BCIRG 001 Molecular Analysis: Prognostic Factors in Node-Positive Breast Cancer Patients Receiving Adjuvant Chemotherapy

Adjuvant polychemotherapy for women with early-stage breast cancer has meaningfully improved disease-free survival (DFS) and overall survival (OS; ref. 1). Although anthracycline-based chemotherapy has produced a modest improvement in DFS and OS over CMF (cyclophosphamide/methotrexate/fluorouracil) polychemotherapy regimens (1), this benefit seems to be largely confined to the subset of patients with amplification of the erb-B2/neu or topoisomerase II α genes or overexpression of the corresponding proteins (2–6). The microtubule-stabilizing taxanes paclitaxel and docetaxel are among the most active single agents in women with advanced breast cancer (7). Because taxanes exhibit incomplete cross-resistance with anthracyclines, and adjuvant anthracycline chemotherapy still leaves considerable residual risk of disease recurrence and death from breast cancer, multiple trials are addressing the value of incorporating taxanes, either sequentially or in combination with anthracyclines, in an effort to improve outcomes (8, 9). The Breast Cancer International Research Group (BCIRG) 001 trial compared the standard fluorouracil-doxorubicin-cyclophosphamide (FAC) regimen to the docetaxel-doxorubicin-cyclophosphamide (TAC) regimen as adjuvant chemotherapy in 1,491 patients with axillary node-positive breast cancer. With a median follow-up of 55 months, a prospectively defined and event-triggered analysis showed that TAC was significantly superior to FAC, both in terms of DFS (75% versus 68% at 5 years) and in OS (87% versus 81% at 5 years; ref. 10).

Evaluation within randomized trials is required to distinguish the predictive value of a marker (its relationship to outcome specifically attributable to a given intervention) from its prognostic value (the relationship to outcome irrespective of treatment; ref. 11). Several candidate predictive markers have been explored in patients receiving taxane therapy for metastatic disease, but none has been fully validated or incorporated into standard practice (12–14). Similarly, markers have been suggested to distinguish those patients who might particularly benefit from adjuvant taxane therapy, in particular estrogen receptor negativity (15, 16), or human epidermal growth factor receptor 2 (HER-2) status (12, 17). Recent immunohistochemical analyses of tissue specimens from 1,322 node-positive breast cancer patients treated with doxorubicin/cyclophosphamide followed by paclitaxel therapy or observation revealed little benefit from additional paclitaxel for HER2-negative, estrogen receptor–positive individuals (18). In an analysis of immunohistochemically defined molecular subtypes using estrogen receptor, progesterone receptor, Her2, and Ki-67 expression, we recently reported that TAC was associated with significantly longer DFS in patients with luminal B subtype (estrogen receptor–positive and/or progesterone receptor–positive and either HER2-positive and/or Ki67-high), but not in those women with luminal A disease (estrogen receptor–positive and/or progesterone receptor–positive and not HER2-positive or Ki67-high; ref. 19).

The BCIRG 001 protocol included centralized assessment of tumor tissue for immunohistochemical assessment of estrogen receptor and HER-2 amplification status determined by fluorescence in situ hybridization (10). We studied immunohistochemical markers including proteins involved in the regulation of apoptosis, cell cycling, proliferation, as well as the microtubule cytoskeleton. A statistical analysis taking into account classical prognostic factors as well as these immunohistochemical parameters was done to determine the potential prognostic and predictive value of these markers in relation to DFS and OS.

In 1997, the BCIRG began a phase III trial to compare the docetaxel-containing regimen TAC with a regimen of FAC as adjuvant treatment for women with operable node-positive breast cancer. In this multicenter, registration study, randomization was stratified according to institution and number of involved axillary lymph nodes per patient (one to three versus four or more). On day 1 of each of six 21-day cycles, eligible patients received either TAC (50 mg of doxorubicin per square meter of body-surface area, 500 mg of cyclophosphamide per square meter, and 75 mg of docetaxel per square meter) or FAC (50 mg of doxorubicin per square meter, 500 mg of fluorouracil per square meter, and 500 mg of cyclophosphamide per square meter).

The primary end point was disease-free survival, defined as the time from randomization to the date of a clinical relapse (with histopathologic confirmation or radiologic evidence of tumor recurrence), a second cancer (with the exceptions of skin cancer other than melanoma, ductal or lobular carcinoma in situ of the breast, or in situ carcinoma of the cervix), or death, whichever occurred first; overall survival was a secondary end point. The protocol defined centralized assessment of tumor tissue for immunohistochemical assessment of biological parameters.

As per the study protocol and with prospective patient consent, slides containing pathologic material and a representative formalin-fixed paraffin block from the primary tumor were requested from each patient's primary institution for central review and analyses. These materials were obtained for 1,350 (91%) of the 1,491 randomized trial patients. All tumor samples were derived from primary breast tumors prior to chemotherapy exposure.

Fourteen pathologic and molecular markers were measured at baseline in each randomized patient. To avoid problems of interobserver variability, all pathologic material relative to a given marker was reviewed by one of the three pathologists in the three participating central laboratories (Cross Cancer Center, Edmonton, Canada; Burnham Institute for Medical Research, La Jolla, CA; Centre Léon Bérard, Lyon, France). Data included distribution and percentage of positive cells for the following parameters: estrogen receptor and progesterone receptor (6F11, Ventana, dilution 1:50 and 636; Dako, dilution 1:100 respectively); Ki-67 (MIB-1 clone, Dako, dilution 1:40); p53 (1801, Novocastra, dilution 1:100); Bcl-2, Bax, Bcl-X (polyclonal antibodies generated by J. Reed, dilutions 1:2,000, 1:2,000, and 1:3000 respectively with high pH retrieval); Bag-1 (cytosolic and nuclear; clone KS-6C8, dilution 1:100 with high pH retrieval); MUC 1 (cytoplasmic luminal, cytoplasmic diffuse, membrane intercellular, membrane peri-aggregate, overall; E29, Dako, dilution 1:50); β tubulin isotype II (clone 7B9 generously provided by A. Frankfurter, dilution 1:50); β tubulin isotype III (clone Tuj1, generously provided by A. Frankfurter, produced by Covalab, Lyon, France; dilution 1:50); β tubulin isotype IV (Biogenex, dilution 1:100); τ protein (Boehringer, dilution 1:40); and glu α tubulin (produced by L. Lafanechère, dilution 1:30). HER2/neu status was considered positive if the HER2/CEP17 ratio was >2 or, if the HER2/CEP17 ratio was not available, DAKO Herceptest intensity was equal to 3+ or, if both HER2/CEP17 ratio and DAKO intensity were unavailable, CB11 intensity was equal to 3+.

The aim of this biological study was to identify biological prognostic and/or predictive factors in a large series of patients receiving adjuvant therapy for node-positive breast cancer. Among the 1,491 randomized patients, 141 (9.5%) had no data on any of the main 16 tumor markers as primary tumor samples were unavailable for analysis. These 141 patients were excluded, leaving a total of 1,350 patients for the present analysis.

The following variables were analyzed: patient characteristics (age, menopausal status, Karnofsky performance status), tumor characteristics (unifocal or multifocal tumor, tumor size, nuclear grade, architectural grade, mitotic grade, overall histologic grade, estrogen receptors, progesterone receptors, vascular invasion, and histologic subtype), lymph node status (micrometastases or macrometastases to resected axillary lymph nodes, number of positive axillary nodes, extra nodal disease), and biological markers HER2, MIB-1, P53, Bax, Bcl2, BAG-1 (expressed in nucleus or cytosol), MUC1, Bcl X_L, tubulin class II, tubulin class III, tubulin class IV, τ protein, and glu α tubulin. Immunohistochemical parameters were analyzed as variables dichotomized at the median value observed in the population. Her2 status was dichotomized as positive or negative. The interaction test between each parameter and treatment arm was done by testing the logarithm of the ratio of two hazard ratios.

Survival curves and probabilities were estimated using the Kaplan-Meier technique. Differences between survival curves were assessed using the log-rank test. The Cox proportional hazards regression model was used for both univariate and multivariate analyses of survival. For the analysis of prognostic factors for survival analysis the proportionality assumption was checked for each of the variables under study by testing the dependency of their hazard ratio over time.

Those covariates of significance identified by univariate analysis (P < 0.10) were included in a multivariate Cox model with four different selection methods to identify the final model of predictive factors. These selection methods (forward selection, backward elimination, stepwise selection, and score for best subsets selection) all led to the same final multivariate Cox model. Thus, only results obtained with the backward elimination procedure are presented. Multivariate analyses were done to validate the previous findings. First-order interaction terms between treatment arms and tumor markers were included in the analysis to identify potential effect modification. In the absence of interaction, exploratory analyses on treatment subgroups were done. As a final step, a bootstrap analysis (20) with resubstitution of all individual samples was done to confirm the independent parameters identified in the multivariate analyses or the exploratory analyses.

Patient characteristics for the 1,350 patients included in the immunohistochemical analysis are presented in Table 1. Data regarding the 9% of patients not included in this analysis were not available. The tumor marker characteristics are shown in Table 2 and Supplementary Fig. S1. As previously reported, with a median follow-up of 55 months, the estimated rates of DFS at 5 years were 75% in the TAC group and 68% in the FAC group (10). The OS at 5 years was 87% in the TAC group and 81% in the FAC group. There were 358 DFS events (201 and 157 events in FAC and TAC arms, respectively). There were 199 deaths, with 115 and 84 in the FAC and the TAC arms, respectively. The distribution of samples available in each treatment arm is shown in Supplementary Fig. S2.

In univariate analyses, treatment arm, tumor size (>2.3 cm), mixed histology, estrogen receptor expression ≤80% (cut-off value chosen due to the median value in the entire cohort), progesterone receptor expression ≤80%, histologic grade 3, presence of extranodal disease, >2 involved axillary lymph nodes, high Ki-67 expression, high p53 expression, low MUC1 expression, high tubulin III, low levels of τ protein, and high glu tubulin were significantly associated with shorter DFS (Table 3). There was no significant interaction between the expression level of any of the markers and the effect of treatment arm on DFS.

In multivariate analysis, treatment arm, estrogen and progesterone receptors, number of involved nodes, Ki-67, and τ protein expression were independently associated with DFS (Table 4). These parameters were confirmed in the bootstrap analysis. In addition, the first-order interaction between treatment arms, Ki-67, and τ protein expression were assessed using a Cox model. There were no significant interactions (ratio of hazard ratios for interaction treatment arms and Ki-67, 1.29; 95% confidence interval, 0.77-2.18; P = 0.34; ratio of hazard ratios for interaction treatment arms and τ protein, 0.75; 95% confidence interval, 0.45-1.25; P = 0.27). Therefore, there was no reason to suspect differences in the treatment effect of TAC and FAC for the Ki-67 and τ protein subgroups. DFS curves according to τ protein and Ki-67 levels are shown in Fig. 1A and B, respectively.

In univariate analyses, tumor size (>2.3 cm), mixed histology, estrogen receptor expression ≤80%, progesterone receptor expression ≤80%, histologic grade 3, >2 involved axillary lymph nodes, high Ki-67 expression, high p53 expression, low BCL-2 expression, high BCL-XL, HER2 positivity, high tubulin III, low levels of τ protein, and high glu tubulin were significantly associated with shorter OS (Table 5). There was no significant interaction between the expression level of any of the markers and the effect of treatment arm on OS.

In multivariate analysis, estrogen and progesterone receptors, number of nodes, p53, tubulin III, and MIB-1 protein expression were independently associated with OS. These parameters were confirmed in the bootstrap analysis. In addition, the first-order interaction between treatment arms and p53 or Ki-67 or tubulin III protein expression was assessed through a Cox model. There were no significant interactions with treatment arm (data not shown). Therefore, there was no reason to suspect differences in the treatment effects of TAC and FAC for p53, Ki-67, or tubulin III subgroups. The OS in the entire patient population according to Ki-67, tubulin III, and p53 levels are shown in Fig. 1.

The BCIRG 001 study was a pivotal trial showing the benefit of incorporating a taxane into adjuvant breast cancer polychemotherapy (10). However, the substitution of docetaxel for fluorouracil increased toxicities, and not all patients remained disease free. Consequently, it was our objective to determine which patients are most likely to benefit from adjuvant therapy with docetaxel. The identification of such molecular predictive factors is, however, complex because most markers correlated with outcome most frequently are prognostic rather than predictive (i.e., related to prognosis only in the setting of treatment with a specific intervention; refs. 11, 21), and the mechanistic determinants of docetaxel response are poorly understood. Evaluation within the context of a randomized trial, such as BCIRG 001, is required to distinguish between the prognostic and the predictive value of a given biomarker.

In this immunohistochemical study, Ki-67 protein was found to be independently associated with both DFS and OS; τ protein was associated with DFS, whereas p53 and tubulin III were associated with OS, even when taking into account classical factors such as hormone receptors, tumor size, extent of nodal involvement, or mitotic grade. These correlations, however, were found both in patients receiving TAC or FAC, and interaction testing revealed no qualitative differences within treatment subgroups, indicating that these biological parameters were not predictive of sensitivity to either regimen.

Ki-67 is correlated with proliferation and has been shown to be correlated with the prognosis of breast cancer patients in several settings, including node-negative disease and in metastatic patients (22–25). Ki-67 correlates with the growth fraction or S-phase fraction and can be evaluated on small samples in which other methods are not applicable (26). We had previously reported that a large majority of triple-negative samples (187 of 192, or 97%) had higher than median expression of Ki-67 (19); similarly, we and others have reported that Ki-67 immunohistochemical staining can be used to differentiate the luminal A and luminal B breast cancers (19, 27). Our results thus show that Ki-67 is independently correlated both with DFS and with OS in patients with node-positive breast cancer, and differs significantly between molecular subtypes. Our findings are also consistent with a recent observation from the Protocole Adjuvant de Cancer du Sein (PACS-01) study of adjuvant docetaxel, in which Ki-67 immunohistochemistry was prognostic in a large series of patients with node-positive early breast cancer, but did not show a statistically significant interaction for the differential treatment effect between the chemotherapy arms (28).

Microtubule-related proteins have been reported to be involved in mechanisms of resistance to tubulin binding agents such as taxanes (29). Insofar as taxanes induce apoptotic cell death by reducing microtubule dynamics, rather than by altering the gross mass of microtubule polymer present in the tumor cells (30), microtubule proteins, such as tubulin, or microtubule associated proteins might play a key role in determining sensitivity to these agents. Tubulin III isotype has been repeatedly reported to correlate with response and outcome in patients receiving taxane-based regimens for lung cancer or ovarian cancer (31, 32). We recently reported that docetaxel was more active than doxorubicin in women with metastatic breast cancers with a high level of tubulin III isotype (33). The level of expression of glu tubulin, a posttranslationally modified form of α tubulin correlated with microtubule stability, has previously been reported to be correlated with tumor aggressiveness in breast cancer patients (34).

In this study we show that high τ protein levels in primary tumors are associated with better outcome in patients receiving adjuvant chemotherapy for node-positive breast cancer, irrespective of the type of chemotherapy administered. This finding is concordant with several other recent reports. Penderouthakis et al. analyzed τ mRNA content in 274 early breast cancer samples and found that high levels were associated with a reduced risk of relapse (35). Veitia et al. reported that τ isoforms were higher in human-xenografts sensitive to docetaxel (36). Several authors have suggested that τ protein was correlated to sensitivity to taxanes, epothilones, or to hormone therapy (36, 37). Pusztai et al. analyzed the level of τ protein by immunohistochemistry in 1,942 patients treated in the National Surgical Adjuvant Breast and Bowel Project (NSABP-B28) trial and found that high τ protein expression was associated with better prognosis, including longer DFS and OS in patients treated with adjuvant anthracyline and paclitaxel chemotherapy and endocrine therapy (38). However, a study by Rouzier et al. used pangenomic transcription chips and immunohistochemistry to show that low τ expression was associated with pathologic complete response in breast cancer patients receiving preoperative paclitaxel-containing chemotherapy (39). To determine whether the differences between the results in the Rouzier study and the present study were due to the choice of antibody, we obtained the T1029 monoclonal antibody used in the Rouzier study (United States Biological, Swampscott, MA) and found no meaningful discordance in staining intensity or pattern compared with our Boeringher anti-τ antibody. A possible explanation for the apparent discrepancy between the results of our own study and the Rouzier study is that the latter did not report correlations with relapse-free or overall survival. Pathologic complete responses in neoadjuvant taxane trials do not necessarily correlate with these more important end points, as shown in the NSABP B-28 study (40), with higher pathological complete response rates in the group with the overall worst prognosis. Low τ protein expression levels could therefore be correlated both with a higher pathologic response rate and a worse overall prognosis in this setting. Another hypothesis is the existence of a qualitative difference between docetaxel and paclitaxel, which is unlikely because the NSABP adjuvant paclitaxel study showed τ-related prognostic signals concordant with our own, but this possibility cannot be ruled out in the absence of a comparative study (41).

In our current study, higher than median p53 protein staining was also found to confer a poor prognosis on univariate analysis (hazard ratio, 1.31; P = 0.039). This finding was independently correlated with both DFS and OS in multivariable analysis taking into account classical prognostic factors. Thus, women with tumors with higher than median p53 expression showed an independently higher risk of relapse (multivariate hazard ratio, 1.47; P = 0.0094).

The relationship between p53 status and sensitivity to taxanes has been previously investigated with conflicting results. Preclinical studies relating p53 status and sensitivity to paclitaxel have shown increased sensitivity in lines having lost normal p53 function after transformed with SV40T antigen, decreased sensitivity after DNA-damaged induction of p53, and decreased sensitivity in the absence of wild-type p53 activity (42–44). Immunopositivity for p53 protein is most frequently, but not always, due to mutated p53 (45). Several smaller immunohistochemical studies of breast cancers from women receiving taxane-based regimens have shown no significant correlation between p53 expression and outcome in patients receiving taxane-based therapy. In a series of 144 metastatic breast cancer patients receiving single-agent taxane, p53, HER2, epidermal growth factor receptor, and p27 were not correlated with outcome (46). In a series of 114 patients receiving doxorubicin or paclitaxel for metastatic disease, p53, HER2, and Bcl-2 were not correlated with clinical response (47). In a series of 60 patients receiving preoperative docetaxel, Tiezzi et al. analyzed hormone receptors, p21, HER2, and p53, and found that only HER2 was associated with outcome in this series (48). In our study of 1,350 patients, p53 was correlated both with DFS and with OS in both treatment arms, suggesting that sensitivity to docetaxel is not dependent on p53 status.

This molecular analysis of BCIRG 001 confirmed the prognostic value of several classical histopathologic assessments in early breast cancer, including number of axillary nodes, tumor size, estrogen receptor status, and progesterone receptor status. Within the context of our trial, we did not find independent prognostic value for MUC1, Bcl-2, Bax, Bag1, tubulin II, or tubulin IV dichotomized at the median value. High Bcl-2 levels were correlated with longer DFS and OS in univariate analysis, but this was not confirmed in the multivariate analysis.

In an exploratory subgroup analysis we determined the distribution of the four most promising exploratory markers (τ protein, tubulin III, Ki 67, and p53) in the molecular subgroups previously described (triple negative, Her 2, luminal A, and luminal B; Supplementary Table S1). These data show that the distribution of each of the markers is significantly different between subgroups, in particular with low τ protein and high tubulin III expression in the triple-negative subgroup and high τ protein and low tubulin III expression in the luminal A subgroup.

The strengths of this study include the assessment of novel molecular markers within a mature randomized study, the large number of samples studied, the high proportion of study subjects for whom tissue was available for analysis (91%), and the prospectively defined nature of this analysis. Limitations of the study include the use of median cut points for most immunohistochemical variables, which may not fully reflect the intrinsic biology of the disease.

In conclusion, our results suggest that Ki-67 and p53 proteins are key independent prognostic factors in node-positive breast cancer patients receiving adjuvant chemotherapy, and that this prognostic property is independent of the type of treatment administered. Microtubule-related parameters such as tubulin III and τ protein also carry independent prognostic value in this setting. These tumor-associated markers can therefore be used to further define subpopulations of node-positive operable breast cancer with different prognoses.

C. Dumontet: consultant/advisory board, Sanofi.

Grant Support: Sanofi-Aventis.

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.