Clonal Evolution

Subclonal Reconstruction

Properties1
tagsmethods, phylogenetics

5 min read

Subclonal Reconstruction

Summary

Subclonal reconstruction is the computational inference of a tumor’s clonal composition — clone genotypes, frequencies, and phylogenetic relationships — from bulk DNA sequencing data (Tarabichi et al. 2021). The standard workflow proceeds through five steps: copy number reconstruction, VAF-to-CCF translation (correcting for purity and copy number), SNV clustering, multiplicity determination, and phylogenetic tree construction. Multi-sample reconstruction adds the crossing rule: if clone A has higher CCF than clone B in one region but lower in another, they are sibling clones in a branching phylogeny. Key tools include PyClone, PhyloWGS, and DeCiFer (Satas et al. 2022), which introduced the descendant cell fraction (DCF) to address the constant mutation multiplicity assumption that biases standard CCF estimates in copy-number-aberrant tumors.

Subclonal reconstruction is the computational inference of a tumor’s clonal composition — clone genotypes, clone frequencies, and phylogenetic relationships — from bulk DNA sequencing data (Tarabichi et al., 2021). It has become a pillar of cancer evolution studies, providing insight into the clonality and relative ordering of mutations and mutational processes.

Three Core Tasks

Subclonal reconstruction involves three key aspects (Tarabichi et al., 2021):

  1. Clone identification: Characterizing the major cell populations by identifying which somatic mutations co-occur in each clone
  2. Frequency estimation: Quantifying the proportion of cells from each clone (its cancer-cell-fraction or cellular prevalence)
  3. Phylogenetic inference: Reconstructing the ancestral relationships between clones — the phylogenetic-tree

The Standard Workflow

For single-sample reconstruction:

  1. Copy number reconstruction: Infer regions of clonal and subclonal copy number change from read depth (logR) and B-allele frequency (BAF) data
  2. VAF-to-CCF translation: Convert observed variant-allele-fraction to cancer-cell-fraction by correcting for sample purity and local copy number
  3. SNV clustering: Group mutations with similar CCFs, assuming they belong to the same clone (using Bayesian Dirichlet processes, MCMC sampling)
  4. Multiplicity determination: Infer how many allelic copies carry each mutation (integer programming)
  5. Phylogenetic tree construction: Build the clonal phylogeny using maximum parsimony or probabilistic models

Multi-sample reconstruction adds the crossing-rule: if clone A has higher CCF than clone B in one region but lower in another, they cannot be in a simple ancestor-descendant relationship — they are sibling clones in a branching phylogeny.

Key Tools

PyClone (Roth et al., 2014), PhyloWGS (Deshwar et al., 2015), SciClone (Miller et al., 2014), and ABSOLUTE (Carter et al., 2012) are the most widely used tools. Tarabichi et al. (2021) recommend comparing results from multiple tools and validating with orthogonal data (e.g., FISH for CNAs) whenever possible.

DeCiFer and the DCF. Satas et al. (2022) introduced DeCiFer, an algorithm that addresses a fundamental limitation of standard CCF estimation: the assumption that every cell with an SNV contains the same number of copies of the SNV (the constant mutation multiplicity assumption, or CMM). In tumors with copy-number aberrations, SNV multiplicity varies across cells, making CCF estimates unreliable. DeCiFer replaces CMM with the single-sample copy-number (SSCN) assumption and introduces the descendant cell fraction (DCF) — a statistic that quantifies both the current prevalence of an SNV AND its past evolutionary history, including mutation losses due to copy-number aberrations. On 49 prostate cancer samples, DeCiFer reclassified over 23,000 SNVs and produced more parsimonious phylogenetic reconstructions than PyClone or PhyloWGS. The DCF is a more principled metric than standard CCF for tumors with substantial CNA burden, and the ITH empirical test design should consider DCF-based SMF as a sensitivity analysis when copy-number calls are available. See satas2022-decifer.

SNV Discarding in CNA Regions

A critical and often underappreciated aspect of the standard workflow is that SNVs in CNA-affected regions are systematically excluded from phylogenetic inference. This is not a bug — it is a rational response to the multiplicity underdetermination problem copy-number-alteration §5.4 — but its scale in high-aneuploidy tumors can decimate the usable SNV set.

Where discarding happens:

  • Step 2 → 3 (CCF → Clustering): After VAF-to-CCF translation with the m = 1 assumption, SNVs with unreliable CCF estimates are pre-filtered before clustering. This is a hard exclusion — they are not sent to the Bayesian mixture model (PyClone, PhyloWGS) because they would create spurious clusters or fragment real ones.
  • Step 4 → 5 (Multiplicity → Tree): SNVs whose multiplicity cannot be resolved at a confidence threshold are excluded from phylogenetic tree construction. They have a CCF estimate but no clone assignment — they contribute nothing to the tree topology.

DeCiFer and the DCF. Satas et al. (2022) introduced DeCiFer specifically to rescue CNA-region SNVs by replacing the constant mutation multiplicity (CMM) assumption with the single-sample copy-number (SSCN) assumption and introducing the descendant cell fraction (DCF). On 49 prostate cancers, DeCiFer reclassified >23,000 SNVs and produced more parsimonious trees than PyClone or PhyloWGS (satas2022-decifer).

Consequences:

EffectMechanism
Molecular clock decimationMost passenger SNVs — the clock mechanism — are discarded, destroying temporal information (passenger-mutation)
Diploid-region biasSurviving SNVs disproportionately sample genomically stable regions; pre-CNA history is preferentially lost
CNA-region subclone blindnessSubclones defined by mutations in amplified/deleted regions are invisible to the tree
Tool-dependent severityPyClone pre-filters; PhyloWGS co-estimates but with wide intervals; DeCiFer rescues some

Limitations

The process is error-prone. Key challenges include:

  • Copy number confounds VAF interpretation — misclassification can erroneously create or merge clones
  • SNVs in CNA-affected regions are systematically discarded from clustering and tree-building (see above)
  • Detection limits (~0.05–0.10 CCF at standard depths) mean minor subclones are invisible
  • The “infinite sites” assumption (each mutation occurs exactly once) can be violated in large tumors
  • Single-sample reconstruction undersamples the tumor’s spatial diversity

Graph View

Backlinks