Charting the Landscape of Genetic Overlap Between Mental Disorders and Related Traits Beyond Genetic Correlation

Abstract

Objective: Mental disorders are heritable and polygenic, and genome-wide genetic correlations (r_g) have indicated widespread shared genetic risk across multiple disorders and related traits, mirroring their overlapping clinical characteristics. However, r_g may underestimate the shared genetic underpinnings of mental disorders and related traits because it does not differentiate mixtures of concordant and discordant genetic effects from an absence of genetic overlap. Using novel statistical genetics tools, the authors aimed to evaluate the genetic overlap between mental disorders and related traits when accounting for mixed effect directions.

Methods: The authors applied the bivariate causal mixture model (MiXeR) to summary statistics for four mental disorders, four related mental traits, and height from genome-wide association studies (Ns ranged from 53,293 to 766,345). MiXeR estimated the number of "causal" variants for a given trait ("polygenicity"), the number of variants shared between traits, and the genetic correlation of shared variants (r_gs). Local r_g was investigated using LAVA.

Results: Among mental disorders, ADHD was the least polygenic (5.6K "causal" variants), followed by bipolar disorder (8.6K), schizophrenia (9.6K), and depression (14.5K). Most variants were shared across mental disorders (4.4K-9.3K) and between mental disorders and related traits (5.2K-12.8K), but with disorder-specific variations in r_g and r_gs. Overlap with height was small (0.7K-1.1K). MiXeR estimates correlated with LAVA local r_g (r=0.88, p<0.001).

Conclusions: There is extensive genetic overlap across mental disorders and related traits, with mixed effect directions and few disorder-specific variants. This suggests that genetic risk for mental disorders is predominantly differentiated by divergent effect distributions of pleiotropic genetic variants rather than disorder-specific variants. This represents a conceptual advance in our understanding of the landscape of shared genetic architecture across mental disorders, which may inform genetic discovery, biological characterization, nosology, and genetic prediction.

Keywords: Attention Deficit Hyperactivity Disorder (ADHD); Bipolar and Related Disorders; Depressive Disorders; Genetics/Genomics; Neurodevelopmental Disorders; Schizophrenia Spectrum and Other Psychotic Disorders.

Figure 1.

MiXeR model concepts. Features of genetic overlap beyond genetic correlation (r_g) characterized by MiXeR (Venn diagrams). a) Mixed effect directions. While r_g captures genetic overlap with (i.) predominantly concordant or (ii.) discordant effects, it is incapable of differentiating (iii.) genetic overlap with a balance of concordant and discordant effects from (iv.) an absence of genetic overlap, returning an estimate of 0 in both scenarios. In contrast, MiXeR quantifies the number of shared ‘causal’ variants (Venn diagrams) and so identifies genetic overlap also in the presence of mixed effect directions. b) Correlation of shared variants (r_gs). r_g does not differentiate (i.) extensive genetic overlap with a small majority of concordant effect directions from (ii.) smaller overlap with a majority of concordant effect directions, returning weak positive r_g in both scenarios. In contrast, MiXeR-estimated r_gs returns an equivalent estimate to r_g in scenario i) but a higher estimate in scenario ii) (same concept applies to weak negative r_g / discordant scenarios).

Figure 2:

Along the diagonal, univariate MiXeR estimates for each mental disorder. h²_SNP=SNP-based heritability estimate; polygenicity⁹⁰= number (in thousands) of causal variants with strongest effects required to explain 90% SNP-based heritability. MiXeR-modelled genome-wide genetic overlap and genetic correlations (top right) and LAVA local correlations (bottom left) across bipolar disorder (BIP), attention-deficit hyperactivity disorder (ADHD), major depression (MD) and schizophrenia (SCZ). Top right: MiXeR Venn diagrams showing the number of shared and disorder-specific ‘causal’ variants in thousands for each pair of disorders. Genome-wide genetic correlation (r_g) and genetic correlation of shared variants (r_gs) are represented by the colour of the disorder-specific (r_g) and shared regions (r_gs), respectively. All analyses besides MD and ADHD had positive AIC differences when comparing modelled estimates to minimum possible overlap but negative compared to maximum overlap, indicating that MiXeR estimates were indistinguishable from maximum overlap. * For MD and ADHD, both AICs were negative, indicating that the analysis was not sufficiently powered to provide precise estimates of genetic overlap. Variation in polygenicity estimates for SCZ are due to variation in univariate MiXeR results across the 20 iterations. Bottom left: Volcano plots of LAVA local genetic correlation coefficients (rho, y-axis) against -log₁₀-p values for each pairwise analysis per locus. Larger dots represent significantly correlated loci after FDR-correction. MiXeR estimated r_g and r_gs, and LAVA estimated rho are represented on the same blue to red colour scale.

Figure 3:

MiXeR Venn diagrams illustrating MiXeR modelled genetic overlap, genome-wide genetic correlation (rg) and genetic correlation of shared variants (r_gs) between mental disorders and a. cognitive traits: intelligence (INT) and educational attainment (EDU), b. personality traits: neuroticism (NEUR) and subjective well-being (SWB), and c. height. The number of unique and shared ‘causal’ variants are presented in thousands and illustrated by the size of the unique and shared regions of the Venn diagrams. Rg and r_gs are provided beneath each diagram and are represented by the shading of the unique (rg) and shared (ρβ) regions, ranging from −1 (dark blue) to +1 (dark red). The SNP-based heritability (h²_SNP ) for each trait is provided beneath each trait label. All analyses besides NEUR and major depression (MD) and all height analyses had positive AIC differences when comparing modelled estimates to minimum possible overlap but negative compared to maximum possible overlap, indicating that the estimates were indistinguishable from maximum overlap. *For NEUR and MD and all analyses involving height, both AIC differences were positive. However, there was unstable model fit for NEUR and MD compared to minimum possible overlap. These results should be interpreted with caution. Variation in polygenicity estimates for SCZ and SWB are due to variation in univariate MiXeR results across the 20 iterations.

Figure 4:

Scatter plot comparing the proportion of LAVA-estimated independent genetic loci with significant positive genetic correlation against MiXeR-estimated proportion of shared ‘causal’ variants with concordant effects. Each individual point represents an individual pairwise analysis. Mental disorder by mental disorder (yellow) are clustered around high concordance consistent with higher genetic correlation, height by mental disorders/mental traits are clustered around 0.5 concordance, consistent with minimal genetic correlation and mixed effects, while disorders by mental traits and mental traits by mental traits (“Other mental traits”) are distributed across the spectrum of concordance. Both methods emphasise the presence of mixed effect directions across most analyses. However, note that LAVA local correlations were generally more extreme than MiXeR, possibly due to LAVA’s tendency to identify the most strongly correlated loci which are most likely to be significant.