Expression of oestrogen receptor-β in oestrogen receptor-α negative human breast tumours

To analyse the phenotype of breast tumours that express oestrogen receptor-β (ERβ) alone tissue microarrays were used to investigate if ERβ isoforms are associated with specific prognostic markers and gene expression phenotypes in ERα-negative tumours. ERα-negative tumours were positive for ERβ1 in 58% of cases (n=122/210), total ERβ in 60% (n=115/192) and ERβ2/cx in 57% of cases (n=114/199). Oestrogen receptor-β1 and total ERβ were significantly correlated with Ki67 (r=0.28, P<0.0001, n=209; r=0.29, P<0.0001, n=191) and with CK5/6, a marker of the basal phenotype (r=0.20, P=0.0106, n=170; r=0.18, P=0.0223, n=158). ERβ2/cx was strongly associated with p-c-Jun and NF-κBp65 (r=0.53, P<0.0001, n=93; r=0.35, P<0.0001, n=176). This study shows that a range of ERβ isoform expression occurs in ERα-negative breast tumours. While expression of ERβ1, total and ERβ2/cx are correlated, individual forms show associations with certain phenotypes that suggest different roles in subsets of ERα-negative cancers. Based on our in vivo observations, ERβ may have the potential to become a therapeutic target in the specific subcohort of ERα-negative breast cancers.

Oestrogen receptor-a (ERa) is an important biomarker of response to endocrine therapy in breast cancer (Osborne, 1998). However, the definition of ER status in breast cancer is potentially more complex, since there are now two known ERs, ERa and ERb. Oestrogen receptor-b is expressed in both normal and neoplastic human breast tissue Mann et al, 2001;Murphy et al, 2002;Fuqua et al, 2003;Skliris et al, 2003) but its role in either tissue remains unknown. Several isoforms of ERb have been identified, which are either exon deletions or products of alternative splicing which result in proteins that are truncated at the C-terminus and do not bind ligand (Lu et al, 1998;Ogawa et al, 1998;Fuqua et al, 1999;Leygue et al, 1999;Saunders et al, 2002). Thirty percent of breast tumours are classified as ER negative at the time of diagnosis and will be mostly resistant to endocrine therapy (Lapidus et al, 1998;Osborne, 1998). However, the previous assays used for ER measurement favoured the detection of ERa (Harvey et al, 1999;Brouillet et al, 2001) and we now know that some of these tumours express ERb . Considering studies where ERb protein expression was determined, the pooled data sets were used to estimate the frequency of ERb and ERa status in breast cancers . The most frequently occurring tumour type is ERa þ /ERb þ (B60%) with similar frequencies of the other three ER phenotypes (ERa þ / ERbÀ; ERaÀ/ERb þ ; ERaÀ/ERbÀ) at 10 -20% . It is important to note that there are two groups of ERb-positive breast tumours, those with coexpression of ERa and those expressing ERb alone. The former is the most frequent and probably dominates the analysis of most previously reported correlative studies, and hence the positive association of ERb expression generally with good prognosis and good clinical outcome with respect to tamoxifen treatment (Mann et al, 2001;Omoto et al, 2001;Murphy et al, 2002;Iwase et al, 2003;Esslimani-Sahla et al, 2004;Fleming et al, 2004;Hopp et al, 2004;Myers et al, 2004;Nakopoulou et al, 2004). There is little data exploring tumours that express ERb alone. Under the current system of determining ER status, these are classified clinically as ER negative, and currently there are few markers for further subclassifying these ERa-negative cancers. Nevertheless recent data show that some invasive breast cancers expressing the basal cytokeratin CK5/6, may represent one ERa-negative subset, known as the basal epithelial phenotype and show a relatively poor prognosis (Perou et al, 2000;Sorlie et al, 2001Sorlie et al, , 2003Nielsen et al, 2004). In the present study, we have investigated the level and frequency of expression of ERb in ER-negative tumours and its association with the basal phenotype and other established markers of prognosis, such as indicators of signal transduction pathways, proliferative and apoptotic markers.

Tissues
All invasive breast cancers used in the current study were obtained from the Manitoba Breast Tumour Bank (MBTB, Department of Pathology, University of Manitoba) (Watson et al, 1996), which operates with approval from the Faculty of Medicine, University of Manitoba, Research Ethics Board. All samples included in the MBTB are rapidly frozen at À701C immediately after surgical removal. A portion of the frozen tissue from each case is then processed to create matched formalin-fixed paraffin-embedded and frozen tissue blocks.

Clinical -pathological characteristics of the patient cohort
Cases selected for this study were on the basis of (a) minimum patient follow-up of 36 months, (b) invasive components occupying more than 20% of the tumour section, while normal epithelial areas comprised no more than 10% of the epithelial content and (c) ER-negative status as defined by ligand binding analysis (LBA) of p3 fmol mg À1 protein. The criteria for interpretation of the variables were as follows: (a) PR-positive status was defined as 415 fmol mg À1 protein by LBA; (b) grade, (Nottingham system), was assigned to low (scores 3 -5), moderate (scores 6 and 7), or high (scores 8 and 9) categories; (c) tumour size, was assigned either small (p2 cm) or large (42 cm) categories; (d) tumour inflammation was assessed by a scale from 1 to 5 and then assigned to low (scores 1 -3) or high (scores 4 and 5) categories. All patients were treated with surgery and for 29 patients this was the only treatment regimen. The remaining patients received a variety of additional treatments, hormonal therapy (28), chemotherapy (49) or radiotherapy (9) alone, or combination of radiation followed by hormonal therapy (8), hormonal and chemotherapy (16), hormonal and chemotherapy (19) or chemotherapy (46), and for 6 patients the treatment regime was unknown.

Tissue microarrays
The histopathology of all MBTB cases has been assessed and entered into a computerised database to enable selection based on composition of the tissue as well as clinical -pathological parameters. After selection, cases were rereviewed on H&E sections by a breast histopathologist (PHW). Tissue microarrays (TMAs) from a total cohort of 255 ERa negative (ERa -255TMA), primary invasive ductal breast carcinomas were constructed. Briefly, duplicate core tissue samples (0.6 mm diameter), were taken from selected areas of maximum cellularity for each tumour with a tissue arrayer instrument (Beecher Instruments, Silver Spring, MD, USA). Although the TMA consisted of 255 cases of ER-negative tumours as determined by LBA (ER þ 43 fmol mg À1 protein), 39 of these were subsequently found to be ERa þ by immunohistochemistry (IHC) and were excluded from the later analysis.

Immunohistochemical assay
Serial sections (5 mm) of the ERa -255TMA were cut, mounted on Fisherbrand Superfrost/plus slides (Fisher Scientific, USA) and stained using IHC with commercially available specific antibodies (Table 1). Further details of the three specific ERb antibodies are as follows: ERb1 (polyclonal, GC17/385P, Biogenex, CA, USA, raised to peptide containing amino acids 449 -465) at 1 : 100 dilution; total ERb (monoclonal, 14C8, Genetex, TX, USA, raised to peptide containing amino acids 1 -153) at 1 : 100; ERb2/cx (mouse monoclonal, clone 57/3, raised to synthetic peptide derived from the specific C-terminus of hERb2/cx isoform; Serotec, UK) used at 1 : 20. Briefly, sections were dewaxed in two xylene baths (5 min each), taken through a series of alcohols (100, 95, 70%), rehydrated in distilled water and then submitted to heat-induced antigen retrieval for 8 min in the presence of a citrate buffer (CC1 mild/ standard, Ventana Medical Systems, AZ, USA) using an automated tissue immunostainer (Discovery Staining Module, Ventana Medical Systems, AZ, USA). The staining protocol was set to 'Mild and Standard Cell Conditioning' procedure for all antibodies. Primary antibodies were applied for 60 min (except for NF-kBp65 which were applied for 30 min) while secondary antibodies were incubated for 32 min. Initial dilutions quoted above were diluted further 1 : 3 with buffer dispensed onto the slide with the primary antibody. Primary antibodies were omitted for negative controls.
Total ERb IHC was performed manually; sections were microwaved in the presence of 0.01 M citrate buffer, pH 6.0, for 20 min at full power (Danby, ON, Canada, model DMW 1001 W, 800 W maximum output). Sections were blocked and then incubated using an ERb monoclonal antibody (14C8, Genetex, TX, USA) at 1 : 100 dilution in a humidified chamber at 41C overnight, as previously described (Skliris et al, , 2003Fuqua et al, 2003). Following incubation with biotinylated goat antimouse antibody for 60 min at 1 : 200 (Jackson ImmunoResearch Laboratories, PA, USA) and with the Vectastain ABC kit (Vector Laboratories, CA, USA) for 45 min, total ERb protein was visualised with 3,3 0 -diaminobenzidine (DAB, Sigma-Aldrich, ON, Canada). Slides were scored semiquantitatively under a standard light microscope. Images were captured using Polaroid DMC-2 software (version 2.0.1, Polaroid, MA, USA).

Quantification technique and marker selection
The expression of ERb isoforms (full-length-ligand binding ERb1, total ERb and ERb2/cx) and other prognostic markers was assessed using semiquantitative scoring (H-scores). H-scores derive from a semiquantitative assessment of both staining intensity (scale 0 -3) and the percentage of positive cells (0 - 100%), which when multiplied, generates a score ranging from 0 to 300. Tissue microarray staining was evaluated by two authors (GPS, PHW) independently and where discordance was found, cases were re-evaluated together to reach agreement. For the primary categorical analysis, staining and cutoff points to distinguish low from high expression for each marker were as follows: only nuclear staining was evaluated for ERb1, total ERb and ERb2/cx isoforms and since there is no agreement or clinical relevant cutoff IHC-scores for ERb isoforms reported in the literature, several IHC-score cut-points equivalent to absent staining, the 25th percentile and median IHC-score values were tested in statistical analysis. Ki67, caspase-3 (markers of proliferation and apoptosis, respectively) and CK5/6 (a marker of the basal phenotype) were scored as previously described (Perou et al, 2000;Wykoff et al, 2001;Foulkes et al, 2004;El-Rehim et al, 2005). Since NF-kB has been associated previously with more aggressive breast cancer (Biswas et al, 2004) and both NF-kB and AP-1 have been shown to interact differentially with ERa and ERb (Paech et al, 1997;An et al, 1999) we have also assessed the relationship of ERb to these pathways in ERa-negative tumours. For NF-kB/p65 nuclear staining was assessed and multiple H-score cutoffs were tested. P-c-Jun, a marker of AP-1 activity, was defined by nuclear staining and an H-score of 40.

Statistical analysis
Associations between ERb isoforms and other clinical -pathological variables were tested using contingency methods (Fisher's exact test). Correlations were assessed by the Spearman's rank correlation test (r). Mann -Whitney rank sum tests, two-sided were also used to evaluate variables. Survival analyses were perfomed using the log rank test to generate Kaplan -Meier curves. Overall survival was defined as the time from initial surgery to the date of death attributable to breast cancer. Relapsefree survival was defined as the time from initial surgery to the date of clinically documented local or distant disease recurrence or death attributed to breast cancer. GraphPad Prism 4.02 version statistics software (GraphPad, San Diego, CA, USA) was used to perform all analyses.

Validation of ERb antibodies
Three antibodies previously validated to detect ERb related proteins were used in this study (Fuqua et al, 1999;Leav et al, 2001;Saunders et al, 2002). GC17/385P (Leav et al, 2001) was raised to a C-terminal epitope of the wild-type ligand binding isoform of ERb, generally referred to as ERb1. 14C8 antibody (Fuqua et al, 1999) was raised to an N-terminal epitope which would be found in both ERb1 and multiple C-terminal truncated nonligand binding forms of ERb and therefore would detect multiple known ERb isoforms including ERb1 and ERb2cx. Hence we refer to it as detecting 'total' ERb. The antibody used to detect the nonligand isoform ERb2/cx (Saunders et al, 2002) has been previously validated by IHC and immunoblotting (Saunders et al, 2002). However, we have also validated the antibody further at the IHC level, by using MCF7 breast cancer cell lines, which have been engineered to overexpress ERb1 or ERb2/cx, after induction with the tetracycline analogue doxycycline (Murphy et al, 2005). Agar embedded cell pellets (Riera et al, 1999), formalin-fixed and paraffin-embedded (Adeyinka et al, 2002) from only the doxycycline treated cells expressing ERb2/cx but not ERb1 or controls were found to show nuclear staining with the specific ERb2/cx antibody under the same IHC conditions described above for the human breast tumours ( Figure 1A).

ERb isoform expression in ERa-negative human breast tumours
Serial sections of the ERa -255TMA were stained with specific antibodies for ERb1, total ERb, and ERb2/cx using IHC. Nuclear staining could be observed with ERb1 and total ERb antibodies in epithelial cells in our series of invasive cancers ( Figure 2). Strong nuclear staining in both normal and neoplastic breast tissues for ERb2/cx isoform was often observed ( Figure 1B). Using the 25% percentile of IHC-scores to define positive status for ERb1, total ERb and ERb2/cx, we observed that 58% of ERa-negative tumours were positive for ERb1 (n ¼ 122/210), 60% positive for total ERb (n ¼ 115/192) and 57% of cancers were positive for ERb2/cx (n ¼ 114/199; Table 2). ERb1 was significantly correlated with both total ERb and ERb2/ cx (r ¼ 0.28, Po0.0001, n ¼ 189; r ¼ 0.27, P ¼ 0.0002, n ¼ 196, respectively; Table 3). The same relationship was evident in categorical analysis using a variety of cutoff values for contingency analysis, where ERb1 was also significantly associated with ERb2/ cx and total ERb (P ¼ 0.0083, 410; P ¼ 0.0016, 0.0391; 410, 425 respectively, Fishers exact test). Using a cut-point for ERb1 of either 410 or 425, median levels of total ERb expression were significantly higher in ERb1-positive vs -negative tumours (P ¼ 0.0026 and P ¼ 0.011, Mann -Whitney rank sum tests, twosided). Similarly using the same two cut-points for ERb1 positivity median levels of ERb2/cx expression were significantly higher in ERb1-positive vs -negative tumours (P ¼ 0.0024 and P ¼ 0.022, respectively Mann -Whitney rank sum tests). These data suggest frequent coexpression of multiple ERb isoforms in breast tumours.
Relationship of ERb isoform expression with markers of proliferation and apoptosis in ERa-negative human breast tumours ERb1 (r ¼ 0.28, Po0.0001, n ¼ 209) and total ERb (r ¼ 0.29, Po0.0001, n ¼ 191; Table 3) were positively correlated with Ki67, a marker of proliferation, which was detected in the nuclei of ERanegative tumours (Figure 2). Contingency analyses also showed that ERb1 and total ERb were associated with Ki67 (data not shown). Using the median Ki67 IHC-score as a cutoff to define low Ki67 (p25) and high Ki67 (425), the median level of ERb1 expression was significantly lower in low Ki67 expressors (median ERb1 ¼ 25) compared to high Ki67 expressors (median ERb1 ¼ 50; P ¼ 0.0008, Mann -Whitney rank sum test). Similarly the median level of total ERb expression was significantly lower in low Ki67 expressors (median total ERb ¼ 20) compared to high Ki67 expressors (median total ERb ¼ 50; P ¼ 0.0008, Mann -Whitney rank sum test). No significant differences in ERb2/cx were found between the low and high Ki67 groups.
However, high proliferation in primary tumours prior to treatment, is often associated with high levels of apoptosis (Lipponen et al, 1994;Lipponen, 1999;Parton et al, 2002). Therefore, ERb expression was investigated with respect to a marker of apoptosis, active caspase-3 (Parton et al, 2002). No correlations were detected between ERb isoforms and caspase-3. However, Ki67 expression was significantly correlated (r ¼ 0.44, Po0.0001, n ¼ 211, Table 3) and associated (Po0.0001 Fisher's exact test; Mann -Whitney rank sum test) with caspase-3 in this breast tumour cohort. These data suggest that ERb expression in ERa-negative tumours is associated with markers of a high proliferative index.

Relationship of ERb expression to basal epithelial phenotype markers in ERa-negative human breast tumours
Invasive breast cancers expressing the basal epithelial phenotype, based on the consensus of the published literature from cDNA microarray and IHC analyses, are ERa negative (Perou et al, 2000;Sorlie et al, 2001;Vijver et al, 2002;Nielsen et al, 2004;El-Rehim et al, 2005), CK5/6 positive (Sorlie et al, 2001;Korsching et al, 2002;Nielsen et al, 2004;Collett et al, 2005) and/or CK14 (El-Rehim et al, 2005) positive. The basal phenotype has also been associated with mutated BRCA1 (Foulkes et al, 2003(Foulkes et al, , 2004Sorlie et al, 2003;Collett et al, 2005). We were therefore interested to determine the relationship of ERb expression in ERa-negative tumours to markers of the basal epithelial phenotype. ERb1 and total ERb expression were weakly correlated with CK5/6 (r ¼ 0.20, P ¼ 0.010; n ¼ 170; r ¼ 0.18, P ¼ 0.022, n ¼ 158; Table 3). No correlations were seen with ERb2/cx. These data support the conclusion that many ERa-negative tumours expressing ERb are associated with some markers of a basal epithelial phenotype in breast cancer.

ERb2/cx expression in ERa-negative human breast tumours
Despite the correlations and associations of ERb2/cx to ERb1 and total ERb described above, ERb2/cx was not correlated with Ki67 nor activated caspase-3. However, ERb2/cx was strongly correlated with p-c-Jun IHC-score (r ¼ 0.53, Po0.0001, n ¼ 93; Table 3). Contingency analyses for ERb2/cx and p-c-Jun positivity, identified a significant association of ERb2/cx with p-c-Jun (Po0.0001, Fisher's exact test). When p-c-Jun expression level was examined in relation to ERb2/cx status, p-c-jun IHC-score was significantly lower in ERb2/cx-negative tumours (median p-c-Jun ¼ 5) compared to high ERb2/cx expressors (median p-c-Jun ¼ 40; Po0.0001, Mann -Whitney rank sum test). Similarly, ERb2/cx expression was also correlated with NF-kBp65 (r ¼ 0.35, Po0.0001, n ¼ 176; Table 3). Using either the 25% percentile (40) or the median (425) ERb2/cx IHC-score as cutpoints to define negative and positive ERb2/cx status the median level of NF-kBp65 expression was significantly lower in negative/ low ERb2/cx expressors (median NF-kBp65 ¼ 50) compared to high ERb2/cx expressors (median NF-kBp65 ¼ 100; Po0.0001, Mann -Whitney rank sum test). Similar but weaker relationships were found for total ERb. Using the median (425) total ERb IHCscore as a cutoff to define negative and positive total ERb status the median level of NF-kBp65 expression was significantly lower in negative/low total ERb expressors (median NF-kBp65 ¼ 75) compared to high total ERb expressors (median NF-kBp65 ¼ 100; Po0.026, Mann -Whitney rank sum test). These data suggest that ERb2/cx expression is associated with AP1 and NF-kB activity in ERa-negative breast tumours. A relationship between total ERb and p-c-Jun and NF-kBp65 was also demonstrated, and is likely to reflect the influence of the ERb2/cx component of the total ERb signal.

ERb isoform expression in relation to clinical and pathological prognostic variables and survival
Only total ERb was associated with tumour grade (P ¼ 0.03). No other statistically significant associations between ER isoforms and established prognostic variables such as tumour size, age at diagnosis, node status, inflammation or progesterone receptor, were observed (Table 2, showing associations with cut-points equivalent to the 25th percentile).
Univariate survival analyses in relation to axillary nodal status, size, grade, Ki67, active caspase-3, or basal phenotype, showed a significant association only with nodal status (P ¼ 0.024) in this cohort of ERa-negative tumours. Furthermore no difference in disease outcome (overall survival and relapse-free survival) was found between low and high ERb1, total ERb or ERb2/cx ( Figure 3).

DISCUSSION
Several interesting observations have been made in the present study concerning ERb isoform expression in ERa-negative breast tumours. The first is that ERb1, total ERb and ERb2/cx isoforms are frequently expressed in this cohort of ERa-negative breast cancers. The second is that there is a significant correlation of ERb1 and total ERb with Ki67, a marker of proliferation, which is of particular interest. As this was not found when ERb2/cx expression was assessed, it is likely that the correlation with total ERb reflects the ERb1 component, although we cannot exclude the existence of other, as yet unknown variant isoforms. Indeed, the frequent expression of the ERb variant isoform, ERb5, in ERanegative breast tumours has recently been described (Poola et al, 2005), however, we did not have access to specific antibodies to investigate this variant isoform in our breast tumour cohort. Our data confirm and extend an observation made by Jensen et al (2001), where the highest expression of either Ki67 and Cyclin A was found in tumours that only expressed ERb, indicating that ERb may be related to proliferation in breast cancer. Jensen's observation showing an association of ERb with Ki67, using an antibody that recognised total ERb (Jensen et al, 2001), also suggests that ERb isoforms are not only expressed in cells with the potential to cycle but also can be expressed in cells that are cycling. The existence of this relationship was reflected only in a very small subset of seven tumours in the ERa-negative/ERb-positive cohort in his study (Jensen et al, 2001), but a study by O'Neill et al (2004) published during the execution of our study confirmed his observation in a larger cohort (n ¼ 167). However, results from these latter studies came only from subset analysis of mixed cohorts of ERa-positive and -negative tumours. Our study is the only one so far exclusively focusing on ERa-negative cancers to address the issue of ERb expression. The cohort used in our study (n ¼ 216) is the largest so far and included tumours that were all selected to be ERa negative, both immunohistochemically and by LBA. Thus, the relationship of ERb1 alone expression in human breast cancer to Ki67, seems to be highly reproducible and therefore likely offers a new significant insight into the possible role of ERb1 in breast cancer. In contrast, this relationship is generally not seen in ERa-positive/ERb-positive breast tumours (Jarvinen et al, 2000;Mann et al, 2001;Omoto et al, 2001;Murphy et al, 2002;Fuqua et al, 2003;Iwase et al, 2003;Fleming et al, 2004;Hopp et al, 2004;Myers et al, 2004;Nakopoulou et al, 2004) and therefore our data together with two other studies support the conclusion that the role of ERb1 when expressed alone in human breast cancers in vivo is likely quite different to when it is coexpressed with ERa. Such data suggest that ERb1 may have a direct role in proliferation in ERa-negative breast cancers, but this is unproven. The involvement of ERb isoforms in proliferation using cell line models is unclear. Most cell line models in which ERb1 has been stably expressed either inducibly or constitutively show that overexpression of ERb1 inhibits proliferation irrespective of whether it is coexpressed with ERa (Paruthiyil et al, 2004;Strom et al, 2004;Murphy et al, 2005) or not (Lazennec et al, 2001;Cheng et al, 2004). However, two studies using cell line models have been published in which stable constitutive overexpression of ERb1 resulted in increased proliferation (Tonetti et al, 2003;Hou et al, 2004) although in the former publication the short form of ERb1 (truncated by 45 amino acids from the N-terminus) was used. Both breast cancer cell lines used (MDA-MB-231 and MDA-MB-435) are typically ERa negative and therefore can be considered to represent the ERb alone expressing breast tumours cohort in vivo. However, in another constitutive ERb overexpression model based on the MDA-MB-231 cells, little or no effect on proliferation, positive or negative, was seen (Rousseau et al, 2004). Such data indicate that differences in potential cell line background, the type of ERb isoforms expressed and experimental variables including possibly clonal selection can influence the effect of ERb on proliferation. However, in other cancer cells types where ERb1 has been overexpressed, increased ERb1 is most often associated with inhibition of proliferation and/or increased apoptosis (Qiu et al, 2002;Cheng et al, 2004). It is unclear, however, whether the overexpression of ERb1 in experimental cancer cell line models, is relevant to the levels of ERb1 seen in tumours in vivo, especially since generally ERb1 expression is reduced in tumours compared to normal tissues in multiple cancers (Foley et al, 2000;Roger et al, 2001;Skliris et al, 2003) leading to the suggestion that ERb1 is a tumour-suppressor gene, and certainly would be consistent with the hypothesis that it is antiproliferative (Weihua et al, 2000;Forster et al, 2002;Paruthiyil et al, 2004). As well the possibility exists that ERb1 may be frequently mutated and/or altered posttranslationally in breast cancers in vivo, although no published data as yet address this issue to our knowledge.
ERb1 and total ERb isoforms were also significantly correlated with CK5/6, a marker of the basal epithelial phenotype as defined from DNA microarray and IHC analyses, predominantly as ERa negative and CK5/6 positive (Sorlie et al, 2001;Korsching et al, 2002;Collett et al, 2005;El-Rehim et al, 2005). As ERb is found widely expressed in the basal myoepithelium Speirs et al, 2002) as well as in luminal epithelial cells in normal human breast tissues, it is possible that many ER-negative breast cancers expressing ERb are derived from a myoepithelial cell lineage, and that ERb is a marker of this lineage. Interestingly, a reduced myoepithelial cell layer is found in the lactating mammary gland of the ERb knockout mouse in contrast to the wild-type controls (Forster et al, 2002). This led to the hypothesis that ERb may be involved in regulating pathways, which are required for the differentiation of the myoepithelial cell lineage in the mammary gland (Forster et al, 2002).
While proliferation and the basal phenotype have been associated with poor survival, no differences in clinical outcome were identified between high and low Ki67 or any markers of the basal phenotype in our ERa-negative breast cancer cohort. It is possible that the lack of association of any of parameters investigated here (ERb isoforms, Ki67 and caspase-3) with clinical outcome (disease-free survival and overall survival) is confounded by the variety of treatments the patient cohort later received. In addition, most other studies where Ki67 has been examined as a prognostic factor have included both ER-positive and ER-negative tumours in their cohorts (Trihia et al, 2003). It should also be noted that ERa-negative status in our cohort was defined by negative IHC and ligand binding assay. This definition eliminated 15% of an initial ERa-negative series selected only on the basis of ligand binding assay. A similar number of ERa IHC-negative tumours have been found to be positive by ligand binding assay (Huang et al, 1997). The basis for discrepancy between these two ERa assays has been a subject of past discussion in the literature (Huang et al, 1997), but is likely to reflect biological variables rather than tissue selection or composition, because of the design of our tumour bank. Therefore, the current study used stringently defined ERa-negative tumours and so was enriched for a generally more aggressive group of breast tumours. In comparison to ERb1, the role of its variant, ERb2/cx, is even more unclear. Transient expression studies using human ERb2/cx, have shown that human ERb2/cx is unable to bind ligand and when overexpressed sufficiently can inhibit ERa transcriptional activity (Ogawa et al, 1998;Peng et al, 2003) but has little if any effect on ERb1 activity. In breast cancer ERb2/cx has been identified at both the RNA and protein levels (Saji et al, 2002;Esslimani-Sahla et al, 2004), and now with another antibody we have also shown the presence of ERb2/cx in both normal and neoplastic breast tissue. Most studies previously published  Figure 3 Kaplan -Meier graphs for 'overall survival' and 'relapse-free survival-time to progression' with respect to expression of ERb1 (A and B), ERb2cx (C and D) and total ERb isoforms (E and F, respectively). ERb1 overall survival (A), n ¼ 210, low ERb1 events ¼ 47, high ERb1 events ¼ 60. ERb1 time to progression (B), low ERb1 events ¼ 48, high ERb1 events ¼ 60. ERb2cx overall survival (C), n ¼ 199, low ERb2cx events ¼ 44, high ERb2cx events ¼ 53. ERb2cx time to progression (D), low ERb2cx events ¼ 44, high ERb2cx events ¼ 53. Total ERb overall survival (E), n ¼ 192, low total ERb events ¼ 40, high total ERb events ¼ 55. Total ERb time to progression, low total ERb events ¼ 40, high total ERb events ¼ 56.
suggested that ERb2/cx is increased in breast tumours compared to normal breast tissue Palmieri et al, 2004) and the relative expression of the ERb2/cx to ERb1 is likely to change during breast tumourigenesis. However, no studies focusing only on ERa-negative tumours have been published. A few studies have suggested hypotheses as to ERb2/cx function due to observed correlations and association with other prognostic markers and clinical outcome with or without treatment Esslimani-Sahla et al, 2004;Palmieri et al, 2004). Esslimani-Sahla et al (2004) showed that ERb2/cx expression was correlated with total ERb, which is in agreement with our observation in our ERa-negative series. However, among these studies contradictory conclusions have often been reached (Saji et al, 2002;Esslimani-Sahla et al, 2004;Palmieri et al, 2004). Our data suggest that in ERa-negative tumours, ERb2/cx expression is significantly associated with both increased AP-1 and NF-kB expression and that ERb1 may not be associated with these activities. This suggests that the different ERb isoforms may be involved in regulation of distinct pathways in these tumours or alternatively there is differential regulation of ERb isoforms by distinct pathways in these tumours.
The absence of any significant correlations between ERb isoforms and particularly total ERb with either overall or relapse-free survival is also in agreement with some other published studies (Hopp et al, 2004) but disagrees with other studies where increased ERb has been associated with better survival (Nakopoulou et al, 2004) and when patients were treated with tamoxifen alone, where an association was shown with better response to tamoxifen therapy (Murphy and Watson, 2006). However, in these latter studies the majority if not all the tumours assessed were ERa positive and so represent a different context where ERb is coexpressed with ERa. In the current study we have hypothesised that the function of ERb expressed alone will be different to that when ERb is coexpressed with ERa, and therefore we have looked at a distinct cohort of patients where their tumours are ERa negative.
These data support the hypothesis that the role of ERb expression is different when expressed alone, to its role when coexpressed with ERa in human breast cancer. This is specifically reflected in the present study, by the confirmation of a strong relationship of ERb1 with Ki67 in ERa-negative tumours, such that it seems likely that the addition of an ERb1 antagonist could be a potentially useful therapy in specific subsets of breast cancer patients in a clinical setting.