Pleiotropic heritability quantifies the shared genetic variance of common diseases

Common diseases are highly pleiotropic, but the overall contribution of pleiotropy to a target diseases architecture is unknown, as most studies focus on genetic correlations with each auxiliary disease in turn. Here we propose a new method, pleiotropic heritability with bias correction (PHBC), to estimate pleiotropic heritability (h2pleio), defined as the liability-scale genetic variance of a target disease that is shared with a specific set of auxiliary diseases. We estimate h2pleio from GWAS summary statistics by estimating the proportion of variance explained from an estimated genetic correlation matrix and employing a Monte-Carlo bias correction procedure to account for sampling noise in genetic correlation estimates. Simulations showed that PHBC produces approximately unbiased estimates of pleiotropic heritability. The average ratio of pleiotropic heritability vs. total SNP-heritability (h2pleio/h2) across 15 diseases from the UK Biobank (spanning 7 disease categories) was 27% (s.e. 2%). Pleiotropic heritability was broadly distributed across disease categories, with h2pleio/h2 decreasing only slightly when removing all auxiliary diseases in the target disease category (avg = 24% (s.e. 2%)) and only moderately when further removing one other (most informative) category whose removal had the greatest impact (avg = 18% (s.e. 1%)). Average h2pleio/h2 increased to 33% (s.e. 2%) when adding 17 auxiliary quantitative traits in UK Biobank, and 49% (s.e. 4%) when further adding 30 auxiliary diseases from large GWAS meta-analyses--with several diseases dominated by pleiotropic heritability, including depression (74%, s.e. 7%) and type 2 diabetes (59%, s.e. 4%). On average, h2pleio/h2 was 1.51x (s.e. 0.14) larger than the proportion of liability-scale total phenotypic variance explained by the same set of auxiliary diseases, implying higher pleiotropy for genetic effects than the effects of non-genetic exposures. In conclusion, we have uncovered pervasive sharing of genetic aetiologies, with roughly half of common disease heritability being pleiotropic with diseases from a broad range of disease categories, which strongly motivates the importance of multi-disease approaches to risk prediction and therapeutic development.