DeCiFering the Elusive Cancer Cell Fraction in Tumor Heterogeneity and Evolution

Abstract

Most tumors are heterogeneous mixtures of normal cells and cancer cells, with individual cancer cells distinguished by somatic mutations that accumulated during the evolution of the tumor. The fundamental quantity used to measure tumor heterogeneity from somatic single-nucleotide variants (SNVs) is the Cancer Cell Fraction (CCF), or proportion of cancer cells that contain the SNV. However, in tumors containing copy-number aberrations (CNAs) -- e.g., most solid tumors -- the estimation of CCFs from DNA sequencing data is challenging because a CNA may alter the mutation multiplicity, or number of copies of an SNV. Existing methods to estimate CCFs rely on the restrictive Constant Mutation Multiplicity (CMM) assumption that the mutation multiplicity is constant across all tumor cells containing the mutation. However, the CMM assumption is commonly violated in tumors containing CNAs, and thus CCFs computed under the CMM assumption may yield unrealistic conclusions about tumor heterogeneity and evolution. The CCF also has a second limitation for phylogenetic analysis: the CCF measures the presence of a mutation at the present time, but SNVs may be lost during the evolution of a tumor due to deletions of chromosomal segments. Thus, SNVs that co-occur on the same phylogenetic branch may have different CCFs. In this work, we address these limitations of the CCF in two ways. First, we show how to compute the CCF of an SNV under a less restrictive and more realistic assumption called the Single Split Copy Number (SSCN) assumption. Second, we introduce a novel statistic, the descendant cell fraction (DCF), that quantifies both the prevalence of an SNV and the past evolutionary history of SNVs under an evolutionary model that allows for mutation losses. That is, SNVs that co-occur on the same phylogenetic branch will have the same DCF. We implement these ideas in an algorithm named DeCiFer. DeCiFer computes the DCFs of SNVs from read counts and copy-number proportions and also infers clusters of mutations that are suitable for phylogenetic analysis. We show that DeCiFer clusters SNVs more accurately than existing methods on simulated data containing mutation losses. We apply DeCiFer to sequencing data from 49 metastatic prostate cancer samples and show that DeCiFer produces more parsimonious and reasonable reconstructions of tumor evolution compared to previous approaches. Thus, DeCiFer enables more accurate quantification of intra-tumor heterogeneity and improves downstream inference of tumor evolution.

Genomic evolution of non-small cell lung cancer patient-derived xenograft models

“…To reconstruct tumour phylogenetic trees of each tumour from the identified somatic mutations, we developed a novel computational method to address three key challenges in phylogenetic reconstruction: (1) scaling to a high number of primary tumour and metastasis regions per patient, (2) correcting for complex evolutionary events, including mutation losses 50 , (3) removing biologically improbable clusters that either are driven by subclonal copy number or are not biologically compatible with the inferred evolutionary tree. This novel method has been extensively benchmarked and a manuscript detailing the steps as well as its application is currently in preparation.…”

CONIPHER: a computational framework for scalable phylogenetic reconstruction with error correction

“…Due to the presence of somatic copy number alterations (SCNAs) and normal cell admixtures, the VAF is not an accurate estimator of the population frequency of the variant. Therefore, most existing algorithms apply different approaches to correct the VAF for tumour purity and SCNAs to infer estimates of the cancer cell fraction (CCF) of a mutation, which de nes the proportion of cancer cells in the sample that carry the mutation 3,8,14 . To reconstruct clonal evolution, these computational methods cluster together mutations with similar CCFs in all samples sequenced into 'subclonal clusters', under the assumption that they are likely present in a similar set of cells and that represent a clonal expansion at a similar evolutionary time point 9,[11][12][13][14][15] .…”

“…Therefore, most existing algorithms apply different approaches to correct the VAF for tumour purity and SCNAs to infer estimates of the cancer cell fraction (CCF) of a mutation, which de nes the proportion of cancer cells in the sample that carry the mutation 3,8,14 . To reconstruct clonal evolution, these computational methods cluster together mutations with similar CCFs in all samples sequenced into 'subclonal clusters', under the assumption that they are likely present in a similar set of cells and that represent a clonal expansion at a similar evolutionary time point 9,[11][12][13][14][15] . Then, by nesting subclonal cluster CCFs based on evolutionary principles for constraining lineage relationships, such as the 'pigeonhole principle' and 'crossing rule' (Supplementary Methods), algorithms seek to infer the evolutionary ordering of clusters and reconstruct the full tumour phylogenetic tree 3,[11][12][13][16][17][18] (Table 1;…”

DeCiFering the subclonal composition of tumors

“…Previous methods either did not account for this limitation or only partially did so, typically through inelegant hacks. To address this, Satas et al (2021) propose the interesting concept of a descendant cell fraction (DCF). While the CCF is the proportion of cells carrying the SNV, the DCF is the proportion of cells that either themselves have the SNV or whose ancestors carried the SNV.…”

“…Thus, DCF allows the recovery of this fossil information. Satas et al (2021) have implemented these concepts in a novel algorithm called DeCiFer to improve estimation of CCFs in tumor samples. DeCiFer assesses potential genotypes of subclones detected through copy number changes by subjecting different pairs of multiplicities to evolutionary constraints.…”