Background: Although the significance of intratumoral heterogeneity (ITH) has been widely acknowledged in various cancers, its role in pancreatic cancer (PC) remains underexplored and warrants further investigation.
Methods: Pancreatic cancer transcriptomic data were acquired from the TCGA and GEO databases. The DEPTH2 algorithm, in combination with differential expression analysis, was used to identify genes associated with intratumoral heterogeneity (ITH). We applied univariate Cox regression analysis and multiple machine learning techniques to establish a reliable prognostic model. Patients were then stratified according to their ITH scores, and differences between subgroups were examined through pathway enrichment analysis, immune cell infiltration profiling, and drug response prediction. Furthermore, we conducted subcellular localization and differential analysis of ITH using single-cell data, followed by cell-cell communication analysis to explore interaction relationships and identify key pathways.
Results: Patients exhibiting lower intratumoral heterogeneity (ITH) levels demonstrated poorer clinical outcomes. The constructed 11-gene signature successfully differentiated individuals into high- and low-risk categories with significant survival differences. Immune profiling revealed notable differences in immune cell composition between the two groups, with patients in the high ITH cohort exhibiting enhanced immune activation. Drug sensitivity analysis indicated a differential response to therapies, with high-risk patients more resistant to certain drugs. Single-cell RNA sequencing identified a greater ITH score in epithelial cells, highlighting key interactions, particularly involving Galectin signaling pathways.
Conclusion: Our results highlight intratumoral heterogeneity's prognostic and therapeutic relevance in pancreatic cancer, suggesting its potential utility in guiding individualized treatment approaches.
Keywords: Drug sensitivity; Immune infiltration; Intratumoral heterogeneity; Pancreatic cancer; Prognostic model.
© 2025. The Author(s).