REG4 serves as a prognostic biomarker for pancreatic cancer with long-standing diabetes mellitus by modulating chemoresistance

Highlights

• •REG4 is enriched in classical-subtype PDAC tumor cells in diabetic patients.
• •Serum REG4 predicts poor survival and FOLFIRINOX resistance in long-standing DM only.
• •Chronic hyperglycemia induces REG4 via WNT/β-catenin signaling in PDAC organoids.
• •REG4 suppresses apoptosis and drives chemoresistance in long-standing diabetes-PDAC.
• •REG4 is a candidate serum biomarker for stratifying diabetes-associated PDAC.

Abstract

Background

Pancreatic ductal adenocarcinoma (PDAC) in patients with diabetes mellitus (DM) represents a clinically heterogeneous subgroup, yet biomarkers that reflect diabetes-associated tumor biology and chemotherapy response remain limited. In particular, the influence of diabetes duration on treatment resistance in PDAC is poorly understood.

Methods

We performed an integrated translational analysis combining reanalysis of public single-cell RNA sequencing (scRNA-seq) datasets, clinical serum biomarker profiling, and functional validation using pancreatic cancer cell lines and patient-derived organoids. Circulating REG4 concentrations were measured in independent PDAC cohorts and correlated with diabetes duration, overall survival, and response to FOLFIRINOX. Functional relevance was assessed under diabetes-mimicking hyperglycemic conditions.

Results

Single-cell transcriptomic analysis demonstrated enrichment of REG4-expressing tumor cells within the classical PDAC subtype specifically in diabetic patients. Clinically, circulating REG4 concentrations were significantly elevated in PDAC patients with long-standing diabetes, and high REG4 levels were associated with poor overall survival and resistance to FOLFIRINOX exclusively in this subgroup. In contrast, no prognostic association was observed in non-diabetic or new-onset diabetic patients. In patient-derived organoids and pancreatic cancer cell lines, chronic glucose exposure induced REG4 expression, activation of WNT/β-catenin signaling, suppression of apoptotic pathways, and increased resistance to FOLFIRINOX, recapitulating key clinical features of long-standing diabetes-associated PDAC.

Conclusions

These findings suggest REG4 as a candidate diabetes duration–dependent prognostic and predictive biomarker in pancreatic cancer, warranting validation in larger prospective cohorts. By linking metabolic context to chemotherapy resistance through REG4-associated signaling, this study provides a translational framework for patient stratification and treatment optimization in diabetes-associated PDAC.

Introduction

Pancreatic cancer (PC) is notorious for its poor prognosis, primarily due to the lack of effective biomarkers for early detection and prognosis determination [1]. Despite recent advancements in treatment modalities, the 5-year survival rate for pancreatic cancer has remained relatively unchanged over the past few decades, increasing only from 5.26% to 10% globally [2]. While various clinical parameters and laboratory results have been suggested to help stratify prognosis in PC treatment, the neoadjuvant or adjuvant chemotherapy regimen known as FOLFIRINOX—comprising leucovorin (folinic acid), fluorouracil (5-FU), irinotecan, and oxaliplatin—has become the standard treatment for patients with advanced pancreatic cancer [3]. FOLFIRINOX has also demonstrated efficacy in tumor downstaging in the neoadjuvant setting, enabling some initially unresectable tumors to become resectable [4]. However, the variability in patient responses to FOLFIRINOX and the potential for chemoresistance complicates treatment strategies across the general PC patient population [5]. Recent studies have further elucidated adaptive signaling mechanisms underlying chemoresistance in PDAC, including chemo-induced pathway rewiring and molecular subtype-dependent resistance mechanisms [[6], [7], [8]]. As a result, there is a strong need for biomarkers that can predict or provide prognostic insights into a patient’s sensitivity to chemotherapy.

Importantly, while the clinical diagnosis of diabetes mellitus (DM) is straightforward, identifying which PDAC patients with diabetes harbor tumors with diabetes-driven molecular alterations—particularly those associated with chemoresistance—remains an unmet need. Not all PDAC patients with DM share the same tumor biology or treatment response, and a serum-accessible biomarker that reflects the biological consequences of prolonged hyperglycemic exposure could enable more precise patient stratification beyond DM diagnosis alone. Interestingly, emerging epidemiological evidence linking DM to PC has opened up new opportunities for discovering molecular biomarkers [9]. Pancreatic cancer with diabetes mellitus (PCDM) has been observed in many cases within three years prior to the diagnosis of PC. Studies have also shown that the duration of DM in PCDM plays an important role, with new-onset DM (NODM) being associated with a higher risk of PC diagnosis compared to long-standing DM (LSDM) [10,11]. While NODM has been linked to an increased risk of PC development in the general population, the prognostic value of NODM and LSDM in PCDM remains unexplored in both epidemiological and experimental settings.

Among the few candidate molecular biomarkers for PC prognosis, Regeneration Gene 4 (REG4), a member of the calcium-dependent lectin gene superfamily, is abnormally expressed in gastrointestinal (GI) cancers and is involved in promoting tumor aggressiveness, chemoresistance, and cancer stem cell characteristics [12,13]. REG4 has recently been proposed as a potential diagnostic and prognostic biomarker, despite the complex underlying mechanisms contributing to frequent resistance to 5-FU-based chemotherapy [14]. Although intriguing findings have suggested that REG4 may serve as an independent prognostic factor, its overexpression in PC tissues provides an opportunity to reposition this molecular biomarker for optimized use in specific clinical scenarios.

This study investigates the complex prognostic value of REG4 in PCDM using patient cohort analysis, single-cell RNA sequencing (scRNA-seq) analysis, and organoid models. Our results reveal the differential prognostic significance of REG4 in PCDM, particularly regarding overall survival and chemoresistance to FOLFIRINOX. Furthermore, modulation of REG4 under DM-mimicking conditions aggravates chemosensitivity to FOLFIRINOX in PCDM patient-derived pancreatic cancer organoids, highlighting REG4 as a potential therapeutic target to overcome chemoresistance in the classical subtype of DM pancreatic cancer.

Materials and methods

Patient cohorts

Two independent pancreatic cancer (PC) patient cohorts were examined in this study. For cohort 1, biobank serum samples from 99 PC patients were used to measure ELISA levels, correlating with DM history and survival profiles. Cohort 2 consisted of 10 PC patients with detailed clinical information, including diabetes mellitus (DM) history, laboratory results, and PC diagnosis. Patient-derived PC organoids were available for analysis in this cohort. The study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of Yonsei University Health System (IRB no. 4-2021-1383, 4-2021-1395).

Definition of clinical parameters

Type 2 diabetes mellitus was defined using laboratory findings such as HbA1c and fasting glucose, interpreted according to the American Diabetes Association guidelines. To distinguish between NODM and LSDM, we used ICD-10 code E11 or self-reported diabetes confirmed by the period of anti-diabetic medication usage. Patients diagnosed with DM within three years before PC diagnosis were classified as NODM, and those with DM for more than three years as LSDM. This 3-year cutoff was adopted based on prior epidemiological evidence demonstrating that the risk of pancreatic cancer is markedly elevated within the first three years of diabetes onset, consistent with published population-based cohort studies [10,11]. All patients in this study were classified as type 2 DM based on laboratory criteria (HbA1c ≥6.5% or fasting glucose ≥126 mg/dL) per American Diabetes Association guidelines. Information on anti-diabetic medication use and HbA1c levels was collected where available. Overall survival (OS) was determined based on the date of death or last follow-up. Chemotherapy response was assessed using RECIST v1.1 criteria, where a good response included partial response (PR) or stable disease (SD) during treatment, and poor response was defined as progressive disease (PD) or any recurrence or metastasis during follow-up.

Cell and organoid culture

The AsPC-1 pancreatic ductal adenocarcinoma (PDAC) cell line, obtained from the Korean Cell Line Bank, was cultured in RPMI 1640 medium with 2.05 mM L-glutamine, 25 mM HEPES, 15 mg/L L-methionine, 2000 mg/L sodium bicarbonate, and 5 mg/L phenol red (Welgene), supplemented with 10% fetal bovine serum (FBS) and 1% antibiotic (penicillin/streptomycin) (Gibco). Cells were grown at 37°C in a 5% CO2 incubator.

Pancreatic cancer organoids were isolated and cultured as previously described [15]. Organoids were grown in AdDMEM/F12 medium supplemented with GlutaMAX, penicillin/streptomycin, B27, N-acetyl-L-cysteine, Wnt3a-conditioned medium, RSPO1-conditioned medium, recombinant human noggin, recombinant human EGF, gastrin, recombinant human FGF10, nicotinamide, and A83-01. Organoids were incubated for 48 hours for RNA collection or cytotoxicity assays.

Diabetes-Mimicking glucose stimulus

To simulate diabetic microenvironments, AsPC-1 cells were cultured in three media types: standard glucose (11 mM), high glucose (25 mM), and very high glucose (50 mM). Cells were exposed to short-term (2 cycles) and long-term (6 cycles) treatments. Patient-derived organoids were exposed to high glucose (50 mM) for short-term (3 days) and long-term (6 days) treatments.

Chemical reagents and immunofluorescence staining

Oxaliplatin, irinotecan hydrochloride, fluorouracil, folinic acid calcium salt hydrate, and dimethyl sulfoxide (DMSO) were purchased from Sigma-Aldrich. For immunofluorescence staining, pancreatic cancer organoids and cells were cultured in 8-well plates and fixed with cold 4% paraformaldehyde. After washing and blocking with 5% BSA and 0.1% Triton X-100, primary antibodies and secondary antibodies were applied. Antibody details are as follows: REG4 (R&D Systems #AF1379, goat polyclonal, 1:200), CD44 (Biogems #065111-20, rat monoclonal, 1:100), β-catenin (Proteintech #51067-2-AP, rabbit polyclonal, 1:200), and GAPDH (Cell Signaling Technology #5174, rabbit monoclonal, 1:1000). Secondary antibodies used were Alexa Fluor 488 anti-goat IgG, Alexa Fluor 555 anti-rat IgG, Alexa Fluor 647 anti-rabbit IgG (all Thermo Fisher Scientific, 1:500), and Goat anti-rabbit IgG-HRP (GeneTex #GTX213110-01,1:1000). Fluorescence images were obtained using a Zeiss LSM 700 laser scanning confocal microscope and analyzed with ImageJ.

Cytotoxicity assay and IC50 calculation

The concentrations of FOLFIRINOX regimen used for in vitro experiments refer to the clinical practice, thus we set the 1x concentration of each component as oxaliplatin 0.37µM (in DMSO), irinotecan 4.6µM (in DMSO), folinic acid 5.0µM (in H₂O), and 5-fluorouracil 38.0µM (in DMSO). A combination of chemotherapy regimen FOLFIRINOX was serially diluted starting at 100x concentration, oxaliplatin (37.8μM), irinotecan (468.4μM), leucovorin (501.3μM), and fluorouracil (3804.9μM) respectively. These four drugs were 10-fold serial dilutions in pancreatic ductal adenocarcinoma cancer AsPC-1 cell lines. 5 × 10³/well cells were initially seeded into 96-well plates, and the culture medium was removed and replaced with FOLFIRINOX-containing medium after 24h. Cell viability was accessed using Cell Counting Kit-8 (CCK-8, also known as WST-8) (Sigma-Aldrich #96992, Inc., St. Louis, MO, USA) by adding 10μL of CCK-8 into each well after incubation with FOLFIRINOX for 72h. After incubation at 37°C for 3h, the plates were read at 450nm/650nm. For the organoid cell viability assay, 100 μL of CellTiter-Glo® 3D (Promega #G9681) was added to each well after incubation with FOLFIRINOX treatment. According to the manufacturer’s instructions, organoids were mixed by shaking and incubated at room temperature for 30 minutes. Luminescence was measured using a VARIOSKAN LUX reader (Thermo Fisher Scientific). The cell IC50, IC70, and IC80 values were calculated using GraphPad Prism by linear approximation regression of the percentage survival versus the drug concentration (GraphPad Software, Inc.).

Reverse-transcription (RT-PCR) and quantitative PCR analysis (qPCR)

Total RNA was extracted from PC organoids cultured at passage 5, using the RNeasy micro kits (Qiagen #74004, QIAGEN GmbH, Verogen, Inc.) according to the manufacturer’s protocol. To reverse-transcribe the RNA into cDNA, purified RNA samples were combined with the RNA to cDNA EcoDry premix (Takara Bio Inc. #639549). Mixtures were incubated at 42°C for 1 h, then at 70°C for 10 min.

The Applied Biosystem StepOne System (Applied Biosystems) was used to perform qPCR. The real-time PCR reaction was measured by detecting the binding of fluorescent SYBR Green dye to double-stranded DNA. For PCR amplification, the total reaction volume was adjusted to 20 μL with RNase-free water after mixing with 100 ng cDNA, 2 μL primer sets, 10 μL 2 × SYBR premix Ex Taq, and 0.4 μL 50 × ROX reference dye (Takara #RR420L). Amplification was performed under the following cycling conditions: 95°C for 15 minutes, followed by 40 cycles of 95°C for 15 seconds, and 60°C for 40 seconds. Analyses were performed in triplicate for each cDNA. Relative mRNA gene expression was normalized with the housekeeping gene GAPDH, and the ΔC_t value was calculated as follows: ΔC_t = C_t (target gene) - C_t (housekeeping gene). The average ΔC_t was then subtracted from each experimental condition described to yield the ΔΔC_t value. The fold changes in gene expression were calculated as 2^-ΔΔCt relative to control samples after normalization to housekeeping gene. The primers used are listed in Supplementary Table 1.

Bulk RNA-seq of PC organoids

From patient-derived PC organoids, total mRNA was extracted by the standard protocol. We used 100 ng total RNA from all subjects to prepare sequencing libraries with by using the TruSeq stranded mRNA sample preparation kit (Illumina). Quality of these cDNA libraries was evaluated with the Agilent 2100 BioAnalyzer (Agilent). They were quantified with the KAPA library quantification kit (Kapa Biosystems) according to the manufacturer’s library quantification protocol. Following cluster amplification of denatured templates, sequencing was progressed as paired-end (2 × 150 bp) using Illumina NovaSeq6000 platform. For the bioinformatics pipeline, gene expression level was measured with Cufflinks v2.1.1 using the gene annotation database of Ensembl release 77. Non-coding gene region was removed with –mask option. To improve the accuracy of measurement, multi-read-correction and fragbias-correct options were applied. ‘-max-bundle-frags’ option was set to 10,000,000 to estimate the highly expressed genes. All other options were set to default values. Differentially expressed genes were identified using Cuffdiff tool with default parameter setting with a significance of p-value < 0.05

Single cell RNA-seq

For scRNA-seq analyses, publically available dataset (the Genome Sequence Archive PRJCA001063) was used with available clinical dataset annotations provided by Peng et al. in the previous study [16]. Cellranger (v6.0.2) software was used to perform read trimming and alignment through the GRCh38-2020 reference genome. For filtering, normalization, and clustering, we used the standard analysis pipeline in the R package Seurat (v4.3.0.1). Briefly, we filtered cells that expressed fewer than 200 genes, more than 8000 genes, or more than 10% mitochondrial genes. To investigate sequential changes of PDAC cells in diabetic patients, we performed a trajectory analysis in the PDAC cluster using the R package Monocle (v2.22.0). Wilcoxon rank-sum test analysis was used to identify differentially expressed genes in REG4 high and REG4 low cell populations with statistical significances. These genes were analyzed for gene ontology enrichment using the Database for Annotation, Visualization and Integrated Discovery (DAVID). The basal and classical scores in the PDAC subset were calculated using gene sets for each subtype [17] through AddModuleScore of Seurat package.

TCGA dataset analysis

The Pancreatic Adenocarcinoma (PAAD) dataset of The Cancer Genome Atlas (TCGA) were analyzed using the R package TCGAbiolinks (Version 2.29.6). Overall survival (OS) data from TCGA-PAAD were evaluated using the R packages survival (Version 3.2-13) and survminer (Version 0.4.9). Optimal cut-off points for REG4 expression levels were identified using the Log-Rank statistics from the maxstat package (Version 0.7-25). P-values were calculated using the Log-Rank test using the R packages survival. We identified the basal and classical subtypes from TCGA-PAAD through de novo compartment deconvolution and weight estimation of tumor samples (DECODER) [18].

ELISA assay

Serum samples from the patient and control groups were measured for REG4 concentration using commercial ELISA assay (Sino Biological #SEK11186). The verification of linearity within assay range of 3.91-250 pg/mL was performed using patient samples according to CLSI guideline EP09-A3.

Statistical data analysis

The results of multiple experiments are presented as the mean ± standard error of the mean (SEM). Statistical analysis was performed using Mann-Whitney U test and one-way analysis of variance followed by Tukey’s multiple comparison test as appropriate, using GraphPad Prism 8 (GraphPad Software). p<0.05 was considered statistically significant. For OS estimates were compared using the long-rank test between groups. For survival analyses in the clinical cohort (Cohort 1), the optimal REG4 serum concentration cutoff was determined using the maximally selected Log-Rank statistics approach (maxstat package, Version 0.7-25 in R), consistent with the method applied to the TCGA-PAAD dataset. Subgroup analyses by diabetes duration were pre-specified based on the NODM/LSDM classification defined above.

Results

REG4-expressing classical subtype cancer cells are enriched in diabetes mellitus (DM)

To identify candidate biomarkers reflecting diabetes-associated tumor biology in PDAC, we leveraged a publicly available single-cell RNA sequencing (scRNA-seq) dataset (Peng et al. [16]) that includes paired clinical annotation of diabetes status. Reanalysis of this dataset allowed us to interrogate diabetes-associated transcriptomic differences at single-cell resolution within the tumor epithelial compartment. Comparison of major cell types between the DM and non-DM groups in PDAC tissues showed no distinct differences (Fig. 1A, Supplementary Fig. 1). However, when analyzing only the PDAC cell type, different cell populations were identified according to DM status (Fig. 1B). To examine gene expression differences associated with DM in PDAC, we performed a differential expression gene (DEG) analysis. The analysis revealed that REG4 and FABP1 were the most upregulated genes in the DM group (Fig. 1C, Supplementary Table 2). Notably, REG4 was more highly expressed in cancer cells than in the tumor microenvironment cells (Fig. 1D). In the previous report [17], REG4 is suggested as a classical subtype marker for PDAC.

Fig. 1.

Identification of REG4 as pancreatic cancer with diabetes mellitus (PCDM) biomarker through public scRNAseq database analysis, (A) Reanalysis of publicly available single-cell genome data for pancreatic cancer (CRA001160 from the Genome Sequence Archive database) according to the presence of diabetes. (B) UMAP analysis for the non-DM and DM in the pancreatic cancer cells (PDACs) subset. (C) Volcano plot displaying differential gene expression in pancreatic cancer clusters based on DM status. (D) Expression level of REG4 in the all cell population with and without DM. (E) Basal and classical score in the PDAC subset were analyzed using AddModuleScore of Seurat package. (F) UMAP analysis showing REG4 expression levels in pancreatic cancer clusters with and without DM. (G) REG4 expression levels in basal-like and classical subtypes from the TCGA-PAAD dataset. (H) Survival plot analysis for total and classical subtypes from TCGA-PAAD dataset (n=150). P values were calculated using the log-rank test, **p < 0.01.

To investigate REG4 levels in the basal and classical subtypes, we calculated gene signatures for each subtype within the PDAC subset (Fig. 1E). REG4(+) cells were enriched in the classical subtype within the DM group (Fig. 1F). Within the classical subtype, PC cells could distinctively be divided into REG4^high and REG4^low/- populations (Fig. 1F). Consistent with previous reports, REG4 expression levels were also upregulated in the classical subtype of the TCGA-PAAD dataset. Interestingly, high expression of REG4 was associated with poorer survival specifically in the classical subtype (Fig. 1G, H). In summary, our findings indicate that in PDAC with DM (PCDM), the classical subtype is enriched, and high REG4 expression within this subtype is associated with a poor prognosis.

Circulating REG4 concentrations in pancreatic cancer patients predicts survival only in diabetic patients

Using REG4 ELISA results of the 99 serum samples obtained from the biobank, significantly elevated serum REG4 concentrations in PCDM patients were observed when compared to PC without DM (Fig. 2A). There was a positive correlation between REG4 levels and the severity of diabetes, as measured by HbA1c (Fig. 2B). The difference in REG4 levels between the PCDM and non-DM groups was more pronounced in earlier stages of PDAC (Fig. 2C); however, effect sizes were modest at later stages, and these stage-stratified comparisons are limited by the reduced sample size within each subgroup. Interestingly, no relationship was found between initial tumor marker CA19-9 levels and REG4 concentrations (Supplementary Figure 2A). Although the highest quartile of REG4 expression was associated with poorer overall survival in PCDM patients, the difference was not statistically significant (Fig. 2D). However, in LSDM patients with PCDM, higher REG4 concentrations were significantly associated with worse survival, an effect not seen in NODM or PC without DM patients (Fig. 2E, Supplementary Figure 2B). Furthermore, REG4 levels were elevated in chemotherapy-resistant LSDM patients, while no such difference was observed in NODM or non-DM PC patients (Fig. 2F).

Fig. 2.

Relationship between circulating REG4 concentrations and clinical outcomes in pancreatic cancer patients, (A) Serum REG4 concentrations in pancreatic cancer patients stratified by diabetes status (DM, n=50; non-DM, n=49). (B) Correlation between HbA1c, a diabetes severity indicator, and serum REG4 levels (n=84). (C) Serum REG4 concentrations in DM and non-DM pancreatic cancer patients stratified by pathological stage (Stage 1: DM n=4, non-DM n=5; Stage 2A: DM n=17, non-DM n=29; Stage 2B: DM n=25, non-DM n=13). Statistical comparison by Mann-Whitney U test; p-values indicated above each comparison. Note that subgroup sample sizes limit statistical power at individual stages. (D) Survival curves of pancreatic cancer patients by quartile groups of serum REG4 concentration. (E) Survival curves of high and low REG4 concentration groups based on diabetes duration. (F) Distribution of serum REG4 concentration in new-onset diabetes-associated, long-standing diabetes-associated, and non-diabetes-associated pancreatic cancer groups by chemotherapy response. Statistical comparison by Mann-Whitney U test. *p<0.05, **p<0.01.

Deceleration of apoptotic cell death pathway in pancreatic cancers is associated with diabetes

To investigate the molecular pathways linked to diabetes in pancreatic cancer cells, a pseudotemporal trajectory analysis of PDAC cluster from DM patients according to REG4 expression levels were conducted, revealing PAF1 and UGT2B7 genes correlated with REG4 expression patterns (Fig. 3A). Gene ontology (GO) enrichment analysis of DEGs identified from this trajectory highlighted upregulation of ion channel transport, whereas pathways related to antigen binding and immune response were downregulated as REG4 expression was upregulated within the PDAC cell cluster (Fig. 3B). Based on prior evidence that REG4 promotes anti-apoptotic signaling in gastrointestinal cancers, we performed targeted GSEA for apoptotic and cell growth pathways. This analysis revealed significant downregulation of the regulation of apoptotic signaling pathway and cell growth pathway in REG4-high compared to REG4-low PCDM cells (Fig. 3C, normalized enrichment score = −1.44 and −1.53, respectively; adjusted p < 0.05), suggesting that REG4 overexpression in the diabetic context is associated with suppression of programmed cell death.

Fig. 3.

Association of apoptotic cell death pathway downregulation in pancreatic cancers with diabetes, (A) Expression distribution of differentially expressed genes identified in pancreatic cancer clusters related to REG4 expression, based on pseudotime analysis. (B) Upregulated and downregulated gene ontology pathways in pancreatic cancer clusters based on REG4 expression levels. (C) GSEA plots showing downregulation of the regulation of apoptotic signaling pathway (left) and cell growth pathway (right) in REG4-high versus REG4-low PCDM cells. Negative NES values indicate pathway downregulation in the REG4-high group relative to REG4-low. NES: normalized enrichment score.

Glucose stimulus induces REG4 and chemoresistance via Wnt/ß-catenin signaling in pancreatic cancer

Transcriptomic analysis of patient-derived PC organoids confirmed that DM-derived organoids (including both NODM and LSDM) clustered distinctly from control organoids in principal component analysis (PCA), reflecting a broad DM-associated transcriptomic signature that dominates the primary axes of variation (Fig. 4A, 4B). The transcriptomic variance attributable to DM status versus non-DM was substantially larger than that between NODM and LSDM at baseline, such that LSDM-specific transcriptomic features were not captured as a dominant axis of PCA separation. Rather, the functionally relevant differences between NODM and LSDM organoids emerged under ex vivo glucose stimulation conditions, where LSDM-derived organoids exhibited significantly greater WNT1 induction compared to NODM-derived organoids (Fig. 4D), consistent with the LSDM-specific clinical associations observed in Fig. 2. GO terms related to Wnt signaling pathway were upregulated in DM patient-derived organoids compared to controls (Fig. 4C). Given the established role of WNT/β-catenin signaling in cancer stem cell maintenance and chemotherapy resistance in PDAC, this finding suggests that hyperglycemia may promote a pro-resistance transcriptomic state in DM organoids. Additionally, pathways related to chemotaxis and granulocyte migration were downregulated, potentially reflecting reduced chemokine-mediated stromal interactions in the DM context, though the direct relevance of this to chemoresistance requires further investigation. Among the Wnt signaling ligands, WNT1 was consistently upregulated under both short-term and long-term high glucose conditions, with a more pronounced effect in organoids derived from LSDM patient compared to those from NODM patient (Fig. 4D, Supplementary Figure 3). Moreover, WNT1 was over-expressed in PCDM surgical samples compared to non-DM PC samples localizing with REG4 (Fig. 4E, 4F).

Fig. 4.

Transcriptomic analysis of patient-derived pancreatic cancer organoids and validation of Wnt1 expression in organoids and surgically resected samples, (A) REG4 mRNA expression levels of pancreatic cancer organoids from control and diabetes patients. (B) PCA analysis of RNA-seq results for organoids derived from new-onset diabetes-associated, long-standing diabetes-associated, and non-diabetes-associated pancreatic cancer. (C) Gene ontology analysis of increased and decreased pathways in differentially expressed genes between control and diabetes-associated pancreatic cancer organoids. (D) Expression levels of WNT1 in organoids derived from control, NODM, and LSDM pancreatic cancer patients according to ex vivo glucose stimulus. (E) Expression levels of WNT1 in the surgically resected pancreatic cancer tissues from non-DM and PCDM patient groups. (F) Expression patterns of WNT1 and REG4 in the surgically resected pancreatic cancer tissues from the PCDM patient. C: Control, S: Short stimulus of glucose, L: Long stimulus of glucose. Scale bar=50㎛. P values were estimated using one-way ANOVA, *p < 0.05, **p < 0.01.

Furthermore, prolonged glucose exposure significantly increased the expression of REG4 and CD44 both in AsPC-1 cell line and PC organoids (Fig. 5A, 5B, Supplementary Figure 4A). Additionally, the IC50 values and cell viability after FOLFIRINOX treatment in AsPC-1 cell line and PC organoids were consistently increased especially in long-standing diabetic conditions (Fig. 5C, 5D, Supplementary Figure 4B), indicating that long-term hyperglycemia promotes chemoresistance and affects apoptosis. Importantly, β-catenin expression was significantly translocated to the nucleus under prolonged high glucose exposure (Fig. 5E), suggesting that resistance to apoptosis in long-term hyperglycemic environment is mediated by the WNT1/β-catenin signaling pathway. In summary, in vitro glucose stimulus study suggests that prolonged high levels of REG4 in PCDM may influence overall chemosensitivity by affecting the maintenance of drug-resistance cancer cell populations, thereby inducing minimal cell death and leading to poor outcomes (Fig. 5F).

Fig. 5.

Chemoresistance via Wnt/ß-catenin signaling in pancreatic cancer organoid and cell line according to glucose stimulus. (A, B) Expression levels of REG4 and CD44 proteins in the AsPC-1 cell line (A) and organoids from LSDM patient (B) based on glucose stimulation status and duration. (C) Increased IC50 values for FOLFIRINOX treatment on AsPC-1 cell line according to ex vivo glucose stimulus dose levels and durations. (D) Increased resistance to continuous FOLFIRINOX treatment in DM patient-derived organoids. (E) Effect of high glucose condition on β-catenin expression. Long duration of high glucose exposure induces the translocation of β-catenin from the plasma membrane to the nucleus. (F) Schematic illustration of hypothesized mechanism. In pancreatic cancer cells from DM patients with prolonged exposure, markedly elevated REG4 levels exhibit increased WNT1 expression. Activated WNT signaling causes CD44 to bind to secreted REG4, resulting in enhanced anti-apoptotic effects. In contrast, pancreatic cancer cells from non-DM patients respond to FOLFIRINOX and undergo apoptosis. C: Control, S: Short stimulus of glucose, L: Long stimulus of glucose. Scale bar=100㎛. P values were estimated using one-way ANOVA, *p < 0.05, **p < 0.01, ***p < 0.001.

DISCUSSION

Although various underlying mechanisms for chemotherapy resistance in pancreatic cancer have been proposed through diverse experimental designs, identifying a key biomarker for specific cases, such as PDAC with diabetes, remains challenging. In this study, single-cell RNA sequencing identified REG4 as a potential biomarker, which was subsequently validated using clinical samples. Previous studies have reported that Regenerating (REG) proteins, including Reg1A, 1B, 3A, 3G, and 4, play critical roles in pancreatic islet regeneration and PDAC development [13,19]. Despite high expression of REG4 in PDAC tissues, its diagnostic potential has been limited by the heterogeneity of control groups [14]. Notably, REG4 has been shown to interact with CD44, promoting the proliferation and stemness of pancreatic cancer cells via EGFR/Akt pathways [13,20]. As a biomarker for tumorigenesis and poor prognosis in pancreatic cancer, REG4 needs to be examined within the specific context of complex PDAC microenvironment.

An intriguing finding in our study is that REG4 is preferentially enriched in the classical PDAC subtype, yet high REG4 expression within this subtype is associated with poor survival. While the classical subtype is generally considered to carry a more favorable prognosis than the basal-like subtype, our data suggest that REG4 expression may define a clinically aggressive subset within the classical category, potentially driven by hyperglycemia-induced WNT/β-catenin activation. It is possible that REG4-high classical PDAC cells represent a transitional state with features of transcriptional plasticity, and future studies examining basal-like gene signature scores within the REG4-high classical subpopulation would be informative.

Interestingly, REG4 was found to be upregulated in PDAC cases with DM, potentially linked to increased chemosensitivity of pancreatic cancer cells to FOLFIRINOX [21,22]. While association of REG4 expression with diabetes has been rarely reported [23,24], aberrant REG4 expression is associated with the inhibition of apoptosis in pancreatic cancer cells, contributing to a poor clinical prognosis in LSDM patients. Moreover, transcriptomic analysis and RNA validation from patient-derived organoids revealed significant upregulation of WNT1 in LSDM patient organoids, a factor known to contribute to chemoresistance [25]. These findings are consistent with emerging evidence that PDAC chemoresistance is mediated by adaptive activation of oncogenic signaling pathways, as recently demonstrated in multiple experimental and clinical settings [[6], [7], [8]]. Taken together, our results suggest that REG4, in LSDM patients, may mediate chemoresistance through the WNT/ß-catenin signaling pathway.

Since REG4 is continuously suggested as an important driver of the neoplastic process [26,27], further investigation into its influence on tumor growth and treatment response is warranted. Recent studies show that REG4 interacts with CD44, promoting the release of CD44ICD and the expression of Klf4 and Sox2, which are critical for cancer stem cell pluripotency [13,20]. Additionally, increased WNT1 expression has been linked to elevated cancer stem cell characteristics, including CD44 expression, in cancer models [28]. With previous studies demonstrating that REG4 silencing reduces stem-like properties and increases sensitivity to chemoradiation-induced cell death [29], our findings support REG4 as a strong biomarker for chemoresistance, particularly within the PCDM context. Furthermore, our pseudotime analysis data also strongly support that REG4 is crucial in maintaining the tumorigenic properties of the favorable microenvironment accompanying upregulation of PAF1 [30] and UGT2B7 [31]. Overall, our study sheds light on REG4-mediated pathways that contribute to chemoresistance in pancreatic cancer patients with diabetes.

Although clinical studies have examined the differential roles of NODM versus LSDM in PDAC development [10,32], a specific molecular biomarker has yet to be identified, owing to the complexity of these conditions. Recently developed indices, such as ENDPAC and GLMI scores [33,34], should be validated against REG4 expression levels. Among diabetes-associated cancers, pancreatic cancer uniquely presents a heightened risk within three years of diabetes onset, likely tied to patterns of aggravated insulin resistance. Consistent with our finding of high REG4 expression linked to insulin resistance, diabetes duration and impaired fasting glucose have been identified as key factors for increased pancreatic cancer risk [35].

While researchers have explored REG4 inhibition through monoclonal antibodies and genetic knockdowns in various gastrointestinal cancers [12], these strategies have not been widely studied in pancreatic cancer. Additional studies on REG4 knockdown or knockout are necessary to clarify the interactions among WNT1, CD44, and REG4 in pancreatic cancer. The limitation of a small number sample size, which does not fully reflect the variability in patient status, also needed for further studies. Specifically, the small size of the organoid cohort (n=10) reflects the technical challenges inherent in patient-derived pancreatic organoid establishment, and limits the statistical power of transcriptomic and functional analyses in Cohort 2. The subgroup survival analyses in Cohort 1 are similarly constrained by sample size after stratification, and should be regarded as exploratory findings requiring validation in larger prospective cohorts. Additionally, complete data on potential confounding variables—including BMI, detailed anti-diabetic medication history, glycemic control trajectories, and history of pancreatitis—were not uniformly available across all patients in Cohort 1, which precludes formal multivariate adjustment for these metabolic factors. Future prospective studies incorporating comprehensive metabolic profiling will be necessary to fully delineate the independent contribution of REG4 to outcomes in diabetes-associated PDAC. The current study also lacks an independent external validation cohort for serum REG4, and the clinical endpoint analyses are limited to overall survival and binary chemotherapy response (PR/SD vs. PD) due to incomplete progression-free survival data in the biobank cohort. Formal multivariate Cox regression analysis was not performed owing to the limited number of events following subgroup stratification. These analyses, together with external validation in a prospective multicenter cohort, are essential to establish REG4 as an independent predictive biomarker beyond established clinical factors and are planned as a next step.

The in vitro glucose stimulation model used in this study has several inherent limitations. Supraphysiologic glucose concentrations (25–50 mM) do not fully recapitulate the complexity of the diabetic tumor microenvironment, which involves not only hyperglycemia but also hyperinsulinemia, dyslipidemia, and chronic inflammation. Osmolality was not formally matched across glucose conditions, and the potential contribution of hyperosmotic stress to the observed phenotypes cannot be excluded. However, we observed that different glucose conditions did not significantly affect baseline proliferation rates prior to FOLIFIRINOX treatment, suggesting that glucose-abundant microenvironment could negatively affect the anticancer response in vitro. Future studies should incorporate osmolality-matched controls and more physiologically relevant models of hyperglycemia. Furthermore, while our data demonstrate activation of WNT/β-catenin signaling via nuclear β-catenin translocation and WNT1 upregulation in hyperglycemic conditions, the precise mechanistic link between REG4 and this pathway—and whether β-catenin inhibition can reverse REG4-associated chemoresistance—remains to be formally demonstrated. Dedicated apoptosis assays such as Annexin V/PI staining or caspase activity measurements were not performed in this study, although apoptotic suppression was inferred from transcriptomic pathway analyses and cell viability data. These mechanistic questions represent important directions for future investigation.

Despite these limitations, to the best of our knowledge, our findings offer the first experimental evidence of a potential chemoresistance mechanism underlying the poor prognosis of LSDM in pancreatic cancer. As chemotherapeutic regimens including FOLFIRINOX are modified and optimized based on molecular subtypes [36], correctly identifying target patient populations for therapies involving REG4 modulation is crucial. In conclusion, our findings suggest that manipulating REG4 expression in pancreatic cancer cells exposed to long-term high glucose may help mitigate chemoresistance and improve outcomes for pancreatic cancer patients with diabetes.

Conclusion

In conclusion, our study demonstrates that REG4 is preferentially enriched in the pancreatic cancer cells with diabetes mellitus and is closely associated with poor prognosis and resistance to FOLFIRINOX in patients with LSDM. Through integrated single-cell transcriptomic analysis, serum biomarker profiling, and patient-derived organoid validation, we showed that chronic hyperglycemic exposure promotes REG4 expression and activates WNT/β-catenin-associated signaling pathways linked to chemoresistance. These findings suggest that REG4 may serve as a clinically accessible prognostic and predictive biomarker for diabetes-associated pancreatic cancer and provide a potential therapeutic framework for overcoming chemotherapy resistance in this clinically challenging subgroup of PDAC patients.

Funding

This work was supported by the National Research Foundation of Korea (NRF) grants funded by the Korea government (MSIT) (No. 2022R1A2C1008868, 2022R1A2C1013380) and by the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health& Welfare, Republic of Korea (RS-2025–25459146). This study was also supported by a faculty research grant of Yonsei University College of Medicine (6-2023-0123, 2024-32-0076).

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of Yonsei University Health System (IRB no. 4-2021-1383, 4-2021-1395). The biospecimens were provided by a human biobank, and the requirement for informed consent was waived by the Institutional Review Board. Written informed consent for the use of organoids derived from patient specimens was obtained from each participant.

CRediT authorship contribution statement

Jae-Il Choi: Writing – original draft, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Hee Jung Park: Visualization, Methodology, Investigation, Formal analysis. Hak Park: Methodology. Yonggeun Cho: Investigation. Hyo Shik Shin: Methodology. Young Ik Koh: Methodology. See Young Lee: Resources. Sung Ill Jang: Resources. Jae Hee Cho: Resources. Hyung Sun Kim: Resources. Ho Kyoung Hwang: Resources. John Hoon Rim: Writing – original draft, Supervision, Methodology, Investigation, Funding acquisition, Formal analysis, Conceptualization. Jong-Baeck Lim: Writing – review & editing, Supervision, Resources, Funding acquisition, Conceptualization.

Declaration of competing interest

All authors (Jae-Il Choi, Hee Jung Park, Hak Park, Yonggeun Cho, Young Ik Koh, Hyo Shik Shin, See Young Lee, Sung Ill Jang, Jae Hee Cho, Hyung Sun Kim, Ho Kyoung Hwang, John Hoon Rim, Jong-Baeck) declare none of competing interests.

Appendix. Supplementary materials

Supplementary Figure 1. Proportions of subcellular clusters according to diabetes mellitus status.

Supplementary Figure 2. (A) Correlation between circulating REG4 concentrations and CA19-9 levels at PDAC diagnosis. (B) Survival curves of high and low REG4 concentration groups in PDAC patients without diabetes mellitus.

Supplementary Figure 3. Expression levels of WNT family in AsPC-1 cell line according to ex vivo glucose stimulus. C: Control, S: Short stimulus of glucose, L: Long stimulus of glucose. P values were estimated using one-way ANOVA, *p < 0.05

Supplementary Figure 4. (A) Expression levels of REG4 and CD44 proteins in the AsPC-1 cell line based on glucose stimulation status and duration. (B) Viability of the AsPC-1 cell line by FOLFIRINOX treatment according to ex vivo glucose stimulus. C: Control, S: Short stimulus of glucose, L: Long stimulus of glucose.

Supplementary Table 1. PCR primer sequences for all genes analyzed by qPCR

Supplementary Table 2. Differentially expressed gene list according to diabetes mellitus states