We used Metal 56 to conduct an inverse variance-weighted meta-analysis of the summary statistics from the three studies (using the log of the odds ratios and the standard errors of the log of the odds ratio), conditional on each variant being available in all three of the contributing cohorts. This provided 8,098,588 genetic variants and up to 246,363 cases and 561,190 controls (n = 807,553) for the meta-analysis. Linkage disequilibrium score (LDSC) regression intercepts 10 were used for genomic inflation control of the three contributing cohorts and the final meta-analysis results. Clumping and merging were used to identify the basepair positions of loci containing depression-associated variants. Clumping of the meta-analysis results was conducted using Plink v1.90b4 57 applying the following parameters: --clump-p1 1e-4 --clump-p2 1e-4 --clump-r2 0.1 --clump-kb 3000, with merging of the clumped loci conducted using bedtools 58. A conditional analysis 59 was used to identify independently-segregating variants in each of the merged loci that was genome-wide significant, using the linkage disequilibrium structure of UK Biobank as a reference panel. Genome-wide statistical significance was defined as P < 5
