The Prognostic Value of BCAR1 in Patients with Primary Breast Cancer

6194 Vol. 10, 6194–6202, September 15, 2004 Clinical Cancer Research

Lambert C. J. Dorssers,1 When added together with uPA and PAI-1 in the multiva- Nicolai Grebenchtchikov,2 Arend Brinkman,1 riate model, BCAR1 contributed independently of PAI-1 Maxime P. Look,3 Simone P. J. van Broekhoven,1 and was favored over uPA. Interaction tests allowed for 1 3 additional analyses of BCAR1 protein levels in clinically Danielle de Jong, Harry A. Peters, relevant subgroups stratified by nodal and menopausal 3 3 Henk Portengen, Marion E. Meijer-van Gelder, status. Jan G. M. Klijn,3 Doorlene T. H. van Tienoven,2 Conclusions: The quantitative BCAR1 protein level Anneke Geurts-Moespot,2 Paul N. Span,2 represents a prognostic factor for RFS and OS in primary John A. Foekens,3 and Fred C. G. J. Sweep2 breast cancer, independent of the traditional prognostic factors and the other novel marker PAI-1. 1Department of Pathology, Division of Molecular Biology, Erasmus MC Rotterdam, Rotterdam; 2Department of Chemical Endocrinology, University Medical Center Nijmegen, Nijmegen; and 3Department INTRODUCTION of Medical Oncology, Erasmus MC Rotterdam, Rotterdam, the Netherlands Breast cancer will affect one in ten women in the Western world and 5-year survival may be expected for only ϳ60% of the patients using current treatments. The heterogeneity of the ABSTRACT disease additionally complicates the identification of patients Purpose: BCAR1, the human homologue of the rat with high risk for recurrence and disease progression and for p130Cas protein, was identified in a functional screen for sensitivity to systemic treatment. Many research efforts are human breast cancer cell proliferation resistant to anties- focused on the identification of suitable markers for therapy trogen drugs. Here, we study the prognostic value of quan- selection for different risk groups of patients. titative BCAR1 levels in a large series of breast cancer In a functional screen for genes involved in breast cancer specimens. resistance to antiestrogenic drugs, we previously reported the Experimental Design: A specific ELISA was developed identification of the BCAR1 gene, which induces estrogen inde- to measure BCAR1 protein levels in 2593 primary breast pendence in estrogen-dependent human breast cancer cells (1). tumor cytosols. Tumor levels of BCAR1 were correlated BCAR1 is the human homologue of the rat p130Cas gene, a with relapse-free survival (RFS) and overall survival (OS) prominent tyrosine phosphorylation target in transformed rat and compared with collected data on urokinase-type plas- cells (2). Additional studies have shown that BCAR1/p130Cas minogen activator (uPA) and plasminogen activator inhibi- is a cytoplasmic adaptor protein, capable of associating with tor 1 (PAI-1). many different proteins and contributing to various cellular Results: In tumor cytosols, BCAR1 protein levels varied processes, including adhesion, migration, cytoskeletal reorgani- between 0.02 and 23 ng/mg protein. BCAR1 levels exhibited zation, host-pathogen interaction, survival, proliferation, trans- a positive correlation with steroid hormone receptor levels, formation, and possibly metastasis (for review, refs. 3, 4). On age and menopausal status, and uPA and PAI-1 levels. The the basis of semiquantitative Western blot detection of BCAR1 level of BCAR1 (continuous or categorized as low, interme- (using an antibody directed against rat p130Cas) in 775 breast diate, or high) was inversely related with RFS and OS time. tumor specimens, we previously showed that high levels of Multivariate analysis showed that BCAR1 levels contributed BCAR1 were significantly associated with early disease recur- independently to a base model containing the traditional rence and with poor response to tamoxifen treatment of recur- prognostic factors for both RFS and OS (both P < 0.0001). rent disease (5, 6). Because of the limitations of the Western blot assay, an extended analysis of the role of BCAR1 in breast cancer was hampered. The development of a quantitative BCAR1-ELISA (7) now offers the opportunity to address this question. Here, we report the prognostic value of BCAR1 pro- Received 3/5/04; revised 5/25/04; accepted 6/4/04. tein levels in a large series of 2593 patients with primary breast Grant support: Dutch Cancer Society Grant DDHK 2000-2256, Asso- ciation for International Cancer Research Grant 00-38, and Revolving cancer. Fund of the Erasmus MC Grant 99,760. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked MATERIALS AND METHODS advertisement in accordance with 18 U.S.C. Section 1734 solely to Patients and Tissue Samples. BCAR1 levels were de- indicate this fact. termined in cytosol preparations (as described below) from Correspondence: Lambert C. J. Dorssers, Department of Pathology, 2593 primary invasive breast tumors collected between 1978 Josephine Nefkens Institute, Be 432, Erasmus MC, P. O. Box 1738, 3000 DR, Rotterdam, the Netherlands. Phone: 31-10-4088378; Fax and June 1995. Our study design was approved by the med- 31-10-4088377; E-mail: [email protected]. ical ethical committee of the Erasmus MC (Rotterdam, the ©2004 American Association for Cancer Research. Netherlands). Selection of samples for analysis of relapse-

free survival (RFS) and overall survival (OS) according in European Organization for Research and Treatment of Cancer to the following criteria: at least 5 years of follow-up; no receptor buffer [10 mmol/L K2HPO4 buffer containing 1.5 metastatic disease at diagnosis; no previous diagnosis of carci- mmol/L dipotassium EDTA, 3 mmol/L sodium azide, 10 noma, with the exception of basal skin carcinoma or cervical mmol/L monothioglycerol, and 10% v/v glycerol (pH 7.4)]. The cancer stage 1; no evidence of disease within 1 month of suspension was centrifuged for 30 minutes at 100,000 ϫ g to surgery; and was additionally based on the availability of stored obtain the supernatant fraction (cytosol). ER and PgR levels cytosol extracts (in liquid nitrogen) which remained after rou- were determined by ligand binding assay or with enzyme im- tine estrogen (ER) and progesterone receptor (PgR) analyses. munoassay as described previously (9). The cutoff point used to Inoperable T4 tumors were not included. Tissue specimens that classify tumors as ER or PgR positive and negative was 10 were sampled after neoadjuvant treatment, or obtained from a fmol/mg protein. The cytosolic levels of uPA and PAI-1 were biopsy, were excluded. Patients with distant metastasis at the determined with ELISAs as described before (11, 12). Total time of primary surgery (M patients; staging according to the 1 cytosolic protein was quantified with the Coomassie brilliant International Union against Cancer Tumor-Node-Metastasis blue method (Bio-Rad Laboratories, Hercules, CA) with human classification; ref. 8) were excluded. Median age of the patients at the time of surgery (modified serum albumin as a standard. mastectomy ϭ 1466 patients, breast-conserving treatment ϭ For quantification of BCAR1 in our ELISA, cytosolic ϳ 1127 patients) was 56 years (range, 22 to 90 years). One breast cancer samples containing 1 mg/mL protein were di- thousand sixty-three of them were premenopausal, and 1530 luted 11-fold in PBS containing 1% BSA and 0.1% Tween 20 were postmenopausal at the time of primary surgery. Radiother- and incubated overnight at 4°C with microplate immobilized apy was applied to 1831 patients (71%): on the breast/thoracic (with duck antichicken) capture chicken anti-BCAR1 antibody. wall in 1602 patients and/or on the axilla in 660 patients, After washing and trapping with rabbit anti-BCAR1 antibody, respectively, parasternal and/or supraclavicular lymph nodes in immune complexes were detected using goat antirabbit biotin- Յ ␤ 748 patients. T1 tumors ( 2 cm) were present in 1058 patients streptavidin- -galactosidase (7). Different dilutions of a recom- Ͼ Յ (41%), T2 tumors ( 2to 5 cm) in 1257 patients (48%), T3 binant BCAR1 protein (calibrator) and a pool sample of iden- Ͼ tumors ( 5 cm) in 170 patients (7%), and operable T4 tumors in tically prepared breast cancer cytosols (HR-1) were included in 108 patients (4%). Pathological examination was not performed each series and used for quantification of BCAR1 and adjust- centrally and thus reflects daily practice in the participating ment for day-to-day variation of the assay [HR-1 within and regional hospitals as described previously (9). Histologic differ- between assay coefficient of variation of 3.5% (range, 1.7 to entiation grade was coded as poor in 1379 patients (53%), 5.5%) and 4.5%, respectively]. The functional sensitivity of the moderate in 451 patients (17%), good in 40 patients (2%), and assay was 4 pg/mL (7). unknown for 723 patients (28%). None of the 1311 node- Statistics. The strength of the associations of BCAR1 negative patients (51%) received systemic adjuvant therapy. Of with continuous variables was tested with Spearman rank the 1282 node-positive patients, 650 had one to three nodes correlation. The strength of the association of BCAR1 (used (25% of total), and 632 patients had more than three nodes as a continuous variable) with other variables (used as group- (24%) involved. Adjuvant chemotherapy alone (mainly cyclo- ing variable) was tested with the nonparametric Wilcoxon phosphamide/methotrexate/5-fluorouracil) was given to 438 pa- rank-sum test or Kruskal-Wallis test. Survival probabilities tients (mainly premenopausal patients), whereas 322 patients were calculated by the actuarial method of Kaplan and Meier received adjuvant hormonal therapy (mainly postmenopausal (13). Both univariate and multivariate (i.e., multivariable) patients), either alone (294 patients) or in combination with analyses were performed using the Cox proportional hazards chemotherapy (28 patients). All patients were examined routinely every 3 to 6 months model. We first defined the base model, including age, meno- during the first 5 years of follow-up and once a year thereafter. pausal status, tumor and nodal status, grade, and ER/PgR The median follow-up period of patients alive (n ϭ 1571) was status, and then investigated the interactions. To the base 96 months (range, 1 to 221 months). Of the 2593 patients model, we next added separately BCAR1, uPA, and PAI-1 and included, 1257 (48%) showed evidence of disease (including checked the interactions. Because all three variables added 212 locoregional relapses) and counted as failures in the anal- significantly to the base model we included them simulta- ysis of RFS. One hundred fifty-two patients died without evi- neously. By a step-down procedure for these added variables, dence of disease and were censored at last follow-up in the we determined the final model (base model plus BCAR1 and analysis of RFS. Eight hundred seventy patients died after a PAI-1). The likelihood ratio test in the Cox regression models previous relapse. Thus, a total of 1022 patients (39%) failed in was used to test for differences and for interactions. In our the analysis of OS. search for the best categorization of BCAR1, we have used Assays of ER, PgR, Total Protein, Urokinase-Type isotonic regression analysis (12, 14). In the univariate analysis Plasminogen Activator (uPA), Plasminogen Activator Inhib- for RFS, the proportionality assumption was investigated using itor 1 (PAI-1), and BCAR1 in Tumor Tissue Extracts. Tu- a test based on the Schoenfeld residuals (15). In a generalized mor tissues were stored in liquid nitrogen and pulverized in the linear regression of the scaled Schoenfeld residuals on a func- frozen state with a microdismembrator as recommended by the tion of time, the null hypothesis of zero slope is tested. All European Organization for Research and Treatment of Cancer computations were done with the STATA statistical package, for processing of breast tumor tissue for cytosolic ER and PgR release 8.2 (STATA Corp., College Station, TX). All P values determinations (10). The resulting tissue powder was suspended are two-sided.

RESULTS Table 1 Relationships of BCAR1 levels with patient and BCAR1 Levels in Relation to Patient and Tumor Char- tumor characteristics acteristics and uPA/PAI-1 Levels. The distribution of the BCAR1 levels of BCAR1 in 2593 tumor cytosols of primary breast Median value tumors determined with the BCAR1-ELISA is presented in Fig. Characteristic Frequency* (interquartile range)† P 1. The levels ranged from 0.02 to 23.0 ng/mg protein (median, All patients 2593 3.58 (3.18) 3.58 ng/mg protein). There appeared to be an approximately Age (y) 0.002‡ log-normal distribution pattern. Table 1 shows the median and Յ40 338 3.47 (2.75) interquartile range levels of BCAR1 in subgroups of tumors and 41–55 926 3.51 (3.18) their relationship with patient and tumor characteristics. The 56–70 832 3.68 (3.39) Ͼ70 497 3.84 (3.24) tumor level of BCAR1 was slightly lower in poor grade tumors, Menopausal status 0.008§ somewhat higher in postmenopausal patients, and was weakly Premenopausal 1063 3.48 (3.01) ϭ ϭ ϭ positively related with age (rs 0.06, P 0.002), ER (rs Postmenopausal 1530 3.72 (3.30) Ͻ ϭ Ͻ T status 0.73¶ 0.18, P 0.0001), and PgR (rs 0.12, P 0.0001). Compared T1 1058 3.55 (3.21) with ductal, lobular, and mucinous carcinomas, medullary tu- T 1257 3.64 (3.14) mors contained lower levels of BCAR1 [␹2 ϭ 8.9, degrees of 2 T3/4 278 3.67 (3.42) freedom (df) ϭ 3, P ϭ 0.03). There were no significant corre- N status 0.99¶ lations with tumor size status or lymph node status. BCAR1 N0 1311 3.67 (3.26) N 650 3.54 (2.90) levels were positively correlated with those of uPA (r ϭ 0.49, 1–3 s N 632 3.58 (3.29) ϭ Ͻ ϭ ϭ Ͻ Ͼ3 n 2587, P 0.0001) and PAI-1 (rs 0.23, n 2587, P Histology 0.03¶,ʈ 0.0001; Table 1). Invasive ductal 1358 3.56 (3.21) RFS and OS: Univariate Analysis. RFS and OS analy- Invasive lobular 191 3.57 (2.65) ses as a function of log-transformed continuous BCAR1 levels Mucinous 59 3.74 (4.10) Medullary 40 2.28 (2.71) showed statistically significant relationships with a poor prog- Others ϩ unknown 945 3.68 (3.32) nosis in univariate analysis (P ϭ 0.005, P ϭ 0.009, respec- Histologic grade 0.002§,ʈ tively). This justified the search for (a) cut point(s) to allow Poor 1379 3.42 (3.13) visualization with survival curves and analysis as a categorical Moderate/good 491 3.71 (3.19) variable in addition to analysis of BCAR1 as a continuous Unknown 723 3.93 (3.16) ER status** Ͻ0.0001‡ variable. Using the results of the isotonic regression analysis Negative 615 3.20 (3.48) with RFS as an end point, we choose 1.75 and 4.04 ng/mg Positive 1978 3.67 (3.09) cytosolic protein as cut points to classify tumors as BCAR1-low PgR status** Ͻ0.0001‡ (n ϭ 477), BCAR1-intermediate (n ϭ 1023), and BCAR1-high Negative 834 3.50 (3.58) ϭ Positive 1732 3.67 (3.02) (n 1093). The Kaplan-Meier curves for all patients as a uPA status†† Ͻ0.0001‡ function of BCAR1 status showed that intermediate and high Low 393 1.86 (1.85) levels of BCAR1 were significantly associated with a poor RFS Intermediate 1222 3.31 (2.58) (log-rank test for trend, P Ͻ 0.0002; Fig. 2A) and OS (P ϭ High 972 4.88 (3.16) Ͻ 0.0007, Fig. 2B). Taking into account all failures in the analysis PAI-1 status‡‡ 0.0001‡ Low 771 2.85 (3.01) Intermediate 1631 3.85 (3.11) High 185 4.46 (3.58) * Because of missing values, numbers do not always add up to 2593. † All values in ng/mg of protein (range between 25th and 75th percentiles). ‡ P for Spearman rank correlation. § P for Wilcoxon rank-sum test. ¶ P for Kruskal-Wallis test. ʈ Others ϩ unknown not included in the analysis. ** Cut points used for ER and PgR: 10 fmol/mg protein. †† Low, Յ0.19 (n ϭ 393); intermediate, Ͼ0.19 and Յ1.21 (n ϭ 1222); high, Ͼ1.21 ng/mg protein (n ϭ 972) (for cut points; ref. 42). ‡‡ Low, Յ9.33 (n ϭ 771); intermediate, Ͼ9.33 and Յ45.28 (n ϭ 1631); high, Ͼ45.28 ng/mg protein (n ϭ 185) (for cut points; ref. 42).

of RFS, the proportional hazards assumption was not violated, neither when using BCAR1 as a log-transformed continuous variable nor as a categorized variable. Fig. 1 Distribution of total BCAR1 over 2593 primary human breast RFS and OS: Multivariate Analysis. The independent tumor cytosols. Arrow indicates position of the median value at 3.58 relationship of BCAR1 with RFS and OS was studied with Cox ng/mg protein. multivariate regression analysis. In the analysis of RFS, which

Fig. 2 RFS (A) and OS (B) as a function of total BCAR1 status in 2593 primary breast cancer patients. Patients at risk are indicated. Cut points used, 1.75 and 4.04 ng/mg protein.

included all 2593 patients, independent associations with poor and 23.0 (df ϭ 2, P Ͻ 0.0001 for both) in the analyses of RFS prognosis were found for tumor size, nodal status, young pre- and OS, respectively. Analysis of BCAR1 as a continuous and postmenopausal age, premenopausal status, and poor tumor variable again showed a significant contribution to the base grade (Table 2). Steroid hormone receptor positivity (ER and/or models for RFS and OS (⌬␹2 ϭ 12.2, df ϭ 1, P ϭ 0.0005, and PgR positive versus both negative) was not significantly related 18.4, df ϭ 1, P ϭϽ0.0001). Also the established strong with RFS, which could be explained by a violation of the prognostic factors uPA and PAI-1 (17, 18) significantly added to proportional hazards assumption (univariate P Ͻ 0.0001), a base models for RFS and OS when added alone (Table 3), with phenomenon described previously (16). In the multivariate anal- ⌬␹2 ϭ 36.2 (df ϭ 2) and ⌬␹2 ϭ 24.5 (df ϭ 2) for uPA, ysis for OS, taking into consideration the traditional prognostic respectively, and ⌬␹2 ϭ 54.2 (df ϭ 2) and ⌬␹2 ϭ 36.0 (df ϭ 2) factors and the observed significant interactions between men- for PAI-1, respectively. After adding BCAR1, uPA, and PAI-1 opausal status and both nodal status and BCAR1 levels (see next to the base models for RFS and OS, the contribution of uPA was paragraph), two differences were observed compared with the no longer statistically significant, whereas BCAR1 and PAI-1 analysis of RFS, i.e., young postmenopausal age and steroid both independently contributed to the base models that included hormone receptor positivity were significantly related with a the traditional prognostic factors (Table 3). The ⌬␹2 (df ϭ 4) as prolonged survival (Table 2). The multivariate model, which a result of the simultaneous addition of BCAR1 and PAI-1 to the included age and menopausal status, tumor size, nodal status, base model for RFS was 72.4, which indicates a significantly tumor grade, and ER/PgR status, was defined as the base model better fit than after the addition of either factor alone (⌬␹2 ϭ (Table 2). 28.3 and 54.2, respectively, each with df ϭ 2). Similarly in the After inclusion of BCAR1 as a categorized variable in the analysis of OS, after the addition of BCAR1, uPA, and PAI-1 multivariate models (Table 3), the increase in ␹2 (⌬␹2) was 28.3 together, the contribution of uPA was no longer statistically

Table 2 Cox multivariate analysis: base model RFS OS Factor HR* P† HR* P† Age and menopausal status‡ Ͻ0.0001 Ͻ0.0001 Age premenopausal§ 0.71 (0.63–0.80) 0.73 (0.64–0.85) Age postmenopausal§ 0.90 (0.82–0.98) 1.24 (1.13–1.36) Post- versus premenopausal 1.21 (0.94–1.55) 1.46 (1.14–1.87) Tumor size Ͻ0.0001 Ͻ0.0001 2to5cmversus Յ2 cm 1.39 (1.23–1.58) 1.49 (1.29–1.73) Ͼ5cmversus Յ2 cm 1.75 (1.44–2.11) 1.90 (1.55–2.33) Nodal status Ͻ0.0001 Ͻ0.0001

N1–3 versus N0 1.29 (1.04–1.59) 1.85 (1.57–2.17) NϾ3 versus N0 2.27 (1.86–2.77) 3.22 (2.76–3.75) Grade Ͻ0.0002 0.001 Poor versus well/moderate 1.38 (1.17–1.63) 1.36 (1.13–1.63) Unknown versus well/moderate 1.19 (0.99–1.43) 1.14 (0.93–1.40) ER/PgR¶ (positive versus negative) 0.92 (0.79–1.06) 0.24 0.63 (0.54–0.73) Ͻ0.0001 Interaction nodal status ϫ 0.002 menopausal status ϫ N1–3 post versus pre and/or N0 1.44 (1.08–1.92) ϫ NϾ3 post versus pre and/or N0 1.56 (1.20–2.02) NOTE. Final base models included 2593 patients. For RFS, HRs and confidence intervals are corrected for interaction between nodal and menopausal status. For OS, no interactions were observed. * Numbers in parentheses, 95% confidence interval. † P values from likelihood ratio test. ‡ Age and menopausal status combined. § Age in decades for pre- and postmenopausal patients. ¶ Positive, Ն10 fmol/mg protein; negative, Ͻ10 fmol/mg protein; ER/PgR, one or both positive versus both negative. Abbreviation: HR, hazard ratio.

Table 3 Cox multivariate analysis: corrected for base model RFS OS Factors added to base model* HR† P‡ HR† P‡ ϩ BCAR1§ Ͻ0.0001 Ͻ0.0001 Intermediate versus low¶ 1.14 (0.90–1.44) 1.37 (1.03–1.82) High versus low¶ 1.60 (1.27–2.02) 1.81 (1.36–2.40) ϩ BCAR1 (continuous)§,ʈ 1.45 (1.18–1.78) 0.0005 1.68 (1.33–2.14) Ͻ0.0001 ϩ uPA§ Ͻ0.0001 Ͻ0.0001 Intermediate versus low** 1.14 (0.96–1.35) 1.13 (0.94–1.37) High versus low** 1.56 (1.31–1.85) 1.51 (1.24–1.83) ϩ PAI-1§ Ͻ0.0001 Ͻ0.0001 Intermediate versus low** 1.42 (1.24–1.62) 1.26 (1.09–1.46) High versus low** 2.19 (1.77–2.71) 2.11 (1.67–2.66) ϩ BCAR1 ϩ PAI-1†† BCAR1 0.0003 Ͻ0.0001 Intermediate versus low¶ 1.08 (0.85–1.37) 1.31 (0.98–1.74) High versus low¶ 1.51 (1.19–1.91) 1.73 (1.30–2.30) PAI-1 Ͻ0.0001 Ͻ0.0001 Intermediate versus low** 1.35 (1.18–1.55) 1.21 (1.04–1.40) High versus low** 2.04 (1.64–2.53) 1.98 (1.56–2.49) NOTE. Base model as defined in Table 2. For RFS, HRs and CI are corrected for interaction between nodal and menopausal status and menopausal status and BCAR1 levels, if applicable. For OS, correction for interaction between BCAR1 and menopausal status, if applicable. All multivariate models with either uPA or PAI-1 included at least 2587 patients. * Factors added alone or together to the base model. † Numbers in parentheses, 95% CI. ‡ P values from likelihood ratio test. § Added alone to the base model. ¶ Low, Յ1.75 (n ϭ 477); intermediate, Ͼ1.75 and Յ4.04 (n ϭ 1023); high, Ͼ4.04 ng/mg protein (n ϭ 1093). ʈ Log-transformed variable. ** See Table 1. †† Added together to the base model ϭ final model. Abbreviations: HR, hazard ratio; CI, confidence interval.

significant and BCAR1 and PAI-1 together added a ⌬␹2 ϭ 43.3 opausal status with nodal status and menopausal status with (df ϭ 4) compared with either factor alone (⌬␹2 ϭ 23.0 and ⌬␹2 BCAR1 were observed in the analysis for RFS. ϭ 36.0, both df ϭ 2, respectively). Thus, for both the analysis BCAR1 and RFS in Subgroups of Patients. Because of RFS and OS, BCAR1 independently contributed to the prog- there was an interaction between menopausal status and nodal nostic value of the established strong prognostic factor PAI-1, status, we considered the patients stratified by menopausal sta- where BCAR1 was favored over uPA. tus and nodal status as four clinically distinct subgroups. Effects of Adjuvant Treatment and Other Interactions. Kaplan-Meier univariate analysis for RFS showed that in Addition of adjuvant (none, chemo, or hormone) treatment as an premenopausal/node-negative patients (Fig. 3A) and in post- indicator variable to the base model for RFS only marginally menopausal/node-positive patients (Fig. 3D), higher BCAR1 changed the estimates for categorized BCAR1 when added levels were associated with a poor prognosis (for both, P Ͻ alone. As a result of the presence of adjuvant therapy in the base 0.01) with a difference in 10-year RFS between the extreme model, there was no significant difference for both the BCAR1- curves of 17 and 14%, respectively. A similar, nonsignificant high group [hazard ratio (95% confidence interval) ϭ 1.51 trend was observed for postmenopausal/node-negative patients (1.19–1.91) versus 1.53 (1.21–1.93)] and the BCAR1-interme- (Fig. 3B) and premenopausal/node-positive patients (Fig. 3C). diate group (hazard ratio ϭ 1.08 versus 1.04). Similarly, in the Taken together, in the node-negative patients, there was a dif- analysis of OS, no significant effects on the coefficients of ference of 26% at 10-year RFS between the group performing BCAR1-intermediate or BCAR1-high were observed after add- best (69% for BCAR1-low in postmenopausal patients; Fig. 3B) ing adjuvant treatment to the base model. In addition, in the and worst (43% for BCAR1-high in premenopausal patients; analysis of RFS in all patients or in the subgroup of node- Fig. 3A). Furthermore, a striking difference was observed for the positive patients, there were no statistically significant interac- patients with tumors containing intermediate levels of BCAR1 tions between adjuvant endocrine- or chemotherapy with when comparing the premenopausal patients (Fig. 3, A and B) BCAR1 (included either as a categorized or as a log-trans- with the postmenopausal patients (Fig. 3, B and D). For the formed continuous variable). Furthermore, in the analysis for premenopausal subgroups, the prognosis of the patients with RFS, there were no statistically significant interactions between intermediate BCAR1 levels is similar to that with low BCAR1 BCAR1 and nodal status, ER/PgR status, uPA, or PAI-1. In levels, whereas for the postmenopausal patients it is the same as contrast, statistically significant (P Ͻ 0.01) interactions of men- in the group with high BCAR1 levels. These results for RFS are

Fig. 3 RFS as function of BCAR1 status in subgroups of premenopausal/node-negative patients (A), postmenopausal/node-negative patients (B), premenopausal/node-positive patients (C), and postmenopausal/node-positive patients (D). Both univariate and multivariate HRs [95% confidence interval (CI)] are presented. Patients at risk are indicated. Cut points used, 1.75 and 4.04 ng/mg protein.

also visualized in Fig. 4, left panel, in which the univariate levels as previously reported (5) could be confirmed. A strong hazard ratio of the patients with low levels of BCAR1 is set at inverse relation between BCAR1 levels and RFS and OS after 1.0, which is presented as a vertical line. In the analysis of OS 10 years of follow up is evident in both univariate and multi- as well, premenopausal patients with intermediate tumor levels variate analyses. Although uPA and PAI-1 contributed signifi- of BCAR1 behave similarly as those with low levels, whereas in cantly to the base model as expected (18), simultaneous addition postmenopausal patients, they have a similar prognosis as those of BCAR1 causes loss of prognostic power of uPA likely with high levels of BCAR1 (Fig. 4, right panel). Additional because of the strong positive correlation between uPA and exploratory analyses for RFS in the clinically relevant sub- BCAR1 levels. Our data thus show that BCAR1 represents a groups of node-negative and -positive patients and ER-positive prognostic factor independent of both traditional and strong and -negative patients showed that the prognostic value of novel prognostic factors (uPA and PAI-1). Because the tumor BCAR1 did not differ substantially between the nodal sub- content of the tissue specimen may slightly affect the actual groups but that it was more prominent in ER-positive patients level of BCAR1, for future diagnostic studies, it may be bene- compared with ER-negative patients (Fig. 4, left panel). In the ficial to take the tumor percentage into account. analyses for OS, the difference between the ER subgroups was Dissection of the role of BCAR1 in the analysis of RFS in less apparent (Fig. 4, right panel). clinically distinct subgroups shows a correlation between BCAR1 levels and nodal status, ER status, and menopausal DISCUSSION status. In these subgroups, higher BCAR1 levels are associated Previous evaluations of the role of the BCAR1 protein in with increased hazard ratios (Fig. 4). Remarkably, patients with breast cancer progression were hampered by the incomplete breast tumors containing intermediate levels of BCAR1 appear specificity of the antibodies directed against p130Cas in the to have a distinct prognosis depending on their menopausal semiquantitative and laborious features of the Western blot status (Figs. 3 and 4). In premenopausal patients, a similar assay (5). Our current BCAR1 ELISA resolves both problems. prognosis is observed for tumors with low or intermediate The closely related HEF1 protein (also termed NEDD9), which BCAR1 levels. In contrast, in postmenopausal patients, tumors has a similar domain structure as BCAR1 (4), is very well with intermediate and high levels have a similar poor prognosis. discriminated by the BCAR1 ELISA. HEF1 does not contribute Different treatment modalities used for pre- and postmenopausal to the signal of BCAR1 in our assay (7). This ELISA has patients and node-negative and node-positive patients cannot facilitated the rapid quantitative analysis of the BCAR1 levels in explain this phenomenon. Similar profiles for the Kaplan-Meier a large series of breast tumor cytosols. The results presented RFS curves for each of the categorized BCAR1 levels were here extend the results of our previous Western blot data (5). obtained in adjuvant treated and untreated patient subgroups Direct comparison of the BCAR1 levels measured with semi- (data not shown). Furthermore, no significant association be- quantitative Western blotting with the corresponding ELISA tween BCAR1 levels and adjuvant treatment was observed in ϭ ϭ Ͻ data showed good concordance (rs 0.56, n 592, P any analysis. 0.0001) in a random set of cytosol samples measured with both The different prognosis of pre- and postmenopausal pa- assays (7). Association of BCAR1 levels with patient age and tients with intermediate tumor levels of BCAR1 is not caused by menopausal status and tumor differentiation grade and ER/PgR the slightly different median levels of BCAR1 (Table 1) because

Fig. 4 RFS (left) and OS (right)asa function of BCAR1 status in subgroups of patients stratified by nodal status, ER status, and menopausal status. The hazard ratio (HR) for tumors with low BCAR1 levels was set at 1.0 (vertical line). The data presented are the point estimates of the univariate HR and its 95% confidence interval (CI) (horizontal line). The number of patients in each BCAR1-intermediate and BCAR1-high subgroup is indicated. Cut points used, 1.75 and 4.04 ng/mg protein.

the distribution of patients among the BCAR1 categories is cancer (40) and to be inversely associated with survival of breast grossly similar (Fig. 3). We have previously shown that estrogen- cancer patients (41). Whether BCAR1 and KAI1 are counter- independent breast cancer cell proliferation in vitro may be active in breast cancer metastasis remains to be determined. controlled by the level of BCAR1 protein (1). In the parental The results of our quantitative BCAR1 analysis within this ZR-75-1 cells, the basal level of BCAR1 is insufficient to drive large series of breast cancer patients indicate that the prognosis cell proliferation in the absence of estrogen. Either insertion of in the various clinically defined subgroups of patients varies a retrovirus in the proximity of the BCAR1 gene or transfection according to the tumor BCAR1 levels and that BCAR1 may of the BCAR1 cDNA into ZR-75-1 cells allows for estrogen- represent, in combination with other factors, a useful prognostic independent cell proliferation. We may speculate that increased marker for patients with breast cancer. levels of BCAR1 also contribute to the growth potential of breast tumors. Because changes in the hormone regulation are ACKNOWLEDGMENTS very prominent during menopause and tumor growth is at least We gratefully express our thanks to the surgeons, pathologists, and in part dependent on estrogen levels (19), the contribution of internists of the St. Clara Hospital, Ikazia Hospital, St. Franciscus BCAR1 to cell proliferation control may change accordingly. Gasthuis at Rotterdam, and the Ruwaard van Putten Hospital at Spi- Alternatively, the change in hormonal regulation during meno- jkenisse, for the supply of tumor tissues and/or for assisting us in the pause may affect the role of BCAR1 in tumor metastasis. collection of the clinical follow-up data. BCAR1 has been shown to be important in adhesion, motility, and migration of various cell types (3, 4, 20–25) and thus could REFERENCES increase the likeliness or success rate of tumor metastasis. The 1. Brinkman A, Van der Flier S, Kok EM, Dorssers LCJ. BCAR1, a clear absence of an association between BCAR1 levels in the human homologue of the adapter protein p130Cas and antiestrogen primary tumor and nodal status (Table 1) suggests that high resistance in breast cancer cells. J Natl Cancer Inst (Bethesda) 2000;92: BCAR1 levels increase the chances for disease recurrence in- 112–20. dependent of the nodal dissemination pathway. 2. Sakai R, Iwamatsu A, Hirano N et al. A novel signaling molecule, Apart from the patient and breast tumor characteristics p130, forms stable complexes in vivo with v-Crk and v-Src in a tyrosine phosphorylation- dependent manner. EMBO J 1994;13:3748–56. (menopausal status, nodal status, tumor size, and histology) and 3. O’Neill GM, Fashena SJ, Golemis EA. Integrin signalling: a new a limited number of well-established markers (ER and PgR), cas(t) of characters enters the stage. Trends Cell Biol 2000;10(3):111–9. few newly identified marker proteins are applied for clinical 4. Bouton AH, Riggins RB, Bruce-Staskal PJ. Functions of the adapter decision making (26, 27). Prospective studies and/or meta- protein Cas: signal convergence and the determination of cellular re- analyses are required to achieve this position (28). HER2/neu sponses. Oncogene 2001;20(44):6448–58. protein level and (or gene amplification) has proven to be 5. Van der Flier S, Brinkman A, Look MP et al. Bcar1/p130Cas protein important for selection of patients for specific treatment (29). and primary breast cancer: prognosis and response to tamoxifen treat- Recently, meta-analysis of proteins of the plasminogen activator ment. J Natl Cancer Inst (Bethesda) 2000;92:120–7. system has shown that high levels of uPA and PAI-1 are 6. Van der Flier S, Chan CMW, Brinkman A et al. BCAR1/p130Cas expression in untreated and acquired tamoxifen-resistant human breast predictive for disease recurrence in patients with early breast carcinomas. Int J Cancer 2000;89(5):465–8. cancer (17, 18, 30). The prognostic value of most novel markers 7. Grebenchtchikov N, Brinkman A, Van Broekhoven SPJ, et al. De- appears insufficient to justify its single use in diagnosis. How- velopment of an ELISA for measurement of BCAR1 protein in human ever, as exemplified by the prognostic power of the expression breast cancer tissue. Clin Chem. 2004;80:1356–63. levels of a set of mRNAs predicting disease outcome (e.g., ref. 8. Sherman CD, Hossfeld DK. Breast cancer. In: Hossfeld DK, Sher- 31), quantitative assays for a set of relevant proteins may pro- man CD, Lover RR, Bosch FX, editors. Manual of clinical oncology. Geneva, Switzerland: Union International Contre Cancer; 1990. vide detailed insight in the prognosis and treatment response of p. 253–71. individual patients. Because of its independent prognostic value 9. Foekens JA, Portengen H, Van Putten WLJ et al. Prognostic value of for RFS, BCAR1 may contribute in such a panel of analyses. In estrogen and progesterone receptors measured by enzyme immunoas- addition to prognosis, the predictive value of increased levels of says in human breast tumor cytosols. Cancer Res 1989;49:5823–8. BCAR1 in the primary tumor for poor clinical benefit of recur- 10. EORTC Breast Cancer Cooperative Group. Revision of the stand- rent disease on first-line tamoxifen treatment (5) has been con- ards for the assessment of hormone receptors in human breast cancer; firmed with the ELISA (32). Furthermore, our quantitative report of the second EORTC workshop, held on 16–17 March, 1979, in the Netherlands Cancer Institute. Eur J Cancer 1980;16:1513–5. BCAR1-ELISA may prove useful for the analysis of small 11. Foekens JA, Schmitt M, Van Putten WLJ, et al. Prognostic value of biopsy specimen of breast cancer and may also be applied to urokinase-type plasminogen activator in 671 primary breast cancer establish a prognostic value for BCAR1 in other malignancies patients. Cancer Res 1992;52:6101–5. and diseases. In addition to its association with human breast 12. Foekens JA, Schmitt M, Van Putten WLJ, et al. Plasminogen cancer, BCAR1/p130Cas has been linked to canine and feline activator inhibitor-1 and prognosis in primary breast cancer. J Clin mammary cancer (33) and suggested to contribute to malignant Oncol 1994;12:1648–58. melanoma (34) and subsets of leukemia (35, 36). Overexpres- 13. Kaplan EL, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc 1958;53:457–81. sion of BCAR1/p130Cas in prostate cancer cells in vitro coun- 14. Barlow RE, Bartholomew DJ, Bremner JM, Brunk HD. Statistical teracts the inhibition of cell motility by CD82/KAI1 (37), sug- inference under order restrictions. New York: John Wiley & Sons; 1972. gesting that BCAR1/p130Cas plays a role in the metastasis 15. Grambsch PM, Therneau TM. Proportional hazards tests and diag- suppressor function of KAI1 in prostate cancer (38) and other nostics based on weighted residuals. Biometrika 1994;81:515–26. poorly differentiated or metastatic cancers (39). Expression of 16. Hilsenbeck SG, Ravdin PM, de Moor CA, Chamness GC, Osborne KAI1 was also found to be down-regulated in infiltrating breast CK, Clark GM. Time-dependence of hazard ratios for prognostic factors

in primary breast cancer. Breast Cancer Res Treat 1998;52(1–3): 29. Slamon DJ, Leyland-Jones B, Shak S, et al. Use of chemotherapy 227–37. plus a monoclonal antibody against HER2 for metastatic breast cancer 17. Janicke F, Prechtl A, Thomssen C et al. Randomized adjuvant that overexpresses HER2. N Engl J Med 2001;344(11):783–92. chemotherapy trial in high-risk, lymph node-negative breast cancer 30. Harbeck N, Schmitt M, Kates RE, et al. Clinical utility of uroki- patients identified by urokinase-type plasminogen activator and plas- nase-type plasminogen activator and plasminogen activator inhibitor-1 minogen activator inhibitor type 1. J Natl Cancer Inst (Bethesda) 2001; determination in primary breast cancer tissue for individualized therapy 93(12):913–20. concepts. Clin Breast Cancer 2002;3(3):196–200. 18. Look MP, van Putten WL, Duffy MJ et al. Pooled analysis of 31. van ’t Veer LJ, Dai H, van de Vijver MJ, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature (Lond.) prognostic impact of urokinase-type plasminogen activator and its in- 2002;415(6871):530–6. hibitor PAI-1 in 8377 breast cancer patients. J Natl Cancer Inst (Be- thesda) 2002;94(2):116–28. 32. Dorssers LCJ, Grebenchtchikov N, Brinkman A, et al. Application of a newly developed ELISA for BCAR1 protein for prediction of 19. Clarke R, Leonessa F, Welch JN, Skaar TC. Cellular and molecular clinical benefit of tamoxifen therapy in patients with advanced breast pharmacology of antiestrogen action and resistance. Pharmacol Rev cancer. Clin Chem 2004;50:1445–7. 2001;53(1):25–71. 33. Scibelli A, d’Angelo D, Pelagalli, A et al. Expression levels of the 20. Kain KH, Klemke RL. Inhibition of cell migration by Abl family focal adhesion-associated proteins paxillin and p130CAS in canine and tyrosine kinases through uncoupling of Crk-CAS complexes. J Biol feline mammary tumors. Vet Res 2003;34(2):193–202. Chem 2001;276(19):16185–92. 34. Eisenmann KM, McCarthy JB, Simpson MA, et al. Melanoma 21. Ruest PJ, Shin NY, Polte TR, Zhang X, Hanks SK. Mechanisms of chondroitin sulphate proteoglycan regulates cell spreading through CAS substrate domain tyrosine phosphorylation by FAK and Src. Mol Cdc42, Ack-1 and p130cas. Nat Cell Biol 1999;1(8):507–13. Cell Biol 2001;21(22):7641–52. 35. Salgia R, Pisick E, Sattler M, et al. p130CAS forms a signaling 22. Huang J, Hamasaki H, Nakamoto T, et al. Differential regulation of complex with the adapter protein CRKL in hematopoietic cells trans- cell migration, actin stress fiber organization, and cell transformation by formed by the BCR/ABL oncogene. J Biol Chem 1996;271:25198–203. functional domains of Crk-associated substrate. J Biol Chem 2002; 36. de Jong R, van Wijk A, Haataja L, Heisterkamp N, Groffen J. 277(30):27265–72. BCR/ABL-induced leukemogenesis causes phosphorylation of Hef1 and 23. Riggins RB, Quilliam LA, Bouton AH. Synergistic promotion of its association with Crkl. J Biol Chem 1997;272(51):32649–55. c-Src activation and cell migration by Cas and AND-34/BCAR3. J Biol 37. Zhang XA, He B, Zhou B, Liu L. Requirement of the p130CAS-Crk Chem 2003;278:28264–73. coupling for metastasis suppressor KAI1/CD82-mediated inhibition of cell migration. J Biol Chem 2003;278(29):27319–28. 24. Abassi YA, Rehn M, Ekman N, Alitalo K, Vuori K. p130Cas couples the tyrosine kinase Bmx/Etk with regulation of the actin cy- 38. Dong J-T, Lamb PW, Rinker-Schaeffer CW, et al. KAI1, a metas- toskeleton and cell migration. J Biol Chem 2003;278(37):35636–43. tasis suppressor gene for prostate cancer on human chromosome 11p11.2. Science (Wash. DC) 1995;268:884–6. 25. Hsia DA, Mitra SK, Hauck CR et al. Differential regulation of cell 39. Welch DR, Steeg PS, Rinker-Schaeffer CW. Molecular biology of motility and invasion by FAK. J Cell Biol 2003;160(5):753–67. breast cancer metastasis. Genetic regulation of human breast carcinoma 26. Eifel P, Axelson JA, Costa J et al. National Institutes of Health metastasis. Breast Cancer Res 2000;2(6):408–16. Consensus Development Conference Statement: adjuvant therapy for 40. Yang X, Wei L, Tang C, Slack R, Montgomery E, Lippman M. breast cancer, November 1–3, 2000. J Natl Cancer Inst (Bethesda) KAI1 protein is down-regulated during the progression of human breast 2001;93(13):979–89. cancer. Clin Cancer Res 2000;6(9):3424–9. 27. Goldhirsch A, Wood WC, Gelber RD, Coates AS, Thurlimann B, 41. Huang CI, Kohno N, Ogawa E, Adachi M, Taki T, Miyake M. Senn HJ. Meeting highlights: updated international expert consensus on Correlation of reduction in MRP-1/CD9 and KAI1/CD82 expression the primary therapy of early breast cancer. J Clin Oncol 2003;21(17): with recurrences in breast cancer patients. Am J Pathol 1998;153(3): 3357–65. 973–83. 28. Hayes DF, Bast RC, Desch CE, et al. Tumor marker utility grading 42. Foekens JA, Peters HA, Look MP, et al. The urokinase system of system: a framework to evaluate clinical utility of tumor markers. J Natl plasminogen activation and prognosis in 2780 breast cancer patients. Cancer Inst (Bethesda) 1996;88(20):1456–66. Cancer Res 2000;60(3):636–43.

Lambert C. J. Dorssers, Nicolai Grebenchtchikov, Arend Brinkman, et al.

Clin Cancer Res 2004;10:6194-6202.

Updated version Access the most recent version of this article at: http://clincancerres.aacrjournals.org/content/10/18/6194

Cited articles This article cites 23 articles, 7 of which you can access for free at: http://clincancerres.aacrjournals.org/content/10/18/6194.full#ref-list-1

Citing articles This article has been cited by 11 HighWire-hosted articles. Access the articles at: http://clincancerres.aacrjournals.org/content/10/18/6194.full#related-urls

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://clincancerres.aacrjournals.org/content/10/18/6194. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.