Dual-channel digital surface-enhanced Raman spectroscopy resolves GPC1+MUC4+ single-exosome heterogeneity released by PANC-1 and BxPC-3 pancreatic cancer cells.

Extracellular-vesicle populations average molecular signals across particles and can conceal rare, co-expressing phenotypes. We developed a dual-channel digital surface-enhanced Raman spectroscopy platform that partitions individual CD63-captured vesicles and identifies glypican-1 (GPC1) and mucin 4 (MUC4) with 4-mercaptobenzoic acid and 5,5'-dithiobis(2-nitrobenzoic acid) nanotags. The method was evaluated with eight independent batches from each of PANC-1, BxPC-3, HPDE and hTERT-HPNE cell models, seven blinded PANC-1/BxPC-3 mixtures, artificial-serum spikes and defined interferents. Across 49 256 single-vesicle events, beta-binomial modelling estimated GPC1+MUC4+ fractions of 13.8% for PANC-1 and 37.8% for BxPC-3. The adjusted between-model difference was 24.1 percentage points (batch-cluster bootstrap 95% confidence interval, 21.6-26.5; P = 5.47 × 10-9). The response was linear against log10 concentration from 1.0 × 105 to 1.0 × 109 particles per mL; the extrapolated detection-limit estimate was 1.0 × 104 particles per mL, the overall precision was 6.49% coefficient of variation, and the mean artificial-serum recovery was 100.5%. Blinded mixture estimates tracked the nominal BxPC-3 fraction, while grouped nested cross-validation classified the four cell sources with 90.6% accuracy. Digital SERS agreed with nanoflow cytometry, giving a concordance correlation coefficient of 0.980 and a mean relative bias of 1.51%. These results show that marker co-occurrence, rather than bulk abundance alone, can distinguish pancreatic-cell vesicle phenotypes and provide a quantitative route for studying source-dependent vesicle heterogeneity.