Inherited Variation in Hormone-Regulating Genes and Prostate Cancer Survival

Purpose: Hormonal manipulation is the mainstay treatment of prostate cancer, notably in advanced stages. Despite initial favorably response, the cancer eventually develops hormone resistance resulting in disease progression and death. However, little is known about genetic determinants of disease progression and prostate cancer–specific death.

Experimental Design: We analyzed a population-based cohort comprising 2,761 men diagnosed with prostate cancer from March 2001 to October 2003 and with complete follow-up through July 2006. During a median follow-up time of 3.8 years, a total of 300 men had died from prostate cancer. We genotyped 23 haplotype tagging single nucleotide polymorphisms in the genes AR, CYP17, and SRD5A2 and used Cox proportional hazards analyses to quantify associations between genotype and risk of dying from prostate cancer.

Results: The variant ‘A’ allele of an AR promoter single nucleotide polymorphism, rs17302090, was borderline associated with a 50% increased risk of dying from prostate cancer (95% confidence interval, 1.0-2.3; P = 0.07). This finding was more pronounced in patients who received hormonal therapy as primary treatment at diagnosis (hazard ratio, 1.9; 95% confidence interval, 1.3-2.9; P = 0.007). We did not identify any associations between CYP17 or SRD5A2 variation and prostate cancer–specific death.

Conclusions: Our results suggest that inherited genetic variation in the androgen receptor gene affects hormonal treatment response and ultimately prostate cancer death.

Prostate cancer is the most common cause of cancer-related death among men in Sweden (1) and the second most common in the United States (2). In addition, many men experience a latent tumor with no clinical symptoms. This heterogeneity in clinical behavior emphasizes the need to find markers for disease progression, particularly for localized tumors. Today, clinicians rely chiefly on tumor-node-metastasis (TNM) stage, Gleason score, tumor extension biopsy, and serum levels of prostate-specific antigen (PSA) for risk evaluation but these factors only explain a limited fraction of the variability in outcome.

Androgens are essential in prostate cancer biology, early prostate cancer is androgen dependent, and hormonal therapy is the standard treatment for advanced disease (3). Androgen withdrawal results initially in tumor regression but eventually hormone-refractory cancer cells emerge signaling that prognosis has become poor (4). However, the duration of response to hormonal therapy varies widely. Although the causes for this variation remain unknown, several experimental observations suggest that the androgen receptor gene (AR, OMIM 313700) is involved (5). AR expression increases during hormonal treatment failure and AR amplification is found in one third of hormone-refractory tumor cells (6).

Due to its variable clinical course, its androgen dependence and the estimated strong genetic contribution to its etiology, prostate cancer is a promising target for identification of genetic factors that are important for disease progression. For example, genetic variation in hormone-regulating genes may modify the effect of hormonal withdrawal therapy and consequently time to progression (7). Specifically, it has been suggested that a polymorphic CAG repeat in exon 1 of the AR gene affects response to hormonal therapy (8, 9).

To explore the mechanisms behind prostate cancer progression and hormone resistance, we used follow-up data for 2,761 incident case patients included in a Swedish population-based prostate cancer case-control study [Cancer Prostate in Sweden (CAPS)]. In this study, multiple single nucleotide polymorphisms (SNP) and haplotypes in three key hormone-regulating genes (AR, CYP17, and SRD5A2) were shown recently to alter prostate cancer susceptibility (10). We now set out to elucidate if germ-line genetic variation within these genes is important also for prostate cancer outcome.

Study cohort. Prostate cancer cases enrolled in the CAPS study, a population-based case-control study described earlier (10), constituted the cohort. All men living in the northern and central parts of Sweden, under the age of 80 years, and all men living in the Stockholm region and southeastern part of Sweden, under the age of 65 years, with an incident histologically verified prostate cancer diagnosed between March 2001 and October 2003 were eligible to participate. In total, 3,648 prostate cancer patients were identified and invited. Of these men, 3,161 (87%) agreed to participate and for those, a blood sample and a questionnaire about risk factors and family history was collected.

Detailed clinical information, including Gleason score, serum PSA at the time of diagnosis, TNM, and initial primary treatment at time of diagnosis, was collected from the Swedish National Prostate Cancer Register (Table 1; ref. 11).⁶

At the time of this analysis, DNA was available for 2,761 case patients comprising the study cohort. Each participant gave written informed consent. The research ethical committees at the Karolinska Institutet and Umeå University approved the study.

Follow-up. Each study participant is identified through his individually unique national registration number, which includes date of birth. Using this registration number, complete follow-up for prostate cancer–specific mortality was achieved up until July 15, 2006 through record linkage to the Swedish Cause of Death Registry.⁷

For individuals deceased after December 31, 2003, cause of death was established through review of death certificates by an experienced oncologist. We defined prostate cancer–specific death as those who had prostate cancer classified as the underlying cause of death. The average follow-up time was 3.8 years (range, 0.3-5.8 years). A total of 502 (17%) individuals died during follow-up and of those, 347 (11%) had prostate cancer classified as their underlying cause of death. For this study, DNA was available for 300 men who had died from prostate cancer.

SNP selection and tagging methodology. The approach for SNP selection and haplotype tagging has been described earlier (10). AR SNPs were selected from the public database SNPper.⁸

To select SNPs for CYP17 and SRD5A2, phase II data from the International HapMap project⁹ were used.

We first genotyped 52 selected AR SNPs in a randomly selected subset of 94 CAPS control subjects. By haplotype deviation analysis using the haplotype tagging SNP2 software,¹⁰

we selected as haplotype tagging SNPs four SNPs that explained >95% of the haplotype diversity. Two more SNPs were added after comparison with haplotype composition from HapMap data. In total, six haplotype tagging SNPs in the AR gene were genotyped in all case patients. For CYP17 and SRD5A2, haplotype tagging SNPs were selected with tagSNPs software (12). In total, we selected 7 SNPs in CYP17 and 11 SNPs in SRD5A2. These SNPs explained >95% of the haplotypic variation.

Genotyping. Genotyping details have been described earlier (10). Shortly, we used matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (Sequenom, Inc.; ref. 13). PCR assays and associated extension reactions were designed using the SpectroDESIGNER software (Sequenom) and primer sequences are available on request.

Statistical analyses. Follow-up began at date of diagnosis and ended at date of death or last follow-up (July 15, 2006). A likelihood ratio test of a covariate equal to number of rare alleles (0, 1, or 2) based on the Cox proportional hazards model was used to test for association between SNPs and prostate cancer–specific death as implemented in STATA version 8.0 (14). The proportional hazards assumption was tested based on Schoenfeld residuals.

If a SNP was associated with overall risk of dying from prostate cancer, we explored the effect conditional on the following established known prognostic factors grouped as follows: PSA at diagnosis (<20, 20-50, and ≥50 ng/mL), Gleason score (2-7 and 8-10), and TNM stage (T₀-T₂N₀/N_xM₀/M_x, T₃-T₄N₀/N_xM₀/M_x, and T₀-T₄N₁/M₁). To avoid further assumptions about proportional hazards, analysis was stratified on these factors instead of including them as indicator variables in the regression model. To estimate haplotypic effects on survival, we used the THESIAS software (15), which allows analysis of censored data using a standard Cox proportional hazards formulation (16). Haplotype frequencies and association between haplotypes and prostate cancer death were estimated simultaneously using a stochastic EM algorithm for likelihood maximization (17). Simultaneous estimation of haplotype frequencies and haplotype effects results in a better efficiency in parameter estimation (16).

Hazard ratios and corresponding confidence intervals were estimated for each haplotype by comparison to a reference haplotype chosen as the most frequent one. A likelihood ratio test was used as a global test of association between haplotypes and prostate cancer death. Effects associated with rare haplotypes (frequency, <0.05) were not estimated. Analyses were done both unadjusted and with adjustment for prognostic factors grouped as described above.

Details about our genotyped SNPs and haplotype distribution have been published (10). Genotyping failed for one selected haplotype tagging SNP in CYP17, rs17724534. Average genotyping success for the other SNPs was 97.9% (range, 92.3-99.8%). The concordance rate between duplicated samples (n = 330) was 99.96%. All autosomal SNPs were in Hardy-Weinberg equilibrium (P > 0.05) among 1,705 CAPS controls as described earlier (10).

SNP analyses. The proportional hazards assumption was fulfilled in our analyses. Hazard ratios with corresponding confidence intervals for polymorphisms in AR, CYP17, and SRD5A2 are presented in Table 2. Carriers of the variant ‘A’ allele of the AR SNP rs17302090, located 10 kb upstream of AR, were at 50% higher risk of dying form prostate cancer compared with ‘C’ carriers (Table 2). With adjustment for known prognostic factors, such as Gleason score, TNM stage, and PSA at diagnosis, rs17302090 ‘A’ carriers remained at significantly increased risk of death [hazard ratio (HR), 1.7; 95% confidence interval (95% CI), 1.1-2.6; P = 0.02]. In addition, we found no significant association between rs17302090 and Gleason score, TNM stage, and PSA at diagnosis (data not shown).

Among patients who received hormonal therapy as their primary treatment, rs17302090 ‘A’ carriers had an ∼2-fold higher risk to die from prostate cancer compared with noncarriers (HR, 1.9; 95% CI, 1.2-3.0; P = 0.007; Fig. 1). This subgroup comprised a total of 918 patients of whom 269 died from prostate cancer during follow-up. Adjusting the analysis for Gleason score, TNM stage, and PSA levels at diagnosis did not alter this association (HR, 1.9; 95% CI, 1.2-2.9; P = 0.006). In contrast, we found no significant association between rs17302090 SNP and survival among patients who did not receive any hormonal treatment (HR, 1.6; 95% CI, 0.5-5.3; P = 0.46). However, the statistical power in this latter group was limited due to few events (n = 31). No other polymorphism was associated with prostate cancer prognosis (Table 2).

Haplotype analyses. Details about haplotype composition have been described earlier (18). We identified four AR haplotypes with prevalence >5%. Crude analysis did not reveal any global association with survival (P_global = 0.41; Table 3) nor did multivariate regression adjusted for known prognostic factors (P_global = 0.10). However, one single haplotype ‘AAGAGT’ with a frequency of 6% was associated with a 30% increased risk to die from prostate cancer (95% CI, 1.0-1.6; P = 0.02) in the multivariate analysis. This association was similar in a univariate analysis among patients who received hormonal therapy (HR, 1.3; 95% CI, 1.1-1.7; P = 0.00; Table 3).

No common haplotype in CYP17 was significantly associated with prostate cancer death (P_global = 0.56; Table 3). The SRD5A2 locus is divided into two haplotype blocks with four haplotypes >5% in each block. We observed no association in either block (block 1, P_global = 0.98; block 2, P_global = 0.85; Table 3).

In this study, we investigated associations between common germ-line polymorphisms in the hormone-related genes AR, CYP17, and SRD5A2 and prostate cancer survival. Overall, we found no association between polymorphisms within these genes and prostate cancer progression. However, an AR promoter SNP was associated with prostate cancer–specific death, specifically among patients who received hormonal treatment. Overall, men carrying the variant ‘A’ allele of rs17302090 were at a 50% higher risk to die of prostate cancer compared with ‘C’ allele carriers, whereas patients who received initial hormonal treatment the ‘A’ allele carriers were at almost a 2-fold higher risk. Adjusting for known prognostic factors did not change the association, suggesting that this SNP is unrelated to former known prognostic factors. Indeed, there was no correlation between rs17302090 and Gleason score, PSA, or TNM stage. Probably, rs17302090 acts as an effect modifier for hormonal treatment, suggesting that inherited genetic variation at the AR locus may affect response ability to hormonal treatment and ultimately prostate cancer death.

The rs17302090 SNP is located on a single haplotype at the AR locus, reaching over the entire gene. The 180-kb AR gene lies in a conserved region situated in one single haplotype block. This implies that although we observed the strongest association in the promoter region, another genetic variant harboring at the same haplotype may be responsible for the association. We and others have earlier shown that genetic variation at the AR locus is associated with prostate cancer incidence, although different SNPs are associated with risk and survival (10). Probably, different mechanisms in AR and androgen mediation are involved in different steps of the natural history of prostate cancer. For example, the AR signaling pathway with several downstream genes may be involved in malignant transformation, whereas treatment response failure and progression may be a consequence of AR characteristics such as activity and amplification.

Early prostate cancer relies on androgens to develop. If the cancer progresses so that hormonal treatment becomes indicated, refractory cells emerge sooner or later, fostering further disease progression. AR is amplified in a significant fraction of hormone-refractory prostate cancers (6) and such amplification is correlated with cellular proliferation in recurrent prostate cancer (19). These findings suggest that AR amplification is one of the mechanisms that trigger prostate tumor growth in the absence of androgens. Hormone-refractory prostate cancers experience a higher frequency of somatic AR mutations compared with untreated tumors (5). These mutations may affect gene properties by altering the ligand binding domain (20) and thereby broadening AR specificity so that other steroid hormones and even antiandrogens can activate the gene (5).

Although androgen deprivation reduces circulating levels of hormones substantially, these are not completely absent (21). Patients who have undergone hormonal treatment require only a small amount of androgens to initiate androgen action, indicating that mutated ARs are hypersensitive to androgens (3). Furthermore, men with high pretreatment serum levels of testosterone encounter better hormonal treatment response (22) and a better prognosis (23, 24). If germ-line genetic variation affects AR activity and thereby testosterone levels, we expect to see a genetic influence on hormonal therapy response and ultimately survival.

Inherited genetic markers and prostate cancer outcome is a fairly unexplored area. In a few small studies, the majority assessed the AR CAG repeat, located in exon 1. PSA recurrence, failure of hormonal response, and disease-specific survival were primary end points with contradictory results (8, 9, 25–29). We genotyped recently the CAG repeat in the CAPS case-control study (18). Altogether, 199 cases of 1,461 had died from prostate cancer making CAPS the largest study on CAG repeat and clinical outcome thus far. We observed, however, no association between number of CAG repeats and prostate cancer death.

To our knowledge, this is the first comprehensive assessment of genetic variation in AR, CYP17, and SRD5A2 and prostate cancer–specific death. Strengths of our study include its large size and homogeneous population. Despite the relatively short time of follow-up, 11% of the patients had died from prostate cancer, indicating the high proportion of clinical relevant cancers in our study cohort. All men were diagnosed during a limited period (2.5 years), which reduces possible effects of clinical trends in diagnostic activity and disease management. We obtained complete information about primary treatment at diagnosis through the Swedish National Prostate Cancer Register. Assuming an allele frequency of 0.06, CAPS has 70% power to detect a HR of 1.9 or higher. But, as we analyzed a large number of SNPs with no adjustment for multiple testing, our findings may be due to chance. Hence, continued follow-up of our study and confirmation in independent, ideally even larger cohorts of prostate cancer patients is crucial. If confirmed, our results might help targeting men with early treatment failure. To elucidate the relationship between AR, hormonal therapy response, and prostate cancer outcome, we need a better understanding of how the AR gene is involved in the natural history of prostate cancer.

We thank all study participants in the CAPS study; Ulrika Undén for skillfully coordinating the study center at Karolinska Institutet; all urologists who included their patients in the CAPS study or provided clinical data to the national register of prostate cancer; Karin Andersson, Susan Lindh, Gabriella Thorén, and Margareta Åswärd at the Regional Cancer Registries and CAPS steering committee, including Drs. Jan Adolfsson, Jan-Erik Johansson, and Eberhart Varenhorst; Sören Holmgren and the personnel at the Medical Biobank in Umeå for skilfully handling the blood samples; the personnel at Mutation Analysis Facility at Karolinska Institutet and Dr. Cecilia Lindgren for excellent genotyping; and Prof. Paul Dickman for kindly providing scripts for power calculations.