Previous assessments of the role of ERβ2 in breast cancer treatment outcome have been limited, with most clinical studies being performed in broader groups of patients and focussing on other associations, largely related to pathology. Our own previous data [17
] was based on a semi-quantitative RTPCR analysis using an assay in which ERβ5 is co-amplified with ERβ2 and distinguished based on size of the PCR product, similar to the triple-primer assay used elsewhere [22
]. We found that, using the arbitrary cut-off imposed by detection sensitivity, ERβ2 mRNA expression was more closely associated with survival benefit than ERβ1 or ERβ5 mRNA expression. We therefore set out to establish whether ERβ2 protein levels similarly predicted patient outcome. We defined discriminatory cut-points of ERβ2 levels in a non-arbitrary manner, using ROC analysis, and used these to assess the relationship between ERβ2 expression and outcome.
Immunohistochemical staining for ERβ2
A cohort of 141 cases were stained by immunohistochemistry for ERβ2 (Table ) including 98 ERα + cases. ERβ2 staining was assessed by 2 observers (R.V., V.A.) using the Allred scoring system and also as percentage positive cells (%+), with good agreement between observers (Allred Spearman 0.91 P = 1.0 × 106, %+ Pearson 0.92 P = 3.4 × 1059). At the cut-point used for outcome analysis the Kappa score was 0.87. A consensus score was produced and used herein, representative examples of immunostaining are shown in Figure . The frequencies of each score were: score 0, 2 cases (1.4%); 3, 3 cases (2.1%); 4, 9 cases (6.4%); 5, 27 cases (19.1%); 6, 39 cases (27.7%); 7, 61 cases (43.3%).
Figure 1 Immunohistochemical staining for ERβ2. Breast carcinomas showing different levels of staining; examples of Allred score 0 (A), 3 (B), 4 (C), 5 (D), 6 (E), 7 (F) and 8 (G). H-K are low (H, J) and high (I, K) magnification images of the same tumour (more ...)
ERβ2 immunostaining significantly correlated with that for ERα (%+ Pearson 0.42 P = 7.8 × 107
, Allred Spearman 0.40 P = 4.1 × 106
) and to a lesser extent PgR (%+ Pearson 0.18 P = 0.035). ERβ2 immunostaining was greater in ERα + cases (mean %+ = 69) than in ERα- cases (mean %+ = 52; P = 0.00001 T-test) and ERβ2 Allred score was greater in PgR+ cases than PgR- cases (P = 0.033 MW). The percentage of ERβ2 positive cells were somewhat lower in grade 3 cases (P = 0.042 MW), in keeping with the association with ERα status. There was no association with Ki67 staining, vascular invasion, nodal status, age or size, or with ERβ1-specific immunostaining [9
]; most previous studies have similarly failed to show clear links to many clinical and pathological parameters.
The association seen here between ERβ2 and ERα has not always been seen by others. Although case selection and clinical setting may have some bearing on this, it is also possible that such correlations are due to better tissue preservation of antigens in some blocks of tissue. We do not think that this is the case here, as in the same cohort ERα but not ERβ2 inversely correlated with p53 immunostaining (unpublished data) and ERα did not correlate with ERβ1 [9
]. If antigen preservation was a major influence on immunostaining patterns it is unlikely that such complex inter-relationships would be evident.
Association of ERβ2 protein with patient survival
Using the Allred scoring system, tumours were designated as either ERβ2 low (score 5 or lower, n = 39) or ERβ2 high (score 6 or higher, n = 97, 71%). ERβ2 status significantly associated with ERα status (P = 0.001 Chi square) and within the subgroup of ERα positive women who received adjuvant tamoxifen there were 18 ERβ2 low cases and 67 ERβ2 high cases (79%).
Within the group as a whole (ERα + and ERα- cases), high ERβ2 protein levels were significantly related to a better relapse free survival (BCR P = 0.049 Log Rank, Figure ), but not breast cancer survival (BCS P = 0.16, Figure ). However, in both cases the survival curves converge at later time-points; with shorter follow-up time a stronger relationship with outcome was seen (5-year BCR P = 0.018 Log Rank, HR 0.50 CI 0.27–0.90 P = 0.020; 7 year BCS P = 0.048 Log Rank, HR 0.50 CI 0.27–0.90 P = 0.020).
Figure 2 Kaplan Meier plots for breast cancer relapse. Plots are shown for dichotomised levels of ERβ2 immunostaining in the whole cohort (A, 34 events in 97 ERβ2 high cases and 21 events in 40 ERβ2 low cases) and dichotomised levels of (more ...)
Figure 3 Kaplan Meier plots for breast cancer survival. Plots are shown for dichotomised levels of ERβ2 immunostaining in the whole cohort (A, 27 events in 91 ERβ2 high cases and 17 events in 38 ERβ2 low cases) and dichotomised levels of (more ...)
When ERα status and ERβ2 immunoscore were included as the only two variables in multivariate analysis of 5 year BCR, ERα status was independently significant (HR 0.38 CI 0.22–0.66 P = 0.001) whereas ERβ2 did not retain independent significance (HR 0.76 CI 0.43–1.33 P = 0.33). When ERβ2 immunoscore, ERα, grade, size and nodal status were included in this multivariate analysis, only nodal status (HR 3.1 CI 1.5–6.2 P = 0.001) and grade (HR 1.5 CI 1.1–2.2 P = 0.026) were independently significant. When considering only ERα +, tamoxifen-treated cases there was no relationship between ERβ2 immunostaining at outcome (BCR P = 0.95, BCS P = 0.65 Log Rank).
One previous study of only 50 ERα positive cases using immunostaining with a different antibody raised to the same ERβ2-specific epitope [7
] similarly failed to show any predictive association with adjuvant tamoxifen treatment. However this analysis was based on detecting differences in staining between "sensitive" and "resistant" cases using the crude measure of relapse within 5 years of tamoxifen therapy. Unpublished observations [12
] also failed to show any predictive value in an adjuvant setting and a similar lack of association between ERβ2 immunostaining and outcome has recently been demonstrated in the neoadjuvant setting [4
]. Hence the early outcome benefit seen with strong ERβ2 immunostaining was not identified previously. However, an association of ERβ2 protein with a favourable outcome has been seen in a metastatic and locally advanced setting [10
]. In this case, not only was the clinical setting different, but ERβ2 was assessed by western blot. Therefore the present study is the largest to date to assess immunostaining of ERβ2 as a predictive marker of outcome in the postmenopausal, adjuvant endocrine setting. Results indicate that ERβ2 protein levels did not apparently relate closely to outcome for ERα + cases. Rather there was some association of ERβ2 immunostaining with better outcome in broader cohorts of patients (including ERα- cases), due in part to a correlation between ERα and ERβ2 protein levels.
Association of ERβ2 mRNA with patient survival
ERβ2 immunostaining results are at odds with previous semi-quantitative RTPCR results. We therefore performed a repeat RTPCR analysis on a larger series of patients, but with fully quantitative RTPCR using independent cDNA synthesis reactions and different splice variant specific PCR conditions. A subgroup of 100 cases (Table ) with suitable quality mRNA available were used in qRTPCR for ERβ2, ERα and control genes HPRT and GAPDH. Expression of ERβ2 mRNA (mean 0.006 attomoles per μg total RNA) was significantly lower (P < 106 paired T-test) than that of ERα (mean 25 attomoles per μg total RNA). These low levels of ERβ mRNA (also seen with ERβ1 and ERβ5, results not shown) may contribute to technical difficulties in reproducibly measuring ERβ variants and hence to the lack of consistency between different studies.
In the 100 case (ERα + and ERα -) qRTPCR cohort (Table ), high grade (BCR & BCS P ≤ 0.001), positive nodal status (BCR & BCS P ≤ 0.0005), larger size (BCS P = 0.042), ERα negative status (BCR P = 0.009, BCS P = 0.041) and PgR negative status (BCR P = 0.032, BCS P = 0.026) were all associated with poor outcome (Log Rank). Using an ROC-derived optimal cut-point (0.0040 attomoles per μg total RNA) for this 100 case cohort in Kaplan Meier Log Rank analysis, there was a significant association between higher ERβ2 mRNA expression and good outcome (BCR P = 0.046 Log Rank, HR 0.51 CI 0.26–1.00 P = 0.0496). As with ERβ2 immunoscore, this association was stronger at 5 years (5-year BCR P = 0.016 Log Rank, HR 0.39 CI 0.18–0.87 P = 0.020). Notably, unlike ERβ2 immunostaining, no significant association was found between ERβ2 mRNA and ERα immunostaining. Also unlike ERβ2 protein, ERβ2 mRNA and ERα status were independently associated with BCR at 5 years (ERβ2 HR 0.43 CI 0.20–0.95 P = 0.036, ERα HR 0.36 CI 0.18–0.75 P = 0.006). These two measures of ERβ2 expression therefore seem to behave differently in relation to ERα status and treatment outcome.
Further outcome analysis was limited to ERα positive women who received adjuvant tamoxifen and had a defined breast cancer related outcome (n = 62 BCR, n = 58 BCS). High grade (BCR P = 0.006, BCS P = 0.0008) and positive nodal status (BCR P = 0.003, BCS P = 0.007) maintained their association with worse outcome (Log Rank). ROC plots for BCR and BCS at 5 years indicated a significant relationship between good outcome and high qRTPCR values for ERβ2 (BCR area under curve 0.68 CI 0.52–0.84, P = 0.036) and the optimal cut-point was 0.0039 attomoles per μg total RNA. There were significant associations between outcome and ERβ2 mRNA level using the ROC-derived cut-point (Figures and ). High ERβ2 mRNA was significantly associated with better outcome (BCR P = 0.0095 Log Rank, HR 0.32 CI 0.13–0.79; BCS P = 0.011 Log Rank, HR 0.25 CI 0.08–0.79). The 5-year cumulative relapse-free population was 81% in the ERβ2-high group (standard error 8%), compared to 55% in the ERβ2-low group (standard error 10%); the 5-year cumulative BCS was 89% in the ERβ2-high group (standard error 6%), compared to 62% in the ERβ2-low group (standard error 10%).
In Cox multivariate analysis of the ERα + tamoxifen-treated cohort including grade, size, nodal status and PgR status, high ERβ2 mRNA had independent significance for good outcome: for BCR ERβ2 (HR 0.31 CI 0.11–0.86, P = 0.024) and nodal status (HR 3.7 CI 1.2–11.5, P = 0.022) were independently significant; for BCS ERβ2 (HR 0.17 CI 0.05–0.65, P = 0.0095) and grade (HR 1.8 CI 1.03–3.3, P = 0.041) were independently significant. Notably there was no significant association between ERβ2 and grade, size, nodal status or PgR status in this treatment-specific cohort (all P > 0.35 Chi-square). In ERα +, node negative cases (n = 33), using a lower cut-off (0.00185 attomoles per μg total RNA), ERβ2 was significantly associated with better outcome (BCR P = 0.0005, BCS P < 0.00005 Log Rank); the 5 year cumulative relapse-free population was 96% in the ERβ2-high group (standard error 4%), compared to 39% in the ERβ2-low group (standard error 24%).
Our results indicate that ERβ2 isoform mRNAs may be an independent marker for ERα + cases that respond well to adjuvant tamoxifen treatment. In node negative cases, where the need for additional markers of response is greatest, our study shows that low ERβ2 mRNA levels are significantly related to worse outcome; as the cases in this subgroup analysis was small, a larger study of node negative patients is warranted. The fully quantitative nature of the qRTPCR results allows comparison of mRNA levels between different ER isoforms and of variant levels between tumours, but necessitated selection of optimal cut-points (in this case using ROC analysis) for the dichotomization required for standard outcome analysis. It should be noted that, whilst such dichotomization is useful in demonstrating associations with outcome, true utility of ERβ variant mRNA measurement will only be demonstrated with larger patient cohorts and may be better achieved by treating mRNA quantitation as a continuous variable, as in other RTPCR based outcome predictors [24
Association of staining for ERβ2 protein with mRNA expression
Associations between high levels of ERβ2 protein (immunoscore) or mRNA (qRTPCR) and improved outcome have been seen, but only the qRTPCR results are significant in the clinically relevant ERα + cohort. It is therefore important to establish the relationship between mRNA and protein levels in clinical samples. Notably, most previous RTPCR-based analyses have failed to take into account the possible translational control when assigning biological or clinical relevance to ERβ isoform expression.
When assessing the relationship between immunostaining and qRTPCR for paired samples from each case, no correlation was seen between levels of protein and mRNA for ERβ2 [Pearson (%+) -0.12 P = 0.24; and Spearman (Allred) -0.08 P = 0.40]. This is in contrast to ERα in the same cohort [Pearson (%+) 0.30 P = 0.003; Spearman (Allred) 0.50 P = 1.0 × 10-6
], but a similar lack of correlation was seen previously for ERβ1 [9
]. Due to tissue heterogeneity, any mRNA analysis of tissue homogenates without selection can contribute to discordance with immunostaining results that are scored on specific cell types. In order to minimise the impact of such artefacts, we selected cases for mRNA analysis that had high proportions of tumour cells (see Methods). It is known that lymphocytes express ERβ2 mRNA, but when 14 cases with inflammatory infiltrates were excluded there was still no significant correlation between ERβ2 mRNA and protein expression. A major factor in the discordance is that many cases express high levels of protein, but low mRNA levels; a situation that is not likely to arise from expression of mRNA in non-tumour cells. It is however possible that heterogeneity of expression in the different parts of the tumour specimen used for mRNA and protein analysis contributes to the lack of correlation and in situ
analysis of mRNA and protein in adjacent sections might address this.
High ERβ2 protein levels and high ERβ2 mRNA levels, when entered into a Cox multivariate model, were independently associated with better relapse-free survival in the whole cohort (ERβ2 protein HR 0.40 CI 0.20–0.80 P = 0.010, ERβ2 mRNA HR 0.43 CI 0.22–0.83 P = 0.013). In multivariate analysis of mRNA and protein in the ERα + tamoxifen-treated cohort, only high ERβ2 mRNA levels were significantly associated with lower BCR (HR 0.28 CI 0.212–0.72 P = 0.008), but a trend remained for protein (HR 0.42 CI 0.15–1.19 P = 0.10). Similar results were obtained for analysis of BCS. This indicates that both mRNA and protein levels may contribute to the relationship of ERβ2 with improved outcome.
Using the cut-points optimised for outcome analysis, the majority of cases (69%) with high ERβ2 mRNA levels also had high levels of ERβ2 protein. However, only a minority of cases (44%) with high ERβ2 protein were also classified as having high ERβ2 mRNA. Thus ERβ2 mRNA expression is frequently associated with expression of significant levels of ERβ2 protein, but ERβ2 protein expression is often dissociated from mRNA expression. Hence, there is a subset of cases (34%) with concomitant high ERβ2 mRNA and protein and another subset of cases (44%) in which high protein levels are not accompanied by high mRNA levels. The cases with both high ERβ2 protein and mRNA had a significantly better outcome than those with low levels of either mRNA or protein or both (P = 0.011 Log Rank, Figure ). When cases with high ERβ2 protein and RNA were compared a group consisting of all other cases they had significantly better outcome: in the whole cohort of ERα + and ERα- cases, (BCR P = 0.002 Log Rank, HR 0.67 CI 0.51–0.88 P = 0.004; BCS P = 0.003 Log Rank, HR 0.61 CI 0.43–0.87 P = 0.006) and for ERα + tamoxifen-treated cases (BCR P = 0.004 Log Rank, HR 0.61 CI 0.43–0.88 P = 0.009; BCS P = 0.009 Log Rank, HR 0.56 CI 0.34–0.91 P = 0.020). The outcome benefit of concomitant high ERβ2 mRNA and protein levels was particularly marked at shorter follow-up, where this measure was the only independent marker of improved outcome in the ERα + tamoxifen-treated cohort using Cox multivariate analysis including grade, size PgR status and nodal status (5-year BCR HR 0.48 CI 0.24–0.95 P = 0.036, 7 year BCS HR 0.46 CI 0.23–0.92 P = 0.029). In the ERα + tamoxifen-treated, node negative cases, having both high ERβ2 mRNA and protein was significantly related to an improved BCS (P = 0.028 Log Rank).
Although ERβ2 protein levels are apparently not directly related to mRNA levels, expression of ERβ2 protein may be important because good outcome was observed for those cases assessed as having both high mRNA and protein levels and this was independent in multivariate analysis. It is possible therefore that the relatively poor utility of ERβ2 protein assessment by immunostaining as a measure of outcome prediction may be due to high levels of ERβ2 protein in some cases (with lower levels of ERβ2 mRNA) being related to some form of protein stabilization, or detection of inactive ERβ2. The disparity between protein and RNA expression for ERβ2 is even suggestive of an inverse relationship. Nevertheless a significant proportion of cancers (34%) had both high protein and high mRNA levels and these had a significantly better outcome than the remaining cases. This suggests that transcription of ERβ2 mRNA drives ERβ2 protein levels in some cases, and these cases do particularly well on tamoxifen treatment. It is perhaps unsurprising that previous studies of ERβ2 protein expression did not find significant associations between ERβ2 and outcome in ERα + tamoxifen treated cases as these did not include measurement of ERβ2 mRNA levels. They were thus unable to distinguish between ERB2 protein associated with increased transcription and that possibly present due to some form of post-transcriptional control (or perhaps the breakdown of normal control).