GWAS: A linear regression model, in the PLINK (Purcell et al., 2007) software package, was used. Genotype was coded log-additively (i.e. increasing copies of the minor allele, selected from the full sample with ethnicities combined). Analyses were conducted separately in the EA (N=2018) and AA (N=1035) subsamples, adjusting for further ethnic differences via the inclusion of 2 principal components (generated via EIGENSTRAT; Price et al., 2006)). Other covariates included age at interview (dummy-coded to represent the lower three quartiles with the oldest age group used as a reference), sex and study source (whether the participants were drawn from COGA, COGEND or FSCD). The results from the EA and AA subsamples were meta-analyzed in METAL (Willer et al., 2010) using inverse-variance weighting.
