Editorial overview: Rare CNV disorders and neuropsychiatric phenotypes: opportunities, challenges, solutions

Introduction

In this issue, our outstanding colleagues review the recent exponential growth in our understanding of copy number variants (CNV) in neuropsychiatric disorders. The impact of DNA dosage changes on phenotypic traits was first understood, and leveraged for genetic mapping, by Drosophila geneticists nearly a century ago[1]. Evidence for DNA dosage changes influencing human phenotypes came several decades later with the observation that Down Syndrome is caused by an extra copy of chromosome 21 [2,3]. This was followed by discoveries that dosage changes far smaller than the scale of a whole chromosome, limited to smaller sets of genes, could also cause clinical syndromes. These hallmark syndromes have set the stage for our current understanding of CNV disorders, and in this issue, we review several classic syndromes, including Williams syndrome or 7q11.23 deletion, reviewed by Osbourne and Mervis; 22q11 deletion syndrome reviewed by Gur et al; and 16p11.2 deletion/duplication, reviewed by Chung and colleagues. DNA dosage changes can also affect single genes; the impact of NRXN1 dosage changes is reviewed by Fuccillo and Pak.

Conventional methods (used prior to ~2005) to detect DNA dosage changes, or CNVs, suffered from ascertainment problems. First, since assaying even one person for the presence of important CNVs was laborious, time-consuming, and expensive, efforts to find CNVs were confined to individuals with significant pathology. Second, the initial “microscopic” CNV screening technologies (e.g., karyotyping using G-banding) could only detect large CNVs (several megabases or larger). The net result was that DNA dosage changes were exclusively seen alongside relatively severe phenotypes, leading to the assumption that all copy number changes were pathological.

The advent of microarray technology allowed for inexpensive and rapid evaluation of DNA dosage changes at scale and, when applied to large case-control samples, a new universe of DNA dosage changes was revealed[4–6]. Microarrays were also deployed for clinical testing and, because these could rapidly scan the entire genome at once, they replaced laborious cytogenetic techniques or single locus FISH-based tests. Early on, several labs developed significant proficiency with microarrays and became clearinghouse sites for clinical testing, processing thousands of microarrays in a short span of time. With these large banks of clinical data, CNVs that were rare but recurrent – that is, where the same de novo mutation occurs in multiple individuals - could be identified, and new syndromes were recognized, often with prominent multisystem features and usually including neurodevelopmental disorders[7].

In the research space, parallel efforts were underway to detect CNVs in large cohorts of individuals with a specific phenotype, such as autism spectrum disorder (ASD) or schizophrenia (SZ). This led to a period of marked productivity with the identification of robust CNV associations. Somewhat surprisingly, these findings lacked specificity – many loci were found to be associated with multiple disorders. There was also overlap between clinical and research efforts: CNVs that were identified as new genetic syndromes with a set of medical, physical and neurodevelopmental sequelae, were also being identified as enriched in cohorts selected for a specific manifestation. This phenotypic heterogeneity, and the implications for research studies and clinical practice, are discussed in detail by Moreno De Luca and Martin. There are now ~30 CNVs that increase risk for a neurodevelopmental or psychiatric disorder, with phenotypes that have onset across the lifespan: manifestations are seen early in development as pervasive developmental delay, intellectual disability, autism, and epilepsy; childhood onset, such as anxiety disorders and ADHD; and adult onset including schizophrenia and schizoaffective disorder) [Figure 1].

Figure 1.

This ‘CIRCOS’ plot depicts the locations of copy number variants (CNVs) known associated with intellectual disability, autism, and schizophrenia. The genome is shown as a circle, from 1ptel to Yqtel in a clockwise direction from the top center. From outside in, the tracks show: chromosome label, giemsa banding, gene density per mb, and then the positions and sizes of CNVs associated with intellectual disability, autism, and schizophrenia (CNV deletions are red and duplications blue). The innermost track (gold) shows CNVw implicated for more than one condition. Chromosomes without a known CNV are omitted. hg19 coordinates. Data sources are PMID 24150940 24352232 25217958 27869829.

Phenotypes associated with CNVs: Clinical Implications

CNVs confer high risk for multiple developmental brain disorders[8–12]; Rees and Kirov review the burden of neurodevelopmental and psychiatric illness associated with CNV. For individual patients, comorbidity is the rule rather than the exception. These data suggest the compelling hypothesis that phenotypes including cognitive disability, autism spectrum disorder (ASD), ADHD, and schizophrenia may exist along a spectrum with underlying etiologic similarity[13], an idea also supported by the study of common variants[14–16]. This is contrary to some expectations for genetic studies; it was initially anticipated that genetics would diminish heterogeneity by enabling more accurate classification. For example, if we identified a genetic variant for schizophrenia, the expectation was that individuals with that genetic “type” of schizophrenia would resemble one another with respect to onset, course, treatment response, and outcome. Instead, the accumulated data indicate even greater heterogeneity. For example, the clinical phenotypes of individuals carrying 22q11.2 deletions are complex, and vary in severity (none, mild, moderate, or severe) across multiple clinical domains (cognitive, mood, anxiety, and psychosis). There certainly is not single type of schizophrenia in people carrying 22q11.2 deletions, and many in fact never develop schizophrenia.

The suggestion from CNV research that ID, ASD, and schizophrenia may all stem from common etiology has now been observed and reinforced across multiple CNV loci. These observations have created conflict across disciplines. In clinical psychiatry, there has been intentional and deliberate separation of phenotypes. This separation, standardized by the DSM and ICD criteria, imposed rigor and revolutionized the practice of psychiatry, benefitting patients and medical practitioners alike. Results from genetic studies, which promote an overlapping etiology among ID, ASD, ADHD, anxiety disorders, and schizophrenia are generally inconsistent with current clinical nosologies. In the tension between “lumping” and “splitting” psychiatric diagnostic classification, the clinical nosologies split and the genetic data suggest lumping. This apparent conflict requires resolution and should be addressed by ongoing research efforts. Vorstman and Scherer tackle this conflict and propose an integrated clinical framework to accommodate psychiatric and genetic diagnoses simultaneously.

In developed countries, marked developmental delay or neurodevelopmental illness increasingly trigger genetic testing, for which genome-wide CNV analysis is fundamental. To be consistent, later-onset psychiatric disorders with neurodevelopmental roots should also lead to genetic testing. Genetic testing is already recommended as the standard of care for ID and ASD[17] but is often not done. Barriers to testing should be identified and removed, and the standard of care should be broadened. Finucane, Ledbetter, and Vorstman advance a strong case for universal genetic testing for neurodevelopmental psychiatric disorders, particularly early-onset and severely impairing conditions (moderate/severe intellectual disability and autism) as well as chronic psychotic disorders in adults.

The benefits of expanded genetic testing to patients are many. First, in psychiatry, the stigma associated with a diagnosis stubbornly remains. A genetic diagnosis may legitimize this as biological, “real” illness. For patients, this realization may abolish internal barriers to seeking medical care. Second, CNV disorders are highly heterogeneous. In some individuals with a CNV, cognitive ability may be preserved, and a more subtle phenotype like ADHD may be first sign. With CNV diagnosis, multiple subtle signs and symptoms exhibited throughout the lifespan are likely to coalesce (in retrospect) as an obvious syndromic phenotypes[18]. Ultimately, this will lead to better care for patients. Third, most large CNVs are multisystem disorders with increased risk of cardiac, neurological, endocrine, renal, hematological, and digestive complications – comprehensive medical care requires awareness and appropriate management of known additional vulnerabilities. Fourth, personalized medicine continues to expand; mutation-specific treatments already exist for some disorders[19]. Chawner, Watson, and Owen offer preliminary recommendations for multi-system clinical evaluations upon CNV diagnosis, highlighting how genetic testing can lead to improved clinical care.

Research frontiers are broadened by CNV studies

The study of CNVs offer unique opportunities for understanding multiple aspects of neurodevelopmental and psychiatric disorders. Many CNV are associated with large effect sizes for multiple disorders. The magnitude of these effect sizes suggests that the biological impact of a CNV is substantial, particularly on molecular and cellular domains and, once identified, can offer clues toward underlying etiologic mechanisms. These insights are particularly important for idiopathic psychiatric disorders. Indeed, as reviewed by Maury and Walsh, at least 0.3–0.5% of idiopathic illness may be explained by somatic mutations (CNVs that occur post-zygotically).

Although the phenotypic heterogeneity associated with CNVs has caused consternation in the clinical realm, in the research domain CNVs are an unprecedented opportunity to understand the biological linkage among these phenotypes. Risk models to explain this heterogeneity have been previously articulated[13]. In the “specific risk” model, CNVs contribute directly to risk of neuropsychiatric illness. In a “generalized risk” model, CNVs impair brain function in a non-specific (and possibly convergent) way, with additional risk factors (e.g. polygenic risk score (PRS)) determining specific phenotypes. Importantly, CNVs offer a system in which to test these models.

Recent data support the generalized risk model, with both common and rare genetic variation forming “second hits” that impact phenotypic expression and severity[20–22]. Early studies used parental phenotypes as a proxy for genetic background, and found that even when the offspring has a de novo genetic variant, there is a significant influence of parental phenotype on child outcomes across multiple phenotypic domains[23]. More recently, common genetic variants have been identified and quantified as contributors to polygenic risk and these also explain a portion of phenotypic heterogeneity among rare genetic disorders[24]. For example, among individuals with 22q11.2 deletion, there is a significant contribution of the polygenic risk scores (PRS) derived from common variants to schizophrenia risk[25] [22]. These data extend to other phenotypes as well. For example, rare de novo variants and polygenic risk act additively to promote risk for autism[26]. These findings extend to rare variants, with ample evidence for a two-hit model where in the presence of a known rare genetic variant, a second rare variant may influence phenotypic severity[27–30]. These exciting data showcase the progress that has already been made though the study of CNV.

What advantage do CNV offer over single gene disorders? Single gene disorders are attractive for study: they reduce complexity; may localize to a single circuit, cell, or developmental time point; and offer single molecular targets which may speed development of therapeutics. However, often when a single gene is implicated, although individuals have mutations in the same gene, the mutations themselves are not identical – thus each mutation may have a distinct impact on biology. Two loss of function mutations in different gene domains, which presumably have a similar mechanism of action, may differentially impact splice variants and thus have different biological consequences[31]. When studying heterogeneity of phenotypes, it is impossible to deconvolute genotype-phenotype correlations when every individual has a private mutation. In contrast, the breakpoints of CNV mutations are often identical, thus each individual has the same or a highly similar mutation.

For many CNV loci, it is likely that more than one gene in the interval contributes to risk for neuropsychiatric illness[32]. There is also a possibility that in addition to the genes themselves, regulatory elements exist within the CNV interval that control expression of key genes outside the interval. Thus, CNVs occupy a space between single genes and polygenic risk, where thousands of loci may contribute to risk. CNV offer a constrained disease-relevant system to tackle multi-gene interactions; contributing elements in a CNV interval number in the single digits rather than in the thousands. By studying CNV, we can develop models, learn from these “simplified” systems, and develop algorithms to identify additional interactions, where the interacting elements are not by chance next to one another.

While single genes offer opportunities to understand molecular mechanisms of neuropsychiatric illness, these are not “simple.” For example, SETD1A, CHD8, FMR1, and MECP2 are all are single genes but are known to regulate or interact with hundreds of downstream targets. Thus, the biological impact of even a single gene may be amplified, with enormous complexity. Intellectually we are driven to simplify to “the gene” or “the pathway” – but the biology of neuropsychiatric disorders is highly unlikely to distill down to a single key element. It is probably instead due to a complex, layered process with multiple insults across cells, circuits, and developmental time. Here is where CNV may offer perhaps their most salient advantage: they force us to grapple with a complex system.

Tools needed to study CNV

From a basic science perspective, human cellular models are key resources for understanding the human-specific biology of CNVs, and the importance of the combination of patient-derived samples with iPSC technology cannot be overstated[33,34]. An attractive approach is the combination of iPSC-derived neurons and recent development of high throughput electrophysiological measurements of individually (or a few neurons at a time) grown neurons which allows researchers to study electrophysiological and synaptic properties in great detail. Another important approach holding great promise is the differentiation into human 2D and 3D organoids which allows the observation and biological dissection of living cell and tissues that were previously unattainable. Advantages include the high relevance to human biology, and the ability to observe in real time the formation of cellular circuits in a human “mini-brain.” For the production of tissue, there are still challenges to overcome: difficulty to standardize organoid production leads to variability in the nervous tissue produced, “brain in a jar” – we do not know how circuit maturation is affected by a lack of salient sensory information, and organoids age in real time, and must be maintained in expensive culture media for possibly several years. If one is to systematically investigate changes in cellular connections at timepoints that are relevant to human disease like schizophrenia one would have to grow organoids for decades and maybe even provide relevant activity patterns and hormones. Furthermore, we cannot observe behavior or complex phenotypes. While there are increasing observations of complex circuitry in organoids, complexity is necessarily reduced.

Animal models for the study of CNV-associated neuropsychiatric phenotypes have inspired controversy [35,36]. Advantages of mouse models include analogous neurodevelopment that in key aspects parallels humans; and faster aging than humans such that changes across the lifespan can be observed as well as the ability to measure subtle behavioral changes relevant to learning, memory, and sociability. Importantly, the first two means that we can study functional and molecular impact of CNV mutations on brain development in intact tissue into adolescence and adulthood. In this issue, Hyman thoughtfully reviews this long-standing controversy and provides guidelines for appropriate use of mouse models to study CNV disorders. While mouse models can shed light on CNV biology, a historical problem has been the lack of translational findings. Many treatments that appear to ameliorate disease phenotypes in mouse models have not been effective in humans. Alternative models like zebrafish offer advantages such as rapid generation time, large clutch sizes, and a transparent organism that can allow for high-throughput phenotyping. Even behavioral assays can be done. The disadvantage to zebrafish is that the organism is even further removed from human biology. Katsanis et al. review the benefits of non-rodent models, including zebrafish, to dissect CNV disorders. Non-human primate models have faithful modeling of complex primate neurodevelopment which is closer to humans. However, disadvantages include ethical considerations, expense, and limited availability of primate centers in US and elsewhere.

CNV disorders: the next generation of questions

There are key advantages offered by CNVs toward understanding neurodevelopmental and psychiatric disorders, but additional work is needed to leverage these for maximum benefit. In the first place, human research cohorts are critical. Since many of these disorders are rare, effort is required to amass a large enough sample size for meaningful research studies. This will require collaboration across sites, as has been successfully implemented for studies of 22q11.2 deletion syndrome[25] and 16p11.2 deletion/duplication[11]. Internet-based registries have been highly effective tools for engaging study subjects with rare disorders[37,38]. These cohorts and registries offer ways to understand the human phenotype and can support longitudinal evaluation for natural history studies as well. Understanding human phenotypes in a systematic way is crucial to advance this research.

But how to evaluate the human phenotype? Comprehensive, unbiased characterization of CNV syndromes is needed. Tools are needed to move beyond diagnosis and into dimensional measures and quantitative phenotypes, as proposed by the NIMH RDoCs initiative. Neuroimaging studies can serve as a link between behavioral phenotypes and genetic lesions by identifying altered brain structures or circuits; these data are reviewed by Moreau et al. Longitudinal evaluation will be important to understand phenotypes across the lifespan[39,40]. The use of standardized phenotyping batteries, and natural history studies with longitudinal data collection will aid research into mechanism, and will also have untold benefits in the clinical/translational realm. Medical geneticists and pediatricians are the “first responders” to these patients and need evidence-based guidelines for referral and follow-up care. Furthermore, as treatments are developed, it is imperative to have outcome measures that can detect and resolve phenotypic changes on an appropriate scale.

As we amass human cohorts to understand the phenotypic spectrum of individual CNV disorders, it will become important to launch cross-disorder comparisons[13]. To do this successfully, standardized phenotyping protocols and harmonized data collection are necessary. Mortillo and Mulle present a comparison of intellectual disability measures across 8 genomics disorders, highlighting the value of cross-disorder comparison as well as the limitations that are imposed when evaluation instruments are not harmonized across studies. Efforts are emerging to address these barriers, with partnerships between researchers and parent-led groups, to harmonize data collection instruments and outcome measures (including AGENDA (Alliance for Genetic Etiologies in Neurodevelopmental Disorders and Autism, https://www.alliancegenda.org); and the Commission for Neurodevelopmental CNVs, https://www.cnvscommission.org). The newly-formed NIH-funded Genes 2 Mental Health Network (https://genes2mentalhealth.com) is another means for cross-disorder comparison, showcasing how funding and NIH support can lead to harmonized data collection.

Each CNV increases the risk for several different disorders and several distinct CNVs confer increased risk for overlapping disorders. This observation suggests the possibility of convergent mechanisms downstream of the initial insults-on either molecular- or circuit level. Much can be learned from studying different CNVs in parallel using multiplex read-outs such as modern molecular-, imaging- or electrophysiological tools. Successful molecular neuroscience approaches will likely use a combination of the models described above. As with human phenotyping, to make progress, cross-disorder comparison is of the utmost importance.

We believe that the community should take a systems biology approach and that this necessitates tackling complexity, applying informed reductionistic steps only after broad observations to identify and record the assumptions necessarily made. Efforts are likely to include network biology, but it is crucial to identify and report ingoing assumptions in the dimensionality reduction that network construction entails. The nervous system is a complex structure developing for a prolonged period and entails a critical interaction between genetic information and environmental variability. It is thus necessary to take developmental time into account. Modern molecular transcriptomic approaches in terms of identifying cell types, are nearing the “complete, accurate and permanent” criteria put up by Brenner[41], which drastically increases the possibility for cross-comparison of data across labs, which will be crucial to the field. With this approach it is likely that we will be able identify convergent biology for either CNVs and/or psychiatric disorders that could serve as key perturbation points for the development of therapeutic avenues.

Finally, as we understand how CNVs cross diagnostic boundaries, we may need to approach funding models differently. In the US, the funding models may enforce silos and allow biases to propagate (i.e., “22q11 isn’t real schizophrenia”). The data are pushing us toward convergence, there may be common underlying biology of many phenotypes we consider “distinct.” Investigating multiple phenotypes simultaneously across the continuum of neurodevelopmental time may be what is needed to capture new etiological insights.

Summary

CNVs offer a singular opportunity for insight into the common mechanisms underlying neurodevelopmental and neuropsychiatric illness. Advantages include large effect sizes, complex but model-able loci, a system to test risk models, with existing cohorts available for study. To advance the study of CNV, the field will need to acknowledge and tackle the complexity inherent in these systems rather that leaning on reductionist approaches. Harmonization of human phenotype measures and molecular neuroscience approaches across loci will be imperative for cross-disorder analyses. Study of CNV loci will lead to new etiologic insights into neurodevelopmental and psychiatric disorders.