Fine mapping of a region of chromosome 11q13 reveals multiple independent loci associated with risk of prostate cancer

Abstract

Genome-wide association studies have identified prostate cancer susceptibility alleles on chromosome 11q13. As part of the Cancer Genetic Markers of Susceptibility (CGEMS) Initiative, the region flanking the most significant marker, rs10896449, was fine mapped in 10 272 cases and 9123 controls of European origin (10 studies) using 120 common single nucleotide polymorphisms (SNPs) selected by a two-staged tagging strategy using HapMap SNPs. Single-locus analysis identified 18 SNPs below genome-wide significance (P< 10(-8)) with rs10896449 the most significant (P= 7.94 × 10(-19)). Multi-locus models that included significant SNPs sequentially identified a second association at rs12793759 [odds ratio (OR) = 1.14, P= 4.76 × 10(-5), adjusted P= 0.004] that is independent of rs10896449 and remained significant after adjustment for multiple testing within the region. rs10896438, a proxy of previously reported rs12418451 (r(2)= 0.96), independent of both rs10896449 and rs12793759 was detected (OR = 1.07, P= 5.92 × 10(-3), adjusted P= 0.054). Our observation of a recombination hotspot that separates rs10896438 from rs10896449 and rs12793759, and low linkage disequilibrium (rs10896449-rs12793759, r(2)= 0.17; rs10896449-rs10896438, r(2)= 0.10; rs12793759-rs10896438, r(2)= 0.12) corroborate our finding of three independent signals. By analysis of tagged SNPs across ∼123 kb using next generation sequencing of 63 controls of European origin, 1000 Genome and HapMap data, we observed multiple surrogates for the three independent signals marked by rs10896449 (n= 31), rs10896438 (n= 24) and rs12793759 (n= 8). Our results indicate that a complex architecture underlying the common variants contributing to prostate cancer risk at 11q13. We estimate that at least 63 common variants should be considered in future studies designed to investigate the biological basis of the multiple association signals.

Figure 1.

Association analysis result, recombination hotspots and LD of 11q13 region. The upper panel shows P-values for association testing from stages 1–3 combined CGEMS prostate cancer scan across a region of 11q13 bounded by rs930782 and rs4584599 (Chr11:68 628 370–68 870 174). Shaded regions ‘Centro’ and ‘Telo’ correspond to the centromeric region and the telomeric region, respectively. The line graph shows likelihood ratio statistics for recombination hotspot by SequenceLDhot software and five different colors represent 5 tests of 900 combined controls without resampling. The upper horizontal line indicates a genome-wide significance level (P-value of <10⁻⁸) and the lower horizontal line indicates a likelihood ratio statistic cutoff to predict the presence of a hotspot with a false-positive rate of 1 in 3700 independent tests (26). The lower panel shows an enlarged view of the region bounded by rs1123608 and rs4131929 (Chr11:68 659 896–68 768 714). The pair-wise r² correlation coefficient for SNPs in the region was estimated using TagZilla and plotted using SnpPlotter (41). The 18 SNPs with genome-wide significance were color coded. Light-blue represents correlation bin1 SNPs (rs10896438, rs2924538, rs11228551 and rs11228553), purple represents bin2 SNPs (rs4255548, rs4495900 and rs7950547) and red represents bin3 SNPs (rs4620729, rs7931342, rs10896449, rs9787877, rs7939250, rs10896450 and rs11228583). Green and black represent singleton SNPs with no proxy (r² > 0.8), but colored to show separation by a recombination hotspot (green, rs10896437 and rs2924540; black, rs12793759 and rs12281017). A black arrow indicates the recombination hotspot that separates the region into centromeric and telomeric regions.

Figure 2.

Sequential multi-locus model of SNPs in the 11q13 region. The colored vertical boxed area represents the region of the observed recombination hotspot. Eighteen SNPs that showed genome-wide significance (P< 10⁻⁸) in the single-SNP analysis (Table 1) were color coded comparably with Figure 1. SNPs in gray were observed to be not genome-wide significant in the single-SNP model but of interest in the multi-locus modeling. (A) Shows the two SNP multi-locus analysis conditioned on rs10896449 (a red diamond on the x-axis). Twenty eight SNPs remained significant at an alpha of 0.05 (horizontal line). rs12793759 is the most significant SNP (per allele OR = 1.14, 95% CI: 1.07–1.21, P = 4.76 × 10⁻⁵, adjusted P = 0.004). (B) Shows the three-SNP multi-locus analysis conditioned on rs12793759 and rs10896449 (black diamond and red diamonds on x-axis, respectively). Thirteen SNPs showed significance at an alpha level of 0.05 (horizontal line) with rs10896438 being the most significant SNP (per allele OR = 1.07, 95% CI: 1.02–1.12, P = 5.92 × 10⁻³, adjusted P = 0.054). (C) Depicts the correlation patterns of the 18 genome-wide significant SNPs with color coding as per Figure 1. Correlation bins were defined with an r² > 0.8 threshold and based on the analysis of all controls of European background in this study. Four SNPs that had no proxy with the threshold were denoted as ‘singleton'. The pair-wise (singleton versus singleton) or average (singleton versus bin, SNPs within a bin) correlation values are expressed by r² (D′).