Tumour evolution at single-cell resolution

Tumours are heterogeneous populations of cells that evolve dynamically in response to selective pressures. Bulk genome sequencing has been used successfully to characterize constituent populations in tumours. However, it is difficult to detect very low prevalence mutations in bulk genomes, and even more challenging to determine which variants co-occur in the same cells. These challenges can be overcome by sequencing the genomes of individual cells, but single-cell genome sequencing methods present their own obstacles. Sample contamination, biases introduced through amplification, and high cost have thus far precluded large-scale studies of single-cell genomes. We present, in three chapters, computational and experimental approaches to the study of tumour evolution at single-cell resolution. The first chapter examines clonal dynamics in breast cancer patient-derived xenograft models. Using a Bayesian probabilistic method, we infer the dynamics of sub-clonal populations, and validate our predictions with targeted single-cell measurements. In the second chapter, we present a new method for preparing single-cell libraries for whole-genome sequencing. Using a direct tagmentation and indexing strategy without whole-genome preamplification, we enable cost-effective generation of low-depth single-cell libraries with uniform coverage. We show that by merging single-cell genomes in silico, we can produce clonal and pseudo-bulk genomes with uniformity equivalent to standard bulk genomes. As such, in a single experiment, our direct library preparation (DLP) approach permits detailed characterization of copy number heterogeneity at the single-cell level, and inference of single-nucleotide variants at the clone or population level. The final chapter presents a new probabilistic model for improving inference of copy number variants from low-depth single-cell genomes. Using a mixture of hidden Markov models, we assign each cell to a ploidy sub-population, and borrow strength across cells to jointly estimate the model parameters. This research paves the way for large-scale studies of genomic heterogeneity at single-cell resolution, and presents new avenues of investigation into the reproducibility and predictability of tumour evolution in response to selective pressures.