Ruscheweyh et al., ). The Edinburgh Handedness Inventory assessed all subjects as right-handed (Oldfield, ). The study obtained written informed consent from all participants, and was approved by the Institutional Ethics Committee of National Cheng Kung University.
### Study procedure
Figure presents the procedure of this study. The original cohort consisted of 60 subjects. Based on the assessment of two physicians specializing in geriatric care and physiotherapy, three subjects were excluded due to incomplete individual data, three due to high blood pressure, and six due to musculoskeletal problems, neurological disorders, or psychiatric illness (e.g., scores above 13 on the BDI-II or below 26 on the MMSE), leaving a total of 48 subjects for the present study. Participants also completed the 7-day physical activity recall questionnaire (7-day PAR; Sallis et al., ) to ascertain their previous levels of physical activity, in order to provide sufficient pre-activity screening to lower potential risk factors prior to long-term resistance training. The 48 participants were randomized to the exercise group (i.e., resistance exercise intervention) or control group after matching for age and baseline level of physical activity. The two groups did not significantly differ at baseline in any of the demographic characteristics, including years of formal education, body mass index, years of smoking, MMSE, BDI-II, or systolic and diastolic pressure (see Table ).
The CONSORT (Consolidated Standards of Reporting Trials) flow chart .
Demographic characteristics of the exercise and control groups .
MMSE, mini mental state examination; BDI, Beck Depression Inventory; 7-day PAR,7-day physical activity recall; 1RM: one-repetition maximum .
Two certified fitness instructors completed all assessments of one repetition maximum (1-RM) and peak muscle power for each participant within 1 week of the completion of the baseline evaluation. All the participants in the exercise group were familiarized with the use of free weights and bodybuilding machines before the formal resistance exercise program. On a separate day during the week after the baseline evaluation, the subjects had blood withdrawn between 8:30 and 9:30 AM following overnight fasting, and then performed a cognitive task test with concomitant neuroelectric recording (i.e., ERPs).
After 12 months, the participants completed the same questionnaires, had blood withdrawn, and received measures of neurocognitive parameters over a period of 1 week.
### Cognitive task and experimental procedure
A laptop computer monitor displayed all white stimuli, including oddball stimuli, standard stimuli, and novel stimuli, against a black background. Pontifex et al. ( ) suggested that the three-stimulus oddball task has a higher level of cognitive difficulty with regard to stimulus discrimination among the elderly. The oddball stimulus was the geometric figure “○”, and the standard stimulus was the geometric figure “□”. The novel stimulus category consisted of the following 10 figures: ⋇, ＊, ⊚, ◇, △, ▽, ⨀, ⊕, ♀, ♂, ⊲. The center of the computer screen (width = 43 cm), located directly in front of the participant at face level at a distance of approximately 75 cm, displayed the stimulus (4.09° × 4.09°).
The cognitive test (oddball task) paradigm presented the three types of stimuli in different proportions, with 20% oddball, 60% standard, and 20% novel stimuli. The monitor presented each stimulus for 500 ms, followed by a blank screen for 1500 ms. The stimulus would disappear immediately after the participants responded. If the participants did not respond within 2000 ms, the stimulus would disappear, and the program would advance to the next trial. Participants pressed the “M” key in response to the oddball stimulus, and the “B” key in response to the standard or a novel stimulus. Stimuli were presented in a different, random order for each participant. The entire experiment consisted of three blocks of 100 trials, with the order of the stimulus blocks counterbalanced across participants. Participants were instructed to respond as quickly and accurately as possible. The present study adapted the oddball task from West et al. ( ), wherein its capacity to effectively differentiate lower executive functions of older adults from younger adults was demonstrated.
A trained experimenter blind to group assignment performed the cognitive testing. The experiment was administered in an acoustically shielded room with dimmed lights. On arrival at the laboratory, the experimenter explained the procedure and made sure that the participants were familiar with it. Participants completed 30 practice trials prior to the formal test to ensure they understood the whole process. The electrocap and electro-oculographic (EOG) electrodes were attached to the head and face of the participants before the formal test. Each participant was asked to sit comfortably in an adjustable chair in front of a laptop computer display driven by an IBM-compatible personal computer with a stimulation system (Neuroscan Ltd., EI Paso, USA). During the test, the experimenter sat next to the participant to monitor visual fixation. The experimenter gave verbal encouragement to look at the screen if they detected the participant’s eyes moving away from the central stimulus during the execution of a response. All participants with normal or corrected-to-normal vision acuity performed the oddball task with simultaneous recording of ERPs.
### Electrophysiological recording and analysis
Electroencephalographic (EEG) activity was recorded from 18 electrode sites (F7, F8, F3, F4, Fz, T3, T4, C3, C4, Cz, T5, T6, P3, P4, Pz, O1, O2, and Oz), using an elastic electrode cap (Quik-Cap, Compumedics Neuroscan, Inc., El Paso, TX) designed for the International 10–20 System. Additional ocular electrodes placed on the supero-lateral right canthus and infero-lateral to the left eye monitored horizontal and vertical EOG (i.e., HEOG and VEOG) activity for eye movements. Scalp locations were referred to linked mastoid electrodes, while a ground electrode was placed on the mid-forehead on the Quik-Cap. All electrode impedances were below 5 kΩ. EEG data acquisition employed an A/D rate of 500 Hz/channel, a band-pass filter of 0.1–50 Hz, and a 60 Hz notch filter, with continuous writing to hard disk for off-line analysis using SCAN4.3 analysis software (Compumedics Neuroscan, Inc., El Paso, USA).
The ERP analysis epochs extracted off-line consisted of segments from −100 ms of pre-stimulus activity to 1000 ms of post-stimulus activity. Trials with a response error or EEG artifacts (e.g., VEOG, HEOG, and electromyogram) exceeding peak-to-peak deflections over 100µV were rejected before averaging. The remaining effective data was assembled according to the three different conditions (i.e., oddball, standard, and novel). Measures of peak amplitude were calculated for two components to quantify the effects of long-term exercise intervention and stimulus type on the ERPs. Since P3a has a more anterior distribution than P3b (O’Connell et al., ), West et al. ( ) outlined the following definitions for P3 amplitudes: the novelty P3 amplitude (i.e., P3a) is the major positive deflection over the anterior scalp (F3 and F4), and the most positive point between 300 ms and 400 ms after the stimulus. In contrast, the P3b amplitude is the major positive deflection over the central and posterior scalps (Cz, Pz, and Oz), and the most positive point between 300 ms and 800 ms.
### Resistance training prescription
Resistance training classes for the exercise group began within 1 week after initial 1-RM testing. Two certified fitness instructors formally trained and educated by professional physical fitness courses led the classes at a university fitness center. The exercise group was divided into small subgroups of three to six participants. Each training class lasted approximately 60 min, with 10 min of warm-up, 40 min of core content, and 10 min of cool-down. The warm-up included slow-paced walking and active mobility exercises for the joints of the four limbs. The core resistance exercise content implemented a circuit-training schedule with a progressive, high-intensity protocol (Liu-Ambrose et al., ). The training circuit consisted of the following exercises in the order stated: biceps curls, leg presses, triceps extensions, hamstring curls, latissimus dorsi pull-downs, calf raises, and seated rowing. The training equipment included bodybuilding machines and free weights. The participants performed the resistance training at an intensity of 75–80% 1-RM for three sets of 10 repetitions, at an average speed, with a 90-second rest between sets, and a 3 min interval between each apparatus. The load pressed or lifted for each exercise was recorded in each participant’s exercise log at every class. As each individual’s muscle strength increased, their prescribed training load was also raised to ensure they performed the training at intensity levels corresponding to 75–80% of 1-RM. Such an intensity protocol led to increases in serum IGF-1 levels in elderly subjects in previous studies (Singh et al., ; Cassilhas et al., , ). During the cool-down period, participants used a variety of relaxation techniques, such as controlled breathing and static stretching exercises (i.e., maintaining maximal muscle elongation for 30s to increase range of motion). The fitness instructors recorded adherence, expressed as the percentage of classes attended, in the participants’ exercise logs. The attendance rate was above 90% for all participants, with no subjects dropping out of the study.
The participants in the exercise group were required to participate in 60 min resistance training classes three times a week for a period of 12 months. Participants in the control group received baseline and post-intervention evaluations, but did not receive a specific intervention or group activity that would prevent any potential cognitive benefits from social interactions they might have engaged in.
### Serum analysis
Prior to the two cognitive task tests, a trained phlebotomist withdrew blood from the antecubital vein using an aseptic technique for analysis of serum IGF-1, GH, and homocysteine. The blood was allowed to clot (BD Vacutainer Plus), and then centrifuged at 3000 rpm for 15 min at 4°C (Hettich Mikro 22R, C1110). Each sample was frozen and stored at –80°C for further serum marker assays. Serum values of GH, IGF-I, and homocysteine were determined by a chemiluminescence immunoassay method using an Access Ultrasensitive hGH reagent pack (Beckman Coulter Inc, USA), Liaison IGF-1 reagent (DiaSorin S.P.A., Italy), and Siemens reagents for homocysteine assay (Siemens Healthcare Diagnostics Inc., USA), respectively. The detection limit for GH was 0.002 ng/mL, while that for the IGF-1 was 3 ng/mL, and that for homocysteine was 0.50 mmol/L. All the procedures to assess GH, IGF-I, and homocysteine were performed by the same person to avoid inter-operator bias.
### Statistical analysis
Independent t -tests were used to examine the homogeneity of the demographic backgrounds of the subjects in the exercise and control groups. The accuracy rates and correct-trial RTs were submitted separately to a 2 ( Group : exercise vs. control) × 2 ( Time : pre-exercise vs. post-exercise) × 3 ( Condition : oddball vs. standard vs. novel) mixed repeated measures analysis of variance (RM–ANOVA). P3 amplitudes from ERP recordings were submitted separately to a 2 ( Group : exercise vs. control) × 2 ( Time : pre-exercise vs. post-exercise) × 3 ( Condition : oddball vs. standard vs. novel) × 2 ( Electrode : F3 vs. F4 for the P3a component; Cz vs. Pz vs. Oz for the P3b component) RM–ANOVA. All biochemical markers were submitted separately to a 2 ( Group : exercise vs. control) × 2 ( Time : pre-exercise vs. post-exercise) RM–ANOVA. Appropriate multiple comparisons were performed following any simple main effects. When a significant difference occurred, Bonferroni post hoc analyses were performed. The Greenhouse–Geisser (G–G) correction adjusted the significance levels of the F ratios whenever RM–ANOVA detected a major violation of the sphericity assumption. Partial Eta squared ( ) was used to calculate effect sizes for significant main effects and interactions, with the following standards used to determine the magnitude of mean effect size: 0.01–0.059 represented a small effect size; 0.06–0.139, a medium effect size; and >0.14, a large effect size. Pearson product–moment correlations were used to examine changes in the biochemical markers and cognitive variables. Significance was set at p < 0.05 for all analyses.
## Results
### Accuracy rate
RM–ANOVA performed on the accuracy rates (see Figure ) highlighted a main effect of Condition [ F = 102.72, p < 0.001, = 0.69], with a lower accuracy rate for oddball (88.4%) than standard (97.3%) and novel (96.6%) conditions. The interactions between Time × Group [ F = 5.34, p = 0.025, = 0.10], Condition × Group [ F = 3.42, p = 0.037, = 0.07], and Time × Condition × Group [ F = 5.39, p = 0.006, = 0.11] were also significant. Post hoc analysis showed a lower accuracy rate for the oddball condition in the control group 12 months after baseline [ t = 3.14, p = 0.005]. The exercise group exhibited a significantly higher accuracy rate in the oddball condition [ t = 2.77, p = 0.008] compared to the control group after 12 months.
Accuracy rate (A) and reaction time (B) performance of exercise and control groups during the oddball task at baseline and 12 months later in three conditions . *Significantly different ( p < 0.05). Statistical values expressed as mean ± SE.
### Reaction time
As shown in Figure , RM–ANOVA conducted on mean RTs revealed a main effect of Time [ F = 26.47, p < 0.001, = 0.37], and a main effect of Condition [ F = 210.94, p < 0.001, = 0.82], suggesting that RTs were faster after the exercise intervention (490.51 ms) than before it (514.52 ms), and that RTs were faster in the standard condition (448.35 ms) than in both the oddball (529.67 ms) and novel (529.52 ms) conditions. The interactions of Time × Group [ F = 10.91, p = 0.002, = 0.19], Time × Condition [ F = 8.35, p < 0.001, = 0.15], and Time × Condition × Group [ F = 4.22, p = 0.018, = 0.08] were also significant. Post hoc analysis showed the exercise group responded faster in the oddball [ t = 5.10, p < 0.001] and novel [ t = 5.27, p < 0.001] conditions after exercise intervention compared to baseline. The exercise group only showed significantly faster responses than the control group in the oddball condition [ t = −3.97, p < 0.001] after 12 months.
### P3a amplitude
As illustrated in Figure , RM–ANOVA performed on the P3a amplitudes showed a main effect of Condition [ F = 7.66, p = 0.001, = 0.14], and a main effect of Electrode [ F = 14.64, p < 0.001, = 0.24], suggesting that the P3a amplitude was significantly smaller in the standard condition than in the novel condition, and significantly larger for the F4 electrode than for the F3 electrode. The interactions of Time × Condition [ F = 3.29, p = 0.042, = 0.07], Condition × Electrode [ F = 7.54, p = 0.001, = 0.14], Time × Condition × Electrode [ F = 3.90, p = 0.024, = 0.08], and Time × Condition × Group [ F = 3.49, p = 0.034, = 0.07] were also significant. Post hoc analysis showed that only the P3a amplitude in the oddball condition [ t = 3.19, p = 0.004] was significantly smaller across all electrodes in the control group after 12 months. The exercise group exhibited significantly larger P3a amplitudes in the oddball condition [ t = 2.40, p = 0.001] across all electrodes compared to the control group after 12 months.
Grand average event-related potential waveforms of five electrodes (F3 and F4 for the P3a component; Cz, Pz, and Oz for the P3b component) in the oddball condition during pre- and post-tests of exercise and control groups .
### P3b amplitude
RM–ANOVA performed on the P3b amplitudes showed a main effect of Condition [ F = 4.07, p = 0.020, = 0.08], and a main effect of Electrode [ F = 153.23, p < 0.001, = 0.77], suggesting that the P3b amplitude was significantly larger in the oddball condition than in the standard condition, and significantly smaller for the Oz electrode than for the Cz and Pz electrodes. The interactions of Time × Condition [ F = 4.03, p = 0.021, = 0.08], Condition × Electrode [ F = 15.61, p < 0.001, = 0.25], Condition × Electrode × Group [ F = 2.98, p = 0.021, = 0.06], and Time × Condition × Group [ F = 3.35, p = 0.039, = 0.07] were also significant. Post hoc analysis showed that only P3b amplitude in the oddball condition [ t = 3.61, p = 0.001] was significantly smaller across all electrodes in the control group after 12 months. The P3b amplitude approached significance [ t = 2.02, p = 0.050] between the two groups in the oddball condition across all electrodes after 12 months.
### GH
Figure shows the levels of all biochemical markers before the intervention and after 12 months in the exercise and control groups. RM–ANOVA performed on serum GH levels showed that neither a significant main effect of Group or Time nor a significant interaction of Time × Group was present, indicating that the training effect of serum GH levels did not differ between two groups.
Changes in serum levels of GH (A), IGF-1 (B), and homocysteine (C) before and after 12 months in the exercise or control group . *Significantly different ( p < 0.05) from corresponding baseline values. Statistical values expressed as mean ± SE.
### IGF-1
RM–ANOVA performed on serum IGF-1 levels (see Figure ) showed a main effect of Time [ F = 5.33, p = 0.025, = 0.10], and the interaction of Time × Group [ F = 11.24, p = 0.002, = 0.20] to be significant. Post hoc analysis showed that serum IGF-1 levels significantly increased in the exercise group after 12 months of resistance exercise. The changes in IGF-1 levels in the exercise group were significantly correlated with the changes in RTs ( r = −0.47, p = 0.020) and P3b amplitude ( r = 0.52, p = 0.009) in the oddball condition. However, this effect was not found for the accuracy rates ( r = 0.17, p = 0.939) or P3a amplitudes ( r = 0.32, p = 0.123).
### Homocysteine
As can be seen from Figure , RM–ANOVA performed on serum homocysteine levels showed Time [ F = 9.49, p = 0.003, = 0.71], and the interaction of Time × Group [ F = 4.85, p = 0.033, = 0.10], to produce significant main effects. Post hoc analysis showed that serum homocysteine levels were only significantly reduced in the exercise group after 12 months of resistance exercise. However, there were no significant correlations between changes in homocysteine levels and changes in behavioral and ERPs performances after long-term intervention in the exercise group.
## Discussion
This study aimed to investigate whether a 12 month high-intensity resistance exercise intervention could effectively retard a decline in executive functions in healthy elderly males, and to determine the relationship between changes in IGF-1, GH and homocysteine levels and neurocognitive performance (e.g., neuropsychological and neuroelectric components) during an oddball task. The control group displayed a lower accuracy rate and smaller P3a and P3b amplitudes in the oddball condition when performing the oddball task after 12 months. The results for the exercise group showed that long-term high-intensity resistance exercise can decrease RTs and attenuate decreases in P3a and P3b amplitudes during a stimulus discrimination task, as well as increase serum IGF-1 levels and decrease serum homocysteine levels, and that changes in IGF-1 levels were significantly correlated with changes in RTs and P3b amplitudes in the oddball condition. These findings suggest that long-term resistance exercise could be an effective mechanism for attenuating the age-related decreases in neural efficiency in healthy elderly individuals manifested during the oddball task, possibly modulated by increased IGF-1 levels.
### Neuropsychological index
Participants in the control group had significantly lower accuracy rates in the oddball condition after 12 months than at baseline, indicating that older adults exhibit a reduced ability with aging to differentiate between standard and target stimuli during the three-stimulus oddball task. The results of the present study support those of previous research, as healthy individuals aged 55–80 years showed a decrease of 0.21 in mean MMSE score per year, with 22% of individuals showing a decrease of more than 1 point per year (Kalmijn et al., ), and elderly individuals demonstrated lower accuracy for novel stimuli (Fabiani and Friedman, ). These findings demonstrate that aging is marked by a progressive decline in cognitive functioning, and that information processing is vulnerable to aging. However, the participants in the exercise group did not exhibit a similar trend of decreased performance during the study intervention period, suggesting that these negative effects of aging may be attenuated by regular participation in resistance exercise.
In addition, the exercise group displayed faster RTs in the oddball and novel conditions after the exercise intervention compared with at baseline, and, in particular, this change led to a group difference in the oddball condition after 12 months. These findings suggest that the neuromotor and central processing of cognitive functions used to distinguish the oddball stimulus from a frequent stimulus could be significantly enhanced in elderly males by 12 months of resistance exercise. Indeed, Hatta et al. ( ) found that regular participation in moderate exercise could promote response processing when performing a somatosensory oddball task in elderly adults. In addition, a previous study demonstrated that elderly individuals who are physically active could retain their reaction capacity (Spirduso, ). Based on findings from both the present study and previous research, we postulate that high-intensity resistance exercise could facilitate greater temporal efficiency in the central processing of cognitive functions in healthy elderly individuals.
Interestingly, the group differences seen in task accuracy and RT were driven by different patterns of behavioral changes during a one-year period. That is, the accuracy in the oddball condition declined significantly in the control group, while the RT decreased significantly in the exercise group. Previous studies examined the accuracy rates and RTs during completion of an oddball task among subjects of various ages, and found that elderly adults are less accurate relative to middle-aged and younger adults, whereas no statistically significant differences in RTs were observed, showing that aging has different effects on these two outcomes, with the authors suggesting that this may be due to age-related changes in response strategies (Ford and Pfefferbaum, ; Iragui et al., ). In this study we observed that, relative to baseline performance, the control group exhibited decreased accuracy with maintenance of RT performance. This is probably because the elderly adults in the control group adopted a strategy that favors processing speed over accuracy when performing this type of cognitive task, which diminishes RT delays at the expense of decreased accuracy. Similarly, for the exercise group, the benefits of exercise training were easier to observe with regard to RTs, since these subjects also focused more on speed than accuracy, and this trade-off may thus reduce any benefits with regard to task accuracy.
### Neuroelectric index
P300 reflects attentional processes, indexed by two distinct yet related subcomponents of neural processes (Pontifex et al., ): the P3a component is elicited by a change in the stimulus environment (e.g., an infrequent or novel non-repeating distractor), with the P3a amplitude reflecting a stimulus-driven, or bottom-up, attentional orienting to a salient but irrelevant stimulus (Polich, ; Richardson et al., ); the P3b component is elicited by a rare stimulus within a series of frequent irrelevant stimuli (O’Connell et al., ), with the P3b amplitude serving as a proposed reflection of the top-down allocation of attentional resources to stimulus evaluation when working memory is updated (Polich, ; Verleger, ). In the present study, the control group exhibited significantly smaller P3a and P3b amplitudes in the oddball condition after 12 months, indicating age-related decreases in both attentional orienting/engagement of focal attention, and attentional resource allocation and subsequent memory processing (Polich, ; Verleger, ) in healthy elderly subjects. These findings echo those of a previous study that examined the extent of the decline in attention control efficiency during normal aging, and which suggested that reduced attention control in older adults relative to younger adults, observed in terms of cortical activity, causes greater inefficiency in the tendency to filter out irrelevant information over successive trials (Fabiani et al., ).
In contrast, the exercise group sustained P3 amplitudes in the oddball condition over 12 months of high-intensity resistance exercise, resulting in a significant difference in P3 amplitudes between the exercise and control groups. Similarly, Polich and Lardon ( ) demonstrated that very physically active young adults achieved larger P3 amplitudes compared to relatively inactive counterparts when performing the visual oddball task. However, Pontifex et al. ( ) examined P300 components separately, and found that older adults with high cardiorespiratory fitness only exhibited greater P3b amplitudes, with no significant differences in P3a amplitudes relative to controls, suggesting that fitness-related changes in cognitive aging appear specific to attentional processing. In the present study, the exercise group exhibited significantly larger P3a and P3b amplitudes during the stimulus discrimination task, indicating that 12 months of resistance training can simultaneously maintain the capacities for both orienting and allocating attention. The present neuroelectric findings seem to suggest that chronic resistance exercise might modulate, and in some cases, potentially reverse, age-related decreases in neuronal tissue loss in the brain cortices.
### Neurophysiological index
Although GH stimulation can cause an increase in IGF-1 production (Sonntag et al., ), the healthy elderly males in the current study showed significant increases in basal serum IGF-1 levels after 12 months of high-intensity resistance exercise, without any accompanying elevation in GH levels from baseline. This result supports previous studies demonstrating that spontaneous GH secretion does not positively correlate with basal IGF-1 levels in older adults (Vermeulen, ; Benbassat et al., ). In this study, the slight increase in GH levels in the exercise group after long-term exercise does not indicate a significant relationship with changes in neuropsychological and neuroelectric performance. This substantiates previous studies that investigated neurocognitive performance in subjects with GH replacement. For example, Papadakis et al. ( ) found that 6 months of GH treatment did not result in a significant improvement in neuropsychological performance in healthy elderly men with low baseline IGF-1 levels. Golgeli et al. ( ) also reported that 6 months of GH replacement therapy in Sheehan syndrome patients with severe GH deficiency did not significantly affect P3 amplitudes. More recently, Quik et al. ( ) failed to find a relationship between GH levels and N2b amplitudes in healthy males aged 50–78 years during performance of a selection-potential go/no-go task. However, Rollero et al. ( ) observed that although the MMSE score was not associated with basal GH levels or GH peaks after GH-releasing hormone stimulation in elderly subjects, cognitive performance was positively related with total IGF-1 levels.
Despite a decrease in circulating serum IGF-1 levels paralleling a decline in GH pulses in later life (Corpas et al., ), the close relationship between them seems not to extend to identical effects on cognitive performance. Indeed, only changes in the IGF-1 levels of the exercise group subjects in the current study correlated significantly with changes in RTs and P3b amplitudes, with such an effect not exhibited by the GH parameter. Previous studies have demonstrated that age-related decreases in serum IGF-1 levels could be a potential mechanism for age-related decline in cognitive functions (e.g., processing speed) in the elderly (Sonntag et al., ). However, increased brain uptake of peripheral IGF-1 during exercise could lead to training-induced neuroprotective effects (Carro et al., ). That is, IGF-1 is essential for exercise-induced neurogenesis (Carro et al., ), and acts to mediate exercise-induced angiogenesis (Lopez-Lopez et al., ). In the current study, IGF-1 levels were significantly increased in the exercise group after 12 months of resistance exercise training. The changes in IGF-1 levels significantly correlated with RT performance, reflecting previous research which demonstrated that higher IGF-1 levels in adults with Prader–Willi syndrome are associated with faster temporal memory performance (van Nieuwpoort et al., ). The findings of the present study show that increases in IGF-1 levels after long-term resistance exercise can reduce the time needed for central processing of cognitive functions (e.g., RTs) in healthy elderly males. Similarly, Baker et al. ( ) found that aerobic fitness may improve executive control with an increase in IGF-1 in elderly males at risk of cognitive disorder. However, although Aleman et al. ( ) reported no correlation between IGF-1 levels and memory, attention, or fluid intelligence in healthy elderly males aged 65–76 years, they did observe significant associations for serum IGF-1 levels with both perceptual–motor and information processing speed, which is known to decline significantly with aging. Based on our findings and those of previous studies, serum IGF-1 levels could be the key neurophysiological indicator of improved response processing in healthy elderly males after participation in 12 months of high-intensity resistance exercise. However, such a positive effect did not appear to emerge in the neural system during the aging process, and thus increased serum IGF-1 levels do not increase P3a and P3b amplitudes in healthy elderly males.
The current study found a significant negative correlation between the changes in IGF-1 levels and P3b amplitudes for the exercise group subjects in the oddball condition, suggesting that enhanced IGF-1 levels might have decreased the neural system degeneration, and thus enabled them to better distinguish the oddball stimulus from the standard stimulus. However, the changes in IGF-1 levels were only associated with P3b amplitude, not P3a amplitude, indicating that changes in the growth factor were not related to a general change in the attentional system. Serum IGF-1 levels appear to selectively associate with a particular aspect of attention; specifically, IGF-1 might mediate the neural network involved in the top-down allocation of attentional resources, while not affecting the bottom-up allocation used during attentional orienting. Since P3a and P3b might relate to the dopaminergic and locus-coeruleus-norepinephrine systems, respectively (Nieuwenhuis et al., ; Polich and Criado, ), further research is warranted in this area, with a possible focus on examining the potential interactive mechanisms between IGF-1 and these two biochemical systems.
Overall, the evidence presented above suggests that IGF-1 might be an intermediary for the effects of resistance exercise at central levels, despite previous studies reporting that the upregulation of the GH/IGF-1 axis seemed to produce positive effects on cognitive functions. Indeed, Kalmijn et al. ( ) found that higher total serum IGF-1 concentrations at baseline significantly correlated with less cognitive decline in terms of MMSE score over a two-year period in healthy individuals aged 55–88 years. Serum IGF-1 concentrations might reflect an underlying biological process influencing cognitive decline based on the statistically significant correlations obtained between changes in IGF-1 levels and cognitive performance. Collectively, these results indicate that this biochemical agent can attenuate or minimize the progress of certain aspects of cognitive degeneration in healthy elderly individuals. This might be achieved through various central mechanisms, including actions on neurons, cerebrovasculature, and the number of cells expressing c-fos in neurons and glia, possibly because of its transportation to the central nervous system via the hematoencephalic barrier (Sonntag et al., ). Additionally, these findings indicate that factors other than GH secretion are involved in the relationship between serum IGF-1 levels and cognitive function decline in the elderly.
High homocysteine levels are a risk factor for cognitive impairment in older adults (Ford et al., ). In the present study, serum homocysteine levels were significantly reduced in the exercise group after 12 months of resistance exercise, echoing the findings of a previous study in which serum homocysteine decreased after 6 months of high- or low-intensity resistance exercise (Vincent et al., ). Although the potential mechanisms by which resistance training might prevent cognitive decline in the elderly involve homocysteine (Liu-Ambrose and Donaldson, ), the present study did not show an association between changes in homocysteine levels and changes in neuropsychological and neuroelectric measures. A possible explanation for this is that the cognitive task adopted in this study related to executive functions, with new research suggesting that high homocysteine levels in elderly adults only decrease performance in tests of immediate and delayed memory, not executive functions (Ford et al., ). Although a few studies have demonstrated prolonged potential latencies in P3 amplitude associated with elevated homocysteine levels (Evers et al., ; Díaz-Leines et al., ), as in all experimental studies with a cross-sectional design, it is difficult to infer causal relationships between serum homocysteine levels and neurocognitive performance in healthy elderly males.
### Study limitations
While the present findings shed light on the beneficial effects of 12 months of high intensity resistance training on neuropsychological, neuroelectric, and neurophysiological outcomes, there are some limitations that indicate they should be applied with caution. First, we only recruited male elderly adults in the present study to exclude the influences of gender differences in executive control (Rubia et al., ), myofiber hypertrophy (Bamman et al., ) and endocrine indices (Staron et al., ) in response to resistance training. The results may thus not be generalized to female subjects without more work being done. Second, the results of the present experimental design (i.e., a single laboratory-based cognitive task) might be difficult to apply to daily living activities. An additional virtual reality task (i.e., Chaddock et al., ) may thus help to assess the potential behavioral benefits of resistance exercise in future investigations.
## Conclusions
In conclusion, increasing the level of physical activity via high-intensity resistance exercise could assist in lowering the rate of age-related neurocognitive decreases in healthy elderly males. In addition, increases in basal IGF-1 levels achieved via such an exercise protocol could have positive effects on both neuropsychological (i.e., RT) and neuroelectric (i.e., P3b amplitude) performance in the elderly. This study’s findings imply that healthy elderly individuals who regularly engage in resistance exercise might delay the onset of age-related decline in executive functions, and that this protective effect may be modulated by the growth factor-IGF-1.
## Conflict of interest statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Impulse control disorders (ICDs) are frequent behavioral complications of dopaminergic (DA) replacement therapies (DRTs) in Parkinson’s disease (PD). Impulsive choice, which refers to an inability to tolerate delays to reinforcement, has been identified as a core pathophysiological process of ICDs. Although impulsive choices are exacerbated in PD patients with ICDs under DRTs, some clinical and preclinical studies suggest that the DA denervation of the dorsal striatum induced by the neurodegenerative process as well as a pre-existing high impulsivity trait, may both contribute to the emergence of ICDs in PD. We therefore investigated in a preclinical model in rats, specifically designed to study PD-related non-motor symptoms, the effect of nigrostriatal DA denervation on impulsive choice, in relation to pre-existing levels of impulsivity, measured in a Delay Discounting Task (DDT). In this procedure, rats had the choice between responding for a small sucrose reinforcer delivered immediately, or a larger sucrose reinforcer, delivered after a 0, 5, 10 or 15 s delay. In two different versions of the task, the preference for the large reinforcer decreased as the delay increased. However, and in contrast to our initial hypothesis, this discounting effect was neither exacerbated by, or related to, the extent of the substantia nigra pars compacta (SNc) DA lesion, nor it was influenced by pre-existing variability in impulsive choice. These results therefore question the potential implication of the nigrostriatal DA system in impulsive choice, as well as the DA neurodegenerative process as a factor contributing significantly to the development of ICDs in PD.
## Introduction
Parkinson’s disease (PD) is a neurodegenerative disorder hitherto considered to stem from the loss of dopaminergic (DA) neurons in the substantia nigra pars compacta (SNc) and mainly characterized by cardinal motor symptoms (Samii et al., ). However, PD is also associated with a plethora of neuropsychiatric deficits, ranging from apathy and depression to impulse control disorders (ICDs; Voon et al., ; Sinha et al., ; Sierra et al., ; Houeto et al., ). ICDs are a complex group of impulsive/compulsive behaviors that includes gambling disorder, hypersexuality or compulsive shopping. ICDs are displayed by up to 14%–15% of PD patients under DA replacement therapies (DRTs) of whom quality of life they dramatically affect (Voon et al., ; Weintraub et al., ; Houeto et al., ). At a neurobiological level, despite the suggestion that an overstimulation of the mesocorticolimbic DA system by DRTs promotes ICDs in PD (Dagher and Robbins, ; Tang and Strafella, ), the underlying psychobiological and etiopathogenic factors that contribute to the development of ICDs only in vulnerable individuals remain unclear.
However, impulsive choice, a form of cognitive impulsivity which reflects an inability to tolerate delays to reinforcement, has been identified as a core pathophysiological process of ICDs (Voon et al., ; Houeto et al., ). Indeed, impulsive choice as measured in Delay Discounting Tasks (DDTs), which is characterized by the preference for small, immediate, rewards, over larger, delayed, rewards, is exacerbated in PD patients with ICDs (reviewed in Voon et al., ). Moreover, higher levels of impulsive choice have also been observed in de novo unmedicated or “off” medication PD patients without ICDs compared to healthy controls (Milenkova et al., ; Al-Khaled et al., ), suggesting that nigral DA cell loss itself may contributes to alter impulse control in PD (Voon and Dalley, ).
Recent studies in rats also support this hypothesis. Indeed, bilateral 6-hydroxydopamine (6-OHDA) lesions of the dorsal striatum reduced their tolerance for delayed reinforcers in a DDT (Tedford et al., ). In addition, α-synuclein-induced nigrostriatal neurodegeneration has been shown to increase other forms of impulsive behaviors (Engeln et al., ). However, the lesional approaches used in these studies provoked substantial motor deficits that may have bias measures of impulsivity. In addition, since only a subset of PD patients is affected by ICDs, the degeneration of the nigrostriatal DA system does not appear to be sufficient to promote ICDs, indicating a potential interaction with an endophenotype of vulnerability, as suggested in a previous study (Engeln et al., ).
Because impulsivity is tightly associated with ICDs and represents an endophenotype of vulnerability to develop compulsive behaviors and is a critical factor for the development of compulsive behaviors (Belin et al., ; Ansquer et al., ), it has been hypothesized that a high impulsivity trait may be associated with the disease progression and the emergence of ICDs in those vulnerable PD patients (Dagher and Robbins, ; Voon and Dalley, ; Houeto et al., ). Yet, the potential relation between the DA denervation and impulsivity remains to be established.
We therefore investigated in a longitudinal study (Figure ) the effect of nigrostriatal DA denervation on impulsive choice and its relation with endogenous level of impulsivity in DDTs. For this, we used a preclinical model in rats specifically designed to study PD-related non-motor symptoms (Magnard et al., ). Based on 6-OHDA-induced bilateral but partial lesions of the nigrostriatal DA system, this model has been demonstrated to reveal denervation-induced behavioral impairments, such as motivational- and affective-related deficits, without displaying significant impairments of motor functions (Carnicella et al., ; Drui et al., ; Favier et al., ).
Delay Discounting Task (DDT) paradigms. (A) Experimental schedule and timeline of the experiments. Training in DDTs was performed prior to, and after, intra-substantia nigra pars compacta (SNc) 6-hydroxydopamine (6-OHDA)/saline injection. The two experimental groups of rats followed the same timeline at the exception that they were tested either in a DDT-within or DDT-between procedure. (B) Flow chart of delay discounting blocks. Each block starts with forced choices wherein only one lever is extended at the time in a random order, allowing rats to learn or recall the contingency of that lever (small or large reinforcer; i.e., 120 μL of a 5% or 10% sucrose solution, respectively) and the delay associated with. Then, rats have free choices access to the small or large reinforcer (both levers are extended), associated with the delay experienced during forced choice trials. No lever press during 35 s of light on period leads to an omission and no reinforcer is delivered. After 35 s, the houselight is turned off and levers are retracted, for a 20 s intertrial interval (ITI), until the beginning of a new trial. During the task, each lever press leads also to levers retraction. (C) Schematic representation of the two DDT paradigms. The within-session delay discounting paradigm (within-DDT), is composed of five blocks, each composed of four forced choices (2 per lever) and 10 free choices. The delay increased progressively from one block to another. The task ends after 70 trials. In the between-session delay discounting paradigm (between-DDT), rats first performed 10 forced choices and then 30 free choices, experiencing only one delay per session. The task ends after 40 trials. These two experiments were independent, and two groups of rats were used, one per paradigm.
## Materials and Methods
### Animals
Experiments were performed on male Sprague-Dawley rats (Janvier, France) 8 weeks old (weighting 300 g) at the beginning of the experiment. Twenty-one and 31 rats were used in DDT-within and -between experiments, respectively. They were individually housed under standard laboratory condition (12 h/light/dark cycle, with lights ON at 7 a.m.). They were food restricted at 90% of their free feeding weight during the DDT procedure, but had ad libitum access to water. Protocols complied with the European Union 2010 Animal Welfare Act and the new French directive 2010/63, and were approved by the French national ethics committee n° 004.
### Bilateral 6-OHDA Lesion
This procedure has been described extensively elsewhere (Carnicella et al., ; Drui et al., ; Favier et al., ). Briefly, food restriction was suspended 2 days before the beginning of the surgery. Rats were administered with desipramine hydrochloride (15 mg.kg subcutaneously; Sigma-Aldrich, St. Quentin-Fallavier, France) 30 min before they received an intracerebral infusion of 6-OHDA or vehicle (NaCl) in order to protect noradrenergic neurons. Rats were then anesthetized with a mixture of xylazine (15 mg.kg ) and ketamine (100 mg.kg ) both administered intraperitoneally. Rats were secured on a Kopf stereotaxic apparatus (Phymep, Paris, France) and 6 μg 6-OHDA (Sigma-Aldrich, St. Quentin-Fallavier, France) dissolved in 2.3 μl sterile 0.9% NaCl, or 2.3 μl sterile 0.9% NaCl (sham conditions), were injected bilaterally, through a 26 gauge cannula (Plastic One, Roanoke, USA) in the SNc, at a flow rate of 0.5 μl.min , at the following coordinates relative to bregma: AP, −5.4 mm; ML, ±1.8 mm; DV, −8.1 mm (Paxinos and Watson, ). After recovery from anesthesia, animals were returned to the facilities with food and water available ad libitum during 4 weeks in order to allow recovery and the 6-OHDA lesion to develop and stabilize prior to being re-subjected to food restriction and behavioral training.
### Tyrosine Hydroxylase Immunohistochemistry and Quantification of Striatal DA Denervation
Rats were euthanized under chloral hydrate anesthesia at the end of the behavioral experiments. They were intracardially perfused with paraformaldehyde (PFA; 4%) as in Drui et ( ; DDT-within) or brains were post-fixed with PFA as in Favier et al. ( , DDT-between), then frozen in cooled isopentane (−40°C) and stored at −30°C. Fourteen micrometer thick serial frontal sections of the striatum were processed with a cryostat (Microm HM 500, Microm, Francheville, France), collected on microscopic slides and stored at −30°C.
Immunostaining was carried out as previously described (Favier et al., ). Sections collected on microscope slides were first air-dried and post-fixed with 4% PFA for 10 min, and then washed in PBS. Brain sections were subsequently incubated with an anti-tyrosine hydroxylase (anti-TH) antibody (mouse monoclonal MAB5280, Millipore, France, 1:2,500) and then with a biotinylated goat anti-mouse IgG antibody (BA-9200, Vector Laboratories, Burlingame, CA, USA; 1:500). Immunoreactivity was visualized with avidin-peroxidase conjugate (Vectastain ABC Elite, Vector Laboratories Burlingame, CA, USA).
TH immunoreactivity in the dorsal striatum and the nucleus accumbens (NAcc) was analyzed across the sections ranging from +2.2 to +0.7 mm from bregma with the ICS FrameWork computerized image analysis system (Calopix, 2.9.2 version, TRIBVN, Châtillon, France) coupled to a light microscope (Nikon, Eclipse 80i) and a Pike F-421C camera (ALLIED Vision Technologies, Stadtroda, Germany) for digitalization. Masks from the different striatal subregions were drawn with the computer analysis system to ensure that appropriate comparisons were made between homologous anatomical regions. Optical densities (ODs) were measured for each striatal region, and the mean OD was calculated with ICS FrameWork software (TRIBVN, 2.9.2 version, Châtillon, France). ODs were expressed as percentages relative to the mean optical density values obtained from the homologous regions of sham-operated animals. Only individuals displaying mean bilateral TH immunoreactivity (TH-IR) loss in the range of 50% to 85% in the dorsal striatum and less than 30% in the NAcc, were included in the analysis, as previously described (Drui et al., ; Favier et al., ). Based on these restrictive criteria, six and 13 6-OHDA lesioned rats were excluded for the within-DDT and the between-DDT experiments, respectively.
### Delay Discounting Tasks
Sixteen operant chambers for rats (30 × 24 × 27 cm) from Med-Associates (St. Albans, VT, USA) were used. They were equipped with two retractable levers, a cue light located above each lever, a central house light, as well as a dual cup liquid receptacle (ENV-200R3AM) located between the two levers and connected through tubing (PHM-122-18) to syringe pumps (PHM-100) for the delivery of the sucrose solution (5% or 10% w/v in tap water, Sigma-Aldrich, St. Louis, MO, USA).
Two independent delay discounting procedures were carried out on two different batches of rats: one with the delay increasing within session (within-DDT) and the other with the delay increasing between sessions (one delay at a time, between-DDT; Figure ). The two tasks are described below, and were adapted from Evenden and Ryan ( ) and Mar and Robbins ( ) and based on pilot parametric experiments. The procedures started after 1 week of food restriction. Rats were randomly assigned to one operant chamber and the lever side (left or right) assigned to the large or small reinforcer was counterbalanced across operant chambers to prevent any bias of preference. Rats were exposed to one training session each day.
#### Phase 1: Operant and Forced Choice Training
Each session of this phase was divided in 70 and 40 trials for the within-DDT and between-DDT experiment, respectively. When a trial started, only one lever was extended and the above cue-light as well as the house light were illuminated to signal the opportunity to press. If the rat pressed the extended lever within 35 s, the lever was retracted, the cue-light turned off and the small reinforcer (120 μL of a 5% sucrose solution) delivered. After 35 s, the house light was turned off for a 20 s intertrial interval (ITI). If the rat failed to press the extended lever during the allocated 35 s period, the lever was retracted, the cue-light turned off at the beginning of the ITI, and an omission was counted.
As soon as the contingency between the instrumental response and the delivery of the small reinforcer (>85% of reinforced trials over three consecutive days) was acquired, rats were trained to acquire the other contingency whereby under similar forced choice sessions, they were required to respond on the other lever to obtain the larger reinforcer (120 μL of a 10% sucrose solution), until the same criteria was reached.
#### Phase 2: Discrimination and Free Choice Training
During this phase, rats were trained to choose between the large and the small reinforcer-associated lever, without any delay. Discrimination phase was considered as acquired when preference for the large reinforcer was above 85% during three consecutive days. As for phase 1, the trial started with the illumination of the house light (35 s long) and it is followed by a 20 s-off period signaled by the absence of light. The within- and between-session procedures are detailed below.
#### Within-Session
The task was divided into five blocks of 14 trials (total: 70 trials/session). Each block began with four forced choices (2 per lever) assigned in a random order, in which only one lever was extended at a time. The remaining 10 trials offered free choices during which both levers and cue-lights were extended and illuminated, respectively. Upon pressing one lever, both levers were retracted immediately and the associated reinforcer (120 μL of a 5% or a 10% sucrose solution) delivered. Each block had the same structure.
#### Between-Sessions
The task started with 10 forced choices (five per lever) assigned in a random order, in which only one lever was extended at the time. Then, rats had access to 30 free choices trials (total: 40 trials/session) with both levers extended and cue-lights illuminated as for the within-session protocol. As in this procedure, only one delay per session was experienced (see “Phase 3: Delay Discounting” section), the last 3 days of the discrimination phase were used to determine the preference for the large reinforcer at the 0 s delay.
#### Phase 3: Delay Discounting
The same blocks and structures as in the discrimination phase were used, except that delays between lever press and the delivery of the large reinforcer varied (Figure ).
#### Within-Session (Figure Top)
Rats underwent 10 consecutive delay discounting test sessions, in which delay between lever press and the delivery of the large reinforcer increased between each block in an ascending order (see Tedford et al., for rationale) within the session. The delays per block were set at 0, 5, 10 and 15 s respectively. The last five sessions were averaged to determine the preference for the large reinforcer in function of delay, as performance was stable across the population (<20% variability between the mean of each group across these five sessions).
#### Between-Sessions (Figure Bottom)
The delay preceding the delivery of the large reinforcer increased every five sessions, with the delays being the same as those used in for the within-session procedure. The last three sessions for each delay were averaged to determine the preference for the large reinforcer, as performances were stable (<20% variability of the mean of each group across these three sessions).
The two DDTs were first performed before the lesion procedure in order to determine individual impulsivity traits and equilibrate the different experimental groups (ANOVAs conducted on this pre-surgery period revealed a main effect of delay: F s > 10.56, p s < 0.001, but no group effect: F s < 0.29, p s > 0.59 and no group × delay interaction: F s < 0.14, p s > 0.93). Post-surgery resumption started on phase 2 (discrimination) in order to evaluate the effect of the lesion on impulsive choice and impulsivity trait.
### Data and Statistical Analyses
Data are presented as mean ±SEM or individual datapoints. Calculation of the area under the curves (AUC) in order to measure delay discounting in each individual, were adapted from Myerson et al., . Briefly, delays, and large reinforcer preference were first expressed as the proportion of their maximum value in order to be comprised between 0 and 1. Then, the resulting discounting curve was subdivided discounting graph into a series of half trapezoids, from 0 to 5 s, 5 to 10 s and 10 to 15 s. The area of each trapezoid is thus equal to the following equation: x − x (y + y )/2, where x and x are successive delays associated with the large reinforcer preference y and y respectively. Thus, the area under the discounting curve is the sum of the area of the three trapezoids. In contrast to Myerson et al. ( ), the value of the delay 0 was not normalized to 100% (i.e., 1), because the current strategy is considered better to capture the reaction and adaptation of each individual to the delay with regards to their initial preference for the large reinforcer. Although we found a strong correlation between the two measures of AUC ( R = 0.88 and 0.82 for the within-session and between-sessions DDTs experiments respectively), these two calculation methods may lead to slight differences. Hence, measures of DD-AUC using the normalization described by Myerson et al. ( ), are reported in .
Data were analyzed by t -test or two-way repeated measure ANOVAs, using SigmaStat software (SigmaStat 4.0 2016, Systat software Inc., San Jose, USA). When indicated, post hoc analyses were carried out using the Student-Newman-Keuls test. Correlations were performed and analyzed with parametric linear regression approaches (Pearson product moment correlation and comparison of linear regression coefficients). Assumptions for the normality of the distributions and the homogeneity of variance were verified using the Shapiro-Wilk and Levene test, respectively. Significance for p values was set at α = 0.05. Effect sizes for the ANOVAs are also reported using partial η values (Levine and Hullett, ; Murray et al., ).
## Results
Rats were first trained in a DDT under which the delay increased progressively within each test session (within-DDT), in order to measure individuals’ basal level of impulsive choice and distribute them in the various experimental groups. Rats were subsequently exposed to either sham or bilateral infusions of 6-OHDA into the SNc, which resulted in partial selective DA denervation in the dorsal striatum that relatively spared the NAcc (68.8 ± 2.4% and 4.6 ± 1.8% loss of TH-immunolabeling, respectively, ( t = 21.42, p < 0.001; Figures ). This pattern of denervation is in line with previous studies (Drui et al., ; Favier et al., ) which have demonstrated it circumvents the motor deficits frequently associated with greater or other patterns of nigrostriatal DA lesions.
Striatal tyrosine hydroxylase immunoreactivity (TH-IR) loss induced by 6-OHDA SNc lesion. (A) Representative photomicrographs of TH-immunostained striatal coronal section of sham rat (left picture) and 6-OHDA lesioned rat (right picture). Quantification of TH-IR loss in DDT within-session experiment (B) and between-session experiment (C) . Data are expressed as the percentage of the mean optical density value obtained for sham rats on the AP levels +2, 2; +1, 6; +0, 7. ±SEM. *** p < 0.001. DS, Dorsal striatum; NAcc, Nucleus accumbens. Scale bar: 1 mm.
At resumption of within-DDT training, all rats maintained a preference for the large reinforcer over the smaller one in absence of delay, and irrespective of the lesion (Figure ; delay 0 s), indicating that, in agreement with previous results (Drui et al., ), the relative reinforcing value of natural rewards was not influenced by the nigrostriatal DA denervation. This preference for the large reinforcer decreased as the delay increased (main effect of delay: F = 9.94, p < 0.001, partial η = 0.43; Figure ). However, this discounting effect was not exacerbated in lesioned animals (no effect of lesion: F = 0.05, p = 0.83, partial η = 0.006, and no delay × lesion interaction: F = 0.49, p = 0.69, partial η = 0.04; Figure ), which led eventually to a similar AUC in both groups (no effect of lesion: F = 0.03, p = 0.86, partial η = 0.001, and no period × lesion interaction: F = 0.01, p = 0.92, partial η < 0.001; Figure ; see also ). Level of impulsivity seemed not to be related to the extent of the nigrostriatal DA denervation, as no correlation was evidenced between the percentage of TH-immunolabeling loss in the dorsal striatum and the level of impulsive choice (indexed by the AUC; Figure and ). A positive correlation was nevertheless found between the AUC before and after surgery in sham rats (Figure ). Intriguingly, this relation significantly decreased in the 6-OHDA condition (significant difference in regression slope: t = 3.35, p < 0.05; Figure ), which may indicate a narrowing of the variance of this trait or a change in the strategy used to compute the delay-preference function. However, this relation as well as the differences between sham vs. lesion groups, were not found with the conventional normalization method of Myerson et al. ( ; ).
Partial nigrostriatal denervation does not affect impulsivity in a within-session DDT (within-DDT). (A) Similar discounting pattern between sham and 6-OHDA lesioned rats, expressed as percentage of preference for the large reinforcer over the smaller one, in function of delay. In this procedure, rats experienced all delays in each session. Data are represented as mean ±SEM and were averaged from the last five sessions. (B) Similar AUC, between sham and 6-OHDA lesioned rats, pre- and post-surgery period. The AUC is extrapolated from area under discounting curves and expressed as mean value ±SEM for each group. (C) No correlation between the post-surgical AUC and the degree of dorsal striatum TH-IR loss. Dots represent individual values for the AUC and dorsal striatum TH-IR loss expressed as percentage of sham mean value. (D) Positive correlation between pre and post-surgery of individual AUC values for sham operated rats (empty circle) and 6-OHDA lesioned rats (full circle). Correlation is only significant for sham rats ( p < 0.05; for 6-OHDA rats: p = 0.15). (E) 6-OHDA lesioned rats exhibited a higher percentage of omissions than sham rats during within-DDT. Data are represented as mean of omissions ±SEM in function of delay (averaged for the last five sessions). NS, non-significant, ** p < 0.01; *** p < 0.001; Sham ( n = 9) vs. 6-OHDA ( n = 6), reinf, reinforcer; AUC, area under the discounting curve; a.u, arbitrary units.
In addition, although the number of omissions displayed by the sham group remained very low in both forced- and free-choice trials and across delays ( , Figure , respectively), it was considerably increased in lesioned rats, especially as the session progressed and the delay increased (for free-choice trials, main effect of the lesion: F = 31.83, p < 0.001, partial η = 0.91 and delay × lesion interaction: F = 15.75, p < 0.001, partial η = 0.55). This increase in omissions may reflect an increased aversion to the cognitive/motivational demand when delays are introduced, which results in a delay-dependent disengagement from the task or a more general impairment in maintaining a motivated behavior over prolonged periods of time. Interestingly, a systematicity in the omissions profile was observed, as even within a block, omissions mostly occur at the end. Indeed, such SNc DA lesions have been shown to impaired the maintenance of preparatory and seeking behaviors (Magnard et al., ; Favier et al., ) and nigrostriatal DA denervation can induce profound attentional and/or cognitive deficits (Nieoullon and Coquerel, ; Aarts et al., ).
Because such a high level of omissions may have interfered with the discounting data by biasing the sampling of the choice responses by the rats, we modified the task to limit this effect. We hypothesized that testing only one delay at a time by increasing delays across sessions (between-DDT), would make each session shorter and less taxing. This new procedure was tested with another batch of rats with a similar pattern of DA denervation (Figure ). Even if lesioned rats displayed significantly higher levels of omissions than those of the sham group during forced- ( ) and free-choice trials (main effect of lesion: F = 5.62, p < 0.05, partial η = 0.18, of the delay: F = 126.49, p < 0.001, partial η = 0.89, and delay × lesion interaction: F = 5.64, p < 0.05, partial η = 0.26; Figure ), this effect was markedly reduced, as the difference between the groups was attributable only to performance on the 10 s delay. Even under these conditions, which controlled for the potential confounding influence of high omissions in lesioned rats, these data confirmed that the partial bilateral SNc DA lesion did not exacerbate impulsive choice (main effect of delay: F = 54.12, p < 0.001, partial η = 0.77, but no effect of lesion: F = 0.20, p = 0.66, partial η = 0.01, or delay × lesion interaction: F = 0.09, p = 0.96, partial η = 0.006; Figure ). This was further supported by an absence of difference between sham and lesioned rats in the AUC (no effect of lesion: F = 0.35, p = 0.55, partial η = 0.01, and no period × lesion interaction: F = 0.19 p = 0.66, partial η = 0.001; Figure ; see also ) and, at the population level, by the absence of relationship between DA denervation and AUC (Figure and ).
Partial nigrostriatal denervation does not affect impulsivity in a between-session DDT (between-DDT). (A) Percentage of omissions in sham and 6-OHDA lesioned rats during between-DDT. Data are represented as mean of omissions ±SEM in function of delay (averaged from each delay from the last three sessions). (B) Similar discounting pattern between sham and 6-OHDA lesioned rats, expressed as percentage of preference for the large reinforcer over the smaller one, in function of delay. Data are represented as mean ±SEM and were averaged from the last three sessions for each delay. (C) Similar discounting AUC, between sham and 6-OHDA lesioned rats, pre- and post-surgery period. The AUC is extrapolated from AUC and expressed as mean value ±SEM for each group. (D) No correlation between the post-surgical AUC and the degree of dorsal striatum TH-IR loss. Dots represent individual values for the AUC and dorsal striatum TH-IR loss expressed as percentage of sham mean value. (E) Positive and significant correlation between pre and post-surgery of individual AUC values for sham and 6-OHDA lesioned rats ( p s < 0.05). NS, non-significant, *** p < 0.001; Sham ( n = 12) vs. 6-OHDA ( n = 6), reinf, reinforcer; AUC, area under the discounting curve; a.u, arbitrary units.
In addition, and in contrast with the previous experiment, the lesion did not change the correlation between pre- and post-surgery AUC, indicating no influence on impulsivity trait in individuals (no difference in regression slope: t = 0.78, p > 0.40; Figure ; see also ).
## Discussion
Using a validated 6-OHDA lesion-based rodent model that was specifically designed for the investigation of non-motor, neuropsychiatric impairments related to PD (reviewed in Cenci et al., ; Magnard et al., ), we showed that bilateral and partial denervation of the nigrostriatal DA pathway neither induced nor exacerbate impulsive choice in two different DDTs, taking into account inter-individual variability at baseline.
The AUC, was used here as an empirical objective measure of discounting behavior and impulsivity trait in rats (Myerson et al., ; Odum, ). The discounting rate k factor has hitherto been a preferred index in clinical studies to assess intertemporal choice (e.g., Milenkova et al., ; Al-Khaled et al., ). Here, AUC was preferred to the k factor as an index of impulsivity, because it better accommodates the properties of the current dataset. Indeed, calculation of the k factor derives from the slope of the discounting function and relies on the indifference point (Broos et al., ), which is not necessarily crossed by all rats (especially low impulsive animals) under our experimental conditions. Together with the marked inter-individual differences observed in the present study in the time course of the discounting curve a fit-for-all model could not be implemented here. Nevertheless, the strong correlations observed between AUC obtained at different time points and test phases in sham conditions, especially in the between-sessions DDT, offer further evidence that it is a relevant and reliable measure of impulsive choice for longitudinal studies, as well as a useful and alternative tool for identifying endophenotypes of impulsive choice, in rats.
Although the discounting of the larger reinforcer over increasing delays was not influenced by nigrostriatal DA denervation, its inter-individual variability was markedly decreased by the lesion in the within- but not between-DDT. Interestingly, bilateral excitotoxic lesions of the dorsal striatum have also been shown to discretely flatten delay-preference function without affecting delay discounting in a within-session design version of the task (Dunnett et al., ). Because this effect progressively disappeared with extensive training, it had been attributed to a decrease ability to adjust behavior to the rapid modifications of task parameters across test sessions rather than to an alteration of impulsivity per se (Dunnett et al., ). Using similar experimental conditions (e.g., food restriction, similar pattern of DA denervation), the number of omissions drastically decreased in a between-DDT, in which the behavioral/motivational (less trials and shorter sessions) and cognitive (only one delay tested at any given time) demand, was reduced compared to a within-DDT. Consistently, Tedford et al. ( ) did not report any significant increase in omissions in another between-sessions DDT following dorsostriatal DA lesions. Together, these data suggest that nigrostriatal DA denervation may decrease the ability of the animal to properly engage in the task, and that between-session designs may be more appropriate than within-session designs to investigate the contribution of the nigrostriatal DA system to impulsive choice.
Although some clinical studies suggest a possible role of the neurodegenerative process in the development of impulsive behaviors in PD (Milenkova et al., ; Al-Khaled et al., ), conflicting results have been reported about the potential implication of a nigrostriatal DA deficit in different forms of impulsivity (Rokosik and Napier, ; Tedford et al., ; Engeln et al., ; Carvalho et al., ). This likely stems from the difficulty to disentangle an effect of nigrostriatal DA denervation on impulse control from its often dramatic effect on motor performance, alongside with the cognitive or motivational alterations potentially induced by DA lesions (Cenci et al., ).
Notably, it has been observed that a bilateral DA lesion of the dorsal striatum increased delay discounting in a between-session, wherein a similar range of ascending delays as the one used in the present study was applied (Tedford et al., ). However, substantial methodological differences between the two studies may account for these seemingly contradictory results. First, the retrograde lesion approach used in Tedford’s study led to motor dysfunctions, which could have impacted the coordinated sequences of actions necessary to perform the associated chained scheduled task. In addition, the use of a dorsostriatal retrograde lesion as opposed to the SNc anterograde lesion performed in the present study may lead to a different DA denervation pattern (e.g., Cenci et al., ) or different underlying DA deficits and compensatory mechanisms. Tedford et al. ( ) also used intracranial self-stimulation (ICSS) as a reinforcer in order to avoid satiety and other potential issues associated with food. However, chronic ICSS itself can induce profound neuroadaptations, such as overexpression of DA D receptors in the NAcc (Simon et al., ), a structure in which DA receptors modulate impulsive choice (Basar et al., ). Therefore, the increased delay discounting reported in this study may result from a direct neurobiological interaction between ICSS and the lesion.
Although the present study focuses exclusively on nigrostriatal cell loss and the emergence of impulsive choice, it should be kept in mind that other neurochemical systems, such as the serotoninergic and noradrenergic systems, have been shown to be affected by the neurodegenerative processes of PD (Delaville et al., ; Maillet et al., ). Due to their implication in the control of impulsive behaviors, alteration of these two monoaminergic systems, which, in the case of noradrenaline, may even precede the degeneration of DA neurons (Delaville et al., ), are prone to contribute, independently, or in conjunction with DA denervation, to the development of impulsivity in PD (Dalley and Roiser, ; Kehagia et al., ; Ye et al., ; Dalley and Robbins, ).
Nevertheless, our study provides novel insights into the contribution of the nigrostriatal DA system to impulsive choice and useful methodological considerations for future studies. It also highlights the fact that further investigations are necessary to better apprehend the potential contribution of the DA neurodegenerative process in conjunction with impulsivity trait and DRTs to the development of ICDs in PD.
## Author Contributions
RM, J-LH, MS, DB and SC designed research. RM, YV, CC, SB and SC performed research. RM, CC and SC analyzed data. RM, DB and SC wrote the manuscript with the help of the other authors.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
## Introduction
Cognitive training efficacy is controversial. Although many recent studies indicate that cognitive training shows merit, others fail to demonstrate its efficacy. These inconsistent findings may at least partly result from differences in individuals' ability to benefit from cognitive training in general, and from specific training types in particular. Consistent with the move toward personalized medicine, we propose using machine learning approaches to help optimize cognitive training gains.
## Cognitive Training: State-of-the-Art Findings and Debates
Cognitive training targets neurobiological mechanisms underlying emotional and cognitive functions. Indeed, Siegle et al. ( ) suggested that cognitive training can significantly improve mood, daily functioning, and cognitive domains. In recent years, various types of cognitive training have been researched. Frequently researched training types include cognitive bias modification (CBM) aims to modify cognitive processes such as interpretations and attention, making these more adaptive and accommodating to real-life demands (Hallion and Ruscio, ); inhibitory training seeks to improve inhibitory control and other executive processes, thus helping regulate behavior and emotion (Cohen et al., ; Koster et al., ); working memory training targets attentional resources, seeking to increase cognitive abilities by improving working memory capacities (Melby-Lervåg and Hulme, ). All these types demonstrated major potential in improving psychopathological symptoms or enhancing cognitive functions (Jaeggi et al., ; Hakamata et al., ).
Despite the accumulating body of evidence suggesting that cognitive training is a promising research path with major clinical potential, questions remain regarding its efficacy, and generalizability. Recent meta-analyses further corroborate this (for a discussion, see Mogg et al., ; Okon-Singer, ). For example, several research groups tested CBM studies using meta-analyses. Hakamata et al. ( ) analyzed twelve studies (comprising 467 participants from an anxious population), reporting positive moderate effects of training on anxiety symptom improvement. Yet two other meta-analyses focusing on both anxiety and depression (49 and 45 studies, respectively) demonstrated small effect sizes and warned of possible publication bias (Hallion and Ruscio, ; Cristea et al., ). These inconsistent results raise important questions about training efficacy. Several factors have been suggested as potential sources of this variability in effect size, including differences in inclusion criteria and quality of the studiesincluded (Cristea et al., ).
As in the CBM literature, meta-analyses of working memory training also yielded divergent results. Au et al. ( ) analyzed twenty working memory training studies comprising samples of healthy adults and reported small positive effects of training on fluid intelligence. The authors suggested that the small effect size underestimates the actual training benefits and may result from methodological shortcomings and sample characteristics, stating that “it is becoming very clear to us that training on working memory with the goal of trying to increase fluid intelligence holds much promise” (p. 375). Yet two other meta-analyses of working memory (87 and 47 studies, respectively) described specific improvements only in the trained domain (i.e., near transfer benefits) and few generalization effects in other cognitive domains (Schwaighofer et al., ; Melby-Lervåg et al., ). As with CBM, these investigations did not include exactly the same set of studies, making it difficult to infer the reason for the discrepancies. Nevertheless, potential factors contributing to variability in intervention efficacy include differences in methodology and inclusion criteria (Melby-Lervåg et al., ).
Some scholars suggested that the inconsistent results seen across types of training may be result from the high variability in training features, such as dose, design type, training type, and type of control groups (Karbach and Verhaeghen, ). For example, some studies suggest that only active control groups should be used and that using untreated controls is futile (Melby-Lervåg et al., ), while others discovered no significant difference between active and passive control groups (Schwaighofer et al., ; Weicker et al., ). Researchers have also suggested that the type of activity assigned to the active control group (e.g., adaptive or non-adaptive) may influence effect sizes (Weicker et al., ). Adaptive control activity may lead to underestimation of training benefits, while non-adaptive control activity may yield overestimation (von Bastian and Oberauer, ).
Training duration has also been raised as a potential source of variability. Weicker et al. ( ) suggested that the number of training sessions (but not overall training hours) is positively related to training efficacy in a brain injured sample. While only studies with more than 20 sessions demonstrated a long-lasting effect. In a highly influential working memory paper, Jaeggi et al. ( ) compared different numbers of training sessions (8–19). Outcomes demonstrated a dose-dependency effect: the more training sessions participants completed, the greater the “far transfer” improvements. In contrast, in a 2014 meta-analytical review Karbach and Verhaeghen reported no dose–dependency, as overall training time did not predict training effects. This is somewhat consistent with the findings of Lampit et al. ( ) meta-analysis, which indicated that only three or fewer training sessions per week were beneficial in training healthy older adults in different types of cognitive tasks. Furthermore, even time gaps between training sessions when the overall number of sessions is fixed may be influential. A study that specifically tested the optimal intensity level of working memory training revealed that distributed training (16 sessions in 8 weeks) was more beneficial than high intensity training (16 sessions in 4 weeks) (Penner et al., ). In sum, literature reviews maintain that this large variability in training hampers attempts to evaluate the findings (Koster et al., ; Mogg et al., ).
So far, the majority of studies in the field of cognitive training have been concerned mainly with establishing the average effectiveness of various training methods, with studies based on combined samples comprising individuals who profited from training and those who did not. Therefore, the samples' heterogeneity might be too high to evaluate efficacy for the “average individual” in each sample. We contend that focusing on the average individual contributes to the inconsistent findings, as is also the case with other interventions aimed at improving mental health (Zilcha-Mano, ). We argue that the inconsistent findings and large heterogeneity in studies evaluating cognitive training efficacy do not constitute interfering noise but rather provide important information that can guide us in training selection . In addition to selecting the optimal training for each individual, achieving maximum efficacy also requires adapting the selected training to each individual's characteristics and needs (Zilcha-Mano, ). In line with this notion, training games studies (i.e., online training platforms displayed in a game-like format) showcased different methods which personalized cognitive training by (a) selecting the type of training according to a baseline cognitive strengths and weaknesses evaluation or the intent of the trainee, and (b) adapting the ongoing training according to the individual's performance (Shatil et al., ; Peretz et al., ; Hardy et al., ). Until now, however, training personalization was made by pre-exist defined criteria and rationale (i.e., individual's weaknesses and strengths, individual's personal preference). Additional method for personalization, that is becoming increasingly popular in recent years, is data-driven personalization implemented by machine learning algorithms (Cohen and DeRubeis, ).
The observed variation in efficacy found in cognitive training studies may serve as a rich source of information to facilitate both intervention selection and intervention adaptation—the two central approaches in personalized medicine (Cohen and DeRubeis, ). Intervention selection seeks to optimize intervention efficacy by identifying the most promising type of intervention for a given individual based on as many pre-training characteristics as possible (e.g., age, personality traits, cognitive abilities). Machine learning approaches are especially suitable for such identification because they enable us to choose the most critical items for guiding treatment selection without relying on specific theory or rationale. In searching for a single patient characteristic that guides training selection, most approaches treat all other variables as noise. It is more intuitive, however, to hypothesize that no single factor is as important in identifying the optimal training for an individual as a set of interrelated factors. Traditional approaches to subgroup analysis, which tests each factor as a separate hypothesis, can lead to erroneous conclusions due to multiple comparisons (inflated type I errors), model misspecification, and multicollinearity. Findings may also be affected by publication bias because statistically significant moderators have a better chance of being reported in the literature. Machine learning approaches make it feasible to identify the best set of patient characteristics to guide intervention and training selection (Cohen and DeRubeis, ; Zilcha-Mano et al., ). With that said, given the flexibility of methods like decision tree analyses, there is a risk of overfitting that reduces validity for inference out of sample, such that the model will fit specifically the sample on which it was built and may be therefore unlikely to be generalizable in an independent application (Ioannidis, ; Open Science Collaboration, ; Cohen and DeRubeis, ). Thus, it is important to test out-of-sample prediction, either on a different sample or a sub-sample of the original sample on which the model was not built (e.g., cross-validation).
An example of treatment selection from the field of antidepressant medication (ADM) demonstrates the utility of this approach. Current ADM treatments are ineffective for up to half the patients, despite much variability in patient response to treatments (Cipriani et al., ). Researchers are beginning to realize the benefits of implementing machine learning approaches in selecting the most effective treatment for each individual. Using the gradient boosting machine (GBM) approach, Chekroud et al. ( ) identified 25 variables as most important in predicting treatment efficacy and were able to improve treatment efficacy in 64% of responders to medication—a 14% increase.
Whereas, training selection affects pre-treatment decision-making, training adaptation focuses on continuously adapting the training to the individual (see ). A patient's baseline characteristics (e.g., age, personality traits, cognitive abilities) and individual training performance trajectory can be used to tailor the training parameters (training type, time gaps between sessions, number of sessions, overall training hours) to achieve optimal performance. Collecting information from a sample of patients with similar baseline characteristics that underwent the same intervention yields an expected trajectory. Deviations from this expected trajectory act as warning signs and can help adapt the training parameters to the individual's needs (Rubel and Lutz, ).
Flow diagram of personalized cognitive training process.
An example of treatment adaptation comes from the field of psychotherapy research, where a common treatment adaptation method involves providing therapists with feedback on their patients' progress. This method was developed to address the problem that many therapists are not sufficiently aware of their patients' progress. While many believe they are able to identify when their patients are progressing as expected and when not, in practice this may not be true (Hannan et al., ). Many studies have demonstrated the utility of giving therapists feedback regarding their patients' progress (Lambert et al., ; Probst et al., ). Shimokawa et al. ( ) found that although some patients continue improving and benefitting from therapy (on-track patients—OT), others seem to deviate from this positive trajectory (not-on-track patients—NOT). These studies provided clinicians feedback on their patients' state so they could better adapt their therapy to the patients' needs. This in turn had a positive effect on treatment outcomes in general, especially outcomes for NOT patients, to the point of preventing treatment failure. These treatment adaptation methods have recently evolved to include implementations of the nearest neighbor machine learning approach originating in avalanche research (Brabec and Meister, ), as well as other similar approaches to better predict an individual's optimal trajectory and identify deviations from it (Rubel et al., ).
Machine learning approaches may thus be beneficial in the efforts of progressing toward personalized cognitive training. The inconsistencies between studies in terms of the efficacy of CBM, inhibitory training, and working memory training can serve as a rich and varied source to guide the selection and adaptation of effective personalized cognitive training. In this way, general open questions such as optimal training duration and time gaps between sessions will be replaced with specific questions about the training parameters most effective for each individual.
## Author Contributions
RS managed the planning process of the manuscript, performed all administrative tasks required for submission and drafted the manuscript. HO-S and SZ-M took part in planning, supervision, brainstorming, and writing the manuscript. ST took part in brainstorming and writing the manuscript.
### Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Nicotine is acknowledged as the key addictive compound of tobacco. Varenicline (Champix or Chantix ), mainly acting as a partial agonist at the α4β2 nicotinic receptor, is an approved smoking cessation pharmacotherapy, although with efficacy limited to a portion of smokers. Smokers differ in the motives that drive their drug seeking and Varenicline might be more efficient in some groups more than others. Studies in rodents revealed that nicotine-seeking is strongly supported by complex interactions between nicotine and environmental cues, and notably the ability of nicotine to enhance the reinforcing properties of salient environmental stimuli. It is not yet understood whether the decrease of nicotine-seeking by acute Varenicline in rats results from antagonism of the primary reinforcing effects of nicotine, of the reinforcement-enhancing effect of nicotine on cues, or of a combination of both. Thanks to a protocol that allows assessment of the reinforcement-enhancing effect of nicotine on cues during self-administration in rats, we showed that Varenicline targets both nicotine reinforcing effects and reinforcement-enhancing effect of nicotine on cues. Importantly, individual variations in the latter determined the amplitude of acute Varenicline-induced decrease in seeking. These results suggest that Varenicline might be more beneficial in smokers who are more sensitive to nicotine effects on surrounding stimuli.
## Introduction
Tobacco dependence continues to be a worldwide health burden, being responsible for as many as 7 million deaths per year (WHO, ). More than 70% of smokers wish to quit (U.S. Department of Health and Human Services, ), but less than 10% succeed without medical support (Rigotti, ). Even so, a major obstacle in ceasing to smoke is the limited efficacy of available treatments against tobacco dependence (Schuit et al., ). For instance, from all patients treated with Varenicline (Champix or Chantix ), one of the most effective approved pharmacotherapies in supporting smoking cessation (Cahill et al., ; Hartmann-Boyce et al., ), only 40% remain abstinent at the end of a 12-week-long treatment, while post-treatment abstinence rates drop to 20% in the following months after treatment cessation (Oncken et al., ; Niaura et al., ; Jordan and Xi, ).
Varenicline is a full agonist at the α7-, and a partial agonist at the α4β2-containing nicotinic cholinergic receptors (Coe et al., ; Rollema et al., , ), which mediate the primary reinforcement properties of nicotine, the major psychoactive compound of tobacco (Benowitz, ). However, the relatively weak primary reinforcement of nicotine cannot explain the pervasiveness of tobacco abuse alone (Caggiula et al., ; Rose, ). Recent studies have highlighted that nicotine can increase the reinforcing value of environmental cues that are primary reinforcers by themselves, or that have acquired reinforcing value through pairing with another reinforcer (Caggiula et al., ; Rupprecht et al., ). The interplay between nicotine and environmental cues is complex and difficult to disentangle, but plenty of evidence suggests it is a determinant factor in tobacco seeking (Caggiula et al., , ; Garcia-Rivas and Deroche-Gamonet, ). Importantly, newer evidence suggests that smokers differ in the psychobiological mechanisms that drive their nicotine-seeking (for review, see Garcia-Rivas and Deroche-Gamonet, ). In this regard, understanding the psychopharmacological dimensions of nicotine-seeking that are being affected by Varenicline could clarify its limited efficacy. However, the numerous studies that have shown that Varenicline can acutely decrease nicotine self-administration in rodents (Rollema et al., ; O’Connor et al., ; Le Foll et al., ; Funk et al., ), have done so in experimental conditions that do not clearly allow the disentangling of the psychopharmacology of Varenicline against nicotine and nicotine-cue interactions.
Furthermore, even though the effects of Varenicline on nicotine-cue interactions have also been subject to extensive studies (Levin et al., ; Schassburger et al., ; Barrett et al., ), they have been studied in conditions under which nicotine is not self-administered. Varenicline has been shown to dose-dependently antagonize the reinforcement-enhancing effect caused by nicotine (Levin et al., ). Consistent with its nature as a partial agonist, it has also been shown that Varenicline can enhance responding for a visual cue in a dose-dependent manner, although with a much weaker effect than nicotine (Barrett et al., ). This last result is consistent with a previous study, which used self-administration of Varenicline and a visual cue self-administered through two different levers, to reveal such reinforcement-enhancing effect of Varenicline (Schassburger et al., ). However, since the psychopharmacological actions of Varenicline in humans are of therapeutic relevance when nicotine intake is volitional, the testing of Varenicline effects on passive nicotine administration has weaker face validity when compared to the classical approach of drug self-administration (Panlilio and Goldberg, ).
Thus, the precise psychopharmacological mechanisms through which Varenicline opposes nicotine self-administration in rodents is still not well understood, but warrant further investigation. Because a key determinant of the synergistic interaction between nicotine and a salient cue is the primary reinforcing effects of the cue (Chaudhri et al., ; Caggiula et al., ), we developed an experimental procedure that allows for increasing these primary reinforcing effects during self-administration and tested the effect of Varenicline while contingently manipulating the reinforcing-enhancing effect of nicotine on the cue.
## Materials and Methods
### Subjects
Male Sprague–Dawley rats (Charles River, France), weighing 280–300 g at the beginning of the experiments, were single housed under a 12 h reverse dark/light cycle. In the animal house, temperature (22 ± 1°C) and humidity (60 ± 5%) were controlled. Rats were habituated to environmental conditions and experimental handling for 15 days before surgery. Standard chow food and water were provided ad libitum . All procedures involving animal experimentation and experimental protocols were approved by the Animal Care Committee of Bordeaux (CEEA50, N° 50120168-A) and were conducted in accordance with the guidelines of the European Union Directive 2010/63/EU regulating animal research.
### Surgeries
A silastic catheter (internal diameter = 0.28 mm; external diameter = 0.61 mm; dead volume = 12 μl) was implanted in the right jugular vein under ketamine (80 mg/kg)/xylazine (16 mg/kg) anesthesia. The proximal end reached the right atrium through the right jugular vein, whereas the back-mount passed under the skin and protruded from the mid-scapular region. Rats were given 5–7 days recovery before nicotine self-administration training began.
### Drugs
Ketamine hydrochloride (80 mg/kg; Imalgène 1000; Rhône Mérieux, Lyon, France) and xylazine hydrochloride (16 mg/kg; Rompun; Rhône Mérieux, Lyon, France) were mixed with saline and administered intraperitoneally in a volume of 2 ml/kg of body weight. (-)-Nicotine-Hydrogen-Tartrate (Glentham, UK) was dissolved in sterile 0.9% physiological saline for a final dose of 0.04 mg/kg free base. Nicotine, as well as sterile 0.9% physiological saline in control groups, was self-administered by the rats via intravenous (i.v.) route in a volume of 40 μl per self-infusion. Nicotine solution was adjusted to a pH of 7.
Varenicline or 7,8,9,10-Tetrahydro-6, 10-methano-6H-pyrazino[2,3-h] benzazepine tartrate (Tocris, UK) was dissolved in sterile 0.9% physiological saline for a final dose of 1 mg/kg free base. Varenicline was administered intraperitoneally (i.p.) 30 min prior to self-administration, in a volume of 2.5 ml/kg.
### Intravenous Self-administration
#### Self-administration Apparatus
The self-administration setup consisted of 48 self-administration chambers made of plexiglas and metal (Imetronic, France), and equipped with holes as operant manipulanda. Each chamber (40 cm long × 30 cm width × 36 cm high) was located in an opaque sound-attenuating cubicle equipped with an exhaust fan to assure air renewal and mask background noise ( ). For self-administration sessions, each rat was placed in one chamber where its chronically implanted intravenous catheter was connected to a pump-driven syringe (infusion speed: 20 μl/s). Two holes, located at opposite sides of the chamber at 5.5 cm from the grid floor, were used to record instrumental responding. In given experimental groups and experiments, a common white light (white LED, Seoul Semiconductor, South Korea), 1.8 cm in diameter, located 11.5 cm above the active hole, was used as nicotine (or saline) delivery-associated discrete visual cue, and is named thereafter “cue light” or “cue.” It produced 5 Lux. As well, in given experimental groups and experiments, a blue light (blue LED, Sloan Precision Optoelectronics, Switzerland), 1.8 cm in diameter, located on the opposite wall at 17 cm of the floor on the left side, was used as, and is named thereafter, “Ambient light” and abbreviated AL . It produced 15 Lux at a wavelength of 470 nm, which is known to not affect vision in Sprague–Dawley rats in a similar exposure pattern as in our experimental approach (Tosini et al., ). LED intensities were both measured in the middle of the cage with a Lux-meter (Moineau Instruments, France). Experimental contingencies were controlled and data were collected with a PC-windows-compatible SK_AA software (Imetronic, France).
#### Self-administration Procedures
In the three experiments presented below, self-administration testing began 2 h after the onset of the dark phase. Nose-poke in the active hole under a fixed ratio three schedule of reinforcement (FR3) produced the activation of the infusion pump (40 μl over 2 s). Nose-pokes at the inactive hole were recorded but had no scheduled consequences. Rats in all protocols of self-administration described in this study were placed under an FR3 schedule of reinforcement from the first session onwards, with the reinforcer varying according to the experimental group in which they were allocated ( ). Neither food-training nor FR-1 transition period was used. Whatever the reinforcer, rats were trained 2 h daily, 5 days per week, from Monday to Friday, except for the first session of Experiments 1 and 3 that took place on a Tuesday. To maintain catheter patency, catheters were flushed with ~10 μl of heparinized saline (30 IU/ml) after each self-administration session and before the self-administration sessions run on Monday.
Experimental Protocols. (A) Experiment 1. Three groups of rats ( saline + cue , n = 10; nicotine , n = 25; nicotine + cue , n = 8) were trained for self-administration for 27 sessions. The Nicotine group was substantially larger than the other two experimental groups, as it was expected that only around 40%–50% of animals in this group would acquire self-administration criteria. An acute IP injection of Varenicline was applied 30 min before session 28 of self-administration. (B) Experiment 2. Two groups of rats were trained for saline + cue self-administration. For one group ( AL , n = 15), the Ambient light was on throughout the first seven sessions. For the other group ( No AL , n = 15) the Ambient light was off during the same period. On the eighth session of self-administration, the Ambient light conditions were switched; removed for the AL group and inserted for the No AL group. On sessions 9 and 10, the No AL group was split into two, with half of the rats switched back to their original No AL condition (Single AL insertion, n = 7), while the other half remaining under the new AL condition ( Sustained AL insertion , n = 8). All rats from the AL group remained without the Ambient light for sessions 9 and 10 ( Sustained AL removal ). (C) Experiment 3. Two groups of rats were trained for nicotine + cue self-administration, using the same AL and No AL conditions as in Experiment 2. The AL group was substantially larger ( n = 36) than the control No AL condition ( n = 19), as it was expected that AL could delay acquisition of nicotine + cue self-administration. Similar to Experiment 2, the AL conditions were switched in Session 23, after which rats were returned to basal conditions. On session 28, the switch of AL conditions was re-applied, with the addition of a Varenicline IP injection 30 min before session. Rats were then allowed to return to a stable baseline before a final test using a single Varenicline injection on a basal self-administration.
In Experiment 1, to define a significant self-administration behavior at the individual level, we used a discrimination index between active and inactive holes [(active nose-pokes/total nose-pokes)*100] strictly superior to 50%, together with a minimal number of at least six self-infusions per session over three consecutive sessions and with stability in the number of self-infusions (±10%) over the last two sessions.
### Experimental Procedures
#### Effect of Varenicline on Self-administration Behavior Reinforced by Either a Discrete Cue Light, a Nicotine Infusion or a Combination of Both Nicotine and Cue Light (Experiment 1)
Nose-poking in the active hole at FR3 was reinforced either by an infusion of 0.04 mg/kg nicotine free base ( nicotine , n = 25), a nicotine 0.04mg/kg infusion plus a discrete cue light ( nicotine + cue , n = 8), or a saline infusion plus a discrete cue light ( saline + cue , n = 10; ). For the nicotine group, following nose-poking in the active hole at FR3 the infusion pump was activated for 2 s. For the nicotine + cue and saline + cue groups, nose-poking in the active hole at FR3 turned on the cue light located above the hole, simultaneous to the activation of the infusion pump. The cue light remained on for 4 s in total. Since it is known that nicotine alone is poorly self-administered in the absence of other salient stimuli (Caggiula et al., ), the nicotine group was substantially larger than the other two experimental groups, as it was expected based on preliminary data that only 40%–50% of animals in this group would meet the desired self-administration criteria.
After 27 daily basal sessions ( ), rats showing significant self-administration behavior were administered with Varenicline (1 mg/kg, ip) 30 min prior to a basal self-administration session. The average number of infusions over training sessions 26–27 was used as the baseline. The Varenicline dose was chosen based on previous literature (e.g., O’Connor et al., ).
#### Effect of Varenicline on the Reinforcement-Enhancing Effect of Nicotine During Nicotine + Cue Self-administration
##### A Procedure to Alter the Primary Reinforcing Effects of the Cue Light (Experiment 2)
A key determinant of the interaction between nicotine infusion and an associated discrete cue light relies on the primary reinforcing effect of the cue. A key issue is then to be able to manipulate the reinforcing effect of the cue during nicotine self-administration. The goal of Experiment 2 was to establish a protocol where the reinforcing effects of the cue can be altered. Therefore, we tested in rats self-administering saline + cue whether we could decrease or increase the primary reinforcing effects of the cue by altering its visual salience, by either adding or removing an interfering ambient light, respectively.
Two groups of rats were trained for saline + cue self-administration, as described in Experiment 1 except that for one group ( AL , n = 15), the Ambient light ( AL ) was on throughout the first seven acquisition sessions. For the other group ( No Ambient light , No AL , n = 15) the AL was off during the same period ( ). On the eighth session of self-administration, the Ambient light conditions were switched; turned off for the AL group and on for the No AL one. On sessions 9 and 10, the No AL group was split into two, with half of the rats switched back to their original No AL condition ( Single AL Insertion subgroup, n = 7), while the other half remaining under the new AL condition ( Sustained AL Insertion subgroup, n = 8). All rats from the AL group remained without the AL for sessions 9 and 10.
##### Effect of Varenicline on the Reinforcement-Enhancing Effect of Nicotine During Nicotine + Cue Self-administration (Experiment 3)
Based on the results of Experiment 2, two groups of rats were trained for nicotine + cue self-administration, as described in Experiment 1. As in Experiment 2 the AL was on throughout the basal training self-administration sessions, for one group ( AL , n = 36), and was off for the other one ( No AL , n = 19; ). The AL group was substantially larger than the control No AL condition, as it was expected that the AL could delay acquisition of nicotine + cue self-administration. On session 23, we tested the effect of: (1) suppressing; and (2) adding, the AL on self-administration in the AL and No AL groups, respectively. Rats were then brought back to the respective basal conditions until session 28, when we tested the effect of Varenicline (1 mg/kg, i.p.) administered 30 min prior to session during which the AL was manipulated, i.e., suppressed in the AL group and inserted in the No AL group. Rats were then returned to basal conditions, and once responding was stable over two consecutive sessions and had returned to the level of infusions of sessions 21–22, we tested the effect of Varenicline (1 mg/kg, i.p.) administered 30 min prior to a basal session.
### Data Analyses
#### Self-administration
Total responses in the active and inactive holes and total number of infusions per self-administration session were considered.
#### Effect of Varenicline and/or AL Manipulation
To evaluate Varenicline and/or AL manipulation ( AL removal or AL insertion), delta infusions from baseline (infusions at test − infusions at baseline) were calculated. Baseline infusions correspond to the mean infusions over the two sessions preceding a test.
### Statistical Analyses
Self-administration behavior was analyzed using repeated measures ANOVA with Time (number of sessions), Hole (active vs. inactive), Test (Baseline vs. Test), Condition ( AL On to AL Off, AL Off to AL On, AL On to AL Off+Var, AL Off to AL On+Var), as within-subject factor, and experimental group ( saline + cue / nicotine + cue / nicotine , AL / No AL ) as between-subject factor.
Significant main effects or interactions were explored by pairwise comparisons of means using the Newman Keuls post hoc test. Pearson’s correlation analyses were used to investigate the correlation between variables of interest. A t -test was used to compare the AL Removal effects (or of AL Insertion effects) on saline + cue and nicotine + cue self-administration.
The results are presented as mean ± SEM. Differences were considered significant at p < 0.05.
The statistical analyses were performed using the STATISTICA 13.3.0 (2017) data analysis software system (TIBCO Software Inc., Palo Alto, CA, USA).
## Results
### Nicotine and a Cue Light Contribute Synergistically to Self-administration (Experiment 1)
Over the first 15 self-administration sessions, saline + cue, nicotine + cue and nicotine rats differed significantly regarding number (Group, F = 10.77, p < 0.001) and pattern (Group × Session, F = 6.7, p < 0.0001) of reinforcers earned ( ), as well as number and discrimination in responses ( ).
Nicotine and infusion-associated discrete cue light contribute synergistically to self-administration behavior. Operant nose-poking at FR3 in active hole was reinforced by the delivery of an intravenous infusion of saline associated with the lighting of a salient visual cue above the active hole ( saline + cue ), of a nicotine intravenous infusion associated with the lighting of a salient visual cue above the active hole ( nicotine + cue ) or of the sole delivery of a nicotine intravenous infusion ( nicotine ). (A) Infusions earned per session over the 15 first behavioral sessions. (B) Responses in the active and inactive holes per session over the 15 first behavioral sessions. Symbols denote group mean and error bars denote SEM. (C) Mean infusions earned in basal conditions ( Baseline ) and after Varenicline administration (1 mg/kg i.p., 30 min prior to session) in rats self-administering saline + cue , nicotine + cue or nicotine . For Baseline , infusions are averaged over the two last sessions prior to Varenicline test. (D) Effect of Varenicline as calculated by the delta between infusions earned in baseline and infusions earned under Varenicline effect, in rats self-administering saline + cue , nicotine + cue or nicotine . Symbols and bars denote group mean and error bars denote SEM. p < 0.0001 as compared to respective session 1. * p < 0.05, *** p < 0.001. p < 0.05 and p < 0.001 as compared to respective baseline. p < 0.01, p < 0.001, as compared to zero.
Nicotine first tended to compromise, but secondarily amplified, the reinforcing effects of a discrete cue light. Thus, nicotine + cue rats increased self-infusions from session 1 to session 6 ( p < 0.0001) while the saline + cue rats showed the opposite profile ( p < 0.0001) when the nicotine rats remained stable over the same sessions ( p = 0.87). The compared self-administration patterns of the three groups suggest that nicotine and cue interact synergistically.
### Nicotine and Saline + Cue Are Mild but Different Reinforcers (Experiment 1)
The behavior of the saline + cue and the nicotine groups stabilize at a similar level from session 6 ( ). Observations exclude, however, that the behavior is just driven by the stimulus that is common to the two groups, i.e., intravenous infusion. Indeed, up to session 6, the saline + cue group produced a higher number of self-infusions than the nicotine one (Group, F = 8.5, p < 0.01) and the two profile of self-infusions differ with decrease, and progressive increase, up to stabilization, respectively (Group × Session, F = 5.7, p < 0.0005). Also, in a preliminary experiment, eight rats were trained for saline + cue for 13 sessions in conditions similar to the ones described in Experiment 1. Omission of the cue on session 14 produced a significant decrease in self-administration ( ) supporting that the cue contributes to the reinforcing effects in saline + cue rats.
The mild reinforcing effects in nicotine and saline + cue rats, as compared to nicotine + cue rats, were further confirmed when using threshold criteria for discrimination, i.e., number of infusion and stability in behavior (see “Materials and Methods” section), to define a significant self-administration behavior at the individual level. By session 15, only 40% of the nicotine rats (10/25) had acquired self-administration, compared to 100% of the nicotine + cue rats (8/8), and 50% of the saline + cue rats (5/10; ).
Distribution of the individual scores of self-infusions in the rats showing self-administration based on these criteria ( ) also further supports the difference in nature of the reinforcers acting in the nicotine and the saline + cue groups. show the self-infusions and hole responses in rats, which either reached ( ) or did not reach ( ) these criteria.
### Varenicline Decreases Nicotine + Cue and Nicotine Self-administration (Experiment 1)
After 27 sessions, the effect of Varenicline on self-administration was tested in the saline + cue ( n = 5), nicotine + cue ( n = 8) and nicotine ( n = 11) rats that met self-administration criteria evaluated on behavior during sessions 26 and 27. Varenicline decreased self-administration as measured by the number of self-infusions earned (Test effect, F = 30.6, p < 0.0001). This effect was function of the experimental group (Test × Group, F = 4.71, p < 0.05) with a significant effect in rats self-administering nicotine + cue ( p < 0.0001) and nicotine ( p < 0.05; ). According to the effect on self-infusions, Varenicline decreased nose-poking in a group-dependent (Test effect, F = 22.49, p < 0.0001; Test × Group, F = 4.55, p < 0.05) and hole-dependent manner (Test × Hole, F = 28.4, p < 0.0001), exclusively targeting the active hole ( ).
The effect of Varenicline, as measured by the delta-infusions from baseline (Group effect, F = 3.29, p < 0.05), was higher in the nicotine + cue group than in the saline + cue ( p < 0.05) and nicotine groups ( p < 0.05), in which the delta-infusions were similar ( ). However, the effect of Varenicline was different from zero in the nicotine group ( p < 0.0001), but not in the saline + cue group. Notably, in the nicotine group, the Varenicline effect, as measured by delta-infusions from baseline, did not correlate with basal self-infusions ( data not shown ).
### Varenicline Targets the Reinforcing-Enhancing Effect of Nicotine on Its Associated Salient Cue
Results of Experiment 1 supported that nicotine and the cue interact to produce reinforcing effects, and that Varenicline significantly decreased the nicotine + cue combined reinforcer. However, it did not allow concluding whether Varenicline was specifically targeting this interaction. To further explore this hypothesis, we aimed at testing the effect of Varenicline while manipulating this nicotine-cue interaction in the same individuals. As a first step, we aimed at developing a procedure that would allow promoting (vs. compromise) the nicotine-induced enhancement of the reinforcing properties of its associated cue. As this enhancement is depending on the primary reinforcing effects of the cue, we initially worked on a procedure allowing to increase (vs. decrease) these reinforcing effects.
#### An Interfering Ambient Light (AL) Appears to Alter the Primary Reinforcing Effects of the Discrete Cue Light (Experiment 2)
As in Experiment 1 rats self-administered saline + cue , as shown by a significant discrimination between active and inactive holes over the seven sessions of self-administration (Hole effect, F = 28.7, p < 0.0001; ). However, this discrimination was a function of the experimental group. The AL appears to compromise the expression of the reinforcing effects of the discrete cue light (Group effect, F = 10.4, p < 0.01). In standard conditions ( No AL ), saline + cue induced self-administration behavior, while in the AL condition, with the same saline + cue reinforcer, rats did not discriminate significantly between active and inactive holes (Group × Hole, F = 18.7, p < 0.0001). In the standard No AL condition, although behavior decreased over sessions, discrimination remained significant, up to the last session ( p < 0.005).
Not only No AL rats discriminated between the inactive control hole and the active hole associated with saline + cue delivery ( ), but they also earned significantly more reinforcers than the AL rats (Group effect, F = 8, p < 0.01; ).
An interfering ambient light (AL) appears to alter the primary reinforcing effects of a salient discrete cue light. (A) Infusions earned per session over seven behavioral sessions during which operant nose-poking in the active hole was reinforced at FR3 by the delivery of an intravenous infusion of saline associated with the lighting of a cue light above the active hole. The presence of a 15 Lux Ambient light ( AL ) reduced self-administration behavior as compared to the control condition ( No AL ). (B) Effect on infusions earned of AL Removal and AL Insertion in rats trained for saline + cue self-administration over seven sessions in the AL and No AL conditions, respectively. Basal infusions are averaged over the two last sessions prior to AL Insertion (or Removal) test. The interfering AL delays acquisition of nicotine + cue self-administration. (C) Infusions earned per session over the first 19 behavioral sessions during which operant nose-poking in the active hole was reinforced at FR3 by the delivery of an intravenous infusion of nicotine associated with the lighting of a cue light above the active hole. (D) Effect on infusions earned of AL Removal and AL Insertion in rats trained for nicotine + cue self-administration in the AL and No AL conditions, respectively. The interfering AL procedure allows revealing the reinforcement-enhancing effect of nicotine on its associated salient cue. (E) Comparison of AL Removal and AL Insertion effects in rats trained for saline + cue or nicotine + cue self-administration. While AL Insertion in No AL rats produced a similar decrease in saline + cue and nicotine + cue rats (bottom), AL Removal produced a stronger increase in nicotine + cue rats (top). Symbols and bars denote group mean and error bars denote SEM. * p < 0.05, ** p < 0.01, *** p < 0.001.
It is unlikely that the absence of discrimination and the reduced number of reinforcers in AL rats was due to a non-specific stress-like or aversive effect. First, the number of inactive nose-poking was not affected ( ), suggesting that the AL effect may be targeting the reinforcement of the cue light. Second, the switch of the AL conditions on session 8 further attested that the AL compromises the cue light reinforcing effects. While AL Insertion decreased self-administration, AL removal increased it (Condition × Group, F = 7.7, p < 0.01; ).
To better understand the effect of AL Removal and Insertion, No AL rats were split into two groups for the following two sessions (9 and 10): one group ( Sustained AL Insertion , n = 8), maintained the newly acquired AL condition, while the other ( Single AL Insertion , n = 7) returned to their No AL condition ( ). Sustained AL Insertion further diminished self-administration in sessions 9 and 10, compared to sessions 6 and 7, while rats in the Single AL Insertion group appeared to compensate by increasing their mean infusions, when back to the initial No AL condition ( ). In the case of the Sustained AL Removal rats, the removal of the AL was maintained for sessions 9 and 10, further increasing self-administration in comparison to sessions 6 and 7 ( ).
#### The Interfering AL Procedure Appears to Reveal the Reinforcement-Enhancing Effect of Nicotine on Its Associated Salient Cue During Nicotine Self-administration (Experiment 3)
Having revealed that it was possible to increase the reinforcing effects of the cue by AL Removal, we tested its effect on nicotine + cue self-administration, both on acquisition and once behavior was established.
During acquisition under the No AL condition, the number of nicotine + cue self-infusions was higher than under the AL condition (Group effect, F = 5.36, p < 0.05), but the difference decreased over the 20 self-administration sessions (Group × Session, F = 4.14, p < 0.0001) and the AL group reached and maintained the level of self-infusions of the No AL group by session 15 ( ).
Rats in the AL condition did not discriminate between active and inactive holes in the first session, contrary to No AL condition ( ). Even though inactive nose-poking was similar in the AL and No AL conditions from session 2, in a manner similar to saline + cue self-administration, active responding in the AL condition remained low compared to No AL conditions up to session 5.
Once stabilized, removal of the AL increased self-administration behavior by the AL group (Test effect, F = 47.9, p < 0.0001), while insertion of the AL decreased self-administration behavior by the No AL group (Test effect, F = 24.46, p < 0.001; ).
As for saline + cue , it is unlikely that AL compromised nicotine + cue self-administration due to a non-specific stress-like or aversive effect. Notably, during the first self-administration session ( ), total responses were not lower in AL rats, and absence of discrimination between active and inactive holes resulted from equal high responses in inactive and active holes, and not reduced responses in the active hole.
Critically, as summarized in , the effect of the AL removal was much more pronounced in nicotine + cue conditions compared to saline + cue conditions ( t -test, p < 0.01), suggesting that any increase in visual salience of the cue is magnified by nicotine. By comparison, introduction of the AL had the same effect in both nicotine + cue and saline + cue conditions, suggesting a non-specific effect on visual perception, which is not potentiated by nicotine.
#### Varenicline Targets the Reinforcement-Enhancing Effect of Nicotine on Its Associated Salient Cue (Experiment 3)
Once stabilized, self-administration behavior by the AL group was altered by removal of the AL , by Varenicline or a combination of both (Test effect, F = 64.8, p < 0.0001). According to the condition tested, the test effect was different however (Test × Condition, F = 76.3, p < 0.0001). AL removal alone produced an increase ( , red bar) in nicotine + cue self-administration ( p < 0.001). When AL removal was combined with Varenicline administration, Varenicline abolished completely the effect of AL Removal and decreased nicotine + cue self-administration below AL Baseline ( , dashed red bar, p < 0.01 vs. AL Baseline ). However, this latter effect was of a lower extent than when Varenicline was applied in the basal self-administration conditions, i.e., with maintenance of the AL ( p < 0.001; , gray bar). Critically, Varenicline and AL Removal effects were not simply additive. When evaluating the effect of AL Remov + Var to the effect of AL Remov alone, one yields an effect which is much higher than the one of Varenicline alone on basal self-administration, suggesting that Varenicline specifically abolishes the enhancing effects of the AL Removal ( ). Noteworthy, this interpretation is supported by the correlation analysis ( ) showing a strong inverse correlation between the effect of Increased Cue Salience by AL Removal (Δ AL Remov = AL Remov − AL baseline) and the calculated Var effect during Increased Cue Salience by AL Removal (Δ AL Remov + var − Δ AL Remov). Varenicline treatment during Increased Cue Salience by AL Removal appears to reduce infusions from an amount equivalent to the increase produced by the Increased Cue Salience. In other words, in these AL Removal conditions, Varenicline appears to decrease specifically the individual increase produced by AL Removal, i.e., the individual potentiation of nicotine + cue self-administration produced by the Increased Cue Salience.
Varenicline targets the reinforcement-enhancing effect of nicotine on its associated salient cue. (A) Infusions earned in rats trained for nicotine + cue self-administration in the presence of the interfering AL ( AL Baseline ), in response to Varenicline ( Var ), to AL Removal ( AL Remov ) or a combination of both ( AL Remov + Var ). (B) Comparison of Varenicline effect in AL Baseline condition (Infusions Var AL Baseline − Infusions AL Baseline ) and in Increased Cue Salience condition [by AL Removal ; calculated from the combined effect of AL Removal and Varenicline (Infusions AL Remov + Var − Infusions AL Baseline ) minus the effect of AL Removal (Infusions AL Remov − Infusions AL Baseline )]. Varenicline absolute effect was amplified in the Increased Cue Salience condition (by AL Removal) . (C) Almost 1 to 1 negative correlation between the effect of Increased Cue Salience and the calculated effect of Varenicline on Increased Cue Salience . The individual increase in nicotine + cue infusions by Increased Cue Salience was antagonized by Varenicline. (D) Infusions earned in rats trained for nicotine + cue self-administration in the absence of the interfering AL ( No AL Baseline), in response to Varenicline ( Var ), to AL Insertion ( AL Insert ) or a combination of both ( AL Insert + Var ). (E) Comparison of Varenicline effect in No AL baseline condition (Infusions Var No AL baseline—Infusions No AL Baseline ) and in Decreased Cue Salience condition [by AL Insertion ; calculated from the combined effect of AL Insertion and Varenicline (Infusions AL Insert + Var − Infusions No AL Baseline ) minus the effect of AL Insertion (Infusions AL Insert − Infusions No AL Baseline )]. Varenicline absolute effect was similar in the two conditions. (F) Correlation between the effect of Decreased Cue Salience [by AL Insertion ] and the calculated effect of Varenicline on Decreased Cue Salience (by AL Insertion ). Bars denote group mean and error bars denote SEM. Data points reflect individual scores. ** p < 0.01, *** p < 0.001.
Self-administration behavior by the No AL group was decreased by insertion of the AL , by Varenicline or a combination of both (Test effect, F = 4.4, p < 0.05; ). According to the condition tested, the test effect was different however (Test × Condition, F = 9.3, p < 0.001). Insertion of the AL , in rats trained in absence of it, produces a significant decrease in nicotine + cue self-administration ( , blue bar), which was similar in amplitude to the effect of Varenicline ( , gray bar). When combined with AL Insertion, Varenicline amplified the effect of the AL Insertion ( , dashed gray bar). Notably, the combined effect of AL Insertion and Varenicline were not synergistic but additive as shown in . When subtracting the AL Insert effect from the AL insert + Var effect, to get the Var effect on decreased cue salience , the result was similar to the effect of Varenicline alone ( Var effect alone ; ). Although less strong, similarly to the effect of Varenicline on Increased Cue Salience by AL Removal, there was a correlation between the decreased effect of AL Insertion on self-administration and the effect of Varenicline on this AL Insertion effect ( ), supporting that Varenicline had a bi-directional effect on the nicotine-induced increase cue reinforcement, depending on how the AL manipulation altered said cue reinforcement.
## Discussion
Varenicline is acknowledged as one of the most efficient therapeutic tools for tobacco dependence. However, its efficacy is limited both in time and to a portion of patients (Oncken et al., ; Niaura et al., ; Jordan and Xi, ). Even though the molecular pharmacology of Varenicline is well-known (Coe et al., ; Rollema et al., ), its psychopharmacological actions are still poorly understood. In this study, we evidenced that acute Varenicline reduced nicotine-induced enhancement of the reinforcing properties of a nicotine-paired cue during intravenous self-administration. This effect appeared to depend on how much nicotine-cue interactions were contributing to self-administration behavior at the individual level. Conversely, the decrease by acute Varenicline of self-administration of nicotine alone appeared not related to individual basal levels of self-administration.
### Nicotine Alone Is a Poor Primary Reinforcer, but Is Strong Enough to Drive Self-administration in Certain Individuals, but Not in Others
Nicotine has weak primary reinforcement properties. Hence, classical nicotine self-administration has been developed to pair contingent nicotine IV delivery with the presentation of a salient visual cue light (Caggiula et al., ). A discrete cue light alone can act as a primary reinforcer in drug naïve rats (Deroche-Gamonet et al., ). In our study, we used the saline + cue condition as a control group evidencing the contribution of the cue in driving self-administration behavior. Comparison with the nicotine + cue group reveals the actual contribution of nicotine in nicotine + cue self-administration behavior.
In our study, by session 15, 100% of all rats trained in nicotine + cue condition showed criteria of significant self-administration behavior, but only 40% of all rats trained in the nicotine condition reached the same criteria. These results not only confirm the well-known observation described by Caggiula and colleagues, but it extends it with the observation that some rats appear much more sensitive to the reinforcing properties of nicotine, thus driving nicotine self-administration despite the lack of salient environmental cues, supporting that individuals may vary in the mechanisms that drive their nicotine-seeking (Garcia-Rivas and Deroche-Gamonet, ).
### A Novel Procedure That Allows Targeting the Reinforcing-Enhancing Effects of Nicotine on Its Associated Salient Cue During Nicotine Self-administration
In a previous study, Palmatier et al. ( ) demonstrated that the reinforcement-enhancing effects of nicotine on visual cues are dependent on the strength of the primary reinforcement of such cues in a nicotine-naïve state, with a stronger enhancing effect observed for visual cues with higher primary reinforcement properties. Further studies have assessed the effect of Varenicline on this nicotinic enhancement of cue reinforcement, but in conditions that are different from volitional nicotine intake (Levin et al., ; Barrett et al., ). Here, we developed a novel experimental approach that attempted a sudden increase in the visual salience of the nicotine-paired cue, through the removal of an interfering Ambient light ( AL ). This approach allowed us to explore the observations by Palmatier et al. ( ), but in the context of nicotine self-administration, and within the same individuals.
A possible explanation for the interfering effect of the Ambient Light (AL) in seeking behavior could be a non-specific aversive or stressful effect, rather than a reduction in the reinforcing effects of the cue. However, this explanation appears unlikely. The aversive effect of an ambient stressor would have impacted both active and inactive responding, while this is not the case. Critically, in the first nicotine + cue self-administration session, total responding was similar whether the Ambient Light was present or not. It is noteworthy that the presence of the AL delayed the acquisition of self-administration of nicotine + cue , which became equivalent to that of the No AL condition starting session 17. Overall, this data suggests that the effect of the AL is due to a reduction of the visual salience of the cue through visual interference, rather than a mere stress effect caused by the AL . Further studies, including progressive ratio schedules of reinforcement, could validate the interfering role of AL in cue reinforcement.
Importantly, the increase in self-administration due to removal of the visual interference was much more pronounced in nicotine + cue conditions compared to saline + cue conditions, supporting a nicotine-specific effect. This difference could be explained by the different value of the cue in these two conditions. In the saline + cue condition, the cue is acting as a primary reinforcer (Deroche-Gamonet et al., ). In the nicotine + cue condition, the cue is both a primary and a secondary reinforcer, and both reinforcing effects can be enhanced further by nicotine itself (Caggiula et al., ). However, it is more likely that the strong nicotine-specific increase in responding after AL removal is due to the magnifying effect by nicotine on a sudden increase in cue reinforcing effects, whether primary or secondary in nature. Supporting this view, previous studies show that nicotine can increase the reinforcement and incentive salience of cues that have already reinforcing value, whether primary or secondary (Donny et al., ; Chaudhri et al., ; Palmatier et al., , ; Rupprecht et al., ). It thus follows that any increase in salience of nicotine-paired cues would be magnified even further by nicotine, as supported by our study. No other study to date has specifically addressed this possibility. By comparison, decreasing the cue salience by introduction of the AL has the same decreasing effect on both nicotine + cue and saline + cue self-administration, suggesting in this instance a non-specific decrease in visual perception, which is not altered by nicotine.
### Varenicline Targets the Reinforcing Effects and Reinforcing-Enhancing Effects of Nicotine on Its Associated Cue
In accordance with the literature (Rollema et al., ; O’Connor et al., ; Le Foll et al., ; Funk et al., ), we showed that Varenicline 1 mg/kg reduces nicotine + cue self-administration. We were interested in exploring whether such robust decrease in self-administration is due to Varenicline affecting nicotine reinforcement, nicotine-cue interactions, or a combination of both. Here we demonstrated that acute Varenicline also decreases behavior in rats self-administering nicotine alone, although to a lesser absolute extent. In the same conditions, acute Varenicline has no effect on the self-administration of the salient visual cue by itself.
A limitation in exploring Varenicline effects on the sole reinforcing effects of nicotine is that these are relatively weak, and even for those rats that acquired nicotine self-administration without the presence of a nicotine-paired cue, their baseline nicotine-seeking behavior is substantially lower than for nicotine + cue self-administration. This could compromise the detection of Varenicline effects, as decreases in responding are less evident when the baseline responding is already low. In trying to bypass this limitation, a recent article by Kazan and Charntikov ( ) studied the role of Varenicline in nicotine reinforcement through a behavioral economics approach. Briefly, they trained rats to self-administer nicotine + cue through daily escalated FR schedules of reinforcement, calculated the individual baseline demand for nicotine, and assessed the individual effect of Varenicline as a function of nicotine demand. They show that individual demand for nicotine predicted the individual reduction in self-administration after a Varenicline challenge. This could look contrary to our results (i.e. absence of correlation between basal self-infusions and Varenicline effect on basal self-administration in the nicotine group - experiment 1) because escalation of schedules of reinforcement is supposed to bring into evidence the role of nicotine reinforcement. However, the nicotine + cue protocol used by Kazan and Charntikov ( ) cannot disentangle the primary reinforcement of nicotine from the reinforcement-enhancing effect of nicotine on the associated visual stimulus. The same protocol with nicotine as the sole reinforcer would help clarify the case.
Our study also complements previous findings in clarifying the reinforcing-enhancing effects of Varenicline on a visual cue: namely, that these effects are only observed when individuals have been previously exposed to nAChR agonists. Contrary to our study, Clemens et al. ( ) and Barrett et al. ( ) showed that acute Varenicline increased the self-administration of a visual cue alone in the absence of nicotine. Furthermore, Levin et al. ( ) briefly reports in drug-naïve animals, the reinforcing-enhancing effects of Varenicline on visual cues. However, and differently to our case, in these studies, rats had been previously exposed to either nicotine or Varenicline. In Clemens et al. ( ), rats had been previously trained for nicotine + cue self-administration and Varenicline tested after seven self-administration of the cue alone, through a nicotine extinction-like procedure. In Barrett et al. ( ), Varenicline was tested following a history of repeated passive exposure to nicotine administered after the cue self-administration sessions. In Levin et al. ( ), the authors make a brief comment that the reinforcing-enhancing effects of Varenicline were evident in the first seven sessions of repeated Varenicline exposure, although it remains unknown if the reported effects were already substantial during the first session. It is noteworthy that in these three cases, the reinforcing-enhancing effects of Varenicline appear similar, regardless of whether the nicotinic agonist was present at the moment of cue self-administration (Levin et al., ; Clemens et al., ) or disconnected from it (Barrett et al., ). In our study, the lack of previous history with nAChR agonists in saline + cue rats could thus explain the lack of previously described reinforcing-enhancing effects of Varenicline (Levin et al., ; Clemens et al., ; Barrett et al., ). This temporal requirement could most probably involve upregulation of α4β2-containing nAChRs, caused by chronic exposure to both nicotine (Marks et al., ; Buisson and Bertrand, ; Staley et al., ) and Varenicline (Marks et al., ). Nicotine, however, is known for its acutely reinforcing-enhancing effect of stimuli, even in drug-naïve individuals (Rupprecht et al., ; Perkins et al., ). This supports that Varenicline does not necessarily reproduce a nicotine-like increase in cue reinforcing effects, but requires a cholinergic system already sensitized to nicotinic agonists, which makes rats more sensitive to the reinforcing-enhancing effect of nicotinic agonists to cues. In addition, within the same study by Levin et al. ( ), Varenicline 1 mg/kg both failed and succeeded to increase the reinforcing effects of a visual stimulus in two distinct experiments with similar design, obscuring any consistent interpretation of the effect of Varenicline at this dose. Possibly, the effect of varenicline in enhancing the reinforcement of visual stimuli could be better seen at lower varenicline doses, as reported by Levin et al. ( ), which we failed to observe in this study. Further studies using different varenicline doses are needed to explore this possibility.
### Varenicline Targets the Reinforcement-Enhancing Effect of Nicotine on Its Associated Cue During Self-administration
Using a novel visual interfering procedure, we evidenced that Varenicline appears to specifically reduce the reinforcement-enhancing effects of nicotine on surrounding cues during nicotine self-administration.
Varenicline effect on nicotine self-administration was bi-directional, depending on how individuals responded to the manipulation of the AL : the more AL removal increased self-administration, the stronger the effect of varenicline in opposing cue salience ( ), while the less AL insertion decreased self-administration, the stronger the effect of varenicline in decreasing cue salience ( ). This correlation was stronger for the AL removal condition. It is possible that the weaker correlation in the AL insertion condition is related to a lower number of rats tested. Nevertheless, these results add to the evidence shown by Kazan and Charntikov ( ), that Varenicline’s effects appear dependent on individual differences in nicotine reinforcement. To our knowledge, we are the first to report an effect of Varenicline that is dependent on the strength of nicotine-cue interactions: a stronger nicotine-cue interaction is associated with a stronger Varenicline effect. This observation supports the rationale for individual variations in the mechanisms of nicotine-seeking (Garcia-Rivas and Deroche-Gamonet, ), with some individuals being more sensitive than others to the influence of the reinforcement-enhancing effect of nicotine on environmental cues, and who could differently benefit from Varenicline treatment.
It has been previously shown that the reinforcement-enhancing effect of nicotine on cues is not only dependent on α4β2-containing nAChRs (Liu et al., ), but also on the dopaminergic system (Palmatier et al., ). Given the precise molecular pharmacology of Varenicline, a possible mechanism for Varenicline could be antagonism at the α4β2-containing nAChRs located in the ventral tegmental area (VTA), thus reducing the nicotine-induced tonic firing of dopaminergic neurons, leading to decreased tonic release of dopamine in the nucleus accumbens (NAcc; Crunelle et al., ). Such a mechanism could also be involved in the effect of Varenicline on the primary reinforcing effects of nicotine, which are also thought to be dependent on VTA to NAcc signaling (Di Chiara, ; Picciotto and Corrigall, ). However, acute Varenicline appears to target the former, as a function of individual response, but not the latter. An alternative mechanism could involve α7 nAChRs, or other structures in the circuitry controlling nicotine-cue interactions, such as the basolateral amygdala, an area rich in α4β2- and α7 nAChRs (Feduccia et al., ) and also involved in drug-cue interactions (Janak and Tye, ).
In our study, we have investigated the psychopharmacological targets of Varenicline during early nicotine + cue self-administration. Future studies should address whether prolonged exposure to nicotine changes the way Varenicline affects nicotine and nicotine + cue self-administration. The differential effects of Varenicline in nicotine + cue self-administration in short vs. prolonged exposure to nicotine might depend on the experimental approach: George et al. ( ) reports that Varenicline does not differently affect rats with long access to nicotine (23-h sessions) compared to short access (1-h session). The study by Clemens et al. ( ) on the other hand, shows that after an extended training (40 sessions) with a short access protocol, Varenicline seems to also target the reinforcing properties of nicotine alone, compared to early training (20 sessions). However, the specificity of this Varenicline effect is problematic, as the decrease is seen both in active and inactive responding. These results warrant further exploration.
Furthermore, as a treatment for tobacco cessation, daily doses of Varenicline are recommended in the week leading up to a cessation attempt, with continuous daily administration over the following 11 weeks after cessation (Ebbert et al., ). While our study only assessed the effect of an acute exposure to 1 mg/kg Varenicline, further studies need to assess if prolonged exposure to Varenicline affects the psychopharmacological dimensions of nicotine-seeking during nicotine self-administration in a different way than those after acute exposure. Studies with repeated Varenicline administration have been performed but focused on the reinforcing effects of a visual cue either in rats never exposed to nicotine (Levin et al., ) or previously administered with passive nicotine injections (Barrett et al., ).
Despite this, our results raise therapeutic implications. Increasing clinical and preclinical data suggests that smokers differ in the mechanisms that drive their nicotine-seeking (Garcia-Rivas and Deroche-Gamonet, ), with some smokers having stronger sensitivity to the primary reinforcing actions of nicotine (Hutchison et al., ; Esterlis et al., ), while others being more sensitive to the effects of nicotine on surrounding cues (Perkins, ; Perkins et al., ; Van Heel et al., ). Our results support individual variations in both nicotine reinforcing effects and nicotine-induced enhancement of cue reinforcing effects in the rat. Our data also suggest that individual variations in nicotine-induced enhancement of cue reinforcing effects, but not individual variations in nicotine reinforcing effects, would determine the amplitude of acute Varenicline-induced decrease in seeking during volitional administration of nicotine. Altogether, Varenicline might be more beneficial for smoking cessation in those who are especially sensitive to nicotine effects on surrounding cues, and not for those who are more sensitive to the primary reinforcing effects of nicotine. Further studies need to clarify more precisely the action of Varenicline, using a preclinical model that would allow for the fine exploration of individual differences in the mechanisms that drive nicotine-seeking (Garcia-Rivas et al., ).
## Data Availability
The datasets generated for this study are available on request to the corresponding author.
## Ethics Statement
All procedures involving animal experimentation and experimental protocols were approved by the Animal Care Committee of Bordeaux (CEEA50, N° 50120168-A) and were conducted in accordance with the guidelines of the European Union Directive 2010/63/EU regulating animal research.
## Author Contributions
VG-R, NC and VD-G designed the experiments. VG-R, J-FF, NC, MC-G, PR and JT performed the research. VG-R, J-FF, NC and VD-G analyzed the data. VG-R and VD-G wrote the article.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
The pineal gland transduces photoperiodic changes to the neuroendocrine system by rhythmic secretion of melatonin. We recently provided new evidence that the pineal gland is a major neurosteroidogenic organ and actively produces a variety of neurosteroids de novo from cholesterol in birds. Notably, allopregnanolone is a major pineal neurosteroid that is far more actively produced in the pineal gland than the brain and secreted by the pineal gland in juvenile birds. Subsequently, we have demonstrated the biological action of pineal allopregnanolone on Purkinje cells in the cerebellum during development in juvenile birds. Pinealectomy (Px) induces apoptosis of Purkinje cells, whereas allopregnanolone administration to Px chicks prevents cell death. Furthermore, Px increases the number of Purkinje cells that express active caspase-3, a crucial mediator of apoptosis, and allopregnanolone administration to Px chicks decreases the number of Purkinje cells expressing active caspase-3. It thus appears that pineal allopregnanolone prevents cell death of Purkinje cells by suppressing the activity of caspase-3 during development. This paper highlights new aspects of the biosynthesis and biological action of pineal allopregnanolone. |
The trigger for synaptic vesicle exocytosis is Ca(2+), which enters the synaptic bouton following action potential stimulation. However, spontaneous release of neurotransmitter also occurs in the absence of stimulation in virtually all synaptic boutons. It has long been thought that this represents exocytosis driven by fluctuations in local Ca(2+) levels. The vesicles responding to these fluctuations are thought to be the same ones that release upon stimulation, albeit potentially triggered by different Ca(2+) sensors. This view has been challenged by several recent works, which have suggested that spontaneous release is driven by a separate pool of synaptic vesicles. Numerous articles appeared during the last few years in support of each of these hypotheses, and it has been challenging to bring them into accord. We speculate here on the origins of this controversy, and propose a solution that is related to developmental effects. Constitutive membrane traffic, needed for the biogenesis of vesicles and synapses, is responsible for high levels of spontaneous membrane fusion in young neurons, probably independent of Ca(2+). The vesicles releasing spontaneously in such neurons are not related to other synaptic vesicle pools and may represent constitutively releasing vesicles (CRVs) rather than bona fide synaptic vesicles. In mature neurons, constitutive traffic is much dampened, and the few remaining spontaneous release events probably represent bona fide spontaneously releasing synaptic vesicles (SRSVs) responding to Ca(2+) fluctuations, along with a handful of CRVs that participate in synaptic vesicle turnover. |
Gene expression studies employing real-time PCR has become an intrinsic part of biomedical research. Appropriate normalization of target gene transcript(s) based on stably expressed housekeeping genes is crucial in individual experimental conditions to obtain accurate results. In multiple sclerosis (MS), several gene expression studies have been undertaken, however, the suitability of housekeeping genes to express stably in this disease is not yet explored. Recent research suggests that their expression level may vary under different experimental conditions. Hence it is indispensible to evaluate their expression stability to accurately normalize target gene transcripts. The present study aims to evaluate the expression stability of seven housekeeping genes in rat granule neurons treated with cerebrospinal fluid of MS patients. The selected reference genes were quantified by real time PCR and their expression stability was assessed using GeNorm and NormFinder algorithms. GeNorm identified transferrin receptor (Tfrc) and microglobulin beta-2 (B2m) the most stable genes followed by ribosomal protein L19 (Rpl19) whereas β-actin (ActB) and glyceraldehyde-3-phosphate-dehydrogenase (Gapdh) the most fluctuated ones in these neurons. NormFinder identified Tfrc as the best invariable gene followed by B2m and Rpl19. ActB and Gapdh were the least stable genes as analyzed by NormFinder algorithm. Both methods reported Tfrc and B2m the most stably expressed genes and Gapdh the least stable one. Altogether our data demonstrate the significance of pre-validation of housekeeping genes for accurate normalization and indicates Tfrc and B2m as best endogenous controls in MS. ActB and Gapdh are not recommended in gene expression studies related to current one. |
<i>Tumour suppressor candidate 7</i> (<i>TUSC7</i>) is a novel tumor suppressor gene generating long non-coding RNA (lncRNAs) in several types of human cancers. The expression and function of <i>TUSC7</i> in human brain glioma has yet to be elucidated. In this study, <i>TUSC7</i> was poorly expressed in tissues and cell lines of glioma, and the lower expression was correlated with glioma of the worse histological grade. Moreover, TUSC7 is a prognostic biomarker of glioma patients. Up-regulation of <i>TUSC7</i> suppressed cellular proliferation and invasion of glioma cells, and accelerated cellular apoptosis. Bioinformatics analysis showed that TUSC7 specifically binds to miR-23b. MiR-23b was up-regulated in glioma and negatively correlated with the expression of TUSC7. The miR-23b expression was inhibited remarkably by the upregulation of TUSC7 and the reciprocal inhibition was determined between TUSC7 and miR-23b.RNA pull-down and luciferase reporter assays were used to validate the sequence-specific correlation between miR-23b and TUSC7. TUSC7 inhibited the proliferation, migration and invasion of glioma cells and promoted cellular apoptosis largely bypassing miR-23b. We conclude that the lncRNA TUSC7 acted as a tumor suppressor gene negatively regulated by miR-23b, suggesting a novel therapeutic strategy against gliomas. |
Peripheral nerves have the capacity to conduct action potentials along great distances and quickly recover following damage which is mainly due to Schwann cells (SCs), the most abundant glial cells of the peripheral nervous system (PNS). SCs wrap around an axonal segment multiple times, forming a myelin sheath, allowing for a significant increase in action potential conduction by insulating the axons. Mature myelin consists of compact and non-compact (or cytoplasmic) myelin zones. Non-compact myelin is found in paranodal loops bordering the nodes of Ranvier, and in the inner and outermost cytoplasmic tongues and is the region in which Schmidt-Lanterman incisures (SLI; continuous spirals of overlapping cytoplasmic expansions within areas of compact myelin) are located. Using different technologies, it was shown that the layers of non-compact myelin could be connected to each other by gap junction channels (GJCs), formed by connexin 32 (Cx32), and their relative abundance allows for the transfer of ions and different small molecules. Likewise, Cx29 is expressed in the innermost layer of the myelin sheath. Here it does not form GJCs but colocalizes with K<sub>v</sub>1, which implies that the SCs play an active role in the electrical condition in mammals. The critical role of GJCs in the functioning of myelinating SCs is evident in Charcot-Marie-Tooth disease (CMT), X-linked form 1 (CMTX1), which is caused by mutations in the <i>gap junction protein beta 1</i> (<i>GJB1</i>) gene that codes for Cx32. Although the management of CMT symptoms is currently supportive, there is a recent method for targeted gene delivery to myelinating cells, which rescues the phenotype in KO-Cx32 mice, a model of CMTX1. In this mini-review article, we discuss the current knowledge on the role of Cxs in myelin-forming SCs and summarize recent discoveries that may become a real treatment possibility for patients with disorders such as CMT. |
As a sensitive cold-shock protein, RNA binding protein motif 3 (RBM3) exhibits a neuroprotective function in the condition of brain injury. However, how RBM3 is involved in acute ischemic stroke by affecting stress granules (SGs) remains unclear. Here, we established an oxygen-glucose deprivation/reperfusion (OGD/R) model in rat primary cortical neurons and PC12 cells to explore the potential mechanism between RBM3 and SG formation in acute ischemic/reperfusion (I/R) condition. The immunofluorescence results showed that the SG formation significantly decreased in rat primary cortical neurons and PC12 cells during the reperfusion period after 6 h of OGD stimulation. The western blot results, flow cytometry analysis, and cell viability assessment showed that the RBM3 expression and ratio of cell viability significantly decreased, while the rate of apoptosis increased in PC12 cells during the reperfusion period after 6 h of OGD stimulation. Co-immunoprecipitation (Co-IP) and immunofluorescence indicated that RBM3 and GTPase-activating protein-binding protein 1 (G3BP1) colocalized cytoplasm of PC12 cells after 6 h of OGD stimulation when the SGs formation reached the highest level. Besides, overexpression and knockdown of the RBM3 were achieved via plasmid transfection and CRISPR-Cas9 technology, respectively. The results of overexpression and knockdown of RBM3 gene illustrated the pivotal role of RBM3 in affecting SG formation and apoptosis level in OGD-treated PC12 cells. In conclusion, RBM3 could combine with G3BP1 resulted in increasing stress granules generation in rat primary cortical neurons and PC12 cells after 6 h of oxygen-glucose deprivation (OGD) injury, which ultimately reduced the apoptosis in OGD-induced cells. Our study may enable a new promising target for alleviating ischemia-reperfusion injury in cells. |
[This retracts the article on p. 517 in vol. 9, PMID: 26793066.]. |
Neurodegenerative diseases including Alzheimer (AD) and Parkinson (PD) have attracted attention in last decades due to their high incidence worldwide. The etiology of these diseases is still unclear; however the role of the environment as a putative risk factor has gained importance. More worryingly is the evidence that pre- and post-natal exposures to environmental factors predispose to the onset of neurodegenerative diseases in later life. Neurotoxic metals such as lead, mercury, aluminum, cadmium and arsenic, as well as some pesticides and metal-based nanoparticles have been involved in AD due to their ability to increase beta-amyloid (Aβ) peptide and the phosphorylation of Tau protein (P-Tau), causing senile/amyloid plaques and neurofibrillary tangles (NFTs) characteristic of AD. The exposure to lead, manganese, solvents and some pesticides has been related to hallmarks of PD such as mitochondrial dysfunction, alterations in metal homeostasis and aggregation of proteins such as α-synuclein (α-syn), which is a key constituent of Lewy bodies (LB), a crucial factor in PD pathogenesis. Common mechanisms of environmental pollutants to increase Aβ, P-Tau, α-syn and neuronal death have been reported, including the oxidative stress mainly involved in the increase of Aβ and α-syn, and the reduced activity/protein levels of Aβ degrading enzyme (IDE)s such as neprilysin or insulin IDE. In addition, epigenetic mechanisms by maternal nutrient supplementation and exposure to heavy metals and pesticides have been proposed to lead phenotypic diversity and susceptibility to neurodegenerative diseases. This review discusses data from epidemiological and experimental studies about the role of environmental factors in the development of idiopathic AD and PD, and their mechanisms of action.
## Introduction
Life expectancy has increased in last decades and health care improvements have contributed to people living longer. However, this has also contributed to increase the number of people with chronic disabling diseases such as Alzheimer (AD) and Parkinson (PD). Genesis of both neurodegenerative diseases has not been elucidated and several endogenous (genetic) and exogenous (environment) factors contribute to the onset and/or development of these illnesses, which highlights the necessity to expand the research on identifying the environmental risk factors that predispose to the development of these neurodegenerative diseases.
It is known that the etiology of neurodegenerative diseases is multifactorial, and there is evidence that potential external factors including lifestyle and chemical exposures are linked with the risk of the onset of these diseases. Since the vast majority of AD and PD cases are observed in elderly populations, yet the exposure to risk factors occurred years or decades before the diagnosis, the assessment of chronic exposures is difficult to perform in retrospective studies to associate them with the onset/development of the disease. Therefore more research for better definition of exposure, as well as for the identification of early specific biomarkers for the diagnosis of these diseases is needed. Attention is now focused on environmental factors that potentially damage the developing nervous system through epigenetic mechanisms, resulting in neurodegenerative diseases later in life. In this review we briefly examined the evidence of environmental etiologies related to two of the most common neurodegenerative diseases, AD and PD, from epidemiological as well as experimental studies.
### Alzheimer’s Disease
Alzheimer’s disease is the major form of dementia in elderly and possibly contributes to 60–70% of cases. It is a progressive, disabling and irreversible disease (Goedert and Spillantini, ). There are two recognized forms of AD. The first one is named familial or of early onset (EOAD), which is directly related to specific gene mutations in the amyloid precursor protein (APP) and presenilin (PSEN) 1 and 2 genes, both related to the amyloid-beta (Aβ) peptide synthesis (Piaceri et al., ). The EOAD begins at early age, less than 65 years, and only explains 5% of all cases. The second one, the late-onset or sporadic AD (LOAD) is the most common form of AD with 95% of all cases. This form of AD is not caused by punctual mutations, but some genetic risk factors have also been described such as polymorphisms in ApoE (encoding for apolipoprotein E), SORL1 (encoding for neuronal receptor of ApoE), and GSK3 (encoding for glycogen synthase kinase 3 beta) genes. The ApoE gene is the strongest genetic risk factor for LOAD, although it is not sufficient to explain the occurrence of the disease (Godfrey et al., ). Therefore, the etiology of LOAD remains unclear, and it is suggested that it has a multifactorial etiology.
Two hypotheses have been most studied for AD development. One is related to the overproduction of the Aβ peptide. According to this, neurofibrillary tangles (NFTs) result from the onset of amyloid deposits as Aβ plaques. While the second hypothesis suggests that the hyperphosphorylation of the Tau protein and its subsequent deposition as NFTs is the ultimate responsible for the disease. The amyloid cascade hypothesis establishes that Aβ aggregation initiates the brain damage leading to memory loss and to AD (Hardy and Higgins, ). Aβ is normally produced during aging, mediated by the proteolytic processing of the APP by the amyloidogenic enzymatic pathway. In this pathway, APP is processed by β- and γ-secretase complexes producing Aβ, soluble APPβ (sAPPβ) and the amyloid intracellular domain (AICD). Alternatively, APP can be processed by the non-amyloidogenic pathway leading to the production of AICD, sAPPα but not Aβ (Thinakaran and Koo, ). Thus Aβ increased levels in the brain of LOAD patients could be mediated by: (i) an increase in APP expression; (ii) an increase in the amyloidogenic pathway; or (iii) a reduction in the non-amyloidogenic pathway. It is stablished that the increase of a member of the β secretase complex, BACE1 (beta-site APP cleaving enzyme 1) produces high brain Aβ levels (Sun et al., ). On the other hand, the reduction on the activity of ADAM10 (a desintegrin and metalloproteinase domain-containing protein 10) could also lead to the overproduction of Aβ (Kojro and Fahrenholz, ). Furthermore, some mutations such as those in PSEN 1 or 2, a catalytic member of the γ-secretase complex, can also increase the production of Aβ (Piaceri et al., ).
Another mechanism to increase brain Aβ levels is through a reduction in the Aβ degradation. There are proteins collectively known as Aβ-degrading enzyme (IDE)s that have the ability to degrade Aβ, including insulin-like IDE, angiotensin-converting enzyme (ACE), endothelin-converting enzyme (ECE), plasmin, cathepsin B, aminopeptidase A, matrix metalloproteinase (MMP) 2 and 9, neprilysin (NEP, neutral endopeptidase) and others. These enzymes have been suggested as viable therapeutic targets for AD treatment (Nalivaeva et al., ). Finally, a reduced brain clearance of Aβ can be another pathway for the brain accumulation of Aβ. Some cholesterol transporters such as the low density lipoprotein receptor-related protein 1 (LRP1) are involved in the Aβ export from the brain to the cerebrospinal fluid (CSF). This receptor links the imbalance of cholesterol homeostasis with AD pathogenesis (Zlokovic et al., ).
On the other hand, aggregates of the microtubule (MT)-associated protein Tau observed in cell bodies and apical dendrites as NFTs cause neurofibrillary lesions associated with AD. Tau is a phosphoprotein mainly localized in the axon of neurons for the stabilization of MTs; it contains a high number of serine and threonine residues, and is therefore a substrate of many kinases (Goedert et al., ). The abnormal aggregation of Tau into insoluble paired helical filaments (PHFs), which are the major component of NFTs found in cell bodies and apical dendrites of neurons are lesions associated with AD (Friedhoff et al., ). Under pathological conditions, Tau is hyperphosphorylated at “pathological” sites leading to MTs depolymerization, axonal transport disruption and aggregation (Götz, ). It has been proposed that repeat domains (RD) of the MT-binding domain (MBD) in the C-terminal structure of Tau can rapidly form PHFs compared with the complete protein, suggesting that RDs are indispensable for its aggregation (Wille et al., ), and for Tau filament formation (Tokimasa et al., ).
There is no cure for AD, and therapeutic treatments are basically to ameliorate the symptoms. Therefore, an early and opportune diagnosis is indispensable to slow the progression of the disease. Currently, the determination of Tau and Aβ levels in blood and CSF are broadly used for the diagnosis of AD, and several medical tools are also used to confirm the diagnosis including the medical history, mental status tests, and evaluations of the brain structure and function with neuroimaging techniques (Lewczuk et al., ). However, these biomarkers are not sensitive nor specific for AD. Interestingly, an emerging body of evidence suggests that micro RNAs (miRNAs, small non-coding RNAs involved in the post-transcriptional regulation of gene expression) could be putative biomarkers for detecting neurodegeneration Thus, recent reviews have shown that some miRNAs are differentially associated with AD by modulating the expression of important genes involved in Aβ production (e.g., BACE1 ) or inflammation (Goodall et al., ; Van den Hove et al., ).
#### Aβ Homeostasis as a Target of Environmental Factors
Environmental factors such as diet (fat-rich), heavy metals, biogenic metals and pesticides have been involved in AD development due to their ability to disrupt metabolic pathways involved in the homeostasis of Aβ. In addition, factors such as lifestyle (antioxidants and exercise) can prevent AD development (Figure ). Many of these environmental factors are oxidative agents acting through different mechanisms as discussed later. The brain is particularly vulnerable to oxidative stress do to its high glucose-based metabolic rate, low levels of antioxidants, high levels of polyunsaturated fatty acids, and high enzymatic activities related to transition metals that catalyze the formation of free radicals (Halliwell et al., ). In addition, micromolar concentrations of Aβ induce the formation of H O in culture cells leading to neurotoxicity, and the presence of some antioxidant enzymes prevents the toxicity of the peptide (Butterfield et al., ). The mechanism by which Aβ generates free radicals is not known, and other endogenous factors also generate reactive oxygen species (ROS) in AD. For instance, the ion Fe , which is at high concentrations in NFTs and Aβ-aggregates, catalyzes the formation of reactive species such as H O , as well as advanced glycation end products (AGE) that are related to neurodegeneration (Smith et al., ). On the other hand, activated microglia that surrounds the senile plaques is a source of NO and O (Cras et al., ), which can react to form the peroxinitrite radical (ONOO-) (Smith et al., ). Likewise, inflammation has gained importance in AD pathogenesis (Tuppo and Arias, ). The central nervous system is considered a privileged site with its own immune system and microglia and astrocytes are the principal cells involved in the inflammatory response. It is accepted the microglial chemotaxis of Aβ and the phagocytosis of amyloid fibrils, effects that produce an increase in the secretion of pro-inflammatory cytokines and ROS, which in consequence produces neuronal loss (Rogers and Lue, ). In agreement, astrocytes are also recruited into amyloid plaques for Aβ degradation (Wyss-Coray et al., ), and it is possible that the activation of microglia and astrocytes is a consequence of Aβ aggregation. The role of inflammatory processes in AD is supported by the use of non-steroidal anti-inflammatory drugs (NSAIDs) to reduce the Aβ levels (Weggen et al., ), and the risk of AD (Etminan et al., ).
Environmental factors associated with Alzheimer’s disease (AD) development through different mechanisms . Several factors including metals, pesticides, nanoparticles, and diet can affect the two targets of AD such as Aβ generation and Tau phosphorylation. The figure depicts the molecular targets than can be modified at different levels following the amyloid hypothesis that ends in Aβ senile plaques formation (upper part) or the hyperphosphorylation of Tau protein and its subsequent deposition as neurofibrillary tangles (NFTs) (lower part). For more detail see the text.
##### Metals
Lead (Pb) is a heavy metal well known by its neurological toxic effects, although a direct association with AD development has not been reported. Pb affects cognitive abilities, intelligence, memory, speed processing and motor functions in children (Mason et al., ), while studies in elderly are limited. A cohort study reported that bone Pb levels were associated with poor cognitive performance scores in old workers, suggesting that past Pb exposure can contribute to late cognitive deterioration (Dorsey et al., ). However, a recent study reported no association between serum Pb levels and AD (Park et al., ). Despite the few epidemiological studies relating Pb exposure with AD, the evidence is more solid in experimental studies. The influence of Pb in AD was initially suggested from results in rats early exposed (from postnatal day 1–20) or late exposed (at 18–20 months of age) to Pb (200 ppm, drinking water). Results showed an increase in the APP mRNA expression late in life after the neonatal exposure, but not in rats exposed as adults (Basha et al., ). Similarly, a study performed in non-human primates ( Macaca fascicularis ) exposed to Pb (1.5 mg/Kg/day, from birth to 400 days) showed that monkeys exposed at a young age had an increased number of amyloid plaques late in life (at 23 years old) (Wu et al., ). The increased Aβ levels appear to be mediated by an augmented expression of APP and BACE1 (Wu et al., ). These effects were also observed in differentiated SH-SY5Y cells incubated with Pb (5–100 µM/48 h) and analyzed 6 days later (Bihaqi and Zawia, ). Another study performed in differentiated SH-SY5Y cells showed an increase in Aβ secretion and APP expression, as well as reduced expression and protein levels of NEP (an Aβ IDE), suggesting that Pb can target both the synthesis and degradation of Aβ (Huang et al., ). However, in a recent work conducted in our laboratory, Pb did not show changes in NEP expression in differentiated SH-SY5Y cells exposed to 50 µM Pb, but an increase in APP levels (Chin-Chan et al., ). Another mechanism by which Pb increases Aβ levels is by reducing the brain Aβ clearance. A recent study showed that acute Pb exposure (27 mg/Kg, i.p.) to APP transgenic mice (V717F) reduced the expression of LRP1 , resulting in the accumulation of Aβ in the hippocampus and cortex of treated mice (Gu et al., ). Studies from this group support that Pb can disrupt the brain export of Aβ leading to its accumulation and plaques formation (Behl et al., , ).
Exposure to Pb during development is a good example of an environmental contaminant as a risk factor to promote neurodegenerative diseases such as AD, supporting the hypothesis that many adult diseases have a fetal origin (FeBAD) (Basha et al., ). The group of Zawia has extensively worked on latent responses to prenatal and early postnatal exposures to Pb. Authors exposed male neonatal rats to Pb (200 ppm, drinking water) from postnatal day 1–20, or to aging animals (18–20 months of age), and observed an increase in the APP mRNA expression as well as in the activity of the transcription factor Sp1 (one of the regulators of APP ) in the cortex of neonates, and after 20 months of Pb exposure had ceased. They observed a concomitant increase in Aβ levels in old animals exposed to Pb at birth. Interestingly, APP and Aβ protein levels did not respond to Pb exposure at old age (Basha et al., ). Similarly, a study conducted in cynomolgus monkeys exposed to Pb (1.5 mg/Kg/day, via infant formula) from birth to 400 days of age, and terminated 23 years later showed increased mRNA levels of APP and Sp1 in the frontal cortex compared with control animals, and high levels of the biomarker of oxidative DNA damage, 8-oxo-dG, suggesting an oxidative mechanism (Wu et al., ). Aβ1–42 and Aβ-1–40 levels in aged monkeys were also increased, as well as the intracellular Aβ staining and dense-plaques compared with age-matched controls (Wu et al., ). In addition, a study reported that gestational exposure to Pb (0.1, 0.5 and 1%, drinking water) in mice led to increased brain levels of Aβ and worst spatial memory performance, as well as increased levels of pro-inflammatory agents such as interleukin-1 (IL-1) and tumor necrosis factor alpha (TNF-α; Li et al., ).
Mercury (Hg) is a heavy metal with a high potential to cause neurotoxicity. Early studies about Minamata and Iraq disasters led us to understand the neurotoxicity of this metal. It is widely accepted that Hg disrupts the brain development and produces cognitive and motor disabilities (Johansson et al., ), and in adults, Hg exposure produces memory loss and cognitive alterations (Wojcik et al., ; Chang et al., ). An early study suggested a link between Hg exposure and AD. Authors reported increased levels of Hg (in microsomes) and bromide (in the whole brain), and reduced levels of rubidium (in the whole brain, microsomes, and nuclei), selenium (Se; in microsomes) and zinc (Zn; in nuclei) in AD brains compared with controls (Wenstrup et al., ). A subsequent work reported more than a 2-fold increase in blood Hg levels in AD patients ( n = 33) compared with a control group ( n = 65), as well as a positive correlation between blood Hg concentration and CSF levels of Aβ ( n = 15, Pearson’s correlation r = 0.7440, p = 0.0015) (Hock et al., ). More recently, several metals, including manganese (Mn), nickel, cadmium (Cd), Pb and Hg were determined in plasma and CSF of AD patients ( n = 173) and healthy controls ( n = 54), however only plasma Mn and Hg concentrations were significantly higher in AD patients (Gerhardsson et al., ). On the other hand, ApoE4 is a risk factor for AD, probably because this protein does not have sulfhydryl (SH) groups to scavenge heavy metals like Hg, whereas ApoE2 has four SH groups and then the ability to reduce the metal toxicity in the brain (Mutter et al., ). Therefore, ApoEε2 is considered a protective genotype for AD development (Suri et al., ). Similarly, Godfrey et al. ( ) observed a shift toward the risky ApoEε4 allele in patients with presumptive Hg-related neuropsychiatric symptoms with an elevated Hg body burden ( n = 400; Godfrey et al., ).
The ability of Hg to increase Aβ levels has been studied in vitro and in vivo , and the suggested mechanisms include an increased production, a reduced degradation and/or a diminished brain clearance of the peptide. Olivieri et al. ( ) reported an increased secretion of both Aβ-40 and Aβ-42 when neuroblastoma cells were exposed to 50 µg/dL of inorganic Hg concomitant with ROS overproduction (Olivieri et al., ), and a study conducted in aggregating brain-cell cultures of fetal rat telencephalon showed that MeHg (non-cytotoxic concentrations/10–50 days) produced increased APP levels accompanied by ROS production and glia activation (Monnet-Tschudi et al., ). Rat pheochromocytoma (PC12) cells exposed to both inorganic and organic (MeHg) Hg (10–1000 nM) also showed a dose-dependent overproduction of Aβ-40 probably by an increase in APP levels as well as to a reduction in Aβ degradation by NEP (Song and Choi, ). However, an increase in Aβ-42 levels was observed in differentiated SH-SY5Y cells exposed to Hg (10 and 20 µM) but not in the APP expression, and rather a reduced activity of the Aβ-degrading enzyme, NEP was observed (Chin-Chan et al., ). Negative effects of Hg on Aβ aggregation have also been published, for example, Atwood et al. ( ) studied the role of the pH and the ability of various metals to aggregate Aβ; Zn, Cu and Fe showed the highest potential on Aβ aggregation, and Hg did not show an important effect (Atwood et al., ). Regarding in vivo studies, oral administration of 20–2000 µg/Kg/day/4 weeks of MeHg produced a dose-dependent increase in Aβ-42 in the hippocampus of male rats, but not significant changes in APP or NEP protein levels (Kim et al., ). Interestingly, authors observed a reduced brain expression of the LRP1 receptor, which was positively correlated with increased Aβ levels in the hippocampus and reduced levels in the CSF, suggesting a reduced clearance of the pathogenic peptide from the brain (Kim et al., ).
Inorganic arsenic (As) is a known neurotoxic metalloid with adverse effects in both the neurodevelopment and cognitive function (Tyler and Allan, ), although its effects in elderly have been less studied. There are few papers evaluating the role of As exposure as a risk factor for AD development. A study conducted in rural-dwelling adults and elders in Texas, US (FRONTIER project) reported that long-term exposure to low As levels (3–15 µg/L As in water) correlated (after adjustment by confounders, including the ApoEε4 presence) with a poor score of cognitive abilities and memory, which reflects the earliest manifestations of AD (O’Bryant et al., ). Similarly, a positive correlation was observed between serum As levels and the cognitive ability in AD patients from Hong Kong ( n = 44, Pearson’s correlation coefficient r = 0.55, p < 0.0001) compared with matched controls ( n = 41), lower serum Zn concentrations were observed in AD patients as well (Baum et al., ). On the other hand, the study conducted in several European countries showed a higher prevalence of AD and other dementias in those countries with As levels in topsoils about 18 ppm such as Italy, Switzerland, Spain, France, Belgium and Norway, compared with countries with lower As levels (in the range of 9 ppm), including Luxemburg, Denmark, Finland, UK and Nertherlands (Dani, ). In experimental studies, the administration of inorganic As (20 mg/L, drinking water during the gestation and early postnatal life) to mice produced a significant loss of spatial memory (Ramos-Chávez et al., ).
A plausible mechanism for the cognitive and memory alterations induced by As exposure is by alterating the amyloid pathway. Zarazúa et al. ( ) reported that cholinergic SN56.B5.G4 cells incubated with sodium arsenite or the organic form dimethylarsinic acid (DMA) (5–10 µM/12–24 h) showed an increase in APP and sAPPβ levels, and consequently an increase in Aβ only with DMA. Similar effects were observed in neurons from Tg2576 mice (a murine model that overexpresses a mutant form of APP most used in AD). They suggest that DMA-induced effects may be due by an increased Aβ anabolism (enhanced APP expression), although authors did not discard an alteration in the Aβ degradation pathway (Zarazúa et al., ). The mechanism by which As causes Aβ overproduction has not been determined, but As exposure has been associated with brain inflammatory responses and oxidative stress, which is in agreement with the inflammatory and oxidative hypotheses of AD (Gong and O’Bryant, ).
Cadmium is another toxic heavy metal associated with neurological alterations including memory loss and mental retardation (Wang and Du, ). An early study observed higher plasma levels of various metals including Cd, aluminum (Al), As, and Se in AD patients ( n = 24) compared with healthy volunteers ( n = 28) (Basun et al., ). Also, the liver from autopsied AD patients ( n = 17) had significant higher Cd levels compared with age- and sex-matched control subjects ( n = 17) (Lui et al., ). However, Gerhardsson et al. ( ) did not observe significant differences in Cd concentrations in plasma or CSF in patients with AD ( n = 173) compared with healthy control subjects ( n = 54) (Gerhardsson et al., ). There is evidence linking Cd exposure with Aβ overproduction. Li et al. ( ) observed cognitive alterations accompanied by an increased production of Aβ-42 and increased size and number of senile plaques in the cerebral cortex and hippocampus from APP/PSEN1 transgenic mice treated with Cd (2.5 mg/Kg/4 days, drinking water). These effects were attributed to a reduced expression of ADAM10, sAPPα and NEP proteins, suggesting that the non-amyloidogenic pathway as well as Aβ degradation are target of Cd exposure (Li et al., ). Additionally, authors reported an increase in free-Zn levels, suggesting that Cd displaces Zn from its native enzymes, including NEP.
Recently, Ashok et al. ( ) investigated the role of the exposure to individual metals (As, Pb and Cd) and their combination in the AD-amyloid pathway in male rats exposed from gestational day 5 to postnatal day 80 through drinking water. They reported that metals activated the synthesis of Aβ in the frontal cortex and/or hippocampus, mediated by an increase in APP, and in APP-processing enzymes such as beta secretases BACE1 and PSEN at postnatal days 24 (post-weaning) and 90 (adulthood). Pb was the most potent metal to induce Aβ, followed by Cd, and As had the smallest effect, however all did increase the APP production. Interestingly, they demonstrated a synergic effect of metals mainly due to As, the exposure to these three cations produced a dramatic increase in Aβ, PSEN1, BACE1 and APP, suggesting an enhanced amyloidogenic processing (Ashok et al., ). Authors also observed (Ashok et al., ) increased levels of malondialdehyde (MDA), reduced activity of antioxidant enzymes, and the induction of 1L-1α and IL-1β in the frontal cortex and hippocampus of rats exposed to As + Pb + Cd mixture. Authors suggest that ROS-induced IL-1 overproduction was responsible for the APP expression. This is supported by the fact that the 5’ÚTR region of the mRNA of APP has a responsive element for IL-1 (Rogers et al., ; Ashok et al., ).
Aluminum is a neurotoxic element involved in the etiology of neurodegenerative disorders such as AD; however, there is no consistent evidence. The incident of Al pollution in Cornwall, UK (1998) gave evidence of Al potential neurotoxicity. Similar brain pathological characteristic found in AD patients were observed in subjects exposed to Al in this region (Exley and Esiri, ), as well as alterations in brain functions (Altmann et al., ). A recent study conducted in China reported a marginal positive association between Al levels in soil and the mortality caused by AD (Shen et al., ); while other studies reported no association. Experimental evidence appears to be more consistent. Chronic oral Al administration in rats (20 g/day in the food/twice weekly from 6 months of age to the end of their lives) increases the Aβ production by raising the levels of APP in hippocampal and cortical tissues (brain regions important for the memory process) (Walton and Wang, ). Cultured rat cortical neurons exposed to Al (50 µM/48 days) showed an accumulation of Aβ; furthermore, Al induced conformational changes in Aβ and enhanced its aggregation forming fibrillar deposits on the surface of cultured neurons. The aggregated Aβ was dissolved by the addition of desferroxamine, a chelator of Al (Exley et al., ; Kawahara et al., ). The administration of Al plus D-galactose (an animal model for AD) produced the impairment of memory and increased the production of Aβ in the cortex and hippocampus. Additionally, an augmented expression of BACE1 and a reduction of NEP were observed in this co-treatment (Luo et al., ). Another study showed that Al reduced the Aβ degradation by decreasing the activity of cathepsin B, suggesting the activation of the amyloidogenic pathway and a reduction of the catabolism of Aβ (Sakamoto et al., ). In addition, a reduced expression of LRP1 was also observed in these mice administered with Al plus D-galactose, indicating a possible reduction of the clearance of Aβ as well (Luo et al., ). Transgenic mice (Tg2576) fed with Al (2 mg/Kg, in the diet/9 months) showed an increased production of Aβ and proteins involved in its anabolism; the accumulation of amyloid plaques were reversed by the treatment with vitamin E, suggesting the contribution of Al-induced oxidative mechanism (Praticò et al., ).
As mentioned earlier, miRNA can be biomarkers of early diagnosis of AD, however few studies have reported the involvement of pollutants in the miRNAs homeostasis (Ray et al., ). Interestingly, Tg2576 mice exposed to Al (2 mg/Kg/9 months, through the diet) showed an increased expression of miRNAs (e.g., miR146a and miR125b) involved in a pro-inflammatory response similar to that observed in AD brains (Bhattacharjee et al., ; Zhao et al., ), and treatment of primary human astroglial (HAG) cells with 100 nM of Al + Fe increased the expression of NFκB-induced miR-125b and miR-146a (Pogue et al., ); these miRNAs are reported in AD pathology (Lukiw, ). Further studies are necessary to look for a possible relation between xenobiotic exposures and the deregulation of miRNA expression involved in neurodegeneration.
##### Pesticides
The association between chronic pesticide exposure and the prevalence of dementias, including AD has not been as well studied as with other environmental risk factors, and results are often inconsistent. This is mainly because the difficulty on getting adequate data on the levels of exposure of individual pesticides, which is often indirectly evaluated by structure questionnaires (exposure index). Some of the studies with positive associations include the one performed in the agricultural Cache County, Utah, US in about 3000 occupationally pesticides-exposed participants who were followed-up for 3, 7 and 10 years. The hazard ratio for developing AD was slightly higher for organophosphate (OP) pesticides exposure (HR = 1.53, 95% CI, 1.05–2.23) than to organochlorines (OCl) (HR = 1.49, 95% CI, 0.99–2.24), after adjusting for some variables, including ApoE genotype (Hayden et al., ). Similarly, a recent case-control study observed a 3.8-fold increase in the OCl metabolite DDE in serum from AD patients ( n = 79) compared with control participants ( n = 86); in addition, authors reported that the highest tertile of DDE levels was associated with an increased risk for AD development (odds ratio-OR = 4.18, 95% CI, 2.54–5.82), and carriers of an ApoEε4 allele may be more susceptible (Richardson et al., ). Baldi et al. ( ) evaluated the association between lifelong cumulative exposure to pesticides and neurodegenerative diseases in a subsample from a cohort of elderly people (aged 65 years or older) (PAQUID study) in southwestern France. Authors analyzed 96 incident cases of AD (71 women and 25 men) in a 5-year follow-up approach, and observed a significant association between AD and occupational exposure to pesticides in men with a relative risk of 2.4 (95% CI, 1.0–5.6) after adjusting by education and smoking. Results were not significant in women (Baldi et al., ).
The role of pesticides in alterations observed in cognitive functions and AD has been suggested based on epidemiological studies, but the mechanisms have been poorly explored. In vitro studies performed in differentiated SH-SY5Y cells incubated with OCl pesticides such as DDE and its parent compound DDT (1 µM/48 h) showed increased APP protein levels, although authors did not evaluate Aβ levels (Richardson et al., ). A recent study reported that DDT augmented Aβ levels by increasing APP and BACE1 levels in human neuroglioma H4-AβPPswe cells, as well as by reducing the clearance and degradation of the peptide by targeting the Aβ-degrading enzyme, IDE and the ATP-binding cassette transporter A1 (ABC1; Li et al., ). Regarding in vivo data, chlorpyrifos (CPF), an OP insecticide associated with cognitive impairment, oxidative stress and neuronal damage caused a significant increase in Aβ levels in the cortex and hippocampus, as well as increased memory loss and reduced motor activity in Tg2576 mice 6 months after an acute subcutaneous administration of 50 mg/Kg of CPF (Salazar et al., ). However, Peris-Sampedro et al. ( ) did not find increased Aβ levels neither significant changes in memory acquisition in Tg2576 mice treated with CPF (25 mg/Kg/twice weekly/4 weeks, intragastric) and analyzed 6 months later (Peris-Sampedro et al., ). More studies are needed to better understand the mechanisms by which OCl, OP and other insecticides are linked to AD.
Paraquat (PQ), is a common used herbicide that has been suggested to be involved in AD development. A recent study showed that treatment of wild type and APP transgenic (Tg2576) mice with PQ (10 mg/Kg/twice a week/3 weeks) produced a significant increase in Aβ levels in transgenic mice that was associated with mitochondrial oxidative damage in cerebral cortex leading to the impairment of learning and memory. Interestingly, the overexpression of peroxiredoxin 3, a mitochondrial antioxidant defense enzyme produced an improvement in cognitive functions and a significant reduction in Aβ levels in APP transgenic mice exposed to PQ (Chen et al., ), suggesting that pro-oxidant xenobiotics like PQ can contribute to AD.
##### Nanoparticles
As the synthesis of NPs for different applications, including drug delivery strategies in the treatment of AD is growing, it is necessary to study the potential toxic effects on proteins related to AD development.
There are not epidemiological studies associating the exposure to NPs with AD. However, there is increasing experimental evidence suggesting the potential role of NPs in brain damage. A recent study reported that nasal administration of TiO -NPs (2.5–10 mg/Kg/90 days) to mice caused neuronal death in the hippocampus, oxidative stress and gliosis, and microarray analysis revealed a decline of genes associated with memory and cognition (Ze et al., ). Similarly, rats exposed to CuO-NPs (0.5 mg/Kg/day/14 days, i.p.) showed worst spatial cognition and a reduction in electrophysiological endpoints such as long-term potentiation, which matched with augmented levels of ROS and lipid peroxidation products (MDA and 4-hydroxinonenal-HNE), and reduced levels of antioxidants enzymes (An et al., ). Studies of NPs of Al, Cu and Ag administered at different doses and routes in rats and mice showed that they produce brain alterations such as motor, sensory and cognitive deteriorations (Sharma et al., ; Sharma and Sharma, ). However a recent study did not observe memory loss in adult mice administered with Ag-NPs (10, 25, and 50 mg/Kg/7 days) (Liu et al., ). Regarding in vitro studies, the exposure of human (SK-N-SH) and mouse (Neuro-2a) neuroblastoma cells to silica NPs (SiNPs) (10 µg/mL/24 h) raised the intracellular content of Aβ in both cell lines, which was associated with increased APP and reduced NEP protein levels. These effects may be mediated by ROS production, since SiNPs increased the production of intracellular ROS (Yang et al., ). Likewise, treatment of Neuro-2a cells to silver NPs (AgNPs, 12.5 µg/mL/24 h) showed the deposition of Aβ plaques and an increased expression of APP , while NEP and LPR1 (or LDLR) expression and protein levels were reduced, suggesting that AgNPs can induce AD by altering the amyloidogenic pathway: Aβ synthesis, degradation or clearance (Huang et al., ). Interestingly, authors also reported an increased expression of genes involved in the inflammatory response such as IL-1 , C-X-C motif chemokine 13 ( CXCL13 ), macrophage receptor with collagenous structure ( MARCO ), and glutathione synthetase ( GSS ) (Huang et al., ).
#### Tau Hyperphosphorylation by Environmental Factors
Several environmental factors have shown to mediate AD development through alterations on Tau phosphorylation and/or aggregation (Figure ).
##### Metals
In vivo and in vitro studies have suggested the potential of Hg to induce P-Tau. Fujimura et al. ( ) reported an increased neuronal death and more migrating astrocytes in cerebral cortex of male mice exposed to MeHg (30 ppm, drinking water), as well as increased levels of P-Tau mediated by c-jun N-terminal kinase (JNK; Fujimura et al., ). An in vitro study showed that inorganic Hg (50 µg/dL/30 min) increased P-Tau in SH-SY5Y cells by a ROS-dependent mechanism, which was reverted by the co-treatment with the antioxidant melatonin (Olivieri et al., ). Another study demonstrated that Hg ions coordinate with Cys291 of the second repeated (R2) of the MT-binding domain of Tau increasing the heparin-induced aggregation, and a conformational change in Tau demonstrated by circular dicroism (CD) analysis (Yang et al., ). On the other hand, Cd appears to play a role in Tau hypothesis since it promotes the aggregation of this protein. It was shown that Cd (II) accelerates heparin-induced aggregation of the third repeated (R3) of Tau. The binding of Cd (II) to the dimeric R3 produces changes on its conformation demonstrated by CD (Jiang et al., ). Subchronic As administration to rats (NaAsO at 3 and 10 mg/Kg/day/4–12 weeks, intragastric) induced P-Tau, suggesting that As-destabilization and disruption of the cytoskeletal framework may lead to axonal degeneration (Vahidnia et al., ). Regarding Pb, it was reported that infantile Pb exposure in cynomolgus monkeys elevated mRNA and protein levels of Tau as well as its transcriptional regulators (Sp1 and Sp3) in aged primates (23 years old). An increase in P-Tau phosphorylation and mRNA and protein levels of cyclin dependent kinase 5 (cdk5, a kinase that phosphorylates Tau) were also observed (Bihaqi and Zawia, ). Other studies also reported that maternal (Li et al., ) and early postnatal exposures (Liu et al., ) to Pb produced significant increased P-Tau levels and cognitive impairment in mice. Finally, chronic Al exposure caused Tau aggregation, and it was suggested that Al is bound to P-Tau in the Al-NFTs lesions (Singer et al., ; Shin et al., ). Also, a study showed that Al is able to confer resistance to the degradation of PHFs both in vivo and in vitro (Shin et al., ), and it can inhibit the activity of the protein phosphatase 2 (PP2), which is involved in P-Tau de-phosphorylation (Yamamoto et al., ).
##### Pesticides
There is some evidence suggesting that pesticide exposure can disrupt Tau function. A recent study showed that the administration of the insecticides deltamethrin (pyethroid) and carbofuran (carbamate) to rats (daily administration by gavage/28 days) produced neuronal death in the cortex and hippocampus and a dysfunction in the spatial memory and learning. These alterations were attributed to a reduced expression of synaptic proteins involved in the memory consolidation. Additionally, P-Tau and activation of p-GSK3β (a major kinase that phosphorylates Tau) were observed (Chen et al., ). Similarly, Wills et al. ( ) showed P-Tau in the striatum, through the activation of p-GSK3β, as well as hyperacetylation of α-tubulin in mice treated with PQ (10 mg/Kg, i.p., twice weekly/6 weeks), suggesting a cytoskeleton remodeling (Wills et al., ).
##### Nanoparticles
The effect of NPs on Tau phosphorylation has not been extensively studied. Silica NPs (siNPs) used in medicine are also able to increase P-Tau at Ser262 and Ser396, two phosphorylation sites characteristic of AD. It was demonstrated that this effect was dependent on the activation of the kinase GSK3β in human SK-N-SH and mouse Neuro-2a cells by a mechanism probably mediated by oxidative stress, since ROS were increased in cells exposed to these NPs (Yang et al., ).
### Parkinson’s Disease
Parkinson Disease is a chronic and progressive neurological disorder characterized by the selective loss of dopaminergic neurons of the substantia nigra pars compacta (SNpc). The cardinal features of the syndrome are related to motor dysfunction including tremor at rest, rigidity, akinesia (or bradykinesia), and postural instability. The motor symptoms appear when at least 60% of dopaminergic neurons are lost and 80–85% of dopamine content in the striatum is depleted (Jankovic, ; Wirdefeldt et al., ). Additional to the neuronal loss, the main neuropathological hallmark of PD is the presence of Lewy bodies (LB) in the surviving neurons, which are eosinophilic cytoplasmic inclusions containing aggregates of protein such as α-synuclein (α-syn) (Gibb and Lees, ; Spillantini et al., ). PD is the second most common neurodegenerative disorder after AD. Due to the lack of specific/differential diagnostic biomarkers, the diagnosis of PD is based on clinical criteria of specific cardinal motor signs of the disease and on the response to levodopa. PD diagnosis is confirmed by the depletion of brain stem pigmented neurons and the presence of LB at necropsy, this is the reason of the misclassification of PD cases (about 10–15%) (Schrag et al., ; Jankovic, ). There is no cure for PD, and the existing therapies only provide brief relief of motor symptoms through improving the dopamine deficit or by surgical methods. This highlights the need of research on early specific/differential biomarkers to have more accurate diagnosis of neurodegenerative disorders, as well as biomarkers for the identification of populations at risk to implement neuroprotective therapies (Jankovic, ).
As in the case of AD, circulating miRNAs are being studied as differential biomarkers for PD. Some reviews have recently addressed this topic, showing the association of specific miRNAs for some genes involved in PD, such as SNCA and LRRK2 (encoding for leucine–rich repeat kinase 2) with PD development (Goodall et al., ; Maciotta et al., ). Some studies have reported differentially expressed miRNAs in serum of PD patients not observed in control subjects or in other diseases. For example, Vallelunga et al. ( ) reported two differentially expressed miRNAs (miR-30c and miR-148b) in Italian PD patients ( n = 25 vs. 25 healthy controls) (Vallelunga et al., ), and another study found that serum levels of miR-29c, miR-29a, and miR-19b were down-regulated in PD patients ( n = 65 vs. 65 healthy controls) from Barcelona, Spain (Botta-Orfila et al., ). Also, a reduced expression of miR-34b and miR-34c in several brain areas including the substantia nigra of PD patients ( n = 11 vs. 6 healthy controls) was detected; interestingly the misregulation of miR-34b/c was observed in patients in pre-motor stages of the disease. Additionally, these miRNAs were deregulated in differentiated SH-SY5Y dopaminergic neuronal cells, which was associated with altered mitochondrial function, oxidative stress and ATP depletion, as well as decreased protein levels of DJ1 (a mitochondrial peroxidase) and Parkin (an E3 ubiquitin ligase) that are associated with the familial form of PD (Miñones-Moyano et al., ).
Although the research on PD has rapidly advanced, the molecular mechanisms involved are still unclear and its etiology is complex. Several molecular mechanisms of neuronal death in PD pathogenesis have been described including mitochondrial dysfunction, impairment of protein quality pathways, oxidative/nitrative stress, microglia activation and inflammation. These mechanisms converge and are consistent with a major role of oxidative stress in PD, which damage organelles and proteins leading to increased protein aggregates (e.g., α-syn), that in turn overwhelms the degradation systems leading to a self-perpetuating cycle and further oxidative stress (Wirdefeldt et al., ; Goldman, ). The evidence in postmortem PD brains supports these mechanisms, as well as a decreased in reduced GSH levels, α-syn aggregation, proteasome impairment and autophagy dysfunction (review in Navarro-Yepes et al., ).
A fraction of PD occurrence has a clear familial inheritance and it is related to mutations in at least 6 genes that have been associated with PD onset. The identification of genes such as SNCA or PARK1 encoding for α-syn (maybe involved in the regulation of dopamine release and transport), LRRK2 or PARK8 encoding for LRRK2 (or Dardarin), PARK7 encoding for DJ1, PARK6 or PINK1 encoding for PTEN-induced putative kinase 1 (PINK1, a mitochondrial kinase), and PARK2 encoding for Parkin have provided clues about the molecular mechanisms involved in its pathogenesis (Corti et al., ; Cookson, ). However, 90% of PD cases are sporadic and cannot be attributed only to genetic factors, which suggests that PD have a multifactorial etiology (Goldman, ). In addition to the aging, which is the main risk factor for PD (Tanner and Goldman, ), epidemiological evidence suggests that the exposure to environmental toxicants, mainly pesticides, metals and solvents could increase the risk of developing PD, and factors such as tobacco consumption can protect against PD development (Figure ; Hatcher et al., ; Gao and Hong, ).
Molecular mechanisms altered by environmental factors related to increased Parkinson’s disease risk . Exposure to environmental toxicants mainly pesticides, metals and solvents may lead to the selective loss of dopaminergic neurons on the substantia nigra pars compacta (SNpc) through the dysregulation/alteration of the molecular mechanisms implicated on PD development such as mitochondrial dysfunction, impairment of protein quality pathways, microglia activation and inflammation, which converge in the production of oxidative stress as the main factor in PD. For more detail see the text.
#### Metals
It has been proposed that chronic exposure to heavy metals such as iron, Pb and Mn and their combinations can be associated with an increased risk of developing PD, since they accumulate in the substantia nigra and generate oxidative stress. However, epidemiological evidence is controversial (Lai et al., ). The epidemiological evidence of Pb association with PD is more consistent because the accumulative lifetime exposure can be estimated through Pb concentration in bone that has a half–life of years to decades. Initially, Kuhn et al. ( ) reported that 7 out of 9 postal workers exposed to lead-sulfate batteries for up to 30 years developed parkinsonian symptoms, suggesting that Pb intoxication may play a role in the occurrence of these symptoms (Kuhn et al., ). Coon et al. ( ) evaluated this association in 121 PD patients vs. 414 controls and found that chronic Pb exposure (evaluated by Pb concentrations in tibial and calcaneal bones) increased 2–fold the risk of PD (OR = 2.27, 95% CI, 1.13–4.55) for individuals in the highest quartile of lifetime Pb exposure relative to those in the lowest quartile (Coon et al., ). In the same way, it was reported that the cumulative exposure to Pb increases the risk of PD (OR = 3.21, 95% CI, 1.17–8.83) in 330 PD patients (vs. 308 controls) recruited from 4 clinics for movement disorders in Boston, MA area (Weisskopf et al., ), and the exposure to Pb for more than 20 years showed a stronger association with PD risk in a health system population-based case-control study (144 cases vs. 464 controls) from the metropolitan Detroit area (Gorell et al., ). At the molecular level, Pb exposure significantly decreases the dopamine release and the dopamine D1 receptor sensitivity post-synaptically in microdialysate samples from rats subchronically exposed to Pb (50 ppm/90 days) (Kala and Jadhav, ), and in rats treated with 250 ppm of Pb for 3–6 weeks through drinking water (Tavakoli-Nezhad and Pitts, ). Furthermore, it increases the lipid peroxidation and reduces the antioxidant cell capacity (Sandhir et al., ), and causes fibrillation and aggregation of α-syn (Yamin et al., ), which induces hippocampal injury and decreases the ability of learning and memory in rats exposed to 0–300 ppm of Pb (Yamin et al., ; Zhang et al., ).
Manganese is an essential element with important physiological functions for cellular homeostasis. The epidemiologic evidence does not provide sufficient support for an association between Mn exposure and PD risk (Wirdefeldt et al., ; Mortimer et al., ). Only one case–control study (144 cases vs. 464 controls) in a population from the metropolitan Detroit area reported an increase of PD risk when the exposure to Mn was over 20 years (OR = 10.63, 95% CI, 1.07–105.99) (Gorell et al., ). However, occupational or environmental exposures to Mn have been associated with a neurological syndrome that include cognitive deficits, neuropsychological abnormalities and Parkinsonism (Guilarte, ). Mn was related to PD since 1837, when it was noted that high Mn exposures caused a severe and debilitating disorder known as “manganism” or manganese–induced Parkinsonism, which consists on an extra pyramidal syndrome that resembles the dystonic movements associated with parkinsonian symptoms (Couper, ; Jankovic, ), but it is clinically distinct from PD since patients do not respond to dopamine replacement therapies (Cook et al., ; Huang et al., ; Lu et al., ). Several cases of Mn–induced Parkinsonism have been reported in individuals whose professions involve prolonged contact with high atmospheric levels of Mn such as welders, miners and smelters (Rodier, ; Wang et al., ; Lee, ). Several investigations have shown that sustained exposure to low-concentrations (below the current US standard of 5.0 mg/m ) is consistent with early manganism, suggesting that Mn is a neurotoxic chemical (Park, ). Patients with manganism and primates experimentally intoxicated with Mn consistently show damage to the globus pallidus, which is in contrast with PD where there is a preferential degeneration of dopamine neurons in the SNpc and preservation of the pallidum (Perl and Olanow, ). Likewise, it was observed microglia activation in the substantia nigra pars reticulate (SNpr) and SNpc in Cynomolgus macaques exposed to Mn (5–6.7 mg/Kg/week/10 months) (Verina et al., ). In vitro , it has been observed that Mn treatment (50–300 µM MnCl / 3–48 h) induces cytochrome C release and activation of caspases 9 and 3, as well as protein aggregation in N27 dopaminergic neuronal cells that stably express α-syn (Harischandra et al., ).
Iron is an essential element transported into the brain through the transferrin receptor and divalent metal transporter 1 (DMT1; Zheng and Monnot, ). It has been evaluated in relation to the risk of PD in few epidemiological studies without convincing evidence (Rybicki et al., ; Logroscino et al., ; Miyake et al., ; Abbott et al., ). However, iron and its deregulated homeostasis have been proposed to have a role in the pathogenesis of PD because of its pro-oxidants characteristics that may lead to ROS generation via Fenton and Haber–Weiss reactions (Stohs and Bagchi, ; Sian-Hulsmann et al., ). The substantia nigra has the highest levels of iron in the human brain, probably due to the presence of neuromelanin in pigmented SNpc dopaminergic neurons that have an impressive capacity of chelating metals, iron in particular; however, this may be a dual-edged sword that may increase their vulnerability since iron may react with ROS produced from dopamine metabolism and promote the further generation of highly toxic radicals (Zecca et al., , ). Alterations in iron distribution have been observed in the substantia nigra of PD postmortem brains (Dexter et al., , ; Hirsch et al., ). On the other hand, it was observed in postmortem samples that although the total iron concentration in the whole substantia nigra was not significantly different between parkinsonian and control samples, there was an increase in the free-iron concentration and a decrease in iron–binding ferritin levels, ferritin sequestrates the excess of iron under physiological conditions (Wypijewska et al., ). Likewise, it was reported that free-iron induces fibrillation and aggregation of α-syn in a dose- and time-dependent way in SK-N-SH cells incubated with ferric iron (1–10 mmol/L/24–48 h) (Li et al., ). Mice administered with iron (120 µg/g of carbonyl iron, oral gavage) at a dose equivalent to that found in iron-fortified human infant formula (12 mg/L of iron) from days 10 to 17 post-partum (an equivalent period to the first human year of life) showed a progressive midbrain neurodegeneration and enhanced vulnerability to toxic injury at 12 and 24 months of age (Kaur et al., ).
#### Pesticides
The hypothesis that pesticide exposures may be related to PD development was prompted by the discovery that intravenous injection of 1-methy l-4pheny l-1, 2, 3, 6-tetrahydropyridine (MPTP), a byproduct of the synthesis of heroin, developed a Parkinson syndrome clinically indistinguishable from PD (Langston et al., ); subsequent findings showed that MPTP selectively damaged dopaminergic neurons in the substantia nigra (Langston and Ballard, ; Langston et al., ). Since then, environmental factors with similar toxicological profiles have received attention as potential risk factors for PD.
A meta-analysis conducted in 2000 evaluated the association between pesticide exposures and PD in 19 case–control studies published between 1989 and 1999. Authors showed that most studies found an elevated risk of PD with the exposure to pesticides, the calculated combined OR was 1.94 (95% CI, 1.49–2.53); similar ORs were observed in studies conducted in United States, Asia, Europa and Canada. Additionally, it was observed that the risk of PD increased with longer exposure times, with an OR of 5.81 (95% CI, 1.99–16.97) for ≥10 years of exposure; however, specific types of pesticides were not identified (Priyadarshi et al., ). Subsequently, Brown et al. ( ) reviewed 31 case–control studies published until 2003, and found that about half of them reported significant associations between pesticide exposure and PD risk with ORs from 1.6–7. Interestingly, in most studies, authors observed a positive association between the exposure to herbicides and insecticides and PD risk, but not with the exposure to fungicides alone (Brown et al., ).
In line with this, a recent review by Freire and Koifman ( ) analyzed the epidemiological evidence published between 2000 and 2011, including ecological, cross–sectional, prospective and case–control studies. They found that 7 out of the 8 prospective (cohort) studies provided evidence of an association between pesticide exposure and PD, reporting risk estimates of 2-fold or higher. Among 23 case–control studies, 13 studies reported a significant increased risk of PD for the professional use of pesticides in comparison with unexposed controls, with ORs ranging from 1.1 to 1.4, which is in agreement with the review of Priyadarshi in the 1990’s (Freire and Koifman, ). Furthermore, van der Mark et al. ( ) performed a systematic review and calculated the summary risk ratio (sRR) from 39 case–control studies, 4 cohort studies and 3 cross–sectional studies. When a job–exposure matrix was constructed, a higher sRR (2.5, 95% CI, 1.5–4.1) was observed compared with self–reported exposure evaluation (1.5, 95% CI, 1.3–1.8). This meta–analysis found a positive association between PD and insecticides (sRR = 1.50, 95% CI, 1.07–2.11), and herbicides (sRR = 1.40, 95% C, 1.08–1.81), but not with fungicides (sRR = 0.99, 95% CI, 0.71–1.40) (van der Mark et al., ), in agreement with Brown et al. ( ) and Freire and Koifman ( ). Other factors related to pesticide exposure such as well–water consumption, farming, and rural living have been associated with an increased PD risk. The meta–analysis of Priyadarshi et al. ( ) found a combined OR of 1.56 (95% CI, 1.18–2.07) for rural living, 1.42 (95% CI, 1.05–1.91) for farming and 1.26 (95% CI, 0.97–1.64) for well–water consumption. However, whether of these factors are independent risk factors or correlated with pesticide exposure could not be determined (Priyadarshi et al., ).
In support to epidemiological evidence, increased levels of some pesticides have been quantified in postmortem brains from PD patients. High concentrations of some OCl pesticides have been observed in PD cases compared with controls, including dieldrin, lindane, and p-p -DDE (Fleming et al., ; Corrigan et al., , ). In the same way, 2 epidemiologic studies reported a significant association (OR ranging from 1.3 to 1.8) between dieldrin use and PD in farmers participants in the Agricultural Health Study (AHS; Kamel et al., ; Tanner et al., ). Another nested case–control study within the Finnish Mobile Clinic Health Examination Survey in Finland, with serum samples collected during 1968–1972, observed that increasing serum concentrations of dieldrin were associated with an increased PD risk (OR = 1.95, 95% CI, 1.26–3.02) in 68 cases vs. 183 controls restricted to never smokers, while no other OCl pesticide showed an association (Weisskopf et al., ).
The epidemiologic evidence that dieldrin exposure may be associated with PD is supported by toxicological data at molecular level. Dieldrin may cross the blood–brain barrier and remains in lipid-rich tissues such as the brain (Kanthasamy et al., ), and it has been shown that it is selectively toxic to dopaminergic neurons and could induce several of the pathologic mechanisms of PD including the depletion of brain dopamine levels, increased ROS in nigral dopaminergic neurons, inhibition of mitochondrial oxidative phosphorylation that lead to a reduction of cellular ATP production, alteration of the mitochondrial membrane potential and cytochrome C release in animal models such as rats and mice chronically exposed to dieldrin (0.3–3 mg/Kg/day in the diet) (Bergen, ; Wagner and Greene, ; Purkerson-Parker et al., ; Hatcher et al., ), and in primary mesencephalic cultures or dopaminergic cell lines (0.01–300 µM) (Sanchez-Ramos et al., ; Kitazawa et al., , ; Kanthasamy et al., ). Aggregation of α–syn, ubiquitin–proteasome impairment function (Uversky et al., ; Sun et al., ) and microglia activation (Mao and Liu, ) have also been observed.
Paraquat is a quaternary nitrogen herbicide used worldwide. Due to its structural similarity to MPP (the active metabolite of MPTP), it was thought to be toxic to dopaminergic neurons and thus might be related to PD. The possible association between PQ and PD received attention from the study of Liou et al. ( ) performed in PD patients (120 patients and 240 controls) in Taiwan, in which the pesticide use was associated with an increased risk of developing PD, being higher for those individuals who reported using PQ (Liou et al., ). Likewise, Tanner et al. ( ) reported a significant association between PD and the use of oxidative pesticides, including PQ (OR = 2.5, 95% CI, 1.4–4.7) in professional pesticide applicators (110 cases and 358 controls) (Tanner et al., ). Similarly, other epidemiologic studies have associated the exposure to PQ with PD (Hertzman et al., ; Ascherio et al., ; Kamel et al., ; Wang et al., ).
Paraquat is taken up into dopaminergic terminals by the dopamine transport and organic cation transporter 3 (Rappold et al., ), and causes cellular toxicity by oxidative stress through the cellular redox cycling generating superoxide radical by the oxidation of NADPH, which in turn impairs the restauration of GSH levels and thus the activity of several antioxidant systems (Berry et al., ; Franco et al., ). It has been observed that repeated administrations of PQ to adult mice and rats (5–10 mg/Kg/ week/at least 3 weeks, i.p.) increase ROS levels in the striatal homogenate, induce a dose-dependent decrease in dopaminergic neurons from the substantia nigra, a decline in striatal dopamine nerve terminal density, and a neurobehavioral syndrome characterized by reduced ambulatory activity (Brooks et al., ; McCormack et al., ; Kuter et al., ). PQ also reproduces other biochemical and neuropathological characteristics of human Parkinsonism such as microglia activation (Wu et al., ; Purisai et al., ), α-syn up-regulation and fibrillation (Uversky et al., ; Manning-Bog et al., ), increases lipid peroxidation (increase of 4-hydroxynonenals) (McCormack et al., ), alters parkin solubility promoting its intracellular aggregation (Wang et al., ), induces a proteasome dysfunction in SH-SY5Y cells (Ding and Keller, ; Yang and Tiffany-Castiglioni, ), as well as in homogenates from postmortem PD brains (McNaught and Jenner, ; McNaught et al., ), impairs mitochondrial function at the level of complex III to generate ROS (Castello et al., ; Drechsel and Patel, ), promotes cytochrome C release (González-Polo et al., ; Fei et al., ), induces GSH depletion (Schmuck et al., ; Kang et al., ), and causes cell injury leading to apoptotic cell death (Berry et al., ; Franco et al., ). PQ has been used as a toxicological model for PD that has permitted getting important information about the mechanisms involved in the neurodegeneration associated with PD (Gao and Hong, ).
Rotenone, an OP insecticide has also been associated with an increased risk of PD. Two epidemiological studies found an association between rotenone exposure and PD risk, reporting an increased risk of 10–fold (OR = 10.0, 95% CI, 2.9–34.3) in East Texas farmers (Dhillon et al., ), and 2.5-fold (OR = 2.5, 95% CI, 1.3–4.7) in PD cases ( n = 110) compared with controls ( n = 358) from professional pesticide applicators participants in the AHS (Tanner et al., ). Rotenone can freely cross the blood–brain barrier and is a well-established mitochondrial toxin that specifically inhibits the complex I (NADH–dehydrogenase) of the electron transport chain leading to ATP depletion, energy failure and mitochondrial ROS production, which in turn induces cytochrome C release and apoptotic cell death (Clayton et al., ; Radad et al., ; Sherer et al., ). It has been shown that, like MPTP, rotenone treatment in animal models (1.5–3 mg/Kg/day/up to 3 weeks) reproduces features of PD such as bradykinesia, postural instability and/or rigidity, reduces the tyrosine hydroxylase-positive neurons in the substance nigra, induces a loss of striatal dopamine, and the accumulation of α-syn and poly-ubiquitin positive aggregates in remaining dopaminergic neurons (Betarbet et al., ; Sherer et al., ; Cannon et al., ). Likewise, Betarbet et al. ( ) observed that chronic administration of 3.0 mg/Kg/day of rotenone for up to 5 weeks to male rats caused the oxidation of DJ-1, accumulation of α-syn, and proteasomal impairment (Betarbet et al., ). These effects were also observed in neuroblastoma SK-N-MC cells treated with rotenone (5 nM/4 weeks), as well as a loss of GSH, oxidative DNA and protein damage and caspase-dependent death (Sherer et al., ; Betarbet et al., ). Rotenone has also the capacity to activate microglia (Sherer et al., ); Gao et al. ( ) demonstrated that the addition of microglia to primary neuron-enriched cultures (neuron/glia cultures) markedly increased the dopaminergic neurodegeneration induced by rotenone (1 nM/8 days), and this neurotoxicity was attenuated by the inhibition of NADPH oxidase or scavenging the superoxide radical that is liberated from the microglia (Gao et al., ). Since rotenone recapitulates several mechanisms of PD pathogenesis, this pesticide is currently used as a toxicological model to study the underlying mechanisms on the PD development.
Despite the widespread use of OP insecticides such as malathion, methyl parathion, chlorpyriphos and diazinon, not many studies have evaluated the association between specific OP and PD risk. Dhillon et al. ( ) found a 2–fold increase (OR = 2.0, 95% CI, 1.02–3.8) in the risk of PD in Texan agricultural workers exposed to chlorpyriphos (cases = 100, controls = 84) (Dhillon et al., ). An increased risk of PD was also observed in rural residents from California possibly exposed to high levels of chlorpyriphos (OR = 1.87, 95% CI, 1.05–3.31) and diazinon (OR = 1.75, 95% CI, 1.12–2.76) through the consumption of contaminated well–water (Gatto et al., ). One study conducted in a population from the Group Health Cooperative (GHC) in Western Washington State occupationally exposed to methyl parathion found a high risk of PD (OR = 8.08, 95% CI, 0.92–70.85), although the association was not statistically significant (Firestone et al., ). This is particularly relevant, because parkinsonian effects have been reported in cases of patients intoxicated with OP (Bhatt et al., ).
#### Solvents
Solvents are widespread used due to their commercial applications, including metal degreasing, dry cleaning, and as ingredients of paint thinners and detergents. Some solvents are lipophilic and thus easily absorbed by the central and peripheral nervous system tissues (Lock et al., ). There are isolated cases of acute Parkinsonism associated with large solvent exposures such as in workers exposed to n -hexane (Pezzoli et al., ), and toluene (Papageorgiou et al., ), among others. There is no consistent evidence of the association of solvent exposure and PD (Wirdefeldt et al., ). One case-control study based on a questionnaire reported an increased risk of PD by the exposure to organic solvents (OR = 2.78, 95% CI, 1.23–6.26) in 86 PD patients and 86 controls from the Emilia-Romagna region in Italy (Smargiassi et al., ). Another case–control study reported an increased risk of PD when the exposure to solvents was above 20 years (OR = 3.59, 95% CI, 1.26–19.26) in 182 cases (vs. 422 controls) identified through death certificates of the Rolls-Royce PLC national pension fund archive from employees of five manufacturing locations in United Kingdom who had any mention of PD (McDonnell et al., ).
Trichloroethylene (TCE) is one of the specific solvents that has been investigated in detail (Goldman, ). Some clinical case reports have reported the onset of PD in workers exposed to TCE through chronic inhalation and dermal exposure by handling TCE, suggesting a potential link between the exposure to TCE and PD (Kochen et al., ; Gash et al., ). More recently, an epidemiologic study in 99 twin pairs discordant for PD showed that the exposure to TCE was associated with a 6–fold increased risk of PD (OR = 6.1, 95% CI, 1.2–33) (Goldman, ). In animal models, TCE may recapitulate several key pathological features of PD. The systemic exposure of adult rats to TCE (1000 mg/Kg/day/5 days a week/2 and 6 weeks, oral gavage) inhibits mitochondrial complex I enzyme activity, increases oxidative stress markers, activates the microglia, induces nigral α-syn accumulation and a significant loss of dopaminergic neurons on the SNpc in a dose-dependent manner, as well as defects in the rotarod behavior test (Liu et al., ). In a similar way, the administration of n- hexane and its metabolite 2, 5-hexanedione (400 mg/Kg/day/5 days a week/6 weeks, i.p.) to mice caused that both chemicals reduced the striatal dopamine concentration by 38 and 33%, respectively, but neuronal cell loss was not confirmed (Pezzoli et al., ). On the other hand, there is no evidence that acute or subchronic exposure to toluene promotes the degeneration of the nigrostriatal dopamine system (Lock et al., ).
#### Nanoparticles and PD
Nanoparticles are an important alternative in the development of treatment strategies for neurodegenerative diseases due to their small particle size, large surface and high drug loading efficiency, which allow them to cross the blood-brain barrier and efficiently release specific drugs (Li et al., ; Leyva-Gómez et al., ). However, their small size allows them to penetrate the cell and organelles, disrupting their normal function (Buzea et al., ).
Although some NPs are being used in therapies for PD, no epidemiological studies are available associating them with PD risk. However, there is evidence suggesting that they could contribute to alter the molecular mechanisms involved in the pathogenesis of PD. Thus, it was reported that intranasal instillation of SiO -NPs (20 µg/day/1–7 days) to rats resulted in their presence in the striatum, the induction of oxidative damage, an inflammatory response, and depleted dopamine concentration and tyrosine hydroxylase levels, suggesting that these NPs have a negative impact on striatal dopaminergic neurons (Wu et al., ). Another report in adult zebrafish exposed to SiO -NPs (300 and 1000 µg/mL; 15 and 50-nm of size) showed alterations in neurobehavioral parameters (general, cognitive behavior and locomotive activity), with the most significant effects observed with the smallest NPs, similar to those observed in neurodegenerative diseases (Li et al., ). In vitro studies also support the potential contribution of NPs in PD development. The exposure of dopaminergic neurons (PC12 cells) to SiO -NPs (25–200 µg/mL/24 h) triggered an oxidative stress, disturbed the cell cycle, induced apoptosis, and activated the p53-mediated signaling pathway (Wu et al., ); while the exposure of these cells to TiO -NPs (50, 100 and 200 mg/mL/24 h) induced a dose–dependent increase in the expression and aggregation of α-syn, as well as a reduction of the expressions of Parkin (E3 ligase), and the ubiquitin C-terminal hydrolase ( UCH-L1 ), these events were associated with increased oxidative stress (Wu and Xie, ). Also, the exposure PC12 cells to iron oxide (Fe O -NPs; 0.15–15 mM) decreased the neurite growth in response to the nerve growth factor (NGF) (Pisanic et al., ). Likewise, citrate-capped gold nanoparticles (Au-NPs; 0.3–32 nM, 10–22 nm) produced a dose-dependent aggregation of purified α-syn, being strongest for the smallest NPs (Alvarez et al., ). In contrast, the administration of Neurotensin (NTS)-polyplex NPs (8.5 nmol/Kg, i.v), a nanocarrier gene with a potential for nanomedicine-based applications for PD treatment, to BALB/c mice does not produce systemic inflammatory (up to 24 h after treatment) nor hepatic cytotoxicity (at 24 and 96 h after treatment), supporting the safety of these NTS-polyplex NPs as a potential therapeutic approach (Hernandez et al., ).
### Early Exposure to Environmental Factors and AD or PD Development: Epigenetic Evidence
Epigenetic DNA modifications include DNA methylation, histone post-translational modifications (mainly acetylation) and miRNAs (Holliday, ). DNA methylation is one of the most studied epigenetic modifications that influence the gene expression. It involves the addition of methyl groups to cytosine bases located at cytosine–phosphate–guanine (CpG) sites by the action of DNA methyltransferases (DNMTs). Alterations in DNA methylation on the promoter regions of genes regulate the gene expression of important processes such as embryonic development, cellular differentiation and aging (Bird, ). Increasing evidence suggests that epigenetic changes in the developing embryo that may play important roles in the susceptibility to diseases in later life (imprinted disease phenotypes) result from maternal exposures to environmental stimuli at critical periods of development. This suggests that a short exposure to chemicals could be memorized through epigenetic mechanisms long after the chemical trigger has gone (Jang and Serra, ), and recent studies have suggested that an epigenetic component could be involved in neurodegenerative diseases related to environmental factors (Marques et al., ).
The latent brain expression of genes observed in animals developmentally exposed to an environmental contaminant may be mediated through epigenetic pathways that are regulated via the DNA methylation. While the conditions leading to early life hypo- or hyper-methylation of specific genes are not known, both can induce oxidative DNA damage; for instance the hypo-methylation of APP gene increases its expression driving the overproduction of APP and Aβ levels, which in turn facilitate the ROS production damaging the DNA, and producing neuronal loss. While the hyper-methylation affects the gene transcription and DNA repair pathways. Therefore, both changes in DNA methylation can impact gene expression and imprint susceptibility to oxidative DNA damage in the aged brain (Zawia et al., ). Thus, it is suggested that Pb interferes with the DNA methylating capacity, thus altering the expression of AD-related genes. The study performed in aged monkeys developmentally exposed to Pb revealed a reduced activity of brain Dnmt, and the exposure of mouse primary cells from the cerebral cortex to Pb (0.1 µM) resulted in a similar effect on Dnmt1 activity a week after 24 h-treatment (Wu et al., ). Also, Bihaqi and Zawia ( ) showed a significant latent increase in AD biomarkers an a reduction in the protein and mRNA levels of DNA methylating enzymes Dnmt1 and Dnmt3a, and methyl CpG binding protein 2 (MeCP2) in differentiated SH-SY5Y cells treated with Pb (5–100 µM/48 h) and analyzed 6 days later (Bihaqi and Zawia, ). Aberrant CpG methylation in APP , Tau and GSK3 β genes was reported in post-mortem brains (Iwata et al., ). In addition, it suggested that reduced levels of CpG methylation in the promoter of APP could be mediated by the oxidation of guanine (8-oxdG) (Zawia et al., ); this is because the oxidation of guanine in CpG dinucleotides inhibits adjacent cytosine methylation (Weitzman et al., ). On the other hand, Cd, another metal involved in AD pathology, reduces the enzymatic activity of Dnmt in rat liver cell cultures (Poirier and Vlasova, ), but this effect has not been evaluated in cerebral cells. While a study showed that subchronic As exposure (3 and 36 ppm/from gestation until 4 months of age) altered the methylation of genes involved in neuronal plasticity, including reelin (RELN) and protein phosphatase 1 (PP1), which was associated with memory deficits (Martínez et al., ). Regarding other compounds, the perinatal exposure to permethrin (34 mg/Kg/daily, by gavage from postnatal day 6–21) to mice showed altered brain functions including biomarkers of maintenance of dopaminergic neurons, and impairment of spatial memory at 6 months of age (Nasuti et al., ).
The relation between epigenetic modifications and PD has been less studied; however, a potential role of DNA methylation in the promoter of α-syn encoding gene ( SNCA ) in the neuropathogenesis of PD has been suggested, considering that α-syn is a fundamental component of LB, the main hallmark of PD (Lu et al., ). A DNA hypomethylation of SNCA was reported in the substantia nigra of sporadic PD patients, suggesting that it might contribute to the dysregulation of SNCA expression in PD (Jowaed et al., ; Matsumoto et al., ). In addition, increased SNCA mRNA levels were observed in SNpc of PD (Chiba-Falek et al., ), and reduced levels of Dnmt1 have been observed in postmortem brains from PD and dementia with LB (DLB) patients, as well as in brains of α-syn transgenic mice; authors suggest that this effect could be a novel mechanism of epigenetic dysregulation in LB-related diseases such as PD (Desplats et al., ). Finally, a lesser degree of methyation of the TNF α promoter, a key inflammatory cytokine associated with dopaminergic cell death was observed in the SNpc from PD patients, predisposing to an increase neuronal vulnerability to inflammatory reactions (Mogi et al., ; Pieper et al., ).
Environmental factors associated with an increased risk of PD such as pesticides can alter the expression of genes by epigenetic mechanisms (Kwok, ). It was reported that pre-treatment with 5-aza-2’deoxycytidine (5’-aza-dC, a DNMT inhibitor) exacerbated the dopaminergic neuron damage induced by PQ, MPP , 6-hydroxydopamine (6-OHDA) and rotenone treatment, and induced oxidative stress, the transcriptional up-regulation of α-syn, and demethylation of the α-syn promoter (Wang et al., ). Likewise, the folate deficiency sensitizes mice to MPTP-induced PD-like pathology and motor dysfunction (Duan et al., ); it is well known that folate deficiency alters the development of human nervous system (Greenblatt et al., ).
On the other hand, it was reported that the exposure to environmental neurotoxicants associated with PD during early life or pregnancy can determine the progressive damage of the substantia nigra years before the onset of clinical parkinsonism, as well as to increase the vulnerability to effects of a second environmental factor (two–hit model) (Logroscino, ). A study in C57BL/6 mice daily treated with pQ (0.3 mg/Kg) or maneb (1 mg/Kg) or PQ + maneb from postnatal day 5–19 and then re-exposed as adults to PQ (10 mg/Kg) or maneb (30 mg/Kg) or PQ + maneb (twice a week/3 weeks) showed that dopaminergic cell loss and decreased dopamine levels were amplified by the adult re-challenge to the pesticides, suggesting that the developmental exposure to neurotoxins enhanced the adult susceptibility to a new toxic insult (Thiruchelvam et al., ). Similarly, prenatal exposure of pregnant C57BL/6J mice to PQ (0.3 mg/Kg) or maneb (1 mg/Kg) altered the development of the nigrostriatal system and enhanced its vulnerability to neurotoxins later in life, which could contribute with the development of PD during aging (Barlow et al., ).
Although there is no direct evidence linking early exposure to environmental pollutants and epigenetic changes with increased susceptibility to LOPD, there is a plausible association based on the following considerations: (1) epigenetic alterations have been observed in PD brains; (2) the exposure to environmental factors is associated with an increased risk of LOPD development and factors such as pesticides and metals can alter mechanisms of epigenetic regulation such as DNA methylation; and (3) early exposure to environmental pollutants might be associated with LOPD later in life. Further studies are needed to confirm this hypothesis in this promising research field to understand the mechanisms underlying the long-term effects of the environment on the PD development.
## Concluding Remarks
The emerging association between exposures to several toxic compounds with neurodegenerative diseases is of considerable public health importance, given the increasing dementia prevalence, the negative social and economic consequences of neurodegeneration-related disabilities, and the increasing environmental pollution in some geographic areas worldwide. Some of the epidemiological studies show not consistent results on getting significant estimates of hazard risk for AD or PD, mainly due to some limitations that include the difficulty on accurate diagnosis for AD or PD cases due to the lack of specific biomarkers, the deficiency to accurately assess chronic exposures, and/or the lack of inclusion of important confounding variables such as co-exposure to toxic compounds, genetic variants and lifestyle among others. Nevertheless, epidemiological studies along with experimental data have led to highlight the potential risk to develop these degenerative diseases due to the exposure to environmental pollutants such as metals, NPs and pesticides, among others. Interestingly, these pollutants show similar mechanisms of toxicity, which converge in a generalized mechanism based on the generation of oxidative stress that leads to common hallmarks of both neurodegenerative disorders. For example, the generation of oxidative stress by increasing the production of ROS and/or deregulating the antioxidant enzymes promotes the formation of protein aggregates such as Tau, Aβ or α-syn. This in turn overwhelms the degradation systems, and produces the activation of the glia inducing neuroinflammation, a process that per se increases the generation of further oxidative stress leading to a self-perpetuating cycle, and finally to neuronal loss of specific brain region such as the hippocampus and cerebral cortex in AD and substantia nigra in PD. The oxidative stress induced by these neurotoxicants activates/inhibits signaling pathways leading to augmented/diminished activity of enzymes that promote the accumulation of toxic materials in neural cells such as damaged/aberrant proteins, Aβ in AD or α–syn in PD and oxidative byproducts, or the oxidation of DNA that can alter genetic or epigenetic regulation. Furthermore, the link between early life exposure to environmental factors and the origin of neurodegenerative diseases is getting attention and can help to clarify the role of the environment on the development of these degenerative diseases. On the other hand, the lack of specific/differential biomarkers for AD or PD limits the early diagnosis and then the timely treatment. In this regard, specific circulating miRNAs have been associated with pathological processes such as AD and PD, therefore they are promising non-invasive biomarkers for these neurological diseases. Additionally, the identification of biomarkers to determine the past exposure to environmental pollutants is of vital importance for a better and opportune management of these diseases. Thus, as we have more knowledge of the risk from the exposure to environmental pollutants, more well-designed epidemiological studies (controlling for as many variables as possible and with high sample sizes) are necessary to improve the quality of life of elderly and to prevent the development of neurodegenerative diseases worldwide.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
D-serine is an endogenous coagonist at the glycine site of synaptic NMDA receptors (NMDARs), synthesized by serine racemase (SR) through conversion of L-serine. It is crucial for synaptic plasticity and is implicated in schizophrenia. Our previous studies demonstrated specific loss of SR, D-serine-responsive synaptic NMDARs, and glutamatergic synapses in cortical neurons lacking α7 nicotinic acetylcholine receptors, which promotes glutamatergic synapse formation and maturation during development. We thus hypothesize that D-serine and SR (D-serine/SR) are associated with glutamatergic synaptic development. Using morphological and molecular studies in cortical neuronal cultures, we demonstrate that D-serine/SR are associated with PSD-95 and NMDARs in postsynaptic neurons and with glutamatergic synapse stability during synaptic development. Endogenous D-serine and SR colocalize with PSD-95, but not presynaptic vesicular glutamate transporter 1 (VGLUT1), in glutamatergic synapses of cultured cortical neurons. Low-density astrocytes in cortical neuronal cultures lack SR expression but contain enriched D-serine in large vesicle-like structures, suggesting possible synthesis of D-serine in postsynaptic neurons and storage in astrocytes. More interestingly, endogenous D-serine and SR colocalize with PSD-95 in the postsynaptic terminals of glutamatergic synapses during early and late synaptic development, implicating involvement of D-serine/SR in glutamatergic synaptic development. Exogenous application of D-serine enhances the interactions of SR with PSD-95 and NR1, and increases the number of VGLUT1- and PSD-95-positive glutamatergic synapses, suggesting that exogenous D-serine enhances postsynaptic SR/PSD-95 signaling and stabilizes glutamatergic synapses during cortical synaptic development. This is blocked by NMDAR antagonist 2-amino-5-phosphonopentanoic acid (AP5) and 7-chlorokynurenic acid (7-CK), a specific antagonist at the glycine site of NMDARs, demonstrating that D-serine effects are mediated through postsynaptic NMDARs. Conversely, exogenous application of glycine has no such effects, suggesting D-serine, rather than glycine, modulates postsynaptic events. Taken together, our findings demonstrate that D-serine/SR are associated with PSD-95 and NMDARs in postsynaptic neurons and with glutamatergic synapse stability during synaptic development, implicating D-serine/SR as regulators of cortical synaptic and circuit development.
## Introduction
NMDARs are glutamate-gated ionotropic channels that are crucial for many physiological processes including neurotransmission, synaptic plasticity, and learning and memory (Waxman and Lynch, ; Hardingham and Bading, ; Vyklicky et al., ). In addition to glutamate, NMDAR activation requires the binding of coagonist, D-serine or glycine, to the glycine site of NR1 subunit of NMDARs (Johnson and Ascher, ; Kleckner and Dingledine, ; Mothet et al., , ; Labrie et al., ; DeVito et al., ; Papouin et al., ; Li and Wang, ; Rosenberg et al., ). In some paradigms, D-serine preferentially gates synaptic NMDARs and glycine preferentially gates extrasynaptic NMDARs (Papouin et al., ). D-serine is synthesized by serine racemase (SR) through conversion of L-serine and degraded by D-amino acid oxidase (DAAO) in various brain regions (Martineau et al., ; Wolosker, , ; Labrie et al., ; Billard, ; Campanini et al., ; Van Horn et al., ). D-serine is now recognized as an important physiological modulator in many NMDAR-dependent processes and functions, including brain development, synaptic transmission and plasticity, learning and memory, and social interactions (Mothet et al., , ; Yang et al., ; Kim et al., ; Labrie et al., ; DeVito et al., ; Papouin et al., ; Li and Wang, ; Rosenberg et al., ).
Abnormal reduction of D-serine levels have been found in schizophrenia patients (Hashimoto et al., , ; Yamada et al., ; Bendikov et al., ) and implicated in the pathogenesis of schizophrenia (Labrie et al., ). Targeted deletion or pathogenic perturbation of SR in mice reduces D-serine production and glutamatergic transmission in the forebrain and leads to schizophrenia-like behavior (Basu et al., ; Ma et al., ). SR null mutant mice, which have less than 10% of normal brain D-serine, have reduced dendritic spine density that can be partially rescued by chronic D-serine treatment (Balu and Coyle, ; Balu et al., , ). Disruption of D-serine/SR has also been associated with schizophrenia during development. Neonatal disruption of SR and D-serine synthesis in mice leads to schizophrenia-like behavioral abnormalities in adulthood (Hagiwara et al., ). Previous studies have demonstrated that glutamatergic synapse formation and maturation are promoted by α7 nicotinic acetylcholine receptor during synaptic development (Lin et al., ; Lozada et al., ). Glutamatergic synapses and D-serine/SR are decreased in the forebrain of α7 nicotinic acetylcholine receptor knockout mice (Lin et al., ). These changes resemble the major neurochemical deficits in schizophrenia (Lin et al., , ). We thus hypothesize that D-serine/SR may be involved in glutamatergic synaptic development, and that D-serine/SR deficiency may thereby disrupt cortical synaptic and circuit development, contributing to permanent deficits in schizophrenia.
The roles of D-serine/SR in neurons and astrocytes have been controversial. D-serine was characterized solely as a gliotransmitter in the initial studies showing D-serine and SR localized to astrocytes (Schell et al., ; Wolosker et al., ; Martineau et al., ; Panatier et al., ). Using more specific SR antibodies and SR-knockout mice as negative controls, SR is found preferentially expressed in excitatory and inhibitory neurons, and D-serine is predominantly produced and released by neurons in rodent and human brains (Kartvelishvily et al., ; Miya et al., ; Wolosker et al., ; Benneyworth et al., ; Ehmsen et al., ; Balu et al., ; Martineau et al., ; Mothet et al., ). SR interacts with DISC1 in astrocytes and pathogenic disruption of SR-DISC1 binding leads to depletion of D-serine levels and schizophrenia-like behavior in mice (Ma et al., ). SR interactions with stargazin and PSD-95 have been suggested to regulate NMDAR-AMPAR cross-talking in neurons (Ma et al., ), however the association of D-serine/SR with PSD-95 in postsynaptic neurons and their association with synapse stability remain unknown. In the current studies, we examined the expression and distribution of SR and D-serine and their association with glutamatergic synapses in primary cortical neuronal cultures containing low-density astrocytes from embryonic mice. Our findings reveal the association of D-serine/SR with PSD-95 and NMDARs in postsynaptic neurons and with glutamatergic synapse stability during cortical synaptic development based on the morphological and molecular evidence.
## Materials and methods
### Materials
Timed-pregnant C57BL/6 mice were purchased from Charles River Laboratories. Biochemicals included 2-amino-5-phosphonopentanoic acid (AP5), D-serine, L-serine, glycine and lithium (Sigma), 7-chlorokynurenic acid (7-CK) (Tocris Bioscience). Antibodies included α-NR1 (BD Pharmagen, mouse monoclonal, Millipore, mouse monoclonal), α-NR2A, α-NR2B (Alomone Labs, rabbit polyclonal), α-PSD-95 (BD Transduction Laboratories, mouse monoclonal; NeuroMab, mouse monoclonal), α-VGLUT1 (Synaptic Systems, polyclonal; mouse monoclonal), α-GAD65 (DSHB, mouse monoclonal), α-gephyrin (Synaptic systems, mouse monoclonal), α-D-serine, α-L-serine, α-Serine Racemase (Abcam, rabbit polyclonal), α-GFAP (Sigma, rabbit polyclonal; NeuroMab, mouse monoclonal), α-GLT1 (obtained from Dr. Michael B. Robinson, a courtesy of Dr. Jeffrey Rothstein, mouse monoclonal; NeuroMab, rabbit polyclonal), α-MAP2 (Abcam, chicken polyclonal), α-GABA (Sigma, rabbit polyclonal), α-VAMP2 (Synaptic systems, mouse monoclonal).
### Neuronal cultures and drug treatment
Primary cortical cultures from E17-19 C57BL/6 mice were prepared as described (Lin et al., ) in accordance with the protocol approved by The Children's Hospital of Philadelphia Institutional Animal Care and Use Committee (IACUC). Briefly, the cortex was dissected, gently minced, trypsinized (0.027%, 37°C; 7% CO for 20 min), and then washed with 1 × HBSS. Neurons were seeded to a density of 3 × 10 viable cells in 35-mm culture dish with five 12-mm glass coverslips (low-density culture, 3 × 10 /cm ) or a density of 1.6 × 10 viable cells in 60-mm culture dishes (high-density culture, 8 × 10 /cm ). The culture dishes were coated with poly-D-Lysine (100 μg/ml) prior to seeding neurons. Neurons were maintained at 37°C with 5% CO in Neurobasal medium with B27 supplement. Cortical cultures contain 5–10% of glia cells and 90–95% cortical neurons. At 17–19 (high-density cultures, 8 × 10 /cm ) or 23–25 (low-density cultures, 3 × 10 /cm ) days in vitro (DIV), cultures were subject to drug treatment for 24 h, western blotting analysis, co-immunoprecipitation and immunocytochemistry. For drug treatment, the cortical cultures were treated with vehicle, D-serine (50 μM), D-serine (50 μM) + AP5 (50 μM), D-serine (50 μM) + 7-CK (50 μM), glycine (50 μM) + lithium (100 μM) for 24 h.
For cell lysate preparation, cultures were lysed in lysis buffer (150 mM NaCl, 1 mM EDTA, 100 mM Tris-HCl, 1% Triton X-100, 1% sodium deoxycholate and 1% SDS, pH 7.4, supplemented the day of use with 1:500 EDTA-free protease inhibitor cocktail III (Calbiochem) for 1 h at 4°C and collected. Debris was cleared by centrifugation at 16,100 × g for 20 min at 4°C. Supernatants were stored at −80°C until use.
### Co-immunoprecipitation and western blotting analysis
Co-immunoprecipitation and Western blotting was performed as described previously (Zhai et al., ). Protein content of cortical lysates was determined using BCA Protein Assay (Thermo Scientific). Equal amounts of total protein lysates (250 μg) were first added 2 μg primary antibody (α-SR, α-NR1, or α-PSD-95) or normal IgG and incubated at 4°C for 2h. Immunocomplexes were then precipitated with protein A or protein G-agarose beads shaking overnight at 4°C, washed twice in lysis buffer, eluted by boiling in SDS-PAGE sample buffer, and subjected to Western blot analysis. Equal volumes of eluted buffers for co-immunoprecipitation assay or equal amounts of total protein (15 μg cell lysate) for protein input analysis were subjected to 4–12% NuPAGE Gel for electrophoresis and transferred to nitrocellulose membranes. Membranes were blocked with 3% nonfat milk and incubated with primary antibody overnight at 4°C. Blots were then incubated with appropriate horseradish peroxidase, HRP-conjugated secondary antibodies (Cell Signaling) for 2 h at room temperature and then washed; Reaction bands were visualized using a luminol-enhanced chemiluminescence (ECL) HPR substrate (Thermo Scientific). Each blot was then incubated with stripping buffer (2% SDS, 50 mM Tris, pH 6.8, and 100 mM β-mercaptoethanol) for 45 min at room temperature to remove the signals and reprobed for other proteins. For quantification analysis, reaction product levels were quantified by scanning densitometry and the ratio of co-precipitated protein was normalized by input levels from 3 different cultures and experiments using NIH Image J software.
### Immunocytochemistry and fluorescence imaging
Primary cultured cortical neurons were fixed for 20 min at 4°C with 4% paraformaldehyde in phosphate-buffered saline (PBS) (pH 7.4), and then subjected to the immunostaining procedure. For immunostaining procedure, after blocking with 5% normal goat serum and 1% bovine serum albumin in combination with 0.3% (vol/vol) Triton X-100 in PBS at room temperature for 1 h, the coverslips or slides were incubated with primary antibodies at 4°C overnight and then secondary antibodies conjugated to Alexa fluor 488 or 568 or 647 (Invitrogen) at room temperature for 60–90 min. Following several washes with PBS, cells or slides were mounted with Vectashield with DAPI (Vector Laboratories). Fluorescence images were obtained with Olympus FluoView and Leica TCS SP8 laser scanning confocal microscope. For quantification analysis of glutamatergic synapses in cortical cultures, neurons were stained for glutamatergic synaptic markers (VGLUT1 and PSD-95). The confocal images were acquired under 40x objectives with zoom x6 from the dendrites of 5 neurons and 3 different cultures for quantification of VGLUT1-positive puncta on the dendrites of cortical neurons. NIH Image J software and the thresholded images were used to quantify the number of glutamatergic synapses on the primary and secondary dendrites of cortical neurons. Since nearly 100% of VGLUT1-positive puncta were colocalized with or adjacent to one or more PSD-95-positive puncta in the dendrites of cortical neurons, the number of VGLUT1-positive puncta reasonably represents the number of glutamatergic synapses on the dendrites (Lin et al., ).
### Statistical analysis
Data was shown as the mean ± S.E.M. Experiments were analyzed using Student's t -test to compare two conditions or ANOVA followed by planned comparisons of multiple conditions. Significance was set at P < 0.05.
## Results
### Serine racemase (SR) colocalizes with PSD-95, but not presynaptic VGLUT1, in glutamatergic synapses of cortical glutamatergic and gabaergic neurons
SR is an endogenous biosynthetic enzyme that converts L-seine to D-serine, and DAAO is an endogenous D-amino acid oxidase that degrades D-serine in the nervous system. To explore the possible role of D-serine in cortical synaptic development, we first examined the expression and distribution of SR and DAAO in primary cortical neuronal cultures at 28 days in vitro (DIV) by immunocytochemical studies using antibodies to SR, DAAO, astrocytic marker (glutamate transporter 1, GLT1) and neuronal markers (Microtubule-associated protein 2, MAP2). Immunostaining with astrocytic and neuronal marker antibodies indicates that cortical neuronal cultures contain 5–10% glial cells and 90–95% cortical neurons. Triple immunostaining with α-SR or α-DAAO and α-MAP2 and α-GLT1 show that SR is abundantly distributed in the soma, nucleus and dendrites of MAP2-positive cortical neurons (N) (Figure ), but absent in the soma and astrocytic terminals of GLT1-positive astrocytes (AS) (Figure ) in cortical neuronal cultures (Figure ). In contrast, DAAO is abundantly distributed in the soma of GLT1-positive astrocytes (AS) (Figure ), but absent in MAP2-positive neurons (N) (Figure ) in cultures (Figure ), suggesting possible synthesis of D-serine in neurons and degradation in astrocytes. High magnification confocal images further show that the majority of SR appears as puncta in the synaptic-like structures (Figures ) on the MAP2-positive dendrites (Figures ). Noticeably, some SR-positive puncta on the MAP2-positive dendrites are close to or overlap with GLT1-positive astrocytic membranes (Figures ).
SR is abundant in cortical neurons but absent in low-density astrocytes in primary cortical neuronal cultures. (A–F) Confocal images of SR or DAAO immunofluorescence (red), GLT1 immunofluorescence (green), MAP2 immunofluorescence (cyan) and DAPI staining (blue) in primary cortical neuronal cultures with low-density astrocytes showing that SR appears abundantly in the MAP2-positive cortical neurons (N) (A) but absent in GLT1-positive astrocytes (AS) (B) . D-serine degrading enzyme DAAO is abundantly present in the GLT1-positive astrocytes (AS) (E) but absent in MAP2-positive cortical neurons (N) (D) . (G–L) Higher magnification of confocal images showing that SR appears abundant in the synaptic-like structures on MAP2-positive dendrites of cortical neurons (N), but absent in GLT1-positive astrocytes (AS). Scale bars as indicated.
To define the nature of SR distribution in the synaptic-like structures on the dendrites, we examined whether SR is located in glutamatergic or GABAergic synapses using antibodies to glutamatergic and GABAergic presynaptic and postsynaptic markers. We first characterized cortical glutamatergic and GABAergic neurons in cultures using the GABAergic neuronal marker GABA as well as glutamatergic markers PSD-95 and VGLUT1. In GABA-positive cortical GABAergic interneurons, GABA is distributed in the soma, dendrites as well as axonal and presynaptic terminals surrounding glutamatergic neurons (Supplemental Figures – ), and PSD-95 is localized in the shaft-like synapses on the somatic and dendritic membrane (Supplemental Figures – ). In cortical glutamatergic neurons surrounded by GABA-positive synaptic terminals, PSD-95 is distributed in the soma and in the spine-like synapses on the dendrites (Supplemental Figures – ). Co-immunostaining with PSD-95 and VGLUT1 antibodies further confirms PSD-95- and VGLUT1-positive shaft-like synapses on cortical GABAergic interneurons (Supplemental Figures – ) as well as dendritic spine-like synapses on cortical glutamatergic neurons (Supplemental Figures – ).
Based on co-immunostaining with antibodies to glutamatergic presynaptic and postsynaptic markers (VGLUT1 and PSD-95), SR colocalizes with PSD-95-positive glutamatergic postsynaptic terminals on the dendrites of cortical glutamatergic (Figures ) and GABAergic neurons (Figures ), but does not colocalize with VGLUT1-positive glutamatergic presynaptic terminals on the dendrites of cortical glutamatergic (Figures ) and GABAergic neurons (Figures ). In addition, co-immunostaining with antibodies to GABAergic presynaptic and postsynaptic markers (GAD65 and gephyrin) shows that SR does not colocalize with GAD65-positve GABAergic presynaptic (Supplemental Figures – ) or gephyrin-positive GABAergic postsynaptic terminals (Supplemental Figures – ) on cortical glutamatergic neurons (Supplemental Figures – , – ) and GABAergic neurons (Supplemental Figures – , – ). Taken together, these results indicate that SR is a postsynaptic protein that colocalizes with PSD-95 in glutamatergic synapses of cortical neurons.
SR is a postsynaptic protein that colocalizes with PSD-95 in the glutamatergic synapses on cortical glutamatergic and GABAergic neurons . Confocal images of SR immunofluorescence (red) and PSD-95 immunofluorescence (green) showing that SR appears as puncta in the soma, nucleus and dendrites of cortical neurons (A,D,G,J) in primary cortical neuronal cultures at 28 DIV. SR colocalizes with PSD-95-positive glutamatergic postsynaptic terminals (B,E,H,K) on the dendrites of cortical glutamatergic (C,F) and GABAergic (I,L) neurons as characterized by PSD-95 immunofluorescence shown in Supplemental Figure . Scale bars as indicated.
SR is absent in glutamatergic presynaptic terminals on cortical glutamatergic and GABAergic neurons . Confocal images of SR immunofluorescence (red) and VGLUT1 immunofluorescence (green) showing that SR appears as puncta in the soma, nucleus and dendrites of cortical neurons (A,D,G,J) . SR does not colocalize with VGLUT1-positive glutamatergic presynaptic terminals (B,E,H, K) on cortical glutamatergic (C,F) and GABAergic (I,L) neurons as characterized by VGLUT1 and PSD-95 immunofluorescence shown in Supplemental Figure . Scale bars as indicated.
### Endogenous D-serine colocalizes with PSD-95-positive glutamatergic postsynaptic terminals on cortical glutamatergic and gabaergic neurons
The distribution of D-serine was directly examined in cortical cultures using triple immunostaining with α-D-serine antibody along with antibodies to neuronal (α-MAP2) and astrocytic markers (α-GFAP). To determine the specificity of α-D-serine antibody, we first compared the immunoreactivities of α-D-serine and α-L-serine antibodies, D-serine appears abundant as puncta in the soma and dendrites of MAP2-positive cortical neurons and in the soma of astrocytes (Supplemental Figures – ), whereas L-serine is diffusely distributed in GFAP-positive astrocytes and very low in cortical neurons (Supplemental Figures – ), suggesting the specificity of D-serine immunoreactivity and matching the findings on astrocytic supply of L-serine (Ehmsen et al., ). Furthermore, D-serine is abundant in the soma and in the synaptic-like structures on the MAP2-positive (Figures ) dendrites of cortical neurons (N) in cortical cultures (Figures ), whereas D-serine appears enriched as large puncta in the soma of GFAP-positive (Figures ) astrocytes (AS). Co-immunostaining with synaptic vesicle membrane associated protein 2 (VAMP2) antibody shows that D-serine appears enriched in large vesicle-like structures (1–3 μm) which are VAMP2-negative in the soma of astrocytes (AS) (Figures ), and that D-serine is abundant in the dendrites but absent in the VAMP2-positve presynaptic terminals on the dendrites of cortical neurons (Figures ).
D-serine is abundant in cortical neurons and enriched in low-density astrocytes in primary cortical neuronal cultures. (A–I) Confocal images of D-serine immunofluorescence (red), GFAP immunofluorescence (green), MAP2 immunofluorescence (cyan), and DAPI nucleus staining (blue) showing that D-serine appears abundantly in the soma and dendrites of MAP2-positive cortical neurons (N) (A) and in the soma of GFAP-positive astrocytes (AS) (B) . Higher magnification of confocal images (D–I) showing that D-serine appears abundant in the synaptic-like structures on MAP2-positive dendrites (D,G) and enriched in large vesicle-like structures (1–3 μm) in the soma of GFAP-positive astrocytes (E,H) . (J–L) Confocal images of D-serine immunofluorescence (red), VAMP2 immunofluorescence (green), and DAPI staining (blue) showing that D-serine appears abundant in the soma and dendrites of cortical neurons surrounded by VAMP2-positive presynaptic terminals (N) and enriched in the soma of VMAP2-negative astrocytes (AS) (J) . Higher magnification of confocal image confirms that D-serine appears in large vesicle-like structures (1–3 μm) in the soma of VMAP2-negative astrocytes (AS) (K) and in the dendrites of cortical neurons surrounded by VAMP2-positive presynaptic terminals (L) . Scale bars as indicated.
We further defined the localization of D-serine within glutamatergic presynaptic and postsynaptic terminals of cortical neurons. Co-immunostaining with α-D-serine and α-PSD-95 antibodies shows that D-serine appears in the synaptic-like structures that colocalize with PSD-95 on the dendrites of cortical glutamatergic (Figures ) and GABAergic neurons (Figures ). High magnification of confocal images confirms that D-serine colocalizes with PSD-95-positive postsynaptic terminals in the spine-like synapses on cortical glutamatergic neurons (Figures ) and in the shaft-like synapses on cortical GABAergic interneurons (Figures ), suggesting the localization of D-serine in postsynaptic neurons. Moreover, co-immunostaining with antibodies to D-serine and the glutamatergic presynaptic terminal marker α-VGLUT1 shows that D-serine is adjacent to, but does not colocalize with, VGLUT-positive glutamatergic presynaptic terminals on cortical glutamatergic (Figures ) and GABAergic neurons (Figures ). Taken together, our findings demonstrate D-serine association with PSD-95 in postsynaptic neurons and possible storage of D-serine in astrocytes.
D-serine colocalizes with PSD-95-positive postsynaptic terminals on cortical glutamatergic and GABAergic neurons. (A–F) Confocal images of D-serine immunofluorescence (red), PSD-95 immunofluorescence (green), and DAPI nucleus staining (blue) showing that D-serine (A,D) colocalizes with PSD-95 (B,E) in the spine-like synapses on the dendrites of cortical glutamatergic neurons in primary cortical neuronal cultures at 28 DIV (C,F) . (G–L) Confocal images showing that D-serine (G,J) colocalizes with PSD-95 (H,K) in the shaft-like synapses on the dendrites of cortical GABAergic neurons (I,L) as characterized by PSD-95 immunofluorescence shown in Supplemental Figure . Scale bars as indicated.
D-serine is absent in glutamatergic presynaptic terminals on cortical glutamatergic and GABAergic neurons . Confocal images of D-serine (B,E,H,K) red immunofluorescence (red) and VGLUT1 (A,D,G,J) immunofluorescence (green) showing that D-serine does not colocalize with VGLUT1-positive glutamatergic presynaptic terminals on cortical glutamatergic (C,F) and GABAergic (I,L) neurons as characterized by VGLUT1 and PSD-95 immunofluorescence shown in Supplemental Figure . Scale bars as indicated.
### Endogenous SR and D-serine colocalize with PSD-95 in the postsynaptic terminals of glutamatergic synapses of cortical neurons during early synaptic development
Previous findings suggest that D-serine/SR may be involved in glutamatergic synaptic development (Balu et al., , ; Ma et al., ; Lin et al., ). To test this, we further examined the association of D-serine/SR with PSD-95 in early cortical cultures (16 DIV and 23 DIV). Co-immunostaining with α-SR and α-PSD-95 antibodies shows that, in cortical neuronal cultures at 16 DIV, SR (Figure ) is located in the postsynaptic terminals of spine-like synapses and colocalizes with PSD-95 (Figure ) on the MAP2-positive dendrites (Figure ) of cortical neurons. At 23 DIV, the number of glutamatergic synapses increases on the MAP2-positive dendrites of cortical neurons, and SR colocalizes with PSD-95 in the postsynaptic terminals of glutamatergic synapses (Figures ). Co-immunostaining with α-D-serine and α-PSD-95 antibodies shows similar co-localization of D-serine (Figures ) and PSD-95 (Figures ) in the postsynaptic terminals of glutamatergic synapses (Figures ) in early cortical cultures at 16 DIV and 23 DIV, suggesting that D-serine/SR may be involved in PSD-95 signaling and glutamatergic synaptic development.
Endogenous SR and D-serine colocalize with PSD-95 in the postsynaptic terminals of glutamatergic synapses during early synaptic development. (A–F) Confocal images of SR immunofluorescence (red), PSD-95 immunofluorescence (green), and MAP2 immunofluorescence (cyan) showing colocalization of SR (A,D) with PSD-95 (B,E) in the spine-like synapses on the MAP2-positive dendrites (C,F) of cortical neurons at 16DIV (A–C) and 23 DIV (D–F) . (G–L) Confocal images of D-serine immunofluorescence (red) and PSD-95 immunofluorescence (green) showing colocalization of D-serine (G,J) with PSD-95 (H,K) in the spine-like synapses on the dendrites (I,L) of cortical neurons at 16DIV (G–I) and 23 DIV (J–L) . Scale bars as indicated.
### Exogenous D-serine, but not glycine, enhances the In vivo interactions of SR with PSD-95 and NMDARs postsynaptically
To explore the possible involvement of D-serine/SR in postsynaptic PSD-95 signaling, we examined the effects of exogenous D-serine application on PSD-95 interactions with SR and NMDARs in cortical cultures. We first confirmed the in vivo interactions of endogenous SR with PSD-95 in cortical neurons using co-immunoprecipitation. In cortical neuronal lysates, α-SR antibody co-precipitates PSD-95 as well as NR1 (Figure ). Similarly, α-NR1 antibody co-precipitates SR and PSD-95 in cortical neuronal lysates (Figure ), and incubation of cortical neuronal lysates with α-PSD-95 antibody co-precipitates NR1 and SR (Figure ). Furthermore, α-SR antibody co-precipitates NR2A and a lesser amount of NR2B in cortical neuronal lysates (Figure ). These findings are not observed with control IgG in these assays (Figures ). This identifies the in vivo interactions of SR with PSD-95, NR1 and NR2 in a postsynaptic protein complex.
In vivo interactions of endogenous SR with PSD-95 and NMDARs postsynaptically in cortical neurons. (A) In normal cortical neurons, co-immunoprecipitation (IP) of endogenous PSD-95 and NR1 with SR antibody in cortical neuronal lysates was observed, but co-precipitation with normal IgG was not; (B) , Co-IP of endogenous PSD-95 and SR with NR1 antibody in cortical neuronal lysates was observed, but co-precipitation with normal IgG was not; (C) . Co-IP of NR1 and SR with PSD-95 antibody was observed, but co-precipitation with normal IgG was not; (D) . Co-IP of NR2A and a lesser amount of NR2B with SR antibody in cortical neuronal lysates was observed, but co-precipitation with normal IgG was not.
To explore the possible role of D-serine in postsynaptic functions, we examined the effects of exogenous application of D-serine on the in vivo interactions of SR with PSD-95 and NMDARs using co-immunoprecipitation assays with α-SR antibody. Exogenous application of D-serine markedly increases the amount of PSD-95 and NR1 co-immunoprecipitated by α-SR antibody from the same amount of cortical neuronal lysate, even though the levels of PSD-95, NR1 and SR inputs remain similar between the control and treatment groups (Figures ). The ratios of co-precipitated NR1 and PSD-95 normalized to input controls are significantly increased in D-serine-treated cortical lysates (Figures ). The results suggest that exogenous D-serine enhances the in vivo interactions of SR with NR1 and PSD-95. This enhancement is blocked by co-application of AP5, an NMDAR competitive antagonist, as well as with 7-chlorokynurenic acid (7-CK), a specific NMDAR competitive antagonist at the glycine site of NMDARs (Figures ), demonstrating that D-serine modulates SR/PSD-95 interactions through NMDAR activation. In contrast, exogenous application of glycine along with glycine transporter inhibitor lithium, does not have the same effect on SR interactions with NR1 and PSD-95 as exogenous D-serine (Figures ), suggesting that D-serine, rather than glycine, modulates postsynaptic SR interactions with PSD-95 and NMDARs.
Exogenous D-serine, but not glycine, enhances the in vivo interactions of endogenous SR with PSD-95 and NMDARs postsynaptically. (A–C) Co-immunoprecipitation of endogenous NR1 and PSD-95 with SR antibody in cortical neuronal lysates from cultures treated with vehicle, D-serine (50 μM), D-serine (50 μM) + AP5 (50 μM), D-serine (50 μM) + 7-CK (50 μM), glycine (50 μM) + lithium (100 μM) for 24 h. Upper panel: Co-immunoprecipitation assays of cortical lysates (250 μg each lane); Lower panel: Western blot analysis of input levels in cortical lysates (15 μg each lane). Graphs (D–F) show quantification analysis of the co-precipitated NR1, PSD-95, and SR with α-SR antibody in the cortical lysates from 3 different cultures and experiments. The co-precipitated products were normalized to inputs levels in cortical lysates.
### Exogenous D-serine, but not glycine, stabilizes cortical glutamatergic synapses
We then examined the effects of exogenous D-serine on glutamatergic synapses in cortical cultures (23-25DIV) by immunocytochemical studies with antibodies to glutamatergic synaptic markers (α-VGLUT1 and α-PSD-95). Exogenous application of D-serine for 24 h dramatically increases the number of VGLUT1- and PSD-95-positive synapses in cortical cultures (Figures ) compared with control cultures (Figures ) with a more pronounced effect on presynaptic terminals. This suggests that D-serine may increase the number of glutamatergic synapses through synapse stabilization. The increase in synapse number is blocked by the NMDAR antagonist 7-CK (Figures ), whereas exogenous application of glycine along with the glycine transporter inhibitor lithium does not increase the number of glutamatergic synapses (Figures ). Higher magnification of confocal images further show that D-serine, but not glycine, increases the number of VGLUT-1- and PSD-95-positive glutamatergic synapses on the dendrites of cortical neurons, which can be blocked by the NMDAR antagonist 7-CK (Figures ). Quantification confirms that D-serine, but not glycine, significantly increases the number of glutamatergic synapses on the dendrites, and that this increase can be blocked by 7-CK (Figure ). The findings thus demonstrate that D-serine, unlike glycine, stabilizes glutamatergic synapses on the dendrites of cortical neurons through NMDARs.
Exogenous D-serine, but not glycine, increases the number of glutamatergic synapses in cortical cultures . Confocal images (A–L) of α-VGLUT1 (A,D,G,J) immunofluorescence (red) and α-PSD-95 (B,E,H,K) immunofluorescence (green) shows that VGLUT1- and PSD-95-positive glutamatergic synapses surrounding cortical neurons are markedly increased by 50 μM exogenous D-serine (D–F) , but not 50 μM glycine + 100 μM lithium (J–L) compared with control cortical neurons (A–C) . This increase can be blocked by co-application of 50 μM 7-CK with 50 μM D-serine (G–I) . Higher magnification of confocal images (A'–L') showing VGLUT1- and PSD-95-positive glutamatergic synapses on the dendrites of cortical neurons in control (A'–C') , D-serine (D'–F') , D-serine+7-CK (G'–I') , glycine+lithium (J'–L') groups. Graph (M) shows quantification of the number of VGLUT1-positive glutamatergic synapses on the dendrites of cortical neurons from 3 different cultures and experiments. Scale bars as indicated.
## Discussion
The present study uses in vitro cultured cortical neurons to show that D-serine and SR are associated with PSD-95 and NMDARs in postsynaptic neurons and with glutamatergic synapse stability. Endogenous D-serine and SR colocalize with PSD-95, but not presynaptic VGLUT1, and is highly enriched in the postsynaptic density of glutamatergic synapses of cortical neurons. Low-density astrocytes in cortical cultures lack SR expression but contain enriched D-serine in large vesicle-like structures, suggesting possible synthesis of D-serine in postsynaptic neurons and separate storage of D-serine in astrocytes. More interestingly, endogenous D-serine and SR colocalize with PSD-95 in the postsynaptic terminals of glutamatergic synapses during early and late synaptic development, implicating D-serine/SR involvement in glutamatergic synaptic development. Exogenous application of D-serine enhances SR interactions with PSD-95 and NMDARs in a postsynaptic protein complex and increases the number of glutamatergic synapses, suggesting that exogenous D-serine enhances postsynaptic SR/PSD-95 signaling and stabilizes glutamatergic synapses during synaptic development. This is blocked by the NMDAR antagonists AP5 and 7-CK, a specific antagonist at the glycine site of NMDARs. In contrast, exogenous application of glycine has no such effects, suggesting that the D-serine effects are mediated through synaptic NMDARs. The present results thus implicate D-serine and SR in control of PSD-95 signaling and glutamatergic synapse stability during cortical synaptic and circuit development.
The localization of SR and D-serine in astrocytes and neurons remains disputed. SR was initially localized to astrocytes, but later found prominently in neurons (Schell et al., ; Wolosker et al., , ; Kartvelishvily et al., ; Martineau et al., , ; Panatier et al., ; Miya et al., ; Benneyworth et al., ; Ehmsen et al., ; Balu et al., ; Mothet et al., ). Our results show that SR is a postsynaptic protein that colocalizes with PSD-95, a glutamatergic postsynaptic terminal marker, but not presynaptic marker VGLUT1, in the glutamatergic synapses of cortical glutamatergic and GABAergic neurons. Co-immunoprecipitation confirms endogenous SR interactions with PSD-95 and NMDARs postsynaptically and co-immunostaining shows that D-serine appears enriched in PSD-95-positive glutamatergic postsynaptic terminals on cortical neurons, suggesting D-serine may be synthesized and produced in the postsynaptic terminals. Low-density astrocytes in primary cortical neuronal cultures lack endogenous SR expression but contain enriched D-serine in large vesicle-like structures (1–3 μm), suggesting possible storage of D-serine in astrocytes. This matches the findings from other labs that D-serine is localized in astrocytic synaptic-like vesicles and released as large vesicles (1–3 μm) in astrocytes (Kang et al., ; Martineau et al., ). The presence of the D-serine degrading enzyme DAAO in astrocytes further suggests possible degradation of D-serine in astrocytes. Our findings thus implicate both neuronal and astrocytic D-serine in synaptic development and postsynaptic functions.
Interactions between SR and PSD-95 are involved in coupling synaptic NMDAR and AMPAR activities and functions (Ma et al., ). We found that exogenous D-serine application modulates postsynaptic SR/PSD-95/NMDAR interactions in cortical neurons, implicating D-serine in SR/PSD-95 signaling and postsynaptic functions. SR is a highly regulated enzyme that interacts with several NMDAR- and AMPA receptor (AMPAR)-interacting proteins including GRIP1, PICK1, DISC1, stargazin, and PSD-95 (Kim et al., ; Fujii et al., ; Hikida et al., ; Ma et al., , ). The physiological interactions of SR with these receptor-interacting proteins may affect SR translocation to the plasma membrane and SR catalytic activity as well as synaptic NMDAR-AMPAR activities. For instance, NMDAR activation promotes translocation of SR to the plasma membrane, which dramatically reduces the enzyme activity (Balan et al., ). AMPAR activation dissociates SR from the protein complex on the membranes and translocates it to the cytosol leading to enhanced activity of SR to generate more D-serine (Ma et al., ). SR interactions with DISC1, PICK1, and GRIP1 also modulate SR activity and D-serine production (Kim et al., ; Ma et al., ; Nomura et al., ). Our findings suggest that D-serine may control postsynaptic SR activity and D-serine production through modulation of SR interactions with PSD-95, NMDAR and other postsynaptic scaffold proteins. Furthermore, PSD-95 is a crucial postsynaptic scaffold protein that regulates glutamatergic synapse formation and maturation (Migaud et al., ; El-Husseini et al., ; Nikonenko et al., ). For example, PSD-95 promotes spine synapse formation through direct interaction with postsynaptic neuronal nitric oxide synthase (nNOS) and drives presynaptic and postsynaptic maturation of glutamatergic synapses (El-Husseini et al., ; Nikonenko et al., ). PSD-95 also controls synaptic AMPAR number through direct interaction with stargazin (Schnell et al., ). Therefore, our findings suggest that D-serine could modulate postsynaptic PSD-95 signaling, possibly through interactions with nNOS and synaptic AMPARs, thereby modulating glutamatergic synaptic development.
Our findings further reveal that endogenous D-serine and SR colocalize with PSD-95 in the postsynaptic terminals of glutamatergic synapses during early and late synaptic development. Exogenous D-serine indeed increases the number of VGLUT1- and PSD95-positive glutamatergic synapses on the dendrites of cortical neurons, implicating D-serine in control of glutamatergic synapse stability during cortical synaptic and circuit development. Our findings also show that the D-serine effects on glutamatergic synapses appears more pronounced on presynaptic than postsynaptic terminals, suggesting that D-serine may stabilize glutamatergic synapses through nitric oxide retrograde signaling via postsynaptic PSD-95/NOS interactions. SR null mutant mice, which have less than 10% of normal brain D-serine and a schizophrenia-like phenotype, have reduced dendritic spine density that can be partially rescued by chronic D-serine treatment (Balu and Coyle, ; Balu et al., , ). In addition, disruption of D-serine synthesis in mice during early postnatal life leads to schizophrenia-like behavioral abnormalities in adulthood, which can be rescued by chronic D-serine treatment during juvenile life (Hagiwara et al., ). Similarly, targeted deletion of the SR-interacting protein PICK1 in mice leads to D-serine deficiency in prefrontal cortex and schizophrenia-like behavioral abnormalities. The abnormalities can be rescued by transient neonatal supplementation of D-serine, but not by a similar treatment in adulthood, implicating D-serine in brain development (Nomura et al., ). Our findings suggest that D-serine/SR may regulate cortical glutamatergic synaptic development, which could be disrupted by abnormal D-serine/SR levels in early life leading to permanent deficits in adulthood in schizophrenia. Hence, D-serine treatment may provide a potential therapeutic for rescuing synaptic deficits in schizophrenia.
Synaptic and extrasynaptic NMDARs are associated with differential gene expression and have different roles in synaptic plasticity and cell death (Hardingham and Bading, ; Gladding and Raymond, ; Kaufman et al., ; Papouin et al., ; Karpova et al., ; Parsons and Raymond, ). D-serine preferentially gates synaptic NMDARs while glycine preferentially gates extrasynaptic NMDARs (Papouin et al., ). Our studies show that D-serine and glycine have differential effects on postsynaptic SR/PSD-95 signaling and glutamatergic synapse stability, implicating D-serine, rather than glycine, in controlling postsynaptic protein signaling and glutamatergic synaptic development. This is likely mediated by synaptic, rather than extrasynaptic, NMDARs. Synaptic NMDARs play crucial roles in many forms of synaptic plasticity, such as LTP and LTD. Similarly, the selective control of synaptic NMDARs may provide a mechanism by which D-serine regulates glutamatergic synaptic development and function.
## Author contributions
HL designed and performed the experiments, analyzed and interpreted the data, and wrote the paper. AJ performed the experiments and analyzed the data. SA wrote the paper. DL designed the experiments, interpreted the data and wrote the paper.
### Conflict of interest statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer NS and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.
|
Neurogenesis, a process of generating functional neurons from neural precursors, occurs throughout life in restricted brain regions such as the subventricular zone (SVZ). During this process, newly generated neurons migrate along the rostral migratory stream to the olfactory bulb to replace granule cells and periglomerular neurons. This neuronal migration is pivotal not only for neuronal plasticity but also for adapted olfactory based behaviors. Perturbation of this highly controlled system by exogenous chemicals has been associated with neurodevelopmental disorders. We reported recently that perinatal exposure to low dose herbicide glufosinate ammonium (GLA), leads to long lasting behavioral defects reminiscent of Autism Spectrum Disorder-like phenotype in the offspring (Laugeray et al., ). Herein, we demonstrate that perinatal exposure to low dose GLA induces alterations in neuroblast proliferation within the SVZ and abnormal migration from the SVZ to the olfactory bulbs. These disturbances are not only concomitant to changes in cell morphology, proliferation and apoptosis, but are also associated with transcriptomic changes. Therefore, we demonstrate for the first time that perinatal exposure to low dose GLA alters SVZ neurogenesis. Jointly with our previous work, the present results provide new evidence on the link between molecular and cellular consequences of early life exposure to the herbicide GLA and the onset of ASD-like phenotype later in life.
## Introduction
Production of neurons is an active ongoing process with 10,000 to 30,000 neurons being produced daily in rodents (Lledo et al., ). While the molecular mechanisms involved are not fully understood, the persistence of neurogenesis in adult subventricular zone (SVZ) is currently well-established (Ghashghaei et al., ). The SVZ newly generated neurons migrate along the rostral migratory stream (RMS) to the olfactory bulb (OB) where they will be used to replace granule cells and periglomerular neurons (Lois and Alvarez-Buylla, ). RMS is characterized by neuroblasts organized into chains continually migrating through an astrocytic tube-like structure called the glial tube (Sun et al., ). This ongoing process is tightly regulated through multiple secretory signals including mitogenic factors (i.e., EGF, Notch) within the SVZ, repulsive factors along the RMS (i.e., Slit, Ephrin, Netrin), and chemoattractive factors (i.e., Reelin, Neuregulin) within the OB (Coskun and Luskin, ; Whitman and Greer, ). In terms of overall brain function, neuroblast migration is of particular relevance to olfactory-based neuronal plasticity and for consequent adapted behavior such as fine-odor discrimination (Gheusi, ).
As for all tightly controlled system, any disturbances could dramatically change the outcome of neuroblast generation or migration (Khodosevich et al., ; Sun et al., ; Young et al., ), and so, would likely lead to neurodevelopmental disorders. In line with this, Autism Spectrum Disorders (ASD) are thought to be associated with alterations in neonatal neurogenesis in the SVZ (Kotagiri et al., ). Indeed, it was shown that neurobehavioral defects concomitant to cellular SVZ abnormalities were induced by perinatal exposure to methotrexate (Seigers et al., ; Hirako et al., ).
Many mediators are involved in the homeostasis of SVZ neurogenesis and neuroblast migration but the neurotransmitter glutamate has been shown to be of particular importance in controlling these processes (Di Giorgi-Gerevini et al., ; Platel et al., , ). The herbicide Glufosinate ammonium (GLA), is the ammonium salt of phosphinothricin (D,L-homoalanin-4-[methyl] phosphinate), an aminoacid structurally related to glutamate and as such is likely to interfere with glutamate signaling. Acute exposure to GLA causes disturbances of the glutamate homeostasis, memory impairments, brain structural modifications, and astrogliosis (Nakaki et al., ; Calas et al., , ; Meme et al., ). Furthermore, we recently showed that GLA has pervasive and harmful effects when administered during the highly sensitive pre- and post-natal periods of brain development (Laugeray et al., ). Neurobehavioral tests revealed significant effects of maternal exposure to GLA on offspring's early reflex development, mother-pup communication and affiliative behaviors later in life. We could also show that perinatal exposure to GLA strongly affected offspring's ability to prefer social olfactory cues over non-social ones. This latter alteration led us to assume that the birth/renewal of SVZ—olfactory bulbs neurons (the main system underlying olfactory-based behaviors in rodents) may be compromised by the herbicide. Interestingly, the expression of two genes involved in the regulation of neuroblast proliferation, migration and apoptosis during brain development (Li et al., ; Wang et al., ), Pten and Peg3 , were dysregulated in GLA-exposed offspring's brain (Laugeray et al., ).
Based on these behavioral and gene expression disturbances, we were interested in the present report in investigating whether GLA-induced-neuropathological conditions may be due to neurogenesis defects and alterations in neuroblast homeostasis by using complementary in vivo and ex-vivo approaches.
## Materials and methods
### Animals and treatments
Seven-week-old female C57Bl/6 mice were purchased from Janvier (Le Genest St Isle, France). All mice were bred and maintained on a 12-h light/dark cycle (lights on from 7:00 a.m. to 7:00 p.m.) with food and water ad libitum in a temperature controlled (21 ± 1°C) room in the animal resource facility. After an acclimatization period of 2 weeks, female mice were mated with male C57Bl6 mice also obtained from Janvier (Le Genest St Isle, France) during 5–6 days. Pregnant mice were then isolated and divided in three experimental groups treated intranasally with either GLA (1 or 0.2 mg/kg; PESTANAL®, analytical standard from Sigma–Aldrich) or saline solution (NaCl 0.9%; 10 μl/30 g mouse) as control. These two doses were not only chosen as they were shown to induce bio-behavioral abnormalities in our previous study (Laugeray et al., ) but also because they are ~5 to 25 times lower than the EPA approved dose (EPA, ). Intranasal exposure was not only chosen as a realistic model of human exposure to volatile toxicants through inhalation but also as it is known that such substances may reach the body via the systemic route without liver interaction (Benson et al., ; Amuzie et al., ). Therefore, many volatile substances may have deleterious consequences, especially in situation of long-lasting or recurrent exposure to low dose pesticides. Dams were treated during pregnancy and lactation periods three times a week from embryonic day 7-10 (E7-10) to postnatal day 15 (PND15) (Figure ). Control animals received a comparable dose of 0.9% saline vehicle. Offspring were maintained in same-sex, litter-mate housed cages with ad libitum access to food and water. All aspects of animal care and experimentation were in accordance with the European Communities Council directive (2010/63/EU). The Ethics committee approved all animal care and use for this study (Approval C45-234-6).
Study experimental design . After 2 weeks acclimation female mice were paired with male mice for 5–6 days to mate. Pregnant mice were treated intranasally with either GLA (0.2 or 1 mg/kg) or saline solution. Dams were treated three times a week from embryonic day 10 (E10) to postnatal day 15 (PND15). At PND5, a group of pups were injected with 100 mg/kg BrdU; 2 h (T0+2 h), 48 h (T0+2 D) or 10 days later (T0+10 D), they were euthanized, and the brains processed for immunohistochemistry. Other pups were euthanized at PND5 and brains processed for culture of SVZ explant, or at PND15 and brains processed for transcriptomic arrays.
### General procedure
General procedure is shown in Figure . From E10 to PND15, we performed preweaning tests to check developmental consequences of perinatal exposure to GLA. With this aim, we investigated neuroblast migration and proliferation by Bromodeoxyuridine (BrdU) labeling in vivo from PND5 to PND15 and ex vivo on Matrigel at PND5. Western Blot, Apoptag®, immunochemistry and transcriptomic arrays were performed on PND15 mice brain.
### Cultures of SVZ explants
Brains from 5 day-old C57bl/6 perinatal treated mice (GLA0.2 or GLA1) or control (CTL) were positioned in ice-cold Leibovitz's L-15 medium (GIBCO). The SVZ from the lateral wall was dissected and sectioned using a vibratome (Leica VT1200S) into pieces 50–300 μm in diameter. The explants were mixed with BD Matrigel Matrix (BD Biosciences) and allowed to solidify in a culture dish (GIBCO). The gel containing the explants was overlaid with 500 μl of Neurobasal medium (GIBCO) containing 10% SVF (Hyclon), B-27 supplement (GIBCO), 0.5 mM L-glutamine (GIBCO), and penicillin–streptomycin antibiotics (GIBCO). Cultures were maintained 3 days in vitro (3DIV) in a humidified, 5% CO2, 37°C incubator (RS Biotech).
### BrdU labeling
Bromodeoxyuridine (BrdU, Sigma), commonly used as a mitotic marker, can inform on neuroblast migration. It is well-described that within the RMS, migrating neuroblasts divide only in the initial portion of the RMS, so BrdU+ cells detected in rostral regions must have proliferated in caudal regions or SVZ, making BrdU labeling a strategy to trace cell migration from the SVZ to the OB. This approach allowed us to evaluate the proliferation and the migration of neuroblasts in the RMS. This kind of analysis had the advantage of looking at the entire population of cells being generated. Thus, cells labeled, although not directly identified, are likely to be primarily precursors of olfactory bulb neurons (Ono et al., ). Migration of cells from the SVZ to the OB involves several processes: the initial decision to exit the SVZ, the migration of cell along the RMS, and the radial migration of individual cells from the end of the RMS into the glomerular layer of the OB (Peretto et al., ).
To evaluate the proliferative and migrating abilities of neuroblasts between the SVZ and the OB, BrdU (100 mg/kg; Figure ) was administered subcutaneous to PND5 GLA-exposed, as well as controls pups. After 2 h, 2 days or 10 days, brains were harvested, fixed, cryoprotected (see below for details) and 14 μm sections were processed (obtained from a cryostat; Leica) using standard immunofluorescence techniques with BrdU monoclonal antibody Alexa Fluor 488 at 1:1000 dilution, counterstained with DAPI (10 μg/ml, Sigma) and coverslipped with fluoromount. Because of the impossibility to determine the number of nuclei within the RMS and the neuroepithelium (NE), we measured integrated intensity (RawIntDen) of BrdU-labeled cells located in the rostral SVZ, the NE and glomerular layer (Glo) with ImageJ® software (NIH) on raw 32 bits imagery according to ImageJ recommendation (Burgess et al., ; McCloy et al., ) (Supplementary Figure ). At least 10 sections per animal and per structure were examined.
### Immunohistochemistry (IHC)
Brains from PND5, 7, and 15 were fixed by immersion in 4% paraformaldehyde (PFA) in 100 mM phosphate buffer, pH 7.4, for 72 h at 4°C and cryoprotected in Tris-buffered saline (TBS) (50 mM Tris-Hcl, pH7.5, 150 mM NaCl) containing 30% sucrose before embedding in Optimum Cutting Temperature (OCT) compound (Sakura Finetek) and frozen in isopentane cooled up to −50°C. Histological coronal sections were mounted onto superfrost+ slides (VWR). Antigen retrieval was performed by incubating the sections in 10 mM sodium citrate solution (pH 9.0) for 30 min in an oven (80°C) (Jiao et al., ) and pre-blocked for IHC for 30 min to 1 h in TBS with 0.3% Triton X-100, 10% normal goat serum and 1% bovine serum albumine (BSA). Sections were incubated overnight at 4°C with the primary antibodies. The following antibodies were used: rabbit polyclonal (pAb) anti-DCX (1:500; Abcam, ab77450), rat monoclonal pAb anti-BrdU (1:500; Abcam, ab6326), mouse monoclonal (mAb) anti-Reelin (1:500; Abcam, ab18570) and species-specific secondary antibodies (Abcam). Sections were counterstained with DAPI (10 μg/ml, Sigma), mounted in Fluoromount-G (SouthernBiotech) and images were captured using a fluorescence microscope (DM6000B; Leica Microsystems) powered by Metamorph software. Measurements were performed with Image J.
In the case of concomitant labeling of BrdU, selected sections from BrdU labeled brains was initiated by the pre-treatment of sections with 10 mM sodium citrate solution (pH 9.0) for 30 min in an oven (80°C). Next, 1 N HCl for 10 min at 4°C, then 2N HCL for 30 min at 37°C.
### Immunocytochemistry
Immunological characterization of cells in chains was performed directly in culture dishes. Samples were fixed with 4% PFA solution in TBS (pH 7.4) for 10 min at room temperature and blocked for 20 min in TBS containing 10% normal goat serum, 1% BSA, and 0.2% Triton X-100. Incubation was with primary monoclonal antibodies (rabbit IgG anti-DCX 1:500 dilution; Abcam), carried out 1 h in humidified box at room temperature (RT). The samples were washed with TBS and incubated with species-specific secondary antibodies (anti-IgG Alexa488 Abcam) for 30 min at RT. Samples washed in TBS and counterstained with DAPI (10 μg/ml, Sigma), mounted in Fluoromount-G (SouthernBiotech) and examined with a fluorescence microscope. Images were captured using a Leica DM6000B microscope powered by Metamorph software. Measurements were performed with Image J.
### TUNEL assay
TUNEL assays were conducted with an ApopTag® Red In Situ Apoptosis Detection Kit (S7165, EMD Millipore) following the indicated protocol. Briefly, sections were treated as indicated above, fixed in 1% PFA, washed in TBS three times, incubated in equilibration buffer (potassium cacodylate; provided in the kit) for 10 min and incubated with terminal deoxynucleotidyl transferase for 60 min at 37°C. After 10 min in stop buffer (provided in the kit), sections were incubated with anti-digoxigenin conjugate overnight at 4°C. After washing in TBS, sections counterstained with DAPI (10 μg/ml, Sigma), mounted in Fluoromount-G (SouthernBiotech) and examined with a fluorescence microscope (Leica).
### Microscopy and imaging
A conventional fluorescence microscope (Leica DM6000B) was used for the rough inspection of stained sections. Images from stained sections were captured with a digital microscope camera (Leica DFC310 FX) and Metamorph software. Selected fluorescently labeled tissues were analyzed with ImageJ software.
### RNA extraction
RNA extraction from PND15 brain murine tissues was carried out using Trizol reagent (Invitrogen, Carlsbad, CA) following the manufacturer's instructions. Quantity and quality of the total RNA were controlled by Nanodrop spectrophotometer (Nanodrop, Wilmington, DE) and Agilent Bioanalyser (Agilent technologies, Palo Alto, CA) in accordance with manufacturer's instructions. Samples A260/A280 absorbance ratio was greater than 1.8 and 28S/18S rRNA ratio greater than 1.5.
### Affymetrix mice exon 1.0 ST array 1.0 and micro-arrays analyses
Gene expression was tested by the Affymetrix Mice Exon 1.0 ST Array (Affymetrix, Santa Clara, CA). A total of 2 μg of RNA from each brain samples was labeled with reagents from Affymetrix according to manufacturers' instructions. Hybridization cocktails containing 5–5.5 μg of fragmented, end-labeled single-stranded cDNA were prepared and hybridized to GeneChip Mouse Exon 1.0 ST arrays. Arrays were washed, stained and scanned on the Affymetrix Fluidics Station and G7 Affymetrix high-resolution scanner (GCOS 1.3). Affymetrix Expression Console Software (version 1.0) was used to perform quality assessment. Raw signals were then transformed into “.CEL” files in GCOS software (Affymetrix, Santa Clara, CA). Probe data were generated using the Robust Multi-chip Average with GC-content Background Correction (GCRMA, ) in Genespring 7 software (Silicon Genetics, Redwood City, CA). This involves background correction, quantile normalization, and summarization of the probe-set values into gene-level expression measurements.
In this study we focus our interest on genes involved in the overall cytoskeleton structure, biogenesis, organization or regulation. Cytoskeleton gene list was obtained from the database GSEA ( ), and injected in Genespring 7 to investigate gene deregulation in our experimental groups (GLA0.2 and GLA1). Thus, differentially regulated cytoskeleton genes were determined using a one-way ANOVA analysis and a Benjamini Hochberg False Discovery Rate (FDR) (< 0.05) method for multiple comparison corrections. This statistic protocol was classically used in the literature for analyzing microarrays expression data (Bittel et al., ; Perche et al., ). Only those genes whose expression changed at least 1.2-fold from the baseline value were selected for downstream analysis. This filter helped maximing the number of genes. PCR quantitative was used to validate the expression arrays, through validation of 45 genes.
Genes that met statistical criteria were analyzed using the DAVID (Database for Annotation, Visualization and Integrated Discovery, Bioinformatics Resources 6.7) for cytoskeleton KEGG pathway visualization.
### Statistical analysis
All data, other than micro array analyses, were analyzed by using parametric procedures: when more than two groups were involved, one-way ANOVA was applied. When appropriate, Dunn's multiple comparison test was performed as post-hoc in order to control the false directory rate. Significance was set at P < 0.05.
Concerning microarrays, differentially regulated cytoskeleton genes were determined using a one-way ANOVA analysis and a Benjamini Hochberg False Discovery Rate (FDR) (< 0.05) method for multiple comparison corrections.
## Results
### Effect of GLA exposure on subventricular zone structure and neuroblast chain migration
With the aim to investigate the effect of GLA exposure on the SVZ, we first carried out morphological measurements of SVZ thickness in PND15 CTL and GLA-exposed pups (Figure ). GLA0.2-exposed offspring displayed a 46.9% increased SVZ thickness compared to control mice (Figure ). This increase was associated with an increased number of DCX-positive neuroblasts in the SVZ (Figure ). Interestingly no difference in SVZ thickness was observed in GLA1-exposed offspring (Figures ). However, we observed that GLA1 treatment promoted ectopic migration of neuroblasts to the surrounding brain regions as, DCX cells were observed in the caudate putamen (Cpu). Then we characterized the effect of GLA ex vivo by using SVZ explants and verified whether neuroblast morphology and/or migration were altered. CTL explants were made of individual long chains of DCX+ cells, extending from the SVZ while explants collected from GLA-treated pups were quite different as chain morphology was completely altered (Figure ). Indeed, SVZ explants coming from GLA0.2 pups displayed chains with a compacted morphology compared to CTL. In explants from GLA1-exposed pups, chains of neuroblast were similar to those of CTL explants in terms of thickness but seemed to be cell-enriched. We quantified this phenomenon by measuring the number of nucleus per chain length. We clearly observed an increase in cell number in explants coming from GLA-exposed pups (Figure ). Moreover, in explants from GLA1 pups, migrating cells displayed abnormal morphology with (1) extension and branching of the growth cone, (2) abnormally long dendrites, and (3) overall loosing of their bipolarity (Supplementary Figure ). These data were corroborated by the measurement of the number of chains and the number of individual cells per surface unit. Indeed, there was a significant decrease of the number of chains/mm for GLA0.2 and GLA1 explants compared to CTL. Further, there was a significant increase in cells migrating individually in the GLA0.2 and GLA1 explants compared to CTL. The effect seemed to be more pronounced at the lowest dose of GLA (Figures ).
Effect of perinatal glufosinate ammonium exposure on neuroblast migration along the SVZ . (A) Diagram of a coronal section at SVZ level (green area). Coronal sections (red square) were stained with DCX (doublecortin; nearly exclusive expression in neuroblasts; green staining) and counterstained with DAPI (nuclear blue staining). Three measurements of thickness were carried for each SVZ (2 SVZ for each coronal section). Three coronal sections were analyzed per animal (Bregma 1.145; 0.745; 0.245); the mean of all values represents one mice. (B) The SVZ thickness of GLA0.2 exposed mice ( n = 5) was significantly increased compared to CTL mice ( n = 5). No difference was found in GLA1 exposed mice ( n = 6). (C) Sections from GLA0.2 mice at PND15 display a more extensive SVZ thickness than CTL. SVZ thickness from GLA1 mice was similar to CTL but with ectopic migration of neuroblasts outside the SVZ (red arrows). Scale bar 100 μm. Each value represents the mean ± SEM ( p < 0.001). lv, lateral ventricle; Cpu, Caudate putamen.
Ex vivo culture of SVZ explant from perinatal GLA exposed mice . Culture on Matrigel of explants from the SVZ of brains from 5 day-old pups treated perinatal with GLA0.2 ( n = 11), GLA1 ( n = 7) or control (CTL, n = 10) was performed for 3 days. We analyzed 6 to 10 explants per animal; the mean of all values represent one mice. (A) The top panel shows a diagram of a coronal section at the SVZ (green area) and the explants were micro-dissected in the party designated by the red square. The photograph on the right of this panel was obtained with an inverted microscope of culture at 3 day in vitro . The morphological appearance of neuroblasts and formed chains were investigated by DCX immunocytology to mark neuroblasts (green) and nuclei with DAPI (blue) (B) . The results show classical bipolar spindle-shaped cells in contact with each other in CTL. Unlike GLA0.2 GLA1 show extensions and ramifications of the growth cone, a loss of bipolarity and abnormal appearance and compact chains. The number of cells per unit distance was measured in formed chains and show a significant increase in this number in GLA0.2 highlighting aggregation (C) . Number of chains formed (D) or individual cells (E) were counted. The number of chains formed in explants exposed mice to GLA0.2 and GLA1 is significantly lower compared to the CTL. Therefore, the number of isolated free cells is significantly higher in exposed mice. These results show the difficulty of neuroblasts to migrate and form chains. Each value is represented by the mean ± SEM ( p < 0.05; p < 0.001). Scale bar 100 μm.
### Perinatal GLA exposure impairs cell proliferation in the sub-ventricular zone and migration toward the olfactory bulb
To verify whether GLA exposure was responsible for changes in the temporal dynamics of neuroblast migration from the SVZ to the OB, BrdU+ cells were determined in the rostral region of the SVZ, the NE and the Glo of the olfactory bulbs at several timepoints post-BrdU injection: 2 h (T0+2 h), 2 days (T0+2 D) or 10 days (T0+10 D) after BrdU injection. Actually, each point gave us information on the location of BrdU-incorporated migrating cells at 5, 7, and 15 days post-natal.
Consistent with previous reports (Luo et al., ), BrdU+ nuclei were found throughout the lateral portions of the SVZ (data not shown) at T0+2 h (PND5). Quantitative analysis of BrdU labeling in the SVZ revealed a significant decrease in cell proliferation in GLA0.2-exposed mice compared to CTL while no effect was observed in GLA1-exposed mice (Figure ). In the NE and Glo, the intensity of BrdU+ cells was similar in the three groups. At T0+2 D (PND7), in CTL group, about half of the BrdU labeled cells reached the NE while a few remained at the rostral SVZ. Interestingly, a third of the BrdU-labeled SVZ level observed at T0+2 h remained labeled at T0+10 D, suggesting that they incorporated BrdU in or near to their final mitosis. At that time, the SVZ-NE-Glo level of BrdU+ dramatically changed in CTL animals (Figure ) reflecting the migration of cells toward and into the olfactory bulb. In GLA0.2-exposed mice, this process appeared to be delayed. Indeed, BrdU levels at T0+2 D remained similar to those found in the SVZ at T0+2 h for GLA0.2 exposed mice. This finding suggested that cells could not leave the SVZ with an adapted timing to reach their final location, the OB. The fact that about two-thirds of the BrdU cells was located in the SVZ at T0+10 D not only strengthened this hypothesis (Figure ) but also indicated that, in GLA0.2-exposed-pups, most of the migrating cells stayed close to their final division location for at least 7–10 days. In GLA1-exposed pups, the temporal dynamics of neuroblast migration was differently altered as, like in CTL pups, BrdU-labeling was significantly decreased in the SVZ at T0+2 D. This result indicated that cells were able to leave the SVZ to reach the OB. However, there was no significant increase of BrdU+ cells in NE or OB at T0+2 D contrary to what we observed in the CTL group (Figures ). To verify whether apoptosis may explain GLA-induced changes in neuroblast migration dynamics, we quantified apoptotic cells within the SVZ at PND15 and we observed a significant increase of apoptotic cells at the highest dose of GLA while no change was noticed at the lowest one (Figures ). Such a phenomenon suggested that apoptotic processes might be involved in GLA1-induced migration defects, and subsequently explain the decreased arrival of BrdU cells in NE at T0+2 D (Figures ), while other mechanisms might be at work in GLA0.2-exposed pups.
BrdU labeling reveals GLA-induced alterations of neuroblast migration in vivo. The top panel shows a diagram of coronal sections at the SVZ and OB, the ventricular walls are colored in green. The BrdU labeling intensity was measured in mouse brain coronal sections of CTL ( n = 3), GLA0.2 ( n = 3) and GLA1 ( n = 3) previously injected with BrdU at PND5 and sacrificed after 2 h (T0+2 h) or 2 days (T0+2 D) or 10 days (T0+10 D). The density of BrdU is measured around the SVZ (A) or around the OB (B) in the neuro-epithelium (NE) (B1) and in the glomerular layer (Glo) (B2) . At T0+2 h, the BrdU density represents the rate of cell proliferation. The results show a decrease of proliferation only in GLA0.2 animals at the SVZ. No proliferation differences were detected at the OB. Measurements performed at 48 h after BrdU injection (T0+2 D) and compared to those performed at T0+2 h indicate cell movements between the SVZ to the OB in CTL animals. At the SVZ, a decrease of intensity is noted in CTL and GLA1 but not in GLA0.2 exposed animals. In the same way, there is an increase of the BrdU density only in CTL OB. We note here that neuroblasts had difficulties to reach the OB in exposed animals to GLA. At T0+10 D, the fluorescence density of BrdU is lowest and similar in all groups in the SVZ. However, at the OB, the fluorescence density of BrdU is greater in GLA0.2 and GLA1 exposed animals than in CTL. Control, white bar chart, GLA0.2, gray bar chart; GLA1, black bar chart. Each value is represented by the mean ± SEM ( p < 0.05; p < 0.01; p < 0.001).
Apoptotic cells labeling in SVZ and in the olfactory bulb . Apoptag® immunostaining of coronal sections of CTL ( n = 3), GLA0.2 ( n = 3) or GLA1 ( n = 3) in the SVZ (A) and in the OB (C) showing a dose effect of GLA in the increase of apoptotic cells in the SVZ of exposed mice (B) . Unlike to SVZ, we show a significant decrease of the number of apoptotic cells in the granular layer (Grl) of exposed mice (D) . No difference was found in neuro-epithelium (NE) and in glomerular layer (Glo). Each value represents the mean ± SEM ( p < 0.05, p < 0.001). Scale bar 100 μm.
At T0+10 D, we observed a drastic decrease of BrdU+ cells in NE and Glo in CTL animals (Figures ). Such a phenomenon is normal as many migrating cells die by apoptosis while only a small percentage of these cells reach the OB and integrates local circuits (Khodosevich et al., ). In GLA0.2-exposed mice, BrdU+ cells continued to increase at T0+10 D in the OB, while in GLA1-exposed pups, the number of BrdU+ cells remained unchanged (Figure ). Concomitantly, apoptosis analysis within the OB revealed a clear decrease of apoptotic cell number in the granular layer in both GLA0.2 and GLA1-exposed pups compared to CTL even if the latter group was minor affected. No changes were observed in NE and Glo (Figures ).
### Anatomical alterations of mitral layer structure after GLA exposure
In control mice, reelin immunohistological staining showed two distinct staining regions clearly expressing reelin in the OB—the glomerular (Glo) and mitral cell layers (MCL) (Figure ). Reelin expression outlined the glomeruli, and was found at high levels in the MCL. In GLA-exposed mice, irrespective of the dose, mitral cells appeared disorganized and the MCL slightly expanded compared to CTL mice (arrows in Figure ), suggesting defects of neuronal lamination (Supplementary Figure ). The other layers appeared undisturbed. In normal OB, the mitral cell bodies were located directly above the Granular cell layer (GrL) and were oriented radially with their primary dendrites projected directly toward the Glo. As shown in Figure , GLA-exposed mice displayed several differences from this normal organization: mitral cell bodies were not located immediately above the GrL and were orientated radially. Furthermore, immunohistology showed an increase of reelin+ cell number both after GLA0.2 and GLA1 perinatal exposure. In addition, analysis of whole brain transcriptomic data highlighted that GLA1-exposed group showed ~40% lower reelin gene expression compared to CTL.
Effect of perinatal glufosinate ammonium exposure on the mitral cells in the OB . (A) In the OB, four coronal sections per mice were stained with Reelin (nearly exclusive expression in mitral cells; red staining) and counterstained with DAPI (nuclear staining; blue staining) [CTL ( n = 3), GLA0.2 ( n = 3), GLA1 ( n = 3)]. (B) The number of mitral cells were determined within the mitral layer. Sections from GLA exposed mice display higher number of mitral cells than CTL. The number of Reelin+ cells/mm of mitral layer is significantly increased in GLA0.2 and GLA1 mice compared to CTL mice. (C) mRNA expression levels of reelin on whole brain measured by qPCR. No difference in expression between GLA0.2 ( n = 8) and CTL ( n = 8), however Reelin is under expressed in GLA1 ( n = 8) compared to the CTL. Each value represents the mean ± SEM ( p < 0.05; p < 0.01). Scale bar 50 μm. Mi, mitral layer; Grl, granular layer; Gl, Glomerular layer.
### Alteration of cytoskeleton regulation after GLA exposure
Since based on our histological analysis showing that GLA exposure altered neuroblast migration and proliferation as well as Pten expression (Laugeray et al., ), we decided to explore cytoskeleton gene expression by transcriptomic.
For this purpose, we focused on gene expression levels of cytoskeleton pathway and selected from GSEA database a list of 494 genes involved in cytoskeleton structure, biogenesis, organization, or regulation. Among this list, one-way ANOVA followed by Benjamini Hochberg multiple testing correction demonstrated that 371 genes were significantly deregulated by GLA parental exposure. Among the 371 genes, we showed that 122 genes were deregulated with at least 1.2-fold change (FC1.2) in GLA0.2 and/or GLA1-exposed pups compared to CTL (Table ). Interestingly, 60 genes (10 Up, 50 Down) were deregulated in GLA0.2-exposed pups whereas 73 genes (37 Up, 36 Down) were affected in GLA1-exposed pups. Only 11 genes, involved in actin cytoskeleton regulation (KEGG pathway, DAVID, Bioinformatics Resources 6.7), were commonly deregulated by both GLA perinatal treatments (Table –lines in gray, and Figure ). However, the deregulation way (up or down-regulated) of these 11 genes were dependent of GLA dose exposure. Indeed, in GLA0.2-exposed pups, 10 out of 11 genes were down-regulated (Table , lines in gray) whereas these same genes were up-regulated in GLA1-exposed pups. Only one gene, Fscn1 , was down-regulated in both conditions.
Cytoskeleton-deregulated genes after perinatal GLA exposure .
FC: Fold Change; Red: Upregulated genes; Blue: downregulated genes.
Signaling pathways involving the cytoskeleton remodeling . Here we show proteins whose expression is deregulated in GLA0.2- (A) and GLA1-exposed pups (B) . Significant number of deregulated genes after perinatal exposure to GLA, involved in cell migration and cytoskeleton dynamics can cause problems at cellular levels of migration and adhesion and even vesicular transport in the cell. Figure from David Software database and adapted by S. Mortaud. Differentially regulated cytoskeleton genes were determined using a one-way ANOVA analysis and a Benjamini Hochberg False Discovery Rate (FDR) (< 0.05) method for multiple comparison corrections.
## Discussion
Although, it is now well-known that GLA is structurally related to glutamate and clearly neurotoxic, its putative aversive effects on neuroblast homeostasis (for which glutamate is of crucial value) remained unexplored. Our recent study suggested that behavioral alterations induced by perinatal exposure to GLA could be related to neurogenesis disturbances due to altered brain expression of relevant genes, like Pten , well-known to be importantly involved in this process (Laugeray et al., ). Herein, we tested this hypothesis thanks to both in vivo and ex vivo analyses.
### SVZ as a potential open window to GLA neurotoxicity
It is now well-established that the neurotransmitters GABA and glutamate are of crucial importance during neuronal migration from the SVZ to the OB and, as such, are likely to contribute to the pathogenesis of neuronal migration disorders (Platel et al., ). Here, we assumed that GLA, as an analog of glutamate, can affect cell proliferation and neuroblast migration by interfering with glutamate signaling, and thus with SVZ functions. In agreement with this hypothesis glutamate homeostasis is known to be importantly involved in RMS migration processes (Platel et al., ). However further experiments are needed to support this statement. We were able to show that exposure to low dose GLA (GLA0.2) led to neuroblast accumulation in the SVZ in vivo at PDN15. Such an alteration has already been reported in some ASD patients (Wegiel et al., ). Together with our data, these findings strengthen our hypothesis of a link between GLA exposure, abnormal neurogenesis in the SVZ and ASD. Interestingly, at the highest dose, GLA did not affect SVZ thickness but, rather, led to ectopic migration of neuroblasts outside the SVZ. These data suggested that the SVZ, lining the walls of the lateral ventricles, is sensitive to exogenous toxicants, and subsequently may constitute the real “window of the brain” as proposed by many authors (Gross and Weindl, ; Moyse et al., ; Joly et al., ; Lin et al., ). Such an assumption is reinforced by a recent study showing that perinatal exposure to another environmental toxicant, methotrexate, also led to SVZ alterations (Hirako et al., ). These aversive effects of exogenous compounds on the SVZ could be inherent to its intrinsic structure characterized by permeable fenestrated capillaries and thus a lack of endothelial blood–brain-barrier (BBB) (Johnson and Gross, ; Tavazoie et al., ). Consequently, the SVZ stem cell niche is in the front line to respond to systemic xenobiotics and is thus likely to be extremely sensitive to environmental toxicants.
### Impaired proliferation, migration, and apoptotic processes after perinatal exposure to GLA
Activation of glutamate receptors causes transient increases in intracellular Ca promoting neuronal migration by acting on the cytoskeleton protein regulators (Luhmann et al., ). Even if our data showed that perinatal exposure to two doses of GLA negatively impacts neuroblast homeostasis within the SVZ-OB system, the present results also show that GLA is able to induce two different patterns of SVZ alterations suggesting that several targets might be impacted depending on the dose. Indeed, cell morphology, considered to be pivotal in the process of migration (Luskin, ; Alvarez-Buylla and Garcia-Verdugo, ), is differentially affected by GLA depending on the dose as illustrated by ex vivo SVZ explants where GLA0.2-exposed neuroblasts display highly compacted chains whereas neuroblasts from GLA1-exposed explants seem less altered in their morphology (although still different from CTL ones). In support of this, our in vivo BrdU labeling experiments also show differential effects of GLA depending on the dose. In agreement with the literature, we observe in CTL offspring that, BrdU+ cells go out from the SVZ and migrate through the RMS to finally reach the OB 2 days later. In GLA0.2-exposed animals, BrdU+ neuroblasts are able to migrate from the SVZ to the OB but our results indicate that changes in SVZ cell proliferation and defects in migration processes significantly delay the temporal dynamics of neuroblast migration On the contrary, no alteration of proliferation is observed in GLA1-exposed offspring in vivo while many cells, despite being able to leave the SVZ as well as CTL ones, are found in an abnormal ectopic location, the caudate putamen (Cpu), a situation never seen in CTL neuroblasts. These results are consistent with the SVZ enlargement observed in vivo and the compacted aspect of chains in SVZ explants, only seen at the lowest dose. Therefore, it is of importance to notice that, for some parameters, the doses have the same effect while for others the effect is qualitatively different. Moreover, we find the OB in GLA0.2 and GLA1-exposed mice to have an apparently normal size and structure. This point is relevant because the number of cells reaching the OB is thought to regulate the size of the bulbs (Gheusi, ). At first glance, the decreased number of TUNEL cells in GLA0.2 and GLA1-exposed mice seem at odds with the lower increase in BrdU cells and the normal bulb morphology. These data are however in line with previous study reporting neuronal migration disorders following in utero exposure to several environmental factors in humans and in animal models (Wisniewski et al., ; Miller, ; Gressens et al., ; Shinmura et al., ).
### Reelin and cytoskeleton alterations as the cause of GLA-induced neurogenesis defects
Reelin expression is essential for neuronal migration in the developing brain, acting as a “detachment signal” of postnatal neuroblast from the RMS (Simo et al., ). In GLA-exposed mice, the OB expression pattern of reelin seem to be altered irrespective of the dose. Such findings are in accordance with our in vivo and ex vivo experiments showing disturbances of neuroblast migration. The reelin pathway corrects the migration of early generated interneurons within the olfactory bulb. This function is a prerequisite for correct OB lamination (Hack et al., ; Hellwig et al., ). GLA-induced alterations of the laminar organization within the OB are clearly consistent with this. In support of this, reeler mice have been shown to exhibit ectopic accumulation of neuroblasts in the RMS, failing to transit from tangential migration in the RMS to radial migration in the OB (Ayala et al., ; Sun et al., ). Interestingly, complex interactions between reeler genotype and early exposure to environmental toxicants have already been demonstrated, (Keller and Persico, ; Laviola et al., ; Persico and Bourgeron, ; Mullen et al., ).
GLA-induced neurogenesis defects was also reinforced by observing our transcriptomic analyses on PND15 brains perinatally exposed to GLA. Indeed, expression pattern of a series of genes regulating the cytoskeleton, cell proliferation and cell migration were affected by GLA exposure. Interestingly, GLA0.2 and GLA1 groups present a different deregulated pattern, and no dose effect was observed. This result was in accordance with our ex vivo data showing that a different cellular morphology observed in explants experiments treated with GLA0.2 or GLA1. Therefore, GLA0.2 and GLA1 specifically induced gene deregulation explaining the compacted neuroblasts chains morphology and abnormal morphology with extension and branching of the growth cone and abnormally long dendrites, respectively. However, 11 genes were deregulated in both GLA0.2 and GLA1 treated groups. Interestingly, these commonly deregulated genes included pivotal representatives for the major processes controlling cell migration such as cytoskeleton rearrangements ( Rac, Rho, Rock ), cell adhesion ( Arp2/3, Pfn, Actn, Erm ), and chemotactic signalization ( Rtk, Itg, Ras ) (Khodosevich et al., ). ARP complex Arpc3 and Arpc4 which nucleates actin filament growth from the minus end and allows rapid elongation at the plus end (Alberts et al., ), PFN ( Pfn1 and Fscn1 ) promoting actin migration/proliferation of non-muscle cells (Wang et al., ), MLC ( My19) regulating cytoskeleton contraction (Saban et al., ), GF ( FgF1 ) regulating cytoskeletal organization collagen contraction (Ding et al., ), GPCR/Chemokine ( Chrm4 and CCl3 ) regulating chemotaxis cell migration and actin cytoskeleton moving machinery (Yang et al., ), and Ras/Rac system ( Arhgdib ) which controls the assembly and disassembly of the actin cytoskeleton in response to extracellular signals (Ory et al., ). RhoGTPase family is one of the major regulators of cytoskeletal properties and plays essential functions in cerebral cortex development. These functions are known to be highly associated with glutamate homeostasis to regulate neuroblast proliferation and migration (Di Giorgi-Gerevini et al., ; Platel et al., ). Here we show, that perinatal organophosphate GLA exposure affect cell migration modulating not only Reelin activity, but also a series of other genes involved in cytoskeleton regulation, thus potentially contributing to the neurodevelopmental basis of autism-like behavior.
### On the unusual “dose-dependent” effects of perinatal exposure to GLA
Most of the results presented here indicate that perinatal exposure to GLA may have different effects depending on the dose to which individuals are exposed. In support of this, it is of interest to see that the lowest dose seems to be more harmful in regards of some parameters we studied here. Such results clearly challenge the prevailing dogma that “the dose makes the poison” as the reported effects show a qualitatively different dose-dependent response. One explanation for this could be related to the structural analogy of GLA to the excitatory neurotransmitter glutamate. Indeed, GLA may have a glutamate-like effect at low dosage whereas, at higher dosage, it may induce detrimental effects on neurons irrespective of its signaling to glutamate receptors. For instance, one can assume that, at high dosage, GLA may be added to α/β tubulin heterodimers instead of glutamate and consequently have disturbing effects on polyglutamylation processes. Some experiments are currently in progress in order to test such a hypothesis.
For a number of years now, there have been reports showing that environmental toxicants can have effects at low dosage and that these cannot be predicted and extrapolated from effects at higher doses (Vandenberg et al., ). The present data strongly suggest that GLA is likely to be one of these toxicant. Our previous study (Laugeray et al., ) considerably strengthens this assumption as some neurodevelopmental outputs were commonly affected by the two doses of GLA while for others, it was not the case. This was especially true for anatomical abnormalities indicating macrocephaly at 1 mg/kg and microcephaly at 0.2 mg/kg. Interestingly, such changes were concomitant with reduced Pten mRNA levels in the brain of GLA1-exposed pups matching the macrocephaly in the same mice, but perinatal exposure to the lower dose had the opposite effect on both Pten expression and brain size (microcephaly). Such a non-classical “dose dependent” effects were also observed on bio-behavioral parameters (Laugeray et al., ). It is therefore essential to build up our knowledge of possible harmful effects of low-level perinatal exposure to pesticides. This is a relevant and topical issue as many governmental reports have noted that early exposure (pre and postnatal) to low or very low doses of pesticides is not usually covered by the tests required for regulatory approval, and therefore that it is impossible to estimate such adverse effects (Bonnefoy, ; Watts, ).
In summary, our work demonstrates for the first time the deleterious effect of perinatal exposure to GLA results in abnormal brain development, both at the cellular and molecular levels, providing a putative structural explanation for GLA-induced ASD-like phenotypes in mice. Our data not only identify the SVZ as a novel target for environmental toxicants, in particular in case of early exposure to low doses but also pave the way for unraveling the molecular events that orchestrate the effect of GLA on cell cytoskeleton.
## Author contributions
Research was designed by AH and SM. Research was realized by AH, AL, JF, OR, and OP. Research was analyzed and discussed by AH, AM, JP, OP, and SM. Paper was written by AH, CM, VQ, OP, and SM.
## Funding
This work was supported by the French National Research Agency – ANR (CESA-10-007 – NEUROPEST), and Region Centre (Doctoral fellowship to Ameziane Herzine).
### Conflict of interest statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
The thyroid hormones (TH) triiodothyronine (T3) and its prohormone thyroxine (T4) are crucial for retinal development and function, and increasing evidence points at TH dysregulation as a cause for retinal degenerative diseases. Thus, precise regulation of retinal TH supply is required for proper retinal function, but knowledge on these mechanisms is still fragmentary. Several transmembrane transporters have been described as key regulators of TH availability in target tissues of which the monocarboxylate transporter 8 (MCT8), a high affinity transporter for T4 and T3, plays an essential role in the central nervous system. Moreover, in the embryonic chicken retina, MCT8 is highly expressed, but the postnatal availability of MCT8 in the mammalian retina was not reported to date. In the present study, spatiotemporal retinal MCT8 availability was examined in mice of different age. For this purpose, we quantified expression levels of Mct8 via Real-Time Reverse-Transcriptase PCR in mouse eyecups (C57BL/6) of juvenile and adult age groups. Additionally, age-dependent MCT8 protein levels were quantified via Western blotting and localized via immunofluorescence confocal microscopy. While no difference in Mct8 expression levels could be detected between age groups, MCT8 protein levels in juvenile animals were about two times higher than in adult animals based on Western blot analyses. Immunohistochemical analyses showed that MCT8 immunoreactivity in the eyecup was restricted to the retina and the retinal pigment epithelium. In juvenile mice, MCT8 was broadly observed along the apical membrane of the retinal pigment epithelium, tightly surrounding photoreceptor outer segments. Distinct immunopositive staining was also detected in the inner nuclear layer and the ganglion cell layer. However, in adult specimens, immunoreactivity visibly declined in all layers, which was in line with Western blot analyses. Since MCT8 was abundantly present in juvenile and about twofold lower in adult retinae, our findings suggest a pivotal role of MCT8 especially during postnatal maturation. The present study provides novel insights into age-dependent retinal TH supply, which might help to understand different aspects regarding retinal development, function, and disorders.
## Introduction
Thyroid hormones, in particular 3,5,3′-triiodothyronine (T3) and its precursor thyroxine (T4), are widely associated with somatic and neuronal development, and metabolism. T3 is a ligand for different THRs, nuclear receptors regulating the expression of a wide range of target genes ( ; ; ). The availability of T3 is mostly regulated by intracellular conversion of T4 by specialized enzymes, deiodinases type 1 and 2 (D1, D2). Inactivation of TH is catalyzed by deiodinases type 1 and 3 (D1, D3; ). In the mammalian retina, a strict TH regulation was found crucial for photoreceptor development and function ( ; ; ; ).
Most mammals possess two types of photoreceptors, rods for dim light vision, and cones for daylight, i.e., color vision. While rods express only one type of opsin (rhodopsin), a light-sensitive molecule enabling phototransduction, mammalian cones usually express short (S) and medium (M) wavelength sensitive opsins, and some primates including humans express an additional long (L) wavelength sensitive opsin ( ). In cones, a particular THR isoform, THRβ2, is expressed and was identified as a crucial upstream factor for M/L-opsin expression ( ; ). While cones first express S-opsins in early postnatal stage ( ), binding of T3 to THRβ2 is required for the onset of M/L-opsin expression, and simultaneous suppression of S-opsin ( ; ; ; ). Consistent with this mechanism, hypothyroidism leads to reduced M/L-opsin expression even in adult age, which can be restored by TH treatment in rodent models as well as human ( ; ). On the other hand, inactivation of TH by D3 was shown crucial for cone development. In D3 knockout mice, about 80% of cones are lost by neonatal cell death ( ). Additional deletion of Thrβ2 leads to cone protection, suggesting that excessive TH signaling via THRβ2 has a deleterious effect on cones. Similar findings were reported in mouse models of retinal dystrophy where TH suppression ( ) and D2 inhibition ( ) have a protective function on cones, while hyperthyroidism leads to cone degeneration ( ). Moreover, in a prospective study, high serum levels of the prohormone T4 positively correlated with a higher prevalence for age-related macular degeneration, while elevated T3 had no impact ( ). The underlying mechanism for TH induced cone degeneration is not known, but TH dysregulation has already been linked to reduced renewal of photoreceptor OS and retinal arteriolar narrowing ( ; ), both processes associated with the pathogenesis of different degenerative diseases. Even in rods, involvement of TH signaling has been associated with development and function ( ; ).
While TH were shown to be essential for photoreceptor development and function, the mechanisms regulating retinal TH supply are still poorly understood. Nourishment of photoreceptors is regulated through the RPE, a monolayer of pigmented cells with its apical membrane lying adjacent to photoreceptor OS. The photoreceptor OS are tightly surrounded by microvilli emerging from the RPE, maintaining a complex of close interaction ( ). The RPE builds a part of the outer BRB, and TH, as many other organic compounds, are transported via transmembrane transporters from the choriocapillaris into photoreceptors and vice versa. Unlike photoreceptors, interneurons (bipolar cells, horizontal cells, and amacrine cells), and ganglion cells located in the inner retinal layers are not connected to the RPE. Nourishment of these layers is thus facilitated through retinal capillaries found in the layers between the OPL and the GCL. Endothelial cells, pericytes, and Müller cells (the most prominent retinal glial cells) build up the inner BRB which regulates nourishment of inner retinal cells ( ).
Several transporters have been described which facilitate the influx and/or efflux of TH across plasma membranes as primary or secondary substrates in diverse tissues (reviewed in ; ). The MCT8 (encoded by Slc16a2 , hereafter: Mct8 ) is a high affinity TH transporter expressed in several tissues ( ; ; ; ). MCT8 transports T4 as well as T3, with a higher affinity for T3 ( ), and is highly expressed in diverse tissues ( ). In the murine brain, Mct8 is highly expressed during the first postnatal weeks ( ), where it possesses a critical role in TH uptake into the brain ( ; ), especially into neurons ( ). TH signaling is required for proper brain development and hypothyroidism in the critical phase of postnatal brain maturation in rodents (e.g., neuronal differentiation, dendritic branching, axon growth and synaptogenesis) leads to drastically diminished neuronal connectivity in rodents ( ). In the embryonic chicken retina, MCT8 mRNA is widely expressed ( ; ). To our knowledge no data is available regarding MCT8 availability in the postnatal vertebrate eye, although MCT8 is one promising candidate for regulation of retinal TH supply to maintain proper retinal maturation and function like in the brain. In order to contribute to a deeper understanding of retinal TH supply, we quantified retinal Mct8 mRNA expression by means of quantitative reverse transcriptase-mediated polymerase chain reaction (qRT-PCR). Protein levels and localization of MCT8 were further investigated by Western blotting and immunohistochemistry, respectively. We focused on juvenile and adult life stages where TH dysregulation has major impact on cone opsin expression and photoreceptor viability.
## Materials and Methods
### Animals
Male mice (C57BL/6) aged 14 days (P14; juvenile), 21 days (P21; juvenile), 28 days (P28; juvenile), 24 weeks (6M; adult), and 24 months (24M; old) were obtained from the Department of Developmental Biology (University of Duisburg-Essen, Essen, Germany) and the Central Animal Laboratory, University Hospital Essen. The three juvenile stages are representative for the onset of vision (P14; ; ), terminal differentiation of retinal structures (P21, e.g., synaptogenesis, pruning, and maturation of vascular pattern; ; ), and a transitional stage between juvenile and adult mice (P28; i.e., onset of puberty; ). 6 and 24M mice represent adult stages, from which 6M retinae can be considered fully mature, and 24M specimen represent aging retinae ( ).
Mice were housed in standard macrolone cages and were fed ad libitum with commercial food pellets. Especially in the 24M group, the animals were checked for obvious degenerative changes of the eye, such as cataracts. For all experimental procedures mice were deeply anesthetized with isoflurane (AbbVie, Wiesbaden, Germany), and subsequently sacrificed by cervical dislocation. Maintenance and all treatments of the animals reported here are in agreement with the North Rhine-Westphalia State Environment Agency.
### RNA Preparation and qRT-PCR
Freshly isolated eyes ( n = 6 per age group) were incised at the corneal rim to remove the cornea, lens, and vitreous. The eyecup was transferred to stabilization reagent RNA later (Qiagen, Hilden, Germany), incubated 24 h at room temperature, as recommended, and stored at -20°C for later expression analyses. Additionally, temporal neocortex including the cerebral membrane was isolated and conserved in RNA later .
For RNA extraction, the eyecups were immersed into lysis buffer (Buffer RLT, Qiagen, Hilden, Germany) and homogenized with a TissueLyser II (Qiagen). Total RNA was extracted by RNeasy Mini Kit (Qiagen) according to manufacturer’s instructions. Complementary DNA (cDNA) was synthesized using AMV Reverse Transcriptase (Promega; catalog no. A3500) primed by random oligomers. We reverse-transcribed 0.25 μg RNA in 20-μl reactions, as recommended by the manufacturer’s instructions, albeit with the addition of 10 μg bovine serum albumin per reaction. We selected three reference genes, Hprt, Atp2b4 , and Mapk1 , in order to obtain normalized mRNA expression data. All these genes showed stable expression levels in previous studies of mammalian retina across different postnatal life stages ( ; ; ). Furthermore, these genes encode proteins with different functional roles and subcellular locations. The validity of this concept was also supported by cross-correlation of the reference genes’ RT-PCR Ct values in this study, which did not reveal any age-related dependencies (results not shown). DNA primer pairs (metabion international, Planegg, Germany) used for qRT-PCR were: 5′-TCC CAT TGC ATT TGA GCT G-3′ and 5′-GGG ACA CCC GCA AAG TAG A-3′ ( Mct8 ), 5′-GCT GGT GAA AAG GAC CTC TC-3′ and 5′-CAA GGG CAT ATC CAA CAA CA-3′ ( Hprt ), 5′-AAC TCA GTG CGC AAG TCC AT-3′ and 5′-TCC TTC CTT GTT CAG GAT TCG-3′ ( Atp2b4 ), 5′-CTC TGG CCC ACC CAT ACC T-3′ and 5′-AAG TCG TCC AAC TCC ATG TCA-3 ( Mapk1 ). Real-time qRT-PCR was performed using SybrGreen (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. The thermal cycler (iQ Cycler, Bio-Rad, Hercules, CA, USA) was programmed to initial 95°C for 90 s and 40 cycles comprising 95°C for 45 s, 55°C for 25 s, and 72°C for 30 s. Each assay, defined by sample/primer combination, was performed in triplicate of independent qPCR reactions, and the Ct values were averaged before normalization. Normalized expression levels were obtained according to the ΔΔ C t method, accounting for primer-specific amplification efficiencies ( ), which were determined from a four-step dilution series of an arbitrary cDNA sample.
The data was kept in C t scale (log ) for a statistical analysis using one-way ANOVA followed by Tukey’s multiple comparison test using GraphPad Prism (GraphPad Software Inc., San Diego, CA, USA), differences being considered significant when p < 0.05.
### Antibodies
For the detection of MCT8 protein we used a commercial rabbit polyclonal IgG (HPA003353, Sigma-Aldrich, Taufkirchen, Germany; dilution Western Blot 1:1000, immunohistochemistry 1:500). For Western blotting, we used a rabbit polyclonal H3 IgG (ab1791, Abcam, Cambridge, UK; dilution 1:5000) as internal loading control. A horseradish peroxidase-conjugated swine anti-rabbit IgG (P0217, Dako, Hamburg, Germany; dilution 1:10000) was used as a secondary antibody. For immunohistological analyses, we labeled S-opsin with a goat polyclonal IgG (sc-14363, Santa Cruz Biotechnology, Dallas, TX, USA; dilution: 1:500), and Glucose transporter 1 (GLUT1) with a mouse monoclonal antibody (ab40084, Abcam; dilution 1:100), to enable a better localization of MCT8 immunostaining in retinal cryosections. To visualize MCT8, GLUT1, and S-opsin in each section, sections were incubated with a mixture of the three primary antibodies, and three secondary antibodies with well separated excitation and emission spectra were chosen. Binding sites of the anti-MCT8 antibody were revealed by Alexa Fluor 647 donkey anti-rabbit (false colored in red), anti-S-opsin by Alexa Fluor 568 donkey anti-goat (false colored in cyan), and anti-GLUT1 antibodies by Alexa Fluor 488 donkey anti-mouse (false colored in green), at a dilution of 1:500 (Abcam).
### Western Blotting
After enucleation, eyecups were transferred to ice cold lysis buffer (20 mM Tris pH 8.0, 1 mM EDTA, 5 mM MgCl , 1 mM DTT, 1% Triton X-100, and 1% protease inhibitor) and homogenized using a Potter-Elvehjem homogenizer. Samples ( n = 4 per age group) were centrifuged (11 min, 13,000 rpm, 4°C) and the protein concentration in the supernatants was determined by a biuret protein assay (Roti-Quant Universal, Carl Roth, Karlsruhe, Germany). For protein separation, 1x Laemmli buffer was added to the samples, vortexed for 5 s and put back on ice. Proteins were separated on an 8% SDS-polyacrylamide gel by applying 15 μg total protein of each sample, and transferred to a polyvinylidene difluoride (PVDF) membrane.
Membranes were blocked with blocking buffer (5% skimmed milk in 0.1% TBS-T) for 1 h at room temperature and incubated overnight at 4°C with rabbit anti-MCT8 antibody and rabbit anti-H3 antibody (internal loading control) diluted in the same blocking buffer (see antibody section for details). We selected H3 for normalization, since other familiar loading controls such as tubulin, actin, or GAPDH are not expressed uniformly across age groups ( ; ). H3 is expressed at stable levels in the postnatal mouse retina ( ), and as a nuclear protein, it additionally allowed validation of efficient dissolution of plasma membranes. Membranes were washed three times in 0.1% TBS-T and incubated with horseradish peroxidase-conjugated swine anti-rabbit secondary antibody for 1 h at room temperature. Binding sites were visualized by means of a chemiluminescence detection kit (AceGlow, VWR International, Langenfeld, Germany). The signal densities obtained for each band were quantified with Bio-1D advanced (Vers. 12.11, Vilber Lourmat, Eberhardzell, Germany), and relative density values were calculated with the first sample set as reference. Relative MCT8 expression was then normalized with relative H3 of respective samples, and statistically analyzed with one-way ANOVA followed by Tukey’s multiple comparison test using GraphPad Prism (GraphPad Software Inc., San Diego, CA, USA).
### Immunohistochemical Analysis
Whole eyes were fixed in 4% paraformaldehyde in 0.1 M phosphate buffer (PB; pH 7.4) at 4°C for 3 h. For further processing, cornea, lens, and vitreous were removed carefully. Then, eyecups were immersed in a successive series of 10, 20, and 30% sucrose in PB for cryoprotection, mounted in Tissue-Tek O.C.T. (Sakura Finetek Germany, Staufen, Germany), and snap-frozen in 2-methylbutane cooled in liquid nitrogen. Frozen sections were cut (sagittal, 16 μm) using a cryostat (CM3000, Leica Biosystems), mounted on silanized glass slides (SuperFrost Ultra Plus, Thermo Fisher Scientific, Carlsbad, CA, USA) and air-dried for 24 h. The sections were washed in PB, and blocked with 10% donkey serum (Sigma-Aldrich, Taufkirchen, Germany) in PB with 1% BSA and 0.5% Triton X-100 (Carl Roth) for 1 h at room temperature. Incubation with primary antibodies (mixture of anti-MCT8, anti-S-opsin, and anti-GLUT1; see antibody section for details) was performed overnight at 4°C in 3% donkey serum with 1% BSA and 0.5% Triton X-100 in PB. After washing in PB, Alexa Fluor conjugated secondary antibodies, appropriately diluted in the same buffer as used for primary antibody, were incubated for 1 h at room temperature. Sections were washed in PB and coverslipped with Roti-Mount FluorCare DAPI (Carl Roth) for supplemental nuclei staining.
Structures labeled by immunofluorescence were visualized using a confocal laser scanning microscope (Zeiss ELYRA PS.1 super resolution microscope combined with a LSM710) equipped with a 405 nm diode laser, an Argon Multiline 458/488/568 and a 633 nm Helium-Neon Laser. Micrographs were captured by means of confocal software (ZEN system 2012 Black Edition, Zeiss), and adjusted for brightness and contrast only.
## Results
### Mct8 Expression
We applied qRT-PCR in order to obtain an overview of Mct8 mRNA expression in the mouse eyecup at different ages. Statistical analysis of Mct8 levels revealed no significant differences between the tested age groups (one-way ANOVA, F = 0.19, p = 0.94, Figure ), while the age-matched, inter-individual variation was about 1.4-fold. This result was independent of the normalization procedure since normalization to total RNA input produced the same qualitative test results as did normalization to reference genes. Thus, qRT-PCR results suggest that Mct8 is expressed at almost equal levels in murine eyecups of different age.
Expression levels of Mct8 mRNA: quantitative reverse transcriptase-mediated polymerase chain reaction (qRT-PCR) was performed with eyecup homogenates at varying age (P14, P21, P28, 6M, and 24M). Mct8 mRNA levels are given as fold-changes relative to P14 expression levels, arbitrarily set to 1. Note that normalization and statistical analysis were performed in the C t scale (log ), as described in “Materials and Methods.” Data are presented as fold changes, median ±25%/75% quartile and maximum/minimum value ( n = 6 per age group).
### Western Blot Analysis
To assess possible differences in MCT8 protein levels, we quantified protein levels in three experimental groups representing three life stages (P14: juvenile; 6M: adult; 24M: old), by means of Western blotting with eyecup homogenates (15 μg protein). The anti-MCT8 antibody recognized a single band in all samples with an apparent size of 60 kDa, which is consistent with the predicted protein size and previous reports ( ; ) ( Figure ; Supplementary Figure for complete gel). Anti-H3 antibody, used for normalization, recognized a single band in each sample with an apparent size of 15 kDa (predicted: 15 kDa) ( Figure ; Supplementary Figure ).
Western blot analysis of MCT8: Western blotting was performed with eyecup homogenates of P14, 6M, and 24M mice with an antibody against MCT8 (predicted: 60 kDa) and an antibody against H3 (predicted: 15 kDa) as internal loading control. Each blot (A) corresponds to the age groups shown in the barplot below. (B) . Each band is representative for one age group and was taken from the same blot, respectively. Barplots are depicted relative to P14 MCT8 protein levels, arbitrarily set to 1. Specific bands for MCT8 and H3 were detected in the Western blots, with significant differences between P14 and 6M/24M. Data are presented as mean ± SD ( n = 4 per age group). p < 0.005; p < 0.05.
Mean normalized MCT8 protein levels were more than twofold higher in P14 compared to 6 and 24M, and revealed significant differences between the juvenile and both adult age groups in the statistical analysis (one-way ANOVA: F = 9.71, p = 0.0057; Tukey’s multiple comparison test: P14 vs. 6 M: p < 0.05; P14 vs. 24M: p < 0.005; Figure ). In contrast, no significant difference was found between 6 and 24M protein levels.
Liver homogenates were used as positive control for the MCT8-antibody, which revealed a broad band with an apparent size of 60 kDa (Supplementary Figure ).
### Immunohistochemical Localization of MCT8 in the Retina
We performed immunohistological stainings on eyecup cryosections of five age groups (P14, P21, P28, 6M, and 24M) to determine MCT8 localization in the mouse retina throughout different postnatal life stages. GLUT1 and DAPI were co-stained to control for normal retinal morphology. Moreover, GLUT1 staining was raised to identify the basolateral and apical RPE membrane, since GLUT1 is localized at both membranes and RPE cells can be very thin especially in juvenile animals.
The overall retinal morphology was intact in all age groups, showing the typical layering known from mammalian retinae (RPE; OS; ONL; OPL; INL; IPL; GCL). However, MCT8 immunoreactivity showed apparent age-dependent changes: In all age groups, MCT8 immunoreactivity was observed in the RPE, and around nuclei of the INL and the GCL in the whole retina, but the intensity visibly declined from juvenile to adult stages ( Figures – ). Detailed inspection of MCT8 immunoreactivity in the RPE revealed that MCT8 is predominantly located at the apical membrane (RPE.ap) with its microvilli, while the basolateral membrane (RPE.ba) showed only faint and irregularly distributed immunopositive signal in all age groups ( Figure ; white arrows). At the RPE.ap of P14 and P21 specimens, MCT8 immunoreactivity reached deep into photoreceptor OS indicated by S-opsin staining, while a slight decline in immunoreactivity was already visible in the RPE.ap of P28 specimens ( Figure ). The decline of MCT8 was even more pronounced in the 6 and 24M retinae, where immunoreactivity in the RPE.ap was restricted to the surface except of short and irregular branches ( Figure ). In the INL and GCL, MCT8 was found at plasma membranes of cell bodies, indicated by nuclei staining ( Figures and ). While in P14, a vast majority of cell bodies showed positive MCT8 staining, the immunoreactivity gradually declined, and was only scattered around single nuclei in the INL and GCL of adult animals.
Overview of MCT8 immunoreactivity in the mouse retina: Eyecup cryosections of five age groups (P14, P21, P28, 6M, 24M) colabeled with antibodies against MCT8 (red), GLUT1 (green), S-opsin (cyan) are shown. Nuclei were counterstained with DAPI (blue). Antibody binding sites were revealed using Alexa Fluor conjugated secondary antibodies and visualized with a Zeiss Elyra PS.1 combined with a LSM710 confocal microscope.
MCT8 immunoreactivity in the mouse RPE: Eyecup cryosections colabeled with antibodies against MCT8 (red), GLUT1 (green), S-opsin (cyan) are shown. Nuclei were counterstained with DAPI (blue). Antibody binding sites were revealed using Alexa Fluor conjugated secondary antibodies and visualized with a Zeiss Elyra PS.1 combined with a LSM710 confocal microscope. (A) MCT8 immunoreactivity in the RPE apical membrane (RPE.ap) and basolateral membrane (RPE.ba). (B) Merged images are shown for better localization of MCT8 in the respective retinal layers. Predominant staining was observed in RPE.ap, whereas RPE.ba showed only faint and irregular immunoreactivity throughout all age groups (white arrows). The signal declined gradually in the apical membrane, with the strongest signal observed in P14 and P21, and only scarcely distributed in 24M. In OS and ONL no specific staining was observed in all age groups.
MCT8 immunoreactivity in the mouse INL: Eyecup cryosections colabeled with antibodies against MCT8 (red) and GLUT1 (green) are shown. Nuclei were counterstained with DAPI (blue). Antibody binding sites were revealed using Alexa Fluor conjugated secondary antibodies and visualized with a Zeiss Elyra PS.1 combined with a LSM710 confocal microscope. (A) MCT8 immunoreactivity in the INL. (B) Merged images are shown for better localization of MCT8 in the respective retinal layers. MCT8 was localized in plasma membranes of cell bodies located in the INL. The signal intensities gradually declined, with the strongest signal observed in P14, and only faintly detectable in 24M. In the IPL no specific staining was observed in all age groups.
MCT8 immunoreactivity in the mouse GCL: Eyecup cryosections colabeled with antibodies against MCT8 (red) and GLUT1 (green) are shown. Nuclei were counterstained with DAPI (blue). Antibody binding sites were revealed using Alexa Fluor conjugated secondary antibodies and visualized with a Zeiss Elyra PS.1 combined with a LSM710 confocal microscope. (A) MCT8 immunoreactivity in the GCL. (B) Merged images are shown for better localization of MCT8 in the respective retinal layers. MCT8 was localized in plasma membranes of cell bodies located in the GCL. The signal intensities gradually declined with the strongest signal observed in P14, and only detectable around single nuclei in 24M. In the IPL no specific staining was observed in all age groups.
Overall, MCT8 immunoreactivity in the mouse retina was strongest during the phase of retinal maturation and showed a gradual decline in adult stages. A difference between the ventral and dorsal retina was not observed in the micrographs, despite of a dorsoventral S-opsin gradient typical for this mouse species (not shown).
A negative control in which we omitted the primary antibodies, showed no staining despite of some faint autofluorescence in the sclera and RPE.ba in the GLUT1 channel (green; Supplementary Figure ). Furthermore, a positive control in which we stained brain cryosections with the same anti-MCT8 antibody was performed. Immunoreactivity was found around neuronal cell bodies similar to the staining observed in the retinal ganglion cells (Supplementary Figure ).
## Discussion
Previous studies showed that TH are crucial for photoreceptor development and function in vertebrates, including human ( ; ; ; ; ; ), and recent studies even showed that TH dysregulation could be involved in photoreceptor degeneration such as age-related macular-degeneration ( ; ). In the present study, we investigated the availability of MCT8, known as an essential TH transporter of the nervous system, in the postnatal murine retina with focus on age-dependent changes.
We could show that MCT8 protein is localized in the RPE, INL, and GCL of all age groups, but we found an age-dependent decrease of MCT8 levels between juvenile and adult age groups. However, on mRNA level no significant differences between age groups were found for Mct8 , suggesting that Mct8 is expressed at almost equal levels throughout life. This is different to findings in the murine brain where Mct8 mRNA was reported to decline with age ( ). Although we homogenized whole eyecups containing RPE and retina for expression analyses and Western blotting, we assume that our data represent MCT8 mRNA and protein levels of the RPE and retina solely, because MCT8 immunoreactivity was not detected in any other eyecup layer by immunohistochemistry, suggesting an absence in the choroid and the sclera lying posterior to the RPE ( Figures and ).
Constant Mct8 expression in combination with decreasing protein levels might be contradictory, but this kind of asymmetry between MCT8 mRNA and protein levels was previously reported in liver and kidney as well ( ). In the study by , Mct8 mRNA of 24M mice exceeded the levels of younger age groups by about 52%, while MCT8 protein levels decreased. In the kidney no age-dependent changes in MCT8 protein levels were observed, but mRNA levels decreased in old animals. Taken together, Mct8 mRNA expression and protein turnover seems to follow a non-linear pattern in diverse tissues. In the retina, excessive TH signaling was reported to have deleterious effects on photoreceptor viability ( ; ), hence downregulation of MCT8 protein levels could be interpreted as a post-transcriptional mechanism to avoid oversupply of TH sensitive photoreceptor cells. A similar protective mechanism was reported for the TH inactivating enzyme D3 in the embryonic mouse retina. D3 deletion leads to neonatal degeneration of immature cones, while additional Thrβ2 deletion prevents early cone loss, suggesting that D3 has a protective function in the immature retina by inhibiting TH signaling via THRβ2 ( ). However, D3 mRNA levels and activity decline rapidly after birth, suggesting that TH signaling contributes to postnatal processes involved in retinal maturation. These processes include differentiation and cell fate determination of retinal precursor cells, axonal growth, M/L-opsin expression, vascularization, and synaptogenesis to form the plexiform layers ( ; ; ). In the brain, axonal growth and synaptogenesis are dependent on proper postnatal TH supply, and impaired TH signaling leads to drastically decreased neuronal connectivity ( ). In the mammalian brain as well as in the retina, many first and second order TH responsive gene products are involved in neuronal maturation, function, and plasticity, such as laminins involved in astrocyte migration and vascularization ( ; ), the neurotrophins BDNF and NT-3 ( ; ; ), or Reelin involved in patterning of synaptic connectivity ( ; ). Moreover, T3 is required to inhibit S-opsin expression and activate M/L-opsin expression ( ; ) which shows a peak at P21 and is still expressed at high levels in adulthood ( ). TH can even promote angiogenesis by upregulating several pro-angiogenic factors, such as vascular endothelial growth factor (VEGF), basic fibroblast growth factor (bFGF) and angiopoietin 1 ( ; ). VEGF is produced in RPE cells and has its peak expression in the mouse retina at P21, which overlaps with vascularization processes ( ; ). Interestingly, T3 signaling leads to an upregulation of angiopoietin 1, while the antagonistic angiopoietin 2 is downregulated ( ). T4 was also shown to initiate angiogenesis via a non-genomic signaling pathway ( ). These pro-angiogenic effects could also play a role in retinal degenerative diseases. For instance, human patients with high T4 levels have a higher risk to develop an age-related macular degeneration via an unknown mechanism, while elevated T3 levels had no effect on its prevalence ( ). Interestingly, one form of age-related macular degeneration is characterized by choroidal neovascularization, which could be attributed to T4 induced angiogenesis. Accordingly, T4 induced neovascularization could be addressed to investigate the pathology of age-related macular degeneration in future studies.
MCT8, as a high affinity TH transporter, could thus be a crucial regulator of TH supply in the first postnatal weeks where many TH dependent processes are upregulated. To avoid degenerative effects in the retina associated with TH oversupply (e.g., excessive choroidal neovascularization), the age-dependent decrease of MCT8 protein levels could represent a protective mechanism.
However, a detailed evaluation of MCT8 immunoreactivity in the RPE ( Figure ) revealed that MCT8 is mainly found at the apical membrane, particularly in the interface between RPE and photoreceptor OS. Therefore, transport of TH through the basolateral membrane, i.e., from the choriocapillaris into the RPE, is likely to be mediated by other transporters in juvenile as well as adult animals. Similar distribution was also reported for two T4 transporters in the adult rat retina, the OATP1C1 and 1A4 ( ; ), but in contrast to MCT8, these two transporters are also evenly distributed along the RPE.ba and endothelial cells of retinal capillaries. Therefore, these two transporters are likely candidates for T4 uptake across the basolateral membrane and endothelial cells of retinal capillaries. However, OATP1C1 and 1A4 were only studied in the retina of adult rats, thus age-dependent analyses are necessary to draw further conclusions. Cell bodies in the INL and GCL also showed strong MCT8 immunoreactivity in the juvenile age groups, which can be explained by the TH dependent postnatal processes involved in retinal maturation stated above (e.g., synaptogenesis and vascularization). Since MCT8 levels decrease after this postnatal phase of maturation, low levels of MCT8 might be sufficient to maintain TH supply in the adult retina. Furthermore, other TH transporters with a lesser affinity might be involved in retinal TH supply, as well.
## Conclusion
The present study along with previous studies suggest that TH transporters in the mouse retina are likely to exhibit a spatiotemporal expression pattern, similar to the mouse brain. The most likely interpretation is a change in retinal TH demand in different life stages. Our findings support a pivotal role of MCT8 in TH supply of photoreceptors, interneurons, and ganglion cells especially during postnatal maturation. However, the primary transport of TH through the outer and inner BRB is likely to be facilitated by transporters other than MCT8. MCT8 was instead localized at transitional zones between the BRB (RPE.ap) and photoreceptor OS, and at the surface of cell bodies (INL, GCL). Thus far, our results along with previous findings suggest a complex network of TH transporters and other TH regulating components in the retina, therefore further investigation is needed to create a scaffold of retinal TH signaling for a deeper understanding of retinal development and function, as well as age-related disorders.
## Author Contributions
Conception of the work: YH. YH and KS contributed to the experimental procedures, data analyses, and drafting of the final manuscript version in equal parts.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
The incorporation of peripheral biomarkers in the treatment of major depressive disorders (MDD) could improve the efficiency of treatments and increase remission rate. Peripheral blood mononuclear cells (PBMCs) represent an attractive biological substrate allowing the identification of a drug response signature. Using a proteomic approach with high-resolution mass spectrometry, the present study aimed to identify a biosignature of antidepressant response (fluoxetine, a Selective Serotonin Reuptake Inhibitor) in PBMCs in a mouse model of anxiety/depression. Following determination of an emotionality score, using complementary behavioral analysis of anxiety/depression across three different tests (Elevated Plus Maze, Novelty Suppressed Feeding, Splash Test), we showed that a 4-week corticosterone treatment (35 μg/ml, CORT model) in C57BL/6NTac male mice induced an anxiety/depressive-like behavior. Then, chronic fluoxetine treatment (18 mg/kg/day for 28 days in the drinking water) reduced corticosterone-induced increase in emotional behavior. However, among 46 fluoxetine-treated mice, only 30 of them presented a 50% decrease in emotionality score, defining fluoxetine responders (CORT/Flx-R). To determine a peripheral biological signature of fluoxetine response, proteomic analysis was performed from PBMCs isolated from the “most” affected corticosterone/vehicle (CORT/V), corticosterone/fluoxetine responders and non-responders (CORT/Flx-NR) animals. In comparison to CORT/V, a total of 263 proteins were differently expressed after fluoxetine exposure. Expression profile of these proteins showed a strong similarity between CORT/Flx-R and CORT/Flx-NR (R = 0.827, p < 1e ). Direct comparison of CORT/Flx-R and CORT/Flx-NR groups revealed 100 differently expressed proteins, representing a combination of markers associated either with the maintenance of animals in a refractory state, or associated with behavioral improvement. Finally, 19 proteins showed a differential direction of expression between CORT/Flx-R and CORT/Flx-NR that drove them away from the CORT-treated profile. Among them, eight upregulated proteins (RPN2, HSPA9, NPTN, AP2B1, UQCRC2, RACK-1, TOLLIP) and one downregulated protein, TLN2, were previously associated with MDD or antidepressant drug response in the literature. Future preclinical studies will be required to validate whether proteomic changes observed in PBMCs from CORT/Flx-R mice mirror biological changes in brain tissues.
## Introduction
Major Depressive Disorders (MDD) are the most frequent mental disorders worldwide (15% lifetime prevalence, 5% 1-year prevalence). MDD can lead to significant mortality, morbidity, reductions in quality of life, and have considerable costs implications to society ( ). In addition, the main risk with MDD is suicide-related mortality. About 6–20% of patients suffering from MDD die by suicide ( ). Selective Serotonin and/or Norepinephrine Reuptake Inhibitors are the most commonly prescribed drugs for the treatment of MDD. Despite recent advances in the pharmacological treatment of MDD, antidepressant drugs are only partially effective, with a 47% response rate and a 30% remission rate after the first-line treatment ( ; ). As most patients fail to enter remission with the first treatment, the incorporation of peripheral biomarkers in the treatment of MDD could supplement clinical observation and increase the remission rate ( ; ; ; ).
The study of proteins as potential disease or treatment biomarkers in the field of psychiatry, seems to be a straightforward approach since they are the main component of the cells and also drug targets ( ). Peripheral blood mononuclear cells (PBMCs) are circulating homogenous cells that can be easily collected and monitored across time in various species including humans and rodents. PBMCs may have a greater diagnostic power than a whole blood signature ( ). Interestingly, parallel transcriptomic changes have been observed using microarray in PBMCs and the brain, including the hippocampus in stressed mice ( ). Yet, no biomarker has proven sufficient validity to be translated to the clinic ( ). However, a few candidates associated with antidepressant response have recently emerged. For instance, similar change in protein expression in PBMCs (β-arrestin 1, a key regulator and scaffolds for G-protein-coupled receptor) has been observed between depressed treated-patients ( ) and depressed-like treated mice ( , ). Recently, showed that reduction in p11 in PBMCs could potentially predict antidepressant response to citalopram. Additionally, there is growing evidence of multiple dysregulated contributing factors, including growth factors, altered endocrine factors ( ) or immune-related pathways ( ; ) in mood disorders and/or in antidepressant responses. Thus, a viable alternative to the single-biomarker approach could be the development of biomarker panels to provide coverage of multiple biological factors that contribute to the heterogeneity of MDD and treatment response ( ).
From “omic” approaches to brain imaging, different strategies may help to identify putative biomarkers of the pathophysiology and antidepressant response. Stimulated by the disappointing results of the genome wide association studies for antidepressant response, the potential of gene expression and proteomics as sources of predictive biosignatures have been explored ( ). Applications of peripheral (PBMC, plasma, serum) proteomics, with the highest potential for having an impact on clinical practice, could be the identification of signatures or biomarkers which may predict antidepressant responses ( ; ).
Using a proteomic approach with high-resolution mass spectrometry technique, this study aimed to identify an indicative biosignature of fluoxetine response, which is commonly used as an antidepressant medication in PBMCs, and its isolated from a well-validated mouse model of anxiety/depression based on elevation of blood levels of glucocorticoids ( ). We proposed that an overall estimation of mouse behavior based on response across complementary tests of anxiety/depressive-like behavior would be able to discriminate fluoxetine responders from non-responders. We hypothesized that a peripheral proteomic signature in isolated PBMCs would help in understanding the common and distinct effects of fluoxetine responders compared to non-responders.
## Experimental Procedures
### Subjects
Adult C57BL/6NTac male mice were purchased from Taconic Farms (Lille Skensved, Denmark). All mice were 7–8 weeks old, weighed 23–25g at the beginning of the treatment and were maintained on a 12L:12 D schedule (lights on at 0600). They were housed in groups of five. Food and water were provided ad libitum . The protocols involving animals and their care were conducted in conformity with the institutional guidelines that are in compliance with national and international laws and policies (Council directive #87-848, October 19, 1987, Ministère de l’Agriculture et de la Forêt, Service Vétérinaire de la Santé et de la Protection Animale, permissions # 92-256B to DJD) and in compliance with protocols approved by the Institutional Animal Care and Use Committee (CEE26 authorization #4747).
### Drugs
Corticosterone (4-pregnen-11b-DIOL-3 20-DIONE 21-hemisuccinate from Sigma (Sigma–Aldrich Saint-Quentin Fallavier, France) was dissolved in vehicle (0.45% hydroxypropyl-β-cyclodextrin, Sigma–Aldrich Saint-Quentin Fallavier, France). Fluoxetine hydrochloride (18 mg/kg per day in the drinking water) was purchased from Anawa Trading (Zurich, Switzerland).
### Corticosterone Model and Treatment
The dose and duration of corticosterone treatment were selected based on previous study (CORT model, ; , ). Corticosterone (35 μg/ml, equivalent to about 5 mg/kg/day) or vehicle (0.45% β-cyclodextrine, β-CD) were available ad libitum in the drinking water in opaque bottles to protect it from light. Corticosterone-treated water was changed every 3 days to prevent any possible degradation. During the last 4 weeks of the protocol, corticosterone was delivered alone ( n = 12 animals, CORT/V) or in the presence of fluoxetine (18 mg/kg/day, n = 46 animals, CORT/Flx) (see the experimental protocol on Figure ). Treatments were maintained until the end of the experiments. Behavioral sessions to assess anxiety/depression-like phenotype and also the antidepressant response to fluoxetine occurred on week 4 and 9, respectively. Control animals received vehicle (vehicle/vehicle, VEH/V).
Timeline of experiments. In place of normal drinking water, grouped-housed male C57BL/6Ntac mice were presented during 10 weeks with vehicle (0.45% hydroxypropyl-β-cyclodextrin) or corticosterone (35 μg/ml) in the presence or absence of an antidepressant (fluoxetine, 18 mg/kg/day) during the last five weeks of the corticosterone regimen. Emotionality z -score was calculated after each behavioral session. Then, we investigated whether the behavioral changes induced after chronic corticosterone (week 4 to 5) were corrected by antidepressant treatment (week 9 to 10). The same animal was successively tested in the Elevated Plus Maze (EPM), the Novelty Suppressed Feeding (NSF), the Splash Test (ST) during both behavioral sessions. Peripheral Blood Mononuclear Cells were isolated from whole blood after each behavioral session.
### Behavioral Experiment
#### Elevated Plus Maze (EPM)
The elevated plus maze (EPM) is a widely used behavioral assay for rodents and it has been validated to assess the anti-anxiety effects of pharmacological agents (for review ). This test was performed as described previously ( ). The maze is a plus-cross-shaped apparatus, with two open arms and two arms closed by walls linked by a central platform 50 cm above the floor. Mice were individually put in the center of the maze facing an open arm and were allowed to explore the maze for a duration of 5 min. The time spent in the maze and the numbers of entries into the open arms were used as an anxiety index. All parameters were measured using a videotracker (EPM3C, Bioseb, Vitrolles, France).
#### Novelty Suppressed Feeding (NSF)
The NSF is a conflict test that elicits competing motivations: the drive to eat and the fear of venturing into the center of a brightly lit arena. The latency to begin eating is used as an index of anxiety/depression-like behavior, because classical anxiolytic drugs as well as chronic antidepressants decrease this measure. The NSF test was carried out during a 10 min period as previously described ( ). Briefly, the testing apparatus consisted of a plastic box (50 cm ×40 cm × 20 cm) the floor of which was covered with approximately 2 cm of wooden bedding. Twenty-four hours prior to behavioral testing, all food was removed from the home cage. At the time of testing, a single pellet of food (regular chow) was placed on a white paper platform positioned in the center of the box. Each animal was placed in a corner of the box, and a stopwatch was immediately started. The latency to eat (defined as the mouse sitting on its haunches and biting the pellet with the use of forepaws) was timed. Immediately afterwards, the animal was transferred to its home cage, and the amount of food consumed by the mouse in the subsequent 5 min was measured serving as a control for change in appetite as a possible confounding factor.
#### Splash Test (ST)
This test consisted of squirting a 10% sucrose solution on the mouse’s snout. This procedure induces grooming behaviors, due to the viscosity and palatability of the sucrose. The grooming behavior is sensitive to chronic stress or chronic corticosterone exposure and antidepressant treatment ( ). The total time spent in different grooming behaviors (i.e., face, paws, hindquarter, and shoulders) was directly recorded for 5 min in the home cage of the animals.
#### Behavioral Emotionality Measurement
Three behavioral tests (i.e., EPM, NSF, and ST) were used to measure components of animal behavioral emotionality. Z-score methodology was used to investigate the potential of combining results within and across the different behavior tests for depressive/ anxious-like behaviors and investigate the treatment effects in the CORT model. The emotionality-related data was normalized as previously described ( ; ). Briefly, z scores are standardized scores (by the group mean and group standard deviation). They indicate how many standard deviations (σ) an observation ( x ) is above or below the mean of a control group (μ).
Z scores for behavioral measures were first averaged within the test and then across the test for equal weighting of the three tests comprising the final emotionality score. The increased behavioral emotionality was defined as decreased activity in the open arms in the EPM, increased NSF latency and decreased grooming in the splash test compared with control group means. The vehicle group was defined as the control. Thus, the emotionality score is not based on a single consistent behavior, but rather by a set of converging behavioral observations that together define an anxiety/depression-like phenotype. Emotionality score was calculated after the first and the second behavioral round. As we were interested in observing the differential antidepressant-response behaviors to global protein expression system level responses, behaviorally ambiguous mice were not used for mass spectrometry analysis described below.
### Isolation of Mouse Peripheral Blood Mononuclear Cells
To determine a biological signature of fluoxetine responders, the “most” affected animals of each group were used for proteomic analysis (5 for corticosterone/Vehicle, CORT/V; 7 for Corticosterone/fluoxetine responders, CORT/Flx-R; 6 for Corticosterone/fluoxetine non-responders, CORT/Flx-NR). The procedure was performed on unaenesthetized mice as previously described ( ). In compliance with the laboratory animal care guidelines, about 0.4 ml of blood per mice was collected in K EDTA tubes using the submandibular bleeding method. The punctures were performed with 5 mm point size sterile lancets (MediPoint, Mineola, NY, United States) where the orbital vein and the submandibular vein join to form the jugular vein ( ). A light pressure with dry gauze was applied to the punctured area for hemostasis. Separation and extractions of PBMCs were done using the iodixanol mixer technique ( ). Separations of mouse PBMCs were purified of mouse whole blood through density centrifugation (1,000 rpm at 20°C for 30 mn) using solution B with the OptiPrep gradient solution (Sigma–Aldrich Saint-Quentin Fallavier, France). After centrifugation, OptiPrep gradient solution separated layers of blood, with PBMCs under a layer of plasma. The PBMCs layers were carefully removed from the tube and transferred to a new 50 mL conical tube and were washed twice with solution B. After centrifugations (1,200 rpm at 20°C for 7 mn) and several washing steps, mouse PBMCs were recovered with a last centrifugation (3,000 rpm at 4°C for 5 mn) and stored at -80°C before subsequent assay.
### Proteomics Analysis
#### Protein Separation
Protein extracts from PBMCs were homogenized in solution solubilization (Urea 7M, Thiourea 2M, CHAPS 3%, Nonidet P-40 1%, DTT 1%). Protein concentration was measured using 2D-Quant kit (GE Healthcare, France) and 15 μg of proteins were loaded and separated by 12 % SDS-PAGE. A short migration was then performed (7 min, 80 V, 25 W followed by 4 min, 200 V, 25 W) and gels were stained with Coomassie colloidal blue (EZblue, Sigma–Aldrich, France).
#### Protein In-Gel Digestion
Portions of gel that contain all proteins were cut and digested as followed: pieces of gel were successively washed and de-stained with water, acetonitrile (ACN) and 25 mM ammonium bicarbonate (NH HCO ). A reduction/alkylation step was performed with dithiothreitol (DTT) 10 mM and iodoacetamide 55 mM. Gels were dehydrated with acetonitrile and rehydrated at 4°C in 12 ng/μL sequencing grade modified trypsin (Promega, France) solubilized in 25 mM NH HCO in 1 h and then digested at 37°C overnight. After tryptic digestion, peptides were extracted by incubating gel pieces in extraction solvent (0.5% trifluoroacetic acid (TFA)/50% ACN) for 15 min and in ACN for 15 min at room temperature. Supernatants were vacuum dried. The dried extract peptides were dissolved in 50 μl of loading buffer (0.08% TFA/2% ACN) just before mass spectrometry analysis.
#### Mass Spectrometry Analysis
Four microliters of sample were loaded on the nano-UPLC Ultimate 3000 RSLCnano (Thermo). Sample was loaded at 20 μL/min on the pre-column cartridge (PepMap 100 C18, 5 μm; 300 μm i.d., 5 mm, Thermo Scientific) and peptides were then separated with a gradient of acetonitrile on the reverse phase column PepMap 100 C18 (stationary phase: C18, 3 μm; column: 75 μm i.d., 500 mm; nanoViper, Thermo Scientific, France). Buffers were 0.1% formic acid in 2% acetronitrile (A) and 0.1% formic acid in 80% acetonitrile (B). The peptide separation was realized during 64 min at 300 nL/min with a linear gradient from 0 to 45% B for 55 min followed by a gradient from 45 to 98% B for 5 min. Eluted peptides were analyzed on-line with a high-resolution mass spectrometer Orbitrap Fusion Lumos Tribrid (Thermo Scientific, France) using a nanoelectrospray interface in positive polarity mode, on PAPPSO platform . Peptide ions were analyzed using Xcalibur 3.0 (Thermo Scientific, France) with following data-dependent acquisition steps: (1) full MS scan in orbitrap (mass-to-charge ratio [ m / z ] = 400–1500; mass tolerance, ±10 ppm) and (2) MS/MS in Ion Trap with CID activation (collision energy, 35%; activation time, 30 ms; centroid mode). Dynamic exclusion time was set to 60 s.
### Statistical Analysis
#### Behavioral Analysis
To assess the behavioral consequence of a chronic corticosterone treatment in the EPM, NSF, and ST or emotionality scores, Student’s t -tests were performed and results were expressed as mean ± SEM values. For the behavioral analysis after chronic fluoxetine treatment, a one-way ANOVAs was applied to the data as appropriate. Significant main effects were followed by Fisher’s Post hoc test. With regards to the NSF test, we used the Kaplan–Meier survival analysis owing to the lack of normal distribution of the data. Mantel–Cox log rank test was used to evaluate differences between experimental groups. Statistical significance was set at P < 0.05. Data were analyzed using Prism 6.0h software (GraphPad, La Jolla, CA, United States).
#### Data Processing and Bioinformatics Analysis
Peak lists were generated as mzXML files using the converter MSConvert (ProteoWizard). A database search was performed using X!TandemPipeline software developed by PAPPSO facility (version 3.4.3) ( ) with search parameters as followed: enzymatic cleavage by trypsin digestion with one possible miscleavage; fixed carbamido-methylation modification on cysteine and variable oxidation on methionine; precursor mass tolerance of ±10 ppm and fragment mass tolerance of 0.5 Da. Several databases were used: the Uniprot KB/SwissProt Mus musculus database (24977 entries, version January 2017) and a homemade contaminant database (trypsin, keratine, etc.). The identified proteins were filtered with a minimum of two different peptides required with a peptide E-value < 0.01, and a protein E-value (product of unique peptide E values ) < 10 . Combine analysis mode with all samples was performed and results collected from grouping proteins: proteins, which have at least one peptide in common. This allowed to group proteins with similar functions. Within each group, proteins with at least one specific peptide relatively to other members of the group were reported as subgroups. One subgroup represents one specific protein. Proteins are characterized with their spectral number. Label free quantification of proteins were achieved with spectral counting approach (SC), which is a strategy to determine a relative quantification of protein from their number of spectra obtained with tryptic peptides in MS. This quantification is based on the fact that more of a particular protein is present in a sample; more MS spectra are detected for peptides of that protein. Statistical analysis was performed using MassChroqR package developed by PAPPSO team (R version 3.3.2). A generalized linear mixed model (GLM) with a Poisson distribution was applied. This model suits in the case of a counting like SC. The principal component analysis was obtained by simulating the kernel densities from group’s means and variances assuming bivariate normal distributions. This distribution was generated using protein abundances as variables. Hierarchical bivariate clustering was performed using Euclidean distances and unweighted pair group averages as the aggregation method. All data analyses and graphical representations were performed using the R package MassChroq. Significant changes in protein abundance were determined by analysis of variance (ANOVA) using a Chi-square test. Treatment effect was considered with an adjusted p -value for multiple testing by a Benjamini–Hochberg procedure ( ). Student’s t -tests were performed to identify proteins with significant differences expressed between groups with the following criteria: p -value was set at < 0.05.
#### Functional Analysis
Selected proteins were overlaid on the global molecular network of Ingenuity Pathway Analysis (Ingenuity Systems) allowing for a generation of gene networks based on their connectivity. Their score takes into account the relative numbers of network eligible molecules, of molecules analyzed and the total number of molecules in Ingenuity’s knowledge base. IPA generates disease links on the literature-based association with illness.
## Results
Detailed statistical results for behavior are provided in Supplementary Table .
### A 4-week Corticosterone Treatment Induced Anxiety/Depression-Like Phenotype
Using a low dose of corticosterone (35 μg/ml) for 4 weeks, we demonstrated that C57BL/6Ntac treated mice developed an anxiety/depression-like phenotype in the EPM, NSF and Spash tests ( Figure ) as previously described ( ; ). Indeed, a decrease in time spent and in entries in the open arms ( Figures , p < 0.05, and p < 0.01, respectively), an increase in the latency to feed ( Figures , p < 0.01) and a decrease in grooming duration ( Figure , p < 0.01) were observed in corticosterone-treated mice. Z -score normalization was then performed, within the respective behavioral parameters, hence transforming absolute values to numbers of standard deviations from the vehicle means. Z -scoring across complementary behavioral dimensions provided a more robust overall assessment of the effect of stress on behavioral emotionality ( ). Thus, chronic corticosterone treatment induced an anxiety/depressive-like phenotype in mice, as measured by an increase in the emotionality score ( Figure , p < 0.01).
A 4-week corticosterone treatment (35 μg/ml) induced an anxiety/depression-like phenotype in C57BL/6Ntac mice. (A,B) Effects of corticosterone (35 μg/ml, CORT) regimen on anxiety behaviors in the Elevated Plus Maze (EPM). Anxiety, measured for various parameters is expressed as mean total time in seconds (A) or entries (B) in open arms of EPM paradigm. (C) Effects of 4 weeks of corticosterone regimen (35 μg/ml) on anxiety- and depression related behaviors in the Novelty Suppressed Feeding paradigm. Results are expressed as cumulative survival with percentage of animals that have not eaten over 10-min. (D) Effects of 4 weeks of corticosterone regimen (35 μg/ml, CORT) on depression related behaviors in the Splash Test (ST). Results are expressed as mean of grooming duration (in seconds). (E) Effects of 4 weeks of corticosterone regimen (35 μg/ml, CORT) on food consumption in the Novelty Suppressed Feeding paradigm. Results are expressed as mean of food consumption (in mg/g of mouse). (F) Effects of 4 weeks of corticosterone regimen (35 μg/ml, CORT) on anxiety/depression-like behaviors on the emotionality score. Test Z -values (elevated plus maze, novelty-suppressed feeding and splash test) are calculated by averaging individual Z -scores to obtain emotionality Z -scores. Values plotted are mean ± SEM [ n = 14 and 59 animals for vehicle (VEH, open circle) and corticosterone (CORT, black dot) per group respectively]. Unpaired t -test ( p < 0.05, p < 0.01 versus VEH group) or Kaplan–Meier survival analysis followed by Mantel–Cox log-rank test were applied ( p < 0.01 versus VEH group).
### Effects of a 4-week Treatment with Fluoxetine in a Stress-Related Model of Anxiety/Depression
We then explored whether a chronic fluoxetine treatment was able to correct the anxiety/depressive-like state induced by chronic corticosterone. In the EPM, a chronic treatment with fluoxetine corrected corticosterone-induced decrease in time and entries in the center ( Figures , p < 0.05, and p < 0.01, respectively). In the NSF, a trend for a reduction of chronic corticosterone-induced increase in latency to feed was observed after chronic fluoxetine treatment ( Figure , Kaplan–Meier survival analysis, Mantel-Cox log-rank test, p < 0.01, p = 0.08 for the insert bar chart). Finally, after squirting a 10% sucrose solution on the mouse’s snout, the decrease in grooming frequency observed in chronic corticosterone animals was reversed by chronic fluoxetine treatment ( Figure , p < 0.05).
Chronic fluoxetine treatment produces anxiolytic-like and antidepressant-like effects in a mouse model of anxiety/depression. (A,B) Effects of 4 weeks of fluoxetine treatment (18 mg/kg/day, Flx) on anxiety behaviors in the Elevated Plus Maze (EPM). Anxiety, measured for various parameters is expressed as mean total time in seconds (A) or entries (B) in open arms of EPM paradigm. (C) Effects of 4 weeks of fluoxetine treatment (18 mg/kg/day, Flx) on anxiety- and depression related behaviors in the Novelty Suppressed Feeding paradigm. Results are expressed as cumulative survival with percentage of animals that have not eaten over 10-min or mean of latency to feed (in seconds) (inset). (D) Effects of 4 weeks of fluoxetine treatment (18 mg/kg/day, Flx) on depression related behaviors in the Splash Test (ST). Results are expressed as mean of grooming duration (in seconds). Values plotted are mean ± SEM [ n = 14, 12, and 46 animals for vehicle/vehicle (VEH/V, open circle) corticosterone/vehicle (CORT/V, black dot) and corticosterone/fluoxetine (CORT/Flx, orange dot) per group respectively]. One-way ANOVA Fisher’s PLSD post hoc analysis ( p < 0.05, p < 0.01 versus Veh/V group; p < 0.05, p < 0.01 versus CORT/V group) or Kaplan–Meier survival analysis followed by Mantel–Cox log-rank test were applied ( p < 0.01 versus Veh/V group).
### Responders and Non-Responders to Chronic Fluoxetine Treatment in Corticosterone-Induced Anxiety/Depression-Like Phenotype
Chronic corticosterone-treated animals were distributed for the last 4 weeks of treatment in a way that no significant difference occurred between the two groups (CORT-V and CORT/Flx) ( Supplementary Figure ). At the end of the study, we also ensured that CORT/Flx-R and CORT/Flx-NR did not differ in their emotionality score before fluoxetine treatment.
A one-way ANOVA after the second behavioral session revealed a significant effect of fluoxetine on emotionality score. Applying z-normalization across tests after the second round of behavior demonstrated that chronic fluoxetine decreased CORT-induced anxiety/depression-like phenotype in mice, as measured by a decrease in emotionality score ( Figure , p < 0.01 versus VEH/V group; p < 0.01 versus CORT/V group), confirming previous studies ( ; ).
Two thirds of chronic fluoxetine-treated animals block stress-induced increase in emotionality z -score. Normalization of data using z -score method was performed for each behavioral parameter in EPM, NSF and ST after the second session of behavior. Test z -values were then calculated by averaging individual z -scores, and averaged to obtain the emotionality score. Values plotted are mean ± SEM [ n = 14, 12, 30, and 16 animals for vehicle/vehicle (VEH/V, open circle), corticosterone/vehicle (CORT/V, black dot), corticosterone/fluoxetine responder (CORT/Flx-R, green dot) and corticosterone/fluoxetine non-responder (CORT/Flx-NR, red dot) per group, respectively]. Colored dot (CORT/V, black dot; CORT/Flx-R, green dot; CORT/Flx-NR; red dot) are the “most” affected animals selected for the proteomic approach. One-way ANOVA Fisher’s PLSD post hoc analysis ( p < 0.05, p < 0.01 versus Veh/V group; p < 0.01 versus CORT/V group; p < 0.01 versus CORT/Flx-R group).
Interestingly, in fluoxetine-treated mice, phenotypic variations were observed as two groups of mice could be distinguished according to their emotionality score ( Figure and Supplementary Figure ). Clinical response to antidepressants in depression is defined as a 50% decrease in rating scale score ( ). Thus, we also analyzed the behavioral data according to the 50% cut-off. A fluoxetine-treated animal with a 50% decrease in emotionality score was defined as CORT/Flx-R, whereas a decrease in this score below 50%, indicates a CORT/Flx-NR animal. Of 46 fluoxetine-treated animals, 65.2% (30 mice) responded to fluoxetine with an average decrease of 1.74 points in emotionality score between both behavioral sessions ( Supplementary Figure , p < 0.01 versus VEH/V group; p < 0.01 versus CORT/V group; p < 0.01 versus corticosterone group). In contrast, in CORT/Flx-NR animals (16 mice out of 46), a 0.82 points increase in emotionality score was observed between the two rounds of behavior, pointing out the absence of anxiolytic/antidepressant-like effect of fluoxetine in these animals ( Supplementary Figure ).
After the behavioral sessions and to determine a biological signature of fluoxetine response, PBMCs from the “most” affected animals of each group (5 for CORT/V, 7 for CORT/Flx-R and 6 for CORT/Flx-NR) were isolated for further proteomic analysis ( Supplementary Figure ).
### Common and Differential Proteomic Changes in Responders and Non-Responders to Fluoxetine
Using a high resolution mass spectrometry analysis by X!Tandem Pipeline 1245 specific proteins ( n = 5–7 samples per group) were detected in PBMCs ( Supplementary Tables , ). Characterized proteins with less than two peptides were excluded, leading to identification of 938 proteins. Hierarchical clustering of the expressed proteins distinguished CORT/V treated animals from CORT/Flx-R and CORT/Flx-NR, however, with some overlap ( Figure ). This aggregate behavior of this large-scale systemic response was quantified with Principal Components Analysis (PCA, Figure ), which confirmed hierarchical clustering analysis. According to PCA, proteins’ abundance separated CORT/V, CORT/Flx-R and CORT/Flx-NR.
Peripheral proteomic changes after fluoxetine exposure in responders and non-responders. (A) Hierarchical bivariate clustering of expression profiles of animals (column) and proteins (rows) depicts the differences between CORT/V (black rectangle), CORT/Flx-R (green rectangle), and CORT/Flx-NR (red rectangle) groups. An animal’s expression is red for above-average values, and blue for below-average values. (B) Principal Component Analysis of expression profiles revealed 2 main axis separating results. (C) Venn diagram of protein levels with significant fluoxetine effect in CORT/Flx-R and CORT/Flx-NR groups. Changes affects massively common proteins between these two groups. Changes associated solely with CORT/Flx-R (38) or CORT/Flx-NR (80) strongly correlated between each other, as indicated by the arrows. Arrows indicate directional correlations between changes in protein expression for protein identified significant in one group (origin of arrow) and changes for the same protein in the other group (end of arrow). (D) Overall directional changes of protein expression were strongly correlated between groups. (E) Hundred proteins were observed significantly differentially expressed between CORT/Flx-R and CORT/Flx-NR groups, however (F) only 19 out of 100 could be associated with response to fluoxetine. p < 0.0001 after Pearson χ correlation analysis.
We observed a group effect for 305 proteins ( p < 0.05, Supplementary Table ). We found changes in expression levels of 183 proteins that were significantly altered by chronic fluoxetine in CORT/Flx-R mice, and 225 in CORT/Flx-NR ( p < 0.05, Figure and Supplementary Table ). Among the proteins affected by fluoxetine, we observed a strong overlap across CORT/Flx-R and CORT/Flx-NR groups, with 145 proteins observed in common. Interestingly, protein changes observed solely in responders displayed a similar trend of expression in the CORT/Flx-NR group (arrow in Figure , Pearson correlation value R = 0.82, p < 1e ), and vice-versa ( R = 0.69, p < 1e ). These response-independent effects of fluoxetine treatment on PBMCs proteome were also reflected in Figure . Indeed, 263 proteins were differentially affected by fluoxetine (compared to CORT) either in responders, non-responders or both with strong similarities of expression between responders and non-responders ( R = 0.827, p < 1e ).
Direct comparison of CORT/Flx-R and CORT/Flx-NR groups revealed 100 proteins differentially expressed between responders and non-responders ( p < 0.05, Figure ). Among these proteins, 19 of them were associated with treatment response. Their expression was significantly different between CORT/Flx-R and CORT/Flx-NR, significantly different between CORT/Flx-R and CORT/V-treated animals and the direction of expression changes was either opposite or of greater amplitude in the CORT/Flx-R group compared to CORT/Flx-NR ( Figure ). Interestingly, an IPA analysis revealed “ protein ubiquitination pathways ”, “ interleukin 1 signaling” and “ metabolic diseases” as the top canonical pathways diseases and biological functions associated with these 19 proteins, respectively ( Table ). Moreover, a literature search performed among these proteins using PubMed and Google Scholar indicated that 9 of them have been previously associated with antidepressant drug response or with MDD in clinical or preclinical studies (8 upregulated: RPN2, HSPA9, NPTN, AP2B1, UQCRC2, RACK-1, TOLLIP, and 1 downregulated protein, TLN2, Table ).
Ingenuity pathway analysis for functional analysis of the mapped biological functions and/or disease categories and canonical pathways for proteins in peripheral blood mononuclear cells.
Identification of the 19 proteins showing significant differential direction of expression in fluoxetine responder (CORT/Flx-R) and fluoxetine non-responder (CORT/Flx-NR) mice.
## Discussion
Current antidepressant drug treatments are not sufficient, as many patients do not adequately respond. Animal models, such as the CORT model, are valuable in providing a translational framework to study SSRI insensitivity and to validate findings as potential biomarkers for treatment responsiveness in humans. This kind of approach will pave the way for novel approaches and therapeutic strategies for relieving depressive disorders. Here, using the mouse CORT-model associated with an emotionality score analysis and proteomic identification in PBMCs, we developed a novel approach to determine a protein expression profile between CORT/Flx-NR and CORT/Flx-R ( ).
### Behavioral Emotionality and PBMCs Isolation from Responders and Non-responders to Fluoxetine in a Mouse Model of Anxiety/Depression
Current animal models are classically used to test whether antidepressant drugs can reverse stress-induced anxiety/depression-like phenotype. In the literature, antidepressant response in these models is defined by comparing the mean results of the behavioral task across control and treated groups, without distinguished responders from non-responders ( ). Applying this classic methodology, we confirmed that chronic corticosterone-induced anxiety/depression-like phenotype is overall corrected by chronic fluoxetine treatment ( , ; , ). Additionally, we used scatterplots representations of the emotionality score to observe the distribution of the results and separate responders and non-responders to chronic antidepressant treatment. Mice that participated in the behavioral tests (EPM, NSF, ST) showed a different distribution in response to antidepressant treatment. Interestingly, previous observations in the NSF already showed that not all CORT-treated mice respond to chronic antidepressant treatment, thus resulting in a bimodal distribution ( ). Here, the selection of animals responding or not responding to chronic fluoxetine was not based on results obtained in a single test, which is classically observed ( ; ; ; ; ). Rather, the use of the behavioral emotionality score covers the multiple aspects of emotional phenotype. Thus, fluoxetine-treated animals were subdivided in CORT/Flx-R and CORT/Flx-NR according to a 50% decrease in emotionality score cut off as defined in clinic ( ). Our study revealed a fluoxetine-responding rate of 65%, which is similar to what is observed in clinic. Using this approach augmented the translational validity of the CORT model to define a biological signature of antidepressant response.
Previously, we showed that PBMCs isolated from a small volume of whole blood in unanesthetized mice using a submandibular bleeding method provided a useful biological tool that assess circulating protein expressions and allowed the screening of potential biomarkers for antidepressant response ( , ). In order to delineate a panel of biomarkers characteristic of fluoxetine response, we isolated PBMCs from the “most” affected CORT/Flx-R and CORT/Flx-NR animals.
### From β-Arrestin 1 Levels to a Differential Biological Signature of Fluoxetine Response in PBMC
Proteomic analysis has been described as a powerful tool for the identification of biomarkers ( ). Using high-resolution mass spectrometry, from 1245 proteins detected among the three conditions (CORT/V; CORT/Flx-R; CORT/Flx-NR), 305 were significantly and different expressed.
Surprisingly, among the proteins exhibiting variations compared to CORT/V group, Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) showed a significant decreased in CORT/Flx-R and NR animals. A previous in vitro study demonstrated that antidepressants such as nortriptyline and escitalopram affect the expression of many housekeeping genes, including GAPDH ( ). From 10 housekeeping genes, GAPDH was the least stable protein detected in Mouse fibroblast L929 cells. Therefore, precautions should be taken in the use of certain housekeeping genes such as GAPDH, as its proteomic expression changes after antidepressant treatment.
Previously, we observed that β-arrestin 1 expression in PBMCs was restored to normal levels after chronic fluoxetine treatment in CORT-treated mice ( ). Here, we confirmed that β-arrestin 1 expression was significantly higher in CORT/Flx-R mice in comparison to CORT animals and a trend for its increase was observed between CORT/Flx-R vs. CORT/Flx-NR mice ( p = 0.07). Although β-arrestin 1 expression in PBMCs has been proposed as a potential marker of antidepressant response in rodents ( ) and in depressed patients ( ; ; ), MDD is more likely a multifactorial and polygenic disorder. Therefore, further investigation is needed to discover and validate other markers ( ).
Hierarchical clustering of proteomic and PCA analyses showed that Flx-NR-animals are neighboring CORT-treated animals and are separate from CORT/Flx-R animals along the axis 1 of the PCA analysis with some overlap. While some animals seemed “mis-categorized” (CORT/Flx-R_28 and CORT/Flx-NR_76), these proteomic “outliers” displayed similar behavioral alterations compared to other animals within the same group. Fluoxetine effects on peripheral proteomic changes were mostly independent from antidepressant response as we observed a strong overlap across CORT/Flx-R and CORT/Flx-NR groups. 145 proteins were commonly observed with similar directional expression between responders and non-responders, even for proteins for which a statistically significant difference was observed between these groups. This is unsurprising as molecular peripheral markers of antidepressant response in clinical studies observed little variation in gene expression between responder and non-responder subjects for the same dosage of antidepressant ( ), thus confirming that the proteins differentially expressed in CORT/Flx-R and CORT/Flx-NR groups compared to the CORT-treated group are mostly witnesses of antidepressant exposure than of treatment response. Preclinical studies in the chronic mild stress model (CMS) also confirmed that overall central transcriptomic changes in the lateral habenula are more similar between responders and non-responders to escitalopram than between antidepressant-treated group compared to CMS or unstressed animals ( ). Unfortunately, the preclinical work that focus on genes/proteins differentially expressed (in peripheral tissue, cell lines, or within brain regions) between responders or non-responders did not evaluate the common effects of antidepressant treatment on gene expression ( ; ; ).
Here, we observed 100 proteins differentially expressed between CORT/Flx-R and CORT/Flx-NR among the 1245 detected proteins (8%), which is in a higher range compared to previous studies using transcriptomic/proteomic approaches in animal models of the disease ( ; ; ), or in human subjects ( ; ; ). However, methodological aspects may explain this difference between studies, as the number of groups, statistical methods and selection of responder/non-responder differ.
Interestingly, we show that the expression profile of these 100 proteins differs between CORT/Flx-R and CORT/Flx-NR ( Figure ). Indeed, the effects caused by fluoxetine in CORT/Flx-NR generated a profile that strongly differed from the CORT-treated group. Importantly, the profile for CORT/Flx-R mice was closer to those found on the CORT/V group. This suggests that even if CORT/Flx-NR mice display a behavioral state similar to CORT/V-treated mice, their peripheral molecular state differs and might play a role in a “decanalization” process. This process has been proposed to explain the origin of complex diseases ( ; ), and may be applicable to a lack of drug treatment-response. This group of proteins may also participate in an iatrogenic effect of fluoxetine.
Among these 100 proteins differentially expressed between CORT/Flx-R and CORT/Flx-NR mice, only 19 proteins were found with a differential direction of expression between CORT/Flx-R and CORT/Flx-NR or with a profile of expression in CORT/Flx-R that droves them away from the CORT/V-treated profile group. This list of proteins may be associated with an improvement in the emotional state of the animals and markers of antidepressant response. As most of them (12/19) showed similar direction of expression in CORT/Flx-R and NR, this also suggests that a threshold of expression needs to be exceeded in order to obtain an efficient antidepressant response.
Interestingly, an IPA analysis on these 19 biomarkers revealed that “ protein ubiquitination pathway ” and “ Interleukin 1 (IL-1) signaling ” are the top canonical pathways associated with this list ( Table ). The predominant role of ubiquitination is to target substrates for rapid degradation within the 26S proteasome, but also to regulate protein function by proteasome-independent processes. Previous clinical findings support the role of ubiquitination not only in the pathophysiology of MDD, but also in the mechanism of antidepressant-like activity ( ). Ubiquitination of proteins such as β-arrestin 1 was demonstrated in PBMCs ( ). In regards to the IL-1 signaling, many evidences support the role of cytokines and, more generally, inflammation in MDD and antidepressant drug treatment responsiveness [( ) for review]. In regards to the top disease and biological functions category, “ metabolic disease ” belongs to the top disease and disorders. A 6-month prospective, multicentric, a real-world treatment observational cohort study of 624 patients (METADAP) suggests that treating MDE with antidepressant drugs can induce or worsen a metabolic syndrome ( ). Actually, biomarkers of antidepressant response might be linked to these side-effects.
Importantly, 8 of the 19 proteins that exhibit differential direction of expression in Flx-responders versus Flx-non-responders have been associated with MDD or antidepressant response (RPN-2, HSPA9, NPTN, AP1B1, UQCRC2, RACK1, TOLLIP, and TLN2 ( ; ; ; ; ; ; ; ; ; ; ; ). Finally, despite the few number of proteins used in this Ingenuity analysis, top networks included molecules belonging to “ Behavior ”, “ Nervous System Development and Function ”, which putatively suggest that we observe a peripheral signature of central differential response to fluoxetine treatment.
### Limitations of the Study
There is a great potential for biomarkers within the field of psychiatry, including diagnosis and/or prediction of treatment responses ( ). Whether or not the proteins involved in the fluoxetine response identified in PBMC are important for the pathophysiology, should also be studied. Moreover, at this stage, our study did not provide the means to directly assess brain markers and respectively, compare then with peripheral markers. Other limitation to address will be the sample size of the pilot study performed in an animal model of anxiety/depression, which was small, and an independent validation cohort was not available; hence, larger prospective pre-clinical studies are warranted.
The design of our study was to unravel the molecular and peripheral mechanisms associated with fluoxetine antidepressant response. Thus, no direct comparison between CORT/V-treated mice and controls was performed at the proteomic level, and the markers differentially expressed between responders and non-responders may not match with a peripheral signature of a behavioral state. Thus, this biosignature only reflects proteins associated with fluoxetine response/non-response. We should also consider that in our study, this peripheral proteomic biosignature of fluoxetine response was determined from the “most” affected mice from each group to increase group differences.
The biological validity of the inflammation and immune activation was not confirmed by other biological tests and is thus speculative at this time. Similarly, effects of fluoxetine on protein expression in naïve, unstressed animals, were not performed here. Thus, the possible iatrogenic effects of this antidepressant drug at the peripheral levels need to be confirmed.
Exploring protein expression profiles in homogeneous cell populations, such as PBMCs may provide a greater diagnostic power than whole blood signature. PBMCs include lymphocyte (T cells, B cells, and NK cells), monocyte and dendritic cells. In our study, the distribution of these cells in our experimental conditions has not been explored. We cannot rule out a decrease in blood PBMCs numbers. Indeed, leukocyte redistribution from the blood to other organs has been described ( ). Further studies should also consider an evaluation of PBMCs composition in the blood using flow cytometry. However, the extraction of more homogeneous cell populations, such as PBMC, which is often laborious and difficult to standardize, involves manipulation of the cells and may influence the expression profiles.
## Conclusion
In this study, we demonstrated that CORT model of anxiety/depression in mice allows the study of response/non-response to chronic fluoxetine treatment. We also provided evidence that even though CORT/Flx-R and CORT/Flx-NR groups share common proteins; a threshold of expression should be reached to categorize them as responders. These proteins represent a combination of markers associated both with the maintenance of a refractory state in these animals, while other may be associated with behavioral improvement. Whether these proteomic changes observed in PBMCs from CORT/Flx-R mice in the CORT model mirror biological changes or not in brain tissues, Further investigation is required.
## Author Contributions
IM-D, AG, DD: designed and performed research. IM-D, CB, VD, DD: performed research. IM-D, J-PG, DD drew figures. IM-D, CB, J-PG, RC, BF, EC analyzed the data. J-PG, RC, FB, EC: contributed to the preparation of the manuscript. IM-D, AG, J-PG, DD wrote the manuscript.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Anoxic depolarization (AD) is a hallmark of ischemic brain damage. AD is associated with a spreading wave of neuronal depolarization and an increase in light transmittance. However, initiation and spread of AD across the layers of the somatosensory cortex, which is one of the most frequently affected brain regions in ischemic stroke, remains largely unknown. Here, we explored the initiation and propagation of AD in slices of the rat barrel cortex using extracellular local field potential (LFP) recordings and optical intrinsic signal (OIS) recordings. We found that ischemia-like conditions induced by oxygen-glucose deprivation (OGD) evoked AD, which manifested as a large negative LFP shift and an increase in light transmittance. AD typically initiated in one or more barrels and further spread across the entire slice with a preferential propagation through L4. Elevated extracellular potassium concentration accelerated the AD onset without affecting proneness of L4 to AD. In live slices, barrels were most heavily labeled by the metabolic level marker 2,3,5-triphenyltetrazolium chloride, suggesting that the highest metabolic demand is in L4 when compared to the other layers. Thus, L4 is the layer of the barrel cortex most prone to AD, which may be due to the highest metabolic demand and cell density in this layer.
## Introduction
The brain is highly metabolically active and particularly vulnerable to metabolic insults. During global or focal ischemia, the limited supply of oxygen and glucose causes a fall in ATP levels, arrest in sodium-potassium pump activity and depolarization of neurons in the metabolically deprived brain regions (Lipton, ; Somjen, ). Release of potassium and glutamate into the extracellular space accelerates depolarization of the adjacent neurons igniting an avalanche-like wave of collective Anoxic Depolarization (AD), which shares many common features with the spreading depression (SD) described by Leao (Leao, ; Nedergaard and Hansen, ; Somjen, ; Pietrobon and Moskowitz, ; Dreier and Reiffurth, ; Hartings et al., ). Near total collective neuronal depolarization during AD is associated with large DC shifts of the extracellular local field potential, and with an increase in tissue light transmittance as a result of cellular swelling (Aitken et al., ; Joshi and Andrew, ; Somjen, ). AD is an initiator of the ischemic damage and irreversible loss of activity in the ischemic core in vivo and oxygen-glucose deprivation (OGD) induced injury in the submerged brain slices in vitro (Rader and Lanthorn, ; Tanaka et al., ; Joshi and Andrew, ) (for reviews, Martin et al., ; Lipton, ; Dreier, ). Factors that increase metabolic activity such as increased neuronal activity or elevated temperature accelerate AD onset and ischemic neuronal death, whereas reduction in metabolic demand is neuroprotective against ischemic damage (Dzhala et al., ; Joshi and Andrew, ; Tyzio et al., ).
Different neuronal populations and structures display different sensitivities to ischemia that may involve different metabolism in different cell types (Kawai et al., ; Lipton, ). In the hippocampus, CA1 pyramidal cells display the highest vulnerability to ischemia, which correlates with the preferential initiation and spread of AD in the CA1 region of the hippocampus (Aitken et al., ; Basarsky et al., ). The neocortex also displays heterogenous incidence and propagation of SD and AD between different cortical regions and layers (Bogdanov et al., ; Kaufmann et al., ). Previous studies using slices of non-identified neocortical areas revealed AD and SD “tropism” for the superficial layers 2/3 (Basarsky et al., ; Joshi and Andrew, ; Kaufmann et al., ). The somatosensory cortex and particularly its whisker-related barrel region are highly sensitive to ischemia (Lin et al., ), and SD and AD preferentially arise in and propagate through the whisker barrel region of the parietal sensory cortex (Bogdanov et al., ; Kaufmann et al., ). However, the initiation and spread of AD across the layers of the somatosensory cortex remain largely unknown.
Here, we addressed initiation and propagation of AD induced by OGD to mimic ischemia-like conditions in slices of the rat barrel cortex. We found that AD was specifically initiated in L4 barrels and its initial front preferentially propagated along layer 4. Preferential initiation and spread of AD in L4 correlated with the most intense L4 staining with TTC, a histological metabolic activity marker. We propose that sensitivity to metabolic insult is non-uniform across layers of the barrel cortex: it is highest in L4 barrels, where AD is preferentially initiated, that may involve the highest metabolic demand and cell density in this layer.
## Materials and methods
### Ethical approval
All animal-use protocols followed the guidelines of the French National Institute of Health and Medical Research (INSERM, protocol N007.08.01) and the Kazan Federal University on the use of laboratory animals (ethical approval by the Institutional Animal Care and Use Committee of Kazan State Medical University N9-2013).
### Brain slice preparation
Wistar rats (16–23 days old) of either sex were used. Animals were decapitated under isoflurane anesthesia (5%), the brain was rapidly removed and placed in ice-cold (2–5°C) slicing solution (modified from Dugué et al., ) of the following composition (in mM): K-Gluconate 140, Na-Gluconate 15, NaCl 4, EGTA 0.2, D-AP5 50 μM and HEPES 10 (pH 7.4). Four hundred μm thick thalamocortical slices were cut using a PELCO easiSlicer® vibratome (Ted Pella, Inc., Redding, CA, USA). Slices containing the barrel cortex were selected by anatomical coordinates (Khazipov et al., ) and the presence of barrel structures in L4 (Figure ). Slices were first kept in oxygenated (95% O -5% CO ) artificial cerebrospinal fluid (ACSF) of the following composition (in mM): NaCl 126, KCl 3.5, CaCl 2, MgCl 1.3, NaHCO 25, NaH PO 1.2 and glucose 11 (pH 7.4) for 30 min at 32°C and then at room temperature (20–22°C) for at least 1 h before use. For recordings, slices were placed into a submerged chamber and superfused with oxygenated ACSF at 30–32°C at a flow rate of 10 ml/min. Oxygen/glucose deprivation (OGD) was induced by superfusion with ACSF in which N replaced O and sucrose replaced glucose at equimolar concentration.
Oxygen-glucose deprivation induced anoxic depololarization in barrel cortex. (A) Scheme of the experimental setup. Submerged cortical slices is exposed to the ACSF in which oxygen is replaced by nitrogen and glucose is replaced by sucrose (oxygen-glucose deprivation, OGD) to mimic ischemic conditions. Local field potential (LFP) recordings and optical intrinsic signal (OIS) imaging in transmittance mode is performed to record anoxic depolarization (AD). (B) Microphotograph of the cortical slice. LFP is recorded from a cortical barrel, regions of interest (ROIs) in the supragranular, granular, and infragranular layers of are indicated by color boxes. (C) Traces of OIS from the ROIs as indicated on panel (B) (top), DC-LFP recordings from L4 and the amplitude of the responses in L4 evoked by stimulation of the white matter during superfusion with OGD solution. (D) Initial phase of AD outlined on panel (C) on expanded time scale. Note that onset of OIS-AD in L4 precedes the OIS in supra- and infragranular layers. Inset shows average white matter-evoked LFP response in L4 before (black) and after (red) OGD. (E) Dependence of the L4 evoked response recovery on the delay of reperfusion with normal ACSF from AD. Each point indicates the scalar integral of the response 15–30 min after reperfusion with oxygenated ACSF normalized to the control values. Black and open circles indicate OGD episodes with and without AD, respectively. The time values for the open circles were deduced from AD values in the experiments with repetitive OGD episodes, where each next OGD increased in duration until the AD was evoked. Red line indicates Boltzmann fit. (F) Snapshots of OIS at different time points after superfusion with OGD-solution. Note that the AD front preferentially propagates along L4.
### Electrophysiological recordings
Extracellular recordings of the local field potentials (LFP) were performed in the barrel cortex using single or 16-site electrodes. Single site glass pipette electrodes were pulled from borosilicate glass capillaries (BF150-86-10, Sutter Instrument, Novato, CA, USA) and had resistances of 2–3 MΩ when filled with ACSF. Electrodes were connected via chlorided silver wire to the headstage of a MultiClamp700B patch-clamp amplifier (Axon Instruments, Union City, CA, USA). Recordings were performed in voltage-clamp mode and currents were inverted and voltage calibrated using 5 mV steps. 16-channel recordings were performed using Menendez-de La Prida style 16 shank silicone probes with a separation distance of 100 μm between electrodes (NeuroNexus, Ann Arbor, MI, USA). The signals from extracellular recordings using silicone probes were amplified and filtered (1,000×; 0–9 kHz) using a Digital Lynx SX amplifier (Neuralynx, Inc., Bozeman, MT, USA), digitized at 32 kHz and saved on a PC for post-hoc analysis. Stimulating bipolar electrodes were placed in the white matter or L6 above the recorded cortical column. Voltage pulses (10–50 V, 50 μs duration, 0.1 Hz) were applied to evoke LFP responses of 100–300 μV in L4.
### Optical intrinsic signal recordings
Optical intrinsic signal (OIS) recordings were performed using slice transillumination as described in Aitken et al. ( ). The slice was illuminated by a halogen lamp with a 775 nm bandpass filter and visualized using a BX51WI upright microscope equipped with a 4×/0.10 Plan N objective (Olympus, Tokyo, Japan). Images were acquired using a QIClick-R-F-M-12 CCD camera (QImaging, Surrey, BC, Canada) usually at 174 × 130 pixel resolution and 5 frames/s acquisition rate. In some experiments a higher resolution of 348 × 260 or 696 × 520 was used.
### TTC-staining
Brain slices were stained with 1% TTC (2,3,5-triphenyltetrazolium chloride) in phosphate-buffered solution (PBS) for 2–3 min at 38°C. Then slices were rinsed in PBS (for 1 min, 3 times). Microphotographs of TTC-stained slices were obtained using a SZX16 wide zoom stereo microscope equipped with a SDF PLAPO 1 × PF objective and SZX2-ILLT LED transmitted light illumination base (Olympus, Tokyo, Japan). Images were acquired at 0.7× -1× magnification using a XC50 CCD camera (Olympus, Tokyo, Japan) at 2,576 × 1,932 pixel resolution.
### Data analysis
Data were analyzed using custom-written procedures in Matlab (MathWorks, Inc., Natick, MA, USA). OIS was calculated using the first-frame subtraction approach: OIS(t) = (I(t) – I )/I , where I(t) – pixel intensity at the moment t, I – time-averaged pixel intensity in the preconditioned baseline period (100 s). Resulting frames were filtered with a 10 × 10 median filter. Regions of interest (ROI) were selected as square areas near recording sites. OIS traces were calculated as the average OIS signal within selected ROIs.
LFP signals were downsampled to 1 kHz. Continuous running line fit was removed using local linear regression in 300 s windows with a 10 s overlap [ locdetrend function from the Chronux toolbox ( )]. Amplitude of the LFP evoked response was calculated as a negative peak value of the LFP in the 100 ms after the stimulus relative to baseline level (average of the LFP in the 10 ms before the stimulus).
Data were smoothed by the 1,000-point moving average filter and the first derivatives were calculated. Local negative peak time of the first LFP derivative was calculated within the 20 s window preceding the negative AD peak. The value of the LFP at this time was taken as 100% and the previous time corresponding to 30% considered as AD onset. Velocity of vertical AD propagation was calculated from onset values as a distance between neighboring recording sites (100 μm) divided by AD onset delays between corresponding channels. The baseline level was calculated for each recording site as the mean value of the LFP in the −20 to −10 s time window preceding AD onset. AD amplitude was calculated as the maximal negative LFP peak from the baseline. Data from different slices were aligned by L4 and average amplitude and onset depth profiles were calculated. Depth profiles of amplitude were smoothed by the 2-point moving average filter. OIS onsets and amplitudes were calculated in same manner.
Microphotographs of TTC-stained slices were analyzed as follows. Pixel intensities were calculated along the barrel cortex and along the perpendicular direction intersecting the barrel and averaged in a 100 μm wide bar. Intensity was converted to a percentage (intensity value of each pixel divided by maximal intensity). Staining efficiency (opacity) was calculated as the value inverse to the calculated intensity.
### Statistical analysis
Statistical analysis was based on the nonparametric Wilcoxon (paired samples) or Mann-Whitney (independent samples) signed rank sum test with the significance level set at p < 0.05. Results are given as means ± SEM.
## Results
### Electrophysiological and optical intrinsic signals during anoxic depolarization
In the present study, we explored spatial-temporal dynamics of the OGD-induced AD in slices of the barrel cortex using extracellular recordings of LFP, and OIS recordings (Figures ). AD was initiated within 6–13 min (9.5 ± 0.5 min; n = 17 slices from 8 rats) and manifested as a sharp increase of light transparency attaining 29.2 ± 3.0% dI/I and negative LFP shift of 8.9 ± 0.6 mV in L4 ( n = 17; Figure ). LFP signals were typically biphasic with an initial sharp negative transient followed by a secondary negative wave. The increase in OIS during AD started in L4 and further spread to L2/3 and L5/6 (Figure ). The responses evoked in L4 by stimulation of the white matter or L6, progressively decreased during OGD and were completely and irreversibly abolished during and after AD (Figure ), while the evoked responses could recover following shorter OGD episodes without AD (Figure ) that is in keeping with the results of previous studies (Rader and Lanthorn, ; Tanaka et al., ; Joshi and Andrew, ). In the experiment illustrated in Figure , OIS recordings revealed that AD unilaterally propagated through the slice with the leading front in L4 and delayed fronts in the supra- and infragranular layers (Figure ). The increase in light transmittance was followed by a decrease in light transmittance probably reflecting cellular swelling followed by dendritic beading (Aitken et al., ; Joshi and Andrew, ; Somjen, ). After AD in the barrel cortex, AD was also observed in the hippocampus and striatum after a several minute delay (Figure ).
### Vertical AD propagation in a cortical barrel column
We also performed simultaneous OIS and multisite LFP recordings from a cortical barrel column using 16-shank silicone probes (Figures ). In keeping with the results described above, AD was initiated in L4 and spread to L2/3 and L5/6 with a velocity of 4.0 ± 0.1 mm/min ( n = 7 slices from 4 rats). The maximal amplitude of negative LFP shift during AD was observed in the superficial layers (13.8 ± 0.8 mV at a depth of 400 μm from the cortical surface; n = 7; Figures ). In L4 and L5/L6 the amplitude of AD was 12.0 ± 0.6 and 10.7 ± 0.7 mV, respectively ( n = 7). The OIS profile of AD was remarkably similar to that of the electrophysiological response including an initial onset in L4 and vertical delays in the supra- and infragranular layers, the speed of vertical propagation, and a maximal amplitude of light transmittance increase of 38.5 ± 10.3% ( n = 7) attained in L2/3 and smaller change in deeper layers (Figures ).
Spread of anoxic depolarization within a cortical barrel column. (A) IR-DIC microphotograph of a slice of the barrel cortex with a 16-shank silicone probe (100 μm electrode separation distance) vertically placed to record DC-coupled LFP from all layers of a cortical barrel column. Corresponding OIS-ROIs are outlined by white boxes. (B) Microphotograph of the brain slice shown on panel (A) on a reduced scale (left) and snapshots of OIS at different time points after OGD induction. (C) Corresponding DC-LFP traces (black) during AD propagation at different depths of the cortical column with the AD onsets marked at each channel with a red circle (left) and the same DC-LFP traces overlaid on the color-coded current-source density map (right). (D) Group data on OGD-induced AD onsets and AD amplitudes as a function of cortical depth (mean ± SE, n = 7). (E) Concomitant OIS recordings of AD (left) and color-coded transparency map calculated from OIS traces (right) in the ROIs indicated on panel (A) . Red circles indicate the OIS-AD onset. (F) Group data on the onsets and amplitudes of OIS associated with OGD-induced AD as a function of cortical depth (mean ± SE, n = 7). Note that AD first occurs in L4 and further spreads to L2/3 and L5/6 and that AD amplitude is maximal in L2/3.
### Patterns of AD initiation and propagation
OIS imaging revealed variability of the AD initiation and propagation patterns, which could be classified in four main groups (Figure and Videos – ):
Single-barrel AD initiation (Figure ; Video ). AD emerges in one barrel within the imaging window and spreads concentrically. The AD spread is often anisotropic with a preferential horizontal propagation along L4 forming a characteristic “bird head” OIS image. This pattern was observed in 11 of 30 slices.
Multiple-barrel AD initiation (Figure ; Video ). AD emerges in two (or more) barrels within the imaging window. AD fronts move concentrically and collide first in L4 and then in the superficial and deep layers ( n = 6 of 30 slices).
One side propagating AD (Figure ; Video ). AD originates on one side of slice but outside of the imaging window and spreads through the slice with a preference to L4 ( n = 8 of 30 slices).
Double-side propagating AD (Figure ; Video ). AD emerges on two sides of slice outside of the imaging window. Two AD waves move toward each other with a preference to the L4 and collide similarly to the multi-barrel initiation pattern ( n = 5 of 30 slices).
Various patterns of AD initiation and propagation in barrel cortex. (A) Microphotograph of a barrel cortex slice in DIC-IR (left) and OIS snapshots at different time points after OGD induction. Below, the corresponding AD fronts are presented as the OIS first derivative. Bottom left panel shows time color coded AD front contours plotted at 3 s intervals. Note that AD is initiated in one barrel and further propagates preferentially through L4. (B) Similar example of AD which is initiated in a barrel located in the middle of the imaging window and which collides with another AD wave arriving from another locus on the right of the imaging window. (C) Example of AD propagating preferentially through layer 4 from the left side of the imaging window. (D) Example of AD propagating from the left and right sides and colliding in the middle of the imaging window in L4.
These results indicate that despite variety in the site of OGD-induced AD initiation in a slice, preference of AD to L4 is a hallmark of all initiation and propagation AD patterns. The rate of AD propagation along L4 was 1.7 ± 0.1 mm/min ( n = 18 slices from 8 rats) which is consistent with the rate of AD and SD propagation in slices and in the intact brain in vivo (Nedergaard, ; Basarsky et al., ; Joshi and Andrew, ). In the cases of single-barrel AD initiation the rate of medial AD propagation along L4 was of 1.8 ± 0.1 mm/min that was not different from the rate of lateral AD propagation of 1.7 ± 0.1 mm/min ( n = 11 slices from 4 rats, p = 0.76).
We next addressed a question of whether AD in L4 is a necessary condition for emergence of AD in the supragranular and infragranular layers. In this aim, we explored OGD-induced AD after surgical cuts made above, below and through the L4 (Figure and Videos – ). We found that AD efficiently invaded supragranular and infragranular layers even after disconnection from L4, with an AD front moving horizontally around the cuts thus indicating that both superficial and deep layers are capable of generating AD independently from L4. We further calculated the time difference between AD in the surface and deep layers at the areas vertically aligned to the middle of the cut. When the cut was made above L4, AD arrived to L2/3 88 ± 19 s later than to L5/6 ( n = 4). When the cut was made below L4, AD in L2/3 emerged 49 ± 6 s earlier than in L5/6 ( n = 5). Thus, AD was generated earlier in the layers maintaining connection with L4 than AD in the layers disconnected from L4. With the cut made through L4, the time delay between AD in L2/3 and L5/6 reduced to 12 ± 4 s ( n = 4). Together, these results indicate that AD in L4 is not a necessary condition for AD in the surface and deep layers, where AD can propagate horizontally. However, vertical AD vector originating from L4 accelerates AD in the supragranular and infragranular layers maintaining their connection with L4.
AD propagation after surgical cut above, below and through L4. (A–C) Example microphotographs of the barrel cortex slices in DIC-IR (left) with the cuts made above (A) , below (B) , and through (C) the layer 4 and OIS snapshots at different time points after OGD induction. Below, the corresponding AD fronts are presented as the OIS first derivative. Bottom left panel shows time color coded SD front contours plotted at 3 s intervals. Note that AD waves invade supragranular and infragranular layers in all cases but AD emerges earlier if connection with L4 is preserved. AD propagation to L2/3 above the cut on panel (A) was too slow and is truncated on the contour map. See Video for the entire AD wave in this experiment.
### Elevated extracellular potassium concentration accelerates the AD onset
Hyperactivity compromises the metabolic state of the tissue under OGD-conditions and accelerates the AD onset in hippocampus (Dzhala et al., ). With the aim of exploring the effect of increased activity on the OGD-induced AD in the barrel cortex, we elevated extracellular potassium concentration in ACSF from 3.5 to 8.5 mM. The high-potassium solution itself induced a slight increase in light transmittance and a negative shift in the LFP baseline in L4 (Figure ). Further superfusion with high-potassium/OGD solution evoked AD with a delay of 5.9 ± 0.4 min ( n = 13 slices from 5 rats), that was almost two-fold quicker ( p < 0.001) than AD evoked by OGD in normal potassium conditions (9.5 ± 0.5 min; n = 17 slices from 8 rats) (Figure ). In the high-potassium/OGD solution the negative LFP shift of 8.5 ± 0.6 mV ( n = 13) in L4 was similar to those in normal potassium conditions ( p > 0.05) while the increase in light transparency of 19.8 ± 1.0% dI/I ( n = 13) was less than in normal conditions ( p < 0.05), due to the progressive increase of dI/I before AD which occurs in elevated potassium conditions (Figure ). OIS imaging revealed that preference of AD initiation and propagation in L4 was maintained under conditions of elevated potassium (Figure ).
Elevation of extracellular potassium accelerates the onset of anoxic depolarization. (A) OIS from L4 (top trace) and nearby L4 DC-coupled LFP (bottom trace) during superfusion with ACSF with elevated (from 3.5 to 8.5 mM) potassium concentration and after OGD induction. (B) Corresponding microphotograph (left) and OIS snapshots at different time points after OGD induction. (C) Group data on AD onsets in control ACSF ( n = 17) and in the ACSF with potassium concentration elevated to 8.5 mM ( n = 13). Each white circle corresponds to one slice and black circles show the mean ± SE. Note that AD onset is accelerated almost two-fold after elevation of extracellular potassium concentration. p < 0.001.
### Highest metabolic activity in L4
Preferential initiation and propagation of AD in the barrels may involve the higher metabolic demand of barrels and therefore their higher vulnerability to metabolic deprivation. We explored this hypothesis using 2,3,5-triphenyltetrazolium chloride (TTC) staining of live slices of the barrel cortex. As shown on Figures , TTC most intensively stained L4. Quantification of TTC-staining along the horizontal projection in L4 revealed peaks in TTC-staining corresponding to neighboring barrels (Figure ). In the vertical projection across cortical depth, TTC staining peaked at the L4 depth (Figure ). A second, less intense peak was found ~0.5 mm deeper at the L5B/L6A border (Figure ). Cross-layer comparisons revealed significantly higher TTC-staining of L4 compared to L2/3 ( p < 0.05) and L5/6 ( p < 0.05) ( n = 7 slices from 4 rats; Figure ). Slices that had been exposed to OGD for 30 min and reperfused with normal ACSF for 2 h displayed only weak non-specific staining (Figure ). TTC-staining of L4 in OGD-exposed slices revealed no difference with L2/3 ( p > 0.05) and L5/6 ( p > 0.05) staining and was significantly lower compared to control slices ( p < 0.01; n = 9 slices from 4 rats) (Figure ).
TTC-staining of a live slice of the barrel cortex shows highest metabolic activity in L4. (A) Microphotograph of a slice of barrel cortex stained with TTC. (B–D) Intensity of TTC-staining within a region outlined by dashed box on panel (A) in (B) horizontal projection along the L4 (indicated by H-line on panel D ) and (C) vertical projection across cortical layers of a barrel column (indicated by V-line on panel D ). Note the horizontal barrel staining pattern and maximal staining of L4 in vertical projection. (E) Group data on the intensity of TTC-staining of different cortical layers in control slices ( n = 7) and slices after OGD exposure ( n = 9). p < 0.05; p < 0.01. (F) Microphotograph of a TTC-stained slice of barrel cortex after OGD exposure. Note weak non-specific staining compared with the live slice.
## Discussion
The principal conclusion emerging from the present study is that different layers of the barrel cortex differ in their propensity to AD and that L4 is the most prone to AD. This conclusion is supported by electrophysiological recordings and OIS imaging indicating that OGD-induced AD is preferentially initiated in, and preferentially spreads through L4. We also found that enhanced L4 susceptibility to OGD correlates with the highest metabolic activity, in L4, revealed with TTC staining of live slices.
Anisotropy is a characteristic feature of the heterogeneous incidence and horizontal spread of cortical SD in vivo (Kaufmann et al., ). Previous studies in non-identified neocortical areas revealed anisotropy of AD and SD across cortical layers with a “tropism” to the superficial layers 2/3 (Basarsky et al., ; Joshi and Andrew, ; Kaufmann et al., ). However, our findings indicate that in the barrel cortex, which contains large barrels and the thickest L4 of all the cortical regions, AD is initiated and preferentially propagates via L4. Onset and preferential propagation of AD in L4 was evidenced by the earliest onset of the negative LFP DC shift and the earliest increase in optical transparency in barrels during the OGD-induced AD. Multisite LFP recordings and simultaneous OIS imaging revealed vertical spread of AD from the L4 to the superficial and deep layers within a column. OIS recordings also enabled us to assess two-dimensional spatial-temporal AD dynamics in the barrel cortex slices revealing a variety of AD initiation and propagation patterns, yet with a common delimiter of the highest proneness of L4 to AD. While AD primarily originated in L4 in the barrel cortex, L4 appeared to be not necessary for the emergence of AD in the supragranular and infragranular layers, however. Indeed, our experiments with surgical cuts above, below and through the L4 revealed that AD may propagate through these layers horizontally around the cuts, although at lower speed. This indicates that supra- and infragranular layers are capable of generating AD independently from the L4. Yet, early ignition of AD in L4 is important for driving AD in the supra- and infragranular layers in the intact slice.
Various factors have been suggested to explain anisotropy of SD and AD (Herreras and Somjen, ; Somjen, ; Canals et al., ; Kaufmann et al., ). High neuronal density in L2/3 has been hypothesized to promote a mutual promotion of depolarization and potassium release and accumulation, making these layers more prone to AD (Joshi and Andrew, ). In the barrel cortex, the highest neuronal density is observed in L4, where it attains 124 thousand neurons/mm compared to 102 and 86 thousand neurons/mm in L3 and L2, respectively (Meyer et al., ). Thus, our findings of the preferential initiation and spread of AD in L4 of the barrel cortex are consistent with the “neuronal density” hypothesis. Interestingly, SD and AD preferentially arise in and propagate through the whisker barrel region of parietal sensory cortex in vivo (Bogdanov et al., ; Kaufmann et al., ). Because the barrel cortex contains large barrels and the thickest L4 of all the somatosensory cortical regions, the elevated proneness of L4 to AD as revealed in the present study may also explain high proneness of the barrel cortex to AD.
Our observations of heterogeneous TTC-staining in different cortical layers with the maximum in L4 suggest that elevated metabolic demand could also be a factor contributing to the particular susceptibility of this layer to AD. TTC staining intensity is determined by the metabolic activity of mitochondrial dehydrogenases, which enzymatically convert colorless TTC to red formazan (Goldlust et al., ). The elevated L4 metabolic activity revealed with TTC-staining is consistent with the highest density of a mitochondrial enzyme cytochrome oxydase and elevated number of mitochondria in L4 of the barrel cortex where they reside mainly in dendrites and axonal terminals (Wong-Riley and Welt, ). Considerable evidence indicates that elevation of metabolic debt strongly aggravates ischemic insults. Indeed, various factors increasing the metabolism such as an increase in neuronal activity caused by adenosine A1 receptor antagonists, blockers of GABA(A) receptors and potassium channels (Dzhala et al., ), elevation of extracellular potassium as in the present study or elevated temperature (Joshi and Andrew, ) strongly accelerate the AD onset. Thus, due to elevated metabolic activity, L4 neurons are most likely to quickly lose ATP, depolarize and ignite AD in metabolically-compromised conditions. The question then arises: why the metabolic activity is highest in L4 barrels? Although the underlying mechanisms are unknown, it could be suggested that it involves a particular cytoarchitectonic and synaptic barrel organization. Indeed, densely packed excitatory and inhibitory neurons form a highly interconnected network in L4 barrels (Feldmeyer et al., ; Lefort et al., ; Valiullina et al., ) that may impose a higher metabolic charge to equilibrate the ionic disturbances caused by the activity in this layer.
Thus, in the present study we have shown that different cortical layers differ in their sensitivity to metabolic insult with L4 being the most prone to AD initiation and preferential propagation. Elevated ongoing metabolic demand of L4 could be a factor contributing to this enhanced sensitivity of L4 to OGD. Our findings also support rationale of the strategies aimed to reduce the metabolic demand as an approach to alleviate ischemic brain damage.
## Author contributions
RK conceived the project. EJ and MM performed the experiments. AN, MM, EJ, AG, and MS analyzed the data. RK wrote the paper.
### Conflict of interest statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
The innate immune Toll-like receptor (TLR) family plays essential roles in cell proliferation, survival and function of the central nervous system. However, the way in which TLRs contribute to the development and maintenance of proper retinal structure and function remains uncertain. In this work, we assess the effect of genetic TLR4 deletion on the morphology and function of the retina in mice. Visual acuity and retinal responsiveness were evaluated in TLR4 knockout and wild type C57BL/6J control mice by means of an optomotor test and electroretinography, respectively, from P20 to P360. Retinal structure was also analyzed in both strains using confocal and electron microscopy. ERG data showed impaired retinal responsiveness in TLR4 KO mice, in comparison to wild type animals. The amplitudes of the scotopic a-waves were less pronounced in TLR4-deficient mice than in wild-type animals from P30 to P360, and TLR4 KO mice presented scotopic b-wave amplitudes smaller than those of age-matched control mice at all ages studied (P20 to P360). Visual acuity was also relatively poorer in TLR4 KO as compared to C57BL/6J mice from P20 to P360, with significant differences at P30 and P60. Immunohistochemical analysis of retinal vertical sections showed no differences between TLR4 KO and C57BL/6J mice, in terms of either photoreceptor number or photoreceptor structure. Horizontal cells also demonstrated no morphological differences between TLR4 KO and wild-type mice. However, TLR4 KO mice exhibited a lower density of bipolar cells (15% less at P30) and thus fewer bipolar cell dendrites than the wild type control mouse, even though both confocal and electron microscopy images showed no morphologic abnormalities in the synaptic contacts between the photoreceptors and second order neurons. Microglial cell density was significantly lower (26% less at P30) in TLR4 KO mice as compared to wild-type control mice. These results suggest that TLR4 deletion causes functional alterations in terms of visual response and acuity, probably through the loss of bipolar cells and microglia, but this receptor is not essential for the processing of visual information in the retina.
## Introduction
Toll-like receptors (TLRs) are type I transmembrane proteins that mediate innate immune responses triggered by pathogen-associated (PAMP) and damage-associated molecular pattern (DAMP) molecules ( ; ; ). Ten human and thirteen mammalian TLRs have been characterized to date ( ; ; ), and are primarily expressed in tissues associated with the immune function, although they are also expressed by other non-immune cells including glia and neurons ( ). It is now evident that, far from their primary well-known function as mediators of the immune response, TLRs play additional relevant roles in the progression of different pathological conditions, such as non-infectious neuronal injuries and neurodegenerative disorders ( ; ; ; ). Interestingly, recent findings show that TLRs also appear to be involved in many physiological cellular processes, such as the development of neuronal circuits during embryogenesis, progenitor cell proliferation and differentiation, CNS plasticity throughout the entire life or drug-reward behavior ( ; ; ; ; ; ). So far, numerous endogenous TLR ligands have been identified that could play a role in the regulation of CNS physiological processes and in the development and progression of neurodegenerative disorders ( ).
Toll-like receptors also influence neuronal survival. This effect depends on the receptor subtype and localization, and can be directly exerted in the brain and the retina through TLR activation in neurons ( , ; ), progenitor cells ( ) and glia ( ). The activation of TLR4 in retinal microglia and Müller cells can directly affect the viability of photoreceptors ( ). In addition, several studies have shown that TLR4 modulates neuronal survival under conditions such as CNS non-infectious diseases and injuries, neuroinflammation and degenerative disorders like multiple sclerosis, Alzheimer’s and Parkinson’s diseases, and amyotrophic lateral sclerosis ( ; ; ; ; ; ). TLR4 might also affect the survival of newborn neurons. Intraperitoneal injection of the natural TLR4 ligand lipopolysaccharide (LPS) during the neonatal period has been shown to decrease the survival of dividing astrocytes and neurons in the hippocampus of mice ( ). But, interestingly, activation of TLR4 before the occurrence of a harmful stimulus might induce preconditioning and has a neuroprotective effect, which opens the door to new therapeutic approaches using TLRs agonists ( ; ).
In the mammalian retina, TLR expression has been demonstrated in astrocytes ( ), microglia ( ), Müller cells ( ; ; ), photoreceptors ( ), bipolar, amacrine and ganglion cells ( ; ), retinal pigmented epithelium ( ; ) and vascular endothelial cells ( ). TLRs mediate the immunological responses in the eye ( ; ; ; ; ; ). But, to date, there is a lack of information regarding their influence on retinal development and neuronal plasticity. A previous work from Shechter and collaborators revealed that TLR4 acts as a negative regulator of the proliferation and neuronal differentiation of retinal progenitor cells in the ciliary margin and ora serrata ( ) in the early postnatal period. However, no studies have yet been conducted to determine whether TLR4 deletion affects the morphology and function of the mouse retina during growth, maturation and aging.
Given the role of TLRs in neurogenesis, proliferation of progenitor cells and neuronal differentiation, the main objective of this work was to determine whether TLR4 deletion had any effects on the structure and/or function of the mouse postnatal retina, and if so, whether these potential effects would vary throughout growth, maturation and aging. Increasing our knowledge of the involvement of TLRs in physiological and pathological conditions may provide new therapeutic options for both infectious and non-infectious diseases.
## Materials and Methods
### Animals
TLR4 KO mice, kindly provided by Dr. M.L. Gil and Dr. D. Gozalbo (Universitat de València, Spain), were employed in this study ( ). Age-matched wild-type C57BL/6J mice (Harlan Laboratories, Barcelona, Spain) represented the control animals. All animals were housed in cages under controlled photoperiod (12 h light/12 h dark), temperature (23 ± 1°C) and humidity (55 to 60%). Food and water were provided ad libitum . All the procedures were performed according to Project License UA-2013-07-22, which was approved by the Ethics Committee for Animal Experimentation from the University of Alicante. All animals were treated according to current regulations on the use of laboratory animals (NIH, ARVO and European Directive 2010/63/UE) in an effort to minimize animal suffering and the number of animals used.
### Electroretinography
We recorded scotopic and photopic ERG responses at ages P20 and 1, 2, and 12 months, essentially as has been previously described ( ). After a period of adaptation to overnight darkness, procedures were carried out to prepare the animals for bilateral ERG recording under dim red light. An intraperitoneal injection of ketamine (100 mg/kg) and xylazine (4 mg/kg) was administered as anesthesia, and the animals were kept on a heating pad at 38°C. A topical application of 1% tropicamide (Alcon Cusí, Barcelona, Spain) was administered to dilate their pupils. A drop of Viscotears 0.2% polyacrylic acid carbomer (Novartis, Barcelona, Spain) was placed on the cornea to facilitate electrical contact with the recording electrodes and to prevent dehydration. DTL fiber electrodes were used, consisting of an X-Static silver-coated nylon conductive strand and supplied by Sauquoit Industries (Scranton, PA, United States). The reference electrode was a 25-gauge platinum needle placed between the eyes under the scalp. A gold electrode placed in the animal’s mouth served as ground. Anesthetized animals were put in a Faraday cage and all experiments were conducted in total darkness. Scotopic flash-induced ERG responses to light stimuli produced by a Ganzfeld stimulator were recorded in both eyes. Light stimuli were administered at 11 different increasing luminances (ranging from -5.2 to 0 log cd⋅s/m ) for 10 ms each. The mean of three to ten consecutive recordings was calculated for each light administration. An interval of 10 s was left between flashes for dim flashes (-5.2 to -1.4 log cd⋅s/m ) and as much as 20 s for higher luminances (-0.8 to 0 log cd⋅s/m ). After a 20-min period of light adaptation at 10 cd/m , photopic responses were obtained using the same stimuli as for scotopic conditions. ERG signals were first amplified and then band-pass filtered (1–1000 Hz, without notch filtering) by means of a DAM50 data acquisition board (World Precision Instruments, Aston, United Kingdom). A PowerLab system (AD Instruments, Oxfordshire, United Kingdom) was used for the administration of stimuli and data acquisition (4 kHz). Recordings were saved to a computer file and were subsequently analyzed off-line. To visualize oscillatory potentials, the signal recorded was filtered between 100 and 1000 Hz. The a-wave amplitude measurement was taken from the baseline 10 ms after the onset of the light stimulus, i.e., prior to the intrusion of the b-wave. The b-wave amplitude measurement was taken from the trough of the a-wave to the peak of the b-wave. For oscillatory potentials the maximum peak-to-trough amplitude was considered.
### Optomotor Test
The spatial frequency threshold was assessed for awake, freely moving TLR4 KO and C57BL/6J mice at ages P20 and 1, 2, and 12 months. The Argos system (Instead, Elche, Spain) was used to observe and score optomotor responses to horizontally drifting, vertically oriented gratings. The spatial frequency threshold for the behavior was considered to be the maximum spatial frequency at maximum contrast that was still capable of inducing smooth head tracking movements. For this test, a mouse was positioned on a platform in the center of a chamber whose sides were four computer monitors. Sinusoidal gratings were projected on all monitors as a virtual cylinder centered on the head and rotating in both horizontal directions. An overhead infrared video camera was used by a trained observer to record the mouse and score smooth head turns in response to the rotating gratings. These tracking responses were observed to be robust at middle spatial frequencies and diminished until they disappeared at the threshold.
### Retinal Histology
Retinal histological studies were carried out at P20, P30 and 12 months of age following well-established procedures ( ). Briefly, the animals were administered a lethal dose of pentobarbital in the mornings. After marking the dorsal margin of the limbus by means of a suture, the eyes were enucleated and fixed in 4% (w/v) paraformaldehyde during 1 h at room temperature. They were then washed in 0.1 M phosphate buffer pH 7.4 (PB) and sequentially cryoprotected in 15, 20, and 30% (w/v) sucrose. The cornea, lens and vitreous body were excised, and the eyecups were then embedded in Tissue-Tek OCT (Sakura Finetek, Zoeterwouden, Netherlands) and frozen in liquid N . Sections with a thickness of 16 μm were obtained at -25°C in a cryostat, mounted on slides (Superfrost Plus; Menzel GmbH and Co. KG, Braunschweig, Germany) and stored at -20°C. Before subsequent use, slides were thawed, washed 3 times with PB and incubated for 1 h with blocking solution [10% (v/v) donkey serum and 0.5% (v/v) triton X-100 in PB]. For single or double immunostaining, sections were incubated overnight at room temperature with combinations of antibodies at different dilutions in PB with 0.5% Triton X-100. The primary antibodies used (see ) have already been characterized in other works. For an objective comparison, TLR4 KO and C57BL/6J retinas were processed in parallel. Donkey anti-mouse or anti-rabbit IgG conjugated to Alexa Fluor 488 or 555 (1:100, Molecular Probes, Eugene, OR, United States) were used as secondary antibodies. In some cases, the nuclear marker TO-PRO 3 iodide (Molecular Probes) was added at a dilution of 1:1000. Images were taken under a Leica (Leica Microsystems, Wetzlar, Germany) TCS SP2 confocal laser-scanning microscope using a spatial resolution of 1024 × 1024. The pinhole was set at 1 airy unit and z-stacks were made of 15 pictures (1.5 μm steps). All stacks were taken under 40× magnification, with an acquisition rate of 16 frames per second. Adobe Photoshop 10 software (Adobe Systems Inc., San Jose, CA, United States) was used to process the final images from C57BL/6J and TLR4 KO groups in parallel.
Primary antibodies employed in this work.
### Transmission Electron Microscopy
Twelve-month-old C57BL/6J and TLR4 KO mice were perfused with 4% paraformaldehyde and 2% glutaraldehyde in 0.1 M PB. Eyes were enucleated and immersed in the same solution for 2 h. After 2 rinses in 0.1 M PB, eyes were dissected, excising the cornea, iris, lens and vitreous body, and cut into four pieces. Each piece was postfixed in 1% osmium tetroxide (OsO ) in 0.1 M PB for 1 h. After being gradually dehydrated in a sequence of ethanol and acetone solutions, the pieces were embedded in EPON 812 overnight. EPON blocks containing the retinal pieces were polymerized at 60°C overnight. Ultrathin sections were obtained with an ultramicrotome (Leica Ultracut R, Leica Microsystems), and lead citrate and uranyl acetate were used for contrast. Transmission electron microscopy images were obtained in a 120 kV JEOL JEM-1400 microscope (JEOL GmbH, München, Germany).
### Morphometric Analysis
Measurement of the thickness of the retinal layers, ribbon density and retinal cell counting were performed using the NIH ImageJ software developed by Wayne Rasband (National Institutes of Health, Bethesda, MD, United States ). Differences in immunofluorescence signals were analyzed by obtaining the corresponding grayscale intensity (range 0–256) profile plots, using the NIH ImageJ software, and quantifying the area (in pixels) under the intensity profile. In all sections analyzed, quantification was performed close to the optic nerve. At least four animals per experimental group were analyzed.
For the quantification of bipolar cell dendrites, four square zones of 4047 pixel square were analyzed per image in the OPL. For ganglion cell counting, in each flat mounted retina twelve equidistant regions of 0.144 mm in the temporal-nasal and superior-inferior axes. Six regions were arranged on the superior-inferior axis and six fields were disposed on the temporal-nasal axis; representative peripheral, medial and central areas of the superior, inferior, temporal and nasal quadrants of each retina were evaluated.
### Statistical Analysis
A two-way ANOVA was applied to assess the effects of genotype (TLR4 KO vs. C57BL/6J) and experimental stage (P20 and 1, 2, and 12 months old), as well as any interactions between them. Post hoc pairwise comparisons using Bonferroni’s test were carried out when a 0.05 level of significance was obtained. Normal distributions and homogeneity of variance were seen for the categories of the previously defined variables. Values of P < 0.05 were considered to be statistically significant. Data were plotted as the average ± standard error of the mean (SEM). All statistical analyses were conducted using SPSS 22.0 software (Statistical Package for Social Sciences, Chicago, IL, United States).
## Results
### TLR4 Deletion Decreases Retinal Responsiveness
To assess the effect of TLR4 deletion on the functioning of the mouse retina, scotopic and photopic flash-induced ERG responses were recorded in both TLR4 KO and wild type animals at P30 ( ), an age when the retina is fully developed. Under scotopic conditions, ERG responsiveness was lower in TLR4 KO mice than in the wild type. The maximum amplitudes observed for scotopic a- and b-waves in TLR4 KO animals were 91 and 87% of the values obtained for wild type animals (ANOVA, Bonferroni’s test; P < 0.05 (scotopic a-waves); P < 0.001 (scotopic b-waves); n = 14 for TLR4 KO mice and n = 21 for wild type animals). No significant differences were observed in the a- and b-wave latencies. Under photopic conditions, no significant differences were found between both groups, with maximum amplitudes of photopic a- and b-waves in TLR4 KO animals being 100 and 102% of the values obtained for wild type animals.
Electroretinographic responses in control and TLR4-deficient mice. Representative scotopic (A) and photopic (B) ERG traces obtained from control (C57BL/6J) and TLR4 KO mice at P30. Units on the left indicate the luminance of the flashes in log cd⋅s/m . Note that scotopic ERG responsiveness was lower in TLR4 KO mice, as compared to the control mice.
To assess whether the detrimental effect of TLR4 deletion on retinal responsiveness was exacerbated with aging, scotopic flash-induced ERG responses were recorded at different ages (P20 and 1, 2, and 12 months) in both TLR4 KO and wild type animals. , show that ERG responsiveness progressively decreased from P20 to 12 months in both control animals ( , ) and TLR4 KO mice ( , ). However, the maximum a-wave amplitudes observed in TLR4 KO mice under scotopic conditions were lower than in control animals at all ages tested ( – ), with a decrease of 3.6, 9.0, 27.7, and 17.2% at P20 and 1, 2, and 12 months, respectively, which were significantly different at 1, 2, and 12 months of age (ANOVA, Bonferroni’s test, P < 0.05 to P < 0.001, as indicated in – ; n = 6 to n = 21 in each group). Also, for scotopic b-waves, the maximum amplitudes recorded were lower in TLR4 KO mice at all ages tested ( – ) (decrease of 15.9, 13.4, 23.4, and 14.7% at the corresponding ages), reaching significance at all ages tested (ANOVA, Bonferroni’s test, P < 0.05 to P < 0.001, as indicated in – ; n = 6 to n = 21 in each group). Oscillatory potentials in the scotopic ERG from TLR4 KO mice also showed a significantly lower amplitude, compared with control animals, at all ages tested ( ).
Scotopic a-wave luminance-response curves in control and TLR4-deficient mice. (A,B) Response curves for mixed scotopic a-waves obtained from control ( A , C57BL/6J) and TLR4 KO ( B , TLR4-/-) mice at different ages, as indicated. (C–F) Paired comparisons between C57BL/6J and TLR4-/- mice responses for each light stimulus at P20 (C) , P30 (D) , P60 (E) , and P360 (F) . The comparison showed significant differences between the two groups tested at P30, P60, and P360 (ANOVA, Bonferroni’s test; n = 6 to n = 21 in each group). P < 0.05, P < 0.01, P < 0.001. Error bars represent the SEM.
Scotopic b-wave luminance-response curves in control and TLR4-deficient mice. (A,B) Response curves for mixed scotopic b-waves obtained from control ( A , C57BL/6J) and TLR4 KO ( B , TLR4–/–) mice at different ages, as indicated. (C–F) Paired comparisons between C57BL/6J and TLR4–/– mice responses for each light stimulus at P20 (C) , P30 (D) , P60 (E) , and P360 (F) . Significant differences are evident between the two groups tested at all conditions (ANOVA, Bonferroni’s test; n = 6 to n = 21 in each group). P < 0.05, P < 0.01, P < 0.001. Error bars represent the SEM.
Oscillatory potentials amplitude and visual acuity in control and TLR4-deficient mice. (A) Amplitude of maximum oscillatory potentials from control and TLR4 KO mice (ANOVA, Bonferroni’s test; n = 6 to n = 21 in each group). (B) Spatial frequency threshold (in cycles per degree) in the optomotor test at different ages in control (C57BL/6J, gray) and TLR4-deficient (TLR4–/–, black) mice. Visual acuity was slightly lower in TLR4 KO mice, with a significant difference observed between the two groups at P30 and P60 (ANOVA, Bonferroni’s test; n = 4 to n = 16 in each group). P < 0.05, P < 0.01, P < 0.001. Error bars represent the SEM.
Visual acuity also changed with age in wild type and TLR4 KO animals (P20 and 1, 2, and 12 months), with the mice reaching maximum acuity at 2 and 1 months, respectively ( ). However, visual acuity values were slightly lower in TLR4 KO mice than in control animals from P20 to 12 months of age, with the differences being statistically significant at 1 and 2 months of age (10.8 and 17.9% less acuity, respectively; ANOVA, Bonferroni’s test, P < 0.05 and P < 0.001, respectively; ; n = 4 to n = 16 in each group). Altogether, these results demonstrate that TLR4 deletion decreases retinal responsiveness in mice.
### TLR4 Deficiency Has No Effect on the Number and Morphology of Photoreceptor Cells
Given the decrease in retinal functionality detected in TLR4 KO mice, we have investigated whether morphological alterations could be behind this abnormal response. Different immunohistochemical analyses were performed on wild type and TLR4 KO animals. We first determined the average number of photoreceptor rows found in the outer nuclear layer (ONL) of each retina, employing the nuclear dye TO-PRO 3 iodide ( ). Because ONL thickness varies throughout the retina, we examined the effects of TLR4 deletion in several retinal areas, ranging from the temporal to the nasal regions. We found that the average number of photoreceptor rows was similar in control and TLR4 KO animals at P30 (14.8 ± 0.1 and 15.3 ± 0.4, respectively, n = 5 in both cases; ). Similar results were found at P20 (not shown). To evaluate photoreceptor morphology, we labeled the cones using antibodies against cone arrestin ( ) and observed no morphological alterations in TLR4 KO mice as compared to the wild type. Immunofluorescence images of rhodopsin did not show differences between TLR4 KO and control mouse retinas ( ).
Number and morphology of photoreceptors in control and TLR4-deficient mice. (A–F) Vertical sections from control ( A,C,E ; C57BL/6J) and TLR4-deficient ( B,D,F ; TLR4–/–) mice retinas stained with the nuclear marker TO-PRO 3 (blue) to visualize all cell nuclei (A,B,E,F) , and labeled for cone arrestin (green) to visualize cone photoreceptors (C,D) or for rhodopsin (red) to evidence rod outer segments (E,F) . (G) Quantitation of photoreceptor rows in the ONL in both C57BL/6J and TLR4–/– retinas ( n = 5 in both cases). The scheme to the right of the panel shows the position of each representative region analyzed in the retina. Error bars represent the SEM. OS, outer segment; IS, inner segment; ONL, outer nuclear layer; OPL, outer plexiform layer; INL, inner nuclear layer; IPL, inner plexiform layer; GCL, ganglion cell layer; ON, optic nerve. Scale bar: 40 μm.
### Absence of TLR4 Reduces the Density of Bipolar Cells and Bipolar Cell Dendrites
Despite the fact that the absence of TLR4 affected neither the number of photoreceptors nor their morphology, retinal responsiveness was impaired, so we wondered whether neurons postsynaptic to photoreceptors, such as bipolar and horizontal cells, could be affected by the lack of TLR4. Horizontal cell bodies are found in the outermost part of the inner nuclear layer (INL). They establish connections with both rod and cone photoreceptors. The only horizontal cells found in the murine retina belong to a subtype that can be labeled using antibodies against calbindin. Calbindin immunostaining of TLR4 KO and wild type retinas showed no differences between the experimental groups ( ) as regards the morphology and density of horizontal cell dendrites. Interactions between horizontal cell dendrites and photoreceptor synaptic ribbons also showed no differences between control and TLR4-deficient mice ( , respectively). Quantification of the horizontal cell number at P30 showed no significant differences between TLR4 KO and wild type retinas (ANOVA, Bonferroni’s test; , n = 5 in both cases).
Horizontal cells morphology and quantitation in control and TLR4-deficient mice. (A–D) Vertical sections from P30 control ( A,C ; C57BL/6J) and TLR4-deficient ( B,D ; TLR4–/–) mice retinas labeled with antibodies against calbindin (horizontal cells, green) and Bassoon (synaptic ribbons, red). The nuclear marker TO-PRO 3 (blue) was used to visualize all cell nuclei. (E) Quantitation of horizontal cells per mm of retinal section in control (gray) and TLR4 KO (white) retinas at P30 ( n = 5 in both cases). OPL, outer plexiform layer; INL, inner nuclear layer. Scale bar: 40 μm.
We next explored alterations in the number and/or morphology of bipolar cells. Rod bipolar cells were identified using an antibody raised against the α isoform of protein kinase C (PKC). In the mouse retina, dendritic terminals of ON-rod bipolar cells make connections with rod spherules by means of a massive dendritic arbor in the outer plexiform layer (OPL) ( – ), and their axons extend into the inner plexiform layer (IPL), where each ends in a bulbous axon terminal in the S5 stratum. As shown in , the density of ON-rod bipolar cell bodies in the INL was significantly lower in the retinas of P30 TLR4 KO mice as compared to their age-matched controls (85% of the value obtained for wild type animals; ANOVA, Bonferroni’s test, P < 0.05, n = 11 for C57BL/6J mice and n = 12 for TLR4 KO mice). A similar decay in the number of rod bipolar cells was found in the retinas of P20 TLR4 KO animals (71% of the value obtained for age-matched wild type animals; ANOVA, Bonferroni’s test, P < 0.05, n = 4 in both cases; data not shown). The analysis of the retinas of 12-month-old animals also revealed a lower number of bipolar cells in TLR4-/- mice as compared to that observed in age-matched wild type animals (89% of the value obtained for wild type animals; ANOVA, Bonferroni’s test, P < 0.05, , n = 4 in both cases). The lower number of rod bipolar cells in TLR4 KO animals resulted in a lower density of bipolar cell dendrites when compared to wild type animals, with a mean fluorescence value of 83.09% in TLR4 KO animals with respect to that observed in control animals at P30, and 88.75% at the age of 12 months (ANOVA, Bonferroni’s test, P < 0.05, , n = 4 in both cases).
Bipolar cells morphology and quantitation in control and TLR4-deficient mice. (A,B) Vertical sections from P30 control ( A ; C57BL/6J) and TLR4-deficient ( B ; TLR4–/–) mice retinas labeled with antibodies against PKC (rod bipolar cells, green) and Bassoon (synaptic ribbons, red). The nuclear marker TO-PRO 3 (blue) was employed to visualize all cell nuclei. (C–F) Magnification of (A,B) showing the labeling of rod bipolar cells (C,D) or the double labeling of rod bipolar cells and synaptic ribbons (E,F) . Note that both positive cell bodies and dendrites were more numerous in control animals than in TLR4-deficient mice. (G) Quantitation of bipolar cells per mm of retinal section in control (gray) and TLR4 KO (white) retinas at 1 (P30) and 12 months (P360) of age. (H) Quantitation of bipolar cell dendrites in the OPL in control (gray) and TLR4 KO (white) retinas. Data are represented as mean values of fluorescence. P < 0.05; ANOVA, Bonferroni’s test, n = 4 to n = 12 in each group. Error bars represent the SEM. Scale bar: 40 μm (A,B) , 10 μm (C–F) .
No significant differences were found in ganglion cell numbers. We found a mean value number of 2149 ± 194 ganglion cells/mm in C57BL/6J mice and 2044 ± 115 ganglion cells/mm in TLR4 KO mice.
### TLR4 Ablation Does Not Affect Photoreceptor Synaptic Complexes
Given the significant reduction in the number of rod bipolar cells in TLR4 KO mice, we hypothesized that synaptic connectivity between photoreceptors and second order neurons could be impaired in these animals. To explore this possibility, we used an antibody against Bassoon, a protein component of the synaptic ribbons presents in both the rod spherules and cone pedicles of the OPL. As shown in , , the typical Bassoon-immunoreactive spots, with the characteristic horseshoe shape that is typical of rod spherules, were observed in a similar proportion in both control and TLR4-deficient mice ( , for C57BL/6J mice and , for TLR4 KO mice). Quantification of the photoreceptor synaptic ribbons showed no significant differences between TLR4-/- mice and wild type animals at both ages tested, P30 and P360 (in pixels of fluorescence area in the OPL: 13996 ± 1902 vs. 14578 ± 1040 at P30 and 13165 ± 1633 vs. 13784 ± 1274 at P360).
With the purpose of analyzing in greater detail whether photoreceptor synaptic complexes are altered by TLR4 ablation, transmission electron microscopy (TEM) images were used to visualize rod spherules and cone pedicles, as well as their synaptic connections to bipolar and horizontal cells in the OPL of 12 months old TLR4 KO and wild type animals. The typical triads, made up by the lateral elements of horizontal cell processes and the central elements of bipolar cells profiles, presented a normal morphology in the rod spherules and cone pedicles of wild type and TLR4 KO animals ( , respectively). Rod spherules and cone pedicles showed normal electron-dense synaptic ribbons and arciform structures ( – , arrows) in the two experimental groups analyzed. Also, the typical large mitochondrion in rod spherules and the multiple mitochondria in the cone pedicles were found ( , arrowheads). Together, these findings indicate that synaptic contacts between photoreceptors and second order neurons presented the typical ultrastructural features of a normal retina in TLR4 mice.
Ultrastructure of photoreceptor ribbon synapses in control and TLR4-deficient mice. Transmission electron microscopy images from 12 months old control ( A,C,D ; C57BL/6J) and TLR4-deficient ( B,E,F ; TLR4–/–) mice retinas. (A,B) Representative regions of the OPL of a control (A) and TLR4-deficient (B) mouse showing normal rod spherules (blue) and cone pedicles (red). Note the typical large mitochondrion in rod spherules and the multiple mitochondria in the cone pedicles (arrowheads). (C–F) Higher magnification representative triad synaptic complexes in spherules (C,E) and pedicles (D,F) . The ribbon synapse and the arciform density are electron-dense structures (arrows) surrounded by two horizontal cell elements (yellow) and a bipolar cell process (Schafer et al.), forming the typical triad structure. No major differences were observed in terms of synaptic complex structure between controls and TLR4 KO mice. Scale bar: 1 μm (A,B) , 500 nm (C–F) .
### Absence of TLR4 Reduces the Density of Retinal Microglia
Bearing in mind that TLR4 plays a role in the innate immune response in the brain, and that microglia serve as the CNS resident immune cells, which includes the retina, we also explored whether TLR4 deletion affected microglial cell numbers in the mouse retina. To that end, vertical retinal sections of 1- and 12-month-old mice were immunostained with an antibody against the ionized calcium-binding adaptor molecule 1 (Iba1). Microglia and macrophages have been reported to express specifically and ubiquitously this calcium-binding protein ( ; ). As can be seen in , microglial cells were found in the inner and outer plexiform cell layers of both wild type and TLR4 KO mice ( – ). Iba1-positive cells displayed a very small soma, little perinuclear cytoplasm, and very many fine, branched processes with numerous projections, i.e., the usual morphological features of resting microglia. The appearance of Iba-1 positive cells in immunolabeled retinal sections of TLR4 KO mice was similar to that observed in C57BL/6J control mice. The mean density of Iba1-positive cells, however, was found to be significantly lower in TLR4 KO mice than in their age-matched controls at P30 and P360 (72 and 87% of the value obtained for wild type animals; ANOVA, Bonferroni’s test, P < 0.001 and P < 0.05, respectively, , respectively, n = 4 in all cases). In addition, microglial activation was assessed by analyzing the expression of MHCII, a marker of activated microglia and macrophages ( ). The quantity of Iba1-positive cells co-labeled by anti-MHCII antibody was higher in 12-month-old mice than in 1-month-old animals, but no significant differences were observed between wild-type and TLR4 KO mice in terms of the number of MHCII-positive cells ( ). These data indicate that TLR4 deficiency reduces the density of retinal microglia cells, without affecting the state of microglia activation, under normal conditions.
Microglial cells in control and TLR4-deficient mice. (A–D) Vertical retinal sections from control ( A,C ; C57BL/6J) and TLR4-deficient (B,D) mice at P30 (A,B) and P360 (C,D) stained for Iba1 (green) and MHCII RT 1B (red). The nuclear marker TO-PRO 3 (blue) was employed to visualize all cell nuclei. (E,F) Quantification of cells stained with one or both markers (Iba1/MHCII) per mm of retinal section at P30 (E) and P360 (F) . P < 0.05, P < 0.001; ANOVA, Bonferroni’s test, n = 4 in all cases. Error bars represent the SEM. ONL, outer nuclear layer; INL, inner nuclear layer; GCL, ganglion cell layer. Scale bar: 40 μm.
Conversely, the mean number of astrocytes in TLR4 KO mice was 95.2% of that observed in WT mice, without reaching signification (not shown). Müller cells morphology showed similar features in TLR4 KO and control mice (not shown).
## Discussion
Mammalian TLRs have relevant roles in embryogenesis, development of neurons and neural circuits, learning and memory, with regulatory roles on the processes of proliferation, differentiation, outgrowth, plasticity and neuron survival ( ; ; ; ; ; ). Thus, it is expected that TLRs contribute to the development and maintenance of retinal structure and function. Our hypothesis is that TLR4 has a relevant role in the morphology and function of the retina under physiological conditions. TLR deficient mice have extensively proven to be useful to elucidate the role and relevance of TLRs. Thereby, we used a TLR4 KO mouse to assess the influence of TLR4 on postnatal retinal morphology and function.
Our results show that, in healthy conditions, TLR4 deletion alters both the structure and function of the retina in postnatal mice, and that this effect remains throughout life. We have demonstrated that TLR4 deletion reduces the mice retinal responsiveness from P20 to P360, with lower scotopic a- and b- waves, reduced oscillatory potentials and lower visual acuity. The significant reduction in retinal function can be explained, at least in part, by both the reduction in bipolar cells and the loss of their dendritic arbors in the OPL of TLR4 KO mice. We report a significant decrease in the density of bipolar cells at P20, P30, and P360, with a consequent reduction in their dendritic arborization. Our results agree with those of previous studies showing that LPS, acting through TLR4 in astrocytes, can induce changes in the dendritic pattern of hippocampal neurons ( ). More recent studies have also found that TLR4 regulate GABAergic synapse function under pathological situations in psychologically stressed animals ( ). Shechter and collaborators showed that retinal progenitor cell proliferation is enhanced in TLR4-deficient mice during the early postnatal period ( ) in ciliary epithelium and ora serrata . As the number of neuronal cells is not increased in adult retinas, we hypothesize that the known process of programmed elimination of extra neurons in the CNS could be responsible for that ( ; ; ). Despite the discrepancy between the decrease in density and dendritic arborization of bipolar cells and the lack of change in photoreceptor numbers and morphology, images of transmission electron microscopy did not show differences in the structure of the photoreceptor spherules.
Recent data indicate that many molecules associated with the immune system regulate circuit development and plasticity in healthy brains ( ). Microglial cells, the CNS resident immune phagocytes, play an essential role in the formation and maintenance of synaptic networks under physiological conditions ( ; ; ; ). Accordingly, TNF-alpha and different cytokines secreted by fetal microglia have a crucial role in CNS development, and synapse formation, refinement and function. As the brain develops, neurons form an overabundance of synaptic connections, many of which are later removed during what is known as synapse pruning, a process crucial to proper brain connectivity ( ; ; ). Moreover, microglia engulf synaptic material in an active manner and play an important role in postnatal synaptic pruning in mice ( ; , ). In line with this, disrupting microglia-specific signaling produced sustained deficits in synaptic connectivity ( , ; ; ). Also, a recent work by Jobling et al. has shown that in postnatal neural development, the absence of microglial Cx3CR1 signaling induced retinal disfunction and photoreceptor loss ( ). Among the TLRs, TLR1 to TLR9 are expressed in microglia ( ), and microglial activation depends on TLRs ( ; ), including TLR4 ( ; ). Moreover, previous works have shown that TLR4 play a part in the activation of retinal microglial cells ( ; ). Our data indicate that TLR4 deficiency reduces microglial density in the retina of healthy mice. This is consistent with previous data showing that microglial phagocytosis of degenerating axons is reduced in TLR4 KO mice ( ). Given the key role of microglia in developing and maintaining the structure and function of the CNS, the retinal microglia deficiency could be responsible, at least in part, for the observed changes in the structure and function of mouse retinal neurons. In this sense, Kashima and Grueter have recently shown that TLR4 influences NMDA-receptor synaptic transmission and plasticity, likely through TLR4 expressed in microglial cells ( ).
Other works have previously demonstrated a role for TLRs in the development of neuronal circuits ( ). In Drosophila , Rose and collaborators have shown that Toll receptors are involved in the synaptic initiation of motoneurons and reported the presence of anomalous innervations in Toll mutants ( ), which can alter synaptic target recognition ( ). Also, TLRs activation can induce synaptic dysfunction in a pathological situation ( ). The activation of TLRs in an inflammatory process triggers the production of several cytokines, as IL-1beta or IFN-gamma that can regulate synaptic plasticity. In this sense, in the case of Alzheimer’s disease, it has been shown that the IL-1beta secreted by microglia cells decreases the synthesis of synaptophysin ( ) and that IFN-gamma can decrease the rate of synapse formation, as shown in rat-cultured sympathetic and hippocampal neurons ( ).
Nowadays it is accepted that TLRs can have both, beneficial and harmful effects on CNS development and neuroplasticity, which can vary depending on the physiological or pathological state and tissue homeostasis ( ; ). In a balanced scenario, an acute TLR activation may tip the balance toward tissue damage or pathology, while a chronic or limited response might be necessary for maintaining the homeostatic balance ( ).
Our observation that bipolar cell numbers and their dendritic arbors are reduced in TLR4 KO adult mice also suggests that TLR4 is required for the development of a mouse retina with the correct structure and function. Whether this mechanism acts mainly by altering cellular proliferation or differentiation processes during embryogenesis, through errors in synapses formation or both, remains to be elucidated. Although more studies are needed to clearly identify their function, altogether these results indicate that TLR4 may be involved in establishing the correct cellular structure of the retina during embryogenesis.
Toll-like receptors are potential therapeutic targets and there is increasing hope that TLRs agonists and antagonists could be used in the treatment of CNS disorders. Since TLR4 activation is related to the progression of several neurodegenerative disorders, mainly by triggering an inflammatory process through the secretion of proinflammatory cytokines by the microglial cells, the use of TLR4 agonists and antagonists could balance or counteract the progression of the diseases or the therapeutic approach. Beyond its role as an immune system receptor, a complete understanding of the TLR signaling pathways is needed to understand the molecular bases of retinal development and/or degeneration, and also to prevent the effects on CNS balance of the use of TLRs agonists and antagonists to treat immune and non-immune pathologies.
Overall, our study suggests that genetic deletion of toll-like receptor 4 causes functional alterations in terms of visual response and acuity, probably through the loss of bipolar cells and microglia. However, the expression of TLR4 does not appear to be essential for the processing of visual information in the retina. Identification of the role of TLR4 on the structure and function of the adult retina has implications for research into the involvement of TLRs in physiological and pathological conditions, and may contribute to the development of new therapeutic options for both infectious and non-infectious diseases.
## Author Contributions
AN, OK, IO-L, LC, EdJ, and VG-V collected and analyzed the data and revised the manuscript. VM designed the experiments and drafted the manuscript. NC designed the study with the assistance of VM and PL, provided study material, and revised the manuscript. PL designed the experiments, collected and analyzed the data, and revised the manuscript. All authors have read and approved the final submitted manuscript.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
GCaMP6f is among the most widely used genetically encoded calcium indicators for monitoring neuronal activity. Applications are at both the cellular and population levels. Here, we explore two important and under-explored issues. First, we have tested if GCaMP6f is sensitive enough for the detection of population activity with sparse firing, similar to the sensitivity of the local field potential (LFP). Second, we have tested if GCaMP6f is fast enough for the detection of fast network oscillations critical for the encoding and consolidation of memory. We have focused this study on the activity of the hippocampal network including sharp waves (SWs), carbachol-induced theta oscillations, and interictal-like spikes. We compare simultaneous LFP and optical GCaMP6f fluorescent recordings in Thy1-GCaMP6f mouse hippocampal slices. We observe that SWs produce a clear population GCaMP6f signal above noise with an average magnitude of 0.3% Δ F / F . This population signal is highly correlated with the LFP, albeit with a delay of 40.3 ms (SD 10.8 ms). The population GCaMP6f signal follows the LFP evoked by 20 Hz stimulation with high fidelity, while electrically evoked oscillations up to 40 Hz were detectable with reduced amplitude. GCaMP6f and LFP signals showed a large amplitude discrepancy. The amplitude of GCaMP6f fluorescence increased by a factor of 28.9 (SD 13.5) between spontaneous SWs and carbachol-induced theta bursts, while the LFP amplitude increased by a factor of 2.4 (SD 1.0). Our results suggest that GCaMP6f is a useful tool for applications commonly considered beyond the scope of genetically encoded calcium indicators. In particular, population GCaMP6f signals are sensitive enough for detecting synchronous network events with sparse firing and sub-threshold activity, as well as asynchronous events with only a nominal LFP. In addition, population GCaMP6f signals are fast enough for monitoring theta and beta oscillations (<25 Hz). Faster calcium indicators (e.g., GCaMP7) will further improve the frequency response for the detection of gamma band oscillations. The advantage of population optical over LFP recordings are that they are non-contact and free from stimulation artifacts. These features may be particularly useful for high-throughput recordings and applications sensitive to stimulus artifact, such as monitoring responses during continuous stimulation.
## Introduction
Fluorescent calcium signals are widely used for monitoring neuronal activity in the brain thanks to the availability of genetically encoded indicators. GCaMP6f is among the best calcium indicators to date, with high sensitivity, high fluorescent yield and relatively fast response time ( ). There are two principal applications of GCaMP6f: to visualize somatic calcium transients due to bursts of action potentials ( ; ; ; ), and monitoring population activity via calcium fluorometry ( ; ). Two important issues related to the measurement of population activity remain less explored: (1) Is GCaMP6f sensitive enough for detecting population events with sparse neuronal firing and sub-threshold synaptic activity? (2) Is GCaMP6f fast enough for detecting physiologically relevant network oscillations?
For the sensitivity issue, we ask whether the population GCaMP6f signal is comparable to local field potential (LFP) recordings, i.e., capable of detecting network activity in which only a small fraction of neurons fire action potentials, while the majority of neurons only have subthreshold potentials. This is of concern as GCaMP fluorescence appears to primarily reflect supra-threshold somatic Ca influx ( ), together with intracellular calcium induced by action potentials ( ). In contrast, the principal source of the LFP may rather reflect subthreshold dendritic synaptic currents ( ). However, while the Ca signals from individual synapses are small, a population signal from the integration of a large number of dendrites in the neuropil might become detectable, given that the sensitivity and fluorescent yield of GCaMP6f are both excellent.
Concerning the frequency response, the time course of somatic calcium transients is about 1 s ( ; ; ), which is too slow for detecting most network oscillations. In a recent report ( ), 2 Hz but not 10 Hz GCaMP signals were detectable. However, the rise time of calcium transients is much faster than 1 s and may be able to follow fast oscillations. Measurements with the organic calcium indicator magnesium green showed <1 ms rise time ( ). When action potentials occur, the duration of intracellular calcium transients are long and dependent on intracellular buffering and clearance ( ; ). In contrast, during a network oscillation with low calcium influx on average in the population, we argue that the time course will be more related to faster rise time than the slower decay time, and in particular the response time of the calcium indicator, or about 40 ms for GCaMP6f ( ), as the onset time of the channels and the rate of calcium influx would be only a few milliseconds ( ; ). When only a small fraction of neurons fire action potentials in the population, the majority of neurons should have a rate of calcium influx related to the opening probability of low voltage activated channels (LVAs) (Reviewed by ; ). The opening probability of the LVA channels should be correlated to the fluctuation of membrane potential. In theory, recording fast oscillations in neuronal populations with calcium indicators is possible during sparse firing, when only a small fraction of neurons fire action potentials, most neurons have subthreshold membrane potential fluctuations, and the ability of intracellular calcium buffering/clearance is much greater than the calcium influx rate. The response time of GCaMP6f (∼40 ms) could in theory permit the detection of oscillations up to 25 Hz.
Hippocampal sharp waves (SWs) are spontaneous network events in which a small fraction of neurons fire action potentials ( ; , ; ) while the majority of neurons receive subthreshold excitatory and inhibitory synaptic input ( ). The summation of excitatory and inhibitory post-synaptic potentials (EPSPs/IPSPs) generates a clear voltage source/sink pair in LFP recordings ( ), reviewed by . In this report we test if spontaneous SWs can be seen in the GCaMP signal, events with sparse firing and sub-threshold synaptic activity. We speculate that the population summation of calcium influx during these events is detectable in GCaMP6f optical recordings.
We compared simultaneously recorded LFP and fluorescent GCaMP6f signals in Thy1-GCaMP6f mouse hippocampal slices during SWs, interictal spikes and carbachol-induced theta oscillations. Population activation by electric stimulation was also used to test the frequency response characteristics of GCaMP6f population signals.
We observed that SWs can be clearly detected optically in the population GCaMP6f signal. The GCaMP signals were highly correlated with LFP-detected events with a delay of 40.3 ms (SD 10.8 ms). The GCaMP signal followed evoked network activity below 20 Hz with high fidelity, while activity up to 40 Hz were still detectable with reduced amplitude. The population GCaMP6f and LFP signals showed a large amplitude discrepancy. The amplitude of GCaMP6f fluorescence increased by a factor of 28.9 (SD 13.5) between spontaneous SWs and carbachol-induced theta bursts, while the LFP amplitude increased by a factor of 2.4 (SD 1.0).
Our results suggest that the population GCaMP6f signal has a sensitivity comparable to that of the LFP and may be even more sensitive than the LFP for detecting network events with low amplitude but disproportionally large GCaMP6f fluorescence, likely arising from asynchronous activity. We also found that the population GCaMP6f signal is fast enough for monitoring theta (4–7 Hz) and beta (14–25 Hz) oscillations in slice, and the detection limit can be as high as 40 Hz. These results suggest that optical recordings of population GCaMP6f signals may be useful for detecting network activity complementary to LFP recordings. They may have particular utility in situations where the confound of electrode disruption and stimulus artifacts need to be minimized. In addition, they have the power to monitor activity in multiple distinct anatomical sites concurrently. Our results also have implications for the interpretation of in vivo data obtained with the increasingly widespread use of GCaMP based photometry ( ; ).
## Materials and Methods
### Slice Preparation
P21–P33 male and female C57BL/6J-Tg (Thy1-GCaMP6f) GP5.5 Dkim/J. mice (Jax 024276) mice were used to prepare paired hippocampal hemi-slices in accordance with a protocol approved by the Institutional Animal Care and Use Committee at Georgetown University Medical Center. Following deep isoflurane anesthesia, animals were rapidly decapitated. The whole brain was subsequently removed and chilled in iced (0°C) sucrose-based artificial cerebrospinal fluid (sACSF) containing (in mM) 252 sucrose; 3 KCl; 2 CaCl ; 2 MgSO ; 1.25 NaH PO ; 26 NaHCO ; 10 dextrose; bubbled with 95% O , 5% CO . Hippocampal slices (480 μm thick) were cut in horizontal sections from dorsal to ventral brain with a vibratome (Leica, VT1000S). Slices were incubated in ACSF for at least 2 h before each experiment. ACSF used for maintenance and recording contained (in mM) 132 NaCl; 3 KCl; 2 CaCl ; 2 MgSO ; 1.25 NaH PO ; 26 NaHCO ; 10 dextrose; bubbled with 95% O , 5% CO at 26°C.
### Local Field Potential (LFP) Recording
Local Field Potential (LFP) recordings were done in a submerged chamber, and slices were placed on a mesh that allowed perfusion on both sides at a high flow rate (10–30 ml/min) ( ; ). All recordings were done with low resistance glass microelectrodes (∼150 kΩ tip resistance). The electrodes were pulled with a Sutter P87 puller with six controlled pulls and filled with 0.5 M NaCl in 1% agar, which prevents leakage of the electrode solution that could potentially alter the tissue surrounding the electrode tip. The recording electrode was placed in CA1 stratum pyramidale , where SWs have large amplitudes ( ) in healthy slices.
### GCaMP Fluorescent Recording
The GCaMP signals were recorded by a 464-channel photodiode array (WuTech Instruments). The two-stage amplifier circuits in the diode array subtract the resting light intensity and amplify the small optical signals 100 times before digitization. This achieves a 21-bit effective dynamic range to fully digitize a signal of ∼0.5% Δ F / F . [For a recent review of the two-stage imaging system, see ].
The Δ F / F is defined as ( F − F )/ F , where F is the signal trace from each detector and F is the baseline fluorescent intensity. The signals were digitized at 1,616 frames/s. In some experiments, only the center of the field of view was sampled at 3,000 Hz, in order to preserve the high frequency components in the signal. The LFP and stimulation signals were sampled and digitized concurrently with the VSD signals. Optical imaging was performed on an upright microscope (Olympus BX51 WI) with an epi-illumination arrangement: excitation light (470 nm LED, ThorLabs) passes a GFP filter cube (Chroma, excitation 425–475 nm, dichroic mirror 480 and emission filter 485 long pass). The GCaMP signals were imaged at two spatial resolutions: A 20× objective (0.95 NA, Olympus) permitted the concurrent imaging of more localized cellular activity and population signals from the same tissue, and a 10× objective (0.30 NA, AMscopes) allowed for imaging all hippocampal subfields in the same field of view. The aperture of the diode array was 19 mm in diameter; containing a hexagonal arrangement of 464 optic fibers. The diameter of the fiber was 750 μm. Each detector on the array (pixel) collected florescent signals from an area of 37.5 μm in diameter with the 20× objective, and about 75 μm in diameter with the 10× objective. The total beam power of the LED was ∼350 mW at 1A. The fluorescent intensity on each detector was about 20,000 photoelectrons/ms for the 20× objective. Illumination intensity at the sample was <20 mW/mm .
### Voltage-Sensitive Dye (VSD) Imaging
Voltage-sensitive dye imaging was used to validate the response time of the GCaMP signal ( ). In three slices, VSD and GCaMP signals are from the same tissue and imaged by the same diode array. The slices were stained by an oxonol dye, NK3630 (Nippon Kankoh-Shikiso Kenkyusho Co., Ltd., Japan), as an indicator of transmembrane potential. Slices were stained with 5–10 μg/ml of the dye dissolved in ACSF for 120 min (26°C). During staining, the ACSF was circulated and bubbled with 95% O , 5% CO . After staining, the slices were transferred back to the incubation chamber for at least 1 h before each experiment. NK3630 binds to the external surface of the membrane of all cells and reports their membrane potential change [for a recent review of the diode array and NK3630, see ]. The absorption spectrum of the dye shifts linearly with the changes in the membrane potential ( ). The VSD signal in this report is the change in absorption of light with a 705 nm wavelength. The detectable signals are a change in light intensity that is roughly 0.01–0.1% of the resting light intensity. Staining with this dye does not cause noticeable changes in spontaneous or evoked neuronal activity ( ; ), and stained slices maintain viability for up to 24 h. In 705 nm recording light, NK3630 molecules do not generate fluorescence, so no noticeable phototoxicity is detected ( ).
The VSD signals were recorded by the same diode array. With a transillumination arrangement, neurons through the whole thickness of the slice (480 μm) contribute relatively equally to the VSD signal. A tungsten filament lamp was used for illumination and a 705/10 nm interference filter (Chroma) was placed in the illumination path during optical recording.
During imaging experiments, the slice was continuously perfused in a submersion chamber with ACSF (same as the incubation solution) at 26°C and at a rate of more than 20 mL/min. Intermittent VSD imaging trials were performed, with 2–3 min intervals between trials.
### Stimulation
Stimulation to the CA3 area was provided with a concentric metal electrode (FHC CBDSE 75). Stimulation pulse was 0.1 ms wide generated by a Master 8 stimulator (AMPI). The stimulation current was 20–100 μA generated by an isolator (AMPI).
### Data Analysis
Digital filters were applied offline. To automatically detect the amplitude and peak time of the SW events in fluorescent signals (e.g., ), we first digitally filtered the simultaneously recorded LFP signals between 1–30 Hz, then a threshold was set manually above the baseline noise to identify the majority of SW events in the LFP. Using a window between −50 and 100 ms of the LFP peak, the peak of the fluorescent signal was identified as the SW peak. Custom programs were written in MATLAB and Labview for digital filtering, threshold detection, and determining the amplitude and frequency distributions. For figure preparation, various bandpass ranges were chosen for the LFP and GCaMP signals to minimize filtering whenever possible. These specific ranges have been identified in figure legends.
GCaMP population signals of SWs across hippocampal subfields. (A) Slice positioning over the diode array. The end of mossy fibers (M) and lack of GCaMP6f expression in CA2 was used to identify the three hippocampal subfields CA1-3. The yellow hexagon marks the field of view of the diode array. The black dot marks the location of the LFP electrode. These structures were also outlined with an image of transmitted light (gray lines). (B) Δ F / F from all 464 diodes during a SW event. This SW was one of the 9 occurring during a 9-s recording sweep (blue box in C ). Note that GCaMP signals of SWs were seen over a large area across CA1, CA2, and CA3. Str. pyramidale (P, orange band) and mossy fiber bundle (M. f∖green band) are identified overlaying the signals. (C) LFP signals were simultaneously recorded with the GCaMP signals (both sampled at 1,616 Hz). Signals from three detectors in CA1, CA2 and CA3 [red dot/traces in panels (A,B) ] plotted together with the LFP recording (filtered 0.1–30 Hz). The amplitude of SWs in the GCaMP signal were on average 0.3% Δ F / F with a signal-to-noise >10. From 11 slices we recorded ∼6,500 SW events optically, all with a clear one-to-one correspondence between LFP and optical GCaMP signals. (D) Decline of optical signals over long recording periods due to photobleaching. Red dots mark the relative amplitude of individual SWs from one slice recording. For clarity, events are only shown for the first 1,000 and last 500 s. The relative Δ F / F amplitude is normalized to the average amplitude of the first 100 events at the beginning of light exposure. Black and red traces are averages of the LFP and GCaMP signals, respectively, in a sliding 100-event window. Blue and green traces are GCaMP signals from two additional animals. Brown broken line: another slice with exposure at 6 times the light intensity for 660 s. Left and right insets: LFP and GCaMP signals from one slice before and after 4300 s of continuous light exposure. Blue broken line in panel (D) marks the sample time of the two traces. Note that amplitude reduction due to photobleaching is not obvious in individual SWs, as spontaneous SWs have a large variation in amplitude. Scale bar for inset = 1 s.
In experiments with high frequency stimulation, in which the amplitude of the response was too small for quantification ( ), we used the root mean square (RMS) power in place of the amplitude. The RMS was calculated with the following equation:
where X is the signal amplitude at the sample point n , X is the RMS power in a period of N sampling points.
Statistics were conducted in Graphpad Prism 8.0. To compare differences in means we first checked normality and lognormality of data with Shapiro-Wilk tests. Differences in means of two groups were assessed by unpaired t -tests for parametric distributions, and Mann-Whitney for non-parametric distributions. For more than two groups we compared means via 1-way ANOVA with the Tukey multiple comparisons correction, or Kruskal-Wallis with Dunn’s multiple comparisons test as appropriate. Error bars displayed are either SD or SEM, as indicated in figure legend. p < 0.05, p < 0.01, p < 0.001, p < 0.0001.
## Results
### Spontaneous SWs Are Detectable in the Population GCaMP Signal
Spontaneous SWs reliably occur in hippocampal slices as reported by our previous papers and other groups ( ; ; ; ; ; ). Our first goal was to test if SWs can be detected in the GCaMP6f signal in hippocampal slices from Thy1-GCaMP6f mice.
To test this, we positioned the slices with the end of the supra-pyramidal mossy fiber bundle at the center of the field of view of the diode array ( ), as this marks the boundary between CA3 and CA2 ( ; ). In Thy1-GCaMP6f mice the mossy fibers showed bright green fluorescence, due to the high expression in granule cells, providing a landmark for the outside limit of CA3 (“M” in ). CA2 could also be clearly identified as a darker region devoid of fluorescent cell bodies. In this way, CA3, CA2, and CA1 could be clearly delineated. By positioning the end of the mossy fiber bundle at the center of the field of view, the 464 detectors on the diode array covered a large area including CA3, CA2, and CA1 (yellow hexagon in ).
We observed one-to-one correlations between SW events detected in the LFP and GCaMP signals. In contrast to localized cellular calcium transients ( ), the population SW signals were reliably seen over a large area of hippocampal tissue spanning CA3, CA2, and CA1 ( ). Under a 20× objective, each optical detector received light from an area of 37.5 μm in diameter, so that the population signals we refer to in this report are a summation of the Δ F / F from both somatic and dendritic areas for a number of neurons under each optical detector.
Sharp wave peaks in the GCaMP signal were visible across trials, with a range in Δ F / F of 0.1–1.0% ( ). The signal-to-noise ratio was >10, allowing clearly distinguishable events above noise. LFP signals were simultaneously recorded with the GCaMP signals, both sampled at 1,616 Hz. From 11 slices we recorded 6,500 SW events optically, all with a one-to-one correspondence between LFP and optical recording of the GCaMP signals, with an average Δ F / F ∼0.3%.
### Signal Polarity of SWs Across Hippocampal Layers
A notable difference between LFP and GCaMP recordings of SWs is the signal polarity in different laminar areas. LFP signals from str. oriens and str. radiatum have opposite polarities ( ), as they form a current source-sink pair around str. pyramidale ( ). This polarity reversal is obviously not observable in the Δ F / F ( ). GCaMP signals from soma (str. pyramidale ) and neuropil (polymorphic or molecular layers) have the same polarity (increased Δ F / F at SW onset), suggesting that the calcium signals increase irrespective of current flow direction in the population.
### Photo-Bleaching Limits Long Recording Times
Continuous exposure to light while recording causes bleaching of the GCaMP fluorescent protein. We did observe a gradual reduction in SW amplitude with exposure time. To test the limits of optical recordings of SWs, we recorded continuously for over an hour (4,300 s). In a representative experiment ( ), the excitation light intensity was reduced to 1/4 of the intensity used in . Under this light intensity the SWs were still reliably detected, while the dark noise (noise in the electronics) became larger ( , inset red traces). After long exposure, the amplitude of SWs in the optical signal reduced but were still distinguishable from noise. Because amplitude of spontaneous SWs varies over a large range ( , red dots), amplitude reduction by photo-bleaching was often difficult to see from individual SW events. However, when the average of 100 SWs were plotted ( , red curve), a clear trend of amplitude reduction in the optical signal was seen, compared to the simultaneously recorded LFP amplitude ( , black curve). Similar long-duration recordings were performed in three slices from three animals ( , blue and green traces). In these experiments the illumination intensity was kept constant, revealing slightly different rates of amplitude reduction. From these results we determined that reliable optical recordings were possible for at least 1,500 s of recording time with continuous illumination, equivalent to 100 15 s trials or 1,000–2,000 spontaneous SWs. At the illumination intensity 6 times greater than in (red, green, and blue), the bleaching rate was much faster, with the amplitude reduced to 50% in 660 s of exposure ( , brown dashed line).
### Time Delay Between GCaMP and VSD Signals
GCaMP signals showed a significant delay compared to LFPs ( , top traces). The peak of the population GCaMP6f signal lagged behind the peak of the LFP signal by 40.3 ms on average (SD 10.8 ms, n = 84 SWs, in two slices from two animals) ( ). The delay time from two recording locations ( , location a, b, ∼1.2 mm apart along the CA1 str. pyramidale ) was similar ( , top traces), suggesting that the delay was not caused by spreading of the SW along the CA1 zone.
Rise time of the population GCaMP6f signal. (A) LFP and GCaMP/VSD recordings of SWs in the same tissue. LFP (black trace, filtered 0.2–50 Hz) were highly correlated with events detected optically at two different CA1 locations either with GCaMP6f (blue and red traces, top, filtered 0.2–50 Hz) or VSD (blue and red traces, top, filtered 0.2–50 Hz). Note the different y-scale for GCaMP and VSD traces, GCaMP events were about 50 times larger and slower. Black arrows mark the SWs displayed in expanded time scale in panel (B) GCaMP signals showed a marked delay compared to the LFP, which was not observed in the VSD signals from the same tissue. (C) Box and whisker plots of the peak delay time between LFP and optical signals. As expected, GCaMP signals showed a longer delay time of 40.3 ms (SD 10.8 ms, n = 84 SWs, in two slices from two animals) compared to VSD delay time of 5.7 ms (SD 4.1 ms, n = 82 SWs from same slices), a significant difference of p < 0.0001 (Mann-Whitney).
The delay time of the GCaMP6f population signal was verified with voltage-sensitive dye recordings. In three animals we stained the slices with the voltage sensitive dye NK3630. The dye staining did not affect the spontaneous SW rate of occurrence or amplitude. The dye absorption signal measured at 705 nm emission wavelength was able to detect SWs. The VSD signal associated with a SW event was manifested by an increase in absorption at 705 nm, in accordance with the previously established nature of voltage imaging with absorption dyes, where dye molecules bind to the neuronal membrane with depolarized membrane potential ( ). The VSD signal of SWs was fast with the peak correlating well with the LFP ( , bottom traces). The VSD signal showed an insignificant delay to the LFP, 5.7 ms (SD 4.1 ms, n = 82 SWs, in two slices from two animals), demonstrating a good correlation between the population summation of membrane potential and LFP signal during SW events.
The GCaMP and VSD signals in were measured from the same tissue in different recording trials. In the same tissue the VSD and the GCaMP signals were independent as the 705 nm light did not excite the GCaMP6f proteins. The GCaMP signals were measured at 500–530 nm (excited by 470 nm), and there is no significant contribution of VSDs at this wavelength.
A wavelength independent “intrinsic” optical signal was also seen at 705 nm in the VSD measurements. The intrinsic signal was slower and with a reversed polarity compared to the VSD signals at the 705 nm ( ). The downward deflection in the VSD ( bottom traces) were associated with this intrinsic optical signal.
Comparing with the VSD signals, the time delay in population GCaMP6f signals was likely dictated by the response of the GCaMP protein. The delay times in our population measurements were comparable to the delay times obtained using intracellular calcium measurements; a 40 ms rise time ( ).
### Population GCaMP Signal Can Detect 20–40 Hz Evoked Activity
If individual calcium transients are far from saturation, the measured rise time of ∼40 ms of the population GCaMP6f optical signal should in theory allow following up to 20 Hz of network oscillations in the tissue. It may also be possible to detect higher frequency signals at attenuated amplitude. To test the frequency response of GCaMP6f population signals, we applied electrical stimulation to CA3 and measured the evoked population response in CA1. The stimulation intensity was low, adjusted to produce evoked LFP responses within the same amplitude range as spontaneous SWs in the same tissue ( ). Evoked GCaMP signals were observed in the CA3 and CA1 areas ( , red and orange traces). When two stimuli were delivered close in time, the response to the second stimulus was larger ( , arrowhead), suggesting paired-pulse facilitation through buildup of calcium is detectable by population GCaMP signals.
Frequency response of population GCaMP signals. (A) GCaMP signals in response to mild electric stimuli. Left: The arrangement for the stimulation and recording. Stimuli were applied to CA3 and response was recorded in both CA1 and CA3. The stimulation intensity was adjusted so that the evoked response had an amplitude similar to the SW amplitude in the LFP signals. Note that double stimuli induced larger responses to the second shock, even with long inter-stimulus intervals of 500 ms (blue arrow head). (B) High frequency stimuli caused a ramp accumulation of GCaMP signals. Note that the ramp signal was much larger and slower than the response to individual stimuli. (C) Evoked GCaMP signal to high frequency stimuli. A high frequency component can be seen in wide-band filtered signals (red traces, 7–800 Hz). Narrow band filtered (blue traces, 7–55 Hz) improved signal-to-noise ratio. (Inset) In a follow-up experiment, 10 mild stimuli (1.4× threshold) at 40 Hz was given to CA3. A narrow bandpass filter (30–50 Hz) of the CA1 GCaMP signals reveals a clear one-to-one correlation between individual stimulus pulses (filtered 30–1500 Hz) and the GCaMP response. (D) Power spectrum of the GCaMP signals, normalized to the power at 30 Hz. The signal power reduced to ∼50% at 35 Hz and ∼25% at 40 Hz. Green peak at the 50 Hz is the 2nd harmonic of the 25 Hz peak. High frequency, weak stimuli experiments were done on three slices from three animals. Data in panels (A,D) are from different slices receiving stimuli of slightly different frequency. (E) RMS power quickly reduced in high frequency. The RMS power during 20 or 40 Hz stimulations was normalized to the RMS noise in the same trial when there was no stimulation. Blue and left black bars, narrow band-pass filtered between 5–30 Hz, n = 8 trials, three slices from three animals, p < 0.0001 (unpaired t -test). Orange and right black bars, narrow band-pass filtered between 25–50 Hz, n = 9 trials, three slices from three animals, p = 0.0022 (unpaired t -test). Error bars indicate SEM.
With a train of stimuli of identical intensity, the GCaMP signals summated, forming a much larger rising ramp than the response to individual stimuli ( ). The Δ F / F amplitude evoked by individual stimuli were on average 0.1%, while the ramp signal from continuous 40 Hz stimulation was ∼100% (comparing ). The ramp rise time was faster with higher stimulation frequency with the same intensity ( ). The large ramp signal suggests an accumulation of calcium. The time course of these long ramps might be caused by the slow clearance of intracellular calcium, while the individual responses may reflect fast calcium influx.
As the rise time of the long ramp signal was much slower than the rise time of individual responses, the ramp signal could be removed by a digital high-pass filter (7–800 Hz) ( , red traces). The one-to-one relationship between evoked stimulation and GCaMP signal was maintained up to a frequency of 40 Hz. Responses to stimulation <30 Hz were clearly seen in high-pass filtered signals, while frequencies between 30 and 44 Hz were distinguishable by further band-pass filtering between 7 and 55 Hz ( , blue traces). One-to-one correlations between stimulus and GCaMP signals were clearly seen at 31 Hz but not at 44 Hz. In a follow-up experiment, 10 mild stimuli (1.4× threshold) at 40 Hz was given to CA3, and a narrow bandpass filter (30–50 Hz) was used on the CA1 GCaMP signals. Under this condition a clear one-to-one correlation between individual stimulus pulses and the GCaMP response was identified ( , inset). Response to the 40 Hz stimuli was further verified with a fast Fourier transform (FFT) ( ). While the 40 Hz FFT peak was much smaller compared to that of lower frequencies, the peak was clearly distinguishable from background noise. The frequency peaks show about a 50% reduction between 35 and 40 Hz. RMS power was calculated from data collected from three animals receiving 20 and 40 Hz stimulation ( ), demonstrating that the 40 Hz GCaMP6f signal was significantly higher than the RMS power of background noise ( p = 0.0022, unpaired t -test, n = 9 trials from three slices from three mice).
### Amplitude Discrepancy Between GCaMP and LFP Signals
Exceptionally large GCaMP signals were occasionally observed during population events, while the LFP signals of the same events were relatively small. These spontaneous interictal events had an amplitude of Δ F / F = 17% (SD 1%, n = 11), almost 50 times greater than the amplitude of SWs in the same tissue [Δ F / F = 0.36% (SD 0.10%, n = 1847)] ( ). In contrast, the LFP signals of SWs and interictal spikes had similar amplitudes, but interictal spikes exhibited reversed polarity and increased extracellular spiking ( , inset).
Amplitude discrepancy between LFP and GCaMP signals during interictal-like events. (A) Spontaneous interictal events (blue box) rarely occur in healthy slices, and are associated with a low LFP peak amplitude but disproportionally large GCaMP fluorescence. The GCaMP signal was measured from the CA1 area, in a region of interest approximately 0.1 mm in diameter surrounding the location of the LFP electrode. Inset (blue box) : expanded time scale of the event, showing cellular spiking and the LFP peak of the interictal event. (B) Interictal-like events induced by bath supply of 20 μM bicuculline. Bottom traces display event highlighted in blue box above on expanded time scale. (C) Comparison of GCaMP6f Δ F / F amplitude for three types of events. BI: bicuculline-induced interictal-like spike, average Δ F / F = 338 (SD 139%, n = 99 events from four slices from four animals). SI: spontaneous interictal event, average Δ F / F = 17% (SD 1%, n = 11 events from one slice from one animal), SW: sharp wave, average Δ F / F = 0.36% (SD 0.10%, n = 1847 events from the same slice with SIs). Error bars indicate SD. (D) Rise time of sharp wave (SW), spontaneous interictal events (SI) and bicuculline-induced interictal-like spikes (BI). The amplitude of the three events differ by a factor of ∼600 but have similar initial rise times. Note that the decay time of the three events are different. These traces are averages of n = 100 SWs from one animal; n = 11 SIs from the same animal; n = 10 BIs from a different animal.
Spontaneous interictal events only occurred occasionally in 2 out of 11 slices examined, and their occurrence rate was low. In one characteristic slice, 1,847 SWs were recorded over 36 min, with only 11 spontaneous interictal events detected. To further investigate the discrepancy between amplitudes in the LFP and GCaMP6f signals, we examined induced interictal-like events with the GABA receptor antagonist bicuculline. 20 μM bicuculline in ACSF was used to induce interictal-like spikes in four slices from four animals ( n = 99 events, ). The amplitudes of GCaMP events associated with bicuculline-induced interictal spikes were almost 1,000 times greater than the amplitudes of the SW events, while the LFP signal had a relative increase ranging from 5 to 20, demonstrating a large discrepancy in relative signal increases between GCaMP and LFP ( ).
Notably, the rise time, defined as the time for the signal to rise from 10 to 60% of peak, was about 40 ms for all three event types (spont. SW, spont. interictal, bicuculline-induced), despite the large amplitude differences in these events ( ). This suggests that the onset time of the optical population signal in all three cases is limited by the response time of GCaMP6f.
### Population GCaMP Signal Can Detect Carbachol-Induced Theta Oscillations
Carbachol-induced theta oscillations and related population events were next explored to further investigate the ability of GCaMP6f to monitor physiologically relevant hippocampal oscillations. When the perfusant was switched from normal ACSF to one containing 40 μM of the cholinergic agonist carbachol, spontaneous SWs disappeared and short bursts of theta oscillations (4–7 Hz) emerged, as recorded by the LFP electrode ( ). High amplitude GCaMP peaks were observed during these theta bursts, and like interictal-like events, there was a large discrepancy between the change in amplitude in the LFP and GCaMP signals. The GCaMP signal accumulated with successive theta cycles ( ). While both SW and carbachol-induced bursts had a wide range of amplitude in GCaMP signal ( ), the amplitude of the carbachol-induced burst was on average greater than the amplitude of SW events by a factor of 28.9 (SD 13.5, n = 252 bursts, four slices from four animals) ( ). In contrast, the amplitude of LFP events increased by a factor of only 2.38 (SD 1.02, n = 252 bursts, four slices from four animals), further demonstrating the large amplitude discrepancy between LFP and GCaMP signals.
Comparison of LFP and GCaMP signals during carbachol-induced theta oscillations. (A) Spontaneous SWs when the slice is bathed in normal ACSF, compared with (B) Induced theta bursts recorded from the same tissue after bath administration of 40 μM of the cholinergic agonist carbachol. The LFP and GCaMP signals are drawn on the same scales for panels (A,B) . On this amplitude scale the GCaMP SW events are very small, but can be clearly seen with 10× amplification ( A , Inset). Note that the polarity of the LFP signals were reversed during theta bursts, and a group of three or more spikes in the LFP was merged into a single large peak with GCaMP. (C) Amplitude distribution of SWs in the GCaMP signals ( n = 260 SWs, four slices from four animals), first normalized to the average amplitude from each slice and then pooled together in the distribution chart. (D) Amplitude distribution of carbachol-induced theta bursts in GCaMP signals ( n = 252 bursts, four slices from four animals). The amplitude was normalized to the average SW amplitude in each slice. Note that the majority of bursts were 13–30 times larger than the SW in the same tissue. (E) Amplitude discrepancy: When LFP and GCaMP signals during bursts were normalized to the average amplitude of SWs in the same tissue, The increase in GCaMP was 28.9 (SD 13.5, n = 252 bursts, four slices from four animals) while the LFP was only 2.37 (SD 1.02, n = 252 bursts, four slices from four animals). Significant difference in medians from the Kruskal-Wallis test, p < 0.0001 (Dunn’s multiple comparisons correction between LFP and GCaMP during carbachol show p < 0.0001). Error bars indicate SD. (F) Continuous theta cycles developed with continued carbachol perfusion. These ∼8 Hz oscillations were seen in both LFP and GCaMP signals. (G) An FFT of a sub-section of this signal ( F , blue box) revealed a clear peak at 8 Hz in both the LFP and GCaMP6f signals, as well as higher harmonics.
With continued carbachol perfusion, theta oscillations with continuous cycles developed ( ). These ∼8 Hz oscillations were seen in both LFP and GCaMP signals with one-to-one oscillation. An FFT of a sub-section of this signal ( , blue box) revealed a clear peak at 8 Hz in both the LFP and GCaMP6f signals, as well as higher harmonics ( ).
### Population GCaMP Signal During Transition Period of Elevated and Asynchronous Activity
The large amplitude discrepancy between the LFP and GCaMP6f signals led us to investigate if GCaMP6f can be used to detect population activity insensitive to the LFP. Elevation of asynchronous firing in a neuronal population should generate only a nominal LFP, with asynchronous currents canceling each other out in the volume conductor surrounding the neurons. However, the GCaMP6f signal in this population would be expected to be high, due to the accumulation of calcium from elevated activity. With even higher levels of activity, depolarization block can lead to a cessation of firing and detectable activity in the LFP, yet with GCaMP6f, the elevated calcium that results from this should be readily detectable.
There was a transition period between SWs and carbachol-induced bursts, in which the GCaMP signal had large fluctuations while the LFP signal displayed only low amplitude peaks ( ). Upon 40 μM carbachol administration, SWs abruptly stopped ( , gray dashed line). A transition period of 3–4 min occurred, during which the GCaMP signal had slow and large fluctuations up to Δ F / F = 30%, or about 40 times the SW signals ( vs. ). Meanwhile, the amplitude of the small peaks in the LFP was only 1/3–1/2 of that of the SW ( vs. , red traces). These fluctuations only occurred after carbachol was added and SWs stopped (observed in two preparations) and became larger until organized theta bursts emerged ( ). At times, one-to-one correlations were observed between peaks in the GCaMP and LFP signals ( ). At other times however, these GCaMP signals displayed dynamics not readily apparent in the LFP recording ( ). Spikes from nearby neurons were detectable with the LFP electrode in some preparations ( , blue traces, sampled at 3000 Hz and filtered 60–1500 Hz). Spiking rate increased after carbachol, indicating elevated population activity. Peaks in GCaMP6f were also correlated to sudden drops in the firing frequency of neurons ( ). More obvious correlations between reduction in spiking and GCaMP peaks were seen when theta bursts developed ( ), potentially from depolarization block or inhibition. Together these observations suggest that in highly excitable and asynchronous environments, GCaMP6f can reveal dynamics not detectable in the LFP, which alone is an incomplete snapshot of population activity.
GCaMP signals during the transition between SWs and carbachol induced bursts. (A) A 5 min recording of the transition period between spontaneous SWs and carbachol-induced theta oscillations. Blue trace shows the LFP filtered between 60–1,500 Hz (data sampled at 3,000 Hz), to isolate neuronal spikes from nearby cells. Black trace shows the LFP filtered between 0.1–30 Hz to isolate SWs (initial period) and theta bursts (final period). Red trace shows the GCaMP signal filtered between 0.1–30 Hz. Gray broken line marks the onset of 40 μM carbachol perfusion. Green broken line marks the onset of high spiking activity in the filtered blue trace. Particular periods of interest highlighted 1–7 to display on expanded time scale in panels (B) . (B1) Spontaneous SWs with coincident low amplitude GCaMP peaks (green dashed line) and nested ripple oscillations (inset) before carbachol administration. After carbachol administration, the GCaMP signal had slow and large fluctuations up to Δ F / F = 30%, or about 40 times the SW signals (B2–B6) . (B2) A period showing one-to-one correlations between observed peaks in the GCaMP and LFP signals (green dashed lines). (B3–B6) Periods showing minimal correlation between LFP and GCaMP fluctuations (comparing black and red traces, green dashed line marks onset of GCaMP peaks). (B4,B5) Periods showing anti-correlation between spiking rate and GCaMP fluctuations (comparing blue and red traces, green dashed line marks onset of GCaMP peaks). (B7) Onset of theta oscillations, showing large GCaMP peaks, high correlations between LFP and GCaMP, and synchronous bursts of spikes.
### Cellular Transients Compared to Population Signals
Finally, we wanted to verify that compared to the population GCaMP6f signal, cellular calcium transients were large, localized to the soma, and with longer duration as previously reported ( ; ). In contrast, the population GCaMP signals in the same tissue were expected to be small, distributed over a large area, and with shorter duration.
To test this, a region of CA3 str. pyramidale was imaged for cellular calcium transients ( , red hexagon). Five detectors were chosen to show both SW population signals ( ) and cellular transients ( ). In normal ACSF the SW signals were seen in all five detectors ( ) as well as in most of the detectors throughout the imaging area (SW marked in shaded blue box in displayed over all detectors in ). Later, 2 μM carbachol was added to the perfusion solution to promote cellular spiking, a low concentration that was found to be insufficient to induce theta bursts. Under this condition, localized large calcium transients were observed under one or a few detectors, suggesting cellular transients from spiking of individual neurons. Because the diode array had a low spatial resolution, these calcium transients cannot be attributed to individual CA3 neurons. However, these large signals were localized to one or a few detectors; e.g., the signals on neighboring red, blue, and green detectors only showed small crosstalk ( ), suggesting that the source of the signals was highly localized to the soma or dendrites of distinct CA3 neurons. In contrast, SW signals were distributed over the entire field of view.
Comparison between cellular transients and population GCaMP signal. (A) Schematic diagram of the imaging field. Green band marks the mossy fiber bundle. Red hexagon marks a subset of optical detectors imaging the CA3 pyramidal cell layer. Color dots mark five individual optical detectors; signals on these detectors are displayed in the color traces in panels (B,C) . (B) Traces of LFP (black) and GCaMP (colors) signals during SW events. LFP was recorded from a location in CA1, marked by the black dot on the top right of panel (A) . Color traces were from optical detectors (color dots in panel A ). Blue box marks the particular SW event with optical recordings from the all detectors (black traces in the imaging field in panel A ), demonstrating that a large fraction of optical detectors around the mossy fiber bundle show SW signals. (C) Cellular Ca transients recorded from individual detectors in the CA3 pyramidal layer. Color traces are from the same detectors in panels (A) . The signals in panels (B,C) were from the same tissue, during normal ACSF and 2 μM carbachol, respectively, a low concentration to promote activity but insufficient to induce theta busts. Note that the cellular Ca transients were often localized to individual detectors, and of high amplitude and longer duration than the population SW signals in panels (A,B) .
The large and localized calcium transients showed a duration of 1–2 s, consistent with cellular calcium transients reported by other groups [e.g., ( ; )]. The amplitude, spatial distribution, and time course of these local calcium signals all varied, suggesting differing sources.
## Discussion
Our results suggest that Ca imaging of population activity can be a useful tool for monitoring activity critical for the consolidation and encoding of memory, e.g., SWs ( ) and theta oscillations ( ). By reliably detecting SW events, these data demonstrate that GCaMP6f is sensitive enough to detect population activity with sparse spiking and sub-threshold activity. With appropriate amplification and filtering of the population signal, we observed that the temporal limit of population Ca imaging is close to the response time of the GCaMP6f protein (40 ms), enabling detection of oscillations up to 20 Hz ( ). The range in amplitude for detected population signals with our method spans 3000-fold, from Δ F / F = 0.1–300% and exhibits a dynamic range different from the accompanying changes to the LFP. In particular, it may be more sensitive than the LFP during highly elevated and asynchronous activity, where the interpretation of the LFP is often ambiguous ( ).
### Temporal Limitations of Population GCaMP6f Signals
The rise time of the population GCaMP6f signal was measured to be 40.3 ± 10.8 ms, which is slower than organic calcium indicators [e.g., Cal-520, Rhof-4, ( ) reviewed by ]. These data indicate that this response time is the major limitation for GCaMP6f to measure fast oscillations. Our results suggest that the speed of GCaMP6f is sufficient for measuring oscillations below 20 Hz, while this can be pushed up to 40 Hz with offline signal processing. A better solution for detecting gamma oscillations would be faster calcium sensitive proteins ( , ; ).
### Population Signal vs. Cellular Transients
One major question raised by our results is whether the integration of CA1 cellular calcium transient would reproduce the population calcium signal seen during SWs. Cellular calcium imaging during SWs has been performed by the Ikegaya group from a large number of neurons with calcium transients in the CA1 area ( , ; ). A careful analysis in these studies ( ) found that while 79% of neurons displaying calcium transients participated in SW events, each SW event only recruited ∼4% of these neurons. Additionally, each neuron participated in only ∼5% of SW events. Since ∼70% of neurons have calcium transients with and without SWs, it is unclear if integrating the individual cellular calcium transients would generate the population signal we observed in this report [see Figure 4 of ]. A large fraction of uncorrelated cellular transients would only contribute to the background fluorescence.
In contrast, the majority of CA1 neurons receive both excitatory and inhibitory synaptic inputs during every SW event ( ). The calcium influx in the presynaptic compartments of both excitatory and inhibitory neurons in the CA1 neuropil would contribute to the population GCaMP6f signal locked to the LFP, given that each CA3 neuron projects to 2/3 of the CA1 area and makes 30,000 to 60,000 excitatory synapses onto CA1 neurons ( ; ). The post-synaptic response of the CA1 neurons might also have activate low threshold calcium channels [reviewed by ] and contribute to the population calcium signal. The current imaging method cannot distinguish whether the signals are from pre- or post-synaptic calcium influx. Our signals are most likely from both, in addition to the regenerative calcium dynamics in the dendritic tree (dendritic calcium spikes). Further experiments are needed to distinguish the source of the population calcium signal.
Cellular calcium transients reach Δ F / F = 30–2000% when recorded under a dark background with confocal or two photon microscopes. Under bright field fluorescent imaging the background is no longer dark so the fractional change would be greatly reduced. In our wide-field fluorescent imaging the cellular calcium transients range from 2 to 5 times the population GCaMP signals ( ). The population signal of SWs is a small intensity change over a brighter background (a light flux of ∼10,000 photoelectrons/ms) which would saturate EMCCDs.
### Greater Dynamic Range of Population GCaMP Signal
GCaMP6f and LFP signals showed a striking amplitude discrepancy during population events ( – ). Spontaneous and bicuculline-induced interictal events, as well as carbachol-induced theta oscillations showed a much greater Ca than LFP response. We hypothesize this greater dynamic range in Δ F / F values compared to the LFP to be due to the population GCaMP6f signal increasing more linearly with increased cellular participation. The LFP will be limited in magnitude due to precise synchrony of voltage-gated and synaptic currents, as well as volume conduction throughout the slice. In contrast, the much slower kinetics of Ca , and the sensitivity of GCaMP6f to depolarized voltages renders the population GCaMP6f signal more sensitive than the LFP in highly active and/or asynchronous environments.
This high amplitude GCaMP6f signal likely reflects increased population firing rate and not artifact, as we never observed such increases outside of interictal events or carbachol administration. The decay time of GCaMP6f is about 200 ms ( ), therefore, all calcium transients within the decay period should accumulate and contribute to the population signal. In addition to this, the continued accumulation of intracellular calcium will lengthen the population signal. Individual neurons’ calcium transients last on average 1 s for GCaMP6f [ , see also ( ; ) as well as measurements with organic calcium indicators ( )]. These long duration intracellular calcium transients are limited by calcium buffering/elimination processes ( ).
The accumulation of the population GCaMP signal was also clearly seen in the ramp-like signals in , where repetitive stimuli caused a rising ramp, with the ramp slope becoming steeper with higher stimulus frequencies. The ramp signal was much slower and larger compared to the signal induced by individual stimuli. This might partially explain the amplitude discrepancy. In contrast to the population accumulation, on a single cell level, post-synaptic potentials and the GCaMP6 signal time course are much better correlated ( ).
The accumulation of population GCaMP signals may offer a sensitive indicator for the “spiking density” in the population. Spiking density here refers to an increased firing rate on a temporal scale of ∼100 ms, which is distinct from the more common concept of synchrony, or coincident firing on a millisecond temporal scale. With this definition, high spiking density would not necessarily result in high LFP peaks, as sodium and potassium currents could negate each other if the firing rates between neurons are not closely synchronous.
### Population GCaMP Signals for Detecting Asynchronous Population Activity
During the transition between SW and theta oscillations ( ), the LFP showed only a nominal signal while the population GCaMP signal exhibited large fluctuations. Four observations in suggest that the fluctuations are not noise. First, correlations can be seen between small LFP peaks and GCaMP signals ( ), indicating that the fluctuations were not random. Second, the fluctuations only occurred during the transition between the SW and the theta oscillations ( ), but never during SW states. Third, the fluctuations gradually increased to become large peaks with one-to-one correlation with LFP bursts ( ), suggesting the population firing gradually became organized into theta oscillations. Fourth, while the LFP only displayed nominal peaks during the transition, the LFP electrode often picked up spikes from nearby neurons. The high firing rate of nearby neurons suggests an asynchronous state in the population ( ). In addition, the reduction of cellular firing was weakly correlated to GCaMP6f fluctuations ( ).
### SWs vs. Epileptiform Activity
There is an active debate whether in vitro SWs are more reflective of epileptiform or other pathological events ( ; ). We demonstrated that the two events have large differences in GCaMP6f characteristics. Spontaneous interictal events, while rare, can happen without changing the bath solution or the excitability of the slices in our preparation ( ). While the LFP did display altered shapes between the events, the GCaMP6f response was even more highly divergent, indicating that they are different types of events. This provides evidence that in vitro SWs are distinct from epileptiform activity.
### Limitations and Advantages of the Diode Device
In contrast to many calcium imaging experiments, we employed a diode array for our measurements. The limiting factor for the sensitivity of the device is the dark noise of the electronics. The intensity of excitation light needs to be high enough so the signal can be distinguished from the dark noise. Bleaching of GCaMP6f fluorescence is a major limitation for the method. In order to achieve >30 min of optical recording time, the excitation light needs to be adjusted as low as possible while maintaining sufficient signal to noise. Under light illumination intensity, the dark noise becomes a major limitation for small signals. When approximately 100 mW of LED light output was used through a 10 × 0.3 NA objective, which delivers <1 mW/cm onto the tissue, resulting in dark noise about 20% of the SW peaks. Higher excitation power may also get better sensitivity with a trade-off in optical recording time. However, 30 min of light exposure was more than enough for many experiments.
The high dynamic range of the diode array is a main advantage. For detecting 0.1% Δ F / F on top of 100% background fluorescent light, a 16–20 bit effective dynamic range would be needed. Such high dynamic range is necessary for detecting small population signal with a sensitivity comparable to LFP.
Some of our results are compatible with a recent in vivo photometry study ( ), in which high frequency stimuli generated a slow accumulative response and faster individual responses, and epileptiform events displayed high amplitude GCaMP6s signals. We were able to resolve fast signals up to 30 Hz on single trials, and able to record small signals (∼0.1 Δ F / F ) during hippocampal SWs. This is due to the diode array having a higher dynamic range and signal-to-noise ratio than photomultiplier-based devices. Further work is needed to verify if the high signal-to-noise ratio can be achieved under in vivo conditions with optical fibers.
## Conclusion
In conclusion, this work demonstrates that population GCaMP signals offer a useful complementary approach to image small and fast population activity, with comparable sensitivity to LFP recordings. A planned future direction is in vivo imaging of spontaneously occurred theta (4–8 Hz), alpha/mu (7–13 Hz) and beta (15–20 Hz) oscillations. Faster calcium sensors may allow more robust detection and monitoring of gamma oscillations in the 40 Hz range. GCaMP permits multiple-site no-contact recordings, revealing spatiotemporal dynamics of neuronal oscillations. In addition, optical signals are not disturbed by the artifact of electrical stimulation, suitable for applications requiring simultaneous recording and stimulation, e.g., augmenting EEG oscillations by transcranial repetitive AC or magnetic stimulation [Zhai group 2019, current biology 2019], which currently no other methods can achieve.
## Ethics Statement
This study was carried out in accordance with the laboratory animal welfare guidelines, NIH Office of Laboratory Animal Welfare. The protocol was approved by the Institutional Animal Care and Use Committee of Georgetown University Medical Center.
## Author Contributions
PL, HJ, and JYW conducted the experiments. All authors participated in the data analysis and the composition of the manuscript.
## Conflict of Interest Statement
WuTech Instruments is a company owned by JYW. The diode array used in this research is a gift of WuTech Instruments. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Group I metabotropic glutamate receptors (mGluRs) mediate a range of signaling and plasticity processes in the brain and are of growing importance as potential therapeutic targets in clinical trials for neuropsychiatric and neurodevelopmental disorders (NDDs). Fundamental knowledge regarding the functional effects of mGluRs upon pyramidal neurons and interneurons is derived largely from rodent brain, and their effects upon human neurons are predominantly untested. We therefore addressed how group I mGluRs affect microcircuits in human neocortex. We show that activation of group I mGluRs elicits action potential firing in Martinotti cells, which leads to increased synaptic inhibition onto neighboring neurons. Some other interneurons, including fast-spiking interneurons, are depolarized but do not fire action potentials in response to group I mGluR activation. Furthermore, we confirm the existence of group I mGluR-mediated depression of excitatory synapses in human pyramidal neurons. We propose that the strong increase in inhibition and depression of excitatory synapses onto layer 2/3 pyramidal neurons upon group I mGluR activation likely results in a shift in the balance between excitation and inhibition in the human cortical network.
## Introduction
Metabotropic glutamate receptors (mGluRs) form a diverse set of G-protein-coupled receptors that are divided into three groups, based on sequence homology, pharmacological properties, and signal transduction ( ). The most studied of the three is group I, which comprises mGluR1 and mGluR5, both of which act through Gq proteins. Group I mGluRs are located perisynaptically and are involved in a range of signaling and synaptic plasticity processes ( ). They are particularly known for inducing a form of long-term depression (LTD) at glutamatergic synapses, which can be mediated by either mGluR1 or mGluR5, depending on brain region, postsynaptic cell type, and specific pathways in which the synapse is involved ( ; ). In addition to their role in LTD, group I mGluR activation potentiates NMDA-receptor-mediated currents ( ; ), and can depolarize several types of neurons through activation of a Ca -dependent cation conductance and decrease of resting K current ( ; ; , ).
While most studies of mGluR function, as well as its therapeutic effects, have centered upon excitatory signaling and pyramidal neurons ( ; ), mGluRs can induce plasticity at GABAergic synapses through a variety of mechanisms ( ; ). Furthermore, group I mGluRs are expressed in several types of interneurons in both mouse and human brain ( ; ). Consequently, group I mGluRs depolarize specific types of interneurons ( ; ) and increase synaptic inhibition in rodent brain ( ; ). Activation of group I mGluRs can also synchronize network activity by eliciting synchronous spiking in low-threshold spiking interneurons ( ), which include Martinotti cells.
In recent years, group I mGluRs, and mGluR5 in particular, have become of increasing interest as potential therapeutic targets in neuropsychiatric and neurodevelopmental disorders (NDDs) ( ), including schizophrenia ( ), and autistic spectrum disorders (ASDs) ( ; ). For example, dysregulated group I mGluR-mediated plasticity was proposed to underlie the NDD pathophysiology of fragile X syndrome (FXS) ( ), since group I mGluR-mediated LTD is exaggerated in hippocampal pyramidal neurons in the FXS mouse model ( ). Strikingly, mGluR-elicited spiking in Martinotti cells has been shown to be reduced in the Fmr1-KO mouse model for FXS ( ). These findings led to clinical trials targeting mGluR5 in adults with FXS ( ; ). Unfortunately, these trials have thus far been unsuccessful, with reasons given ranging from patient age, and drug dosage level, to incomplete knowledge at a brain circuit rather than at a single cell level ( ; , ). Furthermore, rodent data on mGluR function has rarely been validated in the human brain. New work has started to confirm the existence of some of the effects of mGluRs in human cortex. The influence of group II mGluRs on glutamatergic transmission has recently been shown to be the same in human cortex as it is in rodents ( ), as has mGluR-mediated LTD in fast-spiking interneurons ( ). Given the importance of validation in humans of the basic mechanisms underlying therapies for cognitive disorders, we sought to confirm the effects of group I mGluRs in human cortex. Accordingly, we report that group I mGluRs increase inhibitory transmission onto several types of neurons in human cortex and identify depolarization of Martinotti cells as a potential mechanism. Furthermore, we confirm the existence of mGluR-mediated synaptic depression in human pyramidal neurons. Taken together, these results provide an essential step forward in understanding human mGluR-mediated signaling that may inform our understanding of their therapeutic actions in future clinical trials.
## Materials and Methods
### Acute Slice Preparation From Human Cortex
All procedures carried out involving patient tissue were approved by the VU University Medical Center Medical Ethical Committee and in accordance with the Dutch law and the declaration of Helsinki. All 40 patients provided written informed consent. The majority of cortical samples were taken from patients that suffered from drug-resistant epilepsy, in most cases due to hippocampal sclerosis ( ). During surgery, non-pathological tissue showing no structural abnormalities was resected from anterior and medial temporal cortex ( ) (in this paper shows the exact location and extent of the resection and what tissue block was taken to the lab) in order to reach the pathological focus. Tissue was immediately stored and transported to the physiology laboratory in ice-cold slicing solution containing (in mM) 110 Choline chloride, 26 NaHCO3, 10 D-glucose, 11.6 sodium ascorbate, 7 MgCl2, 3.1 sodium pyruvate, 2.5 KCl, 1.25 NaH2PO4, and 0.5 CaCl2. 350–450 μm thick slices were prepared in the same, carbogenated, solution and were left to recover in aCSF containing (in mM) 125 NaCl, 26 NaHCO3, 10 D-glucose, 3 KCl, 2 CaCl2, 1 MgCl2, and 1.25 NaH2PO4 at 35°C, and then for at least 60 min at room temperature. aCSF in both recovery and recording chambers was continuously bubbled with a mixture of 95% O and 5% CO .
Patient data for all subjects used in this study.
### Electrophysiology
Slices in the recording chamber were perfused with aCSF heated to 31–33°C. Recordings were made using borosilicate (GC150-10, Harvard Apparatus, Holliston, MA, United States) glass pipettes with a resistance of 3 – 5 MΩ, pulled on a horizontal puller (P-87, Sutter Instrument Co., Novato, CA, United States). Signals were amplified (Multiclamp 700B, Molecular Devices), digitized (Digidata 1440A, Molecular Devices), and recorded in pCLAMP 10 (Molecular Devices, Sunnyvale, CA, United States). Access resistance was monitored before, during, and after recording. Cells were discarded if the access resistance deviated more than 25% from its value at the start of recording, or if it exceeded 20 MΩ. For current-clamp recordings and voltage-clamp recordings of excitatory postsynaptic current (EPSCs), pipettes contained intracellular solution consisting of (in mM) 148 K-gluconate, 1 KCl, 10 Hepes, 4 Mg-ATP, 4 K2-phosphocreatine, 0.4 GTP and 0.5% biocytin, adjusted with KOH to pH 7.3 (± 290 mOsm). All EPSC recordings except those shown in were performed in the presence of 10 μM Gabazine (Tocris Bioscience, Bristol, United Kingdom). To measure evoked EPSCs (eEPSCs), a pipette filled with aCSF was placed on a stimulation electrode and positioned within 100 μm from the recorded neuron. Current pulses were applied using an ISO-Flex stimulation box, and timed by a Master 9 (A.M.P.I., Jerusalem, Israel). The stimulation pipette was positioned so that a clear postsynaptic response could be observed with a clear separation from the stimulation artifact ( ). The stimulus intensity was set to evoke a half-maximal current. Pulses were applied every 15 s and a baseline of at least 5 min was recorded after the eEPSC amplitude stabilized. After recording a stable baseline, 25 μM DHPG was perfused into the recording chamber for 5 min. After a 5-min washout period, eEPSCs were measured every 15 s for up to 40 min and responses averaged per 10-min bins. In a subset of experiments, shown in , eEPSCs were recorded during DHPG application. These recordings were performed in the absence of GABAzine, so as not to elicit network events. Spontaneous inhibitory postsynaptic currents (sIPSCs) were measured using an intracellular solution containing (in mM) 70 K-gluconate, 70 KCl, 10 Hepes, 4 Mg-ATP, 4 K2-phosphocreatine, 0.4 GTP and 0.5% biocytin, adjusted with KOH to pH 7.3 (±290 mOsm). IPSC recordings were performed in the presence of 10 μM CNQX (Abcam, Cambridge, United Kingdom) and 50 μM D-APV (Abcam). sIPSCs were recorded from pyramidal neurons located in L2/3 and interneurons located in layer 1.
mGluR activation increases synaptic inhibition onto human pyramidal neurons. (A) Example morphological reconstruction of a human pyramidal neuron (350 μm slice; dendrites in black, axon in gray). Inset: electrophysiological response to negative and positive current steps. (B) Experimental protocol. (C) Example traces showing IPSCs before (Pre), during (DHPG) and after (Post) application of DHPG. (D) DHPG elicited a lasting increase in sIPSC frequency in pyramidal neurons (repeated-measures ANOVA: F (2,10) = 16.84, p = 0.003; Tukey’s post hoc test: Pre vs. DHPG p < 0.05, DHPG vs. Post ns, Pre vs. Post p < 0.05). (E) sIPSC amplitude was not significantly affected by DHPG [ F (2,10) = 0.07, p = 0.929]. (F) Average rise time of sIPSCs in pyramidal neurons was slower after DHPG application [ F (2,10) = 7.22, p = 0.011; Tukey’s post hoc test: Pre vs. DHPG ns, DHPG vs. Post ns, Pre vs. Post p < 0.05]. Right panel, cumulative probability distribution of sIPSC rise times, average of probability distributions calculated for each cell. (G) Decay time of sIPSCs was slower after DHPG application F (2,10) = 5.82, p = 0.021; Tukey’s post hoc test: Pre vs. DHPG ns, DHPG vs. Post ns, Pre vs. Post p < 0.05]. Right panel, cumulative probability distribution of sIPSC decay times, average of probability distributions calculated for each cell.
### Post hoc Morphological Assessment
Slices containing biocytin-filled cells were fixed in 4% paraformaldehyde in PBS for 24 – 48 h at 4°C. Slices were washed at least 3 × 10 min in PBS, and incubated in PBS containing 0.5% Triton X-100 and 1:500 Alexa 488-streptavidin (Invitrogen, Waltham, MA, United States) on a shaker at approximately 18–23°C (room temperature) for 48 h. Slices were then further washed at least 3 × 10 min in PBS and mounted on glass slides in mounting medium containing 0.1M Tris pH 8.5, 25% glycerol, 10% w/v Mowiol (Sigma-Aldrich). The morphology of recorded cells was checked for identification of their cell type (see ; ). Selected cells were imaged using an A1 confocal microscope (Nikon, Tokyo, Japan) using a 10×, NA 0.45 objective, scanned at 0.5 μm × 0.5 μm × 1.0 μm (xyz) resolution. Cellular morphology was reconstructed using NeuroMantic software ( ).
### Immunohistochemistry
To assess the expression of mGluR1α in somatostatin-positive neurons, temporal cortical tissue was used from patients undergoing surgery for mesial temporal lobe epilepsy (MTLE; 1 male, 2 female, 25 – 47 years) and three autopsy controls, displaying a normal cortical structure for the corresponding age and without any significant brain pathology (1 male, 2 female, 25 – 49 years). The control cases included in this study were selected from the databases of the Department of Neuropathology of the Academic Medical Center, University of Amsterdam, Amsterdam, Netherlands. Tissue was obtained during autopsy and used in accordance with the Declaration of Helsinki and the AMC Research Code provided by the Medical Ethics Committee. All autopsies were performed within 24 h after death. Tissue was fixed in 10% buffered formalin and embedded in paraffin. 6 μm sections were incubated overnight at 4°C in primary antibody solution (mGluR1α, 1:100, monoclonal mouse SC-55565, Santa Cruz Biotechnology, Santa Cruz, CA; Somatostatin, 1:300, polyclonal rabbit, AB1595, Chemicon, Temecula, CA, United States). Sections were then incubated for 2 h at room temperature with Alexa Fluor 568-conjugated anti-rabbit and Alexa Fluor 488 anti-mouse immunoglobulin G (IgG, 1:200, Thermo Fisher Scientific, Waltham, MA, United States). Finally, sections were analyzed using a laser scanning confocal microscope (Leica TCS Sp2, Wetzlar, Germany).
### Quantification of GRM1 and GRM5 Expression
GRM1 and GRM5 expression levels were quantified using publicly available Allen Institute for Brain Science (AIBS) database on human single-cell transcriptomics at , where the detailed methods can be found. The transcriptomic data from Allen Institute comes from human temporal cortical tissue, postmortem or surgically resected, sectioned and dissected per layer ( ). The methods include single nuclei fluorescence-activated cells sorting (FACS) isolation based on DAPI and neuronal nuclei staining (NeuN), followed by Smart-seq v4 based library preparation and single-cell deep (2.5 million reads/cell) RNA-Seq.
The data on single nucleus GRM1 and GRM5 mRNA expression in transcriptomic types from AIBS database were pooled to represent higher-order hierarchical clusters (SST, PVALB, PAX6/LAMP5, and excitatory types) from selected cortical layers of interest. Violin plots were made using custom-made Matlab scripts (Mathworks, Natick, MA, United States), the plots represent distribution of mRNA expression on a log scale with counts per million (CPM) value of 4000.
### Analysis and Statistics
Electrophysiological data were analyzed using custom scripts in Matlab. All data are represented as mean ± standard error of the mean (SEM). Normal distribution of the data was tested using Shapiro-Wilk tests. Appropriate statistical tests were performed in Prism 7 (Graphpad, La Jolla, CA, United States), and are mentioned in the figure legends.
## Results
### Group I mGluR Activation Increases Inhibition Onto Human Pyramidal Neurons
Activation of group I mGluRs increases spontaneous inhibition in rodent cortex ( ). To test whether this holds true in human cortex, we recorded spontaneous inhibitory postsynaptic currents (sIPSCs) in pyramidal neurons in layer 2/3 of surgically resected human neocortex and activated group I mGluRs by a 5-min bath application of the agonist (S)-3,5-Dihydroxyphenylglycine (DHPG; ). Application of DHPG led to an increase in the frequency of sIPSCs in pyramidal neurons that lasted after the agonist washout from the bath ( ). Interestingly, while the amplitude of inhibitory events was unaffected ( ), both the rise and decay times were increased after washout of the agonist ( ).
### Group I mGluRs Strongly Activate Martinotti Cells in Human Cortex
A potential cause of the slower kinetics would be a change in membrane time constant caused by DHPG. However, the membrane time constant after completion of the experiment did not differ from that measured before the start of the experiment [before: 17.7 ± 2.2 ms, after: 15.3 ± 3.0 ms, paired t (5) = 0.931, p = 0.395]. As inputs that are further away from the soma appear to have slower kinetics due to the filtering properties of dendrites ( ), we hypothesized that the slower synaptic inputs elicited by DHPG might be onto distal dendrites and were therefore likely coming from Martinotti cells (MCs). We performed current-clamp recordings of putative MCs in layer 2/3 to assess whether group I mGluR activation would elicit a change in membrane potential. Putative MCs were identified by an ovoid-shaped cell body and bitufted proximal dendritic morphology in the DIC microscopic image and by a rebound action potential following a depolarizing current step. Post hoc reconstruction of the morphology of these cells showed that the axon of putative MCs branched out and terminated in layer 1 ( ; ). Application of DHPG caused a depolarization of 7.7 ± 1.4 mV before the start of action potential firing ( ) and led to action potential firing in 6 out of 7 MCs ( ). In one experiment, a connected pair of MC and pyramidal neuron was recorded ( ). Upon DHPG application, the MC started firing action potentials, and the pyramidal neuron received an increased number of inhibitory postsynaptic potentials (IPSPs, ). Analysis of the pyramidal neuron membrane potential following 50 MC action potentials showed distinct IPSPs ( , left panel). Performing the same analysis on randomly generated time points did not show a similar peak ( , right panel; p < 0.001). The latency between the peak of the MC action potential and the onset of IPSPs in the pyramidal neuron was 1.75 ms, with a jitter of 396 μs. Thus, action potentials elicited by DHPG in the presynaptic MC generate time-locked inhibitory responses in postsynaptic pyramidal neurons.
mGluR activation depolarizes Martinotti cells and leads to action potential firing. (A) Morphological reconstruction of an MC in human cortex (350 μm slice). Morphology was recovered post hoc for 5 out of 7 recorded cells. Asterisks denote places where a neurite was cut during slice preparation. Inset: electrophysiological response to negative and positive current steps. (B) Membrane potentials of MCs are depolarized by DHPG (Wilcoxon matched-pairs signed rank test, n = 7, W = 28, p = 0.016). (C) DHPG induced an increase in action potential frequency (Wilcoxon matched-pairs signed rank test, n = 7, W = 21, p = 0.031). (D) Voltage traces of a connected pair consisting of an MC (teal) and a pyramidal neuron (gray). Application of DHPG (orange) induces sustained action potential spiking in the MC. (E) Voltage traces of MC and pyramidal neuron before and during application of DHPG (dashed lines indicate corresponding area of the trace in D ). Dashed boxes denote the area used for the analysis in f. (F) Average pyramidal neuron voltage trace (left panel, 50 events, light gray area shows SEM) around MC action potentials (left panel, teal dash) shows an inhibitory response that is absent in voltage traces centered on random time points during the same period (right panel; Mann-Whitney U = 418, p < 0.001). (G) Immunohistochemical staining for somatostatin (cyan) and mGluR1a (yellow) shows that mGluR1a is present in SST interneurons (arrowheads) in both resected and post-mortem tissue. Scale bar = 10 μm. Right panel: percentage of SST cells positive for mGluR1a per subject. (H) Distribution of GRM1 and GRM5 RNA levels in SST cells. Data taken from the Allen Institute human single-cell RNA-seq database. Here and further, black dot shows the median, n number above is the number of cells (nuclei) plotted.
To confirm that DHPG could mediate its effect on local synaptic inhibition directly via Martinotti cells, we performed double-labeling immunohistochemistry for somatostatin and mGluR1a. We observed near-total colocalization of mGluR1a and somatostatin in samples from both surgically resected (22 out of 22 SST+ neurons from 3 samples) and post mortem (22 out of 23 SST+ neurons from 3 samples) human temporal cortex ( ). In addition, single-cell RNA-sequencing data from the Allen Brain Institute showed strong expression of both GRM1 and GRM5 in human SST interneurons ( ). We therefore conclude that Martinotti cells are equipped with group I mGluRs to directly respond to DHPG and mediate the increase in synaptic inhibition observed in pyramidal neurons in superficial layers of human temporal cortex following group I mGluR activation.
### Synaptic Inhibition Onto Layer 1 Interneurons Is Increased by Group I mGluR Activation
Martinotti cells are known to contact most types of interneurons in addition to pyramidal neurons. Therefore, we tested whether interneurons in layer 1 (L1) of the human cortex also receive more inhibitory input upon group I mGluR activation. To this end, we recorded sIPSCs in L1 interneurons ( ). Similar to pyramidal neurons, sIPSC frequency onto L1 interneurons was increased during and after application of DHPG ( ), without a change in sIPSC amplitude ( ). In addition to increased sIPSC frequency, 2 out of 12 L1 interneurons showed a small increase in holding current after DHPG application ( ). This increase in holding current corresponds to a depolarization of 5.4 and 6.7 mV when taking into account the input resistance of the cells. DHPG-induced depolarization in L1 interneurons is therefore unlikely to elicit action potentials. During current-clamp recordings, L1 interneurons exhibited a small depolarization or no response, but did not fire action potentials in response to DHPG ( , n = 3). Thus, we did not find any evidence that L1 interneurons contribute to the increase in synaptic inhibition upon group I mGluR activation. In accordance with this, human L1 interneurons express GRM5 , but only rarely express GRM1 according to Allen Brain Institute single-cell sequencing data ( ).
mGluR activation increases synaptic inhibition onto layer 1 interneurons. (A) Morphological reconstruction of a human L1 interneuron (350 μm slice). Morphology was recovered post hoc for 11 out of 15 recorded cells. Inset: electrophysiological response to negative and positive current steps. (B) Example traces showing IPSCs before (Pre), during (DHPG) and after (Post) application of DHPG. (C) DHPG elicited a prolonged increase in sIPSC frequency in L1 interneurons (repeated-measures ANOVA; F (2,22) = 12.09, p < 0.001; Tukey’s post hoc test: Pre vs. DHPG p < 0.001, DHPG vs. Post ns, Pre vs. Post p < 0.05]. (D) sIPSC amplitude in L1 interneurons was not significantly affected by DHPG [repeated-measures ANOVA; F (2,20) = 1.16, p = 0.333]. (E) Example current trace showing increased sIPSC frequency and shift in holding current upon DHPG bath application. Right panel, proportion of cells in which the holding current shifted upon DHPG application. (F) L1 interneurons are depolarized (cell 1, upper panel) or were unresponsive (cell 2, lower panel) to DHPG application. (G) GRM1 and GRM5 RNA levels in L1 interneurons. Data taken from the Allen Institute human single-cell RNA-seq database.
### Group I mGluRs Depolarize Fast-Spiking Interneurons, but Do Not Elicit Action Potential Firing
In rodents, fast-spiking (FS) interneurons can be depolarized by activation of group I mGluRs. To assess whether FS interneurons contribute to DHPG-induced inhibition in human cortex, we performed current-clamp recordings of FS interneurons ( ). Application of DHPG led to depolarization of all recorded FS interneurons ( , n = 7), but did not elicit action potential firing. In accordance with these results, analysis of single-cell sequencing data revealed that, similar to L1 interneurons, human PV FS interneurons express GRM5 , rather than GRM1 ( ). DHPG application did lead to an increase in the frequency and amplitude of IPSPs ( ). Although this increase in IPSP frequency is likely due to increased MC activity, it could also be caused by an increase in driving force due to the depolarized membrane potential, which would facilitate detection of events. However, we found no significant correlation between the increase in IPSP frequency and the level of membrane depolarization among FS interneurons (Spearman’s R = −0.26, p = 0.62). Thus, FS interneurons receive increased synaptic inhibition upon group I mGluR activation, but are themselves not likely to contribute to this effect.
mGluR activation depolarizes FS interneurons without leading to action potential firing. (A) Morphological reconstruction of a human fast-spiking basket cell (350 μm slice). Morphology was recovered post hoc for 4 out of 7 recorded cells. Inset: electrophysiological response to negative and positive current steps. (B) Voltage trace showing depolarization of a fast-spiking interneuron in response to DHPG. (C) Fast-spiking interneurons are depolarized by DHPG (Wilcoxon matched-pairs signed rank test, n = 7, W = –28, p = 0.016). (D) GRM1 and GRM5 RNA levels in FS parvalbumin interneurons. Data taken from the Allen Institute human single-cell RNA-seq database. (E) Representative traces showing an increase in inhibitory synaptic potentials during DHPG bath application compared to baseline. (F) DHPG increased the frequency of spontaneous inhibitory events in fast-spiking interneurons (Wilcoxon matched-pairs signed rank test, n = 6, W = 21, p = 0.031). (G) sIPSP amplitudes are increased by DHPG application (Wilcoxon matched-pairs signed rank test, n = 6, W = 21, p = 0.031).
### Excitatory Inputs Onto Human Pyramidal Neurons Exhibit mGluR-Mediated Depression
Finally, we examined whether excitatory inputs were equally affected by group I mGluR activation. In current-clamp, only 2 out of 10 pyramidal neurons responded to DHPG by firing action potentials ( ), although most L2/3 pyramidal neurons express GRM1 and GRM5 ( ).
mGluR activation reduces excitatory inputs to pyramidal neurons. (A) Voltage trace showing DHPG-induced action potential firing in a pyramidal neuron. (B) Proportion of pyramidal neurons that displayed a shift in holding potential in voltage-clamp (top panel, neurons from ). Lower panel, proportion of pyramidal neurons that fired action potentials in response to DHPG in current-clamp. (C) GRM1 and GRM5 RNA levels in L2/3 pyramidal neurons. Data taken from the Allen Institute human single-cell RNA-seq database. (D) Experimental protocol for recording sEPSCs. (E) Example current traces showing sEPSCs. (F) sEPSC frequency was not increased by DHPG [Friedman test, χ (2) = 3, p = 0.223]. (G) sEPSC frequency did not change significantly upon DHPG application [repeated-measures ANOVA: F (2,24) = 2.55, p = 0.122]. (H) Experimental protocol for recording evoked EPSCs, depicting placement of stimulus pipette (left panel), and example evoked responses (right panel) before (black) and after (gray) DHPG application. (I) Example of eEPSC responses during wash-in of DHPG (orange bar). Mean ± SEM of 4 responses binned per min. (J) DHPG decreased eEPSC amplitude up to 10 min after wash-out of DHPG [Friedman test, χ (4) = 11.8, p = 0.019. Post hoc : Bonferroni-corrected Wilcoxon matched-pairs signed rank test, 10 min vs. baseline, n = 8, p < 0.05].
We therefore examined whether DHPG increased excitatory inputs onto pyramidal cells by measuring spontaneous excitatory postsynaptic currents (sEPSCs; ). Application of DHPG transiently increased sEPSCs by 25% or more in 6 out of 14 pyramidal neurons. However, there was no significant increase in sESPC frequency overall ( ).
Group I mGluRs are known to induce depression of excitatory synapses. This is mediated by mGluR5, which virtually all L2/3 pyramidal neurons express ( ). To test whether human pyramidal neuron excitatory synapses undergo mGluR-mediated depression, we evoked EPSCs (eEPSCs) by electrical stimulation ( ). Indeed, application of DHPG acutely decreased the amplitude of eEPSCs relative to baseline ( ). Therefore, we conclude that pyramidal neurons in human cortex exhibit group I mGluR-mediated depression of excitatory synapses.
## Discussion
In this study, we addressed how activation of group I mGluRs affects microcircuits in superficial layers of the human neocortex. Our data demonstrate a cell-type specific recruitment of human cortical interneurons by group I mGluR activation. We find that Martinotti cells are strongly excited by group I mGluR activation, which increases the amount of inhibitory inputs to neighboring L2/3 pyramidal neurons. Somatostatin-positive interneurons in superficial layers of the human neocortex show strong abundance of mRNA for mGluR1 and mGluR5 receptors. Other local interneuron types, including fast spiking interneurons and layer I interneurons are depolarized by group I mGluR activation, but do not fire action potentials in response to this depolarization. Also, these interneuron types show a lower abundance of GRM1 and GRM5 mRNA. Furthermore, excitatory inputs to pyramidal neurons are suppressed by group I mGluR activation. Thus, the large increase in synaptic inhibition across cell types in superficial cortical layers and the depression of excitatory synapses most likely results in a net shift in the balance between excitation and inhibition in the cortical network.
In rodents, layer I interneurons and deep layer fast-spiking interneurons have previously been reported to fire action potentials upon mGluR activation with quisqualic acid ( ). We did not observe induced action potential firing in any human layer I interneuron or fast-spiking interneuron. This discrepancy could be due to the difference in pharmacological ligands used in the earlier study, which also activate ionotropic glutamate receptors in addition to metabotropic receptors. Our data are in agreement with metabotropic-specific ligand effects upon fast-spiking interneurons ( ) and layer 1 cortical interneurons in rodents ( ). Enhanced synaptic inhibition in fast-spiking interneurons and in layer 1 Cajal-Retzius cells is mediated by Martinotti cells in rodents. This effect is mediated by mGluR1a specifically ( ; ). Therefore, we propose that Martinotti cells mediate enhanced synaptic inhibition in human superficial temporal cortex in response to group I mGluR activation. While we did not see direct action potential firing in any other interneuron types besides putative Martinotti cells, we cannot exclude the possibility that other interneuron types may also be involved in the mGluR-mediated increase in synaptic inhibition we observed.
Our results show that activation of group I mGluRs can directly depolarize both Martinotti cells and fast-spiking interneurons. Since group I mGluRs are located mostly perisynaptically and can therefore likely be activated by spillover of glutamate from the synaptic cleft ( ), subsequent depolarization of these interneuron types may constitute a mechanism by which inhibition is increased upon a prolonged or very strong initial excitatory drive. Group I mGluR activation can alter neuronal excitability through a variety of differing mechanisms, including protein kinase C-mediated changes upon ion channels, or through calcium-dependent modulation of ion channels ( ). mGluRs have been proposed to be involved in epileptogenesis ( ) and group I mGluRs are upregulated in the hippocampus of patients with temporal lobe epilepsy ( ). In addition, studies have shown that the activation of mGluRs in hippocampal slices can increase epileptiform activity ( ). However, these studies often block GABAergic signaling in order to induce epileptiform activity, thereby disregarding the strong effect on inhibition we show here, and that is also observed in rodent hippocampus ( ; ). We therefore speculate that increased expression of mGluRs in epilepsy patients could be a homeostatic mechanism, rather than a direct component of the pathophysiology of epileptogenesis. In both cortex and hippocampus, group I mGluR-mediated increase in the frequency of inhibitory events is mediated by mGluR1 ( ; ; ). We observed consistent co-expression of mGluR1a and somatostatin in putative Martinotti cells from both surgically resected tissue and autopsy controls. However, because group I mGluRs have different roles in different populations of neurons ( ; ), it remains to be determined whether mGluR1 or mGluR5 is solely responsible for the functional effects demonstrated here. Specifically, we found that FS and a subset of L1 interneurons are depolarized to some extent by DHPG, an effect that might be due to activation of mGluR5, which both types express.
We found group I mGluR-mediated depression of excitatory synapses received by L2/3 pyramidal neurons, similar to that observed in the rodent brain. Group I mGluR-LTD has previously been shown in human cortex for excitatory synapses onto fast-spiking interneurons ( ). The finding of LTD at excitatory synapses on pyramidal neurons is similar to that in rodent hippocampus ( ). The LTD we observed is not particularly strong and is shorter in duration than has been found previously ( ). It is worth mentioning that while other studies in rodents typically use 100 μM DHPG, we only used 25 μM due to its strong acute excitation of network activity. Since the efficacy of DHPG in inducing LTD is dose-dependent ( ), this may explain why the LTD we observed was relatively small and short-lived. Overall, however, we demonstrate the occurrence of group I mGluR-induced LTD as a plasticity mechanism conserved across species, which means the aberrant LTD underlying the mGluR theory of FXS ( ) may also apply to mature human cortex. However, to test mGluR-mediated LTD in FXS patient brain tissue would require using postmortem brain tissue for neurophysiological recordings ( ), since surgically resected tissue as used in this study is not available from FXS patients.
In contrast to evoked excitatory responses, mean amplitudes of spontaneous events were not decreased by mGluR activation in our experiments. Group I mGluRs have been shown to increase the amplitude of excitatory synaptic spontaneous events in rodent somatosensory cortex ( ) and in rodent hippocampal interneurons ( ). It is possible that in our recordings, mGluR-induced depression of a subpopulation of synapses is masked by a simultaneous global increase in events of a relatively large amplitude ( ), and that synaptic depression is visible only during the simultaneous timed activation of multiple synapses that occurs when synaptic events are evoked using extracellular stimulation. Conversely, a depression of excitatory synapses might cause the smaller responses from these synapses to fall below the detection threshold for spontaneous events. This might also explain why we observed no increase in the frequency of sEPSCs in most pyramidal neurons, even though the increase in action potential firing in a subset of pyramidal neurons is quite robust, and we find an increase in sESPC frequency in superficial interneurons. That only a subset of pyramidal neurons responded to mGluR activation may indicate that there are functional subtypes of pyramidal neurons in superficial human cortex that could be distinguished by differential mGluR expression. Indeed, superficial human pyramidal neurons can be divided into two classes based on morphology and electrotonic properties and their somatic location within the cortex corresponds to specific ion channel expression ( ; ). It remains to be determined whether these subtypes correspond to pyramidal neurons that do or do not respond to mGluR activation, or whether mGluR responsiveness further subdivides one or both of these classes.
Finally, recent studies using human cortical tissue have shown that there are fundamental differences in how rodent and human neurons function ( ; ; ; ; ; ; ). It should be noted that although the human neocortex used shows no structural abnormalities, patients typically had a long history of seizures and had been exposed to a variety of anti-epileptic medications, thus we cannot conclude unequivocally that these factors have not influenced neuronal function in some form. However, specific cholinergic mechanisms and modulation of disynaptic inhibition between cortical pyramidal neurons are conserved between rodents and humans ( ; ), as are the action of group II mGluRs ( ), and group I mGluR-dependent LTD of excitatory synapses onto fast-spiking interneurons ( ). We show here that several aspects of group I mGluR activation in the cortex are preserved across these mammalian species. The balance of synaptic excitation to inhibition and the role for aberrant mGluR signaling is of increasing focus for the synaptic, network and behavioral phenotypes related to rodent NDD and neuropsychiatry models ( ; ; ; ). Notably, the specific aspects of group I mGluR function we validate as occurring in human cortex are also dysregulated in mouse models for FXS, notably enhanced LTD in hippocampal pyramidal neurons ( ) and altered GABAergic inhibitory function specifically mediated by mGluR1 ( ; ). At the start of the 21st century, just over one third of all licensed and approved pharmaceutical drugs directly or indirectly modulated G-protein coupled receptors ( ). However, our fundamental knowledge on the function of G-protein coupled receptors, specifically mGluRs, and their specificity of action upon different neuronal subtypes within the human brain is far from complete. Therefore, we believe that our data have direct implications for interpreting the actions of group I mGluR-mediated signaling not only in human cortical circuits, but for translational approaches when designing clinical models from NDD rodent data to test specific mGluR targets therapeutically.
## Data Availability
The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.
## Ethics Statement
This study was carried out in accordance with the recommendations of the VU University Medical Center Medical Ethical Committee and in accordance with the Dutch law and the declaration of Helsinki. All 40 patients provided written informed consent.
## Author Contributions
TK, RM, JD, IK, and HM designed the study. TK, JD, IK, JA, JO, MV, and RW performed the experiments. TK, JD, IK, NG, and EA analyzed the data. SI and JB performed neurosurgery. TK, RM, and HM wrote the manuscript.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
High concentration of cytoskeletal filaments, organelles, and proteins along with the space constraints due to the axon’s narrow geometry lead inevitably to intracellular physical crowding along the axon of a neuron. Local cargo movement is essential for maintaining steady cargo transport in the axon, and this may be impeded by physical crowding. Molecular motors that mediate active transport share movement mechanisms that allow them to bypass physical crowding present on microtubule tracks. Many neurodegenerative diseases, irrespective of how they are initiated, show increased physical crowding owing to the greater number of stalled organelles and structural changes associated with the cytoskeleton. Increased physical crowding may be a significant factor in slowing cargo transport to synapses, contributing to disease progression and culminating in the dying back of the neuronal process. This review explores the idea that physical crowding can impede cargo movement along the neuronal process. We examine the sources of physical crowding and strategies used by molecular motors that might enable cargo to circumvent physically crowded locations. Finally, we describe sub-cellular changes in neurodegenerative diseases that may alter physical crowding and discuss the implications of such changes on cargo movement.
## Introduction
Neurons are amongst the longest cells in most organisms, with lengths up to 1 m in humans ( ). This necessitates an efficient transport system that can move material from the cell body to distal processes. Some organelles, like endosomes, may have diameters up to 5 μm which are similar to the diameter of neuronal processes (200 nm to 20 μm) within which they move ( ; ; ; ). Thus, neurons may occasionally have local physically crowded regions where cargos that are similar in diameter to the neuronal process may stall.
Physical crowding is an outcome of a large number of macromolecules and organelles present in the cytosol of cells. A dense cytoskeletal network, pre-existing intracellular organelles, and a high concentration of proteins imply that the entire cellular volume in a given region is not always accessible to soluble molecules or organelles that move into a region. In line with previous literature, we refer to organelles, the cytoskeleton, and proteins as crowding agents. The volume not available is known as the excluded volume. The average concentration of proteins within a cell ranges from 17 to 35% by dry cell weight ( ). This concentration, along with an average size of about 50 kDa for typical soluble globular proteins, suggests that proteins are on an average closer to each other than their radius of gyration ( ; ). Macromolecular crowding describes the effect of proteins or complexes (e.g., microtubules, proteasomes) that exclude other small molecules from the space that they occupy. The presence of PEG of ∼2.5 nm or ∼20 nm radius leads to reduced association equilibria between TEM1-β-lactamase and β-lactamase inhibitor protein in vitro compared to solutions lacking PEG ( ; ). This reduced association is thought to arise from the reduced rate of diffusion of the reactants ( ), perhaps due to reduced available solution space. The relative decrease in the diffusive movement for any diffusing molecule in the cytosol as compared to water arising both from macromolecular crowding and the viscosity of the cytosol is termed as microscopic viscosity ( ). The microscopic viscosity is governed both by the concentration and the interactions between constituents of the solution ( ). The microscopic viscosity of the cytosol will directly influence the diffusive properties of each molecule in the cell.
Crowding-related challenges are likely faced by all cells ( ), but are particularly relevant in neuronal processes with a narrow axonal diameter, the narrowest of which can be as low as 160 nm in diameter in vertebrates ( ; ; ) and 100 nm in diameter in invertebrates ( ; ), both smaller than the known diameters of some organelles ( ; ; ). Additionally, neuronal processes have a high density of cytoskeletal elements ( ). Microtubules (MTs) and neurofilaments in neurons have been reported to have an average separation between them of about 25–100 nm ( ; ). By contrast in non-neuronal cells, some MTs can be closely spaced but several are separated by >200 nm ( ). Physical crowding is especially important when molecules, macromolecular complexes, and organelles need to move and position themselves within neuronal processes. Active transport that is dependent on molecular motors ( ; ; ) is one strategy to circumvent crowding that leads to fast transport within a cell. However, active transport also faces physical challenges, such as availability of free tracks for transport ( ) and crowded regions in the cytosol due to organellar exclusion ( ). Models of some neurodegenerative diseases are associated with decreased organelle velocity, organellar displacement, and increased organelle stalls ( ; ; ; ; ; ). Increased physical crowding by proteins, cytoskeletal polymers, and stalled membranous organelles in neurodegenerative diseases can all contribute to reducing cargo movement, thereby exacerbating the progression of neurodegenerative phenotypes.
Although there are many different sources of crowding in a neuron, how each source affects the movement of other classes of moving proteins, polymers, or organelles is currently unclear. This review examines existing evidence that suggests that physical crowding may influence cargo movement and the potential strategies that allow cargo to move despite crowding effects. The review concludes with possible experiments that may help delineate the role of physical crowding in influencing cargo movement in neurons.
## Physical Barriers to Diffusive and Active Cargo Movement
The location where some proteins or organelles function can be distant from the cell body in neurons ( ). Consequently, their transport through either diffusive or active (ATP dependent) mechanisms is essential. The different sources of physical crowding include: microscopic viscosity, macromolecules, organelles, diameter of the neuronal process, and cytoskeletal polymers. Some of these crowding agents may preferentially affect active over diffusive movement and are discussed below.
### Crowding Effects on Diffusion
The local environment can influence the diffusion of molecules via (i) reduction in apparent diffusion coefficients due to increased microscopic viscosity of the cytosol ( ) ( ; ) and (ii) excluded volume effects arising from the presence of macromolecules and polymers such as actin, neurofilaments, and MTs ( , ) ( ). Soluble proteins can diffuse at rates ranging from 0.02 to 50 μm /s ( ). Increasing the viscosity of an aqueous solution by adding a large molecular weight polysaccharide, Ficoll 70, leads to reduced diffusion coefficients for several types of molecules ( ). A similar 4–100 times reduction in diffusion coefficients of proteins compared to their diffusion coefficients in water is observed in the cytosol ( ; ). A decreased diffusion coefficient not only affects the movement of molecules but may also reduce reaction rates of fast reactions owing to reduced rates of bimolecular association ( ).
Schematic representation of sources of physical crowding in the axon. Magnified view of features that contribute to crowding shown in insets. (A) Actin cortical rings, deep actin. (B) Neurofilaments can physically crowd the neuron by excluding organelles and small molecules. (C) High concentration of soluble proteins and narrow axonal geometry can lead to a local increase in viscosity. (D) Stalled cargo can physically impede the movement of motile cargo and diffusive proteins. (E) The MT bundle can exclude organelles but may allow diffusion of small proteins.
Electron micrograph of myelinated axons from Central Nervous System (CNS) and Peripheral Nervous System (PNS). (Top panel) Some myelinated axons in the CNS have mitochondria (M) that can be up to half the diameter of the neuronal process. In the magnified view on the top right, we can see a number of microtubules (solid arrows) and neurofilaments (arrowheads). (Bottom panel) Small caliber axons in the Remak bundles (R) can be as thin as 200 nm. We also see the axon filled with mitochondria (M), microtubules (solid arrows), neurofilaments (arrowheads) and vesicles. Reprinted with permission from Frontiers in Neuroscience , 12, (2018) p. 467 ( ).
Additionally, protein diffusion can be retarded by the cytoskeletal network, such as at the axon initial segment (AIS) of the neuron that has a high concentration of F-actin and β-spectrin ( ; ; ). Dextrans of 70 kDa injected in the cell body are restricted to the somatodendritic region, as opposed to smaller 10 kDa dextrans that freely diffuse in the axon ( ). The exclusion of 70 kDa dextran depends on F-actin that can form a physically crowded barrier or sieve at the AIS. Although as yet unexplored, the bundle of microtubules present in axons may also reduce the available volume for free diffusion ( ). However, microtubule depolymerization by nocodazole has not demonstrated a change in diffusion coefficients of small molecules like water ( ) or GFP ( ). The solute size where microtubules may be able to act as a crowding agent is unclear.
### Crowding Effects on Active Transport: Microtubules
Active transport leads to fast movement of molecules over long distances ( ; ). This type of transport utilizes molecular motors that walk on either actin or MT tracks ( ; ; ). MT motors play a major role in long-distance transport in neurons. There are three major sources of crowding experienced by cargo trafficked on MTs: (i) molecular crowding on tracks, (ii) crowding due to organelles in the vicinity of moving cargo, and (iii) drag from the axonal cortex.
Sources of crowding on MT tracks may include MT-associated proteins (MAPs), cargo stalled along the tracks, or protein/organelles at MT ends. Since the average length of MTs is shorter than that of axons, MTs are present as an overlapping staggered array within the neuronal process ( ; ; ). Thus there are numerous MT ends along the neuronal process, locations where both motors and cargo are shown to stall ( ; ; ). Additionally, neuronal processes in an organism are flexible to allow movement ( ; ). A dynamic reorganization of the MT bundle may lead to non-uniform cargo stalling that further leads to rapid changes in local physical crowding in the axon ( ). To maintain cargo transport with dynamic overlapping MT tracks, a cargo must navigate multiple MTs to reach its destination, for instance, the synapse. Ability to move across tracks in part depends on the availability of a free track for motor attachment and movement.
Microtubule associated proteins, such as Tau and MAP2 ( ), or tubulin post-translational modifications, like polyglutamylation or tyrosination ( ; ), are known to physically compete with or affect the affinity of MT motors for binding sites on the MTs respectively ( ; ; ). However, the extent to which each MAP competes with MT motors can differ ( ). Since MAPs may be distributed differentially on the minus and plus end of a given MT ( ), movement of MT motors may vary depending on their location on the MT ( ; ). As an example, the plus TIP complex that forms at the plus end of MTs may physically compete with MT motor binding, forcing the release of MT motors and associated cargo from the MT plus end ( ). In vitro , kinesin motors are also known to physically crowd the plus end of MTs leading to a slowing of other kinesins approaching the plus end ( ). Therefore, MT motors face many obstacles on the transport path to facilitate the active movement of cargo.
### Physical Crowding Effects on Active Transport: Organelles
Transported cargo that varies in size from 0.04 to 2 μm ( ; ; ) can also be physically obstructed by objects present in its vicinity. These include stationary cargo in close proximity, actin-rich regions along the axon ( ), and trapping of cargo by the actin cytoskeleton in the dendrite ( ). A subset of different stationary cargo has been observed in the neuronal processes of a variety of neurons ( ; ; ). These stationary cargos can themselves physically impede the transport of any moving cargo that encounter them, irrespective of the cargo type, thereby leading to a local build-up of stalled cargo ( ). Moreover, increased cargo stalling and reduced cargo run length are observed in crowded regions that are enriched in actin ( ; ). Therefore, both actin-rich regions and stationary cargo at actin-rich regions may act as local crowding agents for moving membrane-bound organelles (MBOs) irrespective of cargo type.
### Physical Effects on Active Transport: The Axonal Cortex
Cargos may also face increased drag due to interaction with the axonal cortex. In neurons, the axonal diameter ranges from ∼160 to 20 μm in humans ( ; ), and ∼100 to 350 nm in Caenorhabditis elegans ( ; ). Cargos that are transported in axons may occasionally contact the axonal cortex ( ; ; ). For instance, in axons, retrogradely moving lysotracker-marked vesicles induce local stretching of the plasma membrane of the axon ( ). This could lead to increased drag, and may cause moving organelles to encounter greater stiffness from the proximate axonal cortex compared to stiffness encountered within the cytosol. Consistent with this hypothesis, the speed of large organelles in neuronal processes is shown to reduce with increasing organelle size and reducing axonal diameter ( ; ).
In conclusion, there are many crowding agents in the neuron that can impede the movement of molecules within the neuronal process ( ; ; ; ). Cytosolic diffusion is affected by the viscosity of the medium where the cytoskeletal filaments themselves can act as molecular sieves ( ; ). On the other hand, active transport by molecular motors on MT tracks is affected by crowding on MT tracks ( ; ; ), by the axonal plasma membrane ( ; ; ), organelles, and cytoskeletal filaments near the tracks ( ) ( ; ). Both diffusion and motor-dependent movement face physical constraints that cargo must circumvent for steady transport.
## Strategies for Bypassing Crowding Depending on the Type of Neuronal Cargo
Neuronal cargos can vary in size and chemical composition. However, each type of cargo achieves long-range transport that can bypass physical crowding. Small molecules may be transported along neuronal processes by diffusion, active transport, or a combination of both. This may prove sufficient for small proteins, however, large protein complexes with potentially slower diffusion rates may be transported largely by active transport. Strategies to bypass crowding are discussed below.
### Strategies for Small Molecules
We define small molecules as any molecule below the molecular weight of 100 kDa (∼5 nm radius), and here, we discuss movement of this size class of proteins. In neurons, MTs form bundles separated from each other by about 25 nm in the axon (by Tau) and by about 65 nm in the dendrite (by MAP2) ( ; ). Tau cross-links between different MTs are transient, and Tau is known to diffuse along the MT lattice ( ). Separation of 25 nm between axonal MTs is unlikely to allow entry of large organelles or molecular complexes within the MT bundle. Moreover, this deep axonal region is rich in neurofilaments and actin that can be as many and as closely spaced as MTs ( ; ). However, the diameter of neurofilaments (6–10 nm) is lower than that of MTs (25 nm) ( ). Thus, the deep axonal space may preferentially be available for protein diffusion both along MTs and in the cytosolic spaces between MTs and other cytoskeletal elements ( ).
Strategies to maintain cargo movement that can bypass physical crowding. (A) Multiple active motors can increase pulling force or reverse. (B) Motors can switch to a MT track that is less crowded. (C) Macromolecular complexes can be transported by transient association with organelles interspersed potentially with diffusion. (D) Switch to an actin track from a MT track. (E) Small soluble proteins can preferentially diffuse deep within the axon (arrows indicate diffusion) rich in MTs and other cytoskeletal elements. Some MAPs can also diffuse along MTs. These proteins can evade crowding near the axonal cortex.
Soluble proteins, such as Casein Kinase 1 (CK1) and Ca /calmodulin-dependent protein kinase IIa (CamKIIa), are also actively transported. Soluble proteins are actively transported in two ways: by associating with synaptic vesicles as shown for CK1 ( ) and CamKIIa ( ), and by associating with molecular motors as shown for choline acetyltransferase that directly binds to Kinesin-II ( ). Due to transient association with MBOs and their small size, soluble proteins undergo assisted movement characterized by bouts of processive transport followed by periods of classic diffusive behaviors ( ) ( ; ; ). Although intermittent active transport may reduce the overall speed of transport as compared to continuous active transport at large distances (>100 μm), the ability to diffuse buffers soluble proteins from MT-based crowding factors.
Macromolecular crowding can have consequences on proteins that use assisted transport depending on the protein’s turnover rate in the axon. The turnover rate of a protein measures the time when half the existing protein is replaced with newly synthesized protein. This depends on the half-life of the protein, the rate of its synthesis, and rate of its transport. Fast turnover proteins like CK1 (half-life ∼16 h) ( ) may be more susceptible to the slowing down of transport from physical crowding along the transport path compared to CamKIIa, which is known to have a longer half-life (∼2 days) ( ). This difference in susceptibility may arise from the degradation of fast turnover proteins before it reaches the distal part of the neuronal process. Radioisotope pulse-labeling reveals the presence of proteins such as tubulin and neurofilament proteins for at least 45 days post labeling along the neuronal process ( ; ; ; ). One possible explanation for long lifetimes of proteins in the axon as found by radioisotope labeling ( ; used in ; ; ), in contrast to stable isotope labeling with amino acids in cell culture (SILAC) (used in ) based labeling, may be due to the axon having a slower degradation rate compared to the cell body or synapse.
In summary, small molecules may preferentially diffuse deep within the axon, but also associate intermittently with organelles leading to active transport. Depending on the turnover rate of the protein, multiple modes of transport can assist in maintaining the levels of any given protein at the distal end of the neuron.
### Molecular Motor-Based Strategies for Large Molecular Complexes
Large macromolecular complexes may be too large for substantive diffusive transport in axons given cytosolic viscosity. Therefore, the transport of molecular complexes may occur largely by active transport. Some examples of molecular complexes include the ribosomal machinery (25–30 nm) ( ) and the proteasome machinery (20S proteasome ∼15 nm, 60S proteasome diffusion coefficient ∼0.44 μm /s) ( ; ). One way to transport large macromolecular complexes is to move the components individually and then assemble them distally. One such example is the ribosome whose components can be assembled in a nucleolus-independent manner in the axons where both the individual ribosomal proteins and their mRNAs are independently actively transported in the axon ( ; ; ).
Another way to transport large macromolecular complexes includes association with an MBO. An example is a pre-assembled proteasome known to associate with MBOs that leads to its processive transport ( ). The interaction of the pre-assembled proteasome with MBOs is regulated through the adaptor protein PI31 ( ). Since the size of molecular complexes is of the order of small vesicles, the types of crowding agents affecting these molecular complexes while undergoing active transport are likely to be similar to those faced by MBOs. However, due to a half-life of 4–7 days of the proteasome subunits ( ), constant active transport may not be required. Large molecular complexes with slower turnover are likely to be more resilient to local transient crowding.
Unlike large protein complexes, RNAs are transported by aggregating into a Membrane-Less Organelles (MLOs) with the help of RNA binding proteins (RBPs) ( ; ; ). RNAs have a high negative charge and naked RNA has an extended geometry ( ) both of which are unfavorable for diffusion-dependent movement in the cytosol ( ). RNAs such as β-actin >500 kDa in size, ∼1.5 kb in bp length and ∼12 nm in physical length ( ) are known to have very low diffusion coefficients ( ). RBPs such as TDP-43, hnRNP A2, FUS ( ; ; ) can condense RNA and form MLOs. These MLOs are known to be actively transported by MT motors ( ; ). In contrast to protein complexes and organelles, MLOs are highly dynamic and undergo fission and fusion events while interacting with each other ( ). These deformations may allow MLOs to transport mRNAs through crowded locations by changing their size and geometry, thus bypassing some physical crowding bottlenecks.
Active transport appears to be the primary means through which large macromolecular protein complexes or RNA-protein complexes move along axons. Thus, the ability of such complexes to navigate crowding are likely similar to those used by MBOs (see below). Additionally, dynamic MLOs that can deform can permit easier movement across crowded regions in the axons.
### Molecular Motor-Based Strategies to Help Organelles Navigate Crowded Locations
In the above-mentioned cases, physical crowding can act as a deterrent to the movement of proteins and larger cargo. Molecular motors are a primary means to help MBOs and MLOs bypass physical crowding ( ; ). Distinct features of molecular motors allow them to help cargo maneuver across physical obstacles typically along the tracks that they use. These obstacles may be present on the MTs, such as MAPs, or present in the cytosol around the MT such as MLOs and MBOs. The strategy used depends on the types of molecular motors that a specific cargo use.
Motors can be distinguished based on the type of cytoskeletal filament they walk on (MT or F-actin). These motors include kinesins, dyneins (MT dependent) ( ; ) and myosins (F-actin dependent) ( ). Specific properties of motors can help cargo maneuver across crowded regions of the neuron. Mechanisms to overcome crowding on MTs can include: (i) dissociation of kinesin on encountering MAPs or other kinesins on the MT, as shown for Kinesin-1 ( ); (ii) binding to multiple MT protofilaments/MTs, which may allow the cargo to switch to another less physically crowded protofilament on the same MT or to switch to another MT ( ) ( ); (iii) dissociating with an increased frequency from MT plus ends as shown for Kinesin-3, that may prevent traffic jams along the MT ( ; ); (iv) frequent protofilament switching as is shown for dynein, that can help cargo navigate across physical obstacles ( ); and (v) a physically obstructed MT-attached cargo may switch to the F-actin cortical ring via unconventional myosin V that allows cargos to move tangentially to the MT bundle and access a less physically crowded region of the axoplasm ( , ) ( ; ). Thus, a combination of the above-mentioned characteristics of molecular motors can allow transported organelles to circumvent physically crowded locations along MT tracks.
Cargos such as endosomes can recruit multiple motors, such as kinesin and dynein ( ) ( ; ; ). Multiple motors are able to exert a greater pulling force ( ) that may allow organelles to continue moving despite plasma membrane drag or other cytosolic obstacles. Computational modeling suggests that the presence of MTs in close proximity is sufficient to recruit multiple motors on different MTs ( ; ). Moreover, in silico modeling also suggests that a combination of increased pulling force and track switching due to kinesins engaged on multiple MTs is sufficient for sustained cargo transport across organellar traffic jams ( ). Cargos are also known to recruit opposing motors ( ; ; ). This can lead to frequent switching of direction (viz. reversals) of motion ( ; ; ). Reversing cargos may be able to sample many more MTs owing to dynein’s ability to switch MT protofilaments while walking ( ; ; ). Increased MT sampling may, in turn, help cargos maneuver across obstacles present on MTs, and potentially in the cytosol as well. Indeed, reversals have been observed in multiple systems for multiple types of cargo ( ; ). These cargo reversals may also distribute the cargo along the track and maintain a steady supply to the distal ends ( ). The cytoskeletal network physically crowds the neuron and creates space constraints for organelle trafficking while providing the tracks required for movement. Mechanisms such as multiple motors, multiple tracks, and opposing motors are all strategies that use this contiguous cytoskeletal physical crowding in an advantageous manner to circumvent physical crowding on tracks or crowding from other organelles.
The neuronal cytoskeleton consists of regularly spaced (∼200 nm) braid-like actin cortical rings ( ) ( ; ), actin filaments deep in the axoplasm ( ), neurofilaments within the axoplasm ( ; ), and a staggered array of MTs of lengths from 1 to 10 μm in C. elegans and up to 760 μm in Mus musculus ( ; ; ). The cytoskeletal arrangement may be disrupted by external forces caused, for instance, through body movement that subjects the underlying neuronal processes to tensile and torsional forces. These stresses can induce local deformations in the neuronal process, which may disrupt active transport and result in non-uniform crowding ( , ). Neurons can resist deformations by maintaining the stiffness of the MT bundle through cross-linking between MTs ( ). Furthermore, uniform membrane tension allows the axon to regain its shape after deformation, a process thought to depend on axonal spectrin ( , ; ). Thus, recovering from deformation of the axon after movement and reducing deformation maintains neuronal shape and potentially helps in maintaining local cargo transport.
## Effects of Physical Crowding in Neurodegeneration
Neurodegenerative diseases due to a combination of genetic predisposition, environmental factors, and traumatic injuries are a confluence of multiple factors such as pathogenic aggregates, defects in transport, and collapse of the axonal cortex ( ; ; ; ). Many neurodegenerative diseases are known to be initiated by the formation of macromolecular aggregates ( ), mutations in MT motors and their adapters ( ; ; ), and increased concentration of reactive oxygen species (ROS) ( ; ). These defects further lead to (i) mitochondrial and lysosomal dysfunction ( ; ), (ii) changes in the cytoskeleton of the neuron ( ; ), and (iii) stalled MBO and MLO transport ( ; ; ). We discuss below the sub-cellular consequences of some neurological disorders whose pathologies can influence or be influenced in part by physical crowding, and their likely consequences on cargo movement.
### Crowding in Neurodegenerative Diseases
Abnormal aggregate formation of some RBPs (like TDP-43 and FUS) or structural proteins (like Tau and α-Synuclein) are characteristics of neurodegenerative diseases like frontal temporal dementia (FTD) and amyotrophic lateral sclerosis (ALS) for the former two ( ; ; ; ), and familial Alzheimer’s and Parkinson’s diseases for the latter two ( ; ). Increase in the size of aggregates over time is seen with Tau where the formation of 20–40 nm granular aggregates precedes larger 200 nm neurofibrillary tangles ( ). Tau oligomers can form granular aggregates up to 200 nm in diameter, or fibrils that are up to 1 μm in length and 20 nm in diameter ( ). RBPs generally form dynamic aggregates within cells ( ; ; ), which are hypothesized to become gel-like when transitioning to form pathological aggregates in neurons ( ). These RBP containing aggregates are seen to increase in numbers along the axon in diseased neurons ( ). Different types of aggregates (amorphous, ribbon-like, fibril, etc.) ( ) that are formed in the neuron have been thought to lead to divergent disease phenotypes (e.g., ALS, FTLD-TDP-A, and FTLD-TDP-C) ( ) that may arise from altering different cellular processes (e.g., recruitment of proteasome, interaction with MBOs, etc.) ( ; ). Dysfunctional aggregated Tau has been shown to (i) decrease binding of kinesin to MTs ( ; ; ), (ii) disrupt MT bundles ( ) ( ; ), and (iii) crowd the neuronal process ( ; ). Decreased MT bundle stability may also result in the collapse of the space between MTs ( ) that in turn leads to loss of soluble-protein diffusion within the MT bundle, reducing movement of material to the synapse.
Defects observed in neurodegenerative diseases. (A) Defective transport or increased physical crowding leads to increase in stalled cargo by disengaging motor-based transport as indicated by arrows. (B) Increased aggregation of cytosolic proteins lead to increased viscosity in the entire neuron. (C) MLOs may undergo a phase transition from a fluid to a gel-like state which further crowds the neuron as illustrated by the arrow. (D) Discontinuous coverage of MTs throughout the axon or (E) collapsed MT bundle lead to jamming of cargo due to unavailability of tracks and defects in cytosolic protein diffusion respectively in the neuronal process.
Despite independent origins of the above aggregates, all of them have the potential to alter physical crowding in the neuron. Small aggregates of Tau, α-Synuclein, and TDP-43 formed during disease initiation is associated with reduced trafficking of MBOs and MLOs ( ; ; ). Indeed, overexpression of proteins such as α-Synuclein that form pathogenic aggregates leads to a decrease in transmembrane and cytosolic pre-synaptic proteins at synapses ( ; ; ). This is consistent with the hypothesis that the presence of aggregates itself may physically impede cargo movement. The presence of protein aggregates may alter neuronal trafficking through both a disruption of specific cellular pathways as well as an increase in physical crowding.
Many animal models for neurodegenerative disease that resemble human late-onset neurodegenerative diseases are associated with an increase in the number of stalled MBOs, MBOs that travel shorter distances before stalling, and a decrease in the number of transported MBOs ( ; ; ; ; ). Transport defects in models of late-onset neurodegeneration may arise from three different processes: (i) a slow buildup of reduced MBO transport ( ; ), (ii) an increase in intracellular viscosity with age ( ) ( ), and (iii) a loss in regulation of the transport machinery ( ) ( ; ). These altered processes can, independently or synergistically with increased physical crowding, contribute to the progressive nature of the disease with aging. However, all three different processes can also directly increase physical crowding in the neuron and promote further cargo stalling. The sources of age-dependent increase in viscosity are currently unclear. Further investigation is necessary to distinguish between the relative contributions of cargo stalling and viscosity changes in worsening of late-onset neurodegenerative diseases.
Nearly all these diseases progress through an increase in crowding in the neuron. Defects in cargo transport in diseased neurons are thought to starve the distal ends of the neuronal process of freshly synthesized proteins and cargo ( ; ; ; ; ; ). These defects in cargo transport could partly occur through increased physical crowding, which subsequently reduces, over time, the material reaching the synapse. Progression of many neurodegenerative diseases that result in death of a neuron is associated with axonal swellings and “dying-back” of the neuron from its distal end. Dying back may occur due to starvation of the distal neuronal ends of different types of materials transported from the cell body ( ; ).
### Crowding in Brain Injury and Demyelination
Axonal swellings are also seen in cases of physical injury [e.g., traumatic brain injury (TBI)] ( ), or where the myelin sheath is inflamed and destroyed (e.g., demyelinating diseases like multiple sclerosis) ( ; ). These external factors lead to increased swelling due to disruption in the cytoskeleton ( ), or damage due to increased local ROS ( ; ). TBI exerts large rotational and stretch forces that can lead to axonal buckling and local swelling ( ). These local swellings are hypothesized to occur due to loss of membrane tension through disruption of MT-membrane interactions ( ). These swellings may act temporarily as locally uncrowded regions, where material can freely diffuse ( ). However, over time these periodic swellings have bent and non-uniform MTs that begin depolymerizing 3 h post-injury ( ). The disruption of MT tracks leads to loss of transport at this local swelling, leading to an accumulation of vesicles and proteins ( ; ; ). This accumulation of MBOs in the swelling can further reduce transport and diffusion rates due to physical crowding of the region ( ). Severe TBI is associated with a higher risk of Alzheimer’s disease later in life ( ; ). This increased risk may arise from the injury site continuing to act as a locally crowded region that favors aggregation of organelles and aggregation-prone proteins. This potential for increased crowding may be one factor that contributes to the observed increased risk of Alzheimer’s in TBI patients ( ). Demyelination diseases are caused by a loss of myelin ( ), essential for electrical conduction ( ; ). Survival of demyelinated neurons has been shown to require redistribution of mitochondria to the demyelinated region to scavenge locally increased ROS ( ; ). Mitochondria have been shown to slow down or stall transport of other organelles ( ; ). Thus, redistribution of a large organelle like a mitochondrion could physically occupy significant available space in the cross section of a narrow diameter axon ( ), and this crowding may be followed by additional reduction in transport by reducing the available space for other MBOs and MLOs to move through this region ( ). Mitochondrial redistribution is an early event in disease progression and may initiate physical crowding ( ; ). Mitochondria appear sufficient to physically crowd the neuron and stall other cargos in the vicinity ( ; ). As demyelination progresses, ovoids filled with organelles and altered MT structures are formed ( ) ( ; ). The local varicosities filled with organelles, similar to that seen in TBI, lead to secondary axotomy ( ). Therefore, loss in myelin of the neuron can lead to local physical crowding which in turn is one factor among many that may disrupt the trafficking of material from the cell body to the synapses ( ). The protective movement of mitochondria in the early stages of disease comes along with a potentially detrimental effect of increased physical crowding. In space-constrained axonal processes, this balance is a tradeoff between competing effects. Whether this crowding tradeoff is actively monitored, and if so, how it occurs, is worth investigating.
Sub-cellular events after injury/disease leading to axonal swellings. (i.A) Large tensile or torsional forces can buckle the cytoskeleton of the axon leading to local swelling of the axonal membrane. (i.B) Axonal swellings are temporarily locally uncrowded and allow free diffusion of proteins as indicated by lighter green. (i.C) Eventual accumulation at the time scale of minutes of organelles and destabilization of MTs in the swelling can lead to a loss of the movement of organelles and proteins illustrated by a darker shade of green. (i.D) Axonal swelling further may act as a locally crowded region promoting aggregation of proteins that later in life may predispose the neuron to degeneration. (ii.A) In the case of demyelinating diseases, external factors lead to a redistribution of mitochondria along the axons. (ii.B) Increase in mitochondria in a region may lead to increased crowding in its vicinity that reduces movement of organelles and cytosolic proteins as illustrated by a darker shade of green. (ii.C) Reduced movement of organelles and proteins at a region leads to swelling of the plasma membrane. (ii.D) Axonal swelling filled with organelles can lead to increased disruption of transport.
## Perspective
A mounting body of evidence suggests that intracellular physical crowding is an inevitable consequence of the geometry and content of the axon ( ; ; ; ). Physical crowding in healthy neurons is not detrimental to neuronal health ( ), perhaps in part due to molecular motors that have ways to circumvent locally crowded regions while carrying cargo over large distances ( ; ; ; ). However, any increase in physical crowding, as seen in multiple neurodegenerative diseases from classical tauopathies to TBI, may underlie some of the pathological changes that are observed in these conditions. Neurons seem to be the most adversely affected by aggregate-prone proteins such as Tau and RBPs ( ; ), even though these proteins are present in many cells ( ; ; ). The susceptibility of neurons to physical crowding might arise from the narrow diameter of the axon compared to the diameter of transported organelles, bundled MTs that exclude organelles between them, a continuously varying and dynamic cytoskeleton, stalled organelles, and actin-rich regions ( ; ; ; ; ; ). In healthy neurons, local crowding may not have detrimental consequences. However, in unhealthy neurons, increased crowding may explain the observation of reduced cargo movement over time ( ; ; ). Further, local crowding amongst other changes may lead to trafficking defects of multiple cargos over time ( ; ). Delineating the contribution of individual crowding agents toward overall physical crowding as impacts cargo movement or protein diffusion may help shed light on neurodegenerative disease progression.
However, crowding is not all detrimental. Neurons show multiple trade-offs in terms of function and crowding. MTs may crowd the neuronal process by excluding organelles from within the MT bundle, but the presence of the bundle provides structural integrity to the neuronal process ( ) and enables MT motor-dependent transport ( ; ). Organelles by their size and geometry cause crowding, but are essential for neuronal function e.g., synaptic vesicles and mitochondria. Large organelles like mitochondria can physically crowd the neuronal process, but their presence at nodes of Ranvier and at synapses are critical for neuronal function ( ; ). Further, Tau aggregates crowd neurons in tauopathies, but may also protect neurons from ROS-mediated damage ( ). Nonetheless, the density of organelles, and even protective aggregates formed during neurodegenerative diseases, may need to strike a balance. Too much physical crowding over time may adversely affect cargo movement ( ; ) where the ability of molecular motors to circumvent crowding fall short for the degree of physical crowding seen in unhealthy neurons. In vitro studies with defined physical crowding due to Tau have identified a precise degree of physical crowding on MT tracks where molecular motors are unable to sustain cargo displacement ( ; ; ). Similar studies with other MAPs or crowding agents may help delineate the contribution of crowding to progressive slowing down of organelle transport in neurodegenerative diseases.
Physical crowding can also be utilized by the neuron to promote the distribution of cargo throughout the neuronal process via slowing down of transport using physical barriers. Indeed, cargo such as synaptic vesicles and mitochondria are seen to distribute along the entire neuronal process, often at actin-rich regions ( ; ; ; ; ). These can lead to formation of naturally occurring cargo reservoirs that can be mobilized during cellular need ( ; ), such as during neuronal injury, where mitochondria localize to the cut-site ( ; ), or during repeated stimulation of the neuron where synaptic vesicles are mobilized to synapses ( ; ). Therefore, physical crowding is an inevitable consequence of the cellular design of axons but may be co-opted to distribute cargo along the neuronal process.
One way to reduce crowding in neurodegenerative models may be to gently perturb the MT and actin cytoskeleton. Cytoskeletal elements such as actin and MT physically crowd the neuronal process while maintaining the shape of the neuron, as well as enabling transport. Low doses of paclitaxel that stabilize MTs ( ) have been shown to reduce axonal swellings in a hereditary spastic paraplegia model ( ). One potential explanation is that drugs that target the MT cytoskeleton may be able to provide additional tracks for transport, thereby alleviating crowding. Further, a low dose of such drugs in combination with treatments to reduce aggregation have been shown to slow disease progression in patients ( ; ). This slowing of neurodegeneration may in part occur through reduced physical crowding in neurons.
To understand neurodegenerative disease progression, it is valuable to understand the effect of increased molecular crowding on the movement of different proteins and organelles. Currently, most studies have focused on movement of MBOs and MLOs by molecular motors ( ; ; ; ; ). We speculate that diffusive movement may also be altered, contributing to deprivation of proteins at distal ends of unhealthy neurons. It is interesting to note that diffusion at longer distances is slower than predicted by experimentally observed diffusion of the same proteins at shorter distances ( ; ). Therefore, experimental paradigms that assess diffusion at short length scales have to be used in conjunction with those that can assess diffusion at longer length scales. To investigate the contribution of crowding to diffusive transport of proteins at short length scales, one can use fluorescence correlation spectroscopy (FCS) ( ; ) and single particle tracking (SPT) ( ). FCS and SPT can assess differences in rates of diffusion dependent, for instance, on the density of organelles in the vicinity of the diffusing proteins. Further, studies using magnetic resonance imaging (MRI) have been used to calculate the apparent diffusion coefficient of water in vivo ( ; ). The diffusion coefficient of water has been suggested to depend on multiple factors, including the caliber of the axon ( ). MRI based imaging can help assess diffusion at longer length scales and is particularly suited for in vivo studies ( ). To assess long-term rates of active transport of MBOs without using radioactive tracers, experiments such as Retention Using Selective Hooks may be useful ( ; ; ). This method is particularly valuable as it might be able to bridge the gap between vesicle imaging over short-time scales of minutes and movement of cargo over hours that result in observed steady state distributions of cargo. Using this method one can begin to assess the role of crowding, for instance, from microtubule-based crowding agents and actin on long-term cargo movement in neurons. The above methods may also allow screening for drugs that aid in alleviating crowding while concomitantly increasing movement of cargo in neurons.
Several studies are necessary to understand the role of physical crowding in the neuron on both diffusive and active transport. Additionally, understanding how crowding changes during neuronal injury or neurodegenerative disease and whether it has both protective and detrimental effects are all open to investigation. Physical crowding effects may be as important as molecular pathways in progression of neurodegenerative diseases.
## Author Contributions
SK conceived the study. VS and SK wrote the manuscript.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer AK declared a past collaboration with one of the authors, SK, to the handling Editor.
|
Spinal cord injury (SCI) causes loss of normal sensation and often leads to debilitating neuropathic pain (NeP). Chronic NeP develops at or below the SCI lesion in as many as 80% of patients with SCI and may be induced by modulators of neuronal excitability released from activated microglia and macrophages. In the inflammatory response after SCI, different microglia/macrophage populations that are classically activated (M1 phenotype) or alternatively activated (M2 phenotype) have become of great interest. Chemokines have also recently attracted attention in neuron-microglia communication. CCL21 is a chemokine that activates microglia in the central nervous system (CNS) and is expressed only in neurons with an insult or mechanical injury. In this study using an SCI model in mutant ( plt ) mice with deficient CCL21 expression, we assessed post-SCI NeP and expression of microglia/macrophages and inflammatory cytokines at the injured site and lumbar enlargement. SCI-induced hypersensitivities to mechanical and thermal stimulation were relieved in plt mice compared with those in wild-type (C57BL/6) mice, although there was no difference in motor function. Immunohistochemistry and flow cytometry analysis showed that the phenotype of microglia/macrophages was M1 type-dominant in both types of mice at the lesion site and lumbar enlargement. A decrease of M1-type microglia/macrophages was seen in plt mice compared with wild-type, while the number of M2-type microglia/macrophages did not differ between these mice. In immunoblot analysis, expression of M1-induced cytokines [tumor necrosis factor-α (TNF-α), interferon-γ (IFN-γ)] was decreased in plt mice, while that of M2-induced cytokines interleukin-4 (IL-4, IL-10) did not differ in the two types of mice. The results of this study indicate that suppression of expression of inflammatory cytokines by decreasing the number of M1-type microglia/macrophages at the injured site and lumbar enlargement is associated with provision of an environment for reduction of NeP. These findings may be useful for the design of new therapies to alleviate NeP after SCI.
## Introduction
The International Association for the Study of Pain defines neuropathic pain (NeP) as that associated with anatomical or functional abnormalities of the nervous system (Merskey and Bogduk, ). Spinal cord injury (SCI) results in loss of normal sensation and often causes debilitating NeP such as allodynia that is a persistent problem for many patients. Pain at the level of the spinal segment occurs in 37–50% of these patients, while 76–83% have pain below the lesion level (Ravenscroft et al., ; Turner et al., ; Siddall et al., ; Calmels et al., ; Jensen and Finnerup, ; Nagoshi et al., ). These symptoms are associated with significant impairment in health-related quality of life (Woolf and Mannion, ; Jensen et al., ; Doth et al., ; Finnerup et al., ; Inoue et al., ; Nakajima et al., ), but current medications are often ineffective for NeP after SCI. Therefore, greater attention to NeP is required since it is clinically important to lessen pain.
The underlying mechanisms of NeP after SCI are multifactorial and change with time, but spinal and supraspinal lesions are the main causes of NeP. Several studies of the pathomechanism of NeP after SCI have shown that monocytes, macrophages, and especially glial cells play important roles (Watanabe et al., ; Gwak et al., ; Chen et al., ). In particular, two subtypes of macrophages have become of great interest in SCI: classically activated macrophages (M1 phenotype) and alternatively activated macrophages (M2 phenotype; Gordon and Martinez, ; David and Kroner, ). The M1 phenotype is the product of exposure to T helper 1 (Th1) cytokines, such as interferon-γ (IFN-γ) and tumor necrosis factor-α (TNF-α). In contrast, the M2 phenotype is activated via T helper 2 (Th2) cytokines, such as interleukin (IL)-4 and IL-10 (Kigerl et al., ; Gordon and Martinez, ).
The M1 phenotype strongly expresses inflammatory cytokines that may be responsible for NeP, while the M2 phenotype has enhanced anti-inflammatory properties. Microglia can also be induced to M1 and M2 phenotypes under different conditions (David and Kroner, ), and activated microglia and macrophages cause allodynia after SCI at the injured site and at remote sites, such as in the brain and lumbar enlargement (Wu et al., ). We have shown that these M1-type cells are involved in post-SCI dorsal horn hyperexcitability and central NeP associated with pain-related substances (Matsuo et al., ; Watanabe et al., ).
There is increasing recognition of involvement of the chemokine CCL21 in initiation and maintenance of allodynia. In a peripheral nerve injury model, CCL21 is only expressed in damaged neurons and induces upregulation of the P2X4 receptor in microglia and macrophages. These P2X4 receptor-expressing cells are activated by adenosine triphosphate (ATP) and release pain-related factors such as brain-derived neurotrophic factors (BDNF) and/or inflammatory cytokines such as TNF-α, IFN-γ, and IL-6; and these events promote NeP after SCI ( ; Biber et al., ; Tsuda et al., ). These findings indicate that CCL21 has a specific role in neuron-microglia/macrophage communication and is a potential drug target for prevention of NeP.
Potential role of CCL21 in neuropathic pain (NeP) after spinal cord injury (SCI). CCL21 secreted by damaged neurons is the trigger to upregulate P2X4 receptors activated by adenosine triphosphate (ATP) in activated microglia and hematogenous macrophages, which might lead to NeP.
Several studies have analyzed the role of CCL21 in NeP in a peripheral nerve injury model, but little is known in SCI, including the effects on M1- and M2-type microglia/macrophage chemotaxis at the injured site and in remote regions. In this study, we assessed expression of inflammatory cytokines associated with M1- and M2-type activated microglia/macrophages at the injured site and lumbar enlargement after SCI in mutant ( plt ) mice with deficient CCL21 expression, in order to identify the role of CCL21 in NeP after SCI.
## Materials and Methods
### Experimental Animals
The study was conducted using C57BL/6 mice (age 10–12 weeks, male, n = 54) and plt (paucity of lymph node T cell) mice (age 10–12 weeks, male, n = 54; Mori et al., ; Nakano and Gunn, ) purchased from the Central Institute for Experimental Animals (Kawasaki, Japan). The plt mouse is a colony of DDD/1 inbred mice exhibiting greatly diminished T cell numbers in lymph nodes. This spontaneous deletion, designated plt , behaves as a single autosomal recessive allele and deletes a portion of chromosome 4, including both the Ccl19 and Ccl21a loci. The Animal Ethics Review Committee of Fukui University approved the study protocol.
### Contusion SCI Model
After treatment with isoflurane [2% (v/v); Forane; Abbot, Tokyo, Japan] to induce deep anesthesia, the mouse underwent laminectomy at the T9–10 vertebral level to expose the spinal cord, with use of a surgical microscope (Vanox; Olympus Optical, Tokyo, Japan). An Infinite Horizon Impacter (Precision Systems and Instrumentation LLC, Fairfax, VA, USA) was used to produce a contusion SCI model, using an impact force of 60 kilodynes. For sham SCI, laminectomy only was performed at T9-T10, with no SCI. The wound and surrounding skin were sutured with 5–0 silk.
### Behavioral and Sensory Testing
The Basso Mouse Locomotor Scale (BMS) is used to evaluate locomotor function after thoracic spinal cord contusion or transection injury (Basso et al., ). It is scored from 0 (hind limb paralysis) to 9 (normal locomotion). BMS scores were recorded at days 4, 14, and 28 post-SCI for each hind limb, and averaged to give one value per mouse per test. Two independent examiners who were blinded to the experimental results tested the mice at days 1, 14, 21, and 28 post-SCI for mechanical allodynia and thermal sensitivity. Allodynia sensitivity was tested using a Dynamic Plantar Aesthesiometer (Ugo Basile, Comerio, Italy; Martucci et al., ). In this test, the withdrawal threshold (expressed in grams) is determined five times and the mean is reported. Thermal sensitivity at the plantar hindpaws was examined using a Plantar Test Apparatus (Ugo Basile; Hargreaves et al., ). In this test, the time between application of the thermal stimulus until hindpaw withdrawal (latency) is recorded (in seconds), as well as any other reaction to the stimulus (e.g., gazing at the affected paw, sniffing, licking, or attacking the stimulus). The latency was calculated using data from six tests after rejecting the longest and shortest latencies (Hoschouer et al., ).
### Immunohistochemistry
For immunohistochemical analysis, mice were deeply anesthetized and transcardially perfused at days 4, 14 and 28 post-SCI, and the obtained tissues were fixed with 4% paraformaldehyde in 0.1 M phosphate-buffered saline (PBS). The spinal cord and the lumbar enlargement at L3-L4 were dissected out carefully and kept in a similar fixative. After a few hours in the fixative solution, the tissue samples were immersed in a mixture of 10% sucrose/0.1 M PBS and maintained at 4°C for 24 h, and then in another solution of 20% sucrose/0.1 M PBS for another 24 h. The injured site and lumbar enlargement of the spinal cord were embedded in OCT (optimal cutting temperature) compound (Sakura Finetek, Torrance, CA, USA) and then cut into serial 20-μm axial or sagittal frozen sections using a cryostat. The cut sections were serially mounted on glass slides and fixed for 5 min with 2% paraformaldehyde in 0.1 M PBS, followed by rinsing in PBS and storage at −80°C.
Immunohistochemical staining was performed after permeabilization of the frozen sections with 0.1 M Tris-HCl buffer (with 0.3% Triton X-100, pH of 7.6). The sections were treated overnight with the following primary antibodies at 4°C, which were diluted with the Antibody Diluent with Background Reducing Components (Dako Cytomation, Carpinteria, CA, USA): mouse rat anti-CD11b monoclonal antibody (Abcam plc, ab1211, Cambridge, UK, diluted 1:200) for microglia/macrophages; rabbit anti iNOS (Proteintech, 18985-A-P, Chicago IL, USA, dilution 1:50) for M1-type macrophages; and rabbit anti-mannose receptor antibody (CD206; Abcam plc, ab64693, diluted 1:500). The sections were then incubated with Alexa Fluor-conjugated 488 or 568 secondary antibodies (dilution, 1:250, Molecular Probes, Eugene, OR, USA) for 1 h at room temperature. Finally, the sections were washed, wet-mounted, and examined by fluorescence microscopy (Olympus AX80, Olympus Optical, Tokyo) or a confocal laser scanning microscopy (TCS SP2, Leica Instruments, Nussloch, Germany), using an argon/helium-neon laser at 488 and 543 nm for fluorescence excitation.
### Semi-quantitative Analysis
Semi-quantitative analysis of the numbers of CD11b /iNOS and CD11b /CD206 cells (merged cells; yellow) at days 4, 14, 28 post-SCI was performed in five axial sections selected randomly at about ±500 μm from the epicenter of the injured site and at the L3-L4 level for the lumbar enlargement. High magnification (×200) photomicrographs (TCS SP2; Leica Microsystems) of superficial laminae I-III on one side of the spinal dorsal horn were analyzed using grain counting with the light intensity automatically set by the color image analysis software (MacSCOPE; Mitani; Hansen et al., , ; Watanabe et al., ). The light intensity and threshold values were maintained at constant levels when collecting digitized images in all analysis.
### Immunoblot Analysis
For immunoblot analysis, the injured site (tissue harvested from 2.5 mm on either side of the injured site) and lumbar enlargement (between L3 and L4) were carefully dissected en bloc from the area (5 mm; n = 3 mice at each time point) and stored at −80°C. Sections were centrifuged at 15,000× g for 30 s (BioMasher Rapid Homogenization Kit, Funakoshi, Tokyo) and then solubilized in RIPA lysis buffer 1× (Santa Cruz Biotechnology, Santa Cruz, CA, USA), homogenized and stored at −80°C. Protein concentrations in tissue samples were determined by Lowry protein assay (DC Protein Assay Kit; Bio-Rad Laboratories, Hercules, CA, USA). Protein mixtures were mixed with Laemmli sodium dodecylsulfate buffer and boiled prior to immunoblot analysis. The total protein (20 μg/lane) was separated on 12.5% SDS-PAGE and transferred onto a polyvinylidene difluoride membrane (PE Applied Biosystems, Foster City, CA, USA) for 70 min. The membranes were washed twice in PBS solution containing 0.05% Tween 20, then blocked with a mixture of 5% skimmed milk in PBS for 1 h at room temperature, and finally incubated overnight at 4°C with an antibody (all Abcam plc) against the following proteins: TNF-α (ab6671, dilution 1:500), IFN-γ (ab133566, dilution 1:1,000), IL-4 (ab11524, dilution 1:500), and IL-10 (ab33471, dilution 1:500) in blocking solution. After washing three times in 0.1 M PBS, the membranes were immersed in medium from an ECL Advance Western Blot Detection kit (GE Healthcare, Little Chalfont, UK) for 1 min and analyzed by imaging (Image Quant LAS 4000, GE Healthcare Life Science, Piscataway, NJ, USA). Each band intensity was quantified using Image Quant TL software (GE Healthcare Life Science) and expressed relative to the intensity of the band for β-actin. Kaleidoscope Prestained Standards (Bio-Rad Laboratories) were used as molecular weight controls.
### Flow Cytometric Analysis
Flow cytometric analysis was conducted using tissues harvested from 2.5 mm on either side of the injured site and between L3 and L4 of the spinal cord at days 4 and 14 post-SCI, as described previously (Watanabe et al., ). Hematogenous macrophages/activated microglia were detected as CD45 /CD11b /GR-1 cells (Saiwai et al., ). For intracellular staining (Stirling and Yong, ), the harvested cells were resuspended in fixation buffer and then treated with permeabilization buffer (Santa Cruz Biotechnology). They were then resuspended in ice-cold PBS and incubated for 1 h with one of the following antibodies: BV510 rat anti-CD11b (BD Horizon, 562950, Piscataway, NJ, USA, dilution 1:1,000), FITC rat anti-CD45 (Abcam plc, ab25670, dilution 1:1,000), and PerCP/Cy5.5 rat anti-Ly-6G/Ly-6C, equivalent to Gr-1 (BioLegend, 108427, San Diego, CA, USA, dilution 0.25 μg for 10 cells), rabbit anti-iNOS antibody-primary antibody (Abcam plc, ab15323, dilution 1:500), Alexafluor 647 equivalent to EPR25A secondary antibody (Abcam plc, ab199093, dilution 1:2,000) for M1 or EPR6828(B) rabbit anti-mannose receptor antibody equivalent to CD206 (Abcam plc, ab195192, dilution 1:50) for M2. Flow cytometry (FACSCanto™ II; BD Biosciences) was then performed with forward scattering to eliminate cellular debris. CD45 /CD11b /GR-1 /iNOS cells and CD45 /CD11b /GR-1 /CD206 cells were detected as hematogenous M1- and M2-type macrophages/activated microglia, respectively.
### Statistical Analysis
All values are expressed as mean ± standard deviation (SD). Differences between groups were examined for statistical significance using one-way factorial analysis of variance (ANOVA). A p < 0.05 denoted the presence of significant difference with Tukey’s post hoc analysis. The above tests were conducted using SPSS software (version 24.0, SPSS, Chicago, IL, USA).
## Results
### Locomotor Function, Mechanical Allodynia and Thermal Hyperalgesia After SCI
There was no difference in BMS locomotor scores between wild-type and plt mice after contusive SCI ( ). Hind paw mechanical and thermal sensitivity tests were performed from week 2 post-SCI, which was when all mice were able to place the plantar surface of the hind paws and support weight. The mechanical response thresholds at 2 and 4 weeks post-SCI were 3.53 g and 4.82 g, respectively, in wild-type mice, and 6.12 g and 7.53 g, respectively, in plt mice, showing a significant decrease in mechanical hypersensitivity in plt mice ( ). In the thermal sensitivity test, the withdrawal latency of wild-type mice gradually increased from 2.23 s at 2 weeks to 3.25 s at 4 weeks post-SCI. In contrast, a significant and marked improvement in thermal sensitivity was noted in plt mice from 2 weeks post-SCI, with withdrawal latency times of 5.44 s at 2 weeks and 6.55 s at 4 weeks post-SCI ( ).
(A) There was no difference in Basso Mouse Scale (BMS) scores between plt and wild-type mice. (B,C) SCI-induced hypersensitivities to mechanical and thermal stimulation were reduced in plt mice from day 14 post-SCI ( n = 10 for each time point). Data are shown as mean ± standard deviation (SD).
### Expression of M1- and M2-Type Microglia/Macrophages
Expression of M1- and M2-type microglia/macrophages at the injured site and lumbar enlargement after SCI was examined by immunohistochemistry. The number of CD11b /iNOS cells (M1 type) at the injured site peaked at 14 days after SCI in wild-type mice, and some of the CD11b-positive cells seemed to form clusters. There was significant suppression of the increase in these cells in plt mice at 4 and 14 days after SCI ( ). In the lumbar enlargement, the number of CD11b /iNOS cells tended to increase from 4 days to 14 days after SCI in both types of mice, but there was significant suppression of the increase in these cells in plt mice at 4, 14, and 28 days after SCI, compared with the number in wild-type mice ( ). The number of CD11b /CD206 cells (M2 type) after SCI was smaller than that of CDD11b /iNOS cells (M1 type) and did not differ at the injured site in wild-type and plt mice ( ). At the lumbar enlargement, there was a significantly larger number of these cells at 4 days after injury, but no differences at 14 and 28 days after injury ( ).
Immunofluorescent staining showing expression of M1-type microglia/macrophages at the injured site and lumbar enlargement after SCI using CD11b (green), iNOS (red) and DAPI (blue). Suppression of increases in CD11b /iNOS cells (M1-type microglia/macrophages; merged cells; yellow) in plt mice at the injured site (A) and lumbar enlargement (B) . Scale bars = 50 μm. The upper left panels in each image were high-power photomicrograph of the merged cells. The number of CD11b /iNOS cells was significantly decreased at days 4 and 14 post-SCI at the injured site (A) , and at days 4, 14 and 28 post-SCI at the lumbar enlargement ( n = 4 at each time point). Data are shown as mean ± SD. * p < 0.05.
Immunofluorescent staining showing expression of M2-type microglia/macrophages at the injured site and lumbar enlargement after SCI using CD11b (green), CD206 (red) and DAPI (blue). Expression of CD11b /CD206 cells (M2-type microglia/macrophages; merged cells; yellow) did not differ between wild-type and plt mice at the injured site (A) and lumbar enlargement (B) . Scale bars = 50 μm. The upper left panels in each image were high-power photomicrograph of the merged cells. The number of CD11b /CD206 cells did not differ significantly from day 4 to day 28 post-SCI at the injured site (A) and lumbar enlargement ( B ; n = 4 for each time point). Data are shown as mean ± SD. * p < 0.05.
### Flow Cytometry for M1/M2 Hematogenous Macrophages and Activated Microglia
Flow cytometry was used to examine the number of M1- and M2-type hematogenous macrophages/activated microglia in 100,000 sorted cells at the injured site and lumbar enlargement. There was a decrease in M1-type cells (CD11b /CD45 /Gr-1 /iNOS ) at 14 days after SCI at the injured site in plt mice (28,162 ± 8,683 in wild-type vs. 11,815 ± 6,529 in plt mice), but no difference at 4 days (23,357 ± 6,522 vs. 24,622 ± 5,430; ). In the lumbar enlargement, the number of M1-type cells was significantly lower in plt mice than in wild-type mice at 4 days (7,860 ± 1,210 vs. 1,065 ± 202) and 14 days (8,382 ± 810 vs. 1,941 ± 684) after SCI ( ). There were few M2-type cells (CD11b /CD45 /Gr-1 /CD206 ) after SCI among sorted cells, with no difference between wild-type and plt mice at the injured site and lumbar enlargement ( ).
Semi-quantitative flow cytometric analysis of M1-type activated microglia/macrophages at the injured site and lumbar enlargement in wild-type and plt mice. Flow cytometry of CD11b /CD45 /Gr-1 cells showed a significant decrease in iNOS-positive hematogenous M1 type microglia/macrophages at day 14 in the injured site (A) and at day 4 and 14 in the lumbar enlargement (B) after injury in plt mice compared with those in wild-type mice ( n = 3 each). Data are shown as mean ± SD. * p < 0.05.
Semi-quantitative flow cytometric analysis of M2-type activated microglia/macrophages at the injured site and lumbar enlargement in wild-type and plt mice. The number of M2-type cells after SCI was <5% in sorted CD11b /CD45 /Gr-1 hematogenous cells at the injured site (A) and lumbar enlargement (B) , with no difference at days 4 and 14 after injury between wild-type and plt mice ( n = 3 each). Data are shown as mean ± SD.
### Expression of M1- and M2-Induced Inflammatory Cytokines
Western blot analysis showed total protein level of M1-induced pro-inflammatory cytokines (TNF-α, IFN-γ) and M2-induced anti-inflammatory cytokines (IL-4, IL-10) at the injured site and lumbar enlargement after SCI. In plt mice, expression of TNF-α at both the injured site and lumbar enlargement was significantly reduced at 4 and 14 days after SCI, compared with wild-type mice. At 28 days after SCI, there was almost no difference in the expression of TNF-α between the two types of mice ( ). Expression of IFN-γ did not differ at the injured site, but decreased expression was seen at 14 days after injury at the lumbar enlargement in plt mice ( ). Expression of IL-4 and IL-10 did not differ, although these levels were slightly increased in plt mice at both the injured site and lumbar enlargement ( ).
Western blotting showing expression of M1- and M2-induced cytokines at the injured site and lumbar enlargement after SCI in wild-type and plt mice. (A) Expression levels of tumor necrosis factor-α (TNF-α) were significantly decreased from day 4 to day 14 after injury at the injured site and lumbar enlargement in plt mice compared with those in wild-type mice ( n = 3 each). (B) The expression level of interferon-γ (IFN-γ) did not differ at the injured site from day 4 to day 28 after injury, but was significantly decreased at day 14 after injury at the lumbar enlargement in plt mice compared with wild-type mice ( n = 3 each). Expression levels of interleukin-4 (IL-4; C ) and IL-10 (D) did not differ between wild-type and plt mice at the injured site and lumbar enlargement from days 4 to 28 after injury ( n = 3 each). Data are shown as mean ± SD. * p < 0.05.
## Discussion
The main finding in this study was that NeP after SCI was reduced in plt mice, which have deficient CCL21 expression, due to decreased infiltration of M1-type hematogenous macrophages/activated microglia and reduced expression of pro-inflammatory cytokines without affecting M2-type chemotaxis at the injured site and lumbar enlargement. This suggests that CCL21 could be a potential drug target for prevention of NeP in response to SCI.
CCL21 is a chemokine that is present in humans and has an important role in mobilizing normal immune cells in response to tumor cells metastasizing to lymph nodes, via activation of a G protein-coupled receptor, CCR7 (Love et al., ). Following storage in large dense core vesicles, neuronal CCL21 is transported along axons in an anterograde direction to presynaptic terminals, indicating the function of CCL21 in neuron-microglia signaling (de Jong et al., ). The purinergic P2X receptors, of which seven subtypes (P2X1R-P2X7R) have been identified, are a family of ligand-gated cation channels. Activated microglia express several subtypes of these receptors, which play a key role in establishing and maintaining NeP states (Tsuda et al., ). In NeP after SCI, CCL21 is secreted from damaged neurons and induces P2X4 receptor upregulation in microglia activated by extracellular ATP released from dying cells. Release of pain-related factors from activated microglia with P2X4 receptor expression induces hyperexcitability in the dorsal horn pain network, which may be responsible for NeP (Biber et al., ; Tsuda et al., ).
Macrophages in an injured spinal cord have dynamic phenotypes and functions that can change based on the spinal lesion microenvironment (Stout and Suttles, ; Menzies et al., ). CCL21 and its receptor CCR7 are widely expressed in T cells, dendritic cells, fibroblasts, smooth muscle, and intravascular cells, and CCL21 contributes to inflammation and remodeling of the extracellular matrix (Jiang et al., ). CCR7 has an important role to migrate dendritic cells and/or T cells from brain parenchyma to deep cervical lymph nodes for immune response; CCR7 deficient inflammatory cells are retained in the central nervous system (CNS) and exacerbate CNS autoimmune diseases, including multiple sclerosis (Kivisäkk et al., ; Clarkson et al., ). However, CCR7 is absent in the spinal cord and CCL21 has not been detected in healthy neurons, glial cells and other non-neuronal cells in the CNS (Biber and Boddeke, ). Recent reports indicate that CCR7 is specifically expressed in M1 cells and triggers their migration. Expression of CCR7 is not found in M0 microglia/macrophages before differentiation into M1/M2 and M2 phenotypes (Bellora et al., ). The study using cultured human macrophage demonstrated that CCR7 was expressed exclusively on the cell surface of M1 but in the cytosol of M2 macrophages and that both CCL19 and CCL21 activated MEK1-ERK1/2 and PI3K-AKT pathways in M1 but not in M2 macrophages (Xuan et al., ). In the other study, very few T cells were found in the spinal cord after spinal nerve injury with no differences between wild-type controls and plt mice (Biber et al., ). Early and long-lasting microglia reactivity in the spinal cord after nerve injury was found, but there was no lymphocyte infiltration (Gattlen et al., ). Microglia activation in the spinal cord involves both hypertrophy and hyperplasia. This process progresses through a hypertrophic morphology, with thickened and retracted processes, and an increase in cell number (Smith, ). Numerous factors, such as cytokines, chemokines and ATP, can induce morphological microglia activity. These factors are expressed and probably released in the spinal cord after spinal nerve injury. This may account for the morphological activation of microglia. However, microglia morphology is not a sufficient readout of their function, since morphologically activated microglia do not cause development of tactile allodynia in the absence of CCL21 and subsequent P2X4 receptor upregulation (Biber et al., ). These findings suggest a specialized role of CCL21 in the recruitment of M1 macrophages, but not of other inflammatory cells, in the injured spinal cord. However, it should be noted that experiments isolating cells of a particular day need to be provided to analyze M1 and/or M2 chemotaxis towards CCL21. CCL21 has been shown to be involved in below-level allodynia after SCI by causing M1 chemotaxis (Xuan et al., ). We previously showed that M1-type active microglia and macrophages that are increased after SCI express pain-related substance and contribute to NeP, and that a change in polarity of microglia/macrophages (M1/M2 phenotype) has roles as a trigger for worsening neuroinflammation (Matsuo et al., ; Watanabe et al., ). In the current study, we found a decrease in the number of M1-type cells in plt mice after SCI at the injured site and lumbar enlargement, with no difference in the number of M2-type cells. There was no difference in motor function between wild-type and plt mice, but the decrease of M1-type macrophage infiltration may have beneficial effects on motor function in cases with more severe spinal cord damage. This is a matter for a future study.
With regard to below-level NeP, expression of TNF-α, IL-1β and IL-6 in the lumbar enlargement after midthoracic SCI is related to the severity of below-level allodynia. Early increases in TNF-α may promote induction of below-level allodynia (Detloff et al., ), and M1-type macrophages themselves produce high levels of inflammatory molecules such as TNF-α, IFN-γ and iNOS (Zhou et al., ). TNF-α and IFN-γ produced by M1-type microglia/macrophages spread to the lumbar enlargement following SCI, leading to activation of resident microglia and infiltration of inflammatory cells due to the hyperpermeability of the blood spinal cord barrier (Peng et al., ; Liu et al., ). Moreover, extracellular ATP stimulated TNF-α/IFN-γ induced cell communication in microglia, which might serve to amplify inflammatory signals (Davalos et al., ; Sáez et al., ). CCL21 has been detected at the site of injury and also at remote sites in SCI, including in brain regions and the lumbar enlargement, due to transport of CCL21 via axons. Distal release of CCL21 triggers microglial activation at the distant site, which is essential for development of central NeP and below-level pain (Wu et al., ). In this study, expression levels of TNF-α and IFN-γ in plt mice were decreased at the lumbar enlargement from day 4 to day 14 after SCI, which may be associated with the decreased M1-type microglia/macrophage chemotaxis to the lumbar enlargement.
These results indicate that prevention of neuronal CCL21 expression at an injured site could serve as preventive therapy for NeP at a site distant from the lesion. In a peripheral nerve injury model, intrathecal injection of CCL21 in plt mice induces long-lasting formation of tactile allodynia, and CCL21-blocking antibody treatment reduced development of tactile allodynia in wild-type mice. These results indicate that prevention of CCL21 expression after SCI could reduce NeP. There are various cytokines associated with NeP after SCI, and it is important to avoid adverse effects by blocking cytokines with drugs. CCL21 only induced M1-type microglia/macrophage chemotaxis without affecting M2-type chemotaxis in this study; therefore, prevention of overexpression of CCL21 may not induce adverse effects. A phase I clinical trial of antagonists of the P2X4 receptor (CCL21 up-regulates this receptor in microglia/macrophages) in Japan has confirmed safety and delivery to the CNS. However, possible upstream therapy should be considered, along with P2X4 receptor antagonists. CCL21-blocking antibody treatment may be more effective in an earlier phase after injury because expression of CCL21 reached a peak early after SCI and diminished thereafter. In a peripheral nerve injury model, there was no increase in pain for 10 days after withdrawal of a CCL21-blocking antibody (Biber et al., ). CCL21 is an initiation factor for NeP, and it is important to inhibit the inflammatory reaction in an earlier phase to prevent induction of chronic pain, whereas the influence of a CCL21 blocker in the chronic phase may be limited.
## Conclusion
We show that NeP after SCI was reduced in plt mice with deficient CCL21 expression due to decreased infiltration of M1-type hematogenous macrophages/activated microglia and reduced expression of pro-inflammatory cytokines without affecting M2-type chemotaxis up to 28 days after SCI at the injured site and lumbar enlargement. These results suggest that CCL21 is a key cytokine for NeP after SCI. Our findings are potentially useful for design of new therapies to alleviate NeP after SCI.
## Data Availability Statement
The datasets generated for this study are available on request to the corresponding author.
## Ethics Statement
The animal study was reviewed and approved by the Animal Ethics Review Committee of Fukui University.
## Author Contributions
KH and HN designed various aspects of the study. KH, HN and AM wrote the final manuscript text. HN and SW prepared , . KH and SW prepared , , . KH and TH prepared , . All authors reviewed the manuscript.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Lower-grade glioma (LGG) is a group of tumors arising from the cells of the central nervous system. Although various therapy interventions are used, the prognosis remains different. Novel biomarkers are needed for the prognosis of disease and novel therapeutic strategies in LGG. The procollagen-lysine, 2-oxoglutarate 5-dioxygenase (PLOD) family contains three members and is related to multiple cancers, yet it was not investigated in LGG. Data from the Chinese Glioma Genome Atlas (CGGA) and The Cancer Genome Atlas (TCGA) cohorts were used to analyze the role of PLOD in LGG. As the PLOD family is involved in processes, such as tumor formation and cancer metastasis, we focused on its relationship to the tumor microenvironment (TME) in LGG. A high expression of the PLOD family relates to poor prognosis and high infiltration of immune cells within the TME. The expression level of the PLOD family might become a novel biomarker for prognosis and is a potential target for individual treatment decisions in LGG.
## Introduction
Lower-grade glioma (LGG) is a group of tumors arising from the cells of the central nervous system. In 2016, the World Health Organization (WHO) reclassified LGG using typical histopathological features by crucial markers, such as isocitrate dehydrogenase 1 (IDH1), which disclosed the fact that the biological markers were closely associated with the prognosis of LGG ( ). The current LGG treatment involves different methods, including surgery and radiotherapy as well as chemotherapy ( ; ; ; ; ). While intensive therapeutic interventions are applied to LGG, the outcome remains different. The cellular alterations of the glioma and its surrounding tissue build the tumor microenvironment (TME) ( ). Neoplastic and non-neoplastic cells, such as cancer-associated fibroblasts and immune cells, both of which are involved in tumor formation, progression, and especially the response to treatment, are important modulators of the TME ( ). Most of the non-neoplastic cells are tumor-associated immune cells. However, the brain is not easily infiltrated by the immune system due to the blood-brain barrier (BBB). The adaptation of infiltrating immune cells, such as monocytes and T cells to brain tissue or neoplastic tissue, is essential to further understand tumor development and progression. The immunological response to growth factors and cytokines created by neoplastic cells might determine disease progression ( ). Resident immune cells and infiltrated immune cells might determine the disease-associated microenvironment in the brain. While focusing on the TME, rather than glioma cells themselves, recent literature identified the TME as a superior way to improve the prognosis of LGG ( ; ).
One of the main components of the TME is the extracellular matrix (ECM). Especially, collagen plays a crucial role in the physiological tissue function and tumor formation ( ). Genetic defects affect the biosynthesis, assembly, post-translational modification, and secretion of collagen, and can lead to collagen-related diseases or even cancer ( ; ; ). Procollagen-lysine and 2-oxoglutarate 5-dioxygenases (PLODs) catalyze lysyl hydroxylase (LH), participating in the process of covalent cross-links and collagen glycosylation. Deposition and cross-linking of collagen within the ECM offer a chemical and physical support for tumor formation and proliferation ( ). Three members of the PLOD family (i.e., PLOD1, PLOD2, and PLOD3) leading to cancer progression and metastasis when dysregulated have already been identified ( ; ). It has recently been recognized that PLOD1 is overexpressed in glioma ( ). Our previous work demonstrated that PLOD3 is also highly expressed in LGG with a poor prognosis ( ). Here, we focused on the regulation of the TME and prognostic functions of the PLOD family in LGG.
## Materials and Methods
### Datasets
The RNA-Seq data and clinical information of 431 patients with LGG in the discovery cohort were obtained from the Chinese Glioma Genome Atlas (CGGA ) database ( ). As an independent validation cohort, RNA-Seq data and clinical information of 510 patients with LGG were obtained from The Cancer Genome Atlas (TCGA) database ( ). Samples that did not have complete survival information and paired RNA-Seq data were excluded from this study (174 of 605 samples from the CGGA dataset and 3 of 513 samples from the TCGA database were excluded). All data have been normalized and log ( x + 1) transformed.
### Gene Expression Profiling Interactive Analysis 2
Gene expression profiling interactive analysis 2 (GEPIA2) is a website tool that can be used for analyzing the gene expression distinctions between tumor and normal tissues based on TCGA and Genotype-Tissue Expression (GTEx) databases. In this study, the GEPIA2 tool was used to compare the expression levels of PLOD family members between 518 LGG tumor tissues and 207 normal brain tissues. In addition, the top 100 PLOD-related genes were also obtained in GEPIA2. GEPIA2 tool was also used to obtain the expression of PLOD family members in glioblastoma (GBM) compared with normal brain tissues. The survival analysis module of GEPIA2 was employed to acquire the data of overall survival (OS) and disease-free survival (DFS) in GBM.
### Human Protein Atlas Analysis
Human Protein Atlas (HPA ) is a project that focused on exploring the human protein in cells, tissues, and organs based on the combination of several omics technologies. In this study, the protein expression level of all the three PLOD family members and immunological cells (e.g., CD3 and CD68) was retrieved from the LGG tumor tissues and correlative normal tissues of the HPA dataset.
### cBioPortal Analysis
The cBioPortal is a tool for the analysis of genomic datasets. In this study, all the PLOD family members were typed into the cBioPortal website to obtain the genetic alternation data based on the TCGA dataset.
### Gene Set Enrichment Analysis
Gene set enrichment analysis (GSEA) software (version 4.1.0) was used to perform GSEA by using the HALLMARK gene set. The significance threshold was set to p < 0.05, and false discovery rate (FDR) was set to <0.25.
### Gene Enrichment Analysis
In this study, the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database was used to obtain the top 50 PLOD-interacted proteins and construct the protein-protein interaction (PPI) network. Combining the top 50 PLOD-interacted genes and the top 100 PLOD-related genes, the “clusterProfiler” R package was used to perform the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis and the Gene Ontology (GO) enrichment analysis.
### Immune Infiltration Analysis
The R package “GSVA” was used to perform a single-sample GSEA (ssGSEA) to quantify the relative abundance of 28 previously defined immune cells ( ). The R package “ESTIMATE” was used to calculate three scores, namely, ImmuneScore (positively correlated with the level of immune cell infiltration in the tumor), StromalScore (positively correlated with the level of stroma cell in the tumor), and ESTIMATEScore (negatively correlated with tumor purity) ( ). The TIMER tool was used to obtain six immune cells infiltration in GBM.
### Patient-Derived Glioma Tissue
Patient-derived glioma tissue was obtained from the University Hospital Leipzig. Glioma tissue from nine different patients was stained with 3,3′-diaminobenzidine (DAB; Sigma Aldrich, St. Louis, MO, United States) tablets as described previously ( ). The following primary antibodies were used in this study: CD3 antibody (MCA1477, lot no. 149500B, 1:200 dilution) and CD68 antibody (M0876, lot no. 20043031, 1:500 dilution).
### Statistical Analysis
Based on multivariate analysis, the risk score model, namely, PLODscore was calculated. This type of model was already published in our previous work and in high-grade glioma research ( ; ; ). The Spearman’s correlation analysis was used to calculate the correlations between PLODscore and the level of 28 immune cells infiltration, ImmuneScore, StromalScore, and ESTIMATEScore. The Kaplan–Meier (KM) curves were used to estimate the difference in survival between two groups, and significance was calculated using a log-rank test. The receiver operating characteristic (ROC) curve and the Harrell’s concordance index were used to assess the predictive value of the risk model. Differences between the two groups were calculated using the unpaired Student’s t -test or Wilcoxon rank-sum test. Comparisons of more than two groups were calculated using the Kruskal-Wallis test. Multivariate Cox regression was performed to evaluate the independent risk factors on OS. All statistical calculations were performed using R software (version 4.0.3; R Foundation for Statistical Computing, Vienna, Austria), and p < 0.05 was considered statistically significant.
## Results
### The Alternation and Expression of Procollagen-Lysine, 2-Oxoglutarate 5-Dioxygenase Family Members at Transcription and Translation Level
In total, 518 LGG tumor tissues from the TCGA database and 207 correlative normal brain tissues from the GTEx database were included based on GEPIA2. PLOD1, PLOD2, and PLOD3 were highly expressed in LGG tumor tissues compared to normal tissues ( , p < 0.01). The determination of the difference in protein levels was also achieved by using the HPA database. The expression of all PLOD family members was higher in LGG tumor tissues than that of the normal tissues by immunohistochemical staining ( ). The cBioPortal was used to obtain the alternations of PLOD1, PLOD2, and PLOD3, including the alternation types and the correlative rate. The alternation rate for PLOD1, PLOD2, and PLOD3 accounted for 0.6, 0.5, and 1.7%, respectively. The amplification occupied the most part of PLOD3 alternation ( ). In addition, the PLOD family members were also highly expressed in GBM ( ).
The alternation of procollagen-lysine, 2-oxoglutarate 5-dioxygenase (PLOD) family members and the expression of PLODs at transcription and translation level in low-grade glioma (LGG). (A) Gene expression profiling interactive analysis 2 was used to harvest the boxplot results of the expressions of PLODs family members between cancer tissues and correlative normal tissue at transcriptional level. (B) Based on the Human Protein Atlas database, the expression of PLODs was obtained at protein level according to the immunohistochemistry staining. (C) The genetic alternations of PLODs were achieved by cBioPortal. ** p < 0.01.
### The Expression of Procollagen-Lysine, 2-Oxoglutarate 5-Dioxygenase Family Members in Correlation to Different Clinical Parameters
Data from CGGA and TCGA datasets were analyzed. Diverse clinical parameters were considered in this analysis, including LGG grade, IDH1 status, histology of LGG, age, and gender. shows that the expression level of PLOD1 and PLOD3 is higher in grade III than grade II samples but does not alter PLOD2 levels. In addition, the expression level of PLOD1 and PLOD2 was higher in IDH1 wild-type samples than IDH1 mutant samples in both datasets, while PLOD3 was only significantly enhanced in the IDH1 wild-type status group of the TCGA dataset ( ). According to the classification of TCGA and CGGA databases, three main pathological subtypes were included in this study. Among the different types of histology, the expression of all PLOD family members was higher in astrocytoma than in oligoastrocytoma and oligodendroglioma of CGGA and TCGA datasets ( ). indicates that there are no significant differences in PLOD1-3 expressions in different age and sex groups. Furthermore, the expression between the PLOD family was positively correlated with LGG ( ).
Procollagen-lysine, 2-oxoglutarate 5-dioxygenases (PLOD) expressions between diverse clinical stratifications and the relationship of expression levels among PLODs family members in lower-grade glioma (LGG) based on the datasets of Chinese Glioma Genome Atlas (CGGA) and The Cancer Genome Atlas (TCGA). (A) PLODs expression between different grades in LGG. (B) PLODs expression based on the status of is citrate dehydrogenase 1 (IDH1). (C) PLODs expression based on the histology of LGG. (D) The relationship of expression levels among the PLODs family members (e.g., PLOD1, PLOD2, and PLOD3) based on CGGA and TCGA databases. ** p < 0.01, *** p < 0.001, **** p < 0.0001. ns, not significant.
### Kyoto Encyclopedia of Genes and Genomes Pathway and Gene Ontology Enrichment Analyses for Procollagen-Lysine, 2-Oxoglutarate 5-Dioxygenase Family Members
The top 50 PLOD-interacting genes were identified, and a PPI network was constructed based on behalf of the STRING tool, as shown in . In , the top 100 PLOD-related genes are displayed. By combining PLOD-interacting genes with PLOD-related genes, the KEGG pathway and GO enrichment analyses were performed. According to the KEGG pathway analysis, these genes are involved not only in carcinogenesis-related pathways, such as focal adhesion and ECM-receptor interaction, but also in immune-related pathways, such as leukocyte migration ( ). The GO enrichment analysis indicates that PLOD-interaction and related genes are associated with ECM organization, cell adhesion, biological adhesion, and endodermal cell differentiation in the biological process (BP) section ( ). Further analysis revealed a significant relationship of the PLOD family with the endomembrane system and the ECM components of the cellular component (CC) section. Analysis of the molecular function (MF) section revealed the components of the ECM to be of relevance, as shown in .
The protein-protein interaction (PPI) networks, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway, and Gene Ontology (GO) enrichment analysis for PLODs family members in lower-grade glioma (LGG). (A) Search Tool for the Retrieval of Interacting Genes/Proteins tool was used to obtain the PPI networks for PLODs. (B) KEGG pathway of PLODs. (C) GO enrichment of biological process for PLODs. (D) GO enrichment of cellular component for PLODs. (E) GO enrichment of molecular function for PLODs.
### Overall Survival Analysis Based on the Expression of Procollagen-Lysine, 2-Oxoglutarate 5-Dioxygenase Family Members
The data from CGGA and TCGA datasets were analyzed to explore the correlation of the expression of PLOD family members with the OS in LGG. The expression median separates the high and low expression of the PLOD family members, and the KM curves visualize the survival data, as shown in . A high expression of all PLOD family members was correlated with worse OS in both datasets compared to a low expression. In addition, the overexpression of PLOD1 showed a correlation to poor prognosis in GBM, while the other PLOD family members were not linked to prognosis in GBM ( ).
The survival analysis for the expression of procollagen-lysine, 2-oxoglutarate 5-dioxygenases (PLOD) members in lower-grade glioma (LGG). A cutoff value of 50–50% was applied and Kaplan–Meier (KM) curve was used to analyze and visualize the survival analysis based on Chinese Glioma Genome Atlas (CGGA) and The Cancer Genome Atlas (TCGA) datasets. (A) Correlation between survival period and PLOD1 expression level in LGG. (B) Correlation between survival period and PLOD2 expression level in LGG. (C) Correlation between survival period and PLOD3 expression level in LGG.
### Establishment of PLODscore for Lower-Grade Glioma Prognostic Prediction
The gathered results suggest that the investigated PLOD members are related to tumorigenesis and development of LGG. To further quantify the expression pattern of PLOD family members in individual patients and to construct a PLOD-related gene signature, the multivariate Cox regression model was used to establish a scoring system based on the CGGA dataset, termed PLODscore. Specifically, the PLODscore was used to calculate the following formula:
After calculating the PLODscore for each sample, all samples in the CGGA cohort were divided into high and low PLODscore groups using the median as the threshold. As shown in , the high PLODscore group had a lower survival rate and higher PLOD expressions. The KM curve indicated that the OS of patients with a high PLODscore group was significantly reduced compared to the low PLODscore group ( ). Similar outcome could also be obtained if the PLODscore was quartered ( ). The area under the ROC curve (AUC) for 1, 3, and 5 years was 0.68, 0.74, and 0.76, respectively ( ). In addition, multivariate Cox regression analysis suggested that the PLODscore was an independent risk factor in LGG ( ). All the above results were independently validated in the TCGA cohort ( , ). Nomogram was established based on the independent risk factors in the CGGA cohort, and the decision curve analysis (DCA) indicates that the nomogram has a good predictive performance ( ) and the Harrell’s concordance index ( ).
The association of PLODscore with the prognosis in lower-grade glioma (LGG) based on Chinese Glioma Genome Atlas (CGGA) dataset. (A) PLODscore, survival status, and heatmap of mRNA expression of the PLODs members. (B) Kaplan–Meier (KM) curve. (C) Time-dependent survival receiver operating characteristic analysis. (D) Multivariate statistics for procollagen-lysine, 2-oxoglutarate 5-dioxygenase (PLOD) score in LGG. ** p < 0.01, *** p < 0.001.
### Cellular and Molecular Characteristics of PLODscore
Tumor response to immunotherapy is largely dependent on the TME ( ). Therefore, we investigated the potential role of PLOD family members within the TME of LGG. The staining from the HPA dataset and the patient-derived glioma tissues showed that there is a significant number of CD3 positive and CD68 positive cells in all stages of glioma tissue ( and ). However, the expression varies greatly within the patient samples. Using ssGSEA, we inferred the enrichment score of 28 previously reported immune cells of the CGGA cohort. The immune cell components between the high and low PLODscore groups were significantly different. Except for CD4 memory effector cells and CD56 natural killer cells, patients with a low PLODscore showed lower infiltration levels of most immune cells ( ). Similar results were found in the TCGA dataset ( ).
Cellular characteristics of PLODscore based on Chinese Glioma Genome Atlas (CGGA) cohort. (A) Multivariate statistics for PLODscore, including vital status, IDH1 status, gender, grade of lower-grade glioma (LGG), histology of LGG, and the heatmap of 28 previously reported immune cell signature scores. (B) The relationship between PLODscore and enrichment score of immunostimulatory cells. (C) The relationship between PLODscore and enrichment score of immunosuppressive cells. ** p < 0.01, *** p < 0.001, **** p < 0.0001. ns, not significant.
Furthermore, the PLODscore was positively correlated with the infiltration levels of multiple immunostimulatory cells (e.g., natural killer T cells) and immunosuppressive cells (e.g., regulatory T cells) by correlation analysis ( ). Patients with a high PLODscore have a “hotter” but more immunosuppressed TME, which implies that the PLODscore could quantify the TME pattern of individual patients. In addition, the expression of the PLOD family members could be related to infiltrating immune cells, such as B cells, T cells, and macrophages ( ). To further test this inference, we calculated the estimated scores for patients with LGG based on the study by . It was found that the PLODscore was positively related to the ImmuneScore, StromalScore, and ESTIMATEScore ( ), which suggests that the high PLODscore group has higher immune and stromal cell infiltration. In addition, we analyzed the expression of selected cytokine and chemokine mRNAs. We considered CXCL10, CXCL9, GZMA, GZMB, PRF1, CD8A, TBX2, and TNF as immune-activating transcripts; IDO1, CD274, HAVCR2, PDCD1, CTLA4, LAG3, and PDCD1LG2 as immune checkpoint transcripts; and VIM, ACTA2, COL4A1, TGFBR2, ZEB1, CLDN3, SMAD9, TWIST1, and TGRB1 as transforming growth factor (TGF)β/epithelial-mesenchymal transition (EMT) pathway transcripts ( ; ). The high PLODscore group had enhanced expression levels of immune activation-relevant genes, immune-checkpoint-relevant genes, and TGFβ/EMT pathway-relevant genes ( ). These results demonstrated that patients with high PLODscore are characterized by immune activation and stromal activation with significant immunosuppression. Similar results were also obtained in the TCGA dataset ( ).
The immune landscape of procollagen-lysine, 2-oxoglutarate 5-dioxygenases (PLOD) in lower-grade glioma (LGG). The immune landscape of correlation between PLODscore, PLODs members, and immune cells according to the (A) Chinese Glioma Genome Atlas (CGGA) cohort and (B) The Cancer Genome Atlas (TCGA) dataset. * p < 0.05, ** p < 0.01, and *** p < 0.001.
The overview of PLODscore with immune infiltration according to Chinese Glioma Genome Atlas (CGGA) database. (A) The association of PLODscore with ImmuneScore, StromalScore, and ESTIMATEScore. (B) The relationship of PLODscore to immune activation-relevant genes, (C) immune-checkpoint-relevant genes, and (D) transforming growth factor-β/epithelial-mesenchymal transition pathway-relevant genes. ** p < 0.01, **** p < 0.0001. ns, not significant.
### Gene Set Enrichment Analysis of PLODscore
The GSEA was used to explore the potential BPs and signal transduction pathways related to PLODscore. A high PLODscore is related to carcinogenesis-related pathways, including angiogenesis and P53 pathway, as well as immune activation-related pathways, including inflammatory response and interferon-gamma (IFNγ) response ( ). This is consistent with our previous results that patients with high PLODscore have a poor prognosis together with a “hot” but immunosuppressed TME.
Gene set enrichment analysis. The relationship of PLODscore with biological processes and signal transduction pathways using the HALLMARK gene set.
## Discussion
Our previous work indicated that PLOD3 overexpression is linked to poor prognosis in LGG and is associated with immune cell infiltration of the TME ( ). In this study, a comprehensive analysis of the correlation between specific PLOD family members (e.g., PLOD1, PLOD2, and PLOD3) and LGG has been performed using the CGGA and TCGA datasets. These PLOD family members are suggested to be potential biomarkers for gastric cancer and hepatocellular carcinoma ( ; ). A further study has shown that PLOD1 is a predictive biomarker for LGG ( ). Hence, we found that not only PLOD1 but also PLOD2 and PLOD3 were abundantly expressed in tumor tissues.
The PLOD catalyzes the LH, which is involved in the biosynthesis of collagen ( ). Consequently, mutations of the PLOD family lead to connective tissue-related diseases, such as the Ehlers-Danlos and Bruck syndromes ( ). During the period of connective tissue repair, the overexpression of the PLOD family might induce EMT and, consequently, contribute to tumor promotion. In EMT, the epithelial cells become unstable by losing their polarity and their adherence ability, and enhancing cellular migration and cellular invasion ( ; ). This process takes place in important processes, such as wound healing, and can also lead to tumor formation ( ). Supporting this hypothesis, the KEGG pathway and GO enrichment analysis correlate all three PLOD genes to alterations in focal and cellular adhesion and ECM-receptor interaction. The KEGG pathway analysis additionally showed that all PLOD genes are associated with leukocyte transendothelial migration, which might indicate the direct effect of PLOD in the TME. The IDH1 is a known biomarker for patients with glioma. Its mutation is regarded as beneficial, as epidermal growth factor receptor (EGFR) and loss of chromosome 10 are hardly found ( ). Therefore, we tested the dependency of PLOD expression and the mutation status of IDH1. Enhanced PLOD1 and PLOD2 expression was found in the group, displaying wild-type IDH1. Although no significance was reached when compared to age and gender, the results fortified the suggestion that high PLOD expression is negatively linked to patient survival, as the expression of PLOD1 and PLOD3 is higher in grade III than grade II LGG ( ). Still, PLOD3 could not be directly related to the IDH1 mutational status.
In this study, we created a PLODscore that was dependent on the expression of all three PLOD family members based on multivariate analysis. The defined PLODscore is negatively associated with the patient OS and serves as an independent prognostic factor for OS as well as tumor grade in LGG. Yet, it has to be shown whether the PLODscore has a prognostic significance for treatment response in all different glioma grades. In GBM, the PLODscore was, however, not applicable. Nevertheless, we could stain peripheral immune cells in GBM and LGG tissue, demonstrating relevant infiltration in individual tissues. Investigating the TME, a positive correlation with the PLODscore was observed, indicating a high potential relevance of the PLODscore in immunomodulatory therapies of LGG.
The TME consists of endothelial cells, fibroblasts, immune cells, and tumor cells ( ). However, as a tumor of the central nervous system, the TME of LGG has various distinguished features compared to peripheral tumor entities. Thus, almost all mentioned cells are found in glioma, but further unique CCs, such as microglia, astrocytes, and neurons, are also present ( ). Fibroblasts are not present in the central nervous system; however, recent studies suggest that specific pericytes of the brain act as cancer-associated fibrocyte-like stromal cell population ( ). Here, we used patient-derived LGG and GBM samples to determine the amount of immune cell infiltration to affirm clinical relevance. Interestingly, CD3 and CD68 positive cells are observed to a higher extent only in patients with GBM and LGG, supporting the relevance of the TME together with personalized approaches in glioma ( ). Brain macrophages are the most abundant immune cell population in the TME of brain tumors. Even though its ontogenesis reveals a specialized, long-lasting macrophage population, peripheral macrophage invasion is likely to happen ( ; ; ). The BBB is a cellular fence of endothelial cells and pericytes that needs to be conquered by immune cells or molecules to invade the brain parenchyma ( ). The functional unit of the BBB is the so-called neurovascular unit (NVU) and protects the brain from immune invasion ( ). However, the NVU loses its function in some diseases, such as Alzheimer’s disease ( ). Mice studies demonstrated the augmented effects of immune checkpoint inhibition in intracranial tumors when extracranial tumors were present ( ; ). Further studies based on animal models investigating multiple sclerosis demonstrated that immune cells, such as B cells, might activate CNS inflammation ( ; ; ). However, it was shown that monocytes influence B cells to suppress CD8 T cell activation and acquisition of an effector phenotype in GBM studies ( ). Currently, antibodies against the intercellular adhesion molecule-1 were shown to reduce B cell invasion, demonstrating that B cells have the potential to be addressed in tumor progression ( ; ). Hence, the integrity of the BBB is compromised by tumor formation or the progression of the parenchymal function needs to be preserved for the sake of neuronal survival ( ). Consequently, the TME of brain tumors is regarded to be immune-specialized.
Here, we used ssGSEA to investigate the PLODscore with immune cells of the TME in LGG. A positive correlation of the PLODscore with cells of the monocyte lineage and other cells determined as immunosuppressive was observed. Nevertheless, lymphoid cells, including CD4 positive and CD8 positive T cells, B cells, and NK cells, are also related to the PLODscore. A murine glioma study indicates that CD8 T cells alter depending on the microenvironment and might be differentially regulated in brain tissue ( ). Our study demonstrated that the high PLODscore can be related to both an elevated T cell population and a high number of immunosuppressive cells, linking a high PLODscore to a hot but suppressive immune environment.
To further investigate this correlation, we used the method ESTIMATE, allowing for the prediction of stromal and immune fractions in tumor tissues ( ; ). The high PLODscore could be related to immune-activating genes (e.g., CXCL10 , CXCL9 , TBX2 , and TNF ) that are previously mentioned ( ; ).
The correlation of the high PLODscore to immune checkpoint genes demonstrates the potential of immune checkpoint inhibition therapy in LGG. However, mouse models and clinical data demonstrate controversial results that might be resolved with adequate delivery systems ( ; ). Further targetable genes of the TGFβ/EMT pathway are associated with the high PLODscore ( ; ; ; ; ).
Although the mechanism of immune cell invasion is not yet understood in detail, evidence of patient-derived tissue samples provides important insight into the clinical relevance of immune cells spread throughout tumor tissues ( ; ; ; ; ). This analysis was based on two independent cohorts due to database limitations. Thus, the influence of brain-specific cells and the peripheral immune system to brain tumor formation and progression urges for further research ( ; ).
Taken together, this study shows the potential role of PLOD genes in LGG prognosis and its involvement in the TME of LGG. It is suggested that the regulation of PLOD genes and its products could be potential new targets for therapeutic interventions as well as for individual treatment decision in LGG.
## Data Availability Statement
The original contributions presented in the study are included in the article/ , further inquiries can be directed to the corresponding author.
## Ethics Statement
The studies involving human participants were reviewed and approved by the Ethics Committee of the Medical Faculty, University of Leipzig (#144/08-ek; 2019-07-04). The patients/participants provided their written informed consent to participate in this study.
## Author Contributions
SG, CW, NS, and SK: conception and design, data analysis and interpretation, and manuscript writing and revisions. SG, FK, JM, and CW: collection and assembly of data. All authors approved the final manuscript and accounted for all aspects of work, read, and agreed to the published version of the manuscript.
## Conflict of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s Note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
|
The journal retracts the 13 January 2020 article cited above.
Following publication, concerns were raised regarding the integrity of the data in the published figures. The authors failed to provide a satisfactory explanation during the investigation, which was conducted in accordance with Frontiers' policies.
This retraction was approved by the Chief Editors of Frontiers in Cellular Neuroscience and the Chief Executive Editor of Frontiers. The authors agree to this retraction.
|
Parkinson’s disease (PD) is a multifactorial neurodegenerative condition with symptoms such as resting tremor, rigidity, bradykinesia (slowness of moment), and postural instability. Neuroinflammation plays a significant part in the onset and progression of neurodegeneration in a wide range of disorders, including PD. The loss of dopaminergic neurons in the substantia nigra (SN) is thought to be the primary cause of PD disease progression. However, other neurotransmitter systems like serotoninergic, glutamatergic, noradrenergic, adrenergic, cholinergic, tryptaminergic, and peptidergic appear to be affected as well. Epigenetic regulation of gene expression is emerging as an influencing factor in the pathophysiology of PD. In recent years, epigenetic regulation by microRNAs (miRNAs) has been discovered to play an important function in the disease progression of PD. This review explores the role of miRNAs and their signaling pathways in regulating gene expression from development through neurodegeneration and how these mechanisms are linked to the pathophysiology of PD, emphasizing potential therapeutic interventions.
## Introduction
Parkinson’s disease (PD) is the second most common neurodegenerative disorder, affecting 1–2 people per 1,000 at any moment. The prevalence of PD rises with age, and it affects 1% of the population over the age of 60 ( ; ). The origin of PD is still unknown, and it’s postulated to be caused by the lack of dopamine due to the loss or degeneration of dopaminergic neurons in the substantia nigra par compacta (SNpc), a part of the brain that controls movement. Since dopamine acts as an inhibitor in the basal ganglia, it limits extrapyramidal movement ( ). The pathophysiology of PD involves the accumulation of neuronal Lewy Bodies, and the risk factors include aging, family history, pesticide exposure, and environmental pollutants. Bradykinesia (slowness of movement), resting tremor (involuntary twitching or shaking of muscles), rigidity (involuntary stiffening of muscles), postural instability (inability to maintain equilibrium), hyposmia (decreased sense of smell) and rapid eye moment are all effects of PD ( ).
Since the symptoms of PD involve motor functions, most of the drugs used in the treatment of PD are dopamine based. Commercially available dopamine agonists exhibit significant improvement in PD, yet they are also susceptible to many adverse effects. This makes it challenging to use in many patients. The pharmacological drugs lack specificity, leading to severe side effects. All these limitations are due to inadequate knowledge of the molecular mechanisms, signaling pathways and epigenetic regulations involved in the disease progression of PD. Diagnosis of PD is possible only after the onset of symptoms, and there is a lack of validated biomarkers for early diagnosis of PD ( ; ). Research on biomarkers could help in early diagnosis as well as could pave for early treatment of PD.
Interestingly, in recent years, many studies have focused on the molecular mechanisms and epigenetic regulation of disease progression in PD. Yet there are no definitive results due to the complexity of the disease. There are specific molecules that are involved in the regulation of various signaling pathways and genes. One such important regulator is microRNA (miRNAs) which are small non-coding RNAs found to be involved in the expression and suppression of various genes.
This review paper focuses on miRNAs involved in the molecular mechanisms, signaling pathways and epigenetic regulations in the disease progression of PD.
## MicroRNAs and their biogenesis
miRNAs are short non-coding RNA molecules with approximately 17–22 nucleotides in length that controls gene expression post-transcriptionally through either translational repression or mRNA degradation ( ). miRNAs have a significant role in the regulation of cell proliferation, differentiation, and apoptosis in various cellular processes. The biogenesis of miRNAs begins with the processing of RNA polymerase II/III. miRNA biogenesis is divided into two pathways, namely canonical and non-canonical pathways ( ). The canonical biogenesis pathway is the most common approach for miRNA processing. In the canonical pathway, at first, the primary miRNA (pri-miRNA) is transcribed from their gene and processed into precursor miRNA (pre-miRNA) by the microprocessor complex, which consists of an RNA binding protein called DiGeorge Syndrome Critical Region 8 (DGCR8) and a ribonuclease III enzyme called Drosha ( ) which takes place in the nucleus. Once pre-miRNAs are synthesized, an exportin 5 (XPO5)/RanGTP complex transports them to the cytoplasm, where the second processing step occurs, where they are processed by the RNase III endonuclease called Dicer ( ). The terminal loop is removed during this step, which results in the formation of the mature miRNA duplex. The mature miRNA form is named according to the miRNA strand’s directionality ( ). The 5p strand emerges from the pre-miRNA hairpin’s 5’ end, while the 3p strand emerges from the 3’ end. The argonaute (AGO) family proteins play a crucial function in RNA silencing. With the help of ATP- dependent chaperone proteins, the miRNA duplex is loaded into the AGO protein ( ). The passenger strand of the miRNA duplex is released once the AGO is returned to its regular conformation, resulting in single-stranded mature miRNA ( ). After loading, AGO stimulates the formation of a ribonucleoprotein complex known as RNA-induced silencing complex (RISC), which facilitates the recognition of the targeted mRNA ( ). represents the miRNA biogenesis in the canonical pathway.
Biogenesis of miRNA. The miRNA gene in the nucleus is transcribed to primary miRNA which is then converted to precursor miRNA by the enzyme Drosha. The precursor miRNA enters the cytoplasm via Exportin and forms the miRNA duplex with the help of Dicer. The miRNA duplex forms the mature miRNA with the help of RNA-Induced Silencing Complex (RISC).
Apart from the above-mentioned canonical miRNA biogenesis processes, a number of additional methods can produce miRNAs: one such pathway is the non-canonical miRNA biogenesis pathway. Non-canonical miRNA biogenesis can be further divided into two categories, namely Drosha/DGCR8-independent and Dicer-independent pathways ( ). The Drosha/DGCR8-independent process produces pre-miRNAs that look similar to that of Dicer substrates. Mirtrons, formed from the introns of mRNA during splicing, are an example of pre-miRNAs ( ). Spliceosomes and debranching enzymes process mirtrons to produce acceptable dicer substrates, bypassing the microprocessor stage ( ). On the other hand, Drosha processes Dicer-independent miRNAs from endogenous short hairpin RNA (shRNA) transcripts ( ). However, recent research on simtrons and other Dicer-independent miRNAs has uncovered new, non-canonical pathways that are not only unexplored but also crucial in understanding the miRNA synthesis without Dicer. Overall, the findings suggest that Dicer, which is known to be important in both canonical and non-canonical pathways, is not required for some miRNAs that have yet to be investigated.
At times, these miRNAs are encapsulated within the exosomes, which are extracellular vesicles and are termed exosomal miRNAs. Since the exosomes are distributed in the body fluids like blood, saliva, cerebrospinal fluid (CSF) and urine, the exosomal miRNAs are identified to be important liquid biopsy biomarkers. Due to the blood-brain barrier system, it is difficult to identify the miRNAs from the central nervous system (CNF) in the blood. Hence, differential expression of miRNAs could be found in the CSF of neurodegenerative patients. Since the exosomes act as cargo, the exosomal miRNAs travel from the CNS to the peripheral nervous system (PNS) and affect the motor functions in the patients by targeting the important genes in PD ( ). Moreover, since the exosomes can incorporate into the cells through receptor-ligand interaction, it makes it easier for the miRNAs to spread and reach the target genes ( ).
## MicroRNAs involved in signaling mechanisms and epigenetic regulation of Parkinson’s disease
Research on the role of miRNAs in the pathophysiology of PD is vast, and various miRNAs are involved in the regulation of signaling pathways in the disease progression of PD. For instance, miRNAs like miR-133b and miR- 384-5p are found to be regulating apoptosis in PD ( ; ). Likewise, the following miRNAs mediate different signaling pathways and epigenetic regulation in PD. represents the miRNA involved in the signaling pathways of Parkinson’s disease.
miRNAs involved in the signaling pathways of Parkinson’s disease.
### MicroRNAs and inflammation mediating pathways in Parkinson’s disease
In a study by , it was identified that microglial activation is regulated by miR-124’s aberrant expression. Their results suggested miR-124 as a potential therapeutic target for controlling the inflammatory response in PD. It was concluded that by targeting p62, p38, and autophagy, miR-124 might be able to reduce neuroinflammation during the development of PD ( ). Moreover, high amounts of let-7b-5p were discovered in the brain tissues of PD animals and MPP + -treated SH-SY5Y cells. Let-7b-5p knockdown prevented SH-SY5Y cell death. The study found that let-7b-5p contributes to cell apoptosis in PD via targeting HMGA2, thus providing a therapeutic target for PD ( ). In one of the studies, miR-29b2/c was found to be involved in aging and disease progression in PD. Using miR-29b2/c gene knockout mice, the role of miR-29b2/c in aging and PD was investigated. Lack of miR-29b2/c increased AMPK activation while suppressing NF-κB/p65 signaling in glial cells. Thus, the study concluded the detrimental effect miR-29b2/c had on PD ( ).
### MicroRNAs and mitophagy mediating pathways in Parkinson’s disease
In PD, the effects of mitophagy mediated by miR-103a-3p/Parkin/Ambra1 signaling were investigated. Parkin expression was reduced by miR-103a-3p, and miR-103a-3p inhibition had neuroprotective benefits in Parkinson’s disease (PD). This suggests that miR-103a-3p is involved in controlling mitophagy via the Parkin/Ambra1 pathway ( ). Interestingly, a study compared two miRNAs, miR-17 and miR-19a in the regulation of DNA methyltransferases (DNMT) in PD. It was identified that miR-17 regulated DNMT1 and was responsible for abnormal DNA methylation occurring in PD. Thus miR-17 was identified to be a therapeutic modulator of DNA methylation in PD ( ). Interestingly, miR-96 and pramipexole (PPX) which plays a protective role in PD was analyzed in PD cell line (1-methyl 4-phenylpyridinium (MPP ) induced cells) and 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) induced mice. It was identified that to prevent PTEN-induced putative kinase 1 (PINK1)/Parkin signals from regulating mitophagy, miR-96 specifically targeted BNIP3. By reducing BNIP3, miR-96 overexpression facilitated MPP + -induced neuronal damage. By controlling miR-96/BNIP3-mediated mitophagy, PPX reduced MPTP-induced neuronal damage in mice. Thus miR-96 was found to be playing an important role in PD ( ).
### MicroRNAs and ferroptosis mediating pathways in Parkinson’s disease
Fascinatingly, the process of iron (Fe )-dependent programmed cell death is known as ferroptosis. Ferroptosis is strongly related to the pathogenic alterations seen in PD, including nigral iron elevation, glutathione depletion, and enhanced reactive oxygen species (ROS) generation. miR-335 was found to enhance ferroptosis by degrading FTH1. The results of the study showed that miR-335 promotes ferroptosis in both in vitro and in vivo models of PD by targeting FTH1 ( ). Moreover, miR-221 controls PC12 cell viability and apoptosis by targeting PTEN, providing protection against PD. Consequently, miR-221 may be a viable therapeutic target for the treatment of PD ( ).
### MicroRNAs and autophagy mediating pathways in Parkinson’s disease
Autophagy in dopamine neurons is thought to be linked to Parkinson’s disease, but the precise mechanism is uncertain. In one of the studies, the findings showed that autophagy was induced in the SNpc dopaminergic neurons of PD animals due to the downregulation of miR-29c-3p and upregulation of 10–11 translocation 2 (TET2) expression. In PD animals, up-regulation of miR-29c-3p reduced TET2 expression and SNpc autophagy, which includes dopaminergic neurons. Overall, over-expression of miR-29c-3p in PD models reduces autophagy, which may be caused by TET2. The findings exhibited the role of miR-29c-3p in controlling autophagy to accelerate the progression of PD ( ). One of the studies sought to understand the targeting of X-box binding protein 1 (XBP1) by miR-326 in the biological functions of PD. Mice treated with a miR-326 mimic and siRNA-XBP1 displayed improved traction test results, autophagy activation, LC3-II, c-Jun, and p-Syn expression, but decreased climbing time and iNOS, α-Syn, and p-c-Jun expression. According to the study’s findings, overexpression of miR-326, which specifically targets XBP1, reduces iNOS production and encourages autophagy of dopaminergic neurons through JNK signaling ( ).
One study examined how inhibiting miR-15b-5p will suppress cell death in Parkinson’s disease (PD) by focusing on the Akt3-mediated GSK-3/catenin signaling pathway ( ). Another study found that miR-216a controlled the development of PD through controlling Bax, suggesting that miR-216a could be a possible target for PD ( ). It was discovered that increased miR-421 encourages cell death by inhibiting MEF2D expression. In cellular and animal models of PD, miR-421 inhibition or MEF2D restoration protected neurons from neurotoxicity ( ). According to a study, mice treated with MPTP had their miR-425 levels increased, which prevented the defective degradation of dopaminergic neurons and improved behavioral impairments. These results point to miR-425 brain administration as a possible therapeutic strategy for the treatment of Parkinson’s disease ( ). Interestingly, another study demonstrated that blocking miR-497-5p caused PD mice’s bradykinesia to improve, cell apoptosis to be reduced, and autophagy to be activated by FGF2. As a result of inhibiting cell apoptosis and boosting autophagy in a FGF2 dependent manner, miR-497-5p silencing alleviates PD and offers a fresh target for treating PD ( ).
In addition, the role of TREM2 and ULK1 in PD were analyzed. It was discovered that up-regulating TREM2 greatly boosted the expression of p-ULK1. Increased p-ULK1 expression had the ability to suppress inflammatory factor expression while promoting autophagy. Through TREM2, miR-3473b may indirectly control ULK1. In order to control the involvement of autophagy in the pathogenesis of inflammation in PD, miR-3473b may affect TREM2/ULK1 expression, which suggests that it may be a possible therapeutic target to control the inflammatory response in PD ( ).
## MicroRNAs as biomarkers and therapeutic targets in Parkinson’s disease
miRNAs are being studied for their potential role as biomarkers and therapeutic targets in several diseases including neurodegenerative disorders. Once the molecular mechanisms and signaling pathways are elucidated, their role as biomarkers and therapeutic targets could be easily explained.
One of the crucial components for the preservation of dopaminergic function and one that is particularly sensitive in PD is nuclear receptor related 1 protein (Nurr1), which is targeted by miR-132 and their expression is adversely linked (PD). Thus miR-132 could serve as a biomarker as well as potential therapeutic target ( ). In one of the studies, the CSF of PD patients were collected and differential expression of miR-626 was analyzed. The results revealed that miR-626 was significantly downregulated in PD patients when compared to normal controls ( ). Engrossingly, the serum miR-150 levels in PD patients was found to be differentially expressed. The miR-150 levels in the serum of PD patients were less compared to healthy normal. It was also identified that miR-150 induced inflammation by targeting the AKT3 gene ( ).
Another study by showed that miR-29 levels in the serum are related to cognitive dysfunction in PD. It has been demonstrated that miR-29b levels are connected with several subtypes of PD cognition and may successfully distinguish PD associated dementia (PDD) from non-PDD ( ). A study by explored the role of miR-185 in PD. It was identified that miR-185 levels were decreased in PD-affected rats. In the substantia nigra of PD rats, miR-185 restoration improved pathological damage, oxidative stress, and inhibited neuronal death. To stimulate the PI3K/AKT signaling pathway, miR-185 targeted IGF1. IGF1 upregulation reduced the effects of restored miR-185 on PD animals. Thus, the study concluded miR-185 as a potential therapeutic target for PD ( ).
Through the control of LIM and SH3 domain protein 1 (LASP1), one study investigated the protective effect of miR-218-5p on dopaminergic neuron damage in SNpc of rats with PD. It was confirmed that miR-218-5p targets LASP1. In the brain SNpc of PD rats, increased miR-218-5p or decreased LASP1 reduced dopaminergic neurons’ oxidative stress and death. Furthermore, elevated miR-218-5p suppressed LASP1 expression in the SNpc of the brain of PD rats. According to the study, up-regulated miR-218-5p may help PD rats’ damaged dopaminergic neurons ( ). In PD animal and neuronal models, one study found that miR-29c-3p (miR-29c) had anti-inflammatory characteristics. The release of pro-inflammatory cytokines as well as the activation of the NF-êB and TXNIP/NLRP3 inflammasome were inhibited by miR-29c overexpression. Additionally, it was discovered that miR-29c and NFAT5 had a negative correlation. In microglia treated with miR-29c inhibitor, NFAT5 knockdown prevented the inflammation from becoming more severe. Therefore, these results imply that miR-29c targets NFAT5, which is a prospective therapeutic target for PD, and controls the NLRP3 inflammasome to inhibit microglial inflammatory responses ( ).
Thus, the miRNAs played an important role in the signaling pathways involved in PD and they could be used as potential biomarker as well as therapeutic target. represents the miRNAs as biomarkers and therapeutic targets in Parkinson’s disease.
miRNAs as biomarkers and therapeutic targets in Parkinson’s disease.
## MicroRNAs in the pathophysiology of Parkinson’s disease involving genes
miRNAs are found to be involved in the disease pathogenesis of PD by regulating the various genes which play a major role in PD. Following are certain miRNAs regulating important genes involved in PD. represents the miRNA regulating specific genes in Parkinson’s disease.
miRNA regulating specific genes in Parkinson’s disease.
### Brain-derived neurotrophic factor
The growth factor known as Brain-Derived Neurotrophic Factor (BDNF) is a member of the neurotrophin family and is essential for the survival of existing neurons as well as the promotion of neurogenesis. The loss of BDNF in the substantia nigra pars compacta (SNpc) causes a dopaminergic deficit in the striatum, which is a key component of PD ( ). A study discussed that 1-Methyl-4-phenylpyridinium (MPP +) controls the production of BDNF through miR-210-3p. Through a transcription-unrelated mechanism, MPP + prevents the synthesis of BDNF in SH-SY5Y cells. Additionally, MPP + was found to increase the expression of miR-210-3p, which targets the BDNF mRNA in SH-SY5Y cells. Moreover, miR-210-3p suppression increases DA neuron survival in the MPTP animal model and prevents MPP + from reducing BDNF synthesis ( ).
In a study by , it was identified that by decreasing NLRP3 inflammasome activation and negatively regulating Nlrp3 expression in the MPTP-induced PD mouse model, miR-30e reduces neuronal injury, neuroinflammation, and dyskinesia. In this work, it was discovered that miR-30e agomir administration significantly restored the decrease in BDNF secretion in SNpc. A prolonged decline in BDNF mRNA expression can be seen in SNpc of PD patients, and abnormal changes in BDNF expression or signaling may play a role in neurodegeneration ( ). In addition, through an auto-regulatory mechanism, miR-7 controls the expression of BDNF and causes the downregulation of α-synuclein (α-syn), which is associated with the neuropathology of PD. These results suggest that miRNA-7 modulates the BDNF/α-syn axis in the early stages of Parkinson’s disease and may be used as a biomarker or therapeutic target ( ).
Interestingly, miR-494-3p was found to be upregulated in PD and was found to be targeting BDNF leading to neurotoxicity. Thus, inhibition of miR-494-3p was found to reverse the effect and increase BDNF levels. Hence, miR-494-3p is found to be a potential target for PD ( ).
Moreover, BDNF was clinically tested for Alzheimer’s disease and an adenovirus vector (AAV2) was used as a carrier. Since in animal models, BDNF was found to reduce cell loss, promote cell function, and create fresh connections (synapses) between brain cells, it was subjected to phase I clinical trials. The study is currently recruiting candidates for the trials. Thus it could be understood that BDNF is an important gene that plays an important role in neurodegenerative disorders including PD.
### Leucine-rich repeat kinase 2
Leucine-rich repeat kinase 2 (LRRK2) is a key player in the genesis of PD. Mutations in LRRK2 are the main cause of both inherited and spontaneous PD. One of the research examined 45 people who had the LRRK2 gene mutated (LRRK2-PD). It was discovered that miR-155 was upregulated in LRRK-PD and that miR-146a, miR-335-3p, and miR-335-5p were downregulated in LRRK2-PD. Thus, these miRNAs were found to be involved in the mutation of the LRRK2 gene in PD patients ( ). Interestingly, in a study by , hsa-miR-4671-3p, hsa-miR-335-3p, hsa-miR-561-3p, hsa-miR-579-3p, and hsa-miR-3143 were found to be differentially expressed in the whole blood of PD patients and these miRNAs were found to be targeting LRRK2 gene. Among these miRNAs, has-miR-561-3p was found to be the highly specific miRNA targeting LRRK2 in PD ( ).
Moreover, in various PD-mimicking circumstances, miR-335 is markedly downregulated, and miR-335 specifically targeted LRRK2 mRNA. The expression of pro-inflammatory genes induced by α-synuclein was reduced by miR-335. The effects of conventional inflammatory stimuli or LRRK2-Wt overexpression are greatly inhibited by miR-335, which, in turn, attenuates chronic neuroinflammation in both microglia and neuronal cells ( ). Another study discovered that miR-599 protected brain cells by controlling the expression of LRRK2. In the brain tissues of PD animals, LRRK2 was substantially expressed compared to miR-599’s low expression. Additionally, LRRK2 expression at the mRNA and protein levels could be negatively regulated by miR-599. By suppressing the expression of LRRK2, miR-599 overexpression prevented SHP-SY5Y cells from being damaged by MPP + treatment ( ). Thus miR-599 was identified as a potential therapeutic target for PD.
In fact, it was discovered that the conserved binding site at the 3’-untranslated region (UTR) of the LRRK2 gene allowed miR-205 to inhibit the expression of the LRRK2 protein. It is interesting to note that patients with sporadic PD had much lower levels of miR-205 expression in their brains, which was indicative of elevated LRRK2 protein levels. Studies conducted in vitro on primary neuron cultures and cell lines further confirmed the function of miR-205 in regulating the production of the LRRK2 protein. In addition, the injection of miR-205 corrected the neurite outgrowth abnormalities in the neurons expressing a PD-related LRRK2 mutant. Collectively, these results imply that miR-205 may be downregulated, which may contribute to the potential pathogenic elevation of LRRK2 protein in the brains of sporadic PD patients, while miR-205 overexpression may offer a useful therapeutic approach to suppress the abnormal upregulation of LRRK2 protein in PD ( ). Similarly, miR-199a-3p was found to be involved in the regulation of LRRK2 in the disease pathogenesis of PD ( ).
In addition, it was discovered that miR-712 targets the robust inflammatory gene LRRK2 and inhibits p38 and ERK1/2 kinase activation ( ). However, their role in PD is yet to be elucidated. Another study by exhibited that LRRK2 degradation was accelerated by NEDD4 which is targeted by the miR-7/STAT3 axis. Thus, the study proposed the mechanisms of bone marrow mesenchymal stem cells derived exosomal TSG-6 to be a regulator of the STAT3/miR-7/NEDD4/LRRK2 axis ( ). Additionally, miR-30c-5p was identified to target LRRK2 and play an important role in the pathogenesis of PD. It was also postulated that the long non-coding RNA Linc00938 could bind directly to hsa-miR-30c-5p, perhaps influencing LRRK2 expression via the miR-30c-5p sponge ( ).
### Death-associated protein kinase 1
A common serine/threonine (Ser/Thr) kinase known as death-associated protein kinase 1 (DAPK1) is essential for cell death in a number of neurological diseases. One of the research found a favorable correlation between neuronal synucleinopathy and DAPK1 overexpression in PD animals. Additionally, it was discovered that synucleinopathy, DA neuron cell death, and motor impairments arise from downregulating miR-26a or upregulating DAPK1. By directly phosphorylating α-synuclein at Ser129, DAPK1 overexpression encourages PD-like symptoms. In line with this, motor problems, synucleiopathy, and the loss of dopaminergic neurons were all prevented by a cell-permeable competitive peptide that prevents the phosphorylation of α -synuclein ( ). miR-26a is found to be influencing DAPK in PD and hence it could be used as a potential therapeutic target for the treatment of PD. Likewise, miR-151-3p is also found to be increasing the neuroprotection by targeting DAPK ( ).
### Glutamate transporter 1
PD can be characterized by glutamate excitotoxicity, which is brought on by malfunctioning GLT1. However, the processes behind the control of GLT in PD are still not completely understood. One of the studies demonstrated that mice treated with MPTP and astrocytes treated with MPP+, miR-30a-5p reduced GLT-1 expression and function. It was also discovered that miR-30a-5p knockdown boosted GLT-1 expression and accelerated glutamate absorption both in vitro and in vivo by preventing GLT-1 ubiquitination and subsequent degradation in a PKCα-dependent way ( ). Another investigation discovered that in mice treated with MPTP and astrocytes treated with MPP+, miR-543-3p can inhibit GLT-1 expression and function. Suppression of miR-543-3p can alleviate dyskinesia and restore GLT-1 expression and function in the PD model, which raises the possibility that miR-543-3p inhibition could be used as a possible therapeutic target for PD ( ). The studies summarized the role of miRNAs in GLT1 dysfunction which is identified to be playing important role in the disease progression of PD.
### Tumor growth factor-β1
A pleiotropic cytokine with immunosuppressive and anti-inflammatory effects is TGF-β1. Recent research has demonstrated that pretreatment with TGF-β1 in vitro prevents the death of dopaminergic neurons caused by MPP+, which is a hallmark of PD ( ). Thus TGF-β1 is identified to be an important gene involved in dopaminergic inflammation which could be regulated by several miRNAs. But studies are sparse on the role of miRNAs in regulating TGF-β1 in PD. Interestingly, a study by investigated the interactions between LncRNA MIAT and miR-221-3p regulated TGF-β1/Nrf2. The study employed the use of a miR-221-3p mimic, a miR-221-3p inhibitor, an NC-inhibitor, and a shRNA (shTGF-β1) which were transfected into MPP+-treated cells. To examine the interaction of miR-221-3p with MIAT or TGFB receptor 1 (TGFBR1), dual-luciferase reporter gene experiments were done. As a consequence, LncRNA MIAT was abundantly produced in PD animals and cells, but LncRNA MIAT downregulation enhanced neuron survival, prevented apoptosis, and reduced oxidative stress in neurons. LncRNA MIAT is bound to miR-221-3p, and the expression of LncRNA MIAT was negatively correlated with miR-221-3p. Furthermore, miR-221-3p inhibited TGF-β1 expression while increasing Nrf2 expression. MIAT, a LncRNA, enhanced MPP+-induced neuronal damage in PD through modulating the TGF-β1/Nrf2 axis by interaction with miR-221-3p ( ).
## Future prospects
One of the most widely expressed miRs in the brain, miR-124 is involved in autophagy, inflammation, synapse architecture, neurotransmission, and neurogenesis. miR-124 is found to be acting through the pathways of calpain 1/p25/cyclin-dependent kinases 5 (CDK5), signal transducer and activator of transcription 3 (STAT3) and Bcl-2-interacting mediator of cell death (Bim). Moreover, there is also evidence of miR-124 regulating nuclear factor-kappa B (NF-κB), AMPK and ERK signaling ( ). Thus miR-124 being an important therapeutic target makes it a potential treatment strategy for PD. There are also studies on miR-34a whose reduction or inhibition increased neuronal survival against numerous neurotoxins linked to PD. miR-34a is found to be regulating various signaling pathways like Nrf2, SIRT1/mTOR and Notch signaling in the pathophysiology of PD ( ; ; ). miR-34a not only regulates these signaling pathways but also plays a major role in the disease progression in PD.
Since various miRNAs are being studied for their potential role in regulating various genes and signaling pathways, they also could be used as potential therapeutic targets and biomarkers for PD. miRNA-based therapeutics have been researched on wide-scale nowadays due to their potential therapeutic properties. But the research on miRNA mimics and miRNA inhibitors in PD is sparse. Hence there is a requirement for many validated studies to seek a clinical breakthrough in miRNA-based therapeutics in PD.
Moreover, research on biomarkers could be useful in identifying various blood markers that help in the early diagnosis and monitoring of the disease. Since there is a need for early diagnosis, research on miRNAs could pave way for the prognosis and diagnosis of PD.
## Conclusion
Parkinson’s disease is a commonly found neurodegenerative disease with late-onset symptoms making it difficult to diagnose at earlier stages. Moreover, the pharmacological treatment for PD currently lacks specificity and has adverse effects. Hence there is a need for studies on biomarkers and therapeutic targets. miRNAs being suitable candidates are being researched for their role in disease progression in PD. miRNAs are found to be regulating various genes and their signaling pathways in PD. Hence, they could be used as biomarkers as well as therapeutic targets.
## Author contributions
DS and DT conceived the idea. DS wrote the first draft of the manuscript. SS, KP, DT, and DS wrote the complete manuscript. All authors contributed to the article and approved the submitted version.
|
Auditory verbal hallucinations (AVH) or "voices" are a characteristic symptom of schizophrenia, but can also be observed in healthy individuals in the general population. As these non-psychotic individuals experience AVH in the absence of other psychiatric symptoms and medication-use they provide an excellent model to study AVH in isolation. Indeed a number of studies used this approach and investigated brain structure and function in non-psychotic individuals with AVH. These studies showed that increased sensitivity of auditory areas to auditory stimulation and aberrant connectivity of language production and perception areas is associated with AVH. This is in concordance with investigations that observed prominent activation of these areas during the state of AVH. Moreover, while effortful attention appears not to be related to AVH, individuals prone to hallucinate seem to have an enhanced attention bias to auditory stimuli which may stem from aberrant activation of the anterior cingulated regions. Furthermore, it was observed that decreased cerebral dominance for language and dopamine dysfunction, which are consistently found in schizophrenia, are most likely not specifically related to AVH as these abnormalities were absent in healthy voice hearers. Finally, specific aspects of AVH such as voluntary control may be related to the timing of the supplementary motor area and language areas in the experience of AVH. |
Several neurophysiologic and neuroimaging studies suggested that motor and perceptual systems are tightly linked along a continuum rather than providing segregated mechanisms supporting different functions. Using correlational approaches, these studies demonstrated that action observation activates not only visual but also motor brain regions. On the other hand, brain stimulation and brain lesion evidence allows tackling the critical question of whether our action representations are necessary to perceive and understand others' actions. In particular, recent neuropsychological studies have shown that patients with temporal, parietal, and frontal lesions exhibit a number of possible deficits in the visual perception and the understanding of others' actions. The specific anatomical substrates of such neuropsychological deficits however, are still a matter of debate. Here we review the existing literature on this issue and perform an anatomic likelihood estimation meta-analysis of studies using lesion-symptom mapping methods on the causal relation between brain lesions and non-linguistic action perception and understanding deficits. The meta-analysis encompassed data from 361 patients tested in 11 studies and identified regions in the inferior frontal cortex, the inferior parietal cortex and the middle/superior temporal cortex, whose damage is consistently associated with poor performance in action perception and understanding tasks across studies. Interestingly, these areas correspond to the three nodes of the action observation network that are strongly activated in response to visual action perception in neuroimaging research and that have been targeted in previous brain stimulation studies. Thus, brain lesion mapping research provides converging causal evidence that premotor, parietal and temporal regions play a crucial role in action recognition and understanding. |
Researchers are enthusiastically concerned about neural stem cell (NSC) therapy in a wide array of diseases, including stroke, neurodegenerative disease, spinal cord injury, and depression. Although enormous evidences have demonstrated that neurobehavioral improvement may benefit from NSC-supporting regeneration in animal models, approaches to endogenous and transplanted NSCs are blocked by hurdles of migration, proliferation, maturation, and integration of NSCs. Electrical stimulation (ES) may be a selective non-drug approach for mobilizing NSCs in the central nervous system. This technique is suitable for clinical application, because it is well established and its potential complications are manageable. Here, we provide a comprehensive review of the emerging positive role of different electrical cues in regulating NSC biology in vitro and in vivo, as well as biomaterial-based and chemical stimulation of NSCs. In the future, ES combined with stem cell therapy or other cues probably becomes an approach for promoting brain repair. |
The formation of coherent multisensory percepts requires integration of stimuli across the multiple senses. Patients with schizophrenia (ScZ) often experience a loss of coherent perception and hence, they might also show dysfunctional multisensory processing. In this high-density electroencephalography study, we investigated the neural signatures of the McGurk illusion, as a phenomenon of speech-specific multisensory processing. In the McGurk illusion lip movements are paired with incongruent auditory syllables, which can induce a fused percept. In ScZ patients and healthy controls we compared neural oscillations and event-related potentials (ERPs) to congruent audiovisual speech stimuli and McGurk illusion trials, where a visual /ga/ and an auditory /pa/ was often perceived as /ka/. There were no significant group differences in illusion rates. The EEG data analysis revealed larger short latency ERPs to McGurk illusion compared with congruent trials in controls. The reversed effect pattern was found in ScZ patients, indicating an early audiovisual processing deficit. Moreover, we observed stronger suppression of medio-central alpha-band power (8-10 Hz, 550-700 ms) in response to McGurk illusion compared with control trials in the control group. Again, the reversed pattern was found in SCZ patients. Moreover, within groups, alpha-band suppression was negatively correlated with the McGurk illusion rate in ScZ patients, while the correlation tended to be positive in controls. The topography of alpha-band effects indicated an involvement of auditory and/or frontal structures. Our study suggests that short latency ERPs and long latency alpha-band oscillations reflect abnormal multisensory processing of the McGurk illusion in ScZ. |
Earlier studies have revealed cross-modal visuo-tactile interactions in endogenous spatial attention. The current research used event-related potentials (ERPs) and virtual reality (VR) to identify how the visual cues of the perceiver's body affect visuo-tactile interaction in endogenous spatial attention and at what point in time the effect takes place. A bimodal oddball task with lateralized tactile and visual stimuli was presented in two VR conditions, one with and one without visible hands, and one VR-free control with hands in view. Participants were required to silently count one type of stimulus and ignore all other stimuli presented in irrelevant modality or location. The presence of hands was found to modulate early and late components of somatosensory and visual evoked potentials. For sensory-perceptual stages, the presence of virtual or real hands was found to amplify attention-related negativity on the somatosensory N140 and cross-modal interaction in somatosensory and visual P200. For postperceptual stages, an amplified N200 component was obtained in somatosensory and visual evoked potentials, indicating increased response inhibition in response to non-target stimuli. The effect of somatosensory, but not visual, N200 enhanced when the virtual hands were present. The findings suggest that bodily presence affects sustained cross-modal spatial attention between vision and touch and that this effect is specifically present in ERPs related to early- and late-sensory processing, as well as response inhibition, but do not affect later attention and memory-related P3 activity. Finally, the experiments provide commeasurable scenarios for the estimation of the signal and noise ratio to quantify effects related to the use of a head mounted display (HMD). However, despite valid a-priori reasons for fearing signal interference due to a HMD, we observed no significant drop in the robustness of our ERP measurements. |
Under natural behavioral conditions, visually guided eye movements are linked to direction-specific modulations of cortico-spinal system (CSS) excitability in upper-limb muscles, even in absence of a manual response. These excitability changes have been shown to be compatible with a covert motor program encoding a manual movement toward the same target of the eyes. The aim of this study is to investigate whether this implicit oculo-manual coupling is enforced following every saccade execution or it depends on the behavioral context. Twenty-two healthy young adults participated in the study. Single-pulse transcranial magnetic stimulation was applied to the motor cortex at nine different time epochs during a double-choice eye task, in which the decision to execute a prosaccade or an antisaccade was made on the color of a peripheral visual cue. By analyzing the amplitude of the motor evoked potentials (MEP) in three distal muscles of the resting upper-limb, a facilitation peak of CSS excitability was found in two of them at 120 ms before the eyes begin to move. Furthermore, a long-lasting, generalized reduced corticomotor excitability develops following the eye response. Finally, a quite large modulation of MEP amplitude, depending on the direction of the saccade, is observed only in the first dorsal interosseous muscle, in a narrow time window at about 150 ms before the eye movement, irrespective of the type of the ocular response (pro-/anti-saccade). This change in CSS excitability is not tied up to the timing of the occurrence of the visual cue but, instead, appears to be tightly time-related to the saccade onset. Observed excitability changes differ in many respects from those previously reported with different behavioral paradigms. A main finding of our study is that the implicit coupling between eye and hand motor systems is contingent upon the particular motor set determined by the cognitive aspects of the performed oculomotor task. In particular, the direction-specific modulation in CSS excitability described in this study appears to be related to perceptual and decision-making processes rather than representing an implicit upper-limb motor program, coupled to the saccade execution. |
To characterize each cognitive function <i>per se</i> and to understand the brain as an aggregate of those functions, it is vital to relate dozens of these functions to each other. Knowledge about the relationships among cognitive functions is informative not only for basic neuroscientific research but also for clinical applications and developments of brain-inspired artificial intelligence. In the present study, we propose an exhaustive data mining approach to reveal relationships among cognitive functions based on functional brain mapping and network analysis. We began our analysis with 109 pseudo-activation maps (cognitive function maps; CFM) that were reconstructed from a functional magnetic resonance imaging meta-analysis database, each of which corresponds to one of 109 cognitive functions such as 'emotion,' 'attention,' 'episodic memory,' etc. Based on the resting-state functional connectivity between the CFMs, we mapped the cognitive functions onto a two-dimensional space where the relevant functions were located close to each other, which provided a rough picture of the brain as an aggregate of cognitive functions. Then, we conducted so-called conceptual analysis of cognitive functions using clustering of voxels in each CFM connected to the other 108 CFMs with various strengths. As a result, a CFM for each cognitive function was subdivided into several parts, each of which is strongly associated with some CFMs for a subset of the other cognitive functions, which brought in sub-concepts (i.e., sub-functions) of the cognitive function. Moreover, we conducted network analysis for the network whose nodes were parcels derived from whole-brain parcellation based on the whole-brain voxel-to-CFM resting-state functional connectivities. Since each parcel is characterized by associations with the 109 cognitive functions, network analyses using them are expected to inform about relationships between cognitive and network characteristics. Indeed, we found that informational diversities of interaction between parcels and densities of local connectivity were dependent on the kinds of associated functions. In addition, we identified the homogeneous and inhomogeneous network communities about the associated functions. Altogether, we suggested the effectiveness of our approach in which we fused the large-scale meta-analysis of functional brain mapping with the methods of network neuroscience to investigate the relationships among cognitive functions. |
Magnetoencephalographic imaging (MEGI) offers a non-invasive alternative for defining preoperative language lateralization in neurosurgery patients. MEGI indeed can be used for accurate estimation of language lateralization with a complex language task - auditory verb generation. However, since language function may vary considerably in patients with focal lesions, it is important to optimize MEGI for estimation of language function with other simpler language tasks. The goal of this study was to optimize MEGI laterality analyses for two such simpler language tasks that can have compliance from those with impaired language function: a non-word repetition (NWR) task and a picture naming (PN) task. Language lateralization results for these two tasks were compared to the verb-generation (VG) task. MEGI reconstruction parameters (regions and time windows) for NWR and PN were first defined in a presurgical training cohort by benchmarking these against laterality indices for VG. Optimized time windows and regions of interest (ROIs) for NWR and PN were determined by examining oscillations in the beta band (12-30 Hz) a marker of neural activity known to be concordant with the VG laterality index (LI). For NWR, additional ROIs include areas MTG/ITG and for both NWR and PN, the postcentral gyrus was included in analyses. Optimal time windows for NWR were defined as 650-850 ms (stimulus-locked) and -350 to -150 ms (response-locked) and for PN -450 to -250 ms (response-locked). To verify the optimal parameters defined in our training cohort for NWR and PN, we examined an independent validation cohort (<i>n</i> = 30 for NWR, <i>n</i> = 28 for PN) and found high concordance between VG laterality and PN laterality (82%) and between VG laterality and NWR laterality (87%). Finally, in a test cohort (<i>n</i> = 8) that underwent both the intracarotid amobarbital procedure (IAP) test and MEG for VG, NWR, and PN, we identified excellent concordance (100%) with IAP for VG + NWR + PN composite LI, high concordance for PN alone (87.5%), and moderate concordance for NWR alone (66.7%). These findings provide task options for non-invasive language mapping with MEGI that can be calibrated for language abilities of individual patients. Results also demonstrate that more accurate estimates can be obtained by combining laterality estimates obtained from multiple tasks. MEGI. |
Biased attention towards emotional stimuli is adaptive, as it facilitates responses to important threats and rewards. An unfortunate consequence is that emotional stimuli can become potent distractors when they are irrelevant to current goals. How can this distraction be overcome despite the bias to attend to emotional stimuli? Recent studies show that distraction by irrelevant flankers is reduced when distractor frequency is high, even if they are emotional. A parsimonious explanation is that the expectation of frequent distractors promotes the use of proactive control, whereby attentional control settings can be altered to minimize distraction before it occurs. It is difficult, however, to infer proactive control on the basis of behavioral data alone. We therefore measured neural indices of proactive control while participants performed a target-detection task in which irrelevant peripheral distractors (either emotional or neutral) could appear either frequently (on 75% of trials) or rarely (on 25% of trials). We measured alpha power during the pre-stimulus period to assess proactive control and during the post-stimulus period to determine the consequences of control for subsequent processing. Pre-stimulus alpha power was tonically suppressed in the high, compared to low, distractor frequency condition, regardless of expected distractor valence, indicating sustained use of proactive control. In contrast, post-stimulus alpha suppression was reduced in the high-frequency condition, suggesting that proactive control reduced the need for post-stimulus adjustments. Our findings indicate that a sustained proactive control strategy accounts for the reduction in both emotional and non-emotional distraction when distractors are expected to appear frequently. |
Tension is one of the core principles of emotion evoked by music, linking objective musical events and subjective experience. The present study used continuous behavioral rating and electroencephalography (EEG) to investigate the dynamic process of tension generation and its underlying neurocognitive mechanisms; specifically, tension induced by structural violations at different music hierarchical levels. In the experiment, twenty-four musicians were required to rate felt tension continuously in real-time, while listening to music sequences with either well-formed structure, phrase violations, or period violations. The behavioral data showed that structural violations gave rise to increasing and accumulating tension experience as the music unfolded; tension was increased dramatically by structural violations. Correspondingly, structural violations elicited N5 at GFP peaks, and induced decreasing neural oscillations power in the alpha frequency band (8-13 Hz). Furthermore, compared to phrase violations, period violations elicited larger N5 and induced a longer-lasting decrease of power in the alpha band, suggesting a hierarchical manner of musical processing. These results demonstrate the important role of musical structure in the generation of the experience of tension, providing support to the dynamic view of musical emotion and the hierarchical manner of tension processing. |
<b>Background:</b> Electroconvulsive therapy (ECT) is one of the most effective treatments for major depressive disorder. Recently, there has been increasing attention to evaluate the effect of ECT on resting-state functional magnetic resonance imaging (rs-fMRI). This study aims to compare rs-fMRI of depressive disorder (DEP) patients with healthy participants, investigate whether pre-ECT dynamic functional network connectivity network (dFNC) estimated from patients rs-fMRI is associated with an eventual ECT outcome, and explore the effect of ECT on brain network states. <b>Method:</b> Resting-state functional magnetic resonance imaging (fMRI) data were collected from 119 patients with depression or depressive disorder (DEP) (76 females), and 61 healthy (HC) participants (34 females), with an age mean of 52.25 (<i>N</i> = 180) years old. The pre-ECT and post-ECT Hamilton Depression Rating Scale (HDRS) were 25.59 ± 6.14 and 11.48 ± 9.07, respectively. Twenty-four independent components from default mode (DMN) and cognitive control network (CCN) were extracted, using group-independent component analysis from pre-ECT and post-ECT rs-fMRI. Then, the sliding window approach was used to estimate the pre-and post-ECT dFNC of each subject. Next, k-means clustering was separately applied to pre-ECT dFNC and post-ECT dFNC to assess three distinct states from each participant. We calculated the amount of time each subject spends in each state, which is called "occupancy rate" or OCR. Next, we compared OCR values between HC and DEP participants. We also calculated the partial correlation between pre-ECT OCRs and HDRS change while controlling for age, gender, and site. Finally, we evaluated the effectiveness of ECT by comparing pre- and post-ECT OCR of DEP and HC participants. <b>Results:</b> The main findings include (1) depressive disorder (DEP) patients had significantly lower OCR values than the HC group in state 2, where connectivity between cognitive control network (CCN) and default mode network (DMN) was relatively higher than other states (corrected <i>p</i> = 0.015), (2) Pre-ECT OCR of state, with more negative connectivity between CCN and DMN components, is linked with the HDRS changes (R = 0.23 corrected <i>p</i> = 0.03). This means that those DEP patients who spent less time in this state showed more HDRS change, and (3) The post-ECT OCR analysis suggested that ECT increased the amount of time DEP patients spent in state 2 (corrected <i>p</i> = 0.03). <b>Conclusion:</b> Our finding suggests that dynamic functional network connectivity (dFNC) features, estimated from CCN and DMN, show promise as a predictive biomarker of the ECT outcome of DEP patients. Also, this study identifies a possible underlying mechanism associated with the ECT effect on DEP patients. |
Neuroscience has had access to high-resolution recordings of large-scale cortical activity and structure for decades, but still lacks a generally adopted basis to analyze and interrelate results from different individuals and experiments. Here it is argued that the natural oscillatory modes of the cortex-cortical eigenmodes-provide a physically preferred framework for systematic comparisons across experimental conditions and imaging modalities. In this framework, eigenmodes are analogous to notes of a musical instrument, while commonly used statistical patterns parallel frequently played chords. This intuitive perspective avoids problems that often arise in neuroimaging analyses, and connects to underlying mechanisms of brain activity. We envisage this approach will lead to novel insights into whole-brain function, both in existing and prospective datasets, and facilitate a unification of empirical findings across presently disparate analysis paradigms and measurement modalities. |
Haptic object recognition is usually an efficient process although slower and less accurate than its visual counterpart. The early loss of vision imposes a greater reliance on haptic perception for recognition compared to the sighted. Therefore, we may expect that congenitally blind persons could recognize objects through touch more quickly and accurately than late blind or sighted people. However, the literature provided mixed results. Furthermore, most of the studies on haptic object recognition focused on performance, devoting little attention to the exploration procedures that conducted to that performance. In this study, we used iCube, an instrumented cube recording its orientation in space as well as the location of the points of contact on its faces. Three groups of congenitally blind, late blind and age and gender-matched blindfolded sighted participants were asked to explore the cube faces where little pins were positioned in varying number. Participants were required to explore the cube twice, reporting whether the cube was the same or it differed in pins disposition. Results showed that recognition accuracy was not modulated by the level of visual ability. However, congenitally blind touched more cells simultaneously while exploring the faces and changed more the pattern of touched cells from one recording sample to the next than late blind and sighted. Furthermore, the number of simultaneously touched cells negatively correlated with exploration duration. These findings indicate that early blindness shapes haptic exploration of objects that can be held in hands. |
In recent years, several theories have been proposed in attempts to identify the neural mechanisms underlying successful cognitive aging. Old subjects show increased neural activity during the performance of tasks, mainly in prefrontal areas, which is interpreted as a compensatory mechanism linked to functional brain efficiency. Moreover, resting-state studies have concluded that elders show disconnection or disruption of large-scale functional networks. We used functional MRI during resting-state and a verbal n -back task with different levels of memory load in a cohort of young and old healthy adults to identify patterns of networks associated with working memory and brain default mode. We found that the disruption of resting-state networks in the elderly coexists with task-related overactivations of certain brain areas and with reorganizations within these functional networks. Moreover, elders who were able to activate additional areas and to recruit a more bilateral frontal pattern within the task-related network achieved successful performance on the task. We concluded that the balanced and plastic reorganization of brain networks underlies successful cognitive aging. This observation allows the integration of several theories that have been proposed to date regarding the aging brain.
## Introduction
Cognitive aging affects a wide range of functions including working memory, processing speed, and inhibitory function (Park et al., ; Reuter-Lorenz and Park, ). Despite the gradual decline described in aging, some seniors are able to keep their cognitive functions with minimal differences in performance compared to healthy young subjects. Several theories have been proposed in attempts to identify the neural correlates of what is known as “successful cognitive aging” (Cabeza et al., ; Park and Reuter-Lorenz, ).
Functional imaging is well suited to the study of changes in brain functionality in advanced age (for a review, see Eyler et al., ). Across various cognitive domains, age-related functional reorganizations have been described as changes in brain responsivity in several brain regions when subjects are scanned during the performance of cognitively demanding tasks (see Spreng et al., ; Turner and Spreng, for recent meta-analyses). Hence, both reductions and increases in activity have been described in different brain regions. Reductions in activity are commonly located in the left prefrontal cortex (PFC) and temporo-occipital areas and are normally associated with less efficient processing in aging (Cabeza et al., ), but the interpretation of increases is less straightforward. However, when associated with better or preserved performance they have been commonly interpreted as evidence of functional compensatory mechanisms (Grady, ; Cabeza et al., ; Grady et al., ; Mattay et al., ; Berlingeri et al., ).
Moreover, recent advances in neuroimaging techniques have made possible the study of functional brain networks that can be observed even during resting-state periods, revealing an intrinsic network-based organization of the brain (Smith et al., ). In this context, cognitive aging has been associated with disruptions/reorganizations within certain brain functional networks (Grady et al., ; Littow et al., ; Tomasi and Volkow, ). Particularly, areas forming part of the default-mode network (DMN), including the posterior cingulate cortex and middle frontal gyrus, are characterized by patterns of age-related decreases in functional connectivity (Damoiseaux et al., ; Hafkemeijer et al., ). These functional correlation reductions have been associated with cognitive decline across multiple domains in healthy old individuals (Andrews-Hanna et al., ).
Despite this evidence of changes in both brain activity and brain connectivity, very few studies have simultaneously investigated the DMN and a task-related brain network in aging in terms of their BOLD responsivity and functional connectivity, and during both resting and cognitive performance. Hence, the objective of our study was to investigate the brain connectivity/activity characteristics of the DMN and a working memory network in a sample of healthy elders (HE) and young adults (YA). In the present report, we placed special emphasis on investigating how these patterns differ between elders who are able to keep their working memory abilities at a level comparable to young subjects during the most demanding conditions, and those who show a decline in this function.
## Material and Methods
### Subjects and scanning
Twenty-nine HE (mean age: 62.55, SD: 9.43, 20 women) and 16 YA (mean age: 21.31, SD: 2.41, 9 women) were included in the study. Old subjects underwent neuropsychological testing, including memory, language, attention, and visuoperceptive/visuospatial functions. The neuropsychological battery was similar to the one used recently in other reports by our group (e.g., Arenaza-Urquijo et al., ). All reported scores were within normal range on the domains tested, and all subjects had scores on the mini-mental state examination ≥ 24 (mean: 28.83, SD: 1.64). All participants in the study were scanned using functional MRI (fMRI) in the resting-state and during the performance of a working memory task, an n -back task including different levels of working memory load ( n = 0, 1, 2, and 3 letters to be retained: see Braver et al., ; Sala-Llonch et al., ). Basically, during the task, blocks of 0-, 1-, 2-, 3-back conditions lasting 26 s were presented four times each in a pseudo-randomized order with inter-block fixation periods (white cross on a black screen) of 13 s. Before any n -back block was presented, an instruction screen appeared to inform the subject about the task. Within each block, a sequence of 12 letters was presented in white in the center of the screen; each letter remained visible for 500 ms, with an inter-stimulus interval of 1500 ms. The subject was asked to press a button when the stimulus on the screen was the same as the one showed n items before. For the 0-back task, subjects were asked to press the button when the letter “X” appeared. All subjects underwent a training session before entering the scanner in order to ensure that they understood the task instructions. All achieved a task accuracy of at least 80%.
Subjects’ responses were collected and the performance of each n -back condition was calculated using the d ′ measure (Z hit rate − Z false alarm rate ), with higher d ′ scores indicating higher performance. Mean reaction time (RT) was also collected for each subject within each load condition.
The HE group was further subdivided into low performers (low-HE) and high-performers (high-HE) according to the score obtained during the performance of the 3-back task (percentile 50 of the distribution). Between-group differences in d ′ measures and RT measures were assessed with one-way ANOVA implemented in PASW vs. 17 (Statistical Package for Social Sciences, Chicago, IL, USA).
Functional MRI images were acquired on a 3T MRI scanner (Magnetom Trio Tim, Siemens Medical Systems, Germany), using a 32-channel coil. During both the resting-fMRI and task-fMRI conditions, a set of T2 -weighted volumes (150 and 336 volumes for resting and task fMRI, respectively) were acquired (TR = 2000 ms, TE = 29 ms, 36 slices per volume, slice thickness = 3 mm, distance factor = 25%, FOV = 240 mm, matrix size = 128 × 128). A high-resolution 3D structural dataset (T1-weighted MPRAGE, TR = 2300 ms, TE = 2.98 ms, 240 slices, FOV = 256 mm; matrix size = 256 × 256; slice thickness = 1 mm) was also acquired in the same scanning session for registration purposes.
### Analysis of resting-fMRI data
Resting-state fMRI images were analyzed with independent component analysis (ICA) and a dual-regression approach. Image preprocessing was carried out in FSL and AFNI softwares. This step included the removal of the first five scans, motion correction, skull stripping, spatial smoothing using a Gaussian kernel of FWHM = 6 mm, grand mean scaling, temporal filtering (low-pass and high-pass filters). Functional scans were then registered to their corresponding individual MPRAGE structural scans using linear registration with 6 df (Jenkinson and Smith, ) and further registered to the standard MNI template by concatenation of both registration matrices. Resampling resolution was set to 3 mm.
We used ICA, as implemented in MELODIC (Beckmann et al., ) from FSL, in order to decompose resting-state data into 25 independent components (ICs) which described common spatio-temporal independent patterns of correlated brain activity across the whole group of subjects in the study. Within the 25 ICs obtained, we identified the common resting-state functional networks (Damoiseaux et al., ; Smith et al., ; van den Heuvel and Hulshoff Pol, ), and selected the DMN, and two components corresponding to the right-lateralized and the left-lateralized fronto-parietal networks (right-FPN, and left-FPN). The selection procedure was performed by visual inspection together with template matching with online available data (Smith et al., ; Biswal et al., ) and with average task-related activation maps obtained from the data-driven analysis (see Figure for a summary of the methods used in the study).
Summary of the methods used for preprocessing and analysis of fMRI data .
Then, we used the spatial patterns of the three selected networks in a dual-regression approach (as described in Filippini et al., ; Leech et al., ) in order to explore between-group differences. In the dual-regression analysis, we first regressed each subject’s resting-state functional data against the spatial IC maps and obtained individual time-series associated to each network (DMN, right-FPN, and left-FPN). These time-courses were then used to regress again the individual preprocessed fMRI data and to obtain individual spatial maps that were also specific for networks. Spatial maps were finally tested for voxel-wise differences between groups using non-parametric testing with 5000 random permutations.
### Analysis of task-fMRI data
Task-fMRI images were analyzed with a model-driven protocol to explore ROI-based BOLD signal change across task conditions. We also used a dual-regression analysis of task-fMRI data to investigate differences in network integration.
First, data preprocessing was performed in FSL and AFNI. It included motion correction, skull stripping, spatial smoothing using a Gaussian kernel of FWHM = 6 mm, grand mean scaling, temporal filtering (high-pass filter of sigma = 80 s), and registration to individual anatomical scans and to MNI standard space (Jenkinson and Smith, ). As in resting-fMRI, resampling resolution was set to 3 mm.
Task-fMRI data were analyzed using standard random-effects general linear model. We used the procedure as implemented in FMRI Analysis Tool (FEAT, Woolrich et al., ) from FSL. Five regressors were used to model the different blocks (0, 1-, 2-, 3-back, and fixation), and five additional regressors, modeling their first derivatives were introduced as nuisance variables. Contrast images were computed from the preprocessed functional data as follows: for the different levels of cognitive load, each condition was evaluated against the 0-back (1-, 2-, and 3-back >0-back), and for the fixation blocks, the signal was compared against the average of the other blocks. Average maps were created including all the subjects in the study. To investigate the load-dependent differences in brain activity especially within the networks of interest, we used peak coordinates of the selected IC spatial maps in order to create a set of spherical ROIs of 6 mm radius and extracted the percentage signal change for each contrast.
Between-group differences on all the quantitative measures of percentage signal change were assessed using one-way ANOVA implemented in PASW. The significance level was set at p < 0.05 (two-tailed).
As with resting-fMRI data, a dual-regression approach was applied to the preprocessed task-fMRI data. The spatial maps of the DMN and the right- and left-frontoparietal networks were used to regress task-fMRI data and to obtain individual patterns of these networks during task-performance. These maps were introduced in a voxel-wise group comparison with 5000 permutations.
## Results
Low-HE and high-HE groups differed in task-performance, but there were no significant differences in the performance of the 3-back task between YA and high-HE subjects. Mean RT was higher in elders than in young subjects for all the conditions, but there were no differences between high-HE and low-HE. Across the two groups of elders, age, gender, and MMSE were comparable, but high-HE had significantly higher education levels (Table ; Figure ).
Group demographics and behavioral results on the n -back task .
MMSE, mini-mental state examination; d ;′, sensitivity index for task-performance; F , result of the analysis of variance (ANOVA) between the three groups; Sig., significance; RT, reaction time; YA, young adults; High-HE, high-performing healthy elders; Low-HE, low-performing healthy elders. Education levels were quantified in a scale from 1 to 4, with: 1, no education; 2, primary school; 3, secondary school, and 4, university studies. All scores are given as mean(SD). Significance levels are given in p values and considered significant when p < 0.05 .
Results of the task-performance during the different working memory loads . (A) Mean and SD values of d prime index, and (B) Mean and SD values of average response time (RT), in seconds. *Indicates p < 0.05 in the ANOVA post hoc analysis of between-group differences. 0B, 0-back; 1B, 1-back, 2B, 2-back, 3B, 3-back, YA, young adults; high-HE, high-performing healthy elders; low-HE, low-performing healthy elders.
The main findings, which are reported in the following sections, are summarized in Table .
Summary of findings .
### Resting-state fMRI analysis
Spatial maps derived from the whole-sample ICA decomposition of resting-state fMRI data corresponding to the DMN, the right-FPN, and the left-FPN are shown in Figure . The component identified as the DMN (Figure A) comprised areas in the frontal pole, middle frontal gyrus, and paracingulate gyrus (BA9, 10), the precuneus and posterior cingulate gyrus (BA7, 18, 23, 30, 31), and bilaterally in the superior occipital and posterior parietal cortices (BA19, 39). The right- and left-lateralized FPN (Figures B,C) involved areas in the middle and inferior frontal cortices (BA8, 9, 10, and 46), the paracingulate and anterior cingulate (BA6, BA8, BA32) and right and left parietal lobes, including the supramarginal and angular gyri. Although the lateralized pattern differed between right- and left-FPN, there was a broad overlap between these two networks.
Identification of the functional networks of interest from the ICA analysis of resting-state fMRI . Spatial maps of the three selected networks. (A) DMN network, (B) right-FPN, and (C) left-FPN.
With the dual-regression approach, we found differences in these three networks during the resting-state. As regards the DMN, low-HE exhibited decreased connectivity during the resting-state in frontal areas compared with YA and high-HE groups (Figure A). In the right-FPN (Figure B), the high-HE group had lower resting-state connectivity than both YA and low-HE groups. Finally, in the left-FPN (Figure C), low-HE, and high-HE had decreased resting-state connectivity in frontal areas. In the low-HE group, this decreased connectivity was observed in the left inferior and middle frontal gyri, left pars opercularis and left frontal pole (BA10, 44, 45, and 46). High-HE subjects showed the same pattern of decreased connectivity, but it was bilateral and extended to the anterior part of the superior temporal gyri, also including the anterior cingulate (BA6) and the insular cortex.
Results of the dual-regression analysis of resting-state fMRI . Maps show voxel-wise group-comparisons thresholded at a FWE corrected significance level of p < 0.05. (A) Group differences in the DMN, (B) Group differences in the right-FPN, and (C) group differences in the left-FPN.
### Analysis of task-related brain activity
We focused the analysis of task-fMRI data on a set of spherical ROIs that were selected from the networks identified in the resting-fMRI analysis (Figure ). Three ROIs were created from the right-FPN: one in the right inferior frontal gyrus: right-IFG ROI (MNI coordinates: x = 42, y = 54, z = −4), one in the right middle frontal gyrus: right-MFG ROI (MNI coordinates: x = 46, y = 34, z = 32) and one in the right superior parietal gyrus: right-PAR ROI (MNI coordinates: x = 42, y = −58, z = 52). Three were created from the left-FPN: one in the left inferior frontal gyrus: left-IFG ROI (MNI coordinates: x = −46, y = 50, z = 0), one in the left middle frontal gyrus: left-MFG ROI (MNI coordinates: x = −46, y = 34, z = 20), and one in the left superior parietal gyrus: left-PAR ROI ( x = −46, y = −50, z = 31). We also selected a region in the anterior cingulate cortex that was common for the right-FPN and the left-FPN: ACC ROI (MNI coordinates: x = −2, y = 26, z = 44). As regards the DMN, we defined four ROIs, one in the precuneus and posterior cingulate cortex: PCC ROI (MNI coordinates: x = 2, y = −66, z = 40), two in the left and right lateral occipital cortices: LLO ROI (MNI coordinates: x = −38, y = −82, z = 32), and RLO ROI (MNI coordinates: x = 42, y = −74, z = 36) and one in the middle frontal cortex: MFC ROI (MNI coordinates: x = 2, y = 58, z = −8).
Results of the ROI-based analysis of brain activity during the performance of the n -back task . Within each defined ROI, percent signal change are plotted for each group and condition. *Indicates that differences have a significance value of p < 0.05.
As shown in Figure , ROIs inside the task-positive networks showed, in general, a positive percentage of signal change during cognitively demanding blocks and negative values during fixation blocks, and ROIs within the DMN had the opposite behavior.
BOLD responsivity scores for the high-HE group were significantly higher ( p < 0.05) than those for the YA group in the ACC , left-IFG , and right-IFG ROIs, for the 1-back and the 2-back conditions, and only in the right-IFG for the 3-back condition. No differences were observed in these regions between the low-HE and the YA groups. Moreover, in the right-IFG ROI, responsivity was also greater in the high-HE group than in the low-HE during 2-back blocks. During fixation, the high-HE group also showed greater deactivation of the right-IFG compared to YA.
No significant group-effects were found in task-related BOLD response in left-MFG , right-MFG , left-PAR , and right-PAR ROIs, during any of the cognitive blocks. During fixation, both high-HE and low-HE groups showed increased deactivation of the right-PAR ROI compared to YA.
Within the MFC ROI, only the YA group showed a clear pattern of deactivation (negative percentage signal change) during cognitive blocks that was not observable in the other two groups. High-HE had significant differences in BOLD response of this region with respect to YA in 2-back condition. Overall, in the MFC ROI we observed positive percentage signal changes associated with cognitive demands in high-HE whereas these values were always negative for the YA group.
In the PCC ROI, high-HE also showed increased task-related activation during cognitive blocks than YA. Although the described effect could be seen at all the load levels of the task, this difference was statistically significant only in 1-back blocks. In Figure , we see that the PCC was moderately activated in the YA group when performing the levels of the task with the highest demand.
Finally, no group differences were found regarding BOLD response in left and right lateral occipital ROIs within the DMN.
### Network analysis on task-fMRI data
We used the dual-regression approach to explore differences in the selected networks during task-fMRI. High-HE showed decreased connectivity of the DMN with respect to YA (Figure A). However, in the right-FPN the same subjects had increased connectivity with respect to YA in several regions (Figure B), including the frontal pole, precentral gyrus, supplementary motor areas, anterior cingulate and paracingulate, insular cortex, and frontal orbital areas (BA6, 9, and 10). Finally, we found no differences in the connectivity of the left-FPN during task-fMRI.
Results of the dual-regression analysis of task-related fMRI . Maps show voxel-wise group-comparisons thresholded at a FWE corrected significance level of p < 0.05. (A) Group differences in the DMN, and (B) Group differences in the right-FPN.
## Discussion
Using ICA we identified intrinsic functional connectivity networks that are in operation during resting-state fMRI and during task fMRI, supporting the idea that brain connectivity has a network-based functional substrate that is not limited to the fact that the brain is functionally active (Smith et al., ). Although the ICA decomposition allowed the identification of other common resting-state brain networks, we only studied the DMN and the networks in the frontoparietal system due to their involvement in the working memory task that we used (similar to the approach considered in Leech et al., ). The three networks were identified in both resting- and task-fMRI, and functional reorganizations were found in both conditions, but in different directions. During the resting-state, elders with lower task-performance showed the largest differences with respect to YA within DMN connectivity, and those with higher task-performance showed reduced connectivity of the frontoparietal system. However, during task-fMRI, high-HE showed decreased connectivity of the DMN and increased connectivity of the FPN, but no differences were found between low-HE and young individuals. The analysis of task-related brain activity helped the interpretation of the results obtained by the ICA and the dual-regression approach, revealing task-related overactivations in frontal areas of the FPN, and the involvement of some DMN areas during task-performance in high-HE. Overall, our data show an age-by-performance modulation of brain networks that depends on external task demands.
### Compensatory role of DMN areas
Our data suggest evidence of reorganizations in the DMN connectivity in elders when compared to the group of YA. The pattern of functional reorganizations varied according to whether subjects are performing a cognitively demanding task. We found decreases in DMN connectivity that are in agreement with previously published work reporting age-related reduced DMN connectivity at rest (Damoiseaux et al., ; Littow et al., ; Wu et al., ; Tomasi and Volkow, ), and during task-fMRI (Sambataro et al., , see Hafkemeijer et al., for a review of DMN and aging). Interestingly, we observed that disruptions in DMN connectivity, when studied at rest, were related with poor cognitive performance in the working memory domain. Studying the DMN connectivity in a group of aged subjects, Andrews-Hanna et al. ( ) also found a relationship between anterior-posterior connectivity and cognitive performance in several cognitive domains. Moreover, during task-fMRI, only the high-HE group showed a reduction in DMN connectivity compared to YA. This latter result suggests dynamic modulations/interactions within networks during the performance of a task: old adults who performed the task successfully needed to recruit/engage additional brain resources which are typically not related to the WM task. Other studies using task-fMRI (Grady et al., , ; Filippini et al., ) reported age-related increases in activity within DMN regions, such as the mPFC, which are not traditionally implicated in task-performance. They also found that these activity increases were accompanied with decreases in task-related functional connectivity in the same areas (Grady et al., ).
The results we obtained with ICA and dual-regression were supported and extended by those of brain responsivity in DMN regions. We found differences in brain activity (measured as percent signal change) in two core regions of the DMN, the PCC, and the MFC. High-HE subjects recruited these areas during task-performance, but YA and low-HE groups did not. The recruitment of the anterior/frontal node of the DMN was specific for the high-HE group (Figure ), a finding that supports the utilization of non-task-related resources as a compensation mechanism in the aged brain (Cabeza et al., ; Mattay et al., ). The fact that this area is located in the frontal node may support the notion of the Posterior-anterior shift in aging (PASA model, see Davis et al., ). As regards the recruitment of the PCC node, this effect was also observable in the YA cohort at high levels of working memory load, in agreement with other studies showing involvement of the precuneus in cognitive control (Leech et al., ). High-performing elders recruited this area at the lowest memory load. This result can be interpreted in the light of the compensation-related utilization of neural circuits hypothesis (CRUNCH) which posits that older adults need to recruit additional neural resources at lower loads than younger adults (Reuter-Lorenz and Cappell, ; Schneider-Garces et al., ).
In summary, age-related DMN disruptions have been discussed either as a compensatory mechanism (Grady et al., ; Filippini et al., ) or as a functional marker of deficits in cognitive control that lead to poorer performance in elder subjects (Persson et al., ; Sambataro et al., ; Hedden et al., ). Our results clearly support the first idea. However it is also plausible that both mechanisms of compensation and dysfunction associated with cognition coexist in the aging brain. Finally, as our elder groups differed in educational attainment, the present results may also be interpreted within the context of neural compensation, as posited by Stern ( ).
### Changes in frontoparietal networks
Focusing on the brain networks responsible for the working memory system, we found functional reorganizations in terms of altered connectivity and greater BOLD response in the high-performing elders. During the resting-state, high-HE showed decreased connectivity of the frontoparietal system. To date, few studies have reported age-related changes in resting-state networks other than the DMN. Filippini et al. ( ) found age-related increases in the executive network at rest but they did not consider the cognitive performance of the subjects. However, Littow et al. ( ) found age-related decreases in some resting-state networks related to executive control.
During task-fMRI, the ROI-based analysis of responsivity showed increased activity in frontal regions, mainly in the inferior frontal gyrus bilaterally and in the anterior cingulate cortex. The same behavior was observed in the left middle frontal gyrus, but did not reach the level of significance established. Although the greater effects were observed in areas of the right-FPN, we interpreted this result as a reduction of asymmetry in task-related networks. The results of the dual-regression on task-fMRI also showed that the spatial pattern of the right-FPN becomes almost bilateral in high-HE subjects when compared to YA. Grady et al. ( ) also found that a greater expression of a network comprising right dorsolateral prefrontal areas predicted better performance in old adults. In some studies, the contralateral PFC activation was interpreted as a result of the difficulty of recruiting specialized neural mechanisms (the dedifferentiation hypothesis, see Persson et al., ; Eyler et al., ); however, our results add evidence for the compensation hypothesis, and more specifically for the hemispheric asymmetry reduction in older adults (HAROLD, Cabeza et al., ) pattern, since this effect was specific to the elders who performed well on the task. In addition, studies with other techniques such as TMS have also supported the HAROLD model and its relationship with successful aging in episodic memory performance (Solé-Padullés et al., ; Manenti et al., ).
Compensation mechanisms in terms of increased task-related BOLD activity have also been described in the pathologic brain in Alzheimer’s disease and mild cognitive impairment (Bokde et al., ). However, these patterns usually reflect activation of additional brain areas due to inefficient functioning of a network that might be compromised by disease; what is more, they have been studied in the context of keeping the same performance level as their age-matched controls, rather than in the context of increased performance as we found in healthy adults.
The interpretation of our results as the specialization of a task-related network also provides an insight into the study of intervention methods that include cognitive training in healthy aging. Although the neural mechanisms of training effects are still unknown, it has been commonly reported that higher BOLD activity is associated with better performance or higher improvement in task-performance (for a review see Klingberg, ). Thus, the study of inter-individual differences in the relationship between brain activity and behavioral outcome may be a key point in the design of effective training programs for patients whose limited cognitive capacities are restricting their daily lives. Moreover, further studies should determine whether these brain-behavior associations are limited to the cognitive task that is being performed in the scanner or whether they can be extrapolated to other cognitive domains. The latter question can be partially answered by examining our results of the resting-state analysis, since we already found brain reorganizations in the resting-state networks in high-performing old subjects.
## Conclusion
Using analysis of functional data during resting-state and during the performance of an n -back task, we provide evidence that the precepts of the principal neurocognitive theories of aging can be accommodated. First, functional compensation mechanisms were found: older people with successful working memory performance utilize different brain regions during cognitive activity than young people, and some of these regions are recruited from the DMN, a brain network that is typically deactivated during working memory performance. The recruitment of the frontal DMN regions was specific to elders with high-performance levels; however, recruitment of the precuneus was also observed in young subjects at high levels of working memory load (supporting the CRUNCH hypothesis, Reuter-Lorenz and Cappell, ). Moreover, within task-related networks, high-performing elders showed both increased connectivity and increased BOLD response bilaterally in frontal regions, supporting the HAROLD model (Cabeza et al., ) as well as the PASA model (Davis et al., ). Moreover, these dynamic network reorganizations were different at rest, when high-performing elders had less disruption of the DMN but greater disruption within the frontoparietal system than low-performing elders. We therefore suggest that successful aging is characterized by a level of brain plasticity that may mediate the efficient recruitment of functional resources in task-relevant areas when the subject is exposed to a task with a high cognitive demand even though this recruitment is not observable at rest. It has been proposed that there is an optimal level of brain plasticity during the age span that varies across subjects and allows this adaptation to a changing environment. Thus, both hypo- and hyperplastic mechanisms may set the stage for dementia or age-related declines in cognitive abilities (Pascual-Leone et al., ).
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Recent formulations of attention—in terms of predictive coding—associate attentional gain with the expected precision of sensory information. Formal models of the Posner paradigm suggest that validity effects can be explained in a principled (Bayes optimal) fashion in terms of a cue-dependent setting of precision or gain on the sensory channels reporting anticipated target locations, which is updated selectively by invalid targets. This normative model is equipped with a biologically plausible process theory in the form of predictive coding, where precision is encoded by the gain of superficial pyramidal cells reporting prediction error. We used dynamic causal modeling to assess the evidence in magnetoencephalographic responses for cue-dependent and top-down updating of superficial pyramidal cell gain. Bayesian model comparison suggested that it is almost certain that differences in superficial pyramidal cells gain—and its top-down modulation—contribute to observed responses; and we could be more than 80% certain that anticipatory effects on post-synaptic gain are limited to visual (extrastriate) sources. These empirical results speak to the role of attention in optimizing perceptual inference and its formulation in terms of predictive coding.
## Introduction
Several years ago, we suggested that attention can be understood as the selection of processing channels that conveyed precise or salient information within the framework of predictive coding (Feldman and Friston, ). The idea is that both the content of visual information and the confidence placed in that information have to be inferred during perception. In predictive coding, top-down predictions of the content are confirmed or disconfirmed by comparison with bottom-up sensory information (Rao and Ballard, ; Friston, ). However, this comparison rests on estimating the reliability or precision of sensory information—or more exactly the residuals or prediction error that cannot be explained. This precision may be itself context sensitive and has to be updated in exactly the same way as predictions of content (Brown and Friston, , ). This leads to view of hierarchical perceptual synthesis in which particular processing channels are selected on the basis of cues that portend spatial locations or featural attributes that are likely to convey precise information. In neuronally plausible implementations of this hierarchical Bayesian inference—namely, generalized Bayesian filtering or predictive coding—expected precision is thought to be encoded by the post-synaptic sensitivity or gain of cells reporting prediction error (Friston and Kiebel, ). Given that prediction error is passed forward from sensory cortex to higher cortical areas by ascending or forward connections, the most likely candidates for reporting prediction error are the superficial pyramidal cells that are the source of ascending connections (Bastos et al., ). This means that one can understand attention as the top-down gain control of superficial pyramidal cells passing information that is yet to be explained (i.e., prediction error) deep into the visual hierarchy.
This normative model and its neuronal implementation have been used to simulate and reproduce both the psychophysical and electrophysiological characteristics of the Posner paradigm (Feldman and Friston, ). In brief, predictive cues engage top-down predictions of increased precision in the left or right hemifield that facilitate the rapid processing of (inference about) valid visual targets. However, when an invalid target is presented in the wrong hemifield, the evidence accumulation implicit in predictive coding is slower, because gain or precision acts as a synaptic rate constant. This leads to protracted reaction times and an invalidity cost. Simultaneously, the scheme infers that prior beliefs about the target have been violated and prediction errors drive higher levels to update both the deployment of attention (i.e., precision) and target predictions per se . This explains the classic electrophysiological correlates of the validity effects in the Posner paradigm—in which invalid targets elicit slightly attenuated P1, N1 and N2 early components and a more pronounced P3b late component (Mangun and Hillyard, ; Hugdahl and Nordby, ; Talsma et al., ). These two electrophysiological characteristics may reflect the initial insensitivity (low precision or gain) of early visual responses and a subsequent post-hoc revision of top-down precision or gain control, when prediction error cannot be resolved by predictions based upon the (invalid) cue.
In this paper, we tried to verify these explanations for electromagnetic responses to valid and invalid targets in the Posner paradigm using magnetoencephalography (MEG) and dynamic causal modeling of differences in effective connectivity. In particular, we hoped to establish that a sufficient explanation for responses evoked by valid and invalid targets would be provided by a difference in the gain or post-synaptic sensitivity of superficial parietal cells following a cue—and a subsequent top-down modulation of this gain from parietal and higher extrastriate sources. To do this, we needed to use dynamic causal models based on canonical microcircuits that distinguish between superficial and deep pyramidal cells (Bastos et al., )—and that explicitly include a top-down modulation of superficial pyramidal cells.
In what follows, we provide a brief description of the dynamic causal models used to address precision or gain control in predictive coding; describe the data and experimental design; and report the results of Bayesian model comparisons that quantify the evidence for condition-specific differences in superficial pyramidal cell gain. Our focus here is on cue-dependent differences in gain prior to the onset of a visual target and subsequent top-down modulation of that gain during target processing. In particular, we asked whether cue-dependent differences in gain, top-down modulation or both were evident in evoked electromagnetic responses—and, whether any differences in gain were restricted to visual sources or extended to the parietal cortex.
## Materials and methods
### Dynamic causal modeling of predictive coding
In predictive coding models of inference in the brain (Mumford, ; Friston, ; Bastos et al., ), prediction error ascends to update representations at higher hierarchical levels. See Figure for a schematic summary. Crucially, the excitability of cells reporting prediction error corresponds (mathematically) to the precision of—or confidence in—the information they convey. This precision has been used to explain the psychophysical and electrophysiological correlates of attention and can be regarded as the basis of selective (predictive or attentional) gain—in which sensory processing channels that convey precise information are enabled.
Schematic detailing the neuronal architecture that might implement generalized predictive coding . This shows the speculative cells of origin of forward driving connections that convey prediction error from a lower area to a higher area and the backward connections that construct predictions (Mumford, ; Friston et al., ). These predictions try to explain away prediction error in lower levels. In this scheme, the sources of forward and backward connections are superficial and deep pyramidal cells respectively. The equations represent a gradient descent on free-energy under a hierarchical dynamic model (see Feldman and Friston, ). State-units are in black and error-units in red. Here, neuronal populations are deployed hierarchically within three cortical areas (or macro-columns). Subscripts denote derivatives.
Neurobiological implementations of predictive coding use superficial pyramidal cells to report precision-weighted prediction error: , where corresponds to representations (posterior expectations) of states of the world at level i in a cortical hierarchy and corresponds to the top-down predictions of these expectations—based upon expectations in the level above. The precision of the ensuing prediction error is modulated by the precision Π to weight prediction errors in proportion to their (expected) reliability (c.f., known uncertainty). From our point of view, the encoding of precision—at each level of the hierarchy—can be associated with the strength of inhibitory recurrent connections by noting that the expression for prediction errors is the solution to the following equation describing neuronal dynamics.
A more complete exposition of these dynamics can be found in Friston ( ). In this equation, γ is the negative log precision.
With Dynamic Causal Modeling (Garrido et al., ; Bastos et al., ), we map this neurobiological implementation of predictive coding onto a neural mass model which is capable of simulating MEG data. The depolarization of the three excitatory cell populations in the model—superficial and deep pyramidal cells, as well as spiny stellate cells, forms the output of the model with the main contribution coming from superficial pyramidal cells. This activity is transformed by an MEG-specific lead-field which describes the translation from source activity to sensor perturbation.
The four-population neural mass model used here has been described before (Brown and Friston, ). In the neural mass models, γ , the negative log precision, corresponds to the strength of recurrent inhibitory connections on superficial pyramidal cells. This means that as preclon increases, the strength of recurrent inhibition decreases. We therefore use the strength of intrinsic (recurrent) self-inhibition (on superficial pyramidal cells) as a proxy for log precision.
One new feature is introduced in this implementation of the neural mass model. To model top-down modulation of this self-inhibition we use the following form of (backward) modulatory connectivity:
Here, γ is self-inhibition when firing rates are at baseline levels σ = σ(0). Firing rates σ( V ) ∈ [0, 1] are a sigmoid function of depolarization V ∈ ℝ of afferent neuronal populations (deep pyramidal cells in other sources). The modulatory connection strength matrix M weights the influence of other sources; such that a high value suppresses self-inhibition and (effectively) increases the gain or precision of the superficial pyramidal cells that are targeted.In what follows, we will model condition (valid or invalid) specific effects on γ to evaluate the evidence for cue-dependent changes in gain at the onset of target processing and test for condition specific changes in M that mediate target-dependent changes in gain as target is processed. Our hope was that we will find evidence for differences in baseline gain and subsequent top-down modulation—and that these would be expressed predominantly in early visual sources.
Specifically, we anticipated that intrinsic self-inhibition would be lower (gain would be higher) in left hemisphere sources after (invalid) cueing of the right hemifield relative to (valid) cueing of the left hemifield, where the target appeared in the left hemifield in both conditions. In other words, we hoped to show differential responses to identical targets could be explained by differences in gain induced by valid and invalid cues. Furthermore, we anticipated differences in descending modulatory effects between valid and invalid trials that would be necessary to reverse the laterality of gain control following an invalid target.
### Participants
Fourteen healthy right-handed subjects participated in the study (8 male; age 20–54). Ethical approval was obtained from the UCL Research Ethics Committee (no. 2715/001). Written informed consent was obtained from all subjects.
### Experimental paradigm
All stimuli were presented using Matlab 7.1 and Cogent ( ). Stimuli were projected onto a screen 70 cm from the subjects. During the task, subjects fixated on a central cross at all times. At the start of each trial, the cross was replaced by an arrow pointing to the bottom left or bottom right corner of the screen, or a double-headed arrow pointing to both (neutral trials). The cues subtended 1.6 degrees of visual angle. After a cue-target interval of 50, 100, 200, or 400 ms, a target appeared either where the arrow had indicated (valid) or at the other side (invalid). The target was a white circle subtending 3.1 degrees of visual angle and presented in the lower left or lower right corners of the screen at 14.7 degrees eccentricity. Participants pressed a button with their right hand as soon as the target appeared. 66% of trials were valid, 17% were invalid and 17% uninformative (neutral cue trials are not considered here). Left and right cues and targets were balanced. Catch trials, in which no target followed the cue, were randomly presented before 10% of trials. 1800 trials were collected over three sessions on two consecutive days.
### Behavioral data
Reaction times were collected by Cogent and analyzed with IBM SPSS 20. A full factorial univariate ANOVA was performed with fixed factors “side” “validity” and “cue-target interval” and random factor “subject.”
### Data collection and processing
MEG data was obtained using a whole-head 275-channel axial gradiometer MEG system (CTF Systems). The sampling rate was 600 Hz and a low-pass filter of 150 Hz was applied. Head position was monitored using three localization coils, placed on the nasion and in front of each ear. An infrared eyetracker (Eyelink 1000) was used to monitor participants' fixation as well as to detect blinks. Stimuli were presented and behavioral data were collected with Cogent.
Data were analyzed using SPM12b for EEG/MEG. Data were down-sampled to 200 Hz and bandpass-filtered between 2 Hz and 32 Hz. Baseline-corrected epochs were extracted from the time series starting at 50 ms before target onset and ending 400 ms after target onset. Trials where the eyetracker detected a blink or saccade were excluded from analysis. Trials were then robustly averaged across cue-target intervals and participants to yield four conditions—left valid cue, right valid cue left invalid cue and right invalid cue. Averaging across participants can reduce the spatial precision of the MEG signal; however, as our hypotheses were not concerned with the spatial location of the signals we chose to combine data across all participants to increase the signal-to-noise ratio of the waveforms.
### Data feature and source specification
We addressed our hypothesis using condition-specific grand average responses over all subjects. Intuitively, this is like treating each subject as if they were the same subject to produce an average ERP. To identify plausible sources we used a distributed source reconstruction (using four grand averages: valid right target, invalid right target, valid left target, and invalid left target) based on multiple sparse priors (with default settings).
The grand average data were bandpass filtered between 2 and 32 Hz and windowed from 0–400 ms of peristimulus time. We used a lead field based upon the standard MRI template and a boundary element model as implemented in SPM12 (Mattout et al., ). After source reconstruction, we quantified the power of evoked responses (over all frequencies and peristimulus time) to produce the maximum intensity projections in Figure . As one would expect, left targets activate right early visual sources and vice versa . Note further, that early visual source responses to valid left targets are greater than the same targets under invalid cues. On the basis of these reconstructions, we identified eight sources corresponding (roughly) to key maxima of source activity. These sources included bilateral early visual sources (V2); bilateral sources near the occipitotemporal-parietal junction (V5); bilateral dorsal (V3) extrastriate sources and bilateral superior parietal sources (PC). The anatomical designation of these sources should not be taken too seriously—they are used largely an aide-memoire for sources at various levels in the visual hierarchy, so that we can discuss the functional anatomy. Clearly, the spatial precision of source localization does not allow us to associate each source with a specific cytoarchitectonic area—and even if we could, there is sufficient intersubject variability in cortical architectures to make this association, at best, heuristic.
Source specification for dynamic causal modeling . A distributed source reconstruction was performed (Mattout et al., ) and the power of evoked responses was quantified over the time course of the trial and all frequencies to yield the maximum intensity projections shown. Eight sources corresponding (roughly) to key maxima of source activity were identified: included bilateral early visual sources (V2); bilateral sources near the occipitotemporal-parietal junction (V5); bilateral dorsal (V3) extrastriate sources and bilateral superior parietal sources (PC).
The distributed network constituting the DCM is shown in Figure . The parietal sources sent backward connections to the extrastriate (V3 and V5) sources that then sent backward connections to the V2 sources. These connections were reciprocated by extrinsic forward connections to produce a simple visual hierarchy with bilateral connections.
The location of the eight sources is shown in the panels on the left . To construct the DCM, these sources were connected in the distributed network shown on the right. The parietal sources sent both driving and modulatory backward connections to the extrastriate (V3 and V5) sources that then sent backward connections to the V2 sources. These connections were reciprocated by extrinsic forward connections to produce a simple visual hierarchy with bilateral connections.
### Model space and Bayesian model comparison
The DCM analyses used data from 0 to 400 ms of peristimulus time. To de-noise the data and improve computational efficiency, we fitted the first eight canonical modes of the scalp data, given the source locations—these can be regarded as the principal components of the data that can be explained by source activity. The sources were modeled as small cortical patches of about 16 mm radius—centered on the source locations in Figure —as described in (Daunizeau et al., ). The vertices of these sources used the same lead fields as in the source reconstruction.
Exogenous (visual target related) input was modeled as a Gaussian function with a prior peak at 120 ms (and a prior standard deviation of 16 ms). This input was delivered to V2 on the appropriate side (left for right target trials and right for left target trials). The ensuing models were optimized to explain sensor responses by adjusting their (neuronal and lead field) parameters in the usual way—this is known as model inversion or fitting. The products of this inversion are posterior estimates of (differences in) intrinsic and extrinsic connectivity and the evidence or marginal likelihood for each model considered.
Our hypothesis centered on the gain of superficial pyramidal cells. We therefore estimated a full model in which all intrinsic gains and their extrinsic (backward) modulation could differ between valid and invalid trials. To ensure the same stimuli were used for assessing these differences we conducted two sets of analyses—one for targets presented to the left visual field and another for targets presented on the right. Each DCM estimated all intrinsic, extrinsic and modulatory connection strengths and any differences in intrinsic and modulatory connections due to invalid cuing.
After inverting the full model we then evaluated the evidence for reduced versions that constitute alternative hypothesizes or models. This model space was created by partitioning connectivity differences into three subsets and considering all eight combinations. These subsets were changes in intrinsic gain in the extrastriate sources (V2, V3, and V5); changes in parietal (PC) gain and changes in extrinsic modulatory connections. This partition was motivated by distinguishing between the effect of the cue on target-related responses—which should be apparent in changes in intrinsic gain in the visual areas—and the effect of the target per se —which should be apparent in changes in backward modulation of gain. To evaluate the ensuing models, we use Bayesian model comparison based upon (a variational free energy) approximation to log evidence. Having identified the model with the greatest evidence, we then examined its posterior parameter estimates. This allowed us to characterize validity effects quantitatively and to interpret them in computational (predictive coding) terms.
## Results
### Behavioral data
The ANOVA demonstratated significant main effects of validity, subject and cue-target interval, with significant interactions between cue-target interval∗validity, cue-target interval∗subject, side∗cue-target interval and validity∗side∗subject. Reaction times to validly cued targets were significantly shorter than to invalidly cued targets [left: mean (SD) 333 ms (42 ms) vs. 355 ms (44 ms), p < 0.001; right: mean (SD) 334 ms (42 ms) vs. 354 ms (44 ms)], Figure .
Reaction times to validly and invalidly cued targets at different cue-target intervals for targets appearing on the left (left panel) and right (right panel), averaged across all participants . Reaction times were faster for validly than invalidly cued targets ( p < 0.0001). Reaction times decreased as cue-target interval increase (all p < 0.05).
### Attentional effects in sensor space
The effects of attention (validity of cueing) on responses to targets presented in the left hemifield are shown—for the first two canonical modes—in Figure . Although these MEG responses are formally distinct from classic EEG results, they speak to similar effects on early and late responses: the blue lines correspond to valid trials and red lines to invalid trials. The response in the first mode shows the early response (just before 200 ms) has a reduced latency and slightly higher amplitude—consistent with an attenuation of N2 response to invalid targets, as seen in classic EEG studies (Mangun and Hillyard, ). In terms of late responses, the second mode shows a protracted and elevated response around 300 ms that is consistent with a P3b component, when the target location is not attended.
The solid lines report the model predictions of observed responses (broken lines) in sensor space after inversion of the DCM. These illustrate the accuracy of model inversion, capturing both the early and late differences to a considerable level of detail. Examples of the underlying source activity that generates these predictions are shown in the lower panel. These traces represent the depolarization of three excitatory populations within the left V2 source, contralateral to the visual input modeling the effects of target presentation. The dotted lines correspond to the spiny stellate and deep pyramidal populations, while the solid lines report the superficial pyramidal cells—that are the predominant contributors to sensor data. Note that this level of reconstructed neurophysiological detail rests on having a biologically plausible forward model.
Somewhat to our surprise, the differential responses to right targets were much less marked (results not shown). Furthermore, model inversion failed to converge for these conditions. Therefore, we restricted our analysis to the left target conditions. The failure to elicit clear validity effects with right targets may relate to the asymmetry of responses—and attentional gain control (see below).
### Bayesian model selection
A provisional Bayesian Model Comparison demonstrated that modeling the validity effect with changes in the strengths of the modulatory backwards connections only had the greatest posterior probability, justifying the investigation of these connections in the following analyses (Figure ). The comparison of different explanations for the validity effects above focused on differences in the gain of superficial pyramidal cells—either intrinsic to extrastriate or parietal sources, or differences in the modulation of gain, mediated by extrinsic top-down connections. The relative log evidences for all combinations of these condition-specific differences are shown in the upper left panel of Figure . The labeling of these models indicates the presence or absence of differences in extrastriate gain, parietal gain and gain modulation. It can be seen that the model with the greatest evidence includes differences in extrastriate gain and gain modulation—but not differences in parietal gain. The corresponding posterior probabilities of these models (assuming all were equally plausible a priori ) are shown in the upper right panel. These suggest that we cannot definitively exclude differences in parietal gain; however, we can be more than 80% confident that parietal effects are not necessary to explain these data, provided we allow for validity effects on extrastriate gain and its top-down modulation.
Results of provisional Bayesian model selection . The (free energy approximation) to log evidence was assessed for models with and without validity–dependent differences in top-down driving and modulatory connections. The log evidences (upper panel) show that the model with differences in modulatory connections has the greatest posterior probability (lower panel) . The log evidences are shown relative to the evidence for a null model with no changes in either driving or modulatory backward connections.
Upper panel: the first two of eight spatial modes (principle components) of the data to which the DCMs were fitted. Observed responses are dashed lines; solid lines show the responses fitted by the winning model (see below), demonstrating a good model fit. Lower panel: reconstructed source activity in left V2.
Upper left panel: relative log evidence for models which fitted differences between conditions through changes in one of three sets of parameters: superficial pyramidal cell gain in visual areas (1 _ _), superficial pyramidal cell gain in parietal areas (_ 1 _) and strength of backwards modulatory connections (_ _ 1). Upper right panel: The winning model had changes in superficial pyramidal cell gain in visual areas and in the strength of backwards modulatory connections, meaning that we can be more than 80% certain that backwards modulatory connections are not necessary explain the electrophysiological signatures of the validity effect. Lower panels show the same data as in the top left panel , but in image format.
The lower panels show the same log evidences but in image format, to illustrate the relative evidence for gain effects. The image on the right is under extrinsic top-down gain modulation and suggests greater evidence than the corresponding results on the left, where modulatory effects are concluded. In both cases, the model with extrastriate—but not parietal—gain differences has the greatest evidence. Having identified the best model, we then quantified the changes in model parameters that explain the validity effect.
### Attentional gain effects
Figure shows the differences in self-inhibition (top left panels) and backwards modulation of self-inhibition (top right panels) for the model with the highest posterior probability above. The upper panels show the differences as connectivity matrices indicating changes in connection strength. This means that differences in self-inhibition are located along the leading diagonal, while differences in backward connections are restricted to the upper diagonal elements. The middle panels show the same results but in terms of the posterior expectations for differences (in connections that changed) and their 90% confidence intervals.
Differences in self-inhibition (upper left panels) and backwards modulation of self-inhibition (upper right panels) between valid and invalid trials for the model with the highest posterior probability above . The lower panels show the gain of the superficial pyramidal cells over time in valid and invalid trials.
As anticipated, the recurrent or self-inhibition of early visual sources showed a highly asymmetrical difference when attending to the right hemifield (during invalid trials), compared to attending to the left hemifield (during valid trials). When attending to the right hemifield the left V2 source shows a profound decrease in the self-inhibition of superficial pyramidal cells—consistent with a disinhibition or increase in gain. This is accompanied by a slight decrease in the gain or sensitivity of the left extrastriate V3 source and an increase in the right V5 source. Note that these gain differences are in place before the target is presented and—presumably—are instantiated by the cue. When the target arrives, it evokes responses throughout the visual hierarchy that modulate the gain of the lower sources. These effects are mediated by the backward modulatory connections.
With the exception of backward connections from the right parietal source, all the differences in backward modulation between valid and invalid trials are positive, speaking to an increase in gain (or a top-down disinhibition of superficial pyramidal populations). However, it is difficult to predict the changes in gain that are produced by modulatory effects, because this disinhibition could itself be inhibited when top-down afference falls below baseline firing rates. Therefore, we evaluated the changes in gain in early visual sources as a function of peristimulus time for the two conditions. This is possible because we have a biologically plausible forward or generative model that allows us to examine changes in both neuronal states and connectivity—over peristimulus time—using the posterior parameter estimates.
Figure shows the log gain or precision of the early visual sources, following target presentation for valid (lower left panel) and invalid trials (lower right panel). As expected, there is a marked asymmetry in gain modulation during the prestimulus period that is revised or updated after the target is processed—through activity dependent modulatory mechanisms. Specifically, during valid trials the gain is greater in the appropriate (right) early visual source and then reaches a peak shortly before 200 ms. This peak is complemented by a suppression of gain in the unattended (left) visual source. This can be contrasted with the gain modulation during invalid trials. Here, the attended left source starts off with a slightly higher gain. Furthermore, the unattended source is suppressed more acutely with the arrival of the target. However, after about 120 ms its gain increases markedly, to peak just before 200 ms. This redeployment of precision (c.f., reorientation of attention) is the largest gain modulation in both sources and conditions. Interestingly, the gain of the left source also enjoys a slight increase but to a substantially lesser degree. In short, the top-down modulation of gain (through modulatory disinhibition of superficial pyramidal cells) appears to exert a dynamic gain control over peristimulus time and shows marked lateralization, when attention is switched from one hemifield to another.
## Discussion
In conclusion, we have used dynamic causal modeling to characterize putative changes in the gain of superficial pyramidal cell populations that might underlie attentional (validity) effects in the Posner paradigm. Our focus on gain mechanisms was motivated by theoretical formulations of attention in terms of optimizing perceptual inference using the expected precision of particular processing streams (Feldman and Friston, ). This formulation rests upon predictive coding schemes that the brain might use to infer the causes of sensory consequences it has to explain (Friston and Kiebel, ). Our model comparison and quantitative analysis of changes in parameter estimates are remarkably consistent with theoretical predictions.
In brief, the modeling results suggest that, following a cue, sensory channels in the appropriate hemisphere are afforded more precision through the disinhibition of recurrent or self-inhibition of superficial pyramidal cells. These cells are thought to pass sensory information (prediction error) to higher levels to inform perception. When a target appears in an unattended location, the misplaced gain or sensitivity of lower areas is revised or updated by top-down modulatory influences from higher extrastriate and parietal sources. Phenomenologically, this increases the latency and reduces the amplitude of early responses to invalid targets—because they are processed by channels that have an inappropriately low gain. The resulting prediction error induces an update response that reverses the misattribution of gain, producing differences in late or endogenous response components—such as the P3b. The P3b is known to be sensitive to probabilistic surprise (Mars et al., ; Kolossa et al., ) as well as to risk (Schuermann et al., ). These results suggest that the larger P300 in response to more unexpected events might be a result of exaggerated precision at lower levels incited by the arrival of an unexpected stimulus.
This application of dynamic causal modeling is slightly more focused than normal applications. We did not explore a large model space but focused on particular synaptic mechanisms as sufficient explanations for condition-specific responses. It is more than likely that there are many models of these differential responses that would produce equally good or better explanations. However, we chose to focus on models that were explicitly informed or constrained by computational and biophysical considerations; namely, that the effects have to be mediated by a neurobiologically plausible gain control that is consistent with normative principles of perceptual inference. This allowed us to validate the theoretical proposals empirically, while providing a principled model space within which to test specific hypotheses about the underlying wetware.
Evidence suggests that gain modulation in pyramidal cells is an important mechanism in visual attention. Electrophysiological studies have demonstrated that attention can enhance the response of visual neurons (likely to be pyramidal cells) by a multiplicative factor (McAdams and Maunsell, ; Treue and Martínez-Trujillo, ). fMRI studies demonstrate increased BOLD response for attended versus unattended stimuli (Kastner et al., ), even if these stimuli are predictable (Kok et al., ), and early visual ERPs, which are most strongly determined by pyramidal cell firing, are enhanced by attention (Rauss et al., ).
Interestingly, although we were almost forced to model gain control using inhibitory self connections—because of the relative simplicity of neuronal mass models used by dynamic causal modeling—this particular mechanism makes a lot of sense in relation to current thinking about attention. Convergent evidence implicates local inhibitory processing, mediated by GABAergic neurotransmission, in attention. Drugs working at GABA receptors, such as benzodiazepines, which are positive allosteric modulators of GABA-A receptors, increase the behavioral effect of cues so that reaction time differences to validly and invalidly cued targets become larger, while overall reaction times are slowed (Johnson et al., ). Nicotine (an agonist at nicotinic acetylcholine receptors) also affects reaction times in the Posner paradigm, but it decreases the validity effect while increasing reaction times (Thiel et al., ; Meinke et al., ), and it is believed that the attentional effects of acetylcholine might be mediated at least partly though depression of inhibitory interneuron activity (Xiang et al., ; Buia and Tiesinga, ). These contrasting effects suggest that the inhibitory interneurons set the gain of their cortical area to determine reaction times. Increasing their effects increases reaction times due to greater overall inhibition, exaggerating the difference between high- and low-gain cortical areas, and vice versa . This is consistent with the “biased activation theory” of selective attention (Grabenhorst and Rolls, ), which suggests that GABA interneurons mediate competition between stimuli which can be biased through top-down signals (the backwards modulatory connections in this DCM).
In summary, the emerging picture is that attention may be mediated through local intrinsic or recurrent inhibitory mechanisms that form a key part of cortical gain control—and that have characteristic signatures in terms of frequency specific induced responses. This fits comfortably with the theoretical perspective provided by predictive coding—that provides a computational role for recurrent inhibition in encoding the gain or precision of prediction errors in hierarchical processing. The results presented in this paper provide an initial link between these computational imperatives and plausible mechanisms at the level of synaptic processing and hierarchical neuronal circuits.
### Conflict of interest statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Aspects of human motor control can be inferred from the coordination of muscles during movement. For instance, by combining multimuscle electromyographic (EMG) recordings with human neuroanatomy, it is possible to estimate alpha-motoneuron (MN) pool activations along the spinal cord. It has previously been shown that the spinal motor output fluctuates with the body's center-of-mass motion, with bursts of activity around foot-strike and foot lift-off during walking. However, it is not known whether these MN bursts are generalizable to other ambulation tasks, nor is it clear if the spatial locus of the activity (along the rostrocaudal axis of the spinal cord) is fixed or variable. Here we sought to address these questions by investigating the spatiotemporal characteristics of the spinal motor output during various tasks: walking forward, backward, tiptoe and uphill. We reconstructed spinal maps from 26 leg muscle EMGs, including some intrinsic foot muscles. We discovered that the various walking tasks shared qualitative similarities in their temporal spinal activation profiles, exhibiting peaks around foot-strike and foot-lift. However, we also observed differences in the segmental level and intensity of spinal activations, particularly following foot-strike. For example, forward level-ground walking exhibited a mean motor output roughly 2 times lower than the other gaits. Finally, we found that the reconstruction of the spinal motor output from multimuscle EMG recordings was relatively insensitive to the subset of muscles analyzed. In summary, our results suggested temporal similarities, but spatial differences in the segmental spinal motor outputs during the step-to-step transitions of disparate walking behaviors.
## Introduction
Muscle activity during human locomotion is coordinated by tens of thousands of alpha-motoneurons (MNs), organized along the spinal cord (Romanes, ; Sharrard, ; Tomlinson and Irving, ). Bio-imaging techniques are being developed to increase our understanding of this spinal neural function, but generally these techniques (e.g., functional magnetic resonance imaging) remain difficult or impossible to use for studying the spinal cord during walking (Harel and Strittmatter, ; Stroman et al., ). Furthermore, most available techniques do not distinguish between activation of sensory neurons and that of MNs in the spinal cord. However, mapping muscle activations onto the rostrocaudal location of MN-pools in the human spinal cord provides a compact representation of the total motor output (Yakovenko et al., ; Ivanenko et al., ; O'Donovan et al., ; Monaco et al., ; Warp et al., ). This mapping also provides a complementary perspective to conventional approaches to understanding neural control, which often rely on detailed analyses of individual muscle activity and inter-muscular coordination (e.g., D'Avella and Bizzi, ; Ting, ; Giszter et al., ; D'Avella et al., ).
In previous studies, the spinal mapping method was used to investigate development and aging (Monaco et al., ; Ivanenko et al., ), as well as the relationship between the spatiotemporal organization of the spinal motor output and the biomechanics of human locomotion (Ivanenko et al., ; Cappellini et al., ; MacLellan et al., ). In particular, Cappellini et al. ( ) found that, during both forward and backward walking on level ground, the spatial activity of the spinal cord fluctuated with the center-of-body-mass (COM) motion, with bursts of activity around touchdown and foot lift-off. However, it is not known whether these bursts of activity around touchdown and toe-off are generalizable to other gaits nor is it clear if the spatial location of the activity (along the rostrocaudal axis of the spinal cord) is fixed or variable for different gaits. A better understanding of spinal motor outputs during different locomotion modes may provide further insights into adaptability and modularity of neural control (Lacquaniti et al., ; Bagnall and McLean, ), interspecies comparison (Carlson-Kuhta et al., ; Yakovenko et al., ), and may thus also have important clinical implications (Grasso et al., ; Scivoletto et al., ; Coscia et al., ; Oetgen and Peden, ; Hoogkamer et al., ).
Thus, the purpose of this study was to investigate these questions about motor output during several different locomotor tasks: forward, backward and digitigrade (tiptoe) walking on level ground, and walking on an inclined surface. These tasks may also be relevant to clinical, rehabilitation or sport applications. For instance, toe walking is observed in patients with various neurologic and developmental abnormalities (Oetgen and Peden, ), backward locomotion is used increasingly in sports and rehabilitation (Hoogkamer et al., ) and uphill walking may be appropriate exercise for obese individuals at risk for musculoskeletal pathology or pain (Haight et al., ). We used the recordings from 26 leg muscles (including intrinsic foot muscles that have not typically been considered) to reconstruct spinal motor outputs with specific interest in identifying common and idiosyncratic features across locomotor gaits.
## Materials and methods
### Experimental protocol
We recorded surface electromyograms (EMGs) and foot motion for 8 subjects (4 males, 4 females, 25.6 ± 2.6 years old, 1.78 ± 0.11 m, 76 ± 16 kg) during 4 ambulation tasks: walking forward, backward, tiptoe and uphill (20% inclined grade), all at 4 km/hr. These tasks were selected to represent biomechanically distinct walking gaits that were cyclic (for EMG analysis purposes) and could be performed at fixed speed on a treadmill. The treadmill speed was selected because it was sufficiently fast to distinguish myoelectric activity from the baseline noise (at slower speeds some muscle EMGs were small and therefore difficult to quantify), but also slow enough that most subjects could perform all the tasks. However, two out of the eight subjects were not able to walk backward at 4 km/hr. Each walking trial lasted 40 s and was performed barefoot on a standard treadmill. Prior to data collection, the subjects were trained on each task, allowing them time to acclimate to the various walking conditions, and all subjects gave informed consent prior to participation. The protocol was approved by the Ethics Committee of the Santa Lucia Institute.
We also collected several additional trials to help identify maximum contraction (MC) magnitude for each muscle EMG. Before collecting the walking data, we asked subjects to perform a set of quasi-static maneuvers against manual resistance. These included: flexing/extending/abducting the toes, plantarflexing/dorsiflexing/inverting/everting the ankle, flexing/extending the knee, flexing/extending/abducting/adducting the hip and flexing the back. Each exercise was performed for 5 s, during which subjects were instructed to perform maximal contractions. Two additional forward walking trials (3 and 5 km/hr) were also recorded and used to help determine MC values.
### Data collection
We recorded kinematic data bilaterally at 100 Hz using a Vicon-612 system (Vicon, Oxford, UK) with nine cameras. Infrared reflective markers (diameter 14 mm) were place bilaterally over the following landmarks: greater trochanter (GT), lateral femural epicondyle (LE), lateral malleolus (LM), heel (HE), and fifth metatarsophalangeal joint (5MP).
We recorded EMG activity by means of surface electrodes from 26 muscles simultaneously on the right side of each subject. These included one muscle from the lower back (erector spinae (ES) at L2 level), two muscles from the buttocks [gluteus maximus (Gmax) and gluteus medius (Gmed)], 11 muscles from the thigh [iliopsoas (Ilio), tensor fasciae latae (TFL), sartorius (Sart), adductor magnus (AddM), adductor longus (AddL), vastus medialis (Vmed), vastus lateralis (Vlat), rectus femoris (RF), biceps femoris long head (BFL), biceps femoris short head (BFS), semitendinosus (Semit)], six muscles from the shank [tibialis anterior (TA), peroneus longus (PerL), peroneus brevis (PerS), medial gastrocnemius (MG), lateral gastrocnemius (LG), soleus (Sol)], and six muscles from the foot [extensor hallucis longus (EHL), flexor digitorum/hallucis longus (FDHL), extensor hallucis brevis (EHB), extensor digitorum brevis (EDB), abductor digit minimi (AbdDM), flexor digitorum brevis (FDB)]. The activations of flexor digitorum longus and flexor hallucis longus were indistinguishable in our surface EMG recordings, due to close proximity of the muscles, and thus are reported together. We placed EMG electrodes based on suggestions from SENIAM (seniam.org), the European project on surface EMG. To this end, we located the muscle bellies by means of palpation and oriented the electrodes along the main direction of the fibers (Winter, ; Kendall et al., ). The placement of EMG electrode for muscles in the foot and shank segments is illustrated in Figure for convenience since some of the foot muscles are less commonly recorded in literature.
Sites of EMG electrode placement for muscles in the foot and shank segments. (A) Plantar surface of the foot and medial aspect of the leg with sites of electrode placement. (B) Dorsal surface of the foot and lateral aspect of the leg with sites of electrode placement.
All EMGs were recorded at 4000 Hz using a Delsys Trigno Wireless System (Boston, MA), except the flexor digitorum brevis which was recorded using a synchronized Delsys Bagnoli System (at 1000 Hz). Due to the recording site of flexor digitorum brevis (on the plantar surface of the foot), the lower-profile Bagnoli electrode was needed. Some electrodes became partially or fully detached during testing, and signals were thus not usable. These EMGs were removed on a subject-specific basis. On average we analyzed 23.4 ± 1.7 muscle EMG from each subject.
### Data processing
The beginning of the gait cycle (foot-strike) was defined based on kinematic events. We used vertical height of the right HE marker for forward walking and limb elevation angle (based on maximum GT-5MP virtual segment displacement) for backward, tiptoe and inclined walking. Similarly, for stance to swing transition (foot-lift) we used limb elevation angle (based on minimum GT-5MP virtual segment displacement) for forward, tiptoe and inclined walking, and minimum vertical height of the HE marker for backward walking. The usage of these criteria was based on the different kinematic endpoint (foot) behaviors for the various gaits (Ivanenko et al., ). While the differences in definition of gait cycle initiation may have introduced minor time shifts between tasks, all muscle EMGs for a single task shift together in time and so for each individual gait this did not impact the fidelity of spinal map reconstruction. General gait parameters [cycle duration, anterior-posterior foot (5MP) excursion] and joint (ankle, knee and hip) angular range of motion were calculated to characterize the kinematics of gaits studied (Figure ).
Kinematic patterns during forward, backward, tiptoe and uphill (20% inclined) walking at 4 km/hr. (A) Ensemble averages (±SD) of hip, knee, and ankle joint angles of the right leg. Hip and knee angles increase in flexion, ankle angle in dorsi-flexion. The dotted region between stance and swing phases depicts inter-subject standard deviation (SD), and is centered at average foot-lift. On the top : stick diagrams for a single stride in one representative subject. (B) Cycle duration (mean +SD) for different gaits. (C) Anterior-posterior foot (5MP marker) excursion. (D) Peak-to-peak amplitudes (+SD) of angular motion.
We processed EMG data using standard filtering and rectifying methods. We applied a 30 Hz high-pass filter, then rectified the EMG signals and applied a 10 Hz low-pass filter (all filters, zero-lag 4th order Butterworth). To reduce residual baseline noise, which appear as offsets in the EMG envelopes, we subtracted the minimum signal from each EMG. This assumes that at some point during walking each muscle is effectively “off” (not actively contracting). Some subjects exhibited artifacts in the foot muscles, generally linked to foot-strike and foot-lift events. In order to remove these artifacts, high-pass filtering of these muscles was performed using a 150 Hz cut-off frequency (rather than 30 Hz). A prior study on cut-off frequency (Potvin and Brown, ) and informal tests on locomotor EMGs (Zelik et al., ) confirmed that this artifact-removal filter had minimal effect on the shape of the muscle activation pattern. For illustrative purposes, the EMGs filtered at higher cut-off frequency were then rescaled to match peak amplitude of the 30 Hz filtered signal (Figures , ). However, this rescaling procedure did not affect the calculation of the motor output since EMGs were eventually normalized to their MC amplitudes before mapping to the spinal cord (detailed below).
EMGs. Subject-averaged patterns of muscle activity during forward, backward, tiptoe and uphill (20% inclined) walking at 4 km/hr . Mean EMGs and inter-subject standard deviations are plotted across a normalized gait cycle from foot-strike to ipsilateral foot-strike. The dotted region between stance and swing phases depicts inter-subject standard deviation (SD), and is centered at average foot-lift.
Peak EMG magnitudes. (A) Maximum contraction (MC) values (+SD) are shown (based on peak EMGs from dynamic and quasi-static trials; see Methods for full details). Peak EMG amplitudes (+SD) are depicted in μV (B) and as a percentage of MC (C) for forward, backward, tiptoe and inclined walking at 4 km/hr.
We divided EMGs into gait cycles based on foot kinematics, then interpolated each stride to 200 time points, and finally averaged across gait cycles (individually for each subject and task). This yielded an ( m × 200) EMG matrix for each task, where m equaled the number of muscles analyzed. Inter-subject mean (and standard deviation) values for EMG were then computed from these subject-specific data. In addition to calculating the ensemble-averaged EMGs (across strides and subjects), we also present some EMG waveforms of individual strides in order to examine inter-stride variability in the spinal motor output.
### EMG normalization
We normalized EMGs by the MC magnitude across all trials. Normalization was performed to account for the differences in μV magnitudes recorded between muscles. We defined MC magnitude as the muscle's maximum EMG signal from either dynamic (walking) or quasi-static trials (during which subjects were instructed to perform maximal contractions against manual resistance, see Experimental Protocol ). Thus, all EMGs were considered on a scale from 0 to 1, where 0 indicates that a muscle is inactive and 1 represents maximum muscle activation. Across all quasi-static trials (EMGs were low-pass filtered as described previously), we looked at a sliding 1-s window (by incrementally shifting each time step) and computed the average EMG during each. The highest average EMG found during any 1-s window was defined as the maximum quasi-static activation magnitude. Similarly, maximum dynamic activation magnitude was defined for each muscle as the peak stride-averaged EMG across all walking tasks. The normalization constant for each muscle was then defined as the larger of the quasi-static and dynamic activation magnitudes.
We note that normalization to muscle physiological cross sectional area (PCSA) was not used in this study for reconstructing the segmental spinal outputs, which has occasionally been done in the past (e.g., MacLellan et al., ; Ivanenko et al., ). This is because the number of motor units for each muscle is not related to PCSA in a simple way (e.g., number of motor units does not scale proportionally with size of muscle; Feinstein et al., ; Christensen, ; McComas et al., ; McComas, ).
### Motor output calculations
To characterize the spinal motor output, EMG-activity was mapped onto the estimated rostrocaudal location of MN-pools in the human spinal cord from L2 and S2 segments. Because this method has been thoroughly documented in previous papers (Ivanenko et al., , ; MacLellan et al., ), we describe it only briefly here. The maps were constructed by adding up the contributions of each muscle to the total activity at each spinal segment, using the myotomal charts of Kendall et al. ( ) to link muscles to their spinal innervation levels (see Figure ). The motor output pattern of each spinal segment S was estimated by the following equation:
where EMG represents the normalized, subject-specific envelope of muscle activity, k is a weighting coefficient for the i -th muscle (to signify if the j -th spinal level is a major, k = 1, or minor, k = 0.5, MN source, see Figure ), m is the number of muscles innervated by the j -th spinal segment, and n is the total number of spinal levels that innervate the i -th muscle, again accounting for major and minor sources (for instance, for the soleus muscle, n = 1 + 1 + 0.5 = 2.5, see Figure ). Thus, the fractional part of Equation 1 can range in value from 0 (inactive) to 1 (maximum activation of that spinal segment). To account for size differences in MN pools at each spinal level, this fractional activity value was then multiplied by the segment-specific number of MNs ( MN ). This MN pool size normalization primarily affects the boundary segments L2 and S2, which contain 2–3 times fewer MNs than the other segments (Table , Tomlinson and Irving, ). We note that Equation 1 is slightly modified with respect to our previous studies (Grasso et al., ; Ivanenko et al., , ) in order to better account for the different number of muscles that innervate each spinal segment and the heterogeneity in the MN pools along the lumbosacral enlargement. Thus, our updated calculation yields spinal motor output in units of number of (active) MNs.
The primary assumptions implicit in this analysis are that (1) the rectified EMG provides an indirect measure of the net firing rate of MNs for each muscle (Yakovenko et al., ), and (2) the set of recorded muscles is representative of the total motor output from each spinal segment. The first assumption seems reasonable given that mean EMG has been found to increase linearly with the net motor unit firing rate (Hoffer et al., ; Day and Hulliger, ). However, a limitation is that this method does not account for confounds due to other physiological properties, such as the effects of muscle length or velocity on the EMG signals. To test the second assumption, we compared the activation maps obtained from all 26 recorded muscles with those obtained from reduced subsets of muscles (detailed in Muscle subset analysis section below).
To obtain the averaged (across subjects) spinal maps, we calculated the spinal motor output for each subject based on stride-averaged EMGs, and then we averaged it across subjects. We computed two summary metrics to describe the spinal maps: mean segmental output and mean temporal output . For each condition, we averaged the motor output patterns over the entire gait cycle to find the subject-specific mean segmental output and then averaged it across subjects to obtain mean ± SD. Similarly, we averaged the motor output across the spinal segments L2 to S2 to find the mean temporal output across the gait cycle. From this mean temporal output waveform, we found the maximum peak in the first half of the stance phase and defined it as activation burst 1 , and the peak in the second half of the gait cycle as burst 2 . To characterize the total intensity of the spinal output for each task, we computed for each subject the mean motor output by averaging across both spatial segments and gait cycle, and then we averaged it across subjects. In addition to creating subject-specific spinal maps from stride-averaged EMG envelopes, we also computed maps for individual strides and compared them with those obtain from ensemble-averaged strides.
### Muscle subset analysis
Practical considerations limit the number of muscles from which we could record. Thus, there is the potential issue of how the specific selection of the muscles affects the resulting spatiotemporal maps of MN activity. To evaluate the sensitivity of the spinal maps approach we compared the motor outputs obtained from analyzing all 26 muscles with those obtained from subsets of these muscles. Subsets were chosen as follows: (1) the 20 non-foot muscles (TA, Sol, MG, LG, RF, Vmed, Vlat, AddL, AddM, ES, TFL, PerL, PerB, BFL, BFS, Semit, Sart, Ilio, Gmax, and Gmed) and (2) 12 commonly recorded muscles (TA, Sol, MG, LG, RF, Vmed, Vlat, ES, TFL, BFL, Semit, and Gmax). For forward walking we also made 26 additional comparisons by correlating maps from each unique set of 25 muscles (i.e., by systematically eliminating each individual muscle) with the map constructed from all muscles. The correlation coefficient (r) was calculated for each subject and condition. Averaged correlation coefficients were then reported for each comparison.
### Statistics
To compare activation waveforms we computed linear correlations ( r -values). For instance, to compare segmental activations of individual subjects with those of averaged maps, correlation coefficient was computed for each subject and each segment, and then they were averaged first across subjects for each segment and then across segments. Similarly, to compare the maps obtained by different sets of muscles, correlation coefficient was computed for each segment, and then the data for all segments were averaged. Since correlation coefficients have non-normal distributions, their mean estimates were computed based on the normally distributed, Z-transformed values.
Repeated measures (RM) ANOVA was used to evaluate differences in the kinematics and the mean motor output across different gaits, and post-hoc Tukey's HSD test was used to determine statistical significance. Since only six out of the eight subjects were able to walk backward at 4 km/hr their missing data for this condition for the ANOVA were replaced by the unweighted mean value estimated from all other subjects. Reported results are considered significant for p < 0.05.
## Results
### Kinematics
General gait parameters and ensemble-averaged joint angular movements are reported in Figure . We observed that cycle duration and foot excursion were slightly but significantly lower for backward walking than for forward and inclined walking ( p < 0.006, Figures ). These two parameters were also larger for inclined walking relative to tip-toe walking ( p < 0.03). The range of hip and ankle angular motion was significantly larger during inclined walking than for the other tasks ( p < 0.001, Figure ). The peak-to-peak amplitude of the knee joint oscillations was significantly smaller for backward and tip-toe walking than for forward and inclined walking ( p < 0.0002, Figure ).
### EMG
Lower-limb EMGs (Figure ) were qualitatively consistent with those reported elsewhere in the literature for forward (Winter, ; Ivanenko et al., ), backward (Thorstensson, ; Grasso et al., ; Ivanenko et al., ), tiptoe (Perry et al., ; Romkes and Brunner, ) and inclined walking (Lange et al., ; Franz and Kram, ). In this study we extended the number of recorded muscles relative to our previous studies (Ivanenko et al., , ). In particular, we included intrinsic foot muscles, which demonstrated their own unique activation patterns with bursts principally around the stance to swing transition of gait (Figure ).
Averaged EMG waveforms for the deeply located and interconnected muscles during forward walking were consistent with those reported in the literature. The deep hip flexors (Ilio) demonstrated the major peak of activity around lift-off (Rab, ; Andersson et al., ; Ivanenko et al., ). EMG recordings of AddL and AddM showed main bursts at foot lift-off and during swing, respectively (Winter, ). The activity of BFL and BFS (at the end of swing and beginning of stance) was similar to that reported by University of California Berkeley ( ). Intramuscular recordings of foot muscle activity (Gersten et al., ; Mann and Inman, ) showed a good correspondence with our data (Figure ). Specifically, EHL activity showed two peaks around foot lift-off and heel strike, respectively, while the FDHL showed activity beginning in early stance and continuing until the foot lift-off (Gersten et al., ). The EDB and AbdDM became active ~20% of the cycle and the FDB at 40% of the cycle, remaining active until just before foot lift-off (Mann and Inman, ).
The amplitude of EMG signals (in μV) varied considerably across muscles, both during walking and in terms of MC (Figures ). We found that normalizing to MC tended to increase the relative activation magnitude of proximal muscles (e.g., ES, Ilio, Gmax, Gmed, TFL, AddL, Sart, Vmed) and some intrinsic foot muscles (e.g., FDHL, AddDM) and thus their contribution to the spinal maps (Figure ).
### Average spinal maps
We observed task-specific spinal motor outputs for each walking condition (Figure ), although with qualitative similarities in temporal profile. In particular, two prominent periods of activity were observed in the mean temporal output of each task (Figure , bottom): the first following foot-strike (~5–10% of the gait cycle) and the second preceding foot-lift (~40–55%). However the timing of the second burst relative to foot-lift varied considerably between tasks, occurring later in tiptoe than forward walking, for example (Figure ). In contrast to the qualitative temporal similarities, we found substantial differences in the spatial localization and intensity of spinal activation for each gait (Figure ). In particular, we found that the mean motor output (spinal activation averaged across the entire gait cycle and all spinal segments) was significantly lower for forward walking than for the other tasks ( p < 0.01, Figure ). We also found that the loci of mean segmental outputs shifted somewhat as a function of gait (Figure , see plots to the right of each spinal map). For instance, forward and tiptoe walking exhibited principle activations in L5 and S1, whereas backward and inclined walking showed a more distributed output with roughly similar intensities from L3 to S1. These differences in spatial level of spinal activation were even more evident during the major “spots” of activity (identified as burst 1 and 2 from the mean temporal output). During burst 1 (after foot-strike, Figure ) peak motor output was at the spinal level L3 for inclined walking, L4 for forward walking, L5 for tiptoe walking and S1 for backwards walking (although this gait exhibited a relatively constant intensity from L3-S1). Differences were less evident for burst 2 (Figure ), when most gaits exhibited peak motor outputs from spinal segments L5 and S1.
Spinal maps . Depicted here are estimates of averaged (across subjects) spatiotemporal spinal motor outputs computed from EMGs for (A) forward, (B) backward, (C) tiptoe, and (D) uphill (20% inclined) walking, all at 4 km/hr. Motor output (reported in units of number of MNs) is plotted as a function of gait cycle and spinal segment level. Waveforms plotted below the maps correspond to the mean temporal output pattern averaged first across all 6 segments and then across subjects (mean ± SD, n = 8 subjects). Note the tendency for peaks to occur around early and late stance (labeled as burst 1 and 2). Curves to the right of maps represent the mean segmental output averaged first across the entire gait cycle and then across subjects. In the gait cycle, the dotted region between stance and swing phases depicts inter-subject standard deviation (centered at average foot-lift).
Spinal motor output. (A) Depicted are mean (+SD) motor outputs (averaged across both gait cycle and spinal levels). Asterisks denote significant differences between conditions. Segment-specific magnitudes of motor output are also shown for (B) burst 1 of spinal activity (occurring after foot-strike; see Figure ), and (C) burst 2 (occurring around foot-lift).
### Subject-specific spinal maps
The major features observed in the average spinal maps were also present in subject-specific maps. In particular, 6 out of the 8 tested subjects exhibited bimodal (two peaked) motor output profiles for all gaits (Figure ). The remaining 2 subjects also showed the bimodal temporal profile for most gaits except for forward walking (s8 subject, Figure ) and backward walking (s5 subject, Figure ). In these cases, mean temporal output was found to have an additional peak at the beginning of the swing phase. Individual subjects also exhibited small differences in the segmental level of spinal activation, particularly during load acceptance following foot-strike (Figures , ). Nevertheless we found a strong correlation between subject-specific and average maps (0.85 ± 0.13 depicted in Figure ), consistent with previously published findings (Ivanenko et al., ).
Spinal maps of MN activity of the lumbosacral enlargement in all subjects for all gaits (A–D) . Note similar temporal features (main peaks around foot-strike and foot-lift) of the segmental output between individual and averaged (Figure ) spinal maps.
EMG profiles exhibit stride-to-stride variability related to dynamic stability and walking speed maintenance (Hausdorff, ; Kang and Dingwell, ). Examples of the spinal maps of individual strides are illustrated in Figure . Despite individual variations in the segmental level of spinal activation, the major features depicted in the stride-averaged maps (Figure ) are representative of the general trends in individual strides (Figure ). We found the mean correlation coefficient between segmental output waveforms of individual and ensemble-averaged strides was 0.90 ± 0.04 (average from Table ).
Examples of spatiotemporal maps of MN activity of the lumbosacral enlargement in two subjects [s2 and s5, (A,B), respectively] for all gaits . For each individual, four individual strides are shown. Note similar spatiotemporal features (activity around foot-strike and foot-lift) of the segmental output between individual and averaged (Figure ) strides for these subjects.
Inter-stride variability in the segmental spinal output for different gaits .
For each subject, correlation coefficients between segmental motor outputs (based on 26 recorded muscles) of individual and ensemble-averaged strides were obtained and averaged across strides. Then, for each spinal segment, these correlation coefficients were averaged across subjects and reported in this table (mean ± SD) .
### Sensitivity to the number of muscles analyzed
We found that the spinal maps were relatively insensitive to the subset of muscles analyzed. Spinal maps computed from 20 and 12 muscle subsets were strongly correlated with the maps computed from the full set of 26 recorded muscles, with average correlation coefficients between 0.98–0.99 and 0.91–0.96, for each task (Table ). The motor outputs evaluated at each individual spinal segment were also found to be in good agreement, with r values (always greater than 0.9 using 20 muscles, and generally greater than 0.85 using 12 muscles). The only exception was that, with 12 muscles, the L5 segment correlation dropped to 0.74–0.90. Forward walking maps obtained by excluding a single recorded muscle were also highly correlated with those obtained from the full set of 26 muscles ( r = 0.99 ± 0.01).
Sensitivity of spinal maps to the muscle subset analyzed .
Correlation coefficients are reported for comparisons between motor outputs based on all 26 recorded muscles and those obtained from subsets of 20 or 12 muscles. For each gait, correlations are reported for individual spinal segments and for the total spinal map (mean ± SD) .
## Discussion
The overall behavior of the body and limbs during walking is determined by the interplay of neural and mechanical factors. Here we observed that spinal motor outputs corresponded to the major phases of biomechanical force production during diverse walking tasks (Winter, ; Perry et al., ; DeVita et al., ; Franz and Kram, ). Specifically, the elevated MN outputs during the gait cycle produce muscle contractions during the step-to-step transition, in which both limbs act to redirect the body's velocity in a way that is thought to improve walking economy (Donelan et al., ). However, during the step-to-step transition, we observed differences in the loci of the segmental spinal activity across gaits (Figures , ). This suggests that even if similar biomechanical functions are performed by the limbs (i.e., redirection of the body during the transition), it may be accomplished differently, through a gait-specific coordination of muscles. Thus, high-level features of locomotion may be flexibly encoded by neural circuits to generate muscle activation patterns based on gait-specific constraints and feedback.
Various neural control strategies have been proposed for transforming such task-level goals to muscle-level execution, for example using a hierarchical, modular architecture under feedback control (Ting et al., ). The pulse-like features of the spinal motor output observed in this study may be consistent with “drive-pulse” rhythmic elements or neural primitives, which have been hypothesized to underlie the spinal circuitry of animals (Giszter et al., ). Although the precise neuronal substrates remain largely unknown (but see Hart and Giszter, ), it is believed that a crucial role is played by central pattern generators (Grillner, ). Specifically, it has been proposed that motor activation patterns may emerge from a multi-layered organization of the spinal neural networks with two functionally distinct levels, one for rhythm generation and the other for muscle pattern generation (McCrea and Rybak, ). In this study we found that the spatial loci of MN pool activations depends greatly on the walking task (Figures , ), indicating that “drive pulse” rhythmic elements may be significantly modulated by task-specific sensory feedback. Since muscle activation timing was linked to major force production events around foot touchdown and foot-lift, it suggests that pre-programmed motoneuronal drive may be principally mediated by afferent force and kinematic-related feedback (Duysens et al., ; Nielsen and Sinkjaer, ; Pearson, ). There is also supporting evidence from a previous study on cats that neuromotor coordination may be modulated by critical points that correspond to key biomechanical events (Saltiel and Rossignol, ).
It is worth noting that we observed similarities in the spinal activation maps across subjects (Figure ) and strides (Figure , Table ), specifically in terms of temporal activation peaks around foot-strike and foot-lift. Thus the spinal mapping methodology seem to provide a robust and repeatable means to reconstruct MN pool activity. Meanwhile, previous literature has demonstrated that the spinal maps do vary for individuals with neuromotor impairments (Grasso et al., ; Coscia et al., ; Ivanenko et al., ) and throughout the aging process (Monaco et al., ) and during childhood development (Ivanenko et al., ). Taken together, the robustness of the methodology and the population-specific activations suggest that spinal mapping approach may be useful for assessing or differentiating gait performance in clinical populations.
It is also interesting to compare spinal maps between plantigrade and digitgrade gaits. Human adults typically walk with a characteristic heel-to-toe progression (plantigrade gait), whereas many animals walk only on their toes (digitgrade gait). In this study we observed roughly a doubling of the intensity of spinal motor output during tiptoe walking (Figure ), which is known to incur increased energetic costs compared to plantigrade gait (Cunningham et al., ). This increase in motor output was due, in part, to differences in the spinal activity after foot-strike, which was both increased in magnitude and spatially shifted toward more distal segments (L5/S1). The spinal maps for human tiptoe walking were, however, qualitatively different from maps constructed from digitigrade feline locomotion (Yakovenko et al., ). In cats, the primary MN activation during walking occurs during midstance and with roughly constant intensity, likely the result of neuromechanical differences associated with their flexed limb posture and quadrupedal gait. This comparison also highlights the potential utility of spinal maps for studying interspecies motor control.
There are several limitations to the spinal mapping approach, many of which have been previously documented (Cappellini et al., ). Briefly, the reconstruction and interpretation of spinal maps assume anatomical similarity of motor pools across individuals and that rectified EMG provides an indirect measure of the net MN firing rate (Yakovenko et al., ). Another potential concern is related to EMG cross-talk, which is always a potential issue with surface EMG recordings: in particular for deep muscles like Ilio that have a relativity small superficial region for recording and for smaller foot muscles (e.g., flexor digitorum longus) that are in close proximity to larger calf muscles. In the previous study (Ivanenko et al., ), the cross-talk issue was addressed by modeling the potential effect of different levels of cross-talk in the EMG profiles. The spinal segmental output was reconstructed by adding up incrementally the magnitude of cross-talk from adjacent muscles (from 10 to 100%). While the intensity and the width of the main loci of activation could be affected by adding cross-talk, this procedure did not give rise to the appearance of new loci of activation or significant time shifts in the spinal maps. Given the similar spinal mapping methodology in this study, we do not expect that the similarities in spinal maps reconstructions (based on different set of muscles EMGs) were due to cross-talk. Furthermore, the spinal maps during walking have been shown to be similar when reconstructed from EMGs obtained using surface and intramuscular electrodes (Ivanenko et al., ). Consistent maps have also been produced in different studies (Ivanenko et al., ; Cappellini et al., ; Monaco et al., ; Coscia et al., ; MacLellan et al., ). We only tested four walking tasks (Figure ), but other gaits may show additional (e.g., skipping) or temporally shifted (e.g., running) spots of activity specific for force production in those gaits (Ivanenko et al., ). Finally our analysis was also based on a limited set of muscles. However, we found the spinal maps to be relatively robust and insensitive to the subset of muscles analyzed (see Results, Table ), presumably because the lumbosacral enlargement innervates numerous muscles and each muscle is innervated by several segments.
In summary, we found that the MN activation patterns exhibited two major bursts during diverse walking tasks, one around foot-strike and the other around foot-lift, but with differences in the segmental level and intensity of the spinal activity. We also found further evidence that spinal MN mapping provides a robust method for estimating spinal motor output, which is relatively insensitive to the subset of EMGs analyzed. We also suggest that spinal motor mapping can be used to assess the recruitment of specific motor pools when using epidural electrical stimulation or corticospinal neuroprostheses for restoring locomotor functions (Capogrosso et al., ; Borton et al., ).
### Conflict of interest statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Comprehension of conventional and novel metaphors involves traditional language-related cortical regions as well as non-language related regions. While semantic processing is crucial for understanding metaphors, it is not sufficient. Recently the precuneus has been identified as a region that mediates complex and highly integrated tasks, including retrieval of episodic memory and mental imagery. Although the understanding of non-literal language is relatively easy for healthy individuals, people with schizophrenia exhibit deficits in this domain. The present study aims to examine whether people with schizophrenia differentially recruit the precuneus, extending to the superior parietal (SP) cortex (SPL), to support their deficit in metaphor comprehension. We also examine interregional associations between the precuneus/SPL and language-related brain regions. Twelve people with schizophrenia and twelve healthy controls were scanned while silently reading literal word pairs, conventional metaphors, and novel metaphors. People with schizophrenia showed reduced comprehension of both conventional and novel metaphors. Analysis of functional connectivity found that the correlations between activation in the left precuneus/SPL and activation in the left posterior superior temporal sulcus (PSTS) were significant for both literal word pairs and novel metaphors, and significant correlations were found between activation in the right precuneus/SPL and activation in the right PSTS for the three types of semantic relations. These results were found in the schizophrenia group alone. Furthermore, relative to controls, people with schizophrenia demonstrated increased activation in the right precuneus/SPL. Our results may suggest that individuals with schizophrenia use mental imagery to support comprehension of both literal and metaphoric language. In particular, our findings indicate over-integration of language and non-language brain regions during more effortful processes of novel metaphor comprehension.
## Introduction
Patients with schizophrenia demonstrate pervasive deficits in processing different pragmatic aspects of language, and in particular they show impairments in understanding proverbs, irony, and metaphors (Rapp, ). Comprehension of figurative language relies on effortful cognitive processes in which the non-literal message of the utterance is extracted. People with schizophrenia tend to interpret proverbs literally, a phenomenon termed “concretism”, and clinicians regard proverb interpretation as a potential tool in the diagnosis of schizophrenia (Reich, ). Some researchers have suggested that schizophrenia is associated with more general difficulties in abstract thinking (for a review see Thoma and Daum, ). The present study focuses on the challenges that people with schizophrenia experience when processing metaphor comprehension, especially novel metaphors.
Metaphors do not constitute a homogenous class of expressions but instead there is a continuum from idioms (dead metaphors) at one end to novel metaphors (live metaphors) at the other end (Fraser, ), with metaphors of different levels of conventionality in between. That is, some conventional metaphors had once been novel but due to repeated use have lost their metaphoricity. Metaphor comprehension is said to depend on level of conventionality (Glucksberg and Keysar, ; Giora, ; Giora and Fein, ; Bowdle and Gentner, ). The Career of Metaphor model (Bowdle and Gentner, ) argues that a newly created metaphor is comprehended via a comparison process, whereas a conventional metaphor is understood via a categorization process. Because the meanings of conventional metaphors are already stored in long term memory (i.e., they have been lexicalized), they are retrieved directly from the mental lexicon or via a previously created abstract metaphoric category. Unlike the comprehension of conventional metaphors, comprehension of novel metaphors involves an on-line effortful process of extraction and comparison of features.
Computing the metaphorical interpretation of an utterance relies on additional cognitive processes. Appraisal of the meaning of figurative language seems to be associated with the development of the ability to evoke mental images. Accordingly, school-aged children provide less sophisticated, less comprehensive, and more concrete mental images of idiom content than do adults (Nippold and Duthie, ). Behavioral evidence concerning the role of mental imagery in the comprehension and memory of idioms suggests that when people interpret idioms they construct a general mental image that is strongly constrained by conceptual mappings between base and target domains (for a review see Gibbs and O’Brien, ). For example, the mental image associated with spill the beans is derived from the conceptual mapping between the image of a mind as a container and the image of ideas as physical entities (Lakoff, ). This mapping evokes the image of taking ideas out of the physical container of the mind. According to Gibbs and O’Brien ( ), these images are unconscious, automatic, and independent of modality. With respect to this notion, Bottini et al. ( ) noted that the retrieval of information from episodic memory as well as mental imagery may be necessary to overcome the denotative violation inherited in metaphoric language.
It has been suggested that deficient language comprehension in schizophrenia is associated with right hemisphere involvement (e.g., Kircher et al., ; Mitchell and Crow, ; Bleich-Cohen et al., ; for a review see Rapp, ). According to Mitchell and Crow ( ), the abnormalities in language processing that are typical of schizophrenia reflect activation in right hemisphere homolog regions of key left hemisphere language regions. Furthermore, Mitchell and Crow ( ) argued that these functional changes indicate the loss or reversal of lateralized activation of brain regions associated with particular components of language processing. Although there is behavioral evidence of impairments in non-literal language comprehension in schizophrenia (de Bonis et al., ; Drury et al., ; for a review see Rapp, ; but see also Titone et al., ), only few neuroimaging studies tested metaphoric processing in this population (Kircher et al., ; Mashal et al., ).
If the right hemisphere is deficient in schizophrenia (Mitchell and Crow, ), and since there is some evidence suggesting that processing of novel metaphors involves the right hemisphere (Mashal et al., , ; Schmidt et al., ; Pobric et al., ; Mashal and Faust, , but see Rapp et al., , ), it is especially intriguing to test novel metaphor processing in schizophrenia. Kircher et al. ( ) found disrupted brain activation during an implicit task of metaphor processing in people with schizophrenia. Participants silently read novel metaphoric sentences (e.g., the lovers’ words are harp sounds ) as well as matching literal sentences (e.g., the lovers’ words are lies ), and then decided whether the sentence had a positive or a negative connotation. People with schizophrenia demonstrated increased activation in the left inferior frontal gyrus (IFG) while processing novel metaphors, whereas healthy participants demonstrated stronger signal changes in the right superior/middle temporal gyrus. Interestingly, the severity of concretism, as rated with the Positive and Negative Syndrome Scale (PANSS), was negatively correlated with left IFG activation, suggesting that activation of this region contributes to concrete thinking in schizophrenia. In a recent fMRI study (Mashal et al., ), people with schizophrenia who were asked to silently read novel metaphors demonstrated increased activation in left middle frontal gyrus (MFG) relative to processing of meaningless word pairs. This pattern of activation differed from enhanced brain activation in the right IFG observed in healthy controls. Thus, reversed lateralization patterns were documented in schizophrenia. These results suggest that inefficient processing of novel metaphors in schizophrenia may involve compensatory recruitment of additional brain regions, such as the left MFG, a region known to be involved in executive functioning, and specifically in working memory (e.g., Braver et al., ; Jha and McCarthy, ). Furthermore, direct comparison between the people with schizophrenia and healthy adults on processing of literal expressions and novel metaphors relative to a baseline condition revealed greater activation in left precuneus in the schizophrenia group.
The precuneus has been studied extensively over the past decade as a central hub of the default mode network (DMN), which typically shows deactivation compared to rest for sensory motor tasks in healthy participants (e.g., Fransson, ; Cavanna and Trimble, ; Fransson and Marrelec, ; Margulies et al., ; Zhang and Li, ). It has been observed that tasks that demand much attention are associated with decreased activity in the DMN (e.g., Mazoyer et al., ). The precuneus is interconnected with both cortical and subcortical regions. It is specifically connected to parietal areas, including the inferior and superior parietal (SP) cortex and the intraparietal sulcus, which have been associated with processing of visuo-spatial information (Selemon and Goldman-Rakic, ). Tracer injection studies in non-human primates have shown that the extra-precuneus cortico-cortical connections include the supplementary motor cortex, dorsal premotor area, anterior cingulate, and language related areas such as the prefrontal cortex (BA 8, 9, and 46), as well as the posterior superior temporal sulcus (PSTS) (for a review see Cavanna and Trimble, ). These widespread connections with frontal and temporal regions suggest that the precuneus may be involved in a variety of highly integrated and associative behavioral functions. The precuneus has been linked with language related tasks at word (e.g., Kouider et al., ) and sentence level comprehension (Whitney et al., ). Reviewing 100 studies, Price ( ) concluded that the comparison of comprehensible and incomprehensible sentences is associated with activation in four core regions including the precuneus, the anterior and posterior parts of the left middle temporal gyrus (MTG), bilateral anterior temporal poles, and the left angular gyrus. These regions were also identified in a meta-analysis of 120 studies (Binder et al., ) that pointed out seven brain regions engaged in semantic processing, including the posterior cingulate extending to the precuneus.
The precuneus has also been linked to episodic memory retrieval (Shallice et al., ), processing of mental imagery (Hassabis et al., ; Johnson et al., ; Burgess, ), and visuo-spatial memory functions (Vincent et al., ; Epstein et al., ). Previous studies have found that the retrieval of contextual associations is related to activation in the posterior precuneus and left prefrontal cortex. Lundstrom et al. ( ) suggested that the posterior precuneus is activated during regeneration of previous contextual associations and that the left lateral inferior frontal cortex is engaged in explicit retrieval as well as in integration of the contextual associations. Thus, the precuneus (together with inferior frontal cortex) is implicated in the recollection of past experiences. According to Binder et al. ( ), the precuneus is involved primarily in encoding episodic memories but at the same time it is consistently activated in semantic tasks, as it stores meaningful experiences together with their related associations in order to guide future behavior.
Evidence regarding the role of the precuneus in metaphor comprehension is mixed. Data from an fMRI study with healthy participants showed prominent left precuneus activation when familiar metaphoric sentences were contrasted with literal sentences (Schmidt et al., ). Data from another fMRI study indicated that the right precuneus plays an important role in processing novel metaphors but not in processing familiar metaphors (Mashal et al., ), suggesting that it involves retrieval of information from long-term episodic memory or the use of mental imagery. This interpretation is in line with Lakoff and Johnson’s ( ) idea that metaphoric language comprehension may depend on conceptualizations of personal experiences that are stored in episodic memory. Studies with patients with schizophrenia reported other findings. For instance, Kircher et al. ( ) found that literal sentences elicited greater activation in the left and right precuneus relative to metaphoric sentences. Our previous work documented increased activation in the left precuneus during processing of both literal expressions and novel metaphors in people with schizophrenia relative to healthy control participants (Mashal et al., ). This means that the precuneus appears to be involved not only in metaphor processing but also in processing of literal language. People with schizophrenia appear to recruit the precuneus but the exact role of the precuneus in language processing in this population remains unclear.
The aim of the present study is to define the role of the precuneus/SPL in processing of metaphors in schizophrenia by applying region-of-interest (ROI) analysis to bilateral precuneus/SPL and language regions. Furthermore, the focus is on precuneus/SPL activation and connectivity. We used a functional connectivity method that measures the interaction of one brain region with another. We thus measured the functional connectivity of the precuneus/SPL with language brain regions (IFG, PSTS) with which it is connected (Cavanna and Trimble, ). We also explored whether comprehension of conventional and novel metaphors is associated with signal change in the precuneus/SPL. We hypothesized that the precuneus/SPL would be more strongly activated when participants with schizophrenia processed literal language and novel metaphors relative to healthy participants. Furthermore, we expected to find a correlation between precuneus/SPL response and activation in language brain regions in schizophrenia that would attest for compensation of deficient metaphoric language processing. We also expected to find a positive correlation between signal change in the precuneus/SPL and comprehension of both conventional and novel metaphors.
## Method
### Participants
Twelve outpatients with schizophrenia (mean age = 28.08, SD = 4.34) and 12 healthy volunteers (mean age = 27.08, SD = 4.10) took part in this research. All participants were native Hebrew speakers and right handed according to self-report. The patient group included five women and had a mean of 12.3 years of formal education ( SD = 1.3), and the control group included seven women and had a mean of 13.1 years of formal education ( SD = 1.0). There were no statistically significant group differences in age ( t = 0.58, ns ), gender (χ = 0.67, df = 1, ns ), or education ( t = 1.01, ns ). Patients were recruited through the Tel Aviv Brull Community Mental Health Center, Israel. Two certified psychiatrists verified diagnoses according to the guidelines of the Structured Clinical Interview of the DSM-IV (SCID), Axis I, Patient Edition (First et al., ).
Prior to the imaging session, patients were clinically assessed with the Positive and Negative Syndrome Scale (PANSS; Kay et al., ) by a clinically trained person. The total mean PANSS score was 58.83 ( SD = 12.55), with a score of 11.75 ( SD = 4.29) for positive symptoms, 17.00 ( SD = 6.95) for negative symptoms, and 30.08 ( SD = 5.53) for general symptoms. All participants were on stable doses of atypical antipsychotic medication (mean chlorpromazine equivalents = 440 mg/day). Participants received a full explanation of the nature of the study as well as its potential risks and benefits and then provided written informed consent. The study was approved by the Institutional Review Board of Tel Aviv Sourasky Medical Center.
In the present study we reanalyzed the data collected by Mashal et al. ( ) in which 14 people with schizophrenia and 14 healthy participants were scanned. Two participants in each group showed no significant activation in the precuneus/SPL and were thus excluded from the present study.
### Behavioral testing
Participants completed a multiple-choice metaphor comprehension questionnaire. The questionnaire included 30 word pairs: 10 conventional metaphors, 10 novel metaphors, and 10 meaningless expressions (Mashal et al., ). For each word pair, four interpretations were provided: a correct interpretation, a literal distracter, an unrelated interpretation, and a phrase saying: “this expression is meaningless”. Participants were instructed to select the best response. The questionnaire was administered after the fMRI session.
### fMRI experiment
Data collection was described in Mashal et al. ( ). Here we reanalyzed the data using a ROI analysis and functional connectivity approaches that were based on the extraction of the individual time courses. To provide the reader with all necessary details, we describe all relevant experimental information from our previous paper.
### Stimuli
We selected 96 Hebrew word pairs that formed four types of semantic relations: literal ( birth weight ), conventional metaphors ( sealed lips ), novel metaphors ( pure hand ), or unrelated ( grain computer ). Several pretests were performed prior to the study. The aim of the first pretest was to determine whether each two-word expression was literal, metaphoric, or meaningless. Twenty healthy judges saw a list of expressions and were asked to decide if each expression is literally plausible, metaphorically plausible, or unrelated. For each condition we selected expressions that were rated by at least 75% of the judges as literally or metaphorically plausible, or as meaningless. To distinguish between conventional and novel metaphors, another group of 10 judges saw a list of only the plausible metaphors from the first pretest. They were asked to rate the degree of familiarity of these expressions on a 5-point scale ranging from 1 (highly unfamiliar) to 5 (highly familiar). Expressions with a score higher than three were considered conventional (average rating 4.67), whereas expressions with a score lower than three on the familiarity scale were considered novel metaphors (average rating 1.98). The third pretest assessed subjective rating of word frequency. Thirty-one additional raters were asked to rate all words on a 5-point scale ranging from 1 (infrequent) to 5 (highly frequent). The average rating was 3.45 for literal expressions, 3.79 for conventional metaphors, 3.67 for novel metaphors, and 3.38 for unrelated word pairs.
### Experimental task and procedure
The stimuli were presented in a block design fashion. Each block contained six word pairs in one of the experimental conditions. Each word pair was presented for 3000 ms followed by a 500 ms blank. The blocks were separated by either 6 s or 9 s, in which participants viewed a fixation point on a gray background (baseline). Each experimental condition appeared four times (with a total of 16 blocks) during each scan session. Each block contained one distracter, so that within a block of literal word pairs (or conventional or novel metaphors) there was one expression that was meaningless, and within the block of unrelated word pairs appeared one metaphoric expression. The first 18 s of the scan were excluded to allow for T2* equilibration effects.
Participants were asked to silently read each word pair and decide whether the word pair made sense. Prior to the fMRI scan the task was practiced with stimuli that were not used in the experiment.
### Image acquisition
Imaging measurements were acquired through a 3T GE scanner (GE, Milwaukee, WI, USA). All images were acquired using a standard quadrature head coil. The scanning session included anatomical and functional imaging. A 3D spoiled gradient echo (SPGR) sequence with high resolution (a slice thickness of 1 mm) was acquired for each person, in order to allow volumetric statistical analyses of the functional signal change and to facilitate later coordinate determinations. The functional T2* weighted images were acquired using gradient echo planar imaging pulse sequence (TR/TE/flip angle = 3000/35/90) with FOV of 200 × 200 mm , and acquisition matrix dimensions of 96 × 96. Thirty-nine contiguous axial slices with 3.0 mm thickness and 0 mm gap were prescribed over the entire brain, resulting in a total of 159 volumes (6201 images).
### Imaging data analysis
The fMRI data were processed through BrainVoyager software (Version 4.9; Brain Innovation, Maastricht, The Netherlands). Prior to statistical tests, motion correction, high frequency temporal filtering (0.006 Hz), and drift correction (no head movement > 1.5 mm was observed in any participant) were applied to the raw data. Pre-processed functional images were incorporated into the 3D datasets through tri-linear interpolation. Images were smoothed with a 6-mm fullwidth, half-maximum (FWHM) Gaussian filter. The complete dataset was transformed into Talairach space (Talairach and Tournoux, ). To allow for T2* equilibrium effects, the first six images of each functional scan were excluded.
### ROIs analyses
Our ROIs were defined anatomically and functionally. Specific effects were studied in the left and right precuneus extending laterally to the superior parietal lobule (SPL) and in pre-determined regions that are part of the language network: the left and right IFG, the left MFG, and the left and right PSTS. Anatomic definition of ROIs was based on sulci and gyri. The precuneus/SPL (BA 7) is limited anteriorly by the cingulate sulcus, posteriorly by the medial portion of the parieto-occipital fissure, and inferiorly by the subparietal sulcus and the intraparietal sulcus; the pars triangularis (BA 45/46) in the IFG (left and right), and the area near or at the PSTS between the superior temporal gyrus and the MTG BA 22 (left and right). Our ROIs were also functionally selected by calculating three-dimensional statistical parametric maps, separately for each participant, using a general linear model in which all three meaningful experimental conditions (literal expressions, conventional metaphors, novel metaphors) were positive predictors, and resting state was a negative predictor, with an expected lag of 6 s (accounting for the hemodynamic response delay). Thus, for each participant, task related activity within the pre-determined regions was identified by convolving the boxcar function with a hemodynamic function (HRF). Table presents the average Talairach coordinates of each ROI in each group.
Mean Talairach coordinates of activation clusters in regions of interest (ROIs).
* healthy participants only; ** patients only; SP = superior parietal; IFG = inferior frontal gyrus; PSTS = posterior superior temporal sulcus; MFG = middle frontal gyrus .
Time courses of statistically significant voxels were collected in each of the ROIs for each person. Individual averaged MR signals were calculated from all epochs (blocks) of the same condition per activated ROI. Signals were then transformed into percent signal change (PSC) relative to baseline. For all analyses involving the fMRI signal extracted from the ROIs, cluster size involved at least 50 voxels, and the significance threshold was set at p < 0.01, uncorrected. Significance tests were thus performed on the average PSC obtained within the cluster of all ROIs, as determined for each condition. Because we examined seven predefined ROIs, we set a more conservative threshold of p = 0.007 (calculated as 0.05/7) to account for multiple comparisons. The statistical analyses were conducted with STATISTICA software (version 5).
### Functional connectivity analysis
Functional connectivity analyses were performed by computing pair-wise correlations between activation in the precuneus/SPL and activation in language regions (PSTS, IFG). For each participant, fMRI time series (one for each ROI) were averaged separately across voxels within these ROIs for each type of semantic relation (literal word pairs, conventional metaphors, and novel metaphors). Pair-wise Pearson correlation coefficients were computed between each pair of regions (left precuneus/SPL-left PSTS, right precuneus/SPL-right PSTS, left precuneus/SPL-left IFG), using the averaged time series across participants (for each group and condition) during task performance (excluding the between-blocks intervals). Next, we standardized these signals by subtracting them from the mean activation and dividing by the SD, highlighting the specific condition fluctuations (see also Ionta et al., ). The significance of the correlations was evaluated through a random permutation test (for similar bootstrapping analysis see Arzouan et al., ). In this test, Pearson correlation coefficients are calculated from 5,000 random permutations of the averaged time courses, and are then used to construct the distribution and test the significance of the original correlation value. Additional correction was used to compensate for the multiple comparisons (2 groups × 3 semantic relations × 3 pairs of regions), resulting in a conservative threshold of p = 0.002 (calculated as 0.05/18).
#### The relation between metaphor comprehension and precuneus/SPL activation
Next, we evaluated the correlation between behavioral scores on the metaphor comprehension questionnaire and precuneus/SPL activation. We thus calculated Pearson correlations between the PSC elicited by each metaphoric condition (conventional and novel metaphors) and the scores obtained in the metaphor questionnaire, separately for each participant. Then, we tested the significance of these correlations with a random permutation test (Arzouan et al., ) that generated 5,000 random permutations for each condition. This method relies on minimal assumptions and can be applied when the assumptions of a parametric approach are untenable (Nichols and Holmes, ). The 5,000 permutations were used to construct the distribution, and test the significance of the original correlation value with a p value of 0.006 corrected for multiple comparisons (0.05/8 comparisons = 2 groups × 2 semantic relations × 2 pairs of regions).
## Results
### Behavioral results
Metaphoric questionnaire : People with schizophrenia understood fewer conventional metaphors (mean = 81.25%, SD = 18.07) than did healthy individuals (mean = 97.92%, SD = 4.8), t = 3.08, p < 0.01, and fewer novel metaphors (mean = 68.73% correct, SD = 17.10) than did healthy individuals (mean = 88.96%, SD = 11.82), t = 3.42, p < 0.01. No significant group difference was found in comprehension of meaningless word pairs ( p > 0.05). Figure presents questionnaire responses by type of expression and group.
Mean percent (and standard deviation) of correct responses on metaphor questionnaire, by group . CM = conventional metaphors; NM = novel metaphors; UR = unrelated word pairs. * denotes p < 0.05.
### ROI analysis
Average PSC was analyzed in each of the ROIs by a two-way repeated measures ANOVA in regions showing significant activation by both groups or by a one-way repeated measures ANOVA in regions in which there was significant activation in only one group (see Table ).
A two-way repeated measures ANOVA for signal change within the right precuneus/SPL, with the two groups (schizophrenia, healthy) as a between-subject factor and expression type (literal, conventional, novel) as a within-subject factor, revealed a main effect of group, F = 9.29, p = 0.006. A Scheffe post hoc analysis revealed greater signal change in the schizophrenia group than in the healthy group, p < 0.01. The main effect of expression type was also significant, F = 8.74, p = 0.006. A Scheffe post hoc analysis indicated that literal expressions led to greater signal change than did both conventional metaphors, p < 0.01, and novel metaphors, p < 0.05. However, the group X expression type interaction was not significant, F = 3.55, p = 0.007 (see Figure ). A two-way repeated measures ANOVA for signal change within the left precuneus/SPL, with the two groups (schizophrenia, healthy) as a between-subject factor and expression type (literal, conventional, novel) as a within-subject factor, revealed no significant effects ( p s > 0.007).
Whole-brain activation showing signal change for the three conditions (LIT = literal word pairs, CM = conventional metaphors, NM = novel metaphors), vs. baseline using fixed effects analysis ( p < 0.0001, uncorrected) and percent signal change (SE) in right precuneus/superior parietal lobe .
Percent signal change in the right and left PSTS and the right and left IFG are presented in Figure . A two-way repeated measures ANOVA for signal change within the right PSTS with the two groups (schizophrenia, healthy) as a between-subject factor and expression type (literal, conventional, novel metaphors) as a within-subject factor, revealed a main effect of expression type, F = 6.27, p = 0.004. A Scheffe post hoc analysis indicated that literal expressions led to greater signal change than did novel metaphors, p < 0.01 (Figure ). No other effects reached significance ( p > 0.007). A two-way repeated measures ANOVA for signal change within the left PSTS revealed no significant main effects ( p > 0.007).
Whole-brain activation showing signal change for the three conditions (LIT = literal word pairs, CM = conventional metaphors, NM = novel metaphors) vs. baseline using fixed effects analysis ( p < 0.0001, uncorrected) and percent signal change (SE) in left IFG, right IFG, left PSTS, and right PSTS .
Significant activation in the right IFG was seen in healthy participants alone and therefore a one-way repeated measures ANOVA was performed on signal change in this location, with expression type (literal, conventional, novel) as a within-subject factor. This analysis revealed a significant main effect of expression type, F = 10.01, p = 0.0008. Signal change for conventional metaphors was significantly weaker than was signal change for novel metaphors, p < 0.05, and significantly weaker than was signal change for literal expressions, p < 0.01. A two-way repeated measures ANOVA for signal change within the left IFG, with group as a between-subject factor and expression type as a within-subject factor, revealed a significant interaction, F = 8.85, p = 0.0006. A Scheffe post hoc analysis showed that literal expressions led to greater signal change in healthy participants than it did in people with schizophrenia, p < 0.05. All other effects were not significant ( p > 0.007).
Finally, because only people with schizophrenia showed significant activation in the left MFG, a one-way ANOVA was performed on signal change in this location, with expression type as a within-subject factor (literal, conventional, novel). No significant main effect of expression type was found ( p > 0.007) (see Figure ).
Whole-brain activation of people with schizophrenia showing signal change for the three conditions (LIT = literal word pairs, CM = conventional metaphors, NM = novel metaphors) vs. baseline using fixed effects analysis ( p < 0.0001, uncorrected) and percent signal change (SE) in left MFG .
### Functional connectivity analysis
To determine connectivity patterns we calculated pair-wise Pearson correlations between activation in the precuneus/SPL and activation in language regions for each expression type, separately for each group (see Table ).
Pair-wise Pearson correlations between activation in precuneus/SPL and activation in pre-determined ROIs, by expression type and group.
* = statistically significant association at the α = 0.0001 level using permutation test .
People with schizophrenia : A permutation test analysis showed a significant correlation between activation in the right precuneus/SPL and activation in the right PSTS for literal word pairs, p < 0.001, conventional metaphors, p < 0.0001, and novel metaphors, p < 0.0001. The correlations between activation in the left precuneus/SPL and activation in the left PSTS were significant for both literal word pairs, p < 0.0001, and novel metaphors, p < 0.00001. There was also a significant correlation between activation in the left precuneus/SPL and activation in the left IFG, but only for novel metaphors, p < 0.0001. These results point to left precuneus/SPL involvement in processing of both literal expressions and novel metaphoric expressions and literal word pairs and between activation in the right precuneus/SPL and activation in the right PSTS while processing all semantic relations.
Healthy group : No significant correlation between precuneus/SPL activation and activation in the other ROIs was observed within the control group.
### Correlations between performance on the metaphor questionnaire and activation in the precuneus/SPL
To further examine whether metaphor comprehension is related to precuneus/SPL activation, we calculated the correlation between scores on the metaphor questionnaire and the BOLD signal recorded within the left and the right precuneus/SPL.
People with schizophrenia : using the permutation test, the only correlation that was found to be significant was the correlation between the comprehension of novel metaphors and BOLD signal in the right precuneus/SPL, r = 0.83, p < 0.001. Thus, the more correct responses that were given on the questionnaire, the stronger was the BOLD signal within the right precuneus/SPL.
Healthy participants : No significant correlations between questionnaire score and precuneus/SPL activation were found in the control group.
## Discussion
The purpose of the present study was to examine the role of the precuneus/SPL in metaphor comprehension in schizophrenia. Three main findings emerged: (1) people with schizophrenia showed greater activation in the right precuneus/SPL relative to healthy participants; (2) within the schizophrenia group BOLD signal in the left precuneus/SPL and in the left PSTS correlated positively during comprehension of both literal word pairs and novel metaphors. There was also a positive correlation between activation in the right precuneus/SPL and in the right PSTS in all semantic relations. In addition, the left precuneus/SPL was co-activated with the left IFG during novel metaphor processing. No equivalent correlations with activation in the precuneus/SPL were found in the healthy group; and (3) within the schizophrenia group comprehension of novel metaphors, as measured by an off-line questionnaire, was correlated with increased activation in the right precuneus/SPL.
The behavioral results showed that people with schizophrenia understood fewer metaphors than did healthy participants. This reduced accuracy is consistent with previous evidence of difficulties in metaphor comprehension in schizophrenia (e.g., Iakimova et al., ; Kircher et al., ), and is associated with an abnormal pattern of brain activation in schizophrenia (Kircher et al., ; Mashal et al., ). The present study suggests that the right precuneus/SPL is involved in processing linguistic expressions in schizophrenia, and in particular in understanding novel metaphors. People with schizophrenia may recruit this right posterior parietal region to compensate for their deficient metaphor comprehension. It is also possible that metaphor comprehension is deficient in schizophrenia because this area is recruited. However, the current study cannot determine which explanation is correct. Our results also show that increased novel metaphor comprehension (as assessed by the off line questionnaire) was correlated with increased activation in the right precuneus/SPL, consistent with previous views about the central role of the right hemisphere in metaphor comprehension (Bottini et al., ; Giora, , ; Mashal et al., , ).
The precuneus/SPL has been linked to both linguistic and cognitive processes. According to recent meta-analyses, the precuneus is part of the brain networks associated with semantic processing (Binder et al., ; Price, ). Our findings point to increased activation in the right precuneus/SPL in schizophrenia as compared to controls. It is possible that this increased activation reflects the process of linking two words into a meaningful expression. However, because processing novel metaphors is demanding, requiring the extraction of relevant features of two disparate domains (Bowdle and Gentner, ), greater activation is expected when we compare novel metaphors to literal expressions. Nevertheless, the results of the ROI analysis documented similar signal change across different semantic relations. Hence, it is less likely that precuneus/SPL activation reflects semantic processing in general (Binder et al., ). Following Lakoff and Johnson ( ), we assume that people construct mental images in order to use and understand not only figurative language but also literal language. The way in which people construct these mental images differs between the two types of expressions though. While the mental images invoked by figurative language are constrained by conceptual mappings between the base and target domains, the mental images invoked by literal language are based on the understanding of basic level prototypes. Thus, it is possible that people with schizophrenia, unlike healthy participants, either use the right precuneus/SPL to form mental images for both literal and figurative language. Alternatively, it is possible that people with schizophrenia may engage in retrieval of personal experiences from episodic memory (Lakoff and Johnson ( )).
The low activation observed in the right precuneus/SPL within the healthy group may be related to the observation that the precuneus is part of the DMN (e.g., Fransson, ; Cavanna and Trimble, ; Fransson and Marrelec, ; Margulies et al., ; Zhang and Li, ). Indeed, there is evidence suggesting that the precuneus is normally less activated during attention-demanding tasks (Cabeza and Nyberg, ). It is therefore possible that the pattern of right precuneus/SPL activation in the control group reflects reliance on attentional resources during metaphor processing. It is also possible that the increase in activation in the schizophrenia group is due to abnormalities in the resting state network. Bluhm et al. ( ) reported altered spontaneous fMRI signal fluctuations in the precuneus/posterior cingulated cortex in schizophrenia during resting state. Thus, our results may suggest that whereas healthy participants activate the right precuneus/SPL in accordance with its role as a central hub in the DMN, people with schizophrenia fail to use these attentional resources.
Functional connectivity analyses allowed us to detect associations between neural regions that conventional activation-based analyses cannot address. An important finding of our study is the strong functional connectivity between the left precuneus/SPL and the left PSTS during comprehension of literal word pairs and novel metaphors, as well as the strong connectivity between the right precuneus/SPL and the right PSTS during processing of all semantic relations. These findings suggest that the interactions between the precuneus/SPL and the posterior language area, PSTS, may serve to mediate metaphor comprehension in schizophrenia. The fact that the left precuneus/SPL and the left PSTS, to which the precuneus has anatomical connections (Cavanna and Trimble, ), were correlated during literal language comprehension as well indicates that people with schizophrenia may automatically activate mental images in response to both literal and metaphoric expressions (Lakoff and Johnson ( )). The mental images may then be transformed to auditory representations in the left PSTS to enhance comprehension. In addition, the fact that the left precuneus/SPL was co-activated with the left IFG during novel metaphor comprehension suggests that people with schizophrenia use this brain region in collaboration with the IFG to facilitate novel metaphor comprehension. As argued by Lundstrom et al. ( ), this co-activation may reflect reliance on previous contextual associations which are processed in the precuneus/SPL and their integration in the left IFG. Thus, our results may explain some of the inconsistency in previous fMRI studies in which both left and right precuneus involvement was seen in processing of both literal language and metaphors (e.g., Kircher et al., ; Schmidt et al., ; Mashal et al., ). We suggest that people with schizophrenia, but not healthy participants, use the bilateral precuneus/SPL in collaboration with language areas to facilitate both literal and novel metaphor comprehension.
The current ROI analysis revealed abnormal patterns of signal change in people with schizophrenia. Whereas the right IFG was activated in the healthy group, no such activation was recorded in the schizophrenia group. Thus, consistent with the right hemisphere hypothesis (Mitchell and Crow, ), lateralization patterns were different in this group. Interestingly, the ROI analysis found greater activation for processing literal expressions as well as novel metaphors relative to conventional metaphors. This finding indicates that the right lateralized activation observed in healthy individuals was not limited to the interpretation of figurative language but included literal language as well. Unlike the group difference that was documented in the right IFG, both groups activated the left IFG. The ROI analysis demonstrated that the healthy group had greater activation in the left IFG while processing literal word pairs than did the schizophrenia group. Thus, whereas both groups activate the left IFG during metaphor processing to the same extent, the patients show deficient activation of the right IFG.
There are some limitations to our study. First, a larger sample size would have strengthened our conclusions. Given that evidence from different analyses converge in showing involvement of the right precuneus/SPL in novel metaphor processing in schizophrenia, we believe the results will be replicated with a larger group of patients. However, a larger sample size of healthy participants is required to test whether the lack of significant connectivity seen in this group stems from the small sample of healthy participants. Second, we performed the analyses on a subgroup of 24 participants (out of 28 in the original study) who showed significant activation within selected ROIs. It is thus possible that the activation pattern seen here is not universal. We note that the exact role of the precuneus/SPL in language processing is still not entirely clear. However, if the precuneus/SPL activates mental images in response to the current task then we expect to see activation in this area during performance of tasks that explicitly tap into the mental visualization of linguistic expressions. Furthermore, because the expressions used in the current study form a continuum in terms of literality and abstractness (Laor, ), people with schizophrenia evoke different mental images on that continuum. Future studies are needed to shed more light on the type of mental images processed by the precuneus. Finally, we did not control for medication effects. Although there is evidence that atypical antipsychotic medication enhances cognitive performance (e.g., Sumiyoshi et al., ) and specially attention and verbal fluency (Meltzer and McGurk, ), the effects of medication on metaphor processing remain unclear.
In summary, our results shed light on precuneus/SPL involvement in metaphor comprehension in people with schizophrenia. The inefficient processing of metaphors in schizophrenia is related to increased activation in the right precuneus/SPL. It appears that people with schizophrenia recruit the right precuneus/SPL to facilitate novel metaphor comprehension, probably because they rely more on mental imagery and episodic retrieval. Furthermore, people with schizophrenia seem to recruit the bilateral precuneus/SPL while processing novel metaphors, as observed by the co-activation of these regions and both language areas. In contrast, healthy participants seem to rely on the bilateral IFG to process literal expressions and the right IFG to facilitate novel metaphor comprehension. Our results also indicate that the precuneus/SPL contributes to comprehension of literal expressions in schizophrenia, as manifested by tight coupling between the precuneus/SPL and the PSTS during literal language processing.
## Conflict of interest statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Complex regional pain syndrome (CRPS) is a chronic, debilitating pain condition that usually arises after trauma to a limb, but its precise etiology remains elusive. Novel clinical signs based on body perceptual disturbances have been reported, but their pathophysiological mechanisms remain poorly understood. Investigators have used functional neuroimaging techniques (including MEG, EEG, fMRI, and PET) to study changes mainly within the somatosensory and motor cortices. Here, we provide a focused review of the neuroimaging research findings that have generated insights into the potential neurocognitive and neuroplastic mechanisms underlying perceptual disturbances in CRPS. Neuroimaging findings, particularly with regard to somatosensory processing, have been promising but limited by a number of technique-specific factors (such as the complexity of neuroimaging investigations, poor spatial resolution of EEG/MEG, and use of modeling procedures that do not draw causal inferences) and more general factors including small samples sizes and poorly characterized patients. These factors have led to an underappreciation of the potential heterogeneity of pathophysiology that may underlie variable clinical presentation in CRPS. Also, until now, neurological deficits have been predominantly investigated separately from perceptual and cognitive disturbances. Here, we highlight the need to identify neurocognitive phenotypes of patients with CRPS that are underpinned by causal explanations for perceptual disturbances. We suggest that a combination of larger cohorts, patient phenotyping, the use of both high temporal, and spatial resolution neuroimaging methods, and the identification of simplified biomarkers is likely to be the most fruitful approach to identifying neurocognitive phenotypes in CRPS. Based on our review, we explain how such phenotypes could be characterized in terms of hierarchical models of perception and corresponding disturbances in recurrent processing involving the somatosensory, salience and executive brain networks. We also draw attention to complementary neurological factors that may explain some CRPS symptoms, including the possibility of central neuroinflammation and neuronal atrophy, and how these phenomena may overlap but be partially separable from neurocognitive deficits.
## Introduction
Complex Regional Pain Syndrome (CRPS) is a chronic, debilitating pain condition that usually arises after trauma to a limb. It is characterized by disproportionate pain and variable combinations of sensory (allodynia, hyperalgesia), vasomotor (temperature changes or asymmetry, skin color changes or asymmetry), sudomotor (sweating changes or asymmetry), edema, trophic (thin glossy skin, abnormal hair growth, coarse nails), and motor (weakness, decreased range of motion, tremor, dystonia) changes. In up to 10% of cases, there is no obvious trauma reported (Turner-Stokes and Goebel, ). Although CRPS is primarily a limb-confined condition, it has also been reported in other body parts including face (Melis et al., ). The precise etiology of this enigmatic condition remains unexplained. Based on observational evidence of aberrant inflammation, vasomotor dysfunction, and cerebral cortical changes, it has been proposed that these factors account for the main features of CRPS but may occur to a different extent depending on individual susceptibility (Marinus et al., ), thus accounting for the clinical heterogeneity of the condition.
A number of perceptual disturbances reported in patients with CRPS may serve as both novel clinical signs of the condition and markers for underlying biological mechanisms that can be targeted for treatment. Prominent examples include finger misperception, impaired hand laterality recognition, astereognosis, and abnormal body scheme (Förderreuther et al., ; Moseley, ; Cohen et al., ). CRPS patients may also report unusual symptoms such as “feeling of foreignness” and wish to amputate the affected limb (autotomy wish) (Galer and Jensen, ). Overall, the evidence points to patients with CRPS having difficulty with the mental representation of their affected limb. However, as outlined in this review, despite many neuroimaging studies of the sensory and motor systems in CRPS, the pathophysiological mechanisms underlying perceptual disturbances remain unknown. We outline a number of approaches to investigating the origin and role of somatosensory perceptual disturbances in CRPS, with a particular focus on the role of top-down (expectancy-related) mechanisms in perception, a relatively unexplored topic in the CRPS literature. We make recommendations for the investigation of the role of top-down mechanisms in CRPS and suggest that this approach may be useful in the delineation of neurocognitive phenotypes that improve our understanding of the heterogeneity of the condition and its causal mechanisms.
## Disturbances of the Body Scheme in CRPS
“Body scheme” is a term used to define the dynamic, real-time, representation of one’s own body. This representation is generated by proprioceptive, somatosensory, vestibular, and other sensory inputs and is integrated with motor systems for control of action in a way that is normally automatic and seamless. Neurological studies indicate that disturbance of body scheme may be caused by abnormal functioning of various parts of cerebral cortex, including the somatosensory (Hari et al., ), parietal (Daprati et al., ), insular (Karnath and Baier, ), and frontal cortices (Weijers et al., ). It thus seems that body scheme can be disturbed at different stages or levels of neuronal integration, from the processing of the early sensory inputs to the body scheme integration and spatial orientation in the parietal cortex to the reportable conscious awareness of own body supported by the executive frontal functions. Recent electroencephalography (EEG) studies suggest that each of the early sensory, the mid-latency, as well as the late cognitive stages of neuronal processing contribute to somatosensory awareness (Auksztulewicz and Blankenburg, ; Adhikari et al., ), which is one of the prerequisites for an intact body scheme. The presence of a frontal component to the pathophysiology seen in patients with CRPS is indirectly supported by the effectiveness of the treatment of subjective CRPS symptoms by cognitive behavioral therapy (De Jong et al., ) and acceptance-based approaches (Cho et al., ), which function through influences on frontal cortices (Etkin et al., ; Brown and Jones, ). The possibility of a mechanistic role for aberrant frontal cortical processes in CRPS signs and symptoms is a topic we explore in depth in the latter sections of this review.
One aspect of body scheme is representation of the position of the limb in peri-personal space. Many CRPS patients lack awareness of the position of their limb in space (Lewis et al., ) and have difficulty recognizing the laterality of a pictured image of a hand as either left or right (Parsons, ). There is also evidence for delayed hand laterality recognition on the affected side that in one study was related to symptom duration and to the pain that would be evoked by executing the movement (Moseley, ), pointing to deficits in the ability to represent the position of the limb in space. Interestingly, a single-case study (Bultitude and Rafal, ) revealed that unawareness of limb position can occur before the onset of pain symptoms in CRPS, suggesting that perceptual disturbances may be a marker for the pathophysiological mechanisms preceding chronic, limb pain, rather than being a consequence of pain.
Other perceptual assessments have focused on the ability to recognize the somatic location and identity of objects touching the skin of the affected limb, revealing deficits in these finer-grained perceptual judgments in CRPS. Förderreuther et al. ( ) found that 48% of patients with CRPS had an impaired ability to identify the fingers of the affected hand. In contrast, the ability to identify fingers on the unaffected hand was impaired in only 6.5% of patients. Impaired identification of the fingers was not related to the affected side of the body (left vs. right) of CRPS. The study authors also reported that all patients stressed that their difficulties naming the fingers could not be explained by reduced perception of the cotton swab. This provides preliminary evidence for no deficit in afferent transmission of cutaneous sensation, but more likely a change in the way the brain constructs a spatial representation of the limb, similar to the concept of a deficit in the body scheme.
A seemingly related phenomenon, astereognosis, is defined as the inability to identify an object by touch (without visual input) despite having intact cutaneous sensation. Classically, this is reported in patients who have had stroke mainly affecting the parietal lobe, but this has also been reported in some patients with CRPS, including 64% of patients in one study (Cohen et al., ). In addition, neurocognitive dysfunctions thought to be similar to the neurological neglect caused by post-stroke damage to the right parietal lobe have been reported in CRPS, and the term “neglect-like” has been used to describe them (Galer et al., ). For example, some CRPS patients perceive their own affected limb to be “foreign” and not belonging to them and this is referred to as “cognitive neglect.” Similarly, some CRPS patients need to focus mental and visual attention in order to move their affected limb (“motor neglect”).
Together, these findings suggest possible parietal and frontal lobe involvement in the perceptual disturbances of CRPS and specifically deficits in the ability to represent the location, orientation, and structure of the affected limb. However, it is presently unclear the extent to which these different manifestations of neurocognitive dysfunction in CRPS tend to co-occur within individual patients with CRPS, whether they share overlapping mechanisms, or what pathophysiological mechanisms underpin them, e.g., disturbances in the early (parietal) stages of processing or abnormalities in the latter (frontal) stages of bodily awareness. Neuroimaging research, which we discuss in the next sections, has to date been conducted largely independently of observations of the various perceptual disturbances described above. Hence, the pathophysiological mechanisms of these salient perceptual disturbances remain unknown.
## Neuroimaging of Somatosensory Representations in CRPS
Among the more compelling evidence of aberrant neurophysiology in CRPS is that of somatosensory and motor cortical plasticity, which is often assumed to be the underlying biological cause of body perceptual disturbances in CRPS. Studies of cortical changes in regions representing somatic sensation, namely the primary (S1) and secondary (S2) somatosensory cortices, were inspired by earlier studies of the effects of alterations of afferent input (as occur in many types of chronic pain) on cortical reorganization of sensory maps. In monkeys, digit amputation resulted in shrunken representation in area 3b of SI cortex of the corresponding finger (Merzenich et al., ), while subsequent human work using magnetoencephalography (MEG) found that upper limb amputation also caused the face area of S1 cortex to expand into the former hand area (Flor et al., ). Critically, these latter findings predicted the intensity of concurrent phantom limb pain, consistent with previous work finding similar correlations between back pain and S1 cortical reorganization (Flor et al., ) [however, also see Makin et al. ( , ) who present contradictory evidence of a lack of invasion of the former hand area by the lip area and a lack of correlation of cortical reorganization with pain intensity]. Together, these results implicated similar somatosensory cortical changes in patients with other types of chronic pain, including neuropathic pain and CRPS.
Subsequently, a number of studies (reviewed in the next section) used MEG or EEG with source imaging, or functional Magnetic Resonance Imaging (fMRI), to investigate somatosensory cortex representations in CRPS. EEG and MEG data are most commonly interrogated to identify changes in electrical or field potentials generated by the coordinated activity of assemblies of cortical pyramidal cells. The majority of studies adopted similar EEG/MEG methods to that of the aforementioned work in phantom limb and chronic low back pain. Indeed, EEG/MEG methods are naturally powerful techniques to resolve, with high temporal resolution, early somatosensory responses that are most closely related to afferent inputs.
Studies in patients with CRPS have mostly assessed averaged time-locked signals (evoked potentials or fields), which are the stimulus-driven cortical changes that have consistent onset latency in relation to the stimulus across many experimental trials. These averaged stimulus-evoked responses can be quantified or further processed in a number of ways, and studies of CRPS have to date focused on one or a number of outcomes: (1) the amplitude of the evoked signal, representing the sum of the activity of (predominantly excitatory) cortical neurons, to investigate possible deficits in afferent processing of somatosensation, (2) the evoked signal’s latency/timing with respect to the stimulus, to investigate possible delays to the afferent signals reaching the brain, (3) habituation/suppression of the amplitude of the evoked signal by multiple stimulus repetitions, to investigate the possibility of deficits in intracortical inhibition, and (4) source modeling of the evoked responses to reveal the spatial location of cortical generators of the signal, to investigate possible changes in the location of cortical representations of the affected limb. In addition to these methods, non-time-locked neuronal oscillations in different frequencies can be measured. fMRI studies have also been conducted in which the magnitude of the evoked signal and cortical spatial representations have been resolved.
## Somatosensory Spatial Representations
Of the lines of neurological investigation conducted in CRPS as summarized above, a recent meta-analysis by Di Pietro et al. ( ) confirmed that the strongest evidence of aberrant neurological changes in CRPS is plasticity in cortical representations of the affected limb, manifesting as a reversible shrinkage of the somatosensory cortex. In the meta-analysis, pooled data from four MEG studies (Juottonen et al., ; Maihöfner et al., ; Sinis et al., ; Vartiainen et al., ) and one EEG study (Pleger et al., ) were examined. Meta-analysis is highly desirable with respect to the above studies given the small sample sizes, ranging from single-subject analyses (Sinis et al., ) to more commonly between 6 and 12 patients. The authors of the meta-analysis reported that the evidence supported the hypothesis that S1 representations of the body were reduced on the affected hand compared to the unaffected hand in CRPS.
A shrinkage in the Penfield’s homunculus (Penfield and Boldrey, ) would provide a compelling explanation for many of the perceptual disturbances seen in CRPS. Cortical reorganization may disrupt the internal body map and impair performance on the tasks requiring the identification of somatosensory information and coding of body posture. However, the evidence supporting this hypothesis has limitations that should be acknowledged, which we discuss in detail below. First, there are important methodological caveats of EEG/MEG for assessing cortical reorganization. Second, the studies included in the meta-analysis suffered from a high risk of bias, which more recent work [not included in the meta-analysis by Di Pietro et al. ( )] has addressed, finding conflicting results.
Regarding methodological limitations of EEG/MEG, an important issue is that the reported spatial changes in somatosensory responses in comparing thumb and little finger digits (typically in the region of 5 mm on average) are comparable to or smaller than the estimated spatial resolution and accuracy of the best available source modeling methods with MEG and EEG based on simulated data (Darvas et al., ; Yao and Dewald, ), which must therefore be considered optimistic when applied to clinical data. With clinical data, the accuracy of the source model may be affected by unknown/unmodelled concurrent neural responses such as those involved with top-down modulation from higher-order cortical regions. Subject motion during recording/scanning, which is more likely in patients with more severe symptoms, can reduce data quality and introduce artifactual effects that may underestimate the observational parameters. The introduction of “noise” from the above sources risks biasing results, especially in studies with small samples sizes.
Corroboration of representation changes measured with MEG/EEG by complementary techniques, such as fMRI, is essential for the evidence to conclusively converge. However, the results of the relatively few currently published fMRI studies investigating somatosensory cortical plasticity in CRPS are equivocal about the precise cortical changes taking place. Pleger et al. ( ) found support for the EEG/MEG findings already discussed. However, as detailed by Di Pietro and colleagues in their meta-analysis (Di Pietro et al., ), the majority of EEG, MEG, and fMRI studies in CRPS to date have a potential for bias arising from the selective reporting, unclear outcomes and unblinded assessments. In order to address this, data from a more recent fMRI study (Di Pietro et al., ) was analyzed blind to the group (CRPS patients or healthy controls) and hand (affected or unaffected). Contrary to previous findings, CRPS was associated with an enlarged representation of the healthy hand, not a smaller representation of the affected hand. Consistent recent findings from fMRI studies of cortical reorganization in patients with phantom limb pain also shed doubt on the hypothesis that maladaptive plasticity is the cause of phantom limb pain: patients with greater pain intensity had a more greatly preserved representation of the former hand area, with pain thought to arise from nociceptive or top-down inputs rather than maladaptive plasticity (Makin et al., ). Further studies are needed to replicate and confirm these fMRI results in patients with CRPS. As well as minimizing the potential for bias, studies could use multiple converging neuroimaging methods, e.g., EEG combined with fMRI to improve spatial localization of early somatosensory responses.
## The Challenge of Heterogeneity
Another common shortcoming of neuroimaging studies to date is the inclusion of only small numbers of poorly characterized patients. Heterogeneity in terms of clinical presentation is well documented (Marinus et al., ). It also appears that there is significant heterogeneity within the CRPS population on the basis of studies of perceptual disturbance in CRPS. For example, finger misperception occurred in 48% of CRPS patients in one study (Förderreuther et al., ), and it is unknown to what extent finger misperception overlaps with other deficits. It is possible that there are common pathophysiological mechanisms, cutting across the different types of body perceptual disturbance, which manifest differently on an individual patient basis depending on other biological and psychological susceptibility factors. Alternatively, the mechanisms underlying two different manifestations, for example identifying fingers and recognizing the laterality of a presented hand, may be entirely or largely discrete.
To illustrate, in comparing the phenomena of astereognosis with finger misperception, the functional difference in terms of somatosensory processing can be summarized in terms of a “what” (i.e., objective identification) vs. “where” (discrimination of tactile stimulus location) distinction. Investigations of somatosensory processing with fMRI (Reed et al., ) have compared “what” vs. “where” processes, showing differential activation patterns. Tactile object recognition activated frontal as well as bilateral inferior parietal areas. In contrast, tactile object location activated bilateral superior parietal areas. A common gray matter deficit across CRPS patients presenting with different perceptual abnormalities therefore seems unlikely.
On the other hand, investigations of white matter may be warranted. The possibility of a common white matter deficit explaining a constellation of neurocognitive dysfunctions is illustrated by findings from research conducted in patients with Gerstmann syndrome (Gerstmann, ). In this syndrome, parietal lobe lesions lead to a tetrad of finger agnosia (difficulty in the naming of fingers), agraphia (difficulty in writing), acalculia (difficulty in performing calculations), and left to right confusion. Neuropsychological studies during open brain surgery found a relation between the Gerstmann tetrad and left parietal cortex and demonstrated a certain degree of proximity and overlap of those cortical sites where electrical stimulation can elicit these symptoms (Morris et al., ). More recently, Rusconi et al. ( ) used fMRI and diffusion tensor imaging in healthy subjects to identify that the parietal activation patterns across all four domains consistently connected to a small region of subcortical parietal white matter. Hence, Gerstmann syndrome might arise from disconnection, via a lesion, to separate but co-localized fiber tracts in the subcortical parietal white matter.
In a similar fashion, it has been suggested that perceptual disturbances in CRPS arise from changes within the parietal lobe (Cohen et al., ), where a matrix of a coherent body scheme may arise (Daprati et al., ). In support of this hypothesis, an fMRI study of activations relating to cold- or brush-induced allodynia in pediatric CRPS patients identified right parietal lobe involvement (Lebel et al., ). However, it is far from clear that there is a consistent constellation of perceptual disturbances in all patients with CRPS, or within a subgroup that has yet to be defined, that would point to a single unifying mechanism. Indeed, the idea that a parietal lobe deficit might be responsible for CRPS was challenged by a study of gray matter atrophy and white matter reorganization (Geha et al., ), in which atrophy in patients with CRPS was found in cluster encompassing right ventromedial prefrontal cortex (PFC), anterior insula, and nucleus accumbens (Figure A). The study found co-localized decreases in white matter anisotropy and changes in branching and connectivity of white matter tracts linked to these site-specific gray and white matter abnormalities. Deficits in the parietal lobe, however, were not evident in the patient sample studied. Smaller gray matter volume in ventromedial PFC in normal or pathological states has been observed to relate to poorer performance on tasks requiring cognitive control and decision making (Clark et al., ; Boes et al., ). Furthermore, it is interesting to note that recent evidence (Figure C) points to the ventromedial PFC and nucleus accumbens as being important in the ability to self-regulate pain (Woo et al., ) – more on this in later sections of this review. Overall, these findings suggest that while some CRPS symptoms may be associated with parietal deficits, in other patients altered frontal cortex activity is more apparent, variability that remains to be explained.
Evidence for a role of the ventromedial PFC and nucleus accumbens in CRPS and in the self-regulation of pain . (A) Brain regional gray matter density, as measured with voxel-based morphometry (VBM), is decreased in patients with CRPS relative to healthy controls in the right hemisphere (red), spanning the ventromedial PFC, anterior insula (AI), and nucleus accumbens (arrows). The scatter plot shows that this decreased gray matter density is negatively correlated to the number of years the patients have been living with CRPS. Individual healthy control subjects are shown at pain duration = 0. The histogram depicts mean (±SEMs) gray matter density within the cluster in both groups. Reproduced from Geha et al. ( ). (B) The localization of MEG-derived independent components (ICs) for a CRPS patient with pain in her left foot and ankle. Top: the localization of the first IC (with frequency spectra in the delta, theta, and beta range) to right S1 and M1 along the central and post-central sulcus, extending to the mesial surface and over the right SA in the superior parietal cortex (see expanded views). Bottom: localization of an IC in the theta range to orbitofrontal cortex bilaterally and left temporal pole. Reproduced from Walton et al. ( ). (C) Multilevel three-path mediation analysis with the ventromedial PFC and nucleus accumbens as a priori regions-of-interest, showing that these regions formally mediate the effect of instructions to voluntarily upregulate and downregulate pain perception on subjective pain ratings. Reproduced from Woo et al. ( ).
Heterogeneity with the CRPS population highlights the importance of characterization/phenotyping and subgrouping of patients for research studies. However, the possibility of there being separate phenotypes within the CRPS population is rarely considered in neuroimaging studies. To date, patient heterogeneity has been considered only in terms of overt sensory and motor symptoms, such as hyperalgesia/allodynia (Maihöfner et al., ) and dystonia (Van Rijn et al., ). A potentially powerful alternative would be to utilize heterogeneity in body perception that may reveal more subtle and detailed information about the processes and mechanisms of the underlying sensory and motor systems. An analysis of sub-groups of CRPS patients according to the degree of different types of perceptual disturbance would improve our understanding of the pathophysiological underpinnings of these phenomena.
Neuroimaging studies to date have been conducted on small numbers of patients, motivating the need for meta-analytic techniques to draw conclusions. Indeed, CRPS is a rare condition, which makes recruitment of large numbers of patients for research studies in single centers an obvious challenge. The need for identifying and comparing subgroups of patients with CRPS, which will require far larger numbers of patients in a single study than has been conducted to date, underlines this challenge further. Researchers are likely to have to look toward conducting multi-center studies and/or amass databases of patients covering large geographical areas in order to have the scale to compare potentially subtle neurophysiological differences among subgroups. Promising steps have been taken in this regard; for example the CRPS UK Clinical Research Network has established a large registry database of patients (300+ in size at the time of writing) to facilitate epidemiology studies and academic and clinical trials (Shenker et al., ).
## Neurological Explanations for Cortical Plasticity
It would be helpful to put the hypothesis of shrinkage of Penfield’s homunculus in CRPS into a broader context and to consider possible mechanisms. It has been proposed that “blurring” of somatosensory maps, i.e., increased overlap between representations of adjacent skin surfaces, would increase the total number of neurons representing the affected body part, leading to generation of the misperception of that body part being larger (swelling) (Haggard et al., ). This mechanism was originally discussed in relation to the generation of phantom limb pain, in which case the cause of somatosensory “blurring” is proposed to be deafferentation of C-fibers leading to cortical disinhibition, because C-fibers normally provide continuous inhibition to primary somatosensory cortex (Calford and Tweedale, ).
There are numerous lines of evidence supporting the idea that reduced afferent input to the somatosensory cortex increases the perceived size of the corresponding body part and can as a result also increase the perceived painfulness of sensations arising from that body part. Local anesthesia of the thumb produces an increase in the perceived size of the thumb (Gandevia and Phegan, ), while anesthetic injections at the dentist make the mouth feel swollen (Türker et al., ) and anesthesia of the brachial plexus results in the perception of swelling of the entire arm (Paqueron et al., ). Paqueron and colleagues identified that changes in perceived limb size had the same time course as a reduction in sensitivity to pin-prick and thermal sensations, implying the phenomenon is related to reduced cortical input from small diameter Aδ and C-fibers (Paqueron et al., ).
However, the hypothesis that disinhibition of somatosensory cortices may underlie somatosensory cortical reorganization and perceptual disturbances in CRPS is not currently well supported. A promising research direction has been the assessment of cortical excitability with EEG/MEG and TMS using a variety of paired-pulse methods. However, while the results of TMS have largely supported the hypothesis of disinhibition of the motor cortices bilaterally in CRPS (Schwenkreis et al., ; Eisenberg et al., ), EEG/MEG investigations of somatosensory disinhibition in small numbers of CRPS patients have provided mixed results (Van Rijn et al., ; Lenz et al., ). Furthermore, fMRI evidence suggests greater cortical inhibition in response to affected relative to unaffected limb stimulation of allodynia in pediatric patients with CRPS (Lebel et al., ). Future studies should be conducted with larger patient numbers that can identify distinct phenotypic subgroups and associated mechanisms, which may account for some of the variability in the study findings thus far.
The hypothesis of cortical disinhibition in CRPS could be further investigated with respect to possible causal factors. In the case of phantom limb pain, reduced afferent input to the somatosensory cortex is a likely contributing factor. However, there is a lack of consistent evidence suggesting differences in afferent input to the somatosensory cortex in CRPS (Di Pietro et al., ). Two out of five EEG or MEG studies (Maihöfner et al., ; Pleger et al., ) found no consistent differences between affected and unaffected limbs of CRPS patients in the amplitudes of S1 responses while the remainder (Juottonen et al., ; Maihöfner et al., ; Vartiainen et al., ) did find some evidence of greater S1 response for the affected compared to the unaffected side. fMRI evidence is equally mixed: no differences were observed between CRPS patients and healthy controls in S1 activation strength to a variety of stimulations ranging from light touch to tonic pain in two studies (Forster et al., ), including no S1 differences in comparing the hyperalgesic (affected) vs. unaffected limbs (Maihöfner et al., ). However, CRPS patients with allodynia did have augmented S1 and S2 cortical responses in one fMRI study (Maihöfner et al., ). Following this work, Pleger et al. ( ) found overall lower responses in S1 and S2 cortex to tactile stimulation. Hence, the evidence for changes in the amplitude of somatosensory processing in patients with CRPS is inconsistent and so far has not been shown to relate to the degree of cortical reorganization.
If cortical disinhibition and reorganization does not result from changes in afferent inputs in CRPS, then a loss of inhibitory cortical interneurons may occur by another mechanism. A hypothesis gaining traction is neuroinflammation. Evidence supporting the case for neuroinflammatory mechanisms in CRPS includes findings that, first, CRPS patients have elevated levels of the pro-inflammatory cytokines IL-1β and IL-6 in their cerebrospinal fluid, as well as reduced levels of the anti-inflammatory cytokines IL-4 and IL-10 (Alexander et al., ). Second, there is evidence for the spread of microglial and astroglial activation within the spinal cord of CRPS patients (Van Rijn et al., ), which may exacerbate neuroinflammation. Third, recent evidence suggests that a number of CRPS patients have serum antibodies that interact with autonomic receptors, in particular the alpha-1a adrenergic receptor (Dubuis et al., ), beta-2 adrenergic receptor (Kohr et al., ), or the muscarinic acetylcholine receptor (Kohr et al., ; Dubuis et al., ). Serious neuroinflammatory consequences would be expected to arise when autoantibodies against these receptors exudate from blood vessels, together with complement proteins and leukocytes (Cooper and Clark, ).
The cause of such neuroinflammation is unclear but could arise from inflammation within peripheral nerves. From both animal and human studies (e.g., Banati et al., ), evidence is accumulating that neuroinflammation can spread, either anterograde or retrograde, via axonal projections in the CNS, thereby establishing neuroinflammatory tracks and secondary neuroinflammatory foci within the neuraxis (Cooper and Clark, ). Neuroinflammation spreading to second-order synapses in supraspinal centers provides a potent mechanism to destabilize feedback circuits, such as those involved in proprioception, nociception, and autonomic functions, as occurs in CRPS (Cooper and Clark, ). A preclinical model of chronic neuropathic pain has implicated glial activation in hyperalgesia (LeBlanc et al., ): minocycline injected into the somatosensory thalamus (posterolateral nucleus) reversed both microglial activity and hyperalgesia. However, whether neuroinflammation affects the somatosensory cortex and related functions in patients with CRPS remains unknown. Future studies could assess the extent of cortical reorganization and perceptual disturbance in CRPS in relation to the presence of neuroinflammatory markers and specifically those in somatosensory and motor cortex.
## Neurocognitive Models of Somatosensory Perception
While most studies on the functioning of the cerebral cortex in CRPS have largely focused on early somatosensory processing, it is known that intact somatosensory awareness depends also on the late cognitive stages of neuronal processing (Auksztulewicz and Blankenburg, ; Adhikari et al., ) and that neurological disturbances of the body scheme can be caused by the frontal abnormalities (Weijers et al., ). If so, perceptual disturbances in some patients with CRPS may in fact point to disturbed cognitive-executive functioning among individuals with CRPS. The mechanics of somatosensory perception has begun to be investigated in terms of its dependency on the executive functions of frontoparietal networks as well as the “salience network” including anterior insula and midcingulate cortex. Much of this investigation has been based on “Hierarchical Predictive Coding (HPC)” accounts of perception [for example (Rao and Ballard, ; Friston, , )] that may shed light on body misperceptions and neuroplasticity in CRPS. Here, we outline this theoretic approach, and in subsequent sections, we review supporting empirical data from neuroimaging studies of somatosensory perception and finally explore how this perspective could form the basis for identifying neurocognitive phenotypes of patients with CRPS.
Hierarchical predictive coding accounts of perception originate from the work of Hermann von Helmholtz ( ) who proposed that the brain does not represent sensations per se , but rather models the causes of those sensations. Because these causes cannot be perceived directly, they must be inferred from sensory data. However, as Friston discussed (Friston, ), the problem is that sensations can potentially have multiple causes that interact. Taking an example from vision, the retinal image size can be affected both by object size and distance from the observer. There is therefore inherent uncertainty in the causes of sensory impressions, which the brain must deal with to generate perceptions and guide actions.
One solution to this problem is for the brain’s model of the environment to contain prior expectations about how causes interact, for example the expectation that regardless of the distance from the observer, objects maintain a constant size. As lucidly described by Clark ( ), HPC models depict that top-down expectancy-related information is used to predict and “explain away” the sensory inputs, leaving residual “prediction errors.” These prediction errors then propagate information forward within the system – they report the “surprise” induced by a mismatch between sensory signals and predictions of those signals and serve to update the brain’s virtual model of the causes of those sensations so as to improve the reliability of predictions. Such errors can occur at multiple levels of a processing hierarchy, such that higher-level systems generate predictions about the inputs to lower-level systems on the basis of their role in modeling the causal structure of the world.
This scheme is attractive due to being computationally efficient (i.e., it reflects computations that neurons could feasibly produce) and providing a structure reminiscent of cortical circuits. On the basis of empirical evidence, asymmetrical (forward and backward) connections are thought to relate to specific computational variables within HCP models (e.g., predictions, prediction errors, and “precision” – a concept we come to later). For example, the dynamics of mismatch responses (in which the brain receives sensory inputs that are unexpected in relation to prior inputs) are better described by the minimization of prediction error than by other alternative hypotheses (Garrido et al., ; Chennu et al., ; Lieder et al., ). An important avenue of future research will be to evaluate how well computational models explain dynamic changes in somatosensory perception and neural plasticity and to assess the importance of these models for understanding the pathophysiology of chronic pain.
## The Role of Top-Down Predictions in Chronic Pain
If the HCP framework is correct, optimal perception and behavior depends on minimizing prediction error. This can either be achieved by changing the brain’s predictions to explain sensory input through the act of perception and learning or by actively changing sensory input to fulfill the brain’s predictions by acting on the world. In the latter case, the agent can selectively sample the sensory inputs that it expects. This is known as active inference . As Friston explains (Friston, ), an intuitive example of this process would be feeling our way in darkness: we anticipate what we might touch next and then try to confirm those expectations.
Selective sampling of sensory data in order to confirm expectations may help to explain why expectations, as formed by prior experiences, have long been known to modify sensory perception, including the perception of pain. However, it is more recently that functional neuroimaging has begun to delineate the mechanisms by which this occurs and to investigate the role of top-down mechanisms in disease states such as chronic pain. For example, pain expectancies trigger anticipatory neural responses (Ploghaus et al., ; Brown and Jones, ; Brown et al., ; Palermo et al., ) that result in changes in perception, emotion, and behavior (Wager et al., ; Brown and Jones, ; Clark et al., ; Kong et al., ; Seidel et al., ). Such changes are adaptive for avoiding acute injury but are potentially maladaptive in clinical conditions in which pain is chronic. These observations have inspired a body of work over the last two decades focusing on identifying the neural mechanisms by which cognitive expectancies influence pain perception and exploring these mechanisms in chronic pain populations. For example, somatosensory responses in S1 and S2 are modulated by expectations (Langner et al., ). These expectation effects reflect top-down biases, presumably originating from frontoparietal networks that activate during anticipation of stimuli (Brown et al., ; Watson et al., ; Kong et al., ) and explain multimodal expectancy effects (Langner et al., ). Recent evidence suggests, for example, that key nodes of the frontoparietal and salience networks, the dorsolateral PFC and anterior insula cortex (described in more detail below), show aberrant responses during anticipation of pain that are common across chronic pain populations suffering both nociceptive and non-nociceptive (unexplained) pain (Brown et al., ). However, to date these approaches have not been applied to understanding top-down mechanisms in CRPS.
Which aspects of HCP are likely to be of relevance to the pathophysiology of CRPS? According to HCP models, ambiguity in sensory input biases perception toward expectations (Dayan and Abbott, ). A hypothetical scenario in which expectations may exert a greater-than-normal influence on somatosensory perception is the existence of sensory nerve pathology resulting in greater signal “noise,” i.e., uncertainty in sensory inputs. This could occur in patients with type 1 CRPS for whom there is evidence of small-fiber neuropathy (Van der Laan et al., ; Albrecht et al., ; Oaklander et al., ). Such changes could be potentially monitored with psychophysics and neuroimaging. For example, neurobiological theories inspired by HCP generally ascribe functional asymmetry to ascending and descending connections (Bastos et al., ); indeed, a recent study of visual cortex (involving hierarchical processing from V1 to V4) demonstrated ascending prediction errors are related to fast gamma oscillations, while descending predictions are related to slower beta (13–31 Hz) and alpha (8–12 Hz) oscillations (Bastos et al., ). On the other hand, there is some evidence for fast gamma-band activity reflecting recurrent connections between the somatosensory and prefrontal cortices during tactile discrimination (Adhikari et al., ), although it is unknown whether these findings correspond to the coding of predictions and/or prediction errors. Indeed, higher-level representations and predictions (e.g., those reflecting conceptual or semantic information about expected changes in the environment) are thought to involve lower frequency bands (Correia et al., ), consistent with the idea that lower frequencies entrain brain regions across larger spatial and temporal distances in the brain (Canolty and Knight, ). Interestingly, greater spectral power in the EEG in the low-frequency delta (< 4 Hz) and theta (4–9 Hz) ranges, localized to both somatosensory and ventral PFC (orbitofrontal cortex), have been found in CRPS patients compared to control subjects (Walton et al., ) in a similar region to that showing gray matter atrophy in patients with CRPS (Geha et al., ) and that appears to be important for the top-down self-regulation of pain (Woo et al., ) – see Figure . This points to the intriguing possibility that the somatosensory processing abnormalities in CRPS are mediated by the long-range and low-frequency entrainment across frontal and somatosensory cortices, representing the influence of high-level predictions on somatosensory perception. This view is also supported by fMRI evidence of greater functional connectivity patterns between the post-central gyrus and prefrontal, cingulate and thalamic regions to cold allodynia in pediatric patients with CRPS (Linnman et al., ) compared to healthy controls, which persisted after recovery.
## Modeling Recurrent Connections in the Somatosensory System
One approach to investigating the respective roles of somatosensory forward (bottom-up) and backward (top-down) connections in body misperceptions would be through the use of modeling techniques such as Dynamic Causal Modeling (DCM). DCM allows the study the neuronal architecture underlying observed electromagnetic signals (from EEG and MEG) and the effective connectivity between its sources, making it a useful tool in testing alternative models of causal interactions between brain areas that explain the measured data (David et al., ). DCM has been applied to EEG data to assess evidence for feedforward, feedback, and recurrent processing between S1 and S2 in a somatosensory detection task (Auksztulewicz et al., ) – also see Figure . Early ERPs (<80 ms) were well explained by a model assuming only modulation in the feedforward connectivity between S1 and S2 cortices, and this connection was only stronger after stimulus detection for data segments longer than 80 ms. Furthermore, recurrent processing after 80 ms was needed in the model to explain the differences in EEG responses between detected and missed stimuli, and after 140 ms to explain the effect of awareness on ERPs. Therefore, recurrent processing within the somatosensory system, dominated by an enhanced S1–S2 connection, underlies somatosensory detection and awareness. This is consistent with dominant neural models of consciousness suggesting that reportable perceptual experiences depend on (1) sufficient early sensory processing, (2) wide distribution of sensory representations within the executive functions, and (3) recurrent interactions between sensory and frontal brain regions (Lamme, ; Dehaene and Changeux, ). If so, any reported perceptual abnormality may be caused not only by disturbed sensory processing but also by disturbed executive functions, or abnormal interaction between the sensory and executive regions of the brain. Abnormalities in such recurrent connections may underlie body misperceptions in CRPS.
(A) Neural networks and their effective connections underlying somatosensory perception, based on Dynamic Causal Modeling research conducted by Allen et al. ( ) and Auksztulewicz et al. ( ) and work studying anticipatory neural activity prior to pain and somatosensation by Brown et al. ( ), Atlas et al. ( ), and Langner et al. ( ). Frontoparietal executive networks are likely to mediate perceptual predictions while the salience network (aIC and MCC) mediate the effect of predictions on the perception of tactile and pain stimuli, with the aIC acting as a “hub” controlling the balance between bottom-up and top-down information. PFC, refrontal cortex; IPC, Inferior parietal cortex; MCC, Midcingulate cortex; aIC, Anterior insular cortex; iS2, Ipsilateral secondary somatosensory cortex; cS2, Contralateral secondary somatosensory cortex; cS1, Contralateral primary somatosensory cortex. (B) Variables hypothesized to influence the neurocognitive phenotype of CRPS, based on a hierarchical predictive coding (HPC) account of parameters describing the computational function of each neural network. The integrity of somatosensory neurons could be potentially influenced both by neurological factors (e.g., neuroinflammation leading to neuronal atrophy) and neurocognitive factors (i.e., changes in neural plasticity related to attention and learning). Resulting changes in signal quality from early cortical processing could change the precision weights attributed to sensory inputs and thereby the gain on prediction errors, a process balanced by the relative precision weights on top-down predictions. According to HCP models, this balance affects the extent to which predictions are updated according to sensory inputs (thereby determining the acuity of tactile perceptions) and also affects the content and influence of top-down predictions as mediated by anticipatory neural activity prior to expected tactile or nociceptive stimuli. Finally, evidence for neuronal atrophy in the executive and salience networks in CRPS lends to the hypothesis of long-term changes in neuroplasticity related to the weighting of top-down predictions, possible leading to aberrant perceptual decision-making.
There is also a body of evidence suggesting an important role for the anterior insula cortex in the anticipation of pain (Porro et al., ; Wager et al., ; Brown et al., ; Palermo et al., ) and mediating the effect of expectations on pain (Koyama et al., ; Brown et al., ; Atlas et al., ). We have seen that disturbance of body scheme is related to abnormal functioning of insula (Karnath and Baier, ), and that this region show atrophy in CRPS patients (Geha et al., and Figure A). An important question is what role the insula plays in neurocognitive (and particularly HCP) models of somatosensory perception and misperception. The insula is a center of salience processing across multiple sensory, emotional, and cognitive domains (Uddin, ). The anterior insula is thought to be crucial for the hierarchical processing of bodily information, integrating afferent thalamic and sensory inputs with top-down control signals arising in the prefrontal and cingulate cortex (Seth et al., ; Seth, ). The right anterior insula is highly interconnected with primary somatosensory areas such as posterior insula and somatosensory cortex (Cerliani et al., ; Chang et al., ) and anticipates the sensory and affective consequences of pain and touch (Brown et al., ; Lovero et al., ). The anterior insula also projects to the amygdala, forming a network contributing to emotional salience (Seeley et al., ). Functional connectivity between the insula and amygdala is thought to be related to levels of pain-related fear and is dampened by effective psychological treatment in pediatric patients with CRPS (Simons et al., ). Observations of the centrality of the insula in salience processing have led researchers to investigate the role of recurrent connections between the insula and somatosensory cortex in somatosensory perception. DCM has revealed that unexpected somatosensory stimuli increase the strength of forward connections along a caudal to rostral hierarchy – projecting from thalamic and somatosensory regions toward insula, cingulate and prefrontal cortices – reflecting the role of forward connection in conveying prediction error (Allen et al., ). The anterior insula, however, was the only region to show increased backwards connectivity to the somatosensory cortex, augmenting a reciprocal exchange of neuronal signals. These results suggest that the anterior insula acts as a hub for regulating somatosensory responses in a top-down manner (Figure ).
## Neurocognitive Mechanisms of Hemispatial Neglect in CRPS
It has been proposed that the anterior insula and midcingulate cortex form a “salience network” (Seeley et al., ). Salience and attention has been linked to the “precision” (reliability/degree of certainty) of sensory inputs (Feldman and Friston, ). Within the HCP framework, attention serves the function of balancing top-down and bottom-up influences on perception according to their respective precision weights (Figure ). In HCP, precision enhances the influence of ascending prediction errors via the regulation of post-synaptic cortical gain (Moran et al., ). By this means, attention (via the salience network) can drive learning and appropriate plasticity. By extension of this logic, a lack of precision/attention to a particular limb, i.e., cognitive neglect, may result in a relative loss of cortical function akin to disuse, a hypothetical explanation for cortical changes in patients with CRPS in cases in which no other neuropathology can be observed.
A useful illustration of how this might work in relation to CRPS neglect-like symptoms is the rubber hand illusion (RHI). The RHI refers to the illusory sense of ownership of a plastic hand, which is induced by synchronous tactile stimulation of the fake and the participant’s real (but hidden) hand. In order for the brain to assign the experience of ownership to the artificial hand, certain sensory evidence must be suppressed, namely proprioceptive evidence that the two hands are in different positions (Zeller et al., ). In HCP, this corresponds to a reduction in the precision/attention afforded to sensory prediction errors (Feldman and Friston, ; Bastos et al., ). As evidence in favor of this account, an ERP study (Zeller et al., ) identified an attenuation of somatosensory-evoked responses in frontal electrodes that corresponded to cortical sources in the (contralateral) perirolandic area and the parietal lobe. In the absence of an illusion but in the presence of a (perceived) artificial hand, responses were larger in primary somatosensory cortex and inferior parietal lobule. This is consistent with a hypothetical reduction in gain mediated by superficial pyramidal cells in order to resolve the multisensory conflicts arising under the illusion. Should similar multisensory conflicts arise in a patient with CRPS, as implied by the success of mirror therapy in some patients (McCabe et al., ), the brain may naturally attempt to resolve these conflicts by attenuating somatosensory predictions errors, with the consequence of driving hemispatial neglect and body misperceptions.
## Neurocognitive Phenotypes
A number of novel neurocognitive mechanisms have been hypothesized here on the basis of the reviewed literature, which if further investigated could help to delineate different phenotypes within the CRPS population. To summarize these possible mechanisms, three hypothetical phenotypes are described here with reference to Figure B. This illustrates how different phenotypes could potentially emerge with overlapping symptoms but distinct causes:
A patient without sensory misperceptions may experience pain and other overt symptoms for neurological reasons, such as neuroinflammation, which is not severe enough to directly affect neuronal integrity. This patient would be regarded as normal with regard to neurocognitive parameters.
A patient experiencing somatosensory misperceptions may have suffered a loss of neuronal integrity, with possible causes including neuroinflammation leading to gray and white matter atrophy in ascending spino-thalamic tracts and/or sensory cortex. This would result in a reduction in signal quality in somatosensory afferents at one or multiple levels from the spinal cord to the thalamus and cortex. A loss of signal quality would result in uncertainty (reduction in precision) regarding sensory inputs and a weighting of perception toward top-down predictions (by increasing the precision of predictions). The result would be a lack of tactile acuity and greater potential for body image distortions arising from abnormally greater biasing of perception by higher-level expectations.
A patient may experience somatosensory misperceptions but have no evidence of neuroinflammation or other possible causes of neuronal loss, suggesting the possible influence of other causal factors (e.g., psychological factors). For example, maintenance of abnormally rigid high-level beliefs about the body may require the suppression of somatosensory prediction errors (by decreasing the gain on neuronal error units) in a way akin to the RHI. This would suppress learning and associated neuronal plasticity, with possible consequences for neuronal integrity within sensory cortices.
Both phenotypes 2 and 3 may manifest as signs of body misperception and cognitive neglect, with corresponding cortical changes, but result from a different pathophysiology.
In order to identify these three proposed phenotypes (or others than remain to be hypothesized), three lines of enquiry would need to overlap in future studies, which to date have been investigated separately: the assessment of perceptual distortions, the investigation of neuroinflammation and neuronal loss, and the modeling and estimation of parameters defining causal perceptual mechanisms. Due to the complexity of such investigations and the analytic techniques required to measure these processes (as outlined in this review), in the future, the practicality of identifying such phenotypes in large samples of patients will likely depend on the discovery of practical biomarkers of the different pathophysiological processes.
## Conclusion
There is a large and increasing literature on CRPS for which the present review does not attempt to create exhaustive account. Instead, we have focused on lines of enquiry that we believe are likely to lead toward a more integrated understanding of the pathophysiological mechanisms underlying perceptual disturbances in CRPS. To date, perceptual disturbances in CRPS have largely been investigated separately from neurological deficits, a fact we draw attention to in order to encourage more multi-disciplinary research in this area. Neuroimaging studies have begun to identify potential mechanisms but have lacked an appreciation of the heterogeneity of perceptual disturbances and their potential underlying pathophysiological mechanisms. We suggest that the definition of pathophysiological subgroups of CRPS patients can be achieved by matching specific neurocognitive deficits to cortical mechanisms and demonstrating the effects of specific treatments on those mechanisms.
## Author Contributions
AK conducted and documented literature searches for the first draft. VN wrote sections of the article. NS, SC, and TB critically appraised and edited the article. CB wrote the majority of the article and was responsible for the article’s narrative, structure and editing. All authors approved the final version.
## Conflict of Interest Statement
This review was supported by funding from an EFIC-Grunenthal Grant (awarded to Christopher Brown) and from the Addenbrooke’s Charitable Trust. No potential conflicts of interest have been identified. The other co-authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Attention is vital to success in all aspects of life (Meck and Benson, ; Erickson et al., ), hence it is important to identify biomarkers of later attentional problems early enough to intervene. Our objective was to determine if any of 11 genes ( APOE, BDNF , HTR4, CHRNA4, COMT, DRD4, IGF2, MAOA , SLC5A7 , SLC6A3 , and SNAP25 ) predicted the trajectory of attentional development within the same group of children between infancy and childhood. We recruited follow up participants from children who participated as infants in visual attention studies and used a similar task at both time points. Using multilevel modeling, we associated changes in the participant’s position in the distribution of scores in infancy to his/her position in childhood with genetic markers on each of 11 genes. While all 11 genes predicted reaction time (RT) residual scores, only Monoamine oxidase A ( MAOA ) had a significant interaction including time point. We conclude that the MAOA single nucleotide polymorphism (SNP) rs1137070 is useful in predicting which girls are likely to develop slower RTs on an attention task between infancy and childhood. This early identification is likely to be helpful in early intervention.
## Introduction
Because attention is vital to success in all aspects of life (Meck and Benson, ; Erickson et al., ), it is important to identify biomarkers of later problems with attention early enough to intervene. Doing so may alleviate the development of attentional problems in psychological disorders such as attention-deficit disorder (ADHD; Abed, ), anxiety (Bar-Haim et al., ), autism spectrum disorders (Camarata, ), and depression (Mueller et al., ). Longitudinal studies are especially important because they allow us to determine which early symptoms actually lead to later problems. Making an early identification and intervening are consistent pursuits among those researching developments (Moss et al., ; Webb and Jones, ; Camarata, ; DuPaul and Stoner, ). Early identification is possible with some biomarkers.
Genes are a logical choice for early biomarkers because they control the availability of neurotransmitters. The biological pathway from a particular gene to attentional behavior is often hypothesized to include interference with normal neural messaging across a synapse as neurotransmitter availability is affected. The varying availability of these neurotransmitters due to genetic differences could make behaviors like attending less efficient in some individuals, and this, in turn, could be captured by various measures of attention, including outcomes from computer tasks. Based on observing associations between specific genes and early visual attention, some researchers have called for studies on the developmental course of genotypic differences from infant attention (Holmboe et al., ).
Identifying genetic influences on the development of attentional deficits will allow interventions to be created. However, first we must identify early individual differences in behavior to which genetic differences can be associated. One aspect of attention that can be studied in infancy is reflexive attention. Reflexive attention is a component of overall attention that deals with the stimulus-driven information processing. It can be contrasted to sustained attention, the other component of attention. Some studies suggest that reflexive and sustained attention are independent components of overall attention (Barry et al., ; Berger et al., ; Andrade et al., ; Underbjerg et al., ; Dye and Hauser, ).
Studies using habituation and paired comparisons (Fantz, ; Fagan, ; Bornstein and Sigman, ) have found that infant measures of information processing can predict child information processing (Fagan and Singer, ; Rose and Wallace, ; Dougherty and Haith, ; Rose and Feldman, ; Rose et al., ; Kavsek, ). Note that information processing measures are often more related at the two time points than infant IQ (which relies on motor skills) and child IQ (which relies on cognitive skills; Colombo, ; McCall, ; Sigman et al., ; Tucker-Drob et al., ).
For our information processing task, we use a forced-choice preferential looking task with a left or right moving bar in a field of static bars (Dannemiller, ). When movement captures attention, there is “selection” of the moving bar from among the other static bars on the display and the infant looks at the moving bar. This task provides percent correct (PC) and latency (RT) for the adult observer to make left-right looking judgments that reflect the speed and clarity of signals the infant sends (Teller, ).
We could find no studies using infant forced-choice preferential looking to predict future cognitive outcomes. The task has, however, been used to classify toddlers with autism or developmental delays (Pierce et al., ). This suggests that preferential looking may be useful to predict risk for future attention related disorders. Our interest in reaction time (RT) trajectory is to determine at an early age, which children might develop problems with attention and need intervention. Such results would be the first to our knowledge to demonstrate genetic influence on the trajectory of the development of reflexive attention.
Several genes have been identified as associated with poor attention at one point or another during development (Frank et al., ; Brookes et al., ; Ribasés et al., ; English et al., ; Nobile et al., ; Bidwell et al., ; Nymberg et al., ; Gálvez et al., ). Dopamine genes are often associated with attentional deficits (Smith et al., ; Li et al., ; Genro et al., ; Holmboe et al., ).
In addition to dopaminergic genes, other genes are plausibly associated with attentional development. For example, APOE , SLC5A7 , and CHRNA4 are all associated with the neurotransmitter acetylcholine, which has been associated with cognitive development (McKinnon and Nathanson, ) and with visuospatial attention, ADHD, and distractibility (Manuck et al., ; Manor, ; Störmer et al., ; Markant et al., ). Likewise, BDNF , HTR4 , and Monoamine oxidase A ( MAOA ) are associated with serotonin, although MAOA and BDNF also influence dopamine availability (Yu et al., ; Razgado-Hernandez et al., ; Voigt et al., ; Parikh et al., ). Serotonin has been associated with brain and attentional development (Binder and Scharfman, ; Shim et al., ; Faraone and Mick, ) and with ADHD (Poirier, ; Greenwood et al., ; Fisher et al., ; Parasuraman et al., ; Walitza et al., ; Winterer et al., ). Overall, the likelihood of genetic influence on the trajectory of attentional development is high because these genes influence such developmental processes as axon guidance, neuronal cell mobility, synaptic function, and chromosomal remodeling (Gilman et al., ; Douet et al., ). While the “snapshot” associates described above are useful, it would be even more useful to be able to predict the likelihood of declining attentional performance over development to determine who might most need intervention to prevent the development of more serious problems.
Trajectory research is essentially a longitudinal research with an emphasis on individual differences. Studying the trajectory of specific outcomes based on earlier risk factors has been especially important in health research (Henly et al., ). A recent study used the concept of trajectory of development and determined that children with a Williams Syndrome diagnosis are not simply delayed in development, but show a distinctly different path of development (Annaz et al., ). Other researchers have found trajectory research useful in studying genetic influences on cognitive development (Torgersen, ; Plomin et al., ; Sigman et al., ; Tucker-Drob et al., ). Because it tracks the same children at two or more time points (Farrington, ; Selig and Little, ; Grammer et al., ), trajectory research could indicate if genetic variation is associated with developmental change in attention. Such influences on development are plausible because genes influence neuronal and synaptic changes from infancy to childhood.
In particular, we are asking whether genes influence the trajectory of reflexive attention development. To answer this question, we recruited participants from a follow up sample of children who participated as infants in visual attention studies (Dannemiller, ). We use RT and PC to index performance on a reflexive attention task. Better reflexive attention orients more quickly and accurately to a moving bar.
## Materials and Methods
To determine if genetic variation is associated with developmental change in attention, we selected genetic markers based on their biological and functional effects. There are a very large number of genes related to brain development and the function of the majority of these genes is poorly understood (Dixon-Salazar and Gleeson, ). Exploring the influence of several candidate genes with a plausible biological influence on attention to determine if that influence alters the trajectory of development can be useful. We focused our selection on genes that had already been mentioned in the literature as related to some aspect of attention, cognition, growth, or brain development and that had a plausible biological explanation for their potential influence on reflexive attention. For example, some genes were selected for study because they are related to growth generally. Others were selected because they are related to brain growth more specifically. Insulin-like genes are one example of a growth-related gene and there has been some suggestion that these may also influence cognition (Borenstein et al., ). In addition, APOE is highly expressed during development and is associated with both epilepsy and schizophrenia (Ziats and Rennert, ). Another way to select genes related to brain development is by examining the stage at which the gene is likely to affect cortical development. For example, defects in genes involved in controlling neurogenesis are likely to cause more severe brain disorders (such as autosomal recessive primary microcephaly, which involves reduced neuronal numbers) than genes that control the developmentally later process of circuit formation (such as CHRNA4 associated with epilepsy; Dixon-Salazar and Gleeson, ). MAOA is a gene on the X chromosome whose gene product is a mitochondrial enzyme that degrades monoamines in the central nervous system (Seif and De Maeyer, ). More specifically, the enzyme catalyzes the oxidative deamination (removal of an amine group) of dopamine and serotonin (Caspi et al., ) and has been associated with ADHD and impulsivity (Manuck et al., ; Lawson et al., ). The rationale behind the selection of each candidate gene is described in Table .
Candidate genes with their biological effects and functional effects .
Note: Biological effects include changes to the gene that alter the way the protein it codes for is produced (e.g., changes in a synaptic receptor, enzyme, or neurotransmitter). Function effects include associations between this genetic variation and behaviors, including symptoms of disorders associated with attention .
In particular single nucleotide polymorphisms (SNPs) on each gene were selected based on several attributes including citation in the cognitive literature, minor allele frequency (MAF), block structure, and distance between markers. The markers that we have in the current data set include 39 SNPs and two VNTRs on 11 genes. The genes were selected based on evidence in the literature that the gene was related to attentional deficits or a disorder with an attentional component (see Table for illustrative literature). COMT, DRD4, IGF2, SLC6A3 , and SNAP25 are related to the availability of dopamine. APOE, CHRNA4 , and SLC5A7 are related to the availability of acetylcholine. BDNF and HTR4 are related to the availability of serotonin while MAOA is related to the availability of both dopamine and serotonin. Knowledge about how these markers affect attentional development is sparse (Konrad et al., ; Dye and Bavelier, ) and so a study with a longitudinal component is likely to be helpful.
### Participants
Participants were recruited after Institutional Review Board (IRB) approval from Rice University (IRB-Human subjects) and the University of Wisconsin-Madison (the Social and Behavioral Sciences IRB). Additional data analysis was performed under IRB approval from Brigham Young University (the IRB for Human Subjects). We invited 345 eligible children who lived in the Madison, Wisconsin area to participate in the computer task study from a population of children who had participated in studies in the Dannemiller lab at the University of Wisconsin-Madison between 1996 and 2001 when they were infants (ages 2–5 months). The infants were full term (±2 weeks) and had normal birthweight (>2500 g). The children were 9–16 years old at the time of our second contact.
Of the 345 invited to participate, 203 participated. All parents gave written informed consent in accordance with the Declaration of Helsinki. Children signed assent forms. The children completed two computer tasks while their parents filled out a questionnaire. The children also provided a saliva sample. The leukocytes (white blood cells) and buccal epithelial cells (from the inner cheek) found in saliva were then used to obtain DNA suitable for genotyping. Two children were excluded for neurological diagnoses and two children were excluded for uncorrected vision diagnoses. After exclusions, we have complete data at two time points for 199 children: one task in infancy and another with the same subjects in childhood.
### Infant Task
Infants were shown a display (Dannemiller, ) with vertical bars in various configurations (see Figure ). One of these bars (either on the right or the left side of the screen) oscillated in place horizontally through a visual angle of approximately 0.5–1° at a rate from 1.2 to 2.4 Hz. The rest of the bars remained static. An observer who was unaware of the location of the moving bar made a forced choice of the location of the moving bar (right vs. left) based on the infant’s initial orienting behaviors (e.g., eye movement, facial expression, head or body movements). The judgments were made quickly after the onset of the display with the average judgment taking less than 2 s. Infants who looked selectively at the side of the display with the moving bar were judged (in an age-adjusted manner) to have better reflexive attention since reflexive attention is captured by moving stimuli.
Monochromatic representation of infant visual display. The actual presentation had red and green bars.
### Child Task
The child task is similar to the infant task except that the bars were monochromatic because contrast and not color were found to be important in the infant task. The static bars are expected to exert potentially distracting effects on eye movements but the moving bar is expected to capture attention most of the time. E-Prime (Psychology Software Tools, Sharpsburg, PA, USA) was used to present stimuli for this task. Participants were digitally recorded with a low-light video camera in a semi-darkened room while performing the task. There were eight practice trials and 16 actual trials. All trial types (display conditions with a moving bar on either the left or right side of fixation and with various placements of distractor bars) were presented in the same order for each participant. All participants were instructed to initially look at a fixation cross (presented as black on a white background) and told that a field of bars would appear (we used a medium gray background with lighter and darker gray bars; see Figure ). Each child was instructed to look directly at the moving bar as soon as he/she saw it begin to move. The E-Prime code to initiate the onset of movement in one of the bars triggered the initiation of a time stamp imprinted on the digital video file by a FOR-A timer (Tokyo, Japan). The timer is accurate to the nearest 1/100th of a second, but was recorded to the nearest 4/100th of a second during video conversion (required for use of the video editing program described under Preparation of Child Eye Movement Data, below). Two raters coded the 16 trials of each subject for latency (RT) and eye movement direction. Information about eye movement direction was used to later determine PC scores.
### Questionnaire
While the children completed the computer task, the parents completed a questionnaire that included the McArthur Health and Behavior Questionnaire-Parent Version (HBQ-P; Essex et al., ; Armstrong et al., ). We used a modified form of the HBQ-P that asks parents to use a three-point Likert scale (0–2) to indicate agreement with statements about the behavior of their children. The modified version has been successfully used in a variety of studies (Lemery-Chalfant et al., ; Burghy et al., ). We examined only the summary scores for Inattention as a follow up to trajectory analyses by genotype. The questionnaire also included demographic information (including family income and family size, which we used to approximate SES) and a modified version of the Epworth Sleepiness Scale that is appropriate for children. The Epworth is a simple rating of sleepiness during certain hypothetical, common events (Melendres et al., ). In the data analysis, we attempted to determine if either SES or sleepiness (which is associated with not getting enough sleep; Saarenpää-Heikkilä et al., ), changed the significance of the model in which MAOA predicted changes in RT from infancy to childhood.
### Genotyping Methods
Participants produced a saliva sample of approximately 2 ml in an Oragene-500 kit (DNA Genotek, Kanata, ON, Canada), which was used for genotyping. The Oragene-500 is a plastic test tube with a preservative that is released when the lid is closed. Genotyping for the study was performed using the GoldenGate assay on the BeadXpress system (Illumina, Inc., San Diego, CA, USA). Briefly, the GoldenGate assay involves biotin-labeling of genomic DNA followed by the capture of the labeled DNA onto streptavidin-coated sepharose beads. Streptavidin has a very high affinity and specificity for biotin and thus aids in labeling. Sepharose is a form of agarose that is commonly used in chromatographic separations of biomolecules. An artificial nucleotide-based molecule that contains universal priming sequences on either end and is complimentary to the target DNA sequence of interest is then created, amplified and hybridized to holographically-labeled silica bars that form arrays with up to 30-fold redundancy of each target to be interrogated. Once the array has been visualized with the BeadXpress reader, wavelength and intensity values of the fluorescence are used to determine the genotype. A custom Laboratory Information Management System is used to track both samples and laboratory throughput. Allele detection and genotype calling were performed using GenomeStudio software v2011.1 (Illumina, Inc.).
### Statistical Procedures
#### Preparation of Infant Scores
In the infant data set there is a PC score for each infant, which indicates the percentage of trials that a single observer determined that the infant oriented to the side with the moving bar. Each infant also has an average latency to orient to the moving bar that represents the RT of the observer to make a decision concerning orientation. PC scores are proportions and therefore received a logit transform. Infant scores were corrected for gestational age and study conditions by saving standardized residual scores after regressing the variables on the transformed PC and RT values. These two residual measures were used separately in analyses.
#### Preparation of Child Eye Movement Data
Because the child eye movement data was coded by multiple video coders, we used inter-rater reliability (IRR) analyses to determine if it was reasonable to aggregate data from two raters. We used a video editing software (Pinnacle, Corel Corporation, Mountain View, CA, USA) to analyze the latency and direction of the eye movements. Video coders worked independently and each video was coded twice. Video coders were trained to code the very first eye movement for direction and latency even if the child self-corrected for direction after this first look. Focusing on the first eye movement captures distractibility in the children and provides variability in the data. Raters advanced each trial frame-by-frame beginning with the presentation of the stimulus until the first eye movement was noted. Latency was recorded based on the time stamp at this initial eye movement. Training continued until acceptable IRR (set at 0.80). After averaging raters’ data codes, the eye movement direction was compared to the stimulus presentation side and coded as correct or incorrect.
Agreement on eye direction was very high. Discrepancies (where one rater coded left and one coded right) occurred in 3% of trials. An additional 3% could not be coded and are considered missing data. Overall IRR across all dyads was 0.95 for RT and 0.93 for eye movement direction. Therefore, the two ratings were averaged. For each subject, PC was calculated and RT for correct trials were aggregated (averaged) per person. After averaging across raters, all the RTs for left targets and all the RTs for right targets are again averaged within each subject and these are not meaningfully skewed (0.96), making the observations normal enough to use in statistical procedures that require normal data.
Raw PC and RT scores were adjusted for age and (in the case of infant scores) study number by saving standardized residual scores after regressing the variables on the transformed PC and RT values. Residual PC and RT scores were used separately in analyses. We examined trajectory using mixed (multilevel) modeling in SPSS 22 (IBM, 2013) using the MIXED command. Multilevel modeling accounts for the nested nature of the data (i.e., that trials can be attributed to individuals). We used this method because the intraclass correlation indicated that 75% of the variance came from the individual-level data (level 2) and we needed to take the nested nature of the data into account when analyzing time point data (level 1). Each gene had two to seven genetic markers (SNPs or VNTRs) that were included as predictors because they were not closely linked with other available markers on that gene. We used separate analyses per gene because a longitudinal study with only two time points has difficulty with more complex models with more predictors and it was important that we be able to test interactions with time point to answer our research questions. We performed 11 sets of analyses using a backwards design similar to backwards regression. Each set of analyses included an empty model (with no predictors), a full model (with all predictors and interactions as described below), and a reduced model with no interactions. Final models were determined by comparison to an empty model with no predictors and to the full model with the following predictors: all markers on a given gene, time point, lag time, and (in the case of MAOA , which is on the X chromosome) sex. The only interactions that were included were between genetic markers and time point and (in the case of MAOA ) between genetic marker, time point and sex. Essentially, the multilevel model is fitting a line for each person through the transformed scores of each person and then the interaction between time point and genotype indicates if the trajectory is affected by genotype.
We entered each set of markers and each markers interaction with the time component (time 1 for infancy and time 2 for childhood) and the lag time between infant testing and child testing as predictors of (in turn) residual RT and residual PC. The interaction between time point and genetic marker indicates different trajectories depending on genotype. The output of interest when examining outcome trajectories over the course of a task is the interaction between the genetic marker and time point. We are looking for time point by marker interactions by modeling transformed RT and PC separately as the dependent variables (the outcome in the multilevel model). We used the lag time as a covariate to account for variability between participants’ two age point differences. A participant in this study was tested once as an infant and once as a child, so we calculated lag time by subtracting age corrected for gestational age (age at the first time they were tested), from the age of the child (age at the second time they were tested). We kept lag time in units of days, as days in the life of a small child can play a major role in physical and cognitive maturity. Then we used the lag time as a covariate to take into account the role of time and hence increases the potential to view with more clarity the influence that genes play in reflexive attention over time.
## Results
### Hardy-Weinberg Equilibrium
We performed chi-square tests to determine if alleles and genotypes were present in the expected proportions according to the Hardy-Weinberg principle (Hardy, ). Non-significant results indicate the absence of increased evolutionary influence such as genetic drift, mutation, or biases in mate selection. These tests were performed using the whole sample, prior to excluding subjects. All SNPs were in Hardy-Weinberg equilibrium (all P s > 0.39) indicating no cause for concern.
### Behavioral Results
Fifty-one percent ( n = 102) of the sample was male. Infants ranged in age from 1.51 to 5.39 months ( M = 3.63) and had raw PC scores from 42 to 98% ( M = 0.72) and raw RT from 1080.52 to 3180.00 ms ( M = 1804.92 ms). The children ranged in age from 10.58 to 16.55 years ( M = 12.93) and had raw PC scores from 44 to 100% ( M = 0.84) and raw RT from 274.52 to 500.71 ms ( M = 361.74). There was a significant correlation between infant and child standardized residual RT scores ( r = 0.29, p < 0.001). This indicates that the infant and the child constructs for RT are similar. However, infant and child PC scores were not correlated ( r = 0.05, p = 0.48). Note that prior to creating the residual PC scores, we adjusted raw PC scores downward so that perfect (100%) scores were converted to 99% scores. This allowed us to take the logit transform and to perform linear-based regression. Ages by sex and residual PC and RT scores are shown in Table .
Dependent variables after controlling for age by time point and sex .
Abbreviations: PC, percent correct; RT, reaction time .
### Multilevel Analysis
All genetic models with interactions with time point were better than models without interactions with time point ( P s < 0.001). All models with interactions were significant after correcting for multiple comparisons using a sequential Bonferroni procedure (Benjamini and Hochberg, ). Adjusted alpha levels ranged from 0.005 to 0.05. We examined Schwarz’s Bayesian Information Criterion ( BIC ) to compare models using chi-square statistics. Smaller BIC values are better and the chi-square value indicates if a model is significantly better. Once the best model was determined, individual predictors and interactions were examined for significance using the fixed effects output. Only the MAOA model had a significant interaction of interest that included time point.
As can be seen in Table , the MAOA model with interactions between genetic markers, time point, and sex had the smallest BIC at 705.38 and was significantly better than the empty model ( BIC = 1015.65), = 246.55, p < 0.001. This p -value (5.66 × 10 ) is substantially less than the family-wise alpha of 0.004. The model with fewer interactions ( BIC = 705.83) was a little worse than this model. Neither the model with SES ( BIC = 712.76) nor the model with sleepiness ( BIC = 706.76) were better than the model with interactions between the MAOA markers, time point and sex.
Parameter estimates from models predicting residual RT with MAOA .
**Significant at p < 0.01. *Significant at p < 0.05. Significant values indicate that the estimate is different from zero. We report results under tests of fixed effects for gene by trial interactions to determine if trajectories differ by genotype. This parameter has been set to zero. Note that this Table shows model parameters when testing residual RTs .
Examining the best model, we see that an interaction of interest is significant, F = 12.50, p < 0.001. Multilevel modeling is regression based, and so this indicates that even after controlling for all other predictors and interactions in the model, the trajectory of residual RT varies according to sex and genotype on the rs1137070 SNP of MAOA (see Figure ). This information is also presented in tabular form in Table , which includes the genotype counts by sex.
Mean residual reaction time (RT; controlling for age) represents the subject’s position in the distribution of infant or child scores at each time point. Because Monoamine oxidase A (MAOA) is on the X chromosome, it is only possible for boys to have one of two genotypes. Boys with either genotype decrease (improve) in their relative RT between infancy and childhood. Girls with the CT genotype decrease in their relative RT but girls with the CC genotype increase in their relative RT.
Means and standard deviations of RT and residual RT scores by time point and sex .
Note. Age was controlled by residualizing RT within time point and saving the standardized scores. Residual RT can be interpreted as a Z-score .
There were also main effects for lag time ( F = 860.64, p < 0.001), sex ( F = 4.98, p = 0.03), and time point ( F = 488.92, p < 0.001)—but not for the other MAOA SNPs ( P s > 0.21) or their interactions with time point and/or sex ( P s > 0.22). In the interaction with rs1137070, boys in either the zero (T genotype) or one (C genotype) risk-allele group improved their position in the distribution of boys and girls (because the residual RT scores can be interpreted as z -scores controlling for age with lower RTs indicating increased speed). Girls with zero risk alleles (the TT genotype) approximately retained their position in the distribution. Those with one risk allele (the TC genotype) improved their position in the distribution, and those with two-risk alleles (the CC genotype group) sharply increased (worsened) their position in the distribution. It is interesting that girls with TT or CT genotypes start out differently, but converge on similar RTs. This finding is unusual; however, it possibly indicates heterozygote disadvantage (aka underdominance). This phenomenon concerns situations where homozygotes are associated with a more beneficial phenotype but heterozygotes with a less desirable phenotype. This is the reverse situation to that of sickle cell anemia, in which heterozygotes are at an advantage. Two possible human examples of heterozygote disadvantage in humans are Rh factor incompatibility, which can lead to hemolytic anemia in a fetus (Cavalli-Sforza and Bodmer, ), and polymorphisms in the MTHFR gene, which influence embryo implantation (Enciso et al., ). While less fit heterozygotes may eventually become rarer, this is likely to take at least 100 generations (O’Fallon and Adler, ; Peischl and Bürger, ) and takes longer under certain conditions. For example, the stable heterozygote disadvantage is more likely when the heterozygote state has a weak influence on survival, the same allele leads to both non-advantageous and advantageous traits (i.e., is pleiotropic), and it is linked to other loci (Wilson and Turelli, ; Eppstein et al., ; Lawson et al., ). This is interesting because MAOA is a gene with high linkage between loci. In addition, the change we witnessed in the distributional position of RT does not appear to have strong influence on survivability, and (due to the action of the MAOA enzyme) there may be some biological advantage to having a slightly more dopamine available in some situations (Cools and D’Esposito, ). Nevertheless, this explanation should be treated with caution until this study is replicated and shows a pattern suggesting heterozygote disadvantage for RT distributional position on a similar attention task.
As a follow up analysis, we examined if the girls in the CC genotype group had more symptoms of inattention as measured by the HBQ-P. There were more inattention symptoms ( F = 3.29, p = 0.04). This indicates that girls in the CC genotype group are being identified by parents as having trouble with attention and are declining in task performance between infancy and childhood relative to their peers.
## Discussion
### Summary and Implications of Results
We found evidence that the SNP rs1137070 on MAOA predicted poorer developmental course in RT such that girls with the CC genotype show an increase in (slowing of) RT between two time points (infancy and childhood) on a reflexive attention task. In contrast, the analogous genotype on MAOA was not associated with slowing RT across development in boys. The effect of genotype on RT remained even when controlling for SES and sleepiness. We also note that our findings are unlikely to be related to IQ because the children’s academic reading scores (which are frequently used to approximate IQ; Manolakes and Sheldon, ; Kaufman et al., ) were not associated with their positional change in the distribution of scores from infancy to childhood. We also note that a trend for this significant interaction ( p = 0.12) remained when we removed likely prepubertal children (males younger than 11.66 years; all females had likely already begun puberty [i.e., were older than 9.86 years]; Lee and Styne, ). The drop in significance probably represents a loss of power with a smaller sample size. Thus, we do not believe that the prepubertal children are driving significance.
Growing evidence suggests that the MAOA genotype may play a role in attention and cognition (Guinmarães et al., ; Wargelius et al., ; Ernst et al., ; Nymberg et al., ; Piton et al., ). However, to gain confidence in these findings, the behavioral phenotype needs to make sense with biological actions of MAOA . The MAOA gene encodes an enzyme that degrades amines such as dopamine, norepinephrine and serotonin (Garrett and Soares-da-Silva, ). The rs1137070 SNP (formerly rs1801291 and aka codon 1460) is in exon 14. T alleles have been found to be associated with increased MAOA activity (Hotamisligil and Breakefield, ), and may therefore regulate gene expression (Zhang et al., ). Increased levels of MAOA (the enzyme) lead to decreased levels of dopamine. Also, the serotonin system inhibits the firing of the dopaminergic system at the midbrain (Jacobs and Azmitia, ), therefore decreasing the likelihood of activating the frontal dopamine system.
We gain confidence in our finding because it is consistent with previous longitudinal studies showing an association between MAOA and negative outcomes that could have a cognitive component (Edwards et al., ; Enoch et al., ; Belsky and Beaver, ; Daw and Guo, ; Fergusson et al., ; Lee, ; Hill et al., ; Pickles et al., ; Priess-Groben and Hyde, ; Haberstick et al., ; Whelan et al., ). Such findings establish heterotypic continuity (Putnam et al., ; Miller et al., ; Lavigne et al., ) between the infant reflexive attention task and later child behavior.
To the best of our knowledge, however, there has only been one other longitudinal molecular genetic studies of attention children that included MAOA (Zohsel et al., ). This study found that the MAOA VNTR plays a role in determining continuity of parent-rated attention problems during adolescence. Attention problems during early adolescence (at 11 years old) were found to be strong predictors of attention problems in middle adolescence (at 15 years old). However, stability of attention in carriers of the low-activity variant (MAOA-L) was higher than in carriers of the high-activity variant (MAOA-H).
Our finding that MAOA (but not the other genes tested) is associated with RT changes between infancy and childhood suggests that MAOA has a particular role in the development of attention. This role might involve MAOA ’s function in the pre-synaptic membrane release of dopamine (Yu et al., ). In addition, the role could involve MAOA ’s involvement in degrading monoamines such as dopamine in the central nervous system (Seif and De Maeyer, ). These roles could change developmentally over the age range we examined. For example, if dopamine drives axon guidance, cell mobility, and synapse formation early in life, but shifts to synapse function and messaging later in childhood (Gilman et al., ; Douet et al., ), then the change in roles could lead to individual differences in attentional trajectory, which might be adequate at one stage of development but not at another.
Examining the positional changes in the distribution is important because of the large developmental differences between infancy and childhood. For example, note in Table that the absolute magnitude of the differences between infancy and childhood is large in terms of raw RT (the range is from 1377 to 1467 ms, depending on genotype groups compared). This is expected because children are much faster at responding than infants (or, in this case, an observer who is judging infant responses). The large difference is one reason we used standardized residual values as the dependent variable. However, the difference in improvement (infant RT—child RT) is much smaller (varying from 23 to 90 ms depending on genotype groups compared). Note that these values are in the range of noticeable differences (Shackleton et al., ).
Also in Table , it is apparent that the CC genotype group is small (11 girls). This might suggest one possible reason for the group differences in standardized residual RT. However, small group size generally increases variability and decreases the ability to detect a significant difference. In addition, notice that the CC group is the least variable of the three genotype groups for girls.
Our finding that the SNP rs1137070 on MAOA predicted poorer developmental course in girls with the CC genotype does not detract from the relevance for the other genes in association at a single time point with attention or cognition. Several of the genes we tested, including BDNF , CHRNA4 , COMT , DRD4 , HTR4 , SLC6A3 , SLC5A7 , and SNAP25 , have been associated with ADHD. In our study, all 11 genes were associated with task performance in terms of RT. They simply were not associated with trajectory, which is the interaction between time point and genetic marker that we set out to test. It appears, therefore, that these genes do not influence the development of change in position in a distribution of RT scores after controlling for age. Another possibility is that these genes do influence RT score position, but that our sample size was not sufficient to detect the size of the effect. If this is true, it is still important to note that the effect was stronger with MAOA .
In addition to explicating the mechanisms by which symptoms of inattention emerge in a general population of children, there are some intriguing clinical implications of our findings. First, if young girls with the CC genotype on rs1137070 can be identified early, they can be monitored in a new longitudinal study to determine more precisely the course of their development and concomitant risk factors such as those involved with various life experiences at home and school. This study should involve many time points and verify the genes involved in predicting risk and the eventual development of inattentive symptoms. Although it is unclear how reflexive attention and sustained attention are related (Hikosaka et al., ; Suzuki and Cavanagh, ; Barry et al., ; Berger et al., ; Andrade et al., ; Henderickx et al., ; Carrasco, ; Macaluso and Doricchi, ; Underbjerg et al., ; Dye and Hauser, ; Anderson, ), my previous work has indicated that reflexive attention is related to day-to-day attentional performance (Lundwall, ). Second, in conjunction with the above, several comparison intervention programs could be targeted earlier to determine which are most effective. An intervention study would also move our knowledge from observational to causal. Clinical resources such as the type and intensity of intervention ideally need to be tailored according to individual differences. Predictors, both clinical and developmental, could shape resources to become more effective. In addition, identifying genetic and environmental risk factors that influence the developmental course of inattentive symptoms is important because it identifies the mechanisms that lead to poorer clinical outcomes.
Information on the development of attentional pathways and other cognitive development comes from studies of children with ADHD. Marx et al. ( ) tested the domains of working memory (prefrontal cortex) interference control (frontocortical circuits), time perceptions (frontostriatal-cerebellar circuits) and delay aversion (striatal-limbic circuits) in male children, adolescents, and young adults. Although the study was cross-sectional, the authors conclude that cognitive deficits in the domains tested tended to persist across the lifespan. Others have noted that cognitive symptoms of ADHD tend to persist even when the behavioral symptoms resolve (Biederman, ; Biederman et al., ) and that clinical improvements tend to parallel brain volume normalization (Giedd and Rapoport, ). Because ADHD symptoms exist on a continuum in the general population, this suggests the possibility of similar patterns to those described above in children from the general population, which seems to be confirmed by the persistence of the influence of MAOA on attention in a general population of children (Zohsel et al., ). There were no studies extending these findings into young adulthood, and this an important goal of future research.
### Study Limitations and Future Directions
Our study has several limitations. For example, three time points would provide more certainty in the time trends we found. We do note that multilevel modeling of two time points has been used successfully by other researchers in other fields (Bere et al., ; Hearst et al., ; Normand et al., ; Murray et al., ; Nicholson et al., ). However, future replication attempts should include more time points and it would be especially helpful to continue this study into young adulthood. In addition, both time points and sample size matter in determining the complexity of the models that can be tested with the data. The sample size ( N = 199) was somewhat small once genetic subgroups are formed. Scherbaum and Ferreter ( ) discusses several factors related to power in multilevel modeling, which suggest that a minimum of 30 individuals each with 30 time points be used for power to detect cross-level interactions (the kind of interactions we are testing), but indicates that a larger number of individuals can compensate for a smaller number of time points. Because we found an association, this suggests sufficient power. Nevertheless, additional subjects would increase confidence in the results of any replication.
Another limitation concerns the possibility of measurement variance. Measurement invariance is sometimes an issue because the construct of reflexive attention has different meanings given the rapid physical, cognitive, and socio-emotional development between infancy and childhood (Glück and Indurkhya, ; Bontempo et al., ). For example, the development of language means we can give instructions to children that infants cannot follow. In this case, the measurement variance (or construct shift) refers to the lack of similarity between PC measures of attention at the two ages. Infant and child RT scores were correlated. This is, admittedly, a very basic look at establishing that the constructs were substantively similar. A more complete analysis would involve multiple measures of the construct of attention at each age and structural equation modeling. This was not possible with the available data. Tracking construct shift at multiple ages would allow for more through interpretation of any shifts that do occur, such as those that seem likely for the PC scores. Another possibility is to use a few different age-appropriate tasks at each time point in an overlapping design and use structural equation modeling to track constructs through time. This method was successfully used by Petersen et al. ( ). Thus, there are ways to investigate the meaning of the constructs across development in future studies.
It is also important for future studies to expand on our Caucasian sample. This was intentional to avoid issues with population stratification, which is simply confounding between genotype and phenotype that can occur when ancestral heritage influences both. However, using only Caucasians does limit generalizability to other groups. The best way to handle this is by replicating this study in each of the other ethnic groups available in a researcher’s region.
Future replication efforts that identify genetic and other risk factors that influence the developmental course of attention is important for several reasons. First, it is important to identify the mechanisms that lead to poorer clinical outcomes. Second, clinical resources such as intensity of followup ideally need to be tailored according to need. Predictors, both clinical and etiological, could thus help target resources more effectively. The identification of non-genetic factors and how they may interact with genetic variants in influencing the developmental course is also an important area for future research, although larger longitudinal samples will be needed when testing for Gene × Environment interactions.
## Conclusion
In conclusion, we find evidence that MAOA is associated with different developmental patterns from infancy to childhood (over a rather long time span of 15 years). Importantly, we also find evidence that the effect of this gene on attention can vary as a function of sex. Identifying how specific risks are associated with identified genes will be necessary to advance our ability to design more effective prevention and intervention programs for individuals at risk. These analyses underscore the importance of studying genetic effects across development and of identifying factors that influence risk.
## Author Contributions
RAL designed the study, collected the data, supervised the statistical analysis, and wrote the manuscript. CGR performed the statistical analysis and wrote portions of the manuscript. Both authors reviewed and approved the final manuscript.
## Funding
The research was supported in part by grants from the Social Sciences Research Institute at Rice University (Dissertation Improvement Grant to RAL and Seed Money Grant to JLD) and by the Lynette S. Autrey Fund (to JLD). Infrastructure support was provided by the Waisman Center (University of Wisconsin-Madison) via a core grant from National Institute of Child Health and Human Development (NICHD; P30 HD03352).
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
The functions of dance and music in human evolution are a mystery. Current research on the evolution of music has mainly focused on its melodic attribute which would have evolved alongside (proto-)language. Instead, we propose an alternative conceptual framework which focuses on the co-evolution of rhythm and dance (R&D) as intertwined aspects of a multimodal phenomenon characterized by the unity of action and perception. Reviewing the current literature from this viewpoint we propose the hypothesis that R&D have co-evolved long before other musical attributes and (proto-)language. Our view is supported by increasing experimental evidence particularly in infants and children: beat is perceived and anticipated already by newborns and rhythm perception depends on body movement. Infants and toddlers spontaneously move to a rhythm irrespective of their cultural background. The impulse to dance may have been prepared by the susceptibility of infants to be soothed by rocking. Conceivable evolutionary functions of R&D include sexual attraction and transmission of mating signals. Social functions include bonding, synchronization of many individuals, appeasement of hostile individuals, and pre- and extra-verbal communication enabling embodied individual and collective memorizing. In many cultures R&D are used for entering trance, a base for shamanism and early religions. Individual benefits of R&D include improvement of body coordination, as well as painkilling, anti-depressive, and anti-boredom effects. Rhythm most likely paved the way for human speech as supported by studies confirming the overlaps between cognitive and neural resources recruited for language and rhythm. In addition, dance encompasses visual and gestural communication. In future studies attention should be paid to which attribute of music is focused on and that the close mutual relation between R&D is taken into account. The possible evolutionary functions of dance deserve more attention.
## The Evolution of Dance and Music, Current Concepts
The origin of dance and music, beautiful and powerful universals of humankind is a mystery. All over the world there are myths on how humankind received dance and music. In Hindu mythology, the god Shiva Nataraj created the world by dancing. In most traditional cultures dance plays a pivotal role ( ; ; ; ; ; ; ).
In the Western world attention has been usually paid to the origin of language and to its relation to the melodic attribute of music whereas dance and rhythm have been, for long, neglected. This might partly be an unintentional consequence of the duality of body and mind concept of Cartesian philosophy as well as the historical hostility of the Roman Catholic and Protestant Churches toward dance. This had led to the omission of percussion instruments in European classical music, thus, diverting the first attention from rhythm and dance to melody ( ; ; ).
The evolution of music has become an important research topic rather recently ( ; ; ; ; ; ; ; ; ; ; ).
The evolution of music has become an important research topic rather recently. The interest is partly awakened by neuroscience, basically to identify the core components of human cognition ( ; ; ; , , ; ; ; ; ; ) in comparison to animals ( , , ; , ). The evolutionary role of dance is even more enigmatic than that of music considering that who dances dispenses considerably more energy than a singer or a musician. The evolutionary functions of dance have received more attention only recently ( ; ; ; ; ; ; ; ; ; ; ).
### Definition of Dance
Neither the term dance nor the term music as such are precise. “Dance” in Oxford’s dictionary is defined as: “move rhythmically to music, typically following a set sequence of steps.” For our purpose we define as “dance” body movements coordinated to a basic rhythm. Rhythm is constituted by a pulse or sequence of beats which are organized hierarchically. There are four main sub-constituent elements of rhythm: (1) tactus represents identical short-duration periods subdivided into strong beats (“downbeats”) and weaker beats (“offbeats”); (2) tempo: the frequency of the tactus; (3) meter: cyclical groupings of beats into units marked by accents; (4) patterns: sequences of time intervals that may or may not extend across meter units ( ; ). Dance differs from simple synchronization to a simple regular pulse, because dance offers the possibility to vary steps with respect to beats inside the tactus. Nevertheless, the dancer has to respect a basic groove ( ; ; ). We would therefore not consider soldiers marching or harvesters working in synchrony with a beat as dancers: they do not differentiate down- and offbeats of a rhythm and they have a defined purpose. We also see modern non-rhythmic expressive dance as theater rather than dance. We would also not describe as dance the repeated steps without keeping a regular basic pulse as described for some birds ( ).
On the other hand, we extend our definition of dance to the beat-keeping movements of music performers. Embodied perception is the physiological fundament of this phenomenon. Unintentional body movements to a beat reflect the role of our body in rhythm perception. For dancing, the capacity of beat anticipation and of embodied rhythm perception are required ( ; ; ).
### Conceptions of Music and Theories on the Origin of Music
“Music” and “dance” encompass overlapping spectral and temporal attributes. Spectral attributes are pitch, intervals, and harmony. Temporal attributes are covered by rhythm consisting in its sub-constituent elements tactus, meter, tempo, and pattern ( ; ; ; ; Table ). Loudness and dynamics are, albeit important, not specific for music since these are also expressive means of other arts such as theater, poetry, rhetoric or cinema. For music we discern its three main specific attributes: rhythm, melody , and harmony . Rhythm is music’s central organizing structure. Rhythm is indispensable for both, dance, and music ( ; ). Whereas, rhythm can exist without melody or harmony, melody or harmony cannot exist without rhythm. The concepts of melody and harmony are partly defined by their temporal rhythmic fundament: melody is defined as a series of sounds with a different pitch over time. Harmony must be subdivided into “ sequential harmony ” which means the defined pitch intervals in a melodic time line and “ polyphony ” which means simultaneous sounds with different pitches following different melodic lines. Authors of publications on the evolution of music, usually do not mention which of these attributes are precisely meant. Usually, only the melodic and to a lesser extent the harmonic attribute of music have been focused on. On the other hand, rhythm has been relatively neglected. Also in the past, scholars who reflected on the origin of music referred to its melodic attribute. The philosopher argued that ancestral humans would have used a proto-musilanguage and that people would have communicated by singing. The German poet interpreted melodic music as a precursor of language. , p. 572) in “The descent of man” noted: “as neither the enjoyment nor the capacity of producing musical notes are faculties of the least use to man in reference to his daily habits of life, they must be ranked amongst the most mysterious with which he is endowed.”
Relations and overlap between dance and music attributes.
Most anthropological, psychological, and musicological references focus on the evolution of melodic music but not of rhythm nor of dance. A favorite theory on the evolution of music has been that it would have evolved just as a by-product of language evolution ( ; ; ).
### Current Hypotheses on the Evolution of Dance
Would also dance constitute a mere by-product of language evolution? It has been proposed that the capacity to move in time with an auditory pulse, i.e., entrainment would have evolved as a by-product of vocal mimicry ( ). We believe that R&D are not mere evolutionary by-products of language. Scientific studies showed that infants are able to extract and anticipate a rhythmical pulse and that they have a strong impulse to move spontaneously when exposed to an external rhythm ( ; ; ; ; ). We hypothesize that R&D are part of an inborn series of physiological reflexes universal in all humans. The observation that children at a certain age inevitably dance when they are exposed to a rhythm has been supported by a cross-cultural questionnaire study conducted in three continents by our group ( ). If such a reflex was constant and genetically determined, it could join the list of physiological reflexes used in developmental psychology and pediatric neurology. The confirmation of the existence of such an innate reflex would also support the concept that dance have had a considerable importance in the evolution of humankind.
This prompted us to perform a literature search in the fields of medicine and developmental psychology, philosophy, archeology, anthropology, ethnology, and musicology focusing on the evolution of dance in both, children and in humankind. We were surprised about the scarcity of references. Neither in textbooks of pediatrics nor of developmental psychology a physiological dance reflex or dance reaction to a beat occurring during infantile development is mentioned ( ; ). A PubMed search on “dance reflex” yielded a total of 22 hits, of which none dealt with our topic.
### Music and Dance as Pre-verbal Communication Tools
What could be the evolutionary functions of rhythm and dance? One has to bear in mind that people who dance, stomp, and clap hands are noisy and less aware of predators ( ). Such a behavior must not, at least, have had survival disadvantages if it was to be conserved in evolution.
formulated the theory that long before the evolution of language humans were communicating by extra-verbal means which he called mimesis. Mimesis can be imagined as how we communicate in a foreign country without any knowledge of the local language. This theory was later taken up by who called extra-verbal communication the “hmmmm”-communication (hmmmm stands for holistic multi-modal manipulative musical). In this context argued that music would have been a means of pre-verbal communication, calling his book “The Singing Neanderthals.” This title reflects the wide-spread opinion that melody is the main attribute which characterizes music.
Since we believe that rhythm is the fundament of music we would propose for the next book the title “The Swinging Neanderthals.” sustains the view that both language and music evolved from a pre-linguistic communication system which was neither language nor music. Similarly, proposes that music has evolved from a pre-linguistic precursor present also in animals which he called “contagious heterophony.” An alternative account is proposed by where music originated from a more general adaptation known as the “Theory of Mind” which would allow an individual to recognize the mental and emotional state of conspecifics. Underpinned by the mirror neuron system of empathy and imitation, music would achieve engagement by drawing from pre-existing functions across multiple modalities ( ; ). This, in our opinion, applies even more to dance because of its strong interaction between perception and motor response. puts forward the hypothesis that song and dance would have even preceded mimesis in hominid evolution. Since most authors looked for the origin of music in melody and singing, some of them interpreted as their precursors the interaction between pre-musical utterances of the infant by modulation of crying and melodic modulation of language by the caretaker, also called “motherese.” The origins of “motherese” would already be established during prenatal development of the fetus ( ). “Motherese” has been proposed as proto-melody paving the way for the evolution of music. Rhythm in this respect has not been considered ( ; ; ; ; ; ; ). Furthermore, although all mothers of all cultures and all ages know that babies are soothed by being rocked, this is, if ever, only exceptionally mentioned and has also never been proposed as a human universal ( ; ).
Summarizing, the functions of R&D as pre- and powerful extraverbal communication tools have found little attention.
## Defining A Differentiated Framework of Music and Dance
For defining the evolutionary functions of music a more fine-grained concept is required ( ; ; ). Dance is usually conceived as being distinct from music. The close relationship between rhythm and dance has been acknowledged only recently ( ; ; ; ; ; ; ). The pivotal importance of rhythm, synchronized body movements and dancing as prerequisites for the emergence of all attributes of music has found relatively little attention. Whereas, rhythm and dance can exist without melody, there is no music without rhythm ( ). Music has also been considered for a long time from the viewpoint of a unimodal phenomenon (auditory processing), whereas, it is multimodal (through action–perception coupling; ). There is no reason to assume that the different attributes of music evolved at the same time and pace. Only two attributes of music constitute human universals: all cultures have rhythm and almost all have melody (songs; Table ). Polyphonic harmony has developed only some 1000s of years ago and has been explored only since historic times starting in Ancient Greece ( ). Among the attributes of melody and sequential harmony only the octave, the perfect fifth and building a melodic scale by the division of the octave into unequal intervals are human universals [ ; (see also Table )]. Many cultures such as in the Arabic world, in Turkey, in Persia or in India, have, instead of polyphony, pursued to evolve melody by adding particular intervals and bending melody sounds in a sequential harmonic way (modal music).
On the other hand, in most cultures, worldwide, the concept and terms of R&D are not separated from each other (e.g., the terms “Samba,” “Salsa,” “Guaguancó,” “Tango,” “Waltz” apply to both, to the rhythm and to the dance).
In line with this, we propose an alternative conceptual framework which focuses on the mutual co-evolution of R&D ( ; ; ; ; ). In other words, R&D are two sides of the same coin. Not only do we move to what we hear but what we hear depends on how we move ( ). The unintentional body movements when we perceive a “groove” (defined as the aspect of music which compels us to move), among all body parts primarily involve the lower limbs confirming the close relation to dance ( ). The mutual connection between R&D is also reflected by the musical terms “downbeat” and “offbeat,” where the downbeat indicates the dance step which carries the weight of the body when it comes back down onto the legs.
Movement, rhythm and emotional well-being have particular neural pathways involving cerebellar structures which coordinate sensory neuronal inputs with motoric responses ( ; ; ; ; , , ; ; ; see the comprehensive reviews of ; ; ). Even deaf children have the impulse and the capacity to dance by perceiving the beat through their cutaneous pallesthesic and visual receptors ( ).
There are relatively few references on the specific evolutionary functions of rhythm in humans, mainly in comparison with non-human animals ( ; ; , ). Rhythm cognition depends on body movement and vice versa ( ; ). emphasizes that necessary prerequisites of research on rhythm are unambiguous definitions of the terms “rhythm” and its sub-constituent elements “beat,” “pulse,” “meter,” “tempo,” and “tactus” ( Table ). Dancing requires the perception of such hierarchically structured rhythms in order to coordinate and differentiate those steps which carry the whole bodyweight (usually the downbeats) from steps carrying less or no weight (offbeats). Sequences of arbitrary pulses are therefore spontaneously classified into a rhythmical tactus by the dancer, a capacity observed already in infants ( ).
For people of traditional societies the importance of dance and the tight connection between R&D are beyond any doubt ( ; ). In societies with strong dance traditions, singing out of tone is tolerated more easily than drumming only slightly out of beat. This is what the composer Duke Ellington meant with his song: “it don’t mean a thing if it ain’t got that swing.” Instead, passive listening to music without moving, as seen in listeners to Western classical music, requires an educational effort, as it can easily be seen when children are obliged to sit still in a concert. In fact, spontaneous unintentional body movements when listening to a rhythmical beat are difficult to suppress ( ; ).
Moreover, dance is a comprehensive art encompassing attributes which go beyond music such as its external visual signals. The dancer is seen by others acting as a moving picture ( ; ). This also applies to a widespread and very old form of dance, i.e., round dance, where the group dances and different dancers enter into the round to perform their solo before rejoining the round as it is also commonly observed in children. Furthermore, dance may encompass gestural and dramatic codes and, thus, has paved the way for the development of theater. The comprehensiveness of dance made already Curt Sachs argue that dance would be the mother of all arts ( ).
## Connecting Studies in Various Scientific Disciplines with Hypotheses on the Evolution of Dance and Music in Humankind
### Evidence of Dance and Music in Archeological Records
When we look for archeological proof for music and dance we must bear in mind that archeology depends on the finding of artifacts indicating the presence of a certain human behavior at a certain time. The earliest artifacts confirming musical activities are around 45,000 years old. Examples are preserved instruments such as flutes in the neolithic caves of “Hohle Fels” and “Geissenklösterle” ( ). Before, we assume that music and dance did not develop earlier than this, we must acknowledge that in a hunter-trapper-gatherer society not only any artifact constitutes an additional weight to carry but not to leave one’s traces also constitutes part of the survival strategy. Moreover, many instruments are natural objects, such as conch shells, or are made of perishable materials, such as wood and animal skins, which are not preserved for long ( ). Furthermore, humans always possessed a versatile instrument without the necessity of producing a musical instrument, i.e., their own body ( ). Some examples of human behavior appear so omnipresent and obvious to us that it has, in fact, never been investigated if these are universal in all human societies. One of these is accompanying a beat by clapping hands, a behavior which is observable in humans of all ages ( ; ; ). Another actual example of body percussion and dance is “Flamenco” which in its original form was performed only by singing, clapping hands and stepping on the ground without the use of any musical instruments ( ). This practice is also illustrated by the “Akonhoun” dance tradition of Benin, where dancers perform percussion on their body while dancing as well as “Schuhplattler” in German folk music. To support our view, one may consider in analogy the development of painting, where the human body was the first canvas as confirmed by ca. 200,000-year-old red ochre findings in several places where Neanderthals were living ( ; ). Cave paintings and sculptures confirming dancing are relatively recent. The oldest cave paint possibly representing a dancer is the around 35,000-year-old “magician” (French: “sorcier”) of the “trois frères cave” in Southern France, a zoomorphic figure with animal and human characteristics. Such mask dancing is still practiced in traditional societies aiming at being possessed by an animal spirit ( ; ). Unequivocal dancing scenes are represented in paintings in the “Valcamonica” and “Addaura” caves in Italy which are no older than 10,000 years ( ).
### Studies on Infants and Children
Studies on infants, toddlers, and children contribute to elucidate the evolution of dance and music. The study group of Henkjan Honing showed in a very elegant experiment that newborns already perceive and anticipate musical pulses, a phenomenon which was called “beat induction” ( ). Beat processing has been shown to be pre-attentive for metrically simple rhythms with clear accents ( ). Three- to four-month-old infants demonstrate spontaneous limb movements coordinated to a musical pulse ( ). Furthermore, infants are able to stratify musical pulses into meters ( ). This capacity is linked with body movement. Infants use meter to categorize rhythms and melodies and learn more readily to tune into musical rhythms than adults ( ; ). Human infants spontaneously engage in significantly more rhythmic movement to music and other rhythmically regular sounds than to language ( ). The precocity of beat induction already being observed in newborns, the ability to stratify musical beats into tactus and meters, the efficacy of rocking, the unintentional movements to rhythm, the pre-attentive characteristic of beat processing, and the intensity of the emotional impact of R&D on humans, support our view that in human evolution communication through R&D preceded verbal and melodic communication ( ; ; ; ; , , ; ; ; ; ; ; ; ; ). Not only environmental but also genetic factors have been shown to play a role in our ability to perceive rhythm ( ).
## What Could the Evolutionary Functions of Rhythm and Dance Be?
With this differently defined framework one may reformulate the question “what was the function of music in human evolution?” in a more particularized way: “what was the function of rhythm and dance?”
Did this module pave the way for further evolution of other attributes of music and of language? Future studies may elucidate to what extent R&D are exclusively human or to which capacities in this respect non-human animals are capable. Also some birds and a captive sea lion are able to anticipate beat and move to it to some extent, but apparently animals do not subdivide rhythm into more and less accentuated beats ( ; , ; ; ; , ). Chimpanzees display spontaneous rhythmical behaviors (drumming and carnival display; , ; ; ; ) and an experiment has shown that a trained chimpanzee was able to anticipate a rhythmical beat ( ). Dancing, however, contrary to marching in lockstep or synchronized working to a beat, requires a sophisticated hierarchical perception of rhythm ( ). Therefore it is, in our opinion, very unlikely that dancing simply constitutes a mere by-product of entrainment. What are R&D in humans good for? Why is the impulse to dance so powerful? What could be the evolutionary functions of R&D?
## Reproductive Fitness and Sexual Attractiveness
Contrary to other scholars of his time, , p. 880) had also the rhythmic aspect of music in mind as well as its reproductive fitness advantage, as he wrote “we may assume that musical tones and rhythm were used by our half-human ancestors, during the season of courtship.” “Dancing is the vertical expression of a horizontal desire” a quote attributed to Robert Frost to which George Bernard Shaw added: “legalized by music,” confirms this observation in poetry. Sexual attractiveness has been since long hypothesized to be the main evolutionary function of dance but has become a scientific research focus only recently ( ; ; ; ; ). Rhythmicity has been proposed as an indicator of mate quality ( ). Furthermore, dancers are able to communicate subtle non-verbal signals ( ; ; ). The Latin-derived French word “emotion” does not by mere chance contain the word “motion” ( ). R&D move us profoundly. Motion alone can effectively communicate emotion, charisma and sex appeal ( ).
To say it with the words of , p. 2): “Dance and sex both use the same instrument — namely, the human body — and both involve the language of the body’s orientation toward pleasure. Thus, dance and sex may be conceived as inseparable even when sexual expression is unintended. The physicality of dance imbued with “magical” power to enchant performer and observer, threatens some people ( ; ; ). The dancing body is symbolic expression that may embody many notions. Among these are romance, desire, and sexual climax.”
Movement quality not only seems to indicate mate quality, but also the interest of a potential partner, which could denote the probability of successful mating ( ; ).
## Social Fitness
### Synchronization of Many Individuals
Synchronization is a behavior not limited to humans ( ). It may have a direct effect on predators or reflect the general advantages of cooperation via positive social interactions, a finding also observed in macaques ( ). Rhythm enables the synchronization of 1000s of dancing human beings such as in a rock concert ( ). The dynamics of rhythmic synchronization differ fundamentally from that of a swarm: a swarm is coordinated by an energy wave passing very quickly but consecutively through many individuals sensing the movement of the adjacent individual. This confounds a predator on which individual prey to catch. Humans, presenting the simultaneous movement of a stomping crowd screaming and armed with fire, may delude a predator by producing the impression of being a homogeneous enormous animal which would be too powerful to attack. This effect may be taken advantage of also in hunting battues ( ; ; ; ; ).
It has recently been argued that self-generated sounds of locomotion and ventilation interfere with the perception of the surroundings. The synchronization of the movement of a number of individuals would thus increase the duration of the intervals where the surroundings can be heard better ( ). This means that synchronization would constitute a by-product of hunting abilities. To prove this hypothesis one would expect that traditional hunters follow animals to hunt in a kind of lockstep, an observation that has not been provided so far.
### Social Bonding
In humans, the synchronic movement leads to “muscular bonding” which enables to overcome emotional boundaries between individuals and, thus, strengthens the community ( ).
A pivotal fitness strategy of hominids is cooperation ( ; ). Drumming and dancing are profoundly social activities ( ). Some of the countless examples of this way of social engagement are “Samba de Roda,” “Flamenco,” and Senegalese “Sabar,” where the audience supports the musicians and dancers rhythmically, members of the audience enter into the round for dancing and the drummers and/or other musicians interact directly with the dancer. In many musical cultures the dancer is a percussionist at the same time, as it may be observed not only in traditional societies but also on ancient Egyptian and Greek frescoes or in actual tap-dancing ( ; ). In many societies dancing is an integral part of important group ceremonies such as initiation rites or weddings. In hunter-gatherer societies, groups may be limited to 40–50 people. The future spouse has to leave her or his group after the wedding in order to join the partner’s group. By dancing, future spouses demonstrate their ability, strength and elegance not only to the future partner but also to other members of the group which will admit the spouse as a new member. In other words, promised spouses need also to catch the eyes of the mothers and fathers in law or other group members who have a say. This latter aspect has, to our knowledge, not yet been explored. Bonding by R&D strengthens the community. Musicians delight dancers. They offer the fundament for the joy of the dancers. This profoundly emotional type of embodied extra-verbal communication increases the group’s cohesion and the identification with the group ( , ; , , ; ; ; ). Music and especially rhythm constitute a deeply rooted signaling system for extra-verbal communication evoking emotional reactions of other potentially cooperating individuals ( ). The propensity to move in time to rhythmic percussive sounds is manifest from an early age on, as seen in children’s impulsive body movement in response to music ( ). Joint drumming facilitates the synchronization in preschool children ( ). Interpersonal synchrony increases helpfulness already in 14 month-old toddlers and the promotion of prosocial behavior by interpersonal rhythmic synchrony has been confirmed in cross-cultural studies in 4-year-old children as compared to matched controls ( ; ; ; ). R&D are also potent collective mood synchronizers ( ; ; ). The emotional impact of the synchronization of many individuals in military drill has impressively been described by .
### Keeping Peace
There is evidence of intra- and intergroup aggression in primates such as chimpanzees ( ), and hominids ( ; ; ). Hominids possessed spears for more than 400,000 years ( ). The advent of tools of potential use as weapons among hominids required even more effective reconciliation means ( ).
To say it with the words of , p. 6447): “The intentional use of implements in the context of intragroup conflict must have had a major impact during hominid evolution because the availability of highly effective hunting and or food-processing tools in interpersonal conflict created a new and considerable potential for intragroup damage, a potential that required specific behavioral adjustments with which to cope. Intragroup aggression in primate societies must be understood as one specific behavioral option in a complex network of social interactions, which is typically balanced by active reconciliatory behavior […].”
This ability is confirmed by the relative scarceness of traces of violence in prehistoric bone findings as compared to skeletons from historic times ( ). Dancing as an effective reconciliatory means has been well-described among potentially hostile Andaman groups by . Dancing enabled to appease our most dangerous enemies: other men of other tribes or even of the own group ( ; ). Similar to symbolic fights present in many non-human animal species, dance may serve for getting to know who is stronger before undertaking a fight, thus reducing the risk of injury and preventing casualties ( ). As an actual example, ghetto dance battles may contribute to avoid deadly duels ( ).
### Dance Rituals, Trance, Shamanism, and Religion
( ) argued that he would not believe in any god unless this god was able to dance. Dance in many societies is not only delightment, but it means also to enter into contact with spirits and gods ( ; ; ; ; ; ; ). Although, trance may in some cultures be also reached without dancing, rhythmical techniques including breathing, hyperventilation and dance, as in the Indonesian island of Bali, are the means which are used in the majority of societies for entering trance. Some historical and actual examples for trance dances include the medieval European St. Vitus’ dance, the Italian Tarantella, the Brazilian Candomblé, the Cuban Santería, the Japanese Nô, the Senegalese N’doep or the Sufi Dervish dances. Trance dance serves as catharsis reached through ecstasy. An ancestor, a spirit or a god drives the dancer; the dancer is possessed. Mask dances are common throughout societies worldwide including Malian Dogon, Japanese Kabuki, Dan acrobats in Ivory Coast, Egungun in Benin and Nigeria. Pre-Christian religious mask dances are the origin of present time Carnival traditions. Dancers moving like puppets on the strings such as in Indian Kathakali and Japanese Kabuki are the precursors of theater and pantomime. In this respect, it is interesting that a 15,000-year-old marionette puppet with moveable limbs has been found in a grave of an adult man believed to be a shaman in Brno, Czech Republic ( ). In Ethiopian and Greek Orthodox Churches people dance for God. It is still matter of debate whether religion is an adaptive complex itself or a by-product of adaptive behaviors in other non- religious contexts. Since there is no evidence of “natural” non-religious control populations, it cannot be excluded that religious beliefs, at least in hunter-gatherer societies might have provided evolutionary advantages ( ; ; ).
### Embodied Pre-verbal Memorizing and Transfer of Traditions
In a pre-verbal context the importance of dance for individual and collective memorizing cannot be overemphasized. Dance in many traditional societies is an instrument to memorize hunting techniques and to preserve traditions by telling stories about the past of the community. In South India Kathakali is danced to tell tales of the Mahabharata epic ( ). Since the mirror motor neurons of who observes dancers are activated dance is an excellent method to train children and adolescents and to communicate experiences and skills which are later internalized by imitation ( ). Also in this function, dance is the predecessor of theater ( ).
### Paving the Way for Verbal Communication
Language might have evolved alongside melody, possibly passing through a “musilanguage” stage as already argued by Rousseau ( ; ). However, the evolution of language requires an underlying rhythmic and gestural understanding, i.e., embodied communication ( ; ; ; ). Rhythm perception enables to discern words and is necessary to codify and decode language. The observation of a dancer aids to recapitulate and decode gestures ( ; ; ; ; ). Thus, it is likely that R&D paved the way for the evolution of language.
## Individual Fitness
### Individual Psychological Fitness
The individual benefits from R&D in several ways. R&D have anti-depressive effects and divert thoughts from sorrows and boredom. Fetuses are able to hear their mother’s physical functions already from the middle of pregnancy on ( ; ; ). Mother’s breathing and heartbeat may produce an incessant conditioning effect which one could describe as a “soothing fetal brainwash.” Up to here, there is no difference between humans and other mammals. Human babies are, however, especially immature at birth as compared to other animals. Therefore, human infants may require more specific soothing efforts such as rocking. To rock the baby one needs free arms. Soothing a baby by rocking is probably a human universal which, however, has not been investigated in this respect. A recent study comparing cultural effects on rocking a baby for soothing showed more similarities than differences between different cultures ( ). There is some research on the effects of rocking in the medical literature. In PubMed we found 157 hits from 1948 to 2014. Especially premature babies benefit from rocking ( ; ; ). Rocking has a positive effect on the entrainment of respiration as well as on neuromuscular development of infants ( ; ). Intuitively, one may assume that experienced caretakers know that rocking is an effective means to soothe a baby, but if we look very carefully at infants’ behavior we appreciate that the infants themselves induce their caretakers to rock them since other means are less effective. Infants and toddlers exhibit also active physiological stereotypic movements ( ; ; ; ). Interestingly, physiological rhythmical stereotypies not only have a self-soothing effect as reflected by heart rate reduction but frequently involve the legs, a behavior that could be the starting point of dancing ( ). In fact, also later spontaneous unintentional movements to a musical beat most frequently involve the lower limbs reflecting an unconscious proneness to dance ( ; ). Whether or not rocking a baby is a behavior strictly confined to humans is an interesting research question which deserves to be explored by evolutionary biologists. We did not find any report of animals or non-human primates rocking their offspring. Moreover, whereas rhesus monkeys have not been found to be good detectors of beat ( ), chimpanzees display rhythmical behaviors ( ).
It is conceivable that the sensitivity of babies for being rocked and physiological stereotypes paved the way for the evolution of R&D in humans ( ). Rocking may also promote the ability of infants to stratify rhythm ( ). Dance enables to self-induce the soothing effect of being rocked. Dance appeases the tormented soul and leads to the secretion of hormones like dopamine and endorphins ( ; ; ; ). The particularly strong emotional impact of R&D is underscored by recent applications in medicine. Their capacity to influence mood, to reach autistic patients otherwise refractory to any emotional involvement and to make Parkinson patients start moving are taken advantage of in medicine ( ; ; ; ; ; ; ). Playing musical instruments and dancing reduce the risk of dementia in the elderly ( ).
R&D enable to divert the otherwise unstoppable flow of thinking ( ). Dance and music playing enable to psychological “flow” experiences that wipe away unpleasant thoughts, sorrows and boredom ( ; ). Boredom may be not only a phenomenon of modern societies but also a problem of traditional societies. Men seem to be more prone to both, boredom and violence, which also are associated with suicide ( ; ). R&D help to overcome boredom and, thus, contribute to keep peace and save lives ( ; ; ).
R&D are particularly powerful means to express the essential “joie de vivre” (joy of life), i.e., the pure “raison d’être” (reason to exist) a philosophical aspect which has been particularly emphasized by Latin-American and African authors ( ; ; ), as Jean Massoulier texted: “Je danse donc je suis” (I dance therefore I am).
### Individual Physical Fitness
Dance, rhythm, music and being rocked have been shown to have painkilling effects ( ; ; ; ). The capacity of music to reduce the dosage of painkilling medication in intensive care patients is documented in medicine ( ). A recent study has shown that rhythmical music reduces the perceived exertion induced by strenuous physical performance an observation which was well-known to the cotton harvesters in the USA and is reflected by specific working songs. This effect occurs not only on a psychological but also on a proprioceptive level ( ). The pain threshold is elevated more by active drumming, dancing or singing than by passive music listening ( ). Rhythmic movements or breathing into hyperventilation are effective means for entering trance, an effect that Hindu yogis take advantage of when they perforate their skin, tongue, or lips before starting their processions.
Furthermore, active and passive rhythmical movements improve body coordination ( ). Although, a major evolutionary advantage is to be expected from cooperation, in some given moments preparation for fighting may be useful for a given group to succeed in winning against enemies and thereby improving the access to resources ( ). Individual and collective coordination skills are trained in martial dances for example in Brazilian Capoeira and Maculelê, in Sicilian Taratatà, Indian Kalaripayattu ( ).
From an evolutionary perspective, all these more or less overlapping aspects are likely to have played a role although these are not equally important at the same time and age. We would tentatively rank reproductive fitness, cooperation and bonding as the driving evolutionary forces whereas the individual aspects may have further contributed to the evolutionary functions of dance in specific age, gender, and prehistoric contexts ( ; ). Survival is particularly important for children in traditional societies with high infant mortality ( ; ). Reproductive fitness applies to sexually mature individuals who may even risk their lives in order to find potential partners. Peace-keeping and martial dancing could have been particularly important for young men during periods of high violence. On the other hand, martial dances are not human universals and high violence periods have been more widespread in historic times than in prehistory ( ; ).
In summary, dance offers evolutionary advantages to humans by contributing to sexual reproduction signaling, cooperation, social bonding, infant care, violence avoidance as well as embodied individual and social communication and memorization. Anticipating one consequence of our R&D concept we would expect that not only beat induction is innate but that during their development infants and toddlers spontaneously start to dance earlier than to express other musical utterances such as singing and that this behavior does not depend on the cultural background of their parents. For further investigating the specific functions of R&D in humans, it would be highly interesting to compare the timing of their emergence during the lifespan of humans with the emergence of synchronic behavior in non-human animals.
## Conclusion
The main intention of this article is to provide a refined concept for further interdisciplinary research on the evolution of dance and music in humankind. It is proposed that in future studies on the evolution of music, attention should be paid on which attribute of music precisely is focused whether rhythm, melody, or harmony. The same applies to rhythmical attributes, i.e., pulse of beats, stronger or weaker beats (downbeats, offbeats), tactus, tempo, meter, and patterns. The evolutionary functions of dance have been relatively neglected. The close mutual relationship between rhythm and dance and embodied rhythm perception should be fully acknowledged in future research.
## Author Contributions
JR did the literature search, developed the hypothesis and wrote the manuscript; RO contributed to the literature search, to developing the hypothesis and writing the manuscript.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Shifting between one’s external and internal environments involves orienting attention. Studies on differentiating subprocesses associated with external-to-internal orienting attention are limited. This study aimed to reveal the characteristics of the disengagement, shifting and reengagement subprocesses by using somatosensory external stimuli and internally generated images. Study participants were to perceive nociceptive external stimuli (External Low (E ) or External High (E )) induced by electrical stimulations (50 ms) followed by mentally rehearsing learned subnociceptive images (Internal Low (I ) and Internal High (I )). Behavioral responses and EEG signals of the participants were recorded. The three significant components elicited were: fronto-central negativity (FCN; 128–180 ms), fronto-central P2 (200–260 ms), and central P3 (320–380 ms), which reflected the three subprocesses, respectively. Differences in the FCN and P2 amplitudes during the orienting to the subnociceptive images revealed only in the E but not E stimulus condition that are new findings. The results indicated that modulations of the disengagement and shifting processes only happened if the external nociceptive stimuli were of high salience and the external-to-internal incongruence was large. The reengaging process reflected from the amplitude of P3 correlated significantly with attenuation of the pain intensity felt from the external nociceptive stimuli. These findings suggested that the subprocesses underlying external-to-internal orienting attention serve different roles. Disengagement subprocess tends to be stimulus dependent, which is bottom-up in nature. Shifting and reengagement tend to be top-down subprocesses, which taps on cognitive control. This subprocess may account for the attenuation effects on perceived pain intensity after orienting attention.
## Introduction
Orienting attention is crucial for enabling individuals to perceive information from the external environment for internal processing such as discrimination and decision making (Lepsien and Nobre, ; Chun et al., ). The external-to-internal process involves disengagement from the external stimuli followed by engagement with the internal representations generated in the mind. Posner’s model of orienting attention (Posner et al., ; Petersen and Posner, ) theorized three steps of orienting attention between two stimuli; namely, disengagement, shifting and reengagement. The three-step model, however, elucidates orienting only between two external stimuli but not from external stimuli to internal representations. The external-to-internal orienting process is important as information processing in humans mostly requires encoding information available in the external environment for other internal processes, but its mechanism is still not clear.
Previous studies focused on orienting attention on external stimulus (called External Attention, EA) or internal stimulus (called Internal Attention, IA) predominantly involving visual modality. Griffin and Nobre ( ) reported that behavioral performances were improved (e.g., faster reaction time and higher accuracy rate) when participants’ orienting attention to the cued spatial location was enhanced in the EA as well as the IA condition. Nevertheless, previous studies indicated that EA and IA involved different neural processes. Sauce et al. ( ) articulated their functionality in that EA primarily inhibits the perception of distractions and noises associated with external stimulus, whereas IA facilitates the perception against interferences associated with internal representation such as irrelevant memory episodes. Griffin and Nobre ( ) demonstrated that EA and IA differed in the N1 component, of which IA elicited more negative-going waveforms at the frontal (120–200 ms) and central regions (160–200 ms) than EA. Tanoue et al. ( ) reported that performance involving IA was found to be more affected than that of EA when the frontal lobe was under stimulation. Taken together, orienting attention to an internal stimulus involves more frontal control than orienting attention to an external stimulus.
Studies on orienting attention involving somatosensory modality further disentangle the subprocesses associated with external stimuli, but also external-to-internal stimuli. Fronto-centrally distributed N1 (also called FCN; 100–200 ms) has been revealed as a common marker associated with orienting attention among external somatosensory stimuli (Kida et al., ; Dowman, , ; Dowman et al., ). In particularly, less negative-going frontal N1 was associated with the disengaging process among external somatosensory stimuli presented at different spatial locations (Katus et al., ), as well as from a somatosensory to a visual stimulus (Ohara et al., ; Staines et al., ) or vice versa (Dowman, , ; Dowman et al., ). Besides the N1, Dowman ( , ) proposed that more P2 (260–380 ms) elicited at the fronto-central region was related to shifting attention, whereas P3a (320–390 ms) elicited at the centro-parietal region was related to reengagement. Taken together, FCN, fronto-central P2 and central P3a appear to play different roles when an individual disengages, shifts and reengages one’s attention on external somatosensory stimulus.
All the studies reviewed above focused on EA processes. Only a handful of studies addressed neural processes of IA and external-to-internal orienting attention involving somatosensory modality. For instance, Legrain et al.’s ( ) reported that working memory modulated the external-to-internal orienting process. The external stimuli used were nociceptive (by CO laser) or subnociceptive stimulations (by electrical stimulations), and the internal presentations were images of visual dots (Legrain et al.’s ). The external-to-internal shifting processes were associated with less negative-going N1 and less positive-going P2 elicited in the frontal and central regions, respectively. Legrain et al.’s ( ) proposed the N1 component reflected attending to and disengaging from the external nociceptive stimulus, whereas the P2 component reflected the shifting process. The involvement of the centrally distributed P2 in the external-to-internal shifting process was further corroborated by the results revealed in Chan et al.’s ( ) study, which used nociceptive stimulations as the external stimuli and subnociceptive images as the internal representations. Their results, however, are different from those reported by Dowman ( , ), which suggested shifting was associated with FCN, P2 and P3a. The discrepancies in the results further suggest that orienting attention to somatosensory stimuli involves unique neural processes depending on the external or internal environment in which the stimuli are processed.
This study aimed to differentiate the subprocesses associated with external-to-internal orienting attention. In particular, we attempted to reveal the characteristics of the disengagement, shifting and reengagement subprocesses by using somatosensory external stimuli and internally generated images. The external stimuli involved were nociceptive stimulations to be perceived by the participants and the internal representations were subnociceptive images generated by the participants after receiving training. The external-to-internal perceptual processes required the participants to perceive different levels of salience of nociceptive stimulations followed by generating a predetermined subnociceptive image of specific salience level. We hypothesized that perception of more salient nociceptive external stimuli would result in a higher level of bottom-up control for initiating a disengagement process, which would yield an FCN (an earlier component). The shifting to more salient internal subnociceptive images would result in a higher level of top-down control for the shifting and reengagement processes, which would yield fronto-central P2 and centro-parietal P3 (later components). The external-to-internal orienting attention would associate with attenuation of the pain intensity felt by the participants for the external nociceptive stimulations.
## Materials and Methods
### Participants
Twenty-two healthy participants (13 females) were recruited. Their mean age was 39.0 (SD = 11.6). All of them had a high school education or above and scored within the norms on the Stroop Test (measure of executive control functions). The participants did not report any type of pain in the past 6 months. The purpose of the study was explained to and informed consents were obtained from each participant. This study was approved by the research committee of the Department of Rehabilitation Sciences of The Hong Kong Polytechnic University.
### Pre-Experimental Preparation
#### Electrical Stimuli
The nociceptive and subnociceptive stimuli used in the pre-experiment training and the experiment were 50-ms electrical stimulations at different intensity levels (25-pulse train of electrical square-wave pulses with 0.5-ms pulse duration and 500 Hz frequency) generated from an S88K Dual Output Square Pulse Stimulator (Grass Technologies, Grass-telefactor, West Warwick, RI, USA) and controlled by a constant current unit (CCU). These devices were the same as those used in Chan et al.’s ( ) study. Anode and cathode electrodes of the stimulator were attached to the skin of a flat bodily area posterior to the lateral malleolus of the left ankle and along the distribution of the sural nerve (L5-S1 dermatome; Dowman, ). The external nociceptive and subnociceptive stimuli were applied to this site during the pre-experiment training, and only the nociceptive stimuli were applied during the actual experiment.
#### Calibration of Stimuli
The procedure used for calibrating the nociceptive (i.e., painful) and subnociceptive (i.e., not painful) stimuli followed the sequential stepping-up and stepping-down method adopted in previous studies (De Pascalis et al., ; Chan et al.’s ). Three critical sensory thresholds were calibrated for the electrical stimulations generated for each participant’s minimum detectable sensation (MDS; the weakest stimulation intensity level with which participants detected a tactile sensation), just painful sensation (JPS; the minimal intensity with which participant’s perceived stimulation as painful and rated “1” on the 11-point numeric rating scale, or NRS; Jensen et al., ; Williamson and Hoggart, ) and very painful sensation (VPS; the intensity with which participants perceived a stimulation as very painful and rated “7” on the NRS). The mean voltages of the MDS, JPS (NRS = 1), and VPS (NRS = 7) of the participants were 3.32 mA (SD = 5.24 mA), 19.11 mA (SD = 6.38 mA), and 41.27 mA (SD = 12.60 mA), respectively (Table ). For each participant, two levels (one-third and two-thirds) of intensity of the subnociceptive stimuli were determined between the MDS and the JPS (labeled as SN and SN ) and six levels of intensity of the nociceptive stimuli were determined by means of even distribution between the JPS and the VPS (labeled as L1 to L6).
Mean (SD) of intensity, voltages and NRS ratings for the nociceptive (six) and sub-nociceptive (two) stimuli.
Note: SN and SN refer to lower and higher intensity sub-nociceptive stimulus, respectively; L1 to L6 are nociceptive stimulus with L1 being the lowest intensity and L6 being the highest intensity. NRS (Numeric Rating Scale), ranged from 0 to 10, reflects the pain intensity felt by the participant from the stimuli .
#### Training
The participants learned to assign NRS ratings to represent the pain intensity felt for the individualized nociceptive and subnociceptive stimuli. This was an important component in the study as the participants were required to give an NRS rating of the pain intensity level of the brief external nociceptive stimulus (only 50 ms exposure) by the end of each trial. The training was to improve the validity and accuracy of the ratings to be assigned by the participants. Individualized calibrated nociceptive stimuli were randomly delivered to the lateral malleolar of the left leg after which the participants assigned one of the L1 to L6 ratings. There were at least 24 attempts and the training would end after the participants achieved at least 80% accuracy. In addition, as each trial involved generation of specific internal subnociceptive image (high or low) after perceiving the external nociceptive stimulus, the participants learned to associate the somatosensory sensation of the two subnociceptive stimuli with the high and low intensity descriptor. The same training protocol was used for learning the intensity level and association with the intensity descriptor of the nociceptive stimuli.
### Experimental Task
The design of the experimental task made reference to that employed in Chan et al.’s ( ). Among the six nociceptive stimuli, the three lower intensity stimuli were grouped into the E condition (low salience external stimuli) and the three higher intensity stimuli were grouped into the E condition (high salience external stimuli). Between-condition differences in the voltages of ( t = 12.13, p < 0.001) and the NRS scores associated with the stimulation intensity ( t = 16.14, p < 0.001) were statistically significant. The two internally generated subnociceptive image representations were categorized into I (internal low salience) and I (internal high salience), respectively.
Each trial involved the participants’ engagement in three steps: (1) perception—perceiving an external nociceptive stimulus of 50 ms (S1); (2) image generation—generating a learned subnociceptive image; and (3) rating response—recalling the perceived external nociceptive stimulus (S1) and assigning an NRS rating reflecting the pain intensity felt from the nociceptive image (Figure ).
Schematic representation of the experimental paradigm. Left panel is an experimental trial in which the participant was to attend to an external nociceptive stimulus (S1, E or E ) and maintain the image (Step 1), and followed by generating a learnt sub-nociceptive image (I or I ; Step 2). During the response (Step 3), the participant was to assign a rating against Numeric Rating Scale (0 score = “non-painful” to 10 score = “extremely painful”) which represents the intensity of the pain felt for the S1 (in 67% of trials). In some occasions (33% of trials), the participant was to attend to a sub-nociceptive stimulus (S2) and judge whether the intensity of S2 was comparable with that of the sub-nociceptive image last generated. Right panel is a control trial which only contains Steps 1 and 3. In Step 3, the same procedure was followed except that, in some occasions, the S2 delivered to the participant was a nociceptive stimulus with 50% of the time the intensity was comparable to that of S1.
A fixation cross (500 ms) was first presented at the middle of a computer monitor. After a varied interval of 1100–1300 ms, a 50 ms nociceptive stimulus (called S1)—randomized intensity (E : L1–L3; E : L4–L6)—would be delivered from the electrical stimulator at the lateral malleolus of the left ankle. A participant was to attend to the nociceptive stimulus (Step 1). Then, participants generated and rehearsed a learned subnociceptive image representation, SN or SN , depending on the condition (Step 2). This process would require participants to disengage from the relatively strong external nociceptive stimulus, shift their attention and reengage with the relatively weak internally generated somatosensory image. The duration for steps 1 and 2 was 3000 ms. Participants were then asked to recall the pain felt from the external nociceptive stimulus and assign a numeric rating of the intensity of the sensation (Step 3). The response statement “How painful do you feel for S1?” appeared on the screen, which lasted until a response was received or after 6000 ms. The response was to press one of the 1-to-10 keys on a number keyboard that represented the perceived pain intensity. The control task required participants to engage in only steps 1 and 3, and skip Step 2. The participants were to perceive the external nociceptive stimulus, which lasted for 50 ms, and retain the image of the stimulus until end of 3000 ms. The same rating of the pain felt from the stimulus was conducted. As a validity check, one-third of the trials in each block involved participants judging the congruence between the internal image rehearsed with an external stimulus (S2) rather than rating the internal image as in Step 3 in the experiment. The S2 was a subnociceptive stimulation of intensity similar to I or I or a nociceptive stimulation of intensity equivalent to S1. The participants were to make responses to “Is the intensity of the stimulus similar to what you had in mind?”
The task was organized in a 2 × 2 (external stimuli × internal image representations) design. One High (E ) and one Low (E ) condition each manipulated the intensity of the external nociceptive stimuli to be perceived by the participants. Only the L4 to L6 stimuli were delivered in the E condition, and only L1 to L3 stimuli were delivered in the H condition. Additionally, one High (I ) and one Low (I ) condition each manipulated the intensity of the internally generated subnociceptive image representations. In the I condition, only the learned higher N image would be generated and rehearsed by the participants. The lower N image would be generated and rehearsed in the I condition. The 2 × 2 external-to-internal combinations produced a total of 324 trials. The trials were organized into three conditions (namely, I , I , and control) and three blocks were organized in each condition. Completion of 36 trials in each block took around 5 min. The sequence of stimuli presentation was counterbalanced across the participants, and the sequence of the blocks was organized in a pseudorandomized sequence.
### EEG Recording and Preprocessing
The EEG signals were recorded throughout tasks from scalps of the participants with a 64-channel cap based on the 10-20 system. The signals collected were preprocessed by CURRY Neuroimaging Suite software (Neuroscan, Compumedics Ltd., Abbotsford, VIC, Australia). Electrooculograph (EOG) was recorded by two pairs of electrodes located vertically and horizontally around the eyes to detect eye blinks and movements. The two reference electrodes were placed on the left and right mastoid. The EEG signal was sampled at 1024 Hz. All EEG/EOG electrode impedances were set to be less than 10 kΩ. The timing and presentation of all the stimuli were controlled by E-Prime (Psychology Software Tools, Inc.). During the preprocessing, all data were referenced to the average of the two referenced electrodes. The EEG was epoched from 200 ms before and 900 ms after the onset of the delivery of each external nociceptive stimulus. Ocular artifact reduction and a zero-pass filter with a low-pass of 30 Hz and 24 db/oct were applied.
### Data Analysis
Behavioral results were the NRS scores for the perceived pain intensity of the external nociceptive stimuli assigned by the participants by the end of each trial. Two-way, repeated-measures ANOVA was used to test the possible External (E vs. E ) and Internal (I vs. I ) effects on the NRS scores. Post hoc comparisons were conducted in the case of significant interaction effects on the External and Internal factors.
For the EEG data, the time windows of the components were determined according to the somatosensory-evoked potential method described in Dowman ( ). The onset and offset latencies of the negative potentials were first approximated by visual inspection. The stable period of a potential was the time window between the onset and offset latencies and was verified by employing the r statistics derived from the amplitudes of the potential captured from the 29 scalp electrodes. A stable period included the time points at which the r of the amplitudes between the midpoint (or peak) and that of the time point was ≥0.85. According to this method, the time windows for each of the three components were: 128–152 ms and 152–180 ms for the SP3 and SP3/P2 (respectively), 200–260 ms for the P2, and 320–380 ms for the P3 (Figure ). Three-way, repeated-measures ANOVA was conducted to test the effects of External (E vs. E ), Internal (I vs. I ), and Electrode (F3/z/4, FC3/z/4, C3/z/4, CP3/z/4, P3/z/4 vs. PO3/z/4) factors on each of the SP3, SP3/P2, P2 and P3 components. The significance level was set at 0.050 for the full model. Bonferroni adjustments were applied to individual pair-wise comparison for the significant interaction effect. Pearson correlation was used to test the relationships between the amplitude change and the NRS rating change for each component. Amplitude (or NRS) change was computed by subtracting the mean amplitude (or NRS) of the control condition from those of the experimental external and internal conditions.
Event-related potential (ERP) waveform and topographic maps of the identified components. Upper panel: t = 0 corresponds to the onset of the nociceptive stimulus recorded at FCz. Bottom panel: topographic maps (top view) of amplitudes of fronto-central negativity (FCN; SP3, SP3/P2), P2 and P3 waves in their respective time-windows.
## Results
Two participants failed to achieve an average of 80% accuracy identifying both the nociceptive and subnociceptive stimuli in the training. They were excluded from the data analysis. Another participant was also excluded as the EEG data did not have the quality for meaningful interpretation. The final sample size for entering into the analysis was 19 participants.
### NRS Ratings
The External and Internal effects on participants’ NRS ratings were statistically significant ( P ≤ 0.001 and 0.012, respectively); however, their interaction effects were not significant ( P = 0.789; Table ). For the External factor, the NRS ratings were significantly higher for the more salient stimuli (Mean = 5.1) than that for the less salient stimuli (Mean = 2.9; Table ). For the Internal factor, on the contrary, the NRS ratings for the less salient stimuli (Mean = 4.1) were significantly higher than those for the more salient stimuli (Mean = 3.9; Table ).
Tests of within-subject effects for NRS scores and mean amplitudes of event-related potential (ERP) components.
Note: NRS, Numeric Rating Scale which yields scores reflect the pain intensity felt by the participants from nociceptive external stimulus at the beginning of the trial (S1) .
Mean (SD) of behavioral results.
Note: E refers to low salience external stimuli; E refers to high salience external stimuli. I refers to low salience internal sub-nociceptive images; I refers to high salience internal sub-nociceptive images. NRS, Numeric Rating Scale which yields scores reflect the pain intensity felt by the participants from the external stimuli .
### ERP Results
#### FCN
##### SP3 (128–152 ms)
The External and Internal interaction effect was found significant ( P < 0.050; Table ). In the highly salient External (E ) condition, mean amplitude of SP3 was significantly more negative-going in the disengagement in a low salience Internal (I ) condition (Mean = −3.81 μV) than that in a highly salient Internal (I ) condition (Mean = −2.49 μV; P < 0.050; Figure ). Amplitude differences between the I and I conditions were not significant in the E condition. Amplitudes of SP3 were the most negative-going at FC4 (Mean = −5.10 μV; P < 0.050). The External effect was also found significant ( P ≤ 0.001), of which the amplitudes elicited by the E condition (Mean = −3.18 μV) were significantly higher than those elicited by the E condition (Mean = −1.39 μV; P < 0.001; Figure ). Other main and interaction effects were not significant.
Comparisons of ERPs among the two external nociceptive stimulus and two internal sub-nociceptive image conditions recorded at FCz and Cz. Left panel presents results of high salience external condition (E ); and Right panel presents results of low salience external condition (E ). Red line: high salience internal condition (I ); Blue line: low salience internal condition (I ); and Black line: control condition.
Bar charts summarizing the external-to-internal interactive effects for FCN, P2 and P3. (A) Left—SP3 based on average amplitudes; Right—SP3/P2 based on average amplitudes. (B) P2 elicited FCz. (C) P3 elicited at C4. Note: error bars are standard errors. Asterisks refer to P < 0.050.
##### SP3/P2 (152–180 ms)
The results pattern from ANOVA for the SP3/P2 time window was similar to that for the SP3 time window (Figure ). The External and Internal interaction effects were found significant ( P < 0.050; Table ). In the highly salient External (E ) condition, mean amplitude of SP3 was significantly more negative-going in the disengagement in the low salience Internal (I ) condition (Mean = 2.74 μV) than that in highly salient Internal (I ) condition (Mean = 3.99 μV; P < 0.050). In the low salience External (E ) condition, mean amplitude of SP3 was significantly less negative-going in the disengagement in the low salience Internal (I ) condition (Mean = 5.13 μV) than when disengaging to the highly salient Internal (I ) condition (Mean = 3.86 μV; P < 0.050; Figure ). Amplitude of SP3/P2 was the most negative-going at F4 (Mean = −0.60 μV) and most positive-going in Cz (Mean = 9.61 μV; P s < 0.050). The External effect was also found significant ( P < 0.001), of which the amplitudes elicited by the E condition (Mean = 3.37 μV) were significantly less negative-going than those elicited by the E condition (Mean = 4.32 μV; P < 0.001). Other main and interaction effects were not significant.
#### P2 (200–260 ms)
The Electrode, External and Internal interaction effects were found significant ( P < 0.001; Table ; Figure ). In the highly salient External (E ) condition, mean amplitude of P2 was significantly less positive-going in the shifting in the low salience Internal (I ) condition (at FCz: Mean = 21.58 μV) than when shifting to the highly salient Internal (I ) condition (at FCz: Mean = 23.79 μV; P < 0.050) or to the control condition (at FCz: Mean = 23.26 μV; P < 0.050). Similar patterns of results were found in electrodes F3, Fz, F4, FC4, Cz and CPz ( P s < 0.050; Figure ). Amplitude differences between the I and I conditions were not significant in the E condition. Amplitudes of P2 were the most positive-going at the central site (Cz: Mean = 25.61 μV; P s < 0.050). Other main and interaction effects were not significant.
#### P3 (320–380 ms)
The Electrode, External and Internal interaction effects were found significant ( P < 0.001; Table ; Figure ). In the highly salient External (E ) condition, mean amplitude of P3 was marginally more positive-going in the reengagement in the low salience Internal (I ) condition (at C4: Mean = 12.79 μV) than in the highly salient Internal (I ) condition (at C4: Mean = 10.00 μV; P = 0.067) or significantly more positive-going in the control condition (at C4: Mean = 10.58 μV; P = 0.050; Figure ). No significant interaction of External and Internal conditions could be found in other electrodes, and amplitude differences between the I and I conditions were not significant in the E condition. Amplitudes of P3 were the most positive-going at the fronto-central site (FCz: Mean = 15.52 μV; P s < 0.050). The Electrode and Internal interaction effects were found significant ( P < 0.050). At C4, mean amplitude of P3 was significantly more positive-going in reengagement in the low salience Internal (I ) condition (Mean = 13.20 μV) than in the highly salient Internal (I ) condition (Mean = 11.62 μV; P < 0.050) or in the control condition (Mean = 11.18 μV; P < 0.001). The Electrode and External interaction effects were found significant ( P < 0.050), of which the amplitudes elicited by the E condition were significantly more positive-going than those elicited by the E condition at all 14 electrodes except at F3 ( P s < 0.050). The External effect was also found significant ( P < 0.001), of which the amplitudes elicited by the E condition (Mean = 12.19 μV) were significantly more positive-going than those elicited by the E condition (Mean = 10.39 μV; P < 0.001). Other main and interaction effects were not significant.
### Correlations between Changes in NRS Ratings and ERP Amplitudes
Among the four external-to-internal conditions, significant correlations were only revealed in the E /I condition between changes in the amplitudes and those in the NRS ratings. Changes in the amplitudes of the P3 (I minus control) recorded at the centro-parietal electrodes (C3, C4, CP3, CP4 and CPz) were positively and moderately correlated with changes (I minus control) in the NRS scores ( r = 0.537 [at CP3] to 0.638 [at CPz], P s < 0.050).
## Discussion
The present study investigated the characteristics of the processes associated with orienting attention from external nociceptive stimuli to internal subnociceptive images. The elicitation of FCN, P2 and P3 supported the hypothesized disengaging, shifting and reengagement subprocesses. The significant findings in the amplitude change of these components observed in the high but not in the low salience external stimulus conditions suggest interactions of both bottom-up and top-down processes in the external-to-internal orienting attention. The bottom-up process would be primarily stimuli driven, represented by the FCN, whereas the top-down process would be primarily goal directed, represented by the P2 and P3 components.
### Disengagement from External Stimuli
The fronto-centrally distributed FCN elicited by the nociceptive stimuli is consistent with that reported in Dowman ( , ). This component has been associated with a stimulus-driven, bottom-up process in which a higher level of attention was found allocated to higher rather than lower salient nociceptive stimuli. Besides, the FCN was associated with visualization of somatosensory stimuli in working memory (Legrain et al.’s ) suggesting that the disengagement from the nociceptive stimuli would have involved working memory, particularly those of high salience. Different from previous studies, we manipulated both the external and internal conditions by classifying stimuli/images in both environments into high and low salience levels. The significant Condition × Salience on the FCN amplitudes indicated the disengagement subprocess was modulated by the highly salient nociceptive stimuli. Highly salient external stimuli resulted in larger incongruence in intensities between the external stimuli and internal image representations than the low salience stimuli. The larger incongruence would have generated larger mental conflicts and hence stronger top-down cognitive control for resolving them (Kerns et al., ; Egner et al., ).
### External-to-Internal Shifting
Similar to FCN, this study revealed a significant external-to-internal interaction effect on the fronto-centrally distributed P2. The P2 has been previously associated with the shifting attention process involving somatosensory stimuli (Legrain et al.’s ). Nevertheless, the less positive-going P2 yielded in this study is inconsistent with that reported in Chan et al.’s ( ). Chan et al. yielded more positive-going P2 when participants shifted their attention from nociceptive stimuli to self-generated subnociceptive images. An analysis of the task designs among these three studies suggests that the discrepancy in the results is likely to be attributable to whether the somatosensory images to be generated are anticipatory or contingent to the perceived nociceptive stimuli. In Chan et al.’s ( ) study, the participants were to generate subnociceptive images at the salient levels contingent upon those of the nociceptive stimuli perceived at the beginning of the trial. Participants in Legrain et al.’s ( ) study were required to generate dot images, as to generate subnociceptive images in our study, which had been maintained in the working memory throughout the trial. It is therefore plausible that when compared with Chan et al.’s ( ) study, our participants would have involved more top-down influence for generating the subnociceptive images, which evoked less positive-going P2 amplitudes.
It is noteworthy that the modulation effect on P2 was only observed in the high but not in the low salience external condition stimuli and when the external-to-internal incongruence was large. It is plausible that the significant P2 reflected a top-down process that interacted with the prior bottom-up, stimulus-driven process (perception of nociceptive stimulation). In fact, interactions between the top-down and bottom-up processes modulating shifting attention have been reported in behavioral studies on visual perception (Caparos and Linnell, ; Linnell and Caparos, ). The results of our study provide evidence that such interactions occurred at around 200–260 ms after presentation of the external somatosensory stimulus represented by the fronto-central P2.
### Reengagement
Centrally distributed P3a was suggested to reflect reengaging attention (Dowman, ). The significant P3 results obtained in this study can be interpreted as participants reengaged with the internal subnociceptive images after shifting from the external nociceptive stimuli. The marginal significance ( P = 0.067) yielded for the P3 effects suggests caution should be taken when interpreting such effects. Significant correlations were revealed between the changes in the P3 amplitudes at the central and parietal electrodes (C3, C4, CP3, CP4 and CPz) and the attenuation on the perceived pain intensity felt for the external nociceptive stimulus. The attenuation was based on scores assigned by the participants using an NRS of the recall of the nociceptive sensation after generating the internal sub-nociceptive by end of the trial. In other words, less positive-going P3 amplitudes were correlated with larger attenuation of the pain intensity scores. The attenuation effect was only observed in trials involving high salience external stimuli and generating a low salience internal image. Previous studies reported that anterior P3 was related to reengaged attention with external nociceptive stimulus (Dowman, , ; Dowman et al., ), and the posterior P3 was associated with rehearsal of the mental representation (Donchin, ; Pontifex et al., ). Our findings offer a plausible mechanism for explaining the attenuation effects on pain intensity after orienting attention, such as those reported in Fors et al. ( ) and Chan et al.’s ( ), and that such effects would be the result of a top-down-dominated reengaging process in orienting attention. The processes probably would involve mental conflict of cognitive control (Kerns et al., ; Egner et al., ).
### Limitations
This study has several limitations. First, the task design did not control the time at which the participants generated the internal subnociceptive images. This could have confounded the latency of the P3 components that impacted peak amplitude and were subsequently attributable to the marginally significant interaction between the external and internal effects. Future task design should improve the timing of generating the internal somatosensory images and perhaps increase the sample size for increasing the effect size and power of the data analysis, respectively. Second, this study was based on a somatosensory stimulus and images—the results may not be readily generalized to other sensory modalities. Further studies should be conducted to test the robustness of the external-to-internal orienting attention on visual and auditory modalities.
## Conclusion
Salience of external stimulus and internal representation was found to modulate the external-to-internal process. Perception of a high salience external stimulus exerted effects on the disengagement and shifting processes, and subsequently the reengaging process with which internal images were generated. This reengaging process was revealed to relate to the perception of the external stimulus and in this study was attenuation of the pain intensity perceived for the nociceptive sensation felt. Our findings offer a plausible mechanism for explaining attentional dysfunction among patients with chronic pain and its therapeutic intervention. Future studies should be conducted to replicate the experiment on patients with chronic pain to test this proposition.
## Author Contributions
BKHC contributed to interpretation of data and article writing. CCHC contributed to conceptualization, study design, interpretation of data, and article writing. JP contributed to conceptualization, study design, data collection, data analysis, interpretation of data, and article writing. SCCC contributed to conceptualization, study design, interpretation of data, and article writing. QY contributed to data collection, data analysis, interpretation of data, and article writing. All approved the version to be published and agreed to be accountable for all aspects of the work in ensuring that questions related to accuracy or integrity of any part of the work were appropriately investigated and resolved.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Transferring current research findings on the topic of learning and memory to “brain-based” learning in schools is of great interest among teachers. However, numerous international studies demonstrate that both pre-service and in-service teachers do not always succeed. Instead, they transfer numerous misconceptions about neuroscience, known as neuromyths, into pedagogical practice. As a result, researchers call for more neuroscience in teacher education in order to create a professional understanding of learning and memory. German pre-service science teachers specializing in biology complete neuroscientific modules ( human biology / animal physiology ) during their studies because they are expected to teach these topics to their students. Thus, they are required to demonstrate a certain degree of neuroscience literacy. In the present study, 550 pre-service science teachers were surveyed on neuromyths and scientific concepts about learning and memory. Pre-service science teachers’ scientific concepts increased over the course of their training. However, beliefs in neuromyths were independent of participants’ status within teacher education (first-year students, advanced students, and post-graduate trainees). The results showed that 10 neuromyths were endorsed by more than 50% of prospective science teachers. Beliefs in the existence of learning styles (93%) and the effectiveness of Brain Gym (92%) were most widespread. Many myths were endorsed even though a large share of respondents had thematically similar scientific concepts; endorsement of neuromyths was found to be largely independent of professional knowledge as well as theory-based and biography-based learning beliefs about neuroscience and learning. Our results suggest that neuromyths can exist in parallel to scientific concepts, professional knowledge and beliefs and are resistant to formal education. From the perspective of conceptual change theory, they thus exhibit characteristic traits of misconceptions that cannot simply be counteracted with increased neuroscientific knowledge. On the basis of our study’s findings, it can be concluded that new teacher programs considering neuromyths as change-resistant misconceptions are needed to professionalize pre-service science teachers’ neuroscience literacy. For this, an intensive web of exchange between the education field and neuroscientists is required, not just to deploy the latest scientific insights to refute neuromyths on learning and memory, but also to identify further neuromyths.
## Introduction
Findings from brain research have unleashed a veritable “neuro-boom” in recent years, which has taken the form of numerous publications for teachers as well as learning guides for students (e.g., ). Teachers in all subjects have expressed great interest in neuroscience research findings and find it useful to incorporate them into their instruction ( ). Even incorrectly interpreted research findings have great appeal once images of the brain and/or neuroscientific explanations are added ( ; ). Media and even educational programs make use of this effect; they are filled with bold and eye-catching yet empty promises like ‘learn while you sleep’ or ‘innate intelligence through Brain Gym’. Money, time, and effort are expended integrating so-called “neuromyths” into the school system ( ). It is understandable that people who lack knowledge in the field of neuroscience might struggle to distinguish facts from myths ( ; ). However, even teachers, the alleged experts on learning, endorse misconceptions about neuroscience and base their pedagogical practice on neuromyths (e.g., ).
The term “neuromyth” was coined in the 1980s by the neurosurgeon Alan Crockard, who used it to refer to unscientific understandings of the brain in medical culture ( ). The defines neuromyths as “misconception[s] generated by a misunderstanding, a misreading, or a misquoting of facts scientifically established (by brain research) to make a case for use of brain research in education and other contexts” (p. 111). Neuromyths are thus falsely or overly interpreted neuroscientific research findings that are transferred to applied contexts such as teaching, learning and instruction. Neuromyths are often seen as originating in simplistic language in the reporting of neuroscientific research findings. These research findings are often published at a challenging reading level ( ) and tend to be very complex and difficult for non-neurobiologists to understand, meaning that simplistic formulations are often resorted to. These ‘pop-science’ statements are then falsely interpreted and quickly lose their kernel of truth. They are packed into “low-cost and easily implemented classroom approaches” ( , p. 819) that claim to promote learning. While fun, these approaches also result in the rapid propagation of neuromyths among students, parents, and teachers. This process is strengthened by media, whose simplified and/or overly interpreted portrayals of research findings reach a wide audience ( ; ), and companies looking to offer learning programs that claim to be “brain-based” but usually provide consumers with little hope of learning success ( ; ). The OECD has been calling attention to the problem of neuromyths almost as long as it has been calling for teaching and learning to be based on neuroscience.
Neuromyths often originate from overgeneralizations of empirical research ( ). Today, neuromyths have emerged for many aspects of neuroscience, including specific learning difficulties such as dyslexia ( ) and the influence of nutrition ( ) or music ( ) on the brain. This study focuses on neuromyths related to learning and memory. illustrates with three examples how these neuromyths arise from errors in transferring neuroscientific information (the kernel of truth). The depicted transfer steps (left) as well as their relationships to neuroscientific findings (right) are based on a summary of the current state of theory on neuromyths as well as supplementary literature research.
Errors in argumentation from the neuroscientific kernel of truth to erroneous implications for school instruction compared to the neuroscientific evidence.
The existing research on neuromyths primarily focuses on teachers. Studies investigating the endorsement of neuromyths among teachers of various subjects have been conducted in the Netherlands, England ( ; ), Latin America ( ; ; ), Portugal ( ), Australia ( ; ), Greece ( ), China ( ), Turkey (e.g., ), Switzerland ( ), Spain ( ), the United States ( ), and Canada ( ). All of these studies have found that teachers believe in a large number of neuromyths, although only a few of these, such as Brain Gym, are related to the topic of learning and memory and there are country-specific differences in the endorsement of specific myths. Cultural differences between countries seem to have an influence on which neuromyths spread ( ; ; ).
One study of post-graduate teacher trainees found that 56–83% of respondents encountered educational programs based on neuromyths in their first year working in schools, which was associated with high levels of acceptance of those myths ( ). Studies by , Spain), , Turkey), , Turkey), , Germany), , Australia), , Greece), and , South Korea) indicate that neuromyths are already present during the academic stage of teacher education. However, no studies focusing on neuromyths related to learning and memory have been conducted with pre-service samples either. Nevertheless, it can generally be concluded on the basis of these results that the neuroscience content knowledge necessary to critically evaluate neuromyths does not seem to be integrated into teacher education to a sufficient degree ( ).
Most studies have not found personal characteristics like age, professional experience, teaching subject, school type, school location (urban/rural), and participation in professional development trainings to be associated with endorsement of neuromyths or with scientific concepts about the brain ( ; ; ). Only found a correlation with gender, with female teachers more likely to endorse neuromyths. found evidence that age (being younger), training (having a university degree), and enrollment in neuroscience courses predict reduced endorsement of neuromyths. The topic of neuroscience is of great interest to teachers internationally (e.g., ), but there seems to be a large gap between teachers’ interest and their ability to actually deal with neuroscientific findings in a professional way ( ). Teachers with high levels of scientific concepts about the brain have proven to be more susceptible to neuromyths ( ; ; ; ; ; ) in almost all studies (except ). found the acceptance of neuromyths to be nearly identical between populations of award-winning and non-award-winning teachers. There seems to be a general tendency to agree with neuroscientific statements, but a lack of ability to separate myths from facts ( ). While reading scientific articles can reduce endorsement of neuromyths, teachers tend to use pop-science sources like TV and the Internet as their main sources of information for neuroscientific facts ( ).
demonstrated that taking an educational psychology course only improves neuroscience literacy; it does not reduce beliefs in neuromyths. found differences in endorsement of neuromyths between the general public, teachers, and people with high levels of neuroscientific knowledge, and indicate that general knowledge about the brain is the best “safeguard against believing in neuromyths” (p. 1). German pre-service science teachers specializing in biology receive such knowledge during their university education so that they will later be able to pass it on to their students in their classroom instruction. It can thus be assumed that they develop the theory-based learning beliefs and professional knowledge about neuroscience needed to critically evaluate neuromyths about learning and memory during their university studies. According to , these two factors, along with motivational orientations and self-regulative skills, are prerequisites for reflective instruction, or in other words, the teaching profession. subdivides beliefs into epistemological beliefs and beliefs about learning content and instructional practice. Applying this to the topic of neuroscience and learning, pre-service teachers might have theory-based learning beliefs on the nature of science and teaching and learning. However, they also bring with them beliefs about the definition of learning and learning strategies rooted in their own learning biographies (biography-based learning beliefs). The professional knowledge pre-service science teachers in Germany specializing in biology are expected to acquire during their studies can be subdivided into, inter alia, psychological-pedagogical knowledge (PPK), content knowledge (CK), and pedagogical content knowledge (PCK) ( ; ). Applying this to the topic of neuroscience and learning, pre-service science teachers need to acquire PPK about the psychology of human learning, CK about curricular content related to neuroscience, and PCK about instructional strategies for sustainable learning ( ). According to , pre-service science teachers perform significantly better in terms of neuromyths than pre-service teachers in other subjects. As of yet, there are no studies specifically investigating neuromyths about learning and memory or on their prevalence among pre-service science teachers (specializing in biology) depending on their status within teacher education.
As mentioned above, and other existing studies on neuromyths interpret the frequently found association between endorsement of neuromyths and scientific concepts as a general tendency to agree with neuroscientific statements among teachers. After more intensive theoretical work on neuromyths, we also see these correlations as rooted in the fact that the test instrument asks in some cases about both the neuromyth and the corresponding kernel of truth as scientific concepts. For example: neuromyth = “Individuals learn better when they receive information in their preferred learning style (e.g ., auditory, visual, kinesthetic)“ and scientific concept (kernel of truth) = “Individual learners show preferences for the mode in which they receive information (e.g ., visual, auditory, kinesthetic).“ instrument for scientific concepts, which was applied in many of the previous studies on neuromyths, cannot be seen as an appropriate knowledge test for pre-service science teachers in light of our theoretical perspective on professional knowledge among science teachers. In order to more effectively design professional development offerings, it will be necessary to further clarify the causal relations between misconceptions and aspects of professional competency (beliefs and professional knowledge). The present study is the first to do this.
## Methods
Building upon the aforementioned theoretical work, this study addresses three research questions: (1) How are pre-service science teachers’ misconceptions and scientific concepts about learning and memory associated with their status within teacher education? (2) What misconceptions and scientific concepts do pre-service science teachers have on the topic of learning and memory? (3) How are their misconceptions associated with their beliefs and professional knowledge about the topic of learning and memory?
### Participants
The study was conducted among pre-service science teachers specializing in biology at two German universities (University of Kassel and University of Kiel) as well as several institutes for post-graduate teacher trainees ( Studienseminare ) in the federal state of Hesse ( N = 550). The total sample consisted of 152 first-year students, 260 advanced students (second year and above), and 138 post-graduate teacher trainees ( Referendare , who were an average of 9 months into their training, SD = 4.47). Respondents were 24.8% male and 75.2% female, and were between 18 and 38 years of age ( M = 24 years old, SD = 3.79). 69.4% of respondents were studying to be teachers at college-preparatory secondary schools ( Gymnasium ), and 30.6% were studying to be teachers in lower-track secondary schools ( Lehramt f r Haupt- und Realschulen ). Research Question 3 was investigated with a subsample of 79 advanced students, who had the opportunity to participate in a more extensive testing for organizational reasons. Advanced means that they had already completed a human biology course with neuroscience content during their studies. 21.5% of the respondents in this sample were male, while 78.5% were female. The average age was 25 years ( SD = 2.70) and the respondents were in their eighth semester of studies on average ( SD = 2.56).
### Procedure
The data was collected in 19 courses in the field of instructional methods for science (biology) education. The post-graduate teacher trainees were recruited and surveyed via their supervisors at the teacher training institutes. Participation took the form of a paper-and-pencil test lasting approximately 15 min. The testing time was expanded to 1 h for a subsample of participants ( n = 79) in order to apply further instruments (see Materials). In both cases, the project was introduced as a study on the topic of neuroscience and learning; the term “neuromyths” was not used. Participation in the evaluation was voluntary and the students provided informed written consent to use the data for research purposes. They were informed that the goal of the study was to collect information on their current state of knowledge and attitudes toward the topic of neuroscience and learning, and that the anonymity of their data would be ensured via a coding system. They were further notified that they could withdraw from participation at any time without consequences. The authors strictly handled student anonymity and ethical issues.
### Instruments
The test instrument for Research Questions 1 and 2 consisted of 11 items on scientific concepts and 11 items on misconceptions/neuromyths (α = 0.66 and 0.76, respectively ). 13 of these 22 items (8 items on scientific concepts and 5 on neuromyths) were taken from and translated into German. One item (“Memory is stored in the brain much like as in a computer. That is, each memory goes into a tiny piece of the brain”) was taken from and put into more concrete terms: “The brain works like a hard drive; information is stored in specific locations.” To guarantee the fidelity of the translation, the resultant version was back-translated into English by a native speaker and both English versions were compared by a third person. Five neuromyths-items on development (Myth: most receptive to learning before age 3), hemispheric asymmetry (Myth: logic in the left hemisphere, creativity in the right), memory (Myth: Genetically determined number of cells determines learning), learning while you sleep (Myth: You can learn while you sleep, e.g., via audio recordings) and evidence-based learning techniques (i.e., desirable difficulties, ; Myth: Blocked learning is better than interleaved learning) were newly constructed for this study, as they have been widely publicized in German media and learning guides. We followed ’s methodological recommendations and replaced the three-option answer format Correct / Incorrect / I don’t know used by and other studies of neuromyths with a 4-point Likert scale in order to force respondents to take a position and allow them to specify how sure they were of their answer (4 = Strongly agree/1 = Strongly disagree ) or how torn (3 = Somewhat agree/2 = Somewhat disagree ). Because the German versions of some items from were refined in terms of content and several new items were created, an English version of the instrument has been provided along with this article (see ). Future studies should note that the German version of the instrument was employed in this study (published in ).
Six instruments on professional knowledge, biography-based learning beliefs, and theory-based learning beliefs on neuroscience and learning were used to answer Research Question 3. provides an overview of the instruments and corresponding scales, numbers of items, and reliability coefficients. Example items are provided here; a complete overview of all items can be found in the . Biography-based learning beliefs were measured via 6-point Likert scales and theory-based learning beliefs via 4-point Likert scales. Professional knowledge was measured via three self-constructed knowledge tests. CK about curricular content in neuroscience was measured via six multiple-choice items with four distractors each. PCK about instructional strategies for sustainable learning (including how to deal with students’ misconceptions about the structure and function of the brain) was measured via 12 open-ended and closed-ended questions, and PPK about the psychology of human learning via 17 open-ended and closed-ended items. The differences in test construction are rooted in the project’s research focus. The interrater reliability for all open-ended items was found to be Cohen’s κ = 0.91 ( p < 0.001). This indicates almost perfect agreement ( ).
Overview of the instruments for learning beliefs and professional knowledge.
In addition, information on sociodemographic data (age, gender, field of study, years of study/training, enrolled in university courses on neuroscience and learning) were collected for all participants. Except for years of study/training and enrollment in a human biology course, these demographic data were requested for descriptive purposes only and were not explored further in the subsequent analyses.
### Data Analysis
We used multifactorial analyses of variance to test whether first-year students, advanced students, and post-graduate teacher trainees differed in their endorsement of scientific concepts and neuromyths (Research Question 1). These three groups were considered to be in different stages of teacher education due to differences in educational content: Group 1 was enrolled in introductory courses in instructional methods in science (biology) education, disciplinary content, and education science; Group 2 was enrolled in or had completed more advanced education science and subject-specific instructional methods courses on teaching and learning (in science) as well as modules that covered neuroscience (human biology and/or animal physiology); and Group 3 was in the process of completing a practical training phase after university graduation. One-way analyses of variance and Bonferroni-adjusted post hoc analyses were used to determine the extent to which the groups differed in their endorsement of individual neuromyths. One-way analyses of variance with Welch corrections and Games-Howell post hoc analyses were applied in the case of heterogeneous variance. To this end and to answer Research Question 2, the 4-point Likert scale was recoded into a dummy format ( agree/disagree ) for better comparability with the previously cited studies. First, the percentage of respondents who agreed with each item (both neuromyths and scientific concepts) was calculated. Then, the neuromyths/scientific concepts were grouped by content into different neuroscientific topics (categories) on the basis of theory. If a category had more than one item, the mean of the percentages was taken. Correlation analyses (Pearson) were conducted to determine the associations between endorsement of neuromyths and beliefs about learning and memory (biography-based learning beliefs about the definition of learning at university and use of learning strategies as well as theory-based beliefs about the nature of science and teaching and learning), professional knowledge (CK about curricular content related to neuroscience, PCK about instructional strategies for sustainable learning, and PPK about the psychology of human learning). The significance level for all analyses was p ≤ 0.05.
## Results
### Scientific Concepts and Misconceptions by Status Within Teacher Education
A multifactorial analysis of variance revealed a significant main effect for conception type (misconceptions vs. scientific concepts: F (1,1076) = 311.70, p ≤ 0.001, = 0.225) but not for stage of teacher education [first-year students, advanced students, and post-graduate teacher trainees: F (2,1076) = 1.95, p = 0.143, = 0.004]. There was a statistically significant interaction between stage of teacher education and conception type: F (2,1076) = 8.13, p ≤ 0.001, = 0.015. The mean levels in show that respondents in different stages of teacher education differed from one another in their endorsement of scientific concepts (left) but not in their endorsement of neuromyths (right).
Group comparison on endorsement of scientific concepts (left) and misconceptions (right) (mean and standard deviations are presented; 4 = strongly agree, 3 = somewhat agree, 2 = somewhat disagree, 1 = strongly disagree).
Turning to the percentage agreeing with individual neuromyths (dichotomous answer format), one-way analyses of variance only uncovered differences between the three groups of subjects with respect to the neuromyths on critical periods of childhood development [Welch’s F (2,297) = 11.84, p ≤ 0.001], blocked learning is better than interleaved [Welch’s F (2,305) = 4.80, p = 0.009], the existence of learning styles [Welch’s F (2,325) = 3.58, p = 0.029] and a genetically determined number of cells determines learning [ F (2,533) = 24.29, p ≤ 0.001]. Games-Howell post hoc analyses revealed significantly greater agreement with the myth of critical periods of childhood development among advanced students compared to first-year-students and post-graduate trainees (68 vs. 56%, p = 0.049, 0.12, 95%-CI[0.00, 0.24] and 43%, p ≤ 0.001, 0.25, 95%-CI[0.13, 0.37]). The neuromyth that blocked learning is better than interleaved was more frequently rejected by advanced students and post-graduate trainees than by first-year students (49% p = 0.021, -0.14, 95%-CI[-0.27, -0.02] and 46% p = 0.017, -0.16, 95%-CI[-0.3, -0.02] vs. 62%). The neuromyth on the existence of learning styles was significantly less frequently endorsed by advanced students than by first-year students (90 vs. 97%, p = 0.023, -0.06, 95%-CI[-0.12, -0.01]). A Bonferroni-adjusted post hoc analysis revealed that the neuromyth that a person’s genetically determined number of cells forms an upper limit for learning success was actually endorsed significantly ( p ≤ 0.001) more by post-graduate trainees (64 vs. 33% of first-year students 0.32, 95%-CI[0.18, 0.45], and 31% of advanced students, 0.33, 95%-CI[0.21, 0.45]).
### Endorsement of Misconceptions and Scientific Concepts
Pre-service science teachers’ neuroscience literacy was to a large extent rooted in neuromyths ( ). 10 of 11 misconceptions on the topic of learning and memory were endorsed by more than half of respondents. The existence of learning styles, the effectiveness of Brain Gym, and the notion that information is stored in specific locations (hard drive) were endorsed most frequently (with 93, 92, and 85% of respondents agreeing with these items, respectively). The only neuromyth to be endorsed by fewer than half of the pre-service teachers in the sample was the notion that a person’s genetically determined number of cells determines learning success (with 40% of respondents agreeing).
Agreement with misconceptions (neuromyths) among all participants.
Neuromyths were sometimes endorsed even when respondents had thematically similar scientific concepts ( ). This was seen in the categories of development, memory, learning techniques, brain activity, and sensory modalities. On the other hand, high levels of agreement with neuromyths were found in categories in which fewer respondents had thematically similar scientific concepts. This was the case for the categories of neuroplasticity and hemispheric asymmetry.
Comparing endorsement of scientific concepts and misconceptions.
### Correlations With Beliefs and Professional Knowledge
As can be seen in , the advanced students for whom individual aspects of professional competency were investigated tended toward agreement with respect to biography- and theory-based learning beliefs. Transmissive beliefs were endorsed to a lesser extent than constructivist beliefs. The latter received the highest average agreement alongside nature of science beliefs. On the knowledge tests, the advanced students were able to correctly answer about 50% of the CK, 20% of the PCK, and 30% of the PPK questions. The standard deviations here varied more widely than they did for beliefs.
Correlations of misconceptions with learning beliefs and professional knowledge.
Correlational analyses revealed only a small positive correlation between neuromyths und constructivist beliefs about teaching and learning ( r = 0.313, p = 0.006). This correlation within the area of theory-based learning beliefs means that pre-service science teachers who endorse misconceptions also exhibit a constructivist view of teaching and learning and think of learning as an active, self-directed, constructive process in which knowledge cannot simply be transferred to the learner ( ). No correlations were found between misconceptions and the other theory-based learning beliefs (transmissive and nature of science beliefs) or the three areas of professional knowledge on neuroscience and learning (CK, PCK, PPK). Nor were there any significant correlations between misconceptions and biography-based learning beliefs related to respondents’ subjective definition of learning or inventory of learning strategies ( ).
## Discussion
### Pre-service Science Teachers’ Scientific Concepts
The results of our study demonstrate that pre-service science teachers’ scientific concepts on learning and memory increase over the course of their training. This finding is in accordance with expectations, because German science teachers specializing in biology complete modules containing neuroscientific content (human biology and animal physiology) during their university studies. It should be noted that average endorsement of scientific concepts did not increase dramatically and was quite high even among first-year students. Based on these findings, one could conclude that many pre-service biology teachers have already acquired scientific concepts during school (e.g., in advanced high school biology courses) and bring them with them to university. However, from a critical perspective, it should be noted that our instrument did not allow us to measure what the students actually know and when they simply took a position despite a lack of knowledge (intuiting/guessing). This problem was strengthened by our use of a Likert scale, which was however recommended by . Despite differences to instrument, our survey was able to confirm their finding that there is a general tendency to agree with neuroscientific statements. Our results indicate that this tendency persists despite academic and practical training.
Our results further indicate that the pre-service science teachers in our study have the weakest scientific concepts with respect to neuroplasticity and hemispheric asymmetry (60 and 40%). This could be because these topics tend to be covered only marginally or as an aside in neuroscience courses and textbooks. From a critical perspective, it should be noted that these values are location-specific and could be different at other German universities. There is currently no curriculum stipulating which scientific concepts must be covered as part of science teachers’ training in the fundamentals of neuroscience. One item from each of the two aforementioned categories adopted from have also been employed in other studies of neuromyths. Putting aside the differences in answer format (our 4-point Likert Scale vs. correct/incorrect/I don’t know ) and the slightly more concrete items in our translation, comparing our results to previous studies indicates that German pre-service science teachers have stronger scientific concepts than Turkish and British pre-service teachers with respect to the items “ When one brain region is damaged due to injury, other parts of the brain can take up its function ” and “ The left and right hemispheres of the brain always work together in processing information ” ( : 20 and 15% correct answers, : 12 and 14% correct answers, although around 30% of respondents in both studies selected I don’t know ). The pre-service science teachers in this study endorsed the topics of development, memory, learning techniques, brain activity, and sensory modalities at very high rates (98-92%). The values of all matching items were higher than in the study by , although the instruments’ differences in language and answer format must be taken into account. The presented tendencies concerning overlapping items indicate that pre-service science teachers seem to have stronger scientific concepts related to neuroscience than other pre-service teachers. This should be tested in a study employing the same instruments for both groups of participants.
### Pre-service Science Teachers’ Misconceptions (Neuromyths)
Given that this study’s quasi-longitudinal design found no differences in endorsement of neuromyths between first-year students, advanced students , and post-graduate trainees , teacher education does not seem to be able to successfully professionalize students’ misconceptions about learning and memory. Only two neuromyths about learning and memory (critical periods and blocked learning) were endorsed less by post-graduate teacher trainees, who have already completed their university training in neuroscience and learning, than by students still in university. In fact, the neuromyth that a person’s genetically determined number of cells forms an upper limit for learning success was endorsed more frequently among the group of post-graduate trainees. This increase is alarming, as belief in this type of myth bestows upon or denies learners a pre-determined, non-malleable aptitude for learning. This could have consequences for teachers’ interactions with students and thus also for students’ self-efficacy beliefs. Whether and to what extent these individual neuromyths find their way into pre-service science teachers’ later pedagogical practice is not clear on the basis of our study. We join in arguing that future studies must test the extent to which the endorsement of neuromyths influences teachers’ effectiveness. However, other studies show that teachers’ beliefs and attitudes guide their actions ( ). We assume that university education represents the most significant opportunity for German science teachers to acquire neuroscientific knowledge in a guided way. In light of the previously cited studies of pre-service and in-service teachers in all school subjects (e.g., ) revealing comparatively high levels of endorsement of neuromyths, we do not assume that in-service science teachers endorse misconceptions to a lesser extent than the pre-service teachers in our study. It might even be the case that committed efforts among in-service teachers to optimally guide students’ learning lead to greater use of practical approaches based on neuromyths, such as Brain Gym or learning styles. In this way, neuromyths might be even more widespread among German in-service science teachers than pre-service teachers. However, further comparative and longitudinal studies are necessary to investigate these hypotheses. In any event, teacher education as it currently exists in Germany does not seem sufficient to dismantle misconceptions about learning and memory or replace them with scientific concepts. New, more effective professional development opportunities and learning programs must be created.
The results presented in this study confirm previous findings that it is not just in-service teachers who believe in neuromyths – a large share of pre-service teachers endorse them as well ( ; ; ; ; ; ; ; ). Out of a total of 11 misconceptions (neuromyths) about learning and memory, the existence of learning styles (93%), the effectiveness of Brain Gym (92%), and the assumption that information is stored in specific locations (hard drive) (85%) were endorsed most frequently. Comparing our results to those of other studies (despite the difference in answer format and the slightly more concrete items in our translation), these myths were also quite frequently endorsed by pre-service teachers in Turkish ( : 97, 67, and 79%) and British studies ( : 82, 62, und 36%, : 94, 37% und -%) . There thus appears to be a core group of neuromyths whose prevalence is independent of culture.
Furthermore, German pre-service teachers believe more frequently than Turkish ( ) or British pre-service teachers ( ; ) in the neuromyths of only using 10% of our brain (57% compared to 42, 52 and 47%) and critical periods of childhood development (item = If the brain is not sufficiently supported in early childhood, learning problems that can no longer be remediated by education can occur : 59% compared to 59, 9, and 24%) . Comparing our results to those of , it appears that German pre-service teachers believe more strongly in the myths of learning differences due to the use of different hemispheres (82% vs. 55%, although I don’t know was selected quite frequently in ) and learning while you sleep (56 vs. 38%) . This points to cultural differences in levels of agreement with individual neuromyths among pre-service teachers, just as among in-service teachers. Moreover, it indicates that the characterization of neuromyths as ‘inadequate scientific concepts’ is insufficient. Instead, these cultural differences suggest that neuromyths to a large degree feed off of a socio-cultural discourse that finds its specific expression – cultural differences included – in these neuromyths. Thus, these misconceptions might be better described as scientific myths (cf. ). Empirical inquiries and interventions should take socio-cultural discourses about teaching and learning into consideration as the relevant context of neuromyths. This study also found evidence for endorsement of the myth that blocked learning is more effective than interleaved learning (i.e., desirable difficulties, ) for the first time. Thus, there is a need for action in teacher education with respect to the aforementioned neuromyths, particularly in light of the importance of a professional understanding of learning and memory for instructional content and instructional methods in science.
### Relations Between Neuromyths and Aspects of Professional Competency
Turning to the aspects of professional competency, firstly, our results with respect to theory-based learning beliefs were in accordance with expectations. Our hypothesis that constructivist and nature of science beliefs would be stronger among advanced students were confirmed. In accordance with the existing research literature (e.g., ), transmissive beliefs about teaching and learning were in turn less strong among these students. Thus, for these students, learning is more of an active than a passive process in which knowledge can be generated independently. This can be interpreted together with professional beliefs on the origins of knowledge in biology (nature of science) as evidence in favor of a professional understanding of learning and memory. The students’ biography-based learning beliefs can also be positively interpreted in light of the rather strong deployment of learning strategies and the definition of learning as transformation (professional definition of learning). However, the existing research literature ( ) also indicates that advanced students maintain less professional definitions of learning (learning as reproduction). In fact, in our study, these were even more widespread than the professional ones. Whether and to what extent these beliefs influence the students’ later actions in schools remains an open question. Our results further indicate that students have some knowledge of PCK about instructional strategies for sustainable learning and PPK about the psychology of human learning, but not a great deal. We see this as primarily rooted in our methodological decision to select advanced students who had completed human biology courses with neuroscientific content. We have no information on the extent to which this sample is also at an advanced level with respect to educational science, psychology, and instructional methods course. However, the CK about curricular content in neuroscience we selected for was present to a stronger extent, although this could also be rooted in the closed answer format. The larger standard deviations for the knowledge results point to differences in the students‘ performance.
Our results show that all of the aspects of professional competency we investigated were present among the students. However, there were few correlations with endorsement of neuromyths. Building upon studies that call for more neuroscience in teacher education (e.g., ), we expected negative correlations with students’ professional knowledge, which would mean that students with greater knowledge endorse neuromyths to a lesser extent. However, no such correlations were found for CK about curricular content related to neuroscience, PCK about instructional strategies for sustainable learning or PPK about the psychology of human learning. For PCK and PPK, this might be rooted in the students’ low levels of knowledge or methodologically in the difficulty level of the tests, which was too high. We reject this possibility with respect to CK about curricular content related to neuroscience, as the students had more knowledge here. Our results indicate that neuromyths are independent of CK. However, it must be emphasized that our instrument only asked about curricular content in neuroscience with respect to the topics of brain structure, memory, and long-term potentiation commonly found in school textbooks. General knowledge of neuroscience and the latest research findings were not considered in this study and could still be a predictor of endorsement of neuromyths.
In addition, the results of our study revealed no correlations with respect to a person’s learning at university (definition of learning and learning strategies), despite our theoretical assumptions indicating that less professional biography-based learning beliefs could facilitate the endorsement of neuromyths. Perhaps our scale on learning as reproduction, with α = 0.58, could not measure the theoretical construct sufficiently and accurately enough. Future studies could also ask about concrete learning experiences that could promote neuromyths (e.g., if they conducted learning style tests during their school years). However, this study found a small positive correlation between misconceptions and constructivist beliefs about teaching and learning despite the fact that some neuromyths (e.g., that the brain works like a hard drive) are not theoretically compatible with such beliefs. People with constructivist views of teaching and learning actually view learning as an active, self-directed, constructive process in which knowledge cannot simply be transferred to the learner ( ). Consequently, neuromyths seem to be integrated into the semantic network of theory-based learning beliefs despite their scientific inconsistencies, which can make them more difficult to change. The co-existence or even synthesis of misconceptions and theory-based beliefs are well-established in theories of misconceptions and conceptual change ( ). Studies by describe how university students can stubbornly hold onto their original concepts despite empirical demonstrations and theoretical explanations. demonstrated this specifically for the neuromyth on the existence of learning styles. In addition, the authors warn of a “backfire effect,” a phenomenon in which attempts to address myths and misunderstandings can lead to a strengthening of beliefs in these myths. Thus, the results of this study indicate that interventions against the endorsement in neuromyths must begin deep in participants’ belief systems.
### Implications for Intervention Approaches With Respect to Neuroscience Literacy
Overall, our results for pre-service science teachers show that call to integrate neuroscientific content into teacher education is not in itself sufficient to limit the spread of neuromyths. Even these students, who are taught such content in their courses, must be trained to become critical consumers of neuroscientific research findings (a kind of preventative focus; ).
In line with the previously cited neuromyth studies, our findings confirm the need for action with respect to neuromyths. Few intervention approaches have been proposed. stresses the importance of developing an understanding of how neuroscience research is conducted and presented (e.g., imaging techniques using differential images). This might not be occurring to a sufficient extent in German pre-service science teacher education, which is primarily set up to enhance professional knowledge. Neuroscience will always be in a continuous state of development and progress. Pre-service science teachers should be put in a position in which they are able to follow the latest developments by effectively reading and critically evaluating the information they obtain from various sources.
According to and , ), one of the most effective evidence-based methods of addressing scientific myths consists of directly refuting misunderstandings. have confirmed this for neuromyths. Neuromyths arise and persist from an entire line of argumentation consisting of misinterpretations and exaggerations that can only be refuted with a multitude of neuroscientific facts (examples provided in of this study). This speaks in favor of examining each neuromyth individually, determining its “kernel of truth,” uncovering its argumentation structure, and comparing it to the corresponding scientific concepts. Only by investigating each neuromyth individually and in more detail than previously will we be able to determine why neuromyths have been spreading and develop appropriate interventions to stop them. This requires strong cooperation between education, neuroscience, and cognitive psychology. demonstrate the positive effects of such cooperation. Materials and courses counteracting the false transfer of scientific concepts to classroom teaching and learning need to be jointly developed. A web of exchange between the field of education and neuroscientists is required to put neuromyths into a more scientifically accurate light. Neuroscientists’ assistance is particularly necessary when it comes to incorporating the latest research.
Several neuromyths found among pre-service science teachers in this study have also been frequently demonstrated in international studies of in-service teachers (cf. e.g., ; ). Given that even first-year students exhibit beliefs in neuromyths, it is likely that pre-service teachers encounter these misconceptions even before beginning their university studies, i.e., during school. It is well-known that neuromyths such as the theory of learning styles and Brain Gym exercises are found in a large number of learning guides and educational programs ( ). Students can also encounter neuromyths through their teachers. Studies by provide empirical indications that misconceptions (e.g., that the brain works like a hard drive) are present among school students. To the best of our knowledge, there are not yet any studies systematically investigating the spread of neuromyths about learning and memory among school students. What we do know is that the misconceptions that develop over a person’s school years are difficult to change via formal education at university ( ). Starting intervention during students’ school years or at the beginning of university education seems advantageous.
### Recommendations for Future Studies
As previously discussed, the results of our study show that pre-service science teachers have weak scientific concepts on neuroplasticity and hemispheric asymmetry. Our results further show that the low levels of scientific concepts for these topics were accompanied by high levels of endorsement of thematically similar neuromyths. Based on our findings, one might conclude that these topics need to be more strongly integrated into teacher education and associated neuroscience teaching materials. Even though pre-service science teachers endorsed scientifically accurate statements (scientific concepts) about the topics of development, memory, learning techniques, brain activity, and sensory modalities, the thematically similar neuromyth items were widely endorsed as well (50–93%). In these cases, we concur with and follow-up studies by other authors that there is a lack of ability to differentiate scientific concepts from misconceptions. Endorsement of misconceptions (neuromyths) was not lower for topics in which there was less knowledge of scientific concepts (93% for learning styles despite high endorsement of scientific concepts; 92% for Brain Gym amid low endorsement of scientific concepts). Consequently, the results obtained with our survey suggest two different causes for the emergence of neuromyths: a lack of scientific concepts, but also false transfer of accurate scientific concepts to teaching and learning. Future studies should further clarify these different explanations for the emergence of neuromyths. In doing so, it would be advantageous to not only contrast thematically similar items on scientific concepts and neuromyths, as we did in the questionnaire for this study, but rather to consistently identify and inquire about the kernel of truth and unique argumentation errors for each neuromyth (see ). Given the current state of theoretical work on neuromyths, we were only able to do this for a few categories (e.g., sensory modalities). Much more theoretical work on neuromyths needs to be completed with respect to this issue. With regard to intervention approaches, future studies must also test whether the theoretical argumentation in favor of neuromyths previously described actually conform to those held by pre-service and in-service teachers. Thus, we see a need for more intensive empirical and theoretical research on neuromyths.
In this study, we were able to show that pre-service science teachers endorse a variety number of neuromyths. We did not collect data on the sources of the students’ neuroscientific information or their perceptions of the origin of their misconceptions on the topic of learning and memory. Biography-based learning experiences in the everyday world, independent learning by reading scientific/leisure magazines, or even the structure of university trainings could be potential sources of neuromyths. Following , the readability of neuroimaging articles and their abstracts could be particularly problematic, especially for first-year students. All of these aspects should be more thoroughly investigated in future studies. In our opinion, qualitative studies in which students are asked to describe their arguments in favor of neuromyths seem more worthwhile than surveys of various sources of information.
### Summary and Outlook
In summary, our results indicate that neuromyths can exist in parallel to scientific concepts, professional knowledge and beliefs about neuroscience and learning and are resistant to conventional German teacher education, which promotes many aspects of professional competency on this topic. From the perspective of conceptual change theory, neuromyths thus exhibit characteristic traits of misconceptions that cannot simply be counteracted with increased neuroscientific knowledge. Both neuroscientific knowledge and didactic interventions will be required to effectively and sustainably banish neuromyths from the education system. For this reason, we call for stronger cooperation between neuroscientists and didactics experts. Creating links between education/didactics on the one hand and neuroscience and cognitive psychology on the other seems to be essential for confronting neuromyths’ lines of argumentation with scientific knowledge as well as improving science teachers’ neuroscience literacy. According to the , each discipline has its own specific methods and language, which makes it particularly difficult for experts in one area to apply knowledge from the other. Joint publications and training programs for pre-service and in-service teachers on the cognitive errors involved in neuromyths would be an important first step toward eliminating ‘language barriers’ ( ) and closing the gap between neuroscience and the practice of education ( ), at least with respect to neuromyths. Teachers train the neuroscientists of tomorrow. It is therefore important to take teachers seriously, to investigate how neuroscience can help them better understand learning, and to invest in their ability to optimally use neuroscience in their practice.
On the basis of this study’s results, the University of Kassel has developed a learning environment in accordance with the conceptual change model through interdisciplinary cooperation. This learning environment gives students reasons and opportunities to more closely interlink their professional knowledge in neuroscience, cognitive psychology, and instructional methods in science and to critically question incomplete or incorrect misconceptions and beliefs about learning and memory. An accompanying study demonstrated positive results with respect to pre-service science teachers’ neuroscience literacy ( ).
## Data Availability Statement
The raw data supporting the conclusions of this manuscript are available on Open Science Framework. The link will be provided by the authors upon request.
## Ethics Statement
No ethics approval was required for the reported study as per the guidelines of the University of Kassel or national guidelines. We conducted the study in line with the recommendations of the University of Kassel’s ethics committee. All participants gave written informed consent in accordance with the Helsinki Declaration.
## Author Contributions
JM was responsible for the project administration and funding acquisition. Both authors developed the study concept and the methodology. FG developed the study materials, conducted the data collection, and analyzed the data. This article is part of the Ph.D. thesis of FG supervised by JM. FG drafted the manuscript. JM provided critical revisions. Both authors approved the final version of the manuscript for submission.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
In our daily life, we frequently need to make decisions between competing behavioral options while we are exposed to various contextual factors containing emotional/social information. We examined how changes in emotional/arousal state influence resolving conflict between behavioral rules. Visual stimuli with emotional content (positive, negative and neutral) and music (High/Low tempo), which could potentially alter emotional/arousal states, were included in the task context while participants performed the Wisconsin Card Sorting Test (WCST). The WCST requires the application of abstract matching rules, to resolve conflict between competing behavioral options. We found that conflict influenced both accuracy and response time (RT) in implementing rules. Measuring event-related autonomic responses indicated that these behavioral effects were accompanied by concomitant alterations in arousal levels. Performance in the WCST was modulated by the emotional content of visual stimuli and appeared as a faster response and higher accuracy when trials commenced with negative emotional stimuli. These effects were dependent on the level of conflict but were not accompanied by changes in arousal levels. Here, we report that visual stimuli with emotional content influence conflict processing without trial-by-trial changes in arousal level. Our findings indicate intricate interactions between emotional context and various aspects of executive control such as conflict resolution and suggest that these interactions are not necessarily mediated through alterations in arousal level.
## Introduction
Executive control ( ; ) is essential for optimizing the flexible use of limited cognitive resources to currently prioritized tasks, in order to support goal directed behavior ( , , ; ). This control, amongst other processes, may be achieved via the detection and resolution of conflict between behavioral choices ( ; , ; ; ; ). Conflict-induced behavioral adjustments appear as a decline in performance when conflict arises in a trial (conflict cost) and also as an improved ability to resolve the conflict in the subsequent trial (conflict adaptation) ( , ).
Recent studies indicate that executive control processes are influenced by emotional factors ( , ; ; ). There has been long lasting debate regarding the mechanisms underlying the interaction of emotion and executive control of behavior when a choice between conflicting-competing options becomes necessary ( , ; ). The dual competition model explains the interaction of emotion and executive functions in terms of shared processing resources ( ; ; ; ; ). This model postulates that task performance is impaired in the presence of any task-irrelevant stimuli, whether emotionally salient or not, as resources for the primary task are utilized toward processing this external stimuli ( ). This proposition was validated in a study which revealed that the presentation of task-irrelevant images presented immediately prior to trials, increased reaction time ( ; ). Other studies also suggest that the interaction between executive functions and emotional regulation ( ) might be mediated through overlapping neural substrates as imaging studies have revealed a large overlap in brain areas which support executive functions and those which regulate emotional state ( ). In this view, the demand for attentional-cognitive resources by each regulatory system might shape the outcome of interaction between emotional stimuli and executive functions and influence the ability to resolve conflict in various cognitive tasks.
In contrast to this, other models propose that priming of emotional state may specifically influence conflict monitoring and resolution ( , ). This model stems from the proposal that conflict is innately aversive ( ; ), and postulates that conflict resolution is sensitive to emotional state, as it is dependent on the assessment of conflict as an aversive cue ( ). This proposal was validated in studies where conflict-induced behavioral adjustments were either attenuated when paired with a reward, as the reward diminished the aversive assessment of the conflict ( ), or enhanced when paired with a negatively valenced stimuli, as the stimuli heightened the aversive assessment of the conflict ( ).
Other models also propose that conflict has an affective cost (a negative influence upon emotion) while inducing behavioral adaptation to reduce further costs if this conflict reappears ( ). A related view proposes that resolving conflict is cognitively demanding and has an inherent cost, thus people aim to minimize cognitive efforts and avoid instances which induce them ( ). Complimentary to this, it has been shown that conflict between behavioral choices trigger negative emotional state and avoidance behavior ( ; ). It has been proposed that such negative emotion is an important factor in triggering and recruiting executive control to resolve the conflict ( ). These studies suggest that the emergence of conflict and the efforts to resolve it or avoid its reoccurrence are associated with emotional state change and therefore what appears as interaction of emotion and executive control is an inherent property of conflict processing ( ).
The interaction of emotional state and conflict might also be considered as a subset of a broader model of interaction between arousal-emotional state and decision-making processes. The somatic marker hypothesis postulates that alterations in emotional regulation and arousal state may influence the decision-making process ( , ). In this context, conflict and its associated cognitive difficulty or uncertainty might induce emotional state change with concomitant autonomic and somatic responses and consequently influence executive functions and conflict resolution ( ).
In this study, we aimed at examining the interaction of background emotional information on executive functions, and specifically conflict processing. A prominently used neuropsychological test for assessing executive control, particularly the ability in resolving conflict between competing behavioral rules, is the Wisconsin Card Sorting Test (WCST) ( ; ; ). The WCST assesses the participants’ ability to shift between abstract rules (which change without notice) using trial and error ( ). Previous studies have revealed that this rule-shifting leads to conflict between potential rules, and impairs both accuracy and response time (RT) (conflict cost) ( ; ; ).
We hypothesized that contextual factors with emotional content might convey emotional information and influence cognitive functions, particularly when resolving conflict requires recruitment and allocation of executive control ( ). Such an interaction between emotional regulation and cognitive control might influence the ability to resolve the conflict. Therefore, we introduced two sources (from two different modalities) of information with emotional content, “visual stimuli with emotional/social content” and “music,” while participants performed different versions of the WCST.
Previous studies have revealed that music may improve ( ; ) or reduce ( ; ) performance in various perceptual, motor, and cognitive tasks ( , ; ). It has been proposed that these effects may be mediated through the direct influence of music over executive functions by limiting available cognitive resources ( ; ; ). Music may also influence emotional state and arousal ( ) and indirectly alter the interaction of emotional and executive control processes, consequently modulating performance in cognitive tasks ( ; ). Our previous studies have indicated that music tempo is a crucial component of the acoustic stimuli which alters subjective experience, so that high- and low-tempo music evoke happy and sad feelings, respectively ( Control Study 2). In addition, high-tempo, but not low-tempo music, significantly affected the practice-related alterations in the response inhibition and this was accompanied by concomitant changes in arousal response in the context of Stop signal task ( ). Therefore, in this study we included high-tempo and low-tempo music as background acoustic conditions while participants performed the cognitive task and hypothesized that music would affect the emotional-arousal state and influence conflict resolution.
Previous studies have also shown that visual stimuli with emotional content influence performance in the context of various cognitive tasks ( ; ; ; ). Inclusion of two sources of emotional stimuli from two different modalities (visual stimuli and music) could potentially help delineating the interaction of various emotional stimuli in influencing conflict processing. If both emotional factors elevate the arousal level their additive effects might appear as larger alterations in behavioral measures in the cognitive task.
We also monitored participant’s event-related arousal/emotional response during cognitive task performance in all conditions. Rapid transient shifts in electrodermal activity (EDA) ( ; ; ) can be measured as an index of arousal level and reflect emotional state change by aspects of cognitive task (e.g., conflict and error commission) and/or external modulating factors, such as music ( ; ; ; ; ; ; ; ). We hypothesized that if changes in arousal level mediate the interaction of emotion and conflict processing, this would be reflected in concomitant changes in event-related EDA.
## Materials and Methods
### Participants
Fifty-five Monash University undergraduate (third year) students (35 female) were recruited to complete the WCST in three separate 2-hour sessions, each one week apart. A priori power analysis was performed for the estimate of required sample size based on data obtained and reported in our recently published study ( ). The effect size for conflict-induced behavioral adjustment (conflict adaptation) was estimated using Cohen’s criteria ( ) based on , which was 0.156. With an alpha set at 0.05, and power set at 0.80, the estimated sample size required for this effect size was 36 participants (using G Power 3.1 ). To achieve counterbalancing for all conditions we recruited even more participants. Thus, our sample size of 55 is adequate for the detection of the smallest effect with 80% power, given the current design. All participants (age range between 20 and 27) had similar educational background and no history of any neurological disorders. Approval was obtained from Monash University Human Research Ethics Committee. Written and informed consent was obtained from all participants.
### Apparatus
An automated test apparatus was used to perform the WCST. Participants were seated in front of a switch and a touch-sensitive screen (17 MicroTouch surface capacitive touch display) on which the stimuli were displayed. Participants were approximately 60 cm from the screen and were informed to maintain focus in the middle of the screen. The size of each stimulus ranged between 5 and 7 cm. The switch was fixed on a pad with a wrist rest in line with the bottom center of the screen. The subjects were informed to only use their dominant index finger to press the switch and then the target stimulus on the screen.
Participants were also monitored via camera during the completion of the task. Data acquisition was controlled by CORTEX program (National Institute of Mental Health) at millisecond resolution. Before performing the WCST, participants read an explanatory note regarding the test procedure following this a structured verbal briefing was also provided, using a written script to ensure consistency.
### Procedure
The main task shown in has been described and validated in previous studies ( , , ). The start cue was an emotionally salient image (positive, negative or neutral images) presented randomly at the start of each trial. After the start cue onset, participants had to press the switch with their dominant index finger within 10 s. If the switch was pressed, the start cue was replaced by a sample stimulus (for 400 ms). If the participant kept the switch pressed, three test items appeared surrounding the sample (to the left, right, and below the sample stimulus). The sample stimuli presented were selected at random from a set of 36 stimuli (combination of 6 colors and 6 shapes). The test items were also selected from this pool at random (with restrictions imposed to meet the necessity to generate either a congruent or incongruent condition). Once test items were displayed, participants had to rapidly release the switch and touch the appropriate test item that matched the sample according to the relevant rule. Participants had to respond as quickly as possible within a limited time window (900 ms). Failure to touch the screen within this window was considered as a time-out error. If target selection was correct, confirmation feedback was given to the subject (the target item flashed twice). However, if erroneous, all presented stimuli were replaced with a large error signal (a purple annulus) which was displayed for 500 ms. Early release of the switch during the presentation of the sample stimuli alone was also considered as an error.
Wisconsin Card Sorting Test (WCST). The incongruent (also termed high-conflict) and congruent (also termed low-conflict) trials were randomly presented, in equal proportion, throughout the testing period.
The appropriate matching rule (either color or shape) was maintained in blocks of trials, and changed without any notice when the shift criterion (9 correct responses from 10 trials) was met. Within the task there were two trial types, incongruent and congruent, with two levels of conflict between the behavioral rules ( ). Within incongruent trials the sample stimuli matched one of the test items in color, yet another item in shape. The other test item did not match the sample in either shape or color. Therefore, participants had to resolve the conflict between the two potentially matching targets in applying the appropriate rule and selecting the correct target item. However, within congruent trials, the sample stimuli matched one of the test items in both color and shape, but did not match with the other test items in color nor shape. Thus, in congruent trials there was only one potentially matching target.
Each weekly testing session comprised of two stages of the WCST with a 10-minute break between stages to minimize fatigue effects. Each of these two testing stages ran for approximately 30 to 45 minutes, dependent on performance. The task commenced with three practice blocks, in which data was not analyzed. The first practice block included only congruent trials, while the second and third practice block included only incongruent trials, requiring the application of the color or shape rule, respectively. Within these practice blocks the shift criterion was set at 5 correct responses from 5 trials. Following these practice blocks, 10 blocks were ran, alternating between the color or shape rule with the aforementioned rule shift criterion (9 correct responses from 10 trials).
### Electrodermal Activity Recording
Event-related EDA was recorded, in microsiemens (μS), continuously for the entire duration of the task (at 75 kHz sampling rate using an electrodermal recording unit: ML116 GSR Amp and PowerLab 26T, ADInstruments). Task-relevant events were simultaneously encoded during EDA recording. The electrodes were connected to the palmar surface of the index and ring fingers of the non-dominant hand, and participants were informed to keep this hand stationary during testing.
Amplitude of phasic activity was measured as the difference between the maximum and minimum value of the EDA waveform within an event-related epoch ( ). Within this study, two epochs were used: a post-feedback epoch [response selection onward (4 s)], to examine changes in arousal level after the awareness of trial outcome, and a pre-feedback epoch [From the start cue to response feedback onset (2.4 s window)], to examine changes in arousal level before feedback.
### Visual Images With Emotional Content
We used music and visual stimuli as “background contextual stimuli” without any relation to the rule-based target selection which was the main task in the WCST. For visual stimuli, multi-color images of items and faces were categorized (by one female and one male) to images with positive, negative and neutral emotional content.
At the start of each trial, multi-colored natural or cartoon images including human faces were shown as the Start-cue. These visual images were categorized to three conditions with positive, negative and neutral emotional content by one female and one male before the start of data collection. Visual images were not repeated in the testing session to avoid multiple exposure (trial unique design). Participants performed the test in three consecutive weeks and none of the visual stimuli was repeated. Trial unique design for exposing the participants to emotional stimuli was a crucial aspect of our study because any repeated exposure would have introduced confounding issues such as familiarity and memory processing and indirectly evoked other cognitive processes and associated changes in arousal state. Participants were not aware of the categorization of the visual images based on their emotional content and were instructed to press a switch and initiate the trial as soon as they see any image. The presentation of the images as start cues, enables the examination of the influence of emotional stimuli on behavior and on arousal state. Concomitant alterations in arousal level was assessed in before (pre-feedback epoch) or after feedback (post-feedback epoch) to participants’ responses.
We also replicated the study using stimuli from the Nencki Affective Picture System (NAPS). NAPS is a large collection of multicolor photographs classified into five categories (People, Faces, Animals, Objects, and Landscapes) and are controlled in terms of “dimension,” “luminance,” “entropy,” and “contrast“( ; ). In Control Study 2 a subset of 360 images (same size of image pool as the main study) were randomly selected from the NAPS pool of images (120 for each emotional category: positive, negative, neutral). In NAPS, the emotional aspects of images were rated on a scale of 1–9 by 204 people, on subjective “valence” and “arousal.” We assigned images to three separate categories (Positive, Negative, and Neutral) based on each image’s ranking for Affective valence ( ). Images within a range of 1.00–2.50, 3.75–6.25, and 7.50–9.00 were used for the selection of Negative, Neutral, and Positive image pool, respectively.
### Music
In each of the three weekly testing sessions, participants listened to one of the three types of music: high tempo with lyrics, low tempo with lyrics, or no-lyric music (of mixed tempo) for the duration of the task. We included no-lyric music to control for any effect of lyrics. We first selected a set of pop songs and then classified them based on their tempo: low [80–100 beats per minute (bpm)] and high tempo (120–140 bpm) music. Songs were excluded if they contained any offensive lyrical statements. The volume of the music across sessions was set (at 70 decibels), however, participants were able to change it only if it was too loud or quiet. The order of the three music conditions (low tempo, high tempo and no-lyrics) was counterbalanced across weekly testing sessions (see for the list of songs).
### Data Analyses
Each of the daily testing sessions comprised of two testing stages (first and second), which allowed assessment of within-session practice-related learning ( ). We measured the time from the onset of the target stimuli to the first touch on the screen as RT. Repeated-measure Analysis of Variance tests (ANOVA) were used to assess the effects of practice, emotional images and music on various behavioral measures. All data points were used in data analyses without removal of outliers. In this study, participants had to deliver a response within a limited response window (900 ms) and if participants could not deliver their response within this window the trial was assigned as a timeout trial. Therefore, all RT data used for analyses were within this response window. Implementing arbitrary procedures for the removal of outliers might bias the results and outcome of statistical analyses. Therefore, we followed our previously published approach ( , ) and did not remove any data point as outlier and included all data points in the statistical analyses. Distributional properties for key measures are shown in .
To ease comparison of the emotional conditions (such as positive, negative, neutral) and different sessions, RT and EDA were calculated for each condition in each testing stage and then normalized by dividing each value by the grand average for all conditions. We have implemented such normalization procedure for RT ( , , , ) and EDA ( ) in our previous studies. Early switch release during sample presentation was considered as an error and participants received an error signal to avoid such errors, however, these procedural errors were not included in the calculation of accuracy. When color- or shape-matching was required, the majority of errors were perseverative errors (selecting the target based on irrelevant rule) or timeout trials and participants rarely committed non-perseverative errors (selecting the target that did not match the sample by either color or shape). Accuracy was calculated as the percentage of correct trials [correct trials/(correct + perseverative errors + non-perseverative error + timeout trials)] and analyzed without normalization. All participants successfully completed the required number of rule-shifts and achieved high percentage of correct responses in the WCST and therefore accuracy data were used without normalization ( , , ).
Each participant performed rule-based target selection at different conflict levels and therefore Conflict was included in a ANOVA as a within-subject factor. Each participant was exposed to all categories of visual images and different background Music conditions. Therefore, Emotion (visual stimuli with emotional content: positive/negative/neutral) and Music (high-tempo/Low-tempo/mixed tempo with no-lyrics) were included in the ANOVA as within-subject factors. For repeated-measure ANOVA, sphericity was examined (Mauchly’s test) and Greenhouse-Geisser correction was applied when necessary. The alpha level was set at 0.05 for all statistical tests. For significant effects, was also reported, which indicates the proportion of the variance explained by the effect in ANOVA analysis. Where significant interactions were detected pairwise comparisons were conducted. All pairwise comparisons were two-tailed t test with Bonferroni adjustment for multiple comparison. To ease interpretation when presented graphically: Significance level of p < 0.05, p < 0.01, and p < 0.001. No additional manipulations of data or measures were implemented.
## Results
We first report the results obtained in the Main blocks in which color or shape was the relevant matching rule. Participants rarely (<10%) chose the test item which did not match the current rule (color or shape), and only rarely committed errors in congruent trials. Within incongruent trials, most errors committed were perseverative, defined as selecting the target item which matched the sample by the alternative rule. These perseverative errors were mostly committed immediately after the rule change.
### Emotionally Negative Visual Stimuli Enhance Conflict Resolution
Music and visual stimuli with emotional content can potentially influence the emotional/arousal state. To examine the interaction of these factors with aspects of executive control such as conflict processing and practice-related learning, we applied a multi-factor ANOVA [Conflict (congruent/incongruent trials, within-subject factor) × Music (high tempo with lyrics/low tempo with lyrics/mixed tempo with no-lyrics, within-subject factor) × Practice (first/second stage of testing, within-subject factor) × Emotion (visual stimuli with emotional content: positive/negative/neutral, within-subject factor)] to RT in correct trials. The main effect of Conflict was significant, F (1,54) = 222.09; p < 0.001, = 0.80, indicating that the higher level of conflict in incongruent trials increased RT (conflict cost) ( ). There was a significant main effect of Practice, F (1,54) = 14.51; p < 0.001, = 0.21, indicating that RT decreased from the first to the second stage in the same testing day (within-session learning).
Importantly, the main effect of Emotion was significant, F (2,108) = 9.20; p < 0.001, = 0.15, indicating that the emotional content of the visual stimulus influenced RT. Pairwise comparisons (two-tailed t test with Bonferroni adjustment for multiple comparison) of the differences in RT for each emotional condition revealed that trials which commenced with negative emotional stimuli had the lowest RT in comparison to those which commenced with a positive ( p < 0.001) or neutral ( p = 0.02) stimuli ( ). Furthermore, there was also a significant difference in RT between trials which commenced with positive stimuli and those with neutral stimuli ( p = 0.03). We also used another set of visual stimuli (NAPS image system) in another participant cohort ( Control Study 2). The results with the new image set replicated and confirmed that exposure to the negative stimuli enhanced performance, as indexed by the lowest RT and higher accuracy in comparison to positive and neutral stimuli exposure ( ).
Emotional content of images shown as the start cue influenced performance in the WCST. (A) Response time (RT) in correct trials are shown for each emotional category. RT was lowest in trials which contained a negative visual stimulus (shown as the start cue) (B) The percentage of correct trials are shown for each emotional category. Accuracy was the highest in trials which contained a negative visual stimulus. represents p < 0.05, represents p < 0.001.
There was also a significant interaction between Conflict and Emotion, F (2,108) = 26.37; p < 0.001, = 0.33, indicating that the effect of emotional stimuli was dependent on the level of conflict. RT was shorter in incongruent trials which contained negative stimuli however, such a decline in RT was not observed in congruent trials ( ). shows the difference in RT between congruent and incongruent trials (conflict cost for each emotional category). Pairwise comparisons (two-tailed t test with Bonferroni adjustment for multiple comparison) of the differences in conflict cost for each emotional condition revealed that negative stimuli attenuated the magnitude of conflict cost in comparison to positive ( p < 0.001) and neutral ( p < 0.001) stimuli.
Behavioral effects of conflict were modulated by the emotional content of visual stimuli. (A) RT in correct trials are shown for congruent and incongruent trials for each emotional category. The effects of emotional stimuli was dependent on the conflict level. Incongruent trials, which commenced by negative stimuli had a decreased RT, while in congruent trials this effect was not observed. (B) The difference in RT between congruent and incongruent trials, conflict cost, is shown for each emotional category. Negative stimuli attenuated the magnitude of conflict cost. (C) Rate of learning (the difference in RT between the first to the second stage of testings in the same daily session) is shown for congruent and incongruent trials for each emotional category. Within session learning was dependent on the emotional valence and the conflict level. (D) Difference in learning between incongruent and congruent trials is shown (incongruent - congruent) for different emotional conditions. Learning decreased the conflict cost (difference between incongruent and congruent trials) in negative emotional conditions. represents p < 0.05, represents p < 0.001.
The main effect of Music was not significant, F (2,108) = 1.93; p = 0.15, indicating that music did not influence RT. There was a significant interaction between Conflict, Emotion and Practice, F (2,108) = 3.26; p = 0.04, = 0.06, indicating that the rate of within-session learning was heightened in incongruent trials commencing with an emotionally negative stimulus ( ). To further assess this 3-way interaction, we calculated the difference in learning between congruent and incongruent trials for each emotional condition ( ). The learning related decline in conflict cost was significantly higher in negative condition when it was compared with neutral emotional condition (two-tailed t test with Bonferroni adjustment for multiple comparison p = 0.04). The difference between negative and positive conditions was not significant ( p = 0.08). There were no other significant main effects, nor significant interactions.
The multi-factor ANOVA [Conflict × Music × Practice × Emotion] was also applied to the percentage of correct responses. The main effect of Conflict was significant, F (1,54) = 83.84; p < 0.001, = 0.61, indicating that accuracy was lower in incongruent trials. The main effect of Emotion was also significant, F (2,108) = 10.26; p < 0.001, = 0.16, indicating that the emotional content of the visual stimulus influenced accuracy ( ). Pairwise comparison of the differences in accuracy for each emotional condition (two-tailed t test with Bonferroni adjustment for multiple comparison) revealed that trials which commenced with negative visual stimuli had the highest accuracy in comparison to those which commenced with positive ( p < 0.001) or neutral ( p < 0.001) stimuli. The main effect of Practice, F (1,54) = 9.77; p = 0.003, = 0.15, was also significant. There were no other significant main effects, nor significant interactions. These findings indicate that the presentation of negative emotional stimuli at the beginning of a trial significantly improved the resolution of conflict between behavioral rules.
### Conflict Adaptation Was Influenced by Emotional Content of Visual Stimuli
The behavioral effects of conflict are not just limited to the current trial, and can also be observed in the subsequent trial, manifested as a behavioral improvement if the subject is faced with the same level of conflict again (conflict adaptation) ( ; ; ; ). In the context of the WCST, this can be examined through contrasting incongruent trials that were immediately preceded by another incongruent trial (HH sequence) against those incongruent trials that were immediately preceded by a congruent trial (LH sequence). We examined whether conflict adaptation was influenced by emotion-modulating factors such as music and visual stimuli. A multi-factor ANOVA [Conflict Adaptation (HH/LH sequences, within-subject factor) × Music × Practice × Emotion] was applied to RT in the second trial in both HH and LH sequences. The main effect of Conflict Adaptation was significant, F (1,54) = 80.54; p < 0.001, = 0.60, and manifested as a lower RT in HH sequences than LH sequences ( ). The main effects of Practice, F (1,54) = 13.93; p = 0.01, = 0.21, and Emotion, F (2,108) = 17.27; p < 0.001, = 0.24, were also significant. Importantly, there was a significant interaction between Conflict Adaptation and Emotion, F (2,108) = 3.73; p = 0.03, = 0.07, indicating that the magnitude of conflict adaptation was differentially influenced by the emotional content of the visual stimuli ( ). For ease of comparison presents the normalized RT for both LH and HH sequences, while presents the magnitude of conflict adaptation (LH-HH). Pairwise comparisons (two-tailed t test with Bonferroni adjustment for multiple comparison) of the differences in conflict adaptation for each emotional condition revealed that the magnitude of conflict adaptation was increased in trials which commenced by positive stimuli in comparison to those which started with negative ( p = 0.046) or neutral ( p = 0.003) stimuli. However, this increase was primarily facilitated from the higher RT in LH sequences with positive stimuli ( ). There were no other significant main effects, nor significant interactions.
Conflict adaptation was modulated by the emotional content of visual stimuli. (A) The RT for incongruent trials that are immediately preceded by another incongruent trial (HH sequence), and those incongruent trials that are immediately preceded by a congruent trial (LH sequence) is shown for each emotional category. In both LH and HH sequences, trials which contained a negative start cue had lower RT than those which contained a positive start cue. (B) The magnitude of conflict adaptation (the difference in RT between LH and HH sequences) is shown for each emotional category. The magnitude of conflict adaptation was highest in trials which commenced with positive stimuli. represents p < 0.05, represents p < 0.01.
This ANOVA was also applied to the accuracy in the second trial in both HH and LH sequences. The main effect of Conflict Adaptation was significant, F (1,54) = 493.45; p < 0.001, = 0.90, indicating that accuracy was increased when the high conflict level was presented again (HH sequences). There were no other significant main effects, nor significant interactions.
### Conflict Processing in the WCST Influenced Emotional/Arousal State
We measured event-related EDA within two epochs: a post-feedback epoch [response selection onward (4 s)], to examine changes in arousal level after the participants became aware of their decision outcome, and a pre-feedback epoch [from the start cue to response feedback onset (2.4 s window)], to examine changes in arousal level before feedback.
Participants’ RT was longer in incongruent trials than congruent trials ( ), indicating that conflict between behavioral options adversely affected performance in the WCST (conflict cost). To examine if processing conflict led to a shift in arousal level a multi-factor ANOVA [Conflict × Emotion × Music × Practice] was applied to post-feedback EDA in congruent and incongruent trials. The main effect of Conflict was significant, F (1,54) = 36.14; p < 0.001, = 0.40, and EDA was higher in incongruent trials. This indicates that conflict between behavioral options evoked a higher EDA and presumably a higher arousal level ( ). This suggests that following the behavioral effects of a higher level of conflict ( ), conflict-induced elevation in arousal level occurs as a sustained effect of conflict during the assessment of behavioral outcome ( ).
Conflict induced modulation in both behavior and emotional/arousal state. (A) The RT is shown for correct incongruent and congruent trials. RT was longer in incongruent trials than that congruent trials (conflict cost). (B) Post-feedback EDA is shown for correct incongruent and congruent trials. Post-feedback EDA was significantly higher following incongruent trials.
Although the emotional stimuli influenced behavior, no modulation of pre- or post-feedback arousal level was observed ( p = 0.632 and p = 0.096. respectively). These indicate that the influence of emotional stimuli upon behavior was not necessarily mediated through changes in arousal level, but rather through a direct influence on executive control processes.
### Conflict Influenced Behavior and Arousal in the Following Trial
To examine whether conflict adaptation ( ) was also accompanied by shifts in emotional/arousal state, we applied a multi-factor ANOVA [Conflict Adaptation (HH/LH) × Emotion × Music × Practice] to the pre-feedback EDA in the second trial of HH and LH sequences. Pre-feedback EDA was selected for this analysis as the process of conflict adaptation must be triggered in the preceding trial and continue into the current trial to influence response selection. The ANOVA indicated that the main effect of Conflict Adaptation was highly significant, F (1,52) = 11.70; p = 0.001, = 0.20. This indicates that the process of conflict-induced behavioral adaptation is also reflected in changes in EDA and presumably in emotional/arousal state. In HH sequences, pre-feedback EDA was lower than that of LH sequences, implying that the process of conflict adaptation is accompanied by lower levels of arousal ( ). This suggests that those cognitive processes mediating the conflict-induced recruitment of executive control ( ; ; , ) also influence emotional/arousal state. There were no other significant main effects, nor significant interactions.
Conflict adaptation was accompanied via concurrent shifts in both pre- and post-feedback arousal. (A) The RT is shown for incongruent trials preceded by another incongruent trial (HH) and incongruent trials preceded by a congruent trial (LH). RT was shorter in HH sequences than LH sequences. (B) Pre-feedback EDA in the second trial in both HH and LH sequences are shown. The history of conflict level in the first trial in the sequences modulated pre-feedback EDA in the second, with incongruent trials preceded by another incongruent trial (HH) having lower pre-feedback EDA than those incongruent trials preceded by a congruent trial (LH sequences). (C) Post-feedback EDA in the second trial in both HH and LH sequences are shown. Similarly, to the pre-feedback epoch concomitant shifts in arousal accompanied conflict adaptation, with incongruent trials preceded by another incongruent trial (HH) having lower post-feedback EDA than those incongruent trials preceded by a congruent trial (LH sequences).
Furthermore, to assess if conflict adaptation is also accompanied by concurrent changes in post-feedback arousal level, we applied the multi-factor ANOVA [Conflict Adaptation × Emotion × Music × Practice] to the post-feedback EDA in the second trial in both HH and LH sequences. The main effect of Conflict Adaptation was significant, F (1,53) = 34.03; p < 0.001, = 0.39, indicating that conflict adaptation was also accompanied by post-feedback arousal shifts. The highest level of arousal was observed within LH trial sequences ( ). There were no other significant main effects, nor significant interactions.
These findings indicate that conflict induces alterations in both behavior and arousal level in the following trial. The improvement in behavioral performance seen as a shorter RT ( ) was accompanied by a lower arousal level in the periods before ( ) and after feedback ( ) in the second trial of HH sequences.
## Discussion
Our findings identified intriguing interactions between contextual factors with emotional content and conflict processing in the context of the WCST. We also assessed concomitant alterations in arousal level that reflected such behavioral alterations. We will discuss the significance of these findings in two aspects:
Modulation of conflict resolution and conflict-induced behavioral adjustment by contextual emotional information.
Alterations in arousal level in relation to the interaction between conflict processing and emotional information.
### Conflict Resolution Was Modulated by Contextual Emotional Information
Within the computerized WCST, the start cue was an image with emotional content of either positive, negative or neutral valence. We found that the emotional content of these stimuli induced significant changes in performance. RT was decreased ( ) and accuracy was enhanced ( ) in trials which commenced with negative stimuli. Executive functions and emotional regulation are intrinsically linked in goal-directed behavior ( ; ), with the somatic marker hypothesis proposing that alterations in emotional and arousal state may influence decision making ( , ). In line with this hypothesis, our findings indicate that a brief exposure to emotional stimuli at the start of a trial, induced changes in emotional state and consequently influenced conflict resolution. Negative emotional stimuli distinctly enhanced performance, complimentary to the proposal that negative stimuli may be more salient than equivalent positive stimuli ( ; ), due to evolutionary mechanisms which prioritize threat responses within the environment ( ). In the context of an emotional Stroop task ( ), performance was impaired in trials which contained negative stimuli. Albeit in the opposing direction, this result is compatible with our findings. Within their task, emotional stimuli were presented at response selection, whereas within our task the emotional stimuli were presented at the start of the trial. Therefore, in their study the heightened saliency of negative stimuli might have engaged attentional resources at the point of response selection and consequently disrupted performance. Whereas, within our study the heightened saliency of the negative stimuli presented at the start of the trial may heighten attention and executive control processes, consequently improving behavioral performance, as observed ( ). These findings indicate that the higher saliency of negative stimuli may modulate behavior across various tasks via augmenting attention, having the capacity to either heighten or impair performance dependent on when the stimuli is presented.
In our study, emotional content of visual stimuli modulated behavior, but did not alter EDA in pre- or post-feedback periods. Therefore, the effects of emotional stimuli were not necessarily mediated through alterations in trial-by-trial arousal level. Instead, emotionally salient stimuli might divert attention from extra-task events or inner mental world (attenuating mind wandering) toward the on-going task and therefore enhance upcoming task performance.
Conflict between available behavioral options impairs performance in numerous neuropsychological tests ( ; ; ; ; , , ; ). As demonstrated in this study, and past studies ( , ), RT was longer in incongruent trials than congruent trials ( ). The behavioral effects of the emotional content of the start cue was dependent on the conflict level ( ) and appeared as a larger modulation in incongruent trials. This suggests that the influence of emotional state on performance is exaggerated when there is more demand for executive control of behavior.
We found significant practice-related learning within each testing session, which appeared as an improved ability in resolving conflict. Interestingly, this practice-related learning was also modulated by emotional state and was dependent on the conflict level ( ). Negative stimuli enhanced practice-related learning in incongruent trials, but had an opposite effect in congruent trials ( ). This suggests that within-session learning depends on an interaction between various contextual factors that influence executive and emotional regulation in the context of goal-directed behavior.
Our results do not fit within the emotional priming model which stems from the postulation that conflict is innately aversive ( ; ), and sensitive to emotional state ( ). This idea gained support from a study where conflict-induced behavioral adjustments were attenuated when paired with a reward, as the reward diminished the aversive assessment of the conflict ( ). Furthermore, this model was elaborated upon ( ) suggesting that heightening the aversive assessment of conflict, through the pairing of conflicting behavioral choices and negative stimuli, may increase the magnitude of conflict-induced behavioral adaptations ( ). The emotional priming model predicts that negative stimuli would heighten the aversive nature of the conflict, increasing conflict-induced behavioral adjustment manifested as an increased magnitude of conflict cost.
However, we found that negative stimuli instead attenuated the magnitude of conflict cost ( ), and therefore our results are in contrast to this model. Therefore, an alternative explanation could be that the influence of emotional content on conflict is mainly facilitated through the modulation of attention. As negative stimuli may be more salient than equivalent positive stimuli ( ; ), the heightened level of attention induced by negative stimuli might enhance conflict resolution, and consequently attenuate conflict cost.
We found heightened practice-related learning in incongruent trials that commenced with negative emotional stimulus. It can be suggested that the heightened “attention capture” by negative stimuli promoted an increased rate of learning within incongruent trials ( ).
### Emotional Stimuli Influenced Executive Functions Without Changing Arousal Level
Although emotional visual stimuli influenced behavior, no modulation of pre- or post-feedback arousal was observed, indicating that the influence of emotional stimuli upon behavior was not necessarily mediated through changes in trial-by-trial arousal level. In our study, categorization of visual stimuli based on emotional content was done by independent raters and not by participants. They were also presented in a trial unique design. These were necessary to avoid the effects of repeated exposure to visual stimuli which could have evoked alterations in cognitive processes and arousal level due to familiarity and memory processing. We did not categorize visual stimuli based on the evoked arousal level because it could have introduced a bias in the outcome of our study and prevented assessing whether emotional stimuli induce their cognitive effects through changes in arousal level. We believe this was an important improvement (difference) in our task design, which allowed us to show “emotional stimuli affect executive functions without changing arousal level.” We may also assume that other features in visual stimuli such as complexity or low-level visual features, rather than the emotional content, could have led to behavioral modulations. We used a large set of stimuli, which were matched for their overall size and resolution and then categorized them by two independent assessors. Each participant was exposed to a large pool of stimuli in each category in a trial unique design in three consecutive weeks. Stimuli in each category included images with very different features. For example, the negative category included picture of a wound or image of a crying baby and therefore various colors, objects and complexity levels were included in each category. Therefore, it is highly unlikely that the behavioral effects were due to a particular feature or low-level visual information. The behavioral modulation was seen by negative stimuli and not by neutral or positive stimuli and therefore, attributing the differences to a difference in low-level visual information between the emotional categories and neutral images does not fit the findings.
Executive control and emotional regulation are intrinsically interrelated ( ), with imaging studies revealing a large overlap in brain areas which support executive control functions and those which regulate emotional state such as the amygdala, insula and prefrontal cortex ( ; ; ). As emotional stimuli modulated behavior without concurrent alterations in arousal level it can be proposed that this modulation occurred primarily through a direct influence on these executive control/emotional areas, rather than an alternative indirect influence facilitated via shifts in arousal level. It is important to acknowledge that shifts in arousal level accompanying conflict-induced behavioral adjustments were detected by our recording system. Therefore, the absence of a change in arousal level by the emotional stimuli was not due to a lack of sensitivity in the measurements. This framework diverges from a bulk of past research which has proposed that the influence of emotional content on behavior was primarily facilitated via changes in arousal level. Our findings instead indicate that emotional regulation is directly associated with cognitive flexibility and control of goal-directed behavior, and may be independent of changes in arousal level.
In contrast to the observed effects of emotional visual stimuli, which were dependent on valence, we did not find any significant modulation of performance depending on the music tempo. In Control Study 1 we confirmed that participants could perceive differences between music conditions dependent on tempo, and therefore the absence of music effect was not because of an inability to distinguish differences in music characteristics. These findings suggest that in the context of the WCST emotional content of visual stimuli is more effective in modulating behavior than alterations in music tempo.
### Conflict-Induced Behavioral Adjustments Were Accompanied by Shifts in Arousal Level
We found that conflict-induced behavioral adjustment in the current trial (conflict cost) and in the subsequent trial (conflict adaptation), were accompanied by parallel changes in arousal level. In correct trials, post-feedback EDA was higher in incongruent than congruent trials ( ), indicating that the higher level of conflict present within incongruent trials evoked parallel shifts in both arousal and behavior ( ). This increase in arousal level in response to conflict was previously reported in other autonomic measures of arousal, such as pupil dilation ( ; ; ; ). Furthermore, we found that arousal was also influenced by conflict adaptation and appeared as a difference in EDA between the second trial of HH and LH sequences in both pre- and post-feedback epochs ( ). Pre-feedback arousal was lower in HH sequences than that of LH sequences. This shift in arousal level was dependent upon the conflict level in the preceding trial and presumably occurred concurrently as conflict-induced recruitment of executive control facilitated the conflict resolution in the second trial of HH sequences. Thus, it can be proposed that this arousal shift may be an autonomic manifestation of conflict-induced executive control adjustment. Moreover, within the post-feedback period arousal was still lower in HH sequences than that of LH sequences. These results are complimentary to a study which revealed concomitant shifts in pupil size, as an index of arousal, and conflict adaptation processes in the context of the Simon Task ( ), where the degree of pupil dilation positively correlated with the magnitude of conflict adaptation. An alternative hypothesis is that in the second trial of HH sequences, additional executive control processes are recruited and the system is in a better condition to resolve the conflict. This might decrease conflict-related uncertainty and therefore attenuate arousal levels in HH conditions.
## Limitations
In this study, participants were young adults within a limited age range (20–27 years old) and were all university students and therefore, it was a uniform cohort highly suitable for decreasing between-subject variabilities. However, generalizing our findings to other age ranges or the general population needs to be cautiously done. Future studies will be necessary to examine the interaction of emotional stimuli and conflict processing in children and elderly to verify whether the observed effects are age related. In addition, we used contemporary songs separated based on their tempo as background music. Future studies need to examine a broader range of music. The first set of visual stimuli used in this study were not rated by participants in terms of arousal or valence. However, in Control study 2, we used the second set of visual stimuli (NAPS images), which were rated based on arousal and valence by 204 people. Our findings with the first set were replicated and confirmed in Control study 2. Future studies may assess the correlation between subjective assessment of stimulus valence and arousal and the behavioral effects of emotional stimuli. In our study, the emotional stimuli were irrelevant to the task performance and therefore participants did not need to pay attention to them. It will be crucial to examine the interaction of emotional regulation and conflict processing when resolution of conflict depends on the processing of emotional information.
## Conclusion
We examined the effect of contextual factors with emotional content on conflict processing and arousal in the WCST. Our findings identified that brief exposure to emotional stimuli can modulate behavior, with behavior enhanced following negative stimuli. However, arousal was not modulated by emotional content. We have replicated and validated these findings in a control study ( Control Study 2) with a completely different set of images in another participant cohort. This additional study confirmed that the effects of emotional stimuli do not depend on a particular set or type of images and instead arises from emotional valence of visual images. We propose that the emotional modulation of behavior occurs primarily through a direct influence on highly interrelated brain areas which support conflict resolution and emotional regulation, rather than indirectly via changes in arousal level. Such an interaction might also be mediated through alterations in allocation of attention without altering arousal level. This novel framework diverges from a plethora of past research which postulated that the influence of emotion on behavior was primarily facilitated via changes in arousal level. Future research should more so consider the hypothesis that emotional regulation is directly associated with cognitive flexibility and control of goal-directed behavior, and may be independent of arousal level.
## Data Availability
The datasets for this manuscript are not publicly available because, other factors within the datasets are still being examined. However, datasets are available to editors and reviewers upon request. Requests to access the datasets should be directed to FM, [email protected].
## Ethics Statement
The approval was obtained from Monash University Human Research Ethics Committee. Written consent was obtained from all participants.
## Author Contributions
FM and DF designed the experiment, performed the analyses and wrote the manuscript. DF and RS collected the data. MR contributed to the manuscript preparation. All authors read and approved the final manuscript.
## Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
|
Subsets and Splits