To estimate the prostate cancer risk conferred by individual single nucleotide polymorphisms (SNPs), SNP-SNP interactions, and/or cumulative SNP effects, we evaluated the association between prostate cancer risk and the genetic variants of 12 key genes within the steroid hormone pathway (CYP17, HSD17B3, ESR1, SRD5A2, HSD3B1, HSD3B2, CYP19, CYP1A1, CYP1B1, CYP3A4, CYP27B1, and CYP24A1). A total of 116 tagged SNPs covering the group of genes were analyzed in 2,452 samples (886 cases and 1,566 controls) in three ethnic/racial groups. Several SNPs within CYP19 were significantly associated with prostate cancer in all three ethnicities (P = 0.001-0.009). Genetic variants within HSD3B2 and CYP24A1 conferred increased risk of prostate cancer in non-Hispanic or Hispanic Caucasians. A significant gene-dosage effect for increasing numbers of potential high-risk genotypes was found in non-Hispanic and Hispanic Caucasians. Higher-order interactions showed a seven-SNP interaction involving HSD17B3, CYP19, and CYP24A1 in Hispanic Caucasians (P = 0.001). In African Americans, a 10-locus model, with SNPs located within SRD5A2, HSD17B3, CYP17, CYP27B1, CYP19, and CYP24A1, showed a significant interaction (P = 0.014). In non-Hispanic Caucasians, an interaction of four SNPs in HSD3B2, HSD17B3, and CYP19 was found (P < 0.001). These data are consistent with a polygenic model of prostate cancer, indicating that multiple interacting genes of the steroid hormone pathway confer increased risk of prostate cancer. (Cancer Epidemiol Biomarkers Prev 2009;18(6):1869–80)

The underlying etiology of prostate cancer is poorly understood, with genetic predisposition and environmental factors likely contributing to risk (1). Heritable risk factors can involve highly penetrant susceptibility genes with low frequencies and/or low-penetrant genes with higher frequencies in the population. The majority of prostate cancer cases likely involve more common, low-penetrant to moderate-penetrant alleles in genes that are components of functional prostatic pathways, such as androgen hormones and their metabolizing enzymes (2-4).

Evidence exists that the steroid hormone pathway and genes involved in the metabolism of estrogens and androgens affect the risk of prostate cancer (4). Androgens are essential for the development, growth, and secretory activities of the prostate, whereas estrogens modulate these effects. Several of the cytochrome p450 enzymes participate in the synthesis of testosterone from cholesterol (ref. 5; Fig. 1). The CYP17 gene (10q24.3, MIM 609300) encodes steroid 17α-hydroxylase and is a member of the cytochrome P450 super family. CYP17 mediates both 17α-hydroxylase and 17,20-lyase activity, and is involved in the biosynthesis of both 17α-hydroxylated glucocorticoids and sex steroids (6-8). The final step in the synthesis of testosterone is the conversion of androstenedione to testosterone by 17β-hydroxysteroid dehydrogenase type 3 (HSD17B3, 9q22, MIM 605573). Testosterone, along with estrone and estradiol, interact with the estrogen receptor ESR1 (6q25.1, MIM 133430), leading to the activation of the enzyme steroid 5α-reductase type 2 (SRD5A2, 2p23, MIM 607306). Testosterone is irreversibly metabolized to dihydrotestosterone by SRD5A2. Normal prostate growth and development requires dihydrotestosterone and a functioning of SRD5A2 (4). Dihydrotestosterone stimulates the transcription of several genes with androgen-responsive elements in their promoters and is inactivated in the prostate by hydroxy-δ5-steroid dehydrogenase, 3β and steroid δ-isomerase 1 (HSD3B1, 1p13.1, MIM 109715), hydroxy-δ5-steroid dehydrogenase, and 3β and steroid δ-isomerase 2 (HSD3B2, 1p13.1, MIM 201810). In the prostate, androstenedione and testosterone can also be converted by aromatase (cytochrome p450, family 19, subfamily A, polypeptide 1; CYP19, 15q21.1, MIM 107910) into estrone and estradiol, which are then metabolized by genes from the cytochrome P450 family, including CYP1A1 (15q22-q24, MIM 108330), CYP1B1 (2p21, MIM 601771), and CYP3A4 (7q21.1, MIM 124010) into hydroxy and methoxy estrogens (Fig. 1).

Figure 1.

Summary of the steroid hormone pathway indicating the candidate genes studied in this report and the products from their enzymatic reactions.

Figure 1.

Summary of the steroid hormone pathway indicating the candidate genes studied in this report and the products from their enzymatic reactions.

Close modal

The growth and differentiation of normal prostatic tissue is also promoted by interactions between the vitamin D and dihydrotestosterone pathways (9). Levels of the bioactive form of vitamin D are controlled by both the activating enzyme 1-A-hydroxylase (CYP27B1, 12q13.1-q13.3, MIM 609506) and the deactivating enzyme 24-hyroxylase (CYP24A1, 20q13, MIM 126065), which regulate cell growth and may reduce the risk of malignant transformation.

Functional variants in the genes of these pathways are likely to influence prostate carcinogenesis. Significant associations between variants in one or more of the above genes and risk for prostate cancer have been reported but with inconsistent findings. For example, meta-analyses of single nucleotide polymorphisms (SNPs) within CYP17, including rs743572, or SNPs rs523349 (V89L) and rs9282858 (A49T) within SRD5A2 did not support evidence for association with risk of prostate cancer although smaller studies reported on significant risk effects (10-12). Recent efforts evaluating these risk variants relied on single-gene approaches. With a multicausal etiology of prostate cancer, it is likely that multiple risk variants act simultaneously, synergistically, or additively to influence prostate cancer susceptibility. Additionally, in most studies only a small proportion of the estimated number of genetic variants was analyzed, and the contributions of variants in regulatory, noncoding regions of genes, rather than in exons, were often omitted and probably underestimated because transcriptional regulatory elements are located in introns.

We systematically evaluated the association of genetic variants of 12 key genes within the steroid hormone pathway and prostate cancer risk, including CYP17, HSD17B3, ESR1, SRD5A2, HSD3B1, HSD3B2, CYP19, CYP1A1, CYP1B1, CYP3A4, CYP27B1, and CYP24A1, and genotyped 116 tagged SNPs covering the genes in 2,452 samples (886 cases and 1,566 controls) of non-Hispanic Caucasian, Hispanic Caucasian, or African American origin. We report on the association of individual SNPs as well as higher-order SNP-SNP interactions. We also tested the hypothesis that individuals carrying multiple copies of risk variants are at increased risk for prostate cancer.

Subjects

Study subjects included men from the San Antonio Center for Biomarkers of Risk of Prostate Cancer cohort. The San Antonio Center for Biomarkers of Risk of Prostate Cancer is funded by the National Cancer Institute and has been prospectively enrolling healthy male volunteers since 2001. On each annual visit, a digital rectal examination was done and the serum prostate-specific antigen level was determined. From this cohort, 231 incident cases (135 non-Hispanic Caucasians, 60 Hispanic Caucasians, and 36 African Americans) were available. We also included 655 cases with a known history of prostate cancer that are enrolled within the same time period in a parallel study of prevalent prostate cancer. These prevalent cases were recruited through the same means and from the same metropolitan population as the San Antonio Center for Biomarkers of Risk of Prostate Cancer study and had a median time period of 3 y (range, 0-25 y) between disease diagnosis and enrollment into the study. Cases had biopsy-confirmed prostate cancer, and controls consisted of male volunteers at least 45 y old who had normal digital rectal examinations and prostate-specific antigen levels <2.5 ng/mL on all study visits. Race/ethnicity was self-reported on a questionnaire filled out at a clinic site at the time of recruiting, and the Hazuda algorithm (13) was used to classify race/ethnicity. All participants have established U.S. residency and are believed to be long-term residents. Study age among controls was the age at last follow-up, and age among cases was the age at prostate cancer diagnosis. Our control group was younger than our prostate cancer cases, with a mean age (SD) of 60.8 (8.8) y for the controls and a mean age (SD) of 65.5 (8.5) y for the cases (P < 0.0001). The clinical characteristics of the subjects are summarized in Table 1. A total of 1,452 non-Hispanic Caucasians (609 cases, 843 controls), 709 Hispanic Caucasians (195 cases, 514 controls), and 291 African Americans (82 cases, 209 controls) were included in this analysis. The San Antonio Center for Biomarkers of Risk of Prostate Cancer received Institutional Review Board approval, and informed consent was obtained from all subjects.

Table 1.

Clinical data of the study group

SubgroupCases (N = 886)
Controls (N = 1,566)
N (%)N (%)
Ethnic background    
    Non-Hispanic Caucasian 609 (68.7%) 843 (53.8%)  
    Hispanic Caucasian 195 (22.0%) 514 (32.8%)  
    African American 82 (9.3%) 209 (13.4%)  
Age (y)    
    ≤50 36 (4.1%) 215 (13.7%)  
    51-60 208 (23.6%) 581 (37.1%)  
    61-70 389 (44.2%) 499 (31.9%)  
    >70 248 (28.1%) 271 (17.3%)  
    Mean ± SD 65.5 ± 8.5 60.8 ± 8.8 P < 0.0001 
PSA (ng/mL)    
    ≤4.0 179 1,566  
    4.1-10.0 38  
    10.1-20.0  
    >20.0  
    Mean ± SD 3.1 ± 4.7 0.85 ± 0.44 P < 0.0001 
SubgroupCases (N = 886)
Controls (N = 1,566)
N (%)N (%)
Ethnic background    
    Non-Hispanic Caucasian 609 (68.7%) 843 (53.8%)  
    Hispanic Caucasian 195 (22.0%) 514 (32.8%)  
    African American 82 (9.3%) 209 (13.4%)  
Age (y)    
    ≤50 36 (4.1%) 215 (13.7%)  
    51-60 208 (23.6%) 581 (37.1%)  
    61-70 389 (44.2%) 499 (31.9%)  
    >70 248 (28.1%) 271 (17.3%)  
    Mean ± SD 65.5 ± 8.5 60.8 ± 8.8 P < 0.0001 
PSA (ng/mL)    
    ≤4.0 179 1,566  
    4.1-10.0 38  
    10.1-20.0  
    >20.0  
    Mean ± SD 3.1 ± 4.7 0.85 ± 0.44 P < 0.0001 

Abbreviation: PSA, prostate-specific antigen.

SNP Selection and Genotyping

DNA was isolated from participants' whole blood cells with the use of a QIAamp DNA Blood Maxi Kit (Qiagen).

A total of 120 SNPs within the 12 candidate genes were selected based on the following criteria. We first selected SNPs from available databases, the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov/projects/SNP/) and SNPper (http://snpper.chip.org/), using the following criteria: (a) Within each gene, SNPs with a minor allele frequency >0.05, which leads to an amino acid substitution, and/or are in other coding regions of the gene and thus potentially functionally important were selected, (b) SNPs for which an association with prostate cancer has previously been shown as reported in literature were chosen. After this initial selection, we identified tagging SNPs within each gene using Haploview with the following criteria: (a) a minor allele frequency >0.05 to gain more statistical power, (b) an r2 threshold of 0.8 and a log of odds threshold for multimarker testing of 3.0, (c) a minimum distance between tags of 60 bp, (d) we included our preselected SNPs (see above), (e) for each gene the search for SNPs extended to a 10 kb region surrounding the gene, and (f) we used the 2-marker and 3-marker haplotype tagging option (http://www.broad.mit.edu/mpg/haploview/). The selection was based on information on the European population as provided by HapMap (www.hapmap.org). The SNPs are described in Table 2.

Table 2.

Genes, SNP selection, their location, and minor allele frequencies in cases and controls of each ethnicity

GeneSNPPositionSNP location/AA changeMinor alleleNon-Hispanic Caucasians
Hispanic Caucasians
African Americans
MAF CasesMAF ControlsP*MAF CasesMAF Controlsp*MAF CasesMAF Controlsp*
HSD3B2 rs1819698 chr1:119767042 3′UTR 0.142 0.104 0.002 0.173 0.14 0.140 0.518 0.525 0.894 
 rs10923823 chr1:119789927  0.38 0.408 0.143 0.407 0.429 0.486 0.159 0.154 0.903 
 rs1538989 chr1:119791376  0.487 0.43 0.003 0.428 0.378 0.099 0.817 0.836 0.591 
 rs4659176 chr1:119791924  0.339 0.323 0.375 0.247 0.229 0.484 0.207 0.259 0.206 
HSD3B1 rs947130 chr1:119818255  0.298 0.268 0.082 0.183 0.19 0.771 0.378 0.352 0.569 
 rs6201 chr1:119855738 Val79Ile 0.499 0.5 0.964 0.5 0.5 1.000 0.433 0.427 0.910 
 rs6428830 chr1:119856298  0.304 0.28 0.162 0.155 0.179 0.290 0.091 0.096 0.880 
 rs6203 chr1:119858681 Leu338Leu 0.429 0.442 0.485 0.518 0.522 0.907 0.104 0.108 0.883 
 rs2362819 chr1:119888607  0.327 0.306 0.239 0.246 0.262 0.556 0.549 0.554 0.916 
SRD5A2 rs9332975 chr2:31603921 3′UTR 0.108 0.119 0.351 0.085 0.078 0.659 0.189 0.207 0.644 
 rs2268794 chr2:31632908  0.135 0.139 0.749 0.095 0.081 0.423 0.329 0.333 0.928 
 rs2268796 chr2:31635784  0.416 0.422 0.767 0.479 0.445 0.270 0.646 0.608 0.410 
 rs2208532 chr2:31642493  0.439 0.427 0.496 0.503 0.52 0.566 0.402 0.333 0.132 
 rs4952222 chr2:31653367  0.056 0.058 0.794 0.15 0.137 0.547 0.019 0.034 0.338 
 rs523349 chr2:31659210 Val89Leu 0.271 0.284 0.553 0.388 0.346 0.274 0.343 0.296 0.355 
 rs9282858 chr2:31659330 Ala49Thr 0.034 0.038 0.665 0.023 0.025 0.884 0.009 0.260 
 rs9332960 chr2:31659458 Ter6Gln 0.000 0.000 0.000 
 rs632148 chr2:31659535 5′UTR failed failed failed failed failed failed failed failed failed 
 rs3754838 chr2:31661804  0.113 0.116 0.825 0.083 0.074 0.560 0.112 0.099 0.641 
CYP1B1 rs1800440 chr2:38151643 Ser453Asn 0.194 0.198 0.828 0.164 0.141 0.410 0.032 0.046 0.512 
 rs1056836 chr2:38151707 Leu432Val 0.431 0.44 0.691 0.222 0.278 0.113 0.765 0.741 0.610 
 rs1056827 chr2:38155681 Ser119Ala 0.277 0.278 0.949 0.316 0.321 0.913 0.52 0.44 0.239 
 rs10012 chr2:38155894 Gly48Arg 0.283 0.273 0.630 0.288 0.301 0.719 0.515 0.458 0.303 
 rs2617266 chr2:38156048  0.278 0.275 0.905 0.271 0.269 0.939 0.269 0.214 0.247 
 rs2551188 chr2:38156298  0.281 0.272 0.683 0.288 0.298 0.783 0.493 0.431 0.258 
 rs2567206 chr2:38157035  0.278 0.273 0.832 0.271 0.27 0.971 0.09 0.069 0.493 
ESR1 rs2234693 chr6:152205028  0.476 0.487 0.639 0.379 0.399 0.629 0.534 0.491 0.447 
 rs9340799 chr6:152205074  0.349 0.37 0.348 0.318 0.34 0.580 0.328 0.25 0.132 
CYP3A4 rs4646450 chr7:99104254  0.16 0.154 0.640 0.309 0.314 0.863 0.854 0.852 0.958 
 rs4646437 chr7:99203019  0.099 0.094 0.606 0.214 0.191 0.351 0.732 0.747 0.717 
 rs2740574 chr7:99220032  0.028 0.031 0.711 0.074 0.091 0.585 0.629 0.665 0.602 
HSD17B3 rs10989735 chr9:98017875  0.464 0.485 0.260 0.448 0.483 0.258 0.323 0.34 0.718 
 rs10820020 chr9:98018481  0.215 0.226 0.476 0.24 0.232 0.757 0.079 0.105 0.364 
 rs2479827 chr9:98028250  0.367 0.331 0.051 0.227 0.204 0.370 0.506 0.531 0.605 
 rs2066479 chr9:98037631 Gly289Ser 0.063 0.039 0.049 0.049 0.044 0.835 0.081 0.125 0.342 
 rs1810711 chr9:98048379  0.513 0.495 0.359 0.42 0.387 0.279 0.598 0.568 0.531 
 rs2026001 chr9:98055916  0.35 0.371 0.255 0.487 0.466 0.488 0.262 0.231 0.454 
 rs8190536 chr9:98056741  0.215 0.198 0.280 0.144 0.122 0.287 0.402 0.38 0.625 
 rs2257157 chr9:98058263  0.459 0.432 0.158 0.34 0.388 0.108 0.439 0.488 0.309 
 rs7039978 chr9:98061403  0.472 0.498 0.170 0.626 0.635 0.775 0.171 0.204 0.383 
 rs7037932 chr9:98070492  0.351 0.327 0.216 0.214 0.206 0.768 0.375 0.444 0.179 
 rs13302476 chr9:98075007  0.13 0.162 0.018 0.229 0.238 0.746 0.262 0.284 0.612 
 rs280663 chr9:98089522  failed failed failed failed failed failed failed failed failed 
 rs2479828 chr9:98090615  0.104 0.1 0.726 0.148 0.133 0.483 0.055 0.043 0.566 
 rs9409407 chr9:98095070  0.123 0.101 0.065 0.047 0.055 0.572 0.019 0.015 0.801 
 rs8190495 chr9:98101705  0.366 0.377 0.566 0.232 0.26 0.304 0.244 0.17 0.051 
 rs2066474 chr9:98104246 5′UTR 0.168 0.187 0.197 0.224 0.275 0.061 0.188 0.18 0.844 
 rs2026002 chr9:98124002  0.119 0.107 0.332 0.222 0.227 0.839 0.012 0.068 0.007 
CYP17 rs284851 chr10:104568521  0.124 0.113 0.373 0.173 0.172 0.970 0.278 0.318 0.365 
 rs619824 chr10:104571278  0.415 0.431 0.380 0.42 0.416 0.883 0.701 0.657 0.330 
 rs10883782 chr10:104573922  0.145 0.178 0.029 0.109 0.112 0.880 0.083 0.116 0.281 
 rs1004467 chr10:104584497  0.096 0.085 0.372 0.194 0.197 0.920 0.287 0.194 0.043 
 rs6162 chr10:104586971 His46His 0.392 0.414 0.239 0.392 0.394 0.930 0.439 0.367 0.125 
 rs743572 chr10:104587142 5′UTR 0.366 0.395 0.112 0.391 0.396 0.883 0.432 0.367 0.167 
 rs17115144 chr10:104587298  0.345 0.361 0.567 0.381 0.412 0.578 0.5 0.331 0.020 
 rs2486758 chr10:104587470  0.222 0.199 0.139 0.289 0.295 0.818 0.043 0.052 0.657 
CYP27B1 rs1048691 chr12:56439215  0.21 0.218 0.641 0.33 0.327 0.931 0.329 0.42 0.053 
 rs4646537 chr12:56443548  0.027 0.03 0.651 0.026 0.034 0.429 0.079 0.093 0.624 
 rs4646536 chr12:56444255  0.31 0.326 0.359 0.343 0.323 0.501 0.262 0.293 0.472 
 rs8176345 chr12:56444825 Leu314Leu 0.035 0.032 0.663 0.031 0.01 0.010 0.012 0.153 
 rs10877012 chr12:56448352  failed failed failed failed failed failed failed failed failed 
 rs703842 chr12:56449006 3′UTR 0.309 0.328 0.285 0.34 0.329 0.690 0.3 0.33 0.502 
CYP19 rs9972359 chr15:49279146  0.417 0.414 0.871 0.402 0.383 0.528 0.768 0.759 0.825 
 rs16964189 chr15:49281529  0.202 0.21 0.622 0.16 0.162 0.938 0.207 0.235 0.496 
 rs934632 chr15:49283122  0.173 0.178 0.736 0.286 0.242 0.102 0.195 0.207 0.762 
 rs934634 chr15:49287830  0.203 0.209 0.719 0.17 0.166 0.868 0.323 0.327 0.929 
 rs2255192 chr15:49288127  0.203 0.209 0.709 0.173 0.164 0.712 0.323 0.336 0.769 
 rs4646 chr15:49290136 3′UTR 0.24 0.254 0.384 0.448 0.424 0.420 0.256 0.275 0.662 
 rs10046 chr15:49290278 3′UTR 0.557 0.536 0.270 0.374 0.408 0.257 0.317 0.265 0.232 
 rs17601241 chr15:49295166  0.075 0.082 0.527 0.16 0.181 0.375 0.055 0.062 0.762 
 rs700519 chr15:49295260 Arg264Cys 0.036 0.04 0.784 0.022 0.03 0.676 0.155 0.168 0.812 
 rs2899472 chr15:49303347  0.281 0.26 0.213 0.124 0.188 0.005 0.055 0.04 0.458 
 rs12439137 chr15:49303596  0.153 0.112 0.002 0.194 0.217 0.381 0.058 0.085 0.310 
 rs12592697 chr15:49312465  0.33 0.347 0.345 0.402 0.376 0.388 0.305 0.318 0.770 
 rs700518 chr15:49316404 Val80Val 0.544 0.517 0.160 0.343 0.385 0.154 0.28 0.216 0.114 
 rs10519295 chr15:49319939  0.111 0.108 0.784 0.026 0.055 0.025 0.006 0.009 0.714 
 rs10459592 chr15:49323433  0.418 0.428 0.579 0.523 0.463 0.053 0.61 0.605 0.918 
 rs11636686 chr15:49329358  0.534 0.51 0.201 0.325 0.37 0.126 0.159 0.123 0.285 
 rs12050772 chr15:49332163  0.493 0.468 0.178 0.281 0.328 0.100 0.183 0.151 0.369 
 rs2414099 chr15:49336074  0.138 0.151 0.327 0.258 0.258 0.984 0.165 0.225 0.117 
 rs11636639 chr15:49350384  0.481 0.453 0.130 0.303 0.366 0.032 0.53 0.41 0.012 
 rs2470176 chr15:49371231  0.139 0.146 0.607 0.186 0.156 0.208 0.524 0.525 0.995 
 rs2470158 chr15:49375687  0.093 0.098 0.607 0.075 0.074 0.953 0.116 0.12 0.884 
 rs17523880 chr15:49379835  0.147 0.134 0.314 0.088 0.106 0.334 0.043 0.025 0.277 
 rs2470152 chr15:49382264  0.551 0.524 0.167 0.407 0.43 0.457 0.451 0.457 0.907 
 rs17523922 chr15:49386497  0.097 0.102 0.664 0.064 0.075 0.506 0.018 0.037 0.257 
 rs3751592 chr15:49393870  0.303 0.266 0.042 0.369 0.372 0.942 0.291 0.224 0.143 
 rs3751591 chr15:49394002  0.166 0.16 0.659 0.169 0.189 0.414 0.043 0.049 0.742 
 rs1902584 chr15:49398946  0.082 0.08 0.908 0.131 0.122 0.650 0.037 0.077 0.083 
 rs2445762 chr15:49405000  0.264 0.247 0.315 0.397 0.379 0.556 0.427 0.401 0.587 
 rs7168331 chr15:49408275  0.384 0.362 0.241 0.49 0.485 0.873 0.579 0.525 0.253 
 rs7174997 chr15:49409420  0.168 0.192 0.092 0.157 0.164 0.763 0.037 0.062 0.243 
 rs868475 chr15:49423746  0.05 0.059 0.281 0.12 0.112 0.693 0.299 0.256 0.317 
 rs2470164 chr15:49425853  0.566 0.501 0.003 0.392 0.389 0.911 0.131 0.056 0.027 
 rs2445773 chr15:49430233  0.174 0.181 0.633 0.276 0.261 0.585 0.207 0.213 0.885 
CYP1A1 rs2198843 chr15:72788283  0.166 0.147 0.233 0.389 0.44 0.676 0.5 0.546 0.853 
 rs1048943 chr15:72800038 Val462Ile 0.045 0.037 0.490 0.279 0.278 0.993 0.033 0.011 0.239 
 rs4646421 chr15:72803245  0.162 0.238 0.185 0.643 0.473 0.013 0.7 0.493 0.083 
 rs2470893 chr15:72806502  0.304 0.316 0.504 0.127 0.121 0.778 0.049 0.053 0.850 
CYP24A1 rs2762929 chr20:52199602  0.389 0.399 0.601 0.369 0.387 0.531 0.683 0.725 0.329 
 rs8118441 chr20:52199792  0.139 0.142 0.837 0.186 0.186 0.994 0.311 0.278 0.445 
 rs6068810 chr20:52202758  0.055 0.045 0.251 0.034 0.019 0.128 0.043 0.025 0.277 
 rs6097807 chr20:52202862  0.243 0.255 0.472 0.268 0.247 0.432 0.488 0.448 0.399 
 rs4809957 chr20:52204578 3′UTR 0.215 0.212 0.816 0.247 0.232 0.547 0.438 0.388 0.283 
 rs2762934 chr20:52204668 3′UTR 0.192 0.187 0.779 0.134 0.162 0.210 0.146 0.179 0.362 
 rs1570669 chr20:52207834  0.323 0.337 0.421 0.312 0.3 0.684 0.604 0.58 0.620 
 rs2296239 chr20:52208935 Pro375Pro 0.204 0.213 0.557 0.247 0.233 0.580 0.439 0.392 0.318 
 rs6068816 chr20:52214498 Thr248Thr 0.108 0.109 0.948 0.106 0.079 0.130 0.067 0.046 0.334 
 rs4809958 chr20:52215845  0.156 0.152 0.794 0.16 0.12 0.056 0.085 0.068 0.486 
 rs3787554 chr20:52216087  0.093 0.091 0.795 0.085 0.045 0.005 0.03 0.037 0.709 
 rs2244719 chr20:52216265  0.451 0.484 0.088 0.552 0.569 0.596 0.244 0.262 0.659 
 rs2762941 chr20:52217059  0.388 0.376 0.505 0.296 0.319 0.434 0.506 0.531 0.605 
 rs2181874 chr20:52217885  0.256 0.241 0.355 0.253 0.289 0.192 0.36 0.423 0.179 
 rs4809960 chr20:52219480  0.232 0.237 0.738 0.242 0.209 0.198 0.11 0.154 0.179 
 rs2296241 chr20:52219626 Ala184Ala 0.484 0.47 0.447 0.554 0.559 0.887 0.549 0.497 0.279 
 rs2245153 chr20:52219813  0.181 0.194 0.390 0.18 0.131 0.027 0.315 0.289 0.563 
 rs2585428 chr20:52220304  0.449 0.473 0.194 0.351 0.352 0.949 0.433 0.503 0.143 
 rs13038432 chr20:52220709  0.084 0.083 0.919 0.052 0.045 0.606 0.012 0.022 0.466 
 rs6022999 chr20:52221420  0.241 0.239 0.930 0.225 0.233 0.782 0.594 0.627 0.485 
 rs2248359 chr20:52224925  0.396 0.396 0.991 0.36 0.343 0.566 0.646 0.627 0.668 
GeneSNPPositionSNP location/AA changeMinor alleleNon-Hispanic Caucasians
Hispanic Caucasians
African Americans
MAF CasesMAF ControlsP*MAF CasesMAF Controlsp*MAF CasesMAF Controlsp*
HSD3B2 rs1819698 chr1:119767042 3′UTR 0.142 0.104 0.002 0.173 0.14 0.140 0.518 0.525 0.894 
 rs10923823 chr1:119789927  0.38 0.408 0.143 0.407 0.429 0.486 0.159 0.154 0.903 
 rs1538989 chr1:119791376  0.487 0.43 0.003 0.428 0.378 0.099 0.817 0.836 0.591 
 rs4659176 chr1:119791924  0.339 0.323 0.375 0.247 0.229 0.484 0.207 0.259 0.206 
HSD3B1 rs947130 chr1:119818255  0.298 0.268 0.082 0.183 0.19 0.771 0.378 0.352 0.569 
 rs6201 chr1:119855738 Val79Ile 0.499 0.5 0.964 0.5 0.5 1.000 0.433 0.427 0.910 
 rs6428830 chr1:119856298  0.304 0.28 0.162 0.155 0.179 0.290 0.091 0.096 0.880 
 rs6203 chr1:119858681 Leu338Leu 0.429 0.442 0.485 0.518 0.522 0.907 0.104 0.108 0.883 
 rs2362819 chr1:119888607  0.327 0.306 0.239 0.246 0.262 0.556 0.549 0.554 0.916 
SRD5A2 rs9332975 chr2:31603921 3′UTR 0.108 0.119 0.351 0.085 0.078 0.659 0.189 0.207 0.644 
 rs2268794 chr2:31632908  0.135 0.139 0.749 0.095 0.081 0.423 0.329 0.333 0.928 
 rs2268796 chr2:31635784  0.416 0.422 0.767 0.479 0.445 0.270 0.646 0.608 0.410 
 rs2208532 chr2:31642493  0.439 0.427 0.496 0.503 0.52 0.566 0.402 0.333 0.132 
 rs4952222 chr2:31653367  0.056 0.058 0.794 0.15 0.137 0.547 0.019 0.034 0.338 
 rs523349 chr2:31659210 Val89Leu 0.271 0.284 0.553 0.388 0.346 0.274 0.343 0.296 0.355 
 rs9282858 chr2:31659330 Ala49Thr 0.034 0.038 0.665 0.023 0.025 0.884 0.009 0.260 
 rs9332960 chr2:31659458 Ter6Gln 0.000 0.000 0.000 
 rs632148 chr2:31659535 5′UTR failed failed failed failed failed failed failed failed failed 
 rs3754838 chr2:31661804  0.113 0.116 0.825 0.083 0.074 0.560 0.112 0.099 0.641 
CYP1B1 rs1800440 chr2:38151643 Ser453Asn 0.194 0.198 0.828 0.164 0.141 0.410 0.032 0.046 0.512 
 rs1056836 chr2:38151707 Leu432Val 0.431 0.44 0.691 0.222 0.278 0.113 0.765 0.741 0.610 
 rs1056827 chr2:38155681 Ser119Ala 0.277 0.278 0.949 0.316 0.321 0.913 0.52 0.44 0.239 
 rs10012 chr2:38155894 Gly48Arg 0.283 0.273 0.630 0.288 0.301 0.719 0.515 0.458 0.303 
 rs2617266 chr2:38156048  0.278 0.275 0.905 0.271 0.269 0.939 0.269 0.214 0.247 
 rs2551188 chr2:38156298  0.281 0.272 0.683 0.288 0.298 0.783 0.493 0.431 0.258 
 rs2567206 chr2:38157035  0.278 0.273 0.832 0.271 0.27 0.971 0.09 0.069 0.493 
ESR1 rs2234693 chr6:152205028  0.476 0.487 0.639 0.379 0.399 0.629 0.534 0.491 0.447 
 rs9340799 chr6:152205074  0.349 0.37 0.348 0.318 0.34 0.580 0.328 0.25 0.132 
CYP3A4 rs4646450 chr7:99104254  0.16 0.154 0.640 0.309 0.314 0.863 0.854 0.852 0.958 
 rs4646437 chr7:99203019  0.099 0.094 0.606 0.214 0.191 0.351 0.732 0.747 0.717 
 rs2740574 chr7:99220032  0.028 0.031 0.711 0.074 0.091 0.585 0.629 0.665 0.602 
HSD17B3 rs10989735 chr9:98017875  0.464 0.485 0.260 0.448 0.483 0.258 0.323 0.34 0.718 
 rs10820020 chr9:98018481  0.215 0.226 0.476 0.24 0.232 0.757 0.079 0.105 0.364 
 rs2479827 chr9:98028250  0.367 0.331 0.051 0.227 0.204 0.370 0.506 0.531 0.605 
 rs2066479 chr9:98037631 Gly289Ser 0.063 0.039 0.049 0.049 0.044 0.835 0.081 0.125 0.342 
 rs1810711 chr9:98048379  0.513 0.495 0.359 0.42 0.387 0.279 0.598 0.568 0.531 
 rs2026001 chr9:98055916  0.35 0.371 0.255 0.487 0.466 0.488 0.262 0.231 0.454 
 rs8190536 chr9:98056741  0.215 0.198 0.280 0.144 0.122 0.287 0.402 0.38 0.625 
 rs2257157 chr9:98058263  0.459 0.432 0.158 0.34 0.388 0.108 0.439 0.488 0.309 
 rs7039978 chr9:98061403  0.472 0.498 0.170 0.626 0.635 0.775 0.171 0.204 0.383 
 rs7037932 chr9:98070492  0.351 0.327 0.216 0.214 0.206 0.768 0.375 0.444 0.179 
 rs13302476 chr9:98075007  0.13 0.162 0.018 0.229 0.238 0.746 0.262 0.284 0.612 
 rs280663 chr9:98089522  failed failed failed failed failed failed failed failed failed 
 rs2479828 chr9:98090615  0.104 0.1 0.726 0.148 0.133 0.483 0.055 0.043 0.566 
 rs9409407 chr9:98095070  0.123 0.101 0.065 0.047 0.055 0.572 0.019 0.015 0.801 
 rs8190495 chr9:98101705  0.366 0.377 0.566 0.232 0.26 0.304 0.244 0.17 0.051 
 rs2066474 chr9:98104246 5′UTR 0.168 0.187 0.197 0.224 0.275 0.061 0.188 0.18 0.844 
 rs2026002 chr9:98124002  0.119 0.107 0.332 0.222 0.227 0.839 0.012 0.068 0.007 
CYP17 rs284851 chr10:104568521  0.124 0.113 0.373 0.173 0.172 0.970 0.278 0.318 0.365 
 rs619824 chr10:104571278  0.415 0.431 0.380 0.42 0.416 0.883 0.701 0.657 0.330 
 rs10883782 chr10:104573922  0.145 0.178 0.029 0.109 0.112 0.880 0.083 0.116 0.281 
 rs1004467 chr10:104584497  0.096 0.085 0.372 0.194 0.197 0.920 0.287 0.194 0.043 
 rs6162 chr10:104586971 His46His 0.392 0.414 0.239 0.392 0.394 0.930 0.439 0.367 0.125 
 rs743572 chr10:104587142 5′UTR 0.366 0.395 0.112 0.391 0.396 0.883 0.432 0.367 0.167 
 rs17115144 chr10:104587298  0.345 0.361 0.567 0.381 0.412 0.578 0.5 0.331 0.020 
 rs2486758 chr10:104587470  0.222 0.199 0.139 0.289 0.295 0.818 0.043 0.052 0.657 
CYP27B1 rs1048691 chr12:56439215  0.21 0.218 0.641 0.33 0.327 0.931 0.329 0.42 0.053 
 rs4646537 chr12:56443548  0.027 0.03 0.651 0.026 0.034 0.429 0.079 0.093 0.624 
 rs4646536 chr12:56444255  0.31 0.326 0.359 0.343 0.323 0.501 0.262 0.293 0.472 
 rs8176345 chr12:56444825 Leu314Leu 0.035 0.032 0.663 0.031 0.01 0.010 0.012 0.153 
 rs10877012 chr12:56448352  failed failed failed failed failed failed failed failed failed 
 rs703842 chr12:56449006 3′UTR 0.309 0.328 0.285 0.34 0.329 0.690 0.3 0.33 0.502 
CYP19 rs9972359 chr15:49279146  0.417 0.414 0.871 0.402 0.383 0.528 0.768 0.759 0.825 
 rs16964189 chr15:49281529  0.202 0.21 0.622 0.16 0.162 0.938 0.207 0.235 0.496 
 rs934632 chr15:49283122  0.173 0.178 0.736 0.286 0.242 0.102 0.195 0.207 0.762 
 rs934634 chr15:49287830  0.203 0.209 0.719 0.17 0.166 0.868 0.323 0.327 0.929 
 rs2255192 chr15:49288127  0.203 0.209 0.709 0.173 0.164 0.712 0.323 0.336 0.769 
 rs4646 chr15:49290136 3′UTR 0.24 0.254 0.384 0.448 0.424 0.420 0.256 0.275 0.662 
 rs10046 chr15:49290278 3′UTR 0.557 0.536 0.270 0.374 0.408 0.257 0.317 0.265 0.232 
 rs17601241 chr15:49295166  0.075 0.082 0.527 0.16 0.181 0.375 0.055 0.062 0.762 
 rs700519 chr15:49295260 Arg264Cys 0.036 0.04 0.784 0.022 0.03 0.676 0.155 0.168 0.812 
 rs2899472 chr15:49303347  0.281 0.26 0.213 0.124 0.188 0.005 0.055 0.04 0.458 
 rs12439137 chr15:49303596  0.153 0.112 0.002 0.194 0.217 0.381 0.058 0.085 0.310 
 rs12592697 chr15:49312465  0.33 0.347 0.345 0.402 0.376 0.388 0.305 0.318 0.770 
 rs700518 chr15:49316404 Val80Val 0.544 0.517 0.160 0.343 0.385 0.154 0.28 0.216 0.114 
 rs10519295 chr15:49319939  0.111 0.108 0.784 0.026 0.055 0.025 0.006 0.009 0.714 
 rs10459592 chr15:49323433  0.418 0.428 0.579 0.523 0.463 0.053 0.61 0.605 0.918 
 rs11636686 chr15:49329358  0.534 0.51 0.201 0.325 0.37 0.126 0.159 0.123 0.285 
 rs12050772 chr15:49332163  0.493 0.468 0.178 0.281 0.328 0.100 0.183 0.151 0.369 
 rs2414099 chr15:49336074  0.138 0.151 0.327 0.258 0.258 0.984 0.165 0.225 0.117 
 rs11636639 chr15:49350384  0.481 0.453 0.130 0.303 0.366 0.032 0.53 0.41 0.012 
 rs2470176 chr15:49371231  0.139 0.146 0.607 0.186 0.156 0.208 0.524 0.525 0.995 
 rs2470158 chr15:49375687  0.093 0.098 0.607 0.075 0.074 0.953 0.116 0.12 0.884 
 rs17523880 chr15:49379835  0.147 0.134 0.314 0.088 0.106 0.334 0.043 0.025 0.277 
 rs2470152 chr15:49382264  0.551 0.524 0.167 0.407 0.43 0.457 0.451 0.457 0.907 
 rs17523922 chr15:49386497  0.097 0.102 0.664 0.064 0.075 0.506 0.018 0.037 0.257 
 rs3751592 chr15:49393870  0.303 0.266 0.042 0.369 0.372 0.942 0.291 0.224 0.143 
 rs3751591 chr15:49394002  0.166 0.16 0.659 0.169 0.189 0.414 0.043 0.049 0.742 
 rs1902584 chr15:49398946  0.082 0.08 0.908 0.131 0.122 0.650 0.037 0.077 0.083 
 rs2445762 chr15:49405000  0.264 0.247 0.315 0.397 0.379 0.556 0.427 0.401 0.587 
 rs7168331 chr15:49408275  0.384 0.362 0.241 0.49 0.485 0.873 0.579 0.525 0.253 
 rs7174997 chr15:49409420  0.168 0.192 0.092 0.157 0.164 0.763 0.037 0.062 0.243 
 rs868475 chr15:49423746  0.05 0.059 0.281 0.12 0.112 0.693 0.299 0.256 0.317 
 rs2470164 chr15:49425853  0.566 0.501 0.003 0.392 0.389 0.911 0.131 0.056 0.027 
 rs2445773 chr15:49430233  0.174 0.181 0.633 0.276 0.261 0.585 0.207 0.213 0.885 
CYP1A1 rs2198843 chr15:72788283  0.166 0.147 0.233 0.389 0.44 0.676 0.5 0.546 0.853 
 rs1048943 chr15:72800038 Val462Ile 0.045 0.037 0.490 0.279 0.278 0.993 0.033 0.011 0.239 
 rs4646421 chr15:72803245  0.162 0.238 0.185 0.643 0.473 0.013 0.7 0.493 0.083 
 rs2470893 chr15:72806502  0.304 0.316 0.504 0.127 0.121 0.778 0.049 0.053 0.850 
CYP24A1 rs2762929 chr20:52199602  0.389 0.399 0.601 0.369 0.387 0.531 0.683 0.725 0.329 
 rs8118441 chr20:52199792  0.139 0.142 0.837 0.186 0.186 0.994 0.311 0.278 0.445 
 rs6068810 chr20:52202758  0.055 0.045 0.251 0.034 0.019 0.128 0.043 0.025 0.277 
 rs6097807 chr20:52202862  0.243 0.255 0.472 0.268 0.247 0.432 0.488 0.448 0.399 
 rs4809957 chr20:52204578 3′UTR 0.215 0.212 0.816 0.247 0.232 0.547 0.438 0.388 0.283 
 rs2762934 chr20:52204668 3′UTR 0.192 0.187 0.779 0.134 0.162 0.210 0.146 0.179 0.362 
 rs1570669 chr20:52207834  0.323 0.337 0.421 0.312 0.3 0.684 0.604 0.58 0.620 
 rs2296239 chr20:52208935 Pro375Pro 0.204 0.213 0.557 0.247 0.233 0.580 0.439 0.392 0.318 
 rs6068816 chr20:52214498 Thr248Thr 0.108 0.109 0.948 0.106 0.079 0.130 0.067 0.046 0.334 
 rs4809958 chr20:52215845  0.156 0.152 0.794 0.16 0.12 0.056 0.085 0.068 0.486 
 rs3787554 chr20:52216087  0.093 0.091 0.795 0.085 0.045 0.005 0.03 0.037 0.709 
 rs2244719 chr20:52216265  0.451 0.484 0.088 0.552 0.569 0.596 0.244 0.262 0.659 
 rs2762941 chr20:52217059  0.388 0.376 0.505 0.296 0.319 0.434 0.506 0.531 0.605 
 rs2181874 chr20:52217885  0.256 0.241 0.355 0.253 0.289 0.192 0.36 0.423 0.179 
 rs4809960 chr20:52219480  0.232 0.237 0.738 0.242 0.209 0.198 0.11 0.154 0.179 
 rs2296241 chr20:52219626 Ala184Ala 0.484 0.47 0.447 0.554 0.559 0.887 0.549 0.497 0.279 
 rs2245153 chr20:52219813  0.181 0.194 0.390 0.18 0.131 0.027 0.315 0.289 0.563 
 rs2585428 chr20:52220304  0.449 0.473 0.194 0.351 0.352 0.949 0.433 0.503 0.143 
 rs13038432 chr20:52220709  0.084 0.083 0.919 0.052 0.045 0.606 0.012 0.022 0.466 
 rs6022999 chr20:52221420  0.241 0.239 0.930 0.225 0.233 0.782 0.594 0.627 0.485 
 rs2248359 chr20:52224925  0.396 0.396 0.991 0.36 0.343 0.566 0.646 0.627 0.668 

NOTE: Significant P values are in bold.

Abbreviation: MAF, minor allele frequency.

*

Assumes Hardy-Weinberg equilibrium.

Markers not in Hardy-Weinberg equilibrium (P < 0.01) in both the controls and the cases.

Genotyping of 104 of the 120 SNPs was done with the GoldenGate assay of the VeraCode technology with the use of the BeadXpress Reader System according to the manufacturer's protocol (Illumina). SNPs rs2066479 (HSD17B3), rs1048943 (CYP1A1), rs2740574 (CYP3A4), and rs700519 (CYP19) were done with a standard TaqMan allelic discrimination assay with the use of an ABI 7900HT Sequence Detection System with the SDS 2.1 software (Applied Biosystems). Genotyping of the seven SNPs within CYP1B1 was as previously described (14). For genotyping of rs2234693 and rs9340799 within ESR1 and rs523349 and rs9282858 within SRD5A2, refer to Hernandez et al. (2006; ref. 15) and Torkko et al. (2008; ref. 16), respectively. SNP rs17115144 (CYP17) was analyzed by restriction fragment digestion of the amplified product as outlined by Lunn et al. (1999; ref. 17). Primers and probe sequences are available upon request. To ensure the reliability of the results, duplicate samples and/or known genotyped samples were included in the analysis as quality controls.

Statistics

Haploview version 4 beta 15 was used to check for Hardy-Weinberg equilibrium for each SNP and to measure linkage disequilibrium between the SNPs in the controls and cases of each race/ethnicity (ref. 18; http://www.broad.mit.edu/mpg/haploview/).

The allele frequency for each SNP was determined in each ethnic group, and the frequencies among the case-control groups were compared with the use of the χ2 test. Association analyses were stratified by ethnicity and done with the R statistical software version 2.5.1. The odds ratio (OR) and its 95% confidence interval (CI) were estimated by unconditional logistic regression as a measure of the associations between genotypes and prostate cancer risk. We tested for additive, dominant, and recessive associations. To correct for multiple testing, we used the method of Storey and Tibshirani (2003) based on the concept of false discovery rate (19). This estimation showed that, for P < 0.01, the probability that the association is expected to be a true positive in our sample group is >50%. To estimate the independent effect of a significant SNP while adjusting for other SNPs, we used a generalized linear model function from the R statistical package so that all SNPs are entered into a single multivariate logistic regression model. SNPs in this model were taken to have additive effects.

The random forest (RF) algorithm, implemented in R (http://www.r-project.org/) was applied to assess the importance of the polymorphisms and potential nonlinear interactions with respect to the outcome variable (20). For each RF analysis, the number of trees to grow (“ntree”) was set to 5,000, and the number of predictors to be randomly selected at each tree node (“mtry”) was set to 20. To better characterize and validate the “important” polymorphisms identified by both the RF algorithm as well as single SNP analysis, and to determine possible interactions between the top ranked or significant SNPs within the same or different genes, we used the generalized multifactor dimensionality reduction (GMDR) method (ref. 21; The GMDR P values and other significance tests are representative of consistency with RF but do not indicate independent confirmation of RF findings). An exhaustive search of all possible n models was done for subsets of selected SNPs (with n between 1 and 10). We used 10-fold cross-validations for each set of n SNPs, and the model with the lowest misclassification error was selected. P values were determined by the sign test, a robust nonparametric test implemented in the software. We also calculated empirical P values for interaction based on 1,000 permutations. Missing values were imputed for both the RF and GMDR methods. The RF method imputes missing data by computing a weighted average of subjects with completed data, in which the weights are similarity measures to the subjects with missing data. The MDR data tool was used to impute missing values in the GMDR analysis (http://sourceforge.net/project/showfiles.php?group_id=131001&package_id=144548).

The cumulative effect of combined genotypes on prostate cancer risk was estimated by counting the number of genotypes associated with prostate cancer on the basis of the best-fitting genetic inheritance from single SNP analysis. ORs and their 95% CIs were calculated for men carrying any combination of one, two, or more alleles associated with prostate cancer compared with men carrying none of the risk genotypes with the use of unconditional logistic regression analysis. We selected SNPs that were not in linkage disequilibrium with each other (D' < 0.8). If several SNPs presented higher linkage disequilibrium values, we chose a SNP in a coding region above an intronic SNP and also selected the most significant SNP.

For all statistical analyses, age (modeled as a continuous predictor) was used as covariate. Individuals with missing data for a particular analysis were removed from the analysis. All statistical tests were two-sided, and significance was set at P < 0.05.

A hypothetical function for SNPs that were significant for single SNP analysis and/or selected by RF was assessed with the use of in silico analysis of transcription factor binding sites: both possible alleles of each SNP were tested for their binding capability to human transcription factors with the use of the web tool 'transcription element search system' (TESS; ref. 22; http://www.cbil.upenn.edu/cgi-bin/tess/tess). The options used were 21 bases of genomic sequence around each SNP (10 bases on either side of the SNP) and a string-based search query with default settings. A log-likelihood score of >16 for a pretty good match (deficit ≤1.0) and >18 for a mismatch (deficit >1.0) were used as cut-off values for reporting.

Single SNP Analysis

A total of 120 SNPs were genotyped in 2,452 samples. The majority of SNPs (>80%) were successfully genotyped in >90% of the samples. Three SNPs failed, and one SNP was not polymorphic in our sample. Table 2 gives the minor allele frequencies of the SNPs estimated in each of the three ethnicities in our study sample. Differences in allele frequency distribution among the ethnicities were found for several SNPs (e.g., rs4646450 in CYP3A4 shows a minor allele frequency of ∼15% for the A allele in non-Hispanic Caucasians, of ∼30% in Hispanic Caucasians, and of ∼85% in African Americans). Significant case/control differences of allele frequencies at a level <0.05 were observed for eight polymorphisms in non-Hispanic Caucasians, for seven SNPs within Hispanic Caucasians, and for five SNPs in African Americans (Table 2). One SNP, rs6201, showed deviation from Hardy-Weinberg equilibrium (P < 0.01) in the cases and controls of all three ethnic study groups. In non-Hispanic and Hispanic Caucasians, rs10923823 was also not in Hardy-Weinberg equilibrium in the cases nor the controls, and variant rs3751592 was out of Hardy-Weinberg equilibrium in non-Hispanic Caucasians. Although the error rate was <0.2%, SNPs that were not in Hardy-Weinberg equilibrium were left out for further statistical analyses in the respective study groups.

Logistic regression analysis was done for all SNPs in the context of recessive, dominant, and additive models after adjustment for age. Thirteen SNPs in four genes were significantly associated with prostate cancer risk in non-Hispanic Caucasians at the P < 0.05 level. For Hispanic Caucasians, 19 SNPs in six genes were found to be significant for prostate cancer risk, and in African Americans, 5 SNPs in four genes showed significant results. The estimated ORs and 95% CIs for significant results in all three ethnicities are shown in Supplementary Tables S1, S2, and S3. After correction for multiple testing, estimating a false discovery rate of <50% for P < 0.01, five SNPs remained significant in non-Hispanic Caucasians, rs1819698 and rs1538989 within HSD3B2, rs12439137, and rs2470152 and rs2470164 within CYP19 (P = 0.001-0.008; Table 3). There was a moderate increase (∼1.5) or decrease (∼0.6) in the ORs associated with these significant results. In Hispanic Caucasians, four SNPs remained significant after correction for multiple testing (P = 0.003-0.009), three SNPs within CYP19 (rs934632, rs10519295, and rs10459592) and one SNP within CYP24A11 (rs3787554). The P values were associated with a >2-fold increase or decrease in OR (Table 3). In African Americans, the association with prostate cancer risk remained significant for rs3751592 in CYP19, with P = 0.008 and a >2.5 increase in OR for the AG heterozygote carriers. In all three sample groups, one or multiple single SNPs within CYP19 showed significant association with prostate cancer. Except for rs1819698 (CYP19), located within the 3′ untranslated region of the gene, all other significant SNPs are within intronic regions of the genes.

Table 3.

Significant results from individual SNP effects on prostate cancer in non-Hispanic Caucasians, Hispanic Caucasians, and African Americans after correction for multiple testing

GeneSNPNon-Hispanic Caucasians
P
GenotypeControls (N)Cases (N)OR*95% CI
HSD3B2 rs1819698 GG 673 439 1.00 —  
  AA 15 2.66 1.13-6.24 0.025 
  AG 157 138 1.34 1.03-1.74 0.029 
  AA/AG vs GG 166 153 1.41 1.09-1.82 0.008 
  AA vs AG/GG 15 2.50 1.07-5.85 0.035 
  A# vs G# 839 592 1.41 1.12-1.77 0.003 
HSD3B2 rs1538989 GG 272 157 1.00 —  
  AA 155 142 1.55 1.15-2.11 0.005 
  AG 412 294 1.20 0.93-1.54 0.160 
  AA/AG vs GG 567 436 1.29 1.02-1.64 0.033 
  AA vs AG/GG 155 142 1.39 1.07-1.80 0.014 
  A# vs G# 839 593 1.24 1.07-1.45 0.005 
CYP19 rs12439137 AA 645 362 1.00 —  
  GG 1.10 0.38-3.20 0.863 
  AG 165 145 1.56 1.20-2.03 0.001 
  GG/AG vs AA 174 151 1.53 1.19-1.99 0.001 
  GG vs AG/AA 0.99 0.34-2.87 0.979 
  G# vs A# 819 513 1.45 1.14-1.84 0.003 
        
CYP19 rs2470152 GG 238 166 1.00 —  
  AA 197 106 0.75 0.55-1.02 0.067 
  AG 404 322 1.16 0.90-1.49 0.257 
  AA/AG vs GG 601 428 1.02 0.80-1.29 0.873 
  AA vs AG/GG 197 106 0.68 0.52-0.89 0.005 
  A# vs G# 839 594 0.87 0.76-1.03 0.121 
CYP19 rs2470164 CC 127 191 1.00 —  
  AA 126 114 0.58 0.41-0.82 0.002 
  AC 206 282 0.91 0.68-1.21 0.512 
  AA/AC vs CC 332 396 0.78 0.60-1.03 0.076 
  AA vs AC/CC 126 114 0.62 0.46-0.83 0.001 

 

 
A# vs C#
 
459
 
587
 
0.77
 
0.65-0.91
 
0.003
 
Gene SNP Hispanic Caucasians
 
    P 

 

 
Genotype
 
Controls (N)
 
Cases (N)
 
OR*
 
95% CI
 
 
CYP19 rs934632 GG 228 93 1.00 —  
  AA 25 10 1.03 0.46-2.31 0.939 
  AG 140 91 1.71 1.17-2.51 0.006 
  AA/AG vs GG 165 101 1.61 1.11-2.32 0.011 
  AA vs AG/GG 25 10 0.82 0.37-1.79 0.613 
  A# vs G# 393 194 1.33 0.99-1.78 0.061 
CYP19 rs10519295 AA 350 185 1.00 —  
  GG ND ND 0.978 
  AG 43 0.34 0.15-0.76 0.009 
  GG/AG vs AA 43 0.39 0.18-0.84 0.016 
  GG vs AG/AA ND ND 0.978 
  G# vs A# 393 194 0.45 0.22-0.94 0.034 
CYP19 rs10459592 CC 120 38 1.00 —  
  AA 91 47 1.82 1.06-3.12 0.029 
  AC 182 109 2.03 1.28-3.22 0.003 
  AA/AC vs CC 273 156 1.96 1.27-3.05 0.003 
  AA vs AC/CC 91 47 1.13 0.73-1.73 0.585 
  A# vs C# 393 194 1.34 1.04-1.74 0.026 
CYP24A1 rs3787554 GG 358 162 1.00 —  
  AA ND ND 0.979 
  AG 35 31 2.05 1.18-3.56 0.011 
  AA/AG vs GG 35 32 2.12 1.23-3.66 0.007 
  AA vs AG/GG ND ND 0.979 
  A# vs G# 393 194 2.14 1.25-3.66 0.005 
        
Gene SNP African Americans
 
    P 

 

 
Genotype
 
Controls (N)
 
Cases (N)
 
OR*
 
95% CI
 
 
CYP19 rs3751592 AA 72 40 1.00 —  
  GG 13 0.83 0.29-2.35 0.725 
  AG 22 32 2.55 1.29-5.06 0.008 
  GG/AG vs AA 35 39 1.90 1.03-3.51 0.040 
  GG vs AG/AA 13 0.61 0.22-1.67 0.330 
  G# vs A# 107 79 1.26 0.81-1.96 0.298 
GeneSNPNon-Hispanic Caucasians
P
GenotypeControls (N)Cases (N)OR*95% CI
HSD3B2 rs1819698 GG 673 439 1.00 —  
  AA 15 2.66 1.13-6.24 0.025 
  AG 157 138 1.34 1.03-1.74 0.029 
  AA/AG vs GG 166 153 1.41 1.09-1.82 0.008 
  AA vs AG/GG 15 2.50 1.07-5.85 0.035 
  A# vs G# 839 592 1.41 1.12-1.77 0.003 
HSD3B2 rs1538989 GG 272 157 1.00 —  
  AA 155 142 1.55 1.15-2.11 0.005 
  AG 412 294 1.20 0.93-1.54 0.160 
  AA/AG vs GG 567 436 1.29 1.02-1.64 0.033 
  AA vs AG/GG 155 142 1.39 1.07-1.80 0.014 
  A# vs G# 839 593 1.24 1.07-1.45 0.005 
CYP19 rs12439137 AA 645 362 1.00 —  
  GG 1.10 0.38-3.20 0.863 
  AG 165 145 1.56 1.20-2.03 0.001 
  GG/AG vs AA 174 151 1.53 1.19-1.99 0.001 
  GG vs AG/AA 0.99 0.34-2.87 0.979 
  G# vs A# 819 513 1.45 1.14-1.84 0.003 
        
CYP19 rs2470152 GG 238 166 1.00 —  
  AA 197 106 0.75 0.55-1.02 0.067 
  AG 404 322 1.16 0.90-1.49 0.257 
  AA/AG vs GG 601 428 1.02 0.80-1.29 0.873 
  AA vs AG/GG 197 106 0.68 0.52-0.89 0.005 
  A# vs G# 839 594 0.87 0.76-1.03 0.121 
CYP19 rs2470164 CC 127 191 1.00 —  
  AA 126 114 0.58 0.41-0.82 0.002 
  AC 206 282 0.91 0.68-1.21 0.512 
  AA/AC vs CC 332 396 0.78 0.60-1.03 0.076 
  AA vs AC/CC 126 114 0.62 0.46-0.83 0.001 

 

 
A# vs C#
 
459
 
587
 
0.77
 
0.65-0.91
 
0.003
 
Gene SNP Hispanic Caucasians
 
    P 

 

 
Genotype
 
Controls (N)
 
Cases (N)
 
OR*
 
95% CI
 
 
CYP19 rs934632 GG 228 93 1.00 —  
  AA 25 10 1.03 0.46-2.31 0.939 
  AG 140 91 1.71 1.17-2.51 0.006 
  AA/AG vs GG 165 101 1.61 1.11-2.32 0.011 
  AA vs AG/GG 25 10 0.82 0.37-1.79 0.613 
  A# vs G# 393 194 1.33 0.99-1.78 0.061 
CYP19 rs10519295 AA 350 185 1.00 —  
  GG ND ND 0.978 
  AG 43 0.34 0.15-0.76 0.009 
  GG/AG vs AA 43 0.39 0.18-0.84 0.016 
  GG vs AG/AA ND ND 0.978 
  G# vs A# 393 194 0.45 0.22-0.94 0.034 
CYP19 rs10459592 CC 120 38 1.00 —  
  AA 91 47 1.82 1.06-3.12 0.029 
  AC 182 109 2.03 1.28-3.22 0.003 
  AA/AC vs CC 273 156 1.96 1.27-3.05 0.003 
  AA vs AC/CC 91 47 1.13 0.73-1.73 0.585 
  A# vs C# 393 194 1.34 1.04-1.74 0.026 
CYP24A1 rs3787554 GG 358 162 1.00 —  
  AA ND ND 0.979 
  AG 35 31 2.05 1.18-3.56 0.011 
  AA/AG vs GG 35 32 2.12 1.23-3.66 0.007 
  AA vs AG/GG ND ND 0.979 
  A# vs G# 393 194 2.14 1.25-3.66 0.005 
        
Gene SNP African Americans
 
    P 

 

 
Genotype
 
Controls (N)
 
Cases (N)
 
OR*
 
95% CI
 
 
CYP19 rs3751592 AA 72 40 1.00 —  
  GG 13 0.83 0.29-2.35 0.725 
  AG 22 32 2.55 1.29-5.06 0.008 
  GG/AG vs AA 35 39 1.90 1.03-3.51 0.040 
  GG vs AG/AA 13 0.61 0.22-1.67 0.330 
  G# vs A# 107 79 1.26 0.81-1.96 0.298 

Abbreviation: ND, not determined.

*

Age-adjusted ORs from unconditional logistic regression analyses.

SNPs with main effect independent from other significant SNPs.

We also tested whether the statistically significant associations with prostate cancer represented independent risk factors in our sample groups. In the non-Hispanic Caucasians, rs2470164 remained significant after correction for the other four significant SNPs (P = 0.0007) under the additive model. SNP rs3787554 in the Hispanic Caucasians remained significant after adjusting for the other significant SNPs (P = 0.004; data not shown). In African Americans, only one SNP remained significant after correction for multiple testing.

Cumulative Effect of Significant SNPs

To evaluate possible cumulative effects of the risk alleles defined in the single SNP analysis, we did an age-adjusted multivariate logistic regression on combinations of at-risk alleles compared with the no-risk allele reference. In non-Hispanic Caucasians, the combination of the three risk alleles not in linkage disequilibrium with each other showed a significant association with prostate cancer (Ptrend = 2.9 × 10-4), and a >2-fold increase in risk (OR, 2.20; 95% CI, 1.44-3.38) was observed for individuals carrying all three risk alleles compared with individuals without any risk allele (Table 4). A similar observation was found in the Hispanic Caucasians, in which the significant association between the combination of the risk alleles of two SNPs and prostate cancer increases the risk significantly (OR, 4.29; 95% CI, 2.11-8.72; Ptrend = 6.4 × 10-5). In African Americans, there was only one SNP that remained significant after correction for multiple testing.

Table 4.

Cumulative effects of risk variants

MarkersNo. of risk genotypesControlsCasesOR (95% CI)*P
Non-Hispanic Caucasians      
rs1819698, rs12439137, rs2470152 174 76 Ref — 
 394 245 1.39 (1.0-1.9) 0.044 
 221 154 1.56 (1.11-2.20) 0.011 
 30 38 2.87 (1.64-5.02) 0.0002 
 Trend   2.20 (1.44-3.38) 0.0003 
Hispanic Caucasians      
rs10459592, s3787554 107 32 Ref — 
 264 136 1.88 (1.17-3.02) 0.009 
 22 26 4.58 (2.19-9.61) 6.1x10-5 
 Trend   4.29 (2.11-8.72) 0.00006 
MarkersNo. of risk genotypesControlsCasesOR (95% CI)*P
Non-Hispanic Caucasians      
rs1819698, rs12439137, rs2470152 174 76 Ref — 
 394 245 1.39 (1.0-1.9) 0.044 
 221 154 1.56 (1.11-2.20) 0.011 
 30 38 2.87 (1.64-5.02) 0.0002 
 Trend   2.20 (1.44-3.38) 0.0003 
Hispanic Caucasians      
rs10459592, s3787554 107 32 Ref — 
 264 136 1.88 (1.17-3.02) 0.009 
 22 26 4.58 (2.19-9.61) 6.1x10-5 
 Trend   4.29 (2.11-8.72) 0.00006 
*

Age-adjusted.

rs2470164 was not included because it is in complete linkage disequilibrium with rs2470152.

rs934632 and rs10519295 were not included because they are in complete linkage disequilibrium with rs10459592.

RF Analysis

To classify and predict the underlying relationships between the predictor variables and prostate cancer, we assessed the importance of the polymorphisms studied in this report by running the RF algorithm. For non-Hispanic Caucasians, Hispanic Caucasians, and African Americans, the number of SNPs chosen were 92, 91, and 97, respectively, corresponding to the missing genotypes in each sample group. Supplementary Table S4 (b) shows the top-12 ranked SNPs as calculated by RF in each of the three sample groups (a choice of 12 SNPs is arbitrary albeit motivated by reasoning that we investigated 12 genes in this report). Of the 5 significant SNPs from single SNP analysis in non-Hispanic Caucasians, 2 are ranked within the first 12 SNPs by RF. In Hispanic Caucasians, 2 of the significant SNPs were selected in the 12 most important SNPs. In African Americans, only 1 of the 3 significant SNPs was scored by RF within the 12 most important SNPs. One SNP, rs11636639 (CYP19), was common in the selection by RF in all three race/ethnic groups. SNP rs2470152 (CYP19) was present in the top-12 list of both non-Hispanic and Hispanic Caucasians. Two SNPs, rs10519295 (CYP19) and rs8176345 (CYP27B1), were within the top 12 SNPs for Hispanic Caucasians and African Americans. Although several of the SNPs selected by RF are coding SNPs, no nonsynonymous SNPs were selected.

SNP-SNP Interaction

The RF algorithm is aimed at detecting the variables that have an important role in the prediction of the outcome, but it does not indicate how these predictors act (i.e., through main effects or through interactive effects involving other variables). Furthermore, because the genes studied are all found in one pathway, possible interactions between single SNPs might exist. SNP-SNP interaction was determined with the use of GMDR. For each racial/ethnicity group, we looked at possible interactions between the significant SNPs found in the single SNP analysis before multiple correction as well as the top-12 ranked SNPs from RF. A summary of the significant results for the models that have minimum prediction error (maximum testing accuracy) and/or maximum cross-validation consistency (CVC) is presented in Table 5.

Table 5.

Best multigenic models, testing accuracies, cross-validation consistencies, and P values identified by GMDR

SNPs into the modelEthnicityBest modelTesting accuracyCVCP*PGenes involved
13 significant Non-Hispanic Caucasians rs1538989-rs2479827-rs17523880-rs2470164 0.63 10/10 0.001 <0.001 HSD3B2, HSD17B3, CYP19 
19 significant Hispanic Caucasians rs523349-rs1810711-rs2066474-rs4646-rs10459592-rs2470152-rs2762941 0.60 10/10 0.001 0.005 SRD5A2, HSB17B3, CYP19, CYP24A1 
5 significant African American rs10012-rs17115144-rs1048691-rs11636639-rs3751592 0.60 10/10 0.05 0.04 CYP1B1, CYP17, CYP27B1, CYP19 
Top 12 RF Hispanic Caucasians rs10989735-rs1810711-rs10459592-rs11636639-rs2470152-rs868475-rs2245153 0.61 10/10 0.01 0.001 HSD17B3, CYP19, CYP24A1 
Top 12 RF African American rs2208532-rs2026002-rs6162-rs8176345-rs12439137-rs2414099-rs11636639-rs1902584-rs7174997-rs13038432 0.63 10/10 0.05 0.014 SRD5A2, HSD17B3, CYP17, CYP27B1, CYP19, CYP24A1 
SNPs into the modelEthnicityBest modelTesting accuracyCVCP*PGenes involved
13 significant Non-Hispanic Caucasians rs1538989-rs2479827-rs17523880-rs2470164 0.63 10/10 0.001 <0.001 HSD3B2, HSD17B3, CYP19 
19 significant Hispanic Caucasians rs523349-rs1810711-rs2066474-rs4646-rs10459592-rs2470152-rs2762941 0.60 10/10 0.001 0.005 SRD5A2, HSB17B3, CYP19, CYP24A1 
5 significant African American rs10012-rs17115144-rs1048691-rs11636639-rs3751592 0.60 10/10 0.05 0.04 CYP1B1, CYP17, CYP27B1, CYP19 
Top 12 RF Hispanic Caucasians rs10989735-rs1810711-rs10459592-rs11636639-rs2470152-rs868475-rs2245153 0.61 10/10 0.01 0.001 HSD17B3, CYP19, CYP24A1 
Top 12 RF African American rs2208532-rs2026002-rs6162-rs8176345-rs12439137-rs2414099-rs11636639-rs1902584-rs7174997-rs13038432 0.63 10/10 0.05 0.014 SRD5A2, HSD17B3, CYP17, CYP27B1, CYP19, CYP24A1 
*

P value from the sign test in GMDR.

P value from the permutation test.

Significant before multiple correction.

Considering SNPs that are significant for single SNP analysis before multiple correction (see Supplementary Tables S1, S2, and S3), we found a significant interaction between SNPs rs1538989, rs2479827, rs17523880, and rs2470164 within HSD3B2, HSD17B3, and CYP19 in the non-Hispanic Caucasians (permutation P < 0.001; CVC, 10/10; testing accuracy, 0.63), suggesting that the four SNPs not only have a main effect, some of which are marginal, but also show a joint action of the three genes. In Hispanic Caucasians, a seven-locus model for rs523349, rs1810711, rs2066474, rs4646, rs10459592, rs2470152, and rs2762941, involving genes SRD5A2, HSB17B3, CYP19, and CYP24A1, showed a significant interaction (permutation P = 0.005; CVC, 10/10; testing accuracy, 0.60). A marginal significant interaction between five SNPs, rs10012, rs17115144, rs1048691, rs11636639, and rs3751592, within the genes CYP1B1, CYP17, CYP27B1, and CYP19 is found in African Americans (permutation P = 0.04; CVC, 10/10; testing accuracy, 0.60), indicating that those SNPs act jointly to confer prostate cancer risk.

Considering the 12 top-ranked SNPs by RF in non-Hispanic Caucasians, no significant interactions were found (data not shown). In Hispanic Caucasians, a highly significant interaction (permutation P = 0.001) was found for 7 of the top-12 ranked SNPs with a CVC of 10/10 and a testing accuracy of 0.61. In African Americans, the model with the highest CVC and maximum testing accuracy included 10 of the 12 top-ranked SNPs (permutation P = 0.014; CVC, 8/10; testing accuracy, 0.63).

Alterations of Binding Sites for Transcription Factors

To investigate possible functional implications of the significant SNPs as well as the SNPs identified by RF, a search for transcription factor binding sites with the use of the transcription element search system was done. The motive behind the analysis is that recent findings show that transcription factor binding sites may be commonly located in introns or other noncoding regions of the genome (23). Three of the five SNPs, rs1538989, rs2470152, and rs2470164, that show significant association with prostate cancer in non-Hispanic Caucasians; one significant SNP, rs3787554, in Hispanics; and the significant SNP rs3751592 in African Americans have changes in transcription factor binding properties related to allelic alterations (Supplementary Table S4, a). The presence of the C allele in rs1538989 (HSD3B2) creates a binding site for a transcriptional activator, PU.1. The Maz binding site, which is a transcriptional activator, is present for the T allele in rs2470152 (CYP19). A binding site for nuclear factor κB, activated by the tumor necrosis factor gene and also a transcriptional repressor of the H-2Kb gene in metastatic tumor cells, is present for the G allele in rs2470164 (CYP19). SNP rs3787554 (CYP24A1) has an Sp1 binding site for the G allele. The A allele of rs3751592 (CYP19) has a binding site for a transcriptional modulator of tumor necrosis factor-α.

Considering the SNPs that were prioritized by RF, 4 of the 12 SNPs (rs4809957, rs1538989, rs2470152, and rs703842) were found to display a binding difference of transcription factors in an allele-specific manner in non-Hispanic Caucasians; 4 of the 12 SNPs (rs11636686, rs10989735, rs1810711, and rs2470152) affect a binding site for transcription factors in Hispanic Caucasians; and 2 SNPs (rs2414099 and rs2208532) show allele-specific alterations of binding motives in African Americans (Supplementary Table S4, b). Of particular interest is the finding for rs4809957 (CYP24A1) in non-Hispanic Caucasians, for which the A allele creates a binding site for the retinoic acid–responsive element. In Hispanic Caucasians, 1 SNP is noteworthy: rs10989735 (HSD17B3), in which the G allele contains an Sp1 transcription factor binding site. In African Americans, the G allele at rs2208532 (SRD5A2) has a LEF1 binding site, which induces neoplastic transformation.

Prostate carcinogenesis is a multistep process involving a multifactorial interplay between genetic and environmental factors; as a result, the effect of individual SNPs are unlikely to be substantial and may be of limited value in predicting risk. A comprehensive approach is required, such as the pathway-based multigenic method integrating multiple polymorphisms that interact in the same pathway. Combining multiple low-penetrant to modestly penetrant SNPs may amplify predictive power.

To estimate prostate cancer risk conferred by individual SNPs, SNP-SNP interactions, and/or cumulative SNP effects, we selected 12 key genes of the steroid hormone pathway and report on the contribution of 116 SNPs within those genes to prostate cancer susceptibility. A total of 886 cases and 1,566 healthy controls from three racial/ethnic groups of the San Antonio Center for Biomarkers of Risk of Prostate Cancer cohort were genotyped.

When analyzed individually, a limited number of SNPs conferred increased risk of prostate cancer after correction for multiple testing (five in non-Hispanic Caucasians, four in Hispanic Caucasians, and three in African Americans). SNPs within CYP19 were significantly associated with prostate cancer in all three ethnic groups, indicating that this gene likely plays a consistent role in prostate cancer susceptibility. CYP19 is a key enzyme in the conversion of androstenedione and testosterone into estrone and estradiol, and a lack of normal function could affect testosterone concentrations, in turn influencing risk of prostate cancer (24, 25). Furthermore, rs2470164 (CYP19) is significant in both non-Hispanic Caucasians and African Americans. Importantly, the allele frequency distribution is significantly different between the two ethnicities with C as the major allele in non-Hispanic Caucasians, whereas this allele is the rare allele in African Americans, and the effect of the SNP differs between both races; male homozygotes of the AA allele have a decreased risk in non-Hispanic Caucasians, whereas heterozygote AC increases risk in African Americans. The SNP rs2470164 has not been reported in previous association studies with prostate cancer.

Of the SNPs analyzed in this report, 35 (∼30%) have been previously studied in prostate cancer. Only two of the significant SNPs from single SNP analysis (rs1819698 in HSD3B2 and rs2470152 in CYP19, both significant in non-Hispanic Caucasians) have been previously reported and showed no association with prostate cancer in non-Hispanic Whites (26, 27). Our findings for SNPs within SRD5A2 are consistent with a previous meta-analysis that did not find evidence for a significant main effect of rs523349 (V89L) or rs9282858 (A49T; ref. 12). However, our results suggest the involvement of SRD5A2 in a polygenic model for prostate cancer risk; in particular, rs523349 and rs2208532 are part of a multilocus model that shows significant interaction in Hispanic Caucasians and African Americans, respectively. We also confirmed previous findings from a meta-analysis by Ntais et al. (10) and Setawian et al. (11) that do not support a significant role of variants within CYP17 in prostate cancer susceptibility in non-Hispanic Caucasians. However, consistent with the findings of Ntais et al. (10), our results of SNP-SNP interaction also suggest that variants within CYP17 may play a role in susceptibility to prostate cancer among African Americans. We could not confirm previous significant findings for association with prostate cancer of coding SNPs found in Caucasians, Hispanics, or African Americans, in particular the nonsynonymous SNPs within SRD5A2 (rs523349 and rs9282858; refs. 2, 28-32), CYP1B1 (rs1800440, rs1056836, and rs1056827; refs. 33-35), HSD17B3 (rs2066479; ref. 36), and CYP1A1 (rs1048943; ref. 37). Explanations for these discrepancies include between-study variability in sample size, diagnostic criteria used, and/or ethnic background of the sample studied.

Higher-order SNP-SNP and/or gene-gene interactions among the reported genetic variants indicate that the combined effect of the polymorphisms with or without significant main effects confer risk for prostate cancer. For each race/ethnicity, significant interactions were found between SNPs with main effects (some marginal) for single SNP analysis. These results suggest interactions among HSD3B2, HSD17B3, and CYP19 in non-Hispanic Caucasians; among SRD5A2, HSB17B3, CYP19, and CYP24A1 in Hispanic Caucasians; and among HSD3B2, CYP1B1, CYP17, CYP27B1, and CYP19 in African Americans. Interactions between SNPs selected by RF, several without a main effect, confirm that complex polygenic models are related to prostate cancer risk. These findings show the importance of SNPs that may not have significance at the individual level and which may be overlooked without an interaction analysis. It should be mentioned that the testing accuracies calculated by GMDR, between 60% and 63%, represent an improvement over the rate of 50% expected under random prediction and shows the importance of multigenic models.

Caution should be used when interpreting these results, especially about the biological significance of these multilocus epistatic models. The statistical modeling of interactions may also not correspond to a true biological interaction (38, 39). However, based on the transcription element search system web tool, hypotheses related to molecular mechanisms resulting in differential activity of genes can be speculated. Several SNPs found to be important in this study, either from single SNP analysis or selected by RF method, have an allele-specific change in transcriptional binding sites and could contribute to transcriptional regulation of oncogenes, as in the case of rs2208532 (SRD5A2), or tumor necrosis factors, as in the case of rs2470164 (CYP19) or rs3751592 (CYP19). Further studies may identify the effect of these allele-specific changes and their contribution to prostate carcinogenesis.

Our results indicate a significant synergistic effect for increasing numbers of potential high-risk genotypes in both non-Hispanic and Hispanic Caucasians. Similar findings for a cumulative effect, in which commonly occurring SNPs incrementally contribute to prostate cancer risk, have been shown for genes from the steroid hormone pathway (30, 40).

The GMDR method has many advantages over others, such as the ability to adjust for covariates and applicability to population-based study designs, including unbalanced case-control samples. In addition, GMDR does not rely on binary splits but does a systematic search through all possible genotype combinations of the SNP variables and may reveal more interactions than methods based on, for example, classification and regression trees.

A limitation of the study is that race/ethnicity is self-reported and, as in any association study, population stratification might lead to confounding results. However, misclassifications are equally likely in cases and controls, and should not have a substantial impact on the outcome. In addition, the majority of markers were not significant, suggesting that population stratification does not represent a significant problem. Another limitation is that we only evaluated the trait prostate cancer using the covariate age. Other endophenotypes and/or interactions with nongenetic factors, as well as the risk factor family history may be essential and could be related to the risk of prostate cancer, which may be missed in this study. For this report, we selected 12 key genes from the steroid hormone pathway. Because the exact mechanism by which these genes might act in prostate cancer development and progression is largely unknown, the genes investigated may represent only a fraction of all relevant genes. Moreover, the SNPs selected in the study are not an exhaustive list of all existing common SNPs and may not fully represent the genetic variability of these genes. However, a careful selection of genes and SNPs was made based on previous functional analysis and/or a previous association study related to prostate cancer risk and, thus, high-priority genes and SNPs were included in the analysis.

In summary, our results support the hypothesis that prostate cancer develops through a variety of interconnected genes, some of which belong to the steroid hormone pathway, and that there are synergistic interactions among these SNPs related to increasing prostate cancer risk. This finding is consistent with the polygenic model for cancer susceptibility, in which the combined effect of variants in many genes, each conferring a small or modest increase in risk, together account for a substantial fraction of the risk for prostate cancer. Our approach of evaluating a large number of genetic variants in a large sample size is one of the first efforts in exploring the effect of high-order gene-gene interactions on prostate cancer risk and might help in improving the ability and accuracy in assessing individual risk of prostate cancer.

No potential conflicts of interest were disclosed.

Grant support: National Cancer Institute Early Detection Research Network grant 5U01CA086402, American Cancer Society grant TURSG-03-152-01-CCE entitled “The Role of Genetic Variation in Prostate Cancer among Hispanics and Blacks,” and National Cancer Institute Cancer Support Grant P30CA54174.

Note: Supplementary data for this article are available at Cancer Epidemiology, Biomarkers and Prevention Online (http://cebp.aacrjournals.org/).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank all the study subjects in the San Antonio Center for Biomarkers of Risk of Prostate Cancer and in the prevalent prostate cancer studies at The University of Texas Health Science Center at San Antonio for their participation, and the San Antonio Center for Biomarkers of Risk of Prostate Cancer clinical staff for their skilled assistance. We utilized the Institutional Genomic Resource Core for genotyping with the Illumina system.

1
Schaid DJ. The complex genetic epidemiology of prostate cancer.
Hum Mol Genet
2004
;
13
Spec No 1:
R103
–21.
2
Makridakis NM, Reichardt JK. Molecular epidemiology of hormone-metabolic loci in prostate cancer.
Epidemiol Rev
2001
;
23
:
24
–9.
3
Hsing AW. Hormones and prostate cancer: what's next?
Epidemiol Rev
2001
;
23
:
42
–58.
4
Cheng E, Lee C, Grayhack J. Endocrinology of the prostate. In: Lepor H, Lawson RK, editors. Prostate diseases. Philadelphia (PA): Saunders; 1993. p. 57–71.
5
Waterman MR, Keeney DS. Genes involved in androgen biosynthesis and the male phenotype.
Horm Res
1992
;
38
:
217
–21.
6
Van Den Akker EL, Koper JW, Boehmer AL, et al. Differential inhibition of 17α-hydroxylase and 17,20-lyase activities by three novel missense CYP17 mutations identified in patients with P450c17 deficiency.
J Clin Endocrinol Metab
2002
;
87
:
5714
–21.
7
Kagimoto M, Winter JS, Kagimoto K, Simpson ER, Waterman MR. Structural characterization of normal and mutant human steroid 17 α-hydroxylase genes: molecular basis of one example of combined 17 α-hydroxylase/17,20 lyase deficiency.
Mol Endocrinol
1988
;
2
:
564
–70.
8
Chung BC, Picado-Leonard J, Haniu M, et al. Cytochrome P450c17 (steroid 17 α-hydroxylase/17,20 lyase): cloning of human adrenal and testis cDNAs indicates the same gene is expressed in both tissues.
Proc Natl Acad Sci U S A
1987
;
84
:
407
–11.
9
Leman ES, DeMiguel F, Gao AC, Getzenberg RH. Regulation of androgen and vitamin d receptors by 1,25-dihydroxyvitamin D3 in human prostate epithelial and stromal cells.
J Urol
2003
;
170
:
235
–40.
10
Ntais C, Polycarpou A, Ioannidis JP. Association of the CYP17 gene polymorphism with the risk of prostate cancer: a meta-analysis.
Cancer Epidemiol Biomarkers Prev
2003
;
12
:
120
–6.
11
Setiawan VW, Schumacher FR, Haiman CA, et al. CYP17 genetic variation and risk of breast and prostate cancer from the National Cancer Institute Breast and Prostate Cancer Cohort Consortium (BPC3).
Cancer Epidemiol Biomarkers Prev
2007
;
16
:
2237
–46.
12
Ntais C, Polycarpou A, Ioannidis JP. SRD5A2 gene polymorphisms and the risk of prostate cancer: a meta-analysis.
Cancer Epidemiol Biomarkers Prev
2003
;
12
:
618
–24.
13
Hazuda HP, Comeaux PJ, Stern MP, Haffner SM, Eifler CW, Rosenthal M. A comparison of three indicators for identifying Mexican Americans in epidemiologic research. Methodological findings from the San Antonio Heart Study.
Am J Epidemiol
1986
;
123
:
96
–112.
14
Beuten J, Gelfond JA, Byrne JJ, et al. CYP1B1 variants are associated with prostate cancer in non-Hispanic and Hispanic Caucasians.
Carcinogenesis
2008
;
29
:
1751
–7.
15
Hernandez J, Balic I, Johnson-Pais TL, et al. Association between an estrogen receptor α gene polymorphism and the risk of prostate cancer in black men.
J Urol
2006
;
175
:
523
–7.
16
Torkko KC, van Bokhoven A, Mai P, et al. VDR and SRD5A2 polymorphisms combine to increase risk for prostate cancer in both non-Hispanic White and Hispanic White men.
Clin Cancer Res
2008
;
14
:
3223
–9.
17
Lunn RM, Bell DA, Mohler JL, Taylor JA. Prostate cancer risk and polymorphism in 17 hydroxylase (CYP17) and steroid reductase (SRD5A2).
Carcinogenesis
1999
;
20
:
1727
–31.
18
Barrett JC, Fry B, Maller J, Daly MJ. Haploview: analysis and visualization of LD and haplotype maps.
Bioinformatics
2005
;
21
:
263
–5.
19
Storey JD, Tibshirani R. Statistical significance for genomewide studies.
Proc Natl Acad Sci U S A
2003
;
100
:
9440
–5.
20
Breiman L. Random forests.
Machine learning
2001
;
45
:
5
–32.
21
Lou XY, Chen GB, Yan L, et al. A generalized combinatorial approach for detecting gene-by-gene and gene-by-environment interactions with application to nicotine dependence.
Am J Hum Genet
2007
;
80
:
1125
–37.
22
Schug JaOGC. TESS: Transcription element search software on the WWW. Technical report CBIL-TR-1997-1001-v0.0 Computational Biology and Informatics Laboratory School of Medicine. PA: University of Pennsylvania; 1997.
23
Long F, Liu H, Hahn C, Sumazin P, Zhang MQ, Zilberstein A. Genome-wide prediction and analysis of function-specific transcription factor binding sites.
In Silico Biol
2004
;
4
:
395
–410.
24
Singh PB, Matanhelia SS, Martin FL. A potential paradox in prostate adenocarcinoma progression: oestrogen as the initiating driver.
Eur J Cancer
2008
;
44
:
928
–36.
25
Miner MM, Seftel AD. Testosterone and ageing: what have we learned since the Institute of Medicine report and what lies ahead?
Int J Clin Pract
2007
;
61
:
622
–32.
26
Douglas JA, Zuhlke KA, Beebe-Dimmer J, et al. Identifying susceptibility genes for prostate cancer-a family-based association study of polymorphisms in CYP17, CYP19, CYP11A1, and LH-β.
Cancer Epidemiol Biomarkers Prev
2005
;
14
:
2035
–9.
27
Cunningham JM, Hebbring SJ, McDonnell SK, et al. Evaluation of genetic variations in the androgen and estrogen metabolic pathways as risk factors for sporadic and familial prostate cancer.
Cancer Epidemiol Biomarkers Prev
2007
;
16
:
969
–78.
28
Forrest MS, Edwards SM, Houlston R, et al. Association between hormonal genetic polymorphisms and early-onset prostate cancer.
Prostate Cancer Prostatic Dis
2005
;
8
:
95
–102.
29
Cicek MS, Conti DV, Curran A, et al. Association of prostate cancer risk and aggressiveness to androgen pathway genes: SRD5A2, CYP17, and the AR.
Prostate
2004
;
59
:
69
–76.
30
Cussenot O, Azzouzi AR, Nicolaiew N, et al. Combination of polymorphisms from genes related to estrogen metabolism and risk of prostate cancers: the hidden face of estrogens.
J Clin Oncol
2007
;
25
:
3596
–602.
31
Salam MT, Ursin G, Skinner EC, Dessissa T, Reichardt JK. Associations between polymorphisms in the steroid 5-α reductase type II (SRD5A2) gene and benign prostatic hyperplasia and prostate cancer.
Urol Oncol
2005
;
23
:
246
–53.
32
Nam RK, Toi A, Vesprini D, et al. V89L polymorphism of type-2, 5-α reductase enzyme gene predicts prostate cancer presence and progression.
Urology
2001
;
57
:
199
–204.
33
Tang YM, Green BL, Chen GF, et al. Human CYP1B1 Leu432Val gene polymorphism: ethnic distribution in African-Americans, Caucasians and Chinese; oestradiol hydroxylase activity; and distribution in prostate cancer cases and controls.
Pharmacogenetics
2000
;
10
:
761
–6.
34
Cicek MS, Liu X, Casey G, Witte JS. Role of androgen metabolism genes CYP1B1, PSA/KLK3, and CYP11α in prostate cancer risk and aggressiveness.
Cancer Epidemiol Biomarkers Prev
2005
;
14
:
2173
–7.
35
Chang BL, Zheng SL, Isaacs SD, et al. Polymorphisms in the CYP1B1 gene are associated with increased risk of prostate cancer.
Br J Cancer
2003
;
89
:
1524
–9.
36
Margiotti K, Kim E, Pearce CL, Spera E, Novelli G, Reichardt JK. Association of the G289S single nucleotide polymorphism in the HSD17B3 gene with prostate cancer in Italian men.
Prostate
2002
;
53
:
65
–8.
37
Chang BL, Zheng SL, Isaacs SD, et al. Polymorphisms in the CYP1A1 gene are associated with prostate cancer risk.
Int J Cancer
2003
;
106
:
375
–8.
38
Siemiatycki J, Thomas DC. Biological models and statistical interactions: an example from multistage carcinogenesis.
Int J Epidemiol
1981
;
10
:
383
–7.
39
Greenland S. Modeling and variable selection in epidemiologic analysis.
Am J Public Health
1989
;
79
:
340
–9.
40
Sobti RC, Gupta L, Singh SK, Seth A, Kaur P, Thakur H. Role of hormonal genes and risk of prostate cancer: gene-gene interactions in a North Indian population.
Cancer Genet Cytogenet
2008
;
185
:
78
–85.
41
Pongubala JM, Nagulapalli S, Klemsz MJ, McKercher SR, Maki RA, Atchison ML. PU.1 recruits a second nuclear factor to a site important for immunoglobulin κ3′ enhancer activity.
Mol Cell Biol
1992
;
12
:
368
–78.
42
Pongubala JM, Van Beveren C, Nagulapalli S, et al. Effect of PU.1 phosphorylation on interaction with NF-EM5 and transcriptional activation.
Science
1993
;
259
:
1622
–5.
43
Kennedy GC, Rutter WJ. Pur-1, a zinc-finger protein that binds to purine-rich sequences, transactivates an insulin promoter in heterologous cells.
Proc Natl Acad Sci U S A
1992
;
89
:
11498
–502.
44
Plaksin D, Baeuerle PA, Eisenbach L. KBF1 (p50 NF-κB homodimer) acts as a repressor of H-2Kb gene expression in metastatic tumor cells.
J Exp Med
1993
;
177
:
1651
–62.
45
Hohmann HP, Brockhaus M, Baeuerle PA, Remy R, Kolbeck R, van Loon AP. Expression of the types A and B tumor necrosis factor (TNF) receptors is independently regulated, and both receptors mediate activation of the transcription factor NF-κB. TNFα is not needed for induction of a biological effect via TNF receptors.
J Biol Chem
1990
;
265
:
22409
–17.
46
Ishii S, Kadonaga JT, Tjian R, Brady JN, Merlino GT, Pastan I. Binding of the Sp1 transcription factor by the human Harvey ras1 proto-oncogene promoter.
Science
1986
;
232
:
1410
–3.
47
Shannon MF, Gamble JR, Vadas MA. Nuclear proteins interacting with the promoter region of the human granulocyte/macrophage colony-stimulating factor gene.
Proc Natl Acad Sci U S A
1988
;
85
:
674
–8.
48
Kliewer SA, Umesono K, Mangelsdorf DJ, Evans RM. Retinoid X receptor interacts with nuclear receptors in retinoic acid, thyroid hormone and vitamin D3 signalling.
Nature
1992
;
355
:
446
–9.
49
Qian X, Samadani U, Porcella A, Costa RH. Decreased expression of hepatocyte nuclear factor 3 α during the acute-phase response influences transthyretin gene transcription.
Mol Cell Biol
1995
;
15
:
1364
–76.
50
Peers B, Voz ML, Monget P, et al. Regulatory elements controlling pituitary-specific expression of the human prolactin gene.
Mol Cell Biol
1990
;
10
:
4690
–700.
51
Cai RL. Human CART1, a paired-class homeodomain protein, activates transcription through palindromic binding sites.
Biochem Biophys Res Commun
1998
;
250
:
305
–11.
52
Anderson GM, Freytag SO. Synergistic activation of a human promoter in vivo by transcription factor Sp1.
Mol Cell Biol
1991
;
11
:
1935
–43.
53
Pascal E, Tjian R. Different activation domains of Sp1 govern formation of multimers and mediate transcriptional synergism.
Genes Dev
1991
;
5
:
1646
–56.
54
Lee JS, Galvin KM, Shi Y. Evidence for physical interaction between the zinc-finger transcription factors YY1 and Sp1.
Proc Natl Acad Sci U S A
1993
;
90
:
6145
–9.
55
Kataoka H, Joh T, Miura Y, et al. AT motif binding factor 1-A (ATBF1-A) negatively regulates transcription of the aminopeptidase N gene in the crypt-villus axis of small intestine.
Biochem Biophys Res Commun
2000
;
267
:
91
–5.
56
Quong MW, Massari ME, Zwart R, Murre C. A new transcriptional-activation motif restricted to a class of helix-loop-helix proteins is functionally conserved in both yeast and mammalian cells.
Mol Cell Biol
1993
;
13
:
792
–800.
57
Aoki M, Hecht A, Kruse U, Kemler R, Vogt PK. Nuclear endpoint of Wnt signaling: neoplastic transformation induced by transactivating lymphoid-enhancing factor 1.
Proc Natl Acad Sci U S A
1999
;
96
:
139
–44.

Supplementary data