Abstract 417: A novel predictive biomarker model for MEK sensitivity

Abstract

Abstract KRAS mutations occur in approximately 25% of NSCLC (1). Tumors with these mutations are predicted to be sensitive to MEK inhibition due to activation of MAPK signaling. However, MEK inhibitors in multiple clinical trials, either as a monotherapy or in combination with chemotherapies, have not shown superior efficacy in the KRAS mutant subgroup when compared to the KRAS wild-type subgroup, indicating a limitation of utilizing KRAS mutation status as a predictive biomarker of efficacy to MEK inhibition (2, 3). Furthermore, stratification based on KRAS mutation status may inadvertently miss wild-type KRAS tumors that could be addicted to MAPK signaling regardless of KRAS mutation status, thus denying patients potential benefit from MEK inhibitors. Here we describe a novel predictive model that more accurately forecasts the sensitivity of the KRAS wild-type NSCLC subpopulation to MEK inhibitors such as cobimetinib and trametinib. Cell viability data from cobimetinib or trametinib-treated cells, with concomitant gene expression data (RNAseq), from 46 colon, 106 lung, and 37 pancreatic cell lines were used to create an elastic net regression model trained on gene expression features (alpha = 0.5 and optimal lambda chosen by 5-fold cross validation) (4). From the model, we established two distinct predictive gene lists: (1) a longer low cross-validation list, and (2) a shorter low error list. Initial analysis of the model demonstrated that predicted mean viabilities of the cell lines used to create the model correlated well with their actual mean viabilities (R: 0.65-0.7 for trametinib and cobimetinib respectively). Predicted mean viabilities of 40 previously unscreened NSCLC cell lines were then generated by the model based on their expression features (RNAseq). The 40 cell lines were categorized either as sensitive or resistant by the median of predicted mean viabilities derived from the model. Subsequently, the actual experimental GI50 values of cobimetinib were obtained for each of these cell lines. We found that KRAS mutational status predicted that only 8 of the 40 cell lines screened would be sensitive to MEK inhibition. In contrast, our model predicted that 15 of the 40 cell lines screened would be sensitive. Experimentally, we demonstrated that 24 of the cell lines were sensitive to MEK inhibition with a measured GI50<1uM. In conclusion, we describe a novel predictive model that more accurately predicts sensitivity to MEK inhibition, provides a potentially larger patient segment, and could be translatable to the clinic. References (1) Blumenschein GR. et al., (2015) Annal Oncol. 26(5):894-901. (2) Gandara DL et al., (2013) J Clin Oncol 31, (suppl; abstr 8028). (3) Laethem JLV et al, (2014) J Clin Oncol 32:5s, (suppl; abstr 4025). (4) Barretina J. et al. (2012) Nature 483, 603-607. Citation Format: Marie Wagle, Christiaan Klijn, Bonnie Liu, Shilpi Mahajan, Peter Haverty, John Moffat, Mark Merchant, Bob Yauch, Garret Hampton, Lukas Amler, Mark Lackner, Shih-Min A. Huang. A novel predictive biomarker model for MEK sensitivity. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 417.