Immune depletion of the methylated phenotype of colon cancer is closely related to resistance to immune checkpoint inhibitors

Abstract

Background: Molecular typing based on single omics data has its limitations and requires effective integration of multiple omics data for tumor typing of colorectal cancer (CRC).

Methods: Transcriptome expression, DNA methylation, somatic mutation, clinicopathological information, and copy number variation were retrieved from TCGA, UCSC Xena, cBioPortal, FireBrowse, or GEO. After pre-processing and calculating the clustering prediction index (CPI) with gap statistics, integrative clustering analysis was conducted via MOVICS. The tumor microenvironment (TME) was deconvolved using several algorithms such as GSVA, MCPcounter, ESTIMATE, and PCA. The metabolism-relevant pathways were extracted through ssGSEA. Differential analysis was based on limma and enrichment analysis was carried out by Enrichr. DNA methylation and transcriptome expression were integrated via ELMER. Finally, nearest template or hemotherapeutic sensitivity prediction was conducted using NTP or pRRophetic.

Results: Three molecular subtypes (CS1, CS2, and CS3) were recognized by integrating transcriptome, DNA methylation, and driver mutations. CRC patients in CS3 had the most favorable prognosis. A total of 90 differentially mutated genes among the three CSs were obtained, and CS3 displayed the highest tumor mutation burden (TMB), while significant instability across the entire chromosome was observed in the CS2 group. A total of 30 upregulated mRNAs served as classifiers were identified and the similar diversity in clinical outcomes of CS3 was validated in four external datasets. The heterogeneity in the TME and metabolism-related pathways were also observed in the three CSs. Furthermore, we found CS2 tended to loss methylations while CS3 tended to gain methylations. Univariate and multivariate Cox regression revealed that the subtypes were independent prognostic factors. For the drug sensitivity analysis, we found patients in CS2 were more sensitive to ABT.263, NSC.87877, BIRB.0796, and PAC.1. By Integrating with the DNA mutation and RNA expression in CS3, we identified that SOX9, a specific marker of CS3, was higher in the tumor than tumor adjacent by IHC in the in-house cohort and public cohort.

Conclusion: The molecular subtypes based on integrated multi-omics uncovered new insights into the prognosis, mechanisms, and clinical therapeutic targets for CRC.

Keywords: colorectal cancer; immunotherapy; machine learning; methylation; molecular subtype.