No statistically significant heterogeneity was observed across individual cohorts for the 30 genome-wide significant loci, as assessed with Cochran’s Q-test (Supplementary Fig. 32), the I2 statistic, and GenomicSEM’s QSNP-statistic16 (Supplementary Table 2). Genome-wide analyses of samples grouped by clinical, comorbid, biobank, and 23andMe (Supplementary Table 3 and Supplementary Figs. 33–37) showed evidence that sample ascertainment impacted results at a genome-wide scale, although not beyond what is observed with closely related psychiatric disorders17,18. We observed moderate to high genetic correlations across the subgroups (between 0.63, s.e. = 0.11 for biobanks and comorbid, and 0.92, s.e. = 0.07 for 23andMe and comorbid; Supplementary Table 7), and a satisfactory fit for a one-factor GenomicSEM model (Supplementary Table 8 and Supplementary Fig. 39). A common-factor GWAS based on the one-factor GenomicSEM model resulted in 20 significant loci, all of which were also significant in the primary GWAS (Supplementary Table 8 and Supplementary Fig. 40; analysis details in Supplementary Note 5). SNP-heritability (assuming a 1% population prevalence) was 6.7% (s.e. = 0.3%), with slightly higher estimates for the clinical (hSNP2=16.4% = 16.4%, s.e. = 1.5%) and comorbid (hSNP2=13.3% = 13.3%, s.e. = 1.7%) subgroups (Supplementary Table 1).
