After quality control, we meta-analysed genotype and summary statistics samples together using Ricopili version 2019_Jun_18.00141 with HRC v1.1 as a reference panel. We identified genome-wide significant SNPs at p <= 5 × 10−8 and then identified independently associated index SNPs by clumping SNPs with p <= 1 × 10−8, INFO >= 0.6 that were r2 > 0.1 and were within 3 Mb windows of an index SNP. The extended MHC region was considered as a single region for clumping. We ran a conditional-and-joint (COJO) analysis12 on each region to identify SNPs that were associated after conditioning on the index SNP, using UK Biobank as the reference panel, and retained SNPs selected by COJO with joint p <= 5 × 10−8. To reduce variation in per-SNP effective sample size due to missingness across cohorts, we filtered SNPs to those with effective N >= 80% of the maximum effective N (calculated separately for autosomes and chrX). We examined discovery power in comparison to previous GWAS of depression using genpwr.66 To determine novelty of association findings, we looked for overlap in regions from recent MDD meta-analyses2,6,7,54,67 and associations for unipolar depression (EFO_0003761) in the GWAS Catalog.68
