Seventy-four patients who underwent GC surgery at Yijishan Hospital, the First Affiliated Hospital of Wannan Medical College (Wuhu, China), from 2015–2019 were enrolled. This study was approved by the ethics committee of the institution. During surgery, GC tissue was collected in a timely manner in a tube, and paired paracarcinoma tissue was collected from 5 cm above the edge of the GC tissue. Regular follow-ups were conducted. Patient details are listed in Supplementary Table S7 .

Four GC cell lines (MKN-45, HGC-27, MGC-803, and AGS) and normal human gastric mucosa epithelial (GES-1) cells were purchased from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). The cells were cultured in RPMI 1640 medium (Beyotime, China) containing 10% fetal bovine serum and the antibiotics penicillin and streptomycin (Beyotime, China) at 37°C in a cell incubator containing 5% CO2.

Total RNA was extracted from cells, paracarcinoma tissues, and GC tissues using TRIzol (Beyotime, China) according to the manufacturer’s instructions. cDNA was synthesized with a reverse transcription reagent (Beyotime, China), and qRT−PCR was performed using cDNA as a template. The reaction conditions were 95°C for 2 min, 95°C for 15 s, 60°C for 1 min, and 72°C for 30 s (40 cycles). GAPDH was used as the internal reference gene for circRNA_0001756, and U6 was used as the internal reference gene for miRNA-185-3p; relative expression was calculated by the 2-ΔΔCt method. The primer sequences are shown in Supplementary Table S1 .

siRNAs and miRNA inhibitors and mimics were synthesized by GenePharma (Shanghai, China). An hsa_circ_0001756 overexpression plasmid was synthesized by Shanghai Hanheng Company. The sequences are shown in Supplementary Tables S2 , S3 . One night prior to the experiment, normal cultured cells were digested and spread on six-well plates at a density of 200,000 cells per well. On the second day, the culture medium was removed, and the cells were washed with PBS. Lipo2000 was mixed with 50 µmol siRNA followed by 50 µL of serum-free or antibiotic-free culture medium. Then, 2 µL of Lipo2000 was added to a new EP tube, 50 µL of culture solution was added, and the mixture was gently mixed. The solutions in the two tubes were then mixed and allowed to stand at room temperature for 20 min. Next, 2 mL of medium was added to each well of a 6-well plate, the mixture was added to each well, and the plate was shaken well. The 6-well plate was then cultured in an incubator. After 6–8 h, the transfection medium was discarded, and culture medium containing serum without antibiotics was added. The cells were further cultured, and cell experiments were performed 48 h after transfection.

Two micrograms of RNA was digested with RNase R (Beyotime, China) for 5 min, and qRT−PCR was used to measure hsa_circ_0001756 and linear PGK1 levels.

Different GC cells were treated with 2 μg/mL ActD (Sigma minus;Aldrich, USA) for a period, and qRT−PCR was performed to measure hsa_circ_0001756 and linear PGK1 levels.

Four hundred microliters of RIPA lysis buffer was added to each group of GC cells to extract total protein, and the samples were separated by SDS−PAGE and then transferred to a nitrocellulose (NC) membrane. Rapid blocking solution was used for blocking (Beyotime, China) for 20 min. The membrane was incubated overnight with diluted primary antibodies [against PGK1 (1:1000, ABclonal), HK2 (1:1000, ABclonal), GLUT1 (1:1000, ABclonal), β-actin (1:4000, ABclonal) and PTBP1 (1:1500, ABclonal)] and then with diluted secondary antibodies (1:2000) at 37°C for 2 h and washed with TBST. ECL reagent was added dropwise, and the membrane was exposed in a darkroom. The grey values of each protein band were analyzed by ImageJ software.

Transplanted tumor tissues were embedded in paraffin, dewaxed, hydrated with xylene and graded alcohol solutions, and washed for 3 min with PBS. A primary antibody against PGK1 (1:1000, ABclonal, China) or Ki67 (1:1000, Abcam, USA); BAX (1:1000, Abcam, USA) was added, and the tumor tissues were refrigerated overnight. Secondary antibody was added, and the tumor tissues were incubated at room temperature for 30 min. Fresh diaminobenzidine (DAB) and hematoxylin solution were added for staining. After dehydration, photographs were taken under a microscope at 400×.

hsa_circ_0001756 and miR-185-3p probes were synthesized by RiboBio. The probe were detected with a FISH kit (RiboBio). Immunofluorescence images were taken with a Zeiss laser scanning microscope. The FISH probes used are shown in Supplementary Table S4 .

GC cells (3 × 10⁴) were plated in confocal dishes one night prior to the experiment. On the day of the experiment, the cells were fixed, permeabilized, and blocked with 2% BSA for 2 h at room temperature. The cells were incubated with primary antibody overnight, with a fluorescent secondary antibody for 1 h, and then with DAPI for 10 min. An LSM 710 confocal microscope (Zeiss, Germany) was used for image acquisition.

HGC-27 GC cells were transfected and cultured for 48 h and collected for the isolation of total protein. Each circRNA probe and NC probe (100 pmol) were denatured at 95°C for 3 min and then immediately placed on ice for denaturation and to maintain the linear structure of the probe. One milligram of total protein was placed into two 1.5-mL EP tubes, the denatured circRNA probe and NC probe were added, and the total protein-probe mixture was incubated overnight in RIP buffer at 4°C with vertical mixing. The next day, 60 µL of prewashed streptavidin magnetic beads (Dynabeads™ Streptavidin Trial Kit, Invitrogen) were added to capture the RNA−protein complexes. The magnetic beads were collected on a magnetic stand, washed with 1 mL of RIP buffer, and subjected to five 5-min washes with vertical mixing. The magnetic beads were then collected, 30 µL of SDS loading buffer was added to resuspend the magnetic beads, and the proteins captured by RNA pull-down were eluted at 95°C for 5 min. Ten microliters of the above samples were separated by 8%~12% SDS−PAGE, high-resolution silver staining was performed, and the unique bands in each sample were isolated for mass spectrometry analysis. The sequences of the probes are shown in Supplementary Table S5 .

The hsa_circ_0001756 overexpression vector was cotransfected with a GFP fusion protein. An anti-GFP antibody and an RNA immunoprecipitation kit were used according to the kit instructions, and the magnetic beads were cleaned with RNA lysis solution to extract RNA. The expression of miR-185-3p was measured by PCR with specific primers. The regulatory effect of miR-185-3p on PGK1 expression was verified using a previously described method. The sequences of the biotinylated probes are listed in Supplementary File 1 , Suppplementary Table S6 .

Various plasmids were used for the synthesis of a dual luciferase reporter system (Nanjing Gene Regulation Company); mutant and control miR-185-3p plasmids were transfected into hsa_circ_0001756-overexpressing, hsa_circ_0001756 knockdown and corresponding control cell lines. Forty-eight hours after transfection, the fluorescence intensity was measured by a fluorescence detector; Renilla luciferase activity was used as an internal reference. The regulatory effect of miR-185-3p on PGK1 expression was verified using a previously described method.

The methods used for flow cytometry analysis of apoptosis, the cell proliferation assay, the scratch healing test, migration and invasion analyses, cell cloning experiments (20), and measurement of glycolytic function can be found in the Supplementary Methods .

Nude mice were purchased from Qinglongshan Animal Factory, Nanjing, China, and 1 × 10⁶ lentivirus-transfected GC cells were subcutaneously administered. The animals were injected once a week and sacrificed after 21 days. The tumor size was observed at intervals of 3 days, and the tumors were weighed at the time of sacrifice. The ethics committee approved the experiments, and the protocols were performed in accordance with the relevant regulations of the Animal Protection Association.

The data were analyzed with GraphPad Prism V8.0 software (La Jolla, CA, USA). Kaplan−Meier analysis was used to generate long-term survival curves. The data are presented as the (X ± S) of three independent experiments. Linear regression analysis was used to analyze the correlation between hsa_circ_0001756 and miR-185-3P expression in GC samples. Data were compared between two groups by independent sample t tests, and data were compared among multiple groups by one-way ANOVA. Statistical significance was defined as P<0.05.

Hsa_circ_0001756 is formed by a loop of exon 2 of the HIPK2 gene, located on chromosome 7, and is 1084 bases long ( Figure 1A ). In 74 pairs of GC and adjacent nontumor tissues, hsa_circ_0001756 was overexpressed (P<0.0001) ( Figure 1B ). Patient data are shown in Supplementary Table S7 . Then, we performed statistical analysis. Increased hsa_circ_0001756 expression was correlated with tumor size and T and N stage ( Table 1 ). Sanger sequencing of the PCR products revealed that the circular structure of the circRNA was formed by head-to-tail splicing ( Figure 1C ). One of the key characteristics of circRNAs is their stability (21–23). AGS and HGC-27 cells were digested with the RNase R enzyme, and the qRT−PCR results revealed that the expression of linear HIPK2 decreased, while there was little change in hsa_circ_0001756 expression ( Figures 1D, E ). The mRNA level of linear HIPK2 gradually decreased over time, whereas the level of hsa_circ_0001756 was more stable and not affected by ActD treatment ( Figure 1F ). In addition, we performed long-term postoperative follow-up. A total of 74 GC patients were divided into circ_0001756^low and circ_0001756^high groups according to the median expression level. Statistical analysis indicated that the prognosis of patients with high expression of hsa_circ_00017565 was worse than that of patients with low expression of hsa_circ_00017565 ( Figure 1G ).

To explore the role of hsa_circ_0001756 in GC cells, We measured hsa_circ_0001756 expression levels in different GC cells. AGS cells were found to have the highest hsa_circ_0001756 expression ( Supplementary Figure S1A ).we designed siRNA oligonucleotides (Si-circ-1 and Si-circ-2). Si-circ-1 successfully reduced hsa_circ_0001756 expression, but the transcript levels of endogenous linear HIPK2 in HGC-27 and AGS cells were not affected ( Figure 2A ). We subsequently performed a series of functional experiments in which Si-circ-1 (si-circ_1756) was used to knockdown hsa_circ_0001756. EdU staining, the CCK-8 assay and the cell colony formation assay revealed that silencing hsa_circ_0001756 decreased the proliferative ability of GC cells ( Figures 2B–F ) ( Supplementary Figure S2B ). In addition, Transwell assays ( Figures 2G, H ) revealed reduced migration and invasion of GC cells in which hsa_circ_0001756 was silenced. Subsequent analysis of the Warburg effect revealed that hsa_circ_0001756 knockdown reduced glucose uptake and lactate production ( Figures 2I, J ). The scratch test detected the migration ability of gastric cancer cells,( Figures 2K, L ). HK2 (24), LDHA (25) and GLUT1 (26) are the key enzymes involved in glycolysis. Silencing circ_0001756 reduced the expression of these glucose-metabolizing enzymes( Figure 2M ). Therefore, we conclude that circ_0001756 silencing inhibits proliferation, migration, invasion and glycolysis.

We further explored the effect of hsa_circ_0001756 overexpression. qRT−PCR was used to assess the transfection efficiency in the overexpression experiments ( Supplementary Figure S2A ). EdU staining and the CCK-8 assay confirmed that overexpression of hsa_circ_0001756 increased the proliferation of GC cells ( Supplementary Figures S2B–D ), and the scratch test ( Supplementary Figures S2E, F ) revealed that overexpression of hsa_circ_0001756 increased the migration of GC cells.

Next, we explored whether hsa_circ_0001756 has other effects. CircRNAs have different functions in different subcellular compartments (27). According to FISH and Subcellular Separation Experiment, hsa_circ_0001756 was mainly localized in the cytoplasm ( Figure 3A ) ( Supplementary Figure S1C ). According to previous reports, cytoplasmic circRNAs can often bind miRNAs (28). Therefore, RIP assay were performed to verify the ability of hsa_circ_0001756 to bind to miRNAs in GC cells. We used an anti-AGO2 antibody to pull down and amplify hsa_circ_0001756 and found that hsa_circ_0001756 acted as a molecular sponge and bound AGO2 miRNA ( Figure 3B ) ( Supplementary Figure S1D ).

We predicted the miRNAs that might bind hsa_circ_0001756 ( Supplementary Table S10 ) through various databases (miRanda, RNAhybird and pita) and identified 117 overlapping miRNAs predicted by all three databases. According to previous reports, 5 of these miRNAs have carcinogenic functions (hsa-miR-185-3p, hsa-miR-139-3p, hsa-miR-448, hsa-miR-370-3p, and hsa-miR-3131) ( Figure 3C ). We then knocked down hsa_circ_0001756 to observe the changes in the expression of these miRNAs ( Figure 3D ). A pull-down assay combined with qPCR revealed that hsa_circ_0001756 could specifically bind to these miRNAs, especially miR-185-3p, and miR-448 and miR-370-3p cannot be significantly enriched. (P<0.001) ( Figure 3E ). RNAhybrid prediction analysis revealed that hsa_circ_0001756 and miR-185-3p have 8 stable binding regions [maximum binding free energy (MFE)>-20 kcal/mol] ( Supplementary Figure S4A ). In summary, miR-185-3p is the most likely miRNA to bind hsa_circ_0001756. Therefore, we performed a pull-down assay with a biotin-conjugated hsa_circ_0001756 probe to determine whether hsa_circ_0001756 can interact with miR-185-3p. The results revealed that miR-185-3p levels were obviously increased in the biotin-conjugated hsa_circ_0001756 probe group compared with the negative control group, suggesting that hsa_circ_0001756 could directly interact with miR-185-3p ( Figures 3F–I ). FISH revealed that both hsa_circ_0001756 and miR-185-3p were localized mainly in the cytoplasm ( Figure 3J ). In addition, miR-185-3p level increased after hsa_circ_0001756 was silenced and decreased after hsa_circ_0001756 was overexpressed. However, miR-185-3p expression was almost unchanged in the groups in which hsa_circ_0001756 with mutations in the miR-185-3p binding site mutant was silenced or overexpressed ( Figure 3L ). A dual-luciferase reporter assay revealed that miR-185-3p bound to hsa_circ_0001756-WT (circ-1756-WT) ( Supplementary Figure S3A ) ( Figure 3K ). In addition, we analyzed miR-185-3p expression in GC tissue by qRT−PCR and found that it was decreased in tumors ( Figure 3M ), and statistical analysis revealed that hsa_circ_0001756 expression was negatively correlated with miR-185-3p expression (P<0.0001) ( Figure 3N ). In summary, hsa_circ_0001756 can competitively bind to miR-185-3p to exert its function in GC.

After elucidating the relationship between hsa_circ_0001756 and miR-185-3p, we next studied how miR-185-3p plays a role in GC. qRT−PCR was used to verify the transfection efficiency of miR-185-3p mimics (Additional file 7, Supplementary Figure S6A ). EdU staining, the colony formation assay and the CCK-8 assay revealed that the miR-185-3p mimics inhibited GC cell proliferation (Additional file 7, Supplementary Figures S6B–D, I, J ). Moreover, the miR-185-3p mimics inhibited cell invasion and migration (Additional file 7, Supplementary Figures S6E–H ). The miRNA mimics also inhibited glycolytic function and decreased glucose uptake and lactate production (Additional file 7, Supplementary Figures S6L, M ) and inhibited the protein expression of HK2, LDHA, and GLUT1 (Additional file 7, Supplementary Figure S6K ).

To clarify the mechanism underlying the effect of miR-185-3p, we used the TargetScan database to determine that miR-185-3p can bind to the 3’-UTR of PGK1 (Additional file 5, Supplementary Figure S4B ). This finding indicates that PGK1 may be a target of miR-185-3p. After miR-185-3p overexpression, the expression of PGK1 was significantly decreased, while the protein level of PGK1 was increased by the inhibition of miR-185-3p (Additional file 7, Supplementary Figure S6P ). The miR-185-3p mimic significantly decreased PGK1 mRNA levels, whereas the miR-185-3p inhibitor increased PGK1 mRNA levels (Additional file 7 Supplementary Figure S6N ).

We investigated whether miR-185-3p exerts tumor-suppressive effects through PGK1. We found that miR-185-3p targeted the 3’-UTR of PGK1 ( Supplementary Figure S3B ); in addition, a dual-luciferase reporter assay revealed that miR-185-3p bound to PGK1-WT (Additional file 7, Supplementary Figure S6O ).

We observed the effect of hsa_circ_0001756 on PGK1 expression, and Western blot analysis revealed that the protein expression of PGK1 decreased after hsa_circ_0001756 was silenced (Additional file 7 Supplementary Figure S6Q ). we analyzed PGK1 expression in GC tissue by qRT−PCR(Additional file 7 Supplementary Figure S6R ). PGK1 is positively correlated with circHIPK2., miR185-3p is negatively correlated with the expression level of PGK1(Additional file 7 Supplementary Figures S6S, T ).

To determine whether hsa_circ_0001756 exerts a tumor-promoting effect through miR-185-3p, we conducted a response experiment. We cotransfected an hsa_circ_0001756 siRNA (Si-circ_0001756) and the miR-185-3p inhibitor into GC cells and then observed whether the tumor-suppressive effect of silencing hsa_circ_0001756 was affected by the miR-185-3p inhibitor. The results showed that the miR-185-3p inhibitor could restore the inhibitory effect of hsa_circ_0001756 silencing on GC cells, affecting glucose consumption ( Supplementary Figure S5A ), lactate production ( Supplementary Figure S5B ), proliferative capacity (Additional file 8, Supplementary Figures S7A–C, F, G ), behavior in the scratch test (Additional file 8, Supplementary Figures S7D, E ) and Transwell assay (Additional file 8, Supplementary Figures S7H, I ).

We next examined whether a miR-185-3p inhibitor could attenuate the downregulation of PGK1 expression induced by hsa_circ_0001756. The hsa_circ_0001756-mediated downregulation of PGK1 expression was reversed by the miR-185-3p inhibitor (Additional file 8, Supplementary Figure S7J ). However, interestingly, downregulation of miR-185-3p expression only partially reversed the effect of circ_0001756 silencing in GC cells. This finding also proves that hsa_circ_0001756 can work in different ways.

RBPs extensively affect gene transcription and translation, and interacting with RBPs is also an important function of circRNAs. Therefore, we further explored whether hsa_circ_0001756 acts through RBPs. To this end, hsa_circ_0001756-bound proteins were detected in the human gastric cancer cell line HGC-27 by an RNA pull-down assay, and the differentially expressed proteins in the pulled down samples were qualitatively analyzed by mass spectrometry ( Figure 4A ) ( Supplementary Table S8 ). We also performed GO enrichment analysis of these proteins ( Supplementary Figure S8A ) ( Supplementary Table S9 ). We ultimately selected 10 proteins related to translation or mRNA stability (marked in green in Supplementary Table S8 ). According to naive analysis and prediction via RBPsuite (http://www.csbio.sjtu.edu.cn/bioinf/RBPsuite/), there may be 4 proteins that bind to hsa_circ_0001756 ( Supplementary Figure S8B ). Among these, the protein with the highest mass spectrometry score was PTBP1. Therefore, we further investigated whether PTBP1 can bind to hsa_circ_0001756 by RNA pull-down and RIP assays and found that PTBP1 could be pulled down by the hsa_circ_0001756 probe in HGC-27 and AGS cell lysates ( Figure 4B ). Hsa_circ_0001756 could also be precipitated by an anti-PTBP1 antibody ( Supplementary Figure S8C ). We predicted the binding sites between hsa_circ_0001756 and PTBP1 by catRAPID (http://service.tartaglialab.com) and found that the binding sites between hsa_circ_0001756 and PTBP1 are located at 509-560 nt and 709-760 nt ( Supplementary Figure S8D ). We constructed a plasmid expressing hsa_circ_0001756 mutants lacking these two regions and transfected it into cells. A pull-down assay revealed that when the 709-760-nt fragment was removed, the amount of protein pulled down by the PTBP1 probe was significantly reduced, which indicated that PTBP1 mainly bound to hsa_circ_0001756 via this region ( Figure 4C ). To further verify which region of PTBP1 binds to hsa_circ_0001756, we designed PTBP1 mutants in which one of the binding regions was removed. RIP analysis revealed that hsa_circ_0001756 bound mainly to RNA recognition motif 1 (RRM1) of PTBP1 ( Figure 4D ). PTBP1 is a transnucleoprotein with distinct functions in the cytoplasm and nucleus. We found that hsa_circ_0001756 overexpression promoted the export of the PTBP1 protein from the nucleus to the cytoplasm, whereas hsa_circ_0001756 knockdown reduced the cytoplasmic level of PTBP1 ( Figure 4F ). Consistently, immunofluorescence confirmed that hsa_circ_0001756 expression was positively correlated with the level of PTBP1 in the cytoplasm ( Figure 4E ).

PTBP1 controls the translation and stability of mRNAs. PTBP1 is expressed in almost all human cells and mainly acts as a regulator of glycolysis in tumors. RIP analysis revealed that significantly more PGK1 mRNA was pulled down by an anti-PTBP1 antibody than by IgG ( Figure 5A ). In the experiment involving ActD, knockdown of PTBP1 reduced the mRNA level of PGK1 and shortened the half-life of the PGK1 transcript ( Figures 5B–D ). We then verified whether hsa_circ_0001756 affects the binding of PTBP1 to PGK1. A RIP assay revealed that inHGC-27 cells, knockdown of hsa_circ_0001756 significantly reduced the binding of PTBP1 to PGK1 ( Figure 5E ). Pull-down experiments also verified that reducing the expression of hsa_circ_0001756 reduced the binding of PTBP1 to the 3’UTR of PGK1 ( Figure 5F ). Rescue experiments revealed that the reduction in PGK1 mRNA stability caused by hsa_circ_0001756 knockdown in HGC-27 cells was reversed after PTBP1 overexpression ( Figure 5G ). We then further verified whether hsa_circ_0001756 can act through ceRNAs and RBPs simultaneously. Mutant miR-185-3P (circ-mut-miR) and miR-185-3P lacking the PTBP1-binding region (circ-mut-Δ709-760) were transfected into HGC-27 cells. The results showed that circ-mut-miR and circ-mut-Δ709-760 could still partially increase the expression of PGK1. However, circ-mut-(miR+▴709-760) did not increase the level of PGK1 ( Figures 5H, I ). Flow cytometry analysis of apoptosis also verified that circ-mut-miR and circ-mut-▴709-760 partially inhibited apoptosis, whereas circ-mut-(miR+▴709-760), similar to the empty vector, did not inhibit apoptosis ( Figure 5J )( Supplementary Figure S8E ). Therefore, we found that hsa_circ_0001756 not only acts through ceRNAs but also regulates the level of PGK1 through PTBP1.

To determine whether hsa_circ_0001756 promotes the development of GC in vivo, HGC-27 cells with stable hsa_circ_0001756 knockdown were injected subcutaneously into nude mice. The mean tumor size and mean tumor weight ( Figure 6C ) in the hsa_circ_0001756 silencing group were significantly reduced ( Figures 6A, B ). HE staining revealed the presence of solid tumors ( Figure 6D ). The hsa_circ_0001756 silencing group presented a significantly reduced level of hsa_circ_0001756 in tumors according to qRT−PCR ( Figure 6G ), whereas miR-185-3p expression increased ( Figure 6H ). This finding is consistent with the results from cells and tissues. Immunohistochemical analysis of tumors from mice revealed that hsa_circ_0001756 silencing reduced the level of PGK1 and Ki67, Increased level of Bax( Figures 6E, F ). In conclusion, interfering hsa_circ_0001756 level effectively inhibited GC cell growth in vivo.