A divergent role for estrogen receptor-beta in node-positive and node-negative breast cancer classified according to molecular subtypes: an observational prospective study

ER-β was analyzed by IHC in a series of 936 BC patients subjected to breast surgery at the Regina Elena Cancer Institute (Rome, Italy) between 2001 and 2005. ER-β expression was routinely determined at the time of surgical treatment along with other conventional biological factors before any adjuvant therapy was planned. Patients were subjected to modified radical mastectomy or breast-conserving surgery (quadrantectomy). Radiotherapy was offered to all patients treated with quadrantectomy and to patients with lymph node metastases treated with modified radical mastectomy. Follow-up data were obtained from hospital charts and by corresponding with the referring physicians. The clinicopathological characteristics of these patients are summarized in Table 1. Seven hundred sixty-seven of 936 patients with invasive BC with a median follow-up of 50 months (range 1 to 108 months) were analyzed for disease-free survival (DFS). The remaining 169 patients (18%) were excluded from DFS analysis since follow-up for disease recurrence was not available or they were treated with chemotherapy (CHT) before surgery. The subgroup analyzed for DFS included 665 (86.7%) invasive ductal carcinomas, 9 tubular carcinomas (1.2%), 87 invasive lobular carcinomas (11.3%), and 6 medullary carcinomas (0.8%). In this series, 58.1% were ER-β⁺, 69.4% ER-α⁺, 60.6% PgR⁺, and 31.9% HER2⁺ (19.9% score 2+ and 12% score 3+) (data not shown). We also studied ER-β distribution in the four different molecular subtypes: LA (ER-α/PgR⁺ and HER2^-, n = 447), LB (ER-α/PgR⁺ and HER2⁺, n = 166), TN (ER-α/PgR^- and HER2^-, n = 159), and HS (ER-α/PgR^- and HER2⁺, n = 164). Tumors were graded according to Bloom and Richardson and staged according to the Unione Internationale Contre le Cancer tumor-node-metastasis system criteria and histologically classified according to the World Health Organization [10]. The study was reviewed and approved by the ethics committee of the Regina Elena National Cancer Institute, and written informed consent was obtained from all patients.

Formalin-fixed paraffin-embedded breast specimens were cut on SuperFrost Plus slides (Menzel-Gläser, Braunschweig, Germany). Antigen retrieval was performed by microwave at 430 W (1 mM citrate buffer, pH 6.0) for two cycles of 10 minutes each and one of 5 minutes for anti-ER-β monoclonal antibodies (MoAbs) and by thermostatic bath at 96°C (10 mM/L citrate buffer, pH 6) for 40 minutes for ER-α, PgR, HER2, p53, Bcl2, and Ki67. Sections were incubated with the anti-ER-β MoAbs PPG5/10 (ER-β1, dilution 1:15; GeneTex, Inc., Prodotti Gianni, Milan, Italy) and 14C8 (ER-β total, dilution 1:25; AbCam, Valter Occhiena s.r.l., Turin, Italy) overnight at 4°C, with the anti-ER-α MoAb 6F11 (Novocastra, Menarini, Florence, Italy), the anti-PgR MoAb 1A6 (Menarini), the anti-Ki67 MoAb MIB-1 (Dako, Milan, Italy), the anti-p53 MoAb DO7 (Dako), the anti-Bcl2 MoAb 124 (Dako), and the anti-HER2 polyclonal antibody (A0485; Dako) for 30 minutes at room temperature. Positive and negative controls were included for each antibody and in each batch of staining. The immunoreactions were revealed by a streptavidin-biotin-enhanced peroxidase system (Super Sensitive Link-Label IHC Detection System; BioGenex, Space, Milan, Italy) using 3-amino-9-ethylcarbazole (Dako) as chromogenic substrate. ER-β was defined as negative (ER-β^-) when a weak staining reaction was observed in less than 20% of carcinoma cell nuclei and as positive (ER-β⁺) when a moderate/strong staining reaction was observed in 20% to 100% of neoplastic cell nuclei. This cutoff, which was in agreement with Shaaban and colleagues [11] and Gruvberger-Saal and colleagues [3], was generated using the classification and regression tree (C&RT) analysis (see Statistical methods). We introduced three variables (nodal status, ER-β expression, and relapses) into the model. The model indicated that, in node-negative patients, the highest percentage of relapses occurred when ER-β was positive in less than 17.5% of neoplastic cells whereas, in node-positive patients, the highest percentage of relapses occurred when ER-β was positive in more than 12.5% of neoplastic cells (Additional file 1). ER-α, PgR, and p53 were considered positive when greater than 10% of the neoplastic cells showed a distinct nuclear immunoreactivity whereas Ki67, based on the median value of our series, was regarded as high if greater than 15% of the cell nuclei were immunostained. Bcl2 was recorded as positive when tumor cells exhibited a strong homogeneous cytoplasmic immunoreaction in more than 30% of neoplastic cells [12]. HER2 overexpression was determined as defined in the guide of the HercepTest kit (Dako): scores of 0 or 1+ were considered negative, 2+ weak positive, and 3+ strong positive. Evaluation of the immunohistochemical results, blinded to all patient data, was performed independently and in blinded manner by two investigators (M Mottolese and FN).

Correlation among MoAbs anti-ER-β, PPG5/10, and 14C8 was estimated using the kappa test, whereas the correlation between ER-β and the biopathological characteristic variables was tested by the Pearson chi-square test. Multiple correspondence analysis (MCA), a descriptive/exploratory technique designed to analyze simple two-way and multi-way tables, was used to identify prognostic biological profiles. The results provide information that is similar in nature to that produced by factor analysis techniques and make it possible to explore the structure of categorical variables included in the table. The most common kind of this type is the two-way frequency cross-tabulation table [13, 14]. This representation aims to visualize the similarities and/or differences of profiles, simultaneously identifying those dimensions that contain the majority of the data variability. The positions of the points in the MCA graph are informative. Categories plotting close to each other are statistically related and are similar with regard to the pattern of relative frequencies and this association is statistically valuable (Lebart's statistic) when the points are located far from the origin of the graph which represents a mean uninformative profile. ER-β, p53, Ki67, Bcl2, hormonal receptors, or the BC subtypes were the variables of major interest for the purpose of our study. These factors were introduced in each analysis as active variables whereas the pathological factors (tumor size [T], lymph node status [N], and histological grade [G]) were introduced as supplementary variables. MCA provides a graphical representation of the active and supplementary variables projected on a plane formed by axes 1 and 2, which accounted for 67.6% of total variability, reproducing quite a significant percentage of the total chi-square value of the multi-way frequency table. C&RT analysis, a type of decision tree methodology, is a nonparametric statistical procedure that identifies mutually exclusive and exhaustive subgroups of a population whose members share common characteristics that influence the dependent variable of interest. C&RT uses a binary recursive partitioning method that produces a decision tree that identifies subgroups of patients with a higher likelihood of being found positive in a test for a disease state. For analytical purposes, the patients are split into two groups: a CHT-treated group that included anthracycline (AC) and no AC regimens and an HT-treated group. The procedure examines all possible independent or splitting variables and selects the one that produces the most different binary groups compared with the dependent variable according to a predetermined splitting criterion. The parent node, containing the entire sample, branches into two child nodes according to the independent variable. Within each of the two child nodes, the tree-growing methodology continues by assessing each of the remaining independent variables to determine which variable results in the best split according to the chosen criterion. The improvement in prediction was evaluated by the Gini coefficient. At the point where no further split is made, a terminal node is created [15]. For the purpose of our study, DFS was considered as a measure of outcome. DFS was calculated from the date of tumor diagnosis to the date of first recurrence, including contralateral carcinomas, local relapses, or distant metastases (Additional file 2). Patients without recurrence were censored at the time of last follow-up. The DFS curves were estimated by the Kaplan-Meier product-limit method. The log-rank test was used to assess differences between subgroups, and significance was defined as a P value of less than 0.05. A multivariate Cox proportional hazard model was also developed using stepwise regression (forward selection) with predictive variables that were significant in the univariate analyses. The enter limit and remove limit were P = 0.10 and P = 0.15, respectively. The SPSS (version 14.0) statistical program (SPSS Inc., Chicago, IL, USA) was used for analyses.

In our series of 936 BC patients, 56% of the tumors stained positive for ER-β1 as detected by MoAb PPG5/10. These data were also confirmed using the 14C8 MoAb-directed anti-total ER-β (data not shown) [16], which, in our series, showed a very good correlation with PPG5/10 (κ = 0.80, 95% CI 69 to 92, P < 0.0001). Results reported hereafter refer to those obtained with the PPG5/10 MoAb. No significant correlation was observed between ER-β expression and the other parameters analyzed. In contrast, ER-α and PgR were directly related to each other (P < 0.0001) and to Bcl2 (P < 0.0001) and inversely correlated with T, G, p53, HER2, and Ki67 (P < 0.0001), as expected from previous reports [11, 17]. ER-α, but not PgR, expression was also significantly associated (P = 0.04) with negative nodal status. The same analysis, performed in the cohort of 767 patients with known follow-up, gave superimposable results (data not shown). As shown in Figure 1, 70.3% of cases were ER-α⁺ and 55.5% were ER-β⁺. ER-α and ER-β were coexpressed in 39.0% of cases whereas 31.3% of breast carcinomas were ER-α⁺ and ER-β^- and 16.5% were ER-α^- and ER-β⁺. When ER-β expression was analyzed according to the novel BC classification (Figure 2a), we found that the receptor evenly distributes (P = 0.99) among the four molecular subtypes: indeed, ER-β stained positive in 54.7% of LA, 55.4% of LB, 55.3% of TN, and 56.3% of HS. In contrast, the percentage of p53 and Ki67⁺ tumors significantly increased (P < 0.0001) (Figure 2b,c) and the percentage of Bcl2⁺ tumors significantly decreased moving from the LA phenotype to LB, TN, and HS (P < 0.0001) (Figure 2d).

As shown in Figure 3, the complex interrelationships among the biopathological variables considered in our study, either clustered into phenotypic subtypes (Figure 3a) or considered individually (Figure 3b), can best be evaluated by using MCA. Using this analysis, it is possible to visualize the association of biological factors (ER-α, PgR, HER2, Ki67, p53, and Bcl2) with ER-β and study their link with conventional pathological factors (N, G, and T). As shown in Figure 3a, along the first axis, the test demonstrates the contrast between high Ki67/p53⁺ tumors (upper right quadrant) and LA low Ki67/p53^- tumors (lower left quadrant), suggesting that these two groups represent biopathologically and statistically distinct entities as their defining parameters appear close to each other directly, far from the origin, and diagonally opposite. Similarly, the second axis clearly differentiates Bcl2⁺ tumors/LB subtype (upper left quadrant) from Bcl2^- tumors/HER2⁺ and TN (lower right quadrant), indicating that these two groups, far from the origin and diagonally opposite, can be statistically correlated also. ER-β⁺ (55.5%), though close to the origin, is located in the same quadrant (lower right quadrant) as Bcl2^-/HER2 subtype/LB subtype tumors, whereas ER-β^- (44.5%) is located in the opposite upper left quadrant. Figure 3b shows the contrast between ER-α/PgR/Bcl2^- tumors (lower right quadrant) and ER-α/PgR/Bcl2⁺ tumors (upper left quadrant), whereas the second axis clearly differentiates p53⁺/HER2⁺/high Ki67 tumors (upper right quadrant) from p53^-/HER2^-/low Ki67 tumors (lower left quadrant). Consistent with the lack of correlation with other variables and with its even distribution across different molecular subtypes, ER-β⁺ (55.2%) or ER-β^- (44.8%) did not discernibly cosegregate with any of the other biopathological factors considered.

As summarized in Table 2, in the entire series of 728 patients with complete follow-up data, multivariate analysis revealed that tumor size (HR 2.51, CI 1.40 to 4.50, P = 0.002), nodal status (HR 2.44, CI 1.62 to 3.66, P < 0.0001), and Ki67 proliferation index (HR 1.59, CI 1.08 to 2.34, P = 0.02) were independent prognostic variables influencing DFS. However, in the node-positive subgroup, ER-β positivity (HR 1.55, CI 1.93 to 2.57, P = 0.09) emerged as an adverse prognostic factor for DFS along with tumor size (HR 2.51, CI 1.40 to 4.50, P = 0.002). An opposite effect was observed in the node-negative subgroup, in which ER-β negativity (HR 1.76, CI 0.95 to 3.24, P = 0.07) and elevated Ki67 proliferation index (HR 2.16, CI 1.15 to 4.04, P = 0.02) were associated with a poorer disease outcome.