Reconstructing the Evolutionary History of Cancer using Single-Cell Sequencing Data

Abstract

Intratumoural heterogeneity (ITH) and evolutionary forces are major factors in cancer development,treatment response and recurrence. To better understand these phenomena, evolutionary history reconstruction methods have been created to construct cancer trees, graphical objects that indicate mutation ordering, subclonal genotypes and subclone phylogeny. Existing methods typically suffer from one of several drawbacks, such as performance drop off when constructing large cancer trees or failing to report reconstruction uncertainty. To address this, I developed scPairtree, a novel single cell-based reconstruction method capable of constructing large cancer trees complete with uncertainty. It incorporates a mutation pairs relationship model with which it constructs a pairs tensor, an object that stores estimated probability distributions of mutation pair relationships. The pairs tensor guides a Markov chain Monte Carlo (MCMC)-based tree sampling algorithm such that tree proposals move poorly placed nodes to more suitable locations. This results in the algorithm converging on the tree posterior in fewer iterations than comparable MCMC-based methods. I have also developed a novel tree sampling method by which trees are sampled directly from the pairs tensor (DFPT), which are then used in importance sampling to infer properties of the tree posterior. Trees from either method are used to construct a consensus graph, a graphical object that captures the most likely cancer tree while visually representing uncertainty. I also present my work constructing a cancer tree for a basal breast cancer in collaboration withDr. Elena Kuzmin. Of particular interest was mutation ordering with respect to chr4p loss, which had been identified as recurrent in basal breast cancers. Data was derived from both bulk and low coverage single-cell sequencing performed on samples from a single tumour. I manually constructed a cancer tree by inferring the haplotype-specific copy number profiles for each single cell, clustering these cells into subclones, generating pseudobulk data using the clustered single cell data, inferring subclonal genotypes by performing set analysis on the point mutations called in each sequencing experiment, and defining heuristics to infer the copy number profile of the most recent common ancestor of the sampled cells. From the resulting clone tree we identified chr4p copy-neutral loss of heterozygosity as an early event with additional copy number loss occurring subclonally.