Chemotherapy is often used for breast cancer treatment, but individual outcome varies widely. We hypothesized that tumor proteomic profiles obtained prior to chemotherapy may predict the individual tumor response to treatment. The goal of our study was to explore feasibility of using proteomic profiling to preselect patients for an effective chemotherapeutic regimen. Tumors from 52 patients with T2-T4 breast cancer were prospectively collected before neoadjuvant chemotherapy, and were analyzed using surface-enhanced laser desorption ionization/time of flight (SELDI) mass spectrometry. Mass spectral profiles were obtained from tumors with various sensitivities to chemotherapy. Both non-supervised hierarchical clustering and supervised neural network-based classification approaches were employed to compare the profiles. The first two thirds of the enrolled cases (35) were allocated to a training set to select peaks characteristic of resistant tumors. The candidate peaks were used to develop a predicting rule to evaluate the remaining 17 specimens in the validation set. In the training set, the most prominent differences were found between drug resistant and drug susceptible tumors by non-supervised hierarchical clustering. In the validation set, the supervised classification with the K nearest neighbor (KNN) model correctly classified most tumor responses with an accuracy rate of 92.3% [100% of resistant tumors (4/4), and 84.6% of the tumors with favorable response (11/13)]. In the entire group, a single peak at m/z 16,906 correctly separated 88.9% of the tumors with pathologically complete response, and 91.7% of the resistant tumors. The data suggest that breast cancer protein biomarkers may be used to pre-select patients for optimal chemotherapeutic treatment.

Related Articles