Methods for detecting co-mutated pathways in cancer samples to inform treatment selection

Tingting Jiang, Uri Shaham, Fabio Parisi, Ruth Halaban, Anton Safonov, Harriet M. Kluger, Sherman M. Weissman, Joseph T. Chang, Yuval KlugerbioRxiv 2016 · DOI

TL;DR

This paper introduces a new statistical method for identifying co-mutated pathways in cancer, addressing limitations of existing approaches like Fisher's exact test that struggle with the high variability in cancer genomics data. Their method, applied to TCGA data, reveals key co-mutated pathways like PI3K and IP3 metabolism across cancers and highlights the proteasome and apoptosis pathway co-mutation in colorectal cancer as a potential mechanism for immune evasion.

What Problem Does This Paper Address?

Cancer is driven by accumulated mutations, and these mutations often don't act in isolation. Instead, they can work together (co-mutate) within biological pathways to promote tumor growth and survival. Understanding these co-mutational relationships is crucial for deciphering cancer's complexity and, ultimately, for developing more effective treatments. While Fisher's exact test (FET) is a common tool for identifying co-occurring events, it falters when dealing with the inherent variability in cancer genomics data, such as differences in overall mutation load and sequencing depth across samples. This variability can lead to inflated significance (false positives) or missed true associations. The authors aim to develop a more robust method that can accurately detect co-mutated pathways despite these challenges.

Methods

The core contribution is a novel "patient-specific" statistical method for detecting co-mutated pathways. This method addresses the limitations of FET by:

  • Pathway-level aggregation: Instead of looking at individual genes, mutations are aggregated at the pathway level. A pathway is considered "dysregulated" if at least one of its genes is mutated.
  • Patient-specific scoring: The method quantifies the importance of a mutated pathway in a given tumor by considering the overall mutation load of that sample.
  • Poisson Binomial Distribution: A Poisson binomial distribution is used as a null model to assess the significance of co-mutation between two pathways, accounting for the patient-specific scores.
  • Handling Overlapping Genes: For pathways that share genes, an additional criterion is implemented to exclude overlapping genes and samples mutated in those genes to avoid confounding.

The authors also employed a permutation test as a benchmark to generate empirical p-values and compare their method's performance against FET.

Datasets

The study utilized two primary types of datasets:

  • Simulated Datasets: 100 simulated datasets were generated with varying mutation loads (small dispersion and large dispersion scenarios) and 180 pathways to test the performance of their method and FET under controlled conditions.
  • TCGA (The Cancer Genome Atlas) Datasets: Somatic mutation calls from 1284 tumor samples across six cancer types were used:
  • Uterine Corpus Endometrial Carcinoma (UCEC)
    • Colon/Rectum Adenocarcinoma (COAD/READ)
    • Stomach Adenocarcinoma (STAD)
    • Skin Cutaneous Melanoma (SKCM)
    • Lung Adenocarcinoma (LUAD)
    • Lung Squamous Cell Carcinoma (LUSC)

Pathway information was sourced from the EnrichmentMap pathway repository (2921 pathways) and a subset from the BIOCARTA database (217 pathways).

Key Insights and Ideas

  • Cancer is a network disease: Mutations rarely act in isolation; they often cooperate within pathways to drive cancer progression. Understanding these pathway interactions is key to unlocking cancer's complexity.
  • Variability is a major hurdle: The high variability in mutation load and sequencing depth in cancer data makes standard statistical tests like FET unreliable, often leading to inflated false positives.
  • A new lens for co-mutation: The patient-specific approach, by considering individual tumor characteristics, offers a more robust way to identify true co-mutated pathways.
  • Proteasome and apoptosis: a cancer immune evasion duo? The co-mutation of these pathways in colorectal cancer is a particularly intriguing finding, suggesting a coordinated strategy for tumors to hide from the immune system.

Key Results

  • Improved performance over FET: The proposed patient-specific method demonstrated higher precision and sensitivity in simulated datasets, especially those with high variability in mutation load, compared to Fisher's exact test. The p-values from the patient-specific method showed a strong correlation with empirical p-values from permutation tests (R=0.88), while FET's correlation was much lower (R=0.15).
  • Pan-cancer co-mutations: Across multiple cancer types, the PI3K and IP3 metabolism pathways were identified as the most significantly co-mutated. This aligns with known biological interactions and suggests a fundamental role in cancer progression.
  • Colorectal cancer specific findings: In colorectal cancer, a significant co-mutation was observed between the proteasome and extrinsic apoptosis pathways. This co-mutation was associated with:
  • Higher predicted neoantigen presentation.
  • Higher cytolytic activity (CYT).
  • Potentially, a mechanism for immune evasion by reducing neoantigen presentation or evading apoptosis.
  • Clonal co-occurrence: Analysis of mutations in the proteasome and apoptosis pathways in colorectal cancer suggested that these co-mutations often occur within the same tumor clone, indicating they are acquired during tumor progression.

Limitations and Future Work

  • Limitations of FET: The paper extensively discusses the limitations of Fisher's exact test, particularly its susceptibility to bias from hyper-mutated phenotypes, uneven sequencing, and variable tumor-stromal cellularity.
  • Future Directions:
    • Single-cell RNA sequencing: The authors suggest validating findings using single-cell RNA sequencing to directly assess co-mutated transcript expression within individual cells, which could overcome limitations of deconvolution methods.
  • Other genomic abnormalities: The developed method can be extended to integrate other types of genomic alterations like CNV and DNA methylation to capture more low-frequency co-mutated pathway pairs.
  • Order of mutations: Inferring the order in which mutations occur during cancer evolution is crucial and could inform treatment decisions. This is a promising avenue for future research.
  • Treatment implications: The ultimate goal is to use these insights to inform treatment decisions in individual patients and advance personalized medicine.

Summary updated

Abstract

Abstract Tumor genomes evolve through a selection of mutations. These mutations may complement each other to promote tumorigenesis. To better understand the functional interactions of different processes in cancer, we studied mutation data of a set of tumors and identified significantly co-mutated pathways. Fisher’s exact test is a standard approach that can be used to assess the significance of the joint dysregulation of pathways pairs across a patient population. We developed a robust test to identify co-occurrence using DNA mutations, which overcomes deficiencies of the Fisher’s exact test by taking into account the large variability in overall mutation load and sequencing depth. Applying our method to a study of six common cancer types, we identify enrichment of co-mutated signal transduction pathways such as IP3 synthesis and PI3K and pairs of co-mutated pathways involving other processes such as immunity and development. We observed enrichment of clonal co-mutation of the proteasome and apoptosis pathways in colorectal cancer, which suggests potential mechanisms for immune evasion.
BibTeX
@misc{jiang2016methods,
  author = {Tingting Jiang and Uri Shaham and Fabio Parisi and Ruth Halaban and Anton Safonov and Harriet M. Kluger and Sherman M. Weissman and Joseph T. Chang and Yuval Kluger},
  title = {{Methods for detecting co-mutated pathways in cancer samples to inform treatment selection}},
  year = {2016},
  doi = {10.1101/082552},
  howpublished = {bioRxiv},
}

Loading related content...

Figure