Reconstructing evolutionary trajectories of mutation signature activities in cancer using TrackSig
Yulia Rubanova, Ruian Shi, Caitlin F. Harrigan, Roujia Li, Jeff Wintersinger, Nil Sahin, Amit G. Deshwar, Stefan C. Dentro, Ignaty Leshchiner, Moritz Gerstung, Clemency Jolly, Kerstin Haase, Maxime Tarabichi, Kaixian Yu, Santiago Gonzalez, Geoff Macintyre, David J. Adams, Pavana Anur, Rameen Beroukhim, Paul C. Boutros, David D.l. Bowtell, Peter J. Campbell, Shaolong Cao, Elizabeth L. Christie, Marek Cmero, Yupeng Cun, Kevin J. Dawson, Jonas Demeulemeester, Nilgun Donmez, Ruben M. Drews, Roland Eils, Yu Fan, Matthew W. Fittall, Dale W. Garsed, Gad Getz, Gavin Ha, Marcin Imieliński, Lara Jerman, Yuan Ji, Kortine Kleinheinz, Ju‐Hee Lee, Henry Lee-Six, Dimitri Livitz, Salem Malikić, Florian Markowetz, Iñigo Martincorena, Thomas J. Mitchell, Ville Mustonen, Layla Oesper, Martin Peifer, Myron Peto, Benjamin J. Raphael, Daniel Rosebrock, S. Cenk Sahinalp, Adriana Salcedo, Matthias Schlesner, Steven E. Schumacher, Subhajit Sengupta, Seung Jun Shin, Oliver Spiro, Lincoln D. Stein, Ignacio Vázquez-Garćıa, Shankar Vembu, David A. Wheeler, Tsun-Po Yang, Xiaotong Yao, Ke Yuan, Hongtu Zhu, Wenyi Wang, Quaid Morris, Paul T. Spellman, David C. Wedge, Peter Van Loo, Lauri A. Aaltonen, Federico Abascal, Adam Abeshouse, Hiroyuki Aburatani, Nishant Agrawal, Keun Soo Ahn, Sung‐Min Ahn, Hiroshi Aikata, Rehan Akbani, Kadir C. Akdemir, Hikmat Al‐Ahmadie, Sultan T. Al‐Sedairy, Fátima Al‐Shahrour, Malik Alawi, Monique Albert, Kenneth Aldape, Ludmil B. Alexandrov, Adrian Ally, Kathryn Alsop, Eva G. Álvarez, Fernanda AmaryNature Communications2020 · DOI
The type and genomic context of cancer mutations depend on their causes. These causes have been characterized using signatures that represent mutation types that co-occur in the same tumours. However, it remains unclear how mutation processes change during cancer evolution due to the lack of reliable methods to reconstruct evolutionary trajectories of mutational signature activity. Here, as part of the ICGC/TCGA Pan-Cancer Analysis of Whole Genomes (PCAWG) Consortium, which aggregated whole-genome sequencing data from 2658 cancers across 38 tumour types, we present TrackSig, a new method that reconstructs these trajectories using optimal, joint segmentation and deconvolution of mutation type and allele frequencies from a single tumour sample. In simulations, we find TrackSig has a 3-5% activity reconstruction error, and 12% false detection rate. It outperforms an aggressive baseline in situations with branching evolution, CNA gain, and neutral mutations. Applied to data from 2658 tumours and 38 cancer types, TrackSig permits pan-cancer insight into evolutionary changes in mutational processes.
Introduction
Cancer is an inherently evolutionary process. As a tumor progresses from a single transformed cell to a complex mass, the biological forces driving its mutations—such as environmental carcinogens, aging, or defective DNA repair—often shift. These patterns of mutation are captured as "mutational signatures." While identifying these signatures in a bulk sample is well-established, determining when each process was active remains a significant challenge. Traditional methods often require complex subclonal clustering which can be unreliable with single-biopsy data. TrackSig addresses this by reconstructing continuous trajectories of signature activity without needing to solve the full subclonal architecture first.
Methods: Reconstructing the Timeline
TrackSig introduces a two-stage approach to map the "life history" of a tumor's mutational processes:
Pseudo-time Ordering: The algorithm uses the Cancer Cell Fraction (CCF) of each mutation as a proxy for time. Mutations present in every cancer cell (high CCF) are considered older (clonal) than those present in only a fraction of cells (low CCF). By sorting mutations by CCF, a "pseudo-timeline" is established.
Segmentation and Deconvolution: The timeline is divided into bins. TrackSig applies an optimal joint segmentation algorithm (PELT) to identify "changepoints"—specific moments in the tumor's history where the mixture of mutational signatures shifts significantly. Within each segment, an Expectation-Maximization (EM) algorithm calculates the "activity" or exposure of each signature.
Reconstructed evolutionary trajectories of mutational signature activities across cancer cell fractions (CCF) for two example tumors. Red vertical lines indicate detected changepoints where signature activities shift.
The application of TrackSig to the PCAWG dataset of 2,658 cancers revealed that mutational processes are rarely static. Over 76% of the samples showed a significant change in at least one mutational signature over time.
High Accuracy and Stability: In simulations, TrackSig achieved a 3–5% activity reconstruction error. Bootstrap analysis confirmed that these trajectories are highly stable across different samplings of the same tumor data.
Signatures as Lineage Indicators: Changes in mutational signatures often proved to be more sensitive indicators of new subclonal branches than traditional variant frequency clustering.
Biological Discovery: In chronic lymphocytic leukaemia, TrackSig successfully mapped the dramatic decrease of Signature 9 (somatic hypermutation) during subclonal expansion, a shift previously difficult to time precisely.
Cumulative ratio of tumors showing the stability of signature activity estimates. The low median activity difference across bootstrap samples indicates high consistency in the reconstructed trajectories.
While TrackSig provides a powerful new lens for cancer genomics, it has limitations. Its accuracy in ordering mutations depends heavily on high-quality Copy Number Aberration (CNA) data. Furthermore, a single sample cannot always distinguish between distinct evolutionary branches occurring at the same CCF level. Future work aims to calibrate these "pseudo-time" trajectories to actual chronological time using "clock" signatures (Signatures 1 and 5) and to incorporate other mutation types like indels and structural variants.
Summary updated
Abstract
The type and genomic context of cancer mutations depend on their causes. These causes have been characterized using signatures that represent mutation types that co-occur in the same tumours. However, it remains unclear how mutation processes change during cancer evolution due to the lack of reliable methods to reconstruct evolutionary trajectories of mutational signature activity. Here, as part of the ICGC/TCGA Pan-Cancer Analysis of Whole Genomes (PCAWG) Consortium, which aggregated whole-genome sequencing data from 2658 cancers across 38 tumour types, we present TrackSig, a new method that reconstructs these trajectories using optimal, joint segmentation and deconvolution of mutation type and allele frequencies from a single tumour sample. In simulations, we find TrackSig has a 3-5% activity reconstruction error, and 12% false detection rate. It outperforms an aggressive baseline in situations with branching evolution, CNA gain, and neutral mutations. Applied to data from 2658 tumours and 38 cancer types, TrackSig permits pan-cancer insight into evolutionary changes in mutational processes.
BibTeX
@inproceedings{rubanova2020reconstructing,
author = {Yulia Rubanova and Ruian Shi and Caitlin F. Harrigan and Roujia Li and Jeff Wintersinger and Nil Sahin and Amit G. Deshwar and Stefan C. Dentro and Ignaty Leshchiner and Moritz Gerstung and Clemency Jolly and Kerstin Haase and Maxime Tarabichi and Jeff Wintersinger and Amit G. Deshwar and Kaixian Yu and Santiago Gonzalez and Yulia Rubanova and Geoff Macintyre and David J. Adams and Pavana Anur and Rameen Beroukhim and Paul C. Boutros and David D.L. Bowtell and Peter J. Campbell and Shaolong Cao and Elizabeth L. Christie and Marek Cmero and Yupeng Cun and Kevin J. Dawson and Jonas Demeulemeester and Nilgun Donmez and Ruben M. Drews and Roland Eils and Yu Fan and Matthew W. Fittall and Dale W. Garsed and Gad Getz and Gavin Ha and Marcin Imieliński and Lara Jerman and Yuan Ji and Kortine Kleinheinz and Ju‐Hee Lee and Henry Lee-Six and Dimitri Livitz and Salem Malikić and Florian Markowetz and Iñigo Martincorena and Thomas J. Mitchell and Ville Mustonen and Layla Oesper and Martin Peifer and Myron Peto and Benjamin J. Raphael and Daniel Rosebrock and S. Cenk Sahinalp and Adriana Salcedo and Matthias Schlesner and Steven E. Schumacher and Subhajit Sengupta and Ruian Shi and Seung Jun Shin and Oliver Spiro and Lincoln D. Stein and Ignacio Vázquez-Garćıa and Shankar Vembu and David A. Wheeler and Tsun-Po Yang and Xiaotong Yao and Ke Yuan and Hongtu Zhu and Wenyi Wang and Quaid Morris and Paul T. Spellman and David C. Wedge and Peter Van Loo and Quaid Morris and Lauri A. Aaltonen and Federico Abascal and Adam Abeshouse and Hiroyuki Aburatani and David J. Adams and Nishant Agrawal and Keun Soo Ahn and Sung‐Min Ahn and Hiroshi Aikata and Rehan Akbani and Kadir C. Akdemir and Hikmat Al‐Ahmadie and Sultan T. Al‐Sedairy and Fátima Al‐Shahrour and Malik Alawi and Monique Albert and Kenneth Aldape and Ludmil B. Alexandrov and Adrian Ally and Kathryn Alsop and Eva G. Álvarez and Fernanda Amary},
title = {{Reconstructing evolutionary trajectories of mutation signature activities in cancer using TrackSig}},
booktitle = {Nature Communications},
year = {2020},
doi = {10.1038/s41467-020-14352-7},
}