Network divergence analysis identifies adaptive gene modules and two orthogonal vulnerability axes in pancreatic cancer.

Transcriptional heterogeneity in pancreatic ductal adenocarcinoma (PDAC) arises not only from changes in gene expression but also from dynamic rewiring of gene-gene coordination. Using a divergent-edge framework applied to 77 155 malignant cells from 42 tumors, we identified four reproducible adaptive modules-integrated growth-energy (IGE), stress-adaptive transcription (SAT), IL-2-linked immune evasion (IL2), and multi-pathway collective invasion (MPC)-that cut across canonical PDAC states and reflect distinct regulatory programs. Integrating these modules with CRISPR-Cas9 dependency profiles and PRISM drug-response data revealed that adaptive behaviors collapse into two higher-order axes: a biosynthetic-metabolic IGE axis enriched for translational and DNA-repair dependencies, and a broader SAT-IL2-MPC stress-immune-invasion axis characterized by proteostasis, cytokine-linked, and cytoskeletal vulnerabilities. This architecture emerges only when divergent-edge modules are mapped into functional genomics space. Module activity also carried clinical relevance in PDAC. SAT-high tumors showed poorer survival, while MPC-high tumors exhibited a similar adverse trend; together, these modules defined a stress-immune-invasion poor-prognosis axis. In contrast, IGE activity showed no overall risk association, although an optimal-cut point-defined IGE-high subgroup displayed modestly improved survival.