Modeling Tumor Growth to Characterize Disease Heterogeneity

点击翻译按钮获取中文摘要

ABSTRACT The overall goal of this project is to develop computational methods for studying tumor growth, and to relate growth parameters to patient characteristics and prognosis. We hypothesize that tumor growth parameters will allow us to define cancer phenotypes that help resolve cancer heterogeneity, and thereby improve power in analyses that try to link germline genetic variation and internal/external environment to phenotypic variation (Projects 1 & 3). When discovered, human tumors vary in size and extent of spread. Although it is impossible to look directly back in time to see how the tumor grew, it is possible to reconstruct the past with “molecular phylogeny”. The approach is analogous to reconstructing the genealogy of species using DNA sequences. In previous work, we developed a molecular phylogeny approach to study human cancers using DNA methylation patterns and found that a relatively simple exponential growth model fits most colorectal cancers. We now propose to test and further develop the model by integrating new independent molecular data types. The experimental data sample glands from opposite tumor sides and measures passenger DNA methylation patterns, chromosome copy number, and point mutations. Each data type provides `molecular clocks' with different rates of sequence evolution, such that their joint analysis permits our setting a new goal of characterizing what happens during the first few cell divisions following transformation, even before a tumor is clinically detectable. We hypothesize that abnormal cell mobility, a prerequisite for subsequent invasion and metastasis, is a phenotype that can be measured immediately after tumor initiation in some cancers but not benign tumors (“Born to be Bad”). This work will provide a new understanding of intratumor heterogeneity and cancer cell behavior, and might well be the catalyst for the development of new treatment or prognostic paradigms. We will use approximate Bayesian computation to estimate model parameters in this high-dimensional setting. This requires the development of software, and implementation of methods for choosing an optimal set of statistics and corresponding weights for parameter inference. These tools will be applicable to any ABC analysis, and not just our own. As such, we will make this software publicly available to the wider community. The present application uses data from colon cancer to develop the methods and software tools for inferring tumor growth, but the approach is generalizable to any adenocarcinomas, or tumors with glandular structure (e.g. prostate, kidney, lung, breast and more).