Molecular and cellular heterogeneity of gastric cancer explained by methylation-driven key regulators

Seungyeul Yoo, Quan Chen, Li Wang, Wenhui Wang, Ankur Chakravarthy, Rita A. Busuttil, Alex Boussioutas, Dan Liu, Junjun She, Tim R. Fenton, Jiangwen Zhang, Xiaodan Fan, Suet Yi Leung, Jun ZhubioRxiv 2020 · DOI

TL;DR

Gastric cancer (GC) is a complex disease with significant heterogeneity driven by genetic and epigenetic factors, as well as the tumor microenvironment (TME). This study introduces an "Integrative Sequential Causality Test" (ISCT) to identify key regulatory genes influenced by DNA methylation and copy number variations (CNVs). By applying ISCT to GC data, they identified 11 "methylation-driven key regulators" that can classify tumors into distinct subtypes, predict survival, and even influence chemotherapy response, offering new avenues for understanding and treating GC.

What Problem Does This Paper Address?

Gastric cancer (GC) is notoriously heterogeneous, meaning tumors can vary significantly in their genetic makeup, molecular characteristics, and how they behave. This heterogeneity makes it challenging to understand the underlying mechanisms of cancer development and progression, and consequently, to develop effective treatments. While genetic mutations and epigenetic changes (like DNA methylation) are known contributors, the complex interplay between these factors and the tumor microenvironment (TME) – the cells and molecules surrounding the tumor – is not fully understood. Specifically, pinpointing the key regulatory genes that drive this heterogeneity and how they are influenced by both internal (genomic/epigenetic) and external (TME) factors has been a major challenge.

Methods

The core innovation here is the Integrative Sequential Causality Test (ISCT). Unlike traditional mediation tests that treat DNA methylation and CNVs as covariates without assuming a temporal order, ISCT models these factors sequentially, reflecting biological processes. It simultaneously considers how DNA methylation and CNVs influence gene expression, both in a "cis" (within the same gene) and "trans" (across different genes) manner. This allows for a more nuanced understanding of causal relationships. The study also employed MethylCIBERSORT and CIBERSORT for cell type deconvolution, enabling the analysis of immune, stromal, and cancer cell proportions within tumors.

Datasets

The study utilized three primary GC cohorts for integrative analysis:

  • HKU: Hong Kong University cohort
  • TCGA: The Cancer Genome Atlas Stomach adenocarcinoma cohort
  • Singapore: University of Singapore cohort

For validation, five independent GC cohorts were used:

  • Asian Cancer Research Group (ACRG)
  • Australian cohort
  • CPTAC (Cancer Proteome Tumor Analysis Consortium)
  • Yonsei hospital cohort
  • Samsung Medical Center (SMC) cohort

Additionally, four gene expression datasets from a previous study by Oh et al. were used to investigate epithelial/mesenchymal phenotypes.

Key Insights and Ideas

  • ISCT's Superiority: The ISCT method demonstrated significantly higher power in detecting causal relationships compared to traditional mediation tests, especially in scenarios with collinearity, while maintaining a low false positive rate. This suggests ISCT is a more robust tool for dissecting complex molecular regulatory networks.
  • 11 Key Regulators: The identification of 11 common methylation-driven key regulators across multiple datasets is a significant finding. These genes appear to be central players in GC's molecular landscape.
  • Tumor Subtypes Driven by Methylation: The clustering of tumors based on these 11 regulators revealed distinct molecular and histological subtypes, aligning with known classifications (Lauren, MSI/CIN, etc.) and even predicting clinical outcomes. This highlights the critical role of epigenetic regulation in GC heterogeneity.
  • TME's Influence: The study strongly suggests that both tumor-intrinsic variations and the tumor microenvironment (specifically stromal and immune cell proportions) contribute to the expression of these key regulators and, consequently, to GC heterogeneity.

Key Results

  • 11 Methylation-Driven Key Regulators Identified: ADHFE1, CDO1, CRYAB, FSTL1, GPT, PKP3, PTPRCAP, RAB25, RHOH, SFN, and SORD were consistently identified across three major GC cohorts.
  • Three Distinct Tumor Clusters: Based on the expression of these 11 genes, GC tumors were classified into three clusters (C1, C2, C3) that showed significant overlaps with known molecular subtypes (e.g., intestinal vs. diffuse, MSI/CIN vs. GS, epithelial vs. mesenchymal).
  • Prognostic Significance: The identified tumor clusters demonstrated significant differences in patient survival, with C1 generally associated with better survival and C3 with poorer survival.
  • Chemotherapy Response Differences: The study found that chemotherapy (CTX) benefit was dependent on the tumor cluster and stage, suggesting the potential for personalized treatment strategies. For instance, CTX was effective in C1 tumors at Stage III.
  • TME Correlation: Expression of the key regulators was significantly correlated with cellular compositions (immune, stromal, cancer cells) within the tumor, underscoring the TME's role.

Limitations and Future Work

  • Causality vs. Association: While ISCT aims to infer causal relationships, it's important to remember that these are statistical inferences. Further experimental validation is always needed to confirm causality.
  • Complexity of TME: The study acknowledges that the TME is complex, and while it investigated immune and stromal proportions, other TME components might also play roles.
  • Functional Elucidation: While the study identifies key regulators and their downstream genes, detailed functional characterization of these genes, particularly FSTL1, is suggested for future work to reveal their precise roles in tumorigenesis and progression.
  • Personalized Treatment: The findings on chemotherapy response highlight the need for further research into personalizing treatment decisions based on molecular phenotype and tumor stage.
  • In Vitro/In Vivo Studies: The authors suggest that co-cultured systems might be beneficial for further investigating gene-stroma interactions.

Summary updated

Abstract

Abstract Gastric cancer (GC) is a heterogeneous disease of diverse genetic, genomic, and epigenetic alterations. Tumor microenvironment (TME) also contributes to the heterogeneity of GC. To investigate GC heterogeneity, we developed an Integrative Sequential Causality Test (ISCT) to identify key regulators of GC by integrating DNA methylation, copy number variation, and transcriptomic data. Applying ISCT to three GC cohorts containing methylation, CNV and transcriptomic data, 11 common methylation-driven key regulators ( ADHFE1, CDO1, CRYAB, FSTL1, GPT, PKP3, PTPRCAP, RAB25, RHOH, SFN, and SORD ) were identified. Based on these 11 genes, gastric tumors were clustered into 3 clusters which were associated with known molecular subtypes, Lauren classification, tumor stage, and patient survival, suggesting significance of the methylation-driven key regulators in molecular and histological heterogeneity of GC. We further showed that chemotherapy benefit was different in the 3 GC clusters and varied depending on the tumor stage. Both immune/stromal proportions in TME and tumor cell genomic variations contributed to expression variations of the 11 methylation-driven key regulators and to the GC heterogeneity.
BibTeX
@misc{yoo2020molecular,
  author = {Seungyeul Yoo and Quan Chen and Li Wang and Wenhui Wang and Ankur Chakravarthy and Rita A. Busuttil and Alex Boussioutas and Dan Liu and Junjun She and Tim R. Fenton and Jiangwen Zhang and Xiaodan Fan and Suet Yi Leung and Jun Zhu},
  title = {{Molecular and cellular heterogeneity of gastric cancer explained by methylation-driven key regulators}},
  year = {2020},
  doi = {10.1101/2020.01.27.920744},
  howpublished = {bioRxiv},
}
Figure