Immunotherapy response predictors in colorectal cancer: A multi-omics approach

Predicting immunotherapy (IT) outcomes remains a clinical challenge as approved biomarkers show limited efficacy. Although aberrant glycosylation has been linked to tumor progression, its role in predicting IT outcomes is underexplored. With the goal of aiding in patient stratification, we integrated genomic, transcriptomic and glycomic data to explore the predictive capacity of glycoimmune genes. Using unsupervised machine-learning methods and cross-validation with two independent cohorts, we developed the GlycoImmune Signature (GIS), an 18-gene expression signature associated with improved response to IT and survival outcomes. We applied the GIS to colorectal cancer (CRC) samples from TCGA-COAD and characterized their immune infiltration and proteomic profiles. MSI-H patients and predicted responders (TIDE algorithm) showed higher GIS scores (p<0.0001). High GIS-scoring (GISH) patients exhibited a “hot” tumor microenvironment (TME) and upregulation of immune-related signatures compared to low GIS-scoring (GISL) cases. Interestingly, approximately 40% of MSI-L/MSS patients were GISH, suggesting that the GIS may identify patients who might respond to IT but are not currently considered by clinical guidelines. Single-cell transcriptomics data of CD45+ and tumor cells (GSE200997) of treatment-naïve samples were aggregated per patient to generate pseudobulk profiles and classify them into GISH and GISL. GISH tumors showed an enrichment of effector CD8+ T cells and reduced infiltration of Tregs and Th17 cells, confirming previous findings. Mapping these GISH and GISL-associated cell states onto cells from patients treated with IT (GSE205506) revealed that GISH-associated effector CD8+ T cells were more cytotoxic and prevalent in responders, while those from GISL patients expressed exhaustion markers ( LAG3, PDCD1, CTLA4, EOMES, TOX). In turn, GISH-associated Tregs showed the loss of regulatory markers ( FOXP3 , CTLA4 , TIGIT ). To translate our transcriptomic marker into a proteomic one, we analyzed proteomic data from GISH patients in TCGA-COAD, finding 29 upregulated proteins associated with immune response and cell killing, and 9 downregulated proteins associated with metabolic reprogramming. Among the former, six proteins present relevant biological roles and are highly expressed in CRC tumors, which could serve as cost-effective clinical readouts. Overall, these findings position the GIS as a multi-omics surrogate of IT response in CRC and highlight its potential to expand patient eligibility for immunotherapy.