Anxiety Genetics Goes Genomic

One solution to the sample size challenge has been the pooling of data through meta-analysis and collaborative consortia such as the international Psychiatric Genomics Consortium (PGC) (18). In 2016, the ANGST Consortium reported a GWAS meta-analysis of two anxiety phenotype definitions: 1) a case-control definition comprising 7,016 case subjects with any of the five core anxiety disorders and 14,745 control subjects; and 2) a quantitative factor score indexing anxiety disorder liability (N=18,186) (19). One genome-wide significant locus was identified for each of the phenotype definitions. The PGC recently established an Anxiety Workgroup for meta-analysis of a larger sample, but results have not yet been reported.

A second approach to achieving large sample sizes leverages the growing availability of large-scale population-based cohorts and biobanks that include genomic data and a broad range of phenotypic data. At least four GWASs based on such resources have appeared in the past year, including the report by Levey et al. in this issue of the Journal (1). The iPSYCH study, a large case-control cohort nested within a comprehensive population-based Danish national register, recently reported (20) a large GWAS of anxiety- and stress-related disorders (N=12,655 case subjects and 19,225 control subjects). Case subjects were individuals whose records documented at least one ICD-10 diagnosis of an anxiety disorder or a stress-related disorder (including acute stress disorder, PTSD, and adjustment disorder). The authors found significant association for a locus encompassing PDE4B, a gene previously implicated in schizophrenia. Follow-up experiments suggested reduced Pde4b expression among mice exhibiting anxiety-related behavior in response to chronic social defeat stress.

A still larger GWAS (21) examined anxiety phenotypes available in the UK Biobank, a community-based prospective cohort of 500,000 participants. The authors examined two case definitions: first, lifetime anxiety disorder (N=25,453) based on self-reported history of being diagnosed with any of the five core anxiety disorders and/or likely generalized anxiety disorder based on a self-report measure of anxiety; and second, screening positive for at least moderate symptoms of current generalized anxiety during the past 2 weeks (N=19,012). GWAS revealed significant association for five loci. Notably, one of these, NTRK2, encodes the receptor for brain-derived neurotrophic factor (BDNF), which has long been implicated in anxiety and depressive phenotypes (22). Gene-based analyses (as opposed to analyses of individual SNPs) also revealed association with NTRK2 in addition to several other genes, including GAD2, which encodes a glutamic acid decarboxylase gene that has been implicated in mouse models of anxiety behaviors. Using genome-wide SNP data, this study also found substantial and significant heritability of both the anxiety disorder (25.7%) and current anxiety symptom phenotypes (30.8%), supporting the polygenic basis of pathologic anxiety.

Meta-analysis of data from the UK Biobank, in combination with data from the Genetics of Personality Consortium, also enabled a very large GWAS of neuroticism (total N=449,484) that identified 136 independent loci (23). Using a suite of analytic tools, the authors found evidence linking these SNPs to 599 independent genes. SNP-based heritability was estimated at 10%, and again, the results underscore the polygenicity of anxiety-related phenotypes. Phenotypically, neuroticism is known to be strongly associated with both anxiety and depressive disorders. Using hierarchical clustering, the authors dissected the neuroticism phenotype into two genetically homogeneous item clusters indexing “depressed affect” and “worry,” respectively. Among the 136 identified loci, 32 were found to be significantly associated only with “depressed affect” and 26 only with the “worry” component, suggesting that neuroticism itself may reflect a heterogeneous mixture of depression- and anxiety-related genetic influences.

Finally, as reported in this issue, Levey et al. (1) turned to another unique biobank resource to conduct the largest GWAS of anxiety disorder phenotypes to date. The Million Veteran Program (MVP) is an ambitious observational cohort study of 1 million U.S. military veterans, with phenotypic (from electronic health records and questionnaires) and genomic data. Based on data available for nearly 200,000 participants, the authors conducted GWASs of two anxiety phenotypes. The primary phenotype involved scores on the Generalized Anxiety Disorder 2-item scale (GAD-2), a self-report measure of generalized anxiety and worry symptoms during the past 2 weeks. Responses were treated as a quantitative score (ranging from 0 to 6). A secondary case-control phenotype was derived from self-reported history of diagnosis of “anxiety reaction/panic disorder.” In addition to the size of the cohort, an important contribution of this study is the inclusion of a cohort of African American participants. A previous GWAS of generalized anxiety symptoms focused on Hispanic/Latino adults (14), but apart from this, genetic studies of anxiety have been overwhelmingly based on samples of European ancestry, a situation seen across the GWAS literature (24) and one that risks exacerbating health disparities.

Based on the GAD-2 score, GWAS analyses identified five genome-wide significant loci in participants of European American ancestry and one locus in the smaller African American sample. Two additional loci were significantly associated with the case-control phenotype in European Americans. The authors attempted replication of their top findings in data from the ANGST consortium, the iPSYCH cohort, and the UK Biobank/GPC neuroticism GWAS discussed above. The strongest GAD-2 association was observed in European Americans for SNPs in and around SATB1 and the related antisense gene SATB1-ASI. Replication evidence for this locus was seen in the neuroticism GWAS but not in the ANGST or iPSYCH anxiety GWASs. SATB1 regulates transcription and chromatin structure of several other genes involved in neuronal development. The authors note that one of SATB1’s target genes is CRH, which encodes corticotropin-releasing hormone, a key component of hypothalamic-pituitary-adrenal (HPA) axis function that has been associated with anxiety and stress responses. Intriguingly, the CRH receptor gene CRHR1 was among 31 genes that were associated with the GAD-2 phenotype in gene-based analyses, further supporting a role for genetic effects on generalized anxiety via HPA axis function. Nonsynonymous SNPs in exons of CRHR1 were also associated with neuroticism in the large GWAS of that phenotype (23).

Another genome-wide significant variant was found in an intron of the estrogen receptor gene ESR1, which has also been linked to anxiety/fear behaviors in animal models. This finding is somewhat surprising given that more than 90% of the MVP sample was male; nominal evidence for replication was seen in the iPSYCH sample but not in the other replication cohorts. More consistent evidence for replication (in all three replication samples) was seen for a genome-wide significant variant on chromosome 1 (near LINC01360 and LRRIQ3). A fourth significant locus, in an intron of MAD1L1, was also supported in the iPSYCH and neuroticism GWAS, and another MAD1L1 SNP was associated with the case-control anxiety phenotype. Other variants at MAD1L1 have been associated with bipolar disorder (25) and schizophrenia (26), suggesting that this gene may have pleiotropic effects.

In fact, the existence of pleiotropic loci and genetic overlap among disorders is one of the major themes emerging from psychiatric genetic research more generally (27). Levey et al. explore this in the MVP sample, using available data from a wide range of other GWASs to estimate the genetic correlation between anxiety and other phenotypes. The results show widespread genetic correlation between GAD-2 score and an array of psychiatric (depression, neuroticism, ADHD, and schizophrenia) and nonpsychiatric (including insomnia, lung cancer, obesity, rheumatoid arthritis, and coronary artery disease) traits. Similar patterns of shared genetics have been observed in other anxiety GWASs (20, 21, 23), including a consistent inverse genetic correlation with educational attainment and IQ that remains to be explained. The particularly strong genetic correlation between generalized anxiety and depression seen in these studies (on the order of 0.60–0.80) mirrors results from earlier twin studies in which, for example, generalized anxiety disorder and major depressive disorder were reported to have a genetic correlation as high as 1.0 (28, 29). The implications of such findings for psychiatric nosology are intriguing, although it is unclear to what extent they simply reflect occult comorbidity or overlap in the symptoms used to define these disorders. The MVP study adds to our understanding of the genetic relationship between anxiety and depression by including analyses using a method known as multi-trait-based conditional and joint analysis (mtCOJO) (30) that conditions genetic results for one phenotype on those observed for a second phenotype. When conditioning on GWAS summary statistics for a large GWAS of major depression (31), the authors report that the signals at SATB1 and ESR1 remained significantly associated with GAD-2 scores, suggesting independent effects of these loci on anxiety.

This newest and largest anxiety GWAS adds important elements to the search for anxiety-related genes. In addition to providing evidence supporting specific loci and genes, it contributes a large new resource for further analysis and replication of findings from other studies. The inclusion of participants of African American ancestry represents an important and welcome expansion beyond the usual focus on European-ancestry samples. In fact, the genome-wide significant variant observed in the African American subsample is rare in other populations, which could explain why it has not appeared in other anxiety GWASs to date.

At the same time, this study has several important limitations. First and foremost is the “minimal phenotyping,” a drawback that is also seen in other GWASs of generalized anxiety. The use of two screening questions based on current anxiety symptoms and a single self-report item regarding lifetime anxiety disorder raises questions about the degree to which the results apply to lifetime clinical diagnoses of pathologic anxiety. Second, the sample comprises mostly older male participants; twin studies have suggested that genetic influences on anxiety may vary by age and sex (32–35). Third, the claim of “replication” in other samples is tempered by the caveat that the anxiety phenotypes measured across studies were not the same.

The recent large-scale GWASs of anxiety phenotypes provide tantalizing clues to the genetic basis of anxiety symptoms and disorders. Taking these studies together with previous research, we can now feel confident in drawing several conclusions (Box 1). First, it is clear that anxiety-related traits and disorders, like other psychiatric phenotypes, are heritable and highly polygenic. Second, a small number of specific loci have now shown association at stringent genome-wide thresholds in at least one study and some degree of replication in other GWASs of similar phenotypes. Examples include variants in or around SATB1, PDE4B, MAD1L1, CAMKMT, LINC01360, and TMEM106B (which was identified in the UK Biobank study and replicated in the MVP study). It is fair to say that none of these would have been hypothesized candidate genes, and so, to the extent that they turn out to be true signals, they underscore the value of GWAS in identifying novel biology. Third, genetic influences on anxiety symptoms and disorders transcend our nosologic boundaries. We see substantial genetic correlation between anxiety disorders and normal variation in anxiety-related traits as well as between anxiety phenotypes and other psychiatric disorders, especially depression.

But an even larger number of questions remain to be answered (see Box 1). First, how can we find more of the genetic variation contributing to anxiety-related traits? The loci identified so far represent a tiny fraction of those that are relevant. The most obvious strategy for addressing this is to study ever larger samples. GWASs in other disorders have demonstrated that beyond a certain sample size threshold (often around 15,000 cases), there is a roughly linear relationship between sample size and number of genome-wide significant loci detected. Fortunately, efforts already under way will substantially increase the available data in the next few years. The current MVP analysis comprises only 20% of the ultimate MVP sample, and GWAS data are expected for the full cohort of 1 million. Other community- and population-based biobanks and cohorts linking genomic data to a broad array of phenotypes, including anxiety, are appearing and expanding, including the All of Us Research Program (36), which is enrolling a diverse cohort of 1 million or more Americans. Biobanks in China, Japan, and Taiwan have now also reached substantial size. This will also help address the crucial need to dramatically expand research beyond European ancestry samples if we are to fully characterize the genetic architecture of anxiety and other psychiatric phenotypes. But the interpretability of GWAS results also depends on the phenotypes studied. As noted, anxiety-related traits and disorders are a phenotypically complex mixture of quantitative dimensions and clinically recognized disorders. The available studies have typically relied on a heterogeneous and often limited set of phenotypes. The largest studies often have the most limited measures, reflecting the familiar trade-off between sample size and the feasibility of deeper phenotyping. As a result, harmonizing results for replication or meta-analysis is challenging, and few studies have included gold-standard measures of anxiety disorders themselves. Distinguishing loci that may be specific to any given anxiety disorder has thus not yet been possible. Many of the studies reported to date, based on relatively nonspecific phenotyping, may be primarily detecting the component of genetic risk that is related to a more general anxiety-proneness. Results from the emerging PGC Anxiety Workgroup should help here.

The identification of risk loci is of course only the first step toward understanding how genetic variants act to influence anxiety. Loci discovered by GWAS may be correlated with the true causal variants but are unlikely to be causal themselves. Advances in statistical fine-mapping of GWAS signals (37) are now facilitating the search for the causal culprits. Levey et al. employed one such approach and found, for example, that the lead SNP in ESR1 appeared to be responsible for the signal in that region. However, even if we can identify causal variants, further work is needed to characterize their biological relevance. This can include a range of functional genomic analyses to search for biological pathways enriched in the GWAS results, characterize cell-specific patterns of gene expression, and even leverage model organisms, cellular models, and genome editing to examine phenotypic effects of allelic substitution. Human studies, including neuroimaging, neurophysiology, and other deep phenotyping methods, can also be useful in characterizing the effects of risk variants at the level of observable phenotypes. An alternative avenue of research involves the use of genetic risk variants as biomarkers to stratify patients in terms of risk, course, and treatment response. In recent years, aggregate measures of SNP effects in the form of polygenic risk scores have been used for these purposes across of a range of biomedical disorders. With respect to anxiety phenotypes, however, the evidence to date suggests that available polygenic scores explain no more than 0.5% of the variance in anxiety risk (1, 21).

Clearly, the road connecting genotype to phenotype is a long and winding one. But it is a journey worth taking, as the dividends could be substantial, including the identification of novel treatment targets that might change the landscape of therapies for pathologic anxiety. Currently available drug treatments for anxiety disorders are ineffective for many patients, and no drugs with novel mechanisms of action have been approved in decades. The emergence of large-scale genomic studies of anxiety represents a first important step along the path toward a deeper understanding of this common but complex cause of distress and disability.

Dr. Smoller is an unpaid member of the Bipolar/Depression Research Community Advisory Panel of 23andMe.