Nuclear and Cytoplasmic Expression of ERβ1, ERβ2, and ERβ5 Identifies Distinct Prognostic Outcome for Breast Cancer Patients

Estrogen receptor (ER)-α remains the most important marker of response to hormonal therapy in breast cancer. The role of ERβ is much less clear with many conflicting studies published to date (reviewed in ref. 1). Although structurally related and homologous at the DNA and ligand binding domains, both receptors are genetically distinct (2, 3). Furthermore, ERβ is abundant in normal mammary tissue (4) with loss or diminished expression implicated in mammary carcinogenesis (5–7).

ERβ exists as five full-length versions, ERβ1 to ERβ5 (8, 9), each formed by alternative splicing of the last coding exon. ERβ1 is the only fully functional isoform (10). ERβ2 is identical to ERβcx (11). ERβ3 seems to be testis specific (8) and ERβ4 is not found in breast tissue (9). Although ERβ2 and ERβ5 cannot bind ligand, transient transfection studies suggest that they may antagonize ERα function through heterodimerization (11–13). Functional studies have shown that overexpression of ERβ1 has antiproliferative and proapoptotic effects (14, 15). ERβ2 and ERβ5 have no intrinsic activities with conflicting data on whether they can (8) or cannot homodimerize (10). However, they can heterodimerize with ERβ1 and enhance its transcriptional activity in a ligand-dependent fashion (10). If expressed in breast tumors, these could have important clinical implications in influencing response to hormone therapy.

Data from all previous immunohistochemical studies on ERβ in breast cancers have to be interpreted with caution. First, the number of cases included in many studies was insufficient to infer generalization of results to the general breast cancer population. Second, in many studies, antibodies used did not discriminate between different ERβ isoforms. Third, cutoff values for defining ERβ positivity are often inconsistent, rendering comparison of results difficult.

To determine if ERβ has any clinical utility in breast cancer, the aim of this study was to elucidate the prognostic significance of ERβ1, ERβ2, and ERβ5 by immunohistochemistry in a large well-characterized cohort of breast carcinomas with long-term clinical follow-up, and to additionally evaluate if nuclear or cytoplasmic staining (or a combination of both) can predict outcome.

Ethical approval was obtained from the Leeds (East) Local Research Ethics Committee at St. James's University Hospital (Leeds, United Kingdom) and the Nottingham Research Ethics Committee 2.

Patients. A series of 880 cases of primary operable invasive breast carcinoma from consecutive patients presenting from 1986 to 1998 and entered into the Nottingham Tenovus Primary Breast Carcinoma Series were used. Clinical history and tumor characteristics (age, tumor type, size, histologic grade, lymph node status, and Nottingham Prognostic Index; ref. 16) were available for 842 cases. Survival data including survival time and disease-free interval were maintained on a prospective basis. Disease-free survival (DFS) was defined as the interval (months) from primary surgical treatment to the first locoregional or distant recurrence. Overall survival (OS) was taken as time (months) from primary surgical treatment to time of death from breast cancer. The series has previously been used to study a wide range of breast cancer–related proteins (17); these data were used in the current study. An independent set of 322 cases were identified from the archives of the Leeds Teaching Hospitals NHS Trust to use as a validation set.

Mouse monoclonal antibodies (clones PPG5/10 and 57/3, Serotec) were raised against the unique COOH-terminal peptides of ERβ1 and ERβ2, respectively (18), and a monoclonal antibody was generated in-house against a COOH-terminal peptide of ERβ5 (19). Antibodies were validated by peptide absorption studies and Western blotting. Peptides specific for ERβ1 (CSPAEDSKSKEGSQNPQS), ERβ2 (CMKMETLLPEATMEQ), and ERβ5 (LLSHVRHARYAP) were used in neutralization studies. Each antibody was diluted to its working concentration and incubated with 10× excess of relevant peptide overnight at 4°C, before application to tissue.

For Western blots, proteins were extracted from two ERβ-positive frozen breast carcinomas and from SkBr3 cells (ERβ negative; ref. 6) in 100 mmol/L Tris-HCl (pH 7.5), 300 mmol/L NaCl, 4 mmol/L EDTA, 2% NP40, 0.5% Na deoxycholate, 1 mmol/L sodium orthovanadate, and protease inhibitor cocktail (Roche). Protein was quantified (Bradford reagent, Pierce); Laemlli buffer was added to 10-μg protein; and samples were boiled (5 min). Proteins were resolved (SDS 4-15% gradient polyacrylamide gels; Bio-Rad Laboratories) and transferred onto Hybond-ECL membrane (Amersham). Membranes were blocked and stained with antibodies in PBS, 5% milk, 0.1% Tween 20 and washed in PBS, 0.1% Tween 20. Blots were incubated at 4°C for >15 h with anti-ERβ1 (1:500), anti-ERβ2 (1:25), anti-ERβ5 (1:100), and mouse anti–β-actin (clone AC-15; 1:10,000; Sigma-Aldrich). Following further washing, blots were incubated with horseradish peroxidase–conjugated goat anti-mouse IgG (1:1,000; Santa Cruz Biotechnology). Horseradish peroxidase was visualized with SuperSignal West Pico ECL reagent (Pierce) on Hyperfilm-ECL (Amersham).

Tissue microarrays were prepared as previously described using 0.6-mm cores selected from the most representative tumor areas (17). Immunohistochemical analysis of 4-μm tissue microarray sections was carried out using standard techniques as previously described (20). ERβ1, ERβ2, and ERβ5 were used at 1:5, 1:10, and 1:75, respectively, overnight at 4°C. A negative control (no primary antibody) and three positive controls (breast carcinoma of varying staining intensities) were included in each immunohistochemistry run.

Tissue microarrays were digitized (Aperio Technologies) and nuclear immunoreactivity was scored as percentage of positive cells in relation to total number of tumor cells present and using the Allred score based on nuclear staining intensity and proportion of positively stained nuclei, which generates numerical values from 0 to 8 (21). Expression status was dichotomized using 20% or Allred Score >3 as cutoffs as previously determined (7, 20). Cytoplasmic staining was determined, where 0, no staining; 1, weak; 2, moderate; 3, strong. Cases were scored independently by three authors (S.K., Y.A., and A.M.S., a specialized consultant breast histopathologist). Discordant results were reevaluated jointly to reach consensus.

Pearson χ² square and Spearman correlation were used for statistical analysis using SPSS 12.0.1. Associations with DFS and OS were analyzed initially by Kaplan-Meier plots and log-rank test. The Cox proportional hazards model was used to test the statistical independence and significance of predictors on relapse-free interval and OS adjusted for other prognostic indicators [tumor size (>1.5 and <1.5 cm), grade (1-3), lymph node status, and ERα status]. P values were two-sided, with P < 0.05 considered as significant. Standard cutoffs were used for established prognostic factors according to a previously published series (17).

Details of the breast tumors represented in the tissue microarray are summarized in Supplementary Table S1. From the original 880 cases, immunohistochemical data were available on a maximum of 757 patients due to loss of some tissue microarray cores during antigen retrieval and staining, a well-recognized phenomenon associated with tissue microarray immunohistochemistry. Of these 757 samples, 733 had information on whether or not they received hormone therapy; 250 received hormone therapy (241 tamoxifen, 8 tamoxifen and Zoladex, and 1 patient an undefined regimen), with 65 experiencing recurrence.

Antibody validation. Specific nuclear and cytoplasmic staining was completely abolished when each primary antibody was preabsorbed with its respective blocking peptide (Fig. 1A-F). By Western blot analysis, antibodies against ERβ1, ERβ2, and ERβ5 recognized single specific proteins in two independent ERβ-positive tumor samples but not in SkBr3 cells (Fig. 1G), which are ERβ negative in our hands (6). Antibodies did not cross-react with each other or with ERα (data not shown), confirming our previous data (18).

Expression patterns. All ERβ isoforms investigated were expressed in epithelial and stromal cell nuclei (Fig. 2A-C). In some cases, this was also associated with cytoplasmic expression; in some instances, cytoplasmic expression of ERβ1 (data not shown) and, in particular, ERβ2 was present in the absence of nuclear immunoreactivity (Fig. 2D). ERβ1 and ERβ2 were also expressed in stromal cells (Fig. 2E and F, respectively).

Expression of ERβ isoforms in relation to patient and tumor characteristics. Using percentage positive tumor cells, nuclear ERβ1, ERβ2, and ERβ5 ranged from 0% to 95% (median of 70%, 85%, and 90%, respectively). Using Allred scoring, nuclear expression of ERβ isoforms ranged from 0 to 8 (median of 6, 8, and 8 for ERβ1, ERβ2, and ERβ5, respectively).

When data were dichotomized, positive nuclear ERβ2 expression was observed in 590 of 713 (83%) of breast tumors, whereas nuclear expression of ERβ1 and ERβ5 was seen in the majority of tumors (731 of 740 and 754 of 757, respectively) using a 20% cutoff. A total of 521 cases were positive for all ERβ isoforms. Similar values were obtained when using Allred scoring (data not shown). We observed a significant positive correlation between Allred and percentage staining for ERβ1, ERβ2, and ERβ5 (P < 0.001, Pearson's regression; correlation coefficients, 0.62, 0.89, and 0.61, respectively), highlighting the robustness of our data.

Associations between ERβ isoform expression and a range of standard clinicopathologic parameters were tested, but only those showing statistical significance are shown in Table 1. ERβ2 correlated positively with grade 2 tumors and moderate Nottingham Prognostic Index and inversely with distant metastasis. ERβ2 was also positively associated with ERα, progesterone receptor, androgen receptor, and BRCA1. Moreover, there was generally good concordance between Allred scoring and percentage positive nuclei. Interestingly, ERβ2 was expressed in a large proportion (124 of 180, 68%) of tumors that were ERα and progesterone receptor negative. ERβ1 did not associate with the above or other known clinicopathologic indicators, whereas ERβ5 was only significantly associated with moderate Nottingham Prognostic Index (data not shown).

Nuclear ERβexpression and patient outcome. Kaplan-Meier survival analysis was used to determine survival with respect to nuclear expression of ERβ isoforms. Any amount of nuclear ERβ2 immunoreactivity was significantly associated with better OS, irrespective if this was determined using the Allred score (Fig. 3A) or as 1% or 20% positive nuclei (Fig. 3B and C, respectively). Nuclear ERβ5 staining was also associated with improved survival, but only when 65% positive nuclei were used as a cutoff (Fig. 3D). Nuclear ERβ1 was not significantly associated with improved OS using Allred (Fig. 3E) or 20% positive cells (data not shown). Cases coexpressing ERα and ERβ2 had significantly improved OS (P < 0.001) and DFS (data not shown; Fig. 3F).

Cytoplasmic ERβ expression and patient outcome. Kaplan-Meier analysis showed that cytoplasmic ERβ2 expression was significantly associated with poor OS (Fig. 4A), whereas cytoplasmic ERβ1 immunoreactivity was not (data not shown). Cytoplasmic ERβ5 immunoreactivity was infrequent across the entire cohort. Interestingly, whereas nuclear ERβ2 immunoreactivity was associated with good OS, combined nuclear and cytoplasmic immunoreactivity or no ERβ2 immunoreactivity had significantly reduced survival, and cases showing only cytoplasmic staining had the worst prognosis of all (Fig. 4B). Cytoplasmic ERβ2 expression additionally correlated with distant metastasis (P < 0.001), recurrence (P = 0.003), poor DFS (P = 0.001), and death from breast cancer (P < 0.001). In an independent set of 322 cases, cytoplasmic ERβ2 was significantly associated with high-grade tumors (Supplementary Table S3).

In a multivariate Cox hazard analysis, nuclear, but not cytoplasmic, ERβ2 was a significant predictor of better OS independent of tumor grade, lymph node status, size, or ERα status (Supplementary Table S2). Additionally, expression of nuclear (20% positive nuclei and Allred score >6), but not cytoplasmic, ERβ2 was strongly predictive of response to endocrine therapy by 2-fold (Table 2).

This is the first study aimed at elucidating the prognostic role of ERβ1, ERβ2, and ERβ5 in a large well-validated cohort of breast carcinomas with comprehensive follow-up data, using specific well-validated antibodies. Our results illustrate the importance of evaluating cytoplasmic as well as nuclear ERβ immunoreactivity as these have important and distinct prognostic implications.

Our data show that nuclear ERβ2 is a powerful predictor not only of DFS and OS in breast cancer but also of response to hormone therapy. We believe ERβ2 to be a robust marker because those associations were, in general, observed when its presence was evaluated by two different widely accepted scoring criteria, Allred and percentage positive tumor nuclei, in more than 700 tumors. Published studies on the prognostic significance of nuclear ERβ2 have been mixed. In accord with our data, ERβ2 mRNA expression in 105 patients treated with adjuvant endocrine therapy showed that ERβ2 was associated with significantly improved DFS (22). However, other studies showed no correlation with ERβ2 mRNA or protein expression and tamoxifen response (23, 24). A study of 18 patients treated with neoadjuvant tamoxifen showed that ERβ2 expression correlated with poor clinical response (25), whereas a combined immunohistochemistry/Western blotting study was contradictory, with ERβ2 expression correlated with favorable response to endocrine therapy and improved survival (26). Another study of 50 breast tumors, including 34 tamoxifen sensitive and 16 cases of relapse, failed to show any difference in ERβ2 expression, leading the authors to conclude that ERβ2 was not predictive of tamoxifen relapse (27). The most likely explanation for these contradictory results is the small numbers of cases studied, ranging from 18 to a little more than 100. Moreover, they used different antibodies that we have previously shown to be not directly comparable (20) and the continued comparison of mRNA versus protein (28) further complicates matters. Microarray data revealed a cluster of 20 breast tumors with reduced ERβ2 expression (29, 30). These cases were also associated with node positivity and the risk of distant metastasis (29, 30) and, in agreement with our data, highlight that loss of expression of nuclear ERβ2 is associated with a more aggressive phenotype (i.e., greater incidence of lymph node positivity, higher likelihood of vascular invasion, increased risk of developing metastatic disease, reduced survival, and poor response to endocrine therapy). Since this work was submitted, two related studies have been published. Vinayagam et al. (28) examined 141 primary breast cancers, which had been treated exclusively with adjuvant endocrine therapy. In keeping with our study, nuclear ERβ2 immunoreactivity was associated with better outcome. This was also reported in a cohort of 150 breast tumors (31). These studies further highlight the potential predictive power of nuclear ERβ2.

Whereas our data showed a strong predictive power for nuclear ERβ2, this was less evident for ERβ1, where only a nonsignificant trend was observed, and ERβ5, where Kaplan-Meier analysis showed expression in ≥65% cell nuclei was associated with better OS. Moreover, unlike ERβ2, the predictive power of these biomarkers was only seen using a single method of scoring with no concordance between them. Furthermore, neither biomarker showed any association with accepted prognostic makers and only ERβ5 associated with moderate Nottingham Prognostic Index. Earlier studies indicated that ERβ1 protein expression decreased progressively from normal breast through preinvasive to invasive lesions, although most breast carcinomas expressed at least some ERβ protein (5–7). Previously, ERβ1 has been shown to be predictive of tamoxifen response (27). In our study, nuclear ERβ1 expression was not significantly associated with good overall prognosis, nor did we observe any association with tamoxifen sensitivity. Studies of total breast mRNA have shown that ERβ5 is the most abundantly expressed isoform in both ERα-positive and ERα-negative cases (32), but the potential contribution of nonepithelial ERβ5 expression to these measurements is unknown: We have shown herein that ERβ5 is additionally expressed in stromal and endothelial cells and immune infiltrates. The only other protein study on ERβ5 in breast cancer was observational and conducted on just 17 cases (33). A larger study in colon cancer showed ubiquitous expression of ERβ5, leading the authors to conclude that its presence was uninformative (19).

One of the striking observations of this study was the strong statistical association of cytoplasmic ERβ2 immunoreactivity with reduced survival. There are observational reports of cytoplasmic ERβ immunoreactivity in breast cancer (6, 7, 20, 34), and we have previously shown that cytoplasmic expression is related to neoplastic progression with a higher incidence in in situ/invasive carcinoma compared with normal breast/benign intraductal proliferation (35). We have also shown herein that this staining is specific because it can be blocked by peptide competition, but until now this has not been formally examined. The functional role of cytoplasmic ERβ is currently unknown, although there are reports of mitochondrial ERβ (36). Metabolically active tumor cells may have increased mitochondrial activity, which might reflect increased cytoplasmic ERβ immunoreactivity.

Recent in vitro data have shown that cytoplasmic ERβ is actually more abundant than nuclear ERβ, and this seems to be modulated by 17β-estradiol (37), providing further evidence that the cellular location and/or hormonal environment of a cell may be a critical determinant of the functional role of ERβ. Indeed, in a recent study of 283 invasive breast tumors, cytoplasmic expression of total ERβ (i.e., without discriminating between isoforms) was commonly observed with significantly increased levels seen in high-grade tumors (38). This was also reflected in our second 322-case breast cancer tissue microarray, although the total number of cases is too small for detailed subgroup analysis. These data strongly support our observation that cytoplasmic ERβ2 expression predicts worse OS.

ERβ additionally displays estrogen-independent activity (39), and it is possible that its presence in the cytoplasm may potentially allow bidirectional cross talk with growth factor receptor pathways; evidence is accumulating to suggest that these are important in breast cancer growth and survival (40). This hypothesis remains to be tested but the presence of cytoplasmic ERβ2 supports the existence of novel pathways by which ERs can function as recently reviewed (41).

In ERα-negative breast tumors, no statistically significant associations were observed between ERβ isoforms and established clinical prognostic markers (42), suggesting that when expressed alone, ERβ isoforms have reduced significance than when coexpressed with ERα. Indeed, it has long been proposed and recently reinforced that the relative expression levels of ERα and ERβ could be key determinants of cell responses to agonists and antagonists (39, 43, 44). Notably, in a microarray study, introduction of ERβ to ERα-positive MCF-7 cells could significantly alter the gene expression profiles following hormone treatment (39). Furthermore, ERβ regulated a number of genes associated with cell cycle control and apoptosis (39, 45), which could collectively suppress tumor growth. It would now seem that specific ERβ isoforms may have additional contributions.

In summary, we show that ERβ isoforms have distinct prognostic significance in breast cancer. This depends on their cellular localization, as our study also highlights, for the first time, the importance of cytoplasmic as well as nuclear ERβ expression in dictating clinical outcome, with any form of cytoplasmic immunoreactivity predicting worse survival and, in particular, cytoplasmic ERβ2 staining in the absence of nuclear staining predicting worst survival. ERβ2 expression also adds to the prognostic value of ERα. The availability of robust antibodies, validated herein, should now facilitate measurement of nuclear and cytoplasmic ERβ isoforms, in particular ERβ2, in clinical breast cancer, providing a more comprehensive picture of patient outcome. We suggest that these could now complement current ERα measurements and this should be taken forward by including ERβ2 in large-scale prospective clinical trials to confirm our findings.

No potential conflicts of interest were disclosed.

We thank Mike Shires for technical assistance.