Global similarity with local differences in linkage disequilibrium between the Dutch and HapMap-CEU populations

Abstract

The HapMap project has facilitated the selection of tagging single nucleotide polymorphisms (tagSNPs) for genome-wide association studies (GWAS) under the assumption that linkage disequilibrium (LD) in the HapMap populations is similar to the populations under investigation. Earlier reports support this assumption, although in most of these studies only a few loci were evaluated. We compared pair-wise LD and LD block structure across autosomes between the Dutch population and the CEU-HapMap reference panel. The impact of sampling distribution on the estimation of LD blocks was studied by bootstrapping. A high Pearson correlation (genome-wide; 0.93) between pair-wise r(2) for the Dutch and the CEU populations was found, indicating that tagSNPs from the CEU-HapMap panel capture common variation in the Dutch population. However, some genomic regions exhibited, significantly lower correlation than the genome-wide estimate. This might decrease the validity of HapMap tagSNPs in these regions and the power of GWAS. The LD block structure differed considerably between the Dutch and CEU-HapMap populations. This was not explained by demographic differences between the CEU and Dutch samples, as testing for population stratification was not significant. We also found that sampling variation had a large effect on the estimation of LD blocks, as shown by the bootstrapping analysis. Thus, in small samples, most of the observed differences in LD blocks between populations are most likely the result of sampling variation. This poor concordance in LD block structure suggests that large samples are required for robust estimations of local LD block structure in populations.

Figure 1

Distribution of minor allele frequency (MAF) of single nucleotide polymorphisms (SNPs) across 22 autosomes in the Dutch trios (red), CEU trios (light blue) and Dutch unrelated subjects (black). The frequency distribution of MAF for SNPs was categorized in bins. A full colour version of this figure is available at the European Journal of Human Genetics online.

Figure 2

(a) Pair-wise linkage disequilibrium (LD) (r²) per chromosome for the Dutch trio sample. The median r² per 500 kb window is plotted against physical distance (kb) in each chromosome. (b) Pearson correlation between r² estimates for the CEU and the Dutch trios. The Pearson correlation between r² calculated over 500 kb windows is plotted against physical distance (kb).

Figure 2

(a) Pair-wise linkage disequilibrium (LD) (r²) per chromosome for the Dutch trio sample. The median r² per 500 kb window is plotted against physical distance (kb) in each chromosome. (b) Pearson correlation between r² estimates for the CEU and the Dutch trios. The Pearson correlation between r² calculated over 500 kb windows is plotted against physical distance (kb).

Figure 3

Proportion of Dutch blocks retrieved in 1000 bootstraps. Frequency distribution of the proportion of observed Dutch blocks that were retrieved from the simulated data. The proportion of Dutch blocks that were observed in the CEU sample are depicted in light blue and the proportion of Dutch blocks that were not present in the CEU sample are depicted in grey. A full colour version of this figure is available at the European Journal of Human Genetics online.