Harnessing Progress in Psychiatric Genetics to Advance Population Mental Health

Abstract

Advances in genomics and neuroscience have ushered in unprecedented opportunities to increase our understanding of the biological underpinnings of mental disorders, yet there has been limited progress in translating knowledge on genetic risk factors to reduce the burden of these conditions in the population.

We describe the challenges and opportunities afforded by the growth of large-scale population health databases, progress in genomics, and collaborative efforts in epidemiology and neuroscience to develop informed population-wide interventions for mental disorders. Future progress is likely to benefit from the following efforts: expansion of large collaborative studies of mental disorders to include more systematically ascertained multiethnic samples from biobanks and registries, harmonization of phenotypic characterization in registry and population samples to extend clinical diagnosis to transdiagnostic concepts, systematic investigation of the influences of both specific and nonspecific environmental factors that may combine with genetic susceptibility to confer increased risk of specific mental disorders, and implementation of study designs that can inform gene–environment interactions.

Such data can ultimately be combined to develop comprehensive models of risks of, interventions for, and outcomes of mental disorders. With its focus on phenotypic characterization, sampling, study designs, and analytic methods, epidemiology will be central to progress in translating genomics to public health.

Advances in genomics and neuroscience have ushered in unprecedented opportunities for our understanding of the biological underpinnings of mental disorders. However, this progress has not yet had a major influence on mental health in the general population. Although methodological advances in the descriptive epidemiology of mental disorders have now led to global information on the prevalence, impact, and service patterns (http://www.who.int/news-room/fact-sheets/detail/mental-disorders), there has been limited progress in translating knowledge on genetic risk factors into reductions in the burden of these conditions in the population. We describe the challenges and opportunities afforded by the growth of large-scale population health databases, progress in genomics, and collaborative efforts in epidemiology, clinical research, and neuroscience to develop informed population-wide interventions for mental disorders.

GENOME-WIDE CASE–CONTROL ASSOCIATION STUDIES

Our understanding of risk factors for several noncommunicable chronic diseases has been transformed by the identification of thousands of genetic markers from genome-wide association studies (GWAS). The international data-sharing initiative on mental disorders, the Psychiatric Genomics Consortium (http://www.med.unc.edu/pgc), has now identified more than 150 genome-wide significant hits for schizophrenia, 19 for bipolar disorder, and 44 for major depression in samples of up to 300 000 individuals.¹ Despite the enthusiasm generated by these advances, translation to public mental health interventions has been challenging. GWAS markers still explain only a small proportion of the variance in the etiology of mental disorders, replication of some of the specific markers has not been forthcoming, and identification of biological pathways indexed by genetic markers has been difficult.² This is not surprising in light of the challenges involved in elucidating the complex pathways to mental disorders.

CHALLENGES

One of the most important impediments to progress in psychiatric genetics has been the lack of validity in the classification of psychiatric disorders in terms of both the arbitrary thresholds imposed on their dimensional underpinnings and the blurred boundaries between purportedly distinct disorders. In fact, one of the most notable findings from psychiatric GWAS has been the extent to which there is genetic correlation among mental disorders, as well as shared polygenic heritability across numerous disorders of the brain. For example, there is growing evidence of a continuum of neurodevelopmental conditions—including autism spectrum disorder, attention deficit disorder, intellectual disability, and psychoses—that has provided new models for characterizing multidimensional phenotypes and their developmental manifestations.² Several ongoing activities to refine psychiatric phenotypes may facilitate our ability to tap phenotypes that are more closely related to their genetic liability. These include identification of domains underlying mental disorders (http://www.nimh.nih.gov/research-priorities/rdoc/index.shtml), consideration of multiple phenotypes simultaneously, identification of endophenotypes, and application of sophisticated statistical methods using machine-learning approaches to identify disease subgroups. Likewise, phenome-wide methods and multitrait GWAS approaches are also being employed to identify shared loci across diverse phenotypes.³

Another challenge in translating GWAS to public health and clinical psychiatry is the substantial gap between single-nucleotide polymorphism (SNP)-based heritability (h²_SNP) and phenotypic heritability (h²_phenotype) for many psychiatric disorders. SNP-based heritability is substantially lower than is phenotypic heritability for bipolar disorder, schizophrenia, and major depression: h²_phenotype = 0.54 vs h²_SNP = 0.21, h²_phenotype = 0.57 vs h²_SNP = 0.24, and h²_phenotype = 0.31 vs h²_SNP = 0.12, respectively.⁴ This disparity can in part be attributed to the fact that GWAS do not examine the full range of genetic variation, such as rare or structural variants,⁵ and they do not reflect sources of complexity of the genetic architecture of the diseases, such as pleiotropy (multiple phenotypic effects of single variants), environmental exposures that have been shown to have a potent influence on mental disorders, or epigenetic influences induced by postconception environmental exposures.⁶

Similar to genetic risk factors, environmental factors are also complex. They are characterized by a lack of specificity, dynamic influences across time and development, and subjectivity and bias in measurement.⁷ Most exposures that have been identified for mental disorders (e.g., perinatal risk factors, socioeconomic status, dietary factors, urban residence, stressful or traumatic life events) have also been implicated for other chronic diseases, are highly interdependent, and may reflect effects rather than causal mechanisms. There is a growing effort to establish exposure-wide assessments that can be used in both exploratory studies and prospective confirmatory research with prespecified hypotheses.⁷ The implementation of safeguards—such as more stringent statistical significance thresholds and false discovery rates, prioritizing effect size over significance, and applying alternatives to traditional null hypothesis testing (e.g., use of Bayesian methods)—may enhance our ability to identify specific and nonspecific environmental risk factors for mental disorders, particularly those that may be amenable to intervention.⁷

Advances in our specification of psychiatric phenotypes and identification of the role of environmental exposures across development will facilitate our ability to apply functional genomics to obtain a deeper understanding of causal pathways to mental disorders. There is an array of approaches to identify the biological factors indexed by GWAS and other genetic markers—including whole genome sequencing,⁵ gene set enrichment and pathway analysis,⁸ and studies of tissue-specific expression using expression-quantitative trait analyses to predict complex trait gene targets⁹—that can inform risk prediction and the etiology of psychiatric disorders.

OPPORTUNITIES FOR POPULATION MENTAL HEALTH

Despite the current challenges, there are many opportunities for population mental health. The development of population registries, biobanks, and research consortia will provide larger, more generalized samples for genomics research. These robust samples will provide increased power to estimate the genetic influence on mental disorders and promote the application of innovative study designs to disentangle the multifactorial risk factors for mental disorders. Ultimately, the greatest opportunity is that of collaborative science; cross-disciplinary efforts will be required to conceptualize and identify mechanisms for the development of mental disorders to enhance our ability to define risk and intervene in their incidence and progression.

Population Registries, Biobanks, and Consortia

The samples in most GWAS in psychiatric genetics have not been collected on the basis of systematic ascertainment of either cases, which have been identified primarily in a mixture of clinical settings, or controls, which have often been aggregated from different sampling bases. The increasing focus on the systematic collection of genetics data from general population samples, disease registries, and health care system databases has provided larger and more representative samples with more comprehensive information on phenotypes, genotypes, treatment patterns, laboratory measures, environmental exposures, and health behaviors than have existing studies. These samples will afford greater power to incorporate confounding, which has been one of the most important challenges to approximate causal inferences regarding disease pathways.¹⁰

Population-based registry data, particularly in the Scandinavian countries, have made a growing contribution to our understanding of mental disorders because of their large sample size, the representativeness of their target populations, and creative analyses that link data across multiple domains to identify risk factors and targets for prevention and interventions for mental disorders. Registries in Denmark and Sweden have been particularly informative, not only in estimating both phenotypic and genotypic heritability of common mental disorders but also in examining their genetic and environmental overlap⁴ and the effect of treatment on course and outcome.¹¹

Biobanks have also provided an unprecedented amount of population health data (e.g., behavioral data, laboratory measures, treatment history), which have generally not been available in most case–control genetics studies. Data from the UK Biobank (http://www.ukbiobank.ac.uk/about-biobank-uk) have been particularly informative in our understanding of some of the core features and associated conditions with mental disorders (e.g., cross-disorder cognitive function).¹² There are several well-established health care registries in the United States, such as the pioneer Kaiser-Permanente registry (http://www.dor.kaiser.org/external/DORExternal/rpgeh/index.aspx), the Mayo Clinic (http://mayoresearch.mayo.edu/mayo/research/center-for-individualized-medicine/programs.asp), the Geisinger MyCode Project (https://www.geisinger.org/mycode), the Vanderbilt Precision Medicine Resource (https://www.vanderbilthealth.com/personalizedmedicine/47371), and the Partners HealthCare Biobank (https://biobank.partners.org). The large and diverse enrollment of the Kaiser registry has particularly facilitated evaluation of race/ethnicity in health behaviors and diseases¹³ that may be a major confounder in current research on mental disorders.

The limitations of registries and biobanks should also be considered in evaluating the generalizability and validity of study findings.¹⁴ Cases may not be representative of the general population if the registry targets a specific condition because of the lack of representation of common, less severe, and chronic conditions in service settings. There may also be limited information on symptoms, clinical history, comorbidity, behavioral risk factors, and greater proportions of missing or misclassified data than in planned research. To maximize the value of biobank and registry data on mental disorders, there is a need for a concerted effort to expand information collected on neuropsychiatric disorders beyond categorical diagnostic codes or brief current symptom checklists that tap only the present state. For example, an expert working group has recently added a mental health assessment to the UK Biobank that was vetted through patient samples to address this issue.¹⁵ In US registries, the longitudinal tracking of mental disorders may be particularly limited because they are built on regional health systems rather than on a national registry. Therefore, membership in a registry may depend on employment and health insurance coverage.

Estimating Genetic Influence

Progress in genomics has also spurred the development of new statistical approaches that can be used to incorporate genetic markers in risk analyses. These methods are increasingly being used to estimate the causal influences of SNPs or genes, interrelationships between traits, the extent of genetic influence on traits, shared genetic influences on traits and diseases, genetic risk prediction, and the detection of sources of genetic heterogeneity.⁸ One of the first approaches to incorporate the identification of susceptibility alleles with known effects on specific phenotypes in epidemiologic analyses was Mendelian randomization.¹⁶ Because genotypes are randomly transmitted across generations according to Mendel’s laws, genotypes may be used as the independent variable in study designs that seek to identify influences of modifiable environmental exposures. Genetic randomization eliminates the effects of confounding and reverse causation that require consideration in observational epidemiologic studies. However, this approach requires a priori identification of susceptibility markers with influence on the disease or outcome of interest, which has not been forthcoming for psychiatric disorders. There are also several challenges in the application of Mendelian randomization in the context of complex genetic mechanisms, such as pleiotropy, intercorrelations among multiple health risk factors, and heterogeneity of disease or trait outcomes.^3,17

Two other commonly used methods that incorporate GWAS data to estimate the role of genetic markers and diseases and traits are polygenic (or genomic profile) risk scores and genome-wide complex trait analysis.⁸ Polygenic risk scores summarize genetic effects in a GWAS by computing a weighted sum of associated risk alleles within each subject. Initially, markers (typically SNPs) are selected on the basis of their evidence for association (typically the beta or odds ratio and associated P value) using a training sample, and the weighted score is then constructed in an independent replication sample. If an association is found between a trait or disorder and the polygenic risk score, one assumes that a genetic signal is present among the selected markers. Later, this score can be used to predict individual trait values.⁸ Polygenic risk score analysis can be used to detect shared genetic etiology among traits or disorders, to infer the genetic architecture of a trait, and to establish the presence of a genetic signal in underpowered studies.³

Genome-wide complex trait analysis estimates the proportion of phenotypic variance explained by genetic variants for complex traits by estimating genetic relatedness directly from the SNP data: so-called SNP-based heritability (h²_SNP). In addition to defining the genetic relationship from genome-wide SNPs, genome-wide complex trait analysis can be used to examine the genetic correlation between two traits or diseases using a bivariate SNP-based model that estimates the average genome-wide relationship between two disorders.⁸ This approach has recently been applied to phenome-wide data in the UK Biobank.¹⁸ Limitations of these methods that should be considered in their utility and interpretation include their reliance on the validity of GWAS findings, a lack of coverage of less common alleles in GWAS, nonreplications of some GWAS findings, and difficulty in determining coherent underlying gene networks or their biological pathways.⁸

Study Designs

The bulk of large-scale psychiatric genetics research has not yet incorporated the role of well-established environmental factors that have been shown to contribute to mental disorders. To date, most research on mental disorders designed to link genetic and environmental risk has incorporated the candidate gene by environment interaction approach on the basis of single candidate genetic markers and environmental factors that have not been replicated in subsequent large-scale efforts.⁸ Advances in genomics may ultimately enhance our ability to overcome some of the challenges involved in elucidating causal inferences on these multifactorial influences on disease pathways through the identification of markers with greater attributable risk and elucidation of their underlying biological mechanisms.

The counterfactual approach of Mendelian randomization is likely to have more utility with progress in identification of genetic variants, availability of large, well-characterized samples across development, and greater insight into risk factors for specific mental disorders.¹⁰ The application of the Mendelian randomization methodology in the design of case–control studies is likely to improve our ability to define specific environmental risk factors and intervene in disease onset and progression. For example, Mendelian randomization designs have been recently employed to examine the relationship between cannabis and schizophrenia,¹⁹ tobacco smoking and mental health symptoms,²⁰ and immunologic markers of susceptibility to infections and schizophrenia.²¹

The resurrection of the extended family study, particularly as an extension of case–control GWAS, is also a promising direction for the next generation of genetic studies of psychiatric disorders.²² Family studies can minimize the false-positive risk induced by population stratification, increase the rate of genetic susceptibility variants, identify disease subtypes, and reduce heterogeneity that plagues current GWAS case–control studies. Within-family designs that condition on either genetic or environmental factors may help to identify sources of the polygenicity and heterogeneity of the psychiatric disorders as well as the indisputable role of potential environmental influences.

The utility of progress in genomics for interventions will depend on the elucidation of the causal pathways underlying mental disorders and their treatment. This will involve a broad array of immunologic, metabolic, and pharmacogenetic biomarkers with specificity for particular clinical entities or drug responses that can be used to develop individualized diagnostic profiles and interventions. Even though evidence for the application of the tools of pharmacogenomics to inform the heterogeneity of treatment response for mental disorders is lacking,²³ future progress in genomics is likely to improve our ability to identify markers of drug response and side effects (https://ispg.net/genetic-testing-statement). The characterization of etiologic pathways should also inform prevention efforts, particularly for malleable risk factors such as diet, drug exposures, and risky health behaviors in combination with susceptibility genotypes.

Collaborative Science

With its focus on phenotypic characterization, sampling, study designs, and analytic methods, epidemiology will be central to progress in translating genomics to public health. The box on this page summarizes approaches to the sampling, design, measurement, and analysis of future studies that can inform mental health etiology and prevention in the genomics era. The integration of genetic risk factors with environmental influences will require the collaborative efforts of epidemiologists and other population and behavioral scientists with molecular and clinical geneticists, bioinformaticians, statistical geneticists, neuroscientists, and developmental neurobiologists, among others. The team science model that has been so successful in advancing the application of genomics to mental disorders serves as a model of future cross-disciplinary collaborative efforts that will be required to conceptualize and identify mechanisms for the multifactorial influences on mental disorders to enhance our ability to define risk and intervene in their incidence and progression. The need for larger samples with standardized methodology will involve large-scale networks within domains as well as across relevant domains—including clinical phenotypes (https://bioportal.bioontology.org/ontologies/HP), neuroimaging (e.g., ENIGMA, at http://enigma.ini.usc.edu), environmental factors,^7,24 and treatment consortia (http://www.conligen.org)—that can be used to develop comprehensive models of risk, outcomes, and interventions with mental disorders. Finally, our field can profit from models of prevention and intervention that have emerged from integrative research that has informed risk factors and treatments for other chronic disease phenotypes (e.g., diabetes).²⁵

TOOLS FOR INCORPORATING PROGRESS INTO GENOMICS IN FUTURE RESEARCH ON MENTAL HEALTH.

Samples	Enrich biobanks and registries
	Collect systematic samples of cases and controls defined by genotypes
	Increase inclusion of multiethnic samples
	Combine age-specific samples to construct life-span assessments
Phenotypes	Move beyond dichotomous classification
	Incorporate endophenotypes into genome-wide association studies
	Include multiple disorders in case–control studies to evaluate specificity
Measures	Augment clinical data with standardized supplementary measures of domains underlying mental disorders
	Collect repeated measures, including in-time mobile assessments when relevant
Study designs	Collect retrospective and prospective cohort studies
	Incorporate families in case–control and registry studies
Statistical methods	Include genes as independent variables: Mendelian randomization
	Calculate polygenic risk scores and genome-wide complex trait analysis
	Apply classification models to define phenotypic subtypes and overlap
	Employ models that incorporate clustering and specificity of genes and environmental exposures
Environmental factors	Stratify genetic case–control studies by environmental exposures
	Design studies that test specific vs general effects of environmental risk factors
	Identify “critical” timing environmental exposures
Collaborative efforts	Include epidemiologists in genomics collaborative networks
	Establish networks on relevant domains, including biobanks, registries, environmental exposures, treatment and prevention, neuroimaging, and behavioral measures