Genotyping and imputation

Both discovery and replication cohorts were genotyped on a variety of platforms (Affymetrix 250 K, Illumina 318 K, Illumina 370 K, Illumina 610 k; Perlegen 600 K; Affymetrix 1000 K). Quality control was done in each group separately. The overall criteria were to exclude individuals with low call rate, excess heterozygosity and gender mismatch, and exclude variants that were out of Hardy–Weinberg equilibrium, had low minor allele frequency (MAF) or low call rate (Supplementary Table S2). In EGCUT1 study where the genome-wide data was not available, the two most significant single nucleotide polymorphisms (SNPs) for which a TaqMan assay was available were genotyped (Supplementary Table S4). Imputations of non-genotyped SNPs in the discovery cohorts were carried out within each study using either MACH^{26, 27} or IMPUTE,^{28, 29} and HapMap CEU v21a or v22 as reference (Supplementary Table S2). Genetic imputations in the replication cohorts were performed using MACH, IMPUTE, minimac or BimBam (Supplementary Table S2). Of the three SNPS rs9907432 was genotyped in most replication cohorts, whereas rs9900428 was imputed in all replication cohorts (Supplementary Table S4). The data are available in the GWAS Central database, under the accession number HGVST1836 (http://www.gwascentral.org/study/HGVST1836).