We also investigated whether the extent to which those associations not shared between populations could be attributed to differential allele frequencies across populations as recently reported 16. For each pair of populations we split the associated SNPs (0.001 permutation threshold) into 3 categories: i) SNPs significant for the same gene in both populations (SNP-shared associations); ii) gene associations in both populations but with different SNPs (Gene-shared associated SNPs); iii) population-specific associations (unshared associations). For these 3 categories of SNPs, we computed the difference in expected heterozygosity (2pq) in the same direction (e.g. Hetpopulation1-Hetpopulation2) and compared the distributions of the differences among the three categories. As expected, median difference in heterozygosity was the lowest for SNP-shared associations, with gene-shared associated SNPs exhibiting the second lowest difference (Figure S4). Our results are consistent with the Spielman et al. 16 observation. One small caveat is that because of small sample sizes, there could be slight fluctuations in allele frequencies simply due to sampling variance that may affect detection of associations above a certain threshold.
