Psychiatric Genetics Begins to Find Its Footing

First, a few take-home messages about what we know so far. (More detailed discussion of these themes can be found in recent reviews [e.g., 1–3].)

All major psychiatric disorders have a familial and heritable component. Decades of family and twin studies conducted even before the human genome sequence was mapped have established that psychiatric disorders aggregate in families and that genetic variation contributes to that familiality. The heritability of a disorder refers to the proportion of variability in disorder risk (in a given population at a given time) that is attributable to individual differences in genetic sequence. Twin studies have documented significant heritability across the spectrum of psychopathology, with estimates ranging from 20% to 45% for anxiety disorders, obsessive-compulsive disorder, posttraumatic stress disorder, and major depressive disorder; from 50% to 60% for alcohol dependence and anorexia nervosa; and from 75% upward for autism spectrum disorder (ASD), attention deficit hyperactivity disorder (ADHD), schizophrenia, and bipolar disorder (1, 4). More recently, with the availability of genome-wide genotype data and newer statistical methods, it has become possible to estimate heritability directly from DNA variation. Estimates of heritability based on single-nucleotide polymorphisms (SNPs) are substantially lower than the estimates from twin studies, at least in part because the SNP data only capture common variants. Nevertheless, these data continue to document significant heritability for common psychiatric disorders.

Genomic studies have succeeded in identifying specific genetic risk loci that are robustly associated with risk of psychiatric disorders. For more than a decade, from about 1996 to about 2006, a vast number of genetic studies tried and failed to identify associations between DNA polymorphisms and psychiatric disorders. This failure, which led some to conclude that the quest was hopeless, was the result of at least two methodological limitations. The first was that (by necessity, given existing genetic technology) these association studies were restricted to a limited set of polymorphisms in “candidate genes”—that is, genes selected on the basis of biological hypotheses about disorder etiology. For example, genes encoding neurotransmitter receptors and enzymes involved in the neurotransmitter synthesis and catabolism were commonly tested candidates. Given our poor understanding of the pathogenesis of psychiatric disorders, any of these cherry-picked candidates were unlikely to be truly involved. Indeed, a recent careful and well-powered analysis of the leading candidate genes for depression from this era (5) showed no support for any of these loci acting alone or in interaction with environmental risk factors. Second, there was a vast overestimation of the effect size that such loci were likely to have. More recent studies have demonstrated that the majority of risk loci have individually small effects whose detection requires sample sizes that are orders of magnitude larger than those examined in the earlier studies. Since 2007, the combination of genome-wide approaches and much larger sample sizes has finally produced robust and replicable associations between DNA variations and a range of psychiatric disorders, including more than 100 variants significantly associated with schizophrenia and with major depression (6–10).

The genetic architecture of psychiatric disorders is highly polygenic and includes the full spectrum of DNA variation. The genetic architecture of a trait refers to the full complement of contributing risk variants, as well as their frequency and effect sizes in a given population. Genome-wide association studies (GWASs) of psychiatric disorders have convincingly demonstrated that a major share of genetic influences on psychiatric disorders is attributable to common DNA variants (SNP alleles carried by more than 1% of the population). These risk variants, which may number in the thousands for a given disorder, each confer small effects (with typical odds ratios on the order of 1.05–1.15) and are spread across the genome. Larger effects are seen when common variant risk is combined into polygenic risk scores (PRSs). A PRS is constructed by multiplying, for each variant, the number of risk alleles a person carries by the effect size of that variant (obtained from an independent GWAS) and then summing across all variants included in the score. It can thus represent a useful summary measure of a person’s total genetic risk burden (or at least the component attributable to common variants). At the same time, other classes of DNA variation have also been shown to contribute to risk, to varying degrees, depending on the disorder. These include rare copy number variations (CNVs) that involve the deletion or duplication of large stretches of DNA (e.g., 500 to 1,000 kilobases) and often encompassing many genes. In particular, rare CNVs have been associated with disorders thought to be neurodevelopmental, such as ASD, ADHD, and schizophrenia (11–13). In addition, rare single variant mutations have been identified for ASD, Tourette’s syndrome, and, to a lesser degree, schizophrenia (14–16). These rare risk CNVs and mutations typically occur de novo, arising in the gametes or embryo, rather than being inherited from parents, and they often have a much larger effect than that seen with individual common SNPs. A recent study in the Journal (17) demonstrated that CNV burden and PRS additively contribute to schizophrenia risk.

Genetic discoveries are providing new windows onto the underlying biology of psychiatric disorders. Beyond simply identifying and enumerating risk variants, a major dividend of recent genetic findings is their ability to provide a new understanding of the pathogenesis of psychiatric disorders. An early success in this context was the discovery that functional variation in complement (C4) genes underlies the strongest GWAS signal for schizophrenia and that the risk alleles have an impact on microglia-mediated synaptic pruning (18). Network and pathway analyses using genomic and transcriptomic data are increasingly implicating biological pathways as well as the cell types and developmental time windows in which they act (19–21). As biological mechanisms and pathways come into focus, they provide crucial new leads for drug development and potential repurposing of existing agents that modulate these pathways (22, 23). Such data are particularly valuable in light of evidence that drug targets supported by human genetic evidence of association with disease are twice as likely to progress from phase 1 studies to successful approval (24, 25).

Genetic influences on psychiatric disorders transcend our clinical diagnostic boundaries. One of the most intriguing themes emerging from genomic studies has been the finding that risk genes and pathways are shared across a range of psychiatric disorders. At the level of DNA sequence, there is substantial overlap among disorders (26), with genetic correlations as high as 0.70 between schizophrenia and bipolar disorder, for example. Recent work from the Cross-Disorder workgroup of the Psychiatric Genomics Consortium (PGC) found more than 100 loci associated with two or more (and up to eight) psychiatric disorders; pleiotropic loci were particularly enriched among genes involved in neurodevelopment (27). Such findings demonstrate that our clinical nosology does not align with underlying biology.

As this brief summary of psychiatric genetics suggests, a great deal of progress has been made in the past decade. But as with most things in science, the more you know, the more you know how little you know. The complexity of the genetic architecture of psychiatric disorders remains largely unexplored; the full spectrum of variants and their penetrance across development are yet to be defined. The emerging application of large-scale whole exome sequencing and whole genome sequencing promises to provide crucial new insights. Other outstanding questions relate to the phenotypic dimensions that genetic risk variants affect. If genetic influences do not map cleanly onto clinically defined disorders, might they act in part through cross-diagnostic phenotypes? How does genetic liability interact with environmental risk factors such as trauma and adversity to affect biology and clinical phenomena? These and other issues will be important targets for future research. This issue of the Journal includes several articles that address some of these unknowns—in particular, providing new clues about genetic architecture and how genes affect psychiatric vulnerability through a transdiagnostic and developmental lens.

Two of these papers advance our understanding of the genetic architecture of psychiatric disorders. Davis and colleagues (28) provide compelling evidence that the genetic basis of autism-related traits includes structural variants that have previously been missed by genotyping and whole exome sequencing assays. They focus on highly variable copy number polymorphisms referred to as the Olduvai protein domain family (formerly called DUF1220). As Davis et al. describe, Olduvai domains comprise variable copies (up to 300 in the human genome) of DNA sequence encoding approximately 65 amino acids that are primarily found in the NBPF gene family. Olduvai domain copy number has expanded substantially in humans relative to other primates, and copies have been subtyped into conserved (CON1–3) and human lineage specific (HLS1–3) groups. Using whole genome sequence data from simplex and multiplex autism cases as well as improved methods for measuring Olduvai copy number, the authors were able to dissect previously obscure contributions of this genomic variation. They combined data from three data sets, including a new replication sample of 215 individuals from multiplex families (two or more affected siblings). Increasing copy number of CON1 was associated with greater autism symptom severity on both the social diagnostic score and the communicative diagnostic score of the Autism Diagnostic Interview–Revised. These effects were also seen in the combined sample for individuals (N=460) in multiplex families, but not in the smaller sample (N=64) of individuals from simplex families. The authors then derived a latent genotype representing components of the NBPF1 gene from CON1 domains and the NBPF14 gene from HLS1 and a latent phenotype that mainly comprised variation in social severity scores. The latent genotype explained 6.2% of the variance in the latent social phenotype. These results suggest that variation in both of these genes underlies the association between Olduvai domains and autism seen in multiplex families. In light of previous evidence that expansion of Olduvai copy number is associated with brain size, neuron number, and cognitive function, the authors speculate that there may be an evolutionary trade-off between adaptive brain growth and neurodevelopmental disease. Notably, decreased copy number has been associated with both microcephaly and schizophrenia. Thus, expanded Olduvai copy number may have been advantageous in primate brain growth, but excessive copy number may contribute to autism and reduced copy number may contribute to schizophrenia (29). If so, the authors suggest, autism and schizophrenia may represent opposite ends of a cognitive disease continuum related to increased or decreased Olduvai copy dosage effects on neurogenesis. The validity of this hypothesis will require further neurobiologic studies. In any case, this work adds to a growing catalog of common variants, rare mutations, and rare structural variation implicated in the complex genetic architecture of autism (7, 16, 30, 31).

Another demonstration of how genomic data now allow us to address fundamental questions about the genetic architecture of psychiatric illness comes from Escott-Price and colleagues (32) in this issue. The authors address a puzzling set of observations. Schizophrenia is associated with reduced fecundity (i.e., reduced number of offspring among affected individuals). We might expect, then, that variants contributing to the disorder would remain at low frequency, as they would be subject to negative selection. And yet we know that a substantial component of the genetic liability to schizophrenia is attributable to variants that have remained common in the population. Could it be that common alleles conferring risk for schizophrenia persist at high frequency because they also confer reproductive advantage in unaffected individuals? The availability of PRSs, which index common variant contributions to a phenotype, now permits an empirical test of this hypothesis. Escott-Price et al. apply a schizophrenia PRS (derived from two large-scale GWASs of the disorder) to nearly 140,000 unaffected participants in the UK Biobank study and look for any association with participants’ number of offspring. They find a weak, though statistically significant, positive association between schizophrenia PRS and number of offspring. The magnitude of the effect was consistent with that seen in a previous study of Icelanders (33), although it was not statistically significant in that study. Interestingly, the effect in the UK Biobank data was mainly driven by the association of genetic loading for schizophrenia alleles and an increase in number of children among women, but not among men (and, in fact, men with the highest genetic loading tended to have no offspring). However, the authors go on to show that if schizophrenia risk alleles do confer reproductive advantage at a population level, the effect they see is too small to account for the persistence of common risk variants. That is because, at a population level, the negative effect of schizophrenia itself on fecundity is many times greater than the modest reproductive advantage the authors observed for schizophrenia risk alleles. As the authors note, their findings are based on data from a particular population at a particular time in human history, but the study nicely illustrates how new large-scale genomic resources can be used to test hypotheses that have previously been subject only to thought experiments.

Psychiatric disorders typically begin early in life, with onset of half of cases by age 14 (34). An increasing body of evidence suggests that trajectories of risk emerge from both genetic and early-life environmental risk factors that may begin as early as fetal life (21, 35). Characterizing early risk profiles may hold important clues for understanding the developmental impact of genetic vulnerability and identifying individuals at increased risk, which in turn may inform the design of prevention strategies. The availability of PRSs indexing disorder liability provides one tool for such efforts. Recent reports have shown, for example, that polygenic liability to schizophrenia is associated with impairments in neurocognition, social cognition, and behavior by early to middle childhood (36, 37). In this issue of the Journal, Halldorsdottir and colleagues (38) examine whether a PRS derived from GWASs of depression in adults (39) is associated with risk of depression-related phenotypes early in life. The authors tested seven PRSs based on progressively more liberal p-value thresholds for including SNPs from the adult depression GWASs and found evidence of association in a small clinical sample of case and control subjects with a mean age of 15. The PRS explained up to 5% of the variance in case-control status, which is numerically larger than that reported in the original adult samples (1.9% on the liability scale). The authors also observed modest association of the PRS for depression severity and age at onset. Self-reported exposure to childhood abuse was independently and additively associated with risk of depression. The authors then extended their findings to two prospective cohorts. In the first, constituting 694 adolescents followed for up to 2 years, the depression PRS (at the more liberal p-value thresholds) was modestly associated with self-reported depressive symptoms at baseline and the emergence of moderate to severe symptoms levels prospectively. An association between the depression PRS and parent-reported depressive symptoms was also seen in a second cohort of children studied at ages 8 and 11. The study examined relatively small samples, although it is notable that some degree of association was seen despite this. Taken together, the results suggest that the effects of genetic contributions to adult depression are expressed as early as age 8 and support a growing body of evidence that PRSs are robust biomarkers of risk. However, they also highlight the complexity of depression. The effects observed were modest, and although the authors state that “PRS prospectively predicted” depression, this is not prediction in any clinically meaningful sense. The PRS area under the receiver operator curve, a common index of prediction performance, was less than 0.60 for incident depression in the primary prospective cohort, and the effect size (hazard ratio=1.54) was smaller than that observed for exposure to childhood abuse in the same cohort. Finally, the samples examined in the study were all of European ancestry, so the generalizability of these results to individuals of other ancestries is unknown. It has now been demonstrated that PRSs derived from European ancestry samples have substantially attenuated effects in other groups, and there is an urgent need for greater diversity in genetic samples before these tools can be applied in clinical settings (40).

This issue of the Journal includes another report demonstrating the value of combining genomic and developmental approaches to advance our understanding of psychopathology. As noted earlier, genetic influences often do not map neatly onto our diagnostic categories. There is growing interest—embodied, for example, in the National Institute of Mental Health’s Research Domain Criteria effort—in reconceptualizing psychiatric diagnosis through a more bottom-up understanding of its component domains and dimensions. In this issue, Riglin and colleagues (41) address the dimension of irritability, a feature of multiple psychiatric disorders and a common source of distress and impairment. While psychiatric nosology has constructed diagnostic groupings around core emotions like fear/anxiety and sadness/depression, it has largely treated anger/irritability as a symptom of other primary disorders, including depression and neurodevelopmental disorders. Riglin and colleagues set out to determine whether distinctive subtypes can be resolved within the heterogeneous phenotype of irritability. To do this, they took advantage of a unique prospective birth cohort, the Avon Longitudinal Study of Parents and Children (ALSPAC), for which genetic and longitudinal phenotypic data are available. They used growth mixture modeling to identify distinct developmental trajectories of irritability measurements across ages 7 to 15 among nearly 8,000 participants. Their modeling identified five trajectory classes: 1) a “low” trajectory, comprising 81% of the sample, with persistently low levels of irritability across childhood and adolescence; 2) a “decreasing” trajectory (5.6%) with declining levels after a peak in early childhood; 3) an “increasing” trajectory (5.5%) that showed rising levels after middle childhood; 4) a “late-childhood limited” trajectory (5.2%); and a 5) high-persistent trajectory (2.4%). These classes were associated with differing patterns of internalizing and externalizing disorders, as well as associations with sex and genetic loading for ADHD and depression. In particular, the early-onset high-persistent trajectory was associated with male sex, childhood ADHD, and ADHD genetic risk (PRS). In contrast, the later-onset increasing trajectory was more common among females and was associated with adolescent depression and both depression PRS and ADHD PRS. These findings imply the existence of at least two interpretable subtypes of irritability—which the authors refer to as an early-onset “neurodevelopmental/ADHD-like type” and a later-onset “depression/mood type.” The profiles were not as distinctive as this might suggest, however, as both types were associated with ADHD PRS and with increased rates of adolescent depression, oppositional defiant disorder, and conduct disorder. Nevertheless, the authors make a strong case for their conclusion that “development matters” in efforts to parse the heterogeneity of complex psychiatric phenotypes. Current and evolving classifications of psychiatric disorders have taken varying approaches to the status of irritability as a symptom of mood or externalizing disorders, and the work by Riglin et al. suggests that a developmentally informed approach may help resolve this conundrum. More broadly, their work provides a model for how we might inform our nosology using data-driven approaches.

A transdiagnostic framework is also useful for understanding what is arguably the most catastrophic outcome of psychiatric illness: suicide-related behavior. The public health burden of suicide and suicide attempts is staggering. In the United States, suicide is the 10th leading cause of death, and rates have increased by 30% since 2000 (42–44). Although rates have declined in some regions, globally 817,000 deaths by suicide occurred in 2016 (45), and many more attempt suicide each year. Our understanding of the causes and risk factors for suicidal self-injury remains limited, although a genetic contribution has been documented in twin studies and, more recently, GWASs. The rate-limiting step for genetic studies of suicide-related behavior has been the availability of sufficient sample sizes to identify variants of small effect. In this issue of the Journal, Mullins and colleagues (46) leverage the PGC to power the largest genome-wide analyses of suicide attempts to date. As the overwhelming majority of attempts occur in the context of psychiatric disorders, the authors examined case subjects (attempters) and control subjects (nonattempters) across three disorders associated with high rates of suicide-related behavior: major depressive disorder, bipolar disorder, and schizophrenia. In GWAS analyses, they identified three genome-wide significant loci within analyses of major depression (one locus), bipolar disorder (one locus), and a meta-analysis of the two mood disorders (one locus). Underscoring the complexity of the phenotype and the need for even larger sample sizes, the authors were unable to replicate these loci in two independent mood disorder cohorts, and estimates of heritability based on genome-wide SNP data were nonsignificant in each of the three disorders. However, PRS analyses revealed a cross-disorder relationship between suicide attempt and genetic risk of major depression. Specifically, those who attempted suicide carried a greater burden of major depression risk alleles than nonattempters across all three disorders. These results suggest that genetic liability to depression is a biomarker of risk for suicide attempt among patients with mood and psychotic illness, although it captures only a small component of the overall risk. They are also consistent with another recent study in U.S. cohorts showing an association between depression PRS and suicide attempt severity (47).

This issue of the Journal provides a snapshot of the vibrant state of psychiatric genetic research. The expanding array of tools, technologies, and data resources have begun to provide robust findings while also revealing new layers of complexity and unresolved questions. The coming years will continue to deepen our understanding of the genetic architecture of psychiatric disorders with large-scale genome sequencing efforts. The advent of new single-cell transcriptomic, epigenomic, and other “-omics” methods will allow new insights into the biological mechanisms of risk variants and pathways relevant to mental illness. The growth of population-based and longitudinal cohorts, including the Million Veteran Program, the All of Us Research Program, the UK Biobank, and other international biobanks, will provide unprecedented opportunities for genome- and phenome-wide analyses and clinical translation. There is much work to be done, but having learned to walk, the field is beginning to hit its stride.