The abiding relevance of mouse models of rare mutations to psychiatric neuroscience and therapeutics

Abstract

Studies using powerful family-based designs aided by large scale case-control studies, have been instrumental in cracking the genetic complexity of the disease, identifying rare and highly penetrant risk mutations and providing a handle on experimentally tractable model systems. Mouse models of rare mutations, paired with analysis of homologous cognitive and sensory processing deficits and state-of-the-art neuroscience methods to manipulate and record neuronal activity have started providing unprecedented insights into pathogenic mechanisms and building the foundation of a new biological framework for understanding mental illness. A number of important principles are emerging, namely that degradation of the computational mechanisms underlying the ordered activity and plasticity of both local and long-range neuronal assemblies, the building blocks necessary for stable cognition and perception, might be the inevitable consequence and the common point of convergence of the vastly heterogeneous genetic liability, manifesting as defective internally- or stimulus-driven neuronal activation patterns and triggering the constellation of schizophrenia symptoms. Animal models of rare mutations have the unique potential to help us move from “which” (gene) to “how”, “where” and “when” computational regimes of neural ensembles are affected. Linking these variables should improve our understanding of how symptoms emerge and how diagnostic boundaries are established at a circuit level. Eventually, a better understanding of pathophysiological trajectories at the level of neural circuitry in mice, aided by basic human experimental biology, should guide the development of new therapeutics targeting either altered circuitry itself or the underlying biological pathways.

SCZ is a severe psychiatric disorder characterized by a large etiological heterogeneity (which is primarily, but not exclusively, genetic in nature) and diverse manifestation of core clinical symptoms (Stessman et al., 2014). Available treatments, most of which have been serendipitously found, are rather limited in ameliorating disease symptoms (Goff et al., 2011). Indeed, there is a half century long stagnation in advancing novel psychiatric therapeutics in clinical trials, which points to the urgent need for adapting a more effective approach to identify novel treatments for mental illness focusing on understanding causal mechanisms.

Genetics to the rescue:

In searching for causal mechanisms underlying a disease there is no substitute for genetic analysis, simply because DNA sequence alteration that has been rigorously and unequivocally associated with a disease, such as SCZ, can be inferred to indicate an unambiguous route towards causation. This stands in distinct contrast to other observed biological alterations which could play roles in causation but could also represent epiphenomena (such as remote downstream effects of the disease, adaptations or results of treatment). In that respect, unbiased genetic studies can and have already led to the identification of previously unsuspected mechanisms expanding the space of biological investigation from a small number of hypotheses that once dominated the field to new observations and ideas. In the case of human genetics, the development of powerful genomic tools along with improvements in the quality of the design and computational analysis have made it possible to conduct comprehensive unbiased genetic studies of diverse psychiatric diseases, including SCZ. DNA microarrays made possible genome-wide association studies (GWAS) involving case-control cohorts to identify disease-associated common single-nucleotide polymorphisms (SNPs) and rare copy number variants (CNVs, large deletions, duplications, or more complex structural changes in a segment of the genome), while massively parallel DNA sequencing technology made possible to identify rare risk alleles in protein coding regions of the genome of patients (Rodriguez-Murillo et al., 2012). Initially, modest-scale studies using powerful family-based designs have been instrumental in cracking the genetic complexity of SCZ and other neurodevelopmental and neuropsychiatric disorders (Sebat et al., 2007; Xu et al., 2012; Xu et al., 2011; Xu et al., 2008) but further, more detailed analysis and more accurate risk estimates required formation of large international consortia collecting large case-control samples (tens of thousands of individuals) or family cohorts.

While a rigorous evaluation of the accumulated genetic data is beyond the scope of this article we summarize a number of key findings and observations that are critical for evaluating the feasibility of modeling genetic complexity in mice. GWAS studies in SCZ have identified a large and ever-growing number (over a hundred) of common SNPs that typically have very small contributions to disease risk (Pardinas et al., 2018; Schizophrenia Working Group of the Psychiatric Genomics, 2014). Genome-wide significant hits from GWAS, in combination, explained only a small fraction of the expected genetic component of risk and these estimates have only increased modestly over time, even with much larger sample sizes and many more significant loci. Genomic tools used in GWAS also led to the identification of rare pathogenic CNVs (Malhotra and Sebat, 2012; Marshall et al., 2017), while DNA sequencing led to the identification of rare pathogenic exonic mutations, especially ones that lead to loss of function (LoF) by blocking production of an active protein either by introducing premature stop codons or by affecting splicing in genes that presumably do not tolerate haploinsufficiency, a situation in which both copies of a gene need to be active for healthy functioning (Singh et al., 2016; Takata et al., 2014). It should be noted that while many of the CNVs and single gene mutations initially associated with such cases occur de novo in the affected individual (Fromer et al., 2014; Xu et al., 2011), there is strong evidence that inherited ultra-rare variants play a major role as well (Genovese et al., 2016; Takata et al., 2014; Xu et al., 2009). Overall, unlike common SNPs, unambiguous identification of individual risk genes affected by rare mutations, especially inherited ones, has been more challenging at current sample sizes due to the rarity and incomplete penetrance of the mutations and the large number of target genes.

While recent genetic studies have led to the unequivocal identification of several common or rare (and highly penetrant) risk variants, the synthesis of these observations into cogent paradigm of genetic architecture is a matter of contention. According to one view, existing genetic data is interpreted in a paradigm in which a large fraction of disease risk is primarily due to large numbers of weak common variants with effect sizes well below genome-wide statistical significance at current sample sizes, captured in the form of polygenic risk scores (a weighted sum of allele frequencies, where the weights are effect size estimates from GWAS) (Bigdeli et al., 2016; International Schizophrenia et al., 2009; Visscher et al., 2008). Full-blown SCZ requires loading of many thousands of risk alleles that each pushes brain development or function toward illness ever so slightly. According to this view, the rule for psychiatric phenotypes is fiendish polygenicity, while rare cases, in which SCZ is associated with damaging mutations in single genes or with CNVs represent exceptions rather than the rule. The largely unproven assumption implicit in this dichotomy is that the effect of thousands subtle insults on cellular regulatory pathways may be equivalent to the effect of one strong insult, a postulation that may not be consistent with the well-known robustness and resilience of regulatory pathways that have evolved to efficiently buffer insults imposed by more lenient genetic and environmental variation.

An alternative view states that risk is primarily conferred by one or combinations of few relative severe (and therefore rare) mutations (Rodriguez-Murillo et al., 2012) and that the polygenic component captured by the thousands weak effect size common SNPs simply reflects background effects that modify either the trajectory of severe genetic variation or its penetrance, by weakening the overall regulatory robustness and making the organism more susceptible to its effects. This view is supported by the proven bulk contribution of rare de novo or inherited mutations in SCZ risk (Genovese et al., 2016; Takata et al., 2014). Additional support comes from the not very surprising finding that the thousands of small-effect variants comprising the polygenic component tend to implicate a very large fraction of all genes expressed in brain. While variation in such genes can collectively capture a sizeable proportion of the genetic variance (heritability) of SCZ simply because of their large number, they may not be intimately involved with the core biological disease mechanisms. Such findings were recently formulated in a rigorous manner as the model of “omnigenic” inheritance (Boyle et al., 2017), which postulates that every gene expressed in a disease- relevant tissue will bear some genetic variants that are statistically associated with disease risk. Finally, it has been recently shown that even in cohorts of children with heterogeneous severe neurodevelopmental disorders expected to be almost entirely monogenic, the polygenic background of inherited common variants acts as a modifier of the severity of the effects of rare mutations (Niemi et al., 2018). This finding supports a secondary role of common genetic variation in affecting the overall risk and clinical presentation of disorders originating from rare mutations. The observation that genes harboring SCZ-associated common variants are more likely to interact with genes harboring rare de novo mutations within protein-protein interaction networks may be a reflection of such genetic interactions (Chang et al., 2018; Gilman et al., 2012; Jia et al., 2018).

Despite lingering uncertainties regarding genetic architecture, the information emerging from unbiased, large-scale genetic studies is yielding initial clues to their biological underpinnings that warrant biological follow-up studies designed to understand disease risk mechanisms, develop biomarkers, novel efficacious drug therapies and facilitate preventive interventions. Of course, while genetic studies are not sufficient by themselves to yield detailed mechanistic insights into the nature of the associated conditions, they provide the only means to acquire a handle on experimentally tractable model systems. We and others have argued that such “follow-up studies” require analysis of the intact brain in carefully designed preclinical models of rare mutations, most commonly involving rodents (Arguello and Gogos, 2006; Dawson et al., 2015). Skepticism regarding this approach (addressed in more detail below), was reignited recently by the emergence of alternative approaches utilizing human cellular models generated by reprogramming technologies (Takahashi et al., 2007). A detailed appraisal of the relevant literature is beyond the scope of this review, but we provide an outline of the advantages and limitation of such approaches and a perspective on how they can be useful not as an alternative but a complementary approach.

Human experimental models:

The ability to reprogram human cells such as fibroblasts or lymphocytes, either into pluripotent cells that can then be coaxed to differentiate into any cellular phenotype or directly into neurons, has made it possible to generate and culture in vitro human neurons from patients with either diverse genetic background or carrying specific risk mutations. To enhance the usefulness of traditional two-dimensional cultures and allow more accurate modeling of the structural and functional synaptic deficits underlying neuropsychiatric disorders, additional approaches have been taken towards the production of three-dimensional cultures, in which multiple cell types can be elaborated over long periods in culture and, in principle, permitted to migrate out and form brain organoids (Quadrato et al., 2016). Similar to flat cultures, organoids can be made from any available genetic background as well as from cell lines that have undergone genome engineering to introduce or edit out genes of interest, including genes associated with disease.

One major source of excitement with generating such models was that they offered an experimental system to recapitulate the high-risk polygenic burden of SCZ by generating two or three-dimensional cultures from people with high versus low polygenic risk scores, allowing investigation of the effect of common variants collectively rather than individually, something not possible in animal models. However, such studies have not been reported yet in psychiatry and may be confounded by difficulties with controlling for the different genetic make-up of these cell lines, since each individual carries a unique partially overlapping subset of the thousands of the presumed population risk variants, which may be “tainted” by additional “invisible” rare risk mutations whose effects on the phenotypes may be substantial but impossible to discern. There are additional complexities and limitations even in more ideal setting where, for example, rare large effect risk mutations in isogenic backgrounds are being modeled (Berry et al., 2018; Huch et al., 2017; Sun et al., 2018). For one thing, there is considerable experimental variation due to irreducible, unknown source developmental noise, that likely originates from stochastic events during the initial stages of the organoid assembly. There is also limited evidence so far from microscopy and neurophysiology (Birey et al., 2017; Pasca et al., 2015; Watanabe et al., 2017; Xiang et al., 2017) and a scarcity of systematic studies (Qian et al., 2016) addressing whether human brain organoids that have been grown for several months develop in situ and in a reproducible manner, mature processes (dendrites and axons) as well as mature synapses that accurately reflect structure and function of neural circuits in intact brains. In addition, there are likely several thousand distinct cell types in the human brain but at present it is only possible to generate a small fraction of them for in vitro experimentation. Even if some or all these manifold issues are resolved by optimizing culture conditions and vascularization (Faley et al., 2019), mutant brain organoids will likely recapitulate predominantly early stages of abnormal brain development, while more mature pathophysiological states such as the ones in SCZ brains may prove much harder to model in a dish (Brennand et al., 2015). Indeed, the biggest limitation is the kind of phenotypes that can be assayed in such systems and whether they can provide useful insights into key SCZ disease mechanisms. If, as we argue below, the genetic heterogeneity of SCZ convergences on the activity, connectivity, plasticity and homeostatic patterns of neuronal assemblies in adolescent and adult brain, then relying exclusively on cellular models for understanding the effects of genetic risk on brain circuits, cognition, and behavior will very unlikely suffice.

There have been a number of reports, primarily from two-dimensional cultures, that have compared properties of cultured human neurons derived from relatively few individuals with SCZ and healthy controls with claims of significant phenotypic differences (Casas et al., 2018; Stachowiak et al., 2017). A critical evaluation of these findings is beyond the scope of this review but, given the enormous etiological heterogeneity of the disease and the seemingly irreducible heterogeneity of these cultures, it is precarious to draw conclusions from such small samples. While no major insights into the emergence of psychiatric symptoms have been provided by such studies so far, better powered studies and more optimal culture conditions may be very useful in understanding and modelling proximal molecular and cellular effects of risk mutations (such as effects on gene expression, neuronal proliferation, axonal or dendritic growth) which may in turn facilitate efforts to screen for novel drugs (Readhead et al., 2018), especially ones targeting the effects of specific genetic liabilities. Overall, understanding the full chain of events leading from DNA variation to brain dysfunction and the emergence of clinical symptoms, necessitates analysis of the intact brain in behaving animals. Mouse models will remain, for the foreseeable future, central to the interrogation of genetically heterogeneous psychiatric disease mechanisms although they, almost certainly, will need to be complemented by human experimental biology, especially for translational purposes.

Can we model a predominantly human disease in mice?

Can an animal model reproduce key aspects of human disease mechanisms and be used to predict the efficacy of therapeutics? How does research on animal models of psychiatric disorders fits into the reality of the fast accumulating genetic knowledge? As we argue below, the answer to such questions lies onto the appropriate selection of models, analytical goals and paradigms. Indeed, selection of etiologically valid mouse models of rare mutations, paired with analysis of homologous cognitive and perceptual deficits (that rely on shared neurobiology) and state of the art neuroscience methods has already provided novel insights into the neural substrates of maladaptive disease-related behaviors, rendering such questions obsolete. Nevertheless, as mentioned above, a number of concerns have been raised regarding this approach (Hyman, 2018) related to the following issues.

i. Concerns about feasibility of modeling the large etiological heterogeneity of SCZ in mice.

If indeed polygenic combinations of seemingly vast numbers of alleles with very low penetrance underlie psychiatric disease risk (with a few exceptions of rare mutations), then it would be probably a pointless exercise to model in mice an isolated DNA variation that can produce a disease phenotype only in the context of thousands other human risk alleles and expect to yield a meaningful phenotype. Such a task would be further mitigated by the fact that most of the common disease-associated loci, are located in noncoding regions of the human genome (Zhang and Lupski, 2015), which are poorly conserved as well as by the fact that the majority of common disease-associated loci are not the actual “causal” variation but simply strongly linked via proximity to actual variation which requires challenging steps to be identified. Under this scenario, the possibility of a genetic mouse model of SCZ withers into a hopeless endeavor. Nevertheless, even if such a scenario is true, focusing on “exceptional” rare mutation-linked cases may be a productive line of inquiry because SCZ in cases harboring such rare mutations appears broadly similar to that found in the general population, with respect to core symptoms and treatment response. Therefore, findings arising from efforts focusing on rare genetic lesions could be generalized over a variety of other genetic causes of SCZ. This issue has been addressed systematically for specific rare CNVs such as the 22q11.2 deletion (Bassett et al., 2003; Bassett et al., 1998), but it is likely widely applicable, since cases that were eventually identified as harboring rare CNVs have been traditionally included in cohorts of idiopathic SCZ employed in large scale GWAS studies.

On the other hand, if genetic risk is conferred primarily by one or combinations of more than one rare mutation with relatively large effect size, then animal models where such mutations are introduced via traditional or more recent, advanced gene editing approaches (Birling et al., 2017), offer a straightforward way to model in mice isolated or combinations of risk alleles based on a solid finding from human genetics. As new insights on the specific genetic contribution of various etiological mutations emerge, our ability to generate appropriate animal models with cross-species validity will continue to improve. In that context, it is worth noting that work on mouse models that carry the equivalent of human CNVs initially indicated that disease phenotypes are very likely due to contributions of more than one neighboring genes and often genetic interactions that amplify or mask the effect of individual genes (Karayiorgou et al., 2010; Paterlini et al., 2005; Xu et al., 2013). Such additive or synergistic interaction of multiple genes within a CNV locus suggest that the genetic contribution of CNVs may also involve a degree of “micro-complexity” reflecting the more extensive oligogenic “macro-complexity” inherent in the genetic architecture of SCZ. Subsequent human genetic studies (Sanders et al., 2015) as well as surveys of existing literature (Jensen and Girirajan, 2019) also confirmed that “micro-complexity” is a general principle underlying the disease risk associated with large recurrent CNVs (encompassing many genes, such as 22q11.2 deletions) but not small (encompassing less than 7 genes, Sanders et al., 2015) de novo CNVs. Therefore, mouse models of large rare pathogenic CNVs could also be very instructive for ongoing efforts to decipher the pattern and consequences of the genetic architecture of SCZ.

Concerns about face, construct and predictive validity of rodent models.

The utility of animal models of SCZ has been traditionally rationalized on the basis of three criteria. Face validity refers to whether animal’s phenotype captures important characteristics of the human disease. The criterion of predictive validity refers to whether assays conducted in an animal successfully predicts treatment efficacy in patients. Construct validity is based on whether the nature of genetic, environmental or pharmacological manipulations employed in the construction of the putative model accurately reflect disease etiology in humans (Nestler and Hyman, 2010).

Face validity has been a controversial issue when characterizing mice manipulated by environmental or pharmacological perturbations in the era preceding the recent human genetics advances and has been often used to misleading validation of phenocopies. The interpretation of the behavioral response (i.e locomotor hyperactivity) of normal rodents treated with psychostimulants (such as amphetamine or cocaine, (Featherstone et al., 2007) as a valid representation of positive symptoms of SCZ provides a relevant example. Predictive validity was not very successful in generating new drugs but mostly in improving the efficiency of prototype drugs and was often based on circular logic asserting successful prediction of efficacy of drugs first identified by their effects on human, without demanding that underlying mechanisms are shared between animal models and patients. The tenuous characterization of behavioral deficits of rodent models in assays originally based on existing anti-depressants (such as the forced-swim and tail-suspension tests) as “depressive-like symptoms” offers an often-cited pertinent example (Nestler and Hyman, 2010) and predictably has not yielded novel therapeutic agents. Construct validity, in its limited definition not including reliable etiological validity criteria that would ensure direct homology to the human disorder, also led to many dead ends because it was based on pathophysiological hypotheses derived mostly from epidemiological data (many of which were later disproved) rather than verified mechanisms. Rodent models based on environmental manipulations such as use of viral mimics that activate the immune system in pregnant rodents (to model the effect of maternal infections) or isolation rearing during developmentally critical periods (to model early life and prenatal stressors) offer a relevant example of SCZ models of pathophysiological mechanisms with, at best, weak and nonspecific contributions to disease risk (Selten et al., 2010; Selten and Termorshuizen, 2017; Susser and St Clair, 2013).

None of these considerations represent a major source of concern when referring to animal models based on findings from unbiased genetic studies. Such models have inherently strong construct validity, while concerns about face validity are mitigated by the fact that we are not modeling the constellation of disease symptoms but a proven etiological factor. Concerns about predictive validity are also mitigated by the fact that genetic manipulations perturb pathways that are highly disease relevant and therefore there is a high probability of identifying new treatments rather than fine-tuning existing ones. Our goal is not to construct a model that recapitulates all disease symptoms but to understand and get a handle on mechanisms that we can target for drug development, while at the same time we have readily assayable phenotypes, which depend on these targeted mechanisms and can be used for drug screening purposes.

Concerns about using rodent models in light of genetic background effects:

The phenotypic effects of rare deleterious mutations depend not only on the mutated gene, but also on its interactions with genetic background. This is a consistent and recurrent observation in human patients in which the same penetrant single-gene mutation or CNV yields highly variable phenotypes (Kushima et al., 2018; Stefansson et al., 2014) maybe due to other co-inherited rare variants or the modifying role of the polygenic background (Pizzo et al., 2018; Shohat et al., 2017). Experimental paradigms in wide current use are based on generating a mutant mouse in an inbred genetic background based primarily on the assumption that inbred mice are preferred over outbred mice because it is assumed that they display less trait variability (an assumption recently challenged, (Tuttle et al., 2018) and dictated by observations that the genetic background onto which a mutation is backcrossed can affect the detection of a phenotype (Arguello and Gogos, 2006). Because of this, the genetic background is often chosen based on the postulated effect of a given mutation on elevating or reducing the behavioral phenotype under scrutiny (Crawley et al., 1997; Dulawa and Geyer, 2000). Two recent systematic studies highlight a number of issues related to such experimental schemes. Sittig et al. engineered mice to carry a severe mutation in one of two genes (Cacna1c, an L-type voltage gated calcium channel, and Tcf7l2, a transcription factor related to Wnt signaling pathway) in which milder genetic variants have been found to be associated with psychiatric disorders (Sittig et al., 2016). When they bred the transgene into different inbred mouse lines (i.e. into different genetic backgrounds), they observed marked differences in the phenotypes, including the occasional disappearance of a trait or a change in its directionality. The take-home message of such studies is that the effect of a given mutation should be analyzed in more than one strains, which is often not practical in the context of academic labs. Neuner et al took an alternative approach hypothesizing that, in light of genetic background effects, inclusion of genetic diversity could improve translational validity of genetic mouse models of disorders associated with cognitive decline, such as Alzheimer’s disease (AD) and aid in the identification of specific genes involved in conferring disease resilience (Neuner et al., 2018). To this end, they combined a well-established complex genetic mouse model of AD with a genetically diverse reference panel (BXD genetic reference panel, a large and well-characterized series of recombinant inbred strains derived from the two common inbred strains B6 and DBA/2J) to generate isogenic mouse lines harboring identical high-risk mutations but differing across the remainder of their genome. They showed that genetic variation profoundly modified the phenotypic impact of human AD mutations and resulted in greater genetic, transcriptomic, and phenotypic overlap with human disease. This work suggested that mouse models incorporating genetic diversity may better translate to human disease.

Overall, although the effects of the genetic background against which to study introduced SCZ risk mutations may limit the generalizability of the results, they should be embraced, explored and exploited for at least two reasons: First, in translational venues they can lead to clues for therapies. For example, more than half of the between-strain variation in gene expression and phenotypic diversity has been attributed to structural variants equivalent to human CNVs and additional fraction of this variation may be due to more subtle variation, including LoF mutations (Egan et al., 2007; Quinlan et al., 2010; Watkins-Chow and Pavan, 2008; Yalcin et al., 2011). Such mutations often have protective effects, in the sense that they can prevent emergence of specific phenotypes due to a modeled disease mutation (Lacaria et al., 2012). LoF mutations with protective effect could, in turn, anchor genetics-driven drug discovery efforts designed to inhibit the activity or expression of previously unknown molecular targets. This approach has been proposed as an efficient way to identify new drug targets in human genomic datasets (Harper et al., 2015) and there is no reason why it could not also work with mutations modeled in mice. The fact that a LoF in the genome could have dramatic protective effects in trans was recently demonstrated for de novo microdeletions of the 22q11.2 locus, one of the most common chromosomal abnormalities, are responsible for up to 1–2% of all sporadic SCZ cases (McDonald-McGinn et al., 2001; Xu et al., 2008). A simple engineered heterozygous LoF mutation in Emc10/Mirta22, a gene upregulated in the brains of the Df(16)A^+/− mice that model the 22q11.2 SCZ susceptibility locus, leads to abolishment of a series of cognitive and synaptic deficits previously shown in these mice (Diamantopoulou et al., 2017). Similarly, natural LoF mutations segregating in various strains may underlie the variable penetrance of risk mutations and point to novel drug targets. Genetically diverse reference panels such as the ones used in Neuner et al (Neuner et al., 2018) may be a useful resource that will enable systematic identification of protective loci and previously unappreciated biological pathways involved in determining individual resilience to the effect of rare mutations. Second, it has been shown that at least in some instances, CNVs do not have high diagnostic specificity for psychiatric or neurodevelopmental disorders and an individual CNV can give rise to many different neuropsychiatric phenotypes (Kushima et al., 2018; Stefansson et al., 2014). The discovery of this overlap raised a number of fascinating questions, concerning not only the phenotypic boundaries between major developmental and psychiatric disorders but also the mechanistic factors involved in determining who will develop what disorder. Even if the genetic factors determining the phenotypic trajectory of a given CNV may not be completely conserved, analysis of a given mutation in various strains of mice might uncover important general principles at the level of both genes and neural circuits.

Concerns about using rodent models in light of evolutionary divergence.

It has been argued that the use of mice and rats as translational models of human neuropsychiatric disorders is severely limited by the over 80 million years of evolutionary divergence since the last shared common ancestor of rodents and primates as well as the corresponding expansion of the brain size. Perceived limitations imposed by evolutionary divergence appear even more challenging under the assumption of the polygenic model given the very poor conservation of genomic regions involved in regulating gene expression, where polygenic risk primarily resides. Obviously, the expansion of the brain size is associated with quantitative and qualitative changes in networks at the levels of individual neurons, how they mature during development and they connect with each other. However, the basic organization of nervous systems tends to be highly conservative for a number of reasons. For example, the organizational features of some neuronal structures, such as the laminar organization of neuronal cell types in the cerebral cortex, can accommodate a change in the number of cells of a particular type providing a simple mechanism for addition or subtraction of identical units. In addition, dramatic alterations in activity patterns (intrinsic firing properties, synaptic connections) of component neurons or more complex networks and correspondingly large and important evolutionary changes in the behavior of an organism can emerge with relatively small changes in the density and distribution of ion channels, receptors, synapses or modulatory inputs driven by species-specific genomic variants (Charrier et al., 2012). This flexibility provides a simple mechanism to allow species-specific behavioral differences without invoking dramatic changes in basic synaptic circuitry mechanism and computational principles (Finlay and Darlington, 1995; Turrigiano, 1999). Along these lines, one possible reason that the basic codes of neural circuitry are so well preserved across phylogenetic orders is that neural circuits are designed to “generalize” rather than perform highly specialized functions, in the sense that a single circuit can produce a variety of behavioral patterns under different circumstances or at different times during the development (Douglas and Martin, 2004). Due to this flexibility in the output of neuronal networks there is likely no pressing need to evolve a completely new circuit to produce a new behavior. Accordingly, it is quite plausible that even when new circuits are expanded and morphed into apparently novel entities such as the human prefrontal cortex (PFC) and its projections (which play critical roles in SCZ and other psychiatric disorders), the fundamental underlying principles of network information processing remain the same.

While evolutionary divergence considerations have not been used as an argument against the diverse rodent-based investigations on the nature of human behaviors, they are regrettably recycled as reasons against the acceptance of rodents as authentic models of psychiatric disease. However, while modern neuroscience very often emphasizes study of specific circuits, that may or may not be conserved at a level sufficient to justify extrapolation from rodents to humans, research in diseases such SCZ that, despite traditional focus on specific brain areas, have wider and more distributed effects on cortical brain circuitry (Glantz and Lewis, 2000; Hashimoto et al., 2008; Uhlhaas and Singer, 2010), is more likely to revolve around basic computational mechanisms that are more likely to be conserved and generalizable within mammals (in principle if not in every detail) than specific circuits (Crabtree and Gogos, 2014).

In that context, a good starting point has been focusing on the low hanging fruit of evolutionary conserved cognitive faculties, such as episodic (EM) or working memory (WM), and perceptual functions. Not only, such functions are conserved at least in a rudimentary form across many species (Carruthers, 2013) but critical neurocircuitry substrates have been identified for some of these cognitive domains and appear to be shared among humans, non-human primates and rodents (Khan and Muly, 2011). While such cognitive deficits seem to be prevalent across a wide range of neuropsychiatric disorders and therefore it could be argued that they are non-specific, they can still help illuminate basic circuitry abnormalities within a given disease context from where findings can possibly be thoughtfully extrapolated to more disease-specific symptom domains. Of course, as both rodent cognitive tasks and methods to manipulate in vivo neural activity become more sophisticated and accessible, animal model studies could shift to investigating dysfunction in more complex cognitive domains and perceptual territories, such as executive function, attentional processes and social cognition, different aspects of which are seriously affected in SCZ (Meechan et al., 2015b; Nilsson et al., 2016b). Ultimately, information collected by analysis of dysfunction across various behavioral realms and various genetic animal models could be used to guide drug screening efforts and assess which aspects of underlying pathobiology can be rescued with novel compounds or other physiological manipulations.

COGNITIVE AND SENSORY PROCESSING DEFICITS IN GENETIC MOUSE MODELS OF SCZ

While mouse models are not able to capture the entirety of symptoms of psychiatric disorders they can be used to model disease-associated mutations and focus on assessing core aspects of a disorder that may manifest in an equivalent manner in rodents. Cognitive and perceptual shortfalls, although not part of the diagnostic criteria for SCZ, represent such core features of the illness. They are mediated by the same genetic risk factors and probably by the same abnormal physiological processes as more prototypical symptoms and are more amenable to study in animal models than psychosis. In support of this, it has been shown that severity of cognitive impairment correlates with risk and perseverance of psychosis in schizophrenia (Lam et al., 2018; Olivier et al., 2017; Waltz, 2017). Along these lines, there have been efforts to validate and systematize this approach, such as the Cognitive Neuroscience Treatment Research to Improve Cognition in Schizophrenia (CNTRICS) initiative (Carter and Barch, 2007). This framework aims at selection of specific tasks for measuring affected cognitive and perceptual constructs in humans and development of homologous rodent assays with translational potential, within the domains of Perception, Attention, Executive Control, WM, EM (including object and relational memory) and Social and Emotional Processing. Indeed, the study of core deficits within WM, EM, social cognition and perception in animal models of rare mutations (Table 1) has already yielded important insights into the nature, origin and neurocircuitry underlying those deficits in SCZ.

TABLE 1:

Cognitive, sensory processing, social interaction and electrophysiological alterations in mouse models of rare SCZ-risk mutations.

Schizophrenia-Related Phenotypes in Mouse Models of Rare Mutations
Locus	Cognitive	Sensory	Social	CNS Physiology	Ref.
22q11.2del	↓WM (↓ acquisition, ↔ delay DNMTP test) ↓↑↔ Memory (depending on test): ↔ Spatial and Object memory ↓ Goal-oriented episodic memory ↓ Fear memory ↓↑↔ Executive function (depending on test and mouse model) ↑↔ Attention (depending on test and mouse model)	↓ PPI ↓ Visual mismatch negativity	↔ Social interaction ↓ Social memory	↑ Synaptic depression ↓ Paired-pulse ratio ↓ Hippocampal-prefrontal synchrony ↓ Reduction in feedforward inhibition in CA2 hippocampal area ↓ Reliability of ensemble recruitment ↓ Auditory evoked potentials	(Didriksen et al., 2017; Drew et al., 2011; Fenelon et al., 2013; Hamm et al., 2017; Long et al., 2006; Meechan et al., 2015a; Merscher et al., 2001; Nilsson et al., 2016a; Piskorowski et al., 2016; Sigurdsson et al., 2010; Stark et al., 2008)
1q21.1 del	↔ Learning a ↔ Memorya	↔ PPI (baseline) ↓ PPI (psychostim.)	↔ Social interaction a	↑ Excitability (↑Spontaneously active and ↑bursty DA neurons)	(Nielsen et al., 2017) a: claim in (Forsingdal et al.,2019)-no data
NRXN1 del	↔ WM ↓ Object discrimination memory ↓ Spatial learning ↓ Instrumental learning ↓ Associative learning	↓↔ PPI	↑↓↔ Social interest ↓ Social memory ↓Maternal care	↓Excitatory synaptic transmission (↓spontaneous miniature synaptic events, ↓evoked responses)	(Dachtler et al., 2015; Etherton et al., 2009; Grayton et al., 2013)
15q13.3del	↓ WM ↓ Attention ↓ Spatial Memory (homozygous KO) ↓ Auditory cued fear memory ↔ Reversal learning, paired-associate learning, extinction learning, progressive ratio, trial-unique non-match to sample, object recognition	↓↔ PPI ↓ amplitude of auditory-evoked potential	↑ Social interest ↓↔Social interaction (dependent on different models)	↑Baseline gamma-activity ↓Evoked gamma-activity ↑Connectivity between brain regions	(Fejgin et al., 2014; Forsingdal et al., 2016; Gass et al., 2016; Kogan et al., 2015; Nilsson et al., 2016b)
16p11.2dup	↑ Object Recognition Memory	PPI not tested	↓↔Social interaction (depending on different background strain)	↔↓Excitatory synaptic transmission ↔Paired-pulse ratio ↓LTP induction	(Arbogast et al., 2016; Blizinsky et al., 2016; Blumenthal et al., 2014; Horev et al., 2011)
SETD1A	↓WM (↔acquisition,↓delay DNMTP test) ↔ Object discrimination	↔ PPI	↔ Social interaction ↔ Social memory	↑Synaptic depression ↑Neuronal excitability	Mukai et al., 2019
DISC1	↓WM (↔acquisition, ↓delay DNMTP test, ↔radial arm test) ↑↔Fear memory (↑context, ↔cued ↓↔Spatial reference memory ↔ Object Recognition Memory ↓Visual discrimination	↓↔PPI	↑Social interaction ↓Social preference ↓ Social memory (when mutation combined with adolescent isolation rearing or poly I:C MIA)	↑Synaptic depression ↓Frequency facilitation ↓↔Paired-pulse ratio ↑↓↔ LTP induction (depending on region and protocol) ↑Neuronal excitability	(Cractree et al., 2017; Gomez-Sintes et al., 2014; Koike et al., 2006; Kuroda et al., 2011; Kvajo et al., 2008; Kvajo et al, 2011; Li et al., 2018; Wulaer et al., 2018)

WM deficits.

WM deficits hold a central role within the cognitive impairment in SCZ (Forbes et al., 2009). In patients, they appear to be independent of the patients’ clinical state as they are even present at prodromal phases (Zanello et al., 2009), they appear to distinguish at-risk individuals from those who progress to psychosis (Pukrop et al., 2007) and they are also stable over time, from the first episode to chronic stages (Haenschel et al., 2009; Reilly et al., 2007). WM impairment in SCZ is typically refractory to medication (Carter et al., 1996) and its severity predicts the degree of patients’ functional impairment (Heinrichs et al., 2010). While WM deficits are rarely found in the healthy population, they are present in first-degree relatives of SCZ patients, pointing to their heritability (Thermenos et al., 2004). Components of WM affected in SCZ include encoding and goal maintenance of information, as well as interference control (Lee and Park, 2005) across a variety of modalities including spatial, verbal/auditory and object. WM dysfunction in SCZ is thought to originate from dorsolateral PFC dysfunction and communication with other brain regions (Barch and Ceaser, 2012) and as such it is considered an excellent opportunity to illuminate the neurobiological substrates of the disease (Arguello and Gogos, 2012). Rodent tasks typically used to assay WM deficits include delayed matching/non-matching tasks, stimulus (usually odor)-span tasks and N-back tasks which focus on goal maintenance, memory capacity and interference control, respectively (Dudchenko et al., 2013). Importantly, these tasks assay equivalent human cognitive processes that rely on shared neural substrates among species, with primary focus on the circuitry involving the PFC, HPC and communication between these two regions. Analysis of WM deficits in animal models of rare mutations has provided substantial knowledge on disease mechanisms and has established a framework for the testing of translational, treatment-oriented hypotheses.

One such example is the comprehensive analysis performed in a mouse model of the 22q11.2 microdeletion. Carriers of the 22q11.2 microdeletion exhibit an increased rate of psychiatric disorders, predominantly SCZ (in approximately 30% of the cases) (Pulver et al., 1994; Sobin et al., 2005). Df(16)A^+/− mice were created by deleting the mouse chromosome 16 region syntenic to the 1.5-Mb human 22q11.2 deletion encompassing ~27 genes (Stark et al., 2008). Df(16)A^+/− mice show a cognitive deficit profile reminiscent of SCZ which includes WM deficits (Stark et al., 2008). Specifically, assessment of spatial WM performance in a T-maze delayed non-match to place task revealed that during training Df(16)A^+/− mice cannot acquire the task as fast as WT littermates. Notably, mutant mice that eventually learn the task perform as well as WT littermates upon subsequent WM testing.

PFC dysfunction plays a key role in functional deficits seen in Df(16)A^+/− mice (Fenelon et al., 2013) and WM representations are thought to be maintained by a dynamic population firing pattern through changes in synaptic strength controlled by short-term synaptic plasticity in PFC (Mongillo et al., 2008). Consistently, it was shown that layer 5 pyramidal neurons in the PFC of Df(16)A^+/− mice show marked alterations in short-term synaptic depression and potentiation, as well as more modest alterations in long-term potentiation along with a significant instability of dendritic spines (Fenelon et al., 2013). Spatial WM deficits have also been associated with deficits in functional connectivity between the HPC and PFC (Meyer-Lindenberg et al., 2005). Indeed, Df(16)A^+/− mice have profound deficits in the synchronization of PFC and HPC networks during WM tasks (Sigurdsson et al., 2010). It has been shown that these defects are at least in part due to the deficient growth of pyramidal cell axons at perinatal stages (Mukai et al., 2015). Specifically, Mukai et al., (2015) showed that mice deficient for a single gene within the Df(16)A region, Zdhhc8, also had impaired HPC-PFC synchrony. Notably, hippocampal neurons in Zdhhc8^+/− mice made fewer branches in the PFC and similar disruptions were also observed in interhemispheric projections. Notably, interfering with the signaling cascades controlling axonal growth during early development leads to efficient rescue both of the functional connectivity alterations and the WM deficits in Df(16)A^+/− mice, with important translational implications for preventing disease onset in at risk individuals (Tamura et al., 2016).

Another relevant example comes from the analysis of Setd1a-deficient mice. SETD1A is a confirmed SCZ risk gene and SETD1A mutations confer a large increase in disease risk (Singh et al., 2016; Takata et al., 2014), which provides high confidence that it represents a robust genetic substrate for disease modeling. Assessment of spatial WM performance of a mutant mouse strain carrying a LoF allele in the endogenous Setd1a orthologue (that models the various identified SETD1A LoF risk alleles) using a T-maze delayed non-match to place task revealed no initial learning impairments in task acquisition, as Setd1a^+/− mice performed almost as well as WT littermates. In the WM test, where WM was assessed with increased delays between the sample and the test run, however, Setd1a^+/− mice manifested a marked impairment compared to WT littermate controls. Similar to Df(16)A^+/− mice, WM deficits in Setd1a^+/− mice were associated with alterations in short-term synaptic plasticity and neuronal excitability in the PFC as well as with deficient growth of cortical pyramidal cell axons. Importantly, it was demonstrated that the deficiency of Setd1a during early developmental stages does not irreversibly compromise the plasticity of neural circuits, since WM impairments can be partially corrected by restoration of Setd1a function in adulthood (Mukai, 2019). Because SETD1A encodes for a histone methyltransferase, these findings inspired a series of pharmacological interventions with well-characterized and potent inhibitors to directly target downstream demethylation processes aiming at generating robust methylation changes which could counteract effects of Setd1a deficiency. It was shown that pharmacological antagonism of LSD1, the major demethylase antagonizing SETD1A, resulted in a full rescue of the behavioral abnormalities and axonal arborization deficits in adult Setd1a-deficient mice. Such findings may prove useful for development of new therapeutics via drug repositioning since LSD1 inhibitors are already in clinical trials in oncology (Fu et al., 2017; Gogos, 2018).

Detailed analysis of WM performance in mutant mice, generated to carry a truncating lesion in the endogenous Disc1 orthologue that models a balanced t(1;11) chromosomal translocation linked to mental illness risk in a multigenerational Scottish pedigree, also revealed a PFC-related impairment of WM (Kvajo et al., 2008). The pattern was identical to the one observed in Setd1a-deficient mice. That is, mutant mice acquired the task and performed almost as well as WT littermates during training but demonstrated marked impairment during the WM test. Mutant mice are impaired especially when WM cognitive control demands are high but not in any other HPC-dependent cognitive task employed, suggesting primary impairment in information processing rather than in the ability to store information per se (Kvajo et al., 2008). Electrophysiological analysis in the PFC of mutant mice revealed altered excitability of layer II/III pyramidal neurons as well as altered paired-pulse ratios and short-term depression of layer V synapses (Crabtree et al., 2017). Unlike Df(16)A^+/− mice and Setd1a-deficient mice no efforts have been reported towards restoring WM deficits in this mouse strain. However, the observed electrophysiological alterations were found to be associated with reduced functional expression of the voltage-dependent potassium channel subunit Kv1.1, attributed to increased cAMP in PFC generated by a reduction in phosphodiesterase 4 activity due to the Disc1 deficiency (Crabtree et al., 2017) thus pointing to promising and potentially effective pharmacological interventions.

Episodic memory deficits.

Deficits in long-term episodic memory are frequently observed in patients with SCZ (Dickinson et al., 2007; Schaefer et al., 2013) and play a prominent role in functional outcome (Aleman et al., 1999; Heinrichs and Zakzanis, 1998). Spatial and verbal memory are most affected and recollection impairment appears to be more prominent than familiarity impairment (Libby et al., 2013) consistent with evidence supporting a greater impairment in associative compared to item memory in SCZ (Achim and Lepage, 2003). In contrast to WM, the body of work on animal studies of EM is quite limited. One of reasons for this seems to be the focus on simplistic tasks of long-term memory, such as the Novel or Spontaneous Object Recognition Task, which has been also criticized to lack construct validity (Bussey et al., 2013) and has not revealed deficits reliably in all animal models of SCZ. While new alternative tests are emerging that are more appropriate for studying EM deficits (Bussey et al., 2013), there have only been a handful of studies that have focused on SCZ-related EM deficits in genetic animal models, which however have led to new insights into the neural network dysfunction.

For example, a comprehensive analysis of Df(16)A^+/− mice and their WT littermates during learning and recall of a goal-oriented hippocampal-dependent EM-like task revealed a marked learning impairment that was dependent on task demands. Learning performance was disrupted by changes in context and reward-location, reminiscent of specific EM deficits in SCZ patients and 22q11.2 deletion carriers (McCabe et al., 2011). Monitoring of place cell dynamics during task performance revealed that the learning deficit of the Df(16)A^+/− mice was due to impaired stability and the inability of hippocampal place fields to reorganize in response to salient information. This finding has important implications for elucidating the neuronal correlates of memory deficits associated with SCZ and also, as we argue below, points to the need of shifting the focus of analysis to extended neuronal population dynamics.

Another example is provided by a study on Disc1-mutant mice, adapting a combination of genetic predisposition (Disc1 mutation) and environmental challenge (immune challenge during pregnancy by treatment with the viral mimetic polyriboinosinic-polyribocytidilic acid (Hartung et al., 2016). Combined insults led to poorer cognitive performance in the recency recognition (RR) task, an EM-like temporal order object recognition task, attributed to impaired maturation of long-range interactions between PFC and HPC networks.

Social Cognition deficits.

Social cognition deficits are prominent features in SCZ diagnosis and affect the functional outcome of patients (Javed & Charles 2018) and include impairments in face and voice perception, mentalizing and emotion regulation as well as impairments in the ability to utilize social contextual information and in complex social cognitive functions such as empathy (Green et al 2015). Translational approaches of studying social cognitive functions in genetic models of SCZ are inevitably challenged by the dependence of human social cognition on language and on visual and verbal processing of emotional or social cues, whereas social interaction in rodents is primarily dependent on olfaction. Social functioning in genetic models of SCZ has been studied extensively by social interaction tasks (assaying the repertoire of social behavior exhibited by a subject mouse when exposed to a non-aggressive conspecific) and to a lesser extent by social recognition/memory tasks (assaying the subject’s memory of a previous exposure to a conspecific) (Table 1). Introducing additional element of precise behavioral analysis in social interaction and memory tasks in rodents, may enable a more detailed way to assess recognition and response to social cues and improve translational efforts. To this end, Df(16)A^+/− mice (Piskorowski et al., 2016), but not Setd1a-deficient mice (Mukai, 2019), were found to display impaired social memory. Initial analysis implicated dysfunction of the CA2 area of the HPC as one major contributor to this impairment providing for the first time a route towards elucidating the neural substrates for social cognition deficits in psychiatric disorders and facilitating development of treatments that target such deficits.

Sensory information processing deficits.

Alterations in sensory information processing have been widely recognized in SCZ and are particularly amenable to modeling in mice due to the strongly conserved elementary mechanisms they involve (Javitt and Sweet, 2015). Auditory processing deficits, in particular, may underlie auditory hallucinations and contribute significantly to the social cognition dysfunction and functional outcome of patients (Wynn et al., 2010). Whereas auditory processing deficits have been studied extensively, visual processing deficits have only recently been appreciated as another promising entry point into probing neural dysfunction mechanisms and translational research in SCZ especially since visual cortex circuitry has been extensively studied and understood (Butler et al., 2008; Javitt and Freedman, 2015).

Of the multiple aspects of auditory and visual domains affected in SCZ, gain control and integration of sensory stimuli seem to be the most appropriate for modeling in animals, based on the cross-species compatibility of tests and conserved neuroanatomical substrates (Siegel et al., 2013). Gain control refers to the processes that enable sensory systems to optimally adapt their responses to stimuli, taking into account alterations in the surrounding background context (Green et al., 2009). Tests typically employed to study gain control in human and animal models include the prepulse inhibition (PPI) test, auditory event-related potentials such as such as P50, N100 and mismatch negativity (MMN). Assays of PPI detect sensorimotor gating deficits in animal models but despite their widespread use, studies in etiologically relevant models of SCZ have yet, with few exceptions (Chun et al., 2014), to offer important new insights in relevant circuit abnormalities. Moreover, deficits in PPI are not consistently observed among genetic mouse models of SCZ. For example, Df(16)A^+/− mice and Df(h15q13)^−/− mice (Forsingdal et al., 2016) show robust deficits in PPI whereas no PPI deficits have been observed in Setd1a, Disc1 and Df (h1q21)/⁺ mice (Nielsen et al., 2017) (a mouse model of the 1q21.1 deletion). Apart from work on PPI, insights into auditory perception deficits have emerged from other sensory gating tests. One such example is the use of the “paired click paradigm” in the Df(h15q13)/+ mouse model of the 15q13.3 deletion, which revealed decreased neuronal firing in response to auditory stimuli and reduced neuronal responses to fast 80 Hz steady‐state auditory stimulation in the auditory cortex that was attributed to interneuron-related mechanisms. Treatment with RE01, a positive modulator of the Kv3.1 channels, which are known to be highly expressed in fast-spiking interneurons reversed this effect (Thelin et al., 2017). Deficits in processing self-generated sounds in animal models of SCZ provide another example. Macroscopic measurements of brain activity in healthy human subjects have consistently shown that responses to self-generated sounds are attenuated in amplitude, albeit to a lesser degree in individuals suffering from SCZ (Ford et al., 2001; Ford et al., 2014; Perez et al., 2012). This deficit may contribute to the hallucinations and delusions observed in the disease. In mice, responses of auditory cortical neurons are attenuated to sounds generated manually by the animals’ own behavior establishing an experimental paradigm allowing investigation of neural circuit impairments (Rummell et al., 2016). Preliminary results revealed that Df(16)A^+/− mice show reduced attenuation compared to WT littermates to self-generated sounds compared to randomly occurring stimuli (Rummell, 2018).

Assays of MMN are designed to detect an infrequent deviant stimulus in an ordered array of frequent and predictable stimuli. For example, the standard stimuli are often tones of a specific frequency and the deviants are tones of a different frequency (so-called ‘frequency deviants’). In healthy individuals the deviant stimulus elicits a neural response called the MMN, which can be measured using EEG or MEG. Decreases in MMN, is a very robust finding with a considerable degree of specificity for SCZ and has been identified as one index of vulnerability to disease progression in high risk populations (Erickson et al., 2016; Umbricht et al., 2004). Because MMN is a reliable indicator for exploring pre-attentive processing is particularly suitable to study basic circuit function with little confounding influence from outside brain areas. While MMN is widely considered as one of the best-established measures of auditory sensory dysfunction in SCZ patients and mouse model, visual MMN assays have also been recently adapted and validated in the mouse (Hamm and Yuste, 2016). Specifically, fast two-photon calcium imaging and multielectrode recordings in awake mice demonstrated that visual cortical circuits displayed decreased responses to repeated stimuli while amplified responses to a deviant stimulus. Notably it was shown that pharmacogenetic silencing of somatostatin-expressing interneurons was sufficient to abolish detection of deviant stimuli through elimination of response amplification. Consistent with MMN deficits found in SCZ, preliminary results revealed a marked deficit in Df(16)A^+/− mice during a visual “oddball” MMN task (Hamm, 2017).

Overall, cognitive and sensory information processing deficits are frequently observed in genetic mouse models of SCZ and can guide identification of underlying neural circuity alterations. Among them, deficits in WM or WM-dependent learning, a core cognitive phenotype of psychotic disorders (Arguello and Gogos, 2012), are the most consistently present and efforts here have already led to remarkable progress towards elucidating neurophysiological mechanisms relevant for disease and development of new therapeutics. By contrast, other cognitive and sensory processing deficits are more erratic in their appearance in mouse models. For example, Df(16)A^+/− mice show robust deficits in PPI, contextual and cued fear conditioning task for associative memory and direct interaction test for social memory while such deficits are not consistently observed in other mouse models (Forsingdal et al., 2016; Koike et al., 2006; Nilsson et al., 2016b). Notably, even for convergent cognitive dysfunction the exact pattern of disturbance may not be identical among different mutations. Differences, for example, in affected aspects of the same WM task (learning in the Df(16)A^+/− mice versus performance in Setd1a and Disc1 mutant mice) among mouse models suggest that within the confines of well-circumscribed dysfunctions there may be variable patterns of circuitry upheaval, which may underlie the variable severity of the symptoms associated with specific mutations. Related to this issue are two observations: First, that mutant mice carrying single gene mutations (such as the Setd1a and Disc1 mutant mice) display some but not all of the SCZ-related phenotypic outcomes observed in models of highly penetrant multi-genic CNVs [such as the Df(16)A^+/− mice]. Second, mouse models of highly penetrant multi-genic CNVs [such as the Df(16)A^+/− mice] exhibit, reassuringly, markedly more severe cognitive and sensory processing phenotypes than multi-genic CNV models with weaker disease penetrance [such as the Df(h15q13)^+/− and Df (h1q21)/⁺ mice]. These observations are consistent with the notion outlined above that the entire constellation of phenotypic outcomes and disease symptoms arise as result of the synergistic action among multiple rare risk variants, a condition more faithfully recapitulated in models of highly penetrant multi-genic CNVs compared to single-gene mutations.

NEURAL SUBSTRATES OF COGNITIVE AND SENSORY PROCESSING DEFICITS

The complexity of the neural substrates affected in SCZ offers a large mutational target comprising many genes. One important question is where this diverse genetic risk converges, to generate a common pattern of clinically observed dysfunction. For example, there is bioinformatics evidence from cross-referencing SCZ risk genes lists with other large-scale “omics” datasets, that genes disrupted by common and rare variants appear to functionally converge upon a highly connected network of synaptic proteins, including synaptic cell adhesion molecules, scaffolding proteins as well as neurotransmitter receptors and associated protein complexes (Chang et al., 2018; Gilman et al., 2012). Additional investigations of, primarily, the polygenic component of the risk, attempt to rationalize convergence of genetic heterogeneity in terms of its effects on basic molecular processes such as chromatin conformation, gene expression or RNA splicing (Gandal et al., 2018; Rajarajan et al., 2018). This level of description though, as useful as it may be for deciphering the most proximate molecular effects of risk variation, fails to capture how genetic liability manifests itself within the brain circuitry underlying disease symptoms and how it relates to specific pathophysiological states that differentiate SCZ from other neuropsychiatric syndromes. Therefore, efforts to frame the diversity of risk genes in terms of their consistency in functional domains, expression or splicing patterns are not very informative, in terms of understanding the biology behind emergence of disease symptoms and lack the granularity needed to meaningfully decipher disease-specific dysfunction.

In that context, the employment of etiologically relevant animal modes coupled with core cognitive and perceptual processing tasks and probed with high-resolution systems-neuroscience techniques, such as opto- and chemo-genetic manipulations, neuronal recordings from behaving animals and multiphoton live imaging, has started to provide unprecedented expansion of knowledge that is currently building the foundation of a new biological framework for mental illness, which does not revolve around specific neurotransmitters or neuronal cell types. Prominent among the important findings and principles emerging from animal models of rare mutations are the degradation of the ordered activity and plasticity of both local and long-range neuronal assemblies, the cortical building blocks necessary for stable cognition, perception and reliable information processing. These disruptions thus might be a common point of convergence of genetic liability, leading to unreliable internally- or stimulus-driven neuronal activation patterns that may trigger the constellation of SCZ symptoms, such as perceptual distortions, social withdrawal and cognitive deficits. Open questions include how exactly optimal computational regimes of neuronal networks are altered, when in developmental time they are altered, where in the brain they are primarily altered and how such alterations likely vary from disease to disease contributing to the establishment of nosological boundaries.

Following the successful elucidation of the genetic principles underlying disease risk and the identification of risk loci, answering these questions represents the next big challenge in SCZ research. Animal models of rare penetrant mutations have the unique potential to help us move from “which” (gene) to “how”, “where” and “when” basic computational functions of neural circuit dynamics are affected. Linking these variables should lead to a better understanding of how discrete genetic insults lead to a specific set of symptoms, help re-conceptualize diagnostic boundaries at the circuit level and also facilitate identification of critical developmental windows during which neural circuits are particularly vulnerable to insults (Marin, 2016). A better resolution of pathophysiological trajectories at the level of neural circuitry should also help to rationally guide the development of new therapeutic approaches aiming at targeting either altered circuitry itself or the underlying biological pathways (Carrillo-Reid et al., 2016).

Local cortical circuits alterations.

Two recent groundbreaking studies in the HPC and in the visual cortex of Df(16)A^+/− mice during a goal-directed task for EM or during visual processing, respectively (Hamm et al., 2017; Zaremba et al., 2017) provided the first empirical evidence reinforcing the view that a better understanding of SCZ symptoms will likely emerge from circuit analysis by studying the properties of local neuronal population dynamics during relevant network functional states normally recruited during information-processing tasks. Zaremba et al (2017) attempted to elucidate neuronal correlates of EM-like memory deficits associated with 22q11.2 deletions employing a goal-oriented, HPC-dependent EM-like task. Df(16)A^+/− mice showed a marked impairment in learning performance, which was disrupted by changes in task-context and reward-location. Place cell ensembles in the HPC provide a well-characterized and tractable system to enable systematic dissection of the molecular, cellular and synaptic mechanisms underlying cognitive dysfunction. Monitoring of place cell activity dynamics during task performance revealed that the episodic memory deficit of the Df(16)A^+/− mice was due to impaired stability and plasticity of neuronal ensembles in the HPC, resulting in an inability of place fields to reorganize in response to changing environmental task demands. Notably, baseline properties of CA1 principal neurons, baseline place cell stability and initial learning of a goal location were not affected, consistent with behavioral and electrophysiological analysis of the HPC of Df(16)A^+/− mice (Drew et al., 2011; Stark et al., 2008), which indicated minimal and well-circumscribed alterations in synaptic plasticity, excitability and elementary spatial and context-related cognitive functions.

A parallel study in Df(16)A^+/− mice related to visual processing abnormalities in SCZ (Spencer et al., 2004; Uhlhaas and Singer, 2010) employed long term two-photon calcium imaging of multi-neuronal co-activity dynamics within cortical microcircuits in the primary visual cortex and described a systematic disorganization in the co-activity patterns of neurons within local neuronal assemblies. Specifically, these studies revealed that the distinctness of ensembles is altered and recurrent ensemble activation is less reliable with time upon visual stimulation (Hamm et al., 2017). By contrast, single-cell activity patterns were minimally altered. Importantly, acute chemogenetic suppression of parvalbumin (PV) positive interneurons in WT mice disinhibited neuronal activity but was insufficient to disorganize ensemble-level activity and recreate the higher-order changes observed in Df(16)A^+/− mice. While the function of PV interneurons has not been directly tested in mutant mice, this demonstration suggests either a developmentally confined role for PV interneuron dysfunction or, while being a contributing factor to the microcircuitry deficits in Df(16)A^+/− mice, 22q11.2 deletions affects neuronal dynamics primarily via destabilization of glutamatergic synapses in excitatory neurons (Fenelon et al., 2013; Mukai et al., 2008).

Taken together, these in vivo findings revealed for the first time that a genetic manipulation, modeling key aspects of SCZ risk leads to changes in the activity, plasticity and adaptability to sensory input of local functional ensembles that are markedly more pronounced than alterations in individual neurons possibly suggesting that seemingly small functional changes in individual neurons can lead to dramatic systems-level alterations when they are assembled into functional networks. Notably, analogous destabilization in neuronal dynamics have been subsequently recapitulated in in vitro HPC preparations from another model of the 22q11.2 deletion (Lgdel/+ mice) and correlated with deficits in the synchronized activity of the CA1 neuronal network in awake Lgdel/+ mice, in the form of diminished spike correlation between neuronal pairs and a reduction in neuronal oscillations in the theta frequency range (Marissal et al., 2018). Interestingly, these network deficits as well as various HPC-dependent behavioral deficits were partially or fully restored by acute pharmacological or chemogenetic manipulations of PV interneuron excitability, implicating inhibitory regulation in establishing stability of local circuit dynamics at least in the HPC.

Long-range cortical circuits alterations.

Dysfunction in the long-range functional connectivity between brain regions has been be implicated in the pathophysiology of several psychiatric disorders including SCZ. Dysfunction in these long-range cortical circuits often occurs concomitantly and may be causatively linked, at least in part, with local-circuit alterations. Indeed, impairments in functional connectivity (as reflected in the coordination of neural activity across different brain regions) between PFC and other cortical and subcortical circuits, including those of HPC, amygdala and the basal ganglia, could be a more general disease mechanism through which a range of positive and cognitive symptoms manifest within a given or across different psychiatric disorders. Work on the Df(16)A^+/− mice revealed for the first time how disrupted long-range connectivity is manifested in the activity of neural circuits and can arise at the level of single neurons, as a result of an unequivocal genetic risk variant. Df(16)A^+/− mice performing a delayed alternation T-maze WM task, simultaneously monitored with in vivo recordings from the PFC and the HPC, showed a robust decrease in PFC-HPC synchrony, both in terms of decreased phase-locking of prefrontal cells to hippocampal theta oscillations as well as of reduced coherence of PFC and HPC local field potentials. Together these findings indicate that in addition to short-range synchrony, impaired long-range synchrony of neural activity is likely a core feature of brain dysfunction associated with a SCZ risk mutation (Sigurdsson et al., 2010). In fact, decreased fronto-temporal and limbic connectivity and decreases in resting-state long-range default mode network connectivity have been associated with the 22q11.2 deletion syndrome (Debbane et al., 2012) and are consistent with the growing number of studies which demonstrate macroscopic functional connectivity impairment in patients with SCZ (reviewed by (Dawson et al., 2015). Additional evidence for long-range dysconnectivity emerging as a result of the 22q11.2 deletion comes from a study on thalamocortical glutamatergic projections in the auditory cortex in another mouse model of the 22q11 deletion. In particular, Df(16)1/+ mice carrying a chromosomal deficiency encompassing only a subset of 22q11.2 orthologous mouse genes show a disruption of synaptic transmission in thalamocortical projections to auditory cortex (but not visual or somatosensory cortices), a deficit correlated with deficient PPI responses and linked to an elevation of D2 receptors in the thalamus (Chun et al., 2014). Notably, abnormal long-range communication within HPC-PFC circuits has also been demonstrated in Disc1-deficient mice, which exhibit impaired patterns of long-range synchrony and directed interactions within HPC-PFC networks where decreased coupling of the two areas at neonatal age switches to hyper-communication in pre-juvenile animals (Hartung et al., 2016). Finally, resting-state fMRI analysis of a 15q13.3 CNV mouse model revealed widespread patterns of altered functional connectivity, including many brain networks thought to be dysregulated in SCZ, which were pharmacologically normalized by positive modulation of nAChA7 receptor (Gass et al., 2016). Gauging the extent of long-range synchrony deficits as well as examining whether such deficits in different circuits make distinct contributions to behavioral impairments is an important goal of future studies in animal models along with a detailed mechanistic analysis of such deficits, which has already offered important translational insights into how such deficits can be restored at the circuit or individual-neuron level.

Diverse genes, diverse failures.

A diverse set of failures, reflecting the diversity of genetic risk, could individually or in combinations converge to impaired activity and plasticity of neuronal ensembles by causing sparser structural connectivity, altered synaptic plasticity and excitatory-inhibitory (E-I) imbalances as well as altered neuromodulation. Acting alone or together, such changes may lead to local and long-range dilapidation of the ordered activity and plasticity of neuronal assemblies needed to support the cognitive and perceptual operations affected in SCZ. When considering the diverse ways that a system can fail we need to take into account additional level of diversity introduced by the developmental timing and spatial distribution under which such failures can emerge. Animal models are well suited to assess and characterize the nature of these failures and the spatiotemporal patterns of their emergence.

Failures in structural connectivity, in various anatomical scaffolds of cortical circuitry may include, for example, reduction in terminal arborization of cortical axons as prominently found in the Df(16)A^+/− and Setd1a-deficient models (Mukai et al., 2015; Mukai, 2019). Sparser connectivity may also emerge due to altered dendritic and spine development which has emerged as a common feature of mouse models of psychiatric and neurodevelopmental disorders (Blizinsky et al., 2016; Mukai et al., 2008; Nakai et al., 2018). These results raise the intriguing possibility that deficits in both local and long-range synchrony in neuronal activity could be caused, to a large degree, by disrupted anatomical connectivity, which is consistently been observed in SCZ patients (Cannon et al., 2002; Glantz and Lewis, 2000; Pettersson-Yeo et al., 2011).

Failures in synaptic transmission and plasticity is also a common feature of genetic mouse models of SCZ (Table 1) (Crabtree and Gogos, 2014). For example, the consistently observed deficits in WM interpreted in the context of synaptic theories for WM, which advocate that information entering the WM domain is maintained through short-term synaptic plasticity in population of neurons (Mongillo et al., 2008), have been particularly informative in interpreting consistently observed and marked alterations in short-term synaptic plasticity (depression and facilitation) and cortical neuron excitability. These impairments are thought to contribute to altered circuit dynamics in vivo during WM performance based on the role of these processes in determining temporally precise functional connectivity between cell ensembles and could represent sites of particular vulnerability contributing to the WM deficit seen in the mutant mice. Given its importance in WM, short-term synaptic plasticity can serve as a candidate for explaining other forms of circuitry dysfunction in disorders like SCZ (Crabtree and Gogos, 2014).

Failures in neuromodulatory systems, especially in dopaminergic transmission, are thought to contribute to some SCZ symptoms and have been reported in some mouse models of pathogenic CNVs (Nielsen et al., 2017). Failures of E-I balance are also frequently observed in genetic mouse models of SCZ. One source of such failures could be alteration in intrinsic excitability of cortical pyramidal neurons as demonstrated for example in Setd1a and Disc1-deficient mice or in the number of synapses due to axonal or dendritic arborization as demonstrated Df(16)A^+/− and Setd1a-deficient models (Mukai et al., 2015; Mukai, 2019), leading to altered network excitatory drive. E-I perturbations and cognitive deficits in SCZ models have also been linked to inhibitory deficits in GABAergic interneurons at multiple levels, consistent with findings from human postmortem studies (Chung et al., 2016; Volk et al., 2002). Defects in specific classes of cortical interneurons that affect their recruitment into local networks could lead to altered synchronization of local pyramidal cell activity with distinct implications on network activity and cognition depending on the specific class of interneurons affected (Del Pino et al., 2017). Such deficits could be due to altered intrinsic activity and excitability or impaired formation of inhibitory synapses. Decreased spontaneous firing of GABAergic interneurons in Df(16)A^+/− mice (Drew et al., 2011) and reduced fast-spiking interneuron baseline activity in Df(h15q13)/⁺ mice (Nilsson et al., 2016b) are two relevant examples. In addition, cell adhesion molecules linked to psychiatric disease such as Neurexins are known to regulate the synaptic output of GABAergic interneurons in a cell-specific manner. For example, complete deletion of all neurexin forms decreases dramatically the number of inhibitory synapses made PV neurons on pyramidal cells (Chen et al., 2017). E-I perturbations could also emerge due to altered neuromodulation. One relevant example comes from recent findings that mice carrying mutations in the gene encoding Proline dehydrogenase (Prodh, within the 22q11.2 deletion) and exhibiting elevated levels of the GABA-mimetic neuromodulator L-proline, have decreased levels of GABA production and specific deficits in GABA-ergic transmission and in gamma-band oscillations (Crabtree et al., 2016). Another example of altered interneuron neuromodulation in the same genetic locus is provided by the observation that PV interneurons in Df(16)A^+/− mice are less susceptible to modulation via D2 receptors, which likely disrupts the ability of dopamine to dynamically sculpt E/I balance under conditions of elevated dopaminergic activity (Choi et al., 2018).

Conclusion:

A DNA sequence alteration unequivocally associated with SCZ can be inferred to indicate an unambiguous route towards causation. The distance traveled between a risk gene and a resultant disease phenotype, however, may not always be a short one but, instead, span via a very tortuous and dynamical path, several levels of biological hierarchy and developmental time. Indeed, the detailed route taken by each mutation may be vastly different with all of these diverse routes converging, with some spatiotemporal specificity, towards a small set of altered brain states that underlie a defined set of disease symptoms. In the context of such genetic and mechanistic diversity, models of rare risk variants provide the most tractable way to (i) decipher convergence patterns at the level of neural computation and local and long-range information processing in the brain and (ii) investigate how exactly the ever-expanding range of lower-level deviations in genetic, molecular, cellular wiring, synaptic plasticity and homeostasis mechanisms converge on altering the stability and plasticity of local neuronal microcircuits. These later studies are expected to provide valuable mechanistic insights and, most importantly, identify novel drug targets, which may be tailored to the effects of individual risk mutations or functionally related groups of risk mutations, a novel paradigm in psychiatry.