Results of fine-mapping across the admixture peak, spanning the region 125.6–129.4 Mb defined in ref. 1. (a) The posterior probability distribution for the position of the disease alleles based on our updated admixture mapping scan (Supplementary Table 1). (b) P values from genotyping of 1,521 variants in 1,175 African American affected individuals diagnosed at age <72 years and 837 African American controls, and 465 European American affected individuals and 446 European American controls. This analysis uncovers a ‘region 2’ containing multiple statistically significant associations between 128.14–128.28 Mb, hundreds of kb away from the previously associated ‘region 1’ (128.54–128.62 Mb, identified in ref. 2). (c) Follow-up genotyping of a total of 2,973 alleles in five populations strongly confirms the evidence for association, with the strongest single marker (rs6983561 at 128.176 Mb) significant at P = 7.9 × 10⁻¹⁹. (d) A focused examination of the span 128.1–128.7 Mb uncovers three independent regions of alleles associated with prostate cancer risk, including a new ‘region 3’ of association spanning 128.47–128.54 Mb. (e) We define the boundaries of these regions based on a comparison of genetic and physical maps, which show breakdown of linkage disequilibrium between these regions. The positions of the seven variants (Table 2) we have identified as efficiently capturing the evidence for association are shown in solid boxes. The variants are (from left to right) rs13254738, rs6983561, Broad11934905 (region 2); rs6983267 and rs7000448 (region 3) and DG8S737-8 and rs10090154 (region 1).

Case-control association statistics by logistic regression for the 186 alleles for which we collected data in all five populations and are in the region 128.1–128.7 Mb. (a) The strongest association is at marker rs6983561 (P = 7.9 × 10⁻¹⁹; in region 2). (b) Controlling for this SNP as an additional covariate, the next strongest association is rs10090154 (P = 8.8 × 10⁻¹³). This SNP is in region 1 (originally identified in ref. 2) and is correlated with their most significantly associated SNP (rs1447295) in all populations (r² ≥ 0.64) except African Americans, although it appears to capture significant additional association (P = 5.4 × 10⁻⁴ after controlling for rs1447295; Supplementary Table 5). (c) Controlling for the top two SNPs, the next strongest association is rs13254738 (P = 3.8 × 10⁻⁷), also in region 2. (d) Controlling for the top three SNPs, the next strongest association is rs7000448 in region 3 (P = 5.1 × 10⁻⁶). (e) Controlling for the top four SNPs, the next strongest association is Broad11934905 in region 2 (P = 1.5 × 10⁻⁴). In gray, we highlight two additional markers, DG8S737-8 and rs6983267, which are of borderline statistical significance after controlling for the top five SNPs (P = 0.008 and P = 0.035 respectively; the latter is significant at 0.0031 in African Americans alone). We believe that the latter two variants are likely to predict additional risk, because they have been independently highlighted in other studies (ref. 2, http://cgems.cancer.gov), so we include them in our modeling of prostate cancer risk.

The distributions of relative risks for prostate cancer in each of the populations, compared with the baseline relative risk (relative risk = 1) for individuals who do not carry any of the risk alleles at the seven markers we identified. For each population, we included the alleles into population-specific risk models in a stepwise fashion, in the order rs6983561, rs13254738, rs10090154, rs6983267, DG8S737-8, rs7000448, Broad11934905 (the last SNP was included for African Americans only, as it is appreciably polymorphic only in this population). Based on the assumption of independent multiplicative effects for each marker, we calculated the relative risks for all possible genotype combinations for these seven risk alleles. The estimated proportion of individuals in each population is plotted (based on the frequencies in the controls). For the purpose of curve smoothing, at each point we plot the percentage of the population in a relative risk range within a factor of a square root of 2 above and below the value on the x-axis (in other words, a range of 0.71–1.41 around that value). Genotypes at the extremes of risk are assigned to the lowest or highest category. For African Americans, we also plot the distribution of risk for cases diagnosed at age <72 years, because of our previous observation of significantly strong genetic risk at 8q24 in this group1. Population attributable risk (PAR) is the expected reduction in prostate cancer incidence in the populations if the risk alleles at each polymorphism did not exist.