Molecular Cytogenetic Identification of Subgroups of Grade III Invasive Ductal Breast Carcinomas with Different Clinical Outcomes

Breast cancer is a heterogeneous disease with a disparate variety of histological types and a wide spectrum of responsiveness to different treatments, making clinical management difficult. The majority of breast carcinomas fall into the category of invasive ductal carcinoma, no special type, for which histological grade is one of the best predictors of behavior. Poorly differentiated grade III invasive ductal carcinomas are strongly associated with shorter recurrence-free and overall survival times than lower grade I and II tumors (1). Within these groups, however, considerable heterogeneity still exists, and delineation of the most aggressive subtypes within grades would be of considerable clinical benefit.

Invasive ductal carcinoma-no special type, as determined morphologically, is thought to arise exclusively from the luminal epithelial cells of the breast. It has been known for some time, however, that a proportion (2–18% of all invasive ductal carcinomas and up to 25% of grade III tumors) of these tumors have been demonstrated to show a basal/myoepithelial cell phenotype by immunohistochemical analysis using a range of markers including intermediate filaments cytokeratin (CK) 5 and 14 (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17). Recent microarray studies have also identified a “basal-like” group of breast tumors based on their patterns of gene expression (18, 19).

Although a comprehensive characterization and definition of basal tumors is lacking, there are a number of features reported to be associated with this phenotype of invasive ductal carcinoma-no special type. Morphologically, they appear to be predominantly grade III (14, 16) and are reported to frequently contain large central acellular zones composed of necrosis, tissue infarction, collagen, and hyaline material on their cut surfaces (15). Immunohistochemically, as well as in the expression of a number of myoepithelial markers, these tumors seem to be predominantly estrogen receptor (ER), progesterone receptor (PR), and ERBB2 negative (16), an immunophenotype resembling BRCA1 tumors (20). Microarray analysis has also demonstrated a similarity between sporadic, basal-like tumors and those familial tumors harboring a BRCA1 mutation, based on their patterns of gene expression (21).

The pathogenesis of such lesions appear to indicate a poor prognosis. As well as the association with high histological grade and hormone receptor negativity, the myoepithelial phenotype has been reported to be associated with a high risk of brain and lung metastases and of death by cancer independent of nodal status and tumor size (22). The basal-like phenotype identified by expression profiling experiments conferred a shorter survival time than the other tumor groups described (19). Tissue microarray analysis of basal keratins 5 and 17 showed a poorer clinical outcome in node-negative tumors expressing one or both of these markers (17).

Preliminary analysis has indicated that the pattern and number of alterations, as detected by comparative genomic hybridization (CGH), more closely resembles pure myoepithelial carcinomas than other grade III invasive ductal carcinoma (16). In this study, we have chosen to use CK14 to identify grade III invasive ductal carcinomas as exhibiting a basal/myoepithelial phenotype and have carried out CGH analysis on 43 CK14 positive and 43 age- and grade-matched CK14 negative tumors, in an attempt to investigate the molecular events associated with the different subtypes and to correlate these data to clinical outcome.

Patients diagnosed as having a grade III mammary carcinoma at the Hedley Atkins/Cancer Research UK Breast Pathology Laboratory, Guy’s Hospital between 1975 and 1992 were included in the study and have formed a cohort upon which studies have been published previously (23). Stained sections from the primary tumor were retrieved, and the tumor was regraded according to the modified Bloom and Richardson method (1). Any patient subsequently found to have a grade II tumor on review was excluded from the study. An additional group of patients was also excluded because insufficient tumor tissue remained in the diagnostic block for study. Therefore, a total of 418 patients were available. Of these, 89 were found to be either focally or diffusely positive for CK14 by immunohistochemistry, and 43 of these tumors proved amenable to microdissection, DNA extraction, and CGH analysis. Forty-three age-matched, CK14-negative tumors were chosen from the same cohort as controls and were subject to the same experimental difficulties. Matching for tumor size, stage, and nodal status were also carried out as closely as possible, with the two groups showing no significant differences in distribution of these markers.

Staining was performed on 3-μm-thick sections of formalin-fixed, paraffin-embedded tissues. Sections were cut and placed onto Vectabond-coated slides and allowed to dry overnight at 37°C. Before staining, the sections were melted onto the slide by placing them in an oven at 60°C for 16 h. Sections were dewaxed, and endogenous peroxidase activity was inhibited using methanol and hydrogen peroxide. After thorough washing, heat-mediated antigen retrieval was used to expose sites of immunoreactivity. This was achieved by placing the slides in a pressure cooker containing boiling 0.01 M citrate buffer (pH 6.0) where they were kept under pressure (15 psi) for 2 min. Once the sections were cool, the remaining immunohistochemical staining was carried out using an Optimax Automated Immunostainer (A. Menarini) with a standard peroxidase-conjugated streptavidin-biotin complex method. CK19 (clone RCK108; 1:50), ER (clone 1D5; 1:100), and PR (clone PgR636; 1:400) were obtained from Dako Ltd., CK14 (clone LL002; 1:50) was obtained from Biogenex, and ERBB² (clone 3B5; 1:1500) was obtained from Oncogene Science. Sites of peroxidase activity were detected using 3,3′-diaminobenzidine/hydrogen peroxide, which produced a brown reaction product. The slides were lightly stained with hematoxylin. A positive control, known to express the antigen, was included in every batch of staining.

ER and PR were assessed using the Quick score method (24) with a score of ≤3 considered negative. c-erbB2 was assessed in accordance with the Dako HercepTest guidelines with a score of ≤1 considered negative. Cytokeratin 14 and 19 were scored according to the presence or absence of expression in the invasive component.

Microdissection, DNA extraction, and CGH analysis were carried out as described previously (25). Briefly, microdissection was carried out using the PixCell II Laser Capture Microdissection system (Arcturus, Mountain View, CA) from formalin-fixed, paraffin-embedded tissue and the DNA extracted with 0.5 μg/μl proteinase K. Amplification and fluorescent labeling of the DNA from microdissected tumor and normal tissue was carried out by degenerate oligonucleotide primed-PCR in two rounds as published previously (26) and hybridized to normal male metaphase spreads (Vysis United Kingdom Ltd., Richmond, England) for 2–3 days at 37°C. Metaphase chromosome preparations were captured using a Zeiss Axioskop microscope, Photometrics (Munich, Germany) KAF1400 CCD camera and Vysis SmartCapture software. Image analysis was performed using Vysis Quips CGH software. Between 5 and 10 representative images of high quality hybridizations were analyzed and the results combined to produce an average fluorescence ratio for each chromosome. Control experiments were carried out using normal:normal (microdissected normal lymph node) cohybridizations, of which the average red:green ratio levels and 95% confidence intervals were used to set the lower and upper limits for scoring losses and gains of genetic material as 0.80–1.20.

All of the changes in DNA copy number were reduced to “gains” or “losses” at the chromosomal arm resolution. The data were retabulated as +2 for gains and −2 for losses for supervised and unsupervised analysis. Unsupervised hierarchical clustering was carried out using “hclust” in R 1.7.0,⁶ and plotted using the “heatmap” function in the mva package in R. Hierarchical clustering was based on a Euclidean distance measure using Ward’s minimum variance method. Supervised analysis was carried out using the nearest shrunken centroid classifier (27) implemented in R 1.7.0. Survival analysis was carried out using the statistical platform S-Plus version 6.1 for Windows (Insightful) on our right-censored clinical follow-up data. Kaplan-Meier plots were generated using the function “survfit,” and the log-rank test was carried out to determine whether curves were significantly different from each other using the “survdiff” function. Multivariate analysis was carried out using the Cox proportional hazards model with “coxph” used to investigate the independence of the individual proteins on prognosis.

Of 418 grade III invasive ductal breast carcinomas in our study, 89 (21.2%) were either focally of diffusely positive for CK14. Of these, 43 were amenable to CGH analysis and 43 age-matched CK14-negative grade III tumors were analyzed as controls. Of these 86 tumors analyzed by molecular cytogenetics, 33 were ER positive (38.4%), 25 were PR positive (34.7%), 15 were ERBB2 positive (18.3%), and 83 were positive for CK19 (96.5%). Forty-nine showed positive lymph node metastasis (57.6%). Follow-up data were available for all of the patients who had a mean age at diagnosis of 47.8 years (median, 46.5), mean tumor size of 3.4 cm (median, 3.5 cm), a mean disease-free time of 8.7 years (median, 4.2), and mean survival time of 9.9 years (median, 7.0). Full details of clinicopathological data are available in Supplementary Table S1.

Fig. 1 shows a representative CK14-positive tumor taken for CGH analysis. Positive CK14 tumors were shown to be significantly associated with negative ER status (P < 0.00001), negative PR (P = 0.001), and negative ERBB2 (P = 0.0201), but not with nodal status (P = 0.191) or tumor size (P = 0.9479).

CGH analysis of the 86 tumors showed a mean number of alterations of 8.2 (median, 6.0). Full details of all CGH data for each case are available in Supplementary Table S1. The most frequent alterations over all samples were gains at 1q (21%), 17q (26%), and 20q (41%); and losses at 1p (21%), 16q (20%), 17p (20%), 17q (24%), 19p (21%), and 19q (21%). Differences in CGH profiles between the CK14-positive and -negative tumors are shown in Table 1. Statistical significance was determined by Fisher’s exact test. The CK14-positive tumors showed an increased number of losses at 16p (33% CK14 positive versus 5% CK14 negative; P = 0.016), 17q (37% versus 12%; P = 0.0110), 19q (33% versus 9%; P = 0.0155) and Xp (26% versus 7%, P = 0.0381). CK14-negative tumors displayed an increased number of gains at 17p (5% versus 21%; P = 0.0488), 17q (9% versus 42%; P < 0.0010), 20p (0% versus 21%; P = 0.0025), and 20q (21% versus 60%; P < 0.0004); and losses at 4q (5% versus 23%; P = 0.0261), 9p (0% versus 14%; P = 0.0259), and 13q (0% versus 21%; P = 0.025). Summary karyograms displaying all of the changes in DNA copy number are shown in Fig. 2.

Clinical follow-up data were available for all of the cases, and univariate analysis was carried out to determine which clinicopathological, immunohistochemical, and molecular variables were associated with prognosis (Table 2). In this grade III invasive ductal carcinoma cohort, the most significant prognostic indicator was nodal status, with positive lymph nodes predicting for shorter disease-free time (P = 0.00269) and overall survival (P = 0.00275). Trends were observed linking poor prognosis with ER negativity, ERBB2 positivity, and number of CGH alterations; however, these did not reach statistical significance. A number of chromosomal loci showed a significant association with disease-free and overall survival, including gain of 1q, loss of 3p, loss of 4p, gain of 6p, and loss of 8p. CK14 positivity showed no significant influence on prognosis in this cohort (disease-free survival P = 0.193; overall survival P = 0.385). CK14-positive tumors showed a significantly lower overall mean number of CGH alterations than the CK14-negative group (6.5 versus 10.3; P = 0.0012). Interestingly, the few CK19-negative tumors (3 of 83) showed a significant association with shorter disease-free time (P = 0.0209) and overall survival (P = 0.0402), despite these small numbers.

Because there were a number of alterations that were more or less prevalent in the different groups of grade III tumors, supervised cluster analysis was carried out to determine whether the CGH data would be able to predict for CK14 status on its own. Using the nearest shrunken centroid classifier, 76% of breast tumors were correctly assigned into the relevant group by CGH data alone using leave-one-out cross-validation (cross-validated error-rate, 0.24) after first training the classifier (Fig. 3,A). We next attempted to determine whether we could identify subgroups of these tumors by their CGH data, which may have different biological behaviors. Hierarchical clustering of the CGH data split the tumors into two large branches, containing 38 of 55 (69%) CK14-positive and 26 of 31(84%) CK14-negative tumors, respectively, giving an overall error rate of 0.256 (Fig. 3 B).

These branches could be additionally refined into 6 smaller clusters. When multivariate analysis was carried out using the Cox proportional hazards model, ER negativity, CK14 positivity, positive lymph nodes, and 1 cluster of tumors were found to be independent predictors of poor survival (Table 3). Within the large, predominantly CK14-positive branch, “cluster 1” was composed of 18 of 23 (78%) CK14 positive tumors and was found to indicate shorter overall survival (P = 0.0270) and disease-free time (P = 0.0503).

Univariate analyses showed that cluster 1 tumors predicted shorter overall survival for all of the tumors (P = 0.0414) and when the CK14-positive tumors were considered (P = 0.0475). Shorter disease-free times were also observed, but the differences were not statistically significant (P = 0.175, all tumors; P = 0.127, CK14-positive tumors). The Kaplan-Meier survival curves are shown in Fig. 4.

The cluster 1 tumors, as well as being predominantly CK14 positive (18 of 23; 79%; P = 0.003), were found to be associated with negative ER (P = 0.0231) and negative PR (P = 0.039) but not with ERBB2 (P = 0.7458), nodal status (P > 0.9999), or tumor size (P = 0.6237). The CGH alterations, which were found to be significantly associated with the cluster 1 tumors versus the rest, were gain of 1q (P = 0.004), and losses at 8p (P = 0.0199), 16p (P < 0.0001), 16q (P = 0.0065), 17p (P = 0.0488), 17q (P = 0.0001), 19p (P = 0.001), and 19q (P = 0.0001), and are shown in Table 4.

There is accumulating evidence to suggest that different histological grades of invasive ductal breast carcinomas may have distinct molecular origins and pathogenesis and do not typically progress from one grade group to another (28, 29, 30, 31). The different grades have different clinical behaviors, and within-grade studies to identify the more aggressive subgroups of these classes of breast tumors would be of great assistance in clinical management. The expression of basal/myoepithelial markers has been observed in a proportion of grade III invasive breast tumors, and the spectrum of basal-like tumors, also recognized by morphology (15, 32), molecular cytogenetics (16, 33), and expression profiling (18, 19), has been associated with poor prognosis (17). CGH has the advantage of being applied to archival pathology specimens with long-term follow-up as well as being amenable to microdissection strategies to profile the molecular genetic change occurring in a pure population of tumor cells.

In the cohort of grade III tumors in the present study, CK14 positivity was not associated significantly with prognosis. At the CGH level, the CK14-positive tumors showed fewer overall changes in DNA copy number than the CK14-negative group. This may in part explain the conflict with a previous study (33), which found a higher number of CGH alterations in the basal-like group of tumors as determined by CK5/6 positivity. This incongruity with published data suggesting that basal keratin expression confers a poorer prognosis must be interpreted in the light of the fact that we have focused only on poorly differentiated, high-grade malignancies. Within this cohort, ER negativity is not a significant prognostic indicator by univariate analysis, nor is tumor size or ERBB2 status. Interestingly, the few (3 of 83) tumors that were negative for the luminal epithelial keratin CK19 showed a very poor prognosis, statistically significant despite the small numbers.

The CK14-positive and -negative groups were clearly different in terms of their CGH profiles. In particular, the CK14-positive tumors showed an increased prevalence for losses at 16p, 17q, and 19q, all alterations associated with pure myoepithelial carcinomas (34). None of these alterations on their own conferred any prognostic information, although gain of 1q and 6p, as well as losses of 3p, 4p, and 8p did indicate shorter disease-free and overall survival times in the whole cohort. Taken as a whole, the CGH profiles of the tumors alone were able to predict the CK14 status in approximately three-quarters of cases by supervised analysis, using leave-one-out cross-validation. This demonstrates the inherent differences in the molecular evolution of the tumor groups.

Unsupervised hierarchical clustering has been applied to gene expression profiling data to identify subgroups of breast tumors with different clinical outcomes (19). Such a statistical approach may also be applied to CGH data, with the advantage in this instance that we are profiling pure populations of microdissected tumor cells. In our study, hierarchical clustering revealed two large branches, which predicted for CK status again with approximately three-quarters accuracy. Within the predominantly CK14-positive group, 4 separate clusters were identified, with the cluster 1 tumors accounting for 18 of 43 CK14-positive tumors. This subgroup of tumors had a worse prognosis than the rest of the tumors, both in terms of the whole data set, as well as that stratified purely by CK14 positivity. In multivariate analysis, this cluster of tumors was found to be an independent indicator of shorter overall survival in a model including ER, CK14, and nodal status, identified by the Cox proportional hazards test.

Substratification of CK14-positive breast tumors into two groups, based on their CGH profiles, which is reflected in their biological behavior, is of considerable clinical interest. This poor prognosis group was associated with negative hormone receptor status and exhibited a number of CGH alterations with higher prevalence than the better prognosis tumors, including gain at 1q, and losses at 8p, 16p, 16q, 17p, 17q, 19p, and 19q. The association of these loci with a poor clinical outcome group of tumors will provide clues for targeted studies hoping to unravel the underlying molecular events associated with the pathogenesis of these lesions.

The immunophenotype of these tumors, as has been pointed out in smaller studies (16), exhibits a resemblance to tumors with germ-line BRCA1 mutation (20), and this possible association has been postulated recently by gene expression analysis (21). Investigations into epigenetic mechanisms of BRCA1 inactivation in basal-like breast tumors seem warranted.

The clinical heterogeneity of grade III invasive ductal breast carcinomas is well-known. The idea that we could identify, at diagnosis, subgroups of these patients that will do badly (and thus require aggressive therapy) or relatively well (where the patient may be spared such treatment) is an attractive one. Clearly some form of basal/myoepithelial differentiation is apparent in these lesions, as determined by their expression of some basal markers, and a pattern of CGH alterations at loci associated with pure myoepithelial carcinomas. Accurate characterization of these lesions at the morphological level, along with an additional immunohistochemical refinement using an extensive panel of basal/myoepithelial markers will be necessary to produce a set of criteria that will allow the accurate diagnosis of these tumors, with the implications that will have on patient management. Unraveling the molecular pathways that drive the divergent groups of good- and poor-prognosis tumors will be required to identify potential targets for novel therapeutic strategies.