A Korea–USA multicenter study of duodenal pancreatic fluid–derived exosomal CES1 and MAN2A1 as pancreas-proximal biomarkers for early pancreatic cancer detection.

Abstract

10552 Background: Small pancreatic lesions often yield limited tissue via endoscopic ultrasound (EUS)-guided biopsy, and serum CA19-9 provides poor discrimination in high-risk settings. Duodenal pancreatic fluid (DPF), obtained during routine endoscopy/EUS, is a pancreas-proximal, low-noise biofluid that may enrich tumor-derived exosomal signals. We evaluated whether DPF exosomal proteins complement serum CA19-9 to discriminate PDAC from controls. Methods: DPF was collected during upper endoscopy/EUS with or without secretin. DPF-derived exosomes (n = 26) were profiled by LC-MS/MS, and CES1/MAN2A1 were prioritized by stage-pattern clustering. Candidates were quantified by ELISA in a Korea–USA multicenter cohort (n = 123, prespecified to detect a target AUC of 0.94; 63 PDAC, 60 controls; 4 Korean sites and 1 US site, combined with serum CA19-9 using a locked stacked ensemble model with LightGBM meta-learner trained only on out-of-fold predictions). High-risk controls included IPMN and chronic pancreatitis. Orthogonal validation used pancreatic juice from surgery or ERCP (PDAC n = 20; chronic pancreatitis n = 20). The primary endpoint was AUC for PDAC versus controls, compared with CA19-9 alone and in the CA19-9–normal subgroup; performance was also assessed in early-stage PDAC versus high-risk controls. 95% confidence intervals (CIs) were estimated by cross-validated resampling. Results: LC-MS/MS identified 2,460 proteins; 130 were differentially expressed (> 2-fold, p < 0.05). ELISA confirmed higher exosomal CES1 and MAN2A1 in PDAC versus controls (p < 0.001). The multi-marker model achieved an AUC of 0.995 (95% CI 0.982–1.000), a sensitivity of 98.33% (95% CI 0.931–1.000), and a specificity of 100% (95% CI 0.925–1.000), outperforming CA19-9 alone (AUC 0.63). In patients with CA19-9 in the normal range (< 37 U/mL), the model maintained high discrimination (AUC 0.931, 95% CI 0.746-1.000) with 100% sensitivity. For early-stage PDAC (stage I–II, n = 23) versus high-risk controls (n = 20, including IPMN), AUC was 0.976 (95% CI 0.924-1.000). In orthogonal validation, CES1/MAN2A1 in pancreatic juice were elevated in PDAC patients (p < 0.001) but not in matched serum, supporting pancreas-proximal enrichment. Decision curve analysis showed a higher net benefit than CA19-9 across clinically relevant thresholds. Conclusions: A DPF exosome–derived CES1/MAN2A1 signature integrated with CA19-9 using a locked ensemble model showed high diagnostic discrimination and supportive biological plausibility across a Korea–USA multicenter cohort. These results suggest a pancreas-proximal exosomal biomarker that addresses a critical gap in early detection of pancreatic cancer. Clinical trial information: NCT07030348 .