It would be desirable to be able to use the full sample of all 210 unrelated individuals to detect genetic effects on gene expression that are common but of smaller magnitude than those detected within each individual population. However, one cannot simply pool the samples without appropriate corrections, since population differentiation will generate spurious associations. Conditional permutations allow us to reveal the relevant associations while masking inflated associations 29,30. We repeated the association analysis after pooling unrelated individuals of: i) all four populations, ii) a subset of three (CEU-CHB-JPT) populations, and iii) two (CHB-JPT) populations. The rationale for the choice of population combinations was to pool those sets of populations that are more closely-related. To correct for inflation of the p-values, we performed conditional permutations, such that expression values from an individual of a given population were only assigned to another individual of the same population. This corrects for the p-value inflation since p-values from permuted datasets are also inflated. A total of 803, 735 and 651 genes were detected as significant for the 4-population, 3-population and 2-population pools, corresponding
