T-47D cells were cultured in a mixture of Dulbecco's modified Eagle's medium (DMEM) and Ham's F-12 1:1 supplemented with 5% fetal bovine serum (FBS), penicillin, and streptomycin. For experiments using 17β-estradiol, DMEM without phenol red and FBS treated with dextran-coated charcoal were used.

T-47D cells stably transfected with tetracycline (TET)-regulated ERβ expression plasmid were generated in two steps: (a) The cells were first transfected with pTet-tTAk (Invitrogen Corporation, Carlsbad, CA, USA) modified to contain puromycin resistance using lipofectin according to manufacturer's instructions (Invitrogen Corporation). Selection was performed with 0.5 μg/ml of puromycin in the presence of 1 μg/ml of TET. A good clone showing a high level of induction and low basal activity was selected using the pUHC13-3 control plasmid (Invitrogen Corporation). The short form of ERβ encoding 485 amino acids was fused to the flag tag (ERβ485) and cloned into PBI-EGFP (Clontech, Mountain View, CA, USA). (b) This construct was then transfected into the previously described inducible clone together with a neomycin-resistance plasmid. The selection was performed using 500 μg/ml of G418 (Calbiochem, brand of EMD Biosciences, Inc., San Diego, CA, USA).

Cells were added to 150-mm plates at a confluency of 40%; after 1 day, the normal medium was replaced by the medium described above, supplemented with 5% dextran-coated charcoal-treated FBS (DCCFBS). After 24 hours, 10 nM ICI 182780 (fulvestrant) was added to the cultures and incubation proceeded for an additional 48 hours. For expression of ERβ, TET was removed 12 hours before initiation of treatment with 17β-estradiol. At time 0 hours, the medium was changed to 0.5% DCCFBS and 10 nM of 17β-estradiol or the equivalent volume of dimethyl sulfoxide for mock-treated controls was added. RNA was prepared by adding 10 ml of Trizol (Invitrogen Corporation) to each 150-mm plate at different time points after the start of treatment, and RNA was prepared according to the manufacturer's instructions.

Microarrays were generated by spotting the Compugen USA, Inc. (Jamesburg, NJ, USA) 19 K human oligo library, made by Sigma-Genosys (The Woodlands, TX, USA), on poly-L-lysine-coated glass slides. Twenty-five micrograms each of sample total RNA and human universal reference RNA (Stratagene, La Jolla, CA, USA) were labeled with cyanine (Cy) 5-conjugated dUTP and Cy3-conjugated dUTP (PerkinElmer Life and Analytical Sciences, Inc., Waltham, MA, USA), respectively, and hybridized to the arrays using protocols established by the Patrick O. Brown Laboratory (Stanford University, Palo Alto, CA, USA) accessible at 35. Array images and data were obtained and processed using the GenePix4000B scanner and GenePix Pro software (Axon Instruments, Inc., now part of Molecular Devices Corporation, Sunnyvale, CA, USA). A single sample from each time point and treatment condition was subjected to microarray analysis. Differentially expressed genes were determined using a 1.15-fold difference cutoff at multiple time points and were clustered and visualized using the Eisen Cluster and TreeView programs 36. Gene ontology of responsive genes was derived from annotations (2002 version) made by Compugen USA, Inc. Significant enrichment of functional categories was determined by chi statistics by comparing the distribution of genes in a given functional category for all of the genes represented on the microarray (expected distribution) with the distribution of genes in the same functional category in differentially expressed genes (observed distribution). The p value was then calculated based on a two-tailed chi distribution, and a p value of less than 0.05 was considered a significant difference between the observed and expected values. The data are presented as the log2-transformed ratio between the treated/induced samples and the mock treated-uninduced controls for the same time point. The cell line microarray data have been deposited with the Gene Expression Omnibus under accession number GSE7206, and the normalized log2-transformed fold-change data are presented in Additional file 1.

For validation of microarray data, RNA samples from the 30-hour time point were converted to cDNA by means of the PowerScript (Clontech) reverse transcriptase in accordance with the recommended protocol with 1 μg of total RNA. Quantification of transcript levels was performed using SYBR Green chemistry on the ABI PRISM^® 7900 HT Sequence Detection System (Applied Biosystems, Foster City, CA, USA). Polymerase chain reaction (PCR) amplification was carried out in 20 μl of reaction mixture consisting of 50 nM of primer pairs. The following primers were used for PCRs: cell division cycle 2 (CDC2), CCCAAATGGAAACCAGGAAG (forward), CGTTTGGCTGGATCATAGAT (reverse); cell division cycle 6 (CDC6), GGAAACCCGTTTGACAAAGG (forward), CGGGTGGGGTGTAAGAGAAGAATT (reverse); cell division cycle 28 protein kinase regulatory subunit 2 (CKS2), CTGATGTCTGAAGAGGAGTG (forward), GGAAGAGGTCGTCTAAAGAG (reverse); DNA replication helicase 2-like (DNA2L), GGATGGAACTGTTGGTGAAC (forward), ATTTAGTGAGGGCACACACC (reverse); ERβ, TCCATGCGCCTGGCTAAC (forward), CAGATGTTCCATGCCCTTGTTA (reverse); and β actin, ACCCACACTGTGCCCATCTACGAG (forward), TCTCCTTAATGTCACGCACGATTTCC (reverse). 18S rRNA Pre-developed TaqMan^® Assays by Applied Biosystems were used as reference gene.

The original patient material consisted of freshly frozen breast tumors that came from a population-based cohort of 315 women and that represented 65% of all breast tumors resected in Uppsala County, Sweden, from 1 January 1987 to 31 December 1989 3738. The follow-up RNA expression profiling study was approved by the ethical committee at the Karolinska Institute, Stockholm, Sweden. Affymetrix (Santa Clara, CA, USA) oligonucleotide microarrays (U133A&B) were used to determine gene expression profiles in breast tumor samples. After exclusions based on RNA integrity and array quality control, expression profiles of 260 tumors were deemed suitable for further analysis 37. To assess the role of ERβ-regulated cell cycle and DNA replication genes in breast tumor biology, we selected the 69 ER-positive tumors from patients who underwent adjuvant tamoxifen therapy (see Additional file 2). Clinico-pathological characteristics were derived from the patient records and from routine clinical measurements at the time of diagnosis as described elsewhere 37. Histopathological re-examination and grading according to Elston-Ellis were performed by an experienced breast cancer pathologist without prior knowledge of selected therapies and outcomes. Gene expression data are presented as mean-centered log2-transformed normalized intensity for the probe set. The distances between expression profiles were calculated based on Pearson correlation, and the average-linkage method from the Eisen Cluster program was used to cluster genes and tumor samples. All cluster data were generated in the same way and using the same parameters to maintain the consistency of the analysis.

ERβ expression was induced in the T-47Dbeta subline by TET withdrawal (see Materials and methods), and global gene expression profiles in the presence or absence of hormone were determined by human oligonucleotide microarrays (approximately 19,000 genes; Compugen USA, Inc.) at 1, 6, 12, 16, and 30 hours following induction. The preliminary expression data were subjected to a filter for genes that exhibited at least a 1.15-fold differential expression ratio between induced and treated samples and mock-treated (-E2)-uninduced (+TET) controls at each time point and in the same direction for three consecutive time points or in four out of the five time points. Eight hundred ninety-three upregulated genes and 728 downregulated genes met the selection criteria. These parameters were selected based on previously published time-course data and the ability to capture known responsive genes 28. We then examined the expression profiles of responsive genes by hierarchical clustering. Two clusters of genes emerged which may explain the antagonistic activity of ERβ on cell growth. The first cluster contains genes whose expression levels were upregulated by E2 treatment but downregulated following ERβ expression alone or E2 treatment and ERβ expression (Figure 1a). Among the genes found within this cluster, PCNA (proliferating cell nuclear antigen) and CDC6 are both known to be involved in DNA replication and have been shown to be estrogen-responsive genes 2939. The second cluster includes CDC2 and CKS2, two cell cycle regulatory genes that were also downregulated by ERβ expression in the presence and absence of E2 (Figure 1b). The highly specific contrary regulation of a small number of genes, specifically the downregulation of genes involved in cell cycle regulation and DNA replication, presents a plausible mechanism by which ERβ or α/β heterodimers can block E2-induced cell proliferation.

To determine whether specific molecular pathways and processes are affected by ERβ expression, we assessed in an unbiased fashion whether the genes that were perturbed could be assigned to functional categories by gene ontology and whether genes of a specific function in cell proliferation were statistically significantly enriched, as measured by χ² tests, when compared to the frequency of these genes represented on the microarray (gene ontology annotated by Compugen USA, Inc.; see Additional File 3) (Table 1).

We first observed that ERβ in the presence of E2 did not significantly upregulate any specific functional class of genes. However, in the downregulated genes, we observed 1.7-fold (2.6% versus 1.5%) enrichment in the frequency of cell cycle genes (χ² p = 0.0241) and a 2.8-fold (1.4% versus 0.5%) enrichment in DNA replication genes (χ² p = 0.00175). We then further divided the 728 downregulated genes into expression profiles relative to E2 treatment alone into three models of interactions between ERβ and ERα. In the synergistic model, ERβ overexpression enhances E2 response, whereas in the attenuation model, ERβ diminishes the observed E2-regulated gene expression. When genes exhibit opposite E2-responsive regulation following ERβ expression, they are categorized under the antagonistic model of interaction between ERβ and ERα. All genes were assigned to these categories based on a cutoff of greater than (synergistic) or less than (attenuation) the E2 response or if the fold-change had an opposite (+ or -) notation (antagonistic) after log2 transformation. Table 2 shows the enrichment of responsive genes in the four cell proliferation/growth-associated functional categories. We found that cell cycle genes were over-represented (5.6-fold; χ² p = 0.00077) in the group described by the antagonistic model and that DNA replication genes were enriched 3.2-fold (χ² p = 0.0008) in the group that showed attenuation by ERβ of the E2-mediated transcriptional induction. Thus, it appears that ERβ can specifically regulate the proliferative effects of E2 and ERα by reducing the expression levels of components of the cell cycle and DNA replication machinery.

We sought to address whether this in vitro effect of ERβ on proliferation could explain some of the heterogeneity in biological behavior of ERα-positive breast cancer. We posited that higher ERβ levels would be correlated with lower expression of these specific DNA replication and cell cycle genes and, in turn, would be associated with better clinical outcome. To test this hypothesis, we examined ERβ transcript levels in primary breast tumors with the expression profiles of ERβ-regulated cell cycle genes identified in our in vitro analysis and their association with clinical outcome.

To study these in vitro observations in a biologically and clinically relevant context in which ERα is present and the function of estrogen-responsive pathways was targeted by anti-estrogen treatment, we specifically examined microarray expression data from 69 ERα-positive breast tumors from patients belonging to a previously described cohort 3738 from Uppsala, who had received adjuvant tamoxifen therapy following surgery (see Materials and methods). We extracted the expression data of the six cell cycle and eight DNA replication genes (Table 2) that were downregulated by ERβ in the presence of E2 and classified the tumors based on their expression profiles. Two of the eight DNA replication genes were not found on the arrays used for the tumor studies, so only 12 genes in total were available for analysis. We first assessed whether ERβ expression was associated with a decrease in the expression of the ERβ-regulated cell cycle genes. When we examined these genes, we found that for the majority of them (10/12; 83%), there is negative correlation between ERβ transcript levels and the ERβ-responsive genes (Table 3). This is consistent with our in vitro findings that overexpression of ERβ leads to the downregulation of these genes. Four genes, CDC2, DNA2L, CDC6, and CKS2, were significantly (Student t test p < 0.05) associated with ERβ transcript levels in the microarray data. The estrogen-dependent upregulation of these genes and the suppression of their expression by ERβ overexpression in the cell line were further validated by semi-quantitative PCR (Figure 2) in cells treated for 30 hours with E2 to replicate prolonged estrogen exposure of breast tumor cells in vivo.

We then focused our analysis on these four validated ERβ-regulated genes (CDC2, CKS2, CDC6, and DNA2L) with significant correlation with ERβ expression in the tumors as the functional ERβ-responsive gene set for subsequent clustering and associations with clinical parameters (Figure 3). With the expression of the ERβ-responsive gene set, tumor samples were clustered (average linkage) into two major groups with statistically significant association with 10-year relapse (Fisher exact p = 0.0015), death (Fisher exact p = 0.0401), and tumor grade (Fisher exact p = 0.0128) as compared to the distribution of the parameters tested in the study population.

Using disease-free survival (DFS) (Figure 3b) and disease-specific survival (DSS) (Figure 3c) as clinical endpoints, we performed Kaplan-Meier survival analysis on the two patient clusters to ascertain the correlation between the molecular signature and disease outcome. Patients with relatively higher levels of ERβ and lower expression of the gene set transcript (Figure 3a, cluster 1; Figure 3b,c, black curves) have significantly improved outcomes compared to the group with lower levels of ERβ and higher responsive gene set transcript levels (Figure 3a, cluster 2; Figure 3b,c, red curves) for both endpoints, DFS (likelihood-ratio p = 0.0017) and DSS (likelihood-ratio p = 0.0268). As expected, we did not observe any significant association of the ERβ-responsive gene set expression profiles with disease outcome in 37 ERα-negative tumors from the same patient population (data not shown). These results were further validated in an independent dataset of 45 ERα-positive tumors from patients who also underwent endocrine therapy 40. In this independent dataset of British patients, the ERβ-responsive gene set expression profiles also identified patients in high and low ERβ gene set expression groups (Figure 4a) and to poor and good outcome groups for DFS (Figure 4b, likelihood-ratio p = 0.0162) and DSS (Figure 4c, likelihood-ratio p = 0.0192), respectively. Our findings indicate that the observed in vitro activity of ERβ on downstream target genes is directly recapitulated in primary breast tumors and has utility in further demarcating prognostic and therapeutic subgroups in the clinical setting. In addition, these results implicate the selective ERβ transcriptional effects as identified in vitro in defining clinical behavior of human breast tumors.