Mutant KRAS Mediates circARFGEF2 Biogenesis to Promote Lymphatic Metastasis of Pancreatic Ductal Adenocarcinoma

Pancreatic ductal adenocarcinoma (PDAC) is one of the deadliest digestive system malignancies; its incidence is increasing and it is predicted to be the second most common cause of death within the next decade (1, 2). KRAS mutations were identified in approximately 90% of patients with PDAC and KRAS^G12D was the most common allele in PDAC (3, 4). Gene therapy targeting the KRAS^G12D mutation achieved 70% regression in metastatic PDAC, indicating the therapeutic potential of KRAS^G12D in metastatic PDAC (5). Lymph node (LN) metastasis is an important metastatic route in PDAC. Recent studies demonstrated the increased tendency of LN metastasis in KRAS^G12D PDAC (6). Therefore, investigating the underlying mechanism between KRAS^G12D mutation and LN metastasis in PDAC is of great importance.

Circular RNAs (circRNA) are covalently enclosed, single-stranded RNAs that have emerged as an important class of molecule in tumorigenesis by regulating various biological processes, such as gene expression, protein translation, and binding with RNA-binding proteins (RBP; refs. 7, 8). Recently, a series of studies revealed that circRNAs are significant in KRAS mutant-driven cancers (9). In dividing cells, circRNAs increased the expression level of KRAS, which prevented senescence and enabled a proliferative phenotype of these cells (10). In bladder cancer, circRNAs activated KRAS signaling and decreased the chemosensitivity of bladder cancer cells to cisplatin, promoting bladder cancer invasion (11). Nevertheless, little is known about the relationship between circRNAs and KRAS^G12D mutant-driven LN metastasis in PDAC.

The traditional circRNA research paradigm mainly focused on the downstream targets of circRNAs, such as acting as endogenous competitive RNAs that sponge miRNAs and blocking the inhibitory effect of miRNAs on their target mRNAs (12). However, differing from the traditional research paradigm, recent studies provided great insight into circRNA biogenesis (13, 14). Generally, circRNA biogenesis is mediated by back-splicing of the proximity of a downstream splice site that interacts with an upstream splice site (15, 16). Recent studies demonstrated that KRAS mutant cells have significant circRNA expression profiles compared with other cells and the mechanism underlying back-splicing cannot fully explain why these circRNAs are abundant in KRAS mutant cells (9, 17). Therefore, understanding the circRNA biogenesis mechanism in KRAS^G12D PDAC is of great significance for KRAS^G12D PDAC LN metastasis.

Quaking (QKI) is a member of the signal transduction and activator of RNA (STAR) family, which belongs to the hnRNPK homology (KH)-type family and is a well-known regulator of pre-mRNA alternative splicing, mRNA turnover, and translation (18, 19). Recently, QKI was identified as the culprit protein in the biogenesis of certain circRNAs by regulating pre-mRNA alternative splicing, and the majority of QKI binds within introns were revealed via specific motifs (20–22). Nonetheless, the alternative splicing mechanism is a complex process that requires the cooperation of a series of proteins, except for QKI (23). The proteins associated with QKI in regulating circRNA biogenesis remain unclear and require further investigation.

In this study, we identified a novel circRNA, circARFGEF2 (hsa_circ_0060665), which was upregulated in KRAS^G12D PDAC and positively associated with KRAS^G12D PDAC LN metastasis. Overexpressing circARFGEF2 significantly facilitated KRAS^G12D PDAC LN metastasis in vitro and in vivo. Mechanistically, circARFGEF2 biogenesis was activated by the alternative splicing factor, QKI-5, which recruited U2AF35 to facilitate spliceosome assembly and bind with the QKI binding motifs and neighboring reverse complement sequence in intron 3 and 6 of ARFGEF2 pre-mRNA to induce circARFGEF2 biogenesis. QKI-5-mediated circARFGEF2 sponged miR-1205 and promoted the activation of JAK2, which phosphorylated STAT3 and activated the JAK2–STAT3 signaling pathway to trigger KRAS^G12D PDAC lymphangiogenesis and LN metastasis. Our findings reveal the underlying mechanism of KRAS^G12D mutation-related circRNA biogenesis and demonstrate the crucial role of circARFGEF2 in KRAS^G12D PDAC LN metastasis, indicating that circARFGEF2 might be a potential target for treating LN metastatic KRAS^G12D PDAC.

A total of 395 patients with PDAC who had undergone surgery and 98 patients with PDAC who had undergone biopsy at Guangdong Provincial People's Hospital, Sun Yat-sen Memorial Hospital, Sun Yat-sen University, and The Sixth Affiliated Hospital of Sun Yat-sen University were included. Fifty-two patients with lung adenocarcinoma and 57 patients with colon cancer who had undergone surgery in Sun Yat-sen Memorial Hospital, Sun Yat-sen University were included. The diagnosis of all the patients were confirmed by two professional pathologists independently. Patients with other pathological diagnosis were excluded. All samples were quickly placed in liquid nitrogen and transferred to −80°C for long-term preservation.

Human lymphatic endothelial cells (HLEC) were purchased from ScienCell Research Laboratories. The other cell lines were purchased from ATCC. The PANC-1 (ATCC, CRL-1469MET; RRID: CVCL_A4BT), Capan-2 (ATCC, HTB-80; RRID: CVCL_0026), and hTERT-HPNE (ATCC, CRL-4023TM; RRID: CVCL_C466) cells were cultured in DMEM (Gibco). AsPC-1 (ATCC, CRL-1682; RRID: CVCL_0152) and BxPC-3 (ATCC, CRL-1687; RRID: CVCL_0186) cells were cultured in RPMI1640 medium (Gibco). All the culture media are complete media with 10% FBS (HyClone). The HLECs were cultured with endothelial cell medium (ECM; ScienCell Research Laboratories) with 5% FBS. All cell lines were authenticated by short tandem repeat analysis and were cultured in a humid atmosphere containing 5% CO₂ at 37°C.

RNA immunoprecipitation (RIP) assay was conducted according to the instructions of EZ-Magna RIP Kit (Merck, catalog no. 17-701) to identify the interaction between QKI and ARFGEF2 pre-mRNA. Cells were incubated on ice for 5 minutes with 100 μL RIP lysis buffer, then stored at −80°C for 5 hours. Magnetic beads (50 μL) were mixed with 5 μL anti-QKI-5 antibody (Thermo Fisher, A300–183A, RRID: AB_2173160) and rotated for 30 minutes at room temperature. Then the mixture was placed in a magnetic rack to discard the supernatant and resuspended with 900 μL RIP buffer. The RIP lysis products were thawed, added to RIP buffer containing magnetic beads and antibody complex, and incubated overnight at 4°C. The centrifuge tube was placed on the magnetic rack, the supernatant discarded, the beads washed with 500 μL RIP washing buffer six times, and the RNA bound on the magnetic beads was extracted and stored on ice for further experiments.

The fluorescence assay was conducted according to the instruction of Multiple Fluorescent Immunohistochemical Kit (Panovue). Paraffin tissue sections were placed at 60°C for 2 hours, then dewaxed with xylene and hydrated with decreasing concentrations of ethanol. After washing with PBS (Biological Industries), the sections were immersed in 10% neutral formaldehyde for 10 minutes. Then the sections were placed in ethylene diamine tetraacetic acid (EDTA) buffer solution for antigen repair, then heated for 7 minutes at medium-high temperature and 14 minutes at medium-low temperature. After cooling to room temperature, the slices were blocked with catalase inhibitor for 10 minutes, washed three times with Tris Buffered Saline with Tween-20 (TBST), and sealed with goat serum. Then, the sections were incubated with the primary antibody solution at room temperature for 1 hour and followed by HRP-labeled secondary antibody solution at room temperature for 30 minutes. Subsequently, the sections were incubated with dye working solution under dark conditions. Finally, the samples were stained with 4′,6-diamidino-2-phenylindole (DAPI) for 15 minutes. Images were captured by confocal fluorescence microscopy (Carl Zeiss AG).

WP1066 was purchased from MedChemExpress (catalog no. HY-15312) and administered to the mice intraperitoneally with a dose of 40 mg/kg from Monday to Friday for 6 weeks (24, 25). The toxicity of WP1066 was dose-dependent, when exceeding 50 mg/kg per day causing diarrhea and a high frequency of urination and we did not observe any toxicities of WP1066 in this study (26).

All quantitative data were calculated as the mean ± SD of at least three independent experiments. Cumulative survival time was evaluated with Kaplan–Meier analysis and tested by the log-rank method. Independent prognostic factors were determined with multivariate Cox proportional hazards models. The relationship between unpaired variables was evaluated using the chi-square test. The relationship between paired variables was evaluated using the two-tailed Student t test and one-way ANOVA. P < 0.05 was considered statistically significant. All statistical tests were performed with SPSS (Version 19.0; IBM Corp.).

The clinical data and specimens were obtained from patients with written informed consent. A total of 342 cases of these specimens have received the approval of the Committees for Ethical Review of Research involving Human Subjects at Sun Yat-sen Memorial Hospital [approval no. 2016 (136)]. A total of 158 cases of these specimens have received the approval of the Committees for Ethical Review of Research involving Human Subjects at Guangdong Provincial People's Hospital (approval no. KY2020–348–01). A total of 102 cases of these specimens have received the approval of the Committees for Ethical Review of Research involving Human Subjects at The Sixth Affiliated Hospital of Sun Yat-sen University (approval no. 2022ZSLYEC-295). The animal study was performed after receiving approval from the Guangdong Provincial People's Hospital (approval no. KY2020-348-01).

RNA sequencing data generated in the study are available in Gene Expression Omnibus under accession numbers GSE234760 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE234760) and GSE234927 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE234927). More detailed methods are available in the Supplementary Materials and Methods. The data generated in this study are available within the article and Supplementary data files, as well as upon request from the corresponding author.

To investigate whether circRNAs participate in KRAS^G12D-driven LN metastasis, we conducted next-generation sequencing in three paired PDAC tissues and normal adjacent tissues (NAT; GSE234760) and found 50 circRNAs upregulated, 58 circRNAs downregulated by >10-fold in the PDAC tissues (Supplementary Fig. S1A). Then the elevated circRNAs were evaluated in a larger cohort of 395-case patients with PDAC, which revealed that 11 circRNAs were upregulated in PDAC tissues as compared with NATs (Supplementary Figs. S1B–S1D). Among these 11 circRNAs, circARFGEF2 was most significantly upregulated in KRAS^G12D PDAC tissues as compared with PDAC tissues with other KRAS subtypes (KRAS^G12C, KRAS^G12V, KRAS^WT; Fig. 1A and B; Supplementary Fig. S1E).

Moreover, circARFGEF2 was overexpressed in LN-positive patients with KRAS^G12D PDAC or patients with a highly pathologic tumor-node-metastasis (TNM) stage, and the metastatic LNs possessed higher circARFGEF2 expression compared with the primary tumors (Fig. 1C and D; Supplementary Table S1). In situ hybridization (ISH) and immunofluorescence analysis revealed that circARFGEF2 overexpression was accompanied with increased microlymphatic vessel density (MLD) in KRAS^G12D PDAC tissues (Fig. 1E and F). Furthermore, fluorescence in situ hybridization (FISH) revealed circARFGEF2 enrichment in LN-positive KRAS^G12D PDAC tissues compared with LN-negative (Fig. 1G and H), indicating that circARFGEF2 was positively correlated with LN metastasis of KRAS^G12D PDAC. Kaplan–Meier survival analysis demonstrated that circARFGEF2 overexpression was negatively correlated with overall survival (OS) and disease-free survival (DFS) in patients with KRAS^G12D PDAC (Fig. 1I and J). Univariate and multivariate Cox analysis indicated that circARFGEF2 was an independent prognostic factor of the OS and DFS of patients with KRAS^G12D PDAC (Supplementary Tables S2 and S3). In addition, circARFGEF2 was rarely differentially expressed or correlated with LN metastasis of other common cancers with KRAS^G12D mutation, including colon cancer and lung adenocarcinoma (Supplementary Figs. S1F–S1I), suggesting that the role of circARFGEF2 in KRAS^G12D-induced LN metastasis is specific to PDAC. Taken together, these results suggest that circARFGEF2 is positively associated with KRAS^G12D PDAC LN metastasis.

As the circular construction of circRNA determines their biological functions (12), we investigated the characteristics of circARFGEF2. We found that circARFGEF2 was derived from the back-splicing of exon 4 to 6 of the ARFGEF2 gene (chr20:47567859–47570327; Supplementary Fig. S1J). Moreover, circARFGEF2 was detected in PCR products using complementary DNA (cDNA) as the template rather than genomic DNA (gDNA; Fig. 1K; Supplementary Fig. S1K). In addition, circARFGEF2 could only be amplified from the reverse transcription products of random primers instead of oligo-dT primers, confirming the circular form of circARFGEF2 without poly-A tail (Fig. 1L; Supplementary Fig. S1L). Furthermore, circARFGEF2 exhibited stronger resistance to RNase R compared with ARFGEF2 mRNA, while actinomycin D assay demonstrated that circARFGEF2 had a longer half-life than ARFGEF2 mRNA (Fig. 1M and N; Supplementary Figs. S1M and S1N), indicating that circARFGEF2 was more stable than ARFGEF2 mRNA. Taken together, these findings demonstrate that circARFGEF2 has a highly stable structure.

We verified that circARFGEF2 abundance was increased in KRAS^G12D PDAC tissues compared with other KRAS subtypes. Subsequently, to examine whether circARFGEF2 upregulation in KRAS^G12D PDAC was a consequence of active regulation or alteration of the parent gene expression, we quantified circARFGEF2, ARFGEF2 mRNA, and ARFGEF2 pre-mRNA expression in KRAS^G12D PDAC. circARFGEF2 expression was significantly increased in the KRAS^G12D PDAC tissues compared with other KRAS subtypes (Fig. 1A), whereas ARFGEF2 pre-mRNA expression was not significantly altered (Supplementary Fig. S2A), indicating that substantially more circARFGEF2 was present in KRAS^G12D PDAC. Importantly, the ratio of circARFGEF2 and ARFGEF2 mRNA was increased in KRAS^G12D PDAC compared with other KRAS subtypes (Supplementary Fig. S2B), suggesting increased circARFGEF2 biogenesis in KRAS^G12D PDAC. To confirm this, we constructed a dual-color fluorescence reporter with circRNAs give rise to IRES-mediated translation of GFP and linear mRNA undergoes to mCherry to quantify linear and circRNA splicing of ARFGER2 pre-mRNA simultaneously (Fig. 2A). The illustration of dual color fluorescence reporter assay was shown in Supplementary Fig. S9. Linear ARFGEF2 generation indicated by HA-mCherry expression was increased and circARFGEF2 formation indicated by FLAG-GFP expression as well as GFP–mCherry ratio were significantly diminished after applying KRAS^G12D inhibitor MRTX1133 or KRAS^G12D siRNA in PANC-1 cells, whereas no obvious change was observed after applying MRTX1133 in Mia PaCa-2 (KRAS^G12C) and BxPC-3 (KRAS^WT) cells (Fig. 2B–D; Supplementary Figs. S2C–S2I), indicating that KRAS^G12D mutation is required for circARFGEF2 biogenesis.

The crucial role of QKI in circRNA biogenesis has been demonstrated previously (27). Accordingly, we performed next-generation sequencing in three paired KRAS^G12D PDAC tissues and NATs to evaluate whether QKI regulates circARFGEF2 biogenesis in KRAS^G12D PDAC and determined that QKI was upregulated in KRAS^G12D PDAC tissues (GSE234927; Fig. 2E). Dual-color fluorescence reporter assay showed that downregulating QKI expression significantly decreased the GFP–mCherry ratio (Fig. 2F and G), indicating that QKI is required for circARFGEF2 biogenesis.

QKI comprises three isoforms (QKI-5, QKI-6, QKI-7) that contribute to different RNA processing (28). To confirm the specific isoform that contributed to circARFGEF2 biogenesis in KRAS^G12D PDAC, dual-color fluorescence report assay and Western blotting assay were conducted and revealed that reducing QKI-5 expression significantly decreased FLAG-GFP expression while increased HA-mCherry expression (Fig. 2H; Supplementary Fig. S2J). Moreover, QKI-5 was significantly overexpressed in KRAS^G12D PDAC cells, rather than PDAC cells with other KRAS subtypes (Supplementary Fig. S2K). Increasing QKI-5 expression promoted circARFGEF2 expression in the PDAC cells while reducing QKI-5 expression diminished circARFGEF2 expression (Supplementary Figs. S2L–S2N), suggesting that QKI-5 was involved in circARFGEF2 biogenesis. We further explored the mechanism underlying KRAS^G12D mutation-induced QKI-5 overexpression. Given that KRAS^G12D mutation activated RAF/MEK/ERK, PI3K/AKT, and RAL/TBK1 signaling pathways to promote target genes expression in cancers (29, 30), we investigated whether these pathways were involved in KRAS^G12D-mediated QKI-5 overexpression. Inhibiting RAF/MEK/ERK signaling pathway with MCP110 significantly reduced QKI-5 expression, whereas neither inhibiting PI3K/AKT with LY294002 nor inhibiting RAL/TBK1 with BQU57 affected QKI-5 expression in PANC-1 cells (Supplementary Figs. S2O and S2P). Moreover, blocking KRAS^G12D mutation signaling decreased QKI-5 expression in PANC-1 cells, which was reversed by the activation of RAF/MEK/ERK with C16-PAF (Supplementary Figs. S2Q and S2R), indicating that RAF/MEK/ERK signaling pathway was required for KRAS^G12D mutation-induced QKI-5 expression. Together, these results suggest that KRAS^G12D mutation-driven QKI-5 overexpression promoted circARFGEF2 biogenesis in PDAC.

We investigated the mechanism underlying QKI-5-regulated circARFGEF2 biogenesis. It is well known that circRNA biogenesis is derived from pre-mRNA back-splicing (13). Thus, we investigated whether QKI-5 promotes circARFGEF2 biogenesis by regulating ARFGEF2 pre-mRNA splicing. The results demonstrated that upregulated QKI-5 expression increased the circARFGEF2-ARFGEF2 pre-mRNA ratio whereas decreased the ARFGEF2 mRNA-ARFGEF2 pre-mRNA ratio, indicating that QKI-5 regulated ARFGEF2 pre-mRNA splicing (Supplementary Fig. S3A). Pre-mRNA alternative splicing is generally catalyzed by spliceosome, the assembly of which occurs stepwise along with associated protein cofactors. Coimmunoprecipitation assays followed by MS and Western blotting identified that the QKI-5 interacted with U2 auxiliary factor 1 (U2AF35), a U2AF subunit that directly triggers spliceosome assembly to stimulate splicing (Fig. 2I and J). Moreover, decreasing U2AF35 expression significantly impeded QKI-5-induced circARFGEF2 biogenesis (Fig. 2K and L). The above results indicate that QKI-5 recruits U2AF35 to regulate ARFGEF2 pre-mRNA splicing and promote circARFGEF2 biogenesis.

The effect of QKI-5 on circRNA formation occurred through direct binding with the binding sites on pre-mRNAs (20). RIP was performed to determine that QKI-5 directly bound to ARFGEF2 pre-mRNA (Supplementary Fig. S3B). Because it is well defined that RBPs bind to the flanking introns of circRNA splicing exons and induce circularization, we designed primers for the introns adjacent to circARFGEF2-forming exons and found that QKI-5 bound to the exon-adjacent sites (intron 3 and 6) at much higher level compared with more distal regions elsewhere in ARFGEF2 pre-mRNA (Supplementary Fig. S3C).

To confirm the precise QKI-5 binding sites on ARFGEF2 pre-mRNA, we conducted pulldown assays with truncated sequences of intron 3 and 6, which revealed that QKI-5 was particularly enriched by intron 3^{1000–1200 nt} and intron 6^{7600–7800 nt} (Fig. 2M and N; Supplementary Figs. S3D and S3E). Moreover, RBPmap revealed a reverse complement sequence neighboring the QKI binding motifs at intron 3^{1128–1180 nt} and intron 6^{7672–7740 nt} (Supplementary Figs. S3F–S3I). Mutation of intron 3^{1128–1180 nt} and intron 6^{7672–7740 nt} impaired QKI-5 and ARFGEF2 pre-mRNA binding, suggesting that QKI-5 directly bound to these two ARFGEF2 pre-mRNA regions (Fig. 2O and P). Furthermore, we assessed whether the interaction between QKI-5 and ARFGEF2 pre-mRNA affected circARFGEF2 production in PDAC. CRISPR/Cas9-mediated knockout of intron 3^{1128–1180 nt} and intron 6^{7672–7740 nt} in ARFGEF2 pre-mRNA substantially reduced QKI-5-mediated circARFGEF2 overexpression (Fig. 2Q; Supplementary Fig. S3J). Moreover, knocking out the QKI-5 binding sites on ARFGEF2 pre-mRNA significantly diminished the rescue effect of U2AF35 on QKI-5 induced circARFGEF2 overexpression (Supplementary Figs. S3K and S3L). Together, our results demonstrate that QKI-5 recruits U2AF35 and binds to intron 3^{1128–1180 nt} and intron 6^{7672–7740 nt} in ARFGEF2 pre-mRNA to promote circARFGEF2 splicing (Fig. 2R).

As we observed a close clinical relationship between QKI-5-mediated circARFGEF2 expression and KRAS^G12D PDAC LN metastasis, we explored the biological function of QKI-5-mediated circARFGEF2 in KRAS^G12D PDAC in vitro. Lymphangiogenesis compounds the rate-limiting process of PDAC LN metastasis and we cocultured HLECs with PDAC cell culture medium to evaluate the role of circARFGEF2 in lymphangiogenesis (31, 32). The HLECs cocultured with cell culture medium of KRAS^G12D PDAC cells with higher circARFGEF2 expression had an increased tendency of tube formation and migration compared with those incubated with culture medium of PDAC cells with other KRAS subtypes, which possessed relatively lower circARFGEF2 expression (Supplementary Figs. S4A–S4C). Moreover, circARFGEF2 downregulation in KRAS^G12D PDAC cells suppressed HLEC tube formation and migration ability, whereas circARFGEF2 overexpression promoted KRAS^G12D PDAC cell-induced HLEC tube formation and migration (Fig. 3A and B; Supplementary Figs. S4D and S4E). 5-Ethynyl-2′-deoxyuridine (EDU) and Cell Counting Kit-8 (CCK8) assays showed that circARFGEF2 upregulation significantly increased the proliferation ability of HLECs, whereas circARFGEF2 downregulation exhibited the opposite effects (Supplementary Figs. S4F–S4I). Nevertheless, neither circARFGEF2 overexpression nor downregulation in PDAC cells with other KRAS subtypes affected HLEC tube formation and migration (Supplementary Figs. S5A–S5D). We further evaluated whether QKI-5 regulated the circARFGEF2-induced tube formation and migration in KRAS^G12D PDAC and demonstrated that QKI-5 overexpression enhanced HLEC tube formation and migration ability while downregulating circARFGEF2 expression suppressed the QKI-5-induced lymphangiogenesis of HLECs (Fig. 3C; Supplementary Fig. S5E). These findings suggest that QKI-5-mediated circARFGEF2 mainly promotes lymphangiogenesis of KRAS^G12D PDAC in vitro.

To explore the effect of QKI-5-mediated circARFGEF2 on KRAS^G12D PDAC LN metastasis in vivo, we established popliteal LN metastasis models by injecting vector/GFP-, QKI-5/GFP- or QKI-5+sh-circARFGEF2/GFP-transfected PANC-1 cells into the foot pads of nude mice (Supplementary Fig. S5F). The LN status of the mice was monitored weekly with an In Vivo Imaging System (IVIS). QKI-5 overexpression significantly promoted the LN metastasis of KRAS^G12D PDAC cells, whereas downregulating circARFGEF2 obviously reversed this effect (Fig. 3D). IHC analysis of the primary tumors demonstrated that decreasing circARFGEF2 expression reversed the increment of MLD induced by QKI-5 upregulation (Fig. 3E). Moreover, QKI-5 overexpression shortened the mice survival whereas circARFGEF2 downregulation significantly reversed this effect (Supplementary Fig. S5G).

To simulate the anatomy and physiology of LN metastasis in vivo, we established the orthotopic xenograft model through implanting PANC-1 cells transfected with vector/GFP, QKI-5/GFP, and QKI-5+sh-circARFGEF2/GFP lentiviral plasmids into the pancreas of nude mice. PET-CT scanning was used to evaluate the orthotopic tumorigenicity of nude mice and the ¹⁸F-fluorodeoxyglucose (¹⁸FDG) accumulation of the orthotopic tumors in the pancreas were shown. The results showed that the PANC-1-QKI-5 group had higher accumulation of ¹⁸FDG than the PANC-1-Vector group, while downregulating circARFGEF2 expression significantly reversed QKI-5-induced ¹⁸FDG accumulation, suggesting that QKI-5 promoted the orthotopic tumorigenicity of KRAS^G12D PDAC cells and this effect was impeded when circARFGEF2 was downregulated (Fig. 3F). Given that the pyloric, hilar, and superior mesenteric LNs represent the most common drainage LNs of PDAC in mice, we enucleated these LNs to evaluate the metastasis of LNs (33). The results showed that QKI-5 significantly promoted the metastasis of cancer cells to peripancreatic LNs and increased the number of metastatic LNs in orthotopic xenograft PDAC model while decreasing circARFGEF2 expression reversed these effects (Fig. 3G). Taken together, the above results suggest that QKI-5-mediated circARFGEF2 promotes KRAS^G12D PDAC LN metastasis in vivo.

We investigated the mechanism of circARFGEF2-induced KRAS^G12D PDAC LN metastasis. As circRNA localization within cells is crucial for their function (7), we conducted FISH and subcellular fractionation assays and found circARFGEF2 mainly located in the cytoplasm of KRAS^G12D PDAC cells (Fig. 4A and B; Supplementary Fig. S6A). Given that circRNAs located in the cytoplasm mainly sponge miRNAs, we explored the predicted binding miRNAs of circARFGEF2 with CircInteractome, miRanda, and RNAhybrid and found four miRNAs (hsa-miR-1205, hsa-miR-145–5p, hsa-miR-330–5p, and hsa-miR-636) to be potentially targeted by circARFGEF2 (Supplementary Fig. S6B). Among these miRNAs, miR-1205 was most specifically overexpressed in KRAS^G12D PDAC cell lines rather than PDAC cell lines with other KRAS subtypes (Supplementary Figs. S6C and S6D). RNA pulldown assay revealed that miR-1205 was significantly enriched by circARFGEF2 probes in the KRAS^G12D PDAC cells rather than KRAS^WT PDAC cells as compared with the Oligo probes (Fig. 4C; Supplementary Figs. S6E and S6F). Analysis of the circARFGEF2 and miR-1205 sequences revealed the existence of reverse complement sequences (Fig. 4D; Supplementary Fig. S6G). Dual luciferase reporter assay revealed that miR-1205 dramatically decreased the luciferase activity of the circARFGEF2 group, whereas the mutation of the reverse complement sequences impaired this effect (Fig. 4E; Supplementary Fig. S6H), suggesting that circARFGEF2 sponged miR-1205 via these specific sequences. Pulldown assay with miR-1205 probes demonstrated that miR-1205 specifically captured circARFGEF2 (Supplementary Fig. S6I). Furthermore, FISH assays demonstrated circARFGEF2 and miR-1205 colocalized in the cytoplasm of KRAS^G12D PDAC cells (Fig. 4F), indicating the interaction between circARFGEF2 and miR-1205.

Considering that circARFGEF2 acted as a miR-1205 sponge in KRAS^G12D PDAC cells, we investigated whether circARFGEF2 promoted KRAS^G12D PDAC lymphangiogenesis by sponging miR-1205. In vitro experiments demonstrated that overexpressing circARFGEF2 triggered HLEC tube formation and migration, whereas overexpressing miR-1205 obviously reversed this effect (Fig. 4G; Supplementary Figs. S6J–S6L). Taken together, our results demonstrate that circARFGEF2 acts as a miR-1205 sponge in KRAS^G12D PDAC cells.

As circRNAs act as miRNA sponges to regulate downstream genes expression (34), we conducted mRNA transcriptome analysis in PANC-1 cells to identify the targets of circARFGEF2. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis revealed that the JAK–STAT pathway was the most significantly enriched pathway associated with circARFGEF2 (Fig. 4H). JAK represents the core component for the activation the JAK–STAT pathway and consists of JAK1, JAK2, JAK3, and TYK2 (35, 36). Subsequently, we detected the expression profiles of the four JAK subtypes and found that JAK2 was positively associated with the expression of circARFGEF2 in KRAS^G12D PDAC cells (Fig. 4I; Supplementary Fig. S6M). Moreover, miR-1205 was negatively associated with JAK2 expression in KRAS^G12D PDAC cells (Fig. 4J and K). miRNAs regulate target genes expression mainly by binding to their 3′UTR (37). Analysis of the JAK2 3′UTR region revealed a sequence complementary to miR-1205 (Supplementary Fig. S7A). The dual luciferase reporter assays demonstrated that miR-1205 obviously decreased the luciferase activity in the JAK2 3′UTR luciferase construct rather than that with a mutant sequence in the miR-1205 binding site (Fig. 4L). miR-1205 overexpression obviously inhibited circARFGEF2-induced JAK2 expression, suggesting that circARFGEF2 acted as a miR-1205 sponge to increase JAK2 expression (Fig. 4M and N; Supplementary Fig. S7B). Western blotting assays revealed that circARFGEF2 overexpression promoted JAK2 expression and STAT3 phosphorylation (Fig. 4O; Supplementary Figs. S7C and S7D), whereas circARFGEF2 downregulating reduced JAK2 expression and inhibited STAT3 phosphorylation (Fig. 4P; Supplementary Figs. S7E and S7F), the full uncut gels of Western blotting were shown in Supplementary Fig. S10. In addition, we investigated the effect of circARFGEF2 on other KRAS downstream effectors, including ERK, AKT, PKCs, TBK1, and p65 (30, 38). Neither circARFGEF2 downregulation nor upregulation showed significant effect on the expression of ERK, AKT, PKC, TBK1, and p65 (Supplementary Figs. S7G and S7H), the primers are listed in Supplementary Table S4. Western blotting assays showed that circARFGEF2 was positively correlated with the phosphorylation of AKT (Fig. 4O and P), which was previously demonstrated as the downstream target of JAK2, the antibodies are listed in Supplementary Table S5 (39). Altogether, these findings demonstrate that circARFGEF2 sponges miR-1205 to upregulate JAK2 expression and activate the JAK2–STAT3 pathway in KRAS^G12D PDAC.

Next, we further clarified whether JAK2–STAT3 pathway activation was indispensable for circARFGEF2-induced lymphangiogenesis and LN metastasis in KRAS^G12D PDAC. circARFGEF2 overexpression significantly promoted HLEC tube formation and migration ability, whereas the JAK2–STAT3 pathway inhibitor WP1066 significantly reversed this effect (Fig. 5A and B). Furthermore, in vivo experiments demonstrated that the circARFGEF2-overexpressing nude mice had larger popliteal LNs compared with the control nude mice, and WP1066 significantly impeded the circARFGEF2-induced LN metastasis (Fig. 5C and D). IHC analysis revealed that circARFGEF2 overexpression promoted the metastasis of popliteal LNs and increased the MLD in the foot pad tumors, while inhibiting JAK2–STAT3 pathway reversed these effects (Fig. 5E and F). These results suggest that circARFGEF2 promotes lymphangiogenesis and LN metastasis in KRAS^G12D PDAC via the JAK2-STAT3 pathway.

To better understand the role of circARFGEF2 in KRAS^G12D PDAC LN metastasis in vivo, we used the engineered transgenic Kras^G12D/+Trp53^R172H/+Pdx-1-Cre (KPC) mouse, a well-characterized autochthonous model that develops the full spectrum of KRAS^G12D mutation-induced PDAC (40, 41). The results showed that circARFGEF2 overexpression significantly increased primary PDAC tumor volume and inhibiting the JAK2–STAT3 signaling pathway reversed this effect (Fig. 6A and B). Moreover, circARFGEF2 overexpression significantly increased the number of metastatic abdominal LNs in KPC mice and the MLD in primary PDAC tumors, whereas blocking JAK2–STAT3 signaling pathway impeded these effects (Fig. 6C and D; Supplementary Fig. S7I). We also evaluated the effect of circARFGEF2 downregulation in the LN metastasis of KPC mouse with AAV-sh-circARFGEF2. PET-CT scanning was used to evaluate the tumorigenicity of PDAC tumors in KPC mice and the ¹⁸FDG accumulation of the PDAC tumors in KPC mice were shown. The results of PET-CT scanning showed that downregulating circARFGEF2 expression obviously reduced ¹⁸FDG accumulation in the PDAC tumors of KPC models compared with the control group (Supplementary Fig. S8A). Moreover, the number of metastatic LNs in sh-circARFGEF2 group was lower than sh-NC group (Supplementary Fig. S8B). Then we separated the PDAC tumors in KPC models and evaluated the MLD in PDAC tumors. HE and IHC results revealed that the MLD was significantly decreased in sh-circARFGEF2 group, indicating that circARFGEF2 downregulation impeded the lymphangiogenesis in the primary PDAC tumors of KPC models (Supplementary Figs. S8C and S8D). These results suggest that blockage of JAK2-STAT3 pathway exhibit a suppression effect on circARFGEF2-induced lymphangiogenesis and LN metastasis of KRAS^G12D PDAC.

Given that QKI-5-mediated circARFGEF2 induced KRAS^G12D PDAC lymphangiogenesis and LN metastasis by sponging miR-1205 and activating JAK2–STAT3 signaling pathway, we assessed the clinical significance of the QKI-5-mediated circARFGEF2–miR-1205–JAK2 axis in patients with KRAS^G12D PDAC. The results demonstrated that miR-1205 was downregulated and circARFGEF2, QKI-5, JAK2 were upregulated in KRAS^G12D PDAC tissues in comparison with the NATs (Fig. 6E and F; Supplementary Figs. S8E and S8F), and miR-1205 was upregulated in KRAS^G12D PDAC tissues compared with PDAC tissues with other KRAS subtypes (Supplementary Fig. S8G). Moreover, miR-1205 was negatively associated with LN metastasis while circARFGEF2, QKI-5, and JAK2 were significantly upregulated in the LN metastatic KRAS^G12D PDAC tissues (Fig. 6G and H; Supplementary Figs. S8H–S8J). In addition, we found no significant association between circARFGEF2 or miR-1205 expression and liver or lung metastasis in PDAC patients (Supplementary Figs. S8K–S8N). The above findings suggest that QKI-5-mediated circARFGEF2–miR-1205–JAK2 axis is specifically associated with LN metastasis of patients with PDAC.

IHC analysis revealed that QKI-5 and JAK2 were positively correlated with circARFGEF2 expression in the KRAS^G12D PDAC tissues (Fig. 6I–K). Correlation analysis revealed that JAK2 was negatively associated with miR-1205 expression while positively associated with QKI-5 expression, and a negative correlation was observed between circARFGEF2 and miR-1205 expression (Fig. 6L and M; Supplementary Fig. S8O). The above results indicated the critical role of the QKI-5-mediated circARFGEF2–miR-1205–JAK2 axis in KRAS^G12D PDAC LN metastasis. Together, our findings demonstrate that circARFGEF2 derived from QKI-5-dependent splicing sponged miR-1205 and activated the JAK2–STAT3 signaling pathway to induce KRAS^G12D PDAC LN metastasis.

Aberrant pre-mRNA splicing causes the production of alternative RNA variants, including circRNAs, and leads to human cancer stemness, proliferation, and metastasis (42). Nonetheless, the precise mechanism of circRNA production via pre-mRNA splicing in KRAS^G12D PDAC LN metastasis remains unexplored. In this study, we determined that circARFGEF2 was overexpressed via a KRAS^G12D mutant-related splicing in PDAC and triggered the lymphangiogenesis and LN metastasis of KRAS^G12D PDAC cells both in vitro and in vivo.

We constructed a dual-color fluorescence reporter system to demonstrate that the circARFGEF2 biogenesis was significantly activated by the alternative splicing factor QKI-5. QKI-5 promoted circARFGEF2 biogenesis by recruiting U2AF35 and regulating ARFGEF2 pre-mRNA back-splicing. Subsequently, circARFGEF2 sponged miR-1205 and activated the JAK2-STAT3 signaling pathway to induce KRAS^G12D PDAC lymphangiogenesis and LN metastasis. Our study highlighted a mechanism of QKI-5-mediated circRNA production in regulating KRAS^G12D PDAC LN metastasis, indicating that targeting circARFGEF2 production is a potential therapeutic strategy for LN metastatic PDAC with KRAS^G12D mutation.

KRAS mutation is the most frequently mutated oncogene, occurring in approximately 90% of patients with PDAC, where KRAS^G12D is the most common allele in PDAC (43). Recent studies revealed that KRAS^G12D mutation caused the aberrant gene expression profile by regulating RNA processing to drive tumor progression (9, 17). However, the role of the KRAS^G12D mutation in triggering the circRNA process during tumor development remains unknown. In this study, circRNA production was responsive to the KRAS^G12D mutation in PDAC and was crucial in PDAC LN metastasis. Moreover, we demonstrated that the RNA splicing regulator QKI-5 was associated with KRAS^G12D and revealed that the KRAS^G12D mutation promoted circARFGEF2 biogenesis by triggering QKI-5-induced back-splicing of circARFGEF2. Targeting the KRAS^G12D mutation with a specific inhibitor dramatically abolished circARFGEF2 production and its role in fostering PDAC LN metastasis. Our study clarified the precise mechanism underlying KRAS^G12D mutation-induced circRNA production, suggesting a new perspective in KRAS^G12D PDAC LN metastasis.

QKI is a potential regulator of pre-mRNA alternative splicing in developmental and pathologic processes. QKI has three major alternatively spliced isoforms (QKI-5, QKI-6, QKI-7) that differ at the carboxy-terminal end and contribute to different RNA-processing functions (28). QKI-5 is predominantly located in the nucleus and regulates pre-mRNA splicing (20, 21). QKI-6 and QKI-7 lack nuclear localization signals and regulate posttranscriptional mRNA processing, such as RNA stability and transportation (44, 45). Notably, all three isoforms regulate the biogenesis of certain circRNAs by regulating alternative pre-mRNA splicing (28, 46). Nevertheless, the specific isoform that regulates circRNA biogenesis in LN metastatic KRAS^G12D PDAC is unclear. In this study, we determined that QKI-5 facilitated circARFGEF2 production in KRAS^G12D PDAC by interacting with the splicing factor U2AF35. Moreover, U2AF35 assisted with QKI-5 in recognizing the intron 3^{1128–1180 nt} and intron 6^{7672–7740 nt} of ARFGEF2 pre-mRNA, which benefited ARFGEF2 pre-mRNA cyclization and back-splicing. Decreasing U2AF35 expression significantly impaired the binding between QKI-5 and the flanking introns of ARFGEF2 pre-mRNA and inhibited circARFGEF2 biogenesis. Our study revealed the specific QKI-5 recognition and splicing pattern that regulated circARFGEF2 production in LN metastatic KRAS^G12D PDAC, which expanded our understanding of QKI-5-mediated circRNA splicing.

circRNAs have been acknowledged as feasible therapeutic targets that exhibited satisfactory tumor suppression effects in mouse models, indicating the broad prospect of circRNAs in cancer treatment (47, 48). KRAS mutant cells exhibit different gene expression profiles, including circRNAs, according to a recent study (9). KRAS-responsive circRNAs promote target gene stabilization and control extracellular exportation to promote tumorigenesis (17). However, the relationship between circRNA biogenesis and KRAS^G12D PDAC is still unclear. In our study, we discovered that circRNA biogenesis was strongly related to the KRAS^G12D mutation in PDAC and was essential for the LN metastasis of PDAC. We identified circARFGEF2, a KRAS^G12D-responsive circRNA, was involved in KRAS^G12D-mediated LN metastasis of PDAC. The overproduction of circARFGEF2 was induced by a RNA splicing factor, QKI-5, which was overexpressed in KRAS^G12D PDAC. Importantly, we determined that blocking QKI-5-induced circARFGEF2 biogenesis significantly inhibited KRAS^G12D PDAC LN metastasis in the mice model. Thus, further developing therapeutic strategy targeting KRAS^G12D-driven circRNA biogenesis might achieved satisfied efficiency in the intervention of LN metastatic KRAS^G12D PDAC.

We established a KPC mouse model to evaluate the therapeutic efficiency of circARFGEF2 in KRAS^G12D PDAC LN metastasis. In KPC models, increasing circARFGEF2 expression markedly accelerated tumor growth, whereas inhibiting JAK2-STAT3 reversed this effect. In addition, blocking JAK2-STAT3 obviously impeded the promotion of circARFGEF2 in the lymphangiogenesis of primary PDAC tumors in the KPC model. Overall, our results provided sufficient support for the preclinical treatment effect of circARFGEF2 in KRAS^G12D PDAC, indicating that circARFGEF2 is a potential therapeutic target in KRAS^G12D PDAC.

In summary, we proposed a novel mechanism of KRAS^G12D mutation-driven circARFGEF2 biogenesis via QKI-5-dependent splicing that mediated the miR-1205–JAK2–STAT3 axis to trigger PDAC LN metastasis. Inhibiting circARFGEF2 expression or the JAK2–STAT3 pathway significantly impeded tumor growth in the KPC model. These findings enabled new understanding of the regulatory mechanism in PDAC LN metastasis, suggesting that circARFGEF2 is a potential therapeutic target in KRAS^G12D PDAC.

No author disclosures were reported.

Y. Kong: Writing–original draft. Y. Luo: Investigation, methodology. S. Zheng: Data curation, validation. J. Yang: Project administration, writing–review and editing. D. Zhang: Data curation, formal analysis. Y. Zhao: Supervision, visualization. H. Zheng: Validation, project administration. M. An: Resources, software. Y. Lin: Validation, investigation, methodology. L. Ai: Software, formal analysis, methodology. X. Diao: Supervision, methodology. Q. Lin: Data curation, validation. C. Chen: Writing–original draft, project administration. R. Chen: Conceptualization, funding acquisition.

This study was funded by National Key Research and Development Program of China (grant no. 2022YFA1305500); the National Key Research and Development Program of China (grant no. 2018YFA0902803); National Natural Science Foundation of China (grant nos. 82072639, 82173272, 82103536, 82202276, 82103416, and 82203691); Guangdong Basic and Applied Basic Research Foundation (grant nos. 2023A1515011648, 2021B1515020091, 2022A1515012288, and 2021A1515010355); the Special Fund of “Dengfeng Plan” of Guangdong Provincial People's Hospital, China (grant nos. KJ012019509, DFJH2020027); the National Key Clinical Specialty Construction Project (grant no. 2022YW030009); China Postdoctoral Science Foundation (grant no. 2022T150147); and Guangdong Science and Technology Department (grant no. 2021A1515011089).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.