Abstract

SUMOylation is a critical post-translational modification of proteins. However, the roles of SUMOylation-related genes in regulating the tumor immune microenvironment and their potential as therapeutic targets in CRC remain unclear. We integrated transcriptomic data from TCGA and GEO to identify 200 SUMOylation-related genes associated with prognosis. Among these, 58 genes exhibiting significant survival differences were selected and classified into three distinct clusters. Cluster C patients have the worst prognosis, linked to increased tumor proliferation and weakened immune responses. The LASSO combined with Random Survival Forest (RSF) approach demonstrated optimal performance in ten machine learning algorithms, yielding the highest C-index (0.715). Subsequently, six key genes-SENP7, NUP85, UHRF2, BRCA1, TOPORS, and SENP5-were selected using the LASSO-RSF framework to develop a refined prognostic model. This model demonstrated high predictive accuracy, with 1-, 3-, and 5-year overall survival AUC values of 0.943, 0.961, and 0.981, respectively. Furthermore, the risk score derived from the model was significantly correlated not only with tumor metastasis and recurrence, advanced tumor stage, and patient age, but also with immune cell infiltration, tumor mutational burden, and sensitivity to anticancer drugs. Single-cell analysis revealed that the core gene UHRF2 is highly expressed in CD8 + exhausted T cells, but exhibits low expression in conventional CD8 + T cells and natural killer cells. Experimental data demonstrated that UHRF2 was upregulated in CRC and linked to poor prognosis. Silencing UHRF2 markedly inhibited both proliferation and migration of tumor cells in vitro, as well as tumor growth in vivo. In conclusion, the prognostic model derived from SUMOylation-related genes demonstrates robust predictive accuracy for patient outcomes. UHRF2 is associated with immunosuppression and promotes CRC progression, positioning it as a promising therapeutic target