Assessing intratumor heterogeneity and tracking longitudinal and spatial clonal evolutionary history by next-generation sequencing

Abstract

Cancer is a disease driven by evolutionary selection on somatic genetic and epigenetic alterations. Here, we propose Canopy, a method for inferring the evolutionary phylogeny of a tumor using both somatic copy number alterations and single-nucleotide alterations from one or more samples derived from a single patient. Canopy is applied to bulk sequencing datasets of both longitudinal and spatial experimental designs and to a transplantable metastasis model derived from human cancer cell line MDA-MB-231. Canopy successfully identifies cell populations and infers phylogenies that are in concordance with existing knowledge and ground truth. Through simulations, we explore the effects of key parameters on deconvolution accuracy and compare against existing methods. Canopy is an open-source R package available at https://cran.r-project.org/web/packages/Canopy/.

Keywords: cancer evolution; cancer genomics; clonal deconvolution; intratumor heterogeneity; phylogeny inference.

Fig. 1.

Tumor phylogeny, observed input, and inferred output of Canopy. (A) Phylogeny of tumor progression as a bifurcating tree with SNAs and CNAs along the branches. Longitudinal and/or spatial samples offer different snapshot of subpopulations, represented by tree leaves. The lengths of the branches are arbitrary—because without further strong assumptions, we cannot infer branch length from these data. (B) Observed VAFs, major copies, and minor copies across samples. Matrix $Y$ indicates whether an SNA resides in a CNA. (C) Matrix decomposition by Canopy. Genotyping matrix $Z$ represents the positions of the SNAs in the phylogeny. ${\tilde{C}}^M$ and ${\tilde{C}}^m$ encode major and minor copy number of each clone. $H$ specifies SNA–CNA phasing—whether SNAs reside in major or minor copies. Clonal frequency matrix $P$ is shown as part of A.

Fig. 2.

Three cases of SNA–CNA phase and order. Different phases and orders of CNA and the SNA it affects are shown with clonal histories concordant with Fig. 1. Major and minor copies are in blue and red, respectively; SNA mutational loci are shown as stars. (A) CNA precedes SNA. SNA resides in only one chromosomal copy. (B) CNA and SNA are on two separate branches. SNA is unaffected by CNA. (C) SNA precedes CNA. Scenario where major copies contain the SNA is shown. SNA4 from Fig. 1 is unaffected by CNA and is not shown.

Fig. 3.

Deconvolution accuracy and clustering quality via simulation studies. Various parameters show effects on deconvolution accuracy (measured by the percentage of wrongly labeled $Z$ elements) and preclustering quality (measured by the clustering purity; Methods). Corresponding RMSE of the $P$ matrix is shown in SI Appendix, Fig. S7. Canopy is compared against Clomial and SciClone and is shown to have better performance. (A and B) WGS compensates its low sequencing depth with more profiled mutations. (C) Increasing sample size helps solve reconstruction ambiguity. (D and E) Number of subclones is negatively correlated with deconvolution accuracy and preclustering quality.

Fig. 4.

Clonal history of transplantable metastasis model MDA-MB-231 with validation by SCP samples. (A) Transplantable model system of MDA-MB-231. Parental line is injected into mouse models and induces organ-specific metastasis. Sublines are derived from single or multiple cell(s) from different metastatic sites. (B) Observed VAFs of somatic SNAs, which reside in nested CNAs. CNA input is shown in SI Appendix, Fig. S13. Canopy takes both SNA and CNA input. (C) BIC as a model selection method to determine the number of subclones. (D) Clonal tree reconstructed by Canopy. Organ-specific subclones (clone 2 and 3) acquire additional mutations from the parental clone (clone 1) and dominate the metastasis. SCP samples successfully validate the subclones and confirm Canopy’s inferred phylogeny.

Fig. 5.

Clonal architecture of breast cancer initial engraftment and passage xenograftment. Tumor sample SA494T and its subsequent xenograft SA494X4 are whole-genome sequenced with SNAs validated by deep amplicon resequencing and CNAs inferred by TITAN. (A) SNA and CNA input of Canopy. VAFs of four SNA clusters and three CNA-affected SNAs are shown in the top panel. Heat map of observed major and minor copy numbers are shown in the bottom panel. (B) BIC as a model selection metric to determine the number of subclones. (C) The most likely tree returned by Canopy based on the mutational profiling. Extreme selection of minor clones is imposed on engraftment. SA494T and SA494X4 bear two mutually exclusive sets of mutations in addition to shared ancestral mutations. (D) Mutation clusters inferred by the Pyclone model.