Abstract 5314: A proteomic signature predicts response to a therapeutic vaccine in pancreas cancer; analysis from the GI-4000-02 trial

Background: We have previously reported that adjuvant treatment with a therapeutic vaccine targeting the mutated Ras oncogene product generated mutation-specific T cell responses associated with a trend toward improved survival in patients with post-operative residual disease (R1 resections) but no improvement in the overall population1. Initial analysis of 90 pretreatment plasma samples using matrix assisted laser desorption ionization time of flight (MALDI-TOF) mass spectrometry (MS) showed the potential to predict improved RFS and OS for treatment with GI-4000/gemcitabine, but not placebo/gemcitabine. Methods: We have developed a novel technique, combining methods used in recent advances in learning theory (‘deep learning’) with newly-refined MS techniques that allow exploration deeper into the proteome to create diagnostic tests. Using 500,000 laser shot Deep MALDI spectra2 more than 700 mass spectral features were identified. A subset of these was used to create many multivariate classifiers that were filtered for performance and combined using dropout regularization. This method allows the use of smaller training sets and so left a test set with which performance of the signature could be independently assessed. This new methodology was used to create a test (BDX-001) to identify patients likely to benefit from the addition of GI-4000 to gemcitabine. Results: Using BDX-001 for stratification, subjects who are BDX-001(+) demonstrated a 499 day advantage in median OS when treated with GI-4000/gemcitabine vs. placebo/gemcitabine. Additionally, these subjects demonstrated a 351 day improvement in median RFS. BDX-001 did not predict response for placebo/gemcitabine treated subjects. These results were obtained using only test set data, and although the small sample size prohibited statistical significance, it should give an unbiased test performance estimate to be validated independently. Conclusions: BDX-001 is a test developed using novel proteomic and learning theory methods that appears to predict treatment response to GI-4000 in resected pancreas cancer patients, potentially identifying patients with improved RFS and OS in the GI-4000/gemcitabine arm. We plan to prospectively validate BDX-001 as a companion diagnostic in a future study of GI-4000 in pancreas cancer. References 1. Richards et al, ESMO GI. Annals of Oncology, June 2012 23 (suppl 4) 2. Duncan et al, ASMS 2013, http://asms.inmerge.com/Proceedings/2013Proceedings.aspx. Citation Format: Do