Estrogen Receptor-β Expression of Ovarian Tumors and Its Association with Ovarian Cancer Risk Factors

This article is featured in Highlights of This Issue, p. 2105

Extensive epidemiologic research has demonstrated that hormonal and reproductive exposures are associated with risk of ovarian cancer. However, the mechanisms by which these factors influence ovarian carcinogenesis and resulting tumor development remain poorly understood. Epithelial ovarian cancers are heterogeneous in their natural history, morphology, and gene and protein expression (1–3). These differences may be crucial to understand the etiology, prognosis, and effective treatment of ovarian cancer. It is notable that nearly all established hormonal and reproductive exposures demonstrated differential associations with ovarian cancer histologic subtypes (4); however, limited research is available on whether risk factor profiles also differ by hormonal receptor expression patterns in ovarian tumors.

While previous research has shown differential associations between ovarian cancer risk factors and estrogen receptor-α (ERα) and progesterone receptor (PR) status (5, 6), no previous studies have examined associations of ovarian cancer risk factors with estrogen receptor-β (ERβ) expression. Several studies suggest that ERβ acts as a tumor suppressor in ovarian tissue and ERβ expression may be lost with malignant transformation; however, the loss of ERβ may be dependent on cell localization (nuclear vs. cytoplasmic; refs. 7–10). De Stefano and colleagues noted that ERβ staining was more likely to be localized in the nucleus in normal ovarian tissue and conversely, was more likely to be localized in the cytoplasm in ovarian cancer cells (11). In addition, higher levels of ERβ protein expression have been associated with improved progression-free and overall survival, as well as decreased probability of lymph node metastasis in serous tumors, although the cellular localization of ERβ was not described in these studies (12, 13). Conversely, two recent studies reported less favorable prognosis for ovarian tumors expressing ERβ; however, ERβ localization differed in these two studies, with one study noting worse prognosis with higher ERβ nuclear staining and the other with higher ERβ cytoplasmic staining (14, 15). These conflicting results, potentially due to differences in ERβ cell localization (nuclear vs. cytoplasmic), as well as the limited data on associations between ovarian cancer risk factors and hormone receptor subtypes, suggest that additional research on ERβ expression, both nuclear and cytoplasmic, in ovarian tumors is needed.

Therefore, in exploratory analyses, we examined associations between known risk factors for ovarian cancer and risk of ovarian cancer by expression of nuclear and cytoplasmic ERβ in paraffin-embedded tumor tissue from ovarian cancers diagnosed among women in the Nurses' Health Studies (NHS and NHSII).

In 1976, 121,700 U.S. female nurses, ages 30–55, were enrolled into the NHS after completing a baseline questionnaire collecting data on disease diagnoses and exposures. The NHSII began in 1989 when 116,429 U.S. female nurses, ages 25–42, completed baseline questionnaires. Information on exposures and disease diagnoses has been updated through mailed biennial questionnaires in NHS and NHSII, with a follow-up rate of 85%–90% for both cohorts. We identified incident cases of ovarian cancer through self-reports on the biennial questionnaires, linkage to the National Death Index (16), or report from family members. The study protocol was approved by the Institutional Review Boards of the Brigham and Women's Hospital (Boston, MA) and Harvard T.H. Chan School of Public Health (Boston, MA), and those of participating registries as required.

Cases diagnosed from 1976 to 2006 for NHS and 1989 to 2005 for NHSII, and for whom we obtained a pathology report of the primary surgery, were included in this study. Case pathology reports were reviewed by a gynecologic pathologist, blinded to exposure status, to confirm the epithelial ovarian cancer diagnosis and classify the tumors by histologic type (serous/poorly differentiated, mucinous, endometrioid, clear cell, other), behavior (invasive and borderline), and stage (I/II and III/IV). For confirmed cases, we requested representative paraffin-embedded tissue blocks of the ovarian tumors. Through 2006, there were 1,083 confirmed NHS cases, of which 217 were included on tissue microarrays (TMA). In the NHSII, 46 of the 201 confirmed cases through 2005 were included on TMAs. Details of case exclusions have been described previously (5). Briefly, cases were excluded because of incomplete information to request tumor tissue (NHS, n = 290 and NHSII, n = 77), inability to obtain tumor blocks (NHS, n = 479 and NHSII, n = 65), or no primary tumor blocks available (NHS, n = 97 and NHSII, n = 13). For these analyses, we additionally excluded cases on the TMAs with no tumor in the core slice (NHS, n = 2), noninterpretable stains (NHS, n = 13 and NHSII, n = 2), or who withdrew consent (NHS, n = 1). Separately within each cohort, up to four controls (n = 1,050) were matched to each case on year of birth. The selected controls did not have a prior bilateral oophorectomy, menopause due to irradiation, or prior diagnosis of cancer (except nonmelanoma skin cancer), and were alive at the time of case diagnosis. Matched controls of excluded cases were included in the analyses.

Details of the TMA construction and IHC process have been described previously (6). Briefly, the primary tumor block for each case was selected by the study pathologist (J.L. Hecht), who circled the tumor on the slide, confirmed histology and behavior, assessed grade (1–3), and sent the slide/block to the Dana-Farber/Harvard Cancer Center Specialized Histopathology Services Core for TMA construction. Three cores per case were transferred to a recipient paraffin-embedded block and sections were cut to create array slides. Using IHC, the slides were processed and stained within 2 weeks of creation using the primary antibody ERβ (anti–estrogen receptor beta 1, clone: PPG5/10, Thermo Fisher Scientific Pierce). For nuclear ERβ (ERβ-nuc), the pathologist scored the number of reactive versus total cells (0%, 1%–10%, 11%–50%, 51%–90%, and >90%) for each of the three cores per case independently. For cytoplasmic ERβ (ERβ-cyto), the pathologist scored the cores as having any cytoplasmic versus no cytoplasmic ERβ staining. ERα (primary antibody: rabbit monoclonal; clone SP1; Neomarkers; dilution 1:40) and PR (primary antibody: mouse monoclonal; clone PgR 636; Dako; dilution 1:150) staining were conducted similarly to ERβ-nuc with the pathologist scoring the number of reactive versus total cells (0%, 1%–10%, 11%–50%, 51%–90%, and >90%) for each of the three cores per case independently. If tissue was missing from the slide or only a few cell clusters (<20 cells) were present, then the core was designated as noninterpretable.

For each case and their matched controls, exposure and covariate information were obtained from the biennial questionnaire cycle prior to case diagnosis. We assessed well-established ovarian cancer risk factors including age (years), oral contraceptive (OC) use (ever vs. never; never, >0–<1 year, 1–5 years, and >5 years; continuous), tubal ligation (ever vs. never), parity (parous vs. nulliparous; nulliparous, 1–2 children, and 3+ children; continuous), menopausal status (premenopausal and postmenopausal), hormone therapy (HT; ever vs. never), and family history of breast/ovarian cancer (yes vs. no). HT use was further divided into estrogen-only HT use (ever vs. never) and estrogen plus progesterone HT use (ever vs. never), for which these categories were not mutually exclusive. In analyses on parity, pregnancies lasting 6 months or longer were included. Ovulatory years were calculated as age at natural menopause (current age for premenopausal women) minus age at menarche, 1 year for each pregnancy, and OC duration, and was categorized into quartiles. In addition, we assessed age at menarche (<13 years vs. ≥13 years), age at natural menopause (<50 years vs. ≥50 years), height (continuous), body mass index (BMI; <25, 25–<30, and ≥30 kg/m²; continuous), and breastfeeding among parous women (ever vs. never; 0–<1 month, 1–6 months, 7–11 months, and ≥12 months; continuous). Age at natural menopause excluded women reporting a hysterectomy before menopause.

We pooled the NHS (201 cases and 866 controls) and NHSII (44 cases and 184 controls) data for all analyses. ERβ-nuc and ERβ-cyto were considered positive (+) if ≥1% of cells stained positive in any core and negative (–) if 0% of cells stained positive. Differential expression of ERβ-nuc and ERβ-cyto staining was assessed using a χ² test. In addition, the χ² test was used to assess the distribution of ERβ-nuc and ERβ-cyto staining positivity by histologic subtype, morphology, stage, and ERα, PR, and joint ERα/PR staining positivity. χ² test for trend was used to assess the distribution of ERβ-nuc and ERβ-cyto staining positivity by grade. Spearman correlation coefficients were calculated between the maximum ERβ-nuc, ERα, and PR scores, which were measured as the percentage of cells staining positive (0%, 1%–10%, 11%–50%, 51%–90%, and >90%).

We used polytomous logistic regression (PLR) to assess the relationship between ovarian cancer risk factors and ovarian cancer risk by ERβ-nuc and ERβ-cyto staining positivity using a three category outcome [ERβ-nuc (+), ERβ-nuc (–), and controls, or ERβ-cyto (+), ERβ-cyto (–), and controls]. Women with missing exposure information were excluded for the specific analysis with missing data. For each exposure, we used a likelihood ratio (LR) test to compare an unconstrained model (i.e., the association between the exposure and ovarian cancer risk by hormone receptor expression was allowed to vary) with a constrained model (i.e., the association was not allowed to vary) to assess the heterogeneity in the ORs by ERβ-nuc and ERβ-cyto expression versus controls (17). Analyses were adjusted for age at diagnosis (years) and cohort (NHS/NHSII), with estimates allowed to vary by receptor status, as well as family history of breast and/or ovarian cancer (yes/no), OC use (continuous in months), number of pregnancies, and menopausal/HT status (premenopausal/postmenopausal + no HT use/postmenopausal + HT use/missing), for which estimates were constrained. In models of estrogen-only HT use, we further adjusted for estrogen plus progesterone HT use (ever/never) and other HT use (any HT use that did not include estrogen-only or estrogen plus progesterone; ever/never), for which estimates were constrained. In models of estrogen plus progesterone HT use, we further adjusted for estrogen-only HT use (ever/never) and other HT use (ever/never), for which estimates were constrained. Women missing OC duration (n = 25) or number of pregnancies (n = 19) were set to the cohort-specific median. In secondary analyses, we assessed joint ERβ-nuc and ERβ-cyto staining positivity using PLR with a five category outcome [ERβ-nuc (+)/ERβ-cyto (+), ERβ-nuc (+)/ERβ-cyto (–), ERβ-nuc (–)/ERβ-cyto (+), ERβ-nuc (–)/ERβ-cyto (–), and controls] adjusting for the same covariates as above.

In sensitivity analyses, we used case–case analyses to explore the influence of histology on the differential risk factor associations by ERβ-cyto status, as ERβ-cyto staining differed by histology. “Cases” were defined as receptor-positive tumors, whereas “controls” were defined as receptor-negative tumors. We compared an unconditional logistic regression model without adjustment for histology with a model with adjustment for histology using an LR test, with the exception of breastfeeding and OC use among ever OC users due to the small sample size of nonserous tumors. We adjusted for the same covariates as above in addition to histology (serous, mucinous, endometrioid, clear cell, and other). In addition, as ERβ-cyto staining differed by ERα expression, we assessed joint ERα and ERβ-cyto staining; because of the small sample, we excluded women with ERα (–) and ERβ-cyto (+) staining (n = 15).

We performed analyses using SAS version 9.4 (SAS Institute Inc.) and Stata version 15.1 (Stata Corp LP). All P values were two-sided and considered statistically significant if less than 0.05.

Among the 245 ovarian cancer cases included in the analyses, 43% stained positive for ERβ-cyto receptors and 71% stained positive for ERβ-nuc receptors (Table 1). The distribution of histology differed between ERβ-cyto (+) and ERβ-cyto (–) tumors, with ERβ-cyto (+) tumors more likely to be serous (75% vs. 59%) and less likely to be mucinous (1% vs. 8%) or clear cell (5% vs. 11%; P = 0.008). In addition, ERβ-cyto (+) tumors were more likely to be invasive (P = 0.02) and higher stage (P = 0.09) compared with ERβ-cyto (–) tumors. ERβ-nuc (–) tumors were more likely to be invasive compared with ERβ-nuc (+) tumors (P = 0.01). No other associations with tumor characteristics were observed for ERβ-nuc staining (P ≥ 0.60).

Tumors categorized as ERβ-nuc (–) were more likely to be categorized as ERβ-cyto (+) compared with ERβ-cyto (–), with 63% of ERβ-nuc (–) tumors staining positive for ERβ-cyto and 65% of ERβ-nuc (+) tumors staining negative for ERβ-cyto (P < 0.0001; Table 1). ERβ-cyto (+) tumors were more likely to stain positive for ERα compared with ERβ-cyto (–) tumors (86% vs. 68% stained positive for ERα, respectively; P = 0.001). ERβ-cyto staining was also significantly associated with PR (P = 0.02) and joint ERα/PR staining status (P = 0.007). ERβ-nuc staining was not associated with ERα or PR staining in the dichotomous classification (P > 0.15) or in correlations between the percentage of cells staining positive (Spearman correlation for ERβ-nuc and ERα = 0.03, P = 0.69; Spearman correlation for ERβ-nuc and PR = −0.05, P = 0.47).

An inverse association was observed between parity and risk of ERβ-cyto (–) tumors [OR, 0.42; 95% confidence interval (CI), 0.23–0.78 for parous vs. nulliparous), but no association was observed for ERβ-cyto (+) tumors (OR, 1.08; 95% CI, 0.45–2.63 for parous vs. nulliparous; P_{heterogeneity} = 0.05; Table 2). Interestingly, among parous women, there was a 16% (OR, 0.84; 95% CI, 0.71–0.99) decreased risk of ERβ-cyto (+) tumors per child, but no association with ERβ-cyto (–) tumors (OR, 1.01; 95% CI, 0.89–1.14 per child; P_{heterogeneity} = 0.07). In addition, there was a suggestion of a reduced risk of ERβ-cyto (–) tumors with increasing BMI (OR, 0.96; 95% CI, 0.92–1.00 per kg/m²), but not for ERβ-cyto (+) tumors (OR, 1.00; 95% CI, 0.96–1.04; P_{heterogeneity} = 0.09). The other ovarian cancer risk factors examined, including OC use, tubal ligation, and menopausal status, did not appear to differ by ERβ-cyto staining positivity (P_{heterogeneity} > 0.10).

The ERβ-cyto associations above were essentially unchanged after adjusting for histology in case–case analyses (Supplementary Table S1). Compared with nulliparous cases, parous cases were twice as likely to develop ERβ-cyto (+) tumors as opposed to ERβ-cyto (–) tumors without adjustment for histology (OR, 2.52; 95% CI, 0.88–7.25). With adjustment for histology, the risk estimate decreased slightly to 2.42 (95% CI, 0.82–7.12). In analyses of joint ERα/ERβ-cyto staining among all women, there were decreased risks per child for ERα (–)/ERβ-cyto (–) tumors (OR, 0.77; 95% CI, 0.62–0.97 per child) and ERα (+)/ERβ-cyto (+) tumors (OR, 0.87; 95% CI, 0.75–1.01 per child), but no association was observed for ERα (+)/ERβ-cyto (–) tumors (OR, 1.01; 95% CI, 0.88–1.16 per child; P_{heterogeneity} = 0.08; Supplementary Table S2). Similar associations between number of children and ERα/ERβ-cyto status of the tumor were observed among parous women (P_{heterogeneity} = 0.03). The association between decreased risk of ERβ-cyto (–) tumors and increasing BMI was fairly similar between ERα (+)/ERβ-cyto (–) tumors (OR, 0.95; 95% CI, 0.91–1.00) and ERα (–)/ERβ-cyto (–) tumors (OR, 0.97; 95% CI, 0.91–1.03), but no association was observed for ERα (+)/ERβ-cyto (+) tumors (OR, 1.01; 95% CI, 0.97–1.05; P_{heterogeneity} = 0.18).

A decreased risk of ERβ-nuc (+) tumors was observed among parous women compared with nulliparous women (OR, 0.46; 95% CI, 0.26–0.81), whereas no association was observed for ERβ-nuc (–) tumors (OR, 1.51; 95% CI, 0.45–5.04; P_{heterogeneity} = 0.04; Table 3). The other ovarian cancer risk factors examined, including OC use, breastfeeding, BMI, or ovulatory years, did not appear to differ by ERβ-nuc staining positivity (P_{heterogeneity} ≥ 0.09).

The decreased risk for parity with ERβ-cyto (–) tumors appeared to be slightly stronger for ERβ-cyto (–)/ERβ-nuc (+) tumors (OR, 0.38; 95% CI, 0.20–0.74 for parous vs. nulliparous) compared with ERβ-cyto (–)/ERβ-nuc (–) tumors (OR, 0.69; 95% CI, 0.15–3.10 for parous vs. nulliparous; P_{heterogeneity} = 0.08; Supplementary Table S3). There was a suggestion of heterogeneity across hormone receptor status for duration of OC use, with a stronger decreased risk of ovarian cancer for ERβ-cyto (–)/ERβ-nuc (–) tumors (OR, 0.85; 95% CI, 0.70–1.02 per year) compared with the other tumor subtypes (ORs ranged from 0.98 to 1.02; P_{heterogeneity} = 0.06) and a similar association was observed among ever OC users.

This is the first study to assess the differential associations of ovarian cancer risk factors by ERβ-cyto and ERβ-nuc status of the ovarian tumors. ERβ-nuc was inversely related to ERβ-cyto. Notably, ERβ-cyto expression differed by tumor characteristics, particularly histology and morphology; however, there appeared to be a similar level of ERβ-nuc expression by tumor characteristics with the exception of morphology. Parity was associated with an inverse association for ERβ-cyto (–) as well as ERβ-nuc (+) ovarian tumors. Although the P value for heterogeneity was not significant, there was a stronger inverse association of parity with ERβ-cyto (–)/ERβ-nuc (+) tumors compared with ERβ-cyto (+)/ERβ-nuc (+) and ERβ-cyto (–)/ERβ-nuc (–) tumors. In addition, there was a suggestion of a stronger inverse association between ERβ-cyto (–) versus ERβ-cyto (+) tumors and BMI. Other risk factors, including HT use, did not demonstrate differential associations by ERβ staining.

We observed that 71% of ovarian tumors included in this study expressed nuclear staining and 43% expressed cytoplasmic staining for ERβ. Previous reports of ERβ staining have varied (53%–93% staining positive), with the majority of studies reporting that approximately 60%–80% of ovarian tumors stained positive for ERβ (10, 11, 13–15, 18), which is in-line with our observed expression of ERβ-nuc staining. Differences in ERβ staining between our study and previous studies may be due to the localization of ERβ staining (nuclear vs. cytoplasmic), differences in antibodies used to detect ERβ, and differences in techniques to determine ERβ levels (mRNA vs. IHC). Interestingly, Chan and colleagues observed in a small pilot study that mRNA levels of ERβ did not correlate with IHC staining for nuclear ERβ (18). Similar to our study, the majority of previous studies observed no significant differences of ERβ-nuc staining with histology or grade (10, 11, 13, 19). Contrary to two previous studies, we observed significant differences of ERβ-cyto staining with histology (11, 19). Interestingly, previous studies have noted a shift in ERβ staining with malignant transformation of ovarian cells (9, 11, 18). Lower ERβ-nuc staining was observed in ovarian cancer cells compared with normal ovarian cells, but higher ERβ-cyto staining was observed in ovarian cancer cells compared with normal ovarian cells. This is consistent with our finding of a strong inverse association of staining within these different cellular compartments, suggesting that malignant transformation may lead to a reduction in localization of ERβ in the nucleus and supporting examination of the ratio of ERβ-nuc to ERβ-cyto in future epidemiologic studies.

Our most notable finding was that parity was only inversely related to risk of ovarian cancer for ERβ-nuc (+) and ERβ-cyto (–) tumors. The latter finding is consistent with our prior work demonstrating that increasing number of pregnancies was only inversely associated with ERα (–) tumors, but not ERα (+); ERα was strongly positively related to ERβ-cyto in our data. Interestingly, in a mouse model, parous versus nulliparous mice had reduced nuclear ERα expression in mammary tissue, but increased ERβ-nuc expression; ERβ-cyto expression did not vary after pregnancy (20). In addition, low nuclear ERα and high ERβ-nuc expression in parous mice was associated with lower Ki67, a marker of proliferation (20, 21). Furthermore, in rats, parity was associated with reduced responsiveness to estradiol as demonstrated by lower uterine weight (22). Similar studies of the effect of parity on ERα and ERβ expression and localization in ovarian and fallopian tube tissue would provide potential insight into how parity may lower risk of ovarian cancer.

In addition, increasing BMI was associated with a decreased risk of ERβ-cyto (–) tumors, but was not associated with ERβ-cyto (+) tumors. Interestingly, genetic variation in ERβ is associated with risk of type 2 diabetes (23). Furthermore, in skin punch biopsies from postmenopausal women, higher adiposity was associated with lower expression of ERβ (24). However, these results should be interpreted with caution as most studies suggest that BMI is associated with a modest increased risk of ovarian cancer (25, 26). The other ovarian cancer risk factors did not appear to differ by ERβ staining.

As this is the first study that has assessed ovarian cancer risk factors by ERβ-nuc and -cyto status and due to our small sample size, future consortium studies with more ovarian tumor samples will be needed to confirm our findings. In addition, as ovarian tumors were only available for a subset of the ovarian cancer cases in NHS and NHSII, our study could have selection bias. To assess for selection bias, we compared the distribution of reproductive and hormonal factors and tumor characteristics between confirmed ovarian cancer cases, cases eligible for tumor block collection, cases with collected tumor blocks, and cases included on the TMA (5). Cases on the TMA were similar to all confirmed ovarian cancer cases with the exception that TMA cases were older at diagnosis, had lower stage tumors, and were slightly more likely to be parous (NHS II only), suggesting that the potential for selection bias within our study is low. As with all epidemiologic studies, there is the chance that our results may be due to residual confounding; however, we adjusted for known ovarian cancer risk factors. Our study also had important strengths. It is the first study to assess lifestyle and reproductive factors with the risk of ovarian cancer by ERβ staining, and our sample of 245 tumor blocks is large compared with other similar studies. In addition, exposure and covariate information were collected prospectively, eliminating the possibility of differential recall in cases versus controls. Finally, the use of TMAs limited variability in assay results as many cases could be stained at the same time.

Our results suggest that parity may be differentially associated with ovarian tumors based on both ERβ-cyto and ERβ-nuc expression. Specifically, parity may influence ovarian cancer risk, in part, through alterations in ERβ expression and localization within tumor cells. These results should be confirmed in future studies with a larger sample size. In addition, future studies should assess the biologic mechanisms that may underlie the observed associations.

A.L. Shafrir reports grants from the NIH and the Marriott Family Foundations outside the submitted work. M.S. Rice reports employment with Sanofi (all work on this article was completed prior to employment at Sanofi). J.L. Hecht reports grants from the NIH/NCI (2P01CA87969, principal investigator: M. Stampfer) during the conduct of the study. S.S. Tworoger reports grants from the NCI during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. The authors assume full responsibility for analyses and interpretation of these data.

A.L. Shafrir: Formal analysis, methodology, writing–original draft. A. Babic: Formal analysis, methodology, writing–review and editing. M. Gates Kuliszewski: Writing–original draft, writing–review and editing. M.S. Rice: Data curation, writing–review and editing. M.K. Townsend: Writing–review and editing. J.L. Hecht: Conceptualization, data curation, visualization, writing–review and editing. S.S. Tworoger: Conceptualization, data curation, supervision, methodology, writing–review and editing.

The authors acknowledge the Channing Division of Network Medicine, Department of Medicine, Brigham and Women's Hospital, as the home of the Nurses' Health Study. The authors thank the participants and staff of the Nurses' Health Study and Nurses' Health Study II for their valuable contributions as well as the following state cancer registries for their help: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, and WY. This project was supported by the NIH (P01 CA87969, to M. Stampfer and S.S Tworoger; UM1 CA186107, to M. Stampfer; U01 CA176726, to W.C. Willet; and R01CA54419 and P50CA105009, to D.W. Cramer) and the U.S. Department of Defense (W81XWH-10-1-02802, to K.L. Terry).

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.