Retrospective evaluation of whole exome and genome mutation calls in 746 cancer samples

Matthew H. Bailey, William U. Meyerson, Lewis Jonathan Dursi, Liang-Bo Wang, Guanlan Dong, Wen-Wei Liang, Amila Weerasinghe, Shantao Li, Yize Li, Sean Kelso, Rehan Akbani, Pavana Anur, Alex Buchanan, Kami Chiotti, Kyle R. Covington, Allison Creason, Li Ding, Kyle Ellrott, Yu Fan, Steven M. Foltz, Gad Getz, Walker Hale, David Haussler, Julian M. Hess, Carolyn M. Hutter, Cyriac Kandoth, L. Sylvia, Melpomeni Kasapi, Dave Larson, Ignaty Leshchiner, John Letaw, Singer Ma, Michael D. McLellan, Yifei Men, Gordon B. Mills, Beifang Niu, Myron Peto, Amie Radenbaugh, Sheila M. Reynolds, Gordon Saksena, Heidi J. Sofia, Chip Stewart, Adam J. Struck, Joshua M. Stuart, Wenyi Wang, John N. Weinstein, David A. Wheeler, Christopher K. Wong, Xi Liu, Kai Ye, Matthias Bieg, Paul C. Boutros, Ivo Buchhalter, Adam P. Butler, Ken Chen, Zechen Chong, Oliver Drechsel, Roland Eils, Shadrielle M. G. Espiritu, Robert S. Fulton, Shengjie Gao, Josep L. L. Gelpi, Mark Gerstein, Santiago González, Marta Gut, Faraz Hach, Michael C. Heinold, Jonathan Hinton, Taobo Hu, Vincent Huang, Yi Huang, Barbara Hutter, David Jones, Jongsun Jung, Hyung‐Lae Kim, Kortine Kleinheinz, Sushant Kumar, Yogesh Kumar, Christopher M. Lalansingh, Ivica Letunić, Dimitri Livitz, Z. Eric, Yosef E. Maruvka, R. Jay Mashl, Andrew Menzies, Ana Milovanovic, Morten Muhlig Nielsen, Stephan Ossowski, Nagarajan ParamasivamNature Communications 2020 · DOI

Abstract

The Cancer Genome Atlas (TCGA) and International Cancer Genome Consortium (ICGC) curated consensus somatic mutation calls using whole exome sequencing (WES) and whole genome sequencing (WGS), respectively. Here, as part of the ICGC/TCGA Pan-Cancer Analysis of Whole Genomes (PCAWG) Consortium, which aggregated whole genome sequencing data from 2,658 cancers across 38 tumour types, we compare WES and WGS side-by-side from 746 TCGA samples, finding that ~80% of mutations overlap in covered exonic regions. We estimate that low variant allele fraction (VAF < 15%) and clonal heterogeneity contribute up to 68% of private WGS mutations and 71% of private WES mutations. We observe that ~30% of private WGS mutations trace to mutations identified by a single variant caller in WES consensus efforts. WGS captures both ~50% more variation in exonic regions and un-observed mutations in loci with variable GC-content. Together, our analysis highlights technological divergences between two reproducible somatic variant detection efforts.

Introduction

In cancer genomics, researchers typically choose between two paths: Whole Exome Sequencing (WES), which targets the ~2% of the genome that codes for proteins, and Whole Genome Sequencing (WGS), which sequences the entire genome. WES is cost-effective and has been the workhorse of large projects like The Cancer Genome Atlas (TCGA), while WGS is more comprehensive but significantly more expensive. The central question is: How much biological information is lost by choosing one over the other? Previous studies suggested discrepancies, but they were often confounded by different bioinformatics pipelines or small sample sizes. This paper leverages a unique dataset where the same 746 tumor samples were processed by two massive consortia—MC3 (WES-focused) and PCAWG (WGS-focused)—to pinpoint exactly where and why these two "gold standard" approaches diverge.

Methods and Datasets

The researchers performed a retrospective evaluation of mutation calls from 746 TCGA samples across 22 different cancer types. To ensure a fair "apples-to-apples" comparison, the team restricted the search space by limiting WGS data to the specific exonic regions targeted by WES "baits" and employed a reciprocal coverage strategy, only analyzing mutations in regions with sufficient sequencing depth in both technologies.

Overview of the data integration workflow and cohort distribution across TCGA cancer types.

The study utilized variant calls from two major repositories:

  • MC3 (Multi-Center Mutation Calling in Multiple Cancers): The consensus WES variant calls for TCGA.
  • PCAWG (Pan-Cancer Analysis of Whole Genomes): The consensus WGS variant calls from the ICGC/TCGA consortium.

To facilitate comparison, the researchers developed MAFit, a web-based graphical interface that allows users to interactively compare and filter these merged mutation sets. They also used random forest regression and binomial models to determine if "private" mutations (those found by only one method) were due to sequencing noise, low Variant Allele Fraction (VAF), or biological factors like subclonal heterogeneity.

Key Insights: The Drivers of Discrepancy

Overall, the two platforms showed 76.7% concordance, with over 70% of samples showing an overlap greater than 80%. However, several key factors drive the remaining discrepancies:

The VAF Sensitivity Gap

The single biggest driver of discrepancy is the Variant Allele Fraction (VAF). Mutations with a VAF < 15% accounted for roughly 70% of the "private" mutations. At these low frequencies, the stochastic nature of sequencing means one platform may simply miss a mutation that the other catches.

GC-Content and Capture Bias

WES relies on hybrid capture (using "baits" to grab DNA), a process notoriously biased by GC-content. The study found that WGS provides much more uniform coverage, catching critical driver mutations (such as VHL in kidney cancer) in GC-rich regions that WES consistently overlooked.

The "Dark Matter" of Exons

Even within "exonic" regions, WGS captured ~50% more variation. This is because WGS naturally sequences the flanking regions—such as 5’ and 3’ Untranslated Regions (UTRs)—which are often poorly covered or entirely ignored by standard WES exome kits. The study highlighted recurrent mutations in the 3’UTRs of genes like PGR and ERBB4 that were only visible via WGS and correlated with changes in gene expression.

Volcano plot illustrating the differential representation of mutations between MC3 (WES) and PCAWG (WGS) across various cancer subtypes.

Conclusion

Interestingly, the specific bioinformatics "callers" used were less of a factor than the underlying sequencing technology. When both platforms had high coverage and the mutation frequency was high, they almost always agreed. WGS identified 424 mutations in known cancer genes that were missed by WES, primarily due to low VAF or poor capture bait efficiency.

While the authors acknowledge that spatial heterogeneity (different mutations in different parts of the same tumor) remains a challenge, they conclude that the discrepancies are primarily technological. As sequencing costs fall, a transition toward WGS is expected; in the interim, tools like MAFit are essential for reconciling these two massive datasets and understanding the technological divergences that define modern cancer genomics.

Summary updated

Abstract

The Cancer Genome Atlas (TCGA) and International Cancer Genome Consortium (ICGC) curated consensus somatic mutation calls using whole exome sequencing (WES) and whole genome sequencing (WGS), respectively. Here, as part of the ICGC/TCGA Pan-Cancer Analysis of Whole Genomes (PCAWG) Consortium, which aggregated whole genome sequencing data from 2,658 cancers across 38 tumour types, we compare WES and WGS side-by-side from 746 TCGA samples, finding that ~80% of mutations overlap in covered exonic regions. We estimate that low variant allele fraction (VAF < 15%) and clonal heterogeneity contribute up to 68% of private WGS mutations and 71% of private WES mutations. We observe that ~30% of private WGS mutations trace to mutations identified by a single variant caller in WES consensus efforts. WGS captures both ~50% more variation in exonic regions and un-observed mutations in loci with variable GC-content. Together, our analysis highlights technological divergences between two reproducible somatic variant detection efforts.
BibTeX
@inproceedings{bailey2020retrospective,
  author = {Matthew H. Bailey and William U. Meyerson and Lewis Jonathan Dursi and Liang-Bo Wang and Guanlan Dong and Wen-Wei Liang and Amila Weerasinghe and Shantao Li and Yize Li and Sean Kelso and Rehan Akbani and Pavana Anur and Matthew H. Bailey and Alex Buchanan and Kami Chiotti and Kyle R. Covington and Allison Creason and Li Ding and Kyle Ellrott and Yu Fan and Steven M. Foltz and Gad Getz and Walker Hale and David Haussler and Julian M. Hess and Carolyn M. Hutter and Cyriac Kandoth and L. Sylvia and Melpomeni Kasapi and Dave Larson and Ignaty Leshchiner and John Letaw and Singer Ma and Michael D. McLellan and Yifei Men and Gordon B. Mills and Beifang Niu and Myron Peto and Amie Radenbaugh and Sheila M. Reynolds and Gordon Saksena and Heidi J. Sofia and Chip Stewart and Adam J. Struck and Joshua M. Stuart and Wenyi Wang and John N. Weinstein and David A. Wheeler and Christopher K. Wong and Xi Liu and Kai Ye and Matthew H. Bailey and Beifang Niu and Matthias Bieg and Paul C. Boutros and Ivo Buchhalter and Adam P. Butler and Ken Chen and Zechen Chong and Li Ding and Oliver Drechsel and Lewis Jonathan Dursi and Roland Eils and Kyle Ellrott and Shadrielle M. G. Espiritu and Yu Fan and Robert S. Fulton and Shengjie Gao and Josep L. L. Gelpi and Mark Gerstein and Gad Getz and Santiago González and Marta Gut and Faraz Hach and Michael C. Heinold and Julian M. Hess and Jonathan Hinton and Taobo Hu and Vincent Huang and Yi Huang and Barbara Hutter and David Jones and Jongsun Jung and Hyung‐Lae Kim and Kortine Kleinheinz and Sushant Kumar and Yogesh Kumar and Christopher M. Lalansingh and Ignaty Leshchiner and Ivica Letunić and Dimitri Livitz and Z. Eric and Yosef E. Maruvka and R. Jay Mashl and Michael D. McLellan and Andrew Menzies and Ana Milovanovic and Morten Muhlig Nielsen and Stephan Ossowski and Nagarajan Paramasivam},
  title = {{Retrospective evaluation of whole exome and genome mutation calls in 746 cancer samples}},
  booktitle = {Nature Communications},
  year = {2020},
  doi = {10.1038/s41467-020-18151-y},
}
Figure