In the present study, three human PC cell lines: MIAPaCa-2 (RRID: CVCL_0428) purchased from the Japan Cancer Research Resources Bank, Panc-1 (RRID: CVCL_0480) purchased from the American Type Culture Collection and PSN-1 (RRID: CVCL_1644) obtained from the European Collection of Authenticated Cell Culture. All cell lines were authenticated using STR profiling within the last 3 years and routinely tested for mycoplasma contamination. The cells were cultured at 37°C in 5% CO₂ atmosphere with at least 95% humidity in DMEM supplemented with 10% FBS and 100 U/ml penicillin and streptomycin each (Thermo Fisher Scientific, Inc.).

Gemcitabine-resistant (GR) cell lines were generated by exposure to gradually increasing concentrations of the drug for 2 months as previously described (24). Parental MIAPaCa-2 cells were exposed to gemcitabine at an initial concentration of 1 ng/ml. When the cells adapted to the drug, the gemcitabine concentration was increased incrementally, with a final concentration of 20 ng/ml for MIAPaCa-2. Through this process, GR cells were established (25). A total of three stable GR cell lines were established from MIAPaCa-2, termed GR3, GR8 and GR10. The GR cell lines showed ~100% cell viability even at gemcitabine concentrations of 200 ng/ml (Fig. S1). Gemcitabine was purchased from Eli Lilly (Eli Lilly and Company).

RT-qPCR was performed as previously described (26). Briefly, total RNA from cultured cells was extracted using QIAzol^® reagent and a miRNeasy mini kit (Qiagen GmbH), while total RNA from serum was extracted using a miRNeasy Serum/Plasma Advanced kit (Qiagen GmbH) and reverse transcribed into cDNA using a Reverse Transcription System kit, according to the manufacturer's protocol (Promega Corporation). qPCR was performed using THUNDERBIRD SYBR qPCR Mix (Toyobo Co., Ltd.) on a QuantiStudio7 amplifier (Thermo Fisher Scientific, Inc.). PCR cycling conditions were as follows: An initial holding stage at 94°C for 60 sec, followed by 40 cycles of denaturation at 95°C for 15 sec and annealing/extension at 62°C for 60 sec. The 2^−ΔΔCq method was used to analyze the relative levels of target genes (27). All primer sequences are listed in Table SI. All experiments were performed in triplicate.

For circ72309 overexpression, the identified 580-bp sequence was synthesized as an artificial oligonucleotide and subcloned into pRP[Exp]-EGFP Laccase2 MCS Exon Vector (VectorBuilder Inc.), flanked by a GT-AG splice-site. Constructs were transfected into cells using Lipofectamine^® 3000 (Invitrogen; Thermo Fisher Scientific, Inc.) according to the manufacturer's protocol. Confirmation of circ72309 transfection was assessed using RT-qPCR 48 h after transfection. The GR cell lines overexpressing circ72309 were termed GR_circ72309 overexpressing (OE). Cells transfected with pRP[Exp]-EGFP/Neo-CMV>ORF_Stuffer (vector ID: VB900145-9829wty) served as the negative control and were termed GR_NC.

For RNase R digestion assay, total RNA was treated with or without 5 U/µg RNase R (AR Brown Co., Ltd.) and incubated at 37°C for 2 h. After purifying RNA using an RNA Clean-up XS kit (Takara Bio, Inc.), RT-qPCR was performed.

circRNA sequencing was performed as previously described (13). Briefly, raw sequencing reads underwent quality assessment using FastQC (version 0.11.9; https://www.bioinformatics.babraham.ac.uk/projects/fastqc/, RRID:SCR_014583), followed by adaptor trimming and removal of low-quality reads using fastp (version 0.19.5; https://github.com/OpenGene/fastp). Cleaned reads were aligned to the human reference genome (GRCh38; https://www.gencodegenes.org/human/) with STAR (version 2.7.10a; https://github.com/alexdobin/STAR, RRID:SCR_004463), and circRNAs were identified using CIRCexplorer2 (version 2.4.0; https://circexplorer2.readthedocs.io/) and circRNA_finder (https://github.com/orzechoj/circRNA_finder). Expression quantification and normalization were performed using edgeR (version 3.42.4; https://bioconductor.org/packages/edgeR/, RRID:SCR_012802) prior to differential expression analysis. To compare differences in the circRNA expression profiles between MIAPaCa-2 and GR3 cells, the fold change was calculated between the groups for each circRNA. Student's t-test was used to determine statistical significance and a significant difference in circRNA levels was defined as a false discovery rate (FDR)<0.05.

To compare differences in the mRNA expression profiles between GR3_ NC and GR3_circ72309 OE cells, mRNA sequencing was performed 48 h after transfection. All sequencing libraries were then sequenced at single-end 100 bp on an Illumina NovaSeq 6000 (Illumina, Inc.) at the Research Institute for Microbial Diseases of the University of Osaka (Osaka, Japan). Raw gene expression was quantified across all gene exons using the top-expressed isoform as a proxy for gene expression and differential gene expression analyses were performed using edgeR (version 3.42.4; http://bioconductor.org/packages/edgeR/, RRID: SCR_012802) using replicates to compute within-group dispersion. Undetectable mRNAs were excluded from the analysis. Differentially expressed genes (DEGs) were defined as having an FDR<0.05 and a log₂ fold change >1.0. Gene Set Enrichment Analysis (GSEA) (RRID: SCR_003199) was performed using clusterProfiler (version 4.8.3; https://bioconductor.org/packages/clusterProfiler/, RRID: SCR_016884). The sequencing coverage and quality statistics for each sample are summarized in Table SII.

To assess the drug response of PC cell lines, 3,000 treated cells were seeded per well in 96-well plates. The following day, fresh medium containing gemcitabine at concentrations of 0, 1.5625, 3.125, 6.25, 12.5, 25, 50, 100 or 200 ng/ml was added to the cells and incubated for 72 h. The 12 mM MTT (~10 µl) solution was added to each well and incubated at 37°C for 4 h. Subsequently, 100 µl SDS-HCl solution was added to each well and agitated for 1 h at room temperature on a shaker; subsequently, the absorbance at 570 nm was measured using a Model 680 Microplate Reader (Bio-Rad Laboratories, Inc.) and the IC₅₀ was calculated. Dose-response curves were analyzed using non-linear regression using a four-parameter logistic (4PL) sigmoidal model [log(concentration) vs. normalized cell viability]. IC₅₀ values were derived from the fitted curves and goodness of fit was assessed by the coefficient of determination (R²) (28). The fitted curves are presented as solid lines. For resistant clones that did not display a typical sigmoidal decline, median values are shown using dashed lines.

The Annexin V-FITC Apoptosis Detection Kit (BioVision, Inc.; Abcam) was used to assess apoptosis according to the manufacturer's protocol. The cells were digested with trypsin and washed with PBS twice. The cell count was adjusted to 1×10⁵; subsequently, the cells were stained with propidium iodide and Annexin V-FITC according to the manufacturer's protocol. Analysis was performed using a flow cytometer (BD Biosciences) and the data were measured using FlowJo version 10 (RRID: SCR_008520; BD Biosciences). For each sample, at least 10,000 events were acquired within the live cell gate. All experiments were independently repeated three times (29).

Total protein was extracted from short interfering (si)RNA knockdown and overexpression cells using RIPA lysis buffer containing protease inhibitors (MilliporeSigma). Total protein concentration was determined using a Bradford protein assay kit (Bio-Rad Laboratories, Inc.). The samples were heated at 95°C for 5 min in loading buffer, loaded into wells of 15% acrylamide gels (Bio-Rad Laboratories, Inc.) along with a protein ladder and resolved by SDS-PAGE at 200V. Resolved proteins were subsequently transferred onto PVDF membranes (Bio-Rad Laboratories, Inc.) for 1 h and blocked using Blocking One (Nacalai Tesque, Inc.) at room temperature for 30 min. The blots were treated with an anti-equilibrative nucleoside transporter-1 (ENT1) polyclonal rabbit antibody (ProteinTech Group, Inc.; 1:600; cat. no. 29862-1-AP; RRID: AB_2935484) and an anti-cytidine deaminase (CDA) polyclonal rabbit antibody (Invitrogen; Thermo Fisher Scientific, Inc.; 1:5,000; cat. no. PA5-84630; RRID: AB_2791781) in Can Get Signal^® solution 1 (Toyobo Co., Ltd.) overnight at 4°C on a shaker, while an anti-βactin polyclonal rabbit antibody (Abcam; 1:1,000; cat. no. 1784-1; RRID: AB_598136) was used as the loading control. After washing three times, horseradish peroxidase-conjugated anti-rabbit IgG antibody (Cytiva; cat. no. NA934, RRID: AB_772206) and HRP-conjugated anti-human sheep IgG whole antibody (Cytiva cat. no. NA933; RRID: AB_772208) were used as secondary antibodies. Signals were visualized using an ECL Prime Western Blotting Detection kit (Cytiva) and membranes were scanned using a ChemiDoc™ Touch MP (Bio-Rad Laboratories, Inc.) (30). Band intensities were quantified using ImageJ (version 1.54g; https://imagej.nih.gov/ij/, National Institutes of Health) using a lane-based densitometric method (31). The values were normalized to the corresponding β-actin loading control. Quantitative results were presented as the mean ± SE from at least three independent experiments.

A ROS Assay Kit-Highly Sensitive (2′,7′-Dichlorodihydrofluorescein diacetate) containing DCFH-DA (Dojindo Molecular Technologies, Inc.) was used to calculate the ROS activity according to the manufacturer's protocol. Briefly, ~1×10⁵ cells were treated with DCFH-DA as indicated, incubated for 30 min in PBS, trypsinized and analyzed by flow cytometry. For each sample, at least 15,000 events were acquired within the live cell gate and the mean fluorescence intensity (MFI) of DCF was calculated using FlowJo (version 10; BD Biosciences). Each experiment was performed in triplicate (n=3 independent biological replicates). N-acetylcysteine (NAC; 5 mmol/l) was used as an ROS scavenger. To inhibit ROS, PC cell lines were treated for 36 h with NAC. PBS-treated cells served as the control.

circ72309 siRNA targeting the sequence of the backsplicing junction of circ72309 was synthesized. Silencer™; Select Negative Control #1 siRNA (Invitrogen; Thermo Fisher Scientific, Inc.; cat. no. 4390843) was used as the negative control. This reagent is proprietary and the manufacturer does not disclose its nucleotide sequence; only product identification and catalog information are publicly available. MIAPaCa-2, PANC1 and PSN-1 were transfected with circ72309 siRNAs using Lipofectamine^® RNAiMAX transfection reagent (Invitrogen; Thermo Fisher Scientific, Inc.) according to the manufacturer's protocol. A total of 2×10⁵ cells/well were seeded in a 6-well plate and 50 nM siRNA was added to each well with the transfection reagent and cultured for 24 h at 37°C in 5% CO₂. The sequence of the circ72309 siRNA is listed in Table SI. Successful knockdown of circ72309 by siRNA transfection was confirmed (Fig. S2).

Circular RNA Interactome (RRID: SCR_016304) (32) was used to identify target miRNAs for circ72309 (Table SIII) and among these miRNAs, TargetScan (RRID: SCR_010845, targetscan. org/vert_80/) was used to explore the candidate mRNAs regulating gemcitabine metabolism targeted by relevant miRNAs (Table SIV). Expression of these target miRNAs was compared using RT-qPCR between GR_NC cells and GR_circ72309 OE cells in all three GR cell lines to confirm the effect of circ72309 on the expression of these miRNAs.

Data were retrospectively collected from patients with PC aged ≥20, excluding those with intrapapillary mucinous carcinoma, between January 2016 and December 2023. A total of 31 patients with resectable status according to the definition of the National Comprehensive Cancer Network Guidelines (Pancreatic Adenocarcinoma, version 1.2023) (33), who received gemcitabine plus S-1 for neoadjuvant chemotherapy (excluding chemoradiotherapy), were recruited. The median age was 73 (50-84) years, and the sex ratio was 16 males (53%) and 14 females (47%). Borderline resectable and unresectable cases were excluded as they frequently received heterogeneous treatment regimens, including multiple lines of chemotherapy and/or radiotherapy, which could confound the accurate assessment of chemosensitivity to gemcitabine. By contrast, patients with resectable status were treated under a standardized protocol with gemcitabine plus S-1. This homogeneity provided a consistent clinical background to appropriately evaluate the predictive value of serum circ72309. The levels of circ72309 in the pre-treatment serum were measured using RT-qPCR. Serum samples were stored at -80°C without repeated freeze-thaw cycles. The cases were divided into two groups according to the median circ72309 levels and the relationship between the levels of circ72309 and the therapeutic effect of neoadjuvant chemotherapy was examined. The Response Evaluation Criteria in Solid Tumors (RECIST) classification (34) and progression-free survival (PFS) were used to assess the therapeutic effect. Briefly, in the RECIST classification, target lesions were defined as primary tumors and lymph nodes (with short axes of ≥15 mm) and the response criteria was defined as follows: i) Complete response (CR), the disappearance of all target lesions; ii) partial response (PR), ≥30% decrease in the sum of diameters of target lesions compared to the respective baseline; and iii) progressive disease (PD), ≥20% increase in the sum of diameters of target lesions compared to the respective baseline (including the baseline sum if that was the smallest on study); in addition to the relative increase of 20%, the sum should also have demonstrated an absolute increase of ≥5 mm; and the appearance of one or more new lesions was also considered progression; and iv) stable disease (SD), neither sufficient shrinkage to qualify for PR nor sufficient increase to qualify for PD, taking the smallest sum diameters while on study as the baseline. PFS was calculated from the day of neoadjuvant chemotherapy initiation to the day of mortality from any cause or to the day of tumor progression and was censored on the last day that the patient was documented to be alive without tumor progression. Post-operative recurrence was monitored using regular imaging tests, including enhanced abdominal computed tomography (CT) and tracking changes in the levels of tumor markers. Post-operative adjuvant treatment was routinely performed unless clinical contraindications, such as gemcitabine or S-1, were present for 6 months (35,36). Local recurrence was defined as a recurrence near the resection site, including the lymph node dissection site and the area where the tumor was dissected. Distant recurrence was defined as metastasis to other organs.

Categorical characteristics are presented as frequencies and percentages. Numerical values are presented as the median and range. Normality was tested using a Shapiro-Wilk test and equality of variances was assessed using a Levene's test. Specifically, a Student's t-test was used for RT-qPCR and western blotting data after confirming normality and homogeneity of variance. Wilcoxon tests were used for non-normally distributed continuous variables in clinical datasets. Categorical variables were analyzed using a χ² or Fisher's exact test, as appropriate. Kaplan-Meier analysis with a log-rank test was used for survival comparisons and Cox proportional hazards models were applied for univariate and multivariate survival analyses. For all analyses involving multiple comparisons, appropriate post hoc tests with Bonferroni corrections were applied if differences were considered significant by ANOVA. JMP^® Pro version 14 (SAS Institute Inc.; RRID: SCR_022199) and R Project for Statistical Computing (version 4.3.2; https://www.r-project.org/, RRID: SCR_001905) were used for statistical analyses. P<0.05 was considered to indicate a statistically significant difference.

circRNA sequencing was performed on parent PC cell lines (MIAPaCa-2, gemcitabine IC₅₀=12.5 ng/ml; n=3) and a gemcitabine-resistant cell line (GR3; gemcitabine IC₅₀ not reached; n=3) to extract candidate circRNAs associated with gemcitabine sensitivity (GSE297252, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE297252). A total of 29 differentially expressed circRNAs (20 upregulated and 9 downregulated) were identified by circRNA sequencing (Fig. 1A). Based on a literature search, circ72309 was identified as potentially relevant; it has a full length of 7,248 bp, a spliced length of 580 bp and is located on human chromosome 5 (chr5:38523520-38530768). To confirm the levels of circ72309 by RT-qPCR, the primers specific for circ72309 were designed to amplify the backsplicing junction sequence (Fig. 1B and Table SV) and the primers were subjected to RNase R treatment to ascertain whether they were capable of recognizing circular RNA. The Cq value of GAPDH, a linear RNA, was increased by RNase R treatment; however, the Cq value of circ72309 was unchanged by RNase R treatment, confirming that the primers recognized the backsplicing junction of circ72309 (Fig. 1C). Subsequently, the levels of circ72309 in three gemcitabine-resistant cell lines (GR3, GR8 and GR10) were confirmed by RT-qPCR; circ72309 was markedly lower in GR cell lines than in MIAPaCa-2 (Fig. 1D). Furthermore, the culture supernatant levels of circ72309 in GR3 was observed to be markedly lower than that in MIAPaCa-2 cells, as well as in cells when compared between MIPaCa-2 and GR3 (Fig. 1E).

The identified 580-bp sequence was synthesized as an artificial oligonucleotide, subcloned into the pRP[Exp]-EGFP Laccase2 MCS Exon Vector (vector ID: VB230525-1184qve), flanked by a GT-AG splice site (Fig. 2A) and transfected into GR3, GR8 and GR10 cells. The circ72309 levels in the transfected cell lines were markedly increased following transfection (Fig. S2). Each overexpressed gemcitabine-resistant cell line (GR3_circ72309 OE, GR8_circ72309 OE and GR10_circ72309 OE) was assessed using MTT and apoptosis assays. The pRP[Exp]-EGFP/Neo-CMV>ORF_Stuffer (vector ID: VB900145-9829wty) was used as a control (GR_NC) for comparison. In the MTT assay, a concentration-dependent decrease in cell viability was evident in GR_circ72309 OE, indicating enhanced sensitivity to gemcitabine. The fitted curves followed a typical sigmoidal profile, with IC₅₀ values of 13.3 ng/ml (R²=0.974) for GR3_circ72309 OE, 130 ng/ml (R²=0.951) for GR8_circ72309 OE and 16.9 ng/ml (R²=0.984) for GR10_circ72309 OE (Fig. 2B). In resistant clones, cell viability remained relatively high even at the maximal drug concentrations tested and the curves did not follow a typical sigmoidal profile, consistent with a resistant phenotype. The effect of si_circ72309 on gemcitabine sensitivity was evaluated using MIAPaCa-2 (IC₅₀, 23.3 ng/ml), Panc-1 (IC₅₀, 24.7 ng/ml) and PSN-1 (IC₅₀, 6.0 ng/ml) cells. Next, untreated parent cells, scramble control (Scr) transfected cells and GR3, GR8 and GR10 si_circ72309 transfected cells were assessed. In MIAPaCa-2, no difference in gemcitabine sensitivity was observed between MIAPaCa-2_Parent and MIAPaCa-2_Scr (IC₅₀, MIAPaCa-2_ Parent vs. MIAPaCa-2_Scr, 23.3 ng/ml vs. 23 ng/ml). However, MIAPaCa-2_si_circ72309 demonstrated increased resistance to gemcitabine, with an IC₅₀ of 180 ng/ml. Similarly, gemcitabine resistance was markedly augmented in Panc-1_si_circ72309 (IC₅₀, not reached) relative to Panc-1_Parent (IC₅₀, 24.7 ng/ml) and Panc-1_Scr (IC₅₀, 21.3 ng/ml). No effect on gemcitabine sensitivity was observed in PSN-1 (Fig. 2C). The apoptosis rate after 20 ng/ml gemcitabine treatment was examined using the annexin V assay, which showed a significant increase in early and late apoptotic cells (Q3 and Q2) in GR_circ72309 OE [rate of apoptosis cells (Q2+Q3): GR3_NC vs. GR3_circ72309 OE, 18.0 vs. 38.3%, P=0.002; GR8_NC vs. GR8_circ72309 OE, 18.7% vs. 43.9%, P=0.009; GR10_NC vs. GR10_circ72309 OE, 12.4% vs. 32.6%, P=0.045, Fig. 2D]. This showed that overexpression of circ72309 in gemcitabine-resistant cell lines induced apoptosis in the presence of gemcitabine, thereby enhancing gemcitabine sensitivity.

To investigate the mechanism by which overexpression of circ72309 improved gemcitabine sensitivity, GR3_NC (n=3) and GR3_circ72309 OE (n=3) were submitted for mRNA sequencing (Fig. 3A). The differentially expressed mRNAs (545 in total; 103 upregulated and 442 downregulated) were identified (GSE297082, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE297082). GSEA showed that the oxidative phosphorylation and ATP production pathway in mitochondria were enhanced in GR3_circ72309 OE (Fig. 3B), suggesting that ROS activity may be involved in the increase in gemcitabine sensitivity. The ROS activity in GR_circ72309 OE showed an increase compared to GR_NC (Fig. 3C and D). A previous report revealed that the effect of ROS reduces CDA, which metabolizes gemcitabine to the inactive form 2′,2′-Difluoro-2′-deoxyuridine (dFdU) (37). Subsequently, GR_circ72309 OE were treated with a ROS scavenger, NAC, to suppress ROS activity and the effect on gemcitabine sensitivity was then assessed. As a control, the same volume of PBS was added to GR_circ72309 OE_NAC-. In the MTT assay, GR_circ72309 OE_NAC+ promoted gemcitabine resistance, indicating that NAC counteracted the gemcitabine sensitivity observed following overexpression of circ72309 (IC₅₀: GR3_circ72309 OE_NAC-vs. GR3_circ72309 OE_NAC+, 12.0 ng/ml vs. not reached; GR8_circ72309 OE_NAC-vs. GR8_circ72309 OE_NAC+, 69.6 ng/ml vs. not reached; GR10_circ72309 OE_NAC-vs. GR10_circ72309 OE_NAC+, 20.8 ng/ml vs. not reached; Fig. 3E). When CDA expression was confirmed by RT-qPCR and western blotting, its expression was markedly reduced in GR_circ72309 OE (Fig. 4A, C and D). mRNA sequencing results confirmed the changes in genes related to gemcitabine metabolism and showed a significant increase in SLC29A1 expression (log fold change, 1.16; P<0.001). The gene encodes human equilibrative nucleoside transporter 1 (hENT1), a transmembrane glycoprotein that localizes to the plasma and mitochondrial membranes and mediates the cellular uptake of nucleosides from the surrounding medium. When the expression of SLC29A1 was confirmed by RT-qPCR and western blotting, the expression was markedly increased in GR_circ72309 OE cells (Fig. 4B, C and E).

Based on the CircInteractome database, circ72309 has adsorption sites for 26 miRNAs (Table SIII). Among these miRNAs, miR-1225-3p and miR-1234 were identified by TargetScan as having binding sites with SLC29A1 (Table SIV). Further, it was confirmed that the expression of miR-1225-3p and miR-1234 in GR cell lines was markedly decreased by circ72309 overexpression (Fig. S3).

A total of 31 patients with PC who received gemcitabine plus S-1 for neoadjuvant chemotherapy were included. The levels of circ72309 are shown in Fig. 5A; one case showing an outlier value was excluded (shown as Δ in Fig. 5A). The patients' pre-treatment characteristics (n=30) are shown in Table I. A total of 3 patients could not undergo surgery owing to tumor progression during neoadjuvant chemotherapy and surgery was performed on 27 patients (90%). In 13 patients who underwent resection and for whom post-operative serum samples were available, serum circ72309 levels were measured after tumor resection and were found to be markedly reduced (Fig. S4). The median pre-treatment tumor size was 20 (0-50) mm and the median pre-treatment CA19-9 levels were 53.3 (0.3-2,729) U/ml. The 30 cases were stratified into high and low groups based on the median circ72309 level; Table I shows the pre-treatment characteristics of the two groups. The pre-treatment CA19-9 levels were markedly lower in the high circ72309 group. The pre-treatment tumor size did not differ markedly between the two groups. Fig. 5B shows a waterfall plot of the rate of change in tumor size. ~18 patients (60%) had some degree of tumor shrinkage on CT imaging. The RECIST classification demonstrated a markedly improved response in the high circ72309 group, whereas PD cases were observed only in the low circ72309 group (Table I; PD/SD/PR, high vs. low, 5/10/0 vs. 2/8/5, P=0.016). PFS was significantly higher in the high circ72309 group (Fig. 5C; median PFS rates, high vs. low, not reached vs. 9.5 months, P=0.035). The results of univariate and multivariate analyses of prognostic factors for PFS are shown in Table II. Multivariate analysis revealed that pre-treatment serum circ72309 levels were an independent prognostic factor for PFS (P=0.042).