Unraveling the roles of DKK1 and S100P in pancreatic ductal adenocarcinoma: Insights from machine learning, single-cell/spatial transcriptomics, and functional validation

Pancreatic ductal adenocarcinoma (PDAC) is a highly aggressive malignancy with a 5-year survival rate of only 13%. Despite recent advances in diagnosis and treatment, the prognosis of PDAC remains poor, largely owing to its complex tumor microenvironment, epithelial heterogeneity, and therapeutic resistance. In this study, we integrated bulk transcriptomic datasets, single-cell RNA sequencing (scRNA-seq), spatial transcriptomics (ST), machine learning analysis, and functional validation to investigate candidate biomarkers associated with PDAC progression. By intersecting genes selected using LASSO and SVM-RFE algorithms, four candidate genes, DKK1, S100P, NMU, and IGFBP3, were identified. External validation showed that DKK1 and S100P retained stronger diagnostic performance than NMU and IGFBP3. Survival sensitivity analyses using optimal, median-expression, and quartile-based cutoffs supported DKK1 as the most robust prognostic marker, whereas S100P showed stable diagnostic performance and partial prognostic relevance. Single-cell analysis of inferCNV-defined malignant epithelial cells revealed that these genes were associated with distinct malignant epithelial programs and dynamic state transitions. DKK1 was preferentially linked to high-plasticity, invasive, or proliferative epithelial states, whereas S100P showed broader epithelial expression and was also enriched in differentiated epithelial states. CellChat and RCTD-based spatial analyses further suggested that DKK1-high and S100P-high epithelial states occupied distinct epithelial-microenvironmental communication contexts. Finally, functional experiments showed that knockdown of DKK1 or S100P significantly inhibited PDAC cell proliferation and migration. These findings highlight DKK1 and S100P as clinically and biologically relevant markers in PDAC, while NMU and IGFBP3 may reflect specific malignant epithelial or microenvironmental states requiring further validation.