Integrative inference of subclonal tumour evolution from single-cell and bulk sequencing data

Abstract

Understanding the evolutionary history and subclonal composition of a tumour represents one of the key challenges in overcoming treatment failure due to resistant cell populations. Most of the current data on tumour genetics stems from short read bulk sequencing data. While this type of data is characterised by low sequencing noise and cost, it consists of aggregate measurements across a large number of cells. It is therefore of limited use for the accurate detection of the distinct cellular populations present in a tumour and the unambiguous inference of their evolutionary relationships. Single-cell DNA sequencing instead provides data of the highest resolution for studying intra-tumour heterogeneity and evolution, but is characterised by higher sequencing costs and elevated noise rates. In this work, we develop the first computational approach that infers trees of tumour evolution from combined single-cell and bulk sequencing data. Using a comprehensive set of simulated data, we show that our approach systematically outperforms existing methods with respect to tree reconstruction accuracy and subclone identification. High fidelity reconstructions are obtained even with a modest number of single cells. We also show that combining single-cell and bulk sequencing data provides more realistic mutation histories for real tumours.

Implications of non-uniqueness in phylogenetic deconvolution of bulk DNA samples of tumors

“…The most commonly used evolutionary model in cancer phylogenetics is the two-state perfect phylogeny model, where mutations adhere to the infinite sites assumption [8][9][10][11][12][13][14][15][16]. That is, for each mutation locus the actual mutation occurred exactly once in the evolutionary history of the tumor and was subsequently never lost.…”

Uncovering the subtype-specific temporal order of cancer pathway dysregulation

“…The latest developments are based on single-cell data (e.g. OncoNEM [43], SCITE [27], SiFit [53]) or simultaneously utilize single-cell and bulk sequencing to create synergy between the two data types (B-SCITE [32] and PhISCS [33]).…”

Inferring Cancer Progression from Single-cell Sequencing while Allowing Mutation Losses

“…Note that none of previous metrics account for ISA violations. We decided to compare SASC against SCITE [13]: while B-SCITE [19] is a clear improvement over SCITE, it combines single cell data with bulk sequencing data -since we do not manage the latter kind of data, a fair comparison is not feasible. OncoNEM [27] and SiFit [34] were excluded because they infer cell lineage progressions instead of mutational progression, therefore it is not possible to compare our predictions with theirs.…”

“…We measured the accuracy of SASC with two standard cancer progressions measures used in various studies [19,13] and a novel approach to test the quality of subclonal inference, defined as follows:…”

“…Various methods have been developed for this purpose [13,27,34], some of them introducing a hybrid approach of combining both SCS and VAF data [25,19,29]. Most of these methods, however, rely on the Infinite Sites Assumption (ISA), which states that a mutation is acquired at most once in the tree and is never lost.…”