Predictive Modeling of Tumor Regression Kinetics Using a Murine Model of Oncogene-Addicted Lung Cancers

Abstract

Purpose/Objective(s): Oncogene-addiction or -dependence is a biological phenomenon whereby tumors following oncogene-inactivation exhibit dramatic regression revealing tumor cells that are dependent on their initiating oncogene for survival. This phenomenon has become more widely appreciated clinically with the emergence of oncogene targeted therapies that expose a subset of human malignancies with this phenotype. A human example are oncogene-addicted patients with EGFR-mutated bronchioloalveolar carcinomas (BACs). A critical question is the identification of patients who most benefit from these targeted therapies. Novel methods for determining and predicting treatment response are therefore needed. The Myc and Ras oncogenes have pleiotropic effects on cell autonomous functions such as proliferation, apoptosis and DNA damage repair and cell non-autonomous functions such as angiogenesis, the sum of which can drive the process of oncogenesis. The murine counterpart to oncogene-addicted human BACs are Ras-induced oncogene-addicted non-small cell lung cancers (NSCLCs). Corroborating this supposition are human mutational studies that suggest Ras and EGFR mutations are mutually exclusive in NSCLC. Materials/Methods: We constructed mouse preclinical lung cancer model systems to address these issues. Using a conditional lung specific gene expression system we analyzed transgenic mouse models of Myc, Ras or Myc/Ras induced lung adenocarcinoma. Results: These transgenic mice develop primary lung adenocarcinomas with variable latency (26-51 weeks) and clinical presentation that is dependent on their genotype. Upon oncogene-inactivation in these murine models, we have found that Ras, but not Myc induced lung adenocarcinomas regress in a matter of weeks completely. We measured quantitatively the clinical behavior of murine lung tumors in situ after oncogene-inactivation by use of serial micro-computed tomography (microCT) imaging. By modeling the regression curves of each transgenic system, we were able to express the difference mathematically between the tumor regression of Rasand Myc-induced tumors. The modeling also allowed us to distinguish that double mutant, Ras/Myc-induced tumors, were likely composed of two dominant populations with behavior similar to single Rasand Myc-induced tumors. We then used our data array as a training set for a predictive support vector machine algorithm which is highly accurate at predicting the regression of murine tumors based on only three serial weekly microCT scans at the initiation of oncogeneinactivation. Conclusions: Our work incorporates highly penetrant spontaneously arising murine lung tumor models, non-invasive serial tumor imaging and predictive mathematically modeling. This system serves as a more clinically relevant bridge for studies transitioning from the bench to the clinic and can also be used to probe basic questions of oncogene-addiction. Furthermore, these preclinical studies suggest a novel method for predicting tumor responses in human BACs.