circNLK restrains glioblastoma malignancy by binding to hnRNP L and destabilizing RGS16

Abstract

Glioblastoma (GBM) is the most aggressive primary brain tumor in adults, underscoring the need to identify tumor-suppressive regulators for the development of more effective therapies. Circular RNAs (circRNAs) have emerged as important regulators in cancers, yet their roles in GBM remain incompletely understood. In this study, we identified circular NLK (circNLK; hsa_circ_0002103) as a tumor suppressor in GBM. circNLK expression was downregulated in GBM relative to non-neoplastic brain tissues, and higher circNLK expression associated with better patient outcomes. Ectopic circNLK overexpression inhibited GBM cell proliferation and motility while increasing apoptosis. Mechanistically, circNLK bound to the RNA-binding protein hnRNP L in an m⁶A-associated manner and partially competed with the RGS16 RNA transcript for hnRNP L binding, thereby reducing RGS16 stability and expression. In orthotopic mouse models of GBM, circNLK overexpression delayed tumor growth, prolonged survival, and reduced RGS16 expression in tumor tissues. Together, these findings expand our understanding of circRNAs as tumor suppressors and provide a rationale for exploring circNLK-based therapeutic strategies in GBM.

1. Introduction

Glioblastoma (GBM) is the most aggressive and lethal primary brain tumor in adults. Despite multimodal treatments including maximal safe resection, radiotherapy, and chemotherapy, the median survival of GBM patients remains only 12–14 months [1]. Several new agents, including the anti-angiogenic drug bevacizumab, have been evaluated for GBM treatment; however, clinical benefit has been limited, with little or no improvement in overall survival [2,3]. These observations highlight the urgent need to identify more effective molecular targets for GBM therapy.

Circular RNAs (circRNAs) are single-stranded, covalently closed RNAs molecules that lack 5′ caps and 3′ terminal poly A tail [4]. Increasing evidence indicates that circRNAs regulate diverse cancer-related processes. In GBMs, circMET encoding a 404-amino-acid MET variant (MET404) promoted tumorigenesis by activating MET signaling [5]. Likewise, circNT5E functioned as a molecular sponge of miR-422a and supported GBM growth [6]. Moreover, circRNAs may serve as biomarkers for diagnosis and prognosis of GBM [7,8]. However, most studies to date have focused on oncogenic circRNAs. Several circRNAs are expressed at lower levels in glioma than in non-neoplastic brain tissues, suggesting tumor-suppressive circRNAs in GBM remain underexplored.

Heterogeneous nuclear ribonucleoproteins (hnRNPs) are a family of RNA-binding proteins that regulate multiple aspects of RNA metabolism [[9], [10], [11]]. Several hnRNP family members have been identified as N6-methyladenosine (m⁶A) readers, thereby linking RNA modification to splicing, stability, and nuclear export [12,13]. For example, hnRNPA2B1 and hnRNPC mediate m⁶A-dependent alternative splicing and microRNA processing [14,15]. Similarly, hnRNPG binds both phosphorylated RNAPII and m⁶A-modified nascent pre-mRNAs, thereby regulating alternative splicing on a transcriptome-wide scale [16]. hnRNP L shares structural features with other hnRNP family members and has been implicated in RNA stability and processing. However, whether hnRNP L recognizes m⁶A-modified transcripts and contributes to GBM progression remains largely unknown.

Regulator of G protein signaling 16 (RGS16) is a member of the regulator of G protein signaling family and functions as a GTPase-activating protein for Gα subunits, thereby controlling the magnitude and duration of G protein-coupled receptor signaling [17,18]. Through modulation of G-protein signaling, RGS16 can influence cellular processes relevant to tumor progression, including proliferation, cell-cycle progression, apoptosis, and migration. In GBM, RGS16 is expressed at elevated levels and is associated with unfavorable patient prognosis [19,20]. Additionally, RGS16 may contribute to aggressive tumor growth by accelerating cell cycle and suppressing apoptosis [21]. These observations indicate that aberrant RGS16 expression may contribute to the aggressive biological behavior of GBM. However, the mechanisms responsible for maintaining elevated RGS16 expression, particularly its post-transcriptional regulation and mRNA stability, remain poorly understood.

In the present study, we identified circNLK as a markedly downregulated circRNA in GBM and investigated its biological function and molecular mechanism. We show that circNLK suppresses GBM progression by interacting with hnRNP L in an m⁶A-associated manner and destabilizing RGS16 mRNA, thereby revealing a previously unrecognized circNLK-hnRNP L-RGS16 regulatory axis in GBM.

2. Materials and methods

2.1. Human specimens

40 GBM samples and 22 non-neoplastic brain tissues were collected at the First Affiliated Hospital of the University of Science and Technology of China (USTC) between 2019 and 2022 in accordance with an Institutional Research Ethics Committee-approved protocol. GBM tissues were obtained from 22 male and 18 female patients aged 29–81 years, while then non-neoplastic brain tissues were from 16 male and 6 female patients aged 25–54 years. Informed consent was obtained from all subjects and experiments were approved by the Medical Research Ethics Committee of The First Affiliated Hospital of the University of Science and Technology of China (approval No. 2024KY-293). Participants are not compensated for the involvement in this study. Clinical, surgical, and pathological data were retrospectively reviewed, with histological diagnoses independently confirmed by two certified pathologists. Follow-up information, including patient survival, was recorded up to September 20, 2023.

2.2. Cell lines and cell culture

U251 cells and U87 cells were purchased from the National Collection of Authenticated Cell Cultures (BNCC100123, BNCC353698, Shanghai, China). Normal human astrocytes (NHAs) were purchased from Shenzhen Haodi Huatuo Biotechnology Co., Ltd. (HTX2408, Shenzhen, China). Primary GBM cell line (GBM1214) was kind gift from Dr. Yu Shi (Department of Neurosurgery, Army Medical University, Chongqing, China) [22]. Cells were cultured in DMEM/F12 (12491023, Gibco) supplemented with 10% FBS (A5256901, Gibco) in a humidified incubator with 5% CO2. Mycoplasma testing was routinely performed, and all cultures remained contamination-free throughout the study.

2.3. Quantitative PCR (qPCR)

Total RNA was extracted from tissues and cultured cells using TRIzol reagent (15596026CN, Invitrogen, USA) according to the manufacturer's instruction. RNA concentration and purity were assessed using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, USA), and samples with an A260/A280 ratio of 1.8–2.0 were used for subsequent experiments. Reverse transcription and quantitative PCR were performed using ChamQ SYBR qPCR Master Mix (Without ROX) (Q321, Vazyme, China) on an ABI 7500 Fast Real-Time PCR System (Applied Biosystems, USA). Relative gene expression was calculated using the 2^-ΔΔCt method, with GAPDH as the endogenous reference. All reactions were conducted in triplicate to ensure technical reproducibility. Primer sequences used in this study are listed in Supplementary Table S1.

2.4. Actinomycin D treatment

Actinomycin D treatment was performed to assess the stability of circNLK in GBM cells. Cells were put in medium containing actinomycin D (2 μg/mL) (C0817, NobleRyder, China) to block de novo RNA transcription, and were harvested at different time points after treatment. Total RNA was extracted, and the levels of circNLK transcript were quantified by qPCR.

2.5. Cell proliferation assays

Cell proliferation was assessed using the Cell Counting Kit-8 (A311-02, Vazyme, China) according to the manufacturer's instruction. Briefly, cells were seeded into 96-well plates at a density of 1,000 cells/well in 100 μL complete medium. At 0, 24, 48, and 72 h after seeding, 10 μL of CCK-8 reagent was added to each well, followed by incubation at 37 °C for 2–4 h. The absorbance at 450 nm was then measured using a microplate reader (DeTie, China). Each experiment was repeated three times independently to ensure biological reproducibility.

2.6. Cell motility assays

Cell motility was assessed using Transwell chambers (8-μm pore size) (CLS3527, Corning, USA). For migration assays, 5 × 10⁴ cells in 200 μL serum-free medium were seeded into the upper chamber, with 600 μL complete medium containing 10% FBS in the lower chamber. After 24 h incubation at 37 °C, non-migrated cells on the upper surface were gently removed with a cotton swab, and migrated cells on the lower surface were fixed with 4% paraformaldehyde for 20 min and stained with 0.1% crystal violet for 15 min. Cells in five random fields (×200 magnification) were counted under a light microscope for quantification. For invasion assays, the upper surface of the membranes was pre-coated with Matrigel (354234, Corning, USA), and 1 × 10⁵ cells were seeded. Other settings were the same as migration assays.

2.7. Apoptosis assays

Cell apoptosis was quantified by flow cytometry using the FITC Annexin V Apoptosis Detection Kit I (556547, BD Biosciences, USA) according to the manufacturer's instructions. Briefly, 10 × 10⁵ cells were harvested, washed twice with cold PBS, and resuspended in 100 μL binding buffer. Cells were incubated with 5 μL Annexin V-FITC and 5 μL propidium iodide for 15 min at room temperature in the dark, followed by addition of 400 μL binding buffer. Samples were analyzed on a flow cytometer (CantoⅡ; BD Biosciences, USA), and data were processed with FlowJo software (V10) (BD Biosciences, USA). Percentage of Annexin V-positive cells was quantified as apoptotic cells.

2.8. Orthotopic GBM xenograft

Orthotopic GBM xenografts were established as previously described [23]. Briefly, 4-6 weeks old female BALB/c nude mice (purchased from SPF (Beijing) Biotechnology Co., Ltd.) were anesthetized with intraperitoneal pentobarbital sodium (50 mg/kg) and placed in a stereotactic apparatus (RWD Life Science, China). U251 or GBM1214 cells (5 × 10⁵ in 5 μL PBS) were slowly injected into the right striatum (coordinates: 2 mm lateral, 1 mm anterior to bregma, 3 mm depth) over 5 min using a Hamilton syringe. The needle was left in place for an additional 5 min before withdrawal to prevent reflux. Tumor growth at the indicated time points was monitored by bioluminescence imaging (IVIS Spectrum, PerkinElmer, USA) after intraperitoneal injection of D-luciferin (150 mg/kg) (LUCNA-1G, Goldbio, US). Mice were sacrificed upon reaching humane endpoints, and brains were harvested for hematoxylin-eosin (H&E) staining and immunohistochemistry. The observation period for survival analysis was 100 days. All animal protocols were approved by the Animal Research Ethics Committee of the University of Science and Technology of China (USTC) (approval No. 2024-N(A)-53), and all animal experiments were performed in accordance to the USTC guidelines for the use of laboratory animals.

2.9. Nuclear-cytoplasmic fractionation

Nuclear and cytoplasmic protein fractions were prepared using the Minute™ Cytoplasmic and Nuclear Extraction Kit for Cells (Invent Biotechnologies, SC-003) according to the manufacturer's instructions. The isolated nuclear and cytoplasmic fractions were subsequently subjected to immunoblotting.

2.10. Immunoblotting

Cell pellets were lysed in RIPA buffer (50 mM Tris-Cl pH 8.0, 150 mM NaCl, 5 mM EDTA, 1% NP-40, 0.1% SDS,1 mM Na3VO4, 10 mM NaF, 2 mM DTT, 1 mM PMSF, plus phosphatase/protease inhibitor cocktail (Roche, 11873580001). Equivalent amounts of proteins were resolved by SDS-PAGE and transferred to PVDF membranes (Millipore). After blocking with 5% skim milk for 1 h, membranes were incubated with primary antibodies overnight at 4 °C, followed by three TBST washes and 1-h incubation with HRP-linked secondary antibodies (1:10000) at ambient temperature. Immunoreactive bands were detected by ECL substrate and captured using a ChemiDoc Imaging System (BIORAD). The primary antibodies used for immunoblotting included hnRNP L (Abcam, ab264340, 1:1000 dilution), RGS16 (Abcam, 119424, 1:1000 dilution), β-actin (Proteintech, 66009-1-Ig, 1:5000 dilution).

2.11. RNA pulldown assays

A 5′-biotin labeled antisense oligonucleotide probe complementary to the back-splice junction of circNLK was used as the specific capture probe, whereas the corresponding biotin-labeled sense probe served as the negative control. The probes were synthesized by RiboBio (Guangzhou, China). Each probe (100 pmol) was incubated with cell lysates containing 200 μg of total protein at 4 °C for 50 min. The resultant RNA-protein complexes were captured using streptavidin-coated magnetic beads from Magnetic RNA-Protein Pull-Down Kit (20164, ThermoFisher), washed with wash buffer to remove non-specific bindings, followed by elution with biotin elution buffer. The products were subjected to SDS-PAGE and silver staining, followed by MS analysis of circNLK-binding proteins.

For pull-down assays using synthetic RNAs, wild-type and mutant RNAs corresponding to the full-length circNLK sequence, as well as RNA corresponding to the RGS16 3′UTR, were synthesized (RiboBio, Guangzhou, China). The circNLK-derived RNAs were synthesized as linear RNAs containing the complete circNLK sequence. Where indicated, the synthetic RNAs were biotin-labeled at the 3′end. No m⁶A modification was introduced into these synthetic RNAs.

For comparison of hnRNP L binding to wild-type and mutant circNLK-derived RNAs, equal amounts of 3′ biotin-labeled wild-type or mutant circNLK-derived RNA were incubated with U251 cell lysates. RNA-protein complexes were captured using streptavidin-coated magnetic beads, washed, and eluted, and the amount of associated hnRNP L was determined by western blotting.

For competition assays, a fixed amount of biotinylated circNLK (1 μg) was incubated with U251 whole-cell lysate (500 μg total protein; protein concentration determined by BCA assay) in the presence of increasing amounts of non-biotinylated RGS16 3′UTR (0, 0.5, and 2.0 μg). Alternatively, 1 μg biotinylated RGS16 3′UTR RNA was incubated with cell lysate together with escalating amounts of unlabeled circNLK (0, 0.5, and 2.0 μg). The RNA-protein complexes were captured using streptavidin-coated magnetic beads, washed with wash buffer, followed by elution. The amount of hnRNP L in pulldown product was detected by western blotting.

2.12. RNA immunoprecipitation (RIP) assays

RNA immunoprecipitation (RIP) was performed using the Magna RIP™ RNA-Binding Protein Immunoprecipitation Kit (Cat#17-700, Millipore) according to the manufacturer's protocol with slight modifications. Briefly, cell lysates were incubated with magnetic beads pre-conjugated to either anti-hnRNP L antibody (Abcam, ab264340) or normal rabbit IgG (Abcam, ab172730) at 4 °C for 8 h with gentle rotation. After incubation, the beads were thoroughly washed 5 times with ice-cold RIP wash buffer. Co-precipitated RNAs were then extracted using TRIzol reagent according to the manufacturer's instructions. The purified RNAs were reverse-transcribed, and the enrichment of circNLK was quantified by qPCR using specific divergent primers.

2.13. Analysis of m⁶A modification on circNLK

Potential m⁶A modification sites on circNLK were predicted using the SRAMP (http://www.cuilab.cn/sramp) and circMIR 2.0 (https://sourceforge.net/projects/circmir/). To assess m⁶A enrichment of endogenous circNLK, total RNA was extracted using TRIzol reagent and fragmented by sonication on ice using a Bioruptor Plus sonicator (Diagenode) for 10 cycles of 30 s on and 30 s off. Methylated RNA immunoprecipitation (MeRIP, Cat#17-700, Millipore) was performed according to the manufacturer's instruction. Fragmented RNAs were incubated with anti-m⁶A antibody conjugated magnetic beads or control beads at 4 °C for 2 h. After washing with wash buffer included in the kit, bound RNA was eluted in elution buffer, purified with TRIzol reagent, and subjected to qPCR for quantification of the enriched circNLK.

2.14. m⁶A dot blot assay

Equal amounts of total RNA were spotted onto nylon membranes, air-dried, and UV-crosslinked. m6A was detected using an anti-m6A antibody (Proteintech, 68055-1-Ig) and ECL. Methylene blue staining (Biosharp, BL1137A) was used to verify comparable RNA loading.

2.15. RNA sequencing (RNA-seq)

Total RNA from cells with circNLK overexpression or hnRNP L knockdown was extracted using TRIzol reagent (15596026CN, Invitrogen, USA) and subjected to RNA-seq on the Illumina NovaSeq 6000 platform (Novogene Co., Ltd., China). Gene expression levels were quantified as fragments per kilobase of transcript per million mapped reads (FPKM) and normalized using DESeq2. Differentially expressed genes were defined as those with |log₂ fold change| > 1.0 and adjusted p-value < 0.05. The interactions between hnRNP L and RGS16 transcript were predicted with the online tool catRAPID (http://service.tartaglialab.com/page/catrapid_group).

2.16. Immunohistochemistry (IHC) staining

Paraffin-embedded tumor sections were deparaffinized in xylene and rehydrated through graded ethanol solutions. Antigen retrieval was performed by heating sections in citrate buffer (pH 6.0) for 20 min. Samples were incubated in 3% hydrogen peroxide for 10 min and 5% bovine serum albumin (BSA) for 30 min at room temperature to block endogenous peroxidase activity. Sections were incubated with primary antibodies against RGS16 (1:200) (119424, Abcam) overnight at 4 °C, followed by incubation with HRP-conjugated secondary antibodies for 1 h at room temperature. Staining was visualized using 3,3′-diaminobenzidine (DAB) substrate, and sections were counterstained with hematoxylin (ZLI-9018, ZSGBbio, China). Images were captured using an Olympus BX53 microscope, and positive staining in five random fields was quantified for each sample using Image J software. IHC staining was assessed via H-score system, which accounts for both staining intensity (scored 0-3) and positive cell percentage, with samples categorized into low, moderate, or high expression based on H-score thresholds.

2.17. RNA fluorescence in situ hybridization (FISH) combined with immunofluorescence (IF)

Subcellular localization and co-localization of circNLK and hnRNP L were examined by sequential RNA FISH and protein immunofluorescence. Briefly, GBM cells grown on glass coverslips were fixed with 4% paraformaldehyde for 15 min and permeabilized with 0.5% Triton X-100 for 10 min. FISH was performed first using a biotin-labeled probe specifically targeting the back-splice junction of circNLK (RiboBio, Guangzhou, China) according to the manufacturer's instructions. After stringent washing, slides were blocked with 5% BSA for 1 h at room temperature. Immunofluorescence staining was then conducted directly on the same slides by incubating with primary antibody against hnRNP L (1:200, Abcam, UK) overnight at 4 °C, followed by Alexa Fluor-conjugated secondary antibody (Invitrogen, USA) for 1 h at room temperature in the dark. Nuclei were counterstained with DAPI (Sigma-Aldrich, USA). Images were acquired using a Zeiss LSM 880 confocal laser scanning microscope, and co-localization analysis was performed with Image J software (https://imagej.net/ij/download.html).

2.18. Statistical analysis

Statistical analyses were performed using GraphPad Prism version 10.1 (GraphPad Software, USA) and R software version 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria). Comparisons between two groups were assessed by Student's t-test, while one-way analysis of variance (ANOVA) followed by Tukey's post hoc test was applied for multiple-group comparisons. Results are presented as mean ± standard deviation (SD). Survival curves were estimated by the Kaplan-Meier method with the log-rank test. Cox proportional hazards regression models were used for multivariate survival analyses. All experiments had at least three biologically independent repeats.

3. Results

3.1. CircNLK is downregulated in GBM and positively correlated with prognosis

To systematically identify circRNAs dysregulated in glioblastoma (GBM), the publicly available circRNA profiling dataset (GSE86202) was used as an initial screening dataset to identify candidate circRNAs potentially involved in GBM. The profiling revealed a distinct expression pattern between tumor and matched non-tumor tissues (Fig. 1A). Differential expression analysis identified a total of 662 dysregulated circRNAs in GBM, including 21 upregulated circRNAs and 641 downregulated circRNAs (Fig. 1B), suggesting that global circRNA reduction is a prominent feature of GBM. To prioritize candidate circRNAs for further investigation, we applied a stepwise screening strategy rather than relying on a single ranking parameter. Markedly downregulated circRNAs were first selected using the criteria of fold change >4 and −log10 (adjusted P value) > 2. The top five downregulated candidates were then subjected to RT-qPCR validation in three paired normal brain and GBM samples. Among these candidates, hsa_circ_0002103 showed the most consistent reduction across all three paired samples and displayed the most robust differential expression pattern; therefore, it was selected for further investigation and hereafter termed circNLK (Supplementary Fig. S1A). According to circBase annotation, circNLK is generated from the NLK locus on chromosome 17q11.2 through back-splicing of exons 2-7 (Fig. 1C). To validate the circular structure of circNLK, divergent and convergent primers were designed based on the exon organization of the linear NLK transcript. Divergent primers specifically amplified circNLK from cDNA but not from genomic DNA (gDNA), whereas convergent primers amplified the linear NLK transcript from both cDNA and gDNA. In addition, Sanger sequencing of the divergent PCR product confirmed the expected back-splice junction of circNLK (Fig. 1C). Consistent with the fact that circular RNAs lack a poly(A) tail, RT-qPCR analysis showed that circNLK was readily detected in cDNA synthesized with random primers but was barely detected in oligo(dT)-primed cDNA, whereas linear NLK mRNA was detected at comparable levels in both preparations (Fig. 1D). Moreover, circNLK was resistant to RNase R digestion in both U251 and GBM cell 1214 cells, while linear NLK mRNA was markedly reduced after RNase R treatment (Fig. 1E). RNA FISH further showed that circNLK was predominantly localized in the nucleus of U251 cells (Fig. 1F). Together, these findings verified the circular nature and subcellular localization of circNLK.

Fig. 1.

circNLK is down-regulated in GBMs and positively correlated with prognosis. (A) Ring heatmap showing circRNA expression profiles in three GBM and three matched non-tumor tissues in the GEO dataset GSE86202. Normalized circRNA expression values were log2-transformed. Rows (circRNAs) and columns (samples) were hierarchically clustered in an unsupervised manner. (B) Volcano plot illustrating circRNAs with differential expression between GBM and non-tumor tissues. circNLK was identified as one of the most significantly downregulated circRNAs in GBM. (C) Schematic diagram of circNLK generated by back-splicing of exons 2-7 of the NLK gene, with validation by divergent/convergent PCR and Sanger sequencing. (D) RT-qPCR analysis of circNLK and linear NLK in cDNA synthesized with random primers or oligo(dT). (E) RT-qPCR analysis of circNLK and linear NLK after RNase R treatment in U251 and GBM1214 cells. (F) RNA FISH showing subcellular localization of circNLK. (G) Cox regression analysis of circNLK expression and prognosis in GBM patients. (H) RT-qPCR analysis of circNLK expression in GBM and non-neoplastic brain tissues. (I) qRT-PCR analysis of circNLK expression in normal human astrocytes (NHAs) and glioblastoma cell lines (GBM1214, U251, U87). Expression levels were normalized to GAPDH. (J, K) Kaplan-Meier curves showing progression free survival (J) and overall survival (K) in GBM patients stratified by circNLK expression. Data are presented as mean ± SD unless otherwise indicated. *P < 0.05, **P < 0.01, ***P < 0.001; ns, not significant.

We next examined the clinical relevance of circNLK in GBM. Cox regression analysis further supported a significant association between circNLK expression and patient prognosis (Fig. 1G). RT-qPCR analysis of clinical specimens showed that circNLK expression was significantly lower in GBM tissues than in non-tumor brain tissues (Fig. 1H). To further validate these findings in vitro, we evaluated circNLK expression in normal human astrocytes (NHAs) and glioblastoma cell lines. Consistent with the observation on tissue samples, circNLK was markedly downregulated in an in-house GBM line (GBM1214) and common GBM cell lines (U251 and U87) relative to NHAs (Fig. 1I). Based on the median circNLK expression level, 40 patients with GBM were divided into circNLK-high (n = 20) and circNLK-low (n = 20) groups. Kaplan-Meier survival analysis showed that patients with high circNLK expression had significantly longer progression-free survival (PFS) and overall survival (OS) than those with low circNLK expression (Fig. 1J and K). In line with these findings, no significant association was observed between circNLK expression and conventional clinicopathological characteristics, including age, sex, tumor size, KPS score, or tumor location (Table 1). Collectively, these results indicate that circNLK is frequently downregulated in GBM and may serve as a favorable prognostic biomarker, while also suggesting a potential tumor-suppressive role in GBM.

Table 1.

Correlations between circNLK expression and clinicopathologic features in 40 GBM patients.

Variable	Relative expression of circNLK (Number (% of total))
	High	Low	P-Value
Age year
<55	8 (20)	11 (27.5)	0.527
≥55	12 (30)	9 (22.5)
Gender
Male	12 (30)	9 (22.5)	0.527
Female	8 (20)	11 (27.5)
Tumour Size
<5	8 (20)	9 (22.5)	1.0
≥5	12 (30)	11 (27.5)
KPS score			0.363
<80	2 (8)	4 (16)
≥80	10 (40)	9 (36)
Location
Frontal	3 (7.5)	5 (12.5)	0.729
Temporal	5 (12.5)	2 (5)
Parietal	1 (2.5)	1 (2.5)
Insula	1 (2.5)	1 (2.5)
Central region	0 (0)	1 (2.5)
Multifoci	10 (25)	9 (22.5)
PFS
<1 year	6 (15)	15 (37.5)	0.005**
≥1year	14 (35)	5 (12.5)
OS
<1 year	7 (17.5)	15 (37.5)	0.012*
≥1year	13 (32.5)	5 (12.5)

Asterisks indicate statistical significance.

*P < 0.05.

**P < 0.01.

3.2. CircNLK suppresses GBM cell proliferation and motility and promotes apoptosis

Given that circNLK was markedly downregulated in GBM, we next investigated its biological functions in GBM cells by performing gain and loss of function experiments in U251 and GBM1214 cells. RT-qPCR analysis confirmed the successful overexpression of circNLK in U251 and GBM1214 cells transduced with the circNLK overexpression construct (Fig. 2A and I). Conversely, two independent shRNAs efficiently reduced circNLK expression in both cell lines (Fig. 2B and J). Because circNLK is derived from the NLK locus, we assessed whether circNLK manipulation affected endogenous linear NLK expression. Neither circNLK overexpression nor knockdown significantly changed linear NLK mRNA or NLK protein levels in U251 or GBM1214 cells, suggesting that the observed effects were not attributable to altered host gene expression (Supplementary Fig. S1B–E). We then assessed the effect of circNLK on GBM cell growth. CCK-8 assays showed that ectopic circNLK expression significantly suppressed the proliferative capacity of U251 and GBM1214 cells compared with the empty vector control (Fig. 2C and K). In contrast, circNLK knockdown increased cell growth in both cell lines, as evidenced by the higher absorbance values in sh-circNLK transduced cells than in sh-Ctrl cells (Fig. 2D and L). These data indicate that circNLK negatively regulates GBM cell proliferation. To determine whether circNLK affects cell survival, apoptosis was examined by flow cytometry using Annexin V-FITC/PI staining. Overexpression of circNLK markedly increased the proportion of apoptotic cells in both U251 and GBM1214 cells (Fig. 2E and M). In contrast, silencing circNLK reduced apoptosis in both cell lines (Fig. 2F and N), suggesting that circNLK promotes apoptotic cell death in GBM cells. We further evaluated the effect of circNLK on GBM cell motility using Transwell assays. In both U251 and GBM1214 cells, circNLK overexpression significantly impaired migratory and invasive capacities relative to the corresponding control cells (Fig. 2G and O). Conversely, circNLK knockdown enhanced both migration and invasion in the two GBM cell lines (Fig. 2H and P). Taken together, these results demonstrate that circNLK suppresses malignant phenotypes of GBM cells by inhibiting proliferation, migration, and invasion, while promoting apoptosis.

Fig. 2.

circNLK inhibits GBM cell proliferation and motility and promotes apoptosis. (A, I) RT-qPCR analysis confirming OE group in U251 (A) and GBM1214 (I) cells compared to EV group. (B, J) RT-qPCR analysis confirming circNLK knockdown in U251 (B) and GBM1214 (J) cells. (C, K) CCK-8 assays showing the effect of OE on proliferation of U251 (C) and GBM1214 (K) cells. P values of CCK-8 assays were determined by two-way ANOVA. (D, L) CCK-8 assays showing the effect of circNLK knockdown on proliferation of U251 (D) and GBM1214 (L) cells. P values of CCK-8 assays were determined by two-way ANOVA. (E, M) Flow cytometric analysis of apoptosis in U251 (E) and GBM1214 (M) cells with OE. (F, N) Flow cytometric analysis of apoptosis in U251 (F) and GBM1214 (N) cells after circNLK knockdown. Bar graphics represent 3 independent biological repeats.Unpaired Student's t-test; (G, O) Transwell migration and invasion assays in U251 (G) and GBM1214 (O) cells with circNLK overexpression. (H, P) Transwell migration and invasion assays in U251 (H) and GBM1214 (P) cells after circNLK knockdown. (G and H) Transwell assays of cell migration and invasion of GBM cells with OE. Bar graphics represent 3 independent biological repeats. Scale bar, 200 μm. OE: circNLK overexpression; EV: empty vector. Data are presented as the mean ± SD from three independent biological experiments. Statistical significance in (A, E, G, I, M, and O) was determined using a two-tailed unpaired Student's t-test. Statistical significance in (B, F, H, J, N, and P) was determined using one-way ANOVA followed by Tukey's multiple comparisons test. Statistical significance in (C, D, K, and L) was determined using two-way ANOVA followed by an appropriate multiple-comparisons test. OE, circNLK overexpression; EV, empty vector; sh-Ctrl, control shRNA. P < 0.05, P < 0.01, P < 0.001; ns, not significant.

3.3. circNLK interacts with hnRNP L and exhibits m⁶A enrichment

To identify downstream effectors of circNLK in GBM cells, we performed RNA pull-down assays using a biotin labeled antisense probe complementary to the circNLK back-splice junction, with the sense probe as a negative control, followed by silver staining in U251 and GBM1214 cells (Fig. 3A and B). The enriched proteins were subsequently analyzed by mass spectrometry (MS). MS analysis of the RNA pull-down products identified 313 and 310 circNLK-associated proteins in U251 and GBM1214 cells, respectively, among which 37 proteins were shared by the two cell lines (Fig. 3C). Among these common candidates, heterogeneous nuclear ribonucleoprotein L (hnRNP L), a well-characterized RNA-binding protein, was selected for further investigation. The identification of hnRNP L in the circNLK pull-down complex was supported by the mass spectrometric peptide spectrum (Fig. 3D). Moreover, RNA pull-down followed by immunoblotting confirmed that hnRNP L was specifically enriched by the circNLK antisense probe, but not by the sense control (Fig. 3E), validating the interaction between circNLK and hnRNP L. Consistently, RIP-qPCR using an anti-hnRNP L antibody showed significant enrichment of circNLK compared with the IgG control (Fig. 3F), further confirming the association between endogenous circNLK and hnRNP L in GBM cells. In addition, combined RNA FISH and immunofluorescence analysis revealed colocalization of circNLK with hnRNP L in U251 cells. Subcellular fractionation followed by immunoblotting further confirmed that hnRNP L was mainly detected in the nuclear fraction of GBM cell lines (Fig. 3G). These findings support the spatial relevance of the circNLK and hnRNP L interaction in the nucleus.

Fig. 3.

m⁶A modification on circNLK mediates its interaction with hnRNP L (A–B) Silver staining of proteins retrieved by RNA pulldown using biotin-labeled circNLK sense or antisense probes in U251 (A) and GBM1214 (B) cells. Input lysates and molecular-weight markers are shown. The boxed regions indicate gel areas selected for mass spectrometry analysis. The section containing protein bands of higher staining in products from antisense relative to sense probes (red box) was excised for LC-MS/MS analysis. (C) Venn diagram showing the circNLK associated proteins identified by mass spectrometry in U251 and GBM1214 cells. The lower bubble plot shows representative candidate RNA binding proteins ranked according to mass spectrometry derived statistical significance and Z-score. (D) Peptide spectrum of hnRNP L in mass spectrometry analysis of circNLK-binding proteins. (E) Immunoblot showing the interaction between circNLK and hnRNP L in RNA pulldown assays. RNA pulldown with the circNLK specific antisense probe but not the sense probe precipitated hnRNP L in U251 cells. (F) qPCR analysis of the products from RNA immunoprecipitation with anti-hnRNP L antibody or IgG for circNLK. A much higher level of circNLK was detected in RNA precipitated with anti-hnRNP L antibody relative to IgG. Paired Student's t-test. (G) Combined RNA FISH and immunofluorescence showing colocalization of circNLK and hnRNPL in U251 cells and Immunoblot analysis of hnRNPL in cytoplasmic and nuclear fractions from U251 and GBM1214 cells. (H) Potential m⁶A modification sites on circNLK as predicted by SRAMP and circMIR2.0. A541 and A641 were predicted to be the m⁶A sites (in red) on circNLK. (I) The A541 and A641 sites were individually mutated to generate the circNLK-A1 and circNLK-A2 mutants, respectively, or simultaneously mutated to generate the circNLK-DA mutant. Biotin-labeled linear RNA transcripts corresponding to the full-length WT and mutant circNLK sequences were used to pull down hnRNP L from U251 cell lysates. The RNA-protein complexes were captured with streptavidin-conjugated beads, followed by immunoblotting for hnRNP L. (J) Immunoblot showing the efficient knockdown of hnRNP L in U251 cells using short hairpin RNAs (sh-hnRNP L-1 and sh-hnRNP L-2). (K) MeRIP-qPCR analysis of circNLK enrichment after hnRNPL knockdown. Unpaired Student's t-test. (L) qPCR analysis of circNLK expression upon knockdown of hnRNP L in U251 and GBM1214 cells. Knockdown of hnRNP L showed negligible effects on circNLK levels. Unpaired Student's t-test. (M) Immunoblot analysis of hnRNP L expression upon OE in U251 and GBM1214 cells. Data are presented as mean ± SD unless otherwise indicated. ***P < 0.001; ns, not significant.

To further explore sequence determinants potentially involved in hnRNP L binding, sequence prediction identified two putative high-confidence m⁶A motifs on circNLK, located at A541 and A641 (Fig. 3H). To test whether these sites contribute to hnRNP L binding, linear RNA transcripts corresponding to the full-length WT circNLK sequence and mutants carrying substitutions at A541 or A641 were generated by in vitro transcription and subjected to RNA pull-down assays. Mutation of either site markedly reduced hnRNP L binding to circNLK (Fig. 3I), indicating that these adenosine residues and their associated local RNA sequence context are important for the circNLK-hnRNP L interaction.

We next examined the relationship between hnRNP L and the recovery of circNLK in the anti-m⁶A immunoprecipitated fraction. Efficient knockdown of hnRNP L was confirmed by immunoblotting (Fig. 3J). Notably, hnRNP L depletion significantly reduced the enrichment of circNLK in anti-m⁶A immunoprecipitates (Fig. 3K), which may reflect altered epitope accessibility or RNP conformation. However, silencing hnRNP L did not significantly alter total circNLK expression in either U251 or GBM1214 cells (Fig. 3L). Likewise, ectopic circNLK expression did not affect hnRNP L protein levels in either cell line (Fig. 3M). To examine whether hnRNP L broadly affects cellular m⁶A modification, we performed m⁶A dot blot analysis using equal amounts of total RNA isolated from control and hnRNP L knockdown cells. Comparable RNA loading was confirmed by methylene blue staining. hnRNP L knockdown did not cause a detectable change in global cellular m⁶A levels (Supplementary Fig. S1F). Collectively, these findings support an interaction between circNLK and hnRNP L and indicate that altering either component does not substantially affect the overall expression level of the other.

3.4. circNLK inhibits hnRNP L-mediated stabilization of RGS16 RNA transcript

Given that circNLK interacted with hnRNP L but did not affect its overall protein expression, we hypothesized that circNLK may act as a molecular decoy by sequestering hnRNP L and thereby impairing its ability to regulate downstream target transcripts. On this premise, the biological consequence of ectopic circNLK overexpression should partially phenocopy that of hnRNP L depletion. To test this possibility and precisely pinpoint their shared downstream effectors, we performed RNA sequencing on U251 cells with circNLK overexpression and hnRNP L knockdown. The circNLK overexpression RNA-seq was performed with three independent biological replicates per group (Fig. 4A and B). For the hnRNP L knockdown, three independent biological replicates were analyzed for the control group and for two independent hnRNP L targeting shRNA groups, sh-hnRNP L-1 and sh-hnRNP L-2 (Fig. 4C and D). Using the initial cutoff of |log2 fold change| > 1 and adjusted P < 0.05, cross-comparison of transcripts downregulated after circNLK overexpression and hnRNP L knockdown identified 125 overlapping downregulated transcripts. To further increase the stringency and reduce background transcriptomic noise, we applied an additional threshold of strong downregulation, defined as log2FC < −2.5, which narrowed the list to 14 high confidence common candidate genes (Fig. 4E), which may represent downstream targets jointly affected by circNLK and hnRNP L.

Fig. 4.

circNLK inhibits hnRNP L-mediated stabilization of RGS16 RNA transcript (A, B) Heatmap (A, 3 vs 3 biological replicates) and volcano plot (B) of RNA-seq results in U251 cells with circNLK overexpression. (C, D) Heatmap (C, control n = 3; sh-hnRNP L-1 n = 3; sh-hnRNP L-2 n = 3) and volcano plot (D) of RNA-seq results in U251 cells after hnRNPL knockdown. (E) Venn diagram showing 125 transcripts commonly downregulated after circNLK overexpression and hnRNP L knockdown using the initial differential expression criteria of |log2FC| > 1 and adjusted P < 0.05. An additional stringent cutoff of |log2FC| > 2.5 was subsequently applied, yielding 14 high confidence candidate genes for experimental validation. (F) RT-qPCR validation of common candidate genes in U251 and GBM1214 cells after OE. (G) RT-qPCR analysis of RGS16 mRNA expression after circNLK knockdown in U251 and GBM1214 cells. (H) Immunoblot showing RGS16 protein levels in U251 and GBM1214 cells after circNLK overexpression. CircNLK overexpression reduced RGS16 protein levels. (I, J) Immunoblots showing RGS16 protein levels in U251 (I) and GBM1214 (J) upon knockdown of hnRNP L. Knockdown of hnRNP L reduced RGS16 protein levels. (K) Computational prediction (catRAPID) of potential hnRNP L binding sites on the RGS16 transcript. Each orange dot represents a predicted RNA-protein contact site. Dense clusters of hnRNP L binding sites were predicted within the RGS16 3′UTR region. The RGS16 3′UTR region corresponds to nt 1000-2500 in this transcript. (L) Immunoblot of the products of RNA pulldown with the RGS16 3′UTR or control bait. The RGS16 3′UTR but not the control bait pulled-down hnRNP L proteins in U251 cells. (M) Competitive RNA pull-down assay using biotin labeled RGS16 3′UTR as the bait in U251 and GBM1214 cell lysates. Unlabeled circNLK was added at the indicated amounts (0.5 and 2.0 μg), followed by immunoblotting for hnRNP L in the pull-down and input fractions. (N) Reverse competitive RNA pull-down assay using biotin labeled circNLK as the bait in U251 and GBM1214 cell lysates. Unlabeled RGS16 3′UTR was added at the indicated amounts (0.5 and 2.0 μg), followed by immunoblotting for hnRNP L in the pull-down and input fractions. (O, Q) Actinomycin D chase assays showing decay of RGS16 mRNA after hnRNPL knockdown in U251 (O) and GBM1214 (Q) cells. (P, R) Actinomycin D chase assays showing decay of RGS16 mRNA after OE in U251 (P) and GBM1214 (R) cells. Data are presented as fold change relative to the mRNA expression level at the multiple time poin. Unpaired Student's t-test. Data are presented as mean ± SD unless otherwise indicated. *P < 0.05, **P < 0.01, ***P < 0.001; ns, not significant.

To further validate these candidates, we examined their expression by RT-qPCR in U251 and GBM1214 cells after circNLK overexpression. Several genes showed reduced expression, among which RGS16 exhibited the most robust and consistent decrease in both cell lines (Fig. 4F). Conversely, circNLK knockdown increased RGS16 mRNA expression in U251 and GBM1214 cells (Fig. 4G). At the protein level, immunoblot analysis confirmed that ectopic circNLK expression decreased RGS16 abundance in both cell lines (Fig. 4H). In addition, hnRNP L knockdown also significantly reduced RGS16 protein levels in U251 and GBM1214 cells (Fig. 4I and J). These findings indicate that RGS16 is positively associated with hnRNP L but negatively regulated by circNLK. We next investigated how hnRNP L regulates RGS16 and whether circNLK interferes with this process. It has been reported that hnRNP L binds to the 3′untranslated region (3′UTR) of target transcripts and regulates their stability [24]. Computational predictions with catRAPID revealed multiple putative hnRNP L binding sites across the RGS16 transcript, with a prominent cluster located within the 3′untranslated region (3′UTR) (Fig. 4K). Consistently, RNA pull-down using a biotin labeled RGS16 3′UTR bait efficiently precipitated hnRNP L (Fig. 4L), supporting the interaction between hnRNP L and the RGS16 3′UTR.

Because circNLK also binds hnRNP L, we asked whether circNLK competes with RGS16 for hnRNP L association. In competitive RNA pulldown assays, increasing amounts of unlabeled circNLK markedly reduced the amount of hnRNP L captured by the biotin-labeled RGS16 3′UTR (Fig. 4M). In the reverse assay using biotinylated circNLK, hnRNP L remained associated with circNLK even after addition of increasing amounts of RGS16 3′UTR (Fig. 4N). These results support partial competition between circNLK and the RGS16 3′UTR for association with hnRNP L, although the magnitude of competition differed between the two assay orientations. Since 3′UTR binding RNA binding proteins frequently regulate transcript stability [25], we next examined whether circNLK and hnRNP L affect the stability of RGS16 mRNA. Actinomycin D chase assays showed that silencing hnRNP L accelerated the decay of RGS16 transcripts in both U251 and GBM1214 cells (Fig. 4O and Q). Notably, circNLK overexpression produced a similar effect, significantly shortening the persistence of RGS16 mRNA in both cell lines (Fig. 4P and R). Taken together, these findings support a model in which circNLK partially limits the association of hnRNP L with the RGS16 3′UTR, thereby reducing hnRNP L mediated stabilization of RGS16 mRNA.

3.5. RGS16 promotes tumorigenic phenotypes of GBM cells

Given that RGS16 was identified as a downstream target negatively regulated by circNLK and positively associated with hnRNP L, we next investigated its clinical significance and biological function in GBM. RT-qPCR analysis of our clinical specimens showed that RGS16 expression was significantly higher in GBM tissues than in non-neoplastic brain tissues, (Fig. 5A). To assess its prognostic relevance, patients were stratified into RGS16-high and RGS16-low groups according to the median expression level. Kaplan-Meier analysis of our cohort revealed that high RGS16 expression was associated with significantly shorter overall survival (OS) (Fig. 5B). This adverse prognostic association was further validated in the TCGA and CGGA cohorts, in which patients with high RGS16 expression also exhibited markedly worse OS than those with low RGS16 expression (Fig. 5C and D). These data indicate that elevated RGS16 expression is associated with poor clinical outcome in GBM.

Fig. 5.

RGS16 promotes tumorigenic phenotypes of GBM cells (A) qPCR analysis of RGS16 expression in GBM and non-neoplastic brain tissues. GAPDH was the internal control. (n = 40 for GBM, n = 22 for non-neoplastic brain tissues, mean ± SD. ***P < 0.001; two-tailed unpaired Student's t-test). (B) Kaplan-Meier survival analysis of GBM patients with high or low RGS16 expression. The 40 GBM patients were stratified into high and low RGS16 expression groups using the median RGS16 level as the threshold. (C, D) Kaplan-Meier analyses of OS according to RGS16 expression in the TCGA (C; RGS16-high, n = 184; RGS16-low, n = 189) and CGGA (D; RGS16-high, n = 185; RGS16-low, n = 187) cohorts. Green line represents the RGS16 high expression group, while the orange line represents the RGS16 low expression group. (E, J) Immunoblot analysis showing the efficient knockdown of RGS16 in U251(E) and GBM1214 cells (J) using two independent shRNAs (sh-RGS16-1 and sh-RGS16-2). β-actin was used as the loading control. (F, K) CCK-8 assays evaluating GBM cells proliferation after RGS16 knockdown in U251(F) and GBM1214 cells (K). (n = 3 independent biological repeats; mean ± SD; **P < 0.01; ***P < 0.001; two-way ANOVA with Bonferroni correction). (G, L) Flow cytometric analysis of cell apoptosis with Annexin V/PI staining in U251(G) and GBM1214 cells (L) following RGS16 knockdown. Bar graphs represent 3 independent biological repeats. RGS16 knockdown increased apoptosis of U251 cells. (Unpaired Student's t-test). (H, I) Transwell assays of cell migration and invasion of U251 cells (H) and GBM1214 cells (I) with RGS16 knockdown. Representative fields of transmigrated cells were shown. Bar graphics represent 3 independent biological repeats. Scale bar, 200 μm. (n = 3 independent biological repeats; mean ± SD; ***P < 0.001; two-tailed unpaired Student's t-test). (M) Representative images of immunohistochemical staining of RGS16 and Ki67 in GBM tissues from patients with poor or good overall survival. The 40 GBM patients were stratified into poor or good OS groups using the median survival as the threshold. (N, O) Quantitative comparisons of RGS16 H-scores (N) and Ki67 positivity rates (O) between the poor and good OS groups positive cells in GBM patient samples. (P) Linear regression of correlation between RGS16 and MKI67 transcriptional levels in GBM patients in CGGA database. (n = 650 patient samples; R = 0.526; P < 0.0001; Pearson's correlation analysis). (Q) RGS16 expression and Ki67 staining showed a positive correlation. (n = 40 patient samples; R = 0.749; P < 0.0001; Pearson's correlation analysis).

We next examined the functional role of RGS16 in GBM cells using loss of function approaches. In U251 cells, efficient knockdown of RGS16 by two independent shRNAs was confirmed by immunoblotting (Fig. 5E). CCK-8 assays showed that RGS16 depletion significantly impaired cell proliferation (Fig. 5F). Flow cytometric analysis using Annexin V-FITC/PI staining further demonstrated that silencing RGS16 increased apoptosis (Fig. 5G). In addition, Transwell assays showed that RGS16 knockdown markedly suppressed both migration and invasion in U251 cells (Fig. 5H). Similar inhibitory effects on cell motility were also observed in GBM1214 cells following RGS16 silencing (Fig. 5I). Consistently, immunoblotting confirmed effective RGS16 knockdown in GBM1214 cells (Fig. 5J), and subsequent CCK-8 and flow cytometry analyses showed reduced proliferation and enhanced apoptosis in these cells (Fig. 5K and L). Together, these results demonstrate that RGS16 promotes malignant phenotypes of GBM cells, including proliferation, migration, and invasion, while suppressing apoptosis.

To further explore the clinical relevance of RGS16 at the protein level, we performed immunohistochemical staining in GBM specimens from patients with poor or good OS. Representative images showed that tumors from patients with poor OS displayed stronger RGS16 staining than those from patients with good OS (Fig. 5M). Quantitative analysis confirmed significantly higher H-scores of RGS16 in the poor OS group (Fig. 5N). In the same specimens, Ki67 staining was also more abundant in tumors from patients with poor OS (Fig. 5M and O), indicating increased proliferative activity. Consistent with these observations, correlation analysis of the CGGA dataset showed a positive association between RGS16 and MKI67 mRNA expression (Fig. 5P). Moreover, in our clinical cohort, the H-score of RGS16 was positively correlated with the Ki67 positive rate (Fig. 5Q). Collectively, these findings support that RGS16 is clinically associated with aggressive disease behavior and functionally contributes to the malignant progression of GBM.

3.6. circNLK inhibits GBM growth via repression of RGS16

Given that circNLK inhibited malignant phenotypes of GBM cells and negatively regulated RGS16 expression, we next asked whether RGS16 contributes the tumor suppressive effects of circNLK. Rescue experiments were performed by reintroducing RGS16 into circNLK-overexpressing (OE) cells. Successful ectopic Myc-RGS16 expression was confirmed in U251 cells by immunoblotting (Fig. 6A). In our cohort of 40 GBM specimens, circNLK expression was inversely correlated with RGS16 expression (Fig. 6J), supporting a functional association between these two molecules in clinical samples. CCK-8 assays showed that circNLK overexpression significantly inhibited cell proliferation in both U251 and GBM1214 cells, whereas reintroduction of RGS16 partially restored cell growth (Fig. 6B and K). Consistently, flow cytometric analysis demonstrated that OE group increased apoptosis, and this effect was attenuated by ectopic RGS16 expression in both cell lines (Fig. 6C and L). Moreover, Transwell assays showed that the inhibitory effects of circNLK on cell migration and invasion were also partially reversed by RGS16 re-expression (Fig. 6D and M). Together, these findings indicate that repression of RGS16 is functionally involved in the anti-tumor effects of circNLK in GBM cells.

Fig. 6.

circNLK inhibits GBM growth via repression of RGS16. (A) Immunoblot showing the ectopic expression of Myc-tagged RGS16 (Myc-RGS16) in U251 cells. (B, K) CCK-8 assays showing proliferation of U251 (B) and GBM1214 cells (K) with ectopic overexpression of circNLK alone or together with RGS16. P values of CCK-8 assays were determined by two-way ANOVA. (C, L) Flow cytometric analysis with Annexin V/PI staining of cell apoptosis showing apoptosis of U251 (C) and GBM1214 cells (L) with ectopic overexpression of circNLK alone or together with RGS16. Representative dot plots were shown. Bar graphics represent 3 independent biological repeats. The data are presented as the mean ± SD of three independent experiments; Unpaired Student's t-test; *P < 0.05; **P < 0.01; ***P < 0.001. (D, M) Transwell assays of cell migration and invasion of U251 (D) and GBM cells (M) with ectopic overexpression of circNLK alone or together with RGS16. Representative fields of transmigrated cells were shown. Bar graphics represent 3 independent biological repeats. Scale bar, 200 μm. (n = 3 independent biological repeats; mean ± SD; ns means no significance; ***P < 0.001; Unpaired Student's t-test). (E-F, N-O) Representative image (E) and quantification (F) of bioluminescence signaling from intracranial xenografts derived from U251 (E-F) and GBM1214 cells (N-O) overexpressing circNLK or control cells. (n = 6 mice per group; mean ± SD; ns means no significance; *P < 0.05; **P < 0.01; ***P < 0.001; Unpaired Student's t-test). (G, P) Kaplan-Meier survival curves of mice bearing tumors derived from circNLK-overexpressing or control U251 cells (G) and GBM1214 cells (P). (H, Q) qPCR analysis of circNLK expression in samples from U251 (H) and GBM1214 (Q) orthotopic tumor tissues. (paired Student's t-test; ***P < 0.001).(I, R) Representative images and quantification of immunohistochemical staining of RGS16 in xenografts derived from U251 (I) and GBM1214 cells (R) overexpressing circNLK or control cells. Scale bar, 50 μm. (n = 6 for each group; mean ± SD; ***P < 0.001; Unpaired Student's t-test). (J) Correlation between circNLK and RGS16 expression in U251 intracranial xenograft tissues.

We next evaluated the role of circNLK in vivo using intracranial xenograft models established with U251 or GBM1214 cells stably overexpressing circNLK. Serial bioluminescence imaging showed that circNLK overexpression markedly suppressed orthotopic tumor growth in both models (Fig. 6E, F, N and O). In line with this, mice bearing circNLK OE tumors survived significantly longer than control mice (Fig. 6G and P). RT-qPCR analysis of xenograft tissues confirmed sustained circNLK overexpression in the OE group (Fig. 6H and Q). Importantly, immunohistochemical staining showed that RGS16 expression was significantly reduced in OE xenografts, as reflected by lower H-scores in both U251 and GBM1214 derived tumors (Fig. 6I and R). Collectively, these results support the notion that circNLK-hnRNP L-RGS16 as one mechanism through which circNLK may restrain GBM progression.

4. Discussion

GBM remains to be a formidable malignancy with poor prognosis, which is largely due to the intrinsic aggressiveness of tumor cells. Several oncogenic circRNAs have been identified to promote gliomagenesis [26,27], yet a substantial subset of circRNAs are downregulated in tumor relative to normal brain tissue, highlighting the underestimated roles of circRNAs as tumor suppressors [28,29]. Our study showed that circNLK (hsa_circ_0002103) was markedly reduced in GBM cells, whereas higher circNLK expression correlated with prolonged progression free and overall survival of GBM patients. Functional assays revealed that circNLK overexpression inhibited GBM cell proliferation, migration, and invasion, while promoting apoptosis in vitro and suppressing tumor growth in vivo. Our data support a model in which circNLK directly interacted with the RNA-binding protein hnRNP L in an m⁶A sensitive manner, thereby competing with RGS16 RNA for hnRNP L binding (Fig. 7). This preferential sequestration impaired hnRNP L mediated stabilization of RGS16 mRNA, leading to decreased RGS16 protein levels. In this scenario, circNLK acts as a decoy for RNA-binding proteins, which expands the functional repertoire of circRNAs beyond classical miRNA sponging [30]. These discoveries deepen our understanding of circRNAs as tumor suppressors.

Fig. 7.

Schematic model of the circNLK-hnRNP L-RGS16 mRNA regulatory axis in glioblastoma. circNLK is generated through back-splicing of NLK pre-mRNA and contains m⁶A modifications that facilitate its interaction with hnRNP L. When circNLK expression is low, hnRNP L associates with the 3′UTR of RGS16 mRNA and enhances its stability, resulting in increased RGS16 protein expression and promotion of GBM progression. When circNLK expression is high, m⁶A-modified circNLK binds hnRNP L and partially limits its association with the RGS16 3′UTR, thereby reducing RGS16 mRNA stability and RGS16 protein expression and suppressing GBM progression.

Our study identified RGS16 as a functionally relevant downstream target of the circNLK-hnRNP L regulatory axis. Although RGS16 has been implicated in glioma progression in previous studies, those reports mainly focused on the oncogenic function of RGS16 itself [20,21], whereas the upstream post-transcriptional mechanism responsible for RGS16 dysregulation in GBM remained unclear. In the present study, we provide evidence that circNLK binds to hnRNP L and interferes with its regulatory effect on RGS16, which may contribute to reduced RGS16 expression and attenuation of malignant phenotypes. Therefore, our findings do not simply reproduce the known pro-tumor role of RGS16, but further extend current knowledge by placing RGS16 within a circRNA-RBP-mRNA regulatory network. Nevertheless, the precise downstream pathways by which RGS16 contributes to GBM progression in the context of the circNLK-hnRNP L-RGS16 axis require further investigation.

It should be emphasized that our data do not establish that hnRNP L recognizes RGS16 mRNA through m⁶A modification within the RGS16 3′UTR. Mutation of the two predicted m⁶A motifs on circNLK, A541 and A641, markedly reduced hnRNP L binding, supporting an important role for these sites in facilitating the circNLK-hnRNP L interaction. In line with our discovery, previous studies have shown that hnRNP family members, such as hnRNPA2B1 and hnRNPC, act as m⁶A readers to regulate RNA splicing, export, and stability, known as the “m⁶A-switch” mechanism [12,[31], [32], [33]]. Given that hnRNP L contains multiple RNA recognition motifs, m⁶A may similarly influence the local structure or accessibility of circNLK and thereby facilitate hnRNP L binding. However, because A541 and A641 were computationally predicted and nucleotide substitution itself may affect RNA sequence or secondary structure, our data do not establish that these residues are directly methylated or that hnRNP L functions as a direct m⁶A reader. Instead, our findings support a model in which m⁶A-associated circNLK interacts with hnRNP L and interferes with hnRNP L mediated stabilization of RGS16 mRNA, while the precise molecular basis of this interaction requires further investigation.

The downregulation of circNLK in GBM tissues and the positive association between circNLK levels and patient prognosis underscores the clinical implications of circNLK. circNLK may therefore serve as a prognostic biomarker and potentially as a therapeutic target. However, one limitation of this study is that the initial circRNA profiling dataset used for candidate screening contained a small number of patient samples, larger multicenter cohorts are still needed to confirm the clinical value of circNLK as a prognostic biomarker in GBM. Given its ability to suppress tumor growth by destabilizing RGS16, restoring circNLK expression by circRNA mimetics, especially with nanoparticle-mediated delivery, may represent promising therapeutic avenues for GBM [34,35]. Moreover, our discovery of the circNLK-hnRNP L-RGS16 axis may prompt future identifications of circRNA-mediated decoy mechanisms regulating RNA stability in cancers, which may provide more targets for treatment of different tumors.

Emerging evidence highlights that circRNAs can orchestrate gene expression and cellular homeostasis through various alternative mechanisms. For instance, specific nuclear localized circRNAs can act as transcriptional regulators, modulating gene expression in cis by interacting with the RNA polymerase II machinery [36], or in trans by forming regulatory networks under stress conditions [37]. Furthermore, circRNAs have been critically implicated in the maintenance of genome stability, either by directly modulating DNA damage repair pathways [38] or through the regulation of complex ciR-loops to prevent replication-transcription conflicts [39]. Importantly, defying their historical classification as strictly non-coding transcripts, recent landmark studies have demonstrated that circRNAs harbor robust translational potential. They can utilize specific structured RNA elements or internal ribosome entry sites (IRES) to encode functional micropeptides that govern critical biological processes, including tumor progression and lifespan extension [40,41]. Although our current data strongly support an RBP regulatory mechanism for circNLK, these diverse functional possibilities warrant careful consideration in future studies to fully delineate the complex regulatory network of circNLK in glioblastoma.

5. Conclusions

This study identifies that circNLK as a downregulated circRNA in GBM and shows that higher circNLK expression is associated with a more favorable prognosis. Functionally, circNLK attenuates GBM progression both in vitro and in vivo. Mechanistically, our findings support a model in which circNLK interacts with hnRNP L in an m⁶A sensitive manner and reduces RGS16 expression, at least in part by interfering with hnRNP L mediated stabilization of the RGS16 transcript. These findings reveal a previously unrecognized circNLK-hnRNP L-RGS16 regulatory axis in GBM and suggest that circNLK may represent a candidate prognostic biomarker and a potential therapeutic target. Further studies are warranted to validate this mechanism in larger clinical cohorts and to define additional downstream pathways regulated by circNLK.

CRediT authorship contribution statement

Xue Kong: Writing – review & editing, Writing – original draft, Funding acquisition, Formal analysis, Data curation, Conceptualization. Yuan Li: Resources, Methodology. Wei Wang: Methodology, Formal analysis. Jiaxin Tang: Visualization, Methodology. Minghui Zeng: Resources, Data curation. Fangzhen Li: Visualization. Jing Xie: Resources, Methodology. Yu Xia: Visualization, Validation, Data curation. Wenchao Zhou: Writing – review & editing, Supervision. Haibo Wu: Writing – review & editing, Supervision.

Availability of supporting data

Raw RNA-sequencing data generated in this study have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession PRJNA1400117. The corresponding Gene Expression Omnibus (GEO) accession number will be released upon acceptance of the manuscript. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD081310.

Ethics approval and consent to participate

This research was approved by the Medical Research Ethics Committee of the First Affiliated Hospital of University of Science and Technology of China (Approval No.2024KY-293) for the use of human specimens, and by the Animal Research Ethics Committee of the University of Science and Technology of China (Approval No. 2024-N(A)-53) for all animal experiments. Informed consent was obtained from all subjects.

Funding

This research was supported by the USTC Research Funds of the Double First-Class Initiative (No. YD9110002063), Natural Science Foundation of Anhui Province (No. 2308085MH272), Scientific Research Project of Anhui Provincial Education Department (No. 2023AH040404), and Anhui Provincial Health Research Program (No. 2024Aa30451).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

figs1.

(A) RT-qPCR analysis of five candidate circRNAs in three paired GBM tissues (T1-T3) and corresponding peritumoral brain tissues (N1-N3). RNA expression was normalized to GAPDH. (B) RT-qPCR analysis of linear NLK mRNA expression in U251 and GBM1214 cells transfected with empty vector (EV) or the circNLK overexpression construct (OE). (C) RT-qPCR analysis of linear NLK mRNA expression in U251 and GBM1214 cells transduced with control shRNA (shCtrl) or two independent shRNAs targeting the back-splice junction of circNLK (sh-circNLK#1 and sh-circNLK#2). (D, E) Immunoblot analysis of NLK protein expression following circNLK overexpression (D) or circNLK knockdown (E) in U251 and GBM1214 cells. β-actin was used as the loading control. (F) m6A dot blot analysis of serially diluted total RNA isolated from U251and GBM1214 cells transfected with control siRNA (sh-Ctrl) or two independent siRNAs targeting hnRNP L (sihnRNP L-1 and sihnRNP L-2). Methylene blue staining was used to confirm comparable RNA loading. OE: circNLK overexpression; EV: empty vector. Data are presented as the mean ± SD. ns, not significant; ***P < 0.001.