Introduction: Precision medicine approaches to guide therapy selection require routine sampling of tumors. However, tumor biopsies are not always accessible and may be confounded by spatial heterogeneity. Liquid biopsies, including analysis of cell-free DNA (cfDNA), present a non-invasive alternative which may reflect multiple tumors in the body. Previous studies have demonstrated exome-wide concordance between single-site tumor biopsies and cfDNA, but little is known about how cfDNA reflects multiple lesions within a patient. Here we sought to determine how cfDNA reflects the body-wide tumor phylogeny, which will inform the use of cfDNA for cancer precision medicine.

Methods: We identified 20 patients with pancreatic cancer who had undergone rapid autopsy. We then screened cfDNA tumor fraction and performed whole-exome sequencing of cfDNA and multiple tumor biopsies for 3 patients with cfDNA tumor fraction >10%. We inferred the tumor phylogeny and then developed a statistical approach to deconvolute the contributions to cfDNA from tumor phylogenetic nodes. Finally, we determined whether shared trunk mutations could be detected in cfDNA and tumor biopsies.

Results: For each patient, we found mutations shared between all sites and cfDNA, including putative driver mutations. We found mutations which were clonal in multiple regions were detectable in cfDNA, whereas mutations private to individual sites were never clonal in cfDNA. Through our deconvolution analysis, we found that cfDNA could not be modeled as a simple linear combination of individual sites, but rather that cfDNA represented multiple nodes in the inferred phylogeny. For two pancreatic adenocarcinoma patients, the inferred ancestor of the metastases had high estimated contribution (>70%) to cfDNA, while the ancestors of the primaries had lower contributions (<10%). Next, we considered trunk mutations, which originate earliest in the tumor phylogenetic tree. When we analyzed precision for detection of trunk mutations, we found on average, 71% of clonal mutations in metastases were truncal, while only 55% of clonal mutations in primary tumors were truncal. Due to copy number deletions, not all trunk mutations were detected in metastases. Finally, on average, cfDNA had equal or better precision than 83% of primaries and 88% of metastases, suggesting cfDNA may provide more accurate trunk SSNV calls than tumor biopsies.

Conclusions: Through analyzing cfDNA and synchronous tumor biopsies from the same patient, we find trunk mutations are enriched in cfDNA as compared to the average single-site biopsy. We also predict that cfDNA represents multiple nodes in the inferred phylogeny. In cases where tumor biopsies are inaccessible, we demonstrate that cfDNA might be a promising alternative to detect trunk SSNVs. These results suggest that cfDNA may be complementary to tumor biopsies for disease monitoring and treatment selection in personalized medicine.

Citation Format: Samuel S. Freeman, Ziao Lin, Gavin Ha, Ignaty Leshchiner, Justin Rhoades, Dimitri Livitz, Daniel Rosebrock, Sarah C. Reed, Gregory Gydush, Christopher Lo, Denisse Rotem, Atish D. Choudhury, Daniel G. Stover, Heather A. Parsons, Jesse S. Boehm, J Christopher Love, Matthew Meyerson, Paul Grandgenett, Michael A. Hollingsworth, Viktor A. Adalsteinsson, Gad Getz. Liquid biopsies identify trunk mutations and reflect multiple tumors in a patient [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr LB-225.