Association tests were first performed for each of the 20 SNPs that were implicated in populations of European descent in this combined AA study (, upper panel). For 17 of these 20 SNPs, the implicated risk-associated alleles were found to be more common in these AA cases than controls, and the differences were statistically significant for two of these SNPs (P
< 0.05). The number of SNPs that have the same direction of association as in European populations significantly exceeded the expected number of 10 under the null hypothesis that none of these SNPs are associated with prostate cancer risk in AA (P
= 0.03, goodness-of-fit test, following a chi-squared distribution with 1 degree of freedom), suggesting that some of these SNPs may also be associated with prostate cancer risk in AA. It is estimated that the study has 80% power to detect SNPs with allelic OR ≥ 1.3 and minor allele frequency ≥ 20% at a significance level of 0.05. The limited number of SNPs reaching statistical significance level in this study may reflect the limited statistical power to detect association of moderate effect. We have also compared the association results for the 20 SNPs between this study in African Americans and previous published studies in Caucasians (Supplementary Table 1
). The strength of associations is similar amongst African Americans and Caucasians, although these SNPs have very different allele frequencies in African Americans and Caucasians.
Summary results of prostate cancer association in African Americans
Two of these 17 SNPs, located at 3p12, and Region 2 at 8q24, were significantly associated with prostate cancer risk (P < 0.05) in this AA study. Of these, SNP rs16901979 at Region 2 of 8q24 (nominal P = 1.9 × 10-5) is the only SNP that reached a study-wide significance level of 0.05 after adjustment for 20 independent tests. In addition, considerable but not statistically significant differences in allele frequencies between cases and controls were also found for SNPs at 7p15, 8q24 (region 3), 17q12 (Region 1), and Xp11 (P < 0.1). Notably, SNP rs2735839 at 19q13 (KLK3) showed a reversed direction of association compared to that observed in Caucasians (P = 0.056).
We have also estimated the ORs for each study population separately and tested homogeneity of the OR among four study populations using a Breslow-Day homogeneity test (Supplementary Table 2
). Significant heterogeneity (nominal P
< 0.05) was found for two SNPs, rs11649743 and rs6981122. However, they were not statistically significant after taking multiple testing into account.
When 14 additional SNPs at 8q24 that were previously implicated in several AA study populations were examined in this combined study, 7 were significantly associated with prostate cancer risk at nominal P
< 0.05 (, lower panel). A multivariate analysis of all nominally significant 8q24 SNPs using a step-wise procedure and adjusted for ancestry proportion, age, and study populations revealed three independent prostate cancer risk-associated SNPs; they were rs13254738 (P
= 0.007), rs16901979 (P
= 0.013), and rs10086908 (P
= 0.001). The first two SNPs have been previously implicated in populations of European descent (2
) and AA (2
). They were ~20 kb apart and were in weak LD in the AA control subjects (r2
= 0.23) (). The last SNP (rs10086908), ~100 kb centromeric to the first two SNPs (r2
= 0 with each of the two SNPs), was a novel finding. This SNP was evaluated in this study because it was initially implicated in a subset of the JHH AA study (372 cases and 350 controls, P
= 0.003, unpublished data). The association was replicated in the remaining 481 cases and 524 controls of this study (P
= 0.03). We did not observe any significant pair-wise interaction among the three independent 8q24 risk SNPs, with P
values ranging from 0.34 to 0.65.
A schematic view of genetic association between SNPs at 8q24 and prostate cancer risk in the combined African American study
Several other 8q24 SNPs, although not statistically significant in the multivariate analysis (P
> 0.05), had results that were consistent with previous reports and warrant further evaluations in larger studies. Specifically, SNPs rs6983267 (Region 3), rs7000448 (Region 1), rs7008482, and rs780321 were reported to be associated with prostate cancer risk in at least one previous AA study (9
). On the other hand, consistent with the results from two previous AA studies (11
), we did not find evidence for prostate cancer association with SNP rs1447295.
We have also tested the association between SNPs and disease aggressiveness. Although the association with prostate cancer was generally stronger among aggressive PCa patients, no statistically significant difference in allele frequencies of these SNPs were found between aggressive and non-aggressive prostate cancer patients (Supplementary Table 3
Although AAs have the highest incidence and mortality rate of prostate cancer, this group is severely under studied. GWAS of prostate cancer in AA can be a powerful approach to identify genetic risk variants; however, this is difficult to implement at this time because large sample sizes are needed for both the initial discovery and subsequent confirmation stages. A systematic evaluation of risk variants implicated in populations of European descent among AA is a feasible and effective alternative that has been empirically demonstrated in this study. The discovery of three prostate cancer risk variants in AA, if confirmed, may be substantial, especially when considering that only three factors (age, family history, and race) have been documented in the past. In the future, these risk factors may be used to identify men at increased risk to prostate cancer for early screening, prevention, and diagnosis.
Several additional implications are worth noting. Novel prostate cancer risk variants at 8q24 in AA suggest that additional prostate cancer risk variants in AA may exist in the regions flanking other prostate cancer risk-associated SNPs that have been previously implicated by genome-wide studies of populations of European descent. In addition, the generally smaller LD blocks in AA may provide better positional information on causal, functional variants. Fine mapping association studies of each region harboring known prostate cancer risk-associated SNP among AA study populations are warranted. In addition, the substantially different genetic background and environmental exposures between populations of European and African descent may provide a unique opportunity to study gene-gene and gene-environmental interactions in disease risk (20
). Finally, our results suggest that studies with larger numbers of AA subjects are needed to detect risk variants with moderate effect, and this will likely require a major collaborative effort to combine multiple AA study populations.