Androgen metabolism and JAK/STAT pathway genes and prostate cancer risk

Abstract

Background

Prostate cancer (PC) is the most frequently diagnosed solid tumor in U.S. men. Genome-wide association studies (GWAS) have identified over 40 risk-associated single nucleotide polymorphisms (SNPs), including variants in androgen pathway genes (e.g., KLK3 and AR). Androgens are important in PC and genes involved in this pathway are therefore candidates for conferring susceptibility to PC.

Methods

In this hypothesis-testing study, we evaluated PC risk in association with SNPs in 22 candidate genes involved in androgen metabolism or interactions with the androgen receptor (AR). A total of 187 SNPs were genotyped in 1,458 cases and 1,351 age-matched controls from a population-based study. PC risk was estimated using adjusted unconditional logistic regression and multinomial regression models.

Results

Single SNP analyses showed evidence (p<0.05) for associations with 14 SNPs in 9 genes: NKX3.1, HSD17B3, AKR1C3, SULT2A1, CYP17A1, KLK3, JAK2, NCOA4 and STAT3. The most significant result was observed for rs2253502 in HSD17B3 (odds ratio, OR=0.57, 95% CI: 0.39–0.84). In addition, five SNPs in four genes (CYP17A1, HSD17B4, NCOA4, and SULT2A1) were associated with more aggressive disease (p<0.01).

Conclusions

Our results replicate previously reported associations for SNPs in CYP17A1, HSD17B3, ARK1C3, NKX3.1, NCOA4 and KLK3. In addition, novel associations were observed for SNPs in JAK2, HSD17B4, and SULT2A1. These results will require replication in larger studies.

Introduction

Prostate cancer (PC) is the most commonly diagnosed solid tumor in men with more than 240,000 new diagnoses and more than 33,000 deaths predicted in 2011 [1]. The primary PC risk factors are African ancestry, older age, and a positive family history. However, despite the strong evidence that a hereditary component plays an important role in PC susceptibility from twin studies, linkage, and segregation analyses, specific genetic mutations responsible for inherited risk still remain largely uncharacterized. In recent years, however, major progress in the identification of genetic variants associated with PC has been made possible through genome-wide association studies (GWAS). GWAS studies of PC have led to discovery of over 40 germline genetic variants that are associated with PC susceptibility, explaining approximately 14% of the total genetic variation of PC [2] however, despite these newly identified SNPs large proportion of disease heritability remains unexplained.

Among the numerous biological pathways relevant to PC susceptibility and progression, there is compelling evidence that steroid hormones play a critical role due to affects on prostate cell proliferation and differentiation. The role of testosterone (T) in PC development was proposed 60 years ago by Huggins and Hodges, [3] however the historical view on the relationship between T and PC has been challenged in recent years [4]. Ross et al. demonstrated that the difference in PC incidence among different ethnic groups was correlated with in vivo activity of 5α-reductase and proposed a polygenic model that focused on a series of genes involved in androgen metabolism [5]. Several genes that play important roles in androgen metabolism, SRD5A2, CYP17, AR, and HSD3B1, have been reported previously to have SNPs associated with PC susceptibility, although few such variants have been consistently replicated. In a recent study by Lu et al., a publicly available ChIP-on-chip dataset was used to identify 22,447 regions containing AR-binding sites in the genome [6] and test associations between genetic variants within AR binding sites and PC risk [7]. Interestingly, AR-binding sites can mark novel AR target genes, and the study found statistical evidence (p<1.0×10⁻⁵) for genetic variants identified to have strong associations with PC risk from GWAS (8p21 and 8q24) that were located within AR-binding sites. AR signaling is regulated by its co-regulatory factors and has a central role in androgen-dependent (AD) to androgen-independent (AI) progression of the disease. Thus, genes in the androgen metabolism or AR signaling pathway are important candidates for conferring genetic susceptibility to PC.

In this hypothesis-testing study, we evaluated associations of common genetic variants in 15 key genes involved in androgen metabolism (AKR1C1, AKR1C2, AKR1C3, AR, CYP11A1, CYP17A1, HSD17B3, HSD17B4, HSD17B6, HSD3B1, HSD3B2, SRD5A1, SRD5A2, SULT2A1 and UGT2B11) as well as seven genes that are either AR target genes (KLK3, NKX3.1, let-7, and miR-222) or co-regulators of AR activation (JAK2, NCOA4, and STAT3) with PC risk in a population-based case-control study of Caucasians and African Americans. We also tested the associations between SNPs and clinical characteristics of PC.

Materials and methods

Study Population

The study population consists of Caucasian and African American male residents of King County, WA, who were enrolled in one of two population-based case-control studies (study I and study II). The collection methodologies have been described in detail elsewhere [8, 9]. In brief, study I cases were diagnosed between January 1, 1993 and December 31, 1996 and study II cases were diagnosed between January 1, 2002 to December 31, 2005. For both studies, cases had histologically confirmed PC and were identified from the Seattle-Puget Sound Surveillance, Epidemiology, and End Results (SEER) cancer registry. Overall, 2,244 eligible patients were identified and 1,754 (78.2%) were interviewed. Of those interviewed, blood samples yielding sufficient DNA for genotyping were drawn from 1,458 (83.1%) cases. A comparison group of controls with no self-reported physician’s diagnosis of PC was identified by random digit dialing. Controls were frequency matched to cases by 5-year age groups and were recruited evenly throughout each ascertainment period for cases. Overall, 1,645 (67.2%) of 2,448 identified men who met the eligibility criteria were interviewed and blood samples were drawn and DNA was prepared using standard protocols from 1,351 (81.7%) interviewed controls.

Study participants completed detailed in-person interviews conducted by trained male interviewers using standardized questionnaires. Information was collected on family structure, medical and cancer history, and social and demographic factors. Clinical information on cases (Gleason score, tumor stage, and serum prostate-specific antigen (PSA) level at diagnosis) was obtained from the cancer registry. Plasma PSA levels in controls were measured using stored samples, retrospectively. All study procedures were approved by the Institutional Review Board (IRB) of the Fred Hutchinson Cancer Research Center and the National Human Genome Research Institute and written informed consent was obtained from all study participants.

SNP selection and genotyping

A total of 187 SNPs in 22 genes were genotyped and the majority (n=150) were selected as tagSNPs to capture common haplotype blocks and to maximize coverage for each gene. TagSNP selection parameters were set to minor allele frequency ≥ 5% and a pairwise r² threshold of 0.8 using the phased HapMap CEU population data (version 22; http://www.hapmap.org). Tagger (www.broadinstitute.org/mpg/tagger) was used to select tagSNPs and Haploview[10] was used to visualize linkage disequilibrium (LD) blocks of targeted genomic regions (10kb upstream and 5kb downstream of genes of interest). The majority of the candidate genes (n=15) are involved in androgen metabolism, and seven genes were selected based on their potential interaction with the AR. Thirty-seven SNPs were selected based on functional significance and previously published associations. Genotyping was performed using the Applied Biosystems (ABI, http://www.appliedbiosystems.com) SNPlex Genotyping System and allele assignment was carried out with the use of ABI 3730xl DNA analyzer for detection and proprietary GeneMapper software to assign specific SNP alleles. Three SNPs (rs6152, rs10761581, and rs17464626) failed genotyping using the SNPlex system and were genotyped by 5’-endonuclease assay (Taqman) using the ABIPrism 7900HT sequence detection system, according to the manufacturer’s instructions. In addition, the NKX3.1 gene region was resequenced using seven primer pairs (Table S1) in 94 samples selected at random from our case-control study population to identify 11 additional SNPs that were subsequently genotyped in all samples using the SNPlex Genotyping system. In summary, 15 SNPs failed due to a low call rate (<90%) and 11 SNPs had low minor allele frequencies (MAF <0.01) in Caucasians. In the African American subset, 16 SNPs failed due to a low call rate and 12 SNPs had a low MAF of <1%. The genotyping rates were 96.1% and 95.5% for Caucasians and African Americans, respectively. The remaining SNPs had ~99% agreement between 144 blind duplicates distributed across all genotyping batches.

Statistical analysis

Hardy-Weinberg equilibrium (HWE) [11] test for each SNP was assessed in controls separately by race using a χ² test. SNPs with significant departure from HWE (p<0.05) were excluded from further analysis (n=14 for Caucasians and n=7 for African Americans). Odds ratios (ORs) and 95% confidence intervals (CIs) were estimated using unconditional logistic regression [12]. Potential confounding factors (family history, PC screening history and BMI) were evaluated to test whether these covariates changed the risk estimates by ≥ 10%. All association analyses were adjusted for age and stratified by race. Additive, dominant and recessive genetic models were evaluated for all SNPs except for SNPs on the X chromosome (n=6) and SNPs for which there were no variant allele homozygotes. Polytomous regression models were used for assessing the association between SNP genotypes and cases stratified by Gleason score [2–7(3+4) vs. 7(4+3)-10], diagnostic PSA level [<20 ng/mL vs. ≥20 ng/mL], or tumor stage [local vs. regional or distant]. Statistically significant associations between individual SNPs and PC risk were evaluated using likelihood ratio-based test statistics. Multiple comparisons were adjusted for by using permutation approach where permutated data were generated by permuting SNPs within each gene while keeping other covariates and case-control status unchanged. Then recessive, dominant and additive genetic models were fit on the permutated data to obtain the ordered minimal p-values. Repeating this process 1000 times yields a null distribution of the order statistics. The permutation p-value for each SNP is the probability of observing a p-value that is less than or equal to the nominal p-value for the corresponding order statistic. SNPs were considered to be significantly associated with PC risk if both the nominal p-value and permuted p-value were less than 0.05. Nominally significant SNPs were evaluated for interaction with all other nominally significant SNPs by testing pairwise combinations. Haplotype blocks were defined by the Gabriel et al. analysis method, [13] and frequencies and measures of association for each gene were estimated using Hplus v3.1 (http://qge.fhcrc.org/hplus) [14]. Moreover, we tested the significance of androgen pathway SNPs by comparing the proportion of statistically significant genes in the pathway with a nominal 5% significance level using the method described by Gauvreau [15]. All other statistical analyses were performed using STATA v10.1 or PLINK v1.07 [16].

Results

Selected characteristics of the study population are shown in Table 1. The mean age at diagnosis for cases was 59.9 years, and cases had a higher percentage of men with a first-degree family history of PC compared to controls. Study participants were predominantly Caucasian (91.7%), but a higher proportion of cases than controls were African American. Between cases and controls, cases had more frequent PSA test within 5-year period prior to reference date and there were no significant differences with respect to BMI. The majority of cases had localized stage tumors, PSA values less than 10.0 ng/mL, and Gleason scores less than or equal to 7 (3+4).

Table 1.

Selected characteristics of population-based prostate cancer cases and controls in King County, WA

Characteristic	Cases		Controls		ORa	95% CI
	n=1,458	%	n=1,351	%
Age at reference date
35–49	118	8.1	126	9.3
50–54	215	14.7	209	15.5
55–59	357	24.5	358	26.5
60–64	433	29.7	348	25.8
65–69	177	12.1	164	12.1
70–74	158	10.8	146	10.8
Race
Caucasian	1,309	89.8	1,266	93.7	1.00	Reference
African American	149	10.2	85	6.3	1.74	1.32	2.30
First-degree family history of PC
No	1,145	78.5	1,199	88.8	1.00	Reference
Yes	313	21.5	152	11.3	2.19	1.77	2.71
Body mass index (kg/m2)
Normal (<25)	467	32.0	411	30.4	1.00	Reference
Overweight (25–29.9)	707	48.5	650	48.1	0.95	0.80	1.13
Obese (≥30)	284	19.5	290	21.5	0.83	0.67	1.02
Prostate cancer screening historyb
None	157	10.8	182	13.5	1.00	Reference
Digital rectal examination only	258	17.7	519	38.4	0.57	0.44	0.74
PSA test	1,043	71.5	650	48.1	1.94	1.52	2.47
PSA value (ng/mL)c
0–3.9	189	12.9	1,253	92.8
4.0–9.9	814	55.8	80	5.9
10.0–19.9	208	14.2	16	1.2
≥20.0	138	9.47	2	0.1
Missing	109	7.47	0
Gleason score
2–4	72	5.0
5–6	741	51.0
7 (3+4)	408	28.1
7 (4+3)	91	6.3
8–10	140	9.6
Missing	6
Stage of disease
Local	1,141	78.3
Regional	280	19.2
Distant	37	2.5

^aOR adjusted for age.

^bProstate cancer screening history in five years prior to the reference date.

^cPlasma prostate-specific antigen (PSA) level at diagnosis (cases) or at interview (controls).

SNP association tests

Genotype data were available for 147 SNPs (Caucasians) and 152 SNPs (African Americans). Association results of nominally significant SNPs in candidate genes are shown in Table 2 for Caucasians and in Table S2 for African Americans.

Table 2.

Prostate cancer risk estimates for candidate SNPs in androgen-related genes among Caucasian men

Gene	SNP	Allelea	Chr	BP	Genetic Model	MAF		ORb (95% CI)	p-valuec
						Controls (n=1,266)	Cases (n=1,309)
CYP17A1	rs10786712	C/T	10	104586386	Additive	0.407	0.378	0.88 (0.78–0.99)	0.03
HSD17B3	rs2066485	A/G	9	98043085	Additive	0.172	0.147	0.84 (0.72–0.97)	0.02
HSD17B3	rs2253502	T/C	9	98048264	Recessive	0.240	0.215	0.57 (0.39–0.84)	0.004
HSD17B3	rs2257157	T/C	9	98058263	Dominant	0.432	0.462	1.23 (1.03–1.47)	0.03
AKR1C3	rs4881400	T/G	10	5134037	Recessive	0.236	0.238	0.67 (0.47–0.96)	0.03
SULT2A1	rs182420	A/G	19	53064007	Recessive	0.230	0.252	1.41 (1.02–1.94)	0.04
NKX3.1	rs4872175	G/A	8	23590747	Additive	0.435	0.468	1.14 (1.02–1.27)	0.02
NKX3.1	NKX3.1_5	C/T	8	23593330	Additive	0.008	0.013	1.82 (1.02–3.25)	0.04
KLK3	rs174776	C/T	19	56051664	Additive	0.147	0.122	0.81 (0.69–0.95)	0.01
KLK3	rs1058205	T/C	19	56055210	Dominant	0.196	0.174	0.84 (0.71–0.99)	0.04
NCOA4	rs17178655	G/A	10	51231805	Additive	0.222	0.248	1.16 (1.01–1.32)	0.03
NCOA4	rs7350420	T/C	10	51264468	Additive	0.345	0.309	0.85 (0.75–0.95)	0.007
JAK2	rs10429491	C/T	9	5040706	Dominant	0.315	0.284	0.79 (0.67–0.92)	0.002
STAT3	rs744166	T/C	17	37767727	Dominant	0.425	0.404	0.83 (0.70–0.98)	0.03

^aMajor/minor alleles, based on frequencies in controls.

^bOR adjusted for age.

^cNominal P values based on likelihood ratio-based test.

Of the 147 SNPs analyzed in Caucasians, associations between 14 SNPs in nine genes and PC were found to be nominally significant at the 5% level. These included variants in four genes from the androgen metabolism pathway (AKR1C3, CYP17A1, HSD17B3 and SULT2A1), two (KLK3 and NKX3.1) that are regulated by the AR and three (NCOA4, JAK2 and STAT3) that are predicted to interact with androgen pathway genes for regulation of steroid hormone levels. Notably, three independent tagSNPs located in the HSD17B3 gene region were associated with PC risk, with the strongest association observed for rs2253502 for TT homozygote carriers (OR=0.57; 95% CI 0.38–0.84). Permutation testing revealed only one likely significant association for SNP rs10429491 (OR=0.79, 95% CI 0.67–0.92, p_permuted=0.02), which is in the Janus kinase 2 gene, after adjusting for multiple testing. In addition, rs1058205 in the KLK3 gene demonstrated a borderline significant OR of 0.84 (95% CI 0.71–0.99, p_permuted=0.055). Haplotype analysis of selected tagSNPs did not improve associations compared to the single SNP analyses.

We next evaluated the association of selected SNPs with clinical variables. Due to the small number of African Americans in our dataset, we restricted these analyses to Caucasians. Stratification on Gleason score, diagnostic PSA value, or stage showed six SNPs that were associated with disease features at p <0.01 (Table 3) using an additive model. For two of those six SNPs, (rs10786712 in CYP17A1 and rs10429491 in JAK2), results were also observed in the overall risk analyses. Interestingly, the JAK2 SNP showed the strongest association. Specifically, carriers of the T allele showed a 23 % fold risk reduction in cases with lower Gleason scores (OR=0.77, 95% CI 0.64–0.91). The minor alleles of two SNPs in HSD17B4 (rs10478424 and rs6897978) and a SNP in SULT2A1 (rs2547238) showed associations with increased risk of PC in cases with advanced tumor stage. After accounting for multiple comparisons, none of abovementioned SNPs remained statistically significant.

Table 3.

Prostate cancer risk estimates for candidate SNPs in androgen-related genes among Caucasian men stratified by feature of disease aggressiveness

Gene	SNP	Allele	Chr.	BP	Variable	OR (95% CI)a	p-valueb
CYP17A1c	rs10786712	C/T	10	104586386	Stage	0.67 (0.51–0.88)	0.004
HSD17B4	rs10478424	A/T	5	118816919	Stage	1.35 (1.08 – 1.69)	0.009
HSD17B4	rs6897978	A/G	5	118817237	Stage	1.37 (1.09 – 1.72)	0.007
JAK2c	rs10429491	C/T	9	5040706	Gleason	0.77 (0.64 – 0.91)	0.003
NCOA4	rs10761581	T/G	10	51238384	PSA	0.59 (0.40 – 0.87)	0.007
SULT2A1	rs2547238	C/G	19	53074292	Stage	1.43 (1.09 – 1.86)	0.009

^aPolytomous regression was used to generate OR, adjusted for age.

^bp-value < 0.01

^cSNPs with nominal association with PC risk from Table 2.

In African Americans, ten SNPs were analyzed adjusted for family history since risk estimates were altered (≥10%) compared to risk estimates adjusted for age only. Nominally significant associations with overall PC risk were observed for 15 SNPs in six genes (Table S1). Associations of HSD17B4 SNPs with PC risk were unique to African American men and six out of 15 SNPs that showed an association were located in the SULT2A1 gene region. In KLK3, an intronic variant, rs174776, had the strongest association in African Americans where carriers of the TT and TC genotype had a 3.4-fold greater risk (95% CI 1.47–7.71) compared to men with CC genotype. Three SNPs, rs182420, rs174776 and rs1058205, were associated in both African Americans and Caucasians; however, the direction of the relative risk estimate for each SNP differed by race. None of the other nominally significant SNPs observed in African Americans were seen in the Caucasian population. The results presented for African Americans must be interpreted with caution due to the limited sample size.

Pathway analyses using proportion test

The analysis of single SNPs with overall PC risk in Caucasian men demonstrated that four genes in the androgen pathway contain SNPs associated with PC risk (nominal p-values < 0.05), when only 0.75 genes (15 X 0.05) are expected to be significant at the same threshold level for 16 genes analyzed. We found an excess of statistically significant associations over the proportion expected by chance in the androgen metabolizing pathway for PC risk (observed:expected = 3.8, p = 0.001).

Discussion

The role of androgen and its receptor has been shown to be critical in PC development based on data from animal studies as well as clinical trials, which further support its importance in the etiology of PC [17, 18]. Recent results from multiple GWAS of PC based on Caucasian populations have shown that individual SNPs confer low to moderate risk for PC. An alternative approach to GWAS studies is a comprehensive evaluation of multiple genes with polymorphisms that interact in the same pathway, which has been proposed as a better way to study multifactorial diseases such as PC. For the AR pathway, this should include not just key genes involved in androgen metabolism and biosynthesis, but also its co-regulators [19].

In this study, we tested the hypothesis that genetic variants in androgen biosynthesis pathway genes and co-regulators of the AR are associated with PC risk by evaluating polymorphisms within 22 selected candidate genes. Our results confirmed previously reported associations and identified novel associations between PC risk and androgen-related genes. We observed a strong association with PC risk for a SNP in JAK2, rs10429491, which remained significant after correcting for multiple comparisons. We also found suggestive evidence of associations for genetic variants in NKX3.1, KLK3, NCOA4, STAT3, HSD17B3, AKR1C3, CYP17A1 and SULTA1 in Caucasian men. Additionally, we evaluated SNPs stratified by disease aggressiveness and showed that five SNPs were associated with comparatively more aggressive clinical phenotypes and one SNP which was associated with less aggressive disease. We also utilized genotype data from the National Cancer Institute’s Cancer Genetic Markers of Susceptibility (CGEMS) project for indirect replication of 14 nominally associated SNPs with PC susceptibility [20]. Two SNPs were directly genotyped and one SNP was in LD (r²=0.96 in the Caucasian HapMap population) with a genotyped SNP in the CGEMS dataset (http://cgems.cancer.gov/data). CGEMS data replicated our associations for SNPs in AKR1C3, NCOA4 and NKX3.1 and PC risk.

Surprisingly, we found a synonymous SNP, rs10429491 (H162H), located in exon 4 of the JAK2 gene that was associated with a significant reduction in overall PC risk and the risk of less aggressive PC. To our knowledge, this is the first report of an association for a genetic variant in the JAK2 gene with PC susceptibility. A single gain-of-function nonsynonymous mutation (rs77375493, V617F) in JAK2 has been described in myeloproliferative neoplasia that leads to constitutive activation of JAK2 and subsequent activation of STAT3 [21]. However, the V617F activating mutation was found to be absent in normal and malignant prostate samples analyzed by Gu et al. [22]. JAK2 encodes a cytoplasmic protein tyrosine kinase that is essential to cytokine signaling cascade in interaction with STAT3. Janus kinase-signal transducer and activator of transcription (JAK-STAT) signaling pathway is well known to play an important role in carcinogenesis of several cell types [23]. Recently, studies demonstrating involvement of the JAK/STAT pathway in PC development have accumulated. Numerous in vitro functional experiments using PC cell lines have shown that STAT3 was constitutively active in PC cell lines and promotes metastatic progression of PC [24–26]. Progression of PC is often associated with transactivation of the AR, and STAT3 has been suggested as a mechanism to enhance the transactivation of AR to promote transition from androgen-dependent to androgen-independent prostate tumors [25]. We have previously reported associations of SNPs (rs12949918 and rs744166, r²=0.98) in STAT3 with PC risk in an analysis focused on its role in the IL6 signaling pathway, [27] and we included STAT3 in the current analyses to evaluate its potential gene interaction with AR and JAK2.

Among genes involved in steroid hormone metabolism, germline genetic variants in AR, SRD5A2, CYP17A1 and CYP3A are the most studied genes in relation to PC risk [28]. Several studies have reported conflicting results regarding the putative association of rs743572 in the 5’UTR of CYP17A1 [29, 30]. The tagSNP rs10786712 shows evidence of a reduced risk for the variant allele with PC risk as well as advanced stage in our analyses, and is in perfect linkage disequilibrium (LD) with SNP rs743572 (r²=1.0). The observed association most likely reflects a true causal effect for rs743572, further supporting prior evidence of an association between the CYP17A1 variant and PC susceptibility [31]. CYP17A1 encodes a cytochrome P450 17alpha-hydroxylase, a rate-limiting enzyme that catalyzes steroid 17–20 lyase activities at key steps in the testosterone biosynthesis pathway [5] and rs743572 is predicted to increase transcription activity of CYP17A1 subsequently increasing steroid hormone production.

The HSD17B3 gene encodes for hydroxysteroid (17-beta) dehydrogenase 3 that catalyzes the conversion of androstenedione to testosterone in the testis. There have been three published studies that examined the role of SNPs in HSD17B3 with PC risk [29, 32, 33]. The most studied missense substitution, G289S (rs2066479), showed no association in the Cancer Prostate in Sweden (CAPS) population-based, case-control study. In contrast, Beuten et al. and Margiotti et al. reported a statistically significant increase in the risk of PC with rs2066479. None of three SNPs (rs2066485, rs2253502, and rs2257157) that showed nominal association in our data are correlated with rs2066479, thus our data demonstrates an independent association at this locus. CGEMS data show an association for a variant in HSD17B3; however, the CGEMS marker (rs407179) is only weakly correlated with rs2257157 (r²=0.31). Another hydroxysteroid (17-beta) dehydrogenase, HSD17B4, catalyzes the conversion of testosterone to androstenedione. We identified two SNPs (rs10478424 and rs6897978) for which carriers of variant alleles are at a 1.5-fold increased risk of more advanced PC stage in Caucasians. To date, there are no reports of association of genetic variants in HSD17B4 with PC risk or clinicopathologic features.

AKR1C3 and SULT2A1 have been recently identified to play critical roles in prostate carcinogenesis. AKR1C3 is a member of the aldo-keto reductase superfamily that converts DHT to 3á-diol and regulates AR occupancy and transactivation of the receptor. To date, only one suggestive association with familial PC has been reported with SNPs in the AKR1C3 gene [34], and our finding is the first to be identified in sporadic cases. The CGEMS project directly genotyped rs4881400 and replicated our association with a similar magnitude of risk for overall PC (OR=0.78, 95% CI 0.64–0.95, p=0.016).

Sulfotransferase (SULT) 2A1 catalyzes dehydroepiandrosterone (DHEA) sulfation in adrenal cortex and a study by Wilborn et al. reported an association between genetic variants in the SULT2A1 gene and DHEA:DHEA-S ratio in African Americans [35]. We identified nominally significant associations for six SNPs in the SULT2A1 gene among African Americans and the strongest association was observed for rs2547238 (OR=3.17, 95% CI 1.23–8.14) located in the second intron. In Caucasians, the most significant association was observed for rs182420, which is located 1675 bases downstream of SULT2A1, for which carriers of the ‘GG’ genotype are at a 1.4-fold increase in PC risk. Additionally, rs2547238, located in the third intron of SULT2A1, and 10,285bp downstream of rs182420, was associated with increased risk for more advanced stage. To our knowledge, this is the first study to report on the positive associations of SNPs in SULT2A1 with PC risk in African Americans and Caucasians.

Androgen-related genes include co-regulators of AR as well as genes downstream of the androgen metabolism pathway containing AR binding sites (e.g. KLK3). A recent GWAS identified new loci containing putative susceptibility genes at 19q13.33 (KLK3), 8p21 (NKX3.1) and 10q11 (NCOA4) [36]. Prostate-specific antigen (PSA/KLK3) was shown to positively associate with PC risk in candidate gene studies [37–39] prior to the GWAS study. The classical kallikrein gene expressed exclusively in prostate tissue is widely used in the diagnosis and monitoring of the disease. Expression of PSA is regulated by recruitment of AR and its co-activators at functional androgen-response elements (AREs) at the proximal promoter as well as an enhancer region upstream of the gene [40, 41]. Alterations (A-252G and G-158A) in the ARE1 region can lead to a change in regulation of PSA expression [37]. The most significant SNP in the GWAS (rs2735839) is located 600 bp downstream of the KLK3 gene when cases were compared to selected controls with low PSA levels [36, 42]. Recently, the same group identified a novel PC susceptibility coding SNP in exon 4 (rs17632542, I179T) that causes a nonsynonymous change with the potential to alter RNA splicing and affect protein stability [43]. We identified two SNPs (rs174776 and rs1058205) in the 3’UTR of KLK3 that confer a ~25% reduced risk of PC in Caucasians. One of them, rs1058205, was also found to be associated with PC in the multi-stage study by Kote-Jarai et al., demonstrating a similar risk estimate (OR=0.87, 95% CI 0.77–0.99) as we report here. Thus, we have replicated an association of rs1058201 with overall PC risk to further support the notion that common genetic variants in KLK3 contribute to PC development.

The putative prostate tumor suppressor gene NKX3.1 is an androgen regulated homeobox-containing transcription factor and is subject to loss of heterozygosity in human PC. Androgen stimulation has been shown to increase its expression, and He et al. reported that the NKX3.1 gene plays a role in androgen-driven differentiation in prostate tissue [44]. To determine genetic and functional aspects of the 8p21 region harboring NKX3.1 with PC risk, Akamatsu et al. identified a SNP in the 5’UTR through fine mapping of the initial GWAS peak. The variant rs11781886 was shown to affect transcription activity by altering binding affinity of transcriptional factor Sp1 in prostate cells [45]. We found a moderate association of two SNPs in NKX3.1 that confers increased relative risk of PC in Caucasians. SNP rs4872175 is located 1,426 downstream and the newly identified novel variant (NKX3.1_5) is located within the 3’UTR of NKX3.1. The 3’UTR region has been found to contain functional AREs that are predicted to play a crucial role in the regulation of NKX3.1 expression [46], and future studies should include SNPs identified here.

An androgen receptor co-activator (ARA70/NCOA4) is located on chromosome 10q11.2 and since its initial discovery by GWAS, [36, 47] three studies have evaluated the associations between PC risk and variants at 10q11.2 [48–50]. The most significant GWAS SNP, rs10993994, is associated with reduced expression of MSMB and overexpression of NCOA4 in the prostate, thus promoting anchorage-independent growth of prostate epithelial cells. In addition, there is evidence of an association between PC risk and rs1074005, which is an independent locus on 10q11. In this study we found an association between two SNPs (rs17178655 and rs7350420) and PC risk that are not correlated with the abovementioned SNPs. The CGEMS data confirmed the association of rs7350420 where carriers of the variant ‘C’ allele had a reduced overall risk of PC (OR=0.82, 95% CI 0.70–0.97), further supporting our findings.

Our study includes several strengths and weaknesses that should be considered when interpreting these results. One of the strength of our study is that the majority of genes examined were interrogated by tagSNPs to comprehensively analyze variation across the entire gene region. Selected candidate genes comprise key enzymes involved in the androgen metabolism pathway. Using data available from the CGEMS project, we were also able to infer significance of previously reported associations if the particular SNP is in LD with our selected tagSNP. Our population-based case-control design and the moderate sample size for evaluating common genetic variants in association with overall risk of PC in Caucasians are also strengths of the study. However, our sample size was insufficient to evaluate gene by gene interactions, which is a challenge in many studies. In addition, the analyses of African Americans were underpowered and results in this subset should be interpreted with caution.

In summary, our results support the hypothesis that androgen metabolism pathway genes and AR interacting genes contribute to PC susceptibility. We observed significant associations with several SNPs in steroid hormone pathway genes (CYP17A1, HSD17B3, HSD17B4, AKR1C3 and SULT2A1) and co-regulators of AR (NCOA4 and JAK2/STAT3). Moreover, we replicated recently identified associations at three independent loci (KLK3, NKX3.1 and NCOA4) that are likely regulated by androgens within prostate tissues. Novel risk associations were observed for SNPs in JAK2, HSD17B4, and SULT2A1. Furthermore, variants in several genes (CYP17A1, HSD17B4, JAK2, NCOA4 and SULT2A1) were found to be associated with PC risk when analyses stratified by clinicopathologic features. The results reported here suggest that genetic variants in androgen-related genes are independently associated with PC risk and aggressive PC, but larger studies are needed to confirm our findings considering features of aggressive PC.