The principal components analysis retained data from 1113 of the 1457 SNPs (including AIMs) remaining after data cleaning. As case-non-dependent control differences reached at least a trend level of significance for 4 PCs, a conservative decision was made to include all 4 PCs as covariates in comparisons of these groups. The respective p-values for population differences along these eigenvectors were 1.1 × 10−12, 2.4 × 10−11, 1.1 × 10−4, and 0.064. Although both populations are primarily of European ancestry, the various combinations of PCs 1,2, and 3 (see scatterplots in Supplementary Figures 3A–C) identify individuals of Asian ancestry (i.e., mapping to Han Chinese and Japanese HapMap populations) more prevalent among cases. PC4 appears to be identifying a Northern European ancestry subgroup. A similar examination in cases and neighborhood controls found no significant or trend level population differences and thus no covariates were used for those comparisons.
