Report of the National Advisory Mental Health Council Workgroup on Genomics

Opportunities and Challenges of Psychiatric Genetics

Table of Contents

• Acknowledgement
• Roster
• Introduction
• Background
• The challenge of interdisciplinary exchange
• The challenge of polygenic risk architecture for follow-up biological studies
• Summary and charge
• Recommendations
• References

Acknowledgement

The authors would like to express our heartfelt gratitude to Dr. Pamela Sklar, who contributed significantly to this report, and passed away in November 2017. Dr. Sklar, chair of the Department of Genetics and Genomic Sciences at the Icahn School of Medicine at Mount Sinai and a long-time NIMH grantee, was a clinician scientist dedicated to delineating the genetic risk factors and molecular processes that lead to neuropsychiatric disorders. Pamela was a seminal figure in psychiatric genetics and a leader in many science initiatives that have been the driving force behind the recent era of discovery. She was a founding member and leader of the Psychiatric Genomics Consortium, the Common Mind Consortium, and PsychENCODE, to name just a few.

National Advisory Mental Health Council Workgroup on Genomics - Roster

Co-Chairs

Members

NIMH Staff

Introduction

Steven E. Hyman and John Krystal

After decades of frustration, psychiatric genetics is flourishing. Recent years have witnessed a large and growing number of common and rare DNA variants significantly associated with psychiatric disorders (Ripke et al., 2014; Sanders et al., 2015; Singh et al., 2017; Wray et al., 2018). These advances were possible because of the advent of powerful genomic technologies and new computational resources that permit accurate and cost-effective microarray-based genotyping and DNA sequencing as well as vastly improved analytic tools. The formation of large consortia of investigators willing to share data has been equally important. These consortia have made it possible to aggregate cohorts and perform meta-analyses with the statistical power to identify risk associated alleles with confidence. As has been well documented in other fields of medicine, molecular insights into disease pathogenesis based on genetic information can contribute to the identification of biomarkers and novel therapeutic targets, which are much needed in psychiatry. Moreover, modern large, scale studies, which are unbiased with respect to prior biological hypotheses, can yield new and unexpected insights into pathogenesis as exemplified by the discovery of the role of complement proteins in schizophrenia (Sekar et al., 2016).

The growing successes of large-scale unbiased genetics have also created new challenges that were central to the deliberations of the workgroup. The methodologies and scientific languages of psychiatric genetics and neurobiology are vastly different. As a result, geneticists are often unfamiliar with data formats and explanations that would make their results more intelligible and useful to neurobiologists who often lack the training to select DNA variants and genes for follow-up studies or to understand the genomic context in which these variants might contribute to relevant phenotypes. These cross-disciplinary challenges are significantly exacerbated by the polygenic risk architecture of most psychiatric disorders, which do not fit readily into widely used experimental approaches in neurobiology such as generation of a transgenic animal with a single penetrant allele (Hyman, 2018). Indeed, even when penetrant mutations are identified, as has occurred with syndromic autism spectrum disorders and Tourette disorder (Sanders et al., 2015), the polygenic background contributes strongly to determination of phenotype. There is no established neurobiological ‘playbook’ for the analysis of polygenic disorders. Thus, new approaches will need to be devised that will require extensive collaboration among geneticists, neurobiologists, data scientists and others.

Biological studies aimed at discovering the functional significance of DNA variants for the mechanisms underlying psychiatric disorders represent significant investments of resources and human effort. Thus, it is imperative for investigators and funders to pay close attention to the rigor with which DNA selected variants and genes were identified and to the way they might be expected to influence pathogenesis. This workgroup report is timely precisely because rapidly accumulating genetic data is creating an inflection point for neurobiological investigation of disorder etiology. In this time of scientific change, the workgroup has sought to provide guidance to investigators, journal referees, and grant reviewers and to offer advice to NIMH leadership on how best to navigate this exciting but treacherous scientific transition.

Background

The substantial heritabilities of psychiatric disorders (Sullivan, Daly, & O’Donovan, 2012; Hilker et al., 2017) indicate that molecular clues to pathogenesis are harbored within the nucleotide sequences of human genomes. Given the ready accessibility of blood and saliva, which contain germline DNA, and the wide availability of accurate, scalable, and cost-effect technologies for genotyping and DNA sequencing, investment in genetics is not only important, but also timely. Well-designed, well-powered human genetic studies, carefully interpreted, are proving a direct route to elucidate psychiatric disease mechanisms, which represents a critical step toward the significant goals of advancing biomarker discovery and discovering new, efficacious therapeutics. However, based on the complexity of the results emerging from genetics, the challenges posed by their application to experimental neuroscience are formidable indeed. If we are to succeed in improving treatments and in developing preventive interventions for mental illnesses, we must design and conduct genetic studies with rigor and adequate power, interpret them carefully without over-claiming, and be clear-eyed about the strengths and weaknesses of the technologies and experimental systems available to follow up on emerging genetic information to inform human disease mechanisms and to nominate candidate biomarkers and therapeutic targets.

To date, attempts to discover pathogenic mechanisms of psychiatric disorders have been frustrated by complexity of the human brain, its inviolability in life, limitations in the resolution of noninvasive technologies, and the difficulty of disentangling causal factors from effects of disease (McCarroll, Feng, & Hyman, 2014). Genetics represents a powerful approach to identification of disease mechanisms because a person’s genome is established at fertilization, prior to developmental processes, environmental exposures, and diverse perturbations and adaptations that can confound cause and effect. Thus, germline genetic variants associated--with high levels of statistical confidence—to disease risk can be inferred to be involved in causal processes. This should not be taken to mean that confidently associated variants always act directly in causation: an instructive example of distal causal influence is illustrated by significant allelic variants discovered in a genome-wide association study of lung cancer, linked to the genes encoding subunits nicotinic acetylcholine receptors. These influence lung cancer risk via smoking behavior rather than by direct effects on control of cell division (Hung et al., 2008). In short, interpretation of genetic results must be made carefully and in recognition of confounding factors introduced by the circumstances of ascertainment.

Stochastic developmental events such as somatic mutations that occur in brain development and epigenetic regulation of gene expression might also contribute to pathogenesis of mental disorders. However, this report focuses on the contribution of variation in germline DNA sequences because the significant heritability of these disorders indicates that notwithstanding other risk factors, germline genetics exerts substantial influence on these phenotypes. Another important consideration for the focus of this report is the current feasibility of advancing such genetic studies of mental disorders. Progress in genomic technologies, computing power and data storage, software, and other computational resources to analyze genomes, and significantly, the formation of large, international consortia (‘team’ science) focused on data sharing and achievement of large cohorts have made possible the genetic analysis of complex traits, including several paradigmatic mental disorders (Ripke et al., 2014). To date, unbiased, large-scale genetic studies have yielded highly reliable identification of disorder-associated DNA sequence variation associated with schizophrenia, bipolar disorder, and autism spectrum disorders (ASDs). Nonetheless, significant scientific and organizational challenges remain that we address here with the goal of rapidly and effectively elucidating disease mechanisms and advancing therapeutics. These pressing challenges include:

1. Advancing genetic investigations, already under way, of disorders, such as schizophrenia, bipolar disorder, and ASDs to gain greater insights into their biology and genotype-phenotype relationships. In addition, studies of these highly heritable and paradigmatic disorders will help refine scientific and organizational approaches that might advance the study of other disorders, including those with somewhat lower heritabilities, as well as cognitive, behavioral, and temperamental phenotypes relevant to the understanding of pathological mechanisms that cut across disorders as currently defined.

2. Studying the genetic underpinnings of a wider range of psychiatric conditions, important both in their own right and to gain insight into the biological significance of shared and unshared genetic risk—a goal with clear implications for the targeting of therapeutics.

3. Diversifying the human genomes under study to include those of multiple global populations, both to aid in the critical goal of fine mapping loci to identify causal variation and to advance the important goal of global mental health equity

4. Gaining deeper understanding of how the non-coding genome contributes to disease risk

5. Gaining a better understanding of how somatic mosaicism in the brain might influence psychiatric and neurodevelopmental disorders

6. Engaging with other disciplines (neurobiology, psychology, and clinical disciplines) to ensure that genetic information is shared in forms that are useful and readily interpretable, and to ensure that genetic information that is applied to biological and phenotypic follow-up studies is derived from rigorous, well-powered studies that have been interpreted appropriately

7. Working with the biology community to develop and improve experimental systems, design principles, and computational tools for the conduct of meaningful and insightful follow-up studies of the highly polygenic risk factors that underlie common psychiatric disorders.

The working group spent much time discussing challenges (6) and (7) given their central importance to research on psychiatric disorders and the lack of established solution. These discussions are summarized here, followed by a description of more specific findings.

The challenge of interdisciplinary exchange

Neurobiology and genetics are two highly technical fields with few individuals well-schooled in both. To a great degree neurobiology has prospered by taking hypothesis-driven, reductionist approaches to problems, focusing on the main effects of experimental interventions and holding all other variables constant. In contrast, human psychiatric genetics first succeeded when, after years of failure, it eschewed biological hypotheses (often in the form of biological candidate gene studies). Instead the field turned to unbiased designs powered to confidently identify associations of partially penetrant alleles against highly diverse genetic backgrounds. Even in the case of rare neurodevelopmental syndromes associated with penetrant mutations, successful designs for gene discovery were agnostic to biological hypotheses.

Successfully robust and replicable approaches to gene discovery in psychiatric genetics, such as unbiased, large-scale case-control designs are recent developments. Consequently, the existing literature on putatively disease-associated alleles and genes is of widely varying quality. Many textbooks and even to this day, many journal articles contain outdated information based on outdated and discredited methodologies and weak statistical inference. The persistence of misleading information, reflects a knowledge gap between human genetics and neurobiology that can affect the ability of referees, journal editorial staff, and grant review panels to recognize work that meets modern standards of rigor, robustness, and replicability. As a result, neurobiologists attempting to use genetic findings to investigate disease mechanisms may inadvertently rely on outdated information in designing experiments, and may find it challenging to know which genes to study, which alleles of those genes, and under what circumstances to expect selected alleles to yield disorder-relevant phenotypes. There is a clear and compelling need for increasingly nuanced and sophisticated exchange among disciplines.

The challenge of polygenic risk architecture for follow-up biological studies

The risk architecture of common mental disorders is highly polygenic, likely involving the additive effects of many thousands of common and rare variants linked to or found within many hundreds of different genes and intergenic regions, mostly of low penetrance, and acting in different combinations to produce risk of disorder phenotypes that are now understood to comprise heterogeneous syndromes. Widely used reductionist paradigms for investigating gene function, such as the generation of transgenic mouse lines, can provide important insights into gene function that can provide a basis for study of disease pathological mechanisms. However, the incomplete, and generally modest penetrance of almost all disorder-associated alleles discovered to date indicates that they are not likely to contribute to the disorder-relevant phenotypes under study except in the presence of other risk alleles in yet unknown numbers and combinations. Moreover, the vast majority of common disorder-associated variants are located in noncoding regions of the genome, which exhibit poor evolutionary conservation--a situation not limited to psychiatric disorders. The combination of low penetrance alleles and poor conservation of the noncoding genome indicates that human genetic risk backgrounds are likely to prove necessary in experimental systems designed to recapitulate the pathogenesis of psychiatric disease. Despite the utility and unquestioned importance of mice and other animals in follow-up studies intended to illuminate components of disease pathological mechanisms, genetically engineered animals should no longer be interpreted as veridical models of psychiatric disorder pathogenesis or pathophysiology.

It is important that these concerns about the nature and interpretation of putative disease models not be mistaken as a rejection of studies in model organisms. Studies in model organisms such as C. elegans, Drosophila, mice, and rats provide critically important information about gene function and about how nervous systems and brains work. Without such more basic studies, translational neuroscience would have no basis on which to advance. In contrast with such basic investigation, in which the goal is to demonstrate generalizable findings and principles, attempts to develop disease models intended to recapitulate the precise effects of human allelic variation are constrained by human genetics and human biology.

Further, to exploit the results of genetic discovery, experimental systems for follow-up studies must have the scalability and flexibility to capture the genetic complexity and heterogeneity of psychiatric disorders. Experimental systems that prove useful for the study of polygenic traits will almost certainly have to be amenable to sophisticated computational approaches to data analysis and modeling. The challenge, of course, is that in contrast to older reductionist approaches to studies of gene function that have long been in widespread use in the neurosciences, experimental systems for discovery genetic and genomic studies in the neurosciences are still being developed and evaluated. Put another way, we still lack a tried and true playbook for the conduct of follow-up biological studies of polygenic human brain disorders.

Summary and charge

Several aspects of the challenges identified by the Working Group represent the healthy result of new findings and new questions that lack simple scientific or technological solutions today. Others reflect hurdles created by the need for interdisciplinary scientific exchange between fields that are as conceptually different as neurobiology and human genetics. Finally, some difficulties reflect the inertia of older scientific results, scientific practices, and beliefs despite compelling new information and the important new focus being placed on rigorous study designs, statistical power, replicability, and generalizability.

This Working Group was formed to advise the NIMH director on how best to proceed in recognition of the importance of genetics to the NIMH mission, the significant place of genetics in the NIMH research portfolio, and the challenges enumerated in this introduction. The Working Group was further charged with providing counsel that could aid NIMH staff in advising applicants and in prioritizing research proposals. The Working Group believes it is able to provide significant guidance in many areas relevant to genetics research, while recognizing that many open questions remain, especially with respect to tools and strategies for follow-up studies. This is no small matter, as current ignorance of pathological mechanisms and stasis in therapeutics is a central motivation for the NIMH genetics investment.

Recommendations

1. Adopt rigorous significance standards for disorder association. Over the past two decades, the field of human genetics has learned the hard lesson that widely shared high standards of statistical certainty are necessary for interpreting both common and rare variant studies and to justify biological follow-up studies. High standards allow the community to proceed confidently with investments in follow-up studies. In the absence of rigorously established statistical evidence, no argument for biological plausibility warrants follow-up investments to examine the role of a gene in disease mechanism—genes can, of course, deserve investigation for reasons unrelated to human disease, but then the justification for investment should be expressed accordingly.

   A critical basis for establishing statistical certainty in the analysis of common variation is calculation of the effective number of independent tests being conducted (~1,000,000 on many widely used microarrays), and the same strategy must be applied to replication samples. For exome sequencing, such analyses must take into account not only the number of genes in the human genome (~20,000), but also the number of distinct tests being performed on those genes (e.g., loss of function only, missense only, combined missense and loss of function, and different frequency thresholds). Several well-validated statistical approaches are available to analyze different types of genetic variation and that take appropriate account of potential confounds.

   Whole genome sequencing brings with it new difficulties. For example, the search space of the non-coding genome is orders of magnitude larger than that for the whole exome—even before taking account of the complex spatial and temporal dynamics of utilization of regulatory sequences. Consequently, it is not yet clear exactly how much correction for multiple comparisons will be required to lead to reliable and reproducible results. It is critical that investigators describe the full scope of their analyses and multiple testing procedures.

   Despite the challenges inherent in developing new statistical approaches to complex problems, the field should not relax its vigilance concerning the high risk of false positives. Smuggling biases into the analysis of the noncoding genome, such as limiting studies to specific genomic regions or putative types of regulatory mechanism, risks repetition of the costly failures of the biological candidate gene era.

2. Candidate gene studies of psychopathologic, cognitive, or behavioral phenotypes should be abandoned in favor of well powered, unbiased association studies. If the phenotype is worth studying, it deserves appropriate statistical power and rigorous, sound design. ‘Candidate gene’ is a term with no consensus definition, but can be taken to refer to genes selected for study by means other than an unbiased, genome-wide approach, most often based on prior biological hypotheses. Candidate gene studies attempting to find associations with neuroimaging or other biological phenotypes have historically been vastly underpowered partly because of the high cost of the phenotypic readouts and partly on serious misunderstandings of the influence of sample size on the robustness and significance of results (Button et al., 2013). Candidate Gene-by- Environment (GxE) studies, which might be better described as candidate gene-by- candidate environment studies, have similarly suffered from inadequate power and poor design, including vague definitions of the effective environment. The spawn of candidate gene and candidate GxE studies have been many costly and futile follow-on studies, publication bias, and the propagation of false, if superficially plausible explanations of psychopathology (Duncan & Keller, 2011).

3. Further considerations in selection of variants and genes for further study. The principles discussed under points 1 and 2 are a sine qua non for justifying follow-up studies in humans, human cellular models, or for basic follow-up investigations relevant to disease mechanisms in animals or animal cells.

1. For copy number variants (CNVs), there is mounting evidence that many if not most of these intervals will not resolve to a single causal gene. For this reason, it is critical to distinguish between follow-up biological studies of the entire structural variant -- where statistical association is established -- versus studies of genes mapping within these regions. To assert a causal link for a single gene within a multigenic CNV locus demands the same level of statistical rigor as any other genic association.

2. Regarding exome sequencing studies that identify deleterious mutations. Identification of rare and de novo deleterious mutations within exomes, generally permit a far more straightforward inference from variant to risk gene than do GWAS results. However, it is important to note that such studies often rely heavily on prima face evidence for loss of function variants. Consequently, functional interrogation of putative risk loci is still warranted. Moreover, there is a selection bias toward the identification of haploinsufficiency and away from other possible mechanisms that deserves redress. Finally, as alluded to above, while rare transmitted and de novo mutations are often associated with large effects, there are few findings of true Mendelian inheritance in psychiatric genetics. In general, the link between genotype and phenotype, regardless of effect size, has proven to be highly complex—with a given CNV or single gene containing a damaging mutation leading to a wide range of phenotypic outcomes. Consequently, even in this setting, consideration of the effects of polygenic background in experimental systems remains essential.

1. Common variants: An important caveat should be stated: while a strong consensus exists on criteria for “statistical significance” in the context of GWAS, even the most robust associations with genetic loci do not automatically identify a causal gene or causal variant that warrants functional characterization. Fine mapping of GWAS loci and other methods of identifying causal variants are proceeding apace, but in the intermediate term, caution, and often consultation is warranted before investing in biological follow-up experiments.

2. Rare variants of high penetrance. Childhood onset neuropsychiatric syndromes often referred to as neurodevelopmental disorders, are often associated with disruptive mutations of large effect that often occur de novo in the affected individual. These deleterious mutations may affect the sequences of single genes or may produce larger structural effects such as deletions, duplications, or more complex variation described as copy number variants (CNVs). Rare, penetrant variants have been well documented in association with intellectual disability, some forms of epilepsy, a significant minority of individuals diagnosed with ASDs and Tourette disorder, and also in rarer cases of schizophrenia, bipolar disorder, attention deficit hyperactivity disorder (ADHD), and perhaps obsessive-compulsive disorder. In cases where rare, penetrant variants act, they do not appear to do so alone. A picture is emerging of an important interplay between additive polygenic risk and rare penetrant, mutations in determination of phenotype. Indeed, the presence of both polygenic risk and disruptive mutations are not mutually exclusive, but frequently occur within an individual. Understanding of disease mechanisms and advancement of therapeutics for the substantial number of individuals affected by early onset neuropsychiatric disorders will require insights into the interactions among different types of genetic variation, and of their independent and joint contributions to phenotype.

3. Rare variants of low penetrance. To date it has been difficult to achieve statistical significance for any single gene. There is little question that this reflects inadequate statistical power, which will be resolved as sequencing studies gain increased sample size. In the event that a single rare, low penetrance transmitted variant achieves genome-wide significance with appropriate corrections for multiple comparisons, the path forward for biological investigation is more straightforward than for the noncoding variation found within most GWAS loci. However, a risk gene will likely contain several different low penetrance transmitted risk alleles. In that case, it will be essential to determine the functional overlaps between putative risk mutations in the same gene. This can best be established through direct experimentation. However, there are also resources that increasingly can provide a basis for reliable judgements about the intolerance of a genomic locus to mutation and to find the prevalence of any given variant in healthy populations. These resources include:

   RVIS (Residual Variation Intolerance Score)
   http://genic-intolerance.org 
   ExAC Browser | Exome Aggregation Consortium
   http://exac.broadinstitute.org 
   gnomAD browser | genome Aggregation Database
   http://gnomad.broadinstitute.org 

4. All forms of genomic variation should be explored for disease association. While DNA variation that alters the amino acid sequence of proteins represents lower-hanging fruit for biological follow-up than noncoding variation, in the discovery phase of genetics, it is critical to study both common and rare variation, single nucleotide variants as well as larger and more complex forms of variation. Only then can the risk architecture of disease phenotypes be understood.

5. Expand efforts at discovery of genomic variation beyond DSM and other textbook phenotypes. A significant fraction of heritability for any one diagnostic entity is shared across multiple psychiatric disorders, behavioral and cognitive traits, and many other phenotypes. Fully informative interpretation of genetic associations, and the pathways and biological processes relevant to psychiatric phenotypes, will require aggregation and joint analysis of genetic studies not only across a broad range of traditional disorder phenotypes, but also across diverse phenotypes ascertained at the molecular, cellular, somatic, behavioral, and cognitive levels.

6. Make investments to capture genetic and phenotypic variation across diverse human populations. The importance of capturing human genetic diversity has a strong scientific rationale, but also has the virtue of contributing to global mental health equity. Empirically, different populations often exhibit different allele frequencies, and different patterns of linkage disequilibrium with causal variants. As a result, comparative studies of diverse populations can contribute significantly to fine mapping of disorder-associated loci, can identify biological clues not detectable in currently well-studied European populations, and can identify variation relevant to pharmacogenomic differences and thus to contribute to stratified or ‘precision’ medicine.

7. Development of resources. Resources are needed to support NIMH genomics research that will aid neurobiologists in interpreting genetic data with a view to follow-up. This includes access to and consultation about databases, bioinformatics resources, and statistical methods, both for existing data and for newly emerging whole genome data. This should include resources for storing and manipulating sequence data, collection and storage of DNA for distribution to the community and renewed focus on the development of brain-banks in light of advances in single cell methodologies that work increasingly well for frozen tissues.

8. Experimental systems for biological follow-up studies

   Three predominant strategies have historically been used in psychiatry in attempts to model disease phenotypes or putative endophenotypes thought to represent aspects of disease pathophysiology. These strategies include (1) genetic manipulation of an organism, most often knock-in of a penetrant human disease-associated mutation in an inbred mouse strain; (2) Inbreeding of a model organism for a desired trait (such as alcohol preference); and (3) exposure of a model organism to environmental perturbations or stressors thought to represent risk factors for a disorder. This discussion will focus on the construction of genetic animal models as this is most relevant to the characterization of human allelic variants. It is broadly recognized that approaches (2) and (3) continue to be used to study a variety of biological mechanisms, and that such studies produce information that critically advances our ability to understanding disease. This does not mean, however, that experimental animal studies definitively capture etiologic features or pathophysiology of human diseases. Especially when judged by criteria that this committee considers outdated, such as ‘face validity’, there is concern that the result with be phenocopies, i.e., animals that exhibit desired phenotypes, but with etiological and pathophysiological mechanisms that differ from the human disease. Such phenocopies carry a significant risk of diverting scientific effort to models that ultimately lack translational relevance.

   The committee uniformly supports the need for animal studies in the investigation of gene function and of the contribution of mutations in disease-associated genes to pathophysiological processes. However, for reasons that will be described, such important studies should be understood as basic investigations rather than as the generation of “disease models”. Historically, it was thought that disease models could be developed that would advance understandings of human disease mechanisms and provide tools for the discovery of therapeutics and that animal models of disease could be ‘validated’ by face validity, construct validity, and predictive validity. As alluded to above, face validity is a problematic concept that contributes to acceptance phenocopies that resemble human disease in a superficial and misleading manner. Like face validity, predictive validity is not predicated on recapitulation of human pathophysiology; rather, typical standards of predictive validity can be met with empirical drug assays that in psychiatry have identified only compounds with similar mechanisms to prototype drugs that had initially been identified by their effects on humans. For example, amphetamine-induced hyperlocomotion does not capture actual pathophysiological processes of schizophrenia or indeed any psychotic disorder, but represents a pragmatic approach to identifying compounds that cross the blood-brain barrier and interfere with dopaminergic neurotransmission.

   Modern genetics was initially thought to have produced tools for the generation of animal models with construct validity based on discovery of rare mutations and CNVs that exert significant effects on neuropsychiatric phenotypes, most notably intellectual disability, rare forms of syndromal and idiopathic ASDs and syndromal and idiopathic schizophrenia. Several years of experience and advances in genetics have raised important questions about the best interpretation of such animals. Such questions about how well these animals capture human pathophysiology have not, however, diminished the importance of animal research. However, based on the importance of human genetic background to phenotype determination, and the significant evolutionary divergence of rodents and humans with respect to cerebral cortical cell types, synaptic connections, and circuits, the committee warns against simplistic interpretation of mice and other organisms, even those engineered to express relatively penetrant human disease mutations, as reliable disease models. Instead these animals represent indispensable basic experimental systems in which to study molecular, cellular, and systems level effects of genes and mutations. These studies are important in their own right and also generate hypotheses concerning pathophysiology that can potentially suggest drug candidates or be applied to human biology. Even if the production of transgenic animals is construed in terms of basic investigation rather than disease modeling, justification of a gene or mutation for study based on its relevance to human disease, requires that a disease-association has already been demonstrated in rigorously designed, well-powered studies. Otherwise studies of genes should be proposed as purely basic investigation based on the intrinsic interest of the selected genes.

   The distinction between a basic investigation of the effects of disease associated genes and the generation of putative disease models is not simply a rhetorical matter. Referring to genetically engineered or environmentally perturbed mice or other animals as disease models based on such traditional criteria as construct, face, or predictive validity invites an unwarranted acceptance among members of the translational and clinical research communities that the organism in question exhibits significant correspondences with human pathophysiology that can be used effectively to identify biomarkers and new therapies. In fact, the field has learned the hard lesson that efficacy against an end-point in an animal cannot be expected to predict efficacy in humans. Instead, understanding a mouse line as a basic scientific tool to interrogate the effects of disease associated genes, encourages more open-ended follow-up studies and hypothesis testing in other experimental systems and, where possible, in properly selected patients. The basic science designation invites investigators to identify important mechanisms that may have been conserved in evolution, but also respects the significant divergences between rodents and humans. These reflect not only 90 million years of evolutionary separation, but also the significantly different niches and thus selection pressures that acted on rodents to yield animals that are nocturnal, social in a fairly rudimentary manner, and olfactory compared with the primate ancestors of humans, a species that is diurnal, intricately social, and visual, and that possesses a highly developed prefrontal cortex that differs significantly in its cells, synapses, and circuits from those of rodents. To avoid confusion, we suggest use of the term ‘experimental system’ in place of the term ‘animal model,’ or ‘disease model’. Perhaps the greatest challenge in the use of mice and other animals even for basic follow-up studies of psychiatric genetics is the fact that, as described above, the genetic risk architecture of all common psychiatric disorders is highly polygenic with diverse combinations of risk alleles contributing to similar phenotypes in different individuals who nevertheless vary considerably in phenomenological and biological aspects. To complicate matters further, approximately 90% of significant GWAS loci are found in the noncoding genome, which is poorly conserved across species compared with coding regions. As alluded to, even in the presence of more penetrant mutations within genes or penetrant CNVs, there is significant phenotypic diversity that appears to reflect the polygenic background of the individuals. It is worth noting that similar observations have been made in inbred transgenic mouse lines constructed with a severe mutation: The same transgene bred into different inbred lines gives rise to significantly different phenotypes.

   Given the modest penetrance of the vast majority risk alleles (both common and rare), and the requirement for many other alleles, stochastic developmental processes, and environmental effects to produce disease, a pressing question is how to interrogate psychiatric disease risk alleles in a manner that will be truly informative. Many investigators are currently attempting to answer that question by advancing human cellular models produced through stem cell technologies. Potentially, human two- and three- dimensional cellular models can be coaxed to produce diverse human neural cell types, be constructed from a large number of both affected and unaffected genetic backgrounds, can be genome engineered to add or subtract alleles of interest, and can be employed in assays of relatively high throughput. There is a long way to go, but even when such experimental systems achieve maturity, they will not permit investigation of human neural circuit function, cognition, or behavior. Thus, from the vantage of the present there will always be a need for basic animal research relevant to disease mechanisms. There is also a significant need to improve human experimental biology.

9. Data sharing: Genetic data has a wide range of uses. These include population genetic analyses that can advance the understanding of how natural selection shapes genetic variation—and by extension the genetic architecture of disease traits; statistical analyses tasked with identifying the genetic risk factors that drive disease, and biological follow-up studies to identify disease mechanisms. Data sharing is fundamental to all of these purposes and necessary if we are to fully realize the full value of substantial investments in genetic studies. Investigators must adopt consent language at the beginning of studies that facilitates data sharing, particularly for general research use; data should be shared as widely and rapidly as possible; and it should be in a form that makes it maximally useful.

References

Button, K. S., Ioannidis, J. P.A., Mokrysz, C., Nosek, B. A., Flint, J., Robinson, E. S.J., & Munafò, M. R. (2013). Power failure: why small sample size undermines the reliability of neuroscience. Nature Reviews Neuroscience, 14(5), 365 – 376.

Duncan, L. E., & Keller, M. C. (2011). A critical review of the first 10 years of candidate gene-by-environment interaction research in psychiatry. American Journal of Psychiatry, 168(10), 1041-1049.

Hilker, R., Helenius, D., Fagerlund, B., Axel, S., Christensen, K., Werge, T.M., Nordentoft, M., & Glenthoj, B. (2018). Heritability of schizophrenia and schizophrenia spectrum based on the nationwide Danish twin register. Biological Psychiatry. http://dx.doi.org/10.1016/j.biopsych.2017.08.017. 

Hung, R. J., McKay, J. D., Gaborieau, V., Boffetta, P., Hashibe, M., Zaridze, D., ... & Fabianova, E. (2008). A susceptibility locus for lung cancer maps to nicotinic acetylcholine receptor subunit genes on 15q25. Nature, 452(7187), 633..

Hyman, S.E. (2018). The daunting polygenicity of mental illness: making a new map.  Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences, 373(1742). 20170031.

McCarroll, S. A., Feng, G., & Hyman, S. E. (2014). Genome-scale neurogenetics: methodology and meaning. Nature neuroscience, 17(6), 756.

McConnell, M. J., Moran, J. V., Abyzov, A., Akbarian, S., Bae, T., Cortes-Ciriano, I., ... & Ganz, J. (2017). Intersection of diverse neuronal genomes and neuropsychiatric disease: The Brain Somatic Mosaicism Network. Science, 356(6336), eaal1641.

Ripke, S., Neale, B. M., Corvin, A., Walters, J. T., Farh, K. H., Holmans, P. A., ... & Pers, T. H. (2014). Biological insights from 108 schizophrenia-associated genetic loci. Nature, 511(7510), 421.

Sanders, S. J., He, X., Willsey, A. J., Ercan-Sencicek, A. G., Samocha, K. E., Cicek, A. E., ... & Goldberg, A. P. (2015). Insights into autism spectrum disorder genomic architecture and biology from 71 risk loci. Neuron, 87(6), 1215-1233.

Sekar, A., Bialas, A. R., De Rivera, H., Davis, A., Hammond, T. R., Kamitaki, N., ... & Genovese, G. (2016). Schizophrenia risk from complex variation of complement component 4. Nature, 530(7589), 177.

Singh, T., Walters, J. T., Johnstone, M., Curtis, D., Suvisaari, J., Torniainen, M., ... & Kirov, G. (2017). The contribution of rare variants to risk of schizophrenia in individuals with and without intellectual disability. Nature genetics, 49(8), 1167.

Sullivan, P.F., Daly, M.J., & O’Donovan, M. (2012). Genetic architecture of psychiatric disorders: the emerging picture and its implications. Nature Reviews Genetics 13, 537-551.

Wray, N.R., Ripke, S., Mattheisen, M., Trzaskowski, M., Byrne, E.M., Abdellaoui, A., et al. (in press). Genome-wide association analyses identify 44 risk variants and refine the genetic architecture of major depression. Nature genetics.