miR-4787-3p and miR-581 as predictive biomarkers for fibrosis and prognosis in pancreatic cancer

Abstract

Background

Pancreatic cancer (PC) is characterized by a profoundly immunosuppressive and fibrotic tumor microenvironment, driven largely by cancer-associated fibroblasts (CAFs). While microRNAs (miRNAs) are key regulators in PC, their specific roles in stromal fibrosis remain poorly understood. This study aims to investigate the prognostic and fibrotic relevance of miR-4787-3p and miR-581 in PC.

Methods

We identified differentially expressed miRNAs from The Cancer Genome Atlas (TCGA) database. MiR-4787-3p and miR-581 were selected for their prognostic value using Cox regression and Kaplan-Meier analysis. Their expression was validated in vitro using a co-culture model of PC cells and CAFs, and in clinical blood and tissue samples via RT-PCR, with correlations to clinical-pathological features and fibrosis markers (TGF-β1, COL1A1) assessed.

Results

TCGA analysis revealed significant downregulation of both miRNAs in PC tissues, which correlated strongly with poor patient survival. Mechanistically, their expression was suppressed in PC cell lines co-cultured with CAFs and inversely correlated with TGF-β1 and COL1A1 levels. Consistent with these findings, miR-4787-3p and miR-581 levels were markedly lower in PC patient tissues and blood samples compared to controls, with the lowest expression observed in high-fibrosis tissues. Both miRNAs demonstrated high sensitivity and specificity for distinguishing PC from non-cancerous controls and for identifying fibrosis within tumors.

Conclusion

The downregulation of miR-4787-3p and miR-581 is a hallmark of pancreatic cancer, closely associated with tumor fibrosis and poor prognosis. These miRNAs represent promising circulating and tissue biomarker candidates for diagnosing PC and evaluating its fibrotic burden, suggesting potential utility for early detection and personalized treatment strategies.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-026-04389-3.

Introduction

Pancreatic cancer (PC) is among the most lethal malignancies, frequently termed the “king of cancers,” with a 5-year survival remaining below 6%. Epidemiological forecasts suggest that by 2030, worldwide, PC is anticipated to be the second most frequent cause of cancer-related mortality [1, 2]. The poor prognosis of pancreatic cancer is largely driven by its marked intratumoral heterogeneity, a dense fibrotic stroma, and a profoundly immunosuppressive tumor microenvironment (TME). Among these features, stromal fibrosis represents a distinctive hallmark of pancreatic cancer, setting it apart from most other solid tumors. Activated PSCs and CAFs play central roles in this process by producing large amounts of ECM, particularly collagen. In addition, cancer cells themselves actively promote and sustain fibrogenesis. The resulting desmoplastic barrier not only restricts the penetration of therapeutic agents but also establishes a protective niche that favors tumor cell survival, thereby diminishing treatment effectiveness [3–6].

At present, the standard therapeutic options for pancreatic cancer include surgical resection, chemotherapy, radiotherapy, interventional procedures, and palliative care. Unfortunately, over 80% are identified at an advanced stage, at which point surgery is no longer feasible. Consequently, the majority of individuals depend on chemoradiotherapy-based regimens, which provide only modest clinical benefit. Increasing evidence indicates that the dense stromal fibrosis characteristic of the pancreatic cancer microenvironment is a key factor limiting drug penetration and diminishing the efficacy of systemic therapies. In addition, the late onset of clinical status and the lack of credible early screening instruments further worsen patient outcomes. Hence, the identification of innovative and accurate diagnostic markers enabling early intervention and disease monitoring has become an urgent priority in PC research [7–11].

MiRNAs are a class of short, endogenously expressed non-coding RNA molecules that play an essential role in the post-transcriptional regulation of gene expression. Through base-pairing with target mRNAs, miRNAs facilitate mRNA decay or suppress translational initiation, thereby finely modulating levels of protein production. This regulatory mechanism enables miRNAs to exert widespread influence over numerous fundamental biological processes, including but not limited to: cell proliferation and development, metabolic homeostasis, programmed cell death, as well as innate and adaptive immune responses. Their involvement in these pathways highlights their critical importance in maintaining normal physiological functions and disease pathogenesis [12, 13, 14].

Emerging studies suggest that multiple miRNAs play key roles in driving the oncogenesis and metastatic evolution of PC. Notably, MIR-766-3p expression is markedly reduced in both pancreatic tumor specimens and in vitro cellular models. Gain-of-function experiments reveal that re-expression of this miRNA suppresses malignant proliferation, stem-like properties, and cell cycle advancement through direct interaction with MAPK1 and subsequent modulation of the MAPK/ERK signaling cascade. These findings highlight MIR-766-3p as a promising biomarker and potential therapeutic target in pancreatic cancer [15]. Research has identified miR-210, upregulated by hypoxia-inducible factor-1α under hypoxic conditions, as a pivotal regulator of epithelial-mesenchymal transition, cancer stem cell phenotypes, and therapeutic resistance in pancreatic ductal adenocarcinoma, highlighting its role in driving invasive and treatment-resistant tumor behavior [16]. Studies have revealed that analyzing miR-1247 methylation alongside KRAS and TP53 mutations in pancreatic juice offers a precise diagnostic tool for differentiating pancreatic cancer from low-grade PanIN/IPMN, particularly for detecting intraductal papillary mucinous carcinoma [17]. Notably, while miRNAs like miR-21 and miR-210 have been implicated in PC pathogenesis, their roles are often context-dependent and not specifically linked to stromal fibrosis as a primary focus. In contrast, miR-4787-3p and miR-581 represent relatively unexplored miRNAs in PC. Our study uniquely focuses on their association with the fibrotic tumor microenvironment, a critical but less understood aspect of PC biology. We hypothesize that these miRNAs may serve as key regulators at the interface of tumor cells and cancer-associated fibroblasts, influencing stromal remodeling and fibrosis progression, thereby offering novel insights beyond the functions of previously characterized miRNAs in PC.

This study focuses on elucidating the relationship between miR-4787-3p, miR-581, and PC, investigating their association with tumor fibrosis and evaluating their potential as diagnostic biomarkers. This study focuses on pinpointing novel therapeutic targets that could facilitate earlier diagnosis and more effective therapeutic approaches for PC, with the goal of enhancing patient outcomes and prognosis.

Materials and methods

Detection of dysregulated MiRNAs following data analysis

Transcriptomic data (miRNA-seq) and corresponding clinical information for pancreatic cancer were obtained from The Cancer Genome Atlas Pancreatic Adenocarcinoma (TCGA-PAAD) project (Data Release: https://portal.gdc.cancer.gov/; accessed on May 1, 2024). The analyzed dataset comprised 183 samples (179 primary tumor tissues and 4 solid tissue normal samples). Raw count data were preprocessed and normalized as follows: using R software (v4.3.3) and relevant bioinformatics packages: (1) Removal of miRNAs with zero counts across all samples; (2) Normalization of read counts using the Trimmed Mean of M-values (TMM) method implemented in the edgeR package to account for differences in library sizes and RNA composition; (3) Conversion of normalized counts to Log2 Counts Per Million (Log2CPM) for downstream differential expression analysis. This normalized and log-transformed expression matrix was used for all subsequent bioinformatic analyses.

Given the limited number of normal tissue samples in the TCGA-PAAD cohort (n = 4), a rigorous consensus-based differential expression strategy was adopted to enhance robustness. Differentially expressed miRNAs (DE-miRNAs) between tumor and normal tissues were identified using three independent algorithms-limma (using the voom function for count data), edgeR (using the quasi-likelihood F-test), and DESeq2 (using the Wald test). To ensure robust and high-confidence results, the final list of DE-miRNAs was defined by taking the intersection of significant miRNAs identified by all three packages. A miRNA was considered differentially expressed only when it simultaneously met the significance criteria of adjusted P-value (FDR) < 0.05 and |log2 fold change| > 1.0 across all three methods. This stringent consensus approach was adopted to minimize false-positive discoveries, which is crucial given the limited number of normal tissue samples (n = 4) in the TCGA-PAAD cohort. The |log2 FC| > 1.0 threshold (corresponding to a 2-fold change) was selected to focus on miRNAs with substantial expression alterations likely to possess biological relevance, while the FDR < 0.05 criterion controls the false discovery rate in multiple testing. This stringent multi-algorithm approach minimizes false-positive results and strengthens the reliability of the initial screening.

The final set of DE-miRNAs was visualized using heatmaps (“pheatmap”) and volcano plots (“ggplot2”), providing an overview of global expression alterations. Candidate miRNAs identified through this integrative process were subsequently subjected to prognostic evaluation and experimental validation in vitro and in clinical samples, as described in the following sections.

Kaplan-Meier method

Kaplan-Meier survival analysis was performed to evaluate the prognostic value of the selected miRNAs based on TCGA clinical data. Patients were stratified into high- and low-expression groups based on the median expression level of each miRNA. Differences in overall survival (OS) between the groups were assessed using the log-rank test. Additionally, univariate Cox proportional hazards regression analysis was conducted for each miRNA to calculate hazard ratios (HRs) and 95% confidence intervals (CIs). All statistical computations and data analyses were conducted utilizing R with the “survival” and “survminer” packages, and P < 0.05 was considered statistically significant.

Cell treatment and grouping

Four cell lines (Capan-1, BxPC-3, Panc-1, and MIA-PaCa-2) and the immortalized pancreatic ductal epithelial cell line hTERT-HPNE were obtained from the central laboratory. BxPC-3 cells were cultured in RPMI 1640 medium (Gibco) supplemented with 10% FBS (Gibco), whereas Capan-1, Panc-1, MIA-PaCa-2, and hTERT-HPNE cells were maintained in DMEM (Gibco) containing 10% FBS. All cultures were kept at 37 °C in a humidified incubator with 5% CO₂.

To model stromal fibrosis, cells were mixed-cell culture with CAFs (Beijing Koresibo Biotechnology Co., Ltd., Beijing, China) at a 1:1 ratio in the respective growth medium under the same culture conditions. The normal control group consisted of untreated hTERT-HPNE cells, while the untreated cancer cell group included Capan-1, BxPC-3, Panc-1, and MIA-PaCa-2 cells cultured alone. The treatment group was defined as pancreatic cancer cells co-cultured with CAFs to simulate a fibrotic microenvironment. The identity and purity of the commercial CAFs were verified upon receipt by assessing the expression of established CAF-specific markers. qRT-PCR analysis demonstrated that the mRNA levels of ACTA2 (encoding α-SMA) and FAP were markedly elevated in these CAFs compared with the hTERT-HPNE control cell line [18, 19].

Clinical data

A retrospective, single-center study was conducted, including 37 patients with PC who underwent surgical treatment in the Department of Oncology between May 2018 and May 2024. Blood samples from 37 age- and sex-matched healthy individuals undergoing routine physical examinations were collected as controls. The sample size was determined by the number of eligible patients meeting predefined inclusion and exclusion criteria during the study period; therefore, no formal a priori power calculation was performed. Although adequate for an exploratory investigation, the limited sample size restricts the power of subgroup analyses. To assess the influence of potential confounders on the association between miRNA expression and fibrosis, baseline clinical characteristics were compared between the low-fibrosis and high-fibrosis groups.

Fibrosis grading in pancreatic cancer was determined using an integrated histological and imaging-based evaluation. Histologically, fibrosis severity was assessed according to collagen deposition and tissue architectural changes. Mild fibrosis was characterized by minimal collagen accumulation and preserved acinar structures; moderate fibrosis showed increased fibrous tissue with partial architectural distortion; and severe fibrosis exhibited extensive collagen deposition, marked sclerosis, and near-complete acinar loss. All histological assessments were independently conducted by two experienced pathologists blinded to clinical data, yielding substantial inter-observer agreement (Cohen’s κ = 0.78). Any discrepancies (n = 5) were resolved through joint review to reach consensus. To further support the reliability of the grading system, collagen volume fraction (CVF) was quantified in a subset of cases (n = 20) using Masson’s Trichrome staining combined with ImageJ analysis, and CVF values demonstrated strong concordance with the assigned histological grades.

Imaging evaluation was performed using CT and MRI features associated with stromal fibrosis. Mild fibrosis was defined by minimal fibrotic signal and well-demarcated tumor margins; moderate fibrosis by partially blurred margins with visible fibrotic strands; and severe fibrosis by pronounced scar-like changes, ill-defined tumor borders, and extensive sclerosis. Final fibrosis classification was established by integrating both histological and imaging findings and categorized as low (< 30% fibrotic area), moderate (30–60%), or high (> 60%) fibrosis.

(1) Inclusion Criteria.

(a) Histologically confirmed pancreatic ductal adenocarcinoma; (b) No prior history of pancreatic surgery, radiotherapy, or chemotherapy; (c) Normal hepatic, renal, and cardiac function; (d) Complete clinical and pathological data available.

(2) Exclusion Criteria.

(a) Presence or history of other malignancies, such as duodenal cancer; (b) Pathological evidence indicating non-pancreatic origin of the tumor; (c) Concurrent acute pancreatitis or autoimmune diseases; (d) Incomplete clinical data.

qRT-PCR

Blood samples were processed to isolate plasma and blood cells by centrifugation, while tissue or blood cell samples were lysed using 1 mL of Trizol reagent per 50 µL of blood cells. The lysate was incubated on ice for 10 min with gentle agitation to ensure thorough mixing. Chloroform was subsequently introduced into the lysate to induce liquid-liquid phase separation. The mixture was then subjected to centrifugation, resulting in the isolation of the RNA-enriched aqueous supernatant.To precipitate RNA, the aqueous phase was combined with isopropanol and centrifuged, yielding an RNA pellet. This pellet was washed with 75% ethanol, centrifuged again, air-dried to remove residual ethanol, and dissolved in DEPC-treated water. The purified RNA was aliquoted and cryopreserved at −80 °C to ensure stability until further molecular analysis.

Following RNA extraction, cDNA was synthesized by reverse transcription using a kit according to the manufacturer’s instructions. The synthesized cDNA was combined with gene-specific primers and SYBR Green dye in a 96-well plate for qRT-PCR. Amplification was performed on a StepOnePlus™ Real-Time PCR System. Relative expression levels of target genes were computed using the 2^−ΔΔCt, with appropriate normalization controls. Detailed oligonucleotide sequences are detailed in Table 1.

Table 1.

Oligonucleotide primer sequences

Target object	Primer sequence (5’−3’)
miR-4787-3p	F: GATGCGCCGCCCACTGCCCCGCGC
	R: GTGCAGGGTCCGAGGT
miR-581	F: TCTTGTGTTCTCTAGATCAGT
	R: GTGCAGGGTCCGAGGT
U6	F: GCTTCGGCAGCACATATACTAAAAT
	R: CGCTTCACGAATTTGCGTGTCAT
TGF-β1	F: AGACCTGGGCTTCACTGGTG
	R: CTCGGGAGTGGTGATGGAG
COL1A1	F: CAGGAGGAGAGGAGAGGT
	R: GGAGGTGGAGGAGGAAG
GAPDH	F: TGCACCACCAACTGCTTAG
	R: GGCATGGACTGTGGTCATG

Statistical analysis

Statistical analyses were performed using SPSS 26.0. Categorical variables are presented as counts and percentages, and differences between groups were assessed using the chi-square (χ²) test. Continuous variables are expressed as mean ± standard deviation (x̄ ± SD). The normality of the distribution for continuous variables was assessed using the Shapiro-Wilk test, and the homogeneity of variances was verified using Levene’s test. Based on these assessments, comparisons between two independent groups were conducted using Student’s t-test (for data meeting parametric assumptions) or the Mann-Whitney U test (when assumptions were violated). Analysis of variance (ANOVA) followed by Tukey’s post hoc test (which inherently adjusts for multiple comparisons) was applied for comparisons across three or more groups meeting parametric assumptions; otherwise, the Kruskal-Wallis test followed by Dunn’s post hoc test with Bonferroni correction was used. Relationships between miR-4787-3p and miR-581 expression were evaluated using Pearson correlation analysis (for normally distributed data) or Spearman’s rank correlation (for non-normally distributed data).

For diagnostic performance, ROC curves were generated using the ‘pROC’ package (version 1.18.0) in R, with 95% CI calculated for the area under the curve (AUC) calculated using the DeLong method. Comparisons between ROC curves (e.g., individual miRNA vs. combined model) were performed using the DeLong test. To evaluate model robustness and reduce potential overfitting, internal validation was performed using bootstrapping with 1000 resamples, and optimism-corrected AUC values were reported alongside the apparent AUC. A P < 0.05 was considered statistically significant.

Results

Filtering and identification of MiRNAs and targets

Analysis of TCGA data identified multiple differentially expressed miRNAs (DE-miRNAs), including upregulated and downregulated miRNAs. A total of 75 DE-miRNAs were discovered, comprising 31 upregulated and 44 downregulated miRNAs. Further selection derived from univariate Cox regression forest plots and Kaplan-Meier method highlighted miR-4787-3p and miR-581 as candidate miRNAs for subsequent experiments. The univariate Cox regression analysis indicated that low expression of miR-4787-3p was associated with significantly poorer overall survival (HR = 0.59, 95% CI: 0.38–0.91, P = 0.014), as was low expression of miR-581 (HR = 0.56, 95% CI: 0.36–0.87, P = 0.018). To explore the potential biological roles of miR-4787-3p and miR-581 in pancreatic cancer, we performed in silico functional enrichment analysis. Putative target genes of miR-4787-3p and miR-581 were predicted using the TargetScan and miRDB databases. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses for the commonly predicted high-confidence target genes were conducted using the DAVID bioinformatics resource. The KEGG pathway analysis revealed significant enrichment (FDR < 0.05) for pathways critically involved in cancer, including the “MAPK signaling pathway”. These computationally predicted associations suggest that miR-4787-3p and miR-581 may participate in pancreatic cancer pathogenesis by regulating key oncogenic signaling cascades, providing a plausible mechanistic context for their observed correlation with fibrosis and poor prognosis (Supplementary Tables 1 and Fig. 1).

Fig. 1.

Filtering and identification of miRNAs and Targets. A Heatmap depicting differentially expressed miRNAs recognized using the “limma” package. B Volcano plot displaying the distribution of differentially expressed miRNAs analyzed with the “limma” package. C, E Kaplan-Meier survival curves illustrating overall survival associated with miR-4787-3p and miR-581, respectively. Differences between groups (High vs. Low expression) were assessed using the log-rank test (n = 179 patients from TCGA-PAAD). D Univariate Cox regression forest plot highlighting miR-4787-3p and miR-581 as potential prognostic indicators. F KEGG pathway enrichment analysis. plots show the top enriched pathways, with the significance level (-log10(FDR)) indicated. G GO enrichment analysis

Levels of miR-4787-3p and miR-581 in PC cell lines

A fibrosis model was successfully established in cell lines, as indicated by the upregulation of TGF-β1 and COL1A1, confirming the reliability of the model. qRT-PCR was then performed to measure the levels of miR-4787-3p and miR-581 in cells. The results unveiled that both miR-4787-3p and miR-581 were markedly downregulated in the cell lines relative to the hTERT-HPNE control. Furthermore, co-culture with CAFs led to a further decrease in the expression of these miRNAs in cells (Fig. 2).

Fig. 2.

Expression of miR-4787-3p and miR-581 in Pancreatic Cancer Cell Lines. A, B Relative expression of COL1A1 and TGF-β1 across the different cell lines. C, D miR-4787-3p and miR-581 expression in the indicated cell lines. Data are presented as mean ± SD (error bars represent standard deviation) from n = 3 independent biological replicates (each with technical triplicates). Statistical significance was determined by one-way ANOVA followed by Tukey’s post hoc test. *P < 0.05, **P < 0.01, ***P < 0.001 for Un-treated (cancer cells alone) vs. Treated (co-culture with CAFs); ^#P < 0.05, ^##P < 0.01, ^###P < 0.001 for Control (hTERT-HPNE) vs. Un-treated (respective cancer cell line alone)

Clinical characteristics of participants

Thirty-seven patients with pancreatic cancer were stratified into a low-fibrosis group (n = 20; encompassing fibrosis-free and low-grade fibrosis) and a high-fibrosis group (n = 17; including moderate and high-grade fibrosis) based on integrated pathological and imaging assessments. Baseline characteristics such as age and gender did not differ significantly between groups or between patients and healthy controls. In contrast, significant differences were noted in tumor differentiation, TNM stage, and lymph node metastasis between the low- and high-fibrosis groups (Table 2).

Table 2.

Comparison of clinical data

Indicator	Low-degree(n = 20)	High-degree (n = 17)	t/χ²	P
Age	64.90 ± 5.84	60.50 ± 7.65	1.446	0.165
Gender
Male	14	10	0.504	0.478
Female	6	7
Differentiation Degree
High/Moderate	15	3	12.099	0.001
Low	5	14
Tumor Maximum Diameter (cm)
< 2 cm	5	1	2.472	0.116
≥ 2 cm	15	16
Tumor Location
Pancreatic Head	10	9	0.032	0.858
Pancreatic Body and Tail	10	8
TNM Stage
I ~ IIA Stage	15	3	12.099	0.001
IIB ~ III Stage	5	14
Lymph Node Metastasis
Yes	2	7	4.852	0.028
No	18	10

To provide objective validation of the semi-quantitative histological fibrosis grading, the CVF was measured in a subset of samples (n = 20). CVF values demonstrated a strong positive correlation with the assigned histological fibrosis categories (Low/Moderate/High; Pearson r = 0.82, P < 0.001), supporting the reliability and accuracy of the grading system in reflecting true collagen deposition within the tumor microenvironment.

Levels of miR-4787-3p and miR-581 in blood and tissue samples

Further test was conducted to examine the levels of miR-4787-3p and miR-581 in blood and tissue samples. Both miRNAs were dramatically reduced in tumor tissues. Moreover, the levels were markedly lower in fibrotic tissues compared to non-fibrotic tissues, indicating a potential association with fibrosis progression (Fig. 3).

Fig. 3.

Level of miR-4787-3p and miR-581 in blood and tissue samples. A, B, E, F Relative expression levels of miR-4787-3p and miR-581 in blood samples from healthy controls (n = 37) and PC patients (n = 37). C, D, G, H Relative expression levels of miR-4787-3p and miR-581 in low-fibrosis (n = 20), and high-fibrosis (n = 17) tissue samples. Data are presented as mean ± SD. Statistical significance was determined by unpaired Student’s t-test (A, B, E, F: Healthy vs. PC) or one-way ANOVA followed by Tukey’s post hoc test (C, D, G, H: across fibrosis grades). ^#P < 0.05, ^##P < 0.01, ^###P < 0.001. vs. Control (Healthy blood or Non-fibrotic tissue) or as indicated by brackets

Correlation analysis of miR-4787-3p and miR-581

Pearson’s correlation coefficient analysis displayed a negative dependence between the levels of miR-4787-3p and fibrosis, differentiation degree, TNM staging, and lymphatic metastasis (r = −0.429, P = 0.000;r = −0.594, P = 0.000; r = −0.594, P = 0.000; r = −0.661, P = 0.000). Similarly, miR-581 levels also showed a negative correlation with fibrosis, differentiation degree, TNM staging, and lymphatic metastasis (r = −0.355, P = 0.001; r = −0.581, P = 0.000; r = −0.581, P = 0.000; r = −0.543, P = 0.000).

Diagnostic value of miR-4787-3p and miR-581 in PC and fibrosis

For the diagnosis of PC, the individual diagnostic performance of miR-4787-3p and miR-581 in blood samples yielded areas under the AUC of 0.899 (95% CI: 0.826–0.972) and 0.853 (95% CI: 0.763–0.943) with sensitivities of 85.00% and 92.50%, respectively. The combined diagnostic sensitivity of miR-4787-3p and miR-581 was 90.00%, with a specificity of 92.50%, and an AUC of 0.938 (95% CI: 0.886–0.990). Internal validation via bootstrapping (1000 repetitions) yielded an optimism-corrected AUC of 0.924 for the combined model in PC diagnosis.

Further analysis based on fibrosis status in pancreatic cancer patients revealed that miR-4787-3p and miR-581 alone had AUC values of 0.802 (95% CI: 0.674–0.930) and 0.786 (95% CI: 0.653–0.919) for diagnosing fibrosis, with sensitivities of 78.30% and 56.50%, respectively. The combined diagnostic sensitivity of miR-4787-3p and miR-581 for pancreatic cancer fibrosis was 87.00%, with a specificity of 76.50%, and an AUC of 0.821 (95% CI: 0.701–0.941). The DeLong test revealed that the combined model showed significantly better diagnostic performance than miR-581 alone (P = 0.038) and a trend toward improvement compared to miR-4787-3p alone (P = 0.062). The combined AUC remained robust upon internal bootstrapping validation (optimism-corrected AUC = 0.798) (Fig. 4).

Fig. 4.

Diagnostic Performance of miR-4787-3p and miR-581. A Receiver operating characteristic (ROC) curves for miR-4787-3p, miR-581, and their combination in discriminating PC patients (n = 37) from healthy controls (n = 37). B ROC curves for miR-4787-3p, miR-581, and their combination in discriminating high-fibrosis PC patients (n = 17) from low-fibrosis PC patients (n = 20). The area under the curve (AUC) with 95% confidence interval (calculated by DeLong method) is shown for each marker. Statistical comparisons between ROC curves were performed using the DeLong test. The optimal cut-off value for each curve was determined by maximizing Youden’s index

Discussion

PC continues to be among the most aggressive and deadly types of cancer. While numerous miRNAs, such as the well-established oncomiR miR-21 or the hypoxia-responsive miR-210, have been studied in PC context, their specific and direct connections to stromal fibrosis remain less delineated. Our study advances this field by identifying and characterizing two miRNAs, miR-4787-3p and miR-581, whose expression is profoundly suppressed in PC and exhibits a strong inverse correlation with fibrotic activity within the TME. This distinct association with fibrosis, coupled with their prognostic value and high diagnostic accuracy, sets them apart from many previously reported miRNAs and underscores their potential novelty as biomarkers and regulatory nodes in PC stroma [20]. The dense stroma within the TME has been identified as a major factor underlying the treatment resistance observed in pancreatic cancer. This stromal environment not only results in a hypovascular state but also impedes the effective delivery of therapeutic agents and immune cell infiltration, significantly reducing the efficacy of treatments. The CAFs, as a key element of the TME, accelerate tumor development, metastasis, and drug resistance by secreting various fibrogenic factors. Fibrosis is vital to the pathogenesis of PC, though the precise mechanisms remain incompletely understood [21–24]. Therefore, exploring the cellular mechanisms involved in PC fibrosis and identifying potential biomarkers are of paramount importance for improving early diagnosis and therapeutic strategies. Although the initial identification of miR-4787-3p and miR-581 as downregulated in PC was derived from the TCGA database, which has a small control group, several measures were taken to ensure the validity of this finding. First, the bioinformatic analysis utilized a multi-algorithm consensus to minimize false discoveries. More critically, the significant downregulation of both miRNAs was consistently replicated in our independent clinical cohort comprising patient tissues and blood samples, as well as in an in vitro model of pancreatic cancer fibrosis. This convergent evidence from independent datasets and experimental models strongly corroborates the initial TCGA-based observation and underscores the association of these miRNAs with PC and its fibrotic microenvironment.

In this study, TCGA-based differential expression analysis revealed that miR-4787-3p and miR-581 were markedly downregulated in pancreatic cancer (PC), and their reduced expression was significantly associated with poorer patient survival. Consistently, their suppression correlated with enhanced fibrotic responses in a CAF co-culture model, suggesting that both miRNAs may function as key regulators of stromal remodeling and warrant further investigation as potential biomarker candidates for PC diagnosis and stratification. The functional roles of miRNAs can be highly context-dependent, varying across cancer types. For instance, miR-4787-3p has been reported as upregulated and oncogenic in ovarian and basal-like breast cancers [25, 26]. In stark contrast, our findings establish its significant downregulation in pancreatic cancer, associating its loss with a pro-fibrotic stroma and poor prognosis. This contrasting role highlights the tissue-specific nature of miRNA regulation and suggests that in the unique context of pancreatic cancer, miR-4787-3p may normally function to restrain fibrotic or oncogenic pathways, such as the TGF-β signaling prevalent in PC stroma. Similarly, while miR-581 has been studied in hepatocellular and colorectal carcinomas [27–30], its role and mechanism in the profoundly fibrotic and immunosuppressive microenvironment of pancreatic cancer remain largely unexplored. Our data position it as a potential key regulator within the pancreatic tumor stroma. Therefore, the findings in other cancers, while informative for understanding the general functional capacity of these miRNAs, underscore the critical need to investigate their specific roles within the distinct biological context of pancreatic cancer, particularly its hallmark desmoplastic reaction. Our in silico analyses further demonstrated that predicted targets of these miRNAs are significantly enriched in oncogenic pathways fundamental to PC progression, particularly MAPK. Given that TGF-β is a central driver of fibrosis and tumor–stroma interactions [31], this aligns with our observation that miR-4787-3p and miR-581 expression is inversely correlated with TGF-β1 and COL1A1, two markers of fibrotic severity. Enrichment of MAPK-related genes suggests additional mechanisms by which these miRNAs could influence tumor aggressiveness and stromal dynamics [32, 33]. Integrative KEGG, TargetScan, and miRDB analyses identified TGFBR2 and KRAS as high-confidence targets of miR-4787-3p, and SMAD7 and RAC1 as central predicted targets of miR-581. Their concerted downregulation in the PC microenvironment may thus derepress these genes, amplifying TGF-β/Smad and MAPK signaling, enhancing CAF activation, ECM production, cytoskeletal remodeling, and stromal stiffness. Together, these findings support a coherent molecular model whereby loss of miR-4787-3p and miR-581 contributes to excessive collagen deposition and the development of a rigid, tumor-promoting stroma in PC. Planned studies, including luciferase reporter assays, site-directed mutagenesis, and functional in vitro and in vivo models, will validate the direct interactions between these miRNAs and their targets and delineate their impact on tumor–stroma crosstalk. Although requiring further experimental confirmation, this framework provides mechanistic insight into how coordinated miRNA downregulation may drive fibrotic stroma formation and facilitate pancreatic cancer progression.

In our study, an in vitro fibrosis model simulating pancreatic cancer was successfully constructed by co-culturing PC cells with CAFs. The effectiveness of the fibrosis model was validated through the detection of TGF-β1 and COL1A1, which are key fibrotic markers. TGF-β1, a classical fibrotic factor, promotes extracellular matrix deposition in various tumors, and its activation is central to fibrosis progression in pancreatic cancer [9, 34–38]. We observed significantly elevated levels of TGF-β1 in the co-culture system, leading to enhanced deposition of the extracellular matrix and facilitating the expansion and aggressive spread of cancer cells. COL1A1, a marker of fibrosis, was also upregulated in our model, further confirming the occurrence of fibrosis. These results suggest that CAFs, by secreting fibrogenic factors such as TGF-β1, regulate the interaction between tumor cells and the extracellular matrix, thereby driving the fibrosis process in pancreatic cancer. We hypothesize that the downregulation of miR-4787-3p and miR-581 during this process may alter CAF functionality accelerate the fibrosis process of PC, and impact tumor progression. Thus, the low level of miR-4787-3p and miR-581 may play a significant role in promoting fibrosis and chemoresistance in PC.

In the context of PC diagnosis and precision therapy, identifying novel biomarkers holds significant clinical value [39, 40]. Our study confirmed that both miR-4787-3p and miR-581 were significantly down-regulated. Moreover, their combined detection demonstrated high sensitivity and specificity in diagnosing pancreatic cancer and assessing the level of fibrosis. ROC displayed that the combination of miR-4787-3p and miR-581 outperformed the use of either miRNA alone, highlighting the clinical utility of their joint detection in distinguishing pancreatic cancer patients and accurately evaluating the degree of fibrosis.

Analysis of clinical specimens demonstrated that miR-4787-3p and miR-581 expression levels were strongly associated with the extent of pancreatic fibrosis and tumor progression. Lower levels of these miRNAs correlated with more severe fibrosis, advanced disease stage, and poorer overall survival, suggesting their potential utility as both diagnostic biomarkers and prognostic indicators for risk stratification. These observations highlight the clinical relevance of miR-4787-3p and miR-581 and underscore the need for further validation.

Despite these strengths, several limitations should be recognized. First, the clinical cohort was relatively small, and the effective sample size was further reduced when patients were stratified into fibrosis subgroups. Although adequate for detecting the primary associations in this exploratory study, the limited sample size reduces the robustness of multivariate analyses and increases susceptibility to outlier effects. As such, the generalizability of the sensitivity and specificity estimates remains uncertain and requires confirmation in larger, prospective, multi-center cohorts. Second, although major clinicopathological factors such as age and sex were balanced between fibrosis groups, unmeasured confounding factors cannot be fully excluded. Third, the present study did not include functional experiments-such as miRNA overexpression or knockdown-to determine whether miR-4787-3p and miR-581 directly regulate key fibrotic mediators (e.g., TGF-β1, COL1A1) or act indirectly through upstream pathways. Accordingly, the causal nature of their involvement in fibrosis cannot yet be established.

Future research should incorporate mechanistic studies to clarify the regulatory roles of these miRNAs in the fibrotic tumor microenvironment. Experimental validation of predicted target genes using luciferase reporter assays, together with gain- and loss-of-function models in vitro and in vivo, will be critical for determining whether miR-4787-3p and miR-581 exert direct antifibrotic effects. Such investigations will also help define their potential as therapeutic targets in addition to their emerging diagnostic and prognostic value.

In conclusion, this study identifies miR-4787-3p and miR-581 as promising biomarkers closely associated with pancreatic cancer fibrosis and disease progression. These findings enhance our understanding of the molecular mechanisms underpinning stromal fibrosis in pancreatic cancer and provide preliminary evidence supporting their clinical applicability. Rigorous mechanistic validation and evaluation in large-scale, multi-center prospective cohorts will be essential steps toward translating these miRNAs into clinical practice.

Author contributions

Conception: CP, PM, SZC, ZXL, ZC; Interpretation or analysis of data: ZXL, HGL. Preparation of the paper: ZYX, CZY; Revision for important intellectual content: ZYX, WXM and ZC.

Funding statement

Key Programs of Wannan Medical College (WK2024ZZD36, WK2024ZZD37), The 2022 Climbing Scientific Peak Project for Talents, the Second Affiliated Hospital of Wannan Medical College(No. DFJH2022010, No. DFJH2022012), Clinical and Translational Research Project of Anhui Province (No.202427b10020066, No. 202427b10020049,No. 202427b10020032).

Data availability

The transcriptomic data used for the initial screening of differentially expressed miRNAs are publicly available from The Cancer Genome Atlas (TCGA) repository (https://portal.gdc.cancer.gov/). The de-identified clinical data and the miRNA expression dataset generated during the current study are not yet publicly available because they are part of an ongoing research program. However, the data are available from the corresponding authors upon reasonable request, and the authors are committed to responding to such requests and providing the data in a manner that supports verification while the broader research project is ongoing. The complete dataset will be deposited in a public repository upon completion of the overall research project.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of the Second Affiliated Hospital of Wannan Medical College (Approval No. WYEFYLS2024127) and conducted in accordance with the Declaration of Helsinki.  This study was retrospective in nature. Since the patients have signed the broad consent form,The Ethics Committee of the Second Affiliated Hospital of Wannan Medical College waived the requirement for informed consent in this study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.