Destabilization of hsa_circ_0015508 by YTHDF2 Enhances miR-496-Mediated FOXN3 Suppression to Drive Nasopharyngeal Carcinoma Progression

1 Introduction

Nasopharyngeal carcinoma (NPC) is a distinct type of head and neck cancer that exhibits a unique geographic distribution, with high incidence rates in North Africa, Southeast Asia and Southern China [1,2]. Despite advances in radiotherapy and chemotherapy, local relapses and distant metastasis remain challenges in NPC patients among treatment failures [3,4]. Therefore, elucidating the molecular mechanisms underlying NPC recurrence and metastasis is urgently needed.

Circular RNAs (circRNAs) are a class of covalently closed non-coding RNAs characterized by their lack of 5′ caps and 3′ poly(A) tails, which confer resistance to exonuclease-mediated degradation and result in high stability [5]. Emerging evidence has demonstrated that circRNAs play critical roles in various physiological and pathological processes, including cancer progression [6,7,8]. Functionally, circRNAs can act as microRNA (miRNA) sponges, protein scaffolds, or transcriptional regulators, thereby modulating gene expression networks [9,10]. In NPC, accumulating studies have revealed that dysregulated circRNAs contribute to tumor proliferation, metastasis, chemotherapy and radiotherapy resistance [11,12]. However, the expression patterns, biological functions, and regulatory mechanisms of circRNAs in NPC remain largely unexplored.

N6-methyladenosine (m6A) is the most prevalent internal RNA modification in eukaryotes, regulating RNA metabolism including splicing, export, stability, and translation [13,14]. The m6A modification is dynamically regulated by methyltransferases complex (“writers”) and demethylases (“erasers”), and is recognized by binding proteins (“readers”) [15]. Among the m6A readers, YTH N6-methyladenosine RNA binding protein F2 (YTHDF2) has been extensively characterized as a key mediator of RNA degradation, preferentially binding to m6A-modified transcripts and typically facilitating their decay [16]. Recent studies have shown that YTHDF2 is frequently upregulated in various cancers, where it promotes tumor progression mainly by degrading tumor suppressors [17,18,19]. An increasing number of studies have revealed that circRNAs can undergo m6A modification, and such modification influences circRNA stability, localization, and function [20,21,22]. Recent research suggests that YTHDF2 regulates circRNA expression in an m6A-dependent manner, a finding that has garnered attention. In colorectal cancer, circ_0003215 inhibited colorectal cancer (CRC) malignancy by acting as a miR-663b sponge to regulate DLG4 expression, and further suppressed the pentose phosphate pathway (PPP) through DLG4-mediated K48-linked ubiquitination of glucose-6-phosphate dehydrogenase (G6PD) [23]. However, whether YTHDF2 can exert a role by regulating circRNAs in NPC remains largely unclear.

The objective of this study was to elucidate the functional role and molecular mechanism of YTHDF2-mediated circRNA regulation in NPC migration and invasion. We examined YTHDF2 expression in NPC tissues and its functional role in NPC cells. We identified a circRNA, hsa_circ_0015508 (circ_0015508), derived from the ACBD6 gene, predominantly localized in the cytoplasm and exhibited tumor-suppressive functions in NPC. We demonstrated that YTHDF2 functioned as an oncogene by promoting migration, invasion, and epithelial-mesenchymal transition (EMT). Mechanistically, YTHDF2 bound to the m6A modification site on circ_0015508 and promoted its degradation, thereby relieving the inhibitory effect of circ_0015508 on miR-496.

2 Materials and Methods

2.1 Cell Culture and Reagents

The NPC cell lines 5-8F and HNE1 were derived from the Cancer Research Institute of Central South University. Human embryonic kidney 293T cells (HEK-293T, Cat. #GNHu17) were purchased from Cell Bank of Chinese Academy of Sciences (Shanghai, China), and were authenticated by the supplier via short tandem repeat (STR) profiling. All cells were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) (Gibco, C11995500BT, Shanghai, China) supplemented with 10% fetal bovine serum (FBS) (Lonsera, S711-050S, Suzhou, China), penicillin (10,000 U/mL) and streptomycin (10 mg/mL) mixed solution (Solarbio, P1400, Beijing, China), diluted to final concentration of 100 U/mL penicillin and 100 μg/mL streptomycin at 37°C in a humidified atmosphere containing 5% CO2. All experiments were performed with mycoplasma-free cells.

2.2 Bioinformatics Analysis

The expression profiles of m6A-associated genes in NPC were analyzed using the GSE12452 dataset (https://www.ncbi.nlm.nih.gov/gds/). This dataset, generated using Platform GPL570, included 41 samples (31 NPC and 10 normal nasopharyngeal tissue specimens) with mRNA expression levels measured. For circRNA characterization, SRAMP (http://www.cuilab.cn/sramp) was used to predict m6A modification sites on circ_0015508. In-house miRNA target prediction tools based on the RNAhybrid or miRanda algorithms, and circInteractome database (https://circinteractome.nia.nih.gov/index.html) were used to predict miRNAs binding to circ_0015508. MiRWalk (http://mirwalk.umm.uni-heidelberg.de), microT-CDS algorithms (http://diana.imis.athena-innovation.gr/DianaTools/index.php?r=MicroT_CDS/index), and miRDB databases (http://www.mirdb.org) were used to predict downstream targets of miR-496. Correlation analysis between YTHDF2 and FOXN3 was performed using the GSE12452 dataset. Analyzed using Pearson correlation coefficient. Statistical analyses were performed using GraphPad Prism 10.0 (GraphPad Software, San Diego, CA, USA). A two-tailed p-value < 0.05 was considered statistically significant.

2.3 Plasmid Construction and Transfection

The small interfering RNAs targeting YTHDF2 (si-YTHDF2) were purchased from RiboBio (Guangzhou, China). The small interfering RNAs targeting FOXN3 (si-FOXN3) were purchased from Sangon Biotech (Shanghai, China). The circ_0015508 overexpression plasmid (pcDNA3.1-Circ-0015508-mini) and empty vector (pcDNA3.1-CircRNAmini vector), YTHDF2 overexpression plasmid (pcDNA3.1-YTHDF2) and OE-Ctrl vector (pcDNA3.1) were obtained from Hunan Fenghui Biotechnology Co., Ltd., (Changsha, China). miR-496 mimics, NC mimics, miR-496 inhibitor and NC inhibitor were synthesized by GenePharma (Suzhou, China). Plasmid DNA or siRNA constructs were introduced into cells using Lipofectamine 3000 reagent (Thermo Fisher Scientific, L3000008, USA) according to the manufacturer’s optimized protocol. The sequences are as followes: siYTHDF2-1, sense, 5′-GACCAAGAAUGGCAUUGCATT-3′, and antisense, 5′-UGCAAUGCCAUUCUUGGUCTT-3′, siYTHDF2-2, sense, 5′-GCACAGAAGUUGCAAGCAATT-3′, and antisense, 5′-UUGCUUGCAACUUCUGUGCTT-3′, siYTHDF2-3, sense, 5′-GGUAGCGGGUCCAUUACUATT-3′, and antisense, 5′-UAGUAAUGGACCCGCUACCTT-3′, siFOXN3, sense, 5′-GUACCUUCUUCAAGAGAAA-3′, and antisense, 5′-UUUCUCUUGAAGAAGGUAC-3′, siNC, sense, 5′-UUCUCCGAACGUGUCACGUTT-3′, and antisense 5′-ACGUGACACGUUCGGAGAATT-3′, miR-496 mimics, sense, 5′-UGAGUAUUACAUGGCCAAUCUC-3′, and antisense, 5′-GAUUGGCCAUGUAAUACUCAUU-3′, miR-496 inhibitor, 5′-GAGAUUGGCCAUGUAAUACUCA-3′, siMETTL3, sense, 5′-CCAGUCAUAAACCAGAUGAAATT-3′, and antisense, 5′-UUUCAUCUGGUUUAUGACUGGTT-3′.

2.4 Quantitative Reverse Transcription PCR (qRT-PCR)

Total RNA was extracted using TRIzol reagent (Takara Bio Inc., 9108, Shiga, Japan). For subcellular localization, cytoplasmic and nuclear RNA was fractionated using the Cytoplasmic & Nuclear RNA Purification Kit (Thermo Fisher Scientific, AM1921). Reverse transcription was performed using the RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, K16225) for mRNA and circRNA, and the miRNA 1st Strand cDNA Synthesis Kit (Vazyme, MR201-01, Nanjing, China) for miRNA. qRT-PCR was conducted on an Applied Biosystems QuantStudio system using SYBR Green Master Mix (Mona Biotech, MQ10301, Suzhou, China). The qPCR program was set as follows: predenaturation at 95°C for 30 s and 40 cycles of denaturation at 95°C for 10 s and annealing at 60°C for 10 s, followed by reaction at 95°C for 10 s, at 65°C for 60 s, and at 95°C for 1 s. Gene expression was normalized to GAPDH or U6 snRNA using the 2−ΔΔCt method. The primer sequences are as follows: YTHDF2, forward, 5′-AGTAGGGCAACAGACACAGC-3′, and reverse, 5′-TGGACCGAAGCTTCTCCAAC-3′; U6, 5′-CTCGCTTCGGCAGCACA-3′, and reverse, 5′-AACGCTTCACGAATTTGCGT-3′; GAPDH, forward, 5′-GGAGCGAGATCCCTCCAAAAT-3′, and reverse, 5′-GGCTGTTGTCATACTTCTCATGG-3′; circ_0015508, forward, 5′-GGCCTGTGATCGAGGACATA-3′, and reverse, 5′-ACAATTTCCAACTTTGACCTGAC-3′; ACBD6, forward, 5′-TTGGTGGGCCAGTTATTAGTTC-3′, and reverse, 5′-CCAGTGAAGTAGAGCCCTACC-3′; METTL3, forward, 5′-GTGATCGTAGCTGAGGTTCGT-3′, and reverse, 5′-GGGTTGCACATTGTGTGGTC-3′; miR-496, forward, 5′-TGAGTATTACATGGCCAATCTC-3′, and reverse (provided with the kit, Vazyme, MR201-01), FOXN3, forward, 5′-TCGTTGTGGTGCATAGACCC-3′, and reverse, 5′-GTGGACCTGATGTGCTTTGATA-3′; NPAS3, forward, 5′-CCCAGCCAAATCATGGGTCTC-3′, and reverse, 5′-GCATGGATGAAGTGGTAGCAT-3′; FLRT2, forward, 5′-CTCCCGATCTCCCAGGTACG-3′, and reverse, 5′-CGTTCCAGCTTACGCAGATTTG-3′.

2.5 Wound Healing Assay

Wound healing assays were performed to evaluate cell migration ability. HNE1 and 5-8F cells (3 × 105) were seeded in 6-well plates and grown to 90% confluence. A straight scratch was generated using a sterile 200 μL pipette tip. Cells were washed with PBS to remove debris and incubated in serum-free DMEM (Gibco, C11995500BT). Images of the wound area were captured at 0 and 24 h post-scratch via an inverted microscope (Thermo Fisher Scientific, EVOS XL Core), and the wound closure rates were determined using ImageJ software (version 1.46, National Institutes of Health, USA). The change in wound area was quantified as (Area0h − Area24h)/Area0h. The resulting ratios were then standardized against the control group, which was assigned a value of 1.

2.6 Transwell Migration and Invasion Assays

Transwell assays were conducted to evaluate both cell migration and invasion abilities. In transwell assays, for migration assays, 1 × 104 cells in serum-free DMEM were added to each uncoated Transwell insert (Corning, 3422, Tewksbury, MA, USA). For invasion assays, 1 × 104 cells were added to each Transwell insert pre-coated with Matrigel (Corning, 354248), cells that migrated/invaded to the lower chamber contained DMEM with 10% FBS as a chemoattractant. After 24 h (migration)/36 h (invasion), cells remaining in the upper chamber were gently removed with a cotton swab, and the inserts were fixed with 4% paraformaldehyde for 15 min at room temperature. Migrating/invading cells on the lower surface were stained with 0.1% crystal violet (Solarbio, G1063) for 10 min, washed with PBS, and visualized.

2.7 RNase R Digestion Assay

Total RNA (2 μg) was incubated with 3 U/μg of RNase R (GLPBIO, GE10003, Montclair, USA) and reaction buffer at 37°C for 20 min. A parallel control reaction was performed under identical conditions in the absence of RNase R. Reverse transcription was performed using the RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, K16225). Subsequent qPCR was performed as described in Section 2.4 to detect circ_0015508 and ACBD6 expression. Data were normalized to the mock group, which was set to 1.

2.8 Stable Cell Line Construction and Lentiviral Transduction

For stable overexpression of circ_0015508, lentiviral particles carrying the circ_0015508 sequence were constructed using the GV689 vector (GeneChem, Shanghai, China). The negative control lentivirus was generated using the empty GV689 vector (carrying ZsGreen1 without the circ_0015508 insert) (GeneChem). 3 × 104 HNE1 and 5-8F cells were seeded at 20% confluence and infected with the lentivirus at a multiplicity of infection (MOI) of 10. After 48 h, the culture medium was replaced with fresh medium containing puromycin (2 μg/mL) (Solarbio, P8230) to select stably infected cells. The efficiency of stable transfection was confirmed by GFP fluorescence observation and qPCR analysis.

2.9 Fluorescence In Situ Hybridization (FISH)

A Cy3-labeled DNA probe specifically targeting the back-splice junction of circ_0015508 was designed and synthesized by GENESEED (Guangzhou, China). Cells were seeded onto confocal dishes at a density of 1 × 104 cells/well and cultured overnight. After washing twice with PBS, cells were fixed with 4% paraformaldehyde at room temperature for 15 min, permeabilized with 0.1% Buffer A (freshly prepared) at room temperature for 15 min, and dehydrated through a graded ethanol series (70%, 85%, and 100% ethanol, 3 min each). After air drying, cells were hybridized overnight with the CY3-labeled probe at 37°C according to the instructions of the RNA FISH kit (GenePharma, F40131). Nuclei were counterstained with DAPI working solution (1:1000 dilution in PBS) in the same kit for 20 min in the dark). Images were captured using an inverted fluorescence microscope (Leica, TCS SP8, Wetzlar, Germany). The FISH probe sequence of circ_0015508 was: 5′-ctttgacctgacagttaatgt-3′.

2.10 Dual-Luciferase Reporter Assay

The wild-type (Wt) or mutant (Mut) sequences of circ_0015508 or the FOXN3 3′UTR, containing the predicted miR-496 binding sites, were cloned into the pmirGLO vector HANBIO (GC20250911PC01, Shanghai, China). 293T cells were co-transfected with these reporter plasmids and miR-496 mimics or negative control (NC) were seeded at a density of 2 × 105 cells/well in 6-well plates and cultured overnight. Cells were co-transfected with 1 μg of the respective reporter plasmids (wild-type or mutant circ_0015508/wild-type or mutant FOXN3 3′-UTR) and miR-496 mimics (50 nM final concentration) or NC mimics (50 nM final concentration) using Lipofectamine™ 3000 Transfection Reagent according to the manufacturer’s protocol. After 48 h, System (Vazyme, DL101-01). Firefly luciferase activity was normalized to Renilla activity.

2.11 RNA Immunoprecipitation (RIP)

Briefly, 2 × 107 cell were harvested in RIP lysis buffer including RNase inhibitor (Medchem express, HY-K1033, Monmouth Junction, New Jersey, USA), protease inhibitor and phosphatase inhibitor (Medchem express, HY-K0010) and incubated with magnetic beads conjugated with anti-YTHDF2 antibody (5 μg, Proteintech, 24744-1-AP, Wuhan, China) or normal IgG (5 μg, Proteintech, 10284-1-AP, Wuhan, China) overnight at 4°C. After extensive washing, the co-precipitated RNA was extracted and analyzed by qRT-PCR.

2.12 Methylated RNA Immunoprecipitation qPCR (MeRIP-qPCR)

MeRIP-qPCR was performed to detect m6A modification on circ_0015508. The procedure was adapted from the published reports [24,25]. Total RNA was extracted from cells and was incubated with anti-m6A antibody (5 μg, Synaptic Systems, 202003, Goettingen, Germany) or normal IgG (5 μg, Proteintech, 10284-1-AP) in immunoprecipitation buffer containing RNaseOUT at 4°C for 2 h. Protein A/G magnetic beads (Medchem express, HY-K0202, Monmouth Junction, New Jersey, USA) were added and incubated with rotation at 4°C for 2 h. The beads were washed three times and the captured RNAs were eluted via competitive displacement with IP buffer containing m6A sodium salt (Santa Cruz Biotechnology, sc-215524A, Dallas, Texas, USA). The eluted RNA was precipitated with ethanol and subjected to qPCR analysis.

2.13 Western Blotting

5-8F and HNE1 cells were lysed in RIPA buffer (Solarbio, R0100) supplemented with protease and phosphatase inhibitors (Solarbio, IP0280) at a ratio of 80 μL lysis buffer per well. After complete lysis, cell lysates were centrifuged at 12,000× g for 15 min at 4°C, and the supernatant was collected for protein extraction. Protein concentration was determined using the BCA Protein Assay Kit (GLPBIO, GK10009) according to the manufacturer’s protocol. Protein samples were diluted fivefold with lysis buffer and quantified, then adjusted to a final concentration of 5 μg/μL using lysis buffer. 5 × loading buffer (Solarbio, P1040) was added, and samples were boiled at 100°C for 10 min. For each sample, 30 μg of proteins were separated by 10% SDS-PAGE, transferred to PVDF membranes, and incubated with primary antibodies against YTHDF2 (Proteintech, 24744-1-AP, 1:1000), E-cadherin (Proteintech, 60902-1-Ig, 1:1000), N-cadherin (Proteintech, 66219-1-Ig, 1:1000), Vimentin (Proteintech, 60330-1-Ig, 1:1000), FOXN3 (ABclonal, A15039, Wuhan, China, 1:1000), or GAPDH (Proteintech, 10494-1-AP, 1:10,000) overnight at 4°C, followed by incubation with HRP-conjugated Goat Anti-Rabbit IgG(H+L) (Proteintech, SA00001-2, 1:10,000) or HRP-conjugated Goat Anti-Mouse IgG(H+L) (Proteintech, SA00001-1, 1:10,000). GAPDH was used as a loading control. Signals were detected using an ECL kit (Biosharp, BL520B, Beijing, China) and visualized with the Tanon Chemiluminescence Imaging System (Tanon, 5200, Shanghai, China).

2.14 SELECT Assay

The SELECT method (single-base elongation- and ligation-based qPCR amplification) was performed to identify specific m6A modification sites [26]. Briefly, primers were designed for the predicted sites (387 and 396) and a non-m6A site (391 site, input control). RNase-treated RNA was mixed with Up Primer and Down Primer and dNTP in CutSmart buffer (NEB, B7204S, Ipswich, Massachusetts, USA). The mixture was annealed. Subsequently, Bst 2.0 DNA polymerase (NEB, M0537S), SplintRTM ligase (NEB, M0375S) and ATP were added. The reaction mixture was incubated at 40°C for 20 min and denatured at 80°C for 20 min. qPCR was then performed using the SELECT qPCR primers, and the threshold cycle (Ct) values obtained from this assay showed the SELECT results for detecting 387 site, 396 site, and 391 site (a non-m6A site) within circ_0015508 in control and METTL3-knockdown HNE1 and 5-8F cells. The primer sequences are as follows: SELECT qPCR forward, 5′-ATGCAGCGACTCAGCCTCTG-3′; SELECT qPCR reverse, 5′ TAGCCAGTACCGTAGTGCGTG-3′; Circ_0015508-387-site Up, 5′-TAGCCAGTACCGTAGTGCGTGACACTGTGACTAGTTCCTTATG-3′; Circ_0015508-387-site Down, 5′-CCTCGATCACAGGCCCAGTGAACAGAGGCTGAGTCGCTGCAT-3′; Circ_0015508-396-site Up, 5′-TAGCCAGTACCGTAGTGCGTGGTTGCAGCAACACTGTGACTAG-3′; Circ_0015508-396-site Down, 5′-TCCTTATGTCCTCGATCACAGGCAGAGGCTGAGTCGCTGCAT-3′; Circ_0015508-391-site Up, 5′-TAGCCAGTACCGTAGTGCGTGACACTGTGACTAGTTCCT-3′; Circ_0015508-391-site Down, 5′-ATGTCCTCGATCACAGCAGAGGCTGAGTCGCTGCAT-3′.

2.15 Actinomycin D Stability Assay

Cells were treated with actinomycin D (GLPBIO, GC16866) (5 μg/mL) to block transcription. Total RNA was harvested at 0, 4, 8, 12 and 24 h after treatment, and circ_0015508 expression was measured by qPCR. The RNA decay curves were fitted using a one-phase exponential decay model, and the half-life (t1/2) was calculated using GraphPad Prism 10.0 software.

2.16 Immunohistochemistry (IHC)

Tissue samples (47 normal nasopharyngeal epithelial tissues and 51 NPC tissues) were fixed in 10% formalin, embedded in paraffin, and sectioned at 4 μm thickness. Sections were deparaffinized, rehydrated, and subjected to antigen retrieval in citrate buffer (pH 6.0) in pressure cooker for 3 min. Endogenous peroxidase activity was blocked with 3% H2O2 for 10 min at room temperature, and non-specific binding sites were blocked by goat serum (Sangon Biotech, E510009, Shanghai, China) for 10 min at room temperature, and sections were incubated with anti-YTHDF2 antibody (Proteintech, 24744-1-AP, Wuhan, China, dilution ratio: 1:200) overnight at 4°C. After washing, sections were incubated with MaxVision-HRP (Maxim Biotechnologies, KIT-5030, Fuzhou, China) for 15 min at room temperature, and signals were visualized using diaminobenzidine (DAB) (Maxim Biotechnologies, DAB-1031). Staining intensity and percentage of positive cells were evaluated to calculate immunoreactive score (IRS). The scoring system evaluates immunohistochemical staining as follows: in each of five randomly selected high-power fields, the proportion of positively stained cells and the staining intensity are assessed to calculate an immunoreactive score (IRS = P × I). The percentage of positive cells is graded on a scale of 0 to 4. 0 (no positive cells), 1 (1%–24%), 2 (25%–49%), 3 (50%–74%), or 4 (75%–100%). Staining intensity is scored from 0 to 3. 0 (no staining), 1 (light), 2 (moderate), or 3 (dark). The scoring was performed by two independent pathologists in a blinded manner. Ethical approval for this study was granted by the Ethics Committee of The First Affiliated Hospital of Guilin Medical University (No. 2024GZRLL-04). Informed consent was taken from all the patients. The study was conducted in accordance with the Declaration of Helsinki.

2.17 circRNA Sequencing

Total RNA was extracted from si-NC and si-YTHDF2 HNE1 cells using Magzol Reagent (R4801-03, Magen, Guangzhou, China) following the manufacturer’s instructions with two independent biological replicates per group. The quantity and integrity of RNA yield was assessed by using the K5500 (Beijing Kaiao, Beijing, China) and the Agilent 2200 TapeStation (Agilent Technologies, Santa Clara, CA, USA) separately. Ribosomal RNAs were removed using the RiboCop rRNA Depletion Trial Kit (K14496, LEXOGEN, Vienna, Austria), and then RNA was treated with RNase R (RNR07250, Epicentre, Madison, WI, USA) and fragmented into approximately 200 bp fragments. The purified RNA fragments were then used for first- and second-strand cDNA synthesis, followed by adapter ligation and enrichment with a low-cycle according to instructions of NEBNext® Ultra™ RNA Library Prep Kit for Illumina (E7530L, NEB, Ipswich, MA, USA). Library quality was assessed using the Agilent 2200 TapeStation and Qubit (Thermo Fisher Scientific), and sequencing was performed on an Illumina NovaSeq 6000 platform (Illumina, USA) with paired-end 150 bp reads at Ribobio Co., Ltd. (Ribobio). To identify circular RNAs (circRNAs), two algorithms, CIRI2 and CIRCEXPLORER2, were applied. Reads were mapped to the human reference genome GRCh37/hg19 (http://genome.ucsc.edu/) by BWA-MEM or Tophat, respectively. Only circRNAs detected by both methods were considered as reliably identified. For exploratory screening of candidate circRNAs, differentially expressed circRNAs were identified according to the criteria of |log2 (fold change)| > 1 and p < 0.05. The datasets presented in this study have been deposited in the Gene Expression Omnibus (GEO) repository under accession number GSE335362.

2.18 Single-Cell Communication Analysis

Single-cell RNA-sequencing datasets GSE150430 were obtained from the TISCH2 database [27,28]. Based on the single-cell stochastic gene silencing (scSGS) framework [29], malignant cells were dichotomized into YTHDF2+ (counts > 0) and YTHDF2− (counts = 0) subpopulations according to their endogenous YTHDF2 expression. Cell–cell communication networks were constructed using the CellChat R package with the Secreted Signaling sub-database of CellChatDB.human. Communication probabilities were calculated (min.cells = 10), followed by pathway-level aggregation and centrality analyses to compare the outgoing and incoming signaling strengths between the two YTHDF2-defined malignant subpopulations.

2.19 Statistical Analysis

The experiments were performed by three independent biological replicates unless otherwise specified. Data were normalized to the control group and are presented as means ± standard deviation (SD). GraphPad Prism 10.0 (GraphPad, Inc., San Diego, CA, USA) was used for all statistical analyses. Unpaired t-test or Mann-Whitney U tests was applied to compare two groups as appropriate, while comparisons involving multiple groups were performed using one-way ANOVA with Tukey’s post hoc test (for all pairwise comparisons) or Dunnett’s test (for comparisons with a control group). Correlation analysis was performed using Pearson’s correlation coefficient. Association between YTHDF2 expression and clinicopathological parameters were evaluated using Fisher’s exact tests. Statistical significance was defined as p < 0.05.

3 Results

3.1 The RNA and Protein Expression of YTHDF2 Is Upregulated in NPC

To identify key m6A regulators involved in NPC pathogenesis, we first analyzed the mRNA expression profiles of 16 major m6A-associated regulators in the GSE12452 dataset, including methyltransferase Like 3 (METTL3), METTL14, METTL16, WT1-associated protein (WTAP), KIAA1429, RNA-binding motif protein 15 (RBM15), AlkB homolog 5 (ALKBH5), fat mass and obesity associated (FTO), YTH N6-methyladenosine RNA binding protein F1 (YTHDF1), YTHDF2, YTHDF3, YTHDC1, YTHDC2, insulin-like growth factor 2 mRNA-binding protein 1 (IGF2BP1), IGF2BP2, and IGF2BP3. The heatmap showed that among all m6A regulators, the expression levels of YTHDF2, YTHDF3, and RBM15 are significantly increased in NPC tissues compared to normal controls (Fig. 1A). We focused on YTHDF2 for subsequent experiments based on its highest expression level among the candidates. To further assess the protein expression level of YTHDF2, we collected 47 normal nasopharyngeal epithelial tissues and 51 NPC tissues and performed immunohistochemistry (IHC) detection. Representative IHC images demonstrated stronger YTHDF2 immunoreactivity in NPC tissues, predominantly localized in the cytoplasm (Fig. 1B). Quantitative analysis of IHC immunoreactive score (IRS) revealed that YTHDF2 protein expression was significantly higher in NPC tissues than in nasopharyngeal epithelial tissues (Fig. 1C). These results indicate that YTHDF2 is consistently upregulated at both the mRNA and protein levels in NPC. To further explore the association between YTHDF2 expression and clinicopathological features, we analyzed data from 51 NPC patients, among whom 43 had complete clinicopathological records. Patients were divided into high- and low-expression groups based on the median expression level of YTHDF2. As shown in Supplementary Table S1, a numerically higher incidence of distant metastasis was observed in the high-expression group (5/22) than in the low-expression group (1/21); however, this difference did not achieve statistical significance (p = 0.185). To investigate whether YTHDF2-transcriptionally active malignant cells are exposed to a distinct microenvironmental signaling landscape in vivo, we performed CellChat analysis on the NPC single-cell dataset GSE150430. Major cell populations were annotated by TISCH2 (Fig. S1A). YTHDF2 exhibited high expression levels in malignant cells across multiple cell subpopulations (Fig. S1B). Based on the scSGS framework, malignant cells were dichotomized into YTHDF2-positive (YTHDF2+) and YTHDF2-negative (YTHDF2−) subpopulations. EGF signaling pathway network analysis revealed that YTHDF2+ malignant cells were preferentially targeted by EGF signals derived from Mono/Macro and DCs cells (Fig. S1C). WNT signaling pathway network analysis demonstrated that YTHDF2+ malignant cells were preferentially targeted by WNT signals originating from both YTHDF2+ malignant cells and Mono/Macro cells (Fig. S1D). Considering the importance of EGF and WNT signaling in tumor malignant progression, these results suggest that YTHDF2+ malignant cells not only receive EGF and WNT signals from immune cells such as Mono/Macro, but also secrete WNT signals and respond to them, which may establish a tumor-promoting niche that drives malignant progression.

Figure 1: The RNA and protein expression of YTH N6-methyladenosine RNA binding protein F2 (YTHDF2) is upregulated in nasopharyngeal carcinoma (NPC). (A) The expression profiles of 16 major m6A-associated genes in the GSE12452 dataset were analyzed by heatmap, including Methyltransferase-like 3 (METTL3), METTL14, METTL16, WT1-associated protein (WTAP), KIAA1429, RNA-binding motif protein 15 (RBM15), AlkB homolog 5 (ALKBH5), fat mass and obesity associated (FTO), YTH N6-methyladenosine RNA binding protein F1 (YTHDF1), YTHDF2, YTHDF3, YTH N6-methyladenosine RNA binding protein C1 (YTHDC1), YTHDC2, insulin-like growth factor 2 mRNA-binding protein 1 (IGF2BP1), IGF2BP2, and IGF2BP3. (B) Representative immunohistochemistry (IHC) images of YTHDF2 protein expression in 47 normal nasopharyngeal epithelial tissues and 51 NPC tissues. (C) Quantification of IHC immunoreactive score (IRS) for YTHDF2 expression. Data are shown as means ± SD. *p < 0.05; **p < 0.01; ***p < 0.001 compared with the control group.

3.2 YTHDF2 Promotes Migration, Invasion and EMT in NPC Cells

To investigate YTHDF2 biological function in NPC cells, we knocked down YTHDF2 using three independent siRNAs (siYTHDF2-1, siYTHDF2-2, and siYTHDF2-3) in HNE1 and 5-8F NPC cell lines. Western blot analysis showed that all three siRNAs significantly reduced YTHDF2 protein levels compared with the negative control siRNA (siNC). Among them, siYTHDF2-1 and siYTHDF2-2 showed knockdown and were selected for subsequent experiments (Fig. 2A). Conversely, we overexpressed YTHDF2 by transfecting a YTHDF2 overexpression plasmid (pcDNA3.1-YTHDF2) into HNE1 and 5-8F cells. Western blot confirmed a marked increase in YTHDF2 protein in the overexpression group compared with the empty vector control (OE-Ctrl) (Fig. 2B). We next performed wound healing assays to evaluate the effect of YTHDF2 on NPC cell migration. Knockdown of YTHDF2 using siYTHDF2-1 or siYTHDF2-2 significantly reduced the wound closure rate in both HNE1 and 5-8F cells compared with the siNC group (Fig. 2C,D). Transwell migration assays (without matrigel) demonstrated that the number of migrated cells was markedly reduced in siYTHDF2-1 and siYTHDF2-2 treated cells compared with control cells (Fig. 2E,F). Moreover, transwell invasion assays (with matrigel) showed that YTHDF2 knockdown also significantly decreased the number of invasive cells in both cell lines (Fig. 2G,H). These results indicated that knockdown of YTHDF2 suppressed the migration and invasion of NPC cells. In contrast, wound healing assays revealed that YTHDF2-overexpressing HNE1 and 5-8F cells exhibited accelerated wound closure compared with empty vector-transfected cells (Fig. 2I,J). Transwell migration assays showed that the number of migrated cells was significantly higher in the YTHDF2 overexpression group than in the control group (Fig. 2K,L). Transwell invasion assays demonstrated that YTHDF2 overexpression markedly increased the number of invasive cells (Fig. 2M,N). To explore the molecular mechanism underlying YTHDF2-mediated cell motility, we examined the expression of epithelial-mesenchymal transition (EMT) markers by Western blot. Knockdown of YTHDF2 led to decreased protein levels of N-cadherin and vimentin (mesenchymal markers), while the epithelial marker E-cadherin was increased in HNE1 cells (Fig. 2O). In contrast, YTHDF2 overexpression resulted in upregulation of N-cadherin, vimentin and downregulation of E-cadherin (Fig. 2P). Taken together, these results demonstrate that YTHDF2 enhance the migratory and invasive capacities of NPC cells, which may be attributed to its role in promoting EMT.

Figure 2: YTHDF2 promotes migration, invasion and epithelial-mesenchymal transition (EMT) in NPC cells. (A) Western blot analysis of YTHDF2 protein levels in HNE1 (left) and 5-8F (right) cells transfected with siRNA negative control (siNC) or three independent siYTHDF2 oligos (siYTHDF2-1, siYTHDF2-2 and siYTHDF2-3). (B) Western blot analysis of YTHDF2 expression in HNE1 (left panel) and 5-8F (right panel) cells transfected with empty vector (OE-Ctrl) or YTHDF2-overexpressing plasmid. (C) Representative images of wound healing assays in HNE1 (left panel) and 5-8F (right panel) cells after YTHDF2 knockdown. (D) Relative wound closure rate is presented in HNE1 (left panel) and 5-8F (right panel) cells. (E) Representative images of Transwell migration assays in HNE1 and 5-8F cells after YTHDF2 knockdown. (F) Relative cell migration is presented in HNE1 (left panel) and 5-8F (right panel) cells. (G) Representative images of Transwell invasion assays in HNE1 and 5-8F cells after YTHDF2 knockdown. (H) Relative cell invasion is presented in HNE1 (left panel) and 5-8F (right panel) cells. (I) Representative images of wound healing assays in HNE1 and 5-8F cells after YTHDF2-overexpression (J) Relative wound closure rate is presented in HNE1 (left panel) and 5-8F (right panel) cells. (K) Representative images of Transwell migration assays in HNE1 and 5-8F cells after YTHDF2 overexpression. (L) Relative cell migration is presented in HNE1 (left panel) and 5-8F (right panel) cells. (M) Representative images of Transwell invasion assays in HNE1 and 5-8F cells after YTHDF2 overexpression. (N) Relative cell invasion is presented in HNE1 (left panel) and 5-8F (right panel) cells. (O) Western blot analysis of E-cadherin, N-cadherin, and Vimentin in HNE1 cells after YTHDF2 knockdown. (P) Western blot analysis of E-cadherin, N-cadherin, and Vimentin in YTHDF2 overexpression in HNE1 cells. Data are shown as means ± SD. siNC, siRNA negative control; OE, overexpression; *p < 0.05, **p < 0.01, ***p < 0.001 compared with the control group.

3.3 Circ_0015508 Features a Circular Structure and Predominantly Localizes to the Cytoplasm

The important role of circRNAs in NPC has recently garnered increasing attention. To identify potential downstream targets of YTHDF2 in NPC cells, we performed circRNA-seq on HNE1 cells with YTHDF2 knockdown and control cells. For exploratory screening of candidate circRNAs, a |log2 FC| > 1 and a p-value < 0.05 were considered significantly differentially expressed. Compared with control cells, 98 circRNAs were upregulated and 137 circRNAs were downregulated upon YTHDF2 silencing (Fig. 3A). Considering that YTHDF2 may function as an m6A reader protein by promoting circRNA destabilization, we focused on differentially expressed circRNAs that were upregulated following YTHDF2 knockdown. Circ_0015508 was selected for further investigation according to its fold change, p-value, and predicted m6A modification sites (Fig. 3A and Fig. 4E). Sequence analysis indicated that circ_0015508 originates from the back-splicing of exons 2–6 of the human ACBD6 gene. (Fig. 3B). Sanger sequencing showed that the PCR product amplified by divergent primers corresponded to the head-to-tail junction sequence of circ_0015508, confirming its circular structure (Fig. 3C). To verify the circular feature of circ_0015508, RNase R digestion assay was conducted in HNE1 and 5-8F cells. The results showed that after RNase R treatment, the RNA level of linear ACBD6 mRNA was significantly reduced, whereas circ_0015508 remained unchanged, indicating that circ_0015508 resists RNase R digestion in both cell lines (Fig. 3D). Furthermore, qPCR assay was performed using oligo (dT) and random primers. The results showed that compared with random primers, circ_0015508 expression was significantly reduced when using oligo (dT) primers, while ACBD6 expression was detectable at comparable levels with both primer types, revealing that circ_0015508 does not contain the poly-A tail (Fig. 3E). These results indicate that circ_0015508 harbors a circular RNA structure. FISH assay was further performed to clarify the subcellular localization of circ_0015508. The results indicated that circ_0015508 was predominantly localized in the cytoplasm (Fig. 3F). The localization of circ_0015508 was also confirmed by the cytoplasmic and nuclear fractionation assay (Fig. 3G). These results indicated that circ_0015508 is a circular RNA predominantly present in the cytoplasm of NPC cells.

Figure 3: Circ_0015508 is predominantly localized in the cytoplasm and inhibits migration and invasion of NPC. (A) CircRNA-seq analysis of HNE1 cells after YTHDF2 knockdown and control cells. Heatmap showing differentially expressed circRNAs (|log₂ fold change (FC)| > 1 and p < 0.05). (B) Schematic diagram of the genomic location of circ_0015508. The back-splicing of exons 2–6 of the ACBD6 gene to form circ_0015508 is shown. (C) The PCR product amplified by divergent primers was subjected to Sanger sequencing. (D) Total RNA extracted from HNE1 and 5-8F cells was treated with or without RNase R, followed by qPCR analysis of circ_0015508 and linear ACBD6 mRNA. Relative expression levels were normalized to the Mock group. (E) qPCR analysis of circ_0015508 and ACBD6 using oligo (dT) primers or random primers in HNE1 and 5-8F cells. (F) Fluorescence in situ hybridization (FISH) assay was performed to determine the subcellular localization of circ_0015508 in HNE1-LV-circ_0015508 and 5-8F-LV-circ_0015508 cells. Nuclei were stained with DAPI (blue), and circ_0015508 was detected with specific probes (red). (G) Subcellular fractionation assay followed by qPCR analysis of circ_0015508, GAPDH (cytoplasmic control), and U6 (nuclear control) in HNE1 and 5-8F cells. (H) qPCR analysis of circ_0015508 expression in HNE1 cells stably transfected with lentiviral empty vector (LV-vector) or circ_0015508-overexpressing lentivirus (LV-circ_0015508). (I) Representative images of wound healing assays in HNE1 LV-vector cells or LV-circ_0015508 cells (left panel). Relative wound closure rate is presented in right panel. (J) Representative images of Transwell migration assays in HNE1 LV-vector cells or LV-circ_0015508 cells (left panel). Relative cell migration is presented in right panel. (K) Representative images of Transwell invasion assays in HNE1 LV-vector cells or LV-circ_0015508 cells (left). Relative cell migration is presented in right panel. (L) Western blot analysis of E-cadherin, N-cadherin, and vimentin in HNE1 cells transfected with LV-vector or LV-circ_0015508. Data are shown as the means ± SD. *p < 0.05, **p < 0.01; ***p < 0.001; ns no significance compared with the control group.

3.4 Circ_0015508 Inhibits NPC Migration and Invasion

To investigate the biological function of circ_0015508 in NPC cells, we constructed HNE1 cells stably overexpressing circ_0015508 (LV-circ_0015508) and an empty vector control (LV-vector) by lentivirus transduction. qPCR analysis confirmed that circ_0015508 expression was significantly increased in LV-circ_0015508 cells compared with the control group (Fig. 3H). Wound healing assays were performed to evaluate the effect of circ_0015508 on cell migration. The results showed that overexpression of circ_0015508 significantly reduced the wound closure rate in HNE1 cells compared with the control group (Fig. 3I). Consistently, transwell migration assays (without matrigel) demonstrated that the number of migrated cells was markedly decreased in circ_0015508-overexpressing cells (Fig. 3J). Transwell invasion assays (with matrigel) showed that circ_0015508 overexpression also significantly reduced the number of invasive cells (Fig. 3K). Moreover, Overexpression of circ_0015508 reduced protein levels of N-cadherin and Vimentin but raised those of E-cadherin (Fig. 3L). These results indicate that circ_0015508 is a circular RNA with inhibitory effects on migration, invasion, and EMT in NPC.

3.5 Circ_0015508 Is an m6A-Modified Circular RNA and YTHDF2 Promotes Its Degradation

Next, we examined the regulatory effect of YTHDF2 on circ_0015508. Consistent with the sequencing results, qPCR analysis showed that circ_0015508 expression was significantly increased upon YTHDF2 knockdown in HNE1 and 5-8F cells (Fig. 4A), whereas YTHDF2 overexpression led to decreased circ_0015508 levels (Fig. 4B). We next investigated whether this regulation was mediated by m6A modification. To determine whether YTHDF2 directly binds to circ_0015508, we performed RNA immunoprecipitation (RIP) assay using an anti-YTHDF2 antibody. The results showed that circ_0015508 was significantly enriched by the YTHDF2 antibody compared with the IgG control, indicating a direct interaction between YTHDF2 and circ_0015508 (Fig. 4C). MeRIP-qPCR assay was performed to confirm the presence of m6A modification on circ_0015508. The results demonstrated that circ_0015508 was significantly enriched by the m6A antibody, suggesting that circ_0015508 harbors m6A modification (Fig. 4D). To identify the specific m6A modification site on circ_0015508, we used the SRAMP (http://www.cuilab.cn/sramp), which predicted three potential m6A sites within the circ_0015508 sequence, two of which were classified as high confidence (Fig. 4E). To pinpoint the exact modification site, we performed the single-base elongation- and ligation-based qPCR amplification method (SELECT) according to the previous report [26]. Primers were designed targeting the sites (positions 387 and 396) as well as a non-m6A site (position 391, input control). When METTL3 was silenced in HNE1 and 5-8F cells, the SELECT product at the 396 site was significantly increased, whereas the 387 site was unaffected (Fig. 4F,G and Fig. S2A,B). This suggests that the 396 site may be an m6A modification site within circ_0015508. Considering that YTHDF2 acts as an m6A reader that typically binds to m6A-modified RNAs and regulates their stability, we further investigated whether YTHDF2 affects circ_0015508 stability. Upon treatment with actinomycin D to block transcription, we observed that circ_0015508 stability was significantly decreased in YTHDF2-overexpressing cells compared with control cells (Fig. 4H and Fig. S2C). Taken together, these results raise the possibility that YTHDF2 binds to the m6A-modified circ_0015508 and promotes its degradation in NPC cells.

3.6 Circ_0015508 Overexpression Alleviated the Pro-Migratory and Pro-Invasive Effects of YTHDF2 in NPC

Given that YTHDF2 promotes NPC cell migration and invasion (Fig. 2), and that YTHDF2 binds to circ_0015508 and promotes its degradation (Fig. 4C,H), we further investigated whether circ_0015508 overexpression could attenuate the oncogenic effects of YTHDF2. Rescue experiments were performed by co-overexpressing YTHDF2 and circ_0015508 in HNE1 and 5-8F cells. Wound healing assays revealed that YTHDF2 overexpression significantly enhanced cell migration, whereas this effect was reversed by concomitant overexpression of circ_0015508 (Fig. 4I and Fig. S2D). Similarly, transwell migration assays showed that the increased number of migrated cells caused by YTHDF2 overexpression was reversed by circ_0015508 co-overexpression (Fig. 4J and Fig. S2E). Transwell invasion assays further demonstrated that circ_0015508 restoration reversed YTHDF2-induced enhancement of cell invasion (Fig. 4K and Fig. S2F). Taken together, these results indicate that YTHDF2 may promote NPC migration and invasion by inhibiting tumor suppressive circ_0015508.

Figure 4: YTHDF2 facilitates the m6A-dependent degradation of circ_0015508, thus promoting NPC migration and invasion. (A) Relative expression of circ_0015508 was determined by qPCR in HNE1 cells transfected with siRNA negative control (siNC) or siYTHDF2. (B) Relative expression of circ_0015508 was determined by qPCR in HNE1 cells transfected with empty vector or YTHDF2-overexpressing plasmid. (C) RNA immunoprecipitation (RIP) assay was performed using anti-YTHDF2 antibody or IgG in HNE1 cells. Enrichment of circ_0015508 was detected by qPCR. (D) Methylated RNA immunoprecipitation qPCR (MeRIP-qPCR) assay was used to determine the enrichment of circ_0015508 by anti-m6A antibody compared with IgG control. (E) Schematic diagram of the predicted m6A modification sites on circ_0015508 using SRAMP. (F) The mRNA expression of METTL3 was determined by qPCR in HNE1 cells transfected with siNC or siMETTL3. (G) Single-base elongation- and ligation-based qPCR amplification (SELECT) assay was performed in HNE1 cells to detect m6A modification at positions 387, 396, and a non-m6A site (391) in circ_0015508. The threshold cycle (C_T) of qPCR was shown. (H) circ_0015508 expression was determined by qPCR in control and YTHDF2-overexpressing HNE1 cells treated with actinomycin D (5 μg/mL) for the indicated time points. (I) Representative images of wound healing assays in HNE1 cells transfected with OE-Ctrl + Empty vector, OE-YTHDF2 + Empty vector, or OE-YTHDF2 + OE-circ_0015508 (left panel). Relative wound closure rate is presented in right panel. (J) Representative images of Transwell migration assays in HNE1 cells transfected OE-Ctrl + Empty vector, Empty vector + OE-YTHDF2, or OE-YTHDF2 + OE-circ_0015508 (left panel). Relative cell migration is presented in right panel. (K) Representative images of Transwell invasion assays in HNE1 cells transfected with OE-Ctrl + Empty vector, OE-YTHDF2 + Empty vector, or OE-YTHDF2 + OE-circ_0015508 (left panel). Relative cell invasion is presented in right panel. Data are shown as the means ± SD. siNC, siRNA negative control; OE, overexpression; ns no significance; *p < 0.05; **p < 0.01; ***p < 0.001.

3.7 Circ_0015508 Acts as a miRNA Sponge That Directly Binds to miR-496

Next, we investigated the mechanism by which circ_0015508 exerts its tumor-suppressive role in NPC. Functioning as a miRNA sponge represents one of the primary mechanisms underlying circRNA biology. Using three independent databases (RNAhybrid, MiRanda, and CircInteractome), we predicted potential miRNAs that may bind to circ_0015508. Overlapping the predictions from these databases yielded five candidate miRNAs, including miR-513a-5p, miR-496, miR-942-5p, miR-619-3p and miR-1233-3p (Fig. 5A). Based on binding scores and predicted favorable binding relationship, miR-496 was selected for subsequent investigations. We then constructed a binding region mutant (circ_0015508-mut) (Fig. 5B). Dual-luciferase reporter assays demonstrated that wild-type circ_0015508 significantly bound to miR-496, whereas the binding ability of mutant circ_0015508 was markedly reduced (Fig. 5C). To determine whether circ_0015508 regulates miR-496 expression, we detected miR-496 levels in HNE1 and 5-8F cells stably overexpressing circ_0015508 (LV-circ_0015508) and control cells (LV-vector). qPCR analysis showed that miR-496 expression was significantly decreased in circ_0015508-overexpressing cells compared with the control group (Fig. 5D). These results suggest that circ_0015508 can adsorb and bind to miR-496. We next investigated the functional role of miR-496 in NPC cells. HNE1 cells were transfected with miR-496 mimics or inhibitors. qPCR analysis confirmed efficient overexpression and knockdown of miR-496, respectively (Fig. 5E). Wound healing assays were performed to evaluate the effect of miR-496 on cell migration. The results showed that overexpression of miR-496 significantly enhanced the wound closure rate, whereas knockdown of miR-496 suppressed cell migration (Fig. 5F,G). Transwell assays showed that overexpression of miR-496 promoted migration and invasion in NPC cells (Fig. 5H,I). Western blotting demonstrated that overexpression of miR-496 led to increased protein levels of N-cadherin and Vimentin and decreased expression of E-cadherin (Fig. 5J). Conversely, knockdown of miR-496 resulted in opposite effect (Fig, 5J). These results suggest that miR-496 promotes migration, invasion and EMT in NPC cells, and circ_0015508 directly binds to and downregulates its expression.

To investigate whether circ_0015508 exerts its biological functions through miR-496, we performed rescue experiments by reintroducing miR-496 mimics into circ_0015508-overexpressing HNE1 cells. Wound healing assays showed that circ_0015508 overexpression significantly suppressed cell migration, whereas this inhibitory effect was reversed upon co-transfection with miR-496 mimics (Fig. 5K). Consistently, Transwell migration assays demonstrated that the decreased number of migrated cells caused by circ_0015508 overexpression was reversed by miR-496 mimics (Fig. 5L). Moreover, Transwell invasion assays showed that circ_0015508-mediated suppression of cell invasion was also attenuated by miR-496 restoration (Fig. 5M). Considering the binding relationship and negative regulatory effect of circ_0015508 on miR-496, these results indicate that circ_0015508 suppresses NPC cell migration and invasion partly by sponging miR-496.

Figure 5: Circ_0015508 functions as a miRNA sponge by directly binding to miR-496. (A) Venn diagram showing the overlap of miRNAs predicted to bind circ_0015508 by RNAhybrid, miRanda, and circInteractome databases. (B) Schematic diagram of the predicted binding site between circ_0015508 and miR-496. (C) Dual-luciferase reporter assays showing the relative luciferase activity in HEK293T cells co-transfected with wild-type (Wt) or mutant (Mut) circ_0015508 reporter and negative control (NC) mimics or miR-496 mimics. (D) qPCR was performed to detect miR-496 expression in HNE1 (left panel) and 5-8F (right panel) cells stably overexpressing circ_0015508 (LV-circ_0015508) versus control cells (lentiviral empty vector, LV-vector). (E) qPCR analysis of miR-496 expression in HNE1 cells transfected with NC mimics or miR-496 mimics (left panel), and NC inhibitor or miR-496 inhibitor (right panel). (F) Representative images of wound healing assays in HNE1 cells transfected with NC mimics or miR-496 mimics (left panel). Relative wound closure rate is presented in right panel. (G) Representative images of wound healing assays in HNE1 cells transfected with NC inhibitor, or miR-496 inhibitor (left panel). Relative wound closure rate is presented in right panel. (H) Representative images of Transwell migration assays in HNE1 cells transfected with NC mimic or miR-496 mimic (left panel), NC inhibitor or miR-496 inhibitor (middle panel). Relative cell migration is presented in right panel. (I) Representative images of Transwell invasion assays in HNE1 cells transfected with NC mimics or miR-496 mimics (left panel), NC inhibitor or miR-496 inhibitor (middle panel). Relative cell invasion is presented in right panel. (J) Western blot analysis of E-cadherin, N-cadherin, and vimentin in HNE1 cells transfected with NC mimics or miR-496 mimics, NC inhibitor or miR-496 inhibitor. (K) Representative images of wound healing assays in HNE1 LV-vector cell line transfected with NC mimics, LV-circ_0015508 cell line transfected with NC mimics, or LV-circ_0015508 cell line transfected with miR-496 mimics (left panel). Relative wound closure rate is presented in right panel. (L) Representative images of Transwell migration assays in HNE1 LV-vector cell line transfected with NC mimics, LV-circ_0015508 cell line transfected with NC mimics, or LV-circ_0015508 cell line transfected with miR-496 mimics (left panel). Relative cell migration is presented in right panel. (M) Representative images of Transwell invasion assays in HNE1 LV-vector cell line transfected with NC mimics, LV-circ_0015508 cell line transfected with NC mimics, or LV-circ_0015508 cell line transfected with miR-496 mimics (left panel). Relative cell invasion is presented in right panel. Data are shown as the means ± SD. NC, negative control; ns no significance; *p < 0.05; **p < 0.01; ***p < 0.001.

3.8 FOXN3 Is a Candidate Target of miR-496

To identify the downstream targets of miR-496, we predicted potential target genes using three independent databases: miRWalk, microT-CDS, and miRDB. Overlapping the predictions from these databases yielded 24 candidate genes (Fig. 6A). Based on their reported functions in diseases, we selected three genes with high potential as downstream targets: neuronal PAS domain protein 3 (NPAS3), fibronectin leucine rich transmembrane protein 2 (FLRT2), and forkhead box N3 (FOXN3). We examined the expression of NPAS3, FLRT2, and FOXN3 in HNE1 cells after miR-496 overexpression or knockdown. The results showed that miR-496 overexpression significantly decreased the expression of these three genes (Fig. 6B), whereas miR-496 knockdown increased their expression, indicating a negative correlation between miR-496 and these targets (Fig. 6C). Among these, FOXN3 showed the most significant changes upon both miR-496 overexpression and knockdown and was therefore selected for subsequent validation. In multiple types of cancer, FOXN3 has been reported to function as a tumor suppressor [30,31,32]. To determine whether FOXN3 is a direct target of miR-496, we constructed wild-type (WT) and mutant (Mut) FOXN3 luciferase reporters based on the predicted binding site (Fig. 6D). Dual-luciferase reporter assays demonstrated that co-transfection with miR-496 mimics significantly reduced the luciferase activity of the wild-type FOXN3 reporter, whereas the mutant reporter showed no significant change (Fig. 6E). Consistent with the qPCR results, FOXN3 protein expression was decreased upon miR-496 overexpression and increased upon miR-496 knockdown (Fig. 6F,G). These results indicate that FOXN3 is a target of miR-496. Notably, we also investigated the effect of circ_0015508 on FOXN3 expression at both the RNA and protein levels. We found that increased circ_0015508 expression significantly promoted FOXN3 expression (Fig. 6H,I). Given the circ_0015508’s ability to bind and inhibit miR-496, these observations suggest that circ_0015508 may function as a miRNA sponge, thereby alleviating the miR-496-mediated repression of FOXN3.

Figure 6: FOXN3 is identified as a candidate target of miR-496. (A) Venn diagram showing predicted miR-496 targets by miRWalk, microT-CDS, and miRDB databases. (B) qPCR analysis of neuronal PAS domain protein 3 (NPAS3), fibronectin leucine rich transmembrane protein 2 (FLRT2), and forkhead box N3 (FOXN3) mRNA expression in HNE1 cells transfected with negative control (NC) mimics or miR-496 mimics. (C) qPCR analysis of NPAS3, FLRT2, and FOXN3 mRNA expression in HNE1 cells transfected with NC inhibitor or miR-496 inhibitor. (D) Schematic diagram of the predicted binding site between miR-496 and FOXN3. (E) Dual-luciferase reporter assays showing the relative luciferase activity in 293T cells co-transfected with wild-type (Wt) or mutant (Mut) FOXN3 reporter and NC mimics or miR-496 mimics. (F) Western blot analysis of FOXN3 protein expression in HNE1 cells transfected with NC-mimics or miR-496-mimics. (G) Western blot analysis of FOXN3 protein expression in HNE1 cells transfected with NC inhibitor or miR-496 inhibitor. (H) qPCR analysis of FOXN3 expression in HNE1 lentiviral empty vector (LV-vector) cells or LV-circ_0015508 cells. (I) Western blot analysis of FOXN3 protein expression in HNE1 LV-vector cells or LV-circ_0015508 cells. Data are shown as the means ± SD. NC, negative control; ns no significance; *p < 0.05; **p < 0.01; ***p < 0.001 compared with the control group.

3.9 FOXN3 Knockdown Reversed the YTHDF2 Knockdown-Induced Inhibition of NPC Cell Migration and Invasion

To further elucidate the role of YTHDF2/circ_0015508/miR-496/FOXN3 axis in NPC. We first analyzed the relationship between YTHDF2 and FOXN3 expression in the GSE12452 dataset. The results revealed a significant negative correlation between YTHDF2 and FOXN3 expression (r = −0.4835, p = 0.0014; Fig. 7A). Consistently, Western blot analysis in HNE1 cells showed that YTHDF2 knockdown increased FOXN3 protein expression, confirming a negative correlation between YTHDF2 and FOXN3 (Fig. 7B). To investigate whether YTHDF2 exerts its biological functions through the circ_0015508/miR-496/FOXN3 axis, we performed experiments by knocking down FOXN3 in YTHDF2-knockdown HNE1 cells. The inhibitory effects of YTHDF2 and FOXN3 siRNAs on their corresponding protein expression were confirmed by Western blot analysis. Wound healing assays showed that YTHDF2 knockdown significantly suppressed cell migration, whereas this inhibitory effect was partially reversed upon co-transfection with siFOXN3 (Fig. 7C). Transwell migration assays demonstrated that the decreased number of migrated cells caused by YTHDF2 knockdown was partially reversed by siFOXN3 (Fig. 7D). Moreover, Transwell invasion assays showed that YTHDF2 knockdown-mediated suppression of cell invasion was also attenuated by FOXN3 knockdown (Fig. 7E). In further rescue experiments, wound healing assays and Transwell migration assays revealed that the cell migration-promoting effect induced by YTHDF2 overexpression could be partially reversed by the miR-496 inhibitor in HNE1 and 5-8F cells (Fig. 7F,G and Fig. S3A,B). Similarly, Transwell invasion assays showed that the cell invasion-promoting effect induced by YTHDF2 overexpression could also be partially reversed by the miR-496 inhibitor in HNE1 and 5-8F cells (Fig. 7H and Fig. S3C). Our studies have established that YTHDF2 exerts negative regulation on circ_0015508 expression via an m6A-dependent mechanism, and that circ_0015508 regulates the miR-496/FOXN3 axis by acting as a miR-496 sponge. In conjunction with the validated negative regulation of FOXN3 by YTHDF2, our findings collectively demonstrate that YTHDF2 promotes NPC cell migration and invasion through the circ_0015508/miR-496/FOXN3 axis (Fig. 8).

Figure 7: Knocking down FOXN3 reversed the migration/invasion inhibition in NPC cells caused by YTHDF2 knockdown. (A) Correlation analysis of YTHDF2 and FOXN3 expression in the GSE12452 dataset. (B) Western blot analysis of FOXN3 protein expression in HNE1 cells transfected with siRNA negative control (siNC), siNC + siYTHDF2 or siYTHDF2 + siFOXN3. (C) Representative images of wound healing assays in HNE1 cells transfected with siNC, siNC + siYTHDF2, or siYTHDF2 + siFOXN3 (left panel). Relative wound closure rate is presented in right panel. (D) Representative images of Transwell migration assays in HNE1 cells transfected with siNC, siNC + siYTHDF2, or siYTHDF2 + siFOXN3 (left panel). Relative cell migration is presented in right panel. (E) Representative images of Transwell invasion assays in HNE1 cells transfected with siNC, siNC + siYTHDF2, or siYTHDF2 + siFOXN3 (left panel). Relative cell invasion is presented in right panel. (F) Representative images of wound healing assays in HNE1 cells transfected with OE-Ctrl + NC inhibitor, OE-YTHDF2 + NC inhibitor, or OE-YTHDF2 + miR-496 inhibitor (left panel). Relative wound closure rate is presented in right panel. (G) Representative images of Transwell migration assays in HNE1 cells transfected with OE-Ctrl + NC inhibitor, OE-YTHDF2 + NC inhibitor, or OE-YTHDF2 + miR-496 inhibitor (left panel). Relative cell migration is presented in right panel. (H) Representative images of Transwell invasion assays in HNE1 cells transfected with OE-Ctrl + NC inhibitor, OE-YTHDF2 + NC inhibitor, or OE-YTHDF2 + miR-496 inhibitor (left panel). Relative cell invasion is presented in right panel. Data are shown as the means ± SD. siNC, siRNA negative control; OE, overexpression; **p < 0.01; ***p < 0.001.

Figure 8: YTHDF2/circ_0015508/miR-496/FOXN3 axis regulates tumor migration, invasion, and EMT. Schematic diagram illustrating that the YTHDF2/circ_0015508/miR-496/FOXN3 axis regulates NPC cell migration, invasion, and EMT. In NPC, YTHDF2 expression is upregulated and promotes the degradation of circ_0015508 by binding to its m6A-modified 396 site. The loss of circ_0015508 releases its inhibitory effect on miR-496, which in turn suppresses FOXN3, a tumor suppressor. Consequently, NPC cell migration, invasion, and EMT are enhanced. NPC, nasopharyngeal carcinoma; YTHDF2, YTH N6-methyladenosine RNA binding protein F2; FOXN3, forkhead box N3.

4 Discussion

In recent years, the role of circRNAs in tumor metabolism and progression has attracted increasing attention [8,33]. Unlike linear RNAs, circRNAs are resistant to degradation by exonucleases such as RNase R due to their covalently closed loop structure, which underlies their high stability [34]. In our study, Sanger sequencing confirmed the head-to-tail junction of circ_0015508 (Fig. 3B,C). Compared with the linear ACBD6 mRNA, circ_0015508 exhibited greater resistance to RNase R digestion in HNE1 and 5-8F cells (Fig. 3D). Additionally, circ_0015508 could be amplified by PCR using random primers rather than oligo(dT) primers, indicating its lack of a poly(A) tail (Fig. 3E). These results confirmed that circ_0015508 harbored a bona fide circular structure. Subcellular localization assays revealed that circ_0015508 was predominantly localized in the cytoplasm of NPC cells (Fig. 3F,G), suggesting its potential to function as a competitive endogenous RNA (ceRNA) by sponging miRNAs.

The m6A modification is the most prevalent internal RNA modification in eukaryotes, and its dysregulation has been implicated in various cancers [13,14,35,36]. Among the m6A regulatory machinery, the reader protein YTHDF2 has emerged as a critical player in RNA metabolism, typically mediating the degradation of m6A-modified transcripts [16,37]. In our study, YTHDF2 expression was significantly elevated at both the mRNA and protein levels in NPC tissues compared with normal controls (Fig. 1). Recent studies also reported YTHDF2 upregulation in various malignancies, for example anaplastic thyroid cancer and oral squamous cell carcinoma, where it often correlates with aggressive tumor behavior [17,18]. To investigate the functional role of YTHDF2 in NPC, we performed loss-of-function and gain-of-function experiments in HNE1 and 5-8F cells. Knockdown of YTHDF2 significantly suppressed cell migration and invasion, whereas YTHDF2 overexpression promoted these aggressive phenotypes. Moreover, Western blot analysis revealed that YTHDF2 positively regulated the expression of mesenchymal markers N-cadherin and vimentin, while negatively regulating the epithelial marker E-cadherin (Fig. 2), indicating that YTHDF2 promotes migration, invasion and EMT in NPC cells. Recently, accumulating evidence has revealed the oncogenic role of YTHDF2 in various types of cancer. In anaplastic thyroid cancer, YTHDF2 was upregulated and promoted proliferation, invasion, and migration by degrading DNA damage inducible transcript 4 (DDIT4) mRNA in an m6A-dependent manner, thereby activating the AKT/mTOR pathway and inducing EMT [18]. In triple-negative breast cancer (TNBC), Zhang et al. [38] demonstrated that processing of precursors 1 (POP1) directly bound to the coding sequence (CDS) of cyclin dependent kinase inhibitor 1A (CDKN1A) mRNA and facilitated its degradation, consequently accelerating cell cycle progression and TNBC proliferation. Notably, this degradation was dependent on the m6A modification at site 497 of CDKN1A and its subsequent recognition by YTHDF2. In prostate cancer, YTHDF2 facilitated m6A-dependent decay of CDKN1C mRNA. Fbxo2 inhibited cell proliferation and motility by ubiquitinating YTHDF2 and targeting it for proteasomal degradation. YTHDF2 knockdown partially reversed the enhanced proliferation and motility induced by Fbxo2 depletion, suggesting the oncogenic role of YTHDF2 [39]. In NPC, a recent study also demonstrated that YTHDF2 enhanced proliferation, migration, and invasion phenotypes. Further studies showed that FOXO1 mRNA contained m6A modification. YTHDF2 bound to this transcript and negatively regulated FOXO1 expression [40]. In line with their observations, our study confirmed that YTHDF2 enhanced NPC cell migration and invasion. Mechanistically, we identified an additional mechanism of YTHDF2 function. Specifically, YTHDF2 degraded circ_0015508 via an m6A-dependent mechanism, which relieved the circ_0015508-mediated sponging of miR-496. This increased the inhibitory effect of miR-496 on its target FOXN3, thereby promoting NPC migration and invasion.

We performed circRNA-seq on YTHDF2-knockdown NPC cell line HNE1 and control cells to identify potential target circRNAs of YTHDF2 in NPC. CircRNA-seq analysis following YTHDF2 knockdown revealed 235 differentially expressed circRNAs. Considering the pro-cancer role of YTHDF2 in NPC and its primary mechanism of binding to and suppressing targets via the m6A pathway, we selected circ_0015508 as a research candidate due to its fold change and p-value upon YTHDF2 knockdown, and the presence of predicted m6A binding sites from sequence analysis (Fig. 3A and Fig. 4E). Further experiments demonstrated that circ_0015508 was significantly upregulated upon YTHDF2 silencing (Fig. 4A). Conversely, YTHDF2 overexpression led to decreased circ_0015508 expression (Fig. 4B). These results suggested that YTHDF2 negatively regulated circ_0015508. To determine whether this regulation is direct, we performed RIP assays and confirmed that YTHDF2 bound to circ_0015508 (Fig. 4C). Furthermore, MeRIP-qPCR assays revealed that circ_0015508 was enriched by the m6A antibody, indicating the presence of m6A modifications on this circRNA (Fig. 4D). Using SRAMP prediction combined with SELECT assays, we suggested the 396 site may be the specific m6A modification site on circ_0015508 (Fig. 4E–G). Importantly, actinomycin D assays demonstrated that YTHDF2 overexpression significantly decreased the stability of circ_0015508 (Fig. 4H). These results provide evidence that YTHDF2 may recognize the m6A modification on circ_0015508 and promotes its degradation, thereby negatively regulating circ_0015508 expression. However, the current identification of the m6A modification site on circ_0015508 relied on the SELECT method. Given that circRNAs possess secondary structures, additional evidence from cross-validation, such as mutational analysis, will be beneficial in future work to fully elucidate the m6A-dependent regulatory mechanism of circ_0015508 by YTHDF2. Accumulating evidence supports the role of YTHDF2 in regulating circRNA stability. Recent studies have shown that YTHDF2 can diminish the stability and biological functions of circRNAs by recognizing m6A-modified circRNAs and recruiting degradation complexes. Park et al. [41] demonstrated that YTHDF2 recognized m6A-modified circRNAs and recruited the RNase P/MRP and HRSP12 complex, thereby mediating their cleavage and subsequent degradation. Chen et al. [42] found that circIRF2 exerted anti-fibrotic effects in liver fibrosis mice. Mechanistically, circIRF2 acted as a sponge for miR-29b-1-5p, competitively alleviating its suppression of FOXO3 and thereby promoting FOXO3 nuclear translocation. Notably, YTHDF2 was found to recognize m6A-modified circIRF2 and impair its stability. In hepatitis B virus-related hepatocellular carcinoma, the Hepatitis B protein x (HBx) promoted RBM15 expression, which increased m6A modification of cFAM210A, leading to cFAM210A degradation via the YTHDF2-HRSP12-RNase P/MRP pathway [43]. Our study extends these findings by establishing that YTHDF2 promotes NPC migration and invasion through the degradation of a tumor-suppressive circ_0015508. We noticed from our sequencing results that the expression of circ_0015508 was significantly increased upon YTHDF2 knockdown, but its overall abundance remained low, which may be attributed to its low basal expression in HNE1 cells. This characteristic is consistent with the typical expression pattern of tumor suppressor genes and also suggests that, in addition to YTHDF2-mediated mechanisms, other regulatory mechanisms may be involved in controlling circ_0015508 expression, warranting further investigation.

CircRNAs exert their functional roles through diverse molecular mechanisms, including functioning as ceRNAs to sponge miRNAs, modulating protein activity or localization through direct interactions, regulating the transcription of their host genes, and encoding functional peptides [44,45,46]. Liu et al. [47] revealed that circTGFBR2 was downregulated in NPC and functioned as a tumor suppressor by sponging miR-107, thereby upregulating TGFBR2 expression and inhibiting NPC proliferation and migration. Hong et al. [48] demonstrated that circCRIM1 was upregulated in highly metastatic NPC cells and promoted metastasis, EMT, and docetaxel chemoresistance by sponging miR-422a, thereby relieving its suppression on FOXQ1. Another study reported that circTP63-N suppressed NPC progression by recruiting LATS/YAP1 via HSP90AB1, which drove the phosphorylation- and ubiquitination-dependent YAP1 degradation [49]. In this study, we found that circ_0015508 exerted its inhibitory effect on NPC cell migration and invasion through the mechanism of sponging miR-496. Dual-luciferase reporter assays validated the direct binding between circ_0015508 and miR-496 (Fig. 5C). miR-496 has been reported to exhibit dual roles in tumor progression, acting as either a tumor suppressor or an oncogene. In hepatocellular carcinoma, METTL3-mediated m6A modification stabilized circ_0027791, which sponged miR-496 to upregulate programmed death-ligand 1 (PD-L1), thereby promoting tumor progression and immune escape [50]. However, in colorectal cancer, miR-496 acted as an oncogene by directly targeting RASSF6 and activating the Wnt signaling pathway, thereby promoting cell migration and EMT [51]. Our study provided evidence that in the context of NPC, miR-496 functioned as a pro-migratory and pro-invasive factor. Overexpression of miR-496 significantly promoted cell migration, invasion and EMT, whereas knockdown of miR-496 suppressed these phenotypes (Fig. 5F–J). To further elucidate the downstream mechanism of miR-496, we predicted its target genes. FOXN3, a member of the forkhead box transcription factor family, has been implicated in tumor suppression in various cancers. Its expression showed a significant negative correlation with miR-496 expression. Dual-luciferase reporter assays confirmed that FOXN3 was a target of miR-496 (Fig. 6). In ovarian cancer, FOXN3 was downregulated and exerted tumor-suppressive effects by inhibiting proliferation, migration, invasion, and angiogenesis. Mechanistically, FOXN3 bound to the promoter of RPS15A and negatively regulated its transcription [30]. In papillary thyroid carcinoma (PTC), FOXN3 was markedly downregulated. Overexpression of FOXN3 exerted tumor-suppressive effects in PTC cells, suppressing proliferation, colony formation, migration, and invasion. Further research demonstrated that these effects were mediated by the Wnt/β-catenin pathway [31]. In NPC, FOXN3 was a direct target of miR-574-5p, and its overexpression reversed the pro-migratory, and pro-invasive effects of miR-574-5p by regulating Wnt/β-catenin signaling [32]. Our study provided evidence that, in NPC, FOXN3 was also regulated by a miRNA different from miR-574-5p. Our rescue experiments also confirmed the tumor-suppressive effect of FOXN3 in NPC, as FOXN3 knockdown reversed the inhibitory effects of YTHDF2 knockdown on NPC cell migration and invasion (Fig. 7). Our study demonstrated that FOXN3 was positively regulated by circ_0015508 and negatively regulated by miR-496 (Fig. 6), indicating that FOXN3 was a critical downstream effector in this regulatory axis.

Therefore, our study constructed a comprehensive regulatory YTHDF2/circ_0015508/miR-496/FOXN3 axis that played an important role in the migration, invasion, and EMT of NPC. YTHDF2, which was upregulated in NPC, promoted NPC cell migration and invasion by recognizing the m6A modification on circ_0015508 and promoting its degradation. This degradation relieved the inhibitory constraint of circ_0015508 on miR-496, allowing miR-496 to suppress its downstream target FOXN3, a tumor suppressor, ultimately driving NPC cell migration, invasion, and EMT (Fig. 8). The identification of this regulatory axis provided a mechanistic insight into NPC migration and invasion. Our study found that YTHDF2 bound to circ_0015508 in an m6A-dependent manner and induced its degradation. We also identified the m6A modification site of circ_0015508 at position 396. This suggests that circ_0015508 harboring a mutation at this m6A site may potentially resist YTHDF2-mediated degradation, thereby maintaining its tumor-suppressive function.

However, this study also has several limitations. First, while FOXN3 is a transcription factor, the specific downstream effectors through which it regulates NPC cell migration and invasion remain unclear. Second, Epstein-Barr virus (EBV) infection plays a critical role in the development of NPC, with approximately 96% of NPC incidences in endemic regions being attributable to EBV [52,53]. Nevertheless, our current experimental model does not account for the context of EBV infection. Third, our study is primarily based on in vitro experiments and lacks in vivo validation. Animal models, such as xenograft mouse models, are needed to confirm the functional significance of the YTHDF2/circ_0015508/miR-496/FOXN3 axis in NPC migration and invasion in future studies. Moreover, the effect of this regulatory axis on clinicopathological parameters and tumor metastasis needs to be validated in expanded clinical samples from NPC patients. Resolving these limitations in future studies will further enhance the translational potential of our findings.

5 Conclusion

Altogether, our present study suggests that highly expressed YTHDF2 in NPC promotes the instability of circ_0015508 by recognizing its m6A modification site. This releases the inhibitory effect of circ_0015508 on miR-496, thereby enhancing miR-496 mediated suppression of FOXN3. Consequently, this cascade promotes NPC migration, invasion, and alterations in EMT-related markers.