Purpose: To characterize the prognostic and predictive impact of protein expression profiles in high-risk breast cancer patients who had previously been shown to benefit from high-dose chemotherapy (HDCT) in comparison to dose-dense chemotherapy (DDCT).

Experimental Design: The expression of 34 protein markers was evaluated using tissue microarrays containing paraffin-embedded breast cancer samples from 236 patients who were randomized to the West German Study Group AM01 trial.

Results: (a) 24 protein markers of the initial panel of 34 markers were sufficient to identify five profile clusters (subtypes) by K-means clustering: luminal-A (27%), luminal-B (12%), HER-2 (21%), basal-like (13%) cluster, and a so-called “multiple marker negative” (MMN) cluster (27%) characterized by the absence of specifying markers. (b) After DDCT, HER-2 and basal-like groups had significantly worse event-free survival [EFS; hazard ratio (HR), 3.6 [95% confidence interval (95% CI), 1.65-8.18; P = 0.001] and HR, 3.7 (95% CI, 1.68-8.48; P < 0.0001), respectively] when compared with both luminal groups. (c) After HDCT, the HR was 1.5 (95% CI, 0.76-3.05) for EFS in the HER-2 subgroup and 1.1 (95% CI, 0.37-3.32) in the basal-like subgroup, which indicates a better outcome for patients in the HER-2 and basal-like subgroups who received HDCT. The MMN cluster showed a trend to a better EFS after HDCT compared with DDCT.

Conclusions: Protein expression profiling in high-risk breast cancers identified five subtypes, which differed with respect to survival and response to chemotherapy: In contrast to luminal-A and luminal-B subtypes, HER-2 and basal-like subgroups had a significant predictive benefit, and the MMN cluster had a trend to a predictive benefit, both from HDCT when compared with DDCT.

Primary breast cancer patients with ipsilateral axillary lymph node metastasis can be risk stratified into subgroups according to the number of axillary lymph nodes involved. Previous studies have shown that the outcome in breast cancer patients significantly deteriorates with increasing numbers of involved axillary lymph nodes (1). Thus, patients with more than nine axillary lymph node metastases are stratified into a high-risk subgroup.

In an effort to improve the outcome in high-risk patients, several trials have studied the benefit of high-dose chemotherapy (HDCT) with stem cell support. In the recently published West German Study Group (WSG) AM-01 trial, high-risk breast cancer patients receiving rapidly cycled tandem HDCT were reported to have significantly better event-free survival (EFS) and overall survival (OS) compared with conventional dose-dense regimen (2). Despite these observations, the failure of a number of tumors to respond to adjuvant treatment, the significant toxic side effects, and the elevated therapy-related morbidity or mortality rates associated with this high-dose approach (3, 4) make it necessary to identify parameters that may accurately predict the clinical outcome after HDCT. The results of such studies could thereby define subgroups within the high-risk breast cancer patients who would have the maximum benefit from HDCT and spare those patients from toxic side effects who will not benefit from HDCT. Large-scale molecular techniques such as DNA microarrays are well suited to identify those breast cancer subgroups that show the best response to a given therapy. By analyzing the expression of thousands of genes simultaneously, these studies have already proposed a novel molecular classification of unselected breast cancers based on gene expression profiles (57). Briefly, such studies have identified prognostically different subsets of breast cancer with separate gene expression profiles: mainly luminal, basal-like, normal-like, and HER-2 subtypes (6). Due to the need of frozen tumor material with high-quality RNA, high costs, and interlaboratory divergence, DNA microarrays are currently unsuitable for routine use in standard clinical settings. Additional opportunities to identify and/or validate molecular prognostic and/or predictive tumor signatures are provided by alternative high-throughput techniques such as tissue microarrays (TMA). This technique provides the opportunity to immunohistochemically analyze the expression of large protein panels in hundreds of paraffin-embedded tumor specimens simultaneously (8). By using TMAs and biostatistical techniques developed for gene expression profiling, protein expression profiling has recently allowed the stratification of breast tumors into clinically relevant subgroups. Novel prognostically relevant subsets of unselected breast cancers that are not identifiable on single marker expression analysis have been identified by this approach (9, 10). Until now, few studies have examined the predictive value of tumor- and treatment-specific prognostic indicators of outcome in high-risk breast cancer patients treated with HDCT; most of these studies focused on single or few markers such as estrogen and progesterone receptor, p53, Her-2/neu, and Bcl-2 (4, 11, 12). The potential for combinations of prognostic markers to be superior to any single marker has been observed previously (13). The objective of this study was to determine whether protein expression profiling could be used to identify prognostically relevant subgroups of high-risk breast cancer patients who will benefit from HDCT. Using TMA and immunohistochemistry, we have analyzed the expression of 34 proteins selected for their relevance in breast cancer in a retrospective panel of 236 patients from the WSG AM 01 study from whom paraffin blocks were available. Interpretation of the protein expression data to define patient subgroups was done by K-means cluster analysis.

The West German Study Group AM-01 trial and tumor samples

In the presented study, paraffin-embedded breast cancer specimens were used from 236 randomized high-risk breast cancer patients with more than nine affected axillary lymph nodes; these patients were treated within a prospective multicenter study comparing tandem HDCT with dose-dense conventional chemotherapy. The Institutes of pathology of the participating centers were asked to provide representative breast cancer tissue specimens of the initial 403 high-risk breast cancer patients for central pathologic review and for further scientific analysis. Finally, breast cancer samples from a total of 236 study patients (59%) were available for review. The HDCT arm consisted of a short induction with two cycles of epirubicin and cyclophosphamid (90:600 mg/m2) followed by tandem epirubicin, cyclophosphamid, and thiotepa (90:3000:400 mg/m2) with autologous peripheral-blood-progenitor support. The control arm consisted of dose-dense chemotherapy (DDCT): four courses of epirubicin and cyclophosphamid (90:600 mg/m2) and three courses of cyclophosphamide, methotrexate, and fluorouracil (600:40:600 mg/m2) given with intervals of 2 weeks with growth-factor support. The absence of distant metastases at the time of diagnosis was assumed on the basis of normal findings in chest radiography, liver ultrasonography, and bone scan. The WSG AM-01 study was reviewed by the ethics committee of each participating center, and a written informed consent was obtained from all patients. The details of the study are described elsewhere (2).

Tissue Microarray. Breast cancer TMAs were prepared as described previously (14). Briefly, representative tissue blocks were selected as donor blocks for the TMA. Sections were cut from each donor block and stained with H&E. Using these slides, one morphologically representative region was chosen from each of the 236 tumor samples. One cylindrical core tissue specimen per tumor block (diameter = 2.0 mm) was punched from these regions and precisely arrayed into a new recipient paraffin block (20·× 35 mm) using a custom-built precision instrument (Beecher Instruments, Silver Spring, MD). Nine tissue array blocks were prepared, seven containing 30 and two containing 13 tumor sample cores, respectively (Fig. 1). Each block was subsequently stained with H&E to verify the presence of tumor within each 2.0-mm tissue core.

Fig. 1.

Representative immunostaining of (A) positive estrogen receptor, (B) HER-2 score of 3+, (C) CK5, (D) BCRP, (E) ET-1, and (F) S6 (×40, inset, ×400).

Fig. 1.

Representative immunostaining of (A) positive estrogen receptor, (B) HER-2 score of 3+, (C) CK5, (D) BCRP, (E) ET-1, and (F) S6 (×40, inset, ×400).

Close modal

Immunohistochemistry. For clustering analyses, we used a large panel of 34 protein markers (Table 1). Most of the proteins selected play a well-established role in breast carcinogenesis (5, 6). Furthermore, the gene transcripts of these proteins have been reported to be important candidate discriminator genes in stratifying breast cancer into distinct groups based on previous cDNA microarray studies (6, 15). They include markers related to the function and proliferation/differentiation status of the mammary gland and tumor cell [estrogen receptor α (ER), progesterone receptor (PR), synaptophysin, MUC-1, Bcl-2, Ki67/MIB1, cyclin D1, cyclin E, p27, topoisomerase IIα, cyclooxygenase 2, phospho-p42/44MAPK (ERK1/ERK2), phospho-S6 ribosomal protein (S6)]; markers of cellular origin (CK5, CK8, CK17, vimentin); known or putative oncogenes or tumor suppressors [HER-2, epidermal growth factor receptor (EGFR), p53, p63, E-cadherin, fragile histdine triad, protein tyrosine phosphatase, β-catenin, p16, c-kit]; markers for metastasis in breast cancer (endothelin-1, endothelin receptor-α, endothelin receptor-β, CXCR4); markers involved in multidrug resistance of cancer cells [breast cancer resistance protein (BCRP), ABCA3, O6-methylguanine-DNA-methyltransferase (MGMT)]. Immunohistochemical staining was done on 3-μm paraffin sections. Pretreatment for antigen retrieval was mainly done by pressure cooker except for EGFR and S6, where Pronase and autoclaving was used, respectively. Dilutions of the antibodies are listed in Table 1. After blockage of biotin (by avidin-biotin) and peroxidase (by H2O2), immunohistochemical staining was done on an automated immunostainer (Biogenex, i6000, San Ramon, CA) using a standard labeled streptavidin-biotin method (UltraTek Reagent Detection Kit, Scy Tek, Logan, UT) followed by 3,3′-diaminobenzidine enzymatic development. Sections were counterstained blue with hematoxylin. Omission of the primary antibody as well as anti-isotype antibodies (Supplementary Fig. S1) served as negative control.

Table 1.

Proteins tested in the study: species, source, dilution, and results

NumberProteinsAntibodiesOriginCloneDilutionStaining patternPositive/n (%)
ABCA3 Rabbit polyclonal Origin: N. Inagaki, G.G. Wulf  1:800 Cytoplasmic 21/201 (10.4) 
Bcl-2* Mouse monoclonal DAKO 124 1:50 Cytoplasmic 69/217 (31.8) 
BCRP* Mouse monoclonal Chemicon BXP-21 1:100 Cytoplasmic/membranous 85/212 (40.1) 
β-Catenin* Mouse monoclonal BD Transduction 14 1:200 Membranous 133/217 (61.3) 
C-kit* Rabbit polyclonal DAKO Code number A4502 1:200 Cytoplasmic/membranous 27/223 (12.1) 
CK5* Mouse monoclonal Novocastra XM26 1:600 Cytoplasmic 20/205 (9.8) 
CK8* Mouse monoclonal Biogenex C-51 1:5,000 Cytoplasmic 201/216 (93.1) 
CK17* Mouse monoclonal DAKO E3 1:20 Cytoplasmic 20/224 (9.0) 
COX-2 Mouse monoclonal Cayman  1:400 Cytoplasmic 84/215 (39.0) 
10 CXCR4* Mouse monoclonal Zymed 12G5 1:100 Cytoplasmic 108/189 (57.1) 
      Nuclear 22/189 (11.6) 
11 Cyclin D1 Rabbit monoclonal DCS SP4 1:50 Nuclear 116/223 (52.0) 
12 Cyclin E Mouse monoclonal Novocastra 13A3 1:50 Nuclear 23/218 (10.6) 
13 E-cadherin* Mouse monoclonal Novocastra 36B5 1:50 Membranous 166/216 (76.9) 
14 EGFR* Mouse monoclonal Merck E30 1:100 Cytoplasmic/membranous 31/230 (13.8) 
15 ER* Rabbit monoclonal DCS SP1 1:800 Nuclear 100/226 (44.2) 
16 ERK1/ERK2* Rabbit monoclonal Cell Signaling p44/42 MAPK (20G11) 1:400 Cytoplasmic/nuclear 80/204 (39.2) 
17 ET-1* Mouse monoclonal Alexis Antiendothelin-1 1:200 Cytoplasmic 69/202 (34.2) 
18 ETR-α Sheep polyclonal Alexis ET-α-receptor antiserum 1:200 Cytoplasmic 102/201 (50.7) 
19 ETR-β* Sheep polyclonal Alexis ET-β-receptor antiserum 1:200 Cytoplasmic 138/197 (70.0) 
20 FHIT Rabbit polyclonal Zymed ZR44 1:150 Cytoplasmic 170/215 (79.1) 
21 HER-2* Rabbit polyclonal DAKO c-erbB-2 1:500 Membranous 43/228 (19.0) 
22 MGMT Mouse monoclonal Neomarkers MT3.1 1:250 Cytoplasmic/nuclear 131/213 (61.5) 
23 MIB1/Ki-67* Mouse monoclonal DAKO MIB1 1:1,000 Nuclear 131/226 (58.0) 
24 MUC-1 Mouse monoclonal Novocastra Ma695 1:100 Membranous/cytoplasmic 174/225 (77.3) 
25 p16* Mouse monoclonal Neomarkers 16P07 1:50 Nuclear 78/215 (36.3) 
26 p27 Mouse monoclonal Novocastra 1B4 1:50 Nuclear 96/217 (44.2) 
27 p53* Mouse monoclonal Oncogene DO-1 1:500 Nuclear 67/228 (29.4) 
28 p63* Mouse monoclonal BD PharMingen 4A4 1:200 Nuclear 4/218 (1.8) 
29 PR* Rabbit monoclonal DCS SP2 1:800 Nuclear 86/222 (38.7) 
30 pTEN* Mouse monoclonal Santa Cruz A2B1 1:50 Cytoplasmic 43/220 (19.5) 
31 Synaptophysin Rabbit polyclonal DAKO A0010 1:200 Cytoplasmic 8/221 (3.6) 
32 S6* Rabbit polyclonal Cell Signaling S6 ribosomal protein antibody 1:400 Cytoplasmic 169/197 (85.5) 
33 Topo-IIα* Rabbit polyclonal Novocastra  1:200 Nuclear 65/205 (31.7) 
34 Vimentin* Rabbit monoclonal Biogenex V9 1:20,000 Cytoplasmic 26/209 (12.4) 
NumberProteinsAntibodiesOriginCloneDilutionStaining patternPositive/n (%)
ABCA3 Rabbit polyclonal Origin: N. Inagaki, G.G. Wulf  1:800 Cytoplasmic 21/201 (10.4) 
Bcl-2* Mouse monoclonal DAKO 124 1:50 Cytoplasmic 69/217 (31.8) 
BCRP* Mouse monoclonal Chemicon BXP-21 1:100 Cytoplasmic/membranous 85/212 (40.1) 
β-Catenin* Mouse monoclonal BD Transduction 14 1:200 Membranous 133/217 (61.3) 
C-kit* Rabbit polyclonal DAKO Code number A4502 1:200 Cytoplasmic/membranous 27/223 (12.1) 
CK5* Mouse monoclonal Novocastra XM26 1:600 Cytoplasmic 20/205 (9.8) 
CK8* Mouse monoclonal Biogenex C-51 1:5,000 Cytoplasmic 201/216 (93.1) 
CK17* Mouse monoclonal DAKO E3 1:20 Cytoplasmic 20/224 (9.0) 
COX-2 Mouse monoclonal Cayman  1:400 Cytoplasmic 84/215 (39.0) 
10 CXCR4* Mouse monoclonal Zymed 12G5 1:100 Cytoplasmic 108/189 (57.1) 
      Nuclear 22/189 (11.6) 
11 Cyclin D1 Rabbit monoclonal DCS SP4 1:50 Nuclear 116/223 (52.0) 
12 Cyclin E Mouse monoclonal Novocastra 13A3 1:50 Nuclear 23/218 (10.6) 
13 E-cadherin* Mouse monoclonal Novocastra 36B5 1:50 Membranous 166/216 (76.9) 
14 EGFR* Mouse monoclonal Merck E30 1:100 Cytoplasmic/membranous 31/230 (13.8) 
15 ER* Rabbit monoclonal DCS SP1 1:800 Nuclear 100/226 (44.2) 
16 ERK1/ERK2* Rabbit monoclonal Cell Signaling p44/42 MAPK (20G11) 1:400 Cytoplasmic/nuclear 80/204 (39.2) 
17 ET-1* Mouse monoclonal Alexis Antiendothelin-1 1:200 Cytoplasmic 69/202 (34.2) 
18 ETR-α Sheep polyclonal Alexis ET-α-receptor antiserum 1:200 Cytoplasmic 102/201 (50.7) 
19 ETR-β* Sheep polyclonal Alexis ET-β-receptor antiserum 1:200 Cytoplasmic 138/197 (70.0) 
20 FHIT Rabbit polyclonal Zymed ZR44 1:150 Cytoplasmic 170/215 (79.1) 
21 HER-2* Rabbit polyclonal DAKO c-erbB-2 1:500 Membranous 43/228 (19.0) 
22 MGMT Mouse monoclonal Neomarkers MT3.1 1:250 Cytoplasmic/nuclear 131/213 (61.5) 
23 MIB1/Ki-67* Mouse monoclonal DAKO MIB1 1:1,000 Nuclear 131/226 (58.0) 
24 MUC-1 Mouse monoclonal Novocastra Ma695 1:100 Membranous/cytoplasmic 174/225 (77.3) 
25 p16* Mouse monoclonal Neomarkers 16P07 1:50 Nuclear 78/215 (36.3) 
26 p27 Mouse monoclonal Novocastra 1B4 1:50 Nuclear 96/217 (44.2) 
27 p53* Mouse monoclonal Oncogene DO-1 1:500 Nuclear 67/228 (29.4) 
28 p63* Mouse monoclonal BD PharMingen 4A4 1:200 Nuclear 4/218 (1.8) 
29 PR* Rabbit monoclonal DCS SP2 1:800 Nuclear 86/222 (38.7) 
30 pTEN* Mouse monoclonal Santa Cruz A2B1 1:50 Cytoplasmic 43/220 (19.5) 
31 Synaptophysin Rabbit polyclonal DAKO A0010 1:200 Cytoplasmic 8/221 (3.6) 
32 S6* Rabbit polyclonal Cell Signaling S6 ribosomal protein antibody 1:400 Cytoplasmic 169/197 (85.5) 
33 Topo-IIα* Rabbit polyclonal Novocastra  1:200 Nuclear 65/205 (31.7) 
34 Vimentin* Rabbit monoclonal Biogenex V9 1:20,000 Cytoplasmic 26/209 (12.4) 

Abbreviations: BCRP, breast cancer resistance protein; COX-2, cyclooxygenase 2; ET-1, endothelin-1; ETR-α, endothelin receptor-α; ETR-β, endothelin receptor-β; FHIT, fragile histdine triad; pTEN, protein tyrosine phosphatase; Topo-IIα, topoisomerase IIα.

*

Protein markers used for cluster analysis.

Immunohistochemistry scoring. Staining results were assessed by one pathologist (R. Diallo-Danebrock) and were reevaluated randomly by a second pathologist (C. Poremba) without any knowledge about clinical follow-up. A good correlation (96%) was found between the two observers. Any discrepancies were resolved with a multihead microscope by discussion. The scoring for the single marker evaluation was done according to the literature as follows: Bcl-2 (16), c-kit (17), cyclooxygenase-2 (18), cyclin E (19), CXCR4 (20), endothelin-1, endothelin receptor-α, endothelin receptor-β (21), fragile histdine triad (22), p16 (23), and p27 (24) were scored by assigning a proportion score and an intensity score. ER and PR were also scored by a proportion and intensity score according to ref. 25, respectively: Intensity of the staining was as follows: 1, weakly positive; 2, moderately positive; and 3, strongly positive. The proportion of the stained tumor cells were <10%, 1; 10% to 50%, 2; 50% to 80%, 3; >80%, 4. The intensity score and the proportion score were multiplied to give a final score ranging from 0 to 12, designated as negative or positive as follows: score of 0 to 3, negative; score of 4 to 12, positive. ABCA3, BCRP, β-catenin, cyclin D1, MGMT, MIB1, MUC-1, p53, p63, and topoisomerase IIα were considered positive when >10% of the cancer cells unequivocally showed positive staining. CK5, CK8, CK17, E-cadherin, EGFR, protein tyrosine phosphatase, vimentin, synaptophysin, and S6 were scored positive if any specific staining in the carcinoma cells was observed. For HER-2, only membranous staining was scored using the HercepTest protocol, and a score of 3+ was recorded as a positive result (17). Cases with 2+ scores were further evaluated by fluorescence in situ hybridization to evaluate HER-2/neu gene amplification. ERK1/ERK2 staining was considered positive if a moderate or strong nuclear staining was observed in breast cancer cells. Cases with single or few focally positive tumors cells were assessed negative (26).

Fluorescence in situ hybridization analysis of HER-2/neu.HER-2 gene amplification was analyzed by fluorescence in situ hybridization in those cases that had an immunohistochemical HER-2 score of 2+. Hybridization was done on 4-μm-thick paraffin sections using the Oncor/Ventana INFORM HER-2/neu Gene Detection System (Ventana Medical Systems, Frankfurt, Germany). Assay and scoring were strictly done according to standardized methods (17).

Statistical methods

Cluster analysis (K-means).K-means analysis was done to cluster the patients into subgroups according to their protein expression profiles. Clustering was based on the initial scores reflecting the full dynamic range of the data. Five patients were excluded from further analysis because too many scores were missing for technical reasons (loss of tissue cores from the TMA). We applied K-means clustering with K = 5 and Manhattan distance as similarity measure with 25 iterations as implemented in the Genesis Software Package (27). The five clusters could be readily assigned to the four major biological breast cancer types (which are luminal-A, luminal-B, HER-2, and basal-like) and a fifth, not yet defined subtype.

Kaplan-Meier and Cox regression. The end points for comparing both treatment strategies were EFS and OS. The Kaplan-Meier method was used to estimate cumulative survival time probabilities for predefined subgroups of patients. The standard log-rank test was applied to test whether the populations differ on a significant level (P < 0.05). Cox proportional hazards models were fitted to estimate the effects of different covariates, such as histopathologic parameters and marker expression, on the times to hazard of subgroups of patients. Comparisons of the hazard ratios were done with respect to their 95% confidence intervals (95% CI). All covariates were treated as categorized data. The rules that applied for categorization correspond to the scoring results of the different protein markers (positive/negative). All statistical calculations were done with the statistical software package SPSS 12.0G for Windows.

Patients, follow-up, and treatment arms. The treatment groups of the 236 patients were well balanced in terms of characteristics such as age, menopausal status, tumor size, number of lymph nodes affected, hormone-receptor status, and grading (Table 2). A total of 116 patients were randomized to the high-dose treatment arm, and 120 patients were randomized to the dose-dense treatment arm. The median follow-up at the time of analysis was 61.7 months (range, 4.2-121.9 months; HDCT, 67.9 months; DDCT, 56.3 months).

Table 2.

Baseline characteristics of tumor patients according to treatment groups

HDCT, n = 116 (%)DDCT, n = 120 (%)
Age (y)   
    Mean ± SD 48.1 ± 9.0 48.7 ± 8.5 
Menopausal status   
    Pre 63 (54.3) 59 (49.2) 
    Post 48 (41.4) 58 (48.3) 
    Unknown 5 (4.3) 3 (2.5) 
Surgery   
    Breast-conserving therapy 39 (33.6) 47 (39.1) 
Tumor size (cm)   
    Mean ± SD 3.5 ± 2.1 3.4 ± 2.1 
    Median 2.8 3.0 
Positive nodes   
    Mean ± SD 17.3 ± 7.5 16.9 ± 6.9 
    Median 15.5 15.0 
Grading   
    G1 4 (3.4) 11 (9.2) 
    G2 61 (52.6) 66 (55.0) 
    G3 51 (44.0) 43 (35.8) 
Receptor status   
    ER+ 51 (44.0) 49 (40.8) 
    PR+ 40 (34.5) 46 (38.3) 
    ER+ and/or PR+ 63 (54.2) 62 (51.7) 
HDCT, n = 116 (%)DDCT, n = 120 (%)
Age (y)   
    Mean ± SD 48.1 ± 9.0 48.7 ± 8.5 
Menopausal status   
    Pre 63 (54.3) 59 (49.2) 
    Post 48 (41.4) 58 (48.3) 
    Unknown 5 (4.3) 3 (2.5) 
Surgery   
    Breast-conserving therapy 39 (33.6) 47 (39.1) 
Tumor size (cm)   
    Mean ± SD 3.5 ± 2.1 3.4 ± 2.1 
    Median 2.8 3.0 
Positive nodes   
    Mean ± SD 17.3 ± 7.5 16.9 ± 6.9 
    Median 15.5 15.0 
Grading   
    G1 4 (3.4) 11 (9.2) 
    G2 61 (52.6) 66 (55.0) 
    G3 51 (44.0) 43 (35.8) 
Receptor status   
    ER+ 51 (44.0) 49 (40.8) 
    PR+ 40 (34.5) 46 (38.3) 
    ER+ and/or PR+ 63 (54.2) 62 (51.7) 

Event-free survival according to the treatment. The EFS was defined as the interval from the time of the primary treatment to the first locoregional or distant metastatic recurrence. The median EFS at the time of analysis for the DDCT group was 56 months (range, 0-113 months) and, for the HDCT group, was 70 months (range, 0-111 months). A total of 76 events (63.3%) in the DDCT group and 55 events (47.4%) in the HDCT group had been reported. The 5-year EFS estimates were 41% (95% CI, 32-52%) and 62% (95% CI, 53-73%) for the DDCT group and the HDCT group, respectively. There was a significant better EFS (hazard ratio, 0.62; P = 0.008) in the HDCT treatment group.

Overall survival according to the treatment. The OS was taken as the time from the date of the primary treatment to the time of death. The median OS at the time of analysis for the DDCT group was 76 months (range, 4-122 months) and, for the HDCT group, was 87 months (range, 8-119 months). A total of 99 patients had died (39 in HDCT and 60 in DDCT). The 5-year OS estimates were 61% and 76% for the DDCT group (95% CI, 51-72%) and for the HDCT group (95% CI, 68-85%), respectively. There was a significant better OS (hazard ratio, 0.63; P = 0.027) in the HDCT treatment group.

Breast cancer subtypes. Histologic breast cancer subtypes, according the WHO histologic typing (28), included invasive ductal carcinomas (79%), lobular (17%), tubular (3%), and other types (1%). Tumor grade was assigned based on the criteria of Elston and Ellis (29). Tumor samples included 15 (6.4%) well-differentiated, 127 (53.8%) moderately differentiated, and 94 (39.8%) poorly differentiated breast carcinomas. Tumor classification and grading were done independently by two experienced breast pathologists (R. Diallo-Danebrock and C. Poremba).

Protein expression analysis. A set of 34 proteins was studied by immunohistochemistry on TMA containing 236 paraffin-embedded primary cancer specimens from 236 female patients with high-risk breast cancer (>9 axillary lymph nodes involved). A total of 7,437 (92.7%) of a maximum possible number of 8,024 tissue core sections produced interpretable immunostains and were scored as described in the literature. The number of positive cases of the interpretable cases is seen in Table 1. In the cases with an immunohistochemical HER-2 score of 2+ (20 cases; 8.5%), fluorescence in situ hybridization analysis revealed HER-2 gene amplification in four cases, no HER-2 gene amplification in 12 cases, and four uninterpretable cases because of missing cores on the slide. Missing immunostain data were usually caused by loss of a core from the section, less commonly by exhaustion of tumor material or necrosis in the core on deeper sections. Examples of representative immunostaining results of ER-, HER-2–, CK5-, BCRP-, ET1-, and S6-positive cases are shown in Fig. 1. Additionally, to prove that the immunohistochemical results of one single core with a large diameter of 2.0 mm are representative for the whole tumor, we generated triplicate cores of 15 cases and compared the results of five immunohistochemical markers (ER, PR, HER-2, CK5, and MIB1) among each other. There was concordance among the triplicate scores in 97.3%. Thereafter, the mean score of the triplicates was compared with the core result of the main tissue block, and concordance was found in 93.3% (Supplementary Table S1), thus indicating that one large tissue core (2.0 mm diameter) is representative for the whole tissue section. For statistical cluster analysis, five patients were excluded from the study due to either technical reasons (lack of staining) or lack of representative cancer tissue. In summary, breast cancer samples from 231 patients were analyzed in this study.

Identification of protein clusters. The overall expression patterns for the 231 breast cancer samples were analyzed by K-means clustering and using Manhattan distance as similarity measure. A clear separation of cases into distinct groups with adequate linkage distances was achieved by calculating with different numbers and sets of the protein markers. Finally, clustering was attained by using a combination of 24 marker proteins (labeled in Table 1). Based on the immunohistochemical data of these markers, five main prognostically and predictively relevant clusters could be identified (Figs. 2 and 3). These breast cancer subgroups were widely in accordance with the prognostically significant breast cancer subsets identified by cDNA microarray studies (6, 29): luminal-A (27%), luminal-B (12%), HER-2 (21%), basal-like (13%) type, and a thus far unknown group (27%), which we called the “multiple marker negative” (MMN) group characterized mainly by the strong expression of luminal marker CK8, a lower expression of the hormone receptor ER and PR, and the absence of any further specifying markers. As is shown in Fig. 3, luminal-A samples and basal-like samples are separated by the widest distance along the X-axis (first component), reflecting their biological differences. Luminal-A and luminal-B type samples overlap closely, but the cluster center of the luminal-B type samples is closer to the MMN cluster center. HER2 and basal-like tumors appear as distinct entities as well, although there is some overlap in single tumor samples.

Fig. 2.

Graphical representation of the K-means clustering results. Patients are ordered in columns, proteins in rows. Different shades of red, increasing levels of immunohistochemical staining intensity. Gray boxes, missing values. MMN, multiple marker negative.

Fig. 2.

Graphical representation of the K-means clustering results. Patients are ordered in columns, proteins in rows. Different shades of red, increasing levels of immunohistochemical staining intensity. Gray boxes, missing values. MMN, multiple marker negative.

Close modal
Fig. 3.

Three-dimensional representation of the results from the principal components analysis. Immunohistochemical staining scores were used to visualize the relations between the analyzed tumor samples. Coloring scheme was adopted from the K-means clustering results. The closer two samples appear in the three-dimensional space, the more similar are their respective expression patterns.

Fig. 3.

Three-dimensional representation of the results from the principal components analysis. Immunohistochemical staining scores were used to visualize the relations between the analyzed tumor samples. Coloring scheme was adopted from the K-means clustering results. The closer two samples appear in the three-dimensional space, the more similar are their respective expression patterns.

Close modal

Cluster 1 and 2, luminal-A group, n = 61 (27%) and luminal-B group, n = 28 (12%). As shown in Fig. 3, both groups showed broadly similar protein expression patterns. They were in general characterized by moderate to strong expression of luminal markers such as the estrogen and progesterone receptor and cytokeratin 8. Furthermore, both groups are associated with endothelin-1, endothelin receptor β, and S6 ribosomal protein expression. However, luminal-A group showed a relatively frequent expression of Bcl-2 and β-catenin compared with luminal-B group. On the contrary, luminal-B breast cancers are nearly the only group that shows a significant expression of nuclear CXCR4 compared with all other clusters. Furthermore, both clusters are nearly completely negative for the basal markers CK5, CK17, EGFR, and vimentin.

Cluster 3, multiple marker negative group, n = 63 (27%). The MMN cluster group is mainly characterized by the expression of CK8 and S6 ribosomal protein, although this does not represent a special feature because all other clusters also continuously express these markers. Concerning the luminal markers Bcl-2, ER, and PR, MMN should be placed between the almost luminal marker negative HER-2–/basal-type cluster and the luminal-A/luminal-B cluster groups. Furthermore, this group is completely negative for the basal type markers EGFR, c-kit, vimentin, CK5, and CK17 and also shows a low MIB1-associated proliferation rate.

Cluster 4, HER-2 group, n = 48 (21%) and cluster 5, basal-like group, n = 29 (13%). The HER-2 and the basal-like cluster share a similar protein expression pattern except for the basal-like protein markers (CK5, CK17, c-kit, EGFR, vimentin) and HER-2, which are either negative or positive in the corresponding groups. As opposed to HER-2, basal-like breast cancers are completely negative for steroid hormone receptors ER and PR. Furthermore, both clusters differ from the other clusters by the frequent and strong expression of BCRP, p53, topoisomerase IIα, p16, and a high proliferation rate (MIB1).

Correlation of the cluster groups with different clinicopathologic parameters. As is shown in Table 3, patients were nearly always older than 40 years or mostly postmenopausal in all clusters except for the basal-like group, which contained a strikingly greater number of premenopausal women. The HER-2 and the basal-like cluster comprised a greater number of poorly differentiated breast cancers (G3), whereas the luminal-A and luminal-B, as well as the MMN, clusters included the better differentiated breast cancers (G1, G2). The size of the tumors was equally distributed in all cluster groups. By contrast, patients in the MMN and the basal-like cluster had more often more than 20 lymph node metastases when compared with the other clusters. Within the group of high-risk breast cancer patients with at least nine lymph node metastases, patients could be subdivided into more aggressive subgroups characterized by a trend to excessive lymph node metastasis (cutoff level at 20 lymph nodes was made for statistical reasons). Furthermore, a greater portion of patients belonging to the HER-2 and basal-like cluster (50-60%) died of disease than in the other clusters (28-35%). Altogether, the tumors in the HER-2 and the basal-like cluster are more frequently associated with clinicopathologic parameters characteristic for poor outcome, whereas this was contrary in the luminal-A and luminal-B group. The MMN group should be placed between the luminal-A/luminal-B and the HER-2/basal-like prognosis blocks because it shares features with both of them.

Table 3.

Cluster distribution in relation to different clinicopathologic parameters

FeatureLuminal-A, total (%)Luminal-B, total (%)MMN, total (%)HER-2, total (%)Basal-like, total (%)
Age      
    ≤40 y 14 (23.3) 3 (10.7) 6 (9.5) 7 (14.6) 14 (48.3) 
    >40 y 46 (76.7) 25 (89.3) 57 (90.5) 41 (85.4) 15 (51.6) 
Grade      
    1 + 2 42 (70.0) 23 (82.1) 46 (73.0) 18 (37.5) 6 (20.7) 
    3 18 (30.0) 5 (17.9) 17 (27.0) 30 (62.5) 23 (79.3) 
Size      
    ≤2 cm 28 (46.7) 14 (51.9) 26 (42.6) 23 (48.0) 15 (51.7) 
    >2 cm 32 (53.3) 13 (48.1) 35 (57.4) 25 (52.0) 14 (48.3) 
Number of lymph nodes metastases      
    <20 nodes 46 (76.7) 22 (78.6) 38 (60.3) 38 (79.2) 18 (62.1) 
    >20 nodes 14 (23.3) 6 (21.4) 25 (39.7) 10 (20.8) 11 (37.9) 
Death of disease      
    No 41 (68.3) 20 (71.4) 41 (65.1) 23 (47.9) 11 (37.9) 
    Yes 19 (31.7) 8 (28.6) 22 (34.9) 25 (52.1) 18 (62.1) 
FeatureLuminal-A, total (%)Luminal-B, total (%)MMN, total (%)HER-2, total (%)Basal-like, total (%)
Age      
    ≤40 y 14 (23.3) 3 (10.7) 6 (9.5) 7 (14.6) 14 (48.3) 
    >40 y 46 (76.7) 25 (89.3) 57 (90.5) 41 (85.4) 15 (51.6) 
Grade      
    1 + 2 42 (70.0) 23 (82.1) 46 (73.0) 18 (37.5) 6 (20.7) 
    3 18 (30.0) 5 (17.9) 17 (27.0) 30 (62.5) 23 (79.3) 
Size      
    ≤2 cm 28 (46.7) 14 (51.9) 26 (42.6) 23 (48.0) 15 (51.7) 
    >2 cm 32 (53.3) 13 (48.1) 35 (57.4) 25 (52.0) 14 (48.3) 
Number of lymph nodes metastases      
    <20 nodes 46 (76.7) 22 (78.6) 38 (60.3) 38 (79.2) 18 (62.1) 
    >20 nodes 14 (23.3) 6 (21.4) 25 (39.7) 10 (20.8) 11 (37.9) 
Death of disease      
    No 41 (68.3) 20 (71.4) 41 (65.1) 23 (47.9) 11 (37.9) 
    Yes 19 (31.7) 8 (28.6) 22 (34.9) 25 (52.1) 18 (62.1) 

Treatment effects in subgroups defined by cluster analysis.Table 4 shows the results of the univariate analysis for EFS and OS (Cox model), and Fig. 4 shows the Kaplan-Meier curves concerning the five cluster subtypes in relation to the treatment protocols (i.e., it was analyzed whether the treatment effect of HDCT in comparison to DDCT on EFS and OS was different in breast cancer patient subgroups). For a quantification of treatment effects on EFS and OS, the hazard ratios of the distinct treatment arms with 95% CI are displayed. Luminal-A cluster subtype served as zero standard because it could be shown by Kaplan-Meier analysis that this subgroup had no benefit from HDCT (Fig. 4). In the DDCT treatment arm, HER-2 and basal-like cluster patients had a statistically significant poorer EFS and OS (hazard ratio of 3.6 and 3.7 for EFS and 3.2 and 5.4 for OS in the HER-2 and basal-like cluster, respectively) compared with the luminal-A, luminal-B, and MMN cluster. In the HDCT treatment arm, an improvement of the hazard ratio was noted in the HER-2 and basal-like cluster (1.5 and 1.1 for EFS and 2.0 and 2.0 for OS in the HER-2 and basal-like cluster, respectively) compared with the DDCT group. This represents a notable risk reduction for EFS (factors 2.1 and 2.6 for HER-2 and basal-like, respectively) and for OS (factors 1.2 and 3.4 for HER-2 and basal-like, respectively). Therefore, after HDCT, there is no longer a significantly poorer EFS and OS in these subgroups compared with luminal-A and luminal-B cluster, thus indicating a benefit from HDCT for HER-2 and basal-like cluster. These results are partly supported by single marker analysis. Breast cancer patients whose tumors were positive for basal-type markers CK5, CK17, vimentin, and p53 (mostly positive in the HER-2 and basal-like cluster) had significantly worse OS in the DDCT treatment arm compared with patients whose tumors were negative for these markers. By contrast, in the HDCT treatment arm, patients with positive staining of these single markers had an improved hazard ratio and did not show a worse OS compared with patients whose tumors were negative for these markers, thus displaying a benefit from HDCT (data not shown). By contrast, patients in the luminal-A and luminal-B cluster group did not show any benefit from HDCT. The MMN subgroup shows a slight improvement in the hazard ratio after HDCT compared with DDCT for EFS (1.4 in the DDCT and 0.9 in the HDCT treatment arm) and OS (1.3 in the DDCT and 0.9 in the HDCT treatment arm), thus indicating a trend to benefit from HDCT, but this was not statistically significant.

Table 4.

Univariate analysis for EFS and OS (Cox model) concerning the five cluster subtypes in relation to the treatment protocols

GroupEvent-free survival
Overall survival
DDCT
HDCT
DDCT
HDCT
PHazard ratio (95% CI)PHazard ratio (95% CI)PHazard ratio (95% CI)PHazard ratio (95% CI)
Luminal-A —  —  —  —  
Luminal-B 0.544 1.3 (0.55-3.07) 0.304 1.6 (0.64-4.14) 0.649 0.7 (0.25-2.36) 0.996 1.0 (0.27-3.60) 
MMN 0.205 1.4 (0.76-3.55) 0.970 0.9 (0.47-2.03) 0.521 1.3 (0.55-3.16) 0.833 0.9 (0.36-2.26) 
HER-2 0.001 3.6 (1.65-8.18) 0.235 1.5 (0.76-3.05) 0.010 3.2 (1.32-7.75) 0.092 2.0 (0.89-4.67) 
Basal-like 0.001 3.7 (1.68-8.48) 0.841 1.1 (0.37-3.32) 0.000 5.4 (2.26-13.23) 0.222 2.0 (0.64-6.50) 
GroupEvent-free survival
Overall survival
DDCT
HDCT
DDCT
HDCT
PHazard ratio (95% CI)PHazard ratio (95% CI)PHazard ratio (95% CI)PHazard ratio (95% CI)
Luminal-A —  —  —  —  
Luminal-B 0.544 1.3 (0.55-3.07) 0.304 1.6 (0.64-4.14) 0.649 0.7 (0.25-2.36) 0.996 1.0 (0.27-3.60) 
MMN 0.205 1.4 (0.76-3.55) 0.970 0.9 (0.47-2.03) 0.521 1.3 (0.55-3.16) 0.833 0.9 (0.36-2.26) 
HER-2 0.001 3.6 (1.65-8.18) 0.235 1.5 (0.76-3.05) 0.010 3.2 (1.32-7.75) 0.092 2.0 (0.89-4.67) 
Basal-like 0.001 3.7 (1.68-8.48) 0.841 1.1 (0.37-3.32) 0.000 5.4 (2.26-13.23) 0.222 2.0 (0.64-6.50) 
Fig. 4.

Kaplan-Meier curves: OS of the five cluster subgroups and all patients according to the HDCT and DDCT treatment arm. The respective hazard ratios are presented in Table 4.

Fig. 4.

Kaplan-Meier curves: OS of the five cluster subgroups and all patients according to the HDCT and DDCT treatment arm. The respective hazard ratios are presented in Table 4.

Close modal

In the present study using a panel of 24 protein markers, we identified five subtypes of breast cancers with distinct protein expression patterns within high-risk breast cancer patients with extensive lymph node metastasis. Previous gene expression studies have shown that there is considerable diversity among breast cancers in terms of genetic, biological, and clinical characteristics (6, 7, 30). Sorlie et al. (6, 31) were able to consistently identify five distinct subtypes of unselected breast cancers (luminal-A, luminal-B, normal-breastlike, HER-2, and basal-like) that are associated with different clinical outcomes. Patients with luminal type A tumors had a significant better outcome compared with the basal and HER-2 groups, which showed much shorter disease-free time intervals. The existence of these clinically relevant distinct breast cancer subtypes has been broadly confirmed on the protein level in unselected breast cancer cohorts by using TMA technology (9, 10). Previous protein expression profiling studies have reported two to six different numbers of clusters. This variation is likely due to the marker proteins selected for the cluster analysis, the number of breast cancers samples, the representativeness of TMA, and the type of statistical method used. Nevertheless, all the studies were at least able to differentiate into ER-positive and ER-negative, as well as highly proliferative, subgroups, which could be further subdivided in basal-type and HER-2 clusters (9, 10, 3234).

In our analysis, five statistically relevant subgroups comprised the luminal-A and B clusters, HER-2 cluster, basal-like cluster, and a novel cluster called “multiple marker negative” (MMN) cluster. Tumors included in the luminal clusters are mainly steroid hormone receptor positive. The HER-2 cluster is hormone receptor weak/negative, p53 positive, and shows a high proliferation rate (MIB1). The basal-like cluster is HER-2 weak/negative and is characterized by younger age and the expression of basal type proteins CK5 and CK17, as well as the expression of vimentin, EGFR, p53, and a high MIB1 labelling index. Thus far, the features of these breast cancer subtypes are widely in accordance with the major molecular classes of breast cancer that have been previously detected by cDNA microarray and protein expression profiling. By contrast to these studies, the luminal-B cluster does not show a lower expression of the hormone receptors (6), but is characterized by the weak/negative expression of Bcl-2 and a high expression of nuclear CXCR4 compared with luminal-A cluster, respectively. Regarding the clinical outcome, there was no difference between the two luminal subtypes. In the literature, a second luminal cluster could be identified only in one study (34) by protein expression profiling, whereas in most previous immunohistochemical studies, the identification of a second luminal cluster was not successful (9, 32, 33). This discrepancy may be due to differences in the protein marker sets analyzed. The HER-2 and the basal-like cluster had the poorest outcome with the highest numbers of patients who died of their disease compared with the other clusters. This observation is in perfect agreement with previous gene and protein expression analysis studies. In addition, we identified a novel cluster (multiple marker negative), which expresses luminal CK 8 and hormone receptors to a lesser extent than the luminal clusters, but is mainly characterized by the absence/weak expression of the other selected markers. In terms of outcome, this cluster falls between the luminal-A/luminal-B and HER-2/basal-like cluster with the highest number of lymph node metastases and a median death rate. Our MMN cluster does not seem to correspond to the normal breast-like group described by Sorlie et al. (6) based on gene expression profiling because this breast cancer subtype was characterized by strong expression of basal epithelial genes and low expression of luminal epithelial genes.

To our knowledge, this is the first study confirming the main breast cancer subtypes in a cohort of high-risk breast cancer patients with more than nine lymph node metastases using TMA technology. cDNA microarray analysis has formerly shown the existence of five main molecular subtypes in inflammatory breast cancer (35), which belongs to the poor-prognosis breast cancer subtype. Van Laere et al. (35) could show that a significantly higher fraction of inflammatory breast cancer belongs to the HER-2 and basal-like subtypes. This is in contrast to Bertucci et al. (36), where no difference was observed between inflammatory breast cancer and noninflammatory breast cancer. In our study, the proportion of the above-mentioned molecular breast cancer mainly of the luminal, HER-2, and basal-like subtypes within a high-risk breast cancer population is very similar to those reported for less selected populations (34). In conclusion, our findings suggest that the existence and the proportional distribution of distinct breast cancer subtypes as defined by protein expression profiling is maintained more or less independently of the tumor stage.

On the basis of these results, we analyzed the outcome of the five breast cancer subgroups in the two treatment arms, i.e., whether the treatment effects of two cycles (tandem) of HDCT followed by autologous stem cell transplantation are different in the breast cancer subgroups compared with DDCT. The breast cancer patients in the HER-2 and basal-like subgroups revealed a significant better overall and EFS in the HDCT treatment arm compared with the DDCT treatment arm, whereas the luminal subgroups did not benefit from HDCT. Patients in the MMN cluster showed a trend to benefit from HDCT, which might be due to the fact that at least some of these tumors express single markers (e.g., p53, BCRP, and others) that are associated with a benefit from HDCT. These findings seem to be reflected by the study of Rouzier et al. (37), where different sensitivities to preoperative chemotherapy with anthracycline and taxane in different molecular breast cancer classes as defined by gene expression analysis were found. In their study, the HER-2 and the basal-like subgroups had the highest rates of histopathologic complete response, whereas the luminal and normal-breast-like tumors had low pathologic complete response rates. Taken together, Rouzier's and our study suggest that fundamental differences between the different breast cancer subtypes exist regarding their response to various chemotherapeutic agents and dose intensities. This observation has been recently confirmed by in vitro studies. Luminal and basal epithelial cell lines display unique transcriptional responses to the chemotherapeutics 5-fluorouracil and doxorubicin. Moreover, gene expression signatures, identified in the cell lines, also seem differentially expressed in basal and luminal breast cancer tumors. Although 5-fluorouracil and doxorubicin have different targets in chemotherapeutic action, both drugs affected gene expression in all phases of the cell cycle (38). Furthermore, the dose intensity and the rapid cycling of the high-dose courses seem to increase the sensitivity to HDCT significantly at least in the HER-2 and basal-like breast cancer subtypes. Apart from the high expression of proliferation-associated marker MIB1, both clusters are characterized by the higher expression of BCRP, p53, and topoisomerase IIα compared with the luminal clusters. Especially the expression of p53 might play an important role in HDCT response because Kroger et al. (12) reported recently a higher EFS in p53-positive high-risk breast cancer patients after a single course of HDCT. This interactive effect between treatment and p53 is even more pronounced in our cohort that received tandem HDCT with a 2-fold risk reduction for p53-positive breast cancer patients in the high-dose treatment arm. The potential interactive role of a subset of proteins in the HER-2 and basal-like cluster, such as BCRP, p53, and MIB1, in the complex process of chemotherapy response depending on the dose and cycle time of different CHT courses has to be addressed in further studies.

In summary, these results indicate that within a high-risk breast cancer subgroup, the major molecular classes of breast cancer exist in the same proportional distribution and have prognostic relevance as in an unselected breast cancer group. Furthermore, the different breast cancer subtypes not only have different prognoses, but also show distinct responses to tandem HDCT. The patients in the HER-2 and basal-like breast cancer groups showed a significant benefit from HDCT with a better overall survival, whereas patients in the MMN cluster showed only a slight trend to benefit from HDCT. Patients in the luminal-A and luminal-B clusters do not benefit from this chemotherapy regime. Therefore, cluster analysis by protein expression profiling may help to identify in a simple and cost-effective way those patients who will likely benefit from highly toxic HDCT and to facilitate decisions for particular therapy regimens in high-risk breast cancer patients. The distinct biological mechanisms by which this benefit is achieved in HER-2 and basal-like cluster warrants further examination.

Grant support: Jürgen-Manchot-Stiftung, Düsseldorf, Germany.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).

We thank N. Inagaki (Akita, Japan) and Gerald G. Wulf (Göttingen, Germany) for providing the ABCA3 antibody, Guido Reifenberger and Jürgen Felsberg (Düsseldorf, Germany) for MGMT and protein tyrosine phosphatase immunohistochemistry, Claire Feldhoff and Sabine Schneeloch for excellent technical assistance and the great number of pathologists in the institutes of pathology of the participating centers of the West German Study Group AM 01 trial for providing the paraffin tumor blocks for this study.

1
Carter CL, Allen C, Henson DE. Relation of tumor size, lymph node status, and survival in 24,740 breast cancer cases.
Cancer
1989
;
63
:
181
–7.
2
Nitz UA, Mohrmann S, Fischer J, et al. West German Study Group. Comparison of rapidly cycled tandem high-dose chemotherapy plus peripheral-blood stem-cell support versus dose-dense conventional chemotherapy for adjuvant treatment of high-risk breast cancer: results of a multicentre phase III trial.
Lancet
2005
;
366
:
1935
–44.
3
Peters WP, Ross M, Vredenburgh JJ, et al. High-dose chemotherapy and autologous bone marrow support as consolidation after standard-dose adjuvant therapy for high-risk primary breast cancer.
J Clin Oncol
1993
;
11
:
1132
–43.
4
Rodenhuis S, Bontenbal M, Beex LV, et al. Netherlands Working Party on Autologous Transplantation in Solid Tumors. High-dose chemotherapy with hematopoietic stem-cell rescue for high-risk breast cancer.
N Engl J Med
2003
;
349
:
7
–16.
5
Perou CM, Sorlie T, Eisen MB, et al. Molecular portraits of human breast tumours.
Nature
2000
;
406
:
747
–52.
6
Sorlie T, Perou CM, Tibshirani R, et al. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications.
Proc Natl Acad Sci U S A
2001
;
98
:
10869
–74.
7
van't Veer LJ, Dai H, van de Vijver MJ, et al. Gene expression profiling predicts clinical outcome of breast cancer.
Nature
2002
;
415
:
530
–6.
8
Bubendorf L, Nocito A, Moch H, Sauter G. Tissue microarray (TMA) technology: miniaturized pathology archives for high-throughput in situ studies.
J Pathol
2001
;
195
:
72
–9.
9
Makretsov NA, Huntsman DG, Nielsen TO, et al. Hierarchical clustering analysis of tissue microarray immunostaining data identifies prognostically significant groups of breast carcinoma.
Clin Cancer Res
2004
;
10
:
6143
–51.
10
Jacquemier J, Ginestier C, Rougemont J, et al. Protein expression profiling identifies subclasses of breast cancer and predicts prognosis.
Cancer Res
2005
;
65
:
767
.
11
Somlo G, Simpson JF, Frankel P, et al. Predictors of long-term outcome following high-dose chemotherapy in high-risk primary breast cancer.
Br J Cancer
2002
;
87
:
281
–8.
12
Kroger N, Milde-Langosch K, Riethdorf S, et al. Prognostic and predictive effects of immunohistochemical factors in high-risk primary breast cancer patients.
Clin Cancer Res
2006
;
12
:
159
–68.
13
Tsutsui S, Ohno S, Murakami S, Kataoka A, Kinoshita J, Hachitanda Y. Prognostic significance of the combination of biological parameters in breast cancer.
Surg Today
2003
;
33
:
151
–4.
14
Poremba C, Heine B, Diallo R, et al. Telomerase as a prognostic marker in breast cancer: high-throughput tissue microarray analysis of hTERT and hTR.
J Pathol
2002
;
198
:
181
–9.
15
Iwao K, Matoba R, Ueno N, et al. Molecular classification of primary breast tumors possessing distinct prognostic properties.
Hum Mol Genet
2002
;
11
:
199
–206.
16
Bhatavdekar JM, Patel DD, Shah NG, et al. Prognostic significance of immunohistochemically localized biomarkers in stage II and stage III breast cancer: a multivariate analysis.
Ann Surg Oncol
2000
;
7
:
305
–11.
17
Diallo R, Rody A, Jackisch C, et al. C-KIT expression in ductal carcinoma in situ of the breast: co-expression with HER-2/neu.
Hum Pathol
2006
;
37
:
205
–11.
18
Wulfing P, Diallo R, Muller C, et al. Analysis of cyclooxygenase-2 expression in human breast cancer: high throughput tissue microarray analysis.
J Cancer Res Clin Oncol
2003
;
129
:
375
–82.
19
Donnellan R, Kleinschmidt I, Chetty R. Cyclin E immunoexpression in breast ductal carcinoma: pathologic correlations and prognostic implications.
Hum Pathol
2001
;
32
:
89
–94.
20
Salvucci O, Bouchard A, Baccarelli A, et al. The role of CXCR4 receptor expression in breast cancer: a large tissue microarray study.
Breast Cancer Res Treat
2006
;
97
:
275
–83.
21
Wulfing P, Diallo R, Kersting C, et al. Expression of endothelin-1, endothelin-A, and endothelin-B receptor in human breast cancer and correlation with long-term follow-up.
Clin Cancer Res
2003
;
9
:
4125
–31.
22
Ginestier C, Bardou VJ, Popovici C, et al. Loss of FHIT protein expression is a marker of adverse evolution in good prognosis localized breast cancer.
Int J Cancer
2003
;
107
:
854
–62.
23
Han S, Ahn SH, Park K, et al. P16INK4a protein expression is associated with poor survival of the breast cancer patients after CMF chemotherapy.
Breast Cancer Res Treat
2001
;
70
:
205
–12.
24
Schondorf T, Gohring UJ, Becker M, et al. High apoptotic index correlates to p21 and p27 expression indicating a favorable outcome of primary breast cancer patients, but lacking prognostic significance in multivariate analysis.
Pathobiology
2004
;
71
:
217
–22.
25
Remmele W, Stegner HE. Recommendation for uniform definition of an immunoreactive score (IRS) for immunohistochemical estrogen receptor detection (ERICA) in breast cancer tissue.
Pathology
1987
;
8
:
138
–40.
26
Milde-Langosch K, Bamberger AM, Rieck G, et al. Expression and prognostic relevance of activated extracellular-regulated kinases (ERK1/2) in breast cancer.
Br J Cancer
2005
;
92
:
2206
–15.
27
Sturn A, Quackenbush J, Trajanoski Z. Genesis: Cluster analysis of microarray data.
Bioinformatics
2002
;
18
:
207
–8.
28
Tavassoli AE, Devilee P. World Health Organization classification of tumours. Pathology and genetics of tumours of the breast and female genital organs. Lyon: IARC Press; 2003.
29
Elston CW, Ellis IO. Pathological prognostic factors in breast cancer. I. The value of histological grade in breast cancer: experience from a large study with long-term follow-up.
Histopathology
1991
;
19
:
403
–10.
30
van de Vijver MJ, He YD, van't Veer LJ, et al. A gene-expression signature as a predictor of survival in breast cancer.
N Engl J Med
2002
;
347
:
1999
–2009.
31
Sorlie T, Tibshirani R, Parker J, et al. Repeated observation of breast tumor subtypes in independent gene expression data sets.
Proc Natl Acad Sci U S A
2003
;
100
:
8418
–23.
32
Zhang DH, Salto-Tellez M, Chiu LL, Shen L, Koay ES. Tissue microarray study for classification of breast tumors.
Life Sci
2003
;
73
:
3189
–99.
33
Korsching E, Packeisen J, Agelopoulos K, et al. Cytogenetic alterations and cytokeratin expression patterns in breast cancer: integrating a new model of breast differentiation into cytogenetic pathways of breast carcinogenesis.
Lab Invest
2002
;
82
:
1525
–33.
34
Abd El-Rehim DM, Ball G, Pinder SE, et al. High-throughput protein expression analysis using tissue microarray technology of a large well-characterised series identifies biologically distinct classes of breast cancer confirming recent cDNA expression analyses.
Int J Cancer
2005
;
116
:
340
–50.
35
Van Laere SJ, Van den Eynden GG, Van der Auwera I, et al. Identification of cell-of-origin breast tumor subtypes in inflammatory breast cancer by gene expression profiling.
Breast Cancer Res Treat
2006
;
95
:
243
–55.
36
Bertucci F, Finetti P, Rougemont J, et al. Gene expression profiling identifies molecular subtypes of inflammatory breast cancer.
Cancer Res
2005
;
65
:
2170
–8.
37
Rouzier R, Perou CM, Symmans WF, et al. Breast cancer molecular subtypes respond differently to preoperative chemotherapy.
Clin Cancer Res
2005
;
11
:
5678
–85.
38
Troester MA, Hoadley KA, Sorlie T, et al. Cell-type-specific responses to chemotherapeutics in breast cancer.
Cancer Res
2004
;
64
:
4218
–26.

Supplementary data