CircSPARC promotes esophageal squamous cell carcinoma through an HNRNPC–EPN2 splicing axis that activates Wnt/β-Catenin signaling Authors: Suli Dai, Cong Zhang, Zishuan Wei, Yaxin Liu, Yang Wen, Jinxia Chen, Xiaoya Li, Sisi Wei, Guogui Sun, Lianmei Zhao Published: 2026 ABSTRACT BackgroundPersistent activation of the Wnt/β-catenin signaling pathway is a key driver of esophageal squamous cell carcinoma (ESCC) progression. Circular RNAs (circRNAs) have emerged as critical regulators of oncogenic signaling in cancer. However, how circRNAs contribute to the sustained activation of Wnt/β-catenin signaling in ESCC is poorly defined.MethodsIntegrated analysis of self-sequenced and public circRNA datasets revealed elevated expression of circSPARC in ESCC. The functional effects of circSPARC on cell proliferation, migration, and invasion were evaluated through both in vitro and in vivo assays. RNA pull-down, RNA immunoprecipitation (RIP), alternative splicing analysis, RNA antisense purification (RAP), and cell surface protein biotinylation were used to elucidate the molecular mechanism by which circSPARC mediates aberrant Wnt/β-catenin activation in ESCC.ResultsThis study identifies a circRNA-mediated alternative splicing axis that promotes activation of Wnt/β-catenin signaling and ESCC progression. CircSPARC interacts with HNRNPC, a heterogeneous nuclear ribonucleoprotein involved in RNA processing, to suppress the tumor-suppressive long transcript of epsin 2 (EPN2), thereby impairing clathrin-mediated endocytosis-dependent degradation of the Wnt receptor frizzled 7 (FZD7). This interaction leads to sustained Wnt/β-catenin activation, thus promoting ESCC progression. Conversely, expression of the EPN2 long transcript restores circSPARC silencing-induced inhibition of Wnt/β-catenin signaling and suppresses ESCC cell malignancy both in vitro and in vivo. Clinically, elevated expression of circSPARC in ESCC tissues correlates with poor patient prognosis.ConclusionsThese findings reveal a circSPARC-HNRNPC-EPN2 axis that drives persistent Wnt/β-catenin activation through a mutation-independent mechanism. Targeting this pathway in tumors with high circSPARC expression presents a promising therapeutic strategy.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12943-026-02691-w. Background Esophageal squamous cell carcinoma (ESCC) is an aggressive digestive tract cancer with a poor prognosis [1, 2]. Aberrant Wnt/β-catenin signaling, caused by Wnt ligand binding to the Frizzled/LRP receptor complex, promotes ESCC initiation and progression [3, 4]. Disruption of this pathway is associated with tumor growth, spread, and resistance to chemotherapy [5–8]. Although mutations in key components like APC and CTNNB1 appear in some cancers, the upstream causes of prolonged Wnt/β-catenin activation in ESCC are still unknown. In particular, post-transcriptional mechanisms are not well understood [9–11]. Therefore, developing precise and effective therapies targeting the Wnt/β–catenin pathway is a major challenge. Alternative splicing (AS) modulates gene expression and increases proteomic diversity [12]. Dysregulation of splicing factors and aberrant splicing events are linked to cancer, altering signaling pathway components and promoting malignancy [13–17]. Recent transcriptomic analyses found widespread splicing abnormalities in ESCC that contribute to disease progression [18, 19]. Yet, the splicing regulators responsible for these events in ESCC, and their roles in activating oncogenic pathways such as the Wnt/β-catenin pathway, remain undefined. This gap hinders the development of therapy for splicing-mediated cancer pathways. Among post-transcriptional regulators, circular RNAs (circRNAs) are single-stranded RNAs that form covalently closed loops and have become significant in cancer research [20, 21]. CircRNAs can serve as molecular sponges for microRNAs (miRNAs) or act as protein decoys, thereby influencing protein activity [22]. Increasing evidence indicates that circRNAs regulate alternative splicing, either by directly binding to splicing factors or by interacting with miRNAs that affect splicing [23–25]. Therefore, circRNAs can connect faulty splicing events to oncogenic signaling pathways [26, 27]. However, the mechanism by which circRNAs mediate this connection in ESCC remains unclear. This study identifies circSPARC as a tumor-promoting circular RNA in ESCC and elucidates its regulatory mechanism in the splicing-Wnt/β-catenin axis. Our results demonstrate that circSPARC is significantly upregulated and closely correlates with poor prognosis. CircSPARC interacts with HNRNPC to suppress the generation of the EPN2 long isoform. This disruption affects clathrin-mediated endocytosis and impairs the degradation of FZD7, which is essential for maintaining active Wnt/β-catenin signaling and driving the progression of aggressive ESCC. These results reveal a mutation-independent mechanism for Wnt/β-catenin activation, highlighting the circSPARC-HNRNPC-EPN2 axis as a potential target for ESCC therapy. Materials and methods Patient samples This study included 124 patients with ESCC from three independent cohorts. Cohort 1 consisted of 72 randomly selected ESCC clinical samples and their paired adjacent non-tumor tissues from The Fourth Hospital of Hebei Medical University (January–June 2020). These samples were used to quantify circSPARC, EPN2-L, and EPN2-S. Cohort 2 consisted of a tissue microarray (TMA) with 40 paired ESCC tissues and matched adjacent non-tumor tissues from the same hospital (March–September 2022). The TMA was used to assess the expression of circSPARC and FZD7. Cohort 3 consisted of 12 paired esophageal cancer tissues and adjacent normal tissues, also obtained from the same insitution (June–July 2023). RT-PCR was used to analyze these samples and measure EPN2-L, EPN2-S, and circSPARC. All tissues were confirmed as ESCC through pathology and had not received neoadjuvant therapy prior to surgery. Complete clinicopathological data were available for each specimen. The study adhered to the Declaration of Helsinki and received approval from the Ethics Committee of the Fourth Hospital of Hebei Medical University. All participants provided written informed consent before surgery, allowing the use of their tissue samples for research. Cell culture Seven human ESCC (esophageal squamous cell carcinoma) cell lines were used in this study. All of them, including KYSE30, KYSE150, KYSE450, KYSE510, KYSE410, KYSE180, and YES2 were commercially obtained from Pricella Biotechnology (Wuhan, China). Human esophageal epithelial cells (HEEC, non-cancerous cells from the esophagus) and HEK293T cells (derived from human embryonic kidney cells) were purchased from Pricella Biotechnology. ESCC cells were cultured in RPMI 1640 (a nutrient-rich medium for cell culture) (Gibco, Massachusetts, USA) supplemented with 10% fetal bovine serum (Gibco), 100 µg/ml streptomycin, and 100 U/ml penicillin (Beyotime, Shanghai, China). HEEC and HEK293T cells were cultured in DMEM (Dulbecco’s Modified Eagle Medium, another cell culture medium) (Gibco) supplemented with 10% fetal bovine serum. All cell lines were authenticated using Short Tandem Repeat (STR) analysis, a DNA fingerprinting method, before their use in the experiments. All cell lines were routinely screened for mycoplasma contamination (a common cell culture contaminant) using the MycoBlue Mycoplasma Detector from Vazyme (Nanjing, China), and only mycoplasma-free cells were used in the experiments. Plasmids, reagents, and antibodies Overexpression plasmids for wild-type and mutant circSPARC were constructed by amplifying and cloning the sequences into the pLC5-ciR vector (Geneseed Biotech, Guangzhou, China). Small interfering RNAs (siRNAs) targeting circSPARC and HNRNPC were acquired from GenePharma (Suzhou, China). Short hairpin (sh) sequences targeting circSPARC and HNRNPC were cloned into the pLKO.1 vector. HNRNPC full-length and truncated mutants and FZD7 were inserted into the pcDNA3 vector. EPN2 variants were cloned into the pLVX vector. All plasmids were designed and synthesized by YouBio Biological Technology (Changsha, China). A detailed list of the reagents and antibodies used in this study is presented in Supplementary Tables 1 and 2, respectively. Cell transfection Cells were transfected with plasmids or siRNA using Lipofectamine 2000 (Thermo Fisher Scientific, Massachusetts, USA) according to the manufacturer’s instructions. After 6 h, the medium was replaced with complete growth medium. Cells were cultured for an additional 48 h before being harvested and analyzed. The siRNA sequences used are listed in Supplementary Table 3. Stable expression cells were prepared as follows. Briefly, HEK293T cells were transfected with a lentiviral vector, psPAX2 packaging plasmid, and pMD2.G envelope plasmid in a 4:3:1 ratio for 48 h to produce lentivirus supernatant. For lentiviral transduction, 1 × 106 ESCC cells were infected with 1 mL of the viral supernatant. After 8 h of incubation, the culture medium was replaced with fresh complete medium. Selection started 48 h after transduction by adding puromycin (2 µg/mL, Solarbio, Beijing, China) to establish stable polyclonal populations. Quantitative real-time PCR and RT-PCR Total RNA was extracted from cells and tissue samples using Trizol reagent (Thermo Fisher Scientific). For qRT-PCR, first-strand cDNA synthesis was performed on the total RNA with the GoScript™ Reverse Transcription System (Promega, Wisconsin, USA) following the manufacturer’s protocol. PCR was conducted using SYBR Green PCR Master Mix (Promega), with GAPDH serving as the endogenous control. Relative gene expression was calculated using the 2−ΔΔCt method. For RT-PCR assays, cDNAs were amplified with 2 × Hieff Robust PCR Master Mix (Yeasen Biotechnology, Shanghai, China). The PCR products were then separated by DNA gel electrophoresis to detect alternative splicing transcripts. Splicing outcomes were quantified using Percent Spliced In (PSI) values, calculated as the ratio of the exon-included isoform intensity to the total intensity of both exon-included and exon-excluded isoforms. The primers used for amplification are listed in Supplementary Table 3. Actinomycin D and RNase R treatment assay To evaluate RNA stability, cells were treated with 2 µg/mL actinomycin D (MedChemExpress, New Jersey, USA) to inhibit RNA synthesis. After specified incubation times, total RNA was extracted and analyzed using quantitative RT-qPCR to determine the decay rates of both circSPARC and its linear SPARC mRNA counterpart. Total RNA was digested with bovine pancreas-derived RNase R (3 U/µg RNA) (Geneseed Biotech) at 37 °C for 30 min. Subsequent qRT-PCR analysis assessed the resistance of circSPARC and the sensitivity of linear SPARC mRNA to RNase R treatment. Cell proliferation, colony formation, migration, and invasion assays Cell proliferation was monitored using the IncuCyte Live-Cell Analysis System (Sartorius, Goettingen, Germany). After 24 h post-transfection, cells from each experimental group were trypsinized, resuspended in complete growth medium, and counted. Then, cells were plated in 96-well plates at a density of 2 × 103 cells per well (200 µL total volume per well), with six replicate wells per condition. The plates were then transferred to an IncuCyte Analyzer for real-time proliferation monitoring, with measurements taken every 3 hours over three days. For the colony formation assay, cells from different treatment groups were separately seeded in 6-well plates at a density of 600 cells per well and cultured at 37 °C for approximately 10 days to allow colony formation. After fixation with 4% paraformaldehyde for 15 min (Solarbio, Beijing, China) and crystal violet staining for 30 min (Solarbio), colony images were captured and counted. For Transwell migration and invasion assays, transfected cells were collected 24 h after transfection and resuspended in serum-free RPMI 1640 medium. Cells were seeded into the upper chamber of Transwell inserts (Corning, New York, USA), either coated (for invasion) or uncoated (for migration) with Matrigel (BD Biosciences, New Jersey, USA). A volume of 600 µL of complete medium was added to the lower chamber. After 24 h of incubation, cells that did not migrate or invade on the upper membrane surface were removed with a cotton swab. Cells that migrated or invaded to the lower surface were fixed with 4% paraformaldehyde for 15 min, stained with crystal violet for 30 min, and counted in five random microscopic fields per well. Western blot Western blotting was performed following established procedures. Transfected ESCC cells were lysed with RIPA buffer (Thermo Fisher Scientific) supplemented with protease inhibitor cocktail (Roche, Basel, Switzerland). After denaturing at 100 °C for 5 min, equal amounts of protein were separated by SDS-PAGE and transferred to PVDF membranes (Merck Millipore, Darmstadt, Germany). Membranes were blocked with 5% non-fat milk (Solarbio) for 1 h at room temperature and then incubated overnight at 4 °C with diluted primary antibodies, followed by HRP-conjugated secondary antibodies for 1 h at room temperature. Protein bands were visualized using an ECL substrate (NCM Biotech, Suzhou, China) according to the manufacturer’s protocol. Nucleus and cytoplasm isolation assay The nuclear and cytoplasmic fractions of KYSE30 and KYSE150 cells were isolated using the PARIS™ Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. Briefly, cells were lysed in a cell fractionation buffer and kept on ice for 10 min. After centrifugation at 500 × g for 3 min, the cytoplasmic fraction (supernatant) was collected, while the nuclear fraction (pellet) was washed with the buffer for further analysis. The distribution of circSPARC in both fractions was then examined by qRT-PCR. RNA FISH assay and immunofluorescence (IF) staining ESCC cell lines (KYSE30 and KYSE150) and tissue sections underwent FISH analysis using a commercial kit (Ribobio, Guangzhou, China) as recommended by the manufacturer. Cells in confocal dishes were fixed with 4% paraformaldehyde at room temperature for 15 min. After washing, pre-chilled permeabilization buffer was added, followed by incubation at 4 °C for 5 min. Cells were then treated with pre-hybridization buffer and incubated at 37 °C for 30 min to block. After removing the pre-hybridization solution, samples were incubated with the probe (Ribobio) in hybridization buffer for 16–18 h at 37 °C in the dark. Subsequently, the cells were washed sequentially with 4 × SSC, 2 × SSC, and 1 × SSC. Finally, nuclear staining was performed using DAPI, and the samples were observed under a laser confocal microscope (Nikon A1R HD25, Japan). For immunofluorescence, cells were fixed with 4% paraformaldehyde for 15 min, then permeabilized with 0.3% Triton X-100 for 10 min. After blocking with 10% BSA for 30 min, the samples were incubated with primary antibodies overnight at 4 °C. Following washes, the cells were treated with secondary antibodies at room temperature for 1 h. Finally, samples were mounted in an anti-fade medium with DAPI for nuclear counterstaining. All fluorescent images were captured using a laser scanning confocal microscope with consistent imaging settings across all channels. RNA pulldown assay For the RNA pulldown assay, cellular lysates were prepared by solubilizing cells in IP lysis buffer and pre-cleared by incubation with pre-washed Pierce streptavidin magnetic beads (Thermo Fisher Scientific) at 4 °C for 1 h. The 3’-biotinylated circRNA probe (GenePharma) was then immobilized onto the beads through a 10-minute incubation at room temperature. Next, the biotinylated beads were incubated with cell lysates overnight at 4 °C. After magnetic separation, the bead-bound complexes were thoroughly washed. The retained proteins were eluted for subsequent mass spectrometry or western blot analysis to identify potential interactors. The sequences of the probes are listed in Supplementary Table 3. RNA immunoprecipitation (RIP) RIP assays were conducted on differentially transfected cell groups using the Magna RIP™ RNA-Binding Protein Immunoprecipitation Kit (Sigma-Aldrich), strictly following the manufacturer’s protocol. Cell lysates were prepared with RIP lysis buffer supplemented with protease and RNase inhibitors. Protein A/G magnetic beads were initially conjugated with either HNRNPC-specific antibody or control IgG antibody through 30-minute rotation at room temperature. The antibody-bound beads were then combined with cell lysate and incubated overnight at 4 °C. Subsequently, the complexes were washed. RNA was extracted from the magnetic beads using Trizol reagent, followed by downstream analysis via qRT-PCR and RT-PCR. The primer sequences used for RIP-PCR are listed in Supplementary Table 3. RNA Antisense purification (RAP) A total of 1 × 10⁷ KYSE30 or KYSE150 cells were collected and lysed in a lysis buffer containing protease inhibitors and RNase inhibitors. Following centrifugation, the supernatant was collected and divided into three portions: input, probe group, and negative control group. The probe group and negative control group were incubated with biotin-labeled circSPARC probes and negative control probes, respectively, at 37 °C for 4 h. Subsequently, the mixtures were incubated with streptavidin magnetic beads for 30 min. After washing, proteinase K was added to digest proteins, and RNA bound to circSPARC was purified using the Trizol method. Finally, the retrieved RNA was detected by RT-qPCR. Mass spectrometry analysis CircSPARC pull-down samples were separated by SDS-PAGE, and excised gel bands were digested in-gel with trypsin. The resulting peptides were extracted using 0.1% formic acid (FA), desalted on a C18 column, and lyophilized. Peptide concentration was measured with a BCA assay. Subsequently, 1 µg of peptides was injected into an Easy-nLC 1200 liquid chromatography system and analyzed with a Q Exactive HF mass spectrometer for proteomic profiling. RNA sequencing Total RNA from KYSE30 cells with circSPARC knockdown was isolated using Trizol reagent according to the manufacturer’s protocol. The integrity of the isolated RNA was evaluated with an Agilent Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA, USA). Its concentration and purity were measured using a Qubit 3.0 Fluorometer and a Nanodrop One spectrophotometer (Thermo Fisher Scientific). After constructing the sequencing library, the samples were sequenced on the Illumina NovaSeq 6000 platform by Sinotech Genomics Science and Technology Ltd. (Shanghai, China). For alternative splicing analysis, after assembling transcripts from the samples using StringTie, the events were analyzed with the rMATS software. The AS events were classified into five types: SE (Skipped exon), RI (Retained intron), A5SS (Alternative 5’ splice site), A3SS (Alternative 3’ splice site), and MXE (Mutually exclusive exons). Alternative splicing levels were quantified using the percent spliced in (PSI) metric, calculated as the number of inclusion junction reads divided by the total number of inclusion and skipping junction reads for each exon skipping event. Co-immunoprecipitation Cells were homogenized in IP lysis buffer containing protease and phosphatase inhibitor cocktails. The lysates were clarified by centrifugation at 13,000 rpm for 15 min at 4 °C. The prepared cell lysates were pre-cleared with 10 µL of Protein A/G magnetic beads (Beyotime) by rotation at 4 °C for 3 h. After adding the specified antibodies and control IgG to the pre-cleared lysates, 20 µL of Protein A/G agarose beads were added, followed by incubation on a rotator at 4 °C for 3–5 h per treatment. The protein-bound beads were washed three times with lysis buffer and finally eluted with 50 µL of SDS loading buffer. The eluted proteins or protein complexes were detected by immunoblotting. Internalization of biotinylated FZD7 Primary ESCC cells were subjected to overnight serum starvation, followed by incubation with 1 nM cleavable EZ-Link Sulfo-NHS-S-S-Biotin (Thermo Fisher Scientific) at 4 °C for 30 min with gentle agitation. Unreacted biotin was quenched by washing cells with cold PBS supplemented with 50 mM glycine (Solarbio). Subsequently, cells were transferred to pre-warmed media containing 100 ng/mL Wnt3a (MedChemExpress) and allowed to internalize at 37 °C for 30 min. Residual surface biotin was removed through three sequential 15-minute treatments with cleavage buffer (23 mM Na2HPO4, 27 mM NaH2PO4, 5 mM NaCl, 1% BSA, 10 mM EDTA, pH 8.0, 50 mM DTT) at 4 °C with gentle shaking. Cells were then lysed in RIPA buffer, and biotinylated proteins were isolated using streptavidin magnetic beads. The pulled-down complexes were analyzed by SDS-PAGE and Western blotting. Internalized FZD7 was detected with an anti-FZD7 antibody. Biotinylation of cell surface FZD7 Primary ESCC cells were subjected to overnight serum starvation, followed by treatment with 100 ng/mL Wnt3a and allowed to internalize at 37 °C for 30 min. Subsequently, cells were incubated with non-cleavable biotin to label remaining cell surface FZD7, cell lysates were affinity purified on streptavidin beads, and bound material was analyzed by Western blotting using anti-FZD7 antibodies to visualize cell-surface FZD7. Immunohistochemistry (IHC) Paraffin-embedded tissue sections were subjected to heat-induced epitope retrieval (performed at 60 °C for 2 h) followed by xylene deparaffinization. Antigen retrieval was carried out in EDTA buffer (pH 9.0) (ZSBion, Beijing, China), and endogenous peroxidases were quenched with 3% H₂O₂. After blocking with 5% BSA to reduce nonspecific interactions, sections were incubated with primary antibodies at 4 °C overnight. Antigen-antibody complexes were visualized using HRP-conjugated secondary antibodies with DAB chromogen development, followed by hematoxylin counterstaining. The stained sections were mounted and imaged by bright-field microscopy for pattern analysis. In vivo tumorigenesis and metastasis formation assay KYSE30 cells stably expressing circSPARC, EPN2-L, or short hairpin targeting circSPARC were resuspended in 0.1 mL PBS and injected subcutaneously into the bilateral subscapular regions of 5-6-week-old BALB/c mice (Vital River, Beijing, China). Tumor growth was monitored every 3 days by caliper measurement. After a 5-week period, the mice were euthanized via cervical dislocation, and the xenograft tumors were excised and weighed. Approximately 1 mm³ tumor tissue fragments were immediately snap-frozen at -80 °C for later analysis, while the remaining tissue was fixed in 10% neutral buffered formalin. In the pulmonary metastasis model, BALB/c mice were randomly divided into two experimental groups (n = 5 per group). A suspension containing 1 × 10⁵ KYSE30 cells with stable knockdown or overexpression of circSPARC in 0.1 mL PBS was administered intravenously via the lateral tail vein. After an 8-week observation period, the animals were euthanized by cervical dislocation. The excised lung tissues were fixed in 4% paraformaldehyde, then processed for paraffin embedding. Serial sections were stained with hematoxylin and eosin (HE), and metastatic foci were counted under a microscope. To assess the effect of β-catenin signaling inhibition on circSPARC-driven ESCC progression, mice were randomly divided into four experimental groups. After inoculation with circSPARC-overexpressing KYSE30 cells, peritumoral injections of either XAV-939 (0.2 mM in corn oil; MCE) or a vehicle control were given 3 times /week for 3 weeks, starting on day 9 after cell implantation. Tumor volumes (measured) and weights were recorded at regular intervals. Excised tumors were processed for paraffin embedding and subsequent histopathological examination. Statistical analysis All experiments were performed with at least three biological replicates. Statistical analyses were conducted using GraphPad Prism 10.0 (GraphPad Software) or SPSS 16.0 (IBM). Numerical data are presented as the mean ± standard deviation (SD). Two-group comparisons used two-tailed Student’s t-tests, while multiple-group data were analyzed by one-way ANOVA. Data for categorical variables are expressed as the number of cases (percentage), with group comparisons performed using the chi-square test. Survival distributions were compared using Kaplan-Meier analysis with log-rank tests. Correlations between variables were assessed using Pearson correlation analysis. Statistical significance was set at P < 0.05, with *P < 0.05, **P < 0.01, and ***P < 0.001. Results Identification of circSPARC as a functional oncogenic circRNA contributing to ESCC progression To comprehensively identify functional circRNAs that promote ESCC progression, we integrated previously generated sequencing data from our research team [28] (Supplementary Table 4) with publicly available circRNA datasets-GSE157803 and PMID:33,663,506 and identified four circRNAs consistently upregulated in ESCC tissues (Fig. 1A-B). Among these, hsa_circ_0004104 (circSPARC) was prioritized for further analysis due to its significantly elevated expression in a cohort of 15 ESCC tissue samples and its increasing trend in ESCC cell lines (Supplementary Fig. 1A-B). circSPARC is derived from exons 6–9 of the SPARC gene located on chromosome 5 (q33) with a spliced length of 553 nucleotides and was validated by Sanger sequencing with divergent primers targeting the back-splice junction (Fig. 1C). Additional experimental evidence confirms circSPARC as a bona fide and stable circular RNA transcript in ESCC cells (Supplementary Fig. 1C-F). Fig. 1Identification of circSPARC as a functional oncogenic circRNA contributing to ESCC progression. A. Volcano plot illustrates the circRNA sequencing results of ESCC tissues and paired adjacent normal tissues. Differentially expressed circRNAs were highlighted (up-regulated in ESCC: orange points; down-regulated in ESCC: blue points). B. A Venn diagram illustrates the overlap of differentially expressed circRNAs between our sequencing data and published datasets. C. Genomic characterization and verification of circSPARC. The specific back-splicing junction of circSPARC was identified by Sanger sequencing. D. The expression levels of circSPARC were assessed by RT-qPCR in 72 matched pairs of ESCC and normal adjacent tissues (cohort 1).Statistical significance was calculated using the paired t-test. E. Representative RNA FISH images of circSPARC (red) in ESCC and normal adjacent tissues (Cohort 2). DAPI was used to label cell nuclei. F. The bar graphs showed the expression level of circSPARC in ESCC based on the FISH analysis. Statistical significance was calculated using the chi-square test. G. The box plot shows the correlations between circSPARC expression and clinical stage, T stage, lymph node metastasis, and tumor size of ESCC patients based on the FISH analysis. Statistical significance was calculated using the chi-square test. H. Kaplan-Meier survival curves indicated the correlation of circSPARC expression and overall survival of patients with ESCC. Patients were separated by the median expression level of circSPARC, as determined by RT-qPCR data. I. Representative images of xenograft tumors in nude mice inoculated with circSPARC stably silenced or control KYSE30 cells (n = 8). J-K. Quantification of tumor weight and volume in the circSPARC down regulated and control groups. Statistical significance was calculated using the Mann-Whitney U test. L. Representative images of xenograft tumors in nude mice inoculated with circSPARC stably overexpressed or control KYSE30 cells (n = 10). M-N. Quantification of tumor weight and volume in the circSPARC overexpressed and control groups. Statistical significance was calculated using the Mann-Whitney U test. O-P. Representative IHC images of Ki-67 staining in the subcutaneous xenograft tumor model in nude mice. Q-R. The nude mice lung metastasis model evaluating the impact of circSPARC knockdown on pulmonary metastases. Statistical significance was calculated using the Mann-Whitney U test. S-T. The nude mice lung metastasis model evaluating the impact of circSPARC overexpression on pulmonary metastases. Statistical significance was calculated using the Mann-Whitney U test. *P < 0.05, **P < 0.01, ***P < 0.001 Identification of circSPARC as a functional oncogenic circRNA contributing to ESCC progression. A. Volcano plot illustrates the circRNA sequencing results of ESCC tissues and paired adjacent normal tissues. Differentially expressed circRNAs were highlighted (up-regulated in ESCC: orange points; down-regulated in ESCC: blue points). B. A Venn diagram illustrates the overlap of differentially expressed circRNAs between our sequencing data and published datasets. C. Genomic characterization and verification of circSPARC. The specific back-splicing junction of circSPARC was identified by Sanger sequencing. D. The expression levels of circSPARC were assessed by RT-qPCR in 72 matched pairs of ESCC and normal adjacent tissues (cohort 1).Statistical significance was calculated using the paired t-test. E. Representative RNA FISH images of circSPARC (red) in ESCC and normal adjacent tissues (Cohort 2). DAPI was used to label cell nuclei. F. The bar graphs showed the expression level of circSPARC in ESCC based on the FISH analysis. Statistical significance was calculated using the chi-square test. G. The box plot shows the correlations between circSPARC expression and clinical stage, T stage, lymph node metastasis, and tumor size of ESCC patients based on the FISH analysis. Statistical significance was calculated using the chi-square test. H. Kaplan-Meier survival curves indicated the correlation of circSPARC expression and overall survival of patients with ESCC. Patients were separated by the median expression level of circSPARC, as determined by RT-qPCR data. I. Representative images of xenograft tumors in nude mice inoculated with circSPARC stably silenced or control KYSE30 cells (n = 8). J-K. Quantification of tumor weight and volume in the circSPARC down regulated and control groups. Statistical significance was calculated using the Mann-Whitney U test. L. Representative images of xenograft tumors in nude mice inoculated with circSPARC stably overexpressed or control KYSE30 cells (n = 10). M-N. Quantification of tumor weight and volume in the circSPARC overexpressed and control groups. Statistical significance was calculated using the Mann-Whitney U test. O-P. Representative IHC images of Ki-67 staining in the subcutaneous xenograft tumor model in nude mice. Q-R. The nude mice lung metastasis model evaluating the impact of circSPARC knockdown on pulmonary metastases. Statistical significance was calculated using the Mann-Whitney U test. S-T. The nude mice lung metastasis model evaluating the impact of circSPARC overexpression on pulmonary metastases. Statistical significance was calculated using the Mann-Whitney U test. *P < 0.05, **P < 0.01, ***P < 0.001 Furthermore, a large - sample cohort of 72 paired clinical ESCC tissues confirmed the significant upregulation of circSPARC in tumor tissues compared to adjacent normal tissues (Fig. 1D), which was further substantiated by FISH analysis of tissue microarrays (Fig. 1E-F). Additional analysis of the association between circSPARC expression and clinicopathological characteristics across both cohorts demonstrated that elevated circSPARC levels were positively correlated with advanced clinical stage, higher T stage, lymph node metastasis, and increased tumor size (Fig. 1G, Supplementary Fig. 1G-J). Notably, Kaplan-Meier analysis demonstrated that higher circSPARC expression was associated with poorer overall survival in ESCC patients (Fig. 1H). Taken together, these findings suggest that high circSPARC level contributes to adverse clinical outcomes of ESCC patients. We further examined the potential role of circSPARC in the malignant progression of ESCC in vitro and in vivo. First, junction-targeting siRNAs or a synthetic circular plasmid were used to effectively knock down or overexpress circSPARC without changing SPARC mRNA levels in KYSE30 and KYSE150 cells (Supplementary Fig. 2A-B). Consistently, a series of functional assays showed that silencing circSPARC significantly reduced the proliferation, colony formation, migration, and invasion of KYSE30 and KYSE150 cells (Supplementary Fig. 2C-H). Conversely, overexpressing circSPARC increased these malignant behaviors in ESCC cells (Supplementary Fig. 2I-N). We further examined the in vivo effect of circSPARC on ESCC progression. First, stable KYSE30 cell lines with circSPARC knockdown and overexpression were created using lentiviral transduction (Supplementary Fig. 2O-P). Then, results from the subcutaneous xenograft tumor model showed that circSPARC knockdown significantly decreased tumor weight and volume (Fig. 1I-K), while overexpression had the opposite effect (Fig. 1L-N). Consistent with this, Ki-67 expression was lower in circSPARC-knockdown tumors and higher in overexpressing groups (Fig. 1O-P). Additionally, in the tail vein metastasis model, circSPARC depletion notably reduced the number of lung metastatic nodules, whereas overexpression led to the opposite trend (Fig. 1Q-T). In summary, these results provide evidence that circSPARC plays a key role in promoting the malignant behavior of ESCC cells, both in vitro and in vivo. The tumor-promoting effects of circSPARC in ESCC depend on its interaction with HNRNPC To further elucidate the mechanism by which circSPARC promotes the malignant progression of ESCC, we first examined its subcellular localization using FISH and subcellular fractionation. The results showed that circSPARC is distributed in both the nucleus and cytoplasm of KYSE30 and KYSE150 cells (Supplementary Fig. 3A-B). Next, an Argonaute 2 (AGO2) RNA immunoprecipitation (RIP) assay ruled out the possibility that circSPARC acts as a miRNA sponge (Supplementary Fig. 3C-D). Our data also excluded the potential for circSPARC to encode a functional polypeptide in ESCC (Supplementary Fig. 3E-F). We then hypothesized that circSPARC exerts regulatory effects through interactions with proteins in ESCC. First, biotinylated RNA pull-down assays confirmed probe specificity: compared to the control probe, the circSPARC-specific probe selectively enriched circSPARC without pulling down linear mSPARC (Supplementary Fig. 3G). Silver staining revealed distinct ~ 33kD protein bands in the circSPARC probe group (Fig. 2A), which were identified by mass spectrometry. The results showed multiple proteins involved in mRNA splicing processes as interactors of circSPARC (Fig. 2B, Supplementary Table 5). Based on unique peptide counts, we selected the top three splicing factors for validation. Our findings confirmed that circSPARC specifically interacts with HNRNPC, but not with RBMX or PCBP2 (Fig. 2C-D, Supplementary Fig. 3H). Additionally, colocalization staining and HNRNPC-specific RIP assays further supported this interaction (Fig. 2E-F). Fig. 2The tumor-promoting effects of circSPARC in ESCC depend on its binding to HNRNPC. A. Putative circSPARC-binding proteins enriched by RNA pull-down were resolved by SDS-PAGE and visualized through silver staining. B. Gene Ontology (GO) terms (BP: biological processes, CC: cellular components, MF: molecular functions) were analyzed for proteins interacting with circSPARC. C. Detection of endogenous HNRNPC following circSPARC pull down from KYSE30 and KYSE150 cells. D. The specific peptide map of HNRNPC as identified by mass spectrometry. E. Representative images of circSPARC FISH (red) and HNRNPC immunofluorescence (green) in KYSE30 and KYSE150 cells. Nuclei were stained with DAPI (blue). F. RNA immunoprecipitation (RIP) assays demonstrated the interaction between HNRNPC and circSPARC in KYSE30 and KYSE150 cells. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using Student’s t-test. G. Graphical abstract of 3×Flag-tagged HNRNPC and its truncated forms. H. Anti-Flag immunoprecipitation followed by PCR assay to assess the enriched circSPARC in KYSE30 cells transduced with full-length HNRNPC and its truncated forms. I. CircSPARC pull-down combined with western blot to detect the enriched Flag-HNRNPC in KYSE150 cells transduced with full-length HNRNPC and its truncated forms. J. Predicted binding motif HNRNPC on target RNAs from RBPsuite. K. Schematic diagram illustrating the putative HNRNPC-binding regions and mutations within circSPARC. L. CircSPARC pull-down combined with western blot to detect the enriched HNRNPC in KYSE30 cells transfected with wild-type or mutant HNRNPC-binding site type of circSPARC. M. Anti-HNRNPC immunoprecipitation followed by RT-PCR assay to measure the enriched circSPARC in KYSE150 cells transfected with wild-type or mutant HNRNPC-binding site type of circSPARC. N. The colony formation and transwell assays measured the changes in KYSE30 cells after knockdown of HNRNPC and co-overexpression with circSPARC wild type or HNRNPC binding site mutant type. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. O. The colony formation and transwell assays measured the changes in KYSE150 cells overexpressing circSPARC or co-overexpressing HNRNPC full-length or without the RRM domain. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. *P < 0.05, **P < 0.01, ***P < 0.001, ns: not significant The tumor-promoting effects of circSPARC in ESCC depend on its binding to HNRNPC. A. Putative circSPARC-binding proteins enriched by RNA pull-down were resolved by SDS-PAGE and visualized through silver staining. B. Gene Ontology (GO) terms (BP: biological processes, CC: cellular components, MF: molecular functions) were analyzed for proteins interacting with circSPARC. C. Detection of endogenous HNRNPC following circSPARC pull down from KYSE30 and KYSE150 cells. D. The specific peptide map of HNRNPC as identified by mass spectrometry. E. Representative images of circSPARC FISH (red) and HNRNPC immunofluorescence (green) in KYSE30 and KYSE150 cells. Nuclei were stained with DAPI (blue). F. RNA immunoprecipitation (RIP) assays demonstrated the interaction between HNRNPC and circSPARC in KYSE30 and KYSE150 cells. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using Student’s t-test. G. Graphical abstract of 3×Flag-tagged HNRNPC and its truncated forms. H. Anti-Flag immunoprecipitation followed by PCR assay to assess the enriched circSPARC in KYSE30 cells transduced with full-length HNRNPC and its truncated forms. I. CircSPARC pull-down combined with western blot to detect the enriched Flag-HNRNPC in KYSE150 cells transduced with full-length HNRNPC and its truncated forms. J. Predicted binding motif HNRNPC on target RNAs from RBPsuite. K. Schematic diagram illustrating the putative HNRNPC-binding regions and mutations within circSPARC. L. CircSPARC pull-down combined with western blot to detect the enriched HNRNPC in KYSE30 cells transfected with wild-type or mutant HNRNPC-binding site type of circSPARC. M. Anti-HNRNPC immunoprecipitation followed by RT-PCR assay to measure the enriched circSPARC in KYSE150 cells transfected with wild-type or mutant HNRNPC-binding site type of circSPARC. N. The colony formation and transwell assays measured the changes in KYSE30 cells after knockdown of HNRNPC and co-overexpression with circSPARC wild type or HNRNPC binding site mutant type. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. O. The colony formation and transwell assays measured the changes in KYSE150 cells overexpressing circSPARC or co-overexpressing HNRNPC full-length or without the RRM domain. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. *P < 0.05, **P < 0.01, ***P < 0.001, ns: not significant To further validate the binding interaction between HNRNPC and circSPARC, we designed and synthesized a series of truncated HNRNPC constructs (Fig. 2G), which revealed that the RNA recognition motif (RRM) domain of HNRNPC mediates its interaction with circSPARC (Fig. 2H-I). To identify the specific region within circSPARC responsible for HNRNPC binding, we conducted bioinformatic analysis using the RBPsuite database, which predicts that HNRNPC preferentially recognizes poly-uridine motifs (UUUUU) in target RNA sequences (Fig. 2J). Based on this prediction, we identified a putative HNRNPC-binding site containing the UUUUU motif within circSPARC and created a plasmid with mutations at this site (Fig. 2K). RNA pull-down assays showed that the circSPARC probe efficiently enriched HNRNPC protein in KYSE30 cells transfected with circ-WT, whereas the interaction was significantly reduced in cells expressing circ-Mut (Fig. 2L). Additionally, anti-HNRNPC immunoprecipitation assays showed significantly reduced enrichment of circSPARC enrichment in the circ-Mut group compared to the circ-WT group (Fig. 2M). Overall, these findings confirm that the poly-uridine motif within circSPARC is essential for its interaction with the RRM domain of HNRNPC. To investigate whether HNRNPC participates in the functional regulation of circSPARC function in ESCC, we first analyzed TCGA and GSE53624 datasets, both of which showed significant upregulation of HNRNPC in ESCC tissues compared to normal counterparts (Supplementary Fig. 3I-J). Survival analysis further indicated that higher HNRNPC expression consistently correlates with poorer overall survival in esophageal cancer patients (Supplementary Fig. 3 K-L). We then investigated the reciprocal regulatory relationship between circSPARC and HNRNPC and found that altering circSPARC levels had no noticeable impact on HNRNPC protein expression, and depletion or overexpression of HNRNPC did not significantly affect circSPARC levels (Supplementary Fig. 3M-P). Functional rescue experiments revealed that overexpression of circ-Mut did not promote tumorigenic phenotypes, while the tumor-promoting effects of circ-WT were eliminated after HNRNPC knockdown, indicating that circSPARC’s oncogenic effects depend on its intact binding site (Fig. 2N, Supplementary Fig. 3Q). Additionally, HNRNPC-FL overexpression more strongly enhanced malignant behaviors in ESCC cells than the HNRNPC-Del-1 mutant. Notably, in cells expressing circ-WT, HNRNPC-FL caused significantly greater increases in colony formation and migration compared to HNRNPC-Del-1 (Fig. 2O, Supplementary Fig. 3R). Collectively, these results demonstrate that the pro-tumorigenic function of circSPARC in ESCC cells is mediated through its specific interaction with HNRNPC. circSPARC promotes skipping of EPN2 exon through its interaction with HNRNPC We next sought to elucidate the downstream mechanisms through which the circSPARC/ HNRNPC complex influences ESCC progression. HNRNPC is a well- known RNA- binding protein that plays a critical role in pre- mRNA splicing [29], suggesting that circSPARC may affect alternative splicing (AS) in ESCC via its interaction with HNRNPC. To explore this, we performed high- throughput RNA-seq on KYSE 30 cells with and without circSPARC knockdown to examine changes in alternative splicing on a global scale. The screening threshold was set at a false discovery rate (FDR) < 0. 05 and a percent spliced- in (|ΔPSI|) > 0. 2. Compared to controls, 1,332 differential splicing events were identified in circSPARC- knockdown cells (Fig. 3A). To find potential downstream targets, we followed the workflow outlined in Fig. 3B, identifying 8 overlapping genes. By analyzing alternative splicing events linked to these genes in our sequencing data using the Ensemble database and considering literature evidence, we prioritized epsin 2 (EPN2) for further study (Fig. 3B). EPN2, a classic endocytic adaptor protein, influences multiple signaling pathways and tumor progression by regulating membrane receptor endocytosis; however, its alternative splicing patterns have not been previously reported [30, 31]. As shown in Fig. 3C, exon 3 of EPN2 was differentially skipped upon circSPARC depletion producing two main isoforms: full- length EPN2 (EPN2- L) and a truncated variant (EPN2-S). We validated these splicing events experimentally following circSPARC knockdown or overexpression in ESCC cells. The results showed that circSPARC depletion shifted EPN2 splicing toward the EPN2- L isoform, while overexpression had the opposite effect. Notably, HNRNPC regulated EPN2 splicing similarly to circSPARC (Fig. 3D). Furthermore, RT-qPCR with transcript-specific primers indicated that silencing either circSPARC or HNRNPC decreased mRNA levels of the EPN2-S isoform and increased levels of EPN2-L. Conversely, overexpression produced opposite effects (Supplementary Fig. 4A-D). Furthermore, we examined the association between EPN2 protein levels and circSPARC status. Using an antibody recognizing a shared epitope, we consistently detected a band corresponding to the EPN2-L isoform, whereas the EPN2-S isoform was not reliably detectable at the endogenous protein level. Notably, EPN2-L protein levels were increased upon circSPARC knockdown and decreased upon circSPARC overexpression (Supplementary Fig. 4E). Importantly, we confirmed that circSPARC influences EPN2 splicing in tumor tissues from xenograft models, with results aligning with those from in vitro experiments (Fig. 3E-F). Fig. 3circSPARC promotes skipping of EPN2 exon through its interaction with HNRNPC. A. Dot plot illustrated the PSI (Percent Spliced In) values of alternative splicing (AS) events in control versus circSPARC-silenced KYSE30 cells, with down-regulated (sapphire blue) and up-regulated (orange) AS events prominently marked. B. A flowchart of the identification of AS target genes regulated by circSPARC-HNRNPC interaction. C. Sashimi plots showing increased inclusion of EPN2 exon 3 in the circSPARC-knockdown KYSE30 cells compared with control cells (top) and gene structure of two EPN2 transcripts (bottom). D. RT-PCR verification of EPN2 alternative splicing in circSPARC or HNRNPC knockdown or overexpressed cells. E-F. RT-qPCR assays were conducted with EPN2 isoform-specific primers in subcutaneous xenograft tumor tissues from nude mice. Data are presented as the means ± SEMs. Statistical significance was calculated using the Mann-Whitney U test. G. Predicted HNRNPC binding motifs with scores in the RNA sequence of human EPN2 intron 2, analyzed using the SpliceAid online program. H. Anti-HNRNPC immunoprecipitation followed by RT-qPCR assessed the enriched EPN2 pre-mRNA in the HNRNPC group compared with the IgG group. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using Student’s t-test. I. The heatmap displayed the predicted binding regions between circSPARC and intron 2 of EPN2 according to the IntaRNA database. J. RNA Antisense purification combined with RT-qPCR detected the enriched EPN2 pre-mRNA in the circSPARC probe group compared with the negative control probe. Data are presented as the means ± SEMs. Statistical significance was calculated using Student’s t-test. K-L. RT-qPCR and RT-PCR assays evaluated the influence of circSPARC knockdown on HNRNPC binding to EPN2 pre-mRNA in KYSE30 cells. Data are presented as the means ± SEMs. Statistical significance was calculated using Student’s t-test. M-N. RT-qPCR and RT-PCR analysis assessed HNRNPC-EPN2 pre-mRNA binding following the overexpression of circSPARC wild-type or HNRNPC binding site mutant types. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. O. RT-PCR verification of EPN2 alternative splicing in KYSE30 and KYSE150 cells with HNRNPC full-length or without the RRM domain overexpressed. P. RT-PCR confirmation of EPN2 alternative splicing in KYSE30 and KYSE150 cells with circSPARC wild type or HNRNPC binding site mutant type overexpressed. *P < 0.05, **P < 0.01, ***P < 0.001 circSPARC promotes skipping of EPN2 exon through its interaction with HNRNPC. A. Dot plot illustrated the PSI (Percent Spliced In) values of alternative splicing (AS) events in control versus circSPARC-silenced KYSE30 cells, with down-regulated (sapphire blue) and up-regulated (orange) AS events prominently marked. B. A flowchart of the identification of AS target genes regulated by circSPARC-HNRNPC interaction. C. Sashimi plots showing increased inclusion of EPN2 exon 3 in the circSPARC-knockdown KYSE30 cells compared with control cells (top) and gene structure of two EPN2 transcripts (bottom). D. RT-PCR verification of EPN2 alternative splicing in circSPARC or HNRNPC knockdown or overexpressed cells. E-F. RT-qPCR assays were conducted with EPN2 isoform-specific primers in subcutaneous xenograft tumor tissues from nude mice. Data are presented as the means ± SEMs. Statistical significance was calculated using the Mann-Whitney U test. G. Predicted HNRNPC binding motifs with scores in the RNA sequence of human EPN2 intron 2, analyzed using the SpliceAid online program. H. Anti-HNRNPC immunoprecipitation followed by RT-qPCR assessed the enriched EPN2 pre-mRNA in the HNRNPC group compared with the IgG group. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using Student’s t-test. I. The heatmap displayed the predicted binding regions between circSPARC and intron 2 of EPN2 according to the IntaRNA database. J. RNA Antisense purification combined with RT-qPCR detected the enriched EPN2 pre-mRNA in the circSPARC probe group compared with the negative control probe. Data are presented as the means ± SEMs. Statistical significance was calculated using Student’s t-test. K-L. RT-qPCR and RT-PCR assays evaluated the influence of circSPARC knockdown on HNRNPC binding to EPN2 pre-mRNA in KYSE30 cells. Data are presented as the means ± SEMs. Statistical significance was calculated using Student’s t-test. M-N. RT-qPCR and RT-PCR analysis assessed HNRNPC-EPN2 pre-mRNA binding following the overexpression of circSPARC wild-type or HNRNPC binding site mutant types. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. O. RT-PCR verification of EPN2 alternative splicing in KYSE30 and KYSE150 cells with HNRNPC full-length or without the RRM domain overexpressed. P. RT-PCR confirmation of EPN2 alternative splicing in KYSE30 and KYSE150 cells with circSPARC wild type or HNRNPC binding site mutant type overexpressed. *P < 0.05, **P < 0.01, ***P < 0.001 We further explored how circSPARC regulates exon skipping of EPN2 by interacting with HNRNPC. Using the online SpliceAid program, we predicted that HNRNPC could bind to the intron sequences surrounding exon 3 of EPN2 (Fig. 3G), a prediction confirmed by RNA immunoprecipitation (RIP) assays targeting HNRNPC (Fig. 3H). Subsequent analysis with the IntaRNA program revealed potential interaction sites between circSPARC and the intron regions flanking EPN2 exon 3 (Fig. 3I), which were experimentally validated through a circSPARC pull-down assay (Fig. 3J). Thus we hypothesized that circSPARC might influence HNRNPC’s binding to pre-EPN2. Consistent with this, RIP assays for HNRNPC in control and circSPARC knockdown or overexpressing ESCC cells showed that depleting circSPARC significantly reduced HNRNPC’s binding to pre-EPN2 (Fig. 3K-L). Conversely, overexpression of circ-WT increased the enrichment of pre-EPN2, while circ-Mut had no such effect (Fig. 3M-N). Importantly, neither circSPARC knockdown nor overexpression changed pre-EPN2 expression levels (Supplementary Fig. 4F), ruling out transcriptional regulation as a mechanism underlying changes in pre-EPN2 enrichment. Additionally, overexpression of HNRNPC-Del-1 or circ-Mut eliminated the effects of HNRNPC or circSPARC on EPN2’s alternative splicing (Fig. 3O-P). These results collectively indicate that the interaction between circSPARC and HNRNPC enhances the recruitment of HNRNPC to pre-EPN2, thereby modulating alternative splicing and shifting the balance toward the EPN2-S isoform. Exon skipping of EPN2 mediates the regulatory effect of circSPARC on the malignant phenotypes in ESCC cells We further evaluated the clinical significance of the circSPARC/HNRNPC complex-mediated EPN2 splicing. First, we measured the expression levels of EPN2-L and EPN2-S in 60 paired ESCC and adjacent normal tissues from Cohort 1. Consistently, the EPN2-L/EPN2-S expression ratio was significantly reduced in tumor tissues and showed a strong negative correlation with circSPARC expression levels (Fig. 4A-B). Notably, survival analysis indicated that ESCC patients with a higher EPN2-L/EPN2-S expression ratio tended to have better overall survival (Supplementary Fig. 5A). To comprehensively evaluate the clinical relevance of EPN2 splicing in ESCC, we analyzed the TCGA ESCC dataset.The result demonstrated that the PSI value of EPN2 was decreased in ESCC samples as compared with normal tissues(Supplementary Fig. 5B). Further survival analysis revealed that higher EPN2 PSI levels are closely associated with improved overall survival in ESCC patients (Fig. 4C). Furthermore, RT-PCR analysis of 12 pairs of clinically matched tissue samples revealed a significant decrease in EPN2-L mRNA levels in ESCC tissues compared to the paired normal samples. Meanwhile, circSPARC expression was elevated in tumor samples (Supplementary Figs. 5 C, D). Notably, the increase in circSPARC expression was inversely related to the PSI value of EPN2 (Fig. 4D). Fig. 4Exon skipping of EPN2 mediates the regulatory effect of circSPARC on the malignant phenotypes in ESCC cells. A. RT-qPCR was used to quantify the relative expression ratio of EPN2-L to EPN2-S in 60 paired ESCC tissues and adjacent normal tissues from cohort 1. Statistical significance was calculated using Student’s t-test. B. Pearson correlation analysis was performed between circSPARC expression levels and EPN2-L/EPN2-S in ESCC tissues (n = 60). C. Kaplan-Meier survival analysis was performed to assess the association between EPN2 splicing alterations and overall survival in ESCC patients from the TCGA database. D. Correlations between circSPARC/GAPDH levels and EPN2 PSIs were analyzed (Pearson) in 12 esophageal tissues. E. Western blot analysis of the EPN2 protein levels in KYSE30 and KYSE150 cells overexpressing His-tagged EPN2-L or EPN2-S. F. The colony formation assay evaluated the impact of EPN2-L or EPN2-S overexpression on the clonogenic potential of KYSE30 and KYSE150 cells. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. G-H. The RTCA analysis evaluated cell proliferation in EPN2-L- or EPN2-S-overexpressed KYSE30 and KYSE150 cells. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. I. Representative images of xenograft tumors in nude mice inoculated with EPN2-L stably overexpressed or control KYSE30 cells (n=6). J-K. Quantification of tumor weight and tumor growth in EPN2-L overexpressed or control groups. Data are presented as the means ± SEMs. Statistical significance was calculated using the Mann-Whitney U test. L. Representative IHC images of Ki-67 staining in the subcutaneous xenograft tumor model in nude mice. M-N. The RTCA analyses were performed using KYSE30 and KYSE150 cells stably overexpressing circSPARC or co-overexpressing EPN2-L. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. O. The colony formation assays were conducted using KYSE30 and KYSE150 cells as described in (M-N). Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. **P < 0.01, ***P < 0.001, ns: not significant Exon skipping of EPN2 mediates the regulatory effect of circSPARC on the malignant phenotypes in ESCC cells. A. RT-qPCR was used to quantify the relative expression ratio of EPN2-L to EPN2-S in 60 paired ESCC tissues and adjacent normal tissues from cohort 1. Statistical significance was calculated using Student’s t-test. B. Pearson correlation analysis was performed between circSPARC expression levels and EPN2-L/EPN2-S in ESCC tissues (n = 60). C. Kaplan-Meier survival analysis was performed to assess the association between EPN2 splicing alterations and overall survival in ESCC patients from the TCGA database. D. Correlations between circSPARC/GAPDH levels and EPN2 PSIs were analyzed (Pearson) in 12 esophageal tissues. E. Western blot analysis of the EPN2 protein levels in KYSE30 and KYSE150 cells overexpressing His-tagged EPN2-L or EPN2-S. F. The colony formation assay evaluated the impact of EPN2-L or EPN2-S overexpression on the clonogenic potential of KYSE30 and KYSE150 cells. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. G-H. The RTCA analysis evaluated cell proliferation in EPN2-L- or EPN2-S-overexpressed KYSE30 and KYSE150 cells. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. I. Representative images of xenograft tumors in nude mice inoculated with EPN2-L stably overexpressed or control KYSE30 cells (n=6). J-K. Quantification of tumor weight and tumor growth in EPN2-L overexpressed or control groups. Data are presented as the means ± SEMs. Statistical significance was calculated using the Mann-Whitney U test. L. Representative IHC images of Ki-67 staining in the subcutaneous xenograft tumor model in nude mice. M-N. The RTCA analyses were performed using KYSE30 and KYSE150 cells stably overexpressing circSPARC or co-overexpressing EPN2-L. Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. O. The colony formation assays were conducted using KYSE30 and KYSE150 cells as described in (M-N). Data are presented as the means ± SEMs from three independent experiments. Statistical significance was calculated using One-way ANOVA. **P < 0.01, ***P < 0.001, ns: not significant Given their correlated expression patterns in clinical samples, we next determined whether circSPARC functionally contributes to ESCC aggressiveness through regulating EPN2 alternative splicing. First, we transfected ESCC cells with plasmids to overexpress either EPN2-L or EPN2-S (Fig. 4E). Functional assays, including RTCA, colony formation, and transwell migration and invasion tests, showed that ectopic expression of EPN2-L significantly reduced proliferation, clonogenicity, migration, and invasion of ESCC cells, while overexpression of EPN2-S had no notable effect on these malignant traits (Fig. 4F-H, Supplementary Fig. 5E-F). Moreover, EPN2-L overexpression also slowed tumor growth in vivo (Fig. 4I-L). These results imply that EPN2-L, not EPN2-S, strongly suppresses the malignant behaviors of ESCC cells. Subsequently, we co-overexpressed circSPARC and EPN2-L in KYSE30 and KYSE150 cells and performed functional assays. Consistently, restoration of EPN2-L expression fully counteracts the pro-tumor effects induced by circSPARC overexpression on cell proliferation, colony formation, migration and invasion, demonstrating that EPN2-L is a key molecule for circSPARC to exert its regulatory functions (Fig. 4M-O, Supplementary Fig. 5G). Overall, our findings suggest that circSPARC promotes ESCC progression, at least partly, by altering EPN2 splicing to reduce the expression of the EPN2-L isoform. EPN2-L inactivates the canonical Wnt/β-catenin pathway by promoting FZD7 endocytosis and degradation We next aimed to elucidate the molecular mechanisms underlying the functional differences resulting from EPN2 alternative splicing. Compared to EPN2-L, the EPN2-S variant encodes a protein lacking the Epsin N-Terminal Homology (ENTH) domain, which is a key part involved in clathrin-mediated endocytosis (CME), a well-orchestrated process responsible for packaging, sorting, and internalizing cell surface proteins for degradation or recycling [32, 33] (Fig. 5A). We hypothesized that the absence of the ENTH domain in EPN2-S may impair its ability to mediate membrane protein endocytosis, potentially disrupting signal transduction and contributing to the malignant progression of ESCC. To test this, we performed mass spectrometry analysis on KYSE30 cells overexpressing either EPN2-L or EPN2-S, focusing on proteins that were differentially expressed in the EPN2-L group but unchanged in the EPN2-S group (Fig. 5B, Supplementary Table 6). KEGG pathway enrichment analysis revealed these proteins were significantly enriched in the Wnt/β-catenin signaling pathway (Fig. 5C). Validation experiments showed that overexpression of EPN2-L in KYSE30 and KYSE150 cells reduced levels of active β-catenin and significantly downregulated C-Myc, cyclin D1, and Axin2-three well-known downstream effectors of Wnt/β-catenin signaling. In contrast, EPN2-S overexpression had no such effect (Fig. 5D). Additionally, immunofluorescence analyses demonstrated that ectopic expression of EPN2-L markedly suppressed Wnt3a-induced nuclear accumulation of β-catenin in both cell lines (Fig. 5E). Fig. 5EPN2-L inactivates the canonical Wnt/β-catenin pathway by promoting FZD7 endocytosis and degradation. A. Schematic diagram of EPN2-L and EPN2-S protein structures. B. A Venn diagram displayed the differentially expressed proteins between the EPN2-L and EPN2-S overexpression groups. C. KEGG pathway enrichment analysis of proteomic alterations induced by EPN2-L overexpression relative to EPN2-S overexpression in KYSE30 cells. D. The Western blot assays evaluated the protein levels of active β-catenin (the non-phosphorylated form at Ser33/37/Thr41 residues) and target genes in EPN2-L- or EPN2-S-overexpressed KYSE30 and KYSE150 cells. E. IF staining analysis of the subcellular localization of β-catenin in EPN2-L overexpressed KYSE30 and KYSE150 cells. Scale bars, 100µm. F. The western blot assays examined the protein levels of active β-catenin and target genes in KYSE30 cells after co-treatment with EPN2 overexpression and the endocytosis inhibitor, chlorpromazine (10 µM). G. The western blot analyses evaluated the expression levels of FZD7 in KYSE30 and KYSE150 cells overexpressing EPN2-L or EPN2-S. H. Western blot analysis was conducted to assess the impact of EPN2-L and FZD7 co-overexpression on the protein levels of active β-catenin and target genes in KYSE30 cells. I. Anti-His co-immunoprecipitation was conducted to assess the interaction between EPN2 variants (EPN2-L/EPN2-S) and FZD7 in HEK293T cells individually overexpressing EPN2-L or EPN2-S. J. Western blot assays detected the FZD7 protein levels in KYSE30 and KYSE150 cells overexpressing EPN2-L with or without Chlorpromazine (10 µM) treatment. K. Biotinylated FZD7 internalization was analyzed through streptavidin bead precipitation and subsequent detection using an FZD7 antibody-based western blot in KYSE30 cells overexpressing EPN2-L, TFRC used as a cargo control. L. The cell surface-biotinylated FZD7 was detected via streptavidin bead pull-down followed by FZD7 antibody-based western blot analysis in KYSE150 cells overexpressing EPN2-L. M. Western blot assays detected the FZD7 protein levels in KYSE30 and KYSE150 cells overexpressing EPN2-L with or without bafilomycin A1 (100 nM) treatment, an inhibitor of lysosomal acidification. N. The co-localization of FZD7 with the endosomal marker EEA1 and the lysosomal marker LAMP1 was assessed by immunofluorescence staining. Scale bars, 100µm EPN2-L inactivates the canonical Wnt/β-catenin pathway by promoting FZD7 endocytosis and degradation. A. Schematic diagram of EPN2-L and EPN2-S protein structures. B. A Venn diagram displayed the differentially expressed proteins between the EPN2-L and EPN2-S overexpression groups. C. KEGG pathway enrichment analysis of proteomic alterations induced by EPN2-L overexpression relative to EPN2-S overexpression in KYSE30 cells. D. The Western blot assays evaluated the protein levels of active β-catenin (the non-phosphorylated form at Ser33/37/Thr41 residues) and target genes in EPN2-L- or EPN2-S-overexpressed KYSE30 and KYSE150 cells. E. IF staining analysis of the subcellular localization of β-catenin in EPN2-L overexpressed KYSE30 and KYSE150 cells. Scale bars, 100µm. F. The western blot assays examined the protein levels of active β-catenin and target genes in KYSE30 cells after co-treatment with EPN2 overexpression and the endocytosis inhibitor, chlorpromazine (10 µM). G. The western blot analyses evaluated the expression levels of FZD7 in KYSE30 and KYSE150 cells overexpressing EPN2-L or EPN2-S. H. Western blot analysis was conducted to assess the impact of EPN2-L and FZD7 co-overexpression on the protein levels of active β-catenin and target genes in KYSE30 cells. I. Anti-His co-immunoprecipitation was conducted to assess the interaction between EPN2 variants (EPN2-L/EPN2-S) and FZD7 in HEK293T cells individually overexpressing EPN2-L or EPN2-S. J. Western blot assays detected the FZD7 protein levels in KYSE30 and KYSE150 cells overexpressing EPN2-L with or without Chlorpromazine (10 µM) treatment. K. Biotinylated FZD7 internalization was analyzed through streptavidin bead precipitation and subsequent detection using an FZD7 antibody-based western blot in KYSE30 cells overexpressing EPN2-L, TFRC used as a cargo control. L. The cell surface-biotinylated FZD7 was detected via streptavidin bead pull-down followed by FZD7 antibody-based western blot analysis in KYSE150 cells overexpressing EPN2-L. M. Western blot assays detected the FZD7 protein levels in KYSE30 and KYSE150 cells overexpressing EPN2-L with or without bafilomycin A1 (100 nM) treatment, an inhibitor of lysosomal acidification. N. The co-localization of FZD7 with the endosomal marker EEA1 and the lysosomal marker LAMP1 was assessed by immunofluorescence staining. Scale bars, 100µm Considering the canonical role of EPN2 in CME, we investigated whether EPN2-L-mediated suppression of Wnt/β-catenin signaling is associated with its function in endocytosis. As expected, treatment with chlorpromazine, a well-established inhibitor of CME, reversed the suppression of Wnt/β-catenin signaling caused by EPN2-L overexpression (Fig. 5F, Supplementary Fig. 6A). Recent evidence indicates that CME reduces Wnt signaling by decreasing the membrane localization of the receptor complex [34, 35]. Interestingly, mass spectrometry analysis and validation experiment showed that the Wnt receptor Frizzled-7 (FZD7) was significantly downregulated in the EPN2-L overexpression group (Fig. 5G). Importantly, FZD7 overexpression reversed the inhibitory effects of EPN2-L on Wnt/β-catenin signaling (Fig. 5H, Supplementary Fig. 6B). To explore the underlying mechanism, we next examined the interaction between EPN2 and FZD7. Co-immunoprecipitation assays showed that the EPN2-S variant, which lacks the ENTH domain, has mild interaction with FZD7, suggesting that a complete ENTH domain is necessary for effective binding (Fig. 5I). To assess the functional significance of endocytosis, cells were treated with chlorpromazine, which significantly decreased the downregulation of FZD7 caused by EPN2-L overexpression (Fig. 5J), supporting a role for CME in this process. Finally, to directly assess whether EPN2-L affects FZD7 protein stability, we performed a cycloheximide (CHX) pulse-chase experiment. The results showed that EPN2-L markedly accelerated the degradation of FZD7 compared to control cells (Supplementary Fig. 6C). We also measured FZD7 internalization using a biotinylation-based internalization assay, revealing that EPN2-L overexpression significantly increased the internalized fraction of FZD7, but had no significant effect on the internalization of transferrin receptor (TFRC), a canonical cargo of clathrin-mediated endocytosis [36], indicating that EPN2-L selectively promotes the endocytosis of FZD7 rather than broadly affecting receptor internalization (Fig. 5K). Consistent with this, cell surface biotinylation experiments indicated that EPN2-L upregulation reduced the FZD7 levels at the plasma membrane (Fig. 5L). Since cargo internalized via CME is usually directed toward lysosomal degradation, we tested whether the decrease in FZD7 levels was due to lysosomal processing. Treatment with bafilomycin A1, an inhibitor of lysosomal acidification, reversed the FZD7 reduction caused by EPN2-L overexpression (Fig. 5M). Lastly, confocal microscopy confirmed that EPN2-L overexpression increased the colocalization of FZD7 with the early endosomal marker EEA1 and the lysosomal marker LAMP1 (Fig. 5N). Taken together, our results demonstrate that EPN2-L acts as a negative regulator of Wnt/β-catenin signaling by promoting FZD7 endocytosis and its subsequent lysosomal degradation. circSPARC sustains activation of the canonical Wnt/β-catenin signaling pathway by suppressing EPN2-L expression We next investigated the clinical relevance of FZD7 in the progression of ESCC. The result showed that FZD7 was upregulated in tumor tissues compared to adjacent normal tissues in clinical cohort 2 (Fig. 6A-B). Additionally, a detailed analysis showed that patients with lymph node metastasis, advanced T stage, and larger tumor size exhibited high FZD7 levels (Fig. 6C-E). Consistent with the idea that circSPARC prevents FZD7 degradation, we observed a positive correlation between FZD7 expression and circSPARC levels in ESCC tissues (Fig. 6F). Fig. 6circSPARC sustains activation of the canonical Wnt/β-catenin signaling pathway by suppressing EPN2-L expression. A-B. A comparative evaluation of FZD7 immunostaining scores was conducted between ESCC tissues and matched adjacent normal tissues. Statistical significance was calculated using Student’s t-test. C-E. The correlations between FZD7 expression and lymph node metastasis, T stage, and tumor size of ESCC patients were analyzed and presented using a box plot. Statistical significance was calculated using the Mann-Whitney U test. F. The correlation between FZD7 expression and circSPARC expression in ESCC tissues was assessed. G-H. The protein levels of FZD7, active β-catenin, and target genes were evaluated by western blot analysis in circSPARC knockdown KYSE30 and KYSE150 cells. I-J. The protein levels of FZD7, active β-catenin, and target genes were evaluated by Western blot analysis in circSPARC wild-type or HNRNPC-binding site mutation-type overexpressed KYSE30 and KYSE150 cells. K-L. Western blot assays were performed in KYSE30 and KYSE150 cells co-overexpressing EPN2-L and circSPARC to measure the expression of FZD7, active β-catenin, and target genes. *P < 0.05 circSPARC sustains activation of the canonical Wnt/β-catenin signaling pathway by suppressing EPN2-L expression. A-B. A comparative evaluation of FZD7 immunostaining scores was conducted between ESCC tissues and matched adjacent normal tissues. Statistical significance was calculated using Student’s t-test. C-E. The correlations between FZD7 expression and lymph node metastasis, T stage, and tumor size of ESCC patients were analyzed and presented using a box plot. Statistical significance was calculated using the Mann-Whitney U test. F. The correlation between FZD7 expression and circSPARC expression in ESCC tissues was assessed. G-H. The protein levels of FZD7, active β-catenin, and target genes were evaluated by western blot analysis in circSPARC knockdown KYSE30 and KYSE150 cells. I-J. The protein levels of FZD7, active β-catenin, and target genes were evaluated by Western blot analysis in circSPARC wild-type or HNRNPC-binding site mutation-type overexpressed KYSE30 and KYSE150 cells. K-L. Western blot assays were performed in KYSE30 and KYSE150 cells co-overexpressing EPN2-L and circSPARC to measure the expression of FZD7, active β-catenin, and target genes. *P < 0.05 Since circSPARC functions as a negative regulator of EPN2-L, we investigated whether it influences FZD7 and the Wnt signaling pathway through EPN2-L. Following circSPARC knockdown, the suppression of FZD7 expression was reduced along with decreased levels of active β-catenin, C-Myc, cyclin D1, and Axin2 in KYSE30 and KYSE150 cells (Fig. 6G-H). Moreover, only overexpression of circ-WT, not circ-Mut, increased FZD7, active β-catenin, and the expression of downstream targets (Fig. 6I-J). Additionally, rescue experiments demonstrated that re-expressing EPN2-L reversed the effects of circSPARC on FZD7 expression and Wnt/β-catenin signaling (Fig. 6K-L). Overall, these findings suggest that circSPARC activates Wnt/β-catenin signaling in ESCC by suppressing EPN2-L production, which inhibits clathrin-mediated endocytosis and lysosomal degradation of FZD7. In summary, this study identifies a novel circSPARC-EPN2-L-FZD7 regulatory axis that activates Wnt/β-catenin signaling and is clinically associated with ESCC progression. Inhibition of the Wnt/β-catenin signaling pathway abrogates circSPARC-driven ESCC progression Given that circSPARC promotes activation of the Wnt/β-catenin pathway, we next investigated whether disrupting this pathway could reverse the accelerated tumor progression induced by circSPARC-overexpressing xenograft tumors (Fig. 7A). In line with our predictions, treatment with XAV-939, a tankyrase inhibitor that suppresses β-catenin activity, significantly reduces the accelerated tumor growth induced by circSPARC overexpression in subcutaneous xenografts (Fig. 7B-C). Moreover, immunohistochemical analysis demonstrated elevated levels of Ki-67, active β-catenin, and FZD7 in circSPARC-overexpressing xenografts, all of which were reduced following XAV-939 administration (Fig. 7D). Collectively, our data suggest that blocking the Wnt/β-catenin signaling pathway suppresses the malignant progression of ESCC driven by circSPARC overexpression. Fig. 7Inhibition of the Wnt/β-catenin signaling pathway abrogates circSPARC-driven ESCC progression. A. Schematic illustration of the experimental design for the establishment of the xenograft model followed by XAV-939 treatment. B. The macroscopic images of subcutaneous tumors in BALB/c nude mice subcutaneously injected with circSPARC overexpressed KYSE30 cells, followed by treatment with or without XAV-939 (0.2 mM). C. Quantification of tumor weight and tumor volume in different groups. Data are presented as the means ± SEMs. Statistical significance was calculated using One-way ANOVA. D. Representative histological staining of Ki-67, active β-catenin, and FZD7 in subcutaneous xenograft tumors. Data are presented as the means ± SEMs. Statistical significance was calculated using One-way ANOVA. E. The schematic diagram of this study. The proposed model illustrates that through binding with HNRNPC, circSPARC modulates EPN2 splicing to downregulate the EPN2-L variant. The reduced EPN2-L levels impair FZD7 internalization and degradation, leading to sustained Wnt/β-catenin pathway activation that drives ESCC development. *P < 0.05, **P < 0.01, ***P < 0.001 Inhibition of the Wnt/β-catenin signaling pathway abrogates circSPARC-driven ESCC progression. A. Schematic illustration of the experimental design for the establishment of the xenograft model followed by XAV-939 treatment. B. The macroscopic images of subcutaneous tumors in BALB/c nude mice subcutaneously injected with circSPARC overexpressed KYSE30 cells, followed by treatment with or without XAV-939 (0.2 mM). C. Quantification of tumor weight and tumor volume in different groups. Data are presented as the means ± SEMs. Statistical significance was calculated using One-way ANOVA. D. Representative histological staining of Ki-67, active β-catenin, and FZD7 in subcutaneous xenograft tumors. Data are presented as the means ± SEMs. Statistical significance was calculated using One-way ANOVA. E. The schematic diagram of this study. The proposed model illustrates that through binding with HNRNPC, circSPARC modulates EPN2 splicing to downregulate the EPN2-L variant. The reduced EPN2-L levels impair FZD7 internalization and degradation, leading to sustained Wnt/β-catenin pathway activation that drives ESCC development. *P < 0.05, **P < 0.01, ***P < 0.001 Based on these findings, we delineate a well-defined molecular cascade in which circSPARC recruits HNRNPC to suppress the tumor-suppressive long isoform of EPN2. This suppression impairs clathrin-mediated endocytosis, leading to FZD7 accumulation and sustained activation of the Wnt/β-catenin signaling pathway, thereby driving ESCC progression (Fig. 7E). Clinically, the circSPARC-HNRNPC-EPN2-FZD7 axis is strongly supported by its significant correlation with poor patient prognosis. Together, our study provides a compelling therapeutic rationale for ESCC, especially in tumors with high circSPARC expression. The mechanism identified shows that persistent activation of the Wnt/β-catenin pathway results from this specific post-transcriptional regulation, rather than from conventional genetic alterations. Consequently, therapeutic strategies could be precisely targeted at multiple nodes, initially using Wnt/β-catenin inhibitors in a biomarker-guided approach. More importantly, our results suggest the potential for developing targeted agents that either restore EPN2-mediated FZD7 degradation or directly inhibit the core circSPARC-HNRNPC splicing complex, offering a new approach for precision therapy in ESCC. Discussion The widespread dysregulation of circRNAs in cancers is frequently associated with a poor prognosis; however, the mechanistic basis by which they promote malignancy, especially through modulation of RNA processing, remains largely unclear. Here, we identify circSPARC as a new oncogenic driver in ESCC that functions through a previously unknown mechanism: reprogramming alternative splicing to stabilize a key Wnt receptor. We demonstrate that circSPARC recruits HNRNPC to the EPN2 pre-mRNA, leading to the suppression of the tumor-suppressive long isoform (EPN2-L). This downregulation impairs clathrin-mediated endocytosis and lysosomal degradation of FZD7, causing its accumulation and sustained activation of the Wnt/β-catenin signaling pathway, ultimately leading to ESCC progression. Our findings establish a direct molecular connection between circSPARC dysregulation and a core cancer pathway through a unique circRNA-mediated splicing regulatory mechanism. Beyond defining this pro-tumorigenic axis, our findings reveal distinct and previously unrecognized aspects of circRNA biology. First, the subcellular localization of circRNAs provides critical insights into their functional roles and regulatory mechanisms [22]. Prior studies have primarily characterized cytoplasmic circRNA-HNRNPC interactions as a conserved mechanism regulating mRNA stability [37–39]. We observed predominant nuclear co-localization of circSPARC and HNRNPC. This led us to uncover their involvement in regulating alternative splicing. Collectively, these results suggest that circRNA binding may drive subcellular localization-dependent functional reprogramming of RBPs. Second, we identify a novel “scaffold-recruitment” mechanism of action. Unlike the common model where circRNAs sequester splicing factors [24, 25], circSPARC functions as an active molecular scaffold that facilitates the recruitment of HNRNPC to the EPN2 pre-mRNA, thereby enhancing its ability to suppress inclusion of exons that define the EPN2-L isoform. It should be noted that while our data demonstrate that circSPARC promotes HNRNPC-pre-EPN2 association, we have not formally proven that all three components simultaneously coexist in a stable ternary complex; direct biochemical evidence would be required for such a claim. Nevertheless, this “scaffold-recruitment” model introduces a new paradigm for regulating alternative splicing by circRNAs, expanding the known functional diversity of circRNAs and emphasizing their potential for precise, transcript-specific regulation through the dynamic assembly of macromolecular complexes. Whether this complex exhibits strict specificity for EPN2 remains an intriguing question for future research. Furthermore, we show that increased EPN2-L PSI in ESCC correlates with better patient survival, supporting its clinical relevance. However, public datasets lack concurrent circRNA and splicing data at sufficient resolution, precluding direct assessment of the circSPARC-EPN2 splicing relationship in patient cohorts. Future integrative studies combining circRNA profiling with splicing quantification are needed to address this question. Clathrin-mediated endocytosis (CME) plays a multifaceted role in cancer progression by modulating signal transduction pathways, thereby influencing tumor initiation, development, and metastasis [40, 41]. Epsins, which are ubiquitously expressed adaptor proteins, regulate the endocytosis of plasma membrane receptor complexes, facilitating their degradation and thus modulating cellular signaling [42, 43]. Previous studies have demonstrated that EPN1 and EPN2 play a role in the development of lung cancer and atherosclerosis by modulating the endocytosis of VEGFR2 and FGFR1 [30, 44]. In this study, we demonstrate that the circSPARC/HNRNPC-mediated suppression of EPN2-L impairs the endocytosis and subsequent lysosomal degradation of FZD7. Consequently, knockdown of circSPARC restores EPN2-L expression, leading to reduced FZD7 levels and inhibition of β-catenin activity, ultimately suppressing malignant phenotypes in ESCC cells. Although the involvement of endocytosis in Wnt signaling regulation is recognized as complex and context-dependent, our findings reveal a specific mechanism whereby the circSPARC-HNRNPC-EPN2 axis selectively disrupts CME-mediated downregulation of FZD7, thereby activating Wnt/β-catenin signaling in ESCC. This observation aligns with prior reports that Dkk1 and Dab2 selectively divert coreceptor LRP6 to clathrin-mediated endocytosis, thereby inhibiting Wnt signaling transduction [34, 35]. In contrast, caveolin-dependent internalization has been shown to activate Wnt signaling [45]. This discrepancy may arise from the spatiotemporally distinct sorting mechanisms of different endocytic pathways sort receptor complexes. Interestingly, epsin depletion in colorectal cancer reduces canonical Wnt signaling by destabilizing DVL2 through mechanisms that do not involve its traditional role as an endocytic adaptor [46]. This notable tissue-specific functional divergence underscores the highly context-dependent biological functions of epsins, which are likely influenced by significant differences in their interactomes across various tumor microenvironments. Conclusions In summary, our study reveals a pro-tumorigenic circuit in ESCC driven by circSPARC. By acting as a molecular scaffold that facilitates the recruitment of HNRNPC to the EPN2 pre-mRNA, circSPARC reprograms EPN2 splicing, leading to the suppression of the EPN2-L isoform. This downregulation of EPN2-L reduces the clathrin-mediated endocytic degradation of FZD7, resulting in FZD7 accumulation, activation of Wnt/β-catenin signaling, and promotion of tumor progression. From a therapeutic perspective, our findings suggest that elevated circSPARC expression may be associated with increased reliance on Wnt/β-catenin signaling in ESCC cells. Given the challenges in directly targeting circRNAs, targeting downstream Wnt signaling may represent a more feasible strategy in circSPARC driven ESCC. In this context, circSPARC expression level could potentially serve as a biomarker to identify patient subgroups more likely to benefit from Wnt/β-catenin inhibition therapy. This strategy guides the focus toward a precise targeted treatment, providing new insights for improving the prognosis of patients with ESCC. Supplementary Information Supplementary Material 1. Supplementary Material 1. Supplementary Material 2. Supplementary Material 2. Supplementary Material 3. Supplementary Material 3.