Investigating the heterogeneity of selective pressures and driver events in cancer

Abstract

Inter-tumour heterogeneity is a significant barrier to the effective treatment of cancer. Heterogeneous molecular features of tumours cause patients to respond differently to therapies. The cancer community has addressed this problem to an extent by developing molecular stratifications and targeted therapies. However, formulating effective treatment strategies for individual patients remains challenging, and a deeper molecular understanding of inter-tumour heterogeneity is required to improve patient outcomes. In this thesis, I investigate inter-tumour heterogeneity from two perspectives. First, I consider germline variation as a driving force behind inter-tumour heterogeneity. While heterogeneity is largely the result of stochastic processes, inherited genetic differences between patients can give rise to patient-specific selective pressures acting on somatic alterations during cancer evolution. However, this phenomenon is not yet fully understood. I analyse how germline variants that perturb the function of biological pathways affect the frequency of somatic driver alterations at the gene and pathway levels, using data from oesophageal adenocarcinoma. By addressing the methodological and statistical challenges involved in this analysis, I find evidence that ATM and its interactors play an important and as-yet unreported role in the biology of oesophageal adenocarcinoma. In particular, I find that perturbations to these genes can substitute for driver alterations in TP53, which is by far the most frequently altered gene in oesophageal adenocarcinoma. This analysis also uncovers evidence that ATM acts as a cancer predisposition gene and a tumour suppressor gene in oesophageal adenocarcinoma. Second, I address the question of how to identify the aspects of inter-tumour heterogeneity that are most relevant to cancer biology and therapy, i.e. cancer drivers. The research community has identified many hundreds of driver genes across cancer types, and I describe the curation of a database, the Network of Cancer Genes (NCG), to capture this information. NCG also annotates the systems-level properties of reported cancer genes. I show that cancer genes are distinguished from other human genes by an array of these systems-level properties, and develop a machine learning method to use these properties to identify novel driver genes. This method (sysSVM2) is capable of identifying driver genes at the level of individual patients, which overcomes the persistent problem of a portion of patients having too few driver alterations to explain the onset of cancer. It is also particularly useful in rare cancer types for which large-scale sequencing studies are infeasible. Using the properties of canonical driver genes to identify drivers in individual patients in this way can help to further the goals of precision oncology and overcome the challenge presented by intertumour heterogeneity. Taken as a whole, this thesis presents novel research that sheds light on both the causes of inter-tumour heterogeneity and how to interpret the heterogeneous molecular landscape of cancer.

Date of Award	1 Apr 2021
Original language	English
Awarding Institution	King's College London
Supervisor	Francesca Ciccarelli (Supervisor)