Investigating longitudinal evolution of liquid cancers using computational and mathematical models

Abstract

Cancer can result from a series of driver mutations, alterations in the genome sequence that confer a fitness advantage to cells containing them, resulting in their net proliferation. Mutations that have a neutral fitness effect ("passenger mutations") also accumulate in cancer cells, contributing to the significant heterogeneity observed in tumors. The genetically distinct subpopulations of cancer cells—or subclones—can have different levels of fitness and treatment sensitivity. Study of the individual subclones facilitates understanding the whole tumor's dynamics, treatment resistance, and potential for relapse. In this thesis I discuss several projects that employ mathematical and computational models to investigate subclonal evolution in longitudinal studies of leukemia. First, I discuss work that employs branching processes to reconstruct the evolutionary history of cancers, including when cancer was initiated and when subsequent driver mutations occurred. Next, I present a pipeline I developed in collaboration with a team of physician-scientists and clinical researchers to enable monitoring and interactive visualization of clonal evolution and cancer relapse in the clinic, by clustering mutations and inferring the inter-clonal architecture and relationships. I show several examples of this pipeline applied to data from a clinical trial of hematopoietic cell transplantation to treat acute myeloid leukemia patients. Last, I analyze the evolution of resistance in response to a new targeted therapy for chronic lymphocytic leukemia. Using the high resolution afforded by ultra-deep sequencing data, we show that resistance mutations are present at very low frequencies pre-treatment and expand upon initiation of treatment, with key implications for cancer monitoring and resistance.