This study utilized publicly available data from the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) ( https://www.ncbi.nlm.nih.gov/gds), a platform for high-throughput genomics and gene expression datasets. Data usage complied with the original researchers’ guidelines on sharing, privacy, and publication.

A tissue-specific sampling approach was applied using PDAC and matched non-tumor tissues. Sectioned tissue samples enabled detailed analysis of the tumor microenvironment (TME), immune cell populations, dysregulated miRNAs, and metabolic alterations. Focusing on tissue specificity reduced confounding effects common in systemic sampling methods.

This study did not involve human participants directly and therefore did not require informed consent or IRB approval.

Included datasets featured pancreatic ductal adenocarcinoma (PDAC) and adjacent normal tissues, all with high-quality metadata. Only datasets using single-cell RNA sequencing (scRNA-seq) of PDAC and healthy pancreatic tissues were considered. Additionally, miRNA expression profiles used were either experimentally validated or high-confidence predictions, minimizing error risks and supporting robust analysis of cell-specific regulatory patterns.

Preprocessed scRNA-seq data for PDAC were obtained from the NCBI Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/gds), a public database for high-throughput genomic datasets. The dataset GSE155698 includes 41 samples (tissue and PBMC), from which 19 pancreatic tissue samples (16 PDAC, 3 normal) were selected for analysis. Seurat v4.4.0 was used for quality control and downstream processing. ¹⁵

Low-quality cells were excluded based on gene count, UMI content, and mitochondrial gene percentage. Specifically, cells with >6000 genes (possible doublets), <200 genes (low quality), mitochondrial reads >10%, or nCount_RNA <1000 were filtered out. Normalization was applied using the “LogNormalize” method, followed by data scaling and identification of 2000 highly variable genes using Seurat’s “FindVariableFeatures”. Dimensionality reduction was conducted via t-SNE using the first 20 principal components, and cells were clustered using the Louvain algorithm. Cell type annotation was performed using the scType R package and a validated pancreas marker database ( https://sctype.app/). ¹⁶ Immune cell subsets were further subclustered to refine resolution and characterize immune diversity within the PDAC tumor microenvironment.

To assess differences in cell type proportions between PDAC and healthy pancreatic tissue, percentages were calculated and visualized using ggplot2, with statistical significance determined via t-test using the ggpubr package. ¹⁷ Memory CD4 ⁺ T cells, showing significant differences in abundance, were selected for further analysis. Differential expression was assessed using the Wilcoxon test; genes with logFC > 0.25 or < -0.25 and p < 0.05 were considered significant. Downregulated marker scores were calculated using AddModuleScore in Seurat, and grouped by cell type for comparison between PDAC and healthy samples.

Downregulated genes in Memory CD4 ⁺ T cells were analyzed for potential miRNA targets using three tools: MiRWalk ( http://mirwalk.umm.uni-heidelberg.de/ ), TargetScan ( https://www.targetscan.org/vert_80/ ), and miRDB ( https://mirdb.org/ ). These tools predict miRNA-mRNA interactions based on conservation, experimental validation, or high-throughput data. ¹⁸ Predicted interactions were filtered using pandas in Python, retaining miRNAs with binding probability = 1, energy score ≤ -25, and AU content < 0.6 in the 30-nucleotide flanking regions. De novo miRNAs predicted by MiRWalk were also included. The final shortlist comprised miRNAs meeting all criteria across tools for further downstream analysis.

MicroRNAs (miRNAs) exhibit tissue-specific expression patterns and are involved in gene regulation under both normal and pathological conditions. To identify dysregulated miRNAs in pancreatic ductal adenocarcinoma (PDAC) relative to normal pancreatic tissue, differential expression analysis was conducted using GEO2R ( https://www.ncbi.nlm.nih.gov/geo/geo2r) on dataset GSE207345, which includes 7 PDAC tumor samples and 6 adjacent normal tissue samples. This dataset was generated using the Affymetrix Multispecies miRNA-4 Array. miRNAs with a p-value < 0.05 and a log fold change (logFC) < -1.0 or > 1.0 were classified as significantly dysregulated and selected for further evaluation.

To integrate this miRNA analysis with transcriptomic findings from single-cell RNA sequencing (scRNA-seq), we focused on genes that were downregulated in memory CD4 ⁺ T cells from PDAC samples. These were cross-referenced with the list of upregulated miRNAs identified in the tumor dataset. Only miRNAs that were both upregulated and predicted to target these downregulated genes were retained, highlighting their possible roles in suppressing immune functions in the tumor microenvironment.

We used Cytoscape ( https://cytoscape.org/) to construct and visualize miRNA-mRNA interaction networks. ¹⁹ Each miRNA and its target genes were represented as nodes connected by interaction edges. This visualization helped identify key regulatory miRNAs. Cytoscape also provided network analysis tools to detect highly connected nodes and functional modules, enhancing interpretation of miRNA regulatory roles in PDAC pathogenesis. ²⁰

Multiple statistical methods were applied in this study. Cell type proportions in PDAC and healthy tissues were compared using t-tests and visualized with ggplot2. Memory CD4 ⁺ T cells were analyzed via Wilcoxon test, considering genes with logFC > 0.25 or < -0.25 and p < 0.05 as significant. Seurat’s AddModuleScore function was used to assess downregulated markers, with results visualized using ggpubr. Differentially expressed miRNAs were identified via GEO2R, using thresholds of logFC < -1.0 or > 1.0 and p < 0.05. miRNA-mRNA interactions were predicted using MiRWalk, TargetScan, and miRDB, filtered based on binding score, energy, and AU content. Network visualization was performed using Cytoscape.

The PDAC scRNA-seq dataset initially includ ed32,738 RNA features and 107,298 cells. Following quality control, 71,232 high-quality cells were retained by removing dead cells and those with abnormal gene profiles. Using highly variable genes and dimensionality reduction methods, a total of 31 distinct clusters were identified and annotated ( Figure 1A). Most clusters were broadly represented across samples, though clusters 22, 30, 29, 23, 6, 11, and 21 were predominantly composed of cells from healthy tissues ( Figure 1B–C). Additionally, cluster 4 was found to consist entirely of cells from sample P13 ( Figure 2E).

Pancreatic cell types were identified, including acinar, alpha, beta, ductal, gamma (PP), mast cell, stellate cells and progenitor and immune cells ( Figure 2F). Of interest, the unidentified cells comprised a greater portion (65%) of patient P9 and all other subpopulation were less represented including mast, ductal and acinar cells, immune cells accounted for (20%) similar to in their sample. Similar to P2, small proportions of CREB3L3-expressing cells were also present in other samples (<5%), with the exception of P11B showing higher proportion (~18%) ( Figure 2G).

Acinar cells were the predominant cell type in controls C2 and C3 at approximately 75%, while that of C1 was substantially less (15%). Ductal and immune cells were well-represented in control and PDAC samples. Gamma (PP) cells were observed in samples P1, P10, and P11B. Of interest, specimen P16rapresented the highest concentration of ductal ceIIs as achieved in specimen_P11A. Mast cells were identified in every sample, to different degrees.

When comparing PDAC and healthy samples collectively, immune cells were the most abundant in PDAC, accounting for approximately 45% of the total population versus ~20% in healthy tissue. Conversely, acinar cells were more frequent in healthy samples (~35%) but dropped to ~2% in PDAC. Ductal cells increased in PDAC (~25%) compared to ~5% in controls ( Figure 2H). The distribution of some cell types was uneven across individual samples ( Figure 2G). Statistical analysis confirmed that immune cells were significantly more prevalent in PDAC, underlining their central role in the tumor microenvironment ( Figure 3A).

Due to the statistically significant difference in immune cell abundance between PDAC and healthy samples, immune system cells were subsetted and subclustered to identify their subtypes. Using the scType database with immune markers, the following cell types were annotated: CD8 ⁺ NKT-like cells, Classical Monocytes, Macrophages, Memory CD4 ⁺ T cells, Plasma B cells, Pre-B cells, and Progenitor cells ( Figure 3B).

Overall, macrophages, plasma B cells, classical monocytes, pre-B cells, CD8 ⁺ NKT-like cells, and memory CD4 ⁺ T cells were found at higher proportions in PDAC than in healthy samples ( Figure 3C–D). However, this might partially reflect the lower number of healthy controls. Memory CD4 ⁺ T cells were slightly more abundant in control samples individually, but collectively more enriched in PDAC. Macrophages appeared across all samples, except for P10 and P6, where they were below 5%. CD8 ⁺ NKT-like cells were present in all PDAC samples (5–35%) and in only one healthy sample (C1) at a concentration of ~50% ( Figure 4E). A Wilcoxon test confirmed that memory CD4 ⁺ T cells were significantly more abundant in PDAC samples, with a p-value of 0.0021 ( Figure 4F).

The effects of PDAC on memory CD4 ⁺ T cells was detected by differential expression analysis using the Wilcoxon test in Seurat. D. The analysis was successful in finding 4 upregulated genes and 115 downregulated genes. Among them, RPS26, RPS20, HBA2 and HBB were up-regulated. These genes are not only involved in ribosomal function, but also in oxygen transport underlining higher protein biosynthesis and modified redox processes in PDAC.

The increased expression levels of RPS26 and RPS20 could suggest an increase in the protein synthesis capacity as a putative nutrient metabolic adaptation to the tumor microenvironment. In addition, HBA2 and HBB, which also contribute to hemoglobin production, could indicate a dysregulation in oxygen transportation and the tumor hypoxic state response.

Some of the pancreatic enzyme genes such as PNLIP, PRSS1/PRSS2 AMY2A CLPS CELA 2A CPA1 PLA2G1B CELA3B CTRC were markedly repressed. These are genes that usually aid in releasing digestive enzymes and their inhibition indicates exocrine dysfunction in PDAC. In addition, SPINK1, a tissue-protector gene of pancreatitis, was also down-regulated. Several immune-regulatory genes such as SOCS3, CD69 and CXCR4 were down regulated that may limit the signaling pathway/transduction of the activation and migration of immune cells. Reduction of AREG and C4BPA, also known mediators of inflammation and immune regulation, additionally underscores the immuno-suppression characteristic of the PDAC TME.

Transcription regulating genes like JUN, FOS, NR4A2, HES7 and JUNB were downregulated leading to negative regulation of proliferation/survival/stress/inflammation/apoptosis. And the expression levels of metabolic stress response genes, DUSP1, DDIT4 and MT1A were down-regulated. These genes are involved in the adaptation to oxidative stress and regulation of metal ion homeostasis. Module score analysis validated these results, which also exhibited lower expression levels of genes in PDAC than in normal samples ( Figure 4G-H, Figure 3A-B).

We identified 18 downregulated miRNAs and 593 upregulated miRNAs by the differential expression analysis of PDAC tumor samples with paired adjacent healthy tissues ( Figure 3C). The previously extracted downregulated genes, and the miRNA targets (strong or weak interactions), were integrated for correlation to predict the miRNA-mRNA interactions. Starting from an initial number of 9,279 interactions (without refined nodes), the set of upregulated miRNAs was maintained and thus the network reduced to 2,094 relationships.

Sequential stringency cut-off was imposed: binding < -25 and AU content < 0.6. This narrowed the list to 1,052 high confidence interactions. Ten miRNAs were hot candidates for downstream analysis: hsa-miR-1207-5p, hsa-miR-6805-5p, hsa-miR-149–3p, hsa-miR-762, hsa-miR-6846–5p, hsamiR-7109-5p, hsami-R-5787, hsa-mi R-6848–5 p, hsa -mi R -197 most of which had logFC values ranging from 1.4 to 1.6 ( Figure 3 D).

Functionally, the miRNAs target major genes responsible for immune suppression, inflammation, tumor advancement as well as stress evasion. For instance, hsa-miR-1207-5p targets with AREG, CLPS, CTRC, SPINK1 and SOCS3; whereas hsa-miR-6805-5p interplays with FOS, JUN, PNLIP and NR4A2 which were all essential for immune signaling. hsa-miR-149-3p also targets CXCR4, SOCS3 and CHAC1, thereby affecting migration and stress regulation. In addition, hsa-miR-762 targets DDIT4, DUSP1, NR4A2 and other inflammation- and metabolism-associated genes. These data emphasize heterogeneity of the PDAC regulatory landscape, in which induced miRNAs may serve as master regulators controlling immune evasion and tumor resistance. Targeting these miRNAs may potentially provide novel therapeutic approaches to restore immune responses and inhibit the progression of PDAC.

Additionally, hsa-miR-6846-5p targets CHAC1, CLPS, CPA1, FOS, HES7, NR4A2, SPINK1, and SYCN—genes involved in cellular stress response, apoptosis, and proliferation. Four other miRNAs—hsa-miR-7109-5p, hsa-miR-5787, hsa-miR-6848-5p, and hsa-miR-197-5p—also regulate genes implicated in immune modulation, inflammatory signaling, and cellular stress, hsa-miR-7109-5p targets C4BPA, CEL, CHAC1, CLPS, CXCR4, DDIT4, NR4A2, and SOCS3, which are crucial for immune cell migration and signaling. Similarly, hsa-miR-5787 regulates CLPS, CTRC, DDIT4, FOS, INS, PRSS3, and SPINK1, all of which are important in stress and inflammatory pathways. hsa-miR-6848-5p is predicted to target AREG, BTG1, CTRB2, INS, NR4A2, REG1A, and SPINK1, while hsa-miR-197-5p interacts with CHAC1, CPA1, GP2, INS, JUNB, and SYCN—genes related to immune cell behavior and oxidative stress responses. These interactions are detailed in Table 1.

Interestingly, many miRNAs converge on the same targets, emphasizing their central regulatory roles. For example, CLPS, NR4A2, PRSS3, SPINK1, and CTRC are common targets of multiple miRNAs, underlining their involvement in inflammation, immune signaling, and enzyme regulation. CLPS is co-targeted by hsa-miR-1207-5p, hsa-miR-6805-5p, and hsa-miR-762, while NR4A2 is regulated by hsa-miR-6805-5p, hsa-miR-762, and hsa-miR-7109-5p. PRSS3 and SPINK1 are also recurrent targets of several immune-modulating miRNAs such as hsa-miR-1207-5p, hsa-miR-5787, and hsa-miR-7109-5p. Figure 3E illustrates this overlap.

The top upregulated miRNAs in PDAC and their predicted target interactions are summarized in Table 1.

The overall clustering of PDAC and healthy cells was visualized using UMAP ( Figure 1).

Donor-based clustering and pancreatic cell type annotation are shown in ( Figure 2).

Immune cell enrichment between PDAC and healthy tissues is presented in ( Figure 3).

Differences in immune subpopulations, particularly elevated memory CD4+ T cells, are highlighted in ( Figure 4).

The effects of miRNAs, including a volcano plot and interaction network, are illustrated in ( Figure 5).

This work revealed profound immune deficiency in the PDAC tumor milieu and identified memory CD4+ T cells as the main target of suppression. This paper confirms profound suppression of immune and metabolic genes in these cells accompanied by upregulated miRNAs hsa-miR-1207-5p and hsa-miR-6805-5p. Promising observations reported are that the study of miRNA-mRNA pairs emphasizes potential avenues to replenish the immunologic capacity in PDAC.

Pancreatic ductal adenocarcinoma is still one of the major causes of cancer death as it is highly invasive, rapidly progressing and has low Immunogenicity. This issue generally lacks an explanation of the mechanism of how miRNAs regulate immune suppression in the context of PDAC tumor microenvironment with regard to memory CD4+ T cells. The analyses of scRNA-seq data with miRNA level information provide a multi-layered picture of immune dysregulation and point to tangible targets for immunomodulation.

Therefore, some variability can arise due to the differences in sample handling and sequencing methodologies from those that were used in this study but originated from public repositories. Real-time exclusion of systemic samples such as blood reduces potential understanding of general immune processes. Since the interaction between miRNA and mRNA, as well as the identified therapeutic targets, need further experimental validation, the conclusions are derived from computational predictions.