Haplotype and linkage disequilibrium architecture for human cancer-associated genes

Abstract

To facilitate association-based linkage studies we have studied the linkage disequilibrium (LD) and haplotype architecture around five genes of interest for cancer risk: ATM, BRCA1, BRCA2, RAD51, and TP53. Single nucleotide polymorphisms (SNPs) were identified and used to construct haplotypes that span 93-200 kb per locus with an average SNP density of 12 kb. These markers were genotyped in four ethnically defined populations that contained 48 each of African Americans, Asian Americans, Hispanic Americans, and European Americans. Haplotypes were inferred using an expectation maximization (EM) algorithm, and the data were analyzed using D', R(2), Fisher's exact P-values, and the four-gamete test for recombination. LD levels varied widely between loci from continuously high LD across 200 kb to a virtual absence of LD across a similar length of genome. LD structure also varied at each gene and between populations studied. This variation indicates that the success of linkage-based studies will require a precise description of LD at each locus and in each population to be studied. One striking consistency between genes was that at each locus a modest number of haplotypes present in each population accounted for a high fraction of the total number of chromosomes. We conclude that each locus has its own genomic profile with regard to LD, and despite this there is the widespread trend of relatively low haplotype diversity. As a result, a low marker density should be adequate to identify haplotypes that represent the common variation at a locus, thereby decreasing costs and increasing efficacy of association studies.

Figure 1

Fixation index (F_ST) of SNP allele frequencies across populations. SNP allele frequencies varied between ethnic groups. The standard deviation of that variation across populations was calculated for each SNP that was used in haplotype and LD analyses. The majority of RAD51 SNPs had an F_ST > 0.081. Excluding RAD51, 100% of SNPs had an F_ST < 0.081, and 71% of SNPs had an F_ST < 0.05.

Figure 2

LD intensity across loci. LD was measured for the five loci in all populations using the statistic ‖D‘‖ in a pairwise manner across markers. GOLD generates these plots through interpolation of the resulting triangular matrices. ‖D‘‖ ranges from 1 to 0, with 1 showing in red and 0 in dark blue.

Figure 3

Comparison of LD and recombination measurements at five loci. Pairwise LD was measured by Fisher’s exact P-values and r². Each was plotted separately by GOLD with a score of 1 showing in red and 0 in dark blue. Recombination was determined using the four-gamete test with R, potential recombination sites, indicated by white Xs. Blue boxes indicate site pairs having four gametic types, which implies that recombination has occurred between these two sites. Red boxes indicate site pairs having less than four gametic types.

Figure 4

LD versus intermarker distance. ‖D‘‖ values for all pairwise comparisons are plotted against the physical distance between each pair of markers. LD measurements were made for each population for each locus. African American data are plotted in blue, Asian American in red, European American in yellow, and Hispanic American in green.