Breast cancer is a common but very diverse disease with considerable survival heterogeneity 12345. An ongoing challenge is to find improved methods of identifying and classifying groups of tumours with differing biological behaviours or responsiveness to specific therapies. Expression profiling analysis is reshaping our understanding of breast cancer and has provided a working model for modern breast cancer taxonomy 678. These seminal studies have systematically identified a subgroup of breast tumours that are characterised by the expression of genes normally expressed in basal/myoepithelial cells of normal breast, including high-molecular-weight cytokeratins (CKs) and P-cadherin 678910. These findings have confirmed and expanded the concepts first described as early as 1967 11, and in particular those described over the last decade 12, by conventional histopathological and immunohistochemical examination. Studies have demonstrated that whilst most breast tumours express a purely luminal phenotype, a minority of ductal tumours (4% to 22%) will show ultrastructural features and co-express markers that are characteristic of the outer myoepithelial, or basal, compartment of the normal breast 11121314151617181920212223242526272829303132333435.

Morphologically, these tumours are predominantly of high grade 919232425262935 and more frequently show medullary-like features 3031 and metaplastic elements 1920323336373839 (Figure 1a). Immunohistochemically, they are predominantly oestrogen receptor (ER)-, progesterone receptor (PR)-, and C-erb B2 receptor (HER2)-negative 1922242532. These characteristics also predominate among tumours arising in patients with germline BRCA1 (breast cancer 1, early onset) mutations 4041. Recent studies have indicated that positive staining for 'basal' markers in patients with familial breast and/or ovarian tumours is strongly predictive of carrying a BRCA1 mutation 273441; however, there are several lines of evidence to suggest that sporadic basal-like ductal carcinomas do not harbour BRCA1 somatic mutations 42.

Despite these morphological, immunohistochemical, and molecular features, current routine diagnostic practice does not separately recognise these tumours and management is the same as for other grade- and stage-matched breast tumours. Importantly, however, evidence suggests that these tumours have a metastatic pattern of dissemination and clinical behaviour that are different than comparable ductal carcinomas not showing basal marker expression. Tsuda et al. 19 and Hicks et al. 43 reported a higher incidence of lung and brain metastases associated with a myoepithelial phenotype, as defined by CK14, smooth muscle actin, or S100 protein expression. Furthermore, retrospective, population-based studies have demonstrated that tumours that express these basal markers have a more aggressive clinical behaviour when compared to those with a luminal phenotype and that 'basal' status may be an independent prognostic factor. In contrast, a comparative genomic hybridisation (CGH) analysis performed in our institutions 28 of 86 invasive ductal carcinomas of no special type (IDC-NST) (50% CK14-positive and stage- and age-matched), all of which were grade III, did not suggest a uniformly poor prognosis for the basal-like (CK14-positive) tumours. Instead, the CGH analysis revealed distinct subgroups of basal-like grade III tumours that exhibited a comparatively good prognosis with extended clinical follow-up (>10 years), in comparison to the rest of the grade III group, and a basal-like subgroup with a relatively poor prognosis. In fact, in contrast to the 'pan-grade' studies, the CK14-positive tumours appeared to have an overall better outcome than the rest of the group 28.

To investigate these observations further and explore the hypothesis that basal-like tumours are entities biologically distinct from other ductal carcinomas and that they may themselves have different subgroups, we assessed a large cohort of grade III IDC-NST for expression of CK14 and correlated this with long-term survival and pattern of metastatic disease.

Four hundred and seventy grade III IDC-NST, diagnosed between 1975 and 1991, were identified from the patient database at the Hedley Atkins/Imperial Cancer Research Fund Breast Pathology Laboratory, Guy's Hospital (London, UK), where all the cases were managed. Tumours had been graded according to the modified Bloom and Richardson system 2 and have been previously published 44. Immunohistochemistry was performed on a single tumour block for CK14 as detailed previously 2829. Briefly, 2 minutes of pressure-cooking at pH 6.0 was used for antigen retrieval with CK14 clone LL002 (1:50) (BioGenex Inc., San Ramon, CA, USA). Four hundred and forty-three cases contained tumour on review.

CK14 expression was assessed independently by two pathologists (LGF and JSR-F). A tumour was designated as CK14-positive if it was considered that at least 1% of true invasive tumour cells expressed the CK marker. From our previous experience 2829, and as recently demonstrated by Laakso et al. 2526, it has been apparent that whereas in some tumours virtually every cell expresses the basal marker (CK14, in this instance), others may show a focal or variable percentage of positive cells (Figure 1b,c). We therefore subdivided the CK14-positive tumours into those showing 'diffuse expression' (defined as more than 90% of tumour cells positive; Figure 1b) and those showing 'focal expression' (1% to 90% of cells positive; Figure 1c). Discrepancies between the observers were resolved on a multi-headed microscope. For ease of reading, CK14-positive tumours are referred to henceforth as 'basal' and CK14-negative tumours as 'non-basal' cases. Data on tumour size, axillary lymph node status, ER, and primary treatment were available for 98%, 98%, 93%, and 100% of patients, respectively. Between 1975 and 1991, the methodology for ER measurement changed as newer reagents became available. Initially, ER status was determined by cytosol ligand-binding assay 45, which was then replaced by the Abbott enzyme immunoassay (EIA) (Abbott Laboratories, Abbott Park, IL, USA) 46. A successful comparative study was undertaken between the ligand-binding assay and EIA before transferring to the latter method.

All ER data were recorded on a prospectively acquired and verified clinicopathological database at the time of assay. Full clinical follow-up data, including sites of metastatic disease, were available for all cases.

The distribution of clinicopathological features is detailed in Table 1. Briefly, age at diagnosis ranged from 21 to 85 years (mean 53.2 years). Fifty-seven percent of tumours had axillary nodal metastases, and 46% were ER-positive. Fifty-two percent of cases had received no form of adjuvant treatment. Length of follow-up ranged from 3.2 to 333.5 months, with a mean follow-up of 116.4 months.

These data contrast with much of the recently published data showing entirely poor outcome in basal tumours 78101733. Unlike that of previous pan-grade studies, the aim of this study was to look at the impact of 'basal' phenotype on the long-term outcome in patients with grade III IDC-NST, a subgroup with a traditionally poor prognosis. We have specifically chosen a cohort of cases diagnosed historically as grade III ductal carcinomas, not including any special subtypes such as metaplastic breast carcinoma or medullary carcinoma, because this is the group of tumours in which the majority of basal cases seem to arise 5.

Terminology and definitions surrounding the concept of basal tumours are controversial, and a plethora of different markers and definitions have been employed to identify cases in clinical studies. Nielsen and colleagues 22 have proposed a definition based on ER, HER2, CK5/6, and EGFR (epidermal growth factor receptor) expression; however, even this definition does not show a perfect correlation with microarray data. Indeed, there is no accepted gold standard on which to assess any signature (for example, in a recent study 9, only 80% of tumours classified as HER2 by cDNA microarrays overexpressed HER2 at the immunohistochemical level). Given that there is currently no consensus terminology or internationally accepted definition (morphological or immunophenotypic) for basal-like tumours, we have employed CK14 alone to identify basal tumours in the present study. Our reasons for choosing a single marker are pragmatic: (a) limited material from this unique cohort of patients was available, and (b) CK14 is a marker that we have had considerable experience with and forms the basis for previous studies in the laboratory 282936. Furthermore, it has been demonstrated that CK14 expression correlates strongly with other 'basal' markers and that it identifies basal tumour proportions similar to those of other investigators (for example, CK5/6, CK17, and P-cadherin) 212223242526323335.

Our results have highlighted several important features of basal tumours as identified by CK14 expression. Relative to other grade III tumours, they are much more likely to be ER-negative and less likely to be axillary node-positive. They show lower rates of metastasis to the bone and liver but a higher rate of brain metastasis. Over the first 5 years after diagnosis, the relapse-free survival and OS appear to be similar to other grade III tumours, but their subsequent prognosis is much better. Given that in this study we compared grade III tumours only and that basal-like phenotype and 3+ expression of HER2 are more prevalent in this group but are inversely correlated 9223947, it is likely that the non-basal-like tumours include a high prevalence of HER2-amplified tumours 47, which are reported to have an aggressive clinical behaviour. Further studies to directly compare the contribution of basal-like and HER2 phenotypes to survival of patients with grade III ductal carcinomas are warranted but are outside the remit of this study. The pattern of recurrence in basal tumours, with almost all recurrences occurring in the first 5 years, also differs from non-basal tumours, in which late relapse is widely recognised. Previous studies, many of which have a shorter median follow-up than that of our cohort of patients, have indicated that basal tumours usually have a very poor prognosis when compared to a mixture of grades I, II, and III basal keratin-negative breast tumours 7810173233. These observations are not necessarily inconsistent with this study, because our study does indicate a poor survival for patients with basal tumours for the first 5 years of follow-up.

The findings suggest the existence of two subgroups of basal carcinomas: one exhibiting early relapse and aggressive clinical course and a separate group that despite the traditionally poor prognostic indicators do not relapse. This would be consistent with the findings of our previous CGH analysis 28, which was carried out on a subset of the tumours examined in the present study and was able to split the basal tumours into two groups based on both their molecular genetic alterations and prognosis.

We found that traditional clinicopathological parameters could not discriminate between 'good' and 'bad' basal tumours. In addition, we found that none of the specific morphological features associated with the basal phenotype 36, such as central scar, high mitotic rates, pushing growth pattern, or necrosis, occurred with any significant difference between good and bad basal cases. However, the pattern of CK14 expression shows clear differences in prognosis. Those with a diffuse (>90%; Figure 1b) staining pattern seem to fall largely into the good-prognosis group of tumours. A similar subclassification of basal tumours according to the expression of basal keratins into basal and basoluminal subtypes also suggested that tumours with a pure basal phenotype had a better prognosis 26.

Because the majority of breast tumours in BRCA1 germline mutation carriers are of basal type 273441, our observations are also relevant to the prognosis in these women. Studies of prognosis in BRCA1 germline mutation carriers have been inconsistent; studies based on carriers identified through clinical testing in high-risk families have shown a relatively good prognosis, whereas studies based on testing of population series have shown a poor prognosis 48. This apparent inconsistency could be explained by our results because cases undergoing clinical testing will tend to be identified some time after diagnosis when subsequent prognosis is good.

These concepts have important implications for patient management. This study confirms that there is a group of aggressive basal tumours that are currently not identified in routine diagnostic practice but that show a biological behaviour different from other grade III IDC-NST. They relapse early with short subsequent survival and notably carry a higher risk of brain metastases 1943. These tumours may need different (for example, platinum-based) 49 chemotherapy regimens or more aggressive therapeutic regimens and may need to be specifically identified in laboratory research designed to identify new therapeutic targets.

It is also clear from these results that expression of basal markers alone cannot be used to indicate a tumour with poor prognosis. Indeed, it may be that some of the basal tumours with a better prognosis will require much less aggressive therapy and follow-up and that patients can be reassured of their long-term outlook despite being diagnosed with a grade III cancer.

This study shows that identifying basal tumours as a distinct group has real clinical relevance. They are relatively common, occurring in approximately 20% of grade III IDC-NST and approximately 10% of breast tumours overall. In the UK, with approximately 50,000 new breast tumours diagnosed each year, we may be missing some 5,000 basal tumours that carry distinct, but mixed, prognostic implications and that may respond differently to conventional chemotherapy regimens. The aim now is finding ways to further identify, characterise, and define 'good' and 'bad' basal tumours and to tailor management appropriately. Furthermore, these tumours may express distinct targets for novel therapeutic regimens, and research programmes should bear this in mind. Clinicopathological, therapeutic, and molecular genetic approaches will almost certainly all play a part in this task.

BRCA1 = breast cancer 1, early onset; CGH = comparative genomic hybridisation; CK = cytokeratin; DFS = disease-free survival; EIA = enzyme immunoassay; ER = oestrogen receptor; HER2 = C-erb B2 receptor; HR = hazard ratio; IDC-NST = invasive ductal carcinomas of no special type; OR = odds ratio; OS = overall survival; PR = progesterone receptor.

The authors declare that they have no competing interests.

LGF conceived the study, analysed the immunohistochemical markers, accrued and collated the data, carried out preliminary statistical analysis, supervised the statistical analysis, and drafted the manuscript. JSR-F analysed the immunohistochemical markers, accrued and collated the data, carried out preliminary statistical analysis, and drafted the manuscript. KR and CEG curated the data. CJ and AH participated in the study design. DE supervised and reviewed the statistical analysis. SRL conceived the study, supervised the statistical analysis, and drafted the manuscript. All authors read and approved the final manuscript.