Estimation of cancer cell fractions and clone trees from multi-region sequencing of tumors

Abstract

Multi-region sequencing of one or multiple biopsies of solid tumors from a patient can be used to improve our understanding of the diversity of subclones in the patient’s tumor and shed light on the evolutionary history of the disease. Due to the large number of possible evolutionary relationships between clones and the fundamental uncertainty of the mutational composition of subclones, elucidating the most probable evolutionary relationships poses statistical and computational challenges. We developed a Bayesian hierarchical model called PICTograph to model uncertainty in the assignment of mutations to subclones and an approach to reduce the space of possible graphical models that postulate their evolutionary origin. Compared to available methods, our approach provided more consistent and accurate estimates of cancer cell fractions and better tree topology reconstruction over a range of simulated clonal diversity. Application of PICTograph to whole exome sequencing data of individuals with pancreatic cancer precursor lesions confirmed known early occurring mutations and indicated substantial molecular diversity, including multiple distinct subclones (range 6 - 12) and intra-sample mixing of subclones. As the complete evolutionary history for some patients was not identifiable, we used ensemble-based visualizations to distinguish between highly probable evolutionary relationships recovered in multiple models from uncertain relationships occurring in a small subset of models. These analyses indicate that PICTograph provides a useful approximation to evolutionary inference, particularly when the evolutionary course of a patient’s cancer is complex.