Polygenicity in a Box: Copy Number Variants, Neural Circuit Development, and Neurodevelopmental Disorders

Abstract

Clinically defined neurodevelopmental disorders (cd-NDDs), including Autistic Spectrum Disorder (ASD) and Schizophrenia (Scz), are primarily polygenic: Multiple risk genes distributed across the genome, in potentially infinite combinations, account for variable pathology. Polygenicity raises a fundamental question: Can “core” cd-NDD pathogenic mechanisms be identified given this genomic complexity? With the right models and analytic targets, a distinct class of polygenic mutations—Copy Number Variants (CNVs): contiguous gene deletions or duplications associated with cd-NDD risk—provide a singular opportunity to define cd-NDD pathology. CNVs orthologous to those that confer cd-NDD risk have been engineered in animals as well as human stem cells. Using these tools, one can determine how altered function of multiple genes cause serial stumbles over cell biological steps typically taken to build optimal “polygenic” neural circuits. Thus, cd-NDD pathology may be a consequence of polygenic deviations—stumbles—that exceed limits of adaptive variation for key developmental steps.

INTRODUCTION

At least 80% of the mammalian genome, from rodents to humans, is used to build or maintain the brain. Thus, brain development, function, and dysfunction must reflect activity of multiple genes in broad, dynamic networks. This does not discount “nodal” contributions of single genes [1, 2]; however, the engagement of 80% of the genome to ensure optimal brain assembly and activity emphasizes likely roles for multigene networks in neural circuit construction and function. Accordingly, development and maintenance of neural circuits, fundamental units of brain organization, function, and behavior, must be polygenic. If polygenic circuits ensure optimal function, then clinically-defined (cd) brain diseases including Autistic Spectrum Disorder (ASD) and Schizophrenia (Scz), collectively labeled “neurodevelopmental disorders” (NDDs) [3], may reflect concerted effects of mutations in multiple genes within broad, overlapping gene networks leading to circuit dysfunction. Indeed, single gene variants do not reliably predict cd-NDD pathology; instead, numerous variants confer small, but significant risk [4, 5]. The challenge of multigene cd-NDD pathology has been distilled in the title of a recent review “Polygenicity in psychiatry—like it or not, we have to understand it” [6]. Indeed, one can argue that polygenicity is far from “unlikable”. If studied using valid genomic models, polygenicity most likely provides the best framework to define how flexible multigene networks guide the multi-step process of neural circuit development or its disruption in cd-NDDs.

Polygenic NDDs: So many genes, so few models

Current estimates of statistically verified single gene variants associated with cd-NDD risk include more than 100 for ASD [5] and at least 50 for Scz [4]. These mutations are distributed across all 23 human chromosomes and have relatively modest individual correlations with cd-NDD pathology (Figure 1). There is, however, another class of cd-NDD-associated polygenic disruptions: Copy Number Variants (CNVs; Figure 1; [7]). These mutations—deletions and duplications of small to large chromosome segments (10² > 10⁷ bPs; including single gene deletions or duplications)—encompass multiple single genes for which dosage changes substantially enhance risk for cd-NDD-like cognitive and social behavior deficits. Finally, similar divergent brain structure and function seen in ASD, Scz, or other cd-NDDs, including size differences of key brain regions and altered connections [8] has also been found in CNV syndromes. Thus, CNVs provide an opportunity to analyze polygenic risk and its pathological consequences in genetically distinct, homogeneous populations with similar disease symptoms: core behavioral and connectivity deficits seen across multiple cd-NDDs.

Figure 1:

The 22 human (left; X and Y are excluded based upon the unique nature of X-linked monogenic disorders) and 19 mouse (right) somatic chromosomes. Red X’s (left) indicate the distribution of 20 Autistic Spectrum Disorder (ASD) top risk genes (de novo or inherited variants confer enhanced vulnerability, but do not predict disease; SFARIgene EAGLE score) across 13/22 human chromosomes, and red X’s (left) indicate distribution of orthologues across 12/19 mouse chromosomes. Green boxes (left) indicate four “model” multigene copy number variant (CNV) regions associated with syndromes that include substantially elevated risk for cd-NDD behavioral symptoms including (but not limited to) ASD and/or Schizophrenia (Scz), and green boxes (right) indicate orthologous murine regions. The four CNVs shown here (center) are well-conserved across these two species. 7q11.23 deletion is associated with Williams-Beuren Syndrome, for which hyper-sociability is a primary behavioral symptom. 15q11.2 is associated with elevated ASD risk, and depending on parent of origin for deletion/duplication, results in varying behavioral symptoms. 16p11.2 deletions are associated with ASD, but duplications can be associated with Scz. 22q11.2 deletion is associated with both ASD and Scz risk. Within each CNV, multiple individual genes (bold) are predicted to be haploinsufficient (HiPRED; https://github.com/HAShihab/HIPred), indicating that phenotypic changes are associated with diminished dosage. Another subset of genes (brackets) has rare SNVs independently associated with ASD or Scz risk [NHGRI-EBI GWAS; https://www.ebi.ac.uk/gwas/].

Genetic, imaging and behavioral assessments in humans with cd-NDD-associated monogenic or polygenic syndromes cannot define underlying mechanisms of in vivo neural circuit pathogenesis. At present, this sort of mechanistic insight requires genomically accurate animal models. Models of rare cd-NDD-associated monogenic syndromes, mostly in mice, have yielded significant insight into brain pathology, and in some cases, foundations for therapeutic interventions [1, 2]. Similarly, obligate functions of cd-NDD-associated single gene variants in circuit development, activity, or behavior can be assessed using full loss- or gain-of-function mutants in several model organisms. Unfortunately, studying the concerted impact of multiple single cd-NDD risk variants distributed across the genome (see Figure 1) in mice or other models has proven challenging. Engineering combinations of more than a few variants is at present difficult if not impossible in most whole animal models. In contrast, multigenic CNVs have been engineered successfully, especially in mice, to model genetic changes associated with cd-NDDs. Thus, CNVs reproduced with genomic accuracy in animal models are, at present, a singular resource to analyze how polygenic architecture prefigures cd-NDD circuit pathology.

Comparisons between human neural circuits and behaviors—or their development—with those in laboratory animals require caution: some human circuits, especially cerebral cortical circuits, are not found in model species, and laboratory animal behavioral repertoires do not include key cognitive, affective, or social domains compromised in most cd-NDDS. Indeed, mice—or flies, worms, fish, and rats—carrying cd-NDD-associated CNVs or related mutations do not have ASD, Scz, or any other cd-NDD [9, 10]. Nevertheless, cellular and molecular consequences of polygenicity on circuit development and function can be resolved optimally in vivo using CNV animal models. In addition, one can analyze behaviors that engage circuits whose organization and function are relevant to cd-NDD pathology. Thus, CNV animal models provide otherwise unavailable insight into polygenic circuit pathology in an organismal context relevant—but not identical— to that in multiple cd-NDDs.

“Polygenicity in a Box”: Multigenic CNVs confer substantial NDD Risk

CNVs associated with cd-NDDs, often including 20 to 50 contiguous genes [7], provide “polygenicity in a box” to accurately model cd-NDD pathology (Figure 1). Genes that regulate neural development, including some with single risk variants, are more frequent within cd-NDD related CNVs [11], and individual CNV gene dosage sensitivity—phenotypes that reflect message/protein levels—is an apparent driver of cd-NDD related pathology [12]. CNV animal models therefore offer several advantages for analyzing polygenic cd-NDD pathogenesis: 1. CNVs can be accurately modeled by deletion/duplication of orthologous chromosomal segments. 2. Concerted changes driven by multiple genes can be analyzed simultaneously. 3. Cellular and molecular consequences for circuit development can be resolved mechanistically, and compared with single CNV gene mutations to define polygenic versus monogenic effects. 4. Mechanistic insights are likely relevant for understanding how multiple single risk variants converge upon developmental processes underlying shared cd-NDD pathology.

The advantages of murine CNV “polygenicity in a box” models are balanced by potential disadvantages: 1. CNVs engineered in mice or other models may not fully recapitulate genomic/chromosomal architecture of human counterparts (see Figure 1), even though deleted/duplicated genes are orthologous. These differences may complicate assessing non-coding or higher order chromatin modifications that likely contribute to human cd-NDD pathology [13–15]. 2. cd-NDD-associated CNVs on controlled backgrounds may fail to capture interactions between CNVs and additional single gene variants that define individual human polygenic risk [16, 17]. 3. “Weak alleles” (pathogenic variants of individual CNV genes) in humans are not likely present in animal models. 4. Some CNVs include “contiguous genes” for which single gene dosage changes drive phenotypes [18, 19]. This may complicate interpretating multigene gene effects and their relationship to cd-NDD pathology [7]. 5. Some aspects of human neural development—e.g. unique progenitor classes or morphogenesis of cortical sulci and gyri [20]—are not seen in the mouse. These limitations caution against facile comparisons between neurodevelopmental changes in mice carrying CNVs and related phenotypes in CNV-associated human cd-NDDs.

Multiple Genes, Multiple Steps: CNVs target fundamental developmental mechanisms

CNV mouse models– and single CNV gene mutations in other model organisms [21]– provide a unique opportunity to study in vivo the genetic disruption of developmental trajectories for circuits and behaviors that parallel deficits in ASD, Scz and other cd-NDDs. These step-by-step developmental processes begin with specifying neural stem cells and end with establishing synaptic connections. Each step relies upon multiple fundamental cell biological mechanisms common to all developing cells and tissues (Figure 2). The consequences of polygenic changes on these serial events in the intact developing organism are nearly impossible to analyze in human fetuses, newborns or—at this point—using human in vitro assays [22]. Within the whole developing organism, where brain development is influenced by fetal as well as maternal organ systems and signals [23, 24], polygenic circuit development likely reflects a “stairway” of cell biological steps taken sequentially to generate functional circuits (Figure 2). CNV polygenic circuit pathogenesis may reflect “stumbles” due to aberrant dosage-dependent modulation of multigene networks that facilitate each step. Serial stumbles may lead to a less adaptive “landing” rather than tumbling down or falling off the stairway—equivalent to more severe pathology or lethality—leading to cd-NDD circuit dysfunction and behavioral deficits.

Figure 2:

A developmental stairway for polygenic circuit assembly. Each step can be taken optimally (for example, by remaining in the green region of each individual tread) or sub-optimally (by stumbling into the red zones). Climbing each step in the optimal zone reflects efficient operation of fundamental cell biological mechanisms, indicated on the stairway wall. Each mechanism is regulated by complex multigene networks that engage large proportions of the genome. Reaching the landing: remaining in the optimal zone of each step with no stumbles, yields a maximally adaptive brain ready for lifelong healthy regulation of behaviors. If any—or many—steps and underlying cell biological mechanisms are stumbled upon, as shown on the stair wall for the developing brain in the context of 22q11 deletion, these cumulative stumbles result in a brain that is less well differentiated for maximally adaptive regulation of behavior across the lifespan, establishing fundamental risk for cd-NDD related behavioral deficits. (credits: Stair graphic via iStock; brain clip art via DALL-E 3).

Key steps upon which CNV-dependent, and possibly all, polygenic cd-NDD circuit development stumbles likely include: 1. Axial patterning and related signaling [25]. 2. Progenitor proliferation and genesis of post-mitotic neurons [20, 26]. 3. Migration of neuron subtypes [20, 26, 27]. 4. Axon and dendrite growth [28]. 5. Formation of synaptic connections [29]. Each step, respectively, relies upon discrete cell biological/molecular mechanisms for successful execution: 1. Differential regulation of progenitor cell gene expression by local signaling [30]. 2. Control of cell cycle and modes of division [31]. 3 & 4. Attractive and repulsive molecules that modulate cytoskeletal states [32]. 5. Adhesion, recognition and trophic molecules that regulate cell survival, validate axon/target matching, stabilize synapses or facilitate plasticity [29, 33]. Each cell biological “step” relies upon engaging substantial fractions of the genome to ensure diverse transcriptional states, cell-cell interactions and cellular maturation. Taking one step at a time, in sequence, one can better analyze cumulative, divergent circuit development and its relationship to cd-NDD-related behavioral deficits.

Up the Steps: CNV-induced stumbles, developmental mis-steps, and altered behaviors

Genetically valid CNV animal models provide a currently unique opportunity to integrate in vivo analysis of developmental mis-steps and their consequences for circuit function and behavior. Key developmental steps that establish circuits for two behaviors disrupted frequently in cd-NDDs: nursing and cognitive flexibility, have been characterized in mice that model 22q11.2 Deletion Syndrome (22q11DS), a cd-NDD-associated CNV disorder [34]. First, in 22q11DS mice, mild to severe nursing difficulties—perinatal dysphagia seen in multiple cd-NDDs including 22q11DS [35]—in 22q11DS mice reflect disrupted cranial nerve circuits that mediate this innate, uniquely mammalian behavior. Key aspects of nursing as well as organization and function of cranial nerves and brainstem nuclei critical for nursing are conserved between rodents and humans [35], facilitating careful comparisons. Second, cognitive flexibility likely reflects working memory, executive function and learning via frontal cortical circuits and their connections with other association cortices; therefore, cd-NDD associated deficits likely result from aberrant frontal cortico-cortical connectivity and function [36]. 22q11DS model mice have deficits in a reversal learning task that depends upon the integrity of frontal association cortico-cortical circuits [37]. Murine reversal learning tasks measure cognitive flexibility [38]; however, their limited, focal nature versus far more complex human behaviors [39] cautions against facile comparisons.

There are multiple “stumbles” in the LgDel 22q11DS mouse model over successive developmental steps that typically lead to optimally adaptive circuits for nursing or cognitive flexibility (Figure 2, 3). The first CNV-associated stumble for cranial nerve and association cortical circuit development occurs during hindbrain [40, 41] or forebrain [42, 43] patterning, respectively. Subsequently, CNV-disrupted circuit development trips upon regulation of stem/progenitor cell self-renewal or neurogenesis to generate appropriate numbers and types of nursing-related cranial nerve neurons [44, 45] or cortical projection neurons that make long distance association cortico-cortical connections essential for cognitive behaviors [37, 46]. The next CNV-dependent missteps for each circuit include disrupted cell migration and aberrant axon/dendrite growth [40, 41, 47–49]. The final CNV-associated stumble occurs on the last step: cell-cell interactions that regulate synaptic connectivity, resulting in divergent synapse numbers, structure, function and ultimately behavior [48, 49]. Thus, each 22q11 deletion-dependent developmental “stumble” likely reflects dosage-mediated destabilization of multigene networks that optimize ascending key developmental steps.

Figure 3:

CNVs regulate biological variation. Normal variation of wild type (WT) vs LgDel circuit development for suckling, feeding and swallowing facilitated by optimal cranial nerve circuit development (top) and cognitive behaviors facilitated by optimal association cortico-cortical circuit development (bottom). Top Row: Embryonic trigeminal ganglion (CNgV) gene expression (far left) for which the coefficient of variation of 18,000 significantly transcripts shared by both genotypes is greater for LgDel than WT (5 replicates/genotype, pooled samples, embryos, litters). Variable LgDel gene expression is accompanied by variably diminished LgDel trigeminal nerve (CN V) axon outgrowth (middle, top, white arrows), preceded by a change in retinoic acid (RA)-dependent patterning/signaling in the anterior hindbrain (black arrows), which generates a substantial portion of sensory neuron progenitors via neural crest, as well as progenitors for the trigeminal motor nucleus, whose axons also extend through CN V. Variation in LgDel CN V axon growth returns to WT by rescuing hindbrain patterning via rebalancing RA signaling genetically (middle, bottom). Increased variation as well as means for LgDel CNgV RA-regulated gene expression also diminish to WT levels due to genetic rescue (from [59, 60, 64]). Bottom Row: Significant variation and mean performance change for LgDel mice on a cognitive task (touch screen visual alternation; far left) that relies upon connectivity between frontal and lateral entorhinal association cortices. Performance variability and deficit are correlated (middle left) with variable frequency of Layer 2/3 Projection Neurons (PNs) in the LgDel medial frontal association cortex; Layer 2/3 PNs make most long-distance association cortico-cortical connections. Pharmacological rescue (with antioxidant n-acetyl cysteine, NAC) of Layer 2/3 PN differentiation (middle right) significantly diminishes variation and returns LgDel behavior to WT performance. In parallel, NAC restores 22q11 deletion-associated mitochondrial dysfunction (not shown) and LgDel layer 2/3 cytology back to WT levels while diminishing variability, including density of synaptic vesicles in presynaptic endings in layer 2/3 (from [66, 69]).

Any but the narrowest stairway allows for multiple paths across each tread. Similarly, as a developing brain traverses the cellular/molecular steps necessary for optimal circuit construction, there is likely to be substantial latitude for each step, reflected in significant variability that is nevertheless within a normal “adaptive” range (Figures 2, 3). Genetic rescue of specific “stumbles” that exceed adaptive variation in 22q11-deleted mouse models returns circuit differentiation toward a more optimal path by diminishing variation (Figure 3). Conversely, genetic or teratological enhancement intensifies specific 22q11-dependent missteps in a similar manner. In each instance, rescue or enhancement results from manipulating molecular mechanisms underlying fundamental cell biological processes for circuit construction [40–42, 47] rather than returning single 22q11 genes toward typical expression levels. Thus, adjusting retinoic acid synthesis, sonic hedgehog/BMP and Cxcr4 signaling, or mitochondrial efficiency moves 22q11 deletion phenotypes closer (rescue) or further (enhancement) from wild type by diminishing or amplifying phenotypic variation (Figure 3). Apparently, a polygenic CNV can disrupt not only the ascent, but variable trajectories over multiple cell biological steps necessary to construct circuits that mediate behaviors targeted by cd-NDD pathology.

Genes and Variation: Polygenic influence on adaptive variability of development

The types of mutations—inherited/de novo, truncation, insertion, missense, or somatic/mosaic single gene [50, 51]—associated with polygenic cd-NDD risk vary substantially, as do functions of genes on constantly growing lists of cd-NDD risk loci. cd-NDD-associated CNVs can impact expression and function of additional genes independently associated with Scz or ASD risk and their potential targets, including those involved in patterning and neurogenesis as well as synaptic structure, transmission and plasticity [52]. While functional variation of single cd-NDD-associated mutant genes or those within CNVs may seem like a “bug” that complicates analysis of cd-NDD genotype-to-phenotype relationships, it may actually be a “feature” of the genetic architecture necessary to ascend developmental steps summarized above that rely upon flexible cell biological mechanisms mediated by multigene networks (Figure 4). Multiple pathogenic variants—including multiple CNV genes—may alter resilience of multigene networks to restrict or expand a “dynamic range” that yields adaptive outcomes for multiple fundamental steps of circuit development. Polygenic disruption of these dynamic ranges beyond adaptive limits by functionally significant, but not lethal (“falling off the steps”), changes may cause cumulative stumbles that increase the probability of cd-NDD pathology.

Figure 4:

CNVs, multigene networks, developmental stumbles and the polygenic neural circuit. Multigene networks that ensure optimal execution of the fundamental cell biological mechanisms that underlie the multiple steps necessary to assemble a “polygenic circuit”. The successful completion of each step without stumbles, based upon optimal operation of each multigene network or optimal range of variability, leads to maximally adaptive neural circuits. Stumbles due to multiple single risk genes distributed throughout these networks, or altered dosage of CNV genes similarly distributed across networks, can divert circuit development to a distinct outcome that predisposes to cd-NDD pathology. (brain clip art via DALL-E 3)

Learning to Like Polygenicity: Adaptive and Maladaptive Polygenic Circuits and NDDs

cd-NDDs are polygenic because neural circuits are polygenic: cell biological mechanisms for step-by-step circuit construction rely upon multigene networks that engage substantial proportions of the genome. Inherent variation of these networks, due to sequence variants, epigenetic regulation, chromatin conformation changes, and additional transcriptional, translational, or posttranslational modulation, allows for dynamic regulation of cell states as neural progenitors and their progeny are specified, migrate, and differentiate to form circuits. A “polygenic circuit” is thus defined by multigene networks (Figure 4) that maximize cell biological flexibility for circuit construction as well as maintenance of synaptic connections, plasticity and behavior [45, 53–55]. CNV animal models provide a singular opportunity to better understand typical versus pathologic polygenic variation and circuit construction (see Figure 3). Using CNV models, compensatory capacities—or pathogenic thresholds—for CNV genes, as well as broader networks that include these genes, can be analyzed experimentally or computationally to identify functional redundancy, dosage compensation, genetic noise, and stochastic allelic expression [56–59]. Similar mechanisms may also modulate sensitivity to single risk variants distributed widely across the genome [53]. Resilience to modest change may ensure latitude for ascending developmental steps to build a typical neural circuit. Multiple pathogenic mutations, including multigene CNVs, may exceed the limits of this resilience. This genetic vulnerability may enhance the probability that polygenic circuit construction will stumble repeatedly (Figure 4), reaching a sub-optimal landing recognized as cd-NDD pathology.