CircFOXK2 promotes hepatocellular carcinoma progression and leads to a poor clinical prognosis via regulating the Warburg effect

HCC and adjacent tumor-free liver tissues were obtained from ninety-two patients who underwent radical hepatectomy at the Department of Hepatic Surgery and Liver Transplantation Center of the Third Affiliated Hospital of Sun Yat-sen University (Guangzhou, China) between July 2014 and August 2018. All enrolled patients were histologically diagnosed without any history of other pre-operative treatment, including chemotherapy and radiotherapy. After being collected, the resected specimens were cut, snap-frozen in liquid nitrogen and subsequently stored at -80℃. This study was approved by the Ethics Committee of the Third Affiliated Hospital of Sun Yat-sen University, and the procedures were conducted abiding by the Declaration of Helsinki. All patients provided informed consent. Detailed information on the operated patients is provided in Supplemental Table 2.

The normal human hepatocyte line LO2 and four human HCC lines, Huh7, HepG2, Hep3B, and SK-Hep1, were purchased from the Chinese Academy of Sciences (Shanghai, China) and maintained in high glucose (4.5 g/L)-Dulbecco's Modified Eagle's medium (DMEM, Gibco, Life Technologies, Carlsbad, CA, USA) containing with 10% fetal bovine serum (FBS, PAN-Biotech, Germany). All cells were cultured at 37℃ under an atmosphere of 5% CO₂. All cell lines were tested for mycoplasma and were mycoplasma-free when the experiments were conducted.

The primer sequences used in this study are listed in Supplemental Table 1. Total RNA extraction, reverse transcription and qRT-PCR were conducted as previously described [10]. The tissues were cut into 0.5 cm³ pieces and sequentially triturated in liquid nitrogen, while the cells were digested in 0.1% trypsin (Gibco, Life Technologies, Carlsbad, CA, USA). Then, total RNA extraction was performed using TRIzol (Invitrogen) according to the manufacturer's protocol. After confirming the amount and purity of total RNA using Biophotometer plus (Eppendorf, Germany), the Transcriptor First Strand cDNA Synthesis Kit (Roche Applied Science, USA) was used to reverse-transcribe total RNA to cDNA. cDNA was amplified using PCR Thermal Cycler (Bio-Rad, USA) by first heating at 65℃ for 10 min, incubating at 55℃ for 30 min, deactivating at 85℃ for 5 min and finally storage at 4℃ for 5 min. qRT-PCR was performed by SYBR Master Mix (Roche Applied Science) using a reverse transcription system (LC-480, Roche, USA). Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used to normalize gene expression data.

RNA fluorescence in situ hybridization (RNA-FISH) was conducted to visualize the location of circFOXK2 (RiboBio, Guangzhou, China) and miR-484 (BersinBio, Guangzhou, China) in cells using Fluorescent in Situ Hybridization Kits according to the manufacturer's instructions. The cells were treated with probes targeting circFOXK2 (hsa_circ_0000817) and miR-484 at 4℃ overnight and observed under a Zeiss 880 confocal microscope (Nikon Instruments, Melville, New York, United States) after being stained with DAPI for 2 min.

The proliferation potential of HCC cells was detected using a cell counting kit-8 (CCK8, KeyGEN BioTECH, Jiangsu, China) according to the manufacturer's protocol. After being treated according to the predefined group design, the cells were seeded in a 96-well plate at a density of 4,000 cells per well and cultured for 24, 48, 72, 96 and 120 h. The cell viability of each group was determined by the optical density (OD) values at 450 nm using a microplate reader (Tecan Spark 10 M, Austria).

The colony formation assay was used to evaluate the cloning ability of HCC cells. HCC cells in each group were seeded in 6-well plates at a density of 1,000 cells/well. Then, the cells were cultured in an incubator for 2 weeks. The colonies were visualized and counted after being fixed with 4% paraformaldehyde (PFA) and stained with crystal violet.

To measure cell apoptosis, a PI/Annexin V-FITC Apoptosis Detection Kit (BD, USA) was used according to the manufacturer's instructions. Briefly, 1 × 10⁶ HCC cells in each group were collected and rinsed twice with a washing buffer. Subsequently, the cells were incubated with Annexin V and PI for 15 min in the dark at room temperature. The cell apoptosis rate was determined using flow cytometry (BD Biosciences, San Jose, CA, USA).

2 × 10⁵ HCC cells were resuspended in serum-free DMEM medium and seeded into the upper chamber of Transwell plates with 8.0-μm pores (Corning Costar, Corning, NY, USA). DMEM medium containing 10% FBS was added into the lower chamber. The cells were maintained in an incubator at 37℃ for 24 h. Then, the non-migrated cells on the upper compartment were removed. Subsequently, the cells were fixed with 4% PFA for 30 min and stained with crystal violet for 15 min. Images were captured under an inverted light microscope (Leica, German) and counted using ImageJ software (National Institutes of Health, USA).

The cells in each group were seeded in 6-well plates for 24 h. A straight scratch was created using the tip of a sterilized pipette. Then, cells were washed gently with D-Hanks and cultured in DMEM containing 1% FBS. The samples were detected and photographed at 0 and 24 h under an inverted light microscope (Leica).

The cells were lysed in cold radioimmunoprecipitation assay (RIPA) Lysis Buffer containing Tris–HCL (pH 7.4, 50 mM), NaCl (150 mM), 0.1% Triton (100 ×), sodium dodecyl sulfate (SDS, 10%), sodium deoxycholate (10%), ethylene diamine tetraacetic acid (EDTA, 2 mM) and protease cocktail inhibitor (KeyGEN BioTECH, Jiangsu, China) for 20 min on ice. After total protein quantitation using a bicinchoninic protein assay (KeyGEN BioTECH), 30 μg proteins were electrophoresed using 12% SDS polyacrylamide gel electrophoresis (PAGE) and subsequently transferred onto polyvinylidene difluoride membranes (PVDF, Millipore, Billerica, MA, USA). 5% non-fat milk was used to block the non-specific antigen of the membranes for 1 h at room temperature. Then, the membranes were incubated with the corresponding primary antibodies at 4℃ overnight. After washing three times, the membranes were treated with secondary antibodies (anti-rabbit IgG, 1:5000, Sigma-Aldrich) for 1 h at room temperature. The blots were visualized by a ChemiDoc™ MP Imaging System (Bio-Rad, CA, USA) after treatment with an enhanced chemiluminescence (ECL) substrate. The intensities of the blots were evaluated using ImageJ software.

The NADH/NAD + Quantification Colorimetric Kit (BioVision) was used to determine the NADH/NAD + ratio of HCC cells. All procedures were conducted following the manufacturer's protocol.

The extracellular acidification rate and oxygen consumption rate were detected using the Seahorse XF^e 96 Extracellular Flux Analyzer (Seahorse Bioscience) following the manufacturer's instructions. The results were analyzed using the Seahorse XF-96 Wave software. The procedures for these two assays were as follows:

The mitochondrial morphology in the HCC cells was observed by staining with a MitoTracker Green FM probe (Invitrogen) in the dark for 30 min at 37℃. Images were obtained under a Zeiss 880 confocal microscope to assess the number and morphology of mitochondria. Finally, the percentage of cells with small and round mitochondria was calculated.

The cells were lysed in 4℃ cold co-IP buffer [HEPES (10 mM, pH 8.0), NaCl (300 mM), EDTA (0.1 mM), NP-40 (0.2%), glycerol (20%), protease and phosphatase inhibitors] for 30 min. The lysates were centrifuged at 10,000 g for 5 min and cleared after being treated with protein A/G agarose (Gibco BRL, Grand Island, NY, USA) for 15 min at 4℃. The pre-cleared supernatant was stirred in a horizontal shaker and incubated with the indicated primary antibodies at 4℃ overnight. Next, the protein complexes were harvested and incubated with the protein G beads for 120 min at 4℃. Finally, the samples were separated via SDS-PAGE.

Mass spectrometry analysis was conducted as previously described [10]. In brief, total proteins were first separated by SDS-PAGE (12%). Then, the protein bands near 15 kDa were manually cut according to the size and digested by sequencing-grade trypsin (Promega, Madison, WI, USA). Peptide mixtures were extracted, dried, and finally loaded to LC–MS/MS (Thermo Fisher Scientific, Waltham, MA, USA) for detecting the protein sequences. The National Center for Biotechnology Information nonredundant protein database with Mascot (Matrix Science, Boston, MA, USA) was used to analyze the fragment spectra.

After cross-linking with 1% formaldehyde in ice-cold PBS for 10 min, the cells were collected and lysed in RIP lysis buffer, followed by treatment with Dynabeads protein G conjugated with anti-AGO2 or anti-IgG and rotating overnight at 4℃. The immunoprecipitated RNAs were obtained using TRIzol reagent and measured by RT-qPCR with specific primers.

The sequences of circFOXK2 with mutated miR-484 binding sites or wild-type were established and inserted into luciferase vectors. Then, these vectors and miR-484 mimics were co-transfected into HEK-293 T cells for 48 h. The luciferase activity was detected by a dual-luciferase reporter assay system (Promega, USA) following the manufacturer's instructions. The Renilla luciferase internal control was used to normalize the results.

The RNA pull-down assay was carried out as previously documented [25]. The biotin-labeled miR-484 mimic (GenePharma, China) was transfected into the stably overexpressing circFOXK2 HCC cells for 48 h. After being washed and blocked by yeast tRNA for 2 h at 4℃, the streptavidin-Dyna beads M-280 were incubated with the cell lysates overnight at 4℃ to pull down the biotin-coupled RNA complex. The abundance of circFOXK2 in bound sections was detected by qRT-PCR.

Five-week-old male BALB/c nude mice were purchased from the Biomedical Research Institute of Nanjing University (Jiangsu, China) and housed under conditions following the Guidelines of Sun Yat-sen University for Animal Experimentation. All mice were housed under specific pathogen-free (SPF) conditions at the Laboratory Animal Center of Sun Yat-sen University with 12 h dark/light cycle, 50% humidity, 20℃ temperature and were fed standard laboratory diet and water ad libitum.

All in vivo experiments were approved by the Ethics Committee of Sun Yat-sen University (Guangzhou, China) and complied with the institutional guidelines. The xenograft model was established as previously described [26, 27]. 5 × 10⁶ HCC cells (Huh7) were suspended in Matrigel and injected subcutaneously into the left back of the mice. The length and width of the tumors were measured every three days when the tumors became palpable. Four weeks after implantation, the tumors were harvested for further analysis after sacrificing the mice. The weight and volume of samples were measured. The formula for calculating the tumor volume was as follows: tumor volume = (tumor length × tumor width²)/2. Moreover, lung metastasis models were established to detect the invasion potential of HCC cells in vivo. In brief, an empty vector, a sh-circFOXK2, or sh-circFOXK2 + a FOXK2-142aa overexpression plasmid, was transfected to the GFP-labeled HCC cells followed by intravenous injection through the tail vein at a density of 2 × 10⁶ cells/mouse. The mice were sacrificed after four weeks, and the lungs were visualized and photographed using the IVIS@ Lumina II system.

The protocol of immunohistochemistry staining was conducted as previously described [28]. In brief, 4 μm-thick paraffin-embedded sections were dewaxed and dehydrated, followed by incubation with H₂O₂ (3%) for 10 min at 37℃. For antigen repair, the sections were treated with EDTA (pH 8.0) at 95℃ for 20 min and then cooled down to 25℃. After blocking the non-specific antigens with normal goat serum, the sections were incubated with primary antibodies overnight at 4℃ and the secondary antibody for 30 min at 37℃. Finally, the sections were treated with diaminobenzidine and detected under a light microscope (Leica, Germany).

The control plasmid (vector control), circFOXK2 overexpressing vector, circFOXK2-flag vector, circFOXK2-mut vector, circFOXK2-flag-mut vector and FOXK2-142aa vector were obtained from General Biosystems (Anhui, China). ShRNA knocking down circFOXK2, siRNA targeting LDHA and miR-484 mimics or inhibitors were purchased from GenePharma (GenePharma Corporation, Shanghai, China). After reaching approximately 60% confluence, the vectors were transfected to the cells using a Lipofectamine® 3000 transfection kit (Invitrogen, Carlsbad, CA, USA), while the shRNA, siRNA and miRNA mimics or inhibitors were transfected using Lipofectamine RNAiMax (Invitrogen) according to the manufacturer's protocol. The efficiency of transfection was determined by qRT-PCR.

Statistical analyses were performed by GraphPad Prism (version 5.0, USA) and SPSS 23.0 (IBM, SPSS, Chicago, IL., USA). The relationship between circFOXK2 expression and the clinicopathological characterizations of HCC was analyzed by the Chi-square test. In addition, univariate and multivariate analyses were performed using a Cox proportional hazards regression model to assess the clinical value of circFOXK2. Survival analysis was carried out by the Kaplan–Meier method, and differences in overall survival were analyzed using the log-rank test. All in vitro experiments were independently repeated three times. As appropriate, the significant difference was assessed by the Mann–Whitney U test, Wilcoxon rank-sum test or unpaired two-tailed Student's t-test. A probability (p) value less than 0.05 was statistically significant. *p < 0.05, **p < 0.01, ***p < 0.001.

It has been reported that FOXK2 can yield heterogeneous effects in different types of cancer and may possess various biological functions [22, 23]. In addition, overwhelming evidence substantiates that circRNAs are generated from precursor mRNAs, and some participate in their host gene functions. In the present study, we first quantified the expressions of 14 circRNAs from FOXK2 in our samples. The results revealed that hsa_circ_0000817 (circFOXK2) was significantly increased in the HCC tissues compared to their adjacent liver tissues (Fig. 1A). Moreover, the expression of hsa_circ_0000817 was higher in multiple HCC cell lines (Huh7, HepG2, Hep3B, and SK-Hep1) than in the normal hepatocyte cell line (LO2) (Fig. 1B). We also performed qRT-PCR assays to investigate circFOXK2 expression in ninety-two pairs of clinical specimens from our center and further validated that compared with the counterpart adjacent liver tissues, circFOXK2 was significantly upregulated in the HCC tissues (Fig. 1C), with 84% (77 of 92) of HCC tissues presenting high cirFOXK2 expression (Fig. 1D). To assess the correlation between circFOXK2 expression and clinicopathological features/prognostic outcomes of HCC patients, we classified the HCC patients into high and low circFOXK2 expression groups according to the median circFOXK2 expression value. We found no significant correlation between circFOXK2 expression and the clinicopathological characteristics, indicating taht circFOXK2 was an independent prognostic factor for HCC after radical hepatectomy (Supplemental Table 2). Kaplan–Meier analysis showed that high expression of circFOXK2 was significantly associated with poorer OS compared to the low expression group (Fig. 1E). Next, the characterization of circFOXK2 was conducted. Our bioinformatics analysis showed that circFOXK2 (hsa_circ_0000817) is an exonic circRNA composed of the 2^nd, 3^rd, and 4^th exon of the FOXK2 gene with a length of 490 nucleotides (Fig. 1F). Sanger sequencing demonstrated the back-splicing junction site of circFOXK2, consistent with the annotation from the circRNA website (www.​cirbase.​com) (Fig. 1G). In addition, we designed two kinds of primers and conducted agarose gel electrophoresis for detecting reverse-transcribed DNA (cDNA) and genomic DNA (gDNA). As expected, we found that circFOXK2 could amplify from cDNA but not from the different primers of gDNA (Fig. 1H). Previous studies demonstrated that linear RNA had a 3' to 5' exoribonuclease, which was conveniently digested by RNase R, while circRNA did not due to it did not contain a poly-A tail [29]. Accordingly, RNase R was used before reverse transcription experiments to further verify the circular characterizations of circFOXK2. Our results corroborated that unlike the linear form of FOXK2, circFOXK2 was resistant to RNase R treatment (Fig. 1I). Moreover, the half-life of circFOXK2 and FOXK2 linear RNA was detected using actinomycin D to block their transcription. It was found that circFOXK2 was more stable than FOXK2 linear RNA (Fig. 1J). The FISH assay for HepG2 and Huh7 cells confirmed that circFOXK2 was mainly expressed in the cytoplasm of HCC cells (Fig. 1K). Taken together, these results suggest that the expression of circFOXK2 (hsa_circ_0000817) in HCC tissues is high and correlates with a poor prognosis.

To probe the biological functions of circFOXK2 in HCC, shRNA or a circFOXK2 overexpression vector was transfected into different HCC cell lines to conduct loss-and gain-to-function experiments. Stable cell lines (HepG2 and Huh7) of circFOXK2 knockdown were established by transfecting with shRNA-circFOXK2, while circFOXK2 overexpression cell lines were transfected with circFOXK2 overexpression vector (Supplemental Fig. 1A). As expected, the CCK8 and colony formation assays found that downregulation of circFOXK2 significantly suppressed the proliferation of HCC cells while enhancing their apoptotic rate, and overexpressing circFOXK2 yielded the opposite results (Fig. 2A-C and Supplemental Fig. 1B-D). In addition, we investigated the role of circFOXK2 in the invasion ability of HCC cells by conducting Transwell assays and wound healing assays. The results showed that knockdown of circFOXK2 expression remarkably suppressed HCC cell invasion, while overexpression of circFOXK2 potentiated HCC cell invasion (Fig. 2D and Supplemental Fig. 1E and F).

There is ample literature indicating that FOXK2 plays a critical role in the Warburg effect and is important for regulating the growth and metastasis of HCC. Therefore, we speculated that the protumor activity of circFOXK2 in HCC cells involves modulation of the Warburg effect. To verify this hypothesis, we established the stable HCC lines of circFOXK2 overexpression or knockdown and compared the critical cellular bioenergetics and metabolic parameters between these cell lines and the NC group (Supplemental Fig. 1A and B). As shown in Fig. 2E-F, compared with the NC group, significantly high pyruvate production and lactate accumulation in HCC cells were observed after circFOXK2 overexpression, whereas knockdown of circFOXK2 yielded the opposite results. Moreover, downregulated circFOXK2 significantly increased ATP formation from OXPHOS, which was reversed by circFOXK2 overexpression (Fig. 2G). The NADH/NAD + ratio was detected to further confirm the change of redox reaction after different treatments, and the results showed that compared with the NC group, knockdown of circFOXK2 significantly increased the ratio of NADH/NAD + , and this phenomenon could be reversed after transfection with circFOXK2 overexpressing plasmid (Fig. 2H). In addition, circFOXK2 overexpression enhanced the ECAR and impaired the oxygen consumption of HCC cells (Fig. 2I and J). In contrast, silencing circFOXK2 expression reduced the glycolytic capacity and enhanced the OCR (Fig. 2K and L). Taken together, these findings indicate the pro-oncogenic role of circFOXK2 in HCC and upregulated circFOXK2 could enhance the Warburg effect in HCC cells.

In recent years, the protein-coding capacity of an increasing number of circRNA has gradually been brought to light. According to the original annotation from the circRNADb database, a 142 amino-acid (aa) protein is encoded by circFOXK2 (hsa_circ_0000817) which contains a 429-nt open reading frame (ORF) (Fig. 3A). Bioinformatics analysis showed that human circFOXK2 has a putative sequence of AUG in ORF that initiates the translation of a novel protein (Fig. 3A and Supplemental Fig. 2). Thus, the ORF of circFOXK2 was speculated to encode a putative 142 aa protein with the unique sequence of aa termed "FOXK2-142aa". Growing evidence demonstrates that the internal ribosome entry site (IRES) initiates the cap-independent translation process, which is a critical mechanism for circFOXK2 translation [30]. Thus, we conducted a dual-luciferase assay to verify the predictive effect of circFOXK2. As shown in Fig. 3B and C, the Luc/Rluc activity in circFOXK2 with full-length IRES was significantly higher than in the mutated or truncated IRES, whereas empty vectors did not stimulate Luc activation. In addition, to distinguish the role of FOXK2-142aa from circFOXK2, we established four flag-labeled vectors, including an empty vector, circFOXK2-flag, circFOXK2-flag-mut and FOXK2-142aa-flag (Fig. 3D). Among these, circFOXK2-flag represented a CMV-induced expression vector containing flag-labeled circFOXK2 sequence; circFOXK2-flag-mut represented a CMV-induced expression vector containing flag-labeled circFOXK2 sequence with start codon mutant (ATG → ACG), and FOXK2-142aa-flag represented a CMV-induced expression vector with the flag-labeled FOXK2-142aa sequence. Compared with the empty vector, the upregulated expression of circFOXK2 was investigated in HCC cells by transfecting with circFOXK2-flag vector or circFOXK2-flag-mut vector, while transfecting with FOXK2-142aa vector did not induce any change (Fig. 3E). Besides, these four vectors did not alter the mRNA expression of linear FOXK2 (Fig. 3E). Western blot provided preliminary evidence of the protein-encoding role of circFOXK2, which translated into FOXK2-142aa protein expression. In this regard, FOXK2 protein expression was not affected after transfection with the four vectors in Huh7 and HepG2 cells (Fig. 3F). Moreover, the expression of the flag was only observed in the cells transfected with circFOXK2-flag or FOXK2-142aa-flag, but not in the vector group or the circFOXK2-flag-mut group (Fig. 3F). Additionally, we prepared the specific antibody targeting FOXK2-142aa to verify this phenomenon and found that transfection with circFOXK2-flag or FOXK2-142aa-flag induced the high expression of FOXK2-142aa, and FOXK2-142aa could still be detected after transfection with an empty vector or circFOXK2-flag-mut, which indicated the presence of endogenous FOXK2-142aa (Fig. 3F). Furthermore, we separated the total cell lysate after transfecting circFOXK2 into HEK-293 T cells and then performed mass spectrometry analysis. The 15 kDa protein band obtained during SDS-PAGE was cut and underwent LC–MS/MS (Fig. 3G). The FOXK2-142aa encoded by circFOXK2 was confirmed, and the unique aa sequence of FOXK2-142aa, "TADKGWQVHIQVPEHK", was identified (Fig. 3H). Next, we used this specific antibody to detect the expression of FOXK2-142aa in clinical specimens. As expected, FOXK2-142aa expression was significantly higher in the HCC tissues than in the adjacent liver tissues (Fig. 3I). Finally, higher expression of FOXK2-142aa correlated with a worse OS of HCC patients (Fig. 3J). Collectively, these findings suggest that circFOKX2 is an encoding circRNA that could translate into the protein FOXK2-142aa in HCC.

To further investigate the function of FOXK2-142aa and its potential regulatory mechanism, we performed a co-IP assay and mass spectrum analysis to identify the putative interaction candidates of FOXK2-142aa. FOXK2-142aa-interacting proteins were analyzed by the Proteome Discover software (Thermo Fisher Scientific). As shown in Fig. 4A and Supplemental Fig. 3A, we found that the most abundant protein was lactate dehydrogenase A (LDHA), and FOXK2-142aa could bind to the peptides of LDHA, speculating that LDHA is the potential target of FOXK2-142aa. The interaction between FOXK2-142aa and LDHA was confirmed by the Western blot assay (Fig. 4B). In addition, after a circFOXK2 overexpression plasmid was transfected into HCC cells, we carried out immunofluorescence staining to further visualize the colocalization of FOXK2-142aa with LDHA in the cytoplasm of HCC cells (Fig. 4C). LDHA is well-recognized as an important enzyme contributing to the Warburg effect. Current evidence suggests that LDHA activity tightly depends on its phosphorylation status [31]. Upon phosphorylation at the residue (Tyr10) of LDHA, it promotes the conversion of NADH to NAD + , the generation of pyruvate and lactate, and is directly associated with the activities of several oncogenes to induce cancer progression [32]. Therefore, the four vectors mentioned above without flag-tag were transfected into HCC cells, and the phosphorylation level of LDHA was detected. It was found that transfection with the circFOXK2 or FOXK2-142aa vector significantly enhanced the phosphorylation at the Tyr10 site of LDHA, unlike the empty vector and circFOXK2-mut, suggesting that FOXK2-142aa interacted with LDHA to activate its Tyr10 (phosphorylation) in HCC cells (Supplemental Fig. 3B).

Moreover, we produced a series of Flag-tagged LDHA deletion mutants to delineate the domains of LDHA responsible for its interaction with FOXK2-142aa (Fig. 4D). Co-IP and Western blot assays showed that FOXK2-142aa preferred interacting with the C-terminal subunit of LDHA rather than its N-terminal domain (Supplemental Fig. 3C). In this regard, the △161–218 region of the C-terminal was indispensable for the LDHA-FOXK2-142aa binding (Fig. 4E). In the reciprocal experiments, the regions of FOXK2-142aa involved in interacting with LDHA were additionally analyzed by conducting a co-IP assay with Flag-tagged truncation mutants of FOXK2-142aa (Fig. 4F). As shown in Fig. 4G, FOXK2-142aa (1–71) and FOXK2-142aa (72–142) could be pulled down by LDHA, suggesting the full length of FOXK2-142aa interacted with LDHA. In a further study, increased LDHA phosphorylation was observed by full-length FOXK2-142aa expression rather than two truncation mutants, indicating that full-length FOXK2-142aa was responsible for regulating the phosphorylation of LDHA (Fig. 4H).

To further detect whether the biological effect of FOXK2-142aa on HCC progression was associated with LDHA, two vectors and one siRNA, including an empty vector, a FOXK2-142aa overexpression vector, and a si-LDHA, were transfected into HCC cells, respectively. Both CCK8 and colony formation assays revealed that LDHA knockdown suppressed the growth of HCC cells, while upregulation of FOXK2-142aa promoted HCC cell proliferation (Supplemental Fig. 3D-E). In addition, we transfected the FOXK2-142aa overexpression vector and si-LDHA into HCC cells and found that the exogenous knockdown of LDHA expression significantly weakened the pro-tumor effects of FOXK2-142aa (Supplemental Fig. 3D-E). The alterations in the Warburg effect in HCC cells were also investigated. Significantly higher pyruvate accumulation and lactate production were found after upregulating FOXK2-142aa expression, and these effects were reversed after further knockdown of LDHA (Fig. 4I and J). Moreover, the upregulation of FOXK2-142aa decreased the formation of ATP generated by OXPHOS; however, LDHA knockdown restored ATP formation (Fig. 4K). Consistent results were observed for the extracellular acidification rate. Combined transfection of si-LDHA and FOXK2-142aa overexpression vectors significantly reversed the FOXK2-142aa-induced decrease in OCR (Fig. 4L and M). These results corroborated that FOXK2-142aa plays an important role in affecting the progression and the Warburg effect of HCC cells, which closely correlates with the interaction of LDHA and stimulates its activation.

To verify the in vitro results, we established xenograft tumor models by subcutaneous injection of HCC cells transfected with an empty vector, an shRNA lentiviral plasmid targeting circFOXK2 (sh-circFOXK2), and sh-circFOXK2 + a FOXK2-142aa overexpression plasmid (Fig. 5A and B). As shown in Figs. 5C and D, the tumors formed by circFOXK2-deficient HCC cells exhibited a significantly smaller volume and lower tumor weight than the NC group, which could be restored by further treatment with a FOXK2-142aa overexpression plasmid. PCNA and TUNEL staining confirmed the correlation between FOXK2-142aa and HCC tumorigenicity (Fig. 5E-F). Moreover, Western blot showed that compared with the NC group, FOXK2-142aa expression was reduced, and knockdown of circFOXK2 significantly decreased LDHA phosphorylation at Tyr10 in the tumor tissues while suppressing circFOXK2 and upregulating FOXK2-142aa expression could restore the phosphorylation levels (Fig. 5G). Finally, lung metastasis models were established to assess the metastatic capacity of HCC cells in each group. We found that the fluorescence intensities in the lungs were much weaker in the downregulated circFOXK2 expression group than in the NC group, and adding FOXK2-142aa overexpression plasmid reversed these findings, suggesting that circFOXK2 also promoted lung metastasis of HCC cells via encoding FOXK2-142aa (Supplemental Fig. 4A).

It is well-acknowledged that circRNAs function as competing endogenous RNAs to regulate miRNA expression, resulting in the discharge of specific miRNA-targeted mRNAs [33]. In this study, we sought to explore the functions of circFOXK2. Ten putative miRNAs that could potentially bind to circFOXK2 (including miR-486-3p, miR-410-5p, miR-409-3p, miR-34c-5p, miR-34b-5p, miR-34a-5p, miR-335-3p, miR-23a-5p, miR-762, and miR-484) were predicted by an online database (RegRNA 2.0, http://​regrna2.​mbc.​nctu.​edu.​tw/​) (Fig. 6A). Subsequently, an RNA pull-down assay was conducted to assess whether circFOXK2 could directly interact with these candidate miRNAs. Biotin-labeled probes targeting circFOXK2 were constructed and used to conduct RNA pulldown in HCC cells, followed by qPCR assay to measure these ten miRNAs. Our results showed that miR-484 was the highest enriched miRNA in HepG2 and Huh7 (Fig. 6B). It has been demonstrated that miRNA could function as an RNA-induced silencing complex (RISC) component by interacting with Argonaute-2 (AGO2). Thus, we further conducted a RIP assay with anti-AGO2 and found that circFOXK2 and miR-484 were pulled down by the anti-AGO2 antibody (Supplemental Fig. 5A). In addition, the site of circFOXK2 binding miR-484 was predicted by RNAalifold (http://​rna.​tbi.​univie.​ac.​at/​), which was mutated to perform dual-luciferase reporter assays (Fig. 6C and Supplemental Fig. 5B). Both circFOXK2 wild-type (WT) luciferase reporter vector and circFOXK2 mutant luciferase reporter vector containing the sequence could not bind miR-484 were established and respectively co-transfected with miR-484 mimics into HepG2 and Huh7 cells. As shown in Fig. 6D, circFOXK2 WT luciferase reporter was significantly reduced by miR-484 mimics, while miR-484 did not affect the activity of circFOXK2 mutant luciferase reporter. Consistently, a pull-down assay with a biotin-labeled miR-484 probe found that circFOXK2 was captured by miR-484 (Fig. 6E). Finally, a FISH assay showed the colocalization of circFOXK2 and miR-484 in the cytoplasm of HepG2 and Huh7 cells (Fig. 6F). Taken together, these results indicated that circFOXK2 could directly bind to miR-484 for modulating the miR-484 expression in HCC cells.

It has been demonstrated that miR-484 plays a critical anti-cancer role in various tumors. Qiu et al. previously reported the effect of miR-484 on negatively regulating the progression of HCC cells [34]. In this study, we performed an RT-qPCR assay to detect the expression of miR-484 in several HCC cell lines and showed that miR-484 expression was downregulated in Huh7, HepG2, Hep3B, and SK-Hep1 cells compared with LO2 cells (Fig. 7A). Then, an miR-484 mimic was used to upregulate miR-484 expression, which could significantly reduce the proliferative potential of HCC cells, while treatment with a miR-484 inhibitor yielded the opposite results (Fig. 7B-C and Supplemental Fig. 6A-B).

It has been established that miRNAs can interact with the 3’-UTR of the target mRNAs to regulate their expression. Next, an online database, miRWalk, was applied to analyze the target genes that miR-484 could bind to (Fig. 7D). After miR-484 mimics and an inhibitor were transfected into HepG2 and Huh7 cell lines, the expression of Fis1 was upregulated and downregulated, respectively, at the mRNA and protein levels (Fig. 7E-F). We also constructed luciferase reporter plasmids containing wild-type Fis1 sequence or mutant Fis1 sequence and co-transfected them with miR-484 mimics to confirm whether Fis1 was a direct target of miR-484 in HCC cells (Fig. 7G). The results showed that miR-484 mimics reduced more than 50% of luciferase activity, whereas no effect on luciferase activity was observed after transfecting a vector comprising a mutant Fis1 sequence (Fig. 7H). Fis1 is a key factor affecting mitochondrial fission and subsequently regulating the Warburg effect. Therefore, we further detected the morphological changes of mitochondria in HCC cells. Compared with the control group, after treatment with a miR-484 inhibitor, more fragmented rod-shaped mitochondria were observed, while filamentous mitochondria were observed after transfecting with miR-484 mimics (Fig. 7I). Our findings indicated that the Warburg effect was weakened after transfecting with miR-484 mimics, as evidenced by the reduced levels of pyruvate and lactate and the increase of ATP generation, which were remarkably reversed by Fis1 overexpression (Fig. 7J-L). Similarly, the glycolysis ability in HCC cells, including the increase of ECAR level and the recession of OCR ability, inhibited by miR-484 mimics was also abolished after upregulated Fis1 expression (Fig. 7M-N). In summary, these findings demonstrated that miR-484 could inhibit the Warburg effect of HCC cells via regulating Fis1 expression and subsequently affecting mitochondrial dynamics.

Given that circFOXK2 could directly bind to miR-484 and miR-484 could inhibit the Warburg effect via regulating Fis1 expression, we carried out rescue experiments to detect whether circFOXK2 promotes the Warburg effect of HCC cells via not only encoding FOX2-142aa to regulate LDHA phosphorylation, but also acting as a ceRNA to sponge miR-484 followed by affecting Fis1-related mitochondrial fission. Upregulated miR-484 expression significantly reduced the levels of pyruvate and lactate in HCC cells; however, these effects could be rescued by combined overexpression of circFOXK2 (Fig. 8A and B). Moreover, increased ATP production induced by transfecting the miR-484 mimic was reversed after transfection of the circFOXK2 overexpression plasmids (Fig. 8C). In addition, the ECAR and OCR measurements showed that the glycolytic ability of HCC cells was significantly potentiated by overexpressing circFOXK2 and reversely decreased after further treatment with miR-484 mimics (Fig. 8D and E). The mRNA and protein expression of Fis1 was further detected. As shown in Figs. 8F and G, Fis1 expression could be enhanced by overexpressing circFOXK2, which could be blocked by introducing a miR-484 mimic. Moreover, we directly visualized the morphological changes of mitochondria in each group and found a more fragmented shape of mitochondria following the administration of a circFOXK2 overexpression vector, whereas mitochondria presented as filamentous configurations by combined application of a miR-484 mimic (Fig. 8H). Finally, we assessed the expression of Fis1 in the tumor specimens of the xenograft models. The results confirmed that Fis1 was significantly downregulated in the sh-circFOXK2 group compared with the control group (Supplemental Fig. 7A-B). Taken together, our findings substantiated that circFOXK2 upregulates Fis1 expression and subsequently induces mitochondrial fission to potentiate the Warburg effect in HCC cells by sponging miR-484 (Fig. 8).