Single-cell and spatial transcriptomic analysis reveals tumor cell heterogeneity and underlying molecular program in colorectal cancer

Background Colorectal cancer (CRC) is a highly heterogeneous tumor, with significant variation in malignant cells, posing challenges for treatment and prognosis. However, this heterogeneity offers opportunities for personalized therapy.Methods The consensus non-negative matrix factorization algorithm was employed to analyze single-cell transcriptomic data from CRC, which helped identify malignant cell expression programs (MCEPs). Subsequently, a crosstalk network linking MCEPs with immune/stromal cell trajectory development was constructed using Monocle3 and NicheNet. Additionally, bulk RNA-seq data were utilized to systematically explore the relationships between MCEPs, clinical features, and genetic mutations. A prognostic model was then established through Lasso and Cox regression analyses, integrating clinical data into a nomogram for personalized risk prediction. Furthermore, key genes associated with MCEPs and their potential therapeutic targets were identified using protein-protein interaction networks, followed by molecular docking to predict drug-binding affinity.Results We classified CRC malignant cell transcriptional states into eight distinct MCEPs and successfully constructed crosstalk networks between these MCEPs and immune or stromal cells. A prognostic model containing 15 genes was developed, demonstrating an AUC greater than 0.8 for prognostic evaluation over 1 to 10 years when combined with clinical features. A key drug target gene TIMP1 was identified, and several potential targeted drugs were discovered.Conclusion This study demonstrated that characterization of the malignant cell transcriptional programs could effectively reveal the biological features of highly heterogeneous tumors like CRC and exhibit significant potential in tumor prognosis assessment. Our research provides new theoretical and practical directions for CRC prognosis and targeted therapy.