Current treatments for lung cancer have been not very successful with less than 15% of patients surviving 5 years after diagnosis. Newer drugs such as gefitinib can have dramatic results with near complete regression of the tumor in a small fraction of cases. Acquired mutations in the EGFR are fairly good at predicting objective response to gefitinib, but are poorly predictive of clinical benefit. In addition, sequence determination requires large fresh frozen biopsies from patients with metastatic disease that are clinically difficult to obtain. Proteomic technologies are currently being used to identify diagnostic markers and the serum proteome is considered to be a rich source of undiscovered biomarkers. In this study, we analyzed 26 individual sera by Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) to obtain proteomic profiles. Class prediction models based on the 24 discriminatory protein peaks allowed correct classification of serum from PR and SD patients, and PD patients with 100% leave-one-out accuracy in the training set. Using an initial test set of 32 patients including 3 objective responders treated with rofecoxib and gefitinib, we achieved an accuracy of 73% in predicting response in this test set. We have also analyzed 33 cell lines for response to gefitinib as well as chemotherapy and found highly predictive profiles in leave-one-out analysis. We are in the process of refining our serum profile using sera from 139 NSCLC patients before and 69 patients 2 weeks after gefitinib administration. This serum set was divided into two groups, a training set and a blinded test set. Discriminatory protein peaks were selected based on differential expression among responses to gefitinib. A class prediction model was built based on the weighted flexible compound covariate method (WFCCM) according to the data from the training set. Then this model was applied to the test set. In this study we demonstrate the potential for deriving protein expression patterns from serum or tumor tissue that may predict response of NSCLC patients to therapy.

[Proc Amer Assoc Cancer Res, Volume 46, 2005]