3D tissue model to patient atlas: dynamic network topology reveals PDAC chemoresistance bottlenecks.

Pancreatic ductal adenocarcinoma (PDAC) develops chemoresistance through dynamic cell-state transitions, yet the regulatory network architecture governing drug-tolerant states remains poorly characterised, because static expression snapshots cannot capture how interaction topology evolves along treatment-induced trajectories. We established a 3D PANC-1 tissue model on decellularised porcine intestinal matrix, which reaches patient-like proliferation indices and tissue-niche-specific drug responses, and performed single-cell RNA sequencing across four conditions: control, gemcitabine (GEM), TGF-β1, and their combination. Coupling pseudotime trajectory inference to dynamic protein-protein interaction network analysis, we tracked betweenness centrality continuously across cell-state transitions rather than comparing discrete states. GEM survivors engaged a proliferative programme (E2F1, CDK1, AURKA, TPX2, TOP2A, BIRC5), whereas TGF-β1 induced an independent EMT attractor (SPOCK1, MBOAT2, COL5A1, THBS1, FN1). Where the two converged, an emergent bottleneck appeared: CDK1 betweenness centrality spiked selectively, restrained by CDKN1A (p21) and WEE1, defining an S-phase-persistence state enriched for GEM survivors. Cross-examination in a PDAC atlas of 726,107 cells from 231 patients showed 8.7-fold metastatic enrichment of triple-positive cells. This framework, linking controlled 3D tissue experimentation, single-cell systems analysis and clinical-scale patient data, is applicable to drug resistance across cancer types.