Hypoxia-induced exosomal circPDK1 promotes pancreatic cancer glycolysis via c-myc activation by modulating miR-628-3p/BPTF axis and degrading BIN1

Pancreatic cell lines were incubated under normoxic conditions until they reached 70–80% confluence and then cultured in exosome-free medium under 20% (normoxia) or 1% (hypoxia) O₂ for 48 h. Next, 20 mL of culture medium was harvested on ice and isolated by differential centrifugation [17]. The resulting sediment was resuspended with PBS. Exosome particle size and form, markers, and concentration were identified by transmission electron microscopy (TEM) (TECNAI 20; Philips, Netherland), Western blotting assay, and nanoparticle tracking analysis (NTA), respectively. Cells were collected for subsequent experiments after stimulation with 1 × 10⁸ exosomes for 48 h.

All cells, including Aspc-1, Bxpc-3, CFPAC-1, MIA PaCa-2, PANC-1, PATU-8988, 293 T, and HPNE cells, were purchased from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). Aspc-1 and Bxpc-3 cells were cultured in RPMI-1640 medium, CFPAC-1 in IMDM, MIA PaCa-2, PANC-1, PATU-8988, HPNE, and 293 T cells in DMEM with 10% fetal bovine serum (FBS).

miR-628-3p mimics, miR-628-3p inhibitor, circPDK1 siRNA (si-circPDK1), UBE2O siRNA (si-UBE2O), HIF1A (si-HIF1A), circPDK1-WT, BPTF, circPDK1-MUT, circPDK1 truncations, BIN1 deletion mutations, and BIN1 and UBE2O were synthesized by Bioegene (Shanghai, China) and inserted into either lentivectors or plasmids. After transduction with lentivirus, 2 µg/mL puromycin was added to construct stable cell lines for 24 h. The miRNA mimics and siRNA sequences are listed in Additional file 1: Table S1, Additional file 2: Table S2. The circPDK1-WT and circPDK1-MUT sequences are listed in Additional file 3: Table S5. Lipofectamine 3000 was used as a transfection reagent for cell transfection.

Cell viability was assessed using a CCK-8 assay (Dojindo, Japan). PC cells were harvested after treatment for 48 h, and 2 × 10³ PC cells were seeded in 96-well plates and incubated further. Then, 90 µL of completed medium with 10 µL of CCK-8 reagent was added and incubated for 2 h, after which the absorbance was measured at 450 nm. For the colony formation assay, 2 × 10³ PC cells were seeded in 6-well plates to evaluate PC cell proliferation ability. After 14 days, 1% crystal violet stain solution was used to fix the PC cells, and the number of colonies was counted. For the EdU assay, the EdU labeling kit (Epizyme, Shanghai, China) was used to assess cell proliferative ability. PC cells (3 × 10⁴) were seeded in 12-well plates for 48 h. The PC cells were then incubated with EdU reagent for 2 h, fixed with 4% paraformaldehyde, 0.5% Triton X-100, and followed by Hoechst staining. The EdU incorporation rate was defined as the proportion of EdU-positive cells (GREEN) to total Hoechst33342-positive cells (BLUE).

The migration ability of PC cells was assessed by transwell migration assay. Briefly, transwell chambers were plated with 6 × 10⁴ PC cells suspended in 200 µL serum-free medium and 700 µL complete medium in the lower chambers. After 24 h of culture, the upper chambers with migrated PC cells were fixed and stained with 1% crystal violet stain solution for 20 min, after which the migrated PC cells were counted manually by microscope. Relative proportions of migrated cells were calculated as the ratios (%) of treated to negative control cells.

Cell proteins were boiled with RIPA buffer, loaded, and separated on 6%, 7.5%, and 10% SDS-PAGE gels, followed by their transfer onto polyvinylidene fluoride membranes and incubation with the indicated primary antibodies (Additional file 4: Table S3). β-actin and β-tubulin were used as controls. ECL reagents were used to detect the protein expression. All Western blotting reagents were purchased from EpiZyme (Shanghai, China). The IHC assay was performed using mouse xenograft tumor tissues and human PC tissues, as previously described [18]. IF assay was used to determine the subcellular localization of BIN1 and UBE2O in MIA PaCa-2 cells, as previously described [19]. Using FISH assay, the subcellular location of circPDK1 in MIA PaCa-2 cells was detected, as previously described [18]. Cy3-labeled circPDK1 probes were synthesized by RiboBio (Guangzhou, China). ISH assay was performed on the tissue microarrays to detect circPDK1 in PC tissues, as previously described [17], and the specific digoxin (DIG)-labeled probe of circPDK1 was designed by Servicebio (Wuhan, China). Protein–protein interactions were detected by performing the Co-IP assay, as previously described [19].

The Glucose Uptake Assay Kit (abcam, USA), ATP Assay Kit (Beyotime, Shanghai, China), and Lactate Assay Kit-WST (Dojindo, Shanghai, China) were performed to assess the glycolysis level according to the manufacturer’s guidelines. Using the Seahorse XF96 Glycolysis Analyzer (Seahorse Bioscience, MA, USA), the ECAR and OCAR were measured according to the manufacturer’s guidelines.

RNA was extracted using TRIzol reagent (Invitrogen, USA), and the HiScript III RT SuperMix (TOYOBO, Japan) was used for reverse transcription, followed by the AceQ Universal SYBR qPCR Master Mix (AG, China) to detect the RNA expression levels, and normalized to β-actin or U6. The divergent and convergent primers of circPDK1 and GAPDH were used to detect the circular characteristics of circPDK1 by RT-PCR. The primer sequences are listed in Additional file 5: Table S4. Using the PARIS Kit according to the manufacturer’s guidelines, RNA from subcellular fractionation was separated to assess the relative subcellular localization of circPDK1.

Luciferase reporter vectors containing the 3ʹ-UTR binding site of circPDK1 and BPTF or the matched mutant sequence, c-myc-responsive transcriptional sequence, wild-type miR-628-3p transcriptional promoter or circPDK1 transcriptional promoter, and circPDK1 corresponding promoter mutant sequence were synthesized by Bioegene (Shanghai, China) and cloned into the reporter plasmids. Luciferase activity was assessed using a dual-luciferase reporter assay system (Synergy LX; BioTek, USA) and normalized to Renilla.

The SimpleChIP Plus Sonication Chromatin IP Kit (CST, USA) was used to perform the ChIP assay according to the manufacturer’s instructions. Using 1% formaldehyde, MIA PaCa-2 cells were cross-linked for 10 min, followed by quenching with glycine. DNA fragments were sheared by sonication until they ranged from 200 to 500 bp. The nuclear lysate was immunoprecipitated with anti-HIF1A, anti-Pol II, or IgG antibodies. The purified DNA fragments were analyzed by qRT-PCR with specific primers; the ChIP primers are listed in Additional file 5: Table S4.

RIP experiments were performed using the Magna RNA-binding protein immunoprecipitation kit (Millipore, Bedford, MA, USA), according to the manufacturer’s instructions. The immunoprecipitated RNA in MIA PaCa-2 and 293 T cells was verified by qRT-PCR.

The biotinylated RNAs were pulled down by incubating the MIA PaCa-2 cell lysates with streptavidin–agarose beads (Invitrogen, USA) according to the manufacturer’s instructions. The eluted proteins were identified by mass spectrometry analysis and Western blotting. circPDK1-Positive-probe-biotin (CTGGTGATTTTGCTTAATGTAGAT) and circPDK1-Negative-probe-biotin (GCAGTTATCTACATTAAGCAAAAT) were designed by RiboBio (Guangzhou, China). The mass spectrometry (MS) findings are provided as in Additional file 6: Table S8.

To evaluate the functions of exosomes in vivo, wild-type MIA PaCa-2 (5 × 10⁶ cells/0.1 mL PBS; n = 5 in every group) were injected subcutaneously on the right flank of 4-week-old male BALB/c nude mice. Once the subcutaneous tumor volume reached 100 mm², the mice were randomly divided into four groups: 1 × 10⁸ exosomes/100 μL corresponding exosomes were injected into the tail vein every three days, and tumor volume was measured. After six injections, the mice were killed on the third day after the last injection. The subcutaneous tumors were weighed, fixed, and stained using IHC. To evaluate the migration of exosomes in vivo, a lung metastasis model was developed. Wild-type MIA PaCa-2 (1 × 10⁶ cells/0.1 mL PBS; n = 3 in every group) with indicated 1 × 10⁸ exosomes/100 μL exosomes were injected into the tail vein. The indicated exosomes were injected into the tail vein of the mice every seven days. After six injections, mice were killed on the third day after the last injection, and lung tissue was photographed and stained with hematoxylin and eosin (HE).

To evaluate the functions of circPDK1-WT and circPDK1-MUT in vivo, stably transfected MIA PaCa-2 cells (5 × 10⁶ cells/0.1 mL PBS; n = 5, every group) overexpressing circPDK1-WT or circPDK1-MUT were injected into the right flank of 4-week-old male nude mice. The subcutaneous tumor volume was calculated every 6 days, and the mice were killed on day 30. Subcutaneous tumors were weighed and subjected to immunohistochemistry (IHC). For the lung metastasis model, stably transfected MIA PaCa-2 cells (2 × 10⁶ cells/0.1 mL PBS; n = 3, every group) were injected into the tail vein of 6-week-old mice. The mice were killed on day 42, lung tissue was photographed, stained with HE, and lung metastasis nodes were counted.

BIN1 protein was predicted by the High-Performance Computing Center of Shanghai Jiao Tong University through artificial intelligence simulation and homology modeling. The 3D modeling of RNA and protein-RNA docking steps were consistent with previous reports [20‐22].

Hypoxia is an important characteristic of pancreatic cancer. Primary tumors with PO₂ < 10 mmHg in patients were defined as hypoxic, since oxygen availability would decrease as the distance from tumor cells to the nearest blood vessels increases. This distance is 100–200 μm depending on the local oxygen concentration in the blood and oxygen consumption rates [23‐25]. Cells cultured under 1% and 20% O₂ conditions were defined as hypoxic and normoxic, respectively. Thus, PC cells cultured in different oxygen concentrations could be defined as hypoxic and normoxic, which might cause tumor heterogeneity among them, and the malignant potential of hypoxic PC cells might be higher (Additional file 7: Fig. S1A). To determine whether exosomes secreted from hypoxic PC cells could affect the malignant biological behavior of normoxic PC cells, exosomes derived from hypoxic PC cells and normoxic PC cells were identified by TEM and NTA (Additional file 7: Fig. S1B, C), which demonstrated typical round particles with a diameter of 100 nm. The results of Western blotting showed that the exosome markers TSG101, CD63, and CD81 were substantially upregulated in hypoxic exosomes (Additional file 7: Fig. S1D). These results indicated that hypoxia could increase exosome secretion in hypoxic PC cells, which was consistent with previous studies of other cancers [26, 27].

Next, we detected the malignant biological behavior of PC cells after treatment with normoxic exosomes or hypoxic exosomes by colony formation, EdU, CCK-8, and transwell assays. The results showed that compared with the PBS control or normoxic exosomes, exosomes derived from hypoxic PC cells considerably improved the proliferation and migration of MIA PaCa-2 and PANC-1 cells (Additional file 7: Fig. S1E–H). Western blotting also showed that E-cadherin was significantly inhibited after treatment with hypoxic exosomes compared to PC cells with normoxic exosomes. Conversely, N-cadherin and vimentin protein levels were increased (Additional file 7: Fig. S1I). This suggests that exosomes derived from hypoxic PC cells could promote the viability and migration of PC cells.

It has been widely confirmed that circRNAs are highly abundant in exosomes stably, which could be a key regulator of cancer [28]. To elucidate the role of hypoxia-induced exosomal circRNAs that may affect tumorigenesis, RNA sequencing was used to detect exosomes purified from normoxic and hypoxic PC cells. A total of 78 differentially expressed circRNAs were identified (FC (fold change) ≥ 1.5, P < 0.05) between hypoxic exosomes and normoxic exosomes (Fig. 1A, B).

Next, the top five upregulated and downregulated circRNAs were analyzed by qRT-PCR in 10 cases of PC tumors and paired normal tissues. We discovered that circPDK1 was the most upregulated circRNA in PC (Fig. 1C). Then, an ISH assay in a tissue microarray with 75 pairs of tumor and matched normal tissues from PC patients was performed. The results revealed that circPDK1 was much more abundant in tumor tissues than in paired normal tissues (Fig. 1D), and the high expression of circPDK1 may be correlated with advanced pathological stage (Fig. 1E). Subsequently, 110 paired postoperative PC tissues and matched normal tissues from our center were detected using qRT-PCR to analyze the expression of circPDK1. The results demonstrated that circPDK1 was significantly upregulated in PC and was positively associated with lymph node metastasis, advanced pathological stage, and poor prognosis (Fig. 1F–I). To evaluate the clinicopathological and prognostic significance of circPDK1 in PC, clinical data were collected and analyzed in detail. We found that circPDK1 expression was significantly related to pathological stage (P = 0.019), T stage (P = 0.022), lymph node metastasis (P = 0.012), and distant metastasis (P = 0.034) (Additional file 8: Table S6). Furthermore, univariate analysis results demonstrated that high circPDK1 expression was an independent prognostic marker for patients with PC (Additional file 9: Table S7). CircRNA has been widely detected in serum exosomes and is considered a potential tumor marker in serum exosomes [17]. Thus, to explore whether circPDK1 could be detected in serum exosomes and its potential as a tumor marker, we collected blood samples from 20 patients with PC and 10 healthy individuals as controls. Interestingly, circPDK1 derived from serum exosomes was expressed abundantly in PC patients, while it almost did not exist in healthy controls (Fig. 1J). In addition, circPDK1 expression levels in serum exosomes were consistent with those in matched PC tumors (Fig. 1K), which made it possible to examine circPDK1 expression in serum samples. Taken together, these data reveal that circPDK1 is significantly upregulated in PC tumor tissues, which could be stably delivered and enriched in serum exosomes. Moreover, circPDK1 is barely expressed in serum exosomes and high circPDK1 expression is associated with poor prognosis in PC, which makes it a potentially promising circRNA biomarker for early diagnosis of PC.

circPDK1 (hsa_circRNA_102854, has_circ_0057104, chr2: 173433468–173457776) was generated from exons 7‒10 of PDK1 with a length of 401 nt. The back-splice site of circPDK1 was verified by Sanger sequencing (Fig. 1L). The divergent primers for circPDK1 could be amplified by PCR analysis, from cDNA but not gDNA; however, the divergent primers for GAPDH as a control could not be amplified (Fig. 1M). Compared with linear PDK1 mRNA, treatment with RNase R or actinomycin D demonstrated that circPDK1 was stabilized in PC cells (Fig. 1N, O). Subcellular fractionation and FISH assay results showed that circPDK1 was mainly localized in the cytoplasm (Fig. 1P, Q). Collectively, these results suggest that circPDK1 is an abundant, stably expressed, and cytoplasmic circRNA in PC.

Studies have demonstrated that large amounts of circRNAs are transcriptionally activated by HIF1A under hypoxic conditions, based on exon-derived circRNAs, and their host linear genes would be generated from the same pre-mRNA [29, 30].Thus, we hypothesized that abundant circPDK1 in hypoxic exosomes may be due to the increased transcription of its host linear gene PDK1 activated by HIF1A. To verify our hypothesis, we first detected the co-expression relationship between PDK1 and circPDK1. The results revealed that circPDK1 expression level was positively correlated with its host linear gene PDK1 expression level (Fig. 2A). Moreover, compared with circPDK1 expression across the seven PC cell lines, circPDK1 expression was significantly non-differential (Fig. 2B). Next, we found that circPDK1 expression levels were notably overexpressed in both cells and exosomes during hypoxia in a time-dependent manner (Fig. 2C, D). Using qRT-PCR and IHC, we revealed that circPDK1 expression was positively associated with HIF1A expression, and HIF1A protein level was upregulated in PC (Fig. 2E–G). Using promoter sequence analysis tools (UCSC and JASPAR), two potential hypoxia-responsive elements (HREs) (ACACGTGCCC/GTACGTGAGG) were verified in the host gene of circPDK1 promoter regions (Fig. 2H). To determine whether HIF1A could directly interact with the two transcription start sites, a dual-luciferase reporter assay was performed in 293 T cells. The results confirmed that luciferase activity was enhanced in the reporter containing two individual wild-type HREs, while the reporter containing mutant HREs did not respond to hypoxia or HIF1A knockdown (Fig. 2I). Moreover, ChIP assays performed on the two predicted HREs upstream of the host gene for the circPDK1 promoter area suggested that HIF1A could directly bind to the PDK1 promoter region and enhance the transcription of host gene PDK1 under hypoxic conditions (Fig. 2J). Furthermore, the abundance of hypoxia-induced intracellular and exosomal circPDK1 was notably inhibited by knockdown of HIF1A (Fig. 2K–M). The ChIP assay results also verified that the binding between RNA polymerase II and the host gene PDK1 promoter site was enhanced under hypoxic conditions, which further verified that circPDK1 could be activated under hypoxia (Fig. 2N). In addition, the circPDK1 expression level in PC cells cultured with hypoxic exosomes was partly inhibited by treatment with ActD, and the circPDK1 stability in PC cells did not be affected under hypoxia (Fig. 2O and Additional file 10: Fig. S2A). Overall, these results indicate that the upregulation of circPDK1 in hypoxic exosomes is partly due to HIF1A activation.

To elucidate the role of hypoxia-derived exosomal circPDK1 in PC, exosomes with different abundances of circPDK1 derived from MIA PaCa-2 and PANC-1 cells were isolated and co-cultured with the corresponding donor cells, including four experimental groups: normoxic exosomes, hypoxic exosomes, hypoxic NC exosomes, and hypoxic sh1-circPDK1 exosomes. Using qRT-PCR, the abundance of circPDK1 in exosomes from each group was confirmed (Additional file 10: Fig. S2B). After treatment with the indicated exosomes, the abundance of circPDK1 in PC cells co-cultured with hypoxic sh1-circPDK1 exosomes was downregulated compared with that in the hypoxic NC exosomes group (Additional file 10: Fig. S2C).

Next, colony formation, EdU, CCK-8, and transwell assays were performed to detect the impact of hypoxia-induced exosomes on normoxic PC cell proliferation and migration in vitro. The results showed that PC cells treated with hypoxic exosomes showed enhanced viability and migration, whereas, after the elimination of circPDK1 from the hypoxic exosomes, this phenomenon of promoting proliferation and migration would disappear (Fig. 3A–D). In addition, Western blotting demonstrated that E-cadherin was significantly inhibited after treatment with hypoxic NC exosomes compared to PC cells with hypoxic sh1-circPDK1 exosomes. On the contrary, compared to treatment with hypoxic sh1-circPDK1 exosomes, N-cadherin and vimentin were notably upregulated in PC cells treated with hypoxic NC exosomes (Fig. 3E).

To explore the malignant effects of exosomal circPDK1 in vivo, a mouse xenograft tumor model (MIA PaCa-2 cell line) was established and intravenously injected with MIA PaCa-2 generated exosomes (normoxic exosomes, hypoxic exosomes, hypoxic NC exosomes, and hypoxic sh1-circPDK1 exosomes) once every three days. After six injections, the mice were killed on day 30 with the tumors harvested. As expected, hypoxic NC exosomes increased the volume and weight of xenograft tumors, while hypoxic sh1-circPDK1 exosomes abolished hypoxic exosomes to accelerate tumor growth (Fig. 3F–H). There was no significant difference in the body weights of nude mice among the groups (Fig. 3I). In the lung metastasis model, the number of lung metastatic nodules in the hypoxic sh1 exosome group was lower than that in the hypoxic NC group (Fig. 3J). To validate the functions of exosomal circPDK1 in mediating tumor growth and metastasis, xenograft tumor tissues of MIA PaCa-2 groups were stained with antibodies against PCNA, E-cadherin, and vimentin for IHC. The results showed that the expression of PCNA and vimentin was decreased in hypoxic sh1-circPDK1 exosome groups, whereas the expression of E-cadherin was increased in hypoxic sh1-circPDK1 exosome groups (Fig. 3K). These findings are consistent with the in vitro results showing that hypoxia-induced exosomal circPDK1 could also promote PC cell tumor growth and metastasis in vivo.

Many cytoplasmic circRNAs have been reported to function as ceRNAs by competitively sponging miRNAs [31], and given that circPDK1 is mainly localized in the cytoplasm, we assumed that cytoplasmic localization circPDK1 may act as a ceRNA by competitively sponging miRNAs. By researching the circinteractome online database (https://​circinteractome.​nia.​nih.​gov/​index.​html), we found that circPDK1 may bind to AGO2 (Fig. 4A). Thus, whether circPDK1 was enriched in AGO2-containing microribonucleoprotein complexes was verified by RIP assay (Fig. 4B). The findings revealed that circPDK1 may act as a miRNA sponge. Thus, we performed miRNA sequencing to screen potential differentially expressed miRNAs between the control groups and circPDK1 overexpression groups, which may be sequestered by circPDK1, and the efficiency of circPDK1 overexpression was also detected (Fig. 4C and Additional file 10: Fig. S2D). Pathway enrichment analysis suggested that differentially expressed miRNAs were enriched in ubiquitin-mediated proteolysis, PI3K-AKT signaling pathway, and HIF-1 signaling pathway, among others (Fig. 4D). Seven suppressed miRNAs were selected randomly, and then qRT-PCR was performed, and the expression levels of these miRNAs were confirmed in circPDK1 overexpressing MIA PaCa-2 cells. Among the seven candidates, miR-628-3p was shown to be significantly suppressed in circPDK1-overexpressing MIA PaCa-2 cells (Fig. 4E). RIP assays also demonstrated that miR-628-3p could bind to AGO2 (Fig. 4F). We found that miR-628-3p was notably suppressed in PC tumors in both TCGA data and our center, and inversely correlated with circPDK1 expression (Fig. 4H). Moreover, low miR-628-3p expression was associated with poor prognosis using clinical data from TCGA and our center (Fig. 4I). circRNAs can bind to RNA polymerase II directly, and the transcription of parental genes is activated [32]. To eliminate the possibility that circPDK1 modulates miRNA transcription, pri-miRNA, or pre-miRNA synthesis, we verified the impact of circPDK1 on the promoter regions, pri-miRNA, and pre-miRNA of miR-628-3p and confirmed that circPDK1 did not regulate promoter activity or pri-miRNA and pre-miRNA expression of miR-628-3p (Fig. 4J). miR-628-3p was downregulated after circPDK1 overexpression but upregulated in sh1-circPDK1 group in both MIA PaCa-2 and PANC-1 cells (Fig. 4K). Dual-luciferase reporter assays were performed, and the results showed that miR-628-3p directly binds to circPDK1 in 293 T cells (Fig. 4L). These findings indicate that circPDK1 may interact with miR-628-3p, thus functioning as a ceRNA sponge for miR-628-3p in PC cells.

The underlying functions and mechanisms of action of miR-628-3p in PC remain unknown. Thus, potential targets of circPDK1-miR-628-3p axis were explored. Using the mirDIP, RNAInter, miRDB, miRTarBase, and TargetScan databases, BPTF, PURA, RBFOX2, and ARSJ were predicted to be potential downstream targets of miR-628-3p (Fig. 5A). We used data from TCGA to primarily detect a correlation between the expression of potential downstream targets and miR-628-3p. The results showed that the expression of BPTF was inversely associated with miR-628-3p, whereas that of PURA, RBFOX2, and ARSJ was not (Fig. 5B and Additional file 11: Fig. S3A–C). The RNA-sequencing findings after the loss of circPDK1 revealed 172 differentially expressed circRNAs (FC ≥ 2, adjusted P < 0.0001), including 135 upregulated and 37 downregulated protein-coding genes. BPTF ranked as one of the substantially downregulated protein-coding genes (Fig. 5C and Additional file 12: Fig. S4). We also identified a significant correlation between a c-myc pathway and circPDK1 using Gene Set Enrichment Analysis (GSEA; Fig. 5D). Notably, BPTF is required for c-myc transcriptional activity, and the BPTF-c-myc axis is involved in cell growth in pancreatic cancer [33]. Based on these results, we assumed that circPDK1 acts as a ceRNA to modulate the BPTF/c-myc axis by sponging miR-628-3p. To verify whether miR-628-3p could sponge BPTF directly, we further confirmed an inverse correlation between BPTF and miR-628-3p using data from our center (Fig. 5E). The mRNA expression levels of BPTF decreased after transfection with miR-628-3p mimics in both MIA PaCa-2 and PANC-1 cells, but increased after transfection with the miR-628-3p inhibitor (Fig. 5F). Dual-luciferase reporter assays showed that miR-628-3p could bind to the 3ʹ-UTR of BPTF (Fig. 5G). Therefore, we concluded that BPTF is a functional target gene of miR-628-3p. In addition, BPTF expression levels were positively correlated with circPDK1 expression levels in PC (Fig. 5H). Overexpressed circPDK1 substantially activated the BPTF-c-myc axis and modulated the c-myc downstream targets CCND1 and p21, whereas transfection with miR-628-3p mimics eliminated this effect. The inactivation of BPTF-c-myc axis and modulation of the two c-myc downstream targets induced by circPDK1 knockdown was also abolished by transfection with the miR-628-3p inhibitor (Fig. 5I, J). The same results were obtained using the dual-luciferase reporter assay (Fig. 5K). Functionally, miR-628-3p abolished the circPDK1 overexpression-induced promotion of cell proliferation and migration, similar to BPTF and miR-628-3p (Additional file 13: Fig. S5A–C). Collectively, circPDK1 serves as a ceRNA to bind to miR-628-3p from the sponging BPTF mRNA, thereby releasing BPTF from the inhibiting effects of miR-628-3p.

We investigated whether circPDK1 exerts its functions by interacting with RNA-binding proteins using RNA pull-down assays and silver staining. We found that abundant 200, 120, 75, 65, and 35 kDa proteins were associated with circPDK1 (Additional file 20: Fig. S12A). Mass spectrometry revealed that 65 kDa Myc box-dependent-interacting protein 1 (BIN1) was one of the topmost likely RNA-binding proteins of circPDK1 and that it might be abundant among circPDK1-associated proteins. Thus, we assumed that BIN1 would interact with circPDK1 directly; this hypothesis was next confirmed by RNA pull-down and immunoprecipitation assays, while a control circRNA named ciRS-7 [34], which acts as a miRNAs sponge, would not bind with BIN1 (Fig. 6A). Moreover, FISH assays demonstrated that circPDK1 was co-localized with BIN1 in the cytoplasm (Fig. 6B and Additional file 14: Fig. S6A). Although circPDK1 did not regulate BIN1 mRNA expression level, the BIN1 protein levels were significantly regulated after circPDK1 was altered (Additional file 15: Fig. S7A and Fig. 6C). In addition, circPDK1 expression induced reduced BIN1 protein levels that could be restored by proteasome inhibitor (MG132) (Fig. 6D). After treatment with cycloheximide (CHX), BIN1 had a shorter half-life in PC cells overexpressing circPDK1, while it had a longer half-life in PC cells that downregulated circPDK1 (Fig. 6E). To investigate the circPDK1 regions that interact with BIN1, 293 T cells were co-transfected with various truncations of circPDK1, follow RIP assays, truncation 3 and 4 interacted with BIN1 (Fig. 6F). Furthermore, those truncations were degraded after treatment with RNase R, which demonstrated that those truncations were linear (Additional file 16: Fig. S8). Based on the bioinformatic online database catRAPID, two predicted potential binding sites in truncations 3 and 4 were mutated (Fig. 6G). Using Sanger sequencing, mutations and the back-splice site in circPDK1-MUT were detected, and the results revealed that circPDK1-MUT was circularized correctly and possessed the same back-splice site with circPDK1-WT (Additional file 17: Fig. S9). Moreover, mutations in circPDK1 reduced the interaction between circPDK1 and BIN1 (Fig. 6H, I). This indicated that circPDK1-WT interacted with BIN1 in truncations 3 and 4, but not circPDK1-MUT in PC cells. circPDK1-MUT did not affect BIN1 at either the mRNA or protein level (Fig. 6J). Therefore, we assumed that circPDK1 might destabilize BIN1 proteins by interacting with it. Suppression of circPDK1 significantly inhibited the ubiquitination levels of BIN1, while circPDK1 overexpression notably enhanced the ubiquitination levels of BIN1, but this effect was eliminated by mutation (Fig. 6K). Furthermore, the atomic rotationally equivariant scorer (ARES) was used to accurately construct the 3D structure of the BIN1 regions that interact with circPDK1 computationally. The analysis results showed that circPDK1 could bind to BIN1 only in the BAR domain region, which was consistent with catRAPID (Fig. 6L). To confirm this prediction, four deletion mutants of BIN1 were designed, followed by an RNA pull-down assay, and the results showed that deletion of the 29‒276 aa region (BAR domain) of BIN1 abolished its interaction with circPDK1 (Fig. 6M). IHC assay also revealed that BIN1 protein levels were downregulated in PC tissues and negatively correlated with circPDK1 in tissue microarray (Fig. 6N, O). Together, these data indicate that circPDK1 might modulate the stability of BIN1 by interacting with the BAR domain region of BIN1, thus promoting its ubiquitin-dependent degradation.

To further screen the ubiquitin-enzyme-mediated BIN1 ubiquitination, mass spectrometry was performed after RNA pull-down. Interestingly, only a ubiquitin conjugating enzyme named UBE2O, which displays both E2 ubiquitin-conjugating enzyme and E3 ubiquitin ligase activities [35], was found among all participants. To verify whether UBE2O participates in the ubiquitination effects of circPDK1 on BIN1, RNA pull-down and RIP were performed, and the interaction between circPDK1 and UBE2O was confirmed, while ciRS-7 would not bind to UBE2O (Fig. 7A). Co-immunoprecipitation (Co-IP) analysis was performed, and results indicated that UBE2O bound to BIN1 (Fig. 7B). Furthermore, FISH and IF assays indicated the colocalization of circPDK1, UBE2O, and BIN1 mainly in the cytoplasm (Fig. 7C and Additional file 14: Fig. S6B). Similarly, the qRT-PCR results demonstrated that the RNA levels of BIN1 and circPDK1 were not affected by UBE2O, but BIN1 protein levels were notably increased after UBE2O downregulation (Additional file 15: Fig. S7B, C and Fig. 7D). Next, we investigated whether circPDK1 is necessary for the ubiquitin ligase activity of UBE2O for BIN1. The results demonstrated that the loss of circPDK1 notably weakened the effects of UBE2O on the ubiquitination and degradation of BIN1 (Fig. 7E, F). Furthermore, Co-IP was analyzed to verify whether circPDK1 could function as a scaffold to enhance the interaction between UBE2O and BIN1. The results showed that the interaction between UBE2O and BIN1 would be enhanced in MIA PaCa-2 PC cells transfected with circPDK1-WT, but not transfected with circPDK1-MUT. In addition, the associations between UBE2O and BIN1 were severely destroyed when treated with RNase A, but not treated with RNase R (Fig. 7G). Because circRNAs were resistant to RNase R, they would be degraded by RNase A. IHC assay also indicated that UBE2O protein levels were increased in PC tissues and negatively correlated with BIN1 (Fig. 7H, I). Collectively, circPDK1 could serve as a scaffold to enhance the binding of UBE2O and BIN1, thus facilitating the effects of UBE2O on the ubiquitination and degradation of BIN1.

Functionally, compared with control or circPDK1-MUT, circPDK1-WT significantly promoted PC cell proliferation and migration both in vitro and in vivo. In addition, circPDK1-MUT promoted the proliferation and migration of PC cells. Moreover, circPDK1-WT overexpression-induced PC cell proliferation and migration could be eliminated by BIN1 overexpression in vitro (Additional file 18: Fig. S10A–C, Additional file 19: S11A–D). IHC assays showed that PCNA and vimentin were upregulated in circPDK1-overexpressing mice subcutaneous tumors, whereas E-cadherin was downregulated (Additional file 19: Fig. S11E). Compared with the control, UBE2O accelerated PC cell proliferation and migration, which could be reversed by BIN1 overexpression and loss of circPDK1 (Additional file 20: Fig. S12B–D). These findings indicate that circPDK1 plays an oncogenic role partially by acting as a scaffold between UBE2O and BIN1 to enhance the degradation of BIN1.

Recent evidence has demonstrated that BIN1 is a tumor suppressor that interacts with c-myc, thereby limiting c-myc transcriptional activity [36, 37]. To verify whether UBE2O and circPDK1 are involved in BIN1-induced limited c-myc transcriptional activity in PC, a c-myc-responsive transcriptional luciferase reporter was transfected into MIA PaCa-2 PC cells with the indicated treatment. The c-myc-responsive transcriptional reporter luciferase activity was notably increased after transfection with circPDK1-WT, but not transfected with circPDK1-MUT, and this increase was partly abolished by BIN1. The luciferase activity increased by UBE2O was eliminated by BIN1 overexpression and loss of circPDK1 (Fig. 7J). Taken together, circPDK1 and UBE2O could activate c-myc transcriptional activity by degrading the BIN1 protein. To explore whether there are any cross-activation axes between the two axes, we detected the effect of BIN1 and UBE2O on the expression of miR-628-3p and BPTF, and vice versa. The results demonstrated that BIN1 and UBE2O did not affect miR-628-3p or BPTF expression (Additional file 21: Fig. S13A–C). At the same time, miR-628-3p or BPTF could not regulate BIN1 or UBE2O at the RNA or protein level (Additional file 21: Fig. S13D–E), suggesting that circPDK1 could activate c-myc through two no cross-talk axes.

As a common target of two axes mediated by circPDK1, c-myc is known to be a key mediator of the Warburg effect by directly activating various target glycolytic genes to meet the demands of rapid proliferation and metastasis [38]. Thus, we explored whether circPDK1 is associated with aerobic glycolysis. qRT-PCR and Western blotting were performed to analyze the glycolytic genes, and the results demonstrated that loss of circPDK1 notably inhibited the expression of these glycolytic genes (Fig. 8A, B). Lactate production, ATP production, and glucose uptake could be reduced by loss of circPDK1 (Fig. 8C–F), as well as the extracellular acidification rate (ECAR) and oxygen consumption rate (OCR) (Fig. 8G, H). In vivo, PET-CT scanning indicated that the loss of circPDK1 would significantly inhibited glucose metabolism in mice subcutaneous tumors (Fig. 8I). Furthermore, IHC assays indicated that PDK1, LDHA, and c-myc were upregulated in circPDK1-WT-overexpressing the subcutaneous tumor (Additional file 19: Fig. S11E). To verify whether circPDK1 could promote glycolysis by activating c-myc through two axes, rescue experiments were performed. Functionally, BIN1 reversed the circPDK1-WT overexpression induced the promotion of glycolysis (Additional file 18: Fig. S10D–F), similar to that of miR-628-3p (Additional file 13: Fig. S5D–F). In addition, circPDK1-MUT promoted the enhancement of glycolysis in PC cells. Meanwhile, loss of circPDK1 and increased BIN1 would also abolish the UBE2O induced potentiation in glycolysis (Additional file 20: Fig. S12E–G). Collectively, these data indicate that c-myc mediates the functions of circPDK1 in glycolysis.