Reconstruction of clone- and haplotype-specific cancer genome karyotypes from bulk tumor samples

Abstract

Many cancer genomes are extensively rearranged with highly aberrant chromosomal karyotypes. These genome rearrangements, or structural variants, can be detected in tumor DNA sequencing data by abnormal mapping of sequence reads to the reference genome. However, nearly all cancer sequencing to date is of bulk tumor samples which consist of a heterogeneous mixture of normal cells and subpopulations of cancers cells, or clones, that harbor distinct somatic structural variants. We introduce a novel algorithm, Reconstructing Cancer Karyotypes (RCK), to reconstruct haplotype-specific karyotypes of one or more rearranged cancer genomes, or clones, that best explain the read alignments from a bulk tumor sample. RCK leverages specific evolutionary constraints on the somatic mutation process in cancer to reduce ambiguity in the deconvolution of admixed DNA sequence data into multiple haplotype-specific cancer karyotypes. In particular, RCK relies on generalizations of the infinite sites assumption that a genome rearrangement is highly unlikely to occur at the same nucleotide position more than once during somatic evolution. RCK's comprehensive model allows us to incorporate information both from short and long-read sequencing technologies and is applicable to bulk tumor samples containing a mixture of an arbitrary number of derived genomes. We compared RCK to the state-of-the-art method ReMixT on a dataset of 17 primary and metastatic prostate cancer samples. We demonstrate that ReMixT's limited support for heterogeneity and lack of evolutionary constrains leads to reconstruction of implausible karyotypes. In contrast, RCK's infers cancer karyotypes that better explain read alignments from bulk tumor samples and are consistent with a reasonable evolutionary model. RCK's reconstructions of clone-and haplotype-specific karyotypes will aid further studies of the role of intra-tumor heterogeneity in cancer development and response to treatment. RCK is available at https://github.com/raphael-group/RCK.

Recovering rearranged cancer chromosomes from karyotype graphs

“…Lemma 2. For an IAG = ( , , ) the number | | of paths in any Eulerian decomposition of is determined as follows [1]:…”

“…Below we show the previously known result that for IAG every connected component ∈ + ( ) can be decomposed into paths-only, and every connected components ∈ 0 ( ) can be covered by a single cycle: Lemma 3. For a decomposable IAG = ( , , ) the size | | = | | + | | of any of its minimal (cardinality-wise) Eulerian decompositions is equal to [1]:…”

“…To evaluate the performance of CCR we applied it on cancer karyotype graphs originally produced with RCK on 17 prostate cancer samples [1]. We note that the karyotypes inferred in the previous study are both clone-and haplotypespecific, but for the purposes of this study we do not consider intra-sample relations between distinct clones that comprise it.…”

“…Previously proposed algorithms enabled the inference of copy number aberrations (CNA) (e.g., deletion, amplification) of genomic segments [8,27,7] and novel adjacencies (NA) between genomic loci that are distant in the reference but are adjacent in cancer genomes [13,26,5,22], in sequenced tumor samples. In the more recent study [1] a novel methodological framework RCK was proposed for reconstruction of cancer genomes karyotype graphs, taking into account both the CNA and NA signals, the non-haploid nature of both the reference and the derived genomes, and, when applicable, possible sample heterogeneity. The karyotype graph provides a much more accurate and complete description of the rearranged cancer genome than either of CNA or NA profiles on their own, but it falls short of describing the actual linear/circular structure of the rearranged chromosomes, and instead provides an alignment of the rearranged chromosomes onto the reference.…”

“…Here we consider the problem of inferring sequential structure of rearranged linear/circular chromosomes in a cancer genome given its karyotype graph representation. In previous studies either general observations about necessary and sufficient conditions on the karyotype graphs structure for the existence of the underlying cancer genome [17,1], or attempts at extracting information about specific local rearranged structures (e.g., amplicons, complex rearrangements, etc) [3,16,4] were made. In the present study we formulate a Eulerian Decomposition Problem (EDP) for a cancer karyotype graph with the objective of finding a covering collection of linear and/or circular paths/cycles (i.e., chromosomes).…”