Integrating RAS Status into Prognostic Signatures for Adenocarcinomas of the Lung

Maud H. W. Starmans, Melania Pintilie, Michelle Chan-Seng-Yue, Nathalie C. Moon, Syed Haider, Francis Nguyen, Suzanne K. Lau, Ni Liu, Arek Kasprzyk, Bradly G. Wouters, Sandy D. Der, Frances A. Shepherd, Igor Jurisica, Linda Z. Penn, Ming-Sound Tsao, Philippe Lambin, Paul C. Boutros*

*Corresponding author for this work

Research output: Contribution to journalArticleAcademicpeer-review

38 Downloads (Pure)

Abstract

Purpose: While the dysregulation of specific pathways in cancer influences both treatment response and outcome, few current prognostic markers explicitly consider differential pathway activation. Here we explore this concept, focusing on K-Ras mutations in lung adenocarcinoma (present in 25%-35% of patients). Experimental Design: The effect of K-Ras mutation status on prognostic accuracy of existing signatures was evaluated in 404 patients. Genes associated with K-Ras mutation status were identified and used to create a RAS pathway activation classifier to provide a more accurate measure of RAS pathway status. Next, 8 million random signatures were evaluated to assess differences in prognosing patients with or without RAS activation. Finally, a prognostic signature was created to target patients with RAS pathway activation. Results: We first show that K-Ras status influences the accuracy of existing prognostic signatures, which are effective in K-Raswild-type patients but fail in patients with K-Ras mutations. Next, we show that it is fundamentally more difficult to predict the outcome of patients with RAS activation (RAS(mt)) than that of those without (RAS(wt)). More importantly, we demonstrate that different signatures are prognostic in RAS(wt) and RAS(mt). Finally, to exploit this discovery, we create separate prognostic signatures for RAS(wt) and RAS(mt) patients and show that combining them significantly improves predictions of patient outcome. Conclusions: We present a nested model for integrated genomic and transcriptomic data. This model is general and is not limited to lung adenocarcinomas but can be expanded to other tumor types and oncogenes.
Original languageEnglish
Pages (from-to)1477-1486
JournalClinical Cancer Research
Volume21
Issue number6
DOIs
Publication statusPublished - 15 Mar 2015

Cite this