Estrogen receptor beta as a prognostic factor in breast cancer patients: A systematic review and meta-analysis

Background The prognostic role of estrogen receptor beta (ERβ) in early-stage breast cancer is unclear. We performed a systematic review and meta-analysis to evaluate the prognostic value of ERβ in early-stage breast cancer patients. Method We searched Medline, Embase, and the Web of Science for studies published between 1990 and 2015 that assessed ERβ status in breast cancer patients. A total of 25 studies comprising 9919 patients fitting our inclusion and exclusion criteria were included. The hazard ratios of ERβ status were extracted for diseases free survival (DFS)/) and overall survival (OS). Random or fixed-effects models were used when appropriate, and between-study heterogeneity was assessed. Results In the 20 studies that assessed ERβ status using immunohistochemical (IHC) methods, we observed significantly improved DFS in patients positive for ERβ-1 (HR=0.56, 95%CI 0.40-0.78, P=0.0007) and ERβ-2 (HR=0.67, 95%CI 0.45-1.00, P=0.05). Improved OS was associated with a positive status for pan-ERβ (HR=0.60, 95%CI 0.45-0.80, P=0.0004) and ERβ-2 (HR=0.44, 95%CI 0.31-0.62, P<0.0001). In ERα-positive patients, ERβ positivity was not associated with DFS (HR=0.77, 95%CI 0.46-1.27, P=0.31) or OS (HR=0.64, 95%CI 0.37-1.11, P=0.11). In contrast, ERβ expression was significantly associated with increased DFS (HR=0.37, 95%CI 0.14-0.93, P=0.03) or OS (HR=0.44, 95%CI 0.30-0.65, P<0.0001) in ERα-negative patients. We did not observe an association between ERβ mRNA levels and DFS and OS. Conclusion In this study, we showed that IHC ERβ status, rather than mRNA levels, is a prognostic factor that is associated with DFS and OS in breast cancer patients. The prognostic value of ERβ may be higher in ERα-negative patients than in ERα-positive patients.


INTRODUCTION
Estrogen receptor α (ERα) has been established as a significant predictor of the response to endocrine therapy in breast cancer patients. Immunohistochemical (IHC) examination of ERα status is the standard-of-care pathological evaluation used to guide adjuvant endocrine therapy after surgery. Anti-estrogen approaches are recommended in ERα+ patients. The discovery of a second ER, ERβ, has lead to the re-evaluation of estrogen activity in normal mammary development, breast tumorigenesis and tumor progression. Despite over 15 years of research on ERβ, its clinical significance remains unclear. Mann et al. [1] were the first to report the significance of ERβ in predicting long-term clinical outcomes (e.g., disease-free survival) in breast cancer patients, a result confirmed by other studies [2][3][4]. However, conflicting findings suggest that ERβ status is not associated with survival [5,6]. The aim of the present systematic review and meta-analysis was to investigate the association of ERβ status (positive vs. negative) and long-term clinical outcomes (e.g. disease-free survival, overall survival) of breast cancer patients.

Study characteristics
Twenty-five studies  with the full text available were identified and included in this study ( Figure 1). We examined the reference list of each study and did not identify any further studies for inclusion in our analysis. We included a total of 9919 patients from these studies. All publications were full-text articles. The features of the included studies are summarized in Table  1a, 1b. The mean patient age ranged from 48 to 68 years, and the median follow-up ranged from 27 to 174 months. Nine of the included studies had a quality score≥6 . None of the included studies were prospective, randomized trials. All of the studies were retrospective and did not report any information about allocation concealment or blinding methods. The matching criteria varied among the studies. Most of the studies reported the length of the follow-up period, and 12 of them exhibited a sufficiently long follow-up (defined as a median follow-up time >60 months) for the outcomes to be determined. The treatment of missing data was not sufficiently described in most of the studies.

The effect of ERβ on DFS
A total of 16 studies [2-7, 11-17, 19, 25] with available DFS used IHC as the method of ERβ assessment.

Sensitivity analysis and publication bias
A sensitivity analysis revealed that a positive ERβ (pan ERβ/ERβ-1/ERβ-2) status was significantly associated with improved DFS or OS in studies with a median follow-up time greater than 60 months [1-8, 10, 15, 17, 19, 20, 22, 23] (Table S1). ERβ-1 was not associated with DFS or OS in studies with a sample size ≥ 200 [2, 5-8, 17-19, 25]. The funnel plots for the studies Oncotarget 10377 www.impactjournals.com/oncotarget for DFS were symmetric, indicating no publication bias ( Figure S6). However, the distribution of the OS funnel plots was not symmetric. As shown in Figure S7, the studies focusing on specific ERβ (ERβ1/ERβ2) and pan-ERβ reactivity were mostly located on the left and right sides of the funnel plot, respectively, indicating possible publication biases.

DISCUSSION
ERβ was discovered nearly two decades ago, but its role as a prognostic or predictive factor in breast cancer remains elusive. Most studies examining ERβ as a biomarker have been retrospective, and these studies have used a variety of detection methods, leading to discrepant results. IHC is the most common method employed for ERβ assessment. In this meta-analysis study, we observed that a positive ERβ status, as assessed via IHC, was generally associated with improved DFS and OS. Multiple ERβ isoforms (ERβ-1, ERβ-2/cx) arise via alternative splicing of downstream coding exons or posttranslational proteolysis [26,27]. In this study, we noted that ERβ-2 was associated with improved DFS and OS. In contrast, ERβ-1 was associated with DFS, but not OS, which may be attributed to a study by Qui et al. [18], who provided the only report of an association between positive ERβ-1 status and a poorer OS. After the exclusion of this study, the pooled HR(95%CI) of ERβ-1 for OS changed significantly, from 0.55(95%CI: 0.20-1.50) to 0.38(95%CI: 0.25-0.57). The heterogeneity of the data synthesis was also eliminated. After careful examination, we noted in consistent results within Qui et al.'s study. In their report, they indicated that ERβ-positive patients exhibit a significantly worse overall survival prognosis compared with ERβ-negative patients. However, when stratified by HER2 status, the survival curves of the ERβpositive and ERβ-negative patients overlapped in both strata. The authors did not attempt to explain this result. We therefore suggest that the exclusion of this study from our meta-data analysis is appropriate.

Assessment method and clinical outcomes
Various methods had been used to assess ERβ status. Two studies employed immunoblotting as the detection method and revealed that a positive ERβ status was associated with improved DFS, similar to studies employing IHC [8,21]. However, a study by Wimberly et al. [22]showed no association between ERβ status and DFS when TMA was used to assess ERβ status in four largecohort populations. We speculate that TMA may not be an accurate method for ERβ assessment. Its major limitation is that the small cores employed to construct a TMA may not accurately and comprehensively represent the whole tissue specimen. Eckel-Passow et al. [28]reported that the number of TMA cores necessary to adequately represent the whole tissue specimen is biomarker-specific. They showed that 2-3 cores appeared to be adequate for assessing the status of B7-H3, Ki-67, CAIX, and IMP3 expression in renal cancer patients, whereas as many as 10 cores were insufficient for assessing B7-H1. Thus, the association between B7-H1determined in whole tissue sections and renal cancer-specific death is not easily revealed through TMA assessment.
Several studies found no consistent association between the mRNA and protein levels of ERβ [14,29,30]. Furthermore, an inverse association between ERβ mRNA levels and improved survival has been reported. Speirs et al. [31]noted that ERβ mRNA levels were increased in tamoxifen-resistant breast cancer patients. Similarly, Kim et al. [9] reported that a higher ERβ mRNA level is associated with poorer DFS in patients treated using endocrine therapy. We believe that the assessment of ERβ status based on mRNA levels may be inaccurate because samples from breast tissue might contain cells from surrounding cancerous tissue. Furthermore, post-transcriptional regulation may also compromise the prognostic value of ERβ mRNA [32]. In our meta-analysis, we found no association between ERβ mRNA levels and survival (DFS or OS). Hence, ERβ mRNA status does not appear to be promising for clinical use.

Prognostic role of ERβ varied by ERα status
As noted above, the prognostic value of ERβ varies depending on a patient's ERα status. The mechanism underlying this effect may be the molecular interplay between ERα and ERβ. Charn et al. [33] investigated the effects of ligand-occupied and unoccupied ERα and ERβ on chromatin binding. They showed that although ERα and ERβ restrict each other's binding site occupancy, ERα is dominant. The binding sites of ERα and ERβ overlap substantially when the are present alone. However, when both ERα and ERβ are present, only a few binding sites are shared. When both receptors are present, ERα displaces ERβ and shifts ligand binding to sites that are less enriched in the estrogen response element. This finding supports our observation that in ERα+ patients, the prognostic role of ERβ was less significant than in ERα-patients. Because endocrine therapy is administered to ERα+, but not ERα-, patients, we suggest that endocrine therapy may play a role as an effect modifier. Unfortunately, there are insufficient data to perform a meta-analysis addressing this issue. Novelliet al. [6]reported that in patients who receive endocrine therapy, a positive ERβ status is associated with increased DFS. Similar results have been reported by other investigators [1,8]. However, Yan et al. [23] found that a positive ERβ status was associated with improved OS in univariate, but not multivariate, analyses. O'Neil et al. [12]noted a trend (though not statistically significant) toward poorer DFS in patients with a positive ERβ status.  Hence, the predictive role of ERβ for the endocrine therapy response is unclear, due to the conflicting results provided by different studies [1,6,7,12,23].
No association between ERβ status and DFS/OS was observed in patients who did not receive endocrine therapy [1,7,23]. We believe that the sample sizes of these studies are too small to detect an association. Our group has initiated a multicenter randomized doubleblind prospective clinical trial comparing the efficacy of tamoxifen as an adjuvant endocrine therapy in early-stage ERα/PR-/ERβ+ breast cancer patients (ClinicalTrials.gov Identifier:NCT02062489). Sun et al. has initiated a similar multicenter study, in which early stage, triple-negative breast cancer patients are randomized into a toremifene/ anastrozole group or an observation group (Clinical Trials. gov Identifier: NCT02089854)

DFS and OS as clinical endpoints
We observed heterogeneity in the synthesis of the HR of pan-ERβ or ERβ-1 status for DFS. However, there was no heterogeneity in the synthesis of the HR of pan-ERβ, ERβ-1 (with the exception of Qui's study) or ERβ-2 status for OS. We suggest that this discrepancy may be due to the definition of DFS/OS. OS is a universally accepted measure of the clinical benefit of a treatment and can be precisely measured. As a result, there might be less heterogeneity for OS. In contrast, the definition of DFS varies between studies. For example, in the NSABP B-06 study [34], DFS was defined as the first recurrence of disease at a local, regional, or distant site, and the diagnosis of a second cancer and death without evidence of cancer were considered DFS events. In contrast, the guidelines from the DATECAN initiative (Definition for the Assessment of Time-to-event Endpoints in CANcer trials) [35] recommend that DFS should include death of from breast cancer as an event. Most of our included studies did not specify the definition of DFS, which may have resulted in heterogeneity in the synthesis of HR of DFS.

Publication bias
All of the included studies were retrospective and may be subject to publication bias. Insignificant HRs, especially following multivariate analysis, are less likely to be reported in retrospective studies. In the present study, we obtained asymmetric funnel plots for the synthesis of the HRs for OS. Studies reported significant HRs of ERβ-1 and ERβ-2 for OS tend to fall on the left side of the reference line, indicating that insignificant HRs are less likely to be reported. Several studies [4,6,23,36] reported finding no association between ERβ status and OS, but without an available HR and/or 95%CI. Hence, we must be cautious about the prognostic role of ERβ for overall survival.

Limitations
Several additional limitations should be addressed. First, IHC was commonly used for detecting ERβ status in most studies, but different hospital used varied commercial antibodies and didn't have uniform criteria. Reported data show that many commercially available IHC stains for ERβ have cross-reactivity with ERα [37]. Percentage of immunoreactive cells and allred scores were used to assess ERβ status, while the cut-off values varied from 1% to 25% (Percentage of immunoreactive cells) [6,14,17], and from 2-4 (Allred score) [2,11,15] across different studies. Different cut-off values used by different studies may cause limitation to our analysis. Second, some of the HRs were not available from the full-text of the included study, and were extrapolated from survival curves. Although, this method has been demonstrated to be feasible [38,39], we still consider this as a limitation. Additionally, HRs for synthesis in our analysis were derived from univariate and/or multivariate analysis (Table S2). This is also a major limitation, as the most standard approach should be collecting HRs derived from prospective controlled trials, with multivariate analysis adjustment.

CONCLUSION
In this meta-analysis, we showed that ERβ status, determined via IHC,is generally associated with DFS/OS in breast cancer patients. Assessment of ERβ mRNA levels is not recommended. As a prognostic factor, ERβ may be more important in ERα (+) patients than ERα (-) patients. Based on these findings, we recommend the initiation of a prospective study to confirm the prognostic value of ERβ in breast cancer patients.

MATERIALS AND METHODS
This study was waived the full IRB review of Sun Yat-sen Memorial Hospital, based on the institutional policy. This study was also performed according to the recommendations of the Cochrane Collaboration and the Quality of Reporting of Meta-analysis guidelines (MOOSE) and reported according to the PRISMA statements [40,41].

Study selection
We searched Medline, Embase, and the Web of Science for potentially relevant studies. The following keywords were searched in the "Title" or "Abstract": "Estrogen receptor," "Beta," and "Breast cancer," without restrictions on the region and publication type. English language was requied for publication. We manually www.impactjournals.com/oncotarget searched the retrieved articles to identify relevant studies. When multiple publications reported on the same study population, the report that was most complete or that had the longest follow-up period was used. The last date of the search was May 10 th , 2015.

Inclusion and exclusion criteria
Studies were eligible if they met the following criteria: (1) the main exposure of interest was early-stage breast cancer stratified by ERβ status (negative/positive or low/high expression); (2) the outcome of interest was disease-free survival or overall survival; (3) hazard ratios (HRs) with corresponding 95% confidence intervals (CIs) or survival curves for ERβ were reported; and (4) over 50 patients were enrolled in the study, which did not present redundant data.

Data extraction
Two reviewers (C.G. & W.T.) independently extracted the data from the included studies. Any disagreement was resolved by the third author (E.S.). The following data were collected: first author, year of publication, clinicopathological features of the study population, methods of ERβ assessment, number of included patients, and the reported outcomes. The outcomes assessed included disease-free survival (DFS) and overall survival (OS) in patients with different ERβ statuses. We assessed the quality of the included studies using the Newcastle-Ottawa quality assessment tool [42]. We allocated a score of 0-9 to each included study, and those with a score≥6 were considered to be of high quality.

Statistical analysis
The hazard ratio (HR) was used as a summary statistic for survival analysis, as described by Parmar and colleagues [43]. An HR of less than 1 indicated a survival benefit favoring ERβ+ patients. We used a random-effects model for this meta-analysis. The data were pooled and weighted using generic inverse variance. Heterogeneity between studies was assessed with the χ2 and I2 statistics. When higher values of the χ2 and I2 statistics (>50%) indicated heterogeneity between studies, we applied sensitivity and subgroup analyses to further evaluate the heterogeneity. We performed a sensitivity analysis when the outcome of interest was reported in more than 3 studies. We used funnel plot analyses to analysis to determine publication bias. A two-tailed p value of less than 0.05 was considered statistically significant. Statistical analyses were performed with Review Manager Version 5.3