Population structure in European, African, and Asian individuals. A and B, PCoA results based on average interindividual ASD, with use of 9,100 SNPs. PC1, 2, and 3 explain 11.6%, 3.4%, and 1.4%, respectively, of the variation. C, Bayesian clustering results with use of STRUCTURE14^,15 and the same markers and individuals. Each individual is represented as a vertical line divided into, at most, K colored segments, where K is the prespecified number of populations into which the data are to be divided. STRUCTURE runs consisted of 80,000 iterations, with a previous burn-in of 40,000 steps, and were performed under the admixture model, which allows fractional assignment of the genome to different populations. Because of the historical and geographic proximity of European populations, the correlated-allele-frequencies model was also employed. The STRUCTURE plots shown above were generated using the companion program DiStruct.29

PCoA boxplots for the first six PCs in samples from Europe and neighboring continents. Bold vertical bars represent the median PC values of each group; the two hinges are the first and third quartile, and notches give an ∼95% CI for the difference in two medians. The overall correlation between group membership and PC value is reported by ANOVA’s adjusted r² for each PC. The few subsequent PCs that are also significant pertain to Europe and are best observed in figures 4 and 5.

Stability of the PC1-PC2 distribution of European individuals across marker sets with different minimum intermarker separation (50 kb and 100 kb).

Population structure in European individuals. A, PCoA based on average interindividual ASD across 9,114 SNPs. PC1, 2, 3, and 4 (fig. 5) explain 2.05%, 1.7%, 1.6%, and 1.5%, respectively, of the variation and were highly significant by SKT and ANOVA testing (see table 2). B, Bayesian clustering analysis using STRUCTURE and the same markers and European individuals. Each individual is represented as a vertical line divided into K colored segments, where K is the prespecified number of populations into which the data are to be divided. The clusteredness (measured by G)—the extent to which individuals belong to a single cluster rather than a combination of clusters30—is also given alongside each K value. See the legend of figure 1 for further details of STRUCTURE runs and conditions.

PCoA boxplots for the first six PCs in European samples. Bold vertical bars represent the median PC values of each group; the two hinges are the first and third quartile, and notches give an ∼95% CI for the difference in two medians. The overall correlation between group membership and PC value is reported by ANOVA’s adjusted r² for each PC. The subsequent PCs (up to 10) are not significant. The single French sample and the German outlier were excluded.

Distribution of F_ST between northern (n=36) and southeastern (n=31) cohorts of individuals selected from PC1 values in figure 4 (>0.5 or <−0.5). A, Histogram with use of all 9,721 SNPs available. B, Plot of top 50 SNPs of highest F_ST (also see table 3).

Population structure in panels of European-derived ancestry within the context of European individuals (from fig. 4A). A, PCoA of the Coriell Caucasian panel (n=42), together with Europeans of known ancestry, based on all 9,114 SNPs in common. B, PCoA of the CEPH Utah individuals (n=74) and Europeans with use of all 6,207 SNPs in common. C, STRUCTURE runs using the Coriell Caucasian sample based on the full SNP data set (bottom) as well as sets of different numbers of north-southeast EuroAIMs (available from the Shriver Lab Web site).