Reconstructing metastatic seeding patterns of human cancers

Johannes G. Reiter, Alvin P. Makohon-Moore, Jeffrey M. Gerold, Ivana Božić, Krishnendu Chatterjee, Christine A. Iacobuzio–Donahue, Bert Vogelstein, Martin A. NowakNature Communications 2017 · DOI

Abstract

Reconstructing the evolutionary history of metastases is critical for understanding their basic biological principles and has profound clinical implications. Genome-wide sequencing data has enabled modern phylogenomic methods to accurately dissect subclones and their phylogenies from noisy and impure bulk tumour samples at unprecedented depth. However, existing methods are not designed to infer metastatic seeding patterns. Here we develop a tool, called Treeomics, to reconstruct the phylogeny of metastases and map subclones to their anatomic locations. Treeomics infers comprehensive seeding patterns for pancreatic, ovarian, and prostate cancers. Moreover, Treeomics correctly disambiguates true seeding patterns from sequencing artifacts; 7% of variants were misclassified by conventional statistical methods. These artifacts can skew phylogenies by creating illusory tumour heterogeneity among distinct samples. In silico benchmarking on simulated tumour phylogenies across a wide range of sample purities (15-95%) and sequencing depths (25-800 × ) demonstrates the accuracy of Treeomics compared with existing methods.

Introduction

Metastasis—the spread of cancer from a primary tumor to distant organs—is the primary cause of cancer-related mortality. Understanding the "travel logs" of these cells is essential for developing effective treatments. Does a single subclone seed all metastases? Do metastases seed other metastases? Answering these questions requires reconstructing the evolutionary history of the tumor. However, analyzing "bulk" tumor samples is notoriously difficult due to "noise" in DNA sequencing and the presence of healthy cells within the sample.

The Challenge of Sample Impurity

A major hurdle in metastatic research is the varying purity of clinical samples. Tumor samples are often "impure," containing a mixture of cancer cells and healthy tissue. Standard statistical methods often fail in these scenarios: they might miss a mutation if tumor content is low (false negative) or misidentify a sequencing error as a real mutation (false positive).

Figure 2: Histological comparison of liver and lung metastases showing significant differences in neoplastic cellularity.

As shown in Figure 2, a liver metastasis might have high neoplastic cellularity (>80%), while a lung metastasis from the same patient might have very low cellularity (<20%). These discrepancies create "illusory heterogeneity," making different tumor sites appear more evolutionarily distant than they truly are.

The Treeomics Framework

To address these challenges, researchers developed Treeomics, a computational tool that moves beyond simple binary "present/absent" mutation calls. Treeomics employs a two-step approach:

  1. Bayesian Inference: Instead of a hard yes/no, Treeomics calculates the probability that a mutation exists in a sample, accounting for sequencing depth and estimated tumor purity.
  2. Integer Linear Programming (ILP): Treeomics treats phylogeny reconstruction as a "binary maximum compatibility problem." By assuming a "perfect phylogeny" (where mutations are gained once and never lost), it uses ILP to find the most likely tree that minimizes incompatible mutations, effectively filtering out sequencing artifacts.

Key Insights and Results

Treeomics was validated using both clinical data (pancreatic, ovarian, and prostate cancers) and a massive benchmark of 90,000 simulated tumor phylogenies.

  • Artifact Detection: The study revealed that conventional methods misclassified approximately 7% of genetic variants. Treeomics identifies these "tree-breakers," showing that many metastases are more genetically similar than previously thought.
  • Superior Accuracy: In simulations, Treeomics consistently outperformed existing tools like LICHeE and PhyloWGS, particularly in low-purity samples (under 35% tumor content).
  • Complex Seeding Patterns: In pancreatic cancer, Treeomics revealed that different metastases in the same organ can originate from distinct subclones in the primary tumor.
  • Efficiency: By utilizing ILP instead of complex Markov Chain Monte Carlo (MCMC) simulations, Treeomics processes cases in seconds or minutes.

Discussion and Future Directions

While Treeomics represents a significant leap forward, it currently focuses on single nucleotide variants (SNVs) and small indels. Future iterations aim to integrate large-scale changes like copy number alterations (CNAs). Additionally, while the "perfect phylogeny" assumption holds for most point mutations, large chromosomal deletions can violate it, presenting an area for further refinement. By providing a clearer picture of how cancer spreads, Treeomics offers a powerful new lens for understanding and eventually thwarting metastatic disease.

Summary updated

Abstract

Reconstructing the evolutionary history of metastases is critical for understanding their basic biological principles and has profound clinical implications. Genome-wide sequencing data has enabled modern phylogenomic methods to accurately dissect subclones and their phylogenies from noisy and impure bulk tumour samples at unprecedented depth. However, existing methods are not designed to infer metastatic seeding patterns. Here we develop a tool, called Treeomics, to reconstruct the phylogeny of metastases and map subclones to their anatomic locations. Treeomics infers comprehensive seeding patterns for pancreatic, ovarian, and prostate cancers. Moreover, Treeomics correctly disambiguates true seeding patterns from sequencing artifacts; 7% of variants were misclassified by conventional statistical methods. These artifacts can skew phylogenies by creating illusory tumour heterogeneity among distinct samples. In silico benchmarking on simulated tumour phylogenies across a wide range of sample purities (15-95%) and sequencing depths (25-800 × ) demonstrates the accuracy of Treeomics compared with existing methods.
BibTeX
@inproceedings{reiter2017reconstructing,
  author = {Johannes G. Reiter and Alvin P. Makohon-Moore and Jeffrey M. Gerold and Ivana Božić and Krishnendu Chatterjee and Christine A. Iacobuzio–Donahue and Bert Vogelstein and Martin A. Nowak},
  title = {{Reconstructing metastatic seeding patterns of human cancers}},
  booktitle = {Nature Communications},
  year = {2017},
  doi = {10.1038/ncomms14114},
}
Figure