Background: Aggressive breast tumors are a clinically challenge to manage. A VEGF-signature (VEGF-s) developed in vivo (Hu et al. BMC Med. 09) and the profile of the anti-apoptotic gene and regulator of tumor angiogenesis, CRYAB (Moyano et al. J Clin Invest 06; Dimberg et al. Blood 08), are both associated with distant metastasis. In this study, we sought to test if these two mechanistically related biomarkers predict pathological complete response(pCR) to anthracyline/taxane-based neoadjuvant chemotherapy.

Methods: Gene expression data of pre-treatment fresh frozen tumors were combined, 149 from the ISPY trial (Agilent 44K) and 225 from MD Anderson Cancer Center(MDACC) (Affymetrix HG U133A). Patients from I-SPY were treated with doxorubicin/cyclophosphamide followed by a taxane. Patients from MDACC were treated with paclitaxel followed by sequential fluorouracil-AC. Tumors were classified into intrinsic subtypes using PAM50(Parker et al. J Clin Oncol 09) and Claudin-low(CL) predictor(Prat et al., Submitted). Odds ratios of biomarkers to pCR were determined using logistic regression. Predictive values of multivariable models were estimated using receiver operating characteristic curve. Results: 319 patients had complete data; the pCR rate was similar between the datasets (ISPY 22%, MDACC 20%). Both the VEGF-s and CRYAB were highly expressed within Basal-like(BL) and CL (P<0.0001); these two biomarkers were positively correlated with each other (r: 0.4, P<0.001). In univariate analysis, both VEGF-s (P<0.001) and CRYAB (P<0.001) were significantly associated with pCR. A multivariable test including tumor size, grade, ER, Her2, intrinsic subtypes, VEGF-s and CRYAB, had an AUC value of 0.86 to predict pCR(Table 1). T stage, Her2, intrinsic subtypes and CRYABwere significant.

82 patients with triple-negative tumors within this combined cohort. In univariate analysis, T stage, grade or classifying tumors into BL vs not did not predict pCR; but VEGF-s (p= 0.04) and CRYAB (p = 0.02) were associated with pCR. Multivariable model, including VEGF-s, CRYAB and the clinical variables, had a predictive value of 0.71 to pCR(Table 1), which identified CRYAB as the most significant factor by likelihood ratio test.

Table 1: Multivariable logistic regression models to predict pCR.

Conclusion: VEGF-s and CRYAB expression are associated with aggressive subtypes and may possibly help to predict response of anthracycline/taxane-based neoadjuvant chemotherapy beyond ER, Her2, and intrinsic subtypes.