Phylogenetic Tree

Summary

A phylogenetic tree in cancer genomics is a reconstruction of the ancestral relationships among tumor subclones, inferred from somatic mutation profiles across one or more tumor samples. Each node in the tree represents a clone (a population of genetically identical cells), each branch represents a lineage of descent, and mutations are mapped to branches — truncal mutations on the root-to-tip trunk (present in all cells), private mutations on terminal branches (present only in one subclone). The tree constrains the evolutionary history: it specifies which clones descended from which, the order of mutation acquisition, and the timing of divergence events. Phylogenetic reconstruction is the endpoint of the subclonal-reconstruction pipeline: VAF → CCF → subclone clustering → tree inference (Tarabichi et al., 2021).

Structure

A cancer phylogenetic tree encodes several distinct types of evolutionary information:

Trunk (clonal) mutations. Mutations present in all cancer cells. These occurred before the most recent common ancestor (MRCA) of the sampled population and are shared by all subclones. Trunk mutations include the tumor’s founder driver events.

Branch (subclonal) mutations. Mutations present in a subset of cancer cells. These occurred after the MRCA, in the lineage leading to one or more subclones. Shared branch mutations define clades — groups of subclones descended from a common intermediate ancestor.

Private mutations. Mutations present in only one subclone. These are the most recent events in the tree and define the terminal branches.

The tree topology — whether the tree is linear (one branch at each divergence), branching (multiple co-existing subclones), or star-like (many subclones radiating from the trunk) — corresponds to the evolutionary mode (intratumor-heterogeneity §4).

Inference

Phylogenetic trees are inferred from mutation clusters (subclones) using the following constraints:

CCF constraints. A parent clone cannot have lower CCF than its descendant — the descendant cannot be present in more cells than the ancestor that gave rise to it. This imposes a nesting rule: if clone A is ancestral to clone B, CCF(A) ≥ CCF(B) in every sample.

The crossing rule. In multi-region sequencing data, if clone A has higher CCF than clone B in one region but lower in another, they cannot be ancestor-descendant — they must be sibling clones in a branching phylogeny (Tarabichi et al., 2021). See crossing-rule.

Pigeonhole principle. The sum of CCFs of sibling clones cannot exceed 1.0 (100% of cancer cells). If A and B are sibling subclones descended from the same parent, CCF(A) + CCF(B) ≤ 1.0 (plus the parent’s residual population). This principle constrains how many sibling subclones can coexist at detectable frequency.

Tree-building algorithms. Methods like PhyloWGS, PyClone, and SciClone integrate SNV and CNA data to jointly infer subclone clusters and tree topology. They use Bayesian nonparametric models (Dirichlet processes) to determine the number of subclones and Markov chain Monte Carlo to sample the posterior distribution over tree topologies (Tarabichi et al., 2021).

CNA-Aware Phylogenetics

CNAs complicate phylogenetic inference in two ways:

Multiplicity uncertainty. An SNV on an amplified segment may be present on one, some, or all of the amplified copies. Different multiplicity assumptions produce different CCF estimates and therefore different tree topologies. Without long-read phasing or multi-sample data, multiplicity is underdetermined.

CNA as a phylogenetic character. CNAs can be used as phylogenetic markers in addition to SNVs: a shared CNA event (e.g., 9p loss in two subclones) may indicate common ancestry, while independent CNA events at the same locus indicate convergent evolution. Distinguishing shared ancestry from convergence requires multi-region data — a single sample cannot tell whether the same CNA in two subclones came from one ancestral event or two independent events.

Limitations

The tree is a model, not an observation. Phylogenetic trees are inferred under assumptions (infinite sites model — each mutation occurs exactly once; no back-mutation; no parallel evolution) that are violated in cancer, particularly in CIN-high tumors where convergent CNA is common.

Single-sample trees are underdetermined. Without multi-region data, many tree topologies are consistent with the same CCF distribution. The crossing rule — the strongest constraint on topology — is inapplicable in single-sample data.

Low-frequency subclones are invisible. Subclones below the detection floor (~0.10–0.15 CCF at 100×) are absent from the tree. The tree represents the visible clonal architecture, not the full evolutionary history.

Tree inference is sensitive to CNA correction. Incorrect copy-number calls propagate into CCF estimates and can produce wrong tree topologies — an ancestral clone with an uncorrected CNA may appear to have lower CCF than its descendant, violating the nesting rule and producing an impossible or incorrect tree (Tarabichi et al., 2021).