Sequence Variants of Estrogen Receptor β and Risk of Prostate Cancer in the National Cancer Institute Breast and Prostate Cancer Cohort Consortium

Background: Estrogen receptor β (ESR2) may play a role in modulating prostate carcinogenesis through the regulation of genes related to cell proliferation and apoptosis.

Methods: We conducted nested case-control studies in the Breast and Prostate Cancer Cohort Consortium (BPC3) that pooled 8,323 prostate cancer cases and 9,412 controls from seven cohorts. Whites were the predominant ethnic group. We characterized genetic variation in ESR2 by resequencing exons in 190 breast and prostate cancer cases and genotyping a dense set of single nucleotide polymorphisms (SNP) spanning the locus in a multiethnic panel of 349 cancer-free subjects. We selected four haplotype-tagging SNPs (htSNP) to capture common ESR2 variation in Whites; these htSNPs were then genotyped in all cohorts. Conditional logistic regression models were used to assess the association between sequence variants of ESR2 and the risk of prostate cancer. We also investigated the effect modification by age, body mass index, and family history, as well as the association between sequence variants of ESR2 and advanced-stage (≥T3b, N₁, or M₁) and high-grade (Gleason sum ≥8) prostate cancer, respectively.

Results: The four tag SNPs in ESR2 were not significantly associated with prostate cancer risk, individually. The global test for the influence of any haplotype on the risk of prostate cancer was not significant (P = 0.31). However, we observed that men carrying two copies of one of the variant haplotypes (TACC) had a 1.46-fold increased risk of prostate cancer (99% confidence interval, 1.06-2.01) compared with men carrying zero copies of this variant haplotype. No SNPs or haplotypes were associated with advanced stage or high grade of prostate cancer.

Conclusion: In our analysis focused on genetic variation common in Whites, we observed little evidence for any substantial association of inherited variation in ESR2 with risk of prostate cancer. A nominally significant (P < 0.01) association between the TACC haplotype and prostate cancer risk under the recessive model could be a chance finding and, in any event, would seem to contribute only slightly to the overall burden of prostate cancer. (Cancer Epidemiol Biomarkers Prev 2007;16(10):1973–81)

Estrogen receptor β (ESR2) is located on chromosome 14q23.1 and comprises nine exons spanning about 62 kb. It is a member of the steroid receptor subgroup of the nuclear receptor gene family, which requires ligand activation for DNA binding and transcriptional activity (1). ESR2 has two transcriptional isoforms and is expressed in prostate epithelium and developing spermatids of the testis (2-5).

The ESR2 gene is coexpressed with ESR1, which is mainly expressed in stromal cells in normal prostates. These two receptors share high sequence homology but are different in ligand binding, transactivation, cofactor interactions, and putative heterodimerization (2-4, 6, 7). Large changes in the relative levels of mRNA of ESR2 were observed in prostate cancer tissue as compared with normal tissue, but this was not observed for ESR1 (8-10). Decreased expression of ESR2 mRNA was also observed for breast, endometrium, ovary, and colon cancer when comparing tumor to normal tissue (7). Expression of ESR2 in prostate tumor cells mediates antiproliferative signals, inhibits the invasion and growth of prostate tumor cells, and regulates apoptosis (8, 11). In high-grade prostatic intraepithelial neoplasia, the variation in ESR2 expression suggests that ESR2 may be a target for ligand-specific activation in the treatment of prostate cancer (7-10).

Information from experimental studies suggests that variations in ESR2 expression may regulate prostate carcinogenesis through the regulation of genes related to cell proliferation and apoptosis, raising the possibility that sequence variants of ESR2 may alter the risk of prostate cancer. In addition to the main effects of ESR2 variants, effect modification by age, body mass index (BMI), and family history of prostate cancer remain to be explored. Here, we report a comprehensive evaluation of germ line variations of ESR2 in relation to prostate cancer.

The National Cancer Institute (NCI) Breast and Prostate Cancer Cohort Consortium (BPC3) is a large multicenter association study (12) that systematically explores the role of genetic variations in the steroid hormone pathway, the insulin-like growth factor pathway, and the associated receptor proteins in the etiology of breast and prostate cancer. Study participants were identified from seven large cohorts: the American Cancer Society Cancer Prevention Study II (CPS-II; ref. 13), the Alpha-Tocopherol, Beta-Carotene Cancer Prevention (ATBC) Study (14); the European Prospective Investigation into Cancer and Nutrition (EPIC), including cohorts from Britain, Denmark, Germany, Greece, Italy, the Netherlands, Spain, and Sweden (15); the Health Professionals Follow-Up Study (HPFS; ref. 16); the Hawaii-Los Angeles Multi-ethnic Cohort Study (MEC; ref. 17); the Physicians Health Study (PHS; ref. 18); and the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial (19). This study was approved by the Institutional Review Board at the corresponding institute of each cohort.

We identified 8,323 prostate cancer cases and 9,412 controls. Whites are the predominant ethnic group in all cohorts except the MEC. The MEC and PLCO were the main contributors of non-White participants. The MEC recruited their cases and controls from Los Angeles and Hawaii, including 922 Whites, 1,313 African Americans, 1,290 Latinos, 945 Japanese Americans, and 138 native Hawaiians. The PLCO includes 2,452 Whites and 424 African Americans. In each cohort, controls were individually or frequency matched on age (controls were disease free at case age at diagnosis) and ethnicity. Some cohorts additionally matched on factors irrelevant to the current analysis, such as fasting status at blood draw. Cases and controls were selected using the same protocol as previous BPC3 publications (20).

Prostate cancer was confirmed by medical records, pathology reports, or cancer registries. The cases were categorized into organ-confined or minimal extraprostatic extension (T1b-T3a and N₀M₀), regionally invasive or metastatic (stage ≥T3b, N₁, or M₁), lower grade (Gleason sum <8), and higher grade (Gleason sum ≥8).

To discover novel missense single nucleotide polymorphisms (SNP), we resequenced ESR2 exons in a panel of 190 advanced breast and prostate cancer cases from five ethnic groups. To determine the haplotype block structure, we genotyped a set of dense SNPs from an ∼90-kb region spanning ESR2 (Fig. 1) in screening samples, which included 349 unrelated women from the MEC with no history of cancer (70 Whites, 70 African Americans, 68 Latinos, 72 Japanese Americans, and 69 Hawaiians). This set included 40 common SNPs with minor allele frequency (MAF) >5% overall and >1% in any one ethnic group.

On the basis of these genotyping data from the screening samples, we used the Gabriel et al. (21) algorithm in the Haploview program³²

to determine haplotype blocks, which are regions of high linkage disequilibrium (LD) and a low recombination frequency. R_h², which measured the correlation between observed and predicted haplotypes, is the statistic for selection of haplotype-tagging SNPs (htSNP). The htSNPs were selected using the stepwise algorithm implemented in the tagSNPs program³³ (22). Empirical studies suggest that tagSNPs chosen using a European-descended screening panel perform at near-nominal levels in other European-descended samples, so we are confident that tagSNPs selected from a panel of MEC Whites capture common variation among White subjects from other cohorts (23-25). TagSNPs chosen using a European-descended screening panel may not achieve nominal tagging efficiency in non-European samples, but a decrease in tagging efficiency is not certain and is a function of the genetic diversity in the target population (26, 27).

Genotyping was done in four core laboratories (Harvard School of Public Health, Boston, MA,; Strangeways Research Laboratory, Cambridge, United Kingdom; National Cancer Institute, Rockville, MD; University of Hawaii, Honolulu, HI) using the fluorogenic 5′-endonuclease assay (TaqMan) with the ABI-Prism 7900. TaqMan assays were designed and optimized for each SNP to complete the ESR2 genotyping of all DNA samples in the BPC3. Quality control (QC) was done by the manufacturer (Applied Biosystems) and by the laboratories of the Cohort Consortium for another 500 test reactions. Detailed assay information for the htSNP in ESR2 is available on a public website.³⁴

For each SNP, sequence validation was done, and a 100% concordance was observed³⁵ (28). The interlaboratory variation was assessed in each laboratory by running assays on 94 samples from the Centre d' Etude du Polymorphismes Humains families, and the completion and concordance rates were >99% (28). At each core laboratory, the internal genotyping QC was done by genotyping 5% to 10% of blinded samples in duplicate or triplicate.

A test for deviation from Hardy-Weinberg genotype frequencies was done for each SNP among controls from each cohort by ethnicity (MEC and PLCO) or by country (EPIC). For each cohort (and country or ethnicity within cohort, where applicable), the partition-ligation-expectation-maximization algorithm was applied to estimate haplotype frequencies in each block, as well as subject-specific expected haplotype indicators (22, 29, 30). Conditional logistic regression models (stratified by cohort, ethnicity, and age at diagnosis/selection in 5-year intervals) were used to estimate odds ratios (OR) for prostate cancer in participants carrying either 1 or 2 versus 0 copies of the minor allele/variant haplotype of each SNP/multilocus haplotype. Expectation substitution (31) was used to compare a model with additive effects for each common haplotype (treating the most common haplotype as the reference) to the intercept-only model, while accounting for unknown phase. This provides a global test of association between ESR2 haplotypes and prostate cancer. The cumulative frequency of haplotypes that were rare overall (<5% frequency) was <8% in every ethnic subgroup.

To account for multiple testing, we conducted a single multiple-degree-of-freedom test of association between ESR2 haplotypes and prostate cancer. Given a significant global test, haplotype- and SNP-specific tests can provide some guidance as to which variant(s) contributes to the significant results, although the nominal P values do not control the family-wise error rate for these post hoc comparisons. In addition, we determined the 99% confidence interval (99% CI) and tested for significant associations at the 0.01 level to minimize the chance of false positives.

We used the likelihood ratio test (LRT) to evaluate the heterogeneity in ORs across the cohorts and heterogeneity in the ORs across ethnicity. These tests compared the model with common additive effects for each common haplotype (except the reference) to the model with distinct additive effects for each cohort or ethnicity where the expected numbers of the haplotype in cases and controls under the null were >5.

Age and family history are known risk factors for prostate cancer (32, 33). Serum testosterone peaks from puberty to about age 40 and then declines afterward (34). BMI has also been related to the risk of prostate cancer, although not consistently so in all studies (35). We used the LRT to evaluate whether these factors modified the association between ESR2 SNPs or haplotypes and the risk of prostate cancer by comparing a model with terms for main effects and cross-classified terms for interaction effects to the model with terms for main effects only. Age was dichotomized by the cutpoint 65. BMI was categorized into three groups (<25, 25 to <30, and ≥30). To test the association between ESR2 variants and risk of prostate cancer subtypes (e.g., high-grade or high-stage cancer), we excluded cases without the subtype under analysis from the conditional logistic regression analysis described above. We tested for differences in the strength of association between ESR2 SNPs or haplotypes and advanced-stage (stage ≥T3b, N₁, or M₁) or high-grade (Gleason sum ≥8) prostate cancer. All analyses were conducted with SAS release 9.0 (SAS Institute), and all statistical tests were two sided.

The mean ages at diagnosis and mean BMI were similar by case-control status (Table 1). The study population was composed of 75.3% Whites, 9.9% African Americans, 7.3% Latinos, 5.3% Japanese Americans, 0.78% native Hawaiians, and 1.4% other ethnic groups. SNP genotyping completion rates were above 95% in all cohorts with the exception of ATBC (between 90% and 93%), and we observed no difference in completion rates between cases and controls. For participants with genotyping data, family history was available for 76.3% (n = 6,154) of cases and 70.7% (n = 6,495) of controls; among these, 14% of cases and 9% of controls reported that either his father or at least a brother had prostate cancer. Stage information was available for 72.9% (n = 5,876) of cases with genotyping data; among these, 22% had advanced prostate cancer (stage≥ T3b, N₁, or M₁). Gleason score was available for 67.8% (n = 5,467) of cases with genotyping data, and 18% of them had a Gleason sum of ≥8.

From the resequencing data among 190 cases of advanced breast and prostate cancer, two missense SNPs not previously in dbSNP were identified.³⁶

However, both of these SNPs were monomorphic in the 349 screening samples. A panel of 40 common SNPs (frequency ≥5% overall or >1% in at least one other non-White ethnic group) spanning about 90 kb including ESR2 was genotyped in 349 multiethnic and cancer-free subjects. These SNPs formed one LD block among Whites (Fig. 1), Japanese, native Hawaiians, and Latinos, and three blocks among African Americans. Among Whites, there were four common haplotypes generated by the 40 SNPs in this reference sample (Supplementary Table S1). Four htSNPs (rs3020450, rs1256031, rs1256049, and rs4986938) were selected to predict common haplotypes (5%) with an R_h² of at least 0.7. We also evaluated the performance of these htSNPs in the phase II HapMap CEU samples using R_s² (a measure of correlation between tagSNP haplotypes and individual SNPs). The median R_s² for the 72 HapMap SNPs with MAF >2% in the region 20 kb upstream and 10 kb downstream of ESR2 was 0.83; 59 (82%) of the HapMap SNPs were captured by the four tagSNPs with an R_s² > 0.6. These four htSNPs can also predict the common haplotypes in African Americans, Latinos, Japanese Americans, and native Hawaiians, with a median R_h² of 0.41, 0.23, 0.48, and 0.57, respectively (Supplementary Table S1).

The four htSNPs were genotyped in the main study; genotype frequencies among controls were in Hardy-Weinberg equilibrium (HWE) for every cohort (stratified by ethnicity; Table 2). The four htSNPs predicted five common haplotypes (>5%) in Whites (Table 3). Common haplotypes composed of the four htSNPs had a cumulative frequency of 93.5% in White controls (Table 3). Hap3 (CACT) and Hap4 (CACC) were rare (frequency<0.05) in Japanese Americans and native Hawaiians.

None of the htSNPs was associated with prostate cancer risk in either all subjects or Whites only (Table 4). None of the SNP-specific ORs showed significant different across ethnicities or across cohorts (Supplementary Table S2).

A global test for the influence of any common haplotype on prostate cancer risk was not significant (χ² = 7.12, 6 df, P = 0.31). In pooled analyses, men carrying two copies of variant Hap5 (TACC) had a 1.46-fold increased risk of prostate cancer (99% CI, 1.06-2.01) as compared with men carrying zero copies of this variant haplotype (Table 5). The risk of prostate cancer increased as the number of variant Hap5 increased, and the trend was significant (P = 0.01). The magnitude of prostate cancer risk was similar among Whites but not significant (2 versus 0 copies of Hap5: OR, 1.50; 99% CI, 0.95-2.37). The heterogeneity test for haplotype-specific ORs across ethnicities was also nonsignificant for all haplotypes (LRT χ² = 29.48, 20 df, P = 0.07). A similar result was observed for the heterogeneity test across cohorts (LRT χ² = 33.73, 33 df, P = 0.43). Cohort- and ethnicity-specific haplotype frequencies and results are detailed in Supplementary Tables S3 and S4.

To assess interaction effects, we calculated stratum-specific SNP and haplotype ORs for strata defined by family history (at least one first-degree relative diagnosed with prostate cancer versus none), age at time of diagnosis or time of diagnosis of the matched case for controls (≤65 years old versus >65 years old), and BMI (<25, 25 ≤ BMI < 30, ≥30; Supplementary Tables S5 and S6). The P value for test of interaction by BMI was significant for SNP ESR2_006 (rs1256049, LRT χ² = 11.51, df = 2, P = 0.003, Supplementary Table S6), but the 99% CI for the ORs included 1 for all BMI strata. Homozygote variant carriers of Hap5 were significantly associated with increased risk of advanced prostate cancer (stage >T3b, N₁, or M₁; OR, 2.16; 99% CI, 1.27-3.69), nonadvanced prostate cancer (stage <T3b, N₁, or M₁; OR, 1.47; 99% CI, 1.02-2.12), and high-grade prostate cancer (Gleason sum ≥8, OR, 2.32; 99% CI, 1.31-4.09) as compared with controls (Table 6). No significant heterogeneity was found in the association between Hap5 and advanced-stage (>T3b, N₁, or M₁) and high-grade (Gleason sum ≥8) prostate cancer.

Overall, we have strong evidence against moderate- to high-risk variants in ESR2 (dominant or multiplicative MAFs >5% and dominant or multiplicative effects of 1.2 or greater) being associated with prostate cancer in this study. Homozygosity for Hap5 (TACC) of ESR2 was nominally associated with prostate cancer with a nominal P value before correction of multiple comparisons in the pooled sample including all studies and ethnicities. A test for heterogeneity showed that this association was not different across ethnic groups or cohorts or across prostate cancer stage or grade. However, the global test for the influence of any common haplotype on prostate cancer risk is not significant, and the association of risk with Hap5 (TACC) does not survive correction for multiple testing. ESR2 common variants and haplotypes were not associated with the aggressiveness of prostate cancer (advanced stage or high grade).

The role of genetic polymorphisms of ESR2 in prostate carcinogenesis was explored in a recent study in Sweden of 1,415 incident cases and 801 controls (36). These authors found that carriage of variant rs2987983, a promoter SNP, was associated with a significantly increased risk of prostate cancer (OR, 1.33; 95% CI, 1.08-1.64). SNP rs2987983 was not genotyped in our population due to high genotyping failure rate after sequencing a reference panel of 38 Whites. However, this SNP is in complete LD with another promoter SNP rs3020450 (D′ = 1, R² = 1 in HapMap CEU samples) that was genotyped in our studies, and we did not observe any elevation in prostate cancer risk with this polymorphism. The NCI's Cancer Genetic Markers of Susceptibility (CGEMS) project³⁷

(37) also genotyped six SNPs (rs944045, rs1256062, rs1256044, rs1269056, rs3020450, and rs10137185) in ESR2 as part of a genome-wide association scan in more than 1,100 prostate cancer cases and 1,100 controls. Two of these SNPs (rs10137185 and rs1952586) showed evidence for association with P < 0.01 (P = 0.003 and P = 0.004, respectively). Both of these SNPs are moderately correlated with haplotypes of our four htSNPs (R_s² = 0.63 for each in the HapMap CEU samples). Because we did not find a strong association between these haplotypes and prostate cancer risk in a much larger sample (more than five times as many samples), it is likely that rs10137185 and rs1952586 are also not associated with prostate cancer risk, although it is possible that there is a true effect in the CGEMS cohort that is obscured in this larger data set.

Expression of ESR2 in prostate carcinoma cells triggers apoptosis (11) and, therefore, inhibits the proliferation of cancer cells induced by 17β-estradiol (estrogen) and 5α-dihydrotestosterone in androgen-responsive prostate cancer cells (LNCaP; ref. 38). Both ESR2 and androgen receptor (AR) are located and widely expressed in prostate epithelium (4). The reduction in ESR2 expression in androgen-deprived recurrent tumors or after castration reflects the androgen dependence of expression of ESR2 in human prostate cancer (39, 40). In summary, ESR2 is regulated by AR and interacts with ESR1 to regulate prostate carcinogenesis through the modulation of genes involved in cell proliferation and apoptosis.

This study includes participants from seven cohorts and is the largest investigation of the association between genetic variants in the steroid hormones pathway and the risk of prostate cancer. The strength of the haplotype-tagging approach we used is that within blocks of high LD and limited haplotype diversity, common SNPs tend to lie on a common haplotype (21). Therefore, the selection of four htSNPs effectively reduced the cost of genotyping and accurately represents the common genetic variation spanning ESR2. However, the small sample size of the non-White groups and less-comprehensive tagging are limitations, and larger studies will be needed to test these associations in non-White populations. In addition, the different grading and staging systems and cutpoints used for data collection in each cohort reduced the flexibility of statistical analyses and may sometimes have resulted in the loss of power for some analyses of effect modification. However, the large sample size partially counteracts this limitation.

The BPC3 project includes >50 genes involved in the steroid hormone pathway (12). When all data are available, it will be important to jointly analyze genes in the same biological pathway with respect to prostate cancer risk. In summary, the evidence does not seem to be a major effect of variation in ESR2 on the risk for prostate cancer.

The authors gratefully acknowledge the participants in the component cohort studies and the expert contributions of Hardeep Ranu, Craig Labadie, and Lisa Cardinale (Harvard University) and William Modi, Meridith Yeager, Robert Welch, Cynthia Glaser, and Laurie Burdett (NCI).