Genomic predictors and evolutionary correlates of chemotherapy response in pancreatic ductal adenocarcinoma.

Abstract

767 Background: Pancreatic ductal adenocarcinoma (PDAC) remains lethal. Neoadjuvant therapy (NAT) is increasingly applied to borderline resectable and locally advanced PDAC. However, most patients show only partial or minimal regression. To define molecular correlates of NAT response, we analyzed paired pre-/post-chemotherapy specimens: pre-treatment endoscopic ultrasound-guided fine needle tissue acquisition (EUS-TA) and matched resection tumors. Methods: At Tohoku University Hospital (2010–2020), 220 NAT-treated PDAC resections were graded by the College of American Pathologists (CAP). 75 cases had sufficient DNA; 69 had paired TA–tumor and 6 tumor-only. NAT regimens primarily included gemcitabine plus oral tegafur/gimeracil/oteracil or gemcitabine plus nab-paclitaxel; a minority received single-agent therapy or regimen switches. Detailed distributions of NAC regimens and clinicopathologic variables are summarized in Table. A deep learning framework (DietNet) was applied to pre-treatment TA data to predict response (CAP1–2 vs CAP3) and integrated with post-treatment tumor profiles to assess molecular evolution. Results: DietNet achieved mean cross-validated AUC 0.77 (95% CI 0.57–0.91); permutation p=0.096. Predictive features included amplifications ( AKT1/2, ERBB2, GATA6, HRAS ) and deletions ( CDKN2A/B, SMAD4, MTAP, RNF43) . Delta (tumor–TA) suggested emergent TP53/STK11 mutation loss and GATA6 copy change. CAP3 vs CAP2 was enriched for CDKN2A/IKBKB loss and gains in BCL9; ARNT/SETDB1/MLLT11; NTRK1; FCGR2B/DDR2/PBX1; PRRX1; TLX3; HOXA9; CSMD3; MYC; RAD21/EXT1 —implicating Wnt/β-catenin, chromatin remodeling, receptor tyrosine kinase activation, and stemness programs. Conclusions: Pre-treatment TA-based modeling moderately predicted NAT response. CAP3 tumors showed CDKN2A/IKBKB loss plus multiple oncogenic gains consistent with chemoresistance. Integrating pre-treatment prediction with post-treatment evolution may aid risk stratification and guide therapeutic development. Clinicopathologic characteristics of patients stratified by pathologic response (CAP1/2 vs CAP3). Variable Categories CAP1/2 n = 25 CAP3 n = 50 Total n = 75 p-value Sex Male:Female 13:12 31:19 44:31 0.684 (Chi²) Age (years) Median (range) 67 (50–78) 65 (56–89) – 0.62 (Wilcoxon) NAC Regimen GS:GnP:Multiple:Other 14:3:1:4 39:10:4:0 53:13:5:4 0.421 (Chi²) Recurrence Yes:No 8:17 11:39 19:56 0.355 (Chi²) Prognosis Alive:Dead:non-cancer death 8:10:7 18:29:3 26:43:6 0.675 (Chi²) UICC Grade G1:G2:G3:G4 5:11:9:0 10:26:11:3 15:37:20:3 0.418 (Fisher) UICC 8th Stage IA:IB:IIA:IIB:III:IV 4:0:0:13:2:6 4:6:3:14:19:4 8:6:3:27:21:10 0.0026 (Fisher) Counts shown. GS=gemcitabine+oral tegafur/gimeracil/oteracil; GnP=gemcitabine+nab-paclitaxel; Multiple = regimen switch during treatment. P-value by Chi-square, Fisher, or Wilcoxon as appropriate.