A statistical approach for tracking clonal dynamics in cancer using longitudinal next-generation sequencing data.

Dimitrios V Vavoulis,Jenny C Taylor,Anthony Cutts,Anna Schuh,Peter Robinson

doi:10.1093/bioinformatics/btaa672

Abstract

MotivationTumours are composed of distinct cancer cell populations (clones), which continuously adapt to their local micro-environment. Standard methods for clonal deconvolution seek to identify groups of mutations and estimate the prevalence of each group in the tumour, while considering its purity and copy number profile. These methods have been applied on cross-sectional data and on longitudinal data after discarding information on the timing of sample collection. Two key questions are how can we incorporate such information in our analyses and is there any benefit in doing so?ResultsWe developed a clonal deconvolution method, which incorporates explicitly the temporal spacing of longitudinally sampled tumours. By merging a Dirichlet Process Mixture Model with Gaussian Process priors and using as input a sequence of several sparsely collected samples, our method can reconstruct the temporal profile of the abundance of any mutation cluster supported by the data as a continuous function of time. We benchmarked our method on whole genome, whole exome and targeted sequencing data from patients with chronic lymphocytic leukaemia, on liquid biopsy data from a patient with melanoma and on synthetic data and we found that incorporating information on the timing of tissue collection improves model performance, as long as data of sufficient volume and complexity are available for estimating free model parameters. Thus, our approach is particularly useful when collecting a relatively long sequence of tumour samples is feasible, as in liquid cancers (e.g. leukaemia) and liquid biopsies.Availability and implementationThe statistical methodology presented in this paper is freely available at github.com/dvav/clonosGP.Supplementary information Supplementary data are available at Bioinformatics online.

Highlights

It is well known that cancer cells undergo a process of Darwinian evolution in response to selective pressures in their local micro-environment, for example as a result of therapeutic intervention[1, 2]
We test various model configurations on whole genome (WGS), whole exome (WES) and targeted sequencing (TGS) data from patients with chronic lymphocytic leukaemia (CLL; [23, 24]), on data from the liquid biopsy of a patient with melanoma[25] and on synthetic data, and we demonstrate that incorporating temporal information in our analysis can boost the performance of clonal deconvolution
We conducted a series of computational experiments on WES and WGS data from patients with CLL[23, 24], on TGS data from the liquid biopsy of a patient with melanoma[25] and on simulated data

Summary

Introduction

It is well known that cancer cells undergo a process of Darwinian evolution in response to selective pressures in their local micro-environment, for example as a result of therapeutic intervention[1, 2] This induces cell propagation and diversification during tumour growth, which result in a heterogeneous population of phylogenetically related, but genotypically and phenotypically distinct cancer cell populations, known as clones. Tumour heterogeneity is clinically important because it complicates the molecular profiling of tumours and enables the fittest cancer cells to escape treatment leading to relapse Monitoring this process of continuous adaptation requires a detailed characterisation (through the use of next-generation sequencing, bioinformatics and statistical analysis) of the somatic aberrations harboured by the tumour at various time points over the course of the disease. Current statistical methodologies seek to identify the number of clones in a tumour, their somatic mutation content, prevalence and phylogenetic relations and they can be used for the analysis of cross-sectional data (obtained, for example, through multiple biopsies from the same patient) or longitudinal data after discarding any information on the timing of tissue sample collection[7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22]

Methods

Results

Conclusion