The targeted inactivation of specific oncogenes can result in the dramatic regression of cancers through the phenomenon of oncogene addiction (1-7). However, it is difficult to predict which tumors will exhibit addiction to a particular oncogene and which patients with cancer will respond to particular targeted therapies. We hypothesize that through the use of mathematical and statistical modeling together with quantitative molecular imaginq, we can predict verv early after the initiation of a therapv whether cancer patients will respond to tarqeted oncogene inactivation. As a direct outgrowth of our previous four years work as members of the Stanford ICMIC program, we have developed an approach to mathematically model tumorigenesis in conditional transgenic models of lung adenocarcinoma utilizing quantitative microCT imaging combined with in situ analysis. Our goal was to address two important and related questions: i) what is the biological mechanism governing the dramatic response of oncogene addicted tumors following oncogene inactivation and ii) how can we best predict which tumors are oncogene addicted versus which are not. To address these questions, we have made use of mechanistic modeling based on ordinary differential equations, together with data-driven statistical modeling based on support vector machine classifiers. Using a mechanistic model, we have found that oncogene inactivation in oncogene addicted tumors can be modeled as a differential attenuation of pro-survival and pro-death intracellular signals. Using a data-driven statistical model, we have been able to predict shortly after oncogene inactivation whether tumors are oncogene addicted or not. Moreover, we provide preliminary results showing that our predictive model can be applied to human patients (N=43) with lung tumors treated by EGFR inhibition with eriotinib, in order to predict both genotype and progression free survival. To date, we have focused on the use of anatomical imaging (microCT) and immunohistochemical analyses to provide the data for our models. However, anatomical imaging does not provide information on the biological activity of a tumor, while immunohistochemistry does not allow for serial sampling of a given tumor. We believe that we can now significantly improve our models and enhance their clinical transiational applicability for the analysis of human lung tumors by incorporating PET and SPECT imaging. In particular, we will incorporate molecular imaging of proliferation [FLT microPET] and apoptosis [ [99] Tc-AxV microSPECT] into our mechanistic model of oncogene addiction to replace immunohistochemical analyses, and into our predictive model of oncogene addiction status to improve the predictive power of the model. Furthermore, we will make use of a novel nanoscale proteomic technology, the Nano-lmmuno-Assay (NIA), for interrogating protein expression and phosphorylation in tumor samples (8). Thus, we will significantly enhance and extend our mathematical models and perform a prospective validation of our approach in a clinical study.
Our work will result in the development of new imaging methods and computational strategies to predict when oncogene inactivation will be therapeutically effective in the treatment of human lung cancer.
Showing the most recent 10 out of 424 publications
