Cases had to have a pathologically confirmed diagnosis of prostate cancer. Controls underwent screening, and possibly biopsy, for prostate cancer and had to be negative for this malignancy.

Pre-genotyping quality control metrics excluded six case samples and two controls. Twenty-seven (7 cases, 20 controls) samples were excluded due to low completion rate (lower than 94%) or extreme mean heterozygosity (lower than 13.5% or greater than 16.5%). A further five cases were excluded for having less than 80% African ancestry using the HapMap build 26 data (CEU, JPT + CHB, YRI) as the continental reference populations in a STRUCTURE analysis. Principal components analysis (PCA) identified 16 individuals (5 cases, 11 controls) with significant deviation of eigenvectors and thus, they were excluded. Finally, unexpected relatedness (1st-2nd degree), assessed using the GLU qc.ibds module (http://code.google.com/p/glu-genetics/) with an IBD0 threshold of 0.70, was detected for 11 pairs of full-sibling and one monozygotic twin; one individual randomly chosen from each related group was retained while two cases and nine controls were excluded. Note that one nuclear family involved three individuals and accounted for three related pairs, so, a total of 11 individuals were removed. In addition, one case sample was excluded due to incomplete phenotype with missing age. The final analytic data set included 474 prostate cancer cases and 458 screen-negative controls.