Portal de Periódicos da CAPES

Learning How Neurons Fail Inside of Networks: Nonhuman Primates Provide Critical Data for Psychiatry

2019; Cell Press; Volume: 102; Issue: 1 Linguagem: Inglês

10.1016/j.neuron.2019.02.030

1097-4199

Sarah R. Heilbronner, Matthew V. Chafee,

Neurological disorders and treatments

Advancing psychiatry requires understanding brain malfunction at a microscopic scale, where neurons and synapses operate under constraints imposed by behavior, cognition, and neural architecture. Nonhuman primates are unmatched in approximating the structural and computational environment of the human brain. Advancing psychiatry requires understanding brain malfunction at a microscopic scale, where neurons and synapses operate under constraints imposed by behavior, cognition, and neural architecture. Nonhuman primates are unmatched in approximating the structural and computational environment of the human brain. Mental illness poses a significant global health problem, responsible for 32% of years lived with disability (YLD) caused by any disease worldwide (Vigo et al., 2016Vigo D. Thornicroft G. Atun R. Estimating the true global burden of mental illness.Lancet Psychiatry. 2016; 3: 171-178Abstract Full Text Full Text PDF PubMed Scopus (1087) Google Scholar). Current treatments include psychotherapy, cognitive/behavioral therapy, and pharmacotherapy in the form of antidepressant and antipsychotic medications. Treatments on the horizon include experimental neuromodulatory interventions such as deep brain stimulation (DBS) and transcranial magnetic stimulation. Current treatments do profoundly improve the lives of people living with mental illness but fall short of cure for most, leaving many symptoms unabated. There is no question that we need better treatments. To find better treatments, we first must acquire a more mechanistically detailed understanding of how the diseases that produce mental illnesses work as biological processes. Decades of neuroimaging in patients with mental illness have documented distortions of large-scale brain activity and synchrony patterns in patients. This has provided one of the first crucial inroads into the biology of the disorders. But functional imaging in patients cannot resolve events at cellular and synaptic levels in the brain. The cell and synaptic levels are where genetic mutations interact with patterns of action potentials in specific neural networks during behavior to modify the computational properties of the brain. That interaction, unfolding within neurons and modifying their functions and connections with other neurons, is likely to prove the arena where the crucial events leading to mental illness are first set in motion. According to this hypothesis, mental illness results from a perturbation in the cellular mechanisms that link synaptic connectivity to neural activity patterns in the brain. Once this relationship is knocked out of balance, the brain is synaptically miswired by its own ongoing activity. Over time, this degrades the information processing capacity of brain networks, further distorting patterns of activity that are generated during behavior. Those aberrant electrical activity patterns are then ultimately experienced by patients as altered perceptual, affective, and cognitive states. If the above is true, then the neuron—the cell—is the nexus, the point where risk factors meet and the key causal events leading to disease play out. That presents a considerable problem. We can't study neurons in a direct way in mental illness because the function of neurons and synapses, at the cellular scale, is extremely difficult to measure in humans. Single-neuron recording is used in human patient populations (most commonly with intractable epilepsy) but it is not typically clinically warranted in patients with mental illness. Inducible pluripotent stem cell (IPSC) technology makes it possible to grow neurons from skin samples of patients in tissue culture, but the neural networks that grow do not recapitulate the precise connectional anatomy of in vivo brain networks or the fine pattern of their electrical activation during behavior. Postmortem analysis of brains of patients can resolve changes in fine neural architecture, such as a reduction in dendritic spines, but these studies cannot relate structural changes to causal changes in neural function while the brain was alive and processing information. This means, in essence, that the causal events that lead to mental illness in humans transpire inside an impenetrable black box—the key biological events taking place in cells and synapses within the human brain are, for all intents and purposes, invisible to us. From the perspective of the neuron, the environment can be equated to the spatiotemporal pattern of synaptic inputs to the neuron. This is the information coming into the cell about sensory inputs, behavioral responses, and internal states such as drives or emotions. That incoming information interacts with a complex set of biochemical pathways within the neuron to adjust the strength of its synaptic input from other neurons according either to Hebb's rule ("cells that fire together, wire together") or to reinforcement signaled by neuromodulators such as dopamine, which modulates synaptic strength to increase the future likelihood of reward-producing actions. Such learning mechanisms sculpt the strength and pattern of synaptic connections between neurons, changing the computational properties of neural networks and modifying how they respond to environmental stimuli. After a period of maladaptive activity-dependent synaptic plasticity during disease pathogenesis, incoming sensory information combines with internal state information to produce distorted patterns of neural activation that manifest as the symptoms of mental illness. Thus, we need to see events at the cell and synaptic levels to understand what is going wrong and figure out how to rationally intervene. Rodent and nonhuman primate (NHP) models each offer unique advantages in this effort and should be vigorously pursued. Rodent models bring exquisite capability for circuit, synaptic, and genetic manipulation that cannot yet be fully achieved in NHPs. What the rodent models lack is faithful replication of the electrical environment of neurons likely to occur within the human brain, at least to the degree provided by NHP models. The electrical environment of neurons defined as the spatiotemporal pattern of synaptic inputs is dictated by the neural architecture of the neural systems controlling the synaptic inputs and the computational functions of those neural systems, meaning the types of information (neuronal activity patterns) they generate during behavior as the organism interacts with stimuli in its environment (Figure 1A). The anatomy and computational capability of neural circuits in the NHP cerebral cortex provide the best approximation we have of the human brain. This is the strength of NHP models, that they are likely to capture the interplay between neural architecture, neural activity, and cognitive function necessary to faithfully replicate the electrical environment of human neurons. Understanding the electrical environment of human neurons while the brain is operating to process information takes a crucial step toward being able to trace network and cognitive failure at a global scale back to neuronal and synaptic failure at the cellular scale. The cellular scale is where genetic mutations first exert their maladaptive effects increasing risk of mental illness. The cellular scale is also where treatments have to work to powerfully and effectively modify the course of disease to improve the lives of patients. The majority of the work described here has been in macaques. Because of long generation times and expensive upkeep, macaques are not currently a useful genetic model. By contrast, marmosets (callitrichids) breed frequently and mature more quickly than macaques. Thus, there have been recent advancements in transgenic marmosets, although none in the realm of psychiatric disease. Marmosets do not share all of the cortical architecture of humans and macaques: for starters, their cortices are lissencephalic, meaning they lack the folds seen in macaques and humans. Nevertheless, they clearly possess connectional architecture akin to those in macaques and humans. Therefore, genetic manipulation in marmosets could provide the best of both worlds: controlled genetic mutation to replicate causal risk coupled with brain architecture similar in some respects to humans. However, marmosets have limited cognitive and behavioral capabilities relative to more widely used macaque monkeys, and therefore, it is not yet clear how faithfully marmosets will be able to capture the cognitive deficits and underlying change in neural dynamics that are hallmarks of cortical network failure in human mental illness. Therefore, although marmoset studies hold considerable promise, we believe that, for the foreseeable future, macaque studies will provide an important translational step toward understanding the physical and computational complexity of human brain networks at the cellular scale. Dysfunction of prefrontal-basal ganglia networks is a hallmark of psychiatric disorders. In these networks, distinct prefrontal cortical (PFC) areas project to the striatum, the input structure of the basal ganglia. At this point, their terminal fields converge with other connections, including those from the hippocampus and amygdala, allowing for integration of functionally distinct information. Via projections through the pallidum and thalamus, the integrated information loops back to the cortex. The details of the dysfunction vary according to symptomology and disorder, but mental illness virtually always involves the PFC-basal ganglia network. Thus, this network is the front line for psychiatric research. Some of the regions of this network (particularly the hippocampus, amygdala, and basal ganglia) are highly conserved across rodents, NHPs, and humans. However, this is decidedly not the case for the PFC, which is thought to have undergone more rapid evolutionary change. The PFC is composed of functionally and anatomically distinct subregions in rodents, NHPs, and humans alike. The majority of these regions are highly conserved from NHPs to humans. Multiple sources of evidence support this claim. First, cytoarchitectural studies indicate that NHP and human PFC both contain granular (a distinct layer IV can be observed), dysgranular (a rudimentary layer IV is present), and agranular (lacking a layer IV) subregions in similar topological positions. Specifically, the dorsal and lateral subregions of the PFC in both species are granular, as are the frontal pole and the rostral orbitofrontal cortex. The middle orbitofrontal cortex and portions of the perigenual anterior cingulate cortex are dysgranular. Finally, the caudal orbitofrontal cortex and subgenual and dorsal anterior cingulate cortices are agranular. Even beyond the prominence of layer IV, the distinct PFC subregions in NHPs and humans share most cytoarchitectonic features. Second, noninvasive neuroimaging methods that can be applied to both species have revealed similar connectivity fingerprints. Although the gold standard for studying anatomical connectivity, tract-tracing, can only be performed in nonhuman animals, indirect methods of assessing connectivity (functional connectivity and diffusion tractography-based connectivity) can be applied similarly across species. Such studies have largely found 1:1 correspondence between connectivity patterns in macaque and human PFC subregions, although some subregions have more similar fingerprints than others (Croxson et al., 2005Croxson P.L. Johansen-Berg H. Behrens T.E. Robson M.D. Pinsk M.A. Gross C.G. Richter W. Richter M.C. Kastner S. Rushworth M.F. Quantitative investigation of connections of the prefrontal cortex in the human and macaque using probabilistic diffusion tractography.J. Neurosci. 2005; 25: 8854-8866Crossref PubMed Scopus (341) Google Scholar). Third, functional studies, although more difficult to compare across species, point to PFC homologies between macaque and human. For example, lesions to the dorsolateral PFC (dlPFC) of both species result in working memory deficits, whereas lesions to the orbitofrontal cortex lead to motivation and decision-making deficits. By contrast, the subregion homology between the PFC of the rodent and the human is hotly debated. Unlike the human and macaque PFC, the rodent frontal lobe is entirely agranular. Nevertheless, there may be fundamental connectional similarities between the rodent and primate orbitofrontal and anterior cingulate cortices (Heilbronner et al., 2016Heilbronner S.R. Rodriguez-Romaguera J. Quirk G.J. Groenewegen H.J. Haber S.N. Circuit-based corticostriatal homologies between rat and primate.Biol. Psychiatry. 2016; 80: 509-521Abstract Full Text Full Text PDF PubMed Scopus (194) Google Scholar; Figure 1B). Connectivity with the striatum reveals similar medial-lateral and dorsal-ventral gradients in both species. However, the rostral dorsal anterior cingulate cortex of the NHP does not have a good match in the rodent frontal lobe. Similarly, many studies suggest on cytoarchitectonic and topological grounds that the dorsolateral PFC (dlPFC) of the primate may not be present in the rodent. This highly granular region is capable of representing abstract concepts in the absence of sensory information and may be an important primate adaptation. Although the problem is far from resolved, it does seem that the dlPFC of the human, in particular, is unlikely to be well represented in the rodent. But is the dlPFC important for understanding and treating psychiatric disorders? The dlPFC seems to be preferentially involved in those disorders with prominent deficits in executive cognition, such as schizophrenia, major depressive disorder, attention-deficit hyperactivity disorder, and addiction. One open question is the extent to which the dlPFC is needed for successful treatment. In depressed patients, dlPFC activity normalization is associated with successful treatment regardless of the treatment modality. Moreover, stimulation of the dlPFC is a viable (though not frontline) treatment for an ever-expanding list of psychiatric disorders. This leaves open the possibility that the dlPFC is uniquely important for recovery from mental illness through compensatory mechanisms. Importantly, the major white matter bundles of the human brain are also found in the macaque brain. This is not the case in the rodent brain. For example, the anterior limb of the internal capsule forms a distinct bundle bisecting the caudate and putamen in nonhuman and human primate brains. This bundle contains fibers traveling between the PFC and the thalamus, subthalamic nucleus, and brainstem. Rodents, of course, have such axons; however, they travel in very small fascicles distributed throughout the striatum and thus do not form a discrete bundle. In humans, the anterior limb of the internal capsule has been used as a DBS target for major depressive disorder and obsessive-compulsive disorder. Thus, although NHP white matter organization has been used to identify the specific bundles targeted by internal capsule DBS, the same investigation cannot be undertaken in rodents. The same is true for noninvasive diffusion-weighted MRI studies of white matter abnormalities in psychiatric disorders: the specific connections running through the white matter in question can only be understood via anatomical tract-tracing, a technique not applicable to humans. Psychiatry research is undergoing a conceptual revolution. Increasing evidence that mental disorders are not unitary entities has led to a push to understand and treat impaired biological "domains" of functioning (such as the National Institute of Mental Health's Research Domain Criteria [RDoC]). Furthermore, no single gene disruption has explained the occurrence of any given mental illness (although many have been identified as risk factors). We argue that this shift makes NHP research particularly important, as the cognitive/behavioral abilities and biases found in NHPs can closely mirror those in humans. Thus, because single-unit recordings are difficult and hit-or-miss in humans and behavior is somewhat impoverished in rodents, NHPs offer a unique opportunity to link single-unit neurophysiological activity patterns to complex behavior. Although not specifically modeling a unique disorder, many studies have uncovered the activity patterns that subserve complex cognitive/behavioral domains. For example, risky decision making (choice involving variance in possible outcomes; RDoC subconstruct under the Domain of Positive Valence Systems) is impaired in most mental illnesses. Patients with depression, anxiety, and schizophrenia are typically more risk averse than healthy controls, whereas drug abusers tend to be more risk seeking. Importantly, risky decision making is characterized by similar features in both humans and NHPs: a gap in preference for risk depending on whether the options are described versus experienced, win-stay/lose-shift choice patterns, overweighting of rare events/underweighting of common events, a preference for options in which probabilities are known over those in which probabilities are unknown, preference changes with iterated choice and changes in inter-trial intervals, and so on. This points to shared cognitive and neural processing mechanisms in risky decision making across NHPs and humans. Shared behavioral complexity can also be observed in the construct of cognitive control (RDoC Domain of Cognitive Systems). The AX-Continuous Performance Task (AX-CPT) is used to measure context-processing deficits: the first stimulus in each trial provides the contextual information that subjects must remember in order to appropriately respond to the second stimulus in the trial. Both monkeys given NMDA receptor antagonists and patients with schizophrenia exhibit the same trial-type selective pattern of behavioral errors while performing the same cognitive control task (MacDonald et al., 2005MacDonald 3rd, A.W. Carter C.S. Kerns J.G. Ursu S. Barch D.M. Holmes A.J. Stenger V.A. Cohen J.D. Specificity of prefrontal dysfunction and context processing deficits to schizophrenia in never-medicated patients with first-episode psychosis.Am. J. Psychiatry. 2005; 162: 475-484Crossref PubMed Scopus (278) Google Scholar). Because patients with schizophrenia show a selective deficit on a specific trial type (rather than a global or nonselective impairment), the neural activity patterns supporting behavior on that particular set of trials may have implications for disease. Another feature of the primate world particularly relevant to modeling psychiatric disease is complex social behavior (RDoC Domain: Systems for Social Processes). Although different species of NHPs live in social communities with dramatically varying social structures, macaques, in particular, live in large, hierarchical groups of conspecifics with both males and females. Their social cognitive abilities are impressive and are often quite similar to those found in humans. For example, like humans, macaques follow a conspecific's gaze, can understand third-party interactions, are sensitive to others' rewards and punishments, and may even be able to understand another's intent as separate from one's own. Importantly, we now appreciate the extent to which mental illnesses are shaped by and manifested in social interactions. In a landmark study on drug abuse, Morgan et al., 2002Morgan D. Grant K.A. Gage H.D. Mach R.H. Kaplan J.R. Prioleau O. Nader S.H. Buchheimer N. Ehrenkaufer R.L. Nader M.A. Social dominance in monkeys: dopamine D2 receptors and cocaine self-administration.Nat. Neurosci. 2002; 5: 169-174Crossref PubMed Scopus (517) Google Scholar showed that social housing increased the availability of dopamine D2 receptors in dominant, but not subordinate, NHPs. Moreover, cocaine was not reinforcing for these dominant monkeys, suggesting a protective role of social behavior mediated by D2 receptors. The intersection between human-like neuroanatomy and human-like cognitive abilities in NHPs suggests that neurons in the NHP and human brains experience similar spatial and temporal patterns of synaptic input (the "environment" from the perspective of the cell). This suggests that it may be possible to reproduce patterns of electrical activity that occur in humans at the cell level in nonhuman primates. Then, by manipulating cortical networks in NHP, it may be possible to reveal how these networks fail, which is the first step toward understanding how to fix them. To do that, it is important to study the spatial and temporal structure of activity patterns in the NHP brain during the performance of the same behavioral tasks that measure deficits in patients. Figures 1C and 1D illustrate an example of the approach. Monkeys administered a drug that blocks NMDA receptors (implicated by genetic linkage as causal in schizophrenia) and patients with schizophrenia exhibit the same highly specific, trial-type selective error pattern while performing the same cognitive task in comparison to control subjects or saline injections. In this task, the response required to a probe stimulus is contingent on a preceding cue. On some trials, the cue countermands a habitual response to the probe (Figure 1C, "BX" trials). On these trials, patients with schizophrenia often fail to engage cognitive control and erroneously produce the reflexive response to the probe, increasing errors. Importantly, monkeys given a drug that blocks NDMA receptors in the brain exhibit the same highly selective error pattern (Figure 1C; Blackman et al., 2013Blackman R.K. Macdonald 3rd, A.W. Chafee M.V. Effects of ketamine on context-processing performance in monkeys: a new animal model of cognitive deficits in schizophrenia.Neuropsychopharmacology. 2013; 38: 2090-2100Crossref PubMed Scopus (32) Google Scholar, compare to human data in MacDonald et al., 2005MacDonald 3rd, A.W. Carter C.S. Kerns J.G. Ursu S. Barch D.M. Holmes A.J. Stenger V.A. Cohen J.D. Specificity of prefrontal dysfunction and context processing deficits to schizophrenia in never-medicated patients with first-episode psychosis.Am. J. Psychiatry. 2005; 162: 475-484Crossref PubMed Scopus (278) Google Scholar), providing an uncanny match in the cognitive errors of patients with a mental illness and an animal subjected to a disease-relevant manipulation. Having established this bridge at the level of behavior, we can record neural activity during the behavioral impairment to ask what is going wrong in the underlying neural circuits. Under normal conditions, interspersed among their irregular firing patterns, pairs of prefrontal neurons were seen to fire action potentials ("spikes") simultaneously (0-lag spiking), and interestingly, the frequency of 0-lag spiking increased around the time that monkeys made their response and (potentially) received reward (Figure 1D, left). This raised the possibility that synchronous spiking in PFC neurons related to processing trial outcomes and could provide a teaching signal that potentially "trained" networks during learning. We found that blocking NMDA receptors strongly reduced the frequency of 0-lag synchronous spiking in PFC neurons around the time of the response (Zick et al., 2018Zick J.L. Blackman R.K. Crowe D.A. Amirikian B. DeNicola A.L. Netoff T.I. Chafee M.V. Blocking NMDAR disrupts spike timing and decouples monkey prefrontal circuits: implications for activity-dependent disconnection in schizophrenia.Neuron. 2018; 98: 1243-1255.e5Abstract Full Text Full Text PDF PubMed Scopus (24) Google Scholar). That meant that cells were no longer "firing together" as often. This could be significant from the perspective of Hebbian, spike-timing-dependent synaptic plasticity, because cells that no longer fired together would be predicted to unwire, or synaptically disconnect. We found indirect evidence of this. The number of neurons in PFC found to exhibit statistical dependencies in their spike trains at timescales consistent with synaptic interaction (through one or two synapses) was reduced in the presence of drugs that block NMDA receptors (Figure 1D, right; "Drug"), and this reduction in functional coupling between prefrontal neurons persisted even after the drug itself washed out of the brain (Figure 1D, right; "Saline"), suggesting chronic disconnection (Zick et al., 2018Zick J.L. Blackman R.K. Crowe D.A. Amirikian B. DeNicola A.L. Netoff T.I. Chafee M.V. Blocking NMDAR disrupts spike timing and decouples monkey prefrontal circuits: implications for activity-dependent disconnection in schizophrenia.Neuron. 2018; 98: 1243-1255.e5Abstract Full Text Full Text PDF PubMed Scopus (24) Google Scholar). This led us to an activity-dependent disconnection theory of schizophrenia. Simply, in schizophrenia, cells in PFC circuits stop firing together as often due to a variety of mechanistically convergent insults (NMDA receptor blockade being one of them) and synaptically disconnect over time, further reducing the probability of synchronous discharge. This could account for the apparently stochastic environmental dependency of the disease. Only certain individual brains encountering certain specific environmental conditions would enter the asynchronous activity states that set this cascade into motion. This example shows the kind of theory building, and ultimately theory testing, that can come from cell-level data in primate models and why getting that data may be important in trying to open the black box in mental illness. The above is only one example, and several groups are working on increasing our understanding of the cell and synaptic mechanisms of information processing in NHP prefrontal circuits to reveal how these circuits may fail in diseases such as schizophrenia. For example, Arnsten and colleagues have extended this approach to the microcircuit level. They focally applied minute amounts of drugs onto neurons while they generated electrical signals to process information. When they applied drugs that blocked NMDA receptors, the sustained activity associated with working memory was reduced in a dose-dependent fashion (Wang et al., 2013Wang M. Yang Y. Wang C.J. Gamo N.J. Jin L.E. Mazer J.A. Morrison J.H. Wang X.J. Arnsten A.F. NMDA receptors subserve persistent neuronal firing during working memory in dorsolateral prefrontal cortex.Neuron. 2013; 77: 736-749Abstract Full Text Full Text PDF PubMed Scopus (307) Google Scholar). This localizes the site where specific NMDA synaptic mechanisms work to influence information processing to the prefrontal neuron or local circuit, providing a cell-level description of how changes in synaptic function could give rise to the working memory deficits observed in patients. Everling and colleagues showed that blocking NMDAR in NHPs systemically disrupted cortical synchrony patterns around the time of the motor response that differentiated correct from erroneous responses in a rule-based task (Skoblenick et al., 2016Skoblenick K.J. Womelsdorf T. Everling S. Ketamine alters outcome-related local field potentials in monkey prefrontal cortex.Cereb. Cortex. 2016; 26: 2743-2752Crossref PubMed Scopus (14) Google Scholar), providing an interesting parallel to the cell-level synchrony deficits we have observed at around the same time in the trial (Figure 1D, left). Processing of errors is crucial to adapting behavior to the environment when rules change, and this disruption in oscillatory dynamics could contribute to deficits in executive control observed in patients. Albright and colleagues measured EEG in monkeys performing an auditory "oddball" task and showed that blocking NMDA receptors blunted the brain's neural response to infrequent stimuli (Gil-da-Costa et al., 2013Gil-da-Costa R. Stoner G.R. Fung R. Albright T.D. Nonhuman primate model of schizophrenia using a noninvasive EEG method.Proc. Natl. Acad. Sci. USA. 2013; 110: 15425-15430Crossref PubMed Scopus (118) Google Scholar). Measuring EEG in monkeys provided a neural signal with direct comparability to EEG measurements in patients performing similar tasks that exhibit similarly reduced response to oddball stimuli. These and other studies in NHPs are starting to chart the landscape of disease-relevant manipulations and circuit malfunctions that could combine to produce mental illness in humans. In summary, we need to know how cells malfunction in human mental illness, but we can't readily study cells in the living human brain, necessitating animal models. NHP models optimize the homology to the human brain in terms of cognitive ability and architectural complexity, both strong determinants of neural dynamics in cortical networks, which could drive pathogenesis. However, it is important to qualify what the term "model" can mean applied to NHPs in relation to a disease. NHPs cannot model the human disease. They can model network vulnerabilities as well as cell and synaptic malfunctions that could contribute to cognitive deficits observed in human mental illness. This effectively brings brain malfunction at a microscopic, cellular level into the scope of scientific enquiry. Along those lines, it will be important to continue to study how cortical networks fail as well as how they function in animal models. Ultimately, cell-level data obtained in animal models are likely to give rise to a new class of mechanistic theories (such as activity-dependent disconnection) that may progressively approximate the causal biology. If we can refine these theories, we may discover powerful new ways to restore brain function and quality of life for patients. We would like to thank the University of Minnesota Neuroplasticity Research in Support of Mental Health group (Neuro-PRSMH) for useful discussions and insight. We acknowledge funding from the National Institutes of Health R01MH118257 and MnDrive Brain Conditions (S.R.H.) and National Institutes of Health R01MH107491 (M.V.C.).