18O - Representative sequencing: Profiling extreme tumor diversity

Abstract

Background: While next-generation sequencing (NGS) has been applied to thousands of solid tumors to date, there exists a fundamental undersampling bias inherent in current methodologies. This is caused by a biopsy input sample of fixed dimensions, which becomes grossly under-powered as tumor volume scales. Indeed, analysis of pan-cancer data reveals that current protocols sample on average only 1.5% of cancer cells, decreasing to 0.3% for stage IV tumors. Failure to address this bias risks undermining the clinical utility of genomic medicine in cancer. Methods: Here we demonstrate Representative Sequencing (Rep-Seq), as a novel method to achieve unbiased sampling of solid tumor tissue. The Rep-Seq protocol comprises homogenization of all residual tumor material not taken for pathology into a well-mixed solution, coupled with NGS. Rep-Seq was implemented on a proof of concept basis in > 10 tumors, and benchmarked against single and multi-region sequencing approaches. Results: Rep-Seq achieved a linear rate of novel variant discovery in whole-exome sequencing across 0 to 5,000x coverage, detecting four-fold more mutations as compared to multi-region sequencing at equivalent total read depth. All variants were validated using custom panel and Ion Torrent platforms. Targeted panel Rep-Seq at 50,000x showed sensitivity to detect extreme parallel evolution, with 16 independent mutations in the gene SETD2 observed in a single tumor. Clonal clustering analysis revealed rapid convergence of cancer cell fraction estimates in Rep-Seq towards true values, as validated in > 70 biopsies taken from a single tumor. As a consequence, 97% of variants were correctly classified as clonal by Rep-Seq, compared to > 85% in single biopsy sequencing. Finally, in a rapid autopsy setting Rep-Seq was able to accurately reconstruct the clonal phylogency of advanced stage disease, recovering a high proportion of all primary and metastatic variants, from deep sequencing of primary tissue alone. Conclusions: Rep-Seq effectively implements unbiased tumor sampling, drawing DNA molecules from a well-mixed solution of the entire tumor mass, hence removing spatial bias inherent in current approaches. As a result, Rep-Seq detects more mutations, and achieves greater accuracy in determining clonal from subclonal variants. Legal entity responsible for the study: The Francis Crick Institute. Funding: Roche. Disclosure: C. Swanton: Consulting and speaker fees from Boehringer Ingelheim, Eli Lilly, Novartis, and Roche; Research grants from Roche. S. Turajlic: Research grants from Roche. All other authors have declared no conflicts of interest.