Population structure, migration, and diversifying selection in the Netherlands

Abstract

Genetic variation in a population can be summarized through principal component analysis (PCA) on genome-wide data. PCs derived from such analyses are valuable for genetic association studies, where they can correct for population stratification. We investigated how to capture the genetic population structure in a well-characterized sample from the Netherlands and in a worldwide data set and examined whether (1) removing long-range linkage disequilibrium (LD) regions and LD-based SNP pruning significantly improves correlations between PCs and geography and (2) whether genetic differentiation may have been influenced by migration and/or selection. In the Netherlands, three PCs showed significant correlations with geography, distinguishing between: (1) North and South; (2) East and West; and (3) the middle-band and the rest of the country. The third PC only emerged with minimized LD, which also significantly increased correlations with geography for the other two PCs. In addition to geography, the Dutch North-South PC showed correlations with genome-wide homozygosity (r=0.245), which may reflect a serial-founder effect due to northwards migration, and also with height (♂: r=0.142, ♀: r=0.153). The divergence between subpopulations identified by PCs is partly driven by selection pressures. The first three PCs showed significant signals for diversifying selection (545 SNPs - the majority within 184 genes). The strongest signal was observed between North and South for the functional SNP in HERC2 that determines human blue/brown eye color. Thus, this study demonstrates how to increase ancestry signals in a relatively homogeneous population and how those signals can reveal evolutionary history.

Figure 1

The three PCs showing significant correlations with geography. (a): Map of the Netherlands, its major rivers, and its 26 largest municipalities (population size >100 000 as of April 2012). The gray area represents the highly urbanized Randstad area. (b), (c), and (d): The colors of the points indicate the mean value per postal code of PC1, PC2 and PC3 respectively from the LD-pruned SNP set without long-range LD regions (Panel 3). The plot is based on 4130 unrelated Dutch individuals spread out across 1635 postal codes of their current living address.

Figure 2

F, haplotype block size and F_st with the Luhya people in relation to the PCs. For the first three PCs (a–c, respectively), the Dutch subjects were ordered in an ascending order according to their PC value (=PCs from Panel 3, the LD-pruned dataset without long-range LD regions) and divided into 10 equally sized groups (the first 9 groups with N=444, and group 10 with N=445). For each of the 10 groups, the mean F (inbreeding coefficient) is calculated and plotted with its 95% confidence interval. As an illustration of two related measures, the genome-wide average haplotype block size and the mean F_st with the Luhya people from 1000 Genomes, are plotted as well.