Mechanisms of copy number variants in neuropsychiatric disorders: from genes to therapeutics

Abstract

Copy number variants (CNVs) are genomic imbalances strongly linked to the etiology of neuropsychiatric disorders such as schizophrenia and autism. By virtue of their large size, CNVs often contain many genes, providing a multi-genic view of disease processes that can be dissected in model systems. CNV research provides an important stepping stone towards understanding polygenic disease mechanisms – positioned between monogenic and polygenic risk models. In this review, we will outline hypothetical models for gene interactions occurring within CNVs and discuss different approaches used to study rodent and stem cell disease models. We highlight recent work showing that genetic and pharmacological strategies can be used to rescue important aspects of CNV pathophysiology, which often converge on synaptic pathways. We propose that the rescue approach provides a new path forward for more precise mechanistic understanding of these complex disorders, and a tangible route towards therapeutic development.

Introduction

Neuropsychiatric disorders (NPDs, here used to refer to both neurological and psychiatric disorders) are highly heritable, but their complex genomic architecture complicates a mechanistic understanding of disease [1,2]. Whilst genome-wide association studies have vastly expanded the repertoire of variants linked to NPDs (especially in schizophrenia), common variants have a small impact on disease risk (odds ratio 1–1.2) making them challenging to interpret for mechanistic studies [3]. By contrast, copy number variants (CNVs) can have a major impact on disease susceptibility (odds ratio 2–60) but are individually rare events (Figure 1) [4,5]. Allied with the identification of rare variants within single genes [6,7], rare variants such as CNVs can be viewed as bone fide causal events that precipitate the disease process, making them valuable candidates for genetic modelling.

Figure 1. Frequencies and odds ratios of disease-associated copy number variants.

A. Table showing properties of different types of genetic variants associated with neuropsychiatric disorders. The last column (No. of genes) estimates the minimum number of genes involved in causing illness. B. Bar chart of the allele frequencies in top neurodevelopmental CNVs involved in schizophrenia and autism, ranked from most to least common in schizophrenia. Data are from the largest scale studies available for each disorder (SZ = 21’094, ASD = 20’627, NDD = 29’085) [11–13]. Control frequencies were obtained from a large UK Biobank sample (n = 152’728) [97]. C. Estimated odds ratios for CNVs implicated in SZ and ASD based on frequencies in A. Please note that odds ratios can vary widely between studies depending on prevalence of CNVs in controls populations. Because 22q13 and 16p11.2 deletions are not significantly associated with SZ, the odds ratio of these CNVs in SZ was arbitrarily set to 1. Abbreviations: SNP, single nucleotide polymorphism; SNV, single nucleotide variant; GWAS, genome-wide association studies; SZ, schizophrenia; ASD, autism spectrum disorder; NDD, Neurodevelopmental disorders.

As the pace of genetic discoveries grows exponentially, a precise and thorough understanding of how disease-associated variants lead to phenotypic changes - and ultimately illness - is severely lagging behind. Much of this phenotyping relies on careful analysis of etiologically valid models based on the human disease processes [8,9]. Although rare variant models based on single genes provide an attractive route to study a more simplified genotype-phenotype relationship, they will ultimately never recapitulate the polygenic nature of NPDs that is inherent in their genetic basis. Conversely, creating disease models based on common variants is not currently imaginable due to their low penetrance and often unknown impact on surrounding genes. CNVs therefore occupy a useful space within disease modelling, where they are both highly penetrant but also multigenic, paving the way towards a more complete understanding of gene interactions and polygenic risk.

CNVs are defined as deletions or duplications of DNA sequence. Although CNVs are present in most genomes (~39% of individuals carry at least one [10]), an increased burden of large (>100kb) and rare (<1%) CNVs are observed in a wide array of NPDs [11–13]. Large CNVs are often flanked by low copy repeats (also called segmental duplications) that cause genomic instability and lead to recurrent deletions or duplication in the same chromosomal region [14]. An important feature of CNVs is their pleiotropy, meaning that the same CNV can confer risk of one or more disorders [15,16]. For example, most CNVs implicated in SZ have also been implicated in the etiology of ASD and ID, with only a few exceptions [4,5]. The mechanisms by which this occurs is largely unknown, although a number of hypotheses have been put forward including: the presence of other rare variants, polygenic risk, environmental influences and the inadequacies of current diagnostic criteria [17–21]. To date, ~70 CNVs have been implicated broadly in neurodevelopmental disorders [5,11], which together can explain ~15% of cases [22,23]. Thus, genetic screening programs have confirmed the notion that CNVs are individually rare but collectively common. Assuming neurodevelopmental disorders have a prevalence of 3%, CNV disorders could affect up to ~1.5 million individuals in US.

Mouse models have been created for a growing number of recurrent disease-associated CNVs, a necessary step towards understanding CNV biology [24]. In parallel, stem cell technologies are allowing us to model CNVs directly in human brain cells, either through genetic engineering induced pluripotent stem cells (iPSCs) or by inducing iPSCs from CNV carriers into relevant cell types [25]. With these model systems in place, the scene is set for major advances in mechanistic understanding of CNV disorders. In this review, we will discuss theoretical considerations in CNV-based models using exemplars from the broader CNV literature, and highlight recent work using genetic and pharmacological strategies to reveal novel mechanisms associated with some of the most highly studied multigenic CNVs.

Considerations for CNV modelling – gene dosage and functional interactions

Modelling neuropsychiatric disorders appears to be an insurmountable challenge. Nonetheless, creating models that are as close as possible to the human disease process is a primary goal for psychiatric research [8]. Mouse models and stem cells each have their strengths and weaknesses, which are now well recognized [26,27]. However, the validity of these model systems is also dependent on how they are designed and interpreted

To better understand mechanisms occurring within CNVs, a popular approach has been to create knockouts of individual genes contained in the CNV (single genes models) to dissect their contribution to the overall phenotype [28,29]. The hypothesis in this approach is that each gene will contribute to some or all of the CNV-associated phenotype. In this strategy, knockouts models of a particular CNV gene can be created and assessed for the phenotype of interest, either in isolation, or in direct comparison to the CNV model, which is a more thorough approach [30] . It is worth noting at this point that CNV loci are intrinsically dosage sensitive, so paying close attention to gene copy number is likely to be paramount [31–33]. Therefore, heterozygous deletions provide a more accurate model than full or conditional knockouts. Full knockouts may also have undesirable compensatory events, and could in some extreme examples, have an opposite phenotype to heterozygous cousins [34,35].

So far, single genes models have been useful in revealing previously unknown biology associated with CNV genes (such as a synaptic role for Kctd13, a gene in the 16p11.2 region [36]) and have also gone some way to decipher which genes may be critical within large CNV loci [36–38]. The strategy can be especially valuable if the CNV disorder is largely caused by deletion of one gene, which is often confirmed by refined genetic analysis [39]. A classic example of this might be SHANK3 deletions in Phelan-McDermid syndrome region [40], although a modulatory role for other genes in the 22q13 region is now well recognized [41,42].

It is becoming increasingly clear, however, that heterozygous single gene models can have significant limitations in the study of CNVs. The relative paucity of evidence for single CNV genes being independently associated to NPDs may support the fact that dysregulation of multiple genes is required to cause the underlying pathology [39,43]. Most notably, single gene models may be not able to recapitulate the complex interactions occurring within the CNV (Box 1). This issue was best exemplified in a drosophila model, where a pairwise knockdown of orthologs in the 16p11.2 region could either enhance or suppress the effect of single gene knockdown, on a developmental eye phenotype [44]. Examples like this offer important insight into what is likely to be occurring in mouse models and humans, and highlight the challenges in dissecting CNV mechanisms. Ultimately, loss or gain of a gene may not produce the same phenotypic outcome if it is alone vs in a larger CNV. This will be especially true if the gene in question operates in a shared pathway with other CNV genes. In the context of the hypothetical models in Box 1, the effect of a heterozygous deletion of gene A alone will have different outcome to a deletion of both A and C together. The complex (non-linear) interactions that occur between disease genes have been highlighted by numerous studies [44–49].In conclusion, single gene models remain useful to understand the fundamental biology of CNV genes, and to recapitulate aspects of CNV disorders (especially where single drivers are present), but may have limited utility in CNVs where non-linear effects are occurring.

Box 1. Hypothetical genetic interactions occurring with CNVs.

To better understand mechanisms of disease within CNVs, it is necessary to understand the potential ways genes (and encoded proteins) can interact with one another. As disease-associated CNVs are often multigenic, and the altered proteins they encode can interact, an appreciation of the potential effects of protein interactions on phenotype is warranted. As well as these pathway interactions, it worth noting other types of genetic interactions exists, discussed elsewhere [43,98]. A few hypothetical scenarios can be imagined (A-D). In these examples, four genes contribute to the same phenotype but their encoded proteins interact in different ways, creating either linear or non-linear outcomes. The coloured dial on the top right of each panel indicates the magnitude of effect on the phenotype. In the simplest version of a gene interaction (A), gene products A and B are part of a similar pathway and contribute to the phenotype in a linear fashion. Increasing A alone, or B alone, will have a similar effect, and increasing both will double the effect on the phenotype. In the second scenario (B), gene products A and C interact in a synergistic way (non-linear), such that the effect on the phenotype is amplified: increasing both A and C will more than double the effect on the phenotype. In scenario three (C), gene product D will inhibit the function of protein A. Thus, increasing both proteins will cancel the effect of protein A, resulting in a minimal change in the phenotype. Finally, scenario four (D) explores the effects of several interactions occurring at the same time. For example, if synergistic and inhibitory interactions occur simultaneously, the effect on the phenotype may be complex and will depend on the weight of individual effects on the overall outcome. To put these examples in a more physiological context, one could imagine gene products A-D forming part of the AMPA receptor trafficking pathway and regulating postsynaptic activity (E). Here, protein A promotes exocytosis of the receptor and increases the number of receptors at the membrane. Protein B promotes trafficking from endosomes to exocytotic vesicles, increasing the pool of receptors that can be exocytosed. Protein D inhibits receptor exocytosis, reducing the rate of receptor insertion into the membrane. Protein C phosphorylates the receptor, promoting both forward trafficking and channel activity once at the membrane (phosphorylation can synergize with membrane insertion (A) to amplify synaptic strength).

With the advent of Cre/loxP based chromosome engineering technologies, the creation of model systems closely mimicking full human CNVs was made possible [50]. This allowed the study of a more faithful disease model and opened the door for a number of CNV mouse models to be generated, including those for: 1q21 deletion [51], 3q29 deletion [52,53], 7q11.23 deletion [54,55], 15q11–13 deletions [56,57] and duplications [58], 15q13.3 deletions [59,60], 16p11.2 deletions and duplications [61–63], 22q11.2 deletions [64–66]. For a more detailed review of the phenotypes in these models please see [24]. These models now allow the investigation of genotype-phenotype relationships directly within a native CNV environment, which provides a more contextualized version of their function in disease. This subtle difference could be critical to understanding pathological mechanisms and find novels routes for therapeutic development.

Genetic rescue strategies

To uncover genotype-phenotype relationships within CNVs, a complementary approach to single gene modelling has recently been employed, which we here call the genetic rescue strategy. The genetic rescue strategy involves modulating the expression of individual genes within the complete CNV model, to assess their impact on a given phenotype. For instance, increasing the expression level of a gene within a multigenic deletion model, or reducing the expression of a gene within a duplication model. The genetic rescue approach has several advantages over the single gene model discussed above: First, it considers the function of a gene within the context of the CNV environment, and honors the gene interactions that are occurring. Second, it has strong translational potential, as once identified, the causal gene or pathway can be targeted using drugs or biologics. Modulating gene expression for this approach can be done in several ways, including gene knockdown, overexpression or genetic crosses. In addition, improvements in epigenome editing could make it possible to selectively tune expression of CNV genes using a modified CRISPR-cas9, as has already been demonstrated in neurodevelopmental imprinting disorders [67–70]. A few key examples of genetic rescue strategies are detailed below and summarized in Table 1.

Table 1.

Genetic and pharmacological rescue strategies in copy number variant models.

CNV	Model	Model reference	Genetic rescue	Pharmacological rescue	Link to CNV genes	System level	Corrected phenotype	Rescue reference
7q11.23 deletion	Mouse	[55]	Gtf2i	NA	direct	genetic	Motor coordination, sociability, and anxiety	[99]
15q11.2 duplication	Mouse	[58]	Nedcin	NA	direct	genetic	Dendritic spine dynamics, cortical excitatoryinhibitory balance, locomotion, anxiety, social behaviour, spatial learning	[72]
15q13.3 deletion	Mouse cortical neurons	[59]	Otud7a	NA	direct	genetic	Cortical neuron arborization and mushroom spine density	[73,75]
	Human induced cortical neurons	[73]	OTUD7A	NA	direct	genetic	Cortical neuron arborization	[73]
16p11.2 deletion	Mouse	[61]	Taok2	NA	direct	genetic	Cortical neuron migration	[78]
16p11.2 duplication	Mouse	[61]	Prrt2	NA	direct	genetic	Cortical hypersynchrony, seizure susceptibility, social behaviour	[100]
22q11.2 deletion	Human cortical organoids	[81]	DGCR8	NA	direct	genetic	Depolarization induced-calcium influx, resting membrane potential	[81]
22q13 deletion	Human induced cortical neurons	[101]	SHANK3	NA	direct	genetic	Excitatory synaptic transmission	[101]
1q21.1 deletion	Human induced cortical neurons	[102]	NA	verapamil (VGCC blocker)	indirect	molecular	Calcium event rate	[102]
3q29 deletion	Mouse	[52]	NA	Oxytocin	indirect	molecular	Social behaviour	[103]
	Mouse	[52]	NA	Rispiridone	indirect	molecular	Startle amplitude, pre-pulse inhibition	[52]
7q11.23 deletion	Mouse	[55]	NA	JZL184 (endocannabinoid pathway)	indirect	molecular	Short-term memory and cardiac phenotypes	[104]
	Mouse	[55]	NA	Epigallocatechin-3-gallate (green tea)	indirect	molecular	Social and cognitive phenotypes, cardiovascular function	[105]
	Mouse	[55]	NA	Verapamil (VGCC blocker) and Curcumin (turmeric)	indirect	molecular	Social behaviour and motor coordination	[106]
15q11.2 duplication	Mouse	[58]	NA	Fluoxetine (serotonin reuptake inhibitor)	indirect	molecular	GABAergic synapse function, ultrasonic vocalisations and social behaviour	[107]
	Mouse	[58]	NA	Oxytocin or 8OH-DPAT (5-HT1A receptor agonist)	indirect	molecular	Social behaviour	[108]
15q13.3 deletion	Mouse	[59]	NA	Lu AF58801 (positive allosteric modulator of nAChA7R)	direct	molecular	Resting-state fMRI functional connectivity patterns	[109]
16p11.2 deletion	Mouse	[63]	NA	Fasudil (RhoA inhibitor)	downstream pathway	molecular	Novel object recognition	[30]
	Mouse	[61]	NA	Cell-permeable peptide (Ras-ERK pathway inhibitor)	upstream target	molecular	Cortical cytoarchitechture, ventral hippocampus and lateral septum volume, hyperactivity, olfaction, maternal behaviour, impaired contextual memory	[87]
	Human induced dopaminergic neurons	[110]	NA	Rhosin (RhoA inhibitor)	downstream pathway	molecular	Neuron soma size, network hyperactivity	[91]
	Human cortical organoids	[92]	NA	Rhosin (RhoA inhibitor)	downstream pathway	molecular	Cortical neuron migration	[92]
	Mouse	[61]	NA	CTEP (negative allosteric modulator of mGluR5)	indirect	molecular	Contextual fear conditioning	[111]
	Mouse	[61]	NA	R-baclofen (GABAB receptor agonist)	indirect	molecular	Cognitive and social behaviours	[112]
16p11.2 duplication	Mouse cortical neurons	[61]	NA	U0126 (MEK-ERK pathway inhibitor)	upstream target	molecular	Cortical neuron arborization	[82]
	Mouse	[61]	NA	LMK235 (HDAC5 inhibitor)	indirect	molecular	GABAergic synapse function and social behaviour	[113]
	Human cortical organoids	[92]	NA	Rhosin (RhoA inhibitor)	downstream pathway	molecular	Cortical neuron migration	[92]
22q11.2 deletion	Mouse	[64]	NA	Neuregulin 1 peptide (interneuron excitability)	indirect	cellular/circuit	Hippocampal CA1 dynamics and contextual fear conditioning	[114]
	Mouse	[65]		Spadin (TREK-1 antagonist)	indirect	circuit	Hippocampal CA2 firing and social memory	[115]
	Mouse	[64]		eticlopride (D2R antagonist)	indirect	cellular/circuit	Gamma activity, PV expression, contextual fear conditioning, social and object memory	[116]
	Human cortical organoids	[81]		Raclopride (D2/D3 receptor antagonist)	indirect	molecular	Depolarization induced-calcium influx	[81]
22q13 deletion	Human induced cortical neurons	[101]		IGF1	indirect	molecular	Excitatory synapse number and transmission	[101]

Abbreviations: VGCC, voltage-gated calcium channel

Duplications of the 15q11-q13 region (Dup15q syndrome, OMIM: 608636) are associated with autism and schizophrenia (Figure 1). A recent study looked at genes within the 15q locus capable of regulating synapse dynamics and behavioural phenotypes in a mouse model of Dup15q syndrome (15q dup) [71,72]. First, four genes of interest were overexpressed in utero for three weeks to determine their effects on spine turnover. One gene (Ndn) significantly increased spine formation, causing a net increase in spine density. To determine the effects of Ndn within the 15q dup mouse model, the gene was excised from the 6.3 Mb duplicated region using CRISPR-Cas9 editing. This elegant strategy resulted in the correct copy number of Ndn (two copies) whilst all other genes in the region remained duplicated. Correcting the genomic copy number of Ndn resulted in a normalization of the spine turnover rates and a restoration of cortical excitatory/inhibitory imbalance in 15q dup mice. These cellular improvements were accompanied by a full recovery of abnormal behaviours found in 15q dup mice, including high anxiety, impaired social interaction and reversal learning.

The 15q13.3 CNV is a 1.53 Mb region on chromosome that occurs between breakpoint (BP)4 and BP5. Deletions in this region cause the 15q13.3 microdeletion syndrome (OMIM: 612001). Single gene modelling in mice as well as human genetic studies have already pointed to OTUD7A as a likely driver gene in the region. In 15q13 mice, cortical neurons have a reduced number of mushroom spines, simplified dendritic arbors and reduced network activity [73]. Similarly, induced human neurons generated from probands with the 15q13.3 microdeletion have lower synapse density, dendritic complexity and network activity [73,74]. To determine the role of OTUD7A in the microdeletion, the protein was reintroduced by ectopic expression, resulting in a rescue of reduced dendritic complexity in human neurons, and a normalization of lower dendritic arbors and mushroom spine density in mouse neurons [73,75]. Overexpression of three other proteins in the 15q13.3 region had no effect on spine density, including the candidate gene CHRNA7, indication a critical role for OTUD7A in synaptic phenotypes.

CNVs in the proximal 16p11.2 (BP4-BP5) are frequently observed in NPDs (~1%). Deletion of the 16p11.2 region is associated with ID and ASD, whilst duplications have a much wider phenotypic spectrum including schizophrenia [76]. Two studies have used the genetic rescue strategy to dissect the role 16p11.2 genes in the context of CNV mouse models, including one of our own.

In humans, the 16p11.2 CNV has a well-established link with brain size, with deletions and duplication being associated to macrocephaly and microcephaly respectively [77]. A potential mechanism for changes to cortical structure is altered neuronal migration. Thus, a study looked at neuronal migration defects in 16p del mice in the embryonic cortex [78]. Cortical neurons were found to have reduced migration speed and altered microtubule dynamics, resulting in fewer neurons reaching the cortical plate from the ventricular zone between E15 and E18. Impressively, the defective migration phenotype was completely reversed by ectopic expression of TAOK2 (alpha isoform). As precise neocortical development is required for higher cognitive and associative functions, reduced Taok2 expression is likely to have an essential role in carriers of the 16p11.2 deletion [78].

The 16p11.2 duplication is associated with a myriad of disorders including autism, schizophrenia, epilepsy. We recently showed using quantitative proteomic analysis that 16p dup mice have disrupted protein networks associated with various NPDs, including epilepsy [79]. Consistent with this, analysis of neuronal circuits showed that hypersynchrony was a critical phenotype in 16p dup mice. To determine which genes in the 16p11.2 region may be responsible for the proteomic and circuit dysfunction, a bioinformatic approach was used that identified PRRT2, a regulator of presynaptic vesicle release [80], as a leading candidate. By screening for PRRT2 interacting proteins, we found over 70% of these were disrupted in 16p dup mice. The functional role of PRRT2 with the full CNV was assessed by crossing 16p dup mice with Prrt2^+/− mice to obtain 16p dup mice with a Prrt2 copy number identical to wild type. This strategy ensured that wild type levels of PRRT2 were obtained throughout development, leaving other 16p genes duplicated. We found that genetic correction for Prrt2 dosage rescued several core phenotypes in 16p dup mice, including hypersynchrony, increased seizure susceptibility, and impairments in social behaviour. This study demonstrated that Prrt2 is an important driver gene, regulating circuit and behavioural phenotypes in mice, and that disrupted cortical synchrony could underlie a range of neuropsychiatric symptoms in 16p11.2 duplication carriers.

The 22q11.2 deletion spans a region of 2–3 Mb, encompassing ~40 genes. Because it is a highly penetrant risk factor for schizophrenia, it has been used to model schizophrenia in a number of cellular and animal models. Although rodent models of the 22q11.2 deletion syndrome (22q11DS) have been studied for some time, little is known about how the human neurons are affected. In one study, stem cells from 15 individuals with 22q11DS were differentiated into cerebral cortical organoids [81]. The 22q11DS organoids had transcriptional profiles that were enriched for SZ GWAS loci and presented with deficits in neuronal excitability and calcium signalling. Interestingly, deleting one copy of DGCR8 recapitulated the neuronal phenotypes from the 22q11DS organoid, indicating it could have a primary role in the overall phenotype. Overexpressing DGCR8 in 22qDS organoids resulted in a partial rescue of calcium signalling defects and a full rescue of resting membrane potential in the residing neurons. These studies indicate that miRNA processing protein DGCR8 is likely a critical gene in the 22q11.2 region, and could regulate neuronal function in the brain of 22q11.2 carriers.

Therapeutic strategies in CNV disorders

In addition to using genetic tools to rescue CNV models, pharmacological strategies have already gained momentum, and could provide a further framework to understand pathogenesis and lead to treatments. The pharmacological approach can involve direct modulation of CNV genes (direct targeting), or targeting of downstream effectors and phenotypes linked to the CNV (indirect targeting). Indirect targeting could include signalling pathways or cellular, synaptic and circuit aberrations, providing multiple potential avenues towards disease reversal. Because CNV models are often associated with a large number differentially expressed genes or proteins, it can be challenging to find appropriate targets. Here, network biology can be a useful tool to identify central nodes capable of disease reversal that could have an outsized impact on critical phenotypes [79,82]. An example of the direct and indirect approach is provided for the 16p11.2 CNV below. A more comprehensive summary is provided in Table 1.

The 16p11.2 CNV region contains ERK1, encoded by the gene MAPK3. The ERK signalling pathway is essential for diverse aspects of neuronal development, function, and plasticity [83,84]. Multiple drugs have been designed to modulated the ERK pathway including some used in oncology [85]. We showed that ERK1 and phospho-ERK are increased in the 16p11.2 duplication and ERK is a central node within a network of psychiatric CNV genes [82]. Pharmacological inhibition of ERK1 using U1026 was able to reverse excessive dendritic branching associated with 16p11.2 duplication neurons. Interesting, ERK signalling is also altered (increased pERK1/ERK1 ratio) in the 16p11.2 deletion model, during a critical window of cortical neurogenesis (E14.5) [86]. Treating 16p del mice with an ERK pathway inhibitor (modulating upstream MAPK) during this period reverses abnormal cytoarchitecture and, at least in part, reverses behavioural phenotypes including hyperactivity, memory, olfaction, and maternal behaviour [87]. These studies position the ERK pathway as potentially useful target in 16p11.2 CNV disorders.

Another therapeutic strategy in 16p11.2 CNV models has been pharmacological inhibition of RhoA. RhoA is part of the Rho GTPase family and an important regulator of actin dynamics, neuron growth and synapse plasticity [88]. Although RhoA is not encoded within the 16p11.2 region, it is a downstream effector of several gene products from the 16p11.2 region including ERK1 and TAOK2, and its expression is regulated by KCTD13 [36,89,90]. Hyper-activation of the RhoA pathway has been linked in various ways to 16p del models including increased phosphorylation of myosin light chain (p-MLC), increased phospho-RhoA and increased RhoA protein levels. In mice, chronic administration of fasudil, a RhoA pathway inhibitor, for six weeks restored object recognition memory and p-MLC levels but had no effect on object location memory or hyper locomotion [30]. In human iPSC-derived dopaminergic neurons, long term treatment (~30 days) with the selective RhoA-inhibitor Rhosin, rescued RhoA protein levels, soma area, and hyperactive neuron networks associated with the 16p del [91]. In cerebral organoids models, Rhosin treatment normalized migration deficits in 16p del and 16p dup carriers, which both exhibited increased phospho-RhoA levels [92]. Although the mechanisms affecting the RhoA pathway in 16p11.2 CNV models are complex, it is an excellent example of how gene products within and outside the CNV can impact a shared process. Further exploration will be required to fully dissect the role of each gene in these phenotypes, but several distinct models now support that targeting this pathway could be a promising avenue drug development avenue in 16p11.2 CNV carriers.

In addition to these specific examples, genetic analysis has shown that CNVs are generally enriched for synaptic genes involved in both excitatory and/or inhibitory signalling [93,94]. In agreement with these findings, analysis of CNV models have revealed that synaptic properties are often impacted (see “Genetic rescue strategies”). Thus, modulating synapse structure and function could be a major source of novel therapeutics. To showcase synaptic proteins associated with CNV loci, we generated a list of all genes within the CNVs presented in Figure 1B and determined which ones were synaptic using SynGO [95]. Synaptic CNV genes were then further divided into pre or post synaptic subsets. Notable groups of proteins include regulators of synaptic vesicle release (PRRT2, DOC2A, STX1A, SNAP29, VAMP3, RIMBP3), adhesion molecules (NRXN1, CLSTN1), postsynaptic scaffolds (SHANK3, DLG1), ionotropic (GABRA5, GABRD, CHRNA7, GSG1L) and metabotropic (GABRB3) receptors subunits, voltage-gated ion channel subunits (KCNB2, CACNG3), and regulators of intracellular signalling (PAK2, MTOR, PRKCZ, TAOK2, CYFIP1). We constructed protein-protein interaction networks from these subsets to highlight functional relationships and network connectivity between these CNV gene products and display these in Figure 2. CNV genes were then annotated according to their potential for being drug targets (druggability) as defined by a previous study [96].

Figure 2. Potential synaptic therapeutics in CNV disorders.

Schematic showing synaptic CNV genes organized in pre and post synaptic network. Coloured nodes are CNVs genes and grey nodes are connecting nodes retrieved from the PPI database. Each nodes is colour scale according to druggability (Briefly, Tier 1 – Targets with approved small molecules and biotherapeutic drugs as well as clinical-phase drug candidates, Tier 2 - binding partners as well as those with ≥50% identity to Tier 1, Tier 3 - secreted or extracellular proteins, proteins with more distant similarity, and members of key druggable gene families not already included). For more details on classification please see [96]. Synaptic networks were obtained by compiling a list of genes within top recurrent CNVs described in Figure 1B (n = 483). SynGO was used to annotate presynaptic (n=16) and post synaptic (n = 28) genes. PPI networks were constructed based on pre- and post-synaptic gene lists using GeneMania plugin with Cytoscape.

Conclusions

Understanding CNV disorders is a major challenge for both clinicians and neurobiologists due to their phenotypic and genetic complexity. However, CNV research may represent a microcosm of psychiatry, where complexity is clearly present, but still somewhat limited compared to NPDs at large. Deconstructing CNV mechanisms therefore presents a valuable opportunity to understand multigenic disorders before considering the full extent of polygenicity associated with mental illnesses. Model systems have been invaluable for discovering tractable mechanisms and possible therapeutics targets but integration of multiple approaches and models will be required achieve a complete understanding of how CNVs cause the myriad of symptoms they are associated with. Genetic rescue strategies should provide a unique window from which we can learn and design of novel therapeutics. Once critical genes are identified, novel drugs can aim to mimic rescue effects observed in gene restoration approaches. The focus will likely be on therapeutics which address unmet needs in NPDs, such as cognitive deficits and social impairments. Here, synaptic modulators are likely to play a key role and could pave the way for improved treatment of CNV disorders.