Genes To Mental Health (G2MH): A Framework to Map the Combined Effects of Rare and Common Variants on Dimensions of Cognition and Psychopathology

Rare genomic disorders (RGDs), including structural genomic variants such as copy number variants (CNVs) and sequence-level variants such as single-nucleotide variants (SNVs), are major contributors to neurodevelopmental psychiatric disorders (Box 1, Table 1). In clinical samples, RGDs are identified in 14%−37% of individuals with autism spectrum disorder (ASD), depending on ascertainment (1, 2), and 2%–7% of individuals with schizophrenia (3). Case-control association studies have linked 102 genes and 16 CNVs with ASD (4, 5), as well as 14 CNVs with schizophrenia (6). Among the latter, seven were also linked to ASD. However, for both disorders, a much broader landscape of rare genomic variants is implicated, as suggested by the greater burden of rare CNVs in individuals with ASD and schizophrenia compared with control subjects (5–7). In fact, statistical models trained on large samples suggest that any deletion or duplication ≥1 Mb encompassing coding genes would be associated with an increase in ASD risk and a decrease in cognitive functioning, albeit with a range of effect sizes (8).

In sharp contrast to the accelerated pace of genomic discovery, very little is known about the mechanisms by which genomic variants increase risk for neurodevelopmental psychiatric disorders. In recent years, the National Institute of Mental Health (NIMH) has invested in the study of RGDs. The Genes to Mental Health Network (G2MH) is a newly established NIMH-funded initiative that collects, shares, and analyzes large-scale data sets combining genomics and dimensional measures of psychopathology. The present article is not intended as a comprehensive review but rather a perspective on the current understanding of the contribution of rare variants to measures of cognition, behavior, and risk for psychiatric conditions. To advance the field, we propose strategies that will be pursued within the G2MH network and discuss challenges currently faced by the field and how results from such studies can be translated to the clinic with potential contributions to patient care.

DO RARE VARIANTS EXERT SPECIFIC OR SHARED EFFECTS ON PSYCHOPATHOLOGY?

The nature and specificity of RGD effects on cognitive and behavioral traits (i.e., whether a gene function or a molecular pathway leads to particular cognitive and behavioral profiles) is an area of intense investigation. Multimorbid presentation of neurodevelopmental and psychiatric conditions is common (9–12). All RGDs studied to date affect cognition to varying degrees and affect a broad range of cognitive functions, including reaction time, attention, executive functions, and social cognition (13, 14). Early studies highlighted distinctive behavioral profiles, such as social disinhibition, excessive empathy, and nonsocial anxiety described in 7q11.23 deletions (Williams-Beuren syndrome [15]), insatiable appetite in paternal 15q11.2-q13 deletions (Prader-Willi syndrome [16]), sleep disturbances in 17p11.2 deletions (Smith-Magenis syndrome [17]), and childhood apraxia of speech in 16p11.2 proximal deletions (18).

Recent studies have attempted to quantify distinct and shared patterns of measurable cognitive and behavioral traits altered by specific genotypes (19–21). A study of 258 individuals with neurodevelopmental disorders and 13 CNVs found that 80% of carriers met criteria for one or more psychiatric disorders, but overall, the range of affected behavioral traits was broadly similar for these CNVs, and specific genotypes accounted for a low proportion of the variance (5%−20%) (20). Phenotypes with impairments across all CNVs included mood, sleep, peer relationships, and sustained attention (Figure 1). Recent studies have leveraged the shared effects of RGDs on cognition by developing models that can predict the effect size of any CNV on IQ with close to 80% accuracy (22).

PHENOTYPIC VARIABILITY

The variability of behavioral symptoms observed in individuals who carry the same RGD has been the subject of several studies (11, 23, 24). Because recurrent CNVs typically have the same break points and include the same genes in unrelated individuals, they are well suited to the investigation of such questions. Hypotheses put forward to explain phenotypic variability may be broadly classified in two categories: nonadditive interactions and additive models (Box 1).

Under the former hypothesis, RGDs may interact with other variants, but nonadditive effects involving variants at different loci remain elusive. In rare cases, a deletion may uncover a second variant on the remaining allele (i.e., recessive condition). For example, in five patients with 22q11.2 deletions, missense variants in the CDC45 gene on the remaining allele in 22q11.2 deletion carriers were found to lead to a syndrome associating craniosynostosis, cleft lip and palate, gastrointestinal, genitourinary and skeletal anomalies, and short stature) (25, 26). Likewise, RBM8 missense variants on the remaining allele in individuals with 1q21.1 deletions caused thrombocytopenia-absent-radius syndrome (25, 26). However, recent data for 22q11.2 deletion syndrome provide no evidence for such recessive effects for schizophrenia expression. Larger studies are required to investigate these challenging questions. Indeed, for traits such as cognition and behavior, the genetic contribution to phenotypic variance has mainly been attributed to additive effects (27). This is in line with studies showing that IQ variance within groups of 16p11.2 and 22q11.2 deletion carriers is similar to that in the general population, although the mean IQ is significantly lower. Therefore, variance in CNV groups is thought to be related to additive effects of genetic (and nongenetic) factors that are captured by parental IQ and common variants (11, 23, 28–30) (Figure 2).

Several conclusions can be drawn from the body of literature on RGDs. First, although many RGDs have large effects, in most instances, phenotypic expression of neurodevelopmental psychiatric disorders occurs in the context of additional genomic and environmental factors. Second, similar to studies focused only on common single-nucleotide polymorphisms, as sample sizes increase, an ever-growing number of rare variants are associated with intellectual disability, ASD, schizophrenia, and other neurodevelopmental psychiatric disorders. Third, RGDs are invariably associated with more than one psychiatric condition. Fourth, some RGDs show preferential association with schizophrenia or ASD, such as 22q11.2 deletions and duplications, respectively (6). Conceptual models may be put forward to link these observations (Figure 1A). For example, in a “common dimensionality” scenario, we posit that each RGD alters (with different effect sizes and proportions) the same dimensions of cognition and psychopathology. Variable alterations across each dimension may lead to distinct clinical manifestations classified as overt neurodevelopmental psychiatric diagnoses (Figure 1). A systematic and quantitative standardized phenotyping approach across RGDs would help decompose the contribution of these dimensions to psychiatric diagnoses.

KNOWLEDGE GAPS

An Elusive Phenome: Closing the Gap Between the Tidal Wave of Gene Discovery and Phenotypes

In this era of rapidly evolving gene discovery, many genomic variants conferring varying degrees of risk for neurodevelopmental and psychiatric conditions have been identified. However, there has not been comparable progress in scalable, harmonized phenotyping measures and methods, which generally remain more expensive and labor-intensive than genotyping. As a result, genomic associations have largely been established for categorical diagnoses, but the effects of variants on multiple phenotypes and dimensional traits have not been investigated on a large scale. Dimensional phenotyping across diagnostic categories is needed to identify the underlying mechanisms that may mediate some of the causal pathways between RGDs and diagnostic psychiatric outcomes (31). A barrier to achieving this has been a lack of cohorts providing consistent, dimensional assessments across a broad spectrum of RGDs irrespective of psychiatric diagnoses. The IMAGINE-ID (Intellectual Disability and Mental Health: Assessing the Genomic Impact on Neurodevelopment) cohort in the United Kingdom has been one of the few studies of RGDs to employ a multidimensional deep-phenotyping approach across a range of loci (20). It is critical that multidimensional phenotyping methods that capture the full breadth of the neuropsychiatric phenome be developed and that they be scalable and culturally appropriate for large international multisite studies. Previous publicly available resource efforts, such as the Simons Simplex Collection (32), have focused on specific diagnoses, providing dimensional phenotyping and exome sequencing in probands with ASD. The latter allowed researchers to establish general relationships between RGDs, IQ, gross motor skills, phonological memory, and language skills (33, 34). Although these resources have been instrumental in advancing our knowledge, they include a limited set of RGDs, and many measures used are either disorder specific (e.g., the Autism Diagnostic Observation Schedule) or developmentally appropriate only for a particular age range, making cross-study harmonization of measures a challenge.

The G2MH strategy aims to capture dimensional phenotypes consistently assessed across a range of clinically ascertained and unselected populations worldwide. This approach is optimally suited for the Research Domain Criteria (RDoC) framework (35) because genetic effects are likely to manifest along dimensions of behavior, and individuals who harbor RGDs can be assessed irrespective of whether they present with subthreshold symptoms or meet full criteria for psychiatric diagnoses.

RDoC is a neuroscience-based framework to guide psychiatric research. It was designed to examine and link the distribution of key behavioral domains to genetics and brain circuitry, thereby advancing the mechanistic understanding of psychiatric conditions (36). Within the RDoC framework, RGDs provide a unique opportunity for genome-phenome association. Such efforts may require adaptations of RDoC measures to pediatric populations, because most neurodevelopmental psychiatric disorders present in childhood. The establishment of normative charts for development will contribute to early identification of risk and protective parameters and advance precision psychiatry across the lifespan.

Historically, an important question has been to investigate whether clusters of dimensional measures associated with RGDs are representative of behaviorally defined conditions, such as ASD and schizophrenia, diagnosed in individuals without any identified RGD (Figure 1).

A study comparing individuals with different CNVs with autism to individuals with “idiopathic” autism found evidence of a range of profile differences; however, these differences tended to be subtle (24). Another recent study showed that CNVs including genes intolerant to haploinsufficiency identified in ASD cohorts were not associated with any differences in ASD severity (34). Individuals with de novo SNVs (as a group) in ASD cohorts are more likely to present with motor delay and a less severe symptom profile with respect to social communication and language deficits compared to those with ASD without de novo variants (33).

Studies of 22q11.2 deletion syndrome indicate that representativeness of neurodevelopmental psychiatric disorders by specific RGD may vary by phenotype. For example, the phenomenology of the schizophrenia clinical phenotype in 22q11.2 deletion syndrome is similar to that of idiopathic schizophrenia (37, 38). In contrast, current evidence suggests that attention deficit hyperactivity disorder (ADHD) in 22q11.2 deletion syndrome has a more specific profile characterized predominantly by inattention symptoms rather than hyperactivity symptoms (39). Overall, these findings suggest that considering idiopathic cases as most phenotypically representative of a condition may be questionable. There may or may not be a clear distinction between the phenotypic presentation of individuals with and without a known RGD who meet standard diagnostic criteria for a particular neurodevelopmental psychiatric disorder.

Additionally, because RGDs can be identified early in development, there are novel opportunities to conduct studies using accelerated longitudinal designs to uncover the early developmental correlates of adult dimensional traits. Pioneering work in 22q11.2 deletion syndrome has identified longitudinal dimensional antecedents of psychotic phenomenology, including verbal IQ decline (40), executive functions, spatial working memory, sustained attention, and early language measures (41, 42).

Biases and Challenges in the Field

Ascertainment methods.

As in all research involving human subjects, several sources of ascertainment bias may affect studies of RGDs. Such studies have mainly been conducted in clinically ascertained individuals (i.e., referred clinically for genetic testing), which is expected to favor inclusion of more severely affected individuals. This is particularly problematic for genomic variants that have effects that are disproportionately milder than typical expression required for clinical referral. The 15q11.2 deletion is a prime example because it has mild effects and occurs relatively frequently in the general population and therefore is frequently identified in neurodevelopmental disorder and genetic clinics. The literature is extensive and confusing, with clinical series reporting cases with mild to severe neurodevelopmental symptoms as well as malformations, leading authors to delineate a microdeletion syndrome with considerable incomplete penetrance (43). Yet, studies in unselected populations and meta-analysis of case-control studies have demonstrated that in fact this deletion, on average, confers very small risk for major disease (odds ratio <2 for ASD, schizophrenia, intellectual disability, and congenital heart disease) and has small effects on cognitive ability (a 4-point drop in IQ) (44, 45).

Methods to limit certain ascertainment biases, especially clinical severity biases, include family studies and the use of unselected populations and general population cohorts. Family studies evaluating first-degree relatives who do not carry the variant of interest allow adjustment for genetic and environmental background factors. Such approaches have provided robust estimates of the effects of RGDs on cognitive and behavioral traits (11, 28). Unselected population studies have been successfully used to accurately estimate effect sizes of RGDs on cognitive traits (22, 45–47). As sample sizes rapidly increase, this approach will become relevant even for very rare variants (48). However, such studies introduce other biases because “unselected” cohorts tend to select against the inclusion of the most affected individuals.

Genomic studies in general population cohorts (epidemiological samples representative of the entire populations) remain uncommon and would theoretically represent the gold standard for unbiased estimates. The Integrative Psychiatric Research Consortium (iPSYCH) case-cohort study has examined 30,000 randomly sampled control subjects and 57,377 case subjects born between 1981 and 2005 identified within the Danish health system to have a diagnosis of ADHD, major depressive disorder, schizophrenia, ASD, or bipolar disorder (49). The population prevalence was 1:3672 and 1:1606 for deletions and duplications, and 60% of deletion carriers were missed by clinical ascertainment, as would be expected for a cohort born before the routine use of chromosomal microarray analysis in diagnostics. By age 32, 10% of these CNV carriers had a recorded diagnosis of ADHD, ASD, or intellectual disability. These estimates suggest that previously reported prevalences of ASD and schizophrenia in clinically ascertained samples may have been overestimated (37). However, epidemiological studies, such as iPSYCH, can potentially introduce other biases, such as the lack of systematic standardized assessment methods leading to underestimation of diagnoses, symptoms, and disabilities. In addition, expression of both cognitive and behavioral phenotypes varies widely across the lifespan (50).

The size and number of genomic studies in unselected population cohorts have significantly increased in the past decade. Because CNVs can be identified using data from genotyping arrays, such studies have been informative on the effects of CNVs on cognition and behavior in individuals who were not selected for a particular condition. However, such studies tend to recruit individuals who are, on average, healthier than the general population, and therefore carriers of RGDs can be significantly underrepresented (for example, the frequency of 16p11.2 and 22q11.2 deletion carriers in the UK Biobank database was 1/4000 and 1/30,000, respectively—twofold and sixfold lower than expected) (22, 51, 52). Mortality effects of certain CNVs (53, 54) may also play a role.

Cultural biases.

Improving the representation of diverse populations (i.e., cultural, geographical background, and ancestry) in mental health and genomic research is an urgent scientific priority. However, the majority of assessment tools for cognition and psychiatric and developmental disorders have been developed in high-resource contexts, with a large proportion of validation samples being of anglophone and European ancestry. Studies adapting or validating assessment tools to account for different environments, languages, and cultural backgrounds remain scarce. As a result, the overall validity of phenotypic data outside of these populations can be questioned. In particular, the interpretation of reported or observed symptoms and comparability of data from different contexts and populations remains a challenge (55, 56).

Subjective methods of assessments—diagnostic criteria.

Inherent to all studies in this field is the challenge stemming from the fact that DSM and ICD remain almost entirely descriptive and agnostic to etiology. Consequently, even though based on objective criteria describing symptoms and functional impact, psychiatric nosology does not necessarily coincide with biologically valid distinctions between phenotypic classes (57). While this challenge is not exclusive for psychiatry—in other fields of medicine, there are also diagnoses without established biological correlates—this challenge is substantially more prominent in psychiatry. Whereas for many nonpsychiatric conditions a variety of laboratory examinations are available to confirm a diagnosis, the adequate identification of psychopathology depends on the reporting ability (and willingness) of the patient and/or proxy and the observational skills of the examiner. Subsequently, the obtained behavioral data require interpretation, adding another level of variability. For example, difficulty with eye contact may be interpreted as social anxiety or indicative of communication deficits, such as ASD. Ascertainment may also vary across clinics where patients with RGDs are followed (psychiatry, psychology, developmental pediatrics, neurology, and clinical genetics). These difficulties are pertinent to any study of psychiatric diagnoses, and as a result of these challenges, rates of psychiatric diagnoses and symptoms in patients with RGDs are not definitively established.

PRECISION PSYCHIATRY: TRANSLATING KNOWLEDGE INTO THE CLINICAL SETTING

Predictive Testing

Given the substantial heritability of many psychiatric disorders (Figure 3), there is increasing interest in using genetic variants to identify individuals at high risk of developing psychiatric symptoms. To date, polygenic risk scores (PRSs), which are additive models aggregating the effect of thousands of common variants, remain poor predictors of psychiatric disease in general medical settings (58). For example, individuals with schizophrenia PRS values in the top decile have approximately a 2% risk of developing schizophrenia, compared with a 1% risk in the rest of the population (23, 59). While PRSs will continue to improve as sample sizes increase for genome-wide association studies (GWASs), we are still a long way from filling in the current “missing heritability” gap of psychiatric conditions and cognitive traits (60). In addition, the GWASs used to compute PRSs were performed in individuals of European descent and do not generalize to individuals with other ancestries.

On the other hand, while neurodevelopmental psychiatric disorder RGDs only explain a small proportion of population-based liability for psychiatric risk, their large effect sizes substantially increase baseline risk in carriers. Using PRSs in the context of an RGD offers additional opportunities to refine prediction. A recent study showed that applying schizophrenia and IQ PRS information to high-risk individuals with 22q11.2 deletions provided positive predictive values that approached levels of clinical utility (23). The baseline prevalence of schizophrenia and intellectual disability in 22q11.2 deletion carriers was 23% and 41%, respectively. However, among 22q11.2 individuals in the highest and lowest schizophrenia PRS risk deciles, schizophrenia prevalence was 33% and 9%, respectively. Similarly, 63% of individuals in the lowest IQ PRS decile had intellectual disability, compared with 24% in the highest decile (23).

These results in individuals with 22q11.2 deletions appear to be consistent with additive models, as shown in Figure 2A and 2B, where we estimate positive predictive values by combining previously published data on the effect sizes of 19 CNVs on cognitive ability and risk for schizophrenia and ASD and the effects of PRSs for IQ, ASD, and schizophrenia on their corresponding trait and disease risk. If our assumptions are correct, this approach could eventually provide clinically meaningful stratification for CNVs with large effect sizes in the range of that observed for 22q11.2 deletions (61). In addition, modest increases in PRS performance, which we expect in the near future, could have a significant impact on psychiatric risk stratification for some RGDs. On the other hand, under additive model assumptions, RGDs with smaller effect sizes are unlikely to benefit as much from PRSs unless the field uncovers evidence of nonadditive interactions. Additionally, CNVs with extremely large effect sizes (e.g., the impact of trisomy 21 on IQ) (Figure 2B) may not benefit from PRS stratification because most of the risk is already conferred by the RGD itself.

Preventive Care and Interventions

Guidelines for clinical care and follow-up of individuals who carry specific RGDs have been published for a few high-impact variants (62–64), along with general recommendation for any RGDs (65).

Relevant guidelines include multisystem assessments beyond the initial reason for referral, taking into account associated comorbidities, and recognizing the need for coordinated multidisciplinary care over the lifespan, inclusive of medical, psychological, and social services, as well as genetic counseling (62–64). Individuals with RGDs often have needs that transcend single specialties and can provide challenges to existing services (62–64, 66). Many of the highly penetrant risk variants for neurodevelopmental disorders also seem to confer considerable risk for mood and anxiety disorders (50, 52, 67), which reinforces the need for psychiatric services.

Predictive testing for neuropsychiatric disorders occurs routinely as a by-product of chromosomal microarray and whole-exome or genome sequencing prescribed in prenatal (68), neonatology, and developmental pediatric clinics (although diagnostic practices vary widely across health systems). Although these tests are typically ordered to diagnose malformations or nonspecific, and often transient, motor symptoms, such as neonatal hypotonia, in many cases the diagnosed RGD provides knowledge about increased risk for psychiatric symptoms and disorders that may manifest years or even decades later and about other treatable medical conditions. Currently, however, patients are only seen by psychiatric services if they are experiencing symptoms.

The positive predictive value of RGDs, especially when combined with additional genetic factor PRSs (as discussed above), has the potential to provide great clinical value and represents opportunities to investigate the efficacy of preventive interventions. Information about underlying genetic causes can also be beneficial for the planning of appropriate educational support (62–64). Early involvement and close collaboration of primary care, clinical genetics, and developmental pediatrics with child, adolescent, and adult psychiatry may be beneficial for providing clinical care that can optimize outcomes.

Lack of Routine Genetic Testing in Psychiatric Clinics

Guidelines have been established for clinical genetic testing in patients with developmental delay, intellectual disabilities, ASD, and complex learning disabilities, in addition to patients with major congenital anomalies (69–71). Genetic testing for CNVs and SNVs is primarily carried out in genetic clinics, prenatal clinics, and some developmental pediatric and neonatal clinics. However, knowledge about clinical applications of genetics is modest, and diagnostic genetic testing is still a rare practice in child and adult psychiatry and in many resource-limited settings (72). This has led to lost opportunities for patient care and training for practitioners who may follow many patients with unrecognized RGDs (73). (Clinical vignettes are provided in Box 2.)

Key challenges to implementation of diagnostic genetic testing recommendations include inadequate medical genetics training by psychiatrists, particularly related to practical issues of test selection, results interpretation, and genetic counseling. A recent study showed that the likelihood for child psychiatrists to order genetic testing was related to the clinician’s own perceived knowledge about these aspects (74). Another challenge is the lack of a unified diagnostic approach for patients with RGDs, whereby psychiatric symptoms and genetic etiology are diagnosed and managed as two entirely separate clinical realities, despite their obvious connection (75), except where specialty clinics exist (62–64, 66).

While inconsistencies remain in recommendations for clinical genetic testing and medical education, depending on the professional group cited and the recency of their published statement, initiatives for harmonization are starting to occur. For example, the Residency Education Committee of the International Society of Psychiatric Genetics (ISPG) published recommendations for the medical training of psychiatrists and endorsed consideration of genetic testing (76). Both the American College of Medical Genetics and Genomics and the European College of Medical Genetics recommend chromosomal microarray analyses as a first-tier evaluation of all children with ASD, intellectual disability, neurodevelopmental disorders, or congenital anomalies (71, 77). The United Kingdom’s National Institute for Health and Care Excellence, however, suggests deferring to evaluation by medical genetics specialists for decisions on genetic testing. A separate multidisciplinary consensus statement that included a meta-analysis by an expert consensus group cites evidence for high diagnostic yields and recommends exome sequencing as a first-tier clinical test in individuals with intellectual disability or ASD (78). Individual research groups have also endorsed genetic testing for patients with schizophrenia, based on evidence for clinically relevant CNVs in this population (79, 80). However, practice guidelines from APA suggest “genetic testing” in the context of psychotic symptoms and mention 22q11.2 deletions without further clarification, while the Canadian Psychiatric Association recommends “genetic testing based on the history and physical examination of the patient, especially at the time of the first episode of psychosis.” A position statement by the ISPG opines that diagnostic genetic testing may have value for patients with ASD or intellectual disability, while offering no specific recommendations on schizophrenia.

Lack of testing in the psychiatric clinic also has significant impacts on scientific advances limiting the number of clinically initiated projects and reports. As an example, there are very few large-scale reports on the yield of genetic testing for schizophrenia in a clinical setting. Increasing the implementation of genetic testing in psychiatry will require the integration of medical genetics in clinical training.

G2MH ROADMAP

Given the promise of genomics approaches to neurodevelopmental psychiatric research, but recognizing the need to tackle the knowledge gaps and challenges discussed throughout this review, the G2MH network (https://genes2mentalhealth.com) was established in 2019 with funding from NIMH and the Eunice Kennedy Shriver National Institute of Child Health and Human Development. The network currently includes researchers representing 14 institutions in North America, Europe, and Africa, all with a shared goal of creating a robust phenotypic pipeline to standardize and accelerate the clinical characterization of RGDs. Conceptually, while few clinical conclusions can be drawn by studying a single individual with a specific RGD, aggregated data on standardized measures from a substantial sample of individuals with the same genetic disorder would allow meaningful analyses, fuel hypotheses, and drive discovery. As such, the G2MH network espouses a strong commitment to data sharing as an essential strategy to reach sample sizes required to power analyses. Its initial focus is on establishing processes for the collection, sharing, and harmonization of quantitative data from measures of cognition and behavior (Figure 3) across multiple genomic variants that confer increased risk of adverse developmental and psychiatric outcomes. Currently, the five main activities of the network are new data collection, leveraging archival data, phenotypic harmonization, identifying new satellite projects, and data sharing.

New Data Collection

Cognitive and behavioral measures investigating the dimensions detailed in Figure 1 will be collected in 2,300 probands who carry RGDs at five target loci (1q21.1, 15q13.3, 16p11.2, 22q11.2, and CHD8) and their family members, as well as in a cohort of 1,000 individuals with ASD and related neurodevelopmental disorders and their parents and a matched number of unaffected community control children. Recruitment and assessments will take place in very different health systems across North America, Europe, and Africa. Additional genetic factors present in probands and family members will be characterized via genotyping and whole-genome sequencing. Phenomena such as dynastic effects and assortative mating can induce correlations between genotypes and phenotypes. By using family data, we will investigate how familial (genetic and environmental) factors may have affected the estimated effect sizes of genomic variants in cognitive and behavioral traits (81).

Leveraging Archival Data

There are many opportunities to identify rare CNVs in large unselected and clinical populations, with cognitive and behavioral assessments as well as data from genotyping arrays. Therefore, G2MH will also coalesce data for close to 800,000 individuals across 30 data sets, including individuals from 5 to 80 years old. CNVs will be identified using genotyping array data. The effect of recurrent CNVs, as well as nonrecurrent CNVs, will be consistently measured across cognitive and behavioral dimensions discussed above (Figure 1). Many of these population cohorts will provide much more information on smaller CNVs than pathogenic large-effect-size CNVs that are the focus of this review. However, the latter are typically multigenic and inherently complex (e.g., 22q11.2 deletions include 50 genes, 10 of them being intolerant to haploinsufficiency). Several studies have shown that models trained on observations of small, nonpathogenic variants in the general population can predict the effects of large multigenic pathogenic variants (22, 34, 47). Therefore, studying all recurrent and nonrecurrent CNVs, regardless of their effect size, is a strategy to help decipher the diversity of mechanisms that are at play in large pathogenic CNVs.

Phenotypic Harmonization Strategy

New phenotypic data collection and archival data will require harmonization for joint analyses. Cohort-specific phenotypic data (diagnostic instruments, rating scales, self-reports, and cognitive assessments) will be mapped to the key dimensions shown in Figure 1 and integrated into a normalized data set in a central database. Integrative analyses will be used where factor models estimated in disparate studies are transformed to minimize measurement invariance.

New Satellite Projects

Currently, G2MH focuses on a small proportion of neurodevelopmental psychiatric disorder-associated RGDs. Ongoing and future studies targeting other genomic loci are encouraged to join the network, including studies of monogenic disorders or CNVs. G2MH now includes investigators studying additional loci, such as 3q29, 16p12.1, and 17q12. The harmonization strategies described above will be applied in order to perform analyses across as many genomic loci as possible.

Data Sharing

Data will be made available to the broader research community. The central hub for data sharing will be the NIMH Data Archive. Individual projects will make regular deposits of genomic data. Access to data derived from human samples will be shared using a controlled access mechanism that complies with regulatory requirements and governance policies regarding protection of personal information.

The projects currently included in G2MH are limited to prospective data collection of the most frequent recurrent CNVs, as well as data available from CNV carriers identified in unselected populations. Therefore, a broad spectrum of RGDs will remain undocumented. When fully operational, an expanding number of G2MH member sites will continuously contribute to a central database, phenotype and genotype data from consented families who carry RGDs at additional genomic loci. Data will be derived from a variety of sources, including prospectively collected and archival RGD research. Additionally, some G2MH sites have established processes to capture research-grade data generated as a by-product of clinical care of consented individuals with RGDs, a cost-effective strategy that is not utilized by other large-scale phenotyping efforts. This unique “learning health care” approach (82) capitalizes on freely available clinical information, redirecting valuable research funds away from data generation toward analysis and discovery. Ultimately, the G2MH aims to fuel breakthroughs in treatment by creating a cost-effective and continuously expanding data resource with actively engaged, well-characterized families who would be potentially interested in participating in future drug development and nonpharmacological interventions.

G2MH is complementary to other major psychiatric genomics efforts, such as the PsychENCODE Consortium (https://www.nimhgenetics.org/resources/psychencode), which focuses on noncoding functional genomic elements in the human brain, with the goal of elucidating their role in the molecular pathophysiology of neurodevelopmental psychiatric disorders. Preclinical models (both animal and in vitro/cellular models) of these genetic defects offer “bottom-up” insights into neurobiological mechanisms underlying these human conditions (83).

CONCLUSIONS

With the increasing implementation of genetic testing into clinical care, RGDs are being more routinely identified. While the resolution, scope, and accuracy of genetic testing methods have advanced substantially over the past two decades (84–97), our ability to describe associated neurodevelopmental and psychiatric phenotypes, both categorically and dimensionally in a consistent and unified way, is lagging behind. RGDs represent unique opportunities to elucidate mechanisms underlying risk for neurodevelopmental psychiatric disorders. In order to fully exploit these opportunities, comprehensive and harmonized phenotyping strategies are required. G2MH will provide large-scale data to investigate the effects of RGDs on dimensional phenotypes and additional genetic and environmental factors that modulate these effects. These results will help to refine clinical outcome predictions and design future interventions that can be implemented in the clinic to improve patient outcomes.