Prostate Cancer Susceptibility Variants Confer Increased Risk of Disease Progression

Prostate cancer is the most commonly diagnosed cancer in men in the United States and is the second leading cause of cancer death. The widespread use of serum prostate-specific antigen (PSA) for prostate cancer screening and the use of extended pattern biopsy techniques have led to a profound stage migration, with the majority of prostate tumors detected today being of low grade, limited volume, clinically localized, and associated with low serum PSA (1). Nevertheless, it remains unclear exactly which men should receive the most aggressive treatment and which tumors are most likely to progress following treatment, raising the possibility of overtreatment of prostate cancer (2).

At present, there are clinical and pathologic measures such as Gleason grade, tumor stage, and prognostic nonograms that help predict which prostate tumors have poor prognosis and the highest potential for recurrence and/or death (3). Even with such prognostic measures, most men diagnosed with this disease undergo radical prostatectomy or radiation treatment. Side effects due to treatment and their effect on quality of life highlight the importance of sparing men from unnecessary treatment and the need for specific markers affecting disease prognosis. There is clear clinical utility in identifying additional factors that can distinguish between prostate tumors that will likely recur, progress rapidly, and be life-threatening versus those that might not have substantial effect on morbidity or mortality once treated.

Genetic factors offer a potentially promising avenue for further clarification of prostate cancer aggressiveness (4, 5). Most studies of inherited genetic variants, however, have focused on risk factors for disease development and not progression among men diagnosed with this disease. For example, genome-wide association studies (GWAS) have clearly identified genetic variants that affect prostate cancer susceptibility (5, 6). Some of these promising variants are in genes on chromosome 10q11.23 (MSMB), 17q12 (TCF2, HNFB1), and Xp11.22 (NUDT10, NUDT11) as well as at loci on 8q24 with indeterminate biological effects (7-11). To date, most of these susceptibility loci do not seem to be associated with clinicopathologic features (9-17) or prostate cancer mortality (12, 18). Although these findings suggest that these loci affect the early stages of prostate cancer development, their effects on disease progression have only been recently investigated, as one study reported two variants at 10q11 and one variant at 8q24 influencing biochemical recurrence following radical prostatectomy (19). In the present study, we examined 26 prostate cancer risk loci, previously identified by GWAS, for their association with prostate cancer progression following treatment of aggressive prostate cancer.

Study subjects were identified from two previously described hospital-based prostate case-control studies from Cleveland, OH and Detroit, MI (dates of diagnoses from January 2001 to August 2004 and August 1999 to July 2004, respectively; refs. 20, 21). Cases were restricted to men with histologically confirmed tumors meeting the D'Amico (22) intermediate or high-risk criteria to focus on the most clinically relevant forms of prostate cancer with a high potential for disease progression. Specifically, the prostate cancer cases had a Gleason score of >7, tumor-node-metastasis stage >T2c, or a diagnostic PSA level of >10 ng/mL. For Gleason scores and stage information, we used values from prostatectomy whenever available; otherwise, biopsy values were used. A total of 788 “aggressive” prostate cancer cases were identified that had follow-up PSA and disease management information available from medical records. The overall median follow-up time was 50.2 months (4.2 y).

Biochemical recurrence as manifested by PSA relapse was defined according to treatment regimen. For radical prostatectomy patients, biochemical recurrence was defined as a posttreatment PSA of ≥0.3 ng/mL. For radiotherapy patients, biochemical recurrence was defined by a 2-ng/mL increase in PSA above the posttreatment nadir according to the American Society for Therapeutic Radiology and Oncology Phoenix Consensus definition (23). For both treatments, the initiation of salvage hormones/radiation treatment or the presence of metastases also indicated prostate cancer progression. Herein, we use the term prostate cancer progression to reflect biochemical recurrence, initiation of salvage hormones/radiation treatment, or the presence of metastases following radical prostatectomy or radiation treatment. Time of prostate cancer progression was the earliest time observed for each treatment definition of PSA relapse, initiation of salvage hormones/radiation treatment, or presence of metastases. Patients with no evidence of progression were censored at the date of the last PSA test. Institutional review board approval was obtained for the study, and informed consent was acquired from all participants.

We evaluated 19 single nucleotide polymorphisms (SNP) identified by recent GWAS of prostate cancer susceptibility (Table 1; refs. 9-11, 16, 17, 24) as well as 7 putative risk SNPs from a GWAS of aggressive prostate cancer (25) given our focus on progression of more aggressive forms of disease. All SNPs were genotyped using the TaqMan allelic discrimination assay except rs629242, which was genotyped by the Illumina GoldenGate assay. All assays were undertaken by individuals blinded to the status of prostate cancer progression. The average genotyping success rate across the 26 SNPs was 98.7%, and the average genotype concordance rate for 2% duplicates was 99.0%. All SNP\s were in Hardy-Weinberg equilibrium among all patients for each racial/ethnic group (P > 0.01).

The association between each SNP and prostate cancer progression was assessed with hazard ratios (HR) and 95% confidence intervals (CI) estimated by Cox proportional hazards regression. The SNPs were evaluated with codominant and log-additive genetic models. All analyses were adjusted for age, institution, and estimates of genetic ancestry (discussed below), stage, Gleason score, PSA at diagnosis, and treatment. Progression-free survival rates were estimated using the Kaplan-Meier method, and differences in survival distributions by genotypes were evaluated with log rank tests.

To determine which combination of the 26 SNPs were independently associated with prostate cancer progression, a forward stepwise Cox regression analysis was conducted (α = 0.10 for inclusion and exclusion) with each SNP coded according to a log-additive model. The risk allele was modeled as the allele conferring a significantly increased risk of progression among our study population. To examine the combined effects of these SNPs, a prognosis score was created by adding the alleles of each SNP (log additive) which were independently associated with progression from stepwise regression analysis.

To correct for potential population stratification within our study population, genetic ancestry was estimated by principal component analysis and included in our regression models (26). Specifically, 37 ancestry-informative markers for European and African ancestry were selected from the DNA Test Panel from Illumina and were genotyped on all subjects by the Illumina GoldenGate assay. Genotype data were analyzed by the EIGENSOFT software, and the leading principal components distinguishing European and African ancestry were estimated (26). The first principal component from this analysis delineated individuals of European and African descent and was used as an estimate of genetic ancestry. To investigate the potential heterogeneity of genotype effects across race/ethnicity and treatment regimens, we included interaction terms between genotype and race/ethnicity, and genotype and treatment (radical prostatectomy/radiation therapy) in our regression models. All reported P values are two-sided.

The characteristics of the 788 cases in our study are presented in Table 2. Approximately 70% of this study population is White, and 30% is African-American. Of the 788 cases, 228 patients experienced prostate cancer progression and 560 patients did not progress. Patients experiencing prostate cancer progression were more likely to be younger, have higher Gleason score/tumor stage/PSA levels at diagnosis, and received radical prostatectomy treatment compared with the 560 patients who did not experience prostate cancer progression. The median follow-up times for patients that progressed and those that did not were 26.8 and 56.8 months, respectively.

Nine of the 26 SNPs tested were statistically significantly associated with prostate cancer progression: rs12621278, rs7652331, rs1545985, rs629242, rs9364554, rs12155172, rs1512268, rs10896449, and rs5945572 (Table 3). Of the 26 SNPs, there was no evidence of heterogeneity in effects by race/ethnicity (P > 0.10), and one SNP (rs17021918) displayed heterogeneous effects by treatment regimen (P = 0.02). For rs17021918, there was no association with progression among radical prostatectomy patients (HR per increase in minor allele, 0.85; 95% CI, 0.66-1.10), although a positive association was observed among radiotherapy patients (HR per increase in minor allele, 1.53; 95% CI, 1.02-2.30).

The ITGA6 locus, rs12621278, displayed the most statistically significant association, with a 2.4-fold increased risk of prostate cancer progression associated with the AG genotype compared with the AA genotype (95% CI, 1.51-3.92). Applying a conservative Bonferroni correction to adjust for the 26 tested SNPs (P = 0.002), rs12621278 reached a study-wide significance with P = 2.6 × 10⁻⁴. The median progression-free survival times for patients carrying the rs12621278 AG versus the AA genotypes were 46.4 and 50.3 months, respectively (P = 0.01).

Five of the nine associated SNPs (rs12621278, rs629242, rs9364554, rs4430796, and rs5945572) showed independent associations with progression based on stepwise regression analysis. By summing the alleles of these five independent SNPs (allele range, 0-7), we created a risk score and estimated a 29% increase in risk of prostate cancer progression per additional risk allele (HR, 1.29; 95% CI, 1.12-1.47; P = 3.0 × 10^-4; Table 4). Patients who carried four to seven risk alleles (22.1% of patients that progressed and 15.2% of patients that did not progress) had a 2.19-fold increased risk of prostate cancer progression compared with patients with less than one risk allele (95% CI, 1.40-3.42; P = 6.0 × 10⁻⁴). In a stratified analysis by race/ethnicity, similar associations were seen for the risk score for Whites and African-Americans (per additional risk allele: HR_Whites, 1.27; 95% CI, 1.08-1.48; and HR_{African-Americans}, 1.31; 95% CI, 1.01-1.72).

Statistically significant differences in survival distributions (P = 0.01) were observed when comparing patients with a risk score of <1, 2, 3, and 4 to 7 prognosis alleles (Fig. 1). We observed a decrease in progression-free survival time associated with an increase in the number of risk alleles. Specifically, the median progression-free survival times were 54.3, 49.6, 47.5, and 43.5 months for patients with prognosis scores of <1, 2, 3, and 4 to 7 alleles, respectively.

Our study examined 26 prostate cancer susceptibility loci for their effect on prostate cancer progression. We found that 9 of 26 prostate cancer susceptibility loci were associated with prostate cancer progression after accounting for known predictors of prostate cancer outcomes—diagnostic PSA, Gleason score, stage, and primary treatment. Moreover, five loci (ITGA6, NUDT10/NUDT11, KIAA1211, SLCC22A3, and HNF1B/TCF) showed independent associations with progression.

The strongest associations were observed at the ITGA6 locus (rs12621278) that encodes for α6 integrin, a protein involved in important features of tumor invasion—cell adhesion, migration, and signaling. This locus was recently identified by Eeles et al. (16) in a follow-up study of their stage 1 GWAS, whereby the G allele of rs12621278 was associated with a reduced risk of prostate cancer (odds ratio, 0.75; 95% CI, 0.70-0.80; P = 8.7 × 10⁻²³). Interestingly, the G allele in our study was associated with a poor prognosis of prostate cancer. Although it might be expected that the protective effect of the G allele for prostate cancer development would align with improved prognosis, our association with worse prognosis might relate to pleiotropy in the biological effects of α6 integrin. Alteration in integrin function as a result of genetic variation may affect numerous downstream signaling effects as well as cell adhesion and migration activity. Prostate tumors persistently express α6 integrin (27-29), and in vitro and in vivo studies have linked its expression to increased tumor cell invasion, migration, and metastasis (30-32). This supports the role of ITGA6 in prostate cancer progression, and other downstream effects of integrin signaling might have other specific effects on disease development.

We created a prognosis score to understand the collective effects of the five independent loci on prostate cancer progression which were identified by stepwise regression. Patients with four to seven risk alleles had a 2.2-fold increased risk, and an 11-month shorter time to prostate cancer progression, than patients with no risk alleles. We recognize that a prognosis score based on associated SNPs may exaggerate the level of statistical significance, and replication of the proposed associations is key.

Penney et al. (12) and Wiklund et al. (18) created similar cumulative scores based on prostate cancer susceptibility variants to examine the association with prostate cancer mortality. Both studies reported no associations with either the cumulative score or individual variants for this outcome (12, 18). Penney et al. (12) examined eight previously associated variants at the 8q24 and 17q loci among 2,746 prostate cancer cases with 243 prostate-specific deaths in the Physicians' Health Study and at the Fred Hutchinson Cancer Research Center. We examined two (rs1859962 and rs4430796) of the eight SNPs and found that these SNPs were also not associated with disease progression. Wiklund et al. (18) tested 16 prostate cancer loci among 2,875 prostate cancer cases with 440 prostate-specific deaths in the Cancer of the Prostate in Sweden study. We tested 12 of these 16 loci, observing significant associations between 2 (rs9364554 and rs10896449) of the 12 loci and prostate cancer progression. Although there seems to be a consistency between prostate cancer mortality and progression for the majority of SNPs tested, the inconsistency for rs9364554 and rs10896449 might relate to differences in biological processes for mortality versus progression, heterogeneity in study populations such as treatment regimens, or chance findings. With biochemical recurrence representing the majority (>95%) of outcome events, it is more reliable in predicting survival following prostatectomy and radiation therapy among high-risk patients such as those with a Gleason score of>7 (similar to our study population; ref. 33), yet it is not equivalent to clinical recurrence nor prostate cancer–specific mortality. Clearly, larger studies are warranted in well-characterized study populations that can evaluate the entire course of the disease from diagnosis to biochemical recurrence, metastasis, and death.

To our knowledge, only one previous study has examined prostate cancer risk loci for their associations with biochemical recurrence (34). In a study of 320 localized prostate cancer patients from Taiwan, three prostate cancer risk variants (rs1447295 at 8q24, and rs7920517 and rs10993994 at 10q11) were associated with recurrence (34). In our study, we did not examine rs1447295 and rs7920517, and observed no association with rs10993994. Of the nine variants that were associated with progression in our study, only two (rs9364554 and rs5945572) were tested in the Taiwanese study, and both showed no associations among this study population. The discrepancies in findings may largely reflect heterogeneity in study populations, and/or chance findings. In particular, our study population consisted of more aggressive prostate cancer, receiving radical prostatectomy and radiotherapy treatment, in contrast with the Taiwanese study of localized prostate cancer, which received radical prostatectomy treatment. In addition, heterogeneity in allele frequency is also possible such as rs5945572, which has differing allele frequencies between Asians [minor allele frequency (MAF) = 0.09; ref. 34], African-Americans (MAF = 0.26), and Whites (MAF = 0.38). Furthermore, the number of SNPs (n = 26) tested in our study under both a codominant and log-additive model raises the concern of false-positive associations due to multiple hypothesis testing. False-positive associations are a major concern for genetic association studies, and it is essential that additional studies replicate these findings to establish a clear genetic association.

We investigated whether these 26 prostate cancer risk variants also influenced clinicopathologic features of the disease (data not shown). We observed a few nominal associations with age at diagnosis (rs5945572, P = 0.047; rs10896449, P = 0.031), Gleason score (rs9364554, P = 0.039), and stage (rs2735839, P = 0.006). Three of these SNPs associated with age at diagnosis (rs5945572 and rs10896449) and Gleason score (rs9364554; an important indicator of prostate cancer outcomes), were also associated with progression. In addition, a similar association between rs2735839 (KLK2/KLK3) with stage of disease (13) has been previously reported. Notably though, these findings should be interpreted with caution given the large number of tests that were conducted and the chance of false-positive associations.

Seventeen of the prostate cancer risk loci reported by GWAS that were tested in our study were not associated with prostate cancer progression. An implicit premise is that risk factors for disease development would be the same as those involved in progression and mortality; however, there may be different biological pathways for prostate cancer initiation and progression with distinct underlying risk factors. Thus, a locus associated with prostate cancer risk may not necessarily be associated with progression of disease. Another possibility is that the current study was unable to detect weak associations.

Our focus on patients with more aggressive disease, who are eligible to undergo radical prostatectomy or radiation therapy, allows us to examine pathologic types that are most clinically relevant with the greatest likelihood of poor prognosis; however, this range in pathologic types is not representative of the full array of clinical prostate cancers because most cancers detected in the PSA era are of low volume, low grade, and of uncertain biological significance. Ultimately, our results need to be replicated in independent case populations to fully establish their potential clinical utility. In addition, given the limited power and sample size of African-Americans in our study, we were unable to conduct analyses focused on this ethnic group.

There is a great need to further our understanding of the risk factors for prostate cancer progression. This is especially important in light of possible morbidity and detrimental effects on quality of life arising from the standard treatment options for prostate cancer (that is, surgery, androgen deprivation, and/or radiotherapy). Although there are several established clinical measures that help determine which men might have poor prognosis (e.g., tumor stage and grade), there remains room for improvement. Moreover, in light of the large number of men diagnosed with prostate cancer, and the relatively high proportion of men experiencing biochemical recurrence (∼33% of radical prostatectomy and radiation therapy patients; refs. 35, 36), additional markers for progression that have reasonable predictive power might be clinically useful. Furthermore, by identifying new progression markers, key insights may be gained in understanding the biological mechanism of disease process.

In summary, our study suggests that five of the recently detected prostate cancer susceptibility loci identified through GWAS also influence a patient's risk of progression beyond the known clinicopathologic predictors. This work builds on the importance of these genetic variants in prostate cancer biology, and it is essential that our findings are replicated in larger studies. By examining how inherited variation affects disease progression, we will further advance our understanding of prostate cancer prognosis, with the ultimate goal of better targeting aggressive treatment to patients most likely to suffer from disease recurrence, metastasis, and death.

E. Klein: honoraria from the speakers bureau for Abbott and Genomic Health. The other authors disclosed no potential conflicts of interest.

We thank the participants of this study, who have contributed to a better understanding of the genetic contributions to prostate cancer, and Chandana Reddy, Alwyn Reuth, Ross Boylan, and Joel Medford for their contributions to this project.

Grant Support: NIH grants (CA88164, CA94211, CA98683, and ES011126).

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.