Gene expression studies using DNA microarrays have identified several distinct breast cancer subtypes which differ significantly in prognosis 12. These subtypes include three main subtypes of estrogen receptor (ER) negative tumors – the basal-like (ER-/HER2-), the ER-/HER2+ subtype and the normal-like/unclassified subtype, and at least 2 types of ER+ tumors (luminal A and luminal B) 2. Basal-like tumors typically show high expression of genes characteristic of the basal epithelial cells of the normal mammary gland, including stratified epithelial cytokeratins, such as cytokeratins 5, 14, 15 and 17 3. Basal-like breast cancers account for around 15% of all invasive ductal breast cancers of no special type 4 with a higher prevalence among African-American women 5. Conventional histopathological as well as molecular studies of breast cancers with "basaloid" differentiation have shown that basal-like tumors are often high grade 6, have areas of necrosis 7, may have a typical or an atypical medullary phenotype 8 and have a distinct pattern of genetic alterations 6, including frequent TP53 mutations 2. A high proportion of BRCA1 tumors exhibit the basal-like phenotype and germ-line BRCA1 mutations result in breast cancers that are more likely to be basal-like in nature 9101112.

The breast cancer subtypes have been extensively characterized by gene expression analysis using DNA microarrays and while this remains the gold-standard, it is not currently feasible for large-scale clinical applications or retrospective studies using formalin fixed, paraffin-embedded samples. In these situations the immunohistochemical staining profile (IHC) can be a useful surrogate of gene expression analysis. However the optimum immunohistochemical profile of basal-like breast cancer remains unclear. The "triple negative phenotype", TNP (ER-, PR-, HER-) is increasingly used as a surrogate marker for basal-like breast cancer as it has the advantage that these three stains are already used routinely in clinical work-up of breast cancers 13. Although most basal-like tumors do not express ER, PR, and HER2, some may, and the overlap between basal-like and TNP breast cancer is not complete 14. Moreover, the ER-, HER2-, CK5/6+ and/or EGFR+ profile seems to correlate better with basal-like breast cancer gene expression profiles 315. We sought to clarify how the utilization of different marker combinations affects prognostic outcome.

We have previously shown that tumors expressing the CK5/6 marker are associated with germline BRCA1 mutations based on data on unselected breast cancer cases from a single institution (Jewish General Hospital, JGH) 9. Here we present data on both the JGH series (n = 192) and a second series of 264 breast cancer cases from the Vancouver General Hospital (VGH), focusing specifically correlations between IHC profiles and outcome in basal-like cancer.

At 10 years time, there was no overall difference in survival for all breast cancer types between the JGH and VGH centers (p = .17) and TNP tumors made up 14% of cases in both series (JGH, 27 tumors, VGH, 36 tumors). In the combined series, the median age at diagnosis was 9.4 years younger in the TNP group (p = .0006) and the median length of follow up in survivors was 6.75 years in the TNP groups versus 9.09 years in the non-TNP group, p = .02. Exclusion of subjects who did not die of breast cancer and were lost to follow up before 10 years did not alter the statistical significance of the study results. There was a significant overlap between the TNP and CBP groups with 49/58 (84%) of TNP cases also falling in the CBP group. Comparison of clinical features in TNP and non-TNP cases in the combined series (Table 1) showed that TNP cases had an increased likelihood of a higher histological grade (odds ratio (OR), for grade 3: 17.7 [95% confidence interval, C.I., 6.05–51.5], p < .0001), a larger tumor (OR for tumor >2 cm: 1.85 [95% C.I. 1.04–3.32], p = .04) but had a decreased likelihood of positive lymph nodes (OR = 0.44 [95% C.I. 0.23–0.84], p = .01). While there was a clear correlation between tumor size and the mean number of positive lymph nodes in both the non-TNP and the TNP group, this correlation was less strong with the TNP group (P = 0.01) and the interaction between tumor size and TNP status on lymph node status was of borderline significance (p = 0.10, Figure 1). Breast cancer survival at 3 and 10 years correlated closely with histological grade, size and lymph node involvement (Table 2). The effect of TNP on prognosis was stronger at 3 years than at 10 years, with TNP conferring a univariate RR of 4.06 [95% C.I. 2.11–7.82], p = .0001 at 3 years (Table 3) compared to 1.71 [95% C.I. 1.05–2.78], p = .03 at 10 years (Table 4). Although there is a degree of overlap between the confidence intervals at 3 years and at 10 years, this is small and the fact that the TNP parameter is not present in parsimonious model 1 at 10 years (Table 4) but is present in the same model at 3 years provides further evidence indicating that the differences are real. A similar pattern was seen with the CBP variable. Predictably, TNP cases were less likely to receive hormone therapy and more likely to receive chemotherapy (Table 1). At 10 years, survival was 63% in TNP cases treated with chemotherapy versus 66% in the no treatment group; the corresponding figures for CBP were 68% and 62%. These differences were not significant and there was also no difference in survival with adjuvant hormone therapy.

In the combined JGH/VGH series, the difference in survival between the TNP and non-TNP groups (Figure 2) was most marked at 3 years with an absolute reduction of 16.7% in the TNP group (76.8% versus 93.5%, p < .0001). Although the absolute reduction in survival of 9.2% at 10 years in the TNP group was still significant (p = .03), the difference appeared to be reducing with time, suggesting that long term survivors in the TNP group may have a comparable survival to non-TNP cases. When using CBP instead of TNP a similar overall survival pattern emerged with a significant difference at 3 years (77.4% versus 93.4%, p =< .0001) that also became less marked at 10 years.

However, when tumors negative for CK5/6 and EGFR expression were compared to tumors that expressed either CK5/6 or EGFR (Figure 3) the absolute survival difference was notably greater at 10 years (17.1%, p = .0007), than that at 3 years (7.8%, p = .02). This was reflected in the multivariate parsimonious models (Table 3 and 4) which showed that at 3 years time, both TNP and CBP parameters not only remained in their respective parsimonious models but both also worked well in predicting outcome (models 1 and 2). As both models share ER negative and HER2 negative status, these appear to be the main driving factors predicting early outcome. Indeed, when all components of TNP and CBP are analyzed separately (model 3), only ER negative and HER2 negative status remained in the parsimonious model while positive PR status and [CK5/6+ and/or EGFR+] status fell out of the model.

In contrast, the data at 10 years indicates that ER negative and HER negative status diminishes in influence with increasing time, with [CK5/6+ and/or EGFR+] status becoming the main driving factor. Therefore CBP (which incorporates CK5/6 and/or EGFR+) may be a better model at 10 years.

The data presented here show that different immunohistochemical marker combinations may influence prognosis at different points in time. This is in agreement with the recent findings of Anderson et al. who using the SEER database found that at 17 months, ER- hazard rates peaked at 7.5% per year then declined, whereas ER+ hazard rates were comparatively constant at 1.5–2% per year; falling ER- and constant ER+ hazard rates crossed at 7 years after which time prognosis was better for ER+ cases 17.

TNP had a marginally greater effect on prognosis in lymph node negative patients compared to lymph node positive patients (Figures 4 and 5). The univariate relative risk for breast cancer death at 3 years in the TNP group versus the non-TNP group was 5.40 and 3.48 among lymph node negative patients and lymph node positive patients respectively, giving a magnitude of reduction in survival at 3 years of 0.68 in lymph node positive patients compared to lymph node negative patients. Whether lymph node status has less prognostic value in basal-like breast cancers remains a contentious issue. For example in their recent study of African-American women, Carey et al did not see a poor survival in lymph node negative basal-like breast cancer, but it was very poor in lymph node positive cases 5 while other groups have found that node-negative basal-like breast cancer also carries a poor prognosis 18. In this study the limited correlation between tumor size and mean number of lymph nodes in the TNP group and the modest difference of the effect of TNP between lymph node negative and lymph node positive groups suggest that lymph node involvement is a less reliable predictor of prognosis in the TNP group.

Data from the first cohort studied (JGH) indicate that breast cancers with the TNP have a worse prognosis at least in the first three years after diagnosis but this difference in prognosis may diminish with time from diagnosis. These data were validated in a second large independent data set (VGH). Both data sets showed very similar overall survival curves suggesting that they are generally comparable. Although both data sets were identified retrospectively, this is counterbalanced by the fact that they originate from different centers and the degree of consistency between the two data sets which strengthens the overall findings. However, given the small numbers of cases analyzed, any impact of treatment on survival stratified by TNP status would have had to be very large to become apparent and further larger, preferably randomized, studies are required to assess this in greater detail.

The data presented here add to the growing body of evidence that basal-like breast tumors have a worse prognosis 13151819 and respond less well to chemotherapy at relapse 20, although there remains a significant degree of heterogeneity within this group 621222324. Some of these studies have used microarray-based gene expression studies to identify the basal-like group and, while these studies are likely to be more accurate in delineating the basal phenotype than standard immunohistochemical methods, they are not yet routinely used in clinical practice. The advantage of this study is that it uses markers that are readily available in most pathology departments and is therefore directly translatable into routine clinical management, and can be applied to archival specimens for which long-term follow-up information is already available.

A large number of new markers are emerging which aim to further delineate the basal phenotype 2325. However, ER, PR, HER2, EGFR and CK5/6 are already routinely available in most centers. ER, PR, and HER2 in particular are used to guide treatment decisions in breast cancer 26 and it is therefore important to know exactly how expression of these markers affects prognosis. Haffty et al. have recently published prognosis data on a series of 482 patients, 117 of which had a TNP 13. The median follow up time was 7.9 years and TNP was an independent predictor of disease-specific survival (hazard ratio = 1.79; 95% CI 1.03–3.22). Another recent study showed that in the neoadjuvant setting, patients with ER negative and HER2 negative breast cancer have higher sensitivity to anthracycline-based chemotherapy than the luminal subtype, and have higher rates of pathologic complete response 27. Several new drugs that target the EGFR in breast cancer are currently being evaluated 28 and the observations presented here suggest that the effects of these drugs may become more apparent over the longer term, beyond the time over which a typical drug trial would extend. This would have important implications for trial design and interpretation of results. Finally, an increasing number of immunohistochemical markers are being utilized in the identification of basal-like and BRCA1-related breast cancers 18293031, and further validation of these additional markers will be required if they are also to be used as a guide to clinical prognosis and therapeutic choices 14.

Our data confirms that TNP is useful as a prognostic marker, but also suggests that the effect of TNP on survival reduces over time and that focusing on CK5/6 and/or EGFR expression may provide a better marker for long term prognosis (beyond three years).

The author(s) declare that they have no competing interests.

MT and WDF conceived and designed the study and drafted the manuscript, JSB undertook the statistical analyses, MCUC, LRB, DGH, LAA and TON were involved in data acquisition and interpretation, and critically revising the manuscript.