Background: KRAS mutation status alone is an imperfect predictor of response to anti-EGFR therapy in CRC, motivating efforts to identify additional molecular traits for RAS pathway activity. This study aimed to build a quantitative readout of RAS pathway activity to: (1) unravel molecular surrogates of RAS activity (2) correlate in vitro drug sensitivity in a large drug panel screen of cell lines and in xenografts (3) improve prediction of Cetuximab response at the clinical setting.

Materials and Methods: The CRC data set used for model training (KFSYSCC, n=290) will be made publicly available upon publication. The following data sets were used for validation: TCGA CRC (n=148), Khambata-Ford et al (n=68), Gaedcke, et al (n=65). Penalized logistic regression (ElasticNet) was used to build a model of RAS pathway activity. Drug sensitivity correlation between RAS model and IC50 of 21 drugs was evaluated in the entire panel of CRC cell lines (n=19) of the Cancer Cell Line Encyclopedia (CCLE). Xenograft data (n=54) was obtained from Julien, et al, 2012.

Results: (1) The prediction performance of our model was remarkably robust in predicting KRAS mutated samples with AUCs of 0.82, 0.80, 0.78, 0.90 for KFSYSCC (internal validation), TCGA, Khambata-Ford, and Gaedcke, respectively. Moreover, samples with an NRAS and BRAF mutation had a higher RAS pathway readout compared to wild-type (p < 1e-08 and p < 2e-04, respectively), while PIK3CA mutated/KRAS wt samples were not significantly different from wild-type (TCGA CRC cohort). (2) Applying our model to the CCLE drug sensitivity panel, we were able to predict sensitivity to the experimental MEK inhibitors PD-0325901 and AZD6244 (p < .002). KRAS and BRAF mutation status were not predictive of MEK response (p > .1) This result was confirmed in a mouse xenograft experiment, where KRAS+ mice treated with AZD6244 (n=2) reported a lower RAS readout score compared to control (n=3, p=.08) (3) On a large mouse xenograft cohort derived from human CRC (n=54), our model accurately predicted response to Cetuximab (p < .0002) better than KRAS and/or BRAF mutation alone (p > .1). Finally, analyzing the Khambata-Ford cohort (metastatic CRC patients treated with Cetuximab, n= 57), we were able to predict drug response (AUC=.75) more accurately than KRAS status (AUC=.70). Interestingly, our model correctly identified 5 out of the 6 KRAS mutant patients who benefited from Cetuximab treatment (p < .05).

Conclusion: We provide a gene expression model of the RAS pathway in CRC, suitable for robust interrogation of pathway activity in cell lines, xenografts, and human tumors. The promising utility of our model for predicting response to both Cetuximab and MEK inhibitors warrants further studies to confirm at the clinical setting.

Citation Format: Justin Guinney, Charles Ferte, Erich Huang, Jonathan Derry, Stephen Friend. A model of RAS pathway in colorectal cancer elucidates molecular traits and drug sensitivity: an integrated analysis of gene expression from human tumors, mouse xenografts and the Cancer Cell Line Encyclopedia (CCLE). [abstract]. In: Proceedings of the 104th Annual Meeting of the American Association for Cancer Research; 2013 Apr 6-10; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2013;73(8 Suppl):Abstract nr 2409. doi:10.1158/1538-7445.AM2013-2409