Gene signatures derived from transcriptomic-causal networks stratify colorectal cancer patients for effective targeted therapy

Background: Gene signatures derived from transcriptomic-causal networks offer potential for tailoring clinical care in cancer treatment by identifying predictive and prognostic biomarkers. This study aimed to uncover such signatures in metastatic colorectal cancer (CRC) patients to aid treatment decisions. Methods: We constructed transcriptomic-causal networks and integrated gene interconnectivity into overall survival (OS) analysis to control for confounding genes. This integrative approach involved germline genotype and tumor RNA-seq data from 1165 metastatic CRC patients. The patients were enrolled in a randomized clinical trial receiving either cetuximab or bevacizumab in combination with chemotherapy. An external cohort of paired CRC normal and tumor samples, along with protein-protein interaction databases, was used for replication. Results: We identify promising predictive and prognostic gene signatures from pre-treatment gene expression profiles. Our study discerns sets of genes, each forming a signature that collectively contribute to define patient subgroups with different prognosis and response to the therapies. Using an external cohort, we show that the genes influencing OS within the signatures, such as FANCI and PRC1, are upregulated in CRC tumor vs. normal tissue. These signatures are highly associated with immune features, including macrophages, cytotoxicity, and wound healing. Furthermore, the corresponding proteins encoded by the genes within the signatures interact with each other and are functionally related. Conclusions: This study underscores the utility of gene signatures derived from transcriptomic-causal networks in patient stratification for effective therapies. The interpretability of the findings, supported by replication, highlights the potential of these signatures to identify patients likely to benefit from cetuximab or bevacizumab.