We used KGG 2.5 (refs. 28, 29) to conduct gene-based tests of our GWAS results. This gene-based test examines whether the set of SNPs in a gene is associated with the phenotype at a level greater than chance. We used publicly available 1000 Genomes Phase 1 version 3 (European subsample) linkage disequilibrium (LD) files to build the ‘analysis genome' by position, for autosomes only. We included extended gene lengths of 5 kb at both the 5′ and 3′ ends. SNPs in high LD (r2>0.9) were connected; those in low LD (r2 <0.02) were considered independent. We used the hybrid set-based test for genome-wide association studies29 and report both the uncorrected P-values as well as the q-value based on a Benjamini and Hochberg30 false discovery rate. q-values <0.05 are considered genome-wide significant. Following this, we used i-GSEA4GWAS31 to assess enrichment across canonical pathways (defined by the authors of the i-GSEA4GWAS program) and gene ontologies. In total, 23 153 genes were submitted along with P-values (which were log-transformed by i-GSEA4GWAS). We set the minimum number of genes per category to 20, and the maximum to 200.
