NSUN2-mediated m5C hypermethylation of hsa_circ_0004516 promotes breast cancer brain metastasis by activating AKT signaling

Abstract

Breast cancer brain metastasis (BCBM) remains fatal with elusive mechanisms. Here, we unveil the first circRNA m5C methylation landscape in BCBM through MeRIP-seq (methylated RNA immunoprecipitation next-generation sequencing) identifying 7465 BCBM-specific m5C peaks versus 5929 in primary breast cancer (BC). A total of 48 hypermethylated and 128 hypomethylated m5C sites in BCBM (231-BR) were identified compared to BC. Bioinformatics enrichment analysis revealed hypermethylated circRNAs enriched in ERBB/VEGF signaling pathways. Among 8 validated differentially methylated circRNAs, hsa_circ_0004516 was consistently upregulated in BCBM tissues/cells and exhibited NSUN2-dependent m5C modification. Mechanistically, NSUN2-mediated m5C methylation enhanced hsa_circ_0004516 stability, evidenced by significantly shortened half-life upon NSUN2 depletion. Crucially, catalytic mutant NSUN2 (C271A/C321A) abolished this effect. Functional assays demonstrated that hsa_circ_0004516 knockdown in 231-BR cells suppressed proliferation, migration, and invasion by reducing p-AKT (Ser473) levels. The AKT activator SC79 reversed these phenotypic impairments, definitively linking hsa_circ_0004516-driven metastasis to AKT signaling activation. Our study establishes the NSUN2-m5C-hsa_circ_0004516-AKT axis as a novel therapeutic target and biomarker for BCBM.

1. Introduction

Brain metastasis (BM) occurs in approximately 30 % of patients with metastatic breast cancer (BC) [1,2] and represents a late-stage complication. With a median survival of only 15 months post-detection, breast cancer brain metastasis (BCBM) exhibits a poor prognosis, yet its underlying biological mechanisms remain largely undefined [3]. Deciphering the cellular and molecular mechanisms underlying BCBM is critical for developing innovative diagnostic biomarkers and therapeutic targets.

Circular RNAs (circRNAs), characterized by their covalently closed structure, exhibit high stability in body fluids, making them promising biomarkers for cancer screening via liquid biopsy [4,5]. Their roles in tumorigenesis, including acting as miRNA sponges and modulating protein interactomes, are increasingly recognized [[6], [7], [8], [9], [10]]. Recent studies have unveiled the translation potential of certain circRNAs that were previously considered non-translatable, particularly following RNA methylation [[11], [12], [13]]. Current studies predominantly focus on their regulation of downstream signaling pathways, while the upstream molecular events governing circRNAs in BCBM remain poorly elucidated.

RNA methylation, a reversible gene expression regulator, includes well-studied modifications such as N6-methyladenosine (m6A), C5-methylcytidine (m5C), N7-methylguanosine (m7G), and N1-methyladenosine (m1A) [4,[14], [15], [16]]. m5C methylation, occurring on the fifth carbon atom of cytosine in RNA, has been linked to several cancers, including bladder cancer, non-small-cell lung cancer, and glioma [[17], [18], [19]]. However, the role of m5C modification in non-coding RNAs, particularly circRNAs, is not well understood [[20], [21], [22]]. Consequently, the m5C methylation profile of circRNAs in BCBM remains largely unexplored.

In this study, we delineated the m5C methylation profile of circRNAs in BCBM using MeRIP-seq. MeRIP-qPCR and RT-qPCR validated m5C methylation levels and expression of candidate circRNAs, respectively. Functional enrichment analysis revealed that differentially m5C-methylated genes were significantly enriched in ERBB/VEGF signaling pathways. We also predicted the correlation between NSUN2 and candidate circRNAs in BC tissues. Our data revealed that hsa_circ_0004516 was upregulated in BCBM and underwent NSUN2-mediated m5C modification, which increased its stability. Furthermore, knockdown of hsa_circ_0004516 significantly reduced p-AKT protein levels and inhibited proliferation and migration via the AKT signaling pathway. This study identified a novel circRNA m5C methylation profile associated with BCBM, providing novel insights into the disease's pathogenesis.

2. Materials and methods

2.1. Cell culture

Human brain-targeting breast carcinoma cell line 231-BR and its parental cell line MDA-MB-231 (RRID: CVCL_0062) were cultured following established protocols [23]. Specifically, cells were maintained in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10 % fetal bovine serum (FBS) under a humidified atmosphere containing 5 % CO₂ at 37 °C. All cell lines were authenticated by STR profiling and confirmed to be mycoplasma-free.

2.2. Cell transfection

Stable cell lines were generated as follows: NSUN2 knockdown (shNSUN2) and control (shNC) lines, along with NSUN2 overexpression (oe-NSUN2) and control (oe-NC) lines, were constructed using GV493/GV492 lentiviral vectors and selected with puromycin. For these procedures, 3 × 10⁵, 231-BR cells were seeded in six-well plates and infected with the virus during the log phase. Infection complexes were formed by mixing 500 μL serum-free DMEM with 40 μL Reagent P and 500 μL serum-free DMEM with 1 μL virus, followed by 5-min incubation at room temperature. This reagent mixture was incubated for an additional 20 min, added to each well, and then incubated at 37 °C for 16 h. Transfected cells were selected in DMEM medium containing 1 μg/mL Puromycin Dihydrochloride (ST551-10 mg, Beyotime, China) for 48 h to eliminate non-transfected populations. Knockdown cells were continuously treated with 10 μM SC79 (a specific AKT activator, 05834-79-1, MCE) throughout the experimental period.

Additionally, sequence-specific siRNAs targeting circRNA, along with plasmids, miRNA controls, and mimics, were transfected using LipoBooster 3000 (40801E, Yeasen Biotechnology).

2.3. Tissue collection

BC tissues and adjacent normal tissues (ANTs) were obtained from the same patients undergoing radical BC resection at Liaocheng People's Hospital. Samples were rinsed with 4 °C physiological saline, quickly frozen in liquid nitrogen, and stored at −80 °C for analysis. This study received approval from the Ethics Committee of Liaocheng People's Hospital (the ethical approval number: 2021158), and written informed consent was obtained from patients prior to participation.

2.4. MeRIP library construction and sequencing

The m5C RNA-seq profiling was conducted by Cloud-Seq Biotech (Shanghai, China). RNA immunoprecipitation (IP) was performed using the GenSeq™ m5C RNA IP Kit (GenSeq Inc., China) under standardized conditions. Subsequently, extracted and purified RNA underwent library construction using an RNA library construction kit (NEB). Libraries were then sequenced on an Illumina HiSeq platform, generating paired-end 150 bp reads. Following quality control, we performed 3′-adaptor trimming and removed low-quality reads using Cutadapt (v1.9.3) [24] with parameters cutadapt -m20 -q15 to retain high-confidence sequencing data.

Processed reads were mapped to the GRCh37/hg19 reference genome using Bowtie, enabling systematic circRNA peak detection through the MeRan compare module. Methylation sites within circRNAs were identified using MACS. Differentially methylated regions were identified using DifReps software [25,26] with the following parameters: sliding window analysis (-window 200 -step 20), statistical rigor (-meth gt), library calibration (-frag 200), background scaling (-bkg 2.0), and streamlined processing (-noanno -nohs).

Methylation peaks with fold change >2 (hypermethylated) or <0.5 (hypomethylated) in BCBM versus BC samples, and P-value <0.05 (by a negative binomial distribution model similar to DESeq2, accounting for library size bias and dispersion), were defined as differentially methylated. This model accounted for library size bias, dispersion, and technical variations among biological replicates.

2.5. m5C MeRIP-qPCR

One hundred micrograms of RNA from MDA-MB-231 and 231-BR cells was fragmented into 100 bp fragments using the GenSeq™ m5C RNA IP Kit (GS-ET-003, GenSeq Inc., China). Briefly, protein A/G magnetic beads were pre-washed and incubated with 2 μL anti-m5C antibody (GS-ET-003, GenSeq Inc., China) for 1 h at 37 °C with rotation. After three washes, the antibody-conjugated beads were mixed with fragmented RNA. The beads-RNA complex was purified using magnetic beads. m5C-enriched circRNAs were dissolved in 11 μL RNase-free water. m5C-methylated circRNAs were quantified via qPCR using target-specific primers (Supplementary Table S1) with TB Green Premix Ex Taq II (RR820A, Takara, China) on a 7500 Fast Real-Time PCR System. Enrichment levels were normalized to input controls using β-actin as the reference gene.

2.6. Total RNA sequencing for circRNA input

RNA-seq was performed by Cloud-Seq Biotech (Shanghai, China). Ribosomal RNA was depleted using the GenSeq® rRNA Removal Kit. Libraries were prepared with the GenSeq® Low Input RNA Library Prep Kit, quality-controlled via Agilent 2100 BioAnalyzer, and sequenced on an Illumina NovaSeq 6000 (150 bp PE) for transcriptome profiling.

2.7. Alignment and circRNA identification

Illumina NovaSeq 6000-generated paired-end reads (150 bp) underwent Q30 quality control, followed by cutadapt (v1.9.3) processing for 3′-adapter trimming and low-quality read removal. Reads aligned by STAR (v2.5.1b) enabled circRNA detection via DCC (v0.4.4) [27,28], with differential expression analysis conducted using EdgeR (v3.16.5). Genome-wide read distribution was visualized via IGV [29].

2.8. Bioinformatics analysis

Gene Ontology (GO) analysis was used to elucidate the functional roles of differentially m5C methylation circRNAs in terms of molecular functions, biological processes, and cellular components. The Kyoto Encyclopedia of Genes and Genomes (KEGG) database was utilized to identify enriched biological pathways (P < 0.05).

To explore enriched terms, ontology clusters, and Protein-Protein Interaction (PPI) networks associated with the differentially expressed circRNAs, we utilized the Metascape (https://metascape.org) with a confidence threshold score ≥0.4. This integrated physical interactions (e.g., BioGRID) and functional associations (e.g., STRING). PPI networks were visualized using Cytoscape (v3.9.1).

Integrated prediction of circRNA-miRNA-mRNA interaction networks was performed using the circAtlas 3.0 and miRDB database.

2.9. Quantitative real-time PCR (RT-qPCR)

RNA extraction from tissue or cell samples was performed using the TRIZOL-based method [30]. cDNA synthesis used BeyoRT™ III cDNA First-strand Synthesis Kit (Beyotime, China) following the manufacturer's instructions. RT-qPCR was performed using TB Green Premix Ex Taq II kit (RR820A, Takara, China). β-actin served as the internal reference gene. Primers targeting circRNA backsplicing junctions were used (Supplementary Table S2).

2.10. Immunohistochemistry (IHC) and hematoxylin-eosin (H&E) staining

For IHC, paraffin-embedded tissues were processed as described [31]. Staining was performed using the BOND-MAX-M495343 system (Leica, Germany). Slides were baked and dewaxed at 62 °C for 2 h, then treated with Peroxide Block for 5 min at 37 °C. Samples were incubated with NSUN2 primary antibody (1:3000 dilution) for 59 min, followed by sequential application of Post Primary reagent (8 min) and Polymer reagent (8 min). Visualization used Mixed DAB Define for 8 min, and sections were counterstained with hematoxylin for 8 min.

For H&E staining, sections were incubated in hematoxylin at 37 °C for 3 min, rinsed with water for 4 min, then stained with eosin for 3 min. After dehydration, sections were mounted with neutral balsam. Stained sections were imaged under a light microscope (Vectra).

2.11. Western blot

Cells were treated with RIPA buffer (Beyotime, China) containing PMSF (Beyotime, China). The BCA protein concentration assay kit (P0012, Beyotime, China) was used to detect protein concentrations. Thirty micrograms of protein were separated on an 8 % SDS-polyacrylamide gel at constant voltage of 100 mV for 90 min. Proteins were transferred to PVDF membranes (Millipore, USA) at constant current of 300 mA for 90 min. Membranes were blocked with 5 % milk for 1 h at room temperature. Antibodies against NSUN2, Phospho-AKT(Ser473), AKT, GAPDH, and β-actin (Proteintech) were incubated overnight at 4 °C. PVDF membranes (Millipore, USA) were incubated with anti-rabbit (A0208, Beyotime, China) or anti-mouse (A0216, Beyotime, China) secondary antibodies at room temperature for 1 h.

2.12. RNA decay assay

Cells were seeded in 6-well plates (3 × 10⁴ cells/well) and cultured for 12 h. The baseline control (t = 0) was harvested by 0.1 % trypsinization (2 min, 37 °C) or scraping. Cells were centrifuged (304×g, 3 min, 37 °C; Eppendorf Centrifuge 5810R, Rotor S-4-104), resuspended in 1 ml TRI Reagent, and stored at −80 °C. Remaining wells were treated with 10 μg/mL Actinomycin D. Cells were harvested at 3 h and 6 h post-treatment using identical methods. Total RNA was extracted for RT-qPCR. Ct values at each time point were normalized to the average Ct value at t = 0 to calculate ΔCt, where ΔCt = (Average Ct of each time point – Average Ct of t = 0). Data were visualized with nonlinear regression curves generated in GraphPad Prism 9.0.

2.13. RNA immunoprecipitation quantitative PCR (RIP-qPCR)

RIP-qPCR was performed using an RNA Immunoprecipitation Kit (Bersin Bio™, China). Briefly, 2 × 10⁷ cells were lysed in 1.7 mL polysome lysis buffer containing protease and RNase inhibitors, vortexed, and incubated on ice for 20 min. Genomic DNA was removed by DNase I treatment (37 °C, 10 min), terminated with EDTA, EGTA, and DTT. After centrifugation (12,000×g, 15 min, 4 °C), the supernatant was collected. 40 μL of pre-equilibrated Protein A/G beads were used. Lysate was divided: 0.8 mL for the IP group, 0.8 mL for the IgG control group, and 0.1 mL as input. Samples were incubated with target antibody or IgG overnight at 4 °C. Equilibrated beads were added and incubated for 1 h. Beads were washed three times with washing buffer 1 and twice with buffer 2 (each containing DTT), then eluted in elution buffer. RNA was extracted using Trizol, precipitated, and reverse-transcribed. qPCR used SYBR Green Mix. β-actin served as the internal control for circRNA normalization. Relative enrichment was calculated using the 2^−ΔΔCt method.

2.14. Agarose gel electrophoresis experiment

cDNA products were subjected to electrophoresis in a 3 % agarose/TAE gel at constant voltage of 100 V for 25 min. Bands were visualized using a Gel Doc™ XR + Imaging System.

2.15. Cell proliferation and viability analyses

Cells (8000/well) were seeded in 96-well plates under standard conditions (37 °C, 5 % CO₂). Cell Counting Kit-8 (CCK-8) reagent (10 μL) was added at 0, 12, 24, and 48 h, followed by incubation for 2 h, and absorbance was measured at 450 nm.

2.16. Invasion and migration assay

For migration assay: A suspension of 20,000 cells in 200 μL serum-free DMEM was seeded in Transwell upper chambers, the lower chambers contained 500 μL complete medium. After 24 h incubation, migrated cells were stained and quantified. For transwell migration assays, 20,000 cells suspended in 200 μL serum-free DMEM were plated in the upper chamber, with the lower chamber filled with 500 μL complete medium. Cells were incubated in 0.5 % crystal violet solution for 15 min to fix and stain nuclei.

2.17. Colony formation assay

Seed 1500 cells in 6-well plates with medium replacement every two days. After 10 days incubation, stain colonies with 0.5 % crystal violet for 13 min and image under bright-field microscopy. Colonies ≥1 mm diameter were counted using ImageJ (v1.46).

2.18. Luciferase reporter assay

Sequences containing wild-type (WT) or mutant (MUT) miR-1301-3p binding sites from hsa_circ_0004516 and the AKT1 3′ untranslated region (3′UTR) were cloned into the dual-luciferase reporter vector GV272 (Genechem), generating hsa_circ_0004516-WT/MUT and AKT1-WT/MUT. Constructs were co-transfected with miR-1301-3p mimics into 231-BR cells. Luciferase activities were measured using the dual-luciferase reporter assay system (Promega) and normalized to Renilla luciferase activity.

2.19. Statistical analysis

We evaluated the distribution of quantitative data via the Shapiro-Wilk normality test. For normally distributed data, two-group comparisons were analyzed with unpaired or paired t-tests, while multiple-group pairwise comparisons were examined using one-way ANOVA coupled with Tukey's post-hoc test. For non-normally distributed data, the Mann-Whitney U test (for two groups) or Kruskal-Wallis H test (for multiple groups) was employed, with Dunn's test applied for subsequent pairwise comparisons. Linear associations between variables were evaluated using Pearson correlation analysis with 95 % confidence intervals. Biological replicates (n) represent independent experimental repeats or sample size: n = 3 for all cell-based assays; tissue sample sizes were cohort-dependent (e.g., clinical specimens: n = 24). Statistical significance was defined as ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗P < 0.0001. Analyses were performed using SPSS 28.0, and data visualized with GraphPad Prism 9.0.

3. Results

3.1. General features of circRNA m5C methylation in BC and BCBM

Utilizing MeRIP-seq, we investigated circRNA m5C methylation landscape in the BC and BCBM groups. The BC group exhibited 6625 circRNA m5C methylation peaks compared to 8161 in BCBM, with 696 overlapping peaks. BC-specific (5,929) and BCBM-specific (7,465) peaks showed distinct methylation patterns. Gene annotation revealed 1851 BC-associated and 1986 BCBM-associated genes, sharing 1527 common targets (Fig. 1A). Genomic mapping of m5C methylation peaks demonstrated predominantly exon localization of circRNA host genes in both BC and BCBM. Furthermore, the m5C peaks in BCBM showed significant differences from those in BC, mainly in intronic regions (BCBM: 5.3 %, BC: 0.228 %). Intronic circRNAs are formed by intron lassoing and are in the cell nucleus, regulating transcription (Fig. 1B). Statistical analysis of circRNA methylation peaks revealed predominant single-peak m5C modifications across individual circRNAs in both BC and BCBM (Fig. 1C). Comparative analysis of m5C methylation patterns revealed marked divergence between BC and BCBM samples in chromosomal distribution characteristics. Chromosomes 1–3 demonstrated significantly higher m5C peak density compared to other genomic regions, while chromosomes 22 and X exhibited the lowest methylation accumulation. This differential methylation landscape suggests potential epigenetic regulatory variations associated with metastatic progression (Fig. 1D). Heatmap profiling revealed distinct m5C epigenetic signatures between BCBM and BC cohorts (Fig. 1E). m5C profiling identified 48 hypermethylated and 128 hypomethylated sites in BCBM versus BC (Fig. 1F). Fig. 1G illustrated five characteristic m5C methylation motifs, highlighting differential circRNA epigenetic modification patterns in BCBM versus BC. These differences in methylation profiles are closely associated with BCBM.

Fig. 1.

Characteristics of circRNA m5C methylation in the BC and BCBM groups.

A: Venn diagram of m5C methylation peaks identified in circRNAs and m5C targeted genes. B: Genomic origin of the m5C methylation peaks. C: Number of m5C methylation peaks on each circRNA. D: Visualization of m5C methylation sites at the chromosome level. E: Hierarchical cluster analysis of m5C-methylated circRNAs (BCBM and BC). F: Volcano plot of the differentially m5C-methylated circRNAs (BCBM vs. BC). G: The top five circRNA m5C methylation motifs (BCBM and BC).

3.2. Verification of m5C methylation levels of the candidate circRNAs by MeRIP-qPCR

Table 1 summarized the ten differentially methylated m5C peaks between BCBM and BC. To validate the MeRIP-seq data, we performed MeRIP-qPCR on three hypermethylated (hsa_circ_0118701/0111796/0004516) and five hypomethylated circRNAs (hsa_circ_0023216/0019225/0005347/0068135/0124450) in BCBM versus BC. Consistent with sequencing data, hypermethylated circRNAs showed significantly elevated m5C levels, while hypomethylated circRNAs exhibited reduced methylation (P < 0.05; Fig. 2A). Log10 fold-changes confirmed high concordance between MeRIP-seq and MeRIP-qPCR (Fig. 2B), with strong correlation (r = 0.855, P = 0.007; Fig. 2C).

Table 1.

Top 10 circRNA m5C hypermethylated and hypomethylated peaks.

circBaseID	chrom	txStart	txEnd	Foldchange	P_value	Regulation
hsa_circ_0122355	chr3	148872961	148873340	624.3	2.7E-09	up
hsa_circ_0003235	chr2	25458575	25458580	331.9	2.9E-09	up
hsa_circ_0140329	chrX	46485161	46485400	275.1	2.4E-09	up
hsa_circ_0054322	chr2	43805661	43805746	26.3	1.3E-08	up
hsa_circ_0004516	chr12	110922882	110923000	18.6	5.7E-08	up
hsa_circ_0008647	chr5	145205560	145205620	17.9	2.5E-08	up
hsa_circ_0118701	chr2	203620260	203620404	14.6	5.0E-08	up
hsa_circ_0002558	chr19	58797481	58797571	13.0	7.9E-10	up
hsa_circ_0111796	chr1	208390141	208390340	12.1	4.5E-10	up
hsa_circ_0069249	chr4	17510893	17510986	11.5	4.0E-08	up
hsa_circ_0070049	chr4	77662581	77663079	1116.9	1.1E-08	down
hsa_circ_0001242	chr22	42946881	42947220	1022.7	5.1E-09	down
hsa_circ_0075357	chr5	179956274	179956388	741.4	9.8E-09	down
hsa_circ_0005661	chr1	6021853	6021940	547.1	4.9E-09	down
hsa_circ_0005133	chr6	30856464	30856520	482.7	4.9E-09	down
hsa_circ_0019225	chr10	96005821	96006060	390.7	4.9E-09	down
hsa_circ_0023216	chr11	68193445	68193640	64.9	5.3E-08	down
hsa_circ_0005347	chr17	65888001	65888150	20.8	1.1E-07	down
hsa_circ_0068135	chr3	180685841	180686042	18.2	2.0E-09	down
hsa_circ_0124450	chr3	64132581	64132880	14.7	8.9E-08	down

Fig. 2.

Validation of m5C methylation levels of the candidate circRNAs.

A: Relative m5C methylation levels of the hypermethylated and hypomethylated circRNAs (BCBM vs. BC) measured by MeRIP-qPCR. n = 3. B: Comparison of the mean fold changes (log10 transformed) between the MeRIP-seq data and MeRIP-qPCR results. C: Correlation of the mean fold changes (log10 transformed) between MeRIP-seq data and MeRIP-qPCR results. Data are means ± SD. ∗∗P < 0.01, ∗∗∗∗P < 0.0001.

3.3. Bioinformatics analysis of differentially m5C-methylated circRNAs

Functional annotation of differentially m5C-methylated cricRNAs were performed using GO and KEGG analyses. GO analysis revealed that hypermethylated cricRNAs were enriched in hepatocyte apoptosis (biological process), nucleoplasm (cellular component), and phospholipid transporter activity (molecular function), whereas hypomethylated cricRNAs were linked to developmental processes, nuclear localization, and heterocyclic compound binding (Fig. 3A and B).

Fig. 3.

Enrichment analyses of the m5C hypermethylated and hypomethylated circRNAs in BCBM versus BC.

A–B: GO analyses of the m5C hypermethylated (A) and hypomethylated (B) circRNAs. BP: biological process; CC: cellular component; MF: molecular function. C–D: KEGG pathway analyses of the m5C hypermethylated (C) and hypomethylated (D) circRNAs.

KEGG pathway analysis identified significant associations of hypermethylated cricRNAs with the ErbB and VEGF signaling pathways, while hypomethylated cricRNAs were predominantly enriched in platelet activation and cancer-related proteoglycan pathways (Fig. 3C and D).

3.4. Differentially expressed circRNAs between BCBM versus BC

To identify differentially expressed circRNAs between BCBM and BC, we performed RNA-seq analysis. Hierarchical clustering revealed distinct circRNA expression patterns (Fig. 4A). Comparative analysis identified 75 significantly upregulated and 53 downregulated circRNAs in BCBM versus BC (|fold change| > 2, P < 0.05; Fig. 4B).

Fig. 4.

Differential expression analysis of the input samples in BCBM versus BC by RNA-seq.

A: Hierarchical cluster of the differentially expressed circRNAs. B: Volcano plot of the differentially expressed circRNAs. C: Enrichment analysis of the biological functions and signaling pathways. D: The protein interaction network analysis of the differentially expressed circRNAs.

PPI network analysis of the 128 differentially expressed circRNAs implicated their involvement in Rho GTPase signaling, chromatin organization, cell shape regulation, and cell-cell adhesion (Fig. 4C). MCODE analysis further highlighted functional clusters associated with PKB/AKT signaling regulation and cell-cell adhesion (Fig. 4D).

3.5. Validation of m5C-methylated circRNA expression by RT-qPCR

RT-qPCR validation confirmed differential expression of m5C-modified circRNAs in cell lines. Hypermethylated circRNAs (hsa_circ_0118701/0111796/0004516) showed significantly elevated expression in BCBM versus BC, while hypomethylated circRNAs (hsa_circ_0023216/0005347/0124450) exhibited reduced expression. Notably, two hypomethylated circRNAs (hsa_circ_0019225/0068135) demonstrated increased expression (P < 0.05, Fig. 5A). Expression patterns from RT-qPCR and RNA-seq showed strong concordance for all circRNAs except hsa_circ_0023216 (Fig. 5B). A four-quadrant scatter plot illustrated the correlation between m5C methylation and expression levels, with hsa_circ_0118701/0111796/0004516 showing both hypermethylation and upregulation, whereas hsa_circ_0023216/0124450/0005347 demonstrated hypomethylation-driven downregulation. These findings highlighted the complexity of epigenetic regulation of circRNAs in BCBM, emphasizing the multifaceted interactions and mechanisms that govern their expression and function in this aggressive disease state (Fig. 5C).

Fig. 5.

Validation of candidate circRNAs expression levels in BCBM versus BC.

A: Relative expression levels of candidate circRNAs in BCBM versus BC by RT-qPCR. n = 3. B: Comparison of the mean fold changes (log transformed) between the RNA-seq data and RNA-qPCR results. C: Quadrantal diagram of circRNA m5C methylation and expression based on the MeRIP-qPCR and RT-qPCR data. Data are means ± SD. ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001.

3.6. Catalytic-dependent m5C methylation by NSUN2 enhanced hsa_circ_0004516 stability

To investigate m5C methyltransferase regulation of circRNA methylation, we first analyzed the expression of key methyltransferases (NOP2, NSUN2- NSUN7) in TCGA database. Notably, NSUN2 showed the highest expression in 1092 BC tissues including 3 BCBM cases (P < 0.0001; Fig. 6A). We next validated NSUN2 expression in both cellular and clinical specimens. At the cellular level, NSUN2 was significantly upregulated in BCBM versus BC cells (P < 0.001; Fig. 6B). Consistently, tissue analysis revealed elevated NSUN2 expression in BC specimens compared to ANTs (P = 0.037; n = 24; Fig. 6C). Immunohistochemistry confirmed high NSUN2 expression in BC and BCBM compared to ANTs (Fig. 6D).

Fig. 6.

NSUN2-mediated m5C modification of has_circ_0004516 promotes BCBM by increasing RNA stability.

A: Relative expression levels of NSUN family members in the TCGA database. n = 1092. Red dots: BCBM cases, n = 3. B–C: Relative expression levels of NSUN2 in cell (B) and tissue samples (C). B, n = 3; C, n = 11. D: Immunohistochemical analysis of NSUN2 expression in paired ANTs, primary BC, and BCBM tissues. Scale bars: 50 μm. E: Correlation analyses of NSUN2 with candidate circRNAs at the tissue level. n = 24. F–G: Confirmation of NSUN2 knockdown efficiency by RT-qPCR (F) and Western blot (G). F and G, n = 3. H: Relative expression levels of the candidate circRNAs upon NSUN2 knockdown. n = 3. I: Relative expression levels of hsa_circ_0004516 in BCIS (breast carcinoma) and MBC (metastatic breast cancer) (dataset: GSE101122 and GSE111504). J: Relative expression levels of the candidate circRNAs in BC versus ANTs by RT-qPCR. n = 24. K–L: Relative expression levels of hsa_circ_0004516 in control (oe-NC) vs oe-NSUN2-WT (oe-WT) vs. e-NSUN2-MUT (oe-MUT). M: Detection of the half-life of hsa_circ_0004516 by RNA stability assay. N: Relative expression levels of YBX1 and ALYREF in 1092 BC tissues. Black dots: BCBM cases, n = 3. O: RIP-qPCR validation of YBX1 binding to the m5C methylation sites on has_circ_0004516. n = 3. P: Agarose gel electrophoresis analysis confirming the interaction between YBX1 and the methylation sites within has_circ_0004516. Data are means ± SD. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001, ns: no statistical significance.

To investigate NSUN2-circRNA relationships, we analyzed three hypermethylated circRNAs in 24 paired BC/ANT samples. Significant positive correlations were observed between NSUN2 and hsa_circ_0118701/0004516 (P < 0.05; Fig. 6E), but not hsa_circ_0111796. To explore the impact of NSUN2 on the expression of candidate circRNAs, we established 231-BR NSUN2-knockdown cells. The efficiency of NSUN2 knockdown was confirmed by RT-qPCR and Western blot, demonstrating a significant reduction in NSUN2 expression at both the mRNA and protein levels (P < 0.01; Fig. 6F and G). Using NSUN2-knockdown 231-BR cells, we found corresponding downregulation of hsa_circ_0118701/0004516 (Fig. 6H). Database analysis (GSE101122/GSE111504) showed elevated hsa_circ_0004516 in metastatic versus primary cells (P = 0.0014; Fig. 6I), while hsa_circ_0118701 showed no detectable expression. Furthermore, expanding the cohort to 24 BC specimens demonstrated significant upregulation of hsa_circ_0004516 in BC tissues versus ANTs (P < 0.05, Fig. 6J), hsa_circ_0118701 showed no significant differences.

To further explore the mechanism by which NSUN2 regulates m5C methylation of hsa_circ_0004516, we constructed overexpression vectors for both the wild-type NSUN2 and a mutant with impaired catalytic activity (C271A/C321A). RT-qPCR results showed that compared to the control group, overexpression of wild-type NSUN2 upregulated the level of hsa_circ_0004516 (P < 0.01), whereas overexpression of the NSUN2 mutant did not exhibit this upregulation (Fig. 6K-L). RNA stability analysis revealed that NSUN2 knockdown significantly shortened the half-life of hsa_circ_0004516 compared to the control (P < 0.05, Fig. 6M).

While ALYREF primarily mediates nucleocytoplasmic transport [32,33], YBX1 stabilizes RNA via m5C-binding domains [34,35]. NSUN2 knockdown significantly reduced the stability of hsa_circ_0004516 (Fig. 6M). Analysis of TCGA data (n = 1092 BC tissues) revealed significantly higher expression of YBX1 than ALYREF (P < 0.00001; Fig. 6N). Given YBX1's role as an m5C reader and overexpression in BC, we hypothesized that YBX1 recognizes m5C-methylated hsa_circ_0004516 to enhance its stability. RIP-qPCR assays demonstrated direct binding of YBX1 to hsa_circ_0004516 in 231-BR cells (P < 0.001; Fig. 6O). Agarose gel electrophoresis of RIP products validated specific binding, showing a clear hsa_circ_0004516 band only in the YBX1 pulldown group (vs. IgG control; Fig. 6P).

3.7. hsa_circ_0004516 promoted BCBM progression through AKT signaling pathway

To further validate the biological function of hsa_circ_0004516 in BCBM, we performed knockdown of hsa_circ_0004516 and confirmed silencing efficiency via RT-qPCR (P < 0.001, Fig. 7A). Functional assays demonstrated that hsa_circ_0004516 silencing robustly suppressed proliferation of 231-BR cells by CCK-8 assays (P < 0.05, Fig. 7B) and concordant colony formation assays (Independent samples t-test, P = 0.0357, Fig. 7C). Furthermore, transwell migration and invasion assays revealed marked impairment of si-circ_0004516-transfected 231-BR cells metastatic potential (P < 0.001, Fig. 7D). Bioinformatics analysis of RNA-seq data revealed significant enrichment of differentially expressed circRNAs in PKB/AKT signaling pathways. To investigate whether hsa_circ_0004516 mediates BCBM cellular processes (proliferation, migration, and invasion) by activating the AKT signaling pathway, we performed functional rescue experiments by treating hsa_circ_0004516-knockdown cells with SC79, a specific AKT activator. Western blot analysis demonstrated that hsa_circ_0004516 silencing (si-circ_0004516) significantly reduced the phosphorylation level of p-AKT (Ser473) (P < 0.05, Fig. 7E), while 10 μM SC79 treatment reversed this effect (P < 0.0001, Fig. 7F).

Fig. 7.

hsa_circ_0004516 promoted BCBM progression via AKT pathway in vitro.

A: Assessment of has_circ_0004516 expression in 231-BR cells after has_circ_0004516 knockdown, analyzed by RT-qPCR, n = 3. B–C: CCK-8 (B) and colony formation assays (C) were conducted to assess the proliferation of 231-BR cells with hsa_circ_0004516 knockdown (si-circ-0004516). B and C, n = 3. D: Transwell assays demonstrating the capacity of migration and invasion in 231-BR cells following hsa_circ_0004516 knockdown. Scale bars: 50 μm. n = 3. E: Western blot analysis revealed the protein levels of AKT, and phospho-AKT (Ser473) after hsa_circ_0004516 knockdown. F: Western blot analysis demonstrated that hsa_circ_0004516 silencing modulated phosphorylated AKT (Ser473) levels following 48 h SC79 (10 μM) treatment, n = 3. G–H: Proliferation ability detection: knockdown of hsa_circ_0004516 in 231-BR cells treated with SC79 (10 μM). CCK8 (G) and colony formation assay (H), n = 3. I: The capacity of migration and invasion: knockdown of hsa_circ_0004516 in 231-BR cells treated with SC79 (10 μM), n = 3. Data are means ± SD. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001.

CCK-8 assays demonstrated that SC79 (10 μM) treatment significantly restored proliferation activity in hsa_circ_0004516-knockdown cells, reversing the inhibitory effect of si-circ_0004516 (P < 0.0001, Fig. 7G). Colony formation assays indicated an increase in clonogenic capacity upon SC79 co-treatment (P < 0.001, Fig. 7H). Furthermore, Transwell assay results showed that hsa_circ_0004516 knockdown inhibited 231-BR cell migration and invasion. However, SC79 treatment rescued the invasion and migration of hsa_circ_0004516 knockdown cells (P < 0.001, Fig. 7I). These results confirm that hsa_circ_0004516 mediates proliferation, invasion, and migration of BCBM cells through the activation of the AKT signaling pathway.

3.8. Validation of interactions between miR-1301-3p and hsa_circ_0004516/AKT1 mRNA

miRDB analysis revealed hsa_circ_0004516 enrichment in PI3K pathways, thus prompting functional validation of its role in activating AKT signaling to drive BCBM (Fig. 8A). Dual-luciferase reporter assays in 231-BR cells validated predicted miR-1301-3p complementary binding sites in hsa_circ_0004516 as well as the AKT1 3′UTR (Fig. 8B). Wild-type (WT) and binding site mutant (MUT) luciferase reporter constructs were generated by cloning the corresponding sequences into GV272 plasmids. Co-transfection of hsa_circ_0004516-WT and miR-1301-3p mimics significantly decreased luciferase activity, while no effect was observed with the empty vector or hsa_circ_0004516-MUT (Fig. 8C). Similarly, co-transfection of AKT1-WT and miR-1301-3p mimics significantly decreased luciferase activity, while no effect was observed with AKT1-MUT (Fig. 8D). These results indicate that hsa_circ_0004516 interacts with miR-1301-3p, and miR-1301-3p directly targets AKT1 3′UTR region.

Fig. 8.

Validation of interactions between miR-1301-3p and hsa_circ_0004516/AKT1 via dual-luciferase reporter assay.

A: The core regulatory network of hsa_circ_0004516-miRNA-mRNA. The networks were predicted using the circAtlas 3.0 and miRDB databases and constructed and visualized using Cytoscape software. B: Schematic diagram illustrates the predicted complementary binding sites of miR-1301-3p within hsa_circ_0004516 as well as within the 3′UTR of AKT1. C: Relative luciferase activity in 231-BR cells co-transfected with hsa_circ_0004516-WT/MUT vector and miR-1301-3p mimics or negative control. D: Relative luciferase activity in 231-BR cells co-transfected with AKT1-WT/MUT vector and miR-1301-3p mimics or negative control. Data are means ± SD. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001.

4. Discussion

CircRNAs serve as ideal diagnostic biomarkers for BC metastasis due to their high stability [36,37]. While m6A modification has dominated circRNA epitranscriptomic research, m5C methylation (a critical RNA modification) remains poorly characterized in terms of its regulatory circuitry and mechanistic implications in circular RNAs [[38], [39], [40]]. Current studies remain limited and primarily focus on the preliminary observations of m5C in solid tumors such as esophageal and lung cancer [22,41]; conversely, a systematic analysis of specific circRNAs m5C methylation patterns and their dynamic regulatory networks in BCBM are entirely absent.

Our study provides the first comprehensive m5C methylation landscape of circRNAs in BCBM. Through integrated MeRIP-seq and transcriptomics, we identified 48 hypermethylated and 128 hypomethylated circRNAs in BCBM versus BC, revealing distinct methylation motifs and epigenetic signatures. The chromosome-specific m5C peak densities (enrichment on chromosomes 1–3, with depletion on chromosomes 22 and X) align with BCBM-associated loci. Although the genomic priorities require further elucidation, the spatial distribution demonstrates significant co-localization with key BCBM-associated genes: chr3 contains PIK3CA (3q26.32), a key PI3K signaling regulator linked to BCBM progression [42], while chr2 harbors ERBB4 (2q34), an EGFR family member with established roles in BCBM [43]. Conversely, low m5C density on chr22 coincides with tumor suppressors like CHEK2, often deleted in BCBM [44]—potentially reducing substrate availability for m5C modification. These chromosome-specific m5C preferences may reflect functional targeting of m5C to genes promoting brain metastatic progression.

Bioinformatics analysis revealed that hypermethylated circRNAs in BCBM were predominantly enriched in the ErbB and VEGF signaling pathways. ErbB overexpression, observed in ∼30 % of BCs, serves as a key prognostic and predictive biomarker in BCBM [45]. ErbB-positive BC demonstrated inherent predisposition to BM [46,47]. VEGF critically contribute to BM pathogenesis by promoting tumor cell transendothelial migration and compromising endothelial integrity [48]. MCODE identified functional clusters associated with AKT-mediated survival signaling and cell-cell adhesion. ENPP1 secreted from HER2+ BC cells promotes BM via disruption of insulin/AKT/GSK3β/β-catenin axis, which destabilizes blood-brain barrier integrity [49]. Functional validation pinpointed hsa_circ_0004516 as a key metastasis driver, where NSUN2-mediated m5C hypermethylation enhanced circRNA stability to activate AKT signaling.

CircRNAs expression dynamics are governed by a multifaceted interplay of molecular mechanisms, including reverse splicing efficiency, RNA polymerase II transcription elongation rate, the presence of intron complementary sequences (ICSs), spliceosomes, RNA-binding proteins, and epigenetic modifications [6,50]. While contemporary circRNA research predominantly focuses on elucidating functional consequences and effector pathways, this research focused mainly on the upstream regulatory mechanisms (especially m5C epigenetic modifications). Emerging evidence demonstrated that m5C deposition coordinates oncogenic transformation through spatiotemporal control of circRNA localization, structural stability, and protein interactome dynamics [21,22,41]. RNA m5C methylation modification constitutes a highly dynamic process orchestrated by the coordinated action of methyltransferases, demethylases, and recognition proteins; dysregulation of this machinery is implicated in tumorigenesis [[51], [52], [53]]. TCGA analysis identified NSUN2 as a highly expressed m5C methyltransferase in BC, showing significant positive correlations with hsa_circ_0118701 and hsa_circ_0004516. Functional studies demonstrated that NSUN2 enhanced hsa_circ_0004516 stability through catalytic sites (C271/C321)-dependent m5C modification, critically regulating BCBM progression. Prior studies indicate NSUN2-driven m5C modification augments circRNA stability to promote autophagy and glycolysis, thereby driving disease progression aligning with our findings [22,32].

This study established a multi-dimensional screening framework by integrating GSE246721 methylation profiles with transcriptomic data for initial candidate identification. Co-expression network analysis and functional validation of NSUN2-circRNA interactions were systematically performed, followed by secondary screening using metastasis-specific databases (GSE101122/GSE111504) and clinical sample validation. Ultimately, we confirmed hsa_circ_0004516 as a pivotal regulator of BCBM. CircRNAs collaborate with NSUN2-mediated m5C modification machinery and readers ALYREF/YBX1, facilitating mRNA nuclear export and stabilization, which triggers downstream pathway activation [20]. RIP-qPCR confirmed YBX1 enrichment at hsa_circ_0004516's m5C region. The RNA epitranscriptome, particularly m5C modification, regulates RNA stability and function in BC through dynamic RNA-protein interactions. Notably, natural compounds targeting methyltransferases (e.g., ginsenosides) may disrupt this axis by inhibiting NSUN2 catalytic activity or YBX1 binding, thereby downregulating PI3K/AKT signaling [54,55].

While circPOKE, circBCBM1 and circRRM2 have established roles in BC metastasis through generic miRNA sponging [[56], [57], [58]], hsa_circ_0004516, acting as an oncogenic factor, promoted BCBM cell proliferation, migration, and invasion in vitro. Intriguingly, Western blot analysis demonstrated hsa_circ_0004516 promotes BCBM by activating the AKT signaling pathway. circUCK2 indirectly induced EGFR and downstream AKT signaling activation by enhancing TGFα secretion [59]. Although hsa_circ_0004516 specifically promotes BCBM via activating the AKT pathway, dual-luciferase reporter assays validated its direct binding to miR-1301-3p and miR-1301-3p′s targeting of AKT1. Nevertheless, the molecular mechanism orchestrating this regulatory axis requires systematic elucidation.This study is the first to reveal the role of circRNA m5C in BM; however, the absence of in vivo validation and a moderate sample size limit its clinical translatability. Future work will incorporate patient-derived xenograft (PDX) models and multi-center cohorts.

As the first study to decode circRNA m5C methylation in BCBM, we elucidated circRNA m5C-mediated epigenetic regulation in metastasis, revealing potential diagnostic and therapeutic implications. This study not only expands the mechanistic repertoire of circRNAs but also identifies hsa_circ_0004516 as a promising therapeutic target for BCBM.

CRediT authorship contribution statement

Min Li: Writing – review & editing, Writing – original draft, Validation, Investigation, Funding acquisition, Formal analysis, Data curation. Jiawei Li: Visualization, Investigation, Formal analysis, Data curation. Hongbo Wen: Validation, Investigation, Formal analysis, Data curation. Jun Li: Validation, Investigation, Formal analysis, Data curation. Song Wang: Validation, Investigation, Formal analysis, Data curation. Jianran Guo: Validation, Investigation, Formal analysis, Data curation. Dongyan Zhang: Validation, Investigation, Formal analysis, Data curation. Anqi Zhang: Writing – review & editing, Writing – original draft, Validation, Investigation, Formal analysis, Data curation. Chuanyou Cui: Validation, Investigation, Funding acquisition, Formal analysis, Data curation. Rong Fu: Validation, Investigation, Formal analysis, Data curation. Meng An: Validation, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Wei Zhang: Validation, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Bo Fu: Writing – review & editing, Writing – original draft, Validation, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization.

Funding

This work was supported by Natural Science Foundation of Shandong Province, China (Grant no. ZR2022MH272 and ZR2023QH115); Medicine and Health Science and Technology Foundation of Shandong Province, China (Grant no. 202402060623 and 202202080721); Liaocheng Key Research and Development Project, China (Grant no. 2023YD27 and 2023YD43); Science and Technology Development Project of Shandong Second Medical University, China (Grant no. 2023FYM106, 2023FYQ035, and 2024FYM127).

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 is the supplementary data to this article:

Data availability

The raw sequencing data have been deposited in the NCBI Gene Expression Omnibus under accession number GSE246721.