Fig. 1

ScRNA-seq analysis of PDAC subtypes and their clinical relevance. A Schematic of the experimental design. ScRNA-seq was performed on PDAC samples from 21 patients, including non-metastatic PDACs (N = 6), metastatic PDACs (N = 15), and matched liver metastases (N = 7). B Heatmap showing the expression of signature genes for NMF subtypes. Each column in the heatmap corresponds to one cell and each row of the heatmap corresponds to a signature gene of four NMF subtypes. Origin, patient, NMF subtype, and previously reported PDAC classification schemes for each cell are shown at the top of the plot, and the results of GSEA for each signature gene set are shown on the right side of the plot. C Pathway enrichment analysis focusing on origin-specific differences within classical and basal-like subtypes. D and E Kaplan–Meier overall survival curves for PDAC patients based on the fraction of basal-like subtype in the deconvoluted TCGA PDAC RNA-seq dataset (D), and in their primary PDAC in our dataset (SMC cohort) (E). F Forest plot showing the estimated hazard ratios for the clinicopathologic parameters and the proportions of NMF subtypes by multivariate Cox regression analysis of combined scRNA-seq data from our cohort and the two previously published PDAC cohorts. Data are presented as hazard ratio ± 95% confidence interval. G Waterfall plot showing the best percentage change in the sum of the target lesions according to the RECIST v1.1. Each bar indicates a study sample, and the sample is divided into two groups: those with basal-like proportion above 22% (red) and those below (cyan). H and I The proportion of PDAC NMF subtypes and CT scan images before and after chemotherapy of PDAC patients PB2341 (H) and PB2311 (I). J Boxplot showing the distribution of mean CNV correlation coefficients among malignant ductal cells within origins (two-sided Wilcoxon rank sum test: *P < 0.05, **P < 0.01, ***P < 0.001). K and L Hierarchical clustering of CNV profiles in individual patients PB2155 (K) and PB2191 (L). M and N Unsupervised transcriptional trajectories of ductal cells in individual patients PB2155 (M) and PB2191 (N) colored by sample origin. Trajectory directions were indicated by arrows. O and P Dots on trajectory projections (left) were colored by copy number scores at the cellular level and overlaid with contour plots of cells with the strongest copy number variation for known cancer-associated genes in individual patients PB2155 (O) and PB2191 (P). Violin plots (right) showed copy number scores of genes by origin (two-sided Wilcoxon rank sum test: *P < 0.05, **P < 0.01, ***P < 0.001)