This study evaluated 1837 patients with pancreatic adenocarcinoma from the National Clinical Research Center for Cancer (n = 1567) and Shanxi Province Cancer Hospital (n = 270) between November 1, 2023 and November 30, 2024. The inclusion criteria were patients with a confirmed histopathological diagnosis of PDAC, no current anticoagulant use, and no surgical intervention in the last month. Patients were excluded if they had coagulation dysfunction, were receiving anticoagulant therapy, had incomplete basic information, or were pregnant or lactating.

Blood coagulation and fibrinolysis indicators were detected in these patients. Measurements of activated partial thromboplastin time (APTT), prothrombin time (PT), fibrinogen (FIB), fibrin/fibrinogen degradation products, (FDP) and D-dimer were performed using an automated coagulation analyzer (ACL TOP 750; Werfen, Barcelona, Spain).

Three tissue samples from three patients with pancreatic adenocarcinoma were collected in this study. Each patient underwent surgical resection at Shanxi Cancer Hospital. Subsequent pathological diagnosis revealed PDAC in the three patients. All human samples used in this research received ethical approval from ethics committees.

PDAC tissue samples were washed in cryogenic phosphate-buffered saline (Hyclone, Logan, UT, USA) and dissociated using the SeekMate Tissue Dissociation Reagent Kit A Pro (K01801301, SeekGene, Beijing, China) in accordance with the manufacturer’s instructions. After removing red blood cells, the Fluorescence Cell Analyzer Countstar® Rigel S2 (Ruiyu Biotechnology Co., Ltd., Shanghai, China) was used to estimate the cell count and survival rate. Then those fragments and dead cells were cleared. These fresh cells were washed with RPMI 1640 medium (Gibco, Carlsbad, CA, USA) and resuspended in RPMI 1640 supplemented with 2% fetal bovine serum (Gibco) at a density of 1 $\times$ 10⁶ cells/mL.

A 5^′ library preparation kit (No.K00501, SeekGene, Beijing, China) and V(D)J enrichment kit (human T-cell receptor) (No. K00601, SeekGene, Beijing, China) were used to construct the libraries. First, the cells and reverse transcription reagents were mixed and then added to the sample wells on the Chips. Second, the partitioning oil and gel beads were respectively added to these wells. After the formation of emulsion droplets, reverse transcription was performed at 42 °C for 90 min and inactivation was performed at 85 °C for 5 min. Third, the purified cDNA was amplified in this above reaction system. The cDNA product amplified from 5^′ cDNA was enriched for V(D)J amplification. Finally, the indexed PCR amplification was carried out on these samples to amplify the 5^′ portion of these expressed genes containing unique molecular identifiers (UMIs) and cell barcodes. The indexed sequencing libraries were cleaned and analyzed. Finally, sequencing was performed on the NovaSeq 6000 Sequencing System (Illumina, San Diego, CA, USA) with paired-end read length of 150 base pairs. In this study, all cell lines were validated by STR and tested negative for mycoplasma.

Fastp was used to remove low-quality bases and primer sequences from these original data. Then these sequence data were analyzed using SeekOne®Tools (SeekGene, Beijing, China). Seurat (version 4.0.0) was applied to omit these low-quality cells (gene numbers 200 or $>$5000). Mean absolute deviation, variance, and the principles of normal distribution in single-cell RNA sequencing analysis are a strategy to identify and potentially exclude cells whose gene expression profiles are disproportionately influenced by mitochondrial genes (Supplementary Fig. 1). Then the remaining sequence was further analyzed. All these sequence data were analyzed using SeekOne®Tools (SeekGene, Beijing, China) (https://seeksoul.online/index.html#/).

The NormalizeData function from Seurat (version 4.0.0) was used to normalize each cell. The FindVariableGenes function was used to calculate the variable genes. The FindIntegrationAnchors and IntegrateData functions were applied to combine all libraries together, and the ScaleData function was applied to regress out UMIs. Principal component analysis and uniform manifold approximation and projection were applied to reduce the dimensions. These cells were clustered by the 20 dims at a resolution of 0.8. Then the FindAllMarkers function was applied to determine cluster-specific genes. These selected marker genes exhibited expression in at least 10% of cells and at a minimum logFC of 0.25.

The level of copy number variation (CNV) was estimated using the inferCNV R package. Ductal epithelial cells (excluding endocrine cells) were considered to be the putative tumor epithelial cell dataset. Immunocytes served as the reference group. Denoised CNV profiles, excluding those from sex chromosomes, were used to estimate and identify malignant cells.

Gene Ontology and Kyoto Encyclopedia of Genes and Genomes (KEGG) were used to perform biological process and pathway enrichment analysis using the clusterProfiler R package. Gene set variation analysis (GSVA) was performed on hallmark pathways obtained in the Molecular Signatures Database (version 7.0, https://www.gsea-msigdb.org/gsea/msigdb), and GSVA scores were obtained by the R package GSVA (version 1.34).

The lineage differentiation was evaluated by Monocle 2 (version 2.3.6, https://bioconductor.org/packages/release/bioc/html/monocle.html). Then the original data were converted into CellDataSet through the import CDS function, and genes for pseudotime ordering were selected using the dispersionTable function.

The transcription factors and their downstream motif target genes were calculated using the single-cell regulatory network inference and clustering (SCENIC) computational method; then the activity levels of these transcription factor regulons were quantified within each cell. The pySCENIC was run with its default parameters. The CellPhoneDB tool was employed to analyze ligand–receptor interactions in different cell types. In one cell type, specific interactions identified by ligand–receptor expression were present in more than 10% of the cells. Paired comparisons were further made among these included cell types. Then we selected for biologically relevant interactions.

Single-cell sequencing data from the GSE212966 and GSE197177 profile were downloaded from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) for verification purposes. Coagulation factor mRNA expression profiles in pancreatic adenocarcinoma and normal tissues from The Cancer Genome Atlas (TCGA) database (https://www.cancer.gov/ccg/research/genome-sequencing/tcga) and genotype-tissue expression (GTE) database (https://www.gencodegenes.org/) were accessed using the gene expression profiling interactive analysis (GEPIA) database (http://gepia.cancer-pku.cn/). Coagulation factor protein expression profiles in pancreatic adenocarcinoma and normal tissues from the clinical proteomic tumor analysis consortium (CPTAC) database (https://proteomics.cancer.gov/data-portal) were analyzed using the University of ALabama at Birmingham CANcer (UALCAN) data analysis portal (https://ualcan.path.uab.edu).

Paraffin-embedded specimens were sectioned at a thickness of 3 µm. The sections were placed in a constant temperature wax box at 60 °C for 10–15 minutes and then cooled in a refrigerator at 4 °C. The sections were incubated with Tissue Factor/CD142 Recombinant antibody (1:600; 83776-4-RR; Proteintech, Wuhan, China) at 4 °C overnight, then incubated with horseradish peroxidase-labeled goat anti-mouse secondary antibody (1:600; No.760-700; Roche Diagnostics Products, Shanghai, China) at 37 °C for 30 minutes, and then stained with 3,3^′-diaminobenzidine (DAB, Roche Diagnostics Products, Shanghai, China) at room temperature for 2 minutes. Staining was performed using an automatic immunohistochemical platform (BenchMark ULTRA, Roche Diagnostics Products, Shanghai, China). Then, gradient alcohol dehydration, xylene clearing, and neutral gum mounting were performed. Finally, the results were observed under an optical microscope (OLYMPUS BX41, Olympus Corporation, Shinjuku City, Japan). Negative (–): no color reaction; Positive (+): the cells were stained tan or brown.

Data were checked for normality. For normally distributed continuous data, the results are presented as the mean and standard deviation (SD); otherwise, the median (M) and interquartile range are reported. The t-test was used for statistical comparisons of normally distributed data, whereas the Mann-Whitney U test was applied to data with asymmetrical distributions. The survival rates were estimated using the Kaplan-Meier method and compared using the log-rank test. The hazard ratio (HR) and its 95% confidence intervals (CIs) were calculated using Cox proportional hazards regression models. All statistical analyses were conducted using SPSS software (version 23.0; IBS SPSS Statistics, Armonk, NY, USA). p 0.01 (two-sided) was considered statistically significant.

A total of 1837 patients with pancreatic adenocarcinoma were included in this study, of whom more than half (54.98%) showed abnormal coagulation indicators (Table 1). Among the 1567 patients with pancreatic adenocarcinoma from the National Clinical Research Center for Cancer, 54.05% showed abnormal coagulation indicators. Among the 270 patients with pancreatic adenocarcinoma from Shanxi Province Cancer Hospital, 60.37% showed abnormal coagulation indicators. There was no significant difference in the number of patients with abnormal coagulation indicators (p = 0.053), or in age and sex (p = 0.268 and p = 0.836, respectively) between the two groups of patients (Table 1).

The coagulation factors in the mRNA expression profiles between 179 pancreatic adenocarcinoma tissues and 171 normal pancreatic tissues from TCGA database and GTE database were analyzed (Supplementary Fig. 2). The results showed that the mRNA expression levels of F3, F5, F10, and F12 were higher in pancreatic adenocarcinoma tissues than in normal pancreatic tissues (p 0.01; Fig. 1A–D). Then we analyzed the protein expression levels of F3, F5, F10, and F12 between pancreatic adenocarcinoma tissues (n = 74) and normal pancreatic tissues (n = 137) using the CPTAC database, and found that the levels of F3, F5, and F12 were higher in pancreatic adenocarcinoma tissues than in normal pancreatic tissues (p 0.01; Fig. 1E,F,H and Supplementary Table 1). However, there was no significant difference in F10 protein expression between pancreatic adenocarcinoma and normal pancreatic tissues (p $>$ 0.01; Fig. 1G). Survival analysis indicated that high F3 expression in pancreatic adenocarcinoma tissues was associated with a poor prognosis (p 0.01; Fig. 1I–K).

Fig. 1.

The expression and the survival analysis of coagulation factors in pancreatic adenocarcinoma tissues. (A–D) The mRNA expression levels of F3, F5, F10, and F12 between PAAD tissues and normal pancreatic tissues. The expression data are first log₂ (TPM + 1) transformed for differential analysis on the Y-axis. (E–H) Protein expression levels of F3, F5, F10, and F12 between PAAD tissues and normal pancreatic tissues. Z-values represent standard deviations from the median across samples for the given cancer type. Log2 Spectral count ratio values from CPTAC data were first normalized within each sample profile, then normalized across samples. (I–K) The relationship between the expression of F3, F5, F12 and survival outcomes of patients with PAAD. Statistical significance was defined as * p 0.01. PAAD, pancreatic adenocarcinoma; F3, coagulation factor 3.

Information on the three patients is shown in Supplementary Table 2. These tissues were processed and dissociated into single-cell suspensions, which then were used for single-cell sequencing. After the strict quality control and standardized analysis, we obtained transcriptome profile data. A total of 33,300 cells were analyzed, and 13 distinct cell clusters annotated by gene expression profiles were displayed (Fig. 2A). These clusters included fibroblasts (collagen, type I, alpha 1 [COL1A1] and lumican [LUM]), T cells (cluster of differentiation 3 epsilon [CD3E] and CD2), ductal cells (keratin 19 [KRT19] and trefoil factor 1 [TFF1]), macrophages (CD68 and CD14), B cells (membrane spanning 4-domains A1 [MS4A1] and B cell scaffold protein with ankyrin repeats 1[BANK1]), neutrophils (S100A8 and G0/G1 switch 2 [G0S2]), stellate cells (adipogenesis regulatory factor [ADIRF] and chromosome 11 open reading frame 96 [C11orf96]), endothelial cells (platelet endothelial cell adhesion molecule [PECAM1] and Von Willebrand factor [VWF]), plasma cells (joining chain of multimeric IgA and IgM [JCHAIN]) and immunoglobulin kappa constant [IGKC]), mast cells (tryptase alpha-1 and tryptase beta-1 [TPSAB1] and tryptase beta-2 [TPSB2]), acinar cells (protease serine 1 [PRSS1] and regenerating gene 1 alpha [REG1A]), Schwann cell (cadherin-19 [CDH19] and neurexin [NRXN1]), and beta cells (proprotein convertase subtilisin/kexin type 1 inhibitor [PCSK1N] and chromogranin B [CHGB]) (Fig. 2B). Fig. 2C shows that ductal cells (34.64%), fibroblasts (23.34%), macrophages (14.56%), and T cells (11.81%) were the dominant cell populations. The proportion of each cell type differed across patients (Fig. 2D), likely reflecting interpatient heterogeneity. To further verify the distribution of tumor cells in the entire ductal cell population, gene copy numbers were analyzed using InferCNV software. When using macrophages as a reference, the tumor cells within ductal cells exhibited substantial CNVs (Fig. 3A–C). Fibroblasts express well-known fibroblast-related markers such as LUM, COL1A1, DCN (decorin), and SFRP2 (secreted frizzled-related protein 2) (Fig. 2B). We named these fibroblasts CAFs. Fig. 2G shows that F3 expression was elevated in both CAFs (n = 7417 cells) and ductal cells (n = 11,008 cells).

Fig. 2.

Single-cell analysis reveals the complex landscape of PDAC. (A) UMAP showed 13 distinct cell clusters. (B) Bubble plot showing selected cell type-specific markers across all clusters. Size of dots indicated the fraction of cells expressing selected markers, and intensity of color indicated the mean expression level. (C) The proportion of each cell type in all three PDAC patients. (D) The proportion of each cell type in each PDAC patients. (E) The proportion of each cell type in each PDAC patients from “GSE212966”. (F) The proportion of each cell type in all 6 PDAC patients from “GSE212966”. (G) The enriched expression of F3 in UMAP. (H) The F3 expression in UMAP from “GSE212966”. UMAP, Uniform Manifold Approximation and Projection; PDAC, pancreatic ductal adenocarcinoma; F3, coagulation factor 3.

In addition, we conducted independent validation from the GEO data. Patients with PDAC from the GSE212966 dataset showed ductal cells (18.50%), fibroblasts (19.83%), macrophages (9.89%), and T cells (33.09%) (Fig. 2F). Patients with PDAC from the GSE197177 dataset showed ductal cells (26.06%), fibroblasts (11.53%), macrophages (9.54%), and T cells (38.52%) (Supplementary Fig. 3A). These four cell types accounted for a relatively high percentage of all cell types. Meanwhile, within a single patient, the proportion of each cell type exhibited heterogeneity (Fig. 2E, Supplementary Fig. 3B). CAFs and ductal cells also showed enriched expression of F3 in these patients with PDAC from the GSE212966 dataset and GSE197177 dataset (Fig. 2H, Supplementary Fig. 3C).

Ductal cells from these three patients with PDAC were isolated and subjected to cluster analysis to detect some fine differences. The t-distributed stochastic neighbor embedding (t-SNE) plots revealed three distinct subclusters of ductal cells (Fig. 3D,G). As shown in Fig. 3F, the subpopulation of ductal cell 1 constituted a higher percentage than the other ductal cell subpopulations and exhibited substantial copy number variations (CNVs) (Fig. 3E). Fig. 3K suggested that F3 expression was primarily found in the ductal 1 subpopulation. KEGG pathway enrichment analysis revealed that the subpopulation of ductal cell 1 was mainly enriched in phosphoinositide 3-kinase (PI3K)/Akt signaling, focal adhesion, human papillomavirus infection, and ECM–receptor interaction (Fig. 3M and Supplementary Table 3).

Fig. 3.

Single-cell analysis uncovers the ductal cell landscape and the F3 expression in ductal cell subpopulations of PDAC. (A–C) CNVs analysis of the distribution of tumor cells in the total ductal cells. (D) Bubble plot showing selected cell type-specific markers across three distinct subclusters of ductal cells in this study. (E) Using macrophages and B_cells as the reference, the subpopulations of ductal cells were further conducted by CNV analysis. (F) Proportional distribution of each ductal cell subpopulation by individual sample in this study. (G) The t-SNE plots displayed the distribution of three distinct subclusters of ductal cells. (H) Bubble plot showing selected cell type-specific markers across three distinct subclusters of ductal cells in the data of “GSE212966”. (I) Using macrophages and B cells as the reference, the subpopulations of ductal cells were further conducted by CNV analysis in the data of “GSE212966”. (J) Proportional distribution of each ductal cell subpopulation by individual sample in the data of “GSE212966”. (K) Expression of F3 in ductal cell subpopulations of the three PDAC patients in this study. (L) Expression of F3 in ductal cell subpopulations of the six PDAC patients from “GSE212966”. (M) KEGG pathway enrichment analysis of ductal cell 1 in this study. (N) KEGG pathway enrichment analysis of ductal cell 1 in the data of “GSE212966”. CNVs, copy number variations.

Ductal cells from the GSE212966 dataset clustered into four distinct subpopulations (Fig. 3H), of which the ductal cell 1 subpopulation was present in a significantly higher percentage than the other ductal cell subpopulations (Fig. 3J) and demonstrated substantial CNVs (Fig. 3I). F3 was mainly expressed in the ductal cell 1 and ductal cell 4 subpopulations (Fig. 3L). Ductal cell 1 was the main subpopulation expressing F3, due to the low abundance of ductal cells 4. These findings align with the data obtained from three patients with PDAC in this study. In addition, KEGG pathway enrichment analysis results were consistent with those observed in the three patients with PDAC in this study (Fig. 3N).

Next, we assessed the expression of F3 in CAF subpopulations; specifically, CAFs from the three patients with PDAC were analyzed. Results revealed two distinct subpopulations (Fig. 4A,B, Supplementary Fig. 3D). Subpopulation 1 showed enriched expression of inflammatory factors and chemokines, leading to its designation as iCAFs. Subpopulation 2 was named mCAFs because of its high expression of $\alpha$-smooth muscle actin (encoded by ACTA2) (Fig. 4A,B). The ratio of mCAFs to iCAFs is shown in Fig. 4C. Fig. 4D shows that F3 was mainly expressed in iCAFs. This study also analyzed the expression of F3 in patients with PDAC from the GSE212966 dataset and GSE197177 dataset. The results suggested that the iCAF subpopulation also exhibited enriched expression of F3 in patients with PDAC (Fig. 4E, Supplementary Fig. 3E). KEGG pathway analysis was performed, as shown in Fig. 4F,G. The results showed that the PI3K/Akt, complement, and coagulation pathways were prominent in the iCAF subpopulation (Fig. 4F, Supplementary Fig. 3F and Supplementary Table 4). These results indicate that iCAFs might play an important role in hypercoagulability and tumor progression in the early stages of pancreatic cancer.

Fig. 4.

Single-cell analysis uncovers the CAFs landscape and the F3 expression in CAF subpopulations of PDAC. (A) The t-SNE plots displayed the distribution of two distinct subclusters of CAFs (iCAFs and mCAFs). (B) Bubble plot showing selected cell type-specific markers across two distinct subclusters of CAFs in this study. (C) Proportional distribution of each CAF subpopulation by individual sample in this study. (D) Expression of F3 in CAF subpopulations of the three PDAC patients in this study. (E) Expression of F3 in CAF subpopulations of the six PDAC patients from “GSE212966”. (F) KEGG pathway enrichment analysis of iCAFs in this study. (G) KEGG pathway enrichment analysis of mCAFs in the data of “GSE212966”.

The above mentioned results showed that F3 is mainly expressed in iCAFs; therefore, iCAFs were further analyzed. Pseudotime trajectory analysis positioned iCAFs primarily at the beginning of the trajectory and mCAFs at the end (Fig. 5A,C). This result was also confirmed by patients with PDAC from the GSE212966 dataset and GSE197177 dataset (Fig. 5B,D; Supplementary Fig. 3G). Hallmark pathway enrichment analysis is presented in Fig. 5E,F. Analysis of cells in state 1 at the terminal point of trajectory branch 1 confirmed that the iCAF subpopulation primarily emerges at an earlier stage of the pseudotime trajectory. The key regulons of iCAFs were also analyzed in this study. Fig. 5G shows the top five regulons of iCAFs from the three patients with PDAC in this study, Fig. 5H shows the top five regulons of iCAFs from patients with PDAC from the GSE212966 dataset, and Supplementary Fig. 3H shows the top five regulons of iCAFs from patients with PDAC from the GSE197177 dataset. The results revealed that nuclear factor IA (NFIA) is a key transcription factor for driving the iCAF process in both groups of patients with PDAC. The t-SNE plots in Fig. 5I shows the distribution of NFIA expression in iCAFs.

Fig. 5.

The pseudotime trajectory analysis and transcriptional regulatory network analysis of CAFs. (A) The differentiation trajectory of CAFs in this study. (B) The differentiation trajectory of CAFs in the data of “GSE212966”. (C) The top 50 DEGs with the most significant expression level change over the pseudotime trajectory were classified into 3 clusters in this study. (D) The top 50 DEGs with the most significant expression level change over the pseudotime trajectory were classified into 3 clusters in the data of “GSE212966”. (E) Distribution of branch states in the pseudotime trajectory. (F) GSVA (Gene set variation analysis) of each state in pseudotime trajectory using hallmark 50 gene sets. (G) The regulon specificity score was used to rate various regulators of iCAFs in this study. (H) The regulon specificity score was used to rate various regulators of iCAFs in the data of “GSE212966”. (I) NFIA gene expression and regulon AUC (area under curve) scores were visualized in the t-SNE map.

Intercellular network analysis revealed a close relationship between iCAFs and other cell types, particularly mCAFs, macrophages, and ductal cells (Fig. 6A,B). We further analyzed the interaction among growth factors, ECM, chemokines (Fig. 6C–E). Intercellular network analysis showed that iCAFs interacted closely with mCAFs, macrophages, and ductal cells through growth factor receptor–ligand pairs (Fig. 6C, Supplementary Fig. 4). In particular, iCAFs interact closely with tumor cells (ductal cells), mediated by the receptor–ligand pairs of various ligands, such as vascular endothelial growth factor (VEGF), transforming growth factor (TGF), platelet-derived growth factor (PDGF), insulin-like growth factor (IGF), and fibroblast growth factor (Supplementary Fig. 4). Then, iCAFs interacted closely with mCAFs, and ductal cells through ECMs (Fig. 6D, Supplementary Fig. 5). In addition, iCAFs interacted closely with neutrophils through chemokine receptor–ligand pairs (Fig. 6E, Supplementary Fig. 6).

Fig. 6.

Base on iCAF cell-cell interaction network. (A) Heatmap displaying number of potential ligand-receptor pairs between cell groups. (B) Chord diagram showing potential receptor-ligand pairs between cell groups. (C–E) Heatmap displaying ligand-receptor pairs of growth factors (C), extracellular matrixes (D) and chemokines (E) between iCAFs and other cell groups.

We further performed the difference in F3 expression between PDAC tissues and adjacent non-cancerous tissues using immunohistochemical staining. The results showed that the expression of F3 in PDAC tissues was significantly higher than that in adjacent non-cancerous tissues (Fig. 7).

Fig. 7.

Immunohistochemical staining showed the expression of F3 in PDAC tissues and adjacent control tissues. Negative: no color reaction; Positive: the cells were stained tan or brown. Scale bar = 100 µm.