Genomic Data-Based Models of Growth Factor Signaling for Personalized Cancer Therapy Selection

Embargo until
Date
2014-11-07
Journal Title
Journal ISSN
Volume Title
Publisher
Johns Hopkins University
Abstract
The genetic heterogeneity of cancer creates patient-to-patient variability that makes it difficult to predict whether the patient will respond to a treatment. This is particularly true for VEGF-targeting therapies, for which, in some cancers, response rates have typically been low and overall survival has only been extended slightly if at all. Development of predictive biomarkers for VEGF-targeting therapies has traditionally focused either on measurement of VEGF family ligand concentrations in the plasma or on tumor-derived transcriptomic data. Here, we incorporate both of these types of information into a whole-body computational model of the kinetic interactions between VEGF ligands and receptors. Gene expression data from a population of cancer patients allows us to create a virtual population where the effects of multiple VEGF-targeting drugs can be tested. This approach allows us to limit our analysis to relevant genes instead of the entire genome and allows us to incorporate mechanistic information, both of which should lead to models with better reproducibility. We first examined patterns of expression of VEGF ligands and receptors as well as a related family, the Semaphorins. Several previously defined subtypes of cancer were associated with pro-angiogenic alterations in the expression of these genes. Multivariate biomarkers based on VEGF and Semaphorin gene expression were, in some cases, able to provide better separation of patients according to prognosis. These results provide clinically relevant subtypes and highlight the role that Semaphorins may play in processes that drive tumor angiogenesis and progression. We then used gene expression data to create three virtual populations: patients with breast, kidney, or prostate cancers. Drug response metrics in these populations allowed us to determine characteristics of patients that make them more responsive to treatments. We found that the best biomarkers of response differed between cancer types. This approach can be applied to other families of growth factors as well. Here, we demonstrate its application to the EGFR/ErbB family.
Description
Keywords
Computational biology, bioinformatics
Citation