The GC tissues and adjacent non-tumorous tissues were collected from patients who underwent surgical resection at Shanghai Tongji Hospital. Immediately after resection, tissue specimens were snap-frozen in liquid nitrogen and stored until further analysis. Clinical and pathological data were obtained from the institutional medical records. All patients provided informed consent before participation in the study. The study was approved by the Medical Ethics Committee of Shanghai Tongji Hospital.

Human GC cell lines were obtained from the Cell Bank of the Chinese Academy of Sciences. The HGC-27 cell line was cultured in Roswell Park Memorial Institute 1640 (RPMI-1640) medium (Servicebio), enriched with 10% fetal bovine serum (FBS, FuHeng Biology). The AGS cell line was maintained in Kaighn’s Modified Ham’s F-12K medium (Servicebio) with 10% FBS (FuHeng Biology). All cells were incubated in a humidified atmosphere at 37 °C with 5% CO₂.

The circNF1 overexpression construct was generated by amplifying full-length human circNF1 via PCR and cloning it into the pLO5-ciR vector (Geneseed Biotech, Guangzhou, China), which contains essential elements required for circRNA formation. For functional inhibition of circNF1, a small interfering RNA targeting the back-splice junction (si-circNF1) was designed and synthesized, as previously described (11). Expression plasmids for IGF2BP1, HMGA2, and ZNF460 were obtained from Miaoling Biology (Wuhan, China). SiRNAs targeting IGF2BP1 (si-IGF2BP1#1 and si-IGF2BP1#2) and HMGA2 (si-HMGA2) were synthesized by Sangon Biotech (Shanghai, China). An empty vector (EV) and non-specific siRNA were used as negative controls (NC). All transfections were performed using Lipofectamine 3000 (Invitrogen), following the manufacturer’s protocol. The sequences of all siRNAs are provided in Supplementary Table S1.

Total RNA was extracted using TRIzol reagent (Invitrogen) and reverse transcribed using the PrimeScript™ RT Master Mix (Takara), as per the manufacturer’s instructions. Quantitative real-time PCR (qRT-PCR) was carried out using TB Green Premix Ex Taq™ II (Takara) on a QuantStudio 6 platform (Life Technologies). Sanger sequencing was used to confirm the back-splice junction of circNF1. All primers used in this study are listed in Supplementary Table S1.

Total protein was isolated using Radio-Immunoprecipitation Assay (RIPA) buffer (NCM Biotech) enriched with protease inhibitor cocktail and phosphatase inhibitor cocktail (APExBIO). Protein concentrations were determined using the Bicinchoninic Acid (BCA) assay kit (Beyotime). Equal amounts of protein lysates were separated by Sodium Dodecyl Sulfate-Polyacrylamide Gel Electrophoresis (SDS-PAGE, Beyotime) and transferred onto Polyvinylidene Fluoride (PVDF) membranes (Immobilon). Membranes were blocked with 5% skim milk for 1 h at room temperature (RT), followed by overnight incubation at 4 °C with primary antibodies specific to target proteins. These membranes were washed thrice with Tris-Buffered Saline containing Tween 20 (TBST, Servicebio) and incubated for 1 h at RT with horseradish peroxidase (HRP)-conjugated goat anti-mouse or anti-rabbit IgG secondary antibodies (Beyotime). Immunoreactive bands were visualized using an Enhanced Chemiluminescence (ECL) detection system (Epizyme). Detailed antibody information is provided in Supplementary Table S2.

Nuclear and cytoplasmic fractions were isolated using the Nuclear and Cytoplasmic Protein Extraction Kit (Beyotime) based on the manufacturer’s instructions. RNA was extracted from both fractions using TRIzol reagent (Invitrogen). GAPDH and U6 were used as cytoplasmic and nuclear markers, respectively.

RNA localization was investigated using the Ribo™ Fluorescent In Situ Hybridization Kit (RiboBio), as per the manufacturer’s protocol. Cy3-labeled probes specific for circNF1, 18S rRNA, and U6 were used for hybridization. Cell nuclei were counterstained with 4’,6-diamidino-2-phenylindole (DAPI, Beyotime). Fluorescence signals were observed under a fluorescence microscope. Probe sequences are listed in Supplementary Table S1.

Paraffin-embedded tissue sections were heated at 60 °C, then dewaxed and rehydrated. Nuclear staining was performed using hematoxylin (Solarbio), and cytoplasmic components were stained with an eosin solution (Solarbio). The stained sections were mounted with neutral resin (Biosharp) for preservation.

In situ hybridization targeting circNF1 was conducted according to the manufacturer’s protocol. Sections hybridized with the circNF1 probe were incubated with an anti-Digoxin-alkaline phosphatase (AP) antibody (Roche), followed by staining with 3,3’-diaminobenzidine (DAB, Vector). The ISH staining score was determined by multiplying the staining intensity by the percentage of positive cells. Scores<4 were categorized as low expression, whereas scores ≥ 4 indicated high expression. Probe sequences are detailed in Supplementary Table S1.

GC cells were fixed with 4% paraformaldehyde (PFA) (Beyotime) for 20 min at RT. After blocking with 10% goat serum for 1 h at RT, cells were incubated overnight at 4 °C with the primary antibody. After PBS washes, cells were incubated with the appropriate fluorescently labeled secondary antibodies for 1 h at 37 °C in the dark. Nuclear staining was performed using DAPI at RT. Fluorescence images were captured using an inverted fluorescence microscope. Antibodies used in this assay are listed in Supplementary Table S2.

Biotin-labeled RNA probes were incubated with streptavidin-conjugated magnetic beads (Invitrogen). The bead–probe complexes were incubated with AGS cell lysates. After five washes, the retained proteins were eluted and visualized by silver staining.

The RIP assay was performed using the PureBinding RNA Immunoprecipitation Kit (Geneseed Biotech, Guangzhou, China), as per the manufacturer’s guidelines. Magnetic beads were pre-incubated with anti-IGF2BP1 antibodies (Abcam) or IgG control antibodies (Proteintech). AGS cells were lysed in RIP lysis buffer, and the lysates were incubated with antibody-conjugated beads. The co-precipitated RNA was isolated using TRIzol reagent (Invitrogen) and analyzed via qRT-PCR.

Cells were seeded into 6-well plates and cultured until reaching full confluency. Uniform vertical wounds were induced using 10µL pipette tips, followed by washing with PBS to remove detached cells. The scratch areas were observed at 0, 24, and 48 h under microscope. Cell migration was quantified using ImageJ software.

Approximately 3 × 10⁴ cells suspended in serum-free medium were seeded into the upper chambers of 24-well transwell inserts (8.0 μm pore size, JET Biofil). The lower chambers were filled with RPMI-1640 medium containing 10% FBS to serve as a chemoattractant. After 24 h of incubation, cells that migrated to the lower membrane surface were fixed with 4% PFA and stained using 1% Crystal Violet (Beyotime). All migrated cells were quantified using ImageJ software.

HGC-27 and AGS cells were treated with 5 μg/mL ActD (MCE, China) for 0, 3, 6, 9, and 12 h. Total RNA was isolated at each time point, and relative mRNA expression levels were normalized to baseline (0 h). The mRNA half-life was calculated using GraphPad Prism software.

A luciferase reporter vector containing either the wild-type (WT) or mutant (Mut) NF1 promoter was constructed by cloning the respective sequences into pGL3-basic vectors (Promega). These vectors were co-transfected with a ZNF460 overexpression construct (Miaoling Biology, Wuhan, China). After 36 h, firefly and Renilla luciferase activities were measured using a Dual-Luciferase Reporter Assay Kit (Promega), as per the manufacturer’s instructions.

Four-week-old female BALB/c nude mice were obtained from Vital River Laboratory (Beijing, China) and randomly divided into experimental groups. Each mouse received a tail vein injection of 1 × 10⁶ HGC-27 cells suspended in 100 μL of PBS. After six weeks, the mice were euthanized by CO₂ asphyxiation at a flow rate of 20% of the chamber volume per minute, followed by cervical dislocation to ensure death. Subsequently, lung tissues were harvested and fixed in 4% PFA. Lung metastases were evaluated, and the tissues were paraffin-embedded and examined with H&E staining.

Molecular docking was performed using the HDOCK SERVER (http://hdock.phys.hust.edu.cn/) to analyze the interaction between RNAs and their target proteins. Docking results were visualized and interpreted using PyMOL software. Protein structures were retrieved from the Protein Data Bank (PDB, https://www.rcsb.org), while RNA structures were predicted using the 3DRNA web tool (http://biophy.hust.edu.cn/new/3dRNA/create).

Statistical analyses were conducted using GraphPad Prism and SPSS software. Student’s t-test, one-way ANOVA, and Chi-square test were applied to compare differences among groups. Univariate and multivariate Cox regression models were used to assess the impact of clinical parameters on GC patient survival. Kaplan–Meier analysis was used for survival curves. P < 0.05 was considered statistically significant.

Our previous findings identified circNF1 as a circular RNA derived from exons 2 to 8 of the NF1 gene, comprising 828 nucleotides (Figure 1A) (11). In the present study, circNF1 amplification was observed exclusively in cDNA using divergent primers, whereas linear NF1 transcripts were amplified in both cDNA and gDNA using convergent primers, confirming its circular nature via back-splicing rather than genomic rearrangement or trans-splicing (Figure 1B). This result was further validated by RT-PCR and Sanger sequencing using divergent primers (Supplementary Figure S1A). Actinomycin D treatment revealed a longer half-life and higher stability of circNF1 compared to its linear counterpart (Supplementary Figure S1B). Nuclear-cytoplasmic fractionation demonstrated predominant cytoplasmic localization of circNF1 in AGS and HGC-27 cells (Figure 1C), a finding consistent with RNA FISH analysis (Figure 1D).

To determine the expression pattern and clinical relevance of circNF1, FISH and ISH assays were performed on GC tissues (Figures 1E, F), indicating significantly elevated circNF1 levels in tumor tissues relative to adjacent normal tissues. An analysis of 80 GC samples revealed a strong correlation between increased circNF1 expression and tumor size (P = 0.007), lymph node metastasis (P = 0.016), advanced clinical stage (P = 0.010), and poor survival (P = 0.010), with no significant association observed with age, gender, or tumor grade (Table 1). Moreover, ISH scores for circNF1 positively correlated with tumor size, lymph node involvement, and clinical stage (Figure 1G). Kaplan–Meier survival analysis demonstrated that patients with high circNF1 expression showed significantly shorter cumulative survival (Figure 1H). Multivariate analysis using the Cox proportional hazards model identified high circNF1 expression as an independent prognostic factor for poor outcome (HR = 2.281, P = 0.031; Table 2). These data collectively suggest that circNF1 was upregulated in GC and correlates with disease progression and unfavorable prognosis.

To elucidate the functional role of circNF1 in GC progression, siRNAs targeting the back-splice junction of circNF1 and a circNF1 overexpression construct were developed (Figure 2A). Quantitative RT-PCR confirmed significant downregulation or overexpression of circNF1 in GC cells after transfection, while linear NF1 transcript levels remained unchanged (Figure 2B). PCR assays yielded consistent results (Figure 2C). Functional assays revealed that circNF1 knockdown suppressed, whereas its overexpression enhanced, cell migration as assessed by transwell and wound healing assays (Figures 2D–G). Moreover, circNF1 silencing increased mRNA expression of the epithelial marker E-cadherin and decreased mesenchymal markers N-cadherin and Vimentin; inverse trends were observed upon circNF1 overexpression (Figure 2H). Western blotting confirmed these changes at the protein level (Figure 2I). These findings indicate that circNF1 enhances the migration of GC cells, likely through the regulation of EMT-associated markers.

To further evaluate the metastatic potential of circNF1 in vivo, a lung metastasis model was developed by tail vein injection of HGC-27 cells with or without circNF1 overexpression into BALB/c nude mice. After six weeks, both the number and size of pulmonary metastatic nodules were markedly increased in the circNF1-overexpressing group compared to controls (Figure 2J). These results demonstrate that circNF1 promotes GC cell metastasis in vivo.

Considering the regulatory potential of circRNAs through RBP interactions, candidate RBPs for circNF1 were identified using Circinteractome (https://circinteractome.irp.nia.nih.gov), CSCD (http://gb.whu.edu.cn/CSCD), and RBPDB (http://rbpdb.ccbr.utoronto.ca) databases. IGF2BP1 was highlighted as a potential binding partner (Figure 3A). RNA pull-down assays showed enrichment of a ~70 kDa protein corresponding to IGF2BP1 (Figure 3B). RIP assays confirmed the direct interaction between circNF1 and IGF2BP1 (Figure 3C). However, qRT-PCR and Western blot analyses indicated no significant changes in IGF2BP1 expression after circNF1 knockdown or overexpression (Figure 3D). Co-localization of circNF1 and IGF2BP1 in the cytoplasm was validated through IF-FISH assays (Figure 3E). To assess the functional relevance of this interaction, siRNAs and an overexpression vector targeting IGF2BP1 were used (Figures 3F, G). The silencing of IGF2BP1 significantly reduced the circNF1-induced enhancement of cell migration (Figure 3H). In line with these observations, qRT-PCR and Western blotting revealed that the expression patterns of E-cadherin, N-cadherin, and Vimentin were also modulated after IGF2BP1 knockdown in circNF1-overexpressing cells (Figures 3I, J). These results establish IGF2BP1 as a key mediator of circNF1-driven motility in GC cells.

IGF2BP1 binds to multiple target mRNAs, including c-MYC, CD44, HMGA2, PTEN, and ACTB (10, 18–20). Among these targets, HMGA2 has been well-documented to be specifically associated with metastatic processes (21–23). Notably, bioinformatic analyses first predicted a high binding propensity between IGF2BP1 and HMGA2 (Supplementary Figure S2). RIP assays confirmed the binding of IGF2BP1 to HMGA2 mRNA (Figure 4A). CircNF1 knockdown resulted in decreased stability of HMGA2 mRNA (Figure 4B). However, circNF1 overexpression significantly elevated both mRNA and protein levels of HMGA2, while its silencing led to substantial downregulation (Figures 4C, D), consistent with results from Actinomycin D-based stability assays. To further confirm that HMGA2 is a downstream target of IGF2BP1, GC cell lines with IGF2BP1 overexpression or knockdown were generated. A positive correlation was observed between IGF2BP1 and HMGA2 expression; specifically, IGF2BP1 silencing reduced HMGA2 levels, whereas its overexpression enhanced them at both RNA and protein levels (Figures 4E, F).

Based on these findings, it was hypothesized that circNF1 modulates HMGA2 expression via its interaction with IGF2BP1. To investigate the interaction between IGF2BP1 and circNF1 as well as HMGA2 mRNA, FLAG-tagged truncation mutants of IGF2BP1 were constructed (Figure 5A). RIP assays showed that circNF1 and HMGA2 mRNA bind to the KH3–4 di-domain of IGF2BP1 (Figures 5B, C). Moreover, IGF2BP1 knockdown in circNF1-overexpressing cells decreased HMGA2 expression at both transcript and protein levels (Figures 5D, E). Molecular docking analysis further revealed that circNF1 enhances the interaction between IGF2BP1 and HMGA2 mRNA by increasing binding affinity (Figure 5F).

Rescue assays were conducted to evaluate whether the metastatic effects of circNF1 are mediated through the HMGA2 axis. In vitro transwell migration assays revealed that HMGA2 knockdown significantly attenuated the circNF1 overexpression-enhanced migration (Figure 6A). Conversely, HMGA2 overexpression rescued the suppressed migratory phenotype caused by circNF1 knockdown (Figure 6B). Moreover, HMGA2 silencing reversed the circNF1-induced EMT-like changes, as evidenced by altered expression of E-cadherin, Vimentin, and N-cadherin at both mRNA and protein levels (Figures 6C–E). In vivo analysis using a lung metastasis model further supported these findings. HMGA2 overexpression effectively reversed the reduction in metastatic nodules observed after circNF1 knockdown (Figure 6F). These data demonstrate that circNF1 exerts its pro-metastatic effects in GC via the HMGA2 signaling pathway.

To identify transcriptional regulators of NF1 and circNF1, a bioinformatic analysis was performed using the JASPAR database (https://jaspar.genereg.net/), which predicted several transcription factors potentially binding to the NF1 promoter. Among these, ZNF460 showed the highest prediction score, suggesting a prominent role in NF1 regulation. Correlation analysis using the GEPIA database (http://gepia.cancer-pku.cn) revealed a positive association between ZNF460 and NF1 expression levels in both GC and adjacent non-tumorous tissues (Figure 7A), supporting the initial prediction.

To further elucidate the role of ZNF460, a ZNF460 overexpression plasmid was constructed and its expression was validated via qRT-PCR (Figure 7B). Overexpression of ZNF460 in GC cells significantly upregulated NF1 mRNA expression (Figure 7C). Based on JASPAR predictions (https://jaspar.genereg.net/), six potential ZNF460 binding motifs were identified within the NF1 promoter region, from which three high-probability sites were selected for further functional analysis. Respective WT and Mut NF1 promoter constructs were generated and cloned into luciferase reporter vectors (Figure 7D). Dual-luciferase assays depicted that ZNF460 significantly increased the transcription of the WT NF1 promoter, with its binding not restricted to a single site (Figure 7E). To investigate whether ZNF460-mediated activation of the NF1 promoter also affected circNF1 expression, ZNF460 was overexpressed in GC cells. Next, qRT-PCR analysis indicated a marked elevation of circNF1 levels in ZNF460-overexpressing cells relative to vector controls (Figure 7F). These findings suggest that ZNF460 transcriptionally activates the NF1 promoter, enhancing circNF1 expression.