Estrogen Receptor β (ERβ) Level but Not Its ERβcx Variant Helps to Predict Tamoxifen Resistance in Breast Cancer

Tamoxifen is one of the first-line adjuvant therapy options in women with ER-positive breast cancer. However, in 30 to 40% of cases, these tumors relapse within 5 years of tamoxifen treatment, which requires the cessation of the regimen and the initiation of a second-line therapy. The mechanism of tamoxifen resistance in ER-positive breast cancer is unknown despite extensive studies (1, 2, 3). Tamoxifen either is inactive and unable to block the mitogenic effect of estrogen and growth factors or behaves as an agonist that stimulates the growth of cancer cells and induces growth-associated genes, as shown in different cell lines selected for their ability to grow with this antiestrogen (4, 5). This estrogenic effect of tamoxifen can be blocked by pure antiestrogens (6).

It has been established, however, both in cell lines (7, 8) and in patients (9), that tamoxifen is mostly active in ER-positive breast cancer, and that the assay of ER in cytosol or in tumor section is the first predictive marker used in practice to guide the clinicians in defining systemic therapy (10).

The recent discovery of a second ER, named ERβ (11), and of several of its variants, raised the question of the relative value of ERα and ERβ in predicting tamoxifen resistance or sensitivity in breast cancer patients. ERβ binds antiestrogens and their hydroxylated metabolites (12) with a higher affinity than does ERα (13). Both the full-length ERβ (ERβ1) and its COOH-terminally truncated splice variant (ERβcx or ERβ2), which is unable to bind tamoxifen, have been found in breast cancer (14, 15). They are able to act as dominant negative of ERα after heterodimerization (16), but their significance in antiestrogen resistance is controversial. It has been proposed that the action of tamoxifen on ERβ stimulates tumor growth via AP-1 interactions (17). Conversely, ERβ could inhibit the agonist activity of tamoxifen for instance on AF-1, the activating domain of transcription of ERα (18, 19). Finally, ERβ might have no value in predicting tamoxifen efficacy or resistance.

To discriminate among these possibilities, we have quantified anonymously by immunohistochemistry the expression of total ERβ protein and its variant ERβcx in 50 archival ER-positive breast carcinomas, which had been treated by tamoxifen as the only adjuvant therapy, and we have compared their value in tamoxifen-resistant and tamoxifen-sensitive tumors, defined according to the presence or absence of relapse within 5 years of standard tamoxifen therapy (20).

The files of 850 patients with primary breast carcinomas treated during 1992, at the Val d’Aurelle Cancer Center in Montpellier, France, were considered for this study. Patients were selected according to the following criteria: (a) absence of neoadjuvant therapy; (b) tumor diameter greater than 1 cm allowing biochemical assay; (c) ER-positive tumor according to cytosolic radioligand assay (≥10 fmol/mg protein); (d) adjuvant therapy exclusively by tamoxifen for 5 years (20 mg/d); (e) availability of paraffin blocks for analysis; and (f) complete clinical data and sufficient follow-up. The tamoxifen-resistant patients were defined as those patients who recurred while on adjuvant tamoxifen therapy (up to 5 years). The tamoxifen-sensitive patients were defined as those patients who had not recurred while on tamoxifen therapy during 5 years. Only 50 cases of 850 could be included in this study with 16 tamoxifen-resistant cases and 34 tamoxifen-sensitive cases. Histopathologic grading of tumor was obtained according to Scarff-Bloom-Richardsson (SBR) modified by Elston (21, 22). Nodal status was obtained by histologic analysis of at least eight axillary nodes. Menopausal status was determined by clinical and hormonal analysis.

All of the tumor samples were fixed in formalin-alcohol solution and embedded in paraffin. The archived breast cancer specimens were studied by immunohistochemistry. The pathologist (ME-S) was blinded to the patient characteristics. Immunostaining was performed with ERβ antibodies obtained in Dr. J-Å. Gustafsson’s laboratory (Department of Medical Nutrition and Biosciences, Karolinska Institute, Novum, Huddinge, Sweden). The chicken polyclonal ERβ 503 IgY antibodies recognize total ERβ proteins (both full-length ERβ and its splice variants) and have been previously validated for immunohistochemistry (23, 24), including validation by protein extinction with authentic ERβ protein (23). The ERβcx polyclonal antibodies were raised in sheep against the 14-amino-acid peptides of the COOH-terminal region: MKMETLLPEATMEQ. Analysis was also performed by ERα (clone 6F11, Novocastra, United Kingdom), progesterone receptor [PgR (clone PgR 636, Dako)], Ki67 (clone MIB-1, Dako), and two HER2/neu (c-ErbB2) markers [polyclonal A0485 (Dako, Denmark) and monoclonal CB11 (Novocastra)]. Adjacent sections of 5 μm each were deparaffinized in xylene and rehydrated with graded EtOH concentrations. Before staining, a heat epitope retrieval procedure was performed. Sections were pretreated by pressure cooking for 15 minutes in EDTA buffer (pH 7) for ERβ and ERβcx, and by waterbath for 40 minutes at 95° for the other markers, with citrate buffer (pH 6) for ERα, PgR, and HER2/neu, and Tris-EDTA buffer (pH 8) for MIB-1. For ERα (1:50 dilution), PgR (1:100 dilution), MIB-1 (1:100 dilution), and c-ErbB2 (1:500 dilution for polyclonal antibody and 1:800 dilution for CB11 antibody), immunohistochemical labeling with the “Dako LSAB^R 2 System-HRP” was performed at room temperature with the automated Dako Autostainer (code no. K0675); and 3′,3′-diaminobenzidine tetrahydrochloride (DAB) was used as a chromogen. The immunohistochemical procedure for total ERβ marker was described previously (23). A similar protocol was performed for polyclonal sheep ERβcx antibody (1:300 dilution), except for the use of an appropriate secondary biotinylated antisheep antibody (Santa Cruz Biotechnology, Santa Cruz, CA). Negative controls were performed by the replacement of primary antibody by IgY nonspecific serum (Nordic, Netherlands) for ERβ, mouse IgG1 nonspecific serum (X0931, Dako) for ERα, PgR, and MIB-1 markers, with similar protein concentrations. Positive external controls were used in each experiment, sections of OVCAR cells, pellet-embedded in paraffin, were used for ERβ, and a positive breast cancer sample was used for each other marker. Adjacent normal breast tissue was also used as an internal control for ERα, PgR, MIB-1, and ERβ. ERβcx specificity of immunostaining was established by preincubating the sheep polyclonal ERβcx antibody with a 10-fold excess of ERβcx peptide. It was also shown with pre-adsorbed ERβcx antiserum (1:2100 dilution), and with preimmune sheep serum for ERβcx (1:5000 dilution). In each ERβcx experiment, pre-immune sheep serum was used, in addition to a positive external control (breast cancer tissue overexpressing ERβcx). Archival material of mammary tumor recurrence and/or metastasis was obtained for six resistant patients and was analyzed with the same markers.

Quantification was performed with a computerized image analyzer (Samba 2005 TITN, Alcatel, Grenoble, France) as described previously (23). Ten to twelve microscopic fields (G200) of invasive tumor, representative of all surface cut, were analyzed for ERα, -β, and -βcx and for PgR. The highly stained fields were chosen for MIB-1 proliferation marker assessment. Results were expressed as the percentage of nuclear-stained epithelial cells, or as a quantitative immunocytochemical (QIC) score [(percentage of surface stained in epithelial cells) × (mean staining intensity) × 10] expressed in arbitrary units (AU). The percentage of nuclear staining of negative control was usually nil and, when weak, was subtracted. A semiquantitative method was performed for c-erbB-2 membrane staining, according to the Dako Hercept Test scoring.

All of the parameters were analyzed by continuous values and by expression status. Receptor status were defined taking as cutoff points the median values observed in the 50 ER-positive cases (70% for ERβ, 30% for ERβcx, 50% for ERα, and 20% for PgR). Because the sensitivity of the assay may vary according to the receptors, these values may not indicate their relative level in the tumor. Univariate analysis comparing resistant and sensitive cases were performed by Fisher’s exact test for categorical variables and by the two-sample Wilcoxon test for all continuous parameters. The Wilcoxon test and the Spearman correlation coefficient were used to evaluate the relationship between ERβ and ERβcx expression with the other parameters. P values < 0.05 were considered statistically significant. The multivariate analysis was carried out in two steps by first introducing all of the immunohistochemical variables in a stepwise backward logistic regression model (25) Significant clinical variables were then introduced to investigate the relationships with the immunohistochemical variables. Statistical significance was measured by the likelihood ratio test. Odds ratios were used to summarize the effects. Statistical analyses were performed with Stata software (StataCorp, College Station, TX; ref. 26).

Two groups of patients were compared according to the occurrence of relapse within 5 years of tamoxifen therapy. The 16 tamoxifen resistant cases relapsed within a median of 3 years from surgery (range 14–56 months). Among the 34 tamoxifen sensitive cases, 4 patients relapsed after 80 months, and the other 30 patients were alive and disease free at a median follow up of 9.4 years (range 60–128 months). Most clinical and pathologic characteristics were not different in the resistant and sensitive groups (Table 1). The only differences were SBR grading and tumor size, which were more elevated in resistant cases.

As shown in Fig. 1,A-a, ERβ immunoreactivity was detected in the nuclei of invasive breast cancer cells, where brown staining was totally abolished with an excess of antigen (23). Since the cytoplasmic staining was not fully abolished, only the nuclear staining was quantified. The absence of cross-reactivity between ERβ and ERα antibodies was also confirmed as shown in Fig. 1,A-d and -e. Nuclear intensity and staining distribution of invasive tumors were variable according to the patient. Staining distribution was either diffuse in all of the tumor, or was focal and generally localized in tumoral islets at the periphery of the tumor. Fig. 1,A shows a typical example of a tamoxifen-sensitive (Fig. 1,A- a, -b, and -c) and tamoxifen-resistant (Fig. 1,A-d, -e, and -f) invasive ductal breast carcinoma, with similar SBR grade (II) and size (pT₂) and without nodal invasiveness. In the tamoxifen-sensitive case, there was a strong expression of ERβ (Fig. 1,A-a), with a low MIB-1 proliferation rate (Fig. 1,A-c). The tamoxifen-resistant case showed a low ERβ expression (Fig. 1,A-d) contrasting with a high ERα level (Fig. 1,A-e), and high MIB-1 staining (Fig. 1 A-f). ERβ nuclear staining was also detected in epithelial and myoepithelial cells of normal mammary glands, in stromal and inflammatory cells.

ERβcx immunostaining is shown in Fig. 1,B. The nuclear staining mostly observed in cancer cells (Fig. 1,B-a and -c) contrasted with a weak cytoplasmic staining. Nuclear staining was also observed in some stromal endothelial cells and inflammatory cells such as lymphocytes and macrophages (Fig. 1,B-a). The specificity of the ERβcx immunostaining was evidenced by three criteria: (a) nuclear signal was removed by adding a 10-fold excess of the antigen (Fig. 1,B-b), (b) nuclear signal was removed by using an ERβcx pre-adsorbed antiserum (not shown), and (c) nuclear signal was removed by using the pre-immune serum (Fig. 1,B-d). ERβcx reactivity varied according to patients (Table 2 and Fig. 3).

As shown in Table 2, total ERβ level was significantly higher in sensitive tumors than in resistant cases, when comparing the percentage of stained nuclei or QIC score. The same difference was found with continuous values (Fig. 2,A) or status expression taking the median as a cutoff level of 70% of stained nuclei. Unlike total ERβ, ERβcx level (either with percentage or with QIC score) did not differ between resistant and sensitive tumors. The difference between total ERβ and ERβcx was significantly higher in sensitive tumors. The difference in ERα levels between the two groups was not significant (Fig. 2). However the ERα-positive tumors (≥50% of stained nuclei) were mostly seen in tamoxifen-sensitive patients. The ERα/ERβ ratio, estimated in adjacent sections of each tumor, and PgR expression were not different between the two groups. The proliferation rate assessed by MIB-1 was greater in the resistant group (P = 0.01), with 63% of resistant cases expressing more than 10% of stained nuclei as compared with 32% of sensitive cases. We found no HER2/neu (c-erbB2) overexpression in any of the 50 tumors, which is consistent for ER-positive tumors. We found no significant variation in ERβ, ERα, PgR, and MIB-1 levels between the primary tumor and recurrence or metastasis for the same patient, but the number (six cases) was too small to reach a conclusion.

ERβ expression was independent of all parameters, including PgR (Table 3). It was correlated only with ERβcx expression (Spearman correlation coefficient, r, = 0.63, P < 0.001). The percentage of ERβ-positive cells was always superior to the percentage of ERβcx-positive cells (Fig. 3). Interestingly, MIB-1 was inversely correlated with ERα level (P = 0.003) but not with ERβ levels. A positive correlation was observed, however, between ERβ expression and MIB-1 proliferation index in the tamoxifen-resistant tumors (r = 0.51, P = 0.04), but no relationship was found in the tamoxifen-sensitive group. All 14 patients with a low MIB-1 proliferation index (<10%) and a high ERβ status (≥70%) were tamoxifen sensitive (Fig. 4). ERβcx expression was associated with total ERβ expression but was independent of all other variables.

In the univariate analysis (Tables 1 and 2), SBR grade was found to be the most discriminant variable between the two groups of resistant and sensitive cases (P = 0.001), followed by ERβ expression (P = 0.004), MIB-1 proliferation index (P = 0.02), and tumor size (P = 0.042). ERα expression and nodal status were not significant. In the multivariate analysis, SBR grade (I+II versus III), MIB-1 proliferation index, and ERα and ERβ expression were introduced in a multivariate logistic regression model (25) on a continuous scale (Table 4). Tumor size had no predictive value and was not included in the model. ERβ expression was the most important independent variable (P = 0.001), followed by SBR grade (P = 0.008) and MIB-1 proliferation index (P = 0.016), whereas ERα expression was at the limit of statistical significance (P = 0.060). According to this model, 43 (86%) of the 50 patients were correctly classified. The sensitivity and specificity were 81 and 88%, respectively. The positive and negative predictive values were 76 and 91%, respectively, assuming a prevalence rate of resistance equal to 32% (16 of 50). On the basis of expression status, the logistic regression model identified SBR grade (P = 0.003), followed by ERβ (P = 0.013) and MIB-1 (P = 0.032). ERα level was not significant. In a regrouping of the four variables, grade III tumors with elevated MIB-1 proliferation index and low ERβ level were at a greater risk for tamoxifen-resistance.

In addition to classical prognostic variables associated with aggressive tumors, such as histologic SBR grade and tumor size, the level of ERβ determined by immunohistochemistry in a population of ER-positive tumors treated by tamoxifen was found to be the major variable in predicting tamoxifen sensitivity. ERα had a lower value, and ERβcx had no value. This should clarify the significance of the cytosolic radioligand assay of ER (10) on which most of the clinical studies allowing introduction of this marker to predict breast cancer response to antiestrogen therapy were based (9, 10). According to this pilot study, which should be confirmed prospectively on a larger scale, the assay of ERβ by immunohistochemistry is better than that of ERα in guiding the clinician, at least in HER2/neu-negative tumors. The few studies on the clinical value of ERβ in terms of prediction of response to tamoxifen have been controversial. Our results agree with others reporting an association between ERβ and response to tamoxifen treatment (27, 28). They disagree. however. with the proposal that ERβ overexpression is associated with tamoxifen resistance (29), and that the tamoxifen/ERβ complex increases expression of AP-1-controlled genes involved in cell proliferation (17). Whether ERβ actively protects breast cancer cells against tamoxifen-resistance is unknown. One possible mechanism, however, could be a dominant-negative effect of ERβ after heterodimerization (16) inhibiting the tamoxifen agonist activity of ERα via the AF-1 domain (18, 19).

We have not discriminated between initial and acquired tamoxifen resistance, the median time for relapse being 3 years; some resistant cases could be secondary to the selection of cancer cells stimulated for growth by tamoxifen acting as an agonist via ERα. The four patients who recurred more than 1 year after the 5 years’ therapy were included in the tamoxifen-sensitive group, with the assumption that these breast cancers were initially responsive to tamoxifen. When considering these four patients as tamoxifen resistant, the multivariate analysis gave a similar significance for ERβ expression (P = 0.012).

Among the classical markers of aggressiveness (SBR grade, tumor size, MIB-1 proliferation), only lymph node invasiveness was not associated with tamoxifen resistance. This is in agreement with studies showing that node-positive tumors respond as well as node-negative tumors to tamoxifen therapy (30) and that cancer cells, having migrated to lymph nodes, retain the same antiestrogen responsiveness as the primary tumor.

The fact that ERβcx expression in breast cancer is not predictive of tamoxifen resistance in our study, suggests that the full-length ERβ-1, or another ERβ variant, may be involved in tamoxifen sensitivity. It is not excluded, however, that ERβcx plays a role in the initial tamoxifen resistance as suggested by studies in which tamoxifen responsiveness was evaluated after 3 months of neo-adjuvant therapy (31). The absence of correlation of ERβ with other classic prognostic parameters further supports its interest for breast cancer monitoring. The absence of correlation with PgR disagrees with other studies (29, 32) but was supported by a recent study on 242 breast cancers (33). The reasons for these discrepancies is unknown and could be due to different methods used for quantification and/or different sets of patients.

Our results do not exclude the involvement of other entities able to induce tamoxifen resistance, such as an increased expression of HER-2/neu (34) and an altered expression of coactivator (35) or corepressor (36). However they strongly suggest that the level of ERβ in breast epithelial cancer cells contributes better than the level of ERα in predicting tamoxifen-sensitivity of breast cancer patients. This should stimulate both large-scale clinical studies before entering ERβ assay into clinical practice and basic studies to define the biological significance of the association between ERβ level and tamoxifen responsiveness of breast cancer.

We thank Drs. Philippe Rouanet, Bernard Saint-Aubert, Jean Grenier, François Quenet, and G. Romieu (from the CRLC Val d’Aurelle, Montpellier) for supplying clinical data, and Jean-Yves Cance for preparing the figures.