Biomathematics seminar – Calum Gabbutt (Imperial College London)
Title:
Cancer evolution is encoded within the epigenome, can be measured using Bayesian inference at scale and predicts patient outcomes in blood cancer.
Abstract:
Cancer’s aberrant growth is driven by evolutionary pressures, therefore measuring the evolutionary history of a cancer may be predictive of its future trajectory. However, cancer evolution cannot be observed directly in vivo, and clinical samples represent a snapshot of a tumour. Here, we modelled how the evolution of a cancer is encoded within its epigenome and developed a Bayesian method to precisely characterise a cancer’s evolutionary history from low-cost, bulk methylation array data. Coupling these inferences with clinical data, we showed that the evolutionary history of a cancer is highly predictive of future outcomes.
We recently characterised CpG methylation sites within the genome that are vulnerable to stochastic, selectively-neutral epimutations, and thus record lineage information in human cells, which we termed fluctuating CpGs (fCpGs). In this work, we developed an unsupervised machine learning algorithm to define a set of 978 pan-lymphoid cancer fCpGs. Combining these data with a simulation-based model of how the patterns of fCpGs within a bulk population vary as a function of different evolutionary parameters, we inferred the evolutionary history of 1,976 lymphoid malignancies. The inferred tumour growth rates, ages and epimutation rates of different lymphoid cancers varied across orders of magnitude. In chronic lymphocytic leukaemia, a relatively indolent disease, the cancer growth rate was highly prognostic in two independent clinical cohorts.
Hence, the early evolutionary history of a cancer influences its future disease course, is encoded within the patterns of fCpGs and can be measured using Bayesian methods.