Advancing psychiatric genetics through dissecting heterogeneity

Abstract

There has been substantial progress in psychiatric genetics in recent years, through collaborative efforts to build large samples sizes for case/control analyses for a number of psychiatric disorders. The identification of replicated trait-associated genomic loci represents a large stride forward in a field where little is known about the biological processes involved in disorder. As researchers build on this early foundation, they are beginning to advance the field towards more fine-grained approaches that interrogate the many sources of heterogeneity within psychiatric genetics that can obscure the identification of genotypic influences on disorder. In this review, we provide a brief overview, across a range of psychiatric diagnoses, of recent approaches that have been employed to dissect heterogeneity to give a flavour of the current direction of the field. We group these into three main categories; tackling the heterogeneity in phenotype that is found within the diagnostic categories used within psychiatry, the many different forms of genetic variation that might influence psychiatric traits and then finally, the heterogeneity that is seen across individuals of different ancestries.

Introduction

In the field of psychiatric genetics, researchers are beginning to gain traction in identifying the genetic variants associated with disorder. There is an ever-growing list of findings from genome-wide association studies (GWAS) for various psychiatric disorders, and this has primarily been achieved through collaborative efforts to build very large datasets in order to gain sufficient statistical power to detect the polygenic signals for these complex phenotypes. Most prominent in this area are the findings for schizophrenia, where in 2014 the Psychiatric Genomic Consortium (PGC) reported 128 independent SNPs in 108 loci associated with the disorder. The identified regions showed enrichment for pathways including glutamate transmission, synaptic plasticity and immune functioning (1). Whilst schizophrenia certainly leads in terms of GWAS signals in psychiatry, recent publications in bipolar disorder (2, with 6 significant loci identified), major depression (3, 4, with 15 and 2 loci identified respectively), and anxiety disorders (5, identifying 1 locus) all demonstrate that efforts to accumulate large sample sizes are effective in identifying the common genetic variants associated with disorder in psychiatry. The increased collaboration between researchers in order to achieve these large cohorts has also helped move psychiatric genetics towards a more open model of working, with many researchers making summary statistics from their GWAS analyses openly available to the wider research community. For example, Table 1 shows the summary data publicly available through the PGC.

Table 1.

Publically available summary statistics from PGC GWAS analyses (largest datasets available per phenotype as of 19/06/2017 at https://www.med.unc.edu/pgc/results-and-downloads)

Disorder	Release	Cases	Controls	Genome-wide significant loci in this cohort
Attention Deficit Hyperactivity Disorder	2017 whole sample	20183	35191	12
	2017 European ancestry only	19099	34194	–
Autism Spectrum Disorder	2015 (intermediate data release)	5305	5305 pseudo-controls	1
Bipolar Disorder	2011	7481	9250	4 (2 did not replicate)
Eating Disorders	2016	3495	10982	1
Major Depression	2013	9240	9519	0
Posttraumatic Stress Disorder	2017 whole sample	5183	15547	0
	2017 European ancestry	2489	7465	0
	2017 African American ancestry	2520	7171	1
Schizophrenia	2014	36989	113075	108

Abbreviations: CNV: copy number variant; GWAS: genome wide association study; PGC: Psychiatric Genomic Consortium; PRS: polygenic risk score; SNP: single nucleotide polymorphism; CONVERGE: China, Oxford and Virginia Commonwealth University Experimental Research on Genetic Epidemiology

As these approaches begin to unveil the genetic pathways involved in psychiatric illness, researchers are beginning to look beyond standard case-control GWAS methods and leverage the heterogeneity within psychiatric genetics to advance our understanding of the aetiology of these disorders. In this review, we highlight recent examples of this, organising these efforts into three main categories. Firstly, there are methods that tackle the phenotypic heterogeneity that is seen within psychiatric diagnostic categories. Secondly, there is heterogeneity in terms of genetic architecture, including common to ultra-rare single nucleotide polymorphisms, structural variants and parent-of-origin effects. Thirdly, there is a population-level heterogeneity, investigating the degree to which findings in specific groups generalise across the globe.

Heterogeneity in Phenotype

Definitions of disorder are a key issue in psychiatry where both limited knowledge of the underlying pathophysiology and high symptomatic heterogeneity between patients have led to debate over the accuracy of the diagnostic categories that we use in the field (Fig. 1). Through genetic analysis of symptom-level details, it is hoped that we can maximize the signal to noise ratio and increase statistical power to detect the relevant risk alleles, but also use this genetic knowledge to refine our understanding of these currently inexact disorder models.

Figure 1.

Phenotypic heterogeneity in psychiatry. Whilst all patients much exceed a threshold of symptom severity to be classified as cases, the pattern and severity of these symptoms can vary between patients, as visualised by both hue and intensity of colour in the figure.

High heterogeneity in symptoms and severity is particularly noted amongst depressed patients and this led to the CONVERGE (China, Oxford and Virginia Commonwealth University Experimental Research on Genetic Epidemiology) study (4) focusing on only recurrent major depression in females within a Han Chinese sample. Two genome wide significant loci (both on chromosome 10) were revealed using sparse whole genome sequencing, and replicated in an independent study from China. A number of genome-wide significant associations have also been identified through a more “brute force” approach from the direct-to-consumer testing company, 23andMe (3). They phenotyped a much larger European-ancestry cohort using self-report on a single item of a web-based survey, asking whether an individual had ever been diagnosed with, or treated for, depression. When this data was meta-analysed with previously published findings, a total of 15 genomic regions were identified as reaching genome-wide significance, although the loci identified in CONVERGE were not significant (potentially due to the differences between the cohorts in both ancestry and phenotyping).

Phenotypic heterogeneity can also be tackled by stratifying patients based on key characteristics. The intention of this approach is that the increased homogeneity achieved outweighs the necessary reduction in sample size, and this can be achieved for quite modest increases in effect size in the smaller sample (6). This was successful in the CONVERGE study, where the genetic signal for SIRT1 on chromosome 10 strengthened when only patients with melancholia (severe depression with pronounced biological symptoms such as sleep and appetite disturbance) were considered. The benefits of this approach were also seen when depressed patients were stratified according to age of onset (7). This method identified a genome-wide significant locus on chromosome 3 associated with adult-onset (>27 years) major depression, and also, through polygenic risk scoring (PRS), found that major depression with an earlier age of onset has greater genetic similarity to bipolar disorder and schizophrenia than adult-onset major depression does.

PRS methods allow us to capture the cumulative polygenic signal for a complex phenotype by summing the weighted effect of trait-associated alleles across many genetic loci. The effect size for each locus is estimated in a discovery sample and the PRS is then calculated for each individual within an independent target sample (8–10). As sample sizes for GWAS cohorts grow, the precision of effect size estimates improve and so PRS become more accurate at capturing cumulative genetic liability for a given trait. These methods have become increasingly popular, and are being used in particular to detect differential genetic contribution to disorder sub-groups.

For example, in major depression cases,

• higher symptom severity is associated with increased PRS for major depression, bipolar disorder and schizophrenia (11),

• typical depression (where patients report decreased appetite and sleep) is associated with increased schizophrenia PRS,

• whilst atypical depression (with symptoms of increased appetite and sleep) is additionally associated with increased PRS for metabolic traits (12).

This heterogeneity by phenotype indicates that subgroups within the broad diagnosis of major depression bear different polygenic loadings and closer investigation of these differences may prove fruitful in understanding the aetiology of disorder. Similar approaches have been used in bipolar disorder to show differences in PRS between clinical subtypes (13) and in schizophrenia to demonstrate gene pathways associate differentially with positive and negative symptom profiles (14).

PRS also may have potential predictive utility. In a cohort of first episode psychosis patients, PRS for schizophrenia predicted whether patients would acquire a later diagnosis for schizophrenia compared with other psychotic disorders (15).

Overall, these findings demonstrate how, despite reducing sample size and hence reducing power, study of phenotypic details can reveal variation in the genetic influences on a psychiatric disorder at the level of the individual variant and when aggregating polygenic risk across the genome. Therefore, approaches that focus on phenotypic heterogeneity can help develop our understanding of the overlapping and distinct aetiological pathways involved in psychiatric illness, could be used to guide the creation of more homogenous samples for increased power in genetic cohorts, and may even be useful in tools for prediction within healthcare settings.

Heterogeneity in Genetic Architecture

The GWAS and PRS approaches outlined above focus on common SNP variants. But there is heterogeneity in the genetic architecture of traits with many different forms of genetic variation that play an important role in psychiatric illness. This includes rarer variants, structural variants and also variation arising from parent of origin effects. Efforts to capture the effects of these different forms of genetic influence are important in not only assessing the full picture in psychiatric genetics but also give multiple routes into unpicking the aetiology of disorder.

As the cost of sequencing drops, there has been a growth of both exome and whole genome sequence studies, capturing these different forms of variation. Reference datasets are critical to ascertain the prevalence of the variant beyond the study sample and so aid interpretation of its putative pathogenic role. Initiatives to create a shared resource of observed variant counts available to all researchers resulted in firstly the Exome Aggregation Consortium (ExAC) database (16), which was then expanded to the Genome Aggregation Database (gnomAD, (http://gnomad.broadinstitute.org; date last accessed June 19, 2017).

The value of this resource is highlighted in two recent psychiatric genetic papers that used the ExAC database to understand the population prevalence of identified rare variants. In one paper looking at de novo mutations in a family-based sample of autism spectrum disorder, the authors used ExAC to show that approximately one third of de novo mutations seen in these families were present as standing mutations in the reference dataset, and therefore do not contribute to risk for neurodevelopmental disorders (17).

In the second example, the ExAC database enabled researchers to focus specifically on population-limited ultra-rare protein-compromising variants within a Swedish schizophrenia case/control sample (18). These ultra-rare variants were more frequently observed amongst schizophrenia cases and this excess was largely within neuronally expressed genes with a synaptic function. There was no evidence of an excess within synaptic function pathways when less rare variants were considered, highlighting the benefit of a well-developed reference resource for more accurate identification of truly rare genetic variants in order to allow more precise inference as to the potentially pathogenic role of these variants in disorder. Previous whole exome sequencing studies have also implicated rare variants as important in schizophrenia (19–22), and two of these studies have highlighted post-synaptic genes (19,22). This aligns with findings from GWAS methods in schizophrenia (1). Whilst one study identified SETD1A using whole exome sequencing in schizophrenia (21), in the Swedish case-control sample, no single gene passed statistical significant on an exome-wide search of rare variants, indicating that, as with common genetic influences, rare genetic influences on schizophrenia are highly polygenic (18).

There has also been progress in understanding the pathogenic role of structural variants such as copy number variants (CNVs) in psychiatric illness. Leveraging the large sample size of the PGC schizophrenia cohort (21,094 cases and 20,227 controls with CNV data available), the CNV and Schizophrenia Working Groups of the Psychiatric Genomic Consortium both confirmed previous reports of an enrichment for CNVs amongst schizophrenics as compared with controls (23), and identified new rare CNVs associated with schizophrenia (24). Again, this burden is focussed in genes with synaptic functions.

When examining the findings from common SNPs, rare variants and CNVs in schizophrenia, it is striking that approaches tackling different types of genetic variation are beginning to give converging results of enrichment in synaptic genes (Fig. 2). If the same biological processes are being affected across different forms of genetic variation, cases arising from rare variants of large effect are likely to have overlapping pathophysiology with cases arising from the cumulative impact of many common variants with small effect.

Figure 2.

Convergence of genetic evidence on common biological pathways. Multiple studies in schizophrenia looking at various different forms of genetic variation have implicated common biological pathways, specifically relating to synaptic function. Further investigation is needed to ascertain whether this convergence is seen beyond these specific examples and look at other psychiatric diagnoses. (1) Genovese et al., 2016, (2) Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014 (3) CNV and Schizophrenia Working Groups of the Psychiatric Genomic Consortium 2016 (4) Tansey et al., 2015.

If we consider the interplay between these different types of variation at the level of the individual patient, there is some evidence in schizophrenia that patients with known pathogenic CNVs also show an increased burden of common risk alleles (25). This suggests that CNVs are only one factor (albeit an important one) contributing to increasing risk for schizophrenia, and that schizophrenics with pathogenic CNVs are not a distinct subgroup from those without, instead sharing overlapping genetic risk factors.

There is also research looking at more subtle forms of genetic architecture such as parent of origin effects where the impact of the genetic variant on a phenotype depends on whether an allele is maternally or paternally inherited. This has been of particular interest within autism research, given the autistic features associated with the imprinted disorders of Angelman and Prader Wili syndromes (26). Recent reports have looked at parent of origin effects for autism across various types of genetic variation including common SNPs and CNVs . In one report focussing on the autism-linked 16p11.2 CNV, researchers observe a bias towards de novo deletions of maternal origin in this region (27), whilst GWAS data revealed maternal genetic effects in SHANK3 and WBSCR17, which have both also been previously linked to autism (28).

Bringing these threads together, there is good evidence of heterogeneity in the genetic architecture of psychiatric traits. But early reports within schizophrenia suggest that these various influences might coalesce on the same biological pathways and combine cumulatively to increase risk within a liability threshold model, rather than reflecting distinct mechanisms to illness in distinct subpopulations of patients. Further work is needed to establish if this pattern holds, both with further replication in schizophrenia and across different psychiatric diagnoses.

Heterogeneity in Populations

The final area in which heterogeneity can be leveraged in order to develop an understanding of psychiatric genetics is by looking across different genetic populations to assess the importance of ancestry and ethnicity and examine whether findings generalise or are specific to populations around the globe. This can reveal aspects of the underlying pathophysiology and is also a critical issue when considering how findings from the field of psychiatric genetics might generalise to be used to help patients.

Heterogeneity between populations is evident at a genetic level. Very rare variants are frequently population-limited (29) and so in order to capture the full impact of rare variants on disease in a global context, it is necessary to study a diverse range of populations. Additionally, inclusion of multiple populations in analyses can help distinguish rare risk variants with incomplete penetrance from false positive associations. This has been demonstrated in the case of hypertrophic cardiomyopathy, where clinical genetic tests misclassified benign rare variation as pathogenic. Researchers used simulations to show that the inclusion of more diverse populations in control cohorts would likely have prevented these misclassifications (30).

Looking at more common variants which are present across populations, the evidence suggests that risk variants identified in one population are relevant in other populations as the biological impact on risk is constant even when the allele frequency is different (31). This finding has been replicated when looking specifically at risk variants linked to schizophrenia (32).

Nevertheless, as the allele frequencies of risk variants vary between populations (33), limitations in terms of sampled populations can lead to limitations in discovery. This is demonstrated by the identification of a variant in an Inuit population from Greenland which influences height both in this population and in Europeans (34). This variant had not previously been identified in GWAS despite its large effect size, because of its rarity in European populations.

In addition to differences in allele frequencies, populations also differ in linkage disequilibrium structure. Where causal variants are consistent across populations, this can be leveraged to assist with fine mapping efforts to identify the causal variants driving GWAS signals and give more precise effect size estimates, as has been shown in type 2 diabetes (35). Given particularly low levels of linkage disequilibrium in African populations (36), samples with individuals of African ancestry may be particularly valuable.

Despite these key issues, the field of human genetics as a whole has been criticised for the lack of diversity of populations examined. In many cases, researchers restricted samples or analyses to single ancestry samples in an effort to avoid issues relating to population stratification. However, a survey in 2016 found 81% of GWAS studies were conducted in samples of European ancestry (37), with the majority of other studies using samples of Asian ancestry. This general pattern is also seen in psychiatry (38), for example, only 3.5% of samples in the large PGC schizophrenia mega-analysis were of East Asian ancestry and the remainder were of European ancestry.

This narrow sampling also impacts PRS approaches, as the predictive ability of genetic risk scores is heavily influenced by the ancestral similarity between the discovery cohort used to build the scores and the target cohort used to test them (39). Martin and colleagues demonstrate how a strong bias towards samples with European ancestry is limiting the power of these methods across a range of different phenotypes. Findings specific to psychiatry replicate this; for example PRS for schizophrenia built from the PGC cohort have lower predictive accuracy of schizophrenia case or control status in non-European samples (15,40). Whilst methodologies have been developed that explicitly model ancestry (41) or linkage disequilibrium patterns (40) to improve performance of PRS, these issues of poor between-population prediction still remain even after adjustment.

There may also be a number of non-genetic influences on heterogeneity between populations that could be overlooked without appropriate diversity in samples. For example, there may be cultural influences on symptom manifestation, systematic differences in diagnosis and misdiagnosis patterns between populations and as environmental exposures vary between groups this can lead to different genetic thresholds for illness (42).

These issues of population heterogeneity highlight that whilst discovery has been driven by an emphasis on building large sample sizes, the global relevance of these discoveries are being hindered by a strong skew towards European ancestry samples

Conclusions

The field of psychiatric genetics has made important steps towards revealing the biological basis of disorder in a field where hitherto there have been many unknowns. Here, we have outlined how researchers have begun to dissect the heterogeneity within psychiatry to advance our understanding of illness. But much work remains, beyond efforts to replicate findings in independent and more diverse cohorts.

Most importantly, further development of this theme will ideally consider the potential clinical value of genetic approaches, framing research questions that can be used to predict individual-level risk and guide treatment choices within the clinic, in a manner that has the maximum potential to improve outcomes for patients across the globe through precision medicine.

Conflict of Interest statement. None declared

Funding

National Institute for Health Research (NIHR) Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health.