Introduction

The prognosis of breast cancer has been significantly improved in recent years because of the discovery of specific predictive biomarkers that enable distinct molecular subtyping of this disease for personalized therapy [1]. Estrogen receptor-alpha (ER-α) and progesterone receptor (PR) are well-established biomarkers that predict the response to endocrine therapy and favorable outcomes in woman with breast cancer [2, 3]. Additionally, another type of estrogen receptor, ER-β, was recently shown to be a prognostic marker in breast cancer [4–6]. Since the discovery of wild-type ER-β (ER-β1) in 1996 [7], other splice variants such as ER-β2 (ER-βcx), ER-β3 (ER-βN), ER-β4, and ER-β5 have been sequentially discovered [8–10]. The most notable variant form is ER-β2, which was reported to form heterodimers with ER-α or ER-β1 without ligand binding, and identified as functional modulators of ER-α and ER-β1 [11, 12].

Although ER-β has been discovered for almost 20 years, its prognostic value in breast cancer remains controversial. Several studies have demonstrated that ER-β1 expression was associated with improved outcomes in breast cancer [4, 13–16]. In contrast, others have found either no prognostic value [17–22] or the opposite role of ER-β1 on patient outcomes [23]. The splice variants of ER-β in breast cancer are another confounding issue. Although several studies have associated ER-β2 protein with better prognosis [13, 24], others have suggested that the ER-β2 protein expression had no impact [4, 20, 25] or even a worse effect [26] on patient outcomes. Moreover, a recent study reported a contradictory prognostic value for nuclear and cytoplasmic expression of ER-β2 [17]. In addition to protein expression, ER-β mRNA was also analyzed in several clinical studies, with the results being also contradictory [13, 18, 27].

Since molecular subtyping is a widely accepted technique as a part of in breast cancer management, whether ER-β has prognostic value in breast cancer with different ER-α status becomes an important issue. A few studies have described the effect of ER-β expression in breast cancer without coexpression of ER-α, and the results were not consistent [4, 14, 16, 21, 28–30]. Preclinical research has identified a bi-faceted activity of ER-β in breast cancer: In ER-α positive breast cancer cells or xenografts, ER-β showed an antiproliferative and proapoptotic activity. Whereas in ER-α negative cells or xenografts, an proliferative activity of ER-β has been observed [31–35]. Therefore, ER-β expression may have different effects on the prognosis of ER-α positive and negative breast cancers.

Thus, the aim of this meta-analysis is to comprehensively assess the prognostic value of distinct splice variants of ER-β in breast cancer and to evaluate whether the impact of ER-β expression on breast cancer prognosis varies with the ER-α status.

Materials and methods

The methods of literature search strategies, data extraction and quality assessment, and statistical analysis were performed according to the recommendations of the Cochrane Collaboration and the Quality of Reporting of Meta-analyses guidelines [36, 37].

Data sources and literature search

The systematic literature search was performed independently by two authors(J. Q. L. and H. S. G.) on articles published between January 1990 and August 2015. A computerized search of the Medline, Embase, and Cochrane Library databases was performed with restriction on the English language articles. The search terms included “breast cancer,” “breast neoplasms,” “ERbeta,” “survival,” and “prognosis.” We used the “related articles” to broaden the search. Reference lists of retrieved articles were manually screened to identify related articles. When a study generated multiple publications, the most recent publication with higher quality was included in the analysis.

Study selection

Inclusion criteria included studies on non-metastatic breast cancer with availability of disease-free survival (DFS) or overall survival (OS) for 5 years, and studies including more than 50 patients. The exclusion criteria were listed below: (1) The inclusion criteria were not met. (2) Studies that presented insufficient data. (3) Editorials, letters, review articles, comments, conference abstracts, case reports, and animal experimental studies.

Outcome measures

Data for DFS and OS at 5 years were collected. If the outcomes were available, when tumors were stratified by ER-α status, data for 5-year DFS and/or OS were assessed as well.

Data extraction and quality assessment

Two reviewers (J. Q. L. and H. S. G.) independently considered the eligibility of potential titles and abstracts. Inter-reviewer agreement was assessed using Cohen’s kappa coefficient. Disagreement was resolved by further consensus. Detail data were collected using predesigned abstraction forms. Survival data were extracted from tables or results of included articles or assessed from the Kaplan–Meier (K–M) curves where applicable as previously described [38]. The quality of observational studies was assessed by the modified criteria suggested by the Newcastle-Ottawa quality assessment tool [39]. A score of 0–9 was allocated to each observational study. A score of 7–9 was reflective of high quality (e.g., low risk of bias), a score of 4–6 indicated moderate quality, and a score of 3 or less indicated low quality (e.g., high risk of bias). Assessment was conducted by both reviewers independently.

Statistical analysis

The hazard ratio (HR) was used as summary statistics for long-term survival analysis as described by Parmar et al. [40]. All outcomes were reported with 95 % confidence interval (CI). Statistical heterogeneity between studies was tested using the Chi-squared test with significance set at P < 0.05. The random-effects model was used if there was high heterogeneity, otherwise, the fixed-effects model was reported. Heterogeneity was assessed using the I 2 statistic, and an I 2 value of <25 % was defined to represent low heterogeneity; a value between 25 and 50 % was defined as moderate heterogeneity, while >50 % indicated high heterogeneity. Sensitivity analyses were carried out for different cut-offs for defining ER-β expression, as well as for potential heterogeneity in the definition of ER-β positive expression (nuclear vs. cytoplasmic staining). Subset analysis was performed to investigate potential sources of heterogeneity among studies and to assess the prognostic values of ER-β expression for breast cancers with different ER-α status (ER-α positive or negative). Statistical analysis was conducted by the metareg procedure STATA 12.0 (StataCrop, College Station, TX). All statistical tests were two-sides, and statistical significance was defined as P < 0.05.

Results

Flow of included studies and study characteristics

Among the 464 abstracts identified, 62 full text articles were retrieved for detailed evaluation. Additional 41 articles were excluded, and 21 studies were included in final analysis. Figure 1 illustrates the study screening and selection process. Agreement between the two authors was 97 % for study selection. For 5-year DFS, 15 studies provided data for ER-β1 protein, 6 studies provided data for ER-β2 protein, and 3 studies provided data for ER-β1 mRNA, but no enough studies (less than 3) provided data for the meta-analysis of ER-β2 mRNA, one study with four different study cohorts provided data for ER-β5 protein. For 5-year OS, data were available for ER-β1 protein from 10 studies, and data were available for ER-β2 protein from 7 studies, while no enough studies (less than 3) offered data for the meta-analysis of ER-β1 or β2 mRNA. For other ER-β isoforms, such as ER-β3, there was only one study available for data collection, respectively, so we could not perform meta-analysis for their prognostic values.

Fig. 1
figure 1

Flow diagram of included studies

Full size image

Due to lack of randomized controlled trials addressing the research questions of the current study, all studies included were of retrospective cohort nature. The characteristics of included studies are shown in Table 1. The median age of breast cancer patients in the individual studies ranged from 46 to 68 years, and the median follow-up was reported between 27 and 165 months. There existed heterogeneity in study populations: among the total 21 articles, 17 studies included both premenopausal and postmenopausal breast cancer women, 3 studies included only postmenopausal women, and one study enrolled just premenopausal women; of the 21 included researches, 16 studies enrolled both ER-α positive and ER-α negative breast cancer patients, 3 studies only enrolled ER-α positive breast cancer patients, and two studies only included ER-α negative breast cancer cases.

Table 1 Characteristics of included studies
Full size table

Among the 6.769 patients included in the studies focusing on ER-β1 protein, 4001 patients (59.1 %) had positive ER-β1 protein expression tumors. Of the 2295 patients in the studies about ER-β2 protein, 1463 patients (63.7 %) had breast tumors with positive ER-β2 protein expression. Of the 2271 patients in the studies on ER-β5 protein, 1735 patients (76.4 %) had breast tumors with positive ER-β5 protein expression. Detection of ER-β protein expression was carried out using immunohistochemistry (IHC) in most of studies (85 %), quantitative immunofluorescent staining (QIF) in one study (5 %), immunoblot in one study (5 %), and Western blot in another study (5 %). Of the 17 studies using IHC test, the positive cut-off was 10 % in three studies, 20 % in three studies, 50 % in one study, and 1 % in one study, while the rest 9 studies applied scores (intensity plus proportion of positive cells) as positivity judgment: Allred score >2 in two studies, Allred score >3 in two studies, Allred score >5 in one study, and 3 studies used a semi-quantitative score (0–300) system, another study used a rare z-score system. All of the three studies with available data for ER-β mRNA expression performed real-time PCR for mRNA analysis.

Quality of included studies

We evaluated the risk of bias in the 21 included nonrandomized studies using a modification of the Newcastle–Ottawa scale (data are shown in Table 2). 18 studies scored ≥7 and were considered to be of high quality (low risk of bias), and 3 studies were considered to be of moderate quality because of a score of 6.

Table 2 Risk of bias in the observational studies using Newcastle-Ottawa Scale
Full size table

ER-β1 protein expression and breast cancer outcomes (DFS/OS)

A total of 15 studies provided data for the effect of ER-β1 protein expression on 5-year DFS. There was statistically moderate heterogeneity between studies (P = 0.006, I 2 = 47.7 %), which may be due to the different cut-off of ER-β1 positive expression among studies. There was a significant association between ER-β1 protein expression and improved 5-year DFS (HR 0.690, 95 % CI 0.610–0.779; P < 0.001) (Fig. 2a). Subset analysis showed positive ER-β1 protein expression was associated with improved DFS for ER-α positive breast cancer patients (HR 0.754, 95 % CI 0.605–0.940; P = 0.012) (heterogeneity: P = 0.432, I 2 = 0.0 %) (Fig. 2b), whereas no significant association was found between positive ER-β1 protein expression and 5-year DFS for ER-α negative patients (HR 0.600, 95 % CI 0.336–1.070; P = 0.083) (heterogeneity: P = 0.017, I 2 = 70.7 %) (Fig. 2c). The significantly high heterogeneity between studies for ER-α negative patients resulted from the inclusion of one study [30] with outlying data. This study defined the ER-β1 positive as tumor with ≥1 % nuclear-stained cells, whereas most of the other studies used 10 or 20 % as a cut-off of ER-β1 positive expression. Exclusion of the one study with outlying data eliminated the heterogeneity (P = 0.211, I 2 = 35.7 %), but identified that positive ER-β1 protein expression was associated with improved DFS for ER-α negative patients as well (HR 0.423, 95 % CI 0.224–0.797; P = 0.008) (Fig. 2d). Thus, subset analysis suggested that higher ER-β1 expression was associated with better 5-year DFS in both ER-α positive and negative patients.

Fig. 2
figure 2

Forest plot of association between ER-β1 protein and 5-year DFS for a whole non-metastatic breast cancer population, b ER-α positive breast cancer patients, c ER-α negative breast cancer patients, and d ER-α negative breast cancer patients when one study with outlying data was excluded

Full size image

A total of 10 studies reported data for the impact of ER-β1 protein expression on 5-year OS. There was again statistically heterogeneity between studies (P = 0.009, I 2 = 54.7 %), which might be owing to the different cut-off of ER-β1 positive expression among studies. Similarly, patients with ER-β1 protein positive breast cancer showed significant better 5-year OS compared with those with ER-β1 negative breast cancer (HR 0.632, 95 % CI 0.533–0.749; P < 0.001) (Fig. 3a). Inclusion of studies with the same cut-off (10 %) of ER-β1 eliminated the heterogeneity (P = 0.486, I 2 = 0 %) but remained the statistically significant positive impact of ER-β1 protein on the 5-year OS (HR 0.352, 95 % CI 0.234–0.531; P < 0.001) (Figure S1). Subgroup analysis evaluated positive ER-β1 protein expression was associated with improved 5-year OS for ER-α positive breast cancer patients (HR 0.694, 95 % CI 0.520–0.926; P = 0.013) (heterogeneity: P = 0.108, I 2 = 50.6 %) (Fig. 3b), but no significant association was found between positive ER-β1 protein expression and OS at 5 years for ER-α negative patients (HR 0.859, 95 % CI 0.475–1.554; P = 0.616) (heterogeneity: P = 0.127, I 2 = 47.4 %) (Fig. 3c).

Fig. 3
figure 3

Forest plot of association between ER-β1 protein and 5-year OS for a whole non-metastatic breast cancer population, b ER-α positive breast cancer patients, and c ER-α negative breast cancer patients

Full size image

ER-β1 mRNA expression and breast cancer outcomes (DFS)

Three studies provided data for the prognostic value of ER-β1 mRNA expression on 5-year DFS. Because of the general interstudy heterogeneity without evidence of inclusion of studies with outlying data, there was high heterogeneity between studies (P = 0.029, I 2 = 71.6 %). The general heterogeneity might be due to distinct study populations of these studies: Hiroshi et al. included only ER positive breast cancer patients, PA O’ et al. focused on the postmenopausal women, while the study objects of G. C. et al. were the general breast cancer patients [13, 18, 27]. The result showed no significant association between positive ER-β1 mRNA expression and 5-year DFS (HR 0.915, 95 % CI 0.581–1.440; P = 0.700) (Fig. 4).

Fig. 4
figure 4

Forest plot of association between ER-β1 mRNA and 5-year DFS of breast cancer patients

Full size image

ER-β2 protein expression and breast cancer outcomes (DFS/OS)

Six papers reported data for the relationship between ER-β2 protein and 5-year DFS. There was statistically high heterogeneity between studies (P = 0.003, I 2 = 72.0 %), and this can be explained by different cut-off of ER-β2 positive expression among studies. Similar to the prognostic value of ER-β1 protein, breast cancers with positive ER-β2 protein expression were associated with a statistically significant improvement in DFS (HR 0.799, 95 % CI 0.644–0.992; P = 0. 042) (Fig. 5).

Fig. 5
figure 5

Forest plot of association between ER-β2 protein and 5-year DFS of breast cancer patients

Full size image

A total of 7 studies provided data for the impact of ER-β2 protein expression on OS at 5 years. There was statistically high heterogeneity between studies (P = 0.002, I 2 = 69.4 %). This significant heterogeneity resulted from one study with survival data of both nuclear and cytoplasmic ER-β2 protein expressions [17], while the other 6 studies all focused on nuclear ER-β2 protein expression [4, 13, 24, 25]. When all studies were included in the analysis, the positive prognostic impact of ER-β2 protein on 5-year DFS failed to translate to a significant difference in 5-year OS (HR 0.958, 95 % CI 0.762–1.205; P = 0.712) (Fig. 6). Exclusion of the study contained both nuclear and cytoplasmic data eliminated the heterogeneity (P = 0.134, I 2 = 40.8 %) but remained the conclusion (HR 0.877, 95 % CI 0.664–1.158; P = 0.354) (Figure S2).

Fig. 6
figure 6

Forest plot of association between ER-β2 protein and 5-year OS of breast cancer patients

Full size image

Very few studies reported data for the effect of ER-β2 expression in ER-α negative breast cancer, so it was impossible to assess the associations between ER-β2 and patient outcomes (DFS/OS) of breast cancers with different ER-α status.

ER-β5 protein expression and breast cancer outcomes (DFS)

One study with four distinct breast cancer patient cohorts reported data for the relationship between ER-β5 protein and 5-year DFS, so we defined this study as four separate analyses. There was no significant heterogeneity between study cohorts (P = 0.489, I 2 = 0.0 %). The result showed no significant association between positive ER-β5 protein expression and 5-year DFS (HR 1.070, 95 % CI 0.810–1.410; P = 0.642) (Fig. 7).

Fig. 7
figure 7

Forest plot of association between ER-β5 protein and 5-year DFS of breast cancer patients

Full size image

Sensitivity analysis

Among studies using IHC for evaluation of ER-β1 expression, the cut-off of positive staining varied with studies. Inclusion of studies with the same cut-off (10 %) of ER-β1 remained the positive prognostic impact of ER-β1 expression on OS at 5 years (HR 0.352, 95 % CI 0.234 to 0.531; P < 0.001). And this totally eliminated the heterogeneity (P = 0.486, I 2 = 0 %).

Sensitivity analyses were carried out for potential heterogeneity in the definition of ER-β positive expression (nuclear vs. cytoplasmic staining) as well. Removal of one study [17] with both nuclear and cytoplasmic staining as the definition of positive ER-β2 expression eliminated the heterogeneity (P = 0.134, I 2 = 40.8 %), but remained the conclusion that there was no significant association between positive ER-β2 protein expression and 5-year OS (HR 0.877, 95 % CI 0.664–1.158; P = 0.354).

Discussion

Estrogen is essential for the growth and development of both normal and neoplastic mammary tissues, and exerts most of its effects via two estrogen receptors, ER-α and ER-β. A number of studies have described the association of ER-β expression and patient outcome in breast cancer since its discovery in 1996 [7]. However, the exact role of distinct splice variants of ER-β in breast cancer remains to be established, and whether the outcome is consistent among different breast cancer subtypes is also uncertain. To the best of our knowledge, this is the first meta-analysis that estimated the associations between ER-β expression and clinical outcomes in non-metastatic breast cancer. Our results demonstrated that wild-type ER-β (ER-β1) was a molecular marker of good prognosis for breast cancer patients, with both favorable DFS and OS at 5 years, and ER-β2 was associated with improved 5-year DFS, but not OS. We further identified that the positive prognostic value of ER-β1 for 5-year DFS was independent of ER-α status, but this effect for 5-year OS was dependent on coexpression of ER-α.

Previous studies have reported inconsistent results on the prognostic role of ER-β1. Several studies showed ER-β1 protein was associated with improved outcomes in breast cancer [4, 13–16]; however, others demonstrated either no significant association [17–22] or opposite effect of ER-β1 on prognosis [23]. Here, we show ER-β1 is a positive predictive biomarker for breast cancer, although one Chinese study [23] focused on DFS was not included because the detailed survival data is lacking. Nevertheless, lack of data from this Chinese study may not influence the conclusion due to its outlying data. The positive rate of ER-β1 expression in this study was only 21.1 %, which was much lower than those reported in other studies (average 66.8 %, range 45.1–98.8 %) [4, 6, 13–15, 18, 20, 28, 29, 41–43]. So the ER-β1’s association with DFS showed in this study may not reflect the true clinical situation. Another interesting finding was that ER-β2 expression was also associated with improved 5-year DFS; while we noticed distinct prognostic roles of nuclear and cytoplasmic ER-β2 expression. Since very few studies focused on the effect of cytoplasmic ER-β expression on survival [17, 44], the different prognostic values between nuclear and cytoplasmic ER-β2 should be interpreted with caution. A recent study reported of the ER-β in mitochondria [45], but the exact function of cytoplasmic ER-β remains unclear. Further studies are needed to investigate the clinical prognostic significance of cytoplasmic ER-β and its underlying mechanisms. Very few studies have data on the association between ER-β5 expression and patient outcomes [17, 22]. Here, we defined that there was no significant association between positive ER-β5 protein expression and 5-year DFS, but this meta-analysis only included four study cohorts with very long inclusion periods (from 1962 to 2003). Another retrospective study found that positive ER-β5 protein expression was significantly correlated with improved OS [17]. It is really difficult to draw conclusions on the prognostic effect of ER-β5 expression in breast cancer, which merits further researches.

Preclinical studies have shown a dual activity of ER-β in different molecular subtypes of breast cancer: positive ER-β expression suggested a favorable prognosis in ER-α positive breast cancer, but a poor prognosis in ER-α negative breast cancer [31–35]. Nevertheless, our meta-analysis did not support such a conclusion. Anti-estrogen therapy is currently a standard treatment for ER-α positive breast cancer and normally not used in ER-α negative ones. However, an EBCTCG (Early Breast Cancer Trialists’ Collaborative Group) meta-analysis involving 75,000 women showed that more than 10 % of ER-α negative breast cancer patients could benefit from endocrine therapy [46]. This result might be partially explained by the effect of ER-β. Therefore, the positive prognostic role of ER-β, particularly in ER-α negative breast cancer, suggests the possibility to explore hormone therapy in ER-α negative and ER-β positive breast cancer. A retrospective study reported that the classical ERs antagonist drug tamoxifen had significant clinical efficacy for ER-α negative and ER-β positive early-stage breast cancer patients, although it has not been widely recommended yet [4]. Drugs targeting ER-β may expand the conventional indications of endocrine therapy for breast cancer and might be a novel treatment option available for ER-α negative and ER-β positive cancer patients.

Expression of ER-β in breast cancer has been analyzed at the mRNA or protein level in several studies. The results on the association between ER-β mRNA expression and survivals were conflicting [13, 18, 27]. Our analysis showed no significant correlation between ER-β1 mRNA expression and 5-year DFS. It is possible that mRNA data is from a mixture of tumor cells and stromal cells in breast tissue, which is different from protein expression only in tumor cells. Indeed, a poor correlation between ER-β mRNA and protein levels was observed in several previous studies [13, 19, 47, 48].

In an attempt to review the literature, we noted that no randomized control trials focused on this issue, and all the included studies were retrospective cohort design. Another limitation is the heterogeneity in patient populations, methods for assessing ER-β, different antibodies for the identification of ER-β expression (some of them might be non-specific or insensitive), as well as the definition of ER-β positivity, though we have reduced this heterogeneity by performing sensitivity analysis. Furthermore, Chen et al. reported a novel mechanism through which ER-β might contribute to aggressiveness in HER2-positive breast cancer cell lines [49]. Thus, ER-β might play an important role in HER2 positive breast cancer. But we were unable to identify the effect of ER-β on outcomes in HER2-overexpressing tumors because of the limited number of available studies.

Conclusions

In conclusion, the protein expression of wild-type ER-β (ER-β1) and its variant ER-β2 are associated with improved outcomes, and the positive prognostic role of ER-β1 for 5-year DFS is independent of the ER-α status, yet this effect for OS occurs only in ER-α positive breast cancers. This prognostic significance of ER-β indicates the existence of new molecular subtypes of hormone-sensitive breast cancer. Endocrine therapy drugs targeting ER-β might be a novel treatment option for ER-α negative, ER-β positive breast cancer patients. Further randomized clinical trials are warranted to determine the exact efficacy and safety of these strategies.