Response to Comment on "Genetic Structure of Human Populations"

Abstract

Excoffier and Hamilton (2) performed a complementary variance component analysis, demonstrating that when a subset of our data corresponding to (3) is studied using allele sizes, as was done in (3), similar estimates to (3) are obtained. Their smaller within-population variance component compared with that in (1) is consistent with the smaller estimate of (3) in relation to microsatellite studies that used indicator variables (4–7). However, because previous indicator-based studies of microsatellites and other markers have not all been in full agreement (1, 4–11), a difference in the nature of the variable cannot be the sole source of differing estimates. First, the homogenizing effect of the higher mutation rates of microsatellites, in contrast with those of other markers, probably explains some of the difference of our results from nonmicrosatellite indicator-based studies (12). Second, consistent with past observations (13), the high fraction of tetranucleotide loci in our data contributes to higher within-population variance component estimates (Table 1) than are seen in dinucleotide studies (3, 4, 7). Third, the estimates vary considerably across sampling schemes within regions, and in several cases (3, 6, 7), past microsatellite samples that included multiple groups per region used populations that are among the most differentiated of the 52 groups in our data (Fig. 1). Any estimate computed with the well-separated populations that contribute to the 83.4% within-population variance component obtained by Excoffier and Hamilton (2) should be regarded as a lower bound.