Characterizing genetic intra-tumor heterogeneity across 2,658 human cancer genomes

April 2021Cell

Dentro SC, Leshchiner I, Haase K, Tarabichi M, Wintersinger J, Deshwar AG, Yu K, Rubanova Y, Macintyre G, Demeulemeester J, Vázquez-García I, Kleinheinz K, Livitz DG, Malikic S, Donmez N, Sengupta S, Anur P, Jolly C, Cmero M, Rosebrock D, Schumacher SE, Fan Y, Fittall M, Drews RM, Yao X, Watkins TBK, Lee J, Schlesner M, Zhu H, Adams DJ, McGranahan N, Swanton C, Getz G, Boutros PC, Imielinski M, Beroukhim R, Sahinalp SC, Ji Y, Peifer M, Martincorena I, Markowetz F, Mustonen V, Yuan K, Gerstung M, Spellman PT, Wang W, Morris QD, Wedge DC, Van Loo P, PCAWG Evolution and Heterogeneity Working Group and the PCAWG Consortium.

During the early part of my graduate work, I contributed to the Pan-Cancer Analysis of Whole Genomes (PCAWG), an international effort to use whole-genome sequencing to study 2658 newly sequenced cancers, where I was part of the Evolution and Heterogeneity working group. My primary contribution was to build consensus profiles of copy-number aberrations (CNAs) for each of the cancers. The working group had multiple individual methods for characterizing the CNAs for each cancer, but their results differed radically with respect to the genomic breakpoints where copy-number status changed, the total copy number for each genomic segment, the allele-specific copy number, and even whether there was enough information to infer copy-number status at a given locus. I developed a benchmark suite that revealed considerable differences in precision and recall for each method relative to a ground-truth dataset, then devised a tiered consensus scheme that first produced consensus genomic breakpoints where copy-number status changed, and then used the results of individual methods on each resulting segment to infer consensus CNAs, drawing on the strengths and weaknesses of each method revealed by the benchmark. This work was essential for both of the papers our working group produced, as it lay the groundwork for characterizing the evolutionary history of each cancer. The first ten authors on this paper, including me, are co-first authors.

Colorectal Cancer Cells Enter a Diapause-like DTP State to Survive Chemotherapy

January 2021Cell

Rehman SK, Haynes J, Collignon E, Brown KR, Wang Y, Nixon AML, Bruce JP, Wintersinger JA, Mer AS, Lo EBL, Leung C, Lima-Fernandes E, Pedley NM, Soares F, McGibbon S, He HH, Pollet A, Pugh TJ, Haibe-Kains B, Morris Q, Ramalho-Santos M, Goyal S, Moffat J, O’Brien CA.

This paper, led by Dr. Catherine O'Brien's lab, showed that colorectal cancer cells can enter a reversible drug-tolerant persister state to evade death from chemotherapy and other targeted agents. I contributed to this project by using Pairtree to build clone trees for two cancers using multiple samples from each, then developing two information-theoretic measures to quantify the degree of genomic heterogeneity present in each cancer sample. This showed that multiple genomically distinct cell subpopulations were maintained in every tissue sample, including those taken before treatment, those that developed treatment resistance, and those that were seeded from the treatment-resistant samples. Though genomic heterogeneity was reduced in the treatment-resistant samples, it rebounded in the regrowth tumours seeded from them, lending support to the idea that drug tolerance arose through a non-genetic mechanism.

Relapse-Fated Latent Diagnosis Subclones in Acute B Lineage Leukemia Are Drug Tolerant and Possess Distinct Metabolic Programs.

April 2020Cancer Discovery

Dobson SM, García-Prat L, Vanner RJ, Wintersinger J, Waanders E, Gu Z, McLeod J, Gan OI, Grandal I, Payne-Turner D, Edmonson MN, Ma X, Fan Y, Voisin V, Chan-Seng-Yue M, Xie SZ, Hosseini M, Abelson S, Gupta P, Rusch M, Shao Y, Olsen SR, Neale G, Chan SM, Bader G, Easton J, Guidos CJ, Danska JS, Zhang J, Minden MD, Morris Q, Mullighan CG, Dick JE.

Dr. John Dick's lab analyzed 14 B-progenitor acute lymphoblastic leukemias (B-ALLs), investigating the genomic and metabolic differences that allowed some subclones within a patient's cancer to tolerate treatment and seed disease relapse. To enable this work, his lab xenografted diagnosis and relapse samples from each patient into multiple mice and then subjected them to targeted sequencing, allowing us to analyze patient and mouse xenograft samples jointly as different depictions of a single cancer. This approach yielded up to 90 samples per cancer. My contribution was constructing clone trees for each patient, which revealed the genomic identities of the distinct subclonal populations composing a cancer, and allowed us to track their evolutionary trajectory between diagnosis and relapse. Additionally, by comparing the subclonal composition of mouse xenograft samples seeded from similar initial conditions, we gained insights into the stochasticity of cancer evolution. Existing methods for building clone trees failed when faced with such rich data, composed of up to 90 samples per cancer that revealed the existence of up to 26 subclonal populations. These failures forced us to consider why current approaches failed and how we could improve on them, leading to the creation of Pairtree.

A community effort to create standards for evaluating tumor subclonal reconstruction

January 2020Nature Biotechnology

Salcedo A, Tarabichi M, Espiritu SMG, Deshwar AG, David M, Wilson NM, Dentro S, Wintersinger JA, Liu LY, Ko M, Sivanandan S, Zhang H, Zhu K, Yang T-HO, Chilton JM, Buchanan A, Lalansingh CM, P’ng C, Anghel CV, Umar I, Lo B, Zou W, Simpson JT, Stuart JM, Anastassiou D, Guan Y, Ewing AD, Ellrott K, Wedge DC, Morris Q, Loo PV, Boutros PC.

This paper developed several novel means of comparing cancer subclonal reconstruction methods, then benchmarked leading methods to see how well they could recover a known truth from simulated data. I contributed by assisting with the benchmarking.

Kablammo: an interactive, web-based BLAST results visualizer

April 2015Bioinformatics

Wintersinger JA, Wasmuth JD.

As part of my bachelor's thesis, I developed Kablammo, a web-based tool for visualizing BLAST results. The method has been used by a number of other studies to produce figures for publication. As an extension of this work, I developed a dynamic programming algorithm for computing the optimal set of BLAST hits.