The evolution of metastases is responsible for 90% of cancer-related deaths. Genome wide sequencing and phylogenomic methods enable the reconstruction of the evolutionary history of a patient's cancer at unprecedented depth. However, due to a lack of samples from multiple spatially-distinct metastases from untreated patients and a lack of phylogenomic tools applicable to noisy and impure sequencing samples, the evolutionary rules governing metastatic spread have remained poorly understood.

We performed whole-genome sequencing (coverage: median 51x) as well as deep targeted sequencing (coverage: median 347x) on 21 samples from multiple regions of the primary tumor and many distinct liver and lung metastases of two treatment-naïve pancreatic ductal adenocarcinoma patients. We developed a tool, called Treeomics, that leverages computational and statistical advances to reconstruct the phylogeny of a cancer with commonly available sequencing technologies. Treeomics employs a uniquely-designed Bayesian inference model to account for error-prone sequencing and varying low neoplastic cell content (estimated purities 16-44%) to calculate the probability that a specific variant is present or absent in each sequenced lesion. Based on Mixed Integer Linear Programming, a mathematically guaranteed optimal evolutionary tree is produced.

We obtained robust phylogenies consistent with the biological processes underlying cancer evolution. The reconstructed phylogenies show that advanced cancer cells of related subclones were equally capable of seeding lung and liver metastases. Treeomics identified sequencing and biological artifacts such as those resulting from insufficient coverage or loss of heterozygosity; almost 7% of the variants were misclassified by conventional methods. Among the identified false-negatives was the common clonal driver mutation in KRAS within a region that has low sequencing read alignability and a significantly reduced coverage. Such artifacts can skew phylogenies by creating illusory tumor heterogeneity among distinct samples. Additionally, we reanalyzed publicly available data from ovarian, prostate and skin cancers. We further illuminated evolutionary relationships among some samples in a conclusive fashion and show that classical distance-based phylogenetic methods can produce evolutionarily implausible results. Treeomics avoids these common pitfalls and infers robust phylogenies confirmed by high bootstrapping values.

The new approach described here efficiently reconstructs the evolutionary history of metastases, detects potential artifacts in noisy high-throughput sequencing data, and finds subclones of distinct origin. These phylogenies shed new light on seeding patterns and metastatic progression, which has significant implications for clinical decision making and may provide predictive value for a patient's prognosis.

Citation Format: Johannes G. Reiter, Alvin P. Makohon-Moore, Jeffrey M. Gerold, Ivana Bozic, Krishnendu Chatterjee, Christine A. Iacobuzio-Donahue, Bert Vogelstein, Martin A. Nowak. Reconstructing the evolutionary history of metastatic cancers. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 2374.