We selected HapMap phase II SNPs with minor allele frequency greater than 5% from each of the four populations (CEU, CHB, JPT, and YRI; approximately 2.2 million SNPs per population). We tested for association between SNP variation and expression variation for each of 13,643 genes independently in each population, considering unrelated individuals only. For each population, we employed a linear regression model. To analyze those SNPs potentially acting in cis, we tested only those SNPs located within 1Mb upstream or downstream of the expression probe midpoint (“candidate region approach” 18). For large genes ( > 500 Kb), we also used the transcription start site (TSS) as the center of the 2Mb window, discovering only one additional association relative to those discussed below. To determine significance of the regression p-values, we performed 10,000 permutations of the data independently for each gene (and population) and analyzed in depth those associations significant at the 0.001 permutation threshold (see Supplementary Methods). At this level of significance, we expect approximately 14 genes to have at least one significant association by chance, and we detect 299,
