Intranuclear paraspeckle‐circular RNA TACC3 assembly forms RNA‐DNA hybrids to facilitate MASH‐related hepatocellular carcinoma growth in an m⁶A‐dependent manner

Abstract

Background

Metabolic dysfunction‐associated steatohepatitis (MASH) is anticipated to become the leading cause of hepatocellular carcinoma (HCC). Accumulating evidence indicates that N6‐methyladenosine (m⁶A)‐modified circular RNAs (circRNAs) play key roles in tumor malignant progression. However, the precise molecular mechanisms by which circRNAs and their m⁶A modification regulatory networks respond to metabolic reprogramming, such as lipid overload stress, to drive malignant tumor progression in the context of MASH‐related HCC remain unclear. This study aimed to investigate the role and regulatory network of m⁶A‐modified circRNAs in MASH‐related HCC.

Methods

Epitranscriptomic microarray and in situ hybridization assays were used to validate circTACC3 expression in MASH‐related HCC specimens. Palmitic acid (PA) and oleic acid (OA) was applied to NAC‐organ assembled three‐dimensional‐organoid and HCC cell lines to imitate pathological lipid overload. The circTACC3‐paraspeckle interaction was studied utilizing fluorescence lifetime imaging microscopy‐Forster resonance energy transfer. An integrative analysis combining DNA‐RNA immunoprecipitation combined with chromatin isolation by RNA purification (DRIP‐ChIRP), γH2AX cleavage under target and tagmentation, and high‐throughput/resolution chromosome conformation capture sequencing were used to study chromatin remodeling induced by circTACC3‐formed RNA‐DNA hybrids (R loops) at DNA double‐strand break (DSB) loci during lipid overload.

Results

The most prevalent m⁶A‐modified circRNA in MASH‐related HCC, circTACC3, had a substantial impact on the intracellular lipid accumulation, growth, and environmental adaptive survival of tumor cells. Under lipid overload conditions, circTACC3 interacted directly with non‐POU domain‐containing octamer‐binding protein (NONO/p54^nrb) to assemble intranuclear paraspeckle. This process was dependent on the m⁶A‐modification sites of circTACC3 and facilitated its nuclear retention. Using DRIP‐ChIRP‐sequencing, we demonstrated that circTACC3‐containing paraspeckles were recruited to DSB foci to form R loops (DSB‐circTACC3‐R loops). We discovered 4 highly enriched motifs of DSB‐circTACC3‐R loops. DSB‐circTACC3‐R loops further facilitated the contact and fusion of topologically associated domains (TADs) and selectively activated genes related to the malignant phenotype of MASH‐related HCC. Interestingly, circTACC3‐R loops exerted positive feedback control over the assembly of circTACC3 paraspeckle and clustering of TADs.

Conclusions

The m⁶A modification‐dependent circTACC3‐paraspeckle assembly results in the formation of R loops at DSB foci, leading to chromatin remodeling and the activation of genes involved in MASH‐related HCC malignant progression. This process identifies potential therapeutic targets.

List of Abbreviations

3D

Three dimensions

AFL

Alcoholic fatty liver

ASH

Alcoholic steatohepatitis

BSJ

Back‐splicing junction

ChIRP

Chromatin isolation by RNA purification

circRNA

Circular RNA

CLIP

UV cross‐linking and immunoprecipitation

CUT&Tag

Cleavage under target and tagmentation

DAA

3‐deazaadenosine

ddNTPs

Dideoxynucleotides

DDR

DNA damage response

DILI

Drug‐induced liver injury

dNTPs

Deoxynucleotides

Dox

Doxycycline

DRIP

DNA‐RNA immunoprecipitation

DSB

DNA double‐strand break

FACS

Fluorescence activated cell sorting

FDR

False discovery rate

FISH

Fluorescence in situ hybridization

FLIM‐FRET

Fluorescence lifetime imaging microscopy‐Forster resonance energy transfer

GO

Gene Ontology

H&E

Hematoxylin and eosin

HBc

Hepatitis B virus core antigen

HBs

Hepatitis B virus surface antigen

HBV

Hepatitis B virus

HBx

Hepatitis B virus X protein

HCC

Hepatocellular carcinoma

Hi‐C

High‐throughput/resolution chromosome conformation capture

IF

Immunofluorescence

IGF2BP1

Insulin like growth factor 2 mRNA binding protein 1

ISH

I n situ hybridization

KIAA1429

Vir‐like m⁶A methyltransferase‐associated protein

LC‒MS/MS

Liquid chromatography‒tandem mass spectrometry

LncNEAT1

LncRNA nuclear paraspeckle assembly transcript 1

m⁶A

N6‐methyladenosine

MASH

Metabolic dysfunction‐associated steatohepatitis

MASLD

Metabolic dysfunction‐associated steatotic liver disease

MeRIP

RNA methylation immunoprecipitation

METTL14

Methyltransferase‐like 14

METTL3

Methyltransferase‐like 3

NAFLD

Nonalcoholic fatty liver disease

NAS

Non‐alcoholic fatty liver disease activity score

NC

Nitrocellulose

ncRNA

Noncoding RNA

NONO/p54^nrb

Non‐POU domain‐containing octamer‐binding protein

O/E value

Observed/Expected value

OA

Oleic acid

PA

Palmitic acid

PAGE

Polyacrylamide gel electrophoresis

PC1

First principal component

PCA

Principal component analysis

PT

Peritumoral normal tissue

PVDF

Polyvinylidene difluoride

R loop

RNA‐DNA hybrid

RIP

RNA immunoprecipitation

ROI

Region of interest

ROS

Reactive oxygen species

RPKM

Reads per kilobase per million mapped reads

RT‐PCR

Reverse transcription‐polymerase chain reaction

RT‐qPCR

Reverse transcription‐quantitative Real‐time polymerase chain reaction

SAH

S‐adenosylhomocysteine

STM

STM2457

STX6

Syntaxin 6

T

Tumor tissue

TAD

Topologically associated domain

VCL

Vinculin

γH2AX

Ser‐139 residue of the histone variant H2AX

1. BACKGROUND

Owing to the increasing prevalence of obesity and metabolic syndrome, the global incidence of metabolic‐associated steatotic liver disease (MASLD), formerly called nonalcoholic fatty liver disease (NAFLD), has increased to 20%‐25% within the last decade [1, 2]. Notably, metabolic dysfunction‐associated steatohepatitis (MASH), the more severe manifestation of MASLD, contributes to approximately 2% of hepatocellular carcinoma (HCC) cases annually [3, 4]. MASH is expected to become the primary etiology of HCC by 2030 [3].

Compared with other etiologies, MASH‐related HCC has unique molecular and immune characteristics, influencing its progression and leading to suboptimal therapeutic outcomes [5]. Lipid overload caused by excess lipids in both the microenvironment and hepatocytes induces oxidative stress, replication stress and nucleotide pool imbalance [3, 6], further contributing to DNA damage and mutation accumulation, as well as conferring selective benefits on hepatocytes, as premalignant events in MASH‐related HCC [3, 7].

Noncoding RNAs (ncRNAs) play crucial roles in many biological processes. Circular RNAs (circRNAs) are unique ncRNAs that are abundant and stable in cells [8]. Dysregulated circRNAs are instrumental in HCC progression, and may serve as diagnostic biomarkers or therapeutic targets [9, 10]. N6‐methyladenosine (m⁶A) RNA methylation, the most abundant RNA modification, is crucial in physiological and pathological hepatic immune responses, lipid metabolism, hepatitis virus infection and hepatitis B virus (HBV)‐related HCC [11, 12]. However, further research is needed to elucidate the regulatory network of m⁶A modifications and m⁶A‐modified circRNAs in MASH‐related HCC.

Rather than being equally dispersed throughout the cell, regulatory circRNAs have specific subcellular localizations that match their roles. Intronic and intron lariat‐derived circRNAs preferentially localize to the nucleus, whereas exonic circRNAs primarily localize to the cytoplasm [8]. In addition, circRNA nuclear export pathways are distinct from those of linear mRNAs, which require exportin‐2 as an export receptor, and insulin like growth factor 2 mRNA binding protein 1 (IGF2BP1) as an adapter [13]. However, certain intron containing circRNAs and some exonic circRNAs are predominantly retained in the nucleus, and the mechanisms underlying their nuclear localization, as well as the functions of nucleus‐localized circRNAs remain relatively poorly studied.

The R loop structure contains a DNA‐RNA hybrid and a displaced single‐stranded DNA [14]. Increasing evidence has revealed that R loops function in gene expression, chromatin structure modulation, DNA damage response (DDR) and DNA replication [14, 15]. Moreover, the accumulation of aberrant R loops is closely related to the occurrence and development of cancers; thus, R loops are considered potential targets for molecular diagnosis and treatment. For example, HOTTIP‐dependent R loop formation regulates CTCF boundary activity and topologically associated domain (TAD) integrity in leukemia cells [16]. In breast cancer, R loop accumulation is related to transcription‐replication collisions and excessive genomic instability [17]. In addition, circRNA‐DNA hybrids (circR loops) within the leukemia cell genome regulate the oncogenes by mediating chromatin remodeling, DNA double‐strand break (DSB), and gene transcription, which ultimately promote leukemia development [18]. Nevertheless, the role of circR loops in solid tumor cells remains to be investigated.

Emerging evidence highlights the pivotal regulatory roles of m⁶A‐modified circRNAs in oncogenic progression. Nevertheless, the precise mechanisms through which m⁶A‐circRNA networks orchestrate metabolic perturbations like lipid overload to fuel MASH‐related HCC pathogenesis remain unclear. This study systematically dissected the functional hierarchy, molecular interaction, and disease‐specific regulatory circuitry of intranuclear m⁶A‐circRNAs and circR loops in MASH‐related HCC progression.

2. MATERIALS AND METHODS

2.1. Patient samples

This study used MASLD tissues from patients diagnosed and surgically resected at the First Affiliated Hospital of Shandong First Medical University (Jinan, Shandong, China). The samples are appropriately stored at the hospital's biological sample bank. Patient recruitment occurred from December, 2022 to October, 2023. This study used the 2020 worldwide expert consensus statement and 2023 multisociety Delphi consensus statement on new fatty liver disease nomenclature to select samples. For MASLD, patients were excluded from any other causes of steatosis, and meet at least one of the following characteristics: (a) body mass index (BMI) ≥ 23 kg/m² or waist circumference > 94 cm (male) and > 80 cm (female) or ethnicity adjusted equivalent; (b) fasting serum glucose ≥ 5.6 mmol/L or 2‐hour post‐load glucose levels ≥ 7.8 mmol/L or HbA1c ≥ 5.7% or type 2 diabetes or treatment for type 2 diabetes; (c) blood pressure ≥ 130/85 mmHg or specific antihypertensive drug treatment; (d) plasma triglycerides ≥ 1.70 mmol/L or lipid lowering treatment; (e) plasma high‐density lipoprotein (HDL)‐cholesterol ≤ 1.0 mmol/L (male) and ≤ 1.3 mmol/L (female) or lipid lowering treatment.

For MAFLD tissues, NAS scores were assigned based on their pathological characteristics, following the criteria established and validated by the Pathology Committee of the NASH Clinical Research Network. The NAS scoring system evaluates three histological features semi‐quantitatively: steatosis (0‐3), lobular inflammation (0‐3), and hepatocellular ballooning (0‐2), with a maximum total score of 8. Specifically:

• (1)Steatosis: 0 points (<5% of hepatocytes affected); 1 point (5%‐33%); 2 points (34%‐66%); 3 points (>66%).
• (2)Lobular Inflammation (per 20× field): 0 points (none); 1 point (<2 foci); 2 points (2‐4 foci); 3 points (>4 foci).
• (3)Hepatocellular Ballooning: 0 points (none); 1 point (rare ballooning cells); 2 points (prominent ballooning present).

Liver tumor tissues and matched peritumoral normal tissues were obtained from patients who were diagnosed and received surgical resection at the Eastern Hepatobiliary Surgery Hospital (Shanghai, China) and the First Affiliated Hospital of Shandong First Medical University (Jinan, Shandong, China). The samples are appropriately stored at the hospital's biological sample bank. Patients were recruited from May, 2010 to December, 2020. The criteria of MASH‐related HCC sample inclusion in this study were as follows: the patients were diagnosed with HCC using histopathology or imaging criteria based on American Association for the Study of Liver Disease and European Association for the Study of the Liver guidelines; MASH was documented clinically; patients with MASLD and steatohepatitis will be designated as MASH; other etiological backgrounds, such as HBV/hepatitis C virus (HCV) infection, alcoholic steatohepatitis (ASH) or alcoholic fatty liver (AFL), and drug‐induced liver injury (DILI) were ruled out.

2.2. NAC‐organ assembled 3D organoid culture technology

NAC‐organ assembled 3D organoid culture was performed by Puheng Technology Co (Suzhou, Jiangsu, China). In brief, MASH‐related HCC tumor tissues and paired peritumoral normal tissues of around 3 cm × 3 cm × 3 cm dimensions were digested into single cells by a single cell suspension preparation equipment (DHTEH‐10, Reward, Shenzhen, Guangdong, China), and then seeded onto low‐adhesion 96‐well plates at 3,000 cells per well to form the 3D model (Day 0). On Day 2, MASH medium (#MI01, Puheng Technology Co.) was added to construct the MASH‐related HCC model, and NAC‐Liver liver cancer medium (#MS01, Puheng Technology Co.) was added to construct the control model. The culture medium was changed every 2 days in the following 10 days. STM2457, S‐adenosylhomocysteine (SAH), 3‐deazaadenosine (DAA) or the solvent control DMSO (v/v = 1/1000, #HY‐Y0320, MedChemExpress, NJ, US) was added on Day 4. Continuous dosing treatment was performed for 8 days. NAC‐Organs were collected on Day 12, and were fixed with 4% paraformaldehyde (#G1101, PFA, Sevicebio, Wuhan, Hubei, China) for paraffin, frozen sections and 3D‐fluorescence in situ hybridization (3D‐FISH) staining.

2.3. Cell culture

HCC cell lines HepG2, HCCLM3, Huh7, Hep3B and normal liver cell lines QSG‐7701 and L02 were purchased from Cell Bank of Type Culture Collection of Chinese Academy of Sciences, Shanghai Institute of Cell Biology, and Chinese Academy of Sciences. Cell lines were maintained at 37°C in an atmosphere containing 5% CO₂ in DMEM (Basal Media, Shanghai, China) supplemented with 10% fetal bovine serum (Biological Industries, Kibbutz Beit Haemek, Israel). HCC cell line Hep‐AD38 was kindly provided by Professor Tong Shuping's laboratory at Fudan University (Shanghai, China), and was maintained at 37°C in an atmosphere containing 5% CO₂ in DMEM/F‐12 (Basal Media) supplemented with 10% fetal bovine serum (Biological Industries). Cells were passed every 1‐2 days to maintain logarithmic growth.

2.4. Hematoxylin and eosin (H&E) staining

Tissue sections were deparaffinized in xylene (#10023418, Sinopharm Chemical Reagent Co., Shanghai, China), and rehydrated through graded ethanol (100% → 70%). Nuclei were stained with hematoxylin staining solution (#E607317, BBI Life Sciences Corporation, Shanghai, China) for 5 min, rinsed in tap water, differentiated briefly in 1% acid ethanol for 30 s, and blued in tap water for 1 min. Cytoplasmic counterstaining was performed in eosin staining solution (#E607321, BBI Life Sciences Corporation) for 5 min, followed by dehydration in ethanol (75% → 95% → 100%) (ethanol 75%, #80176961, Sinopharm Chemical Reagent Co.; ethanol 95%, #10009164, Sinopharm Chemical Reagent Co.; ethanol 100%, #10009218, Sinopharm Chemical Reagent Co.), clearing in xylene, and mounting with xylene‐based medium. All steps were performed at room temperature with distilled water rinses between reagents.

2.5. Epitranscriptomic microarray assay for m⁶A‐modified circRNAs

Arraystar human m⁶A‐circRNA epitranscriptomic microarray's detection and analysis were performed as follows: the nanodrop ND‐1000 (Thermo Fisher Scientific, MA, US) machine quantified total RNA from each sample, and the bioanalyzer 2100 electrophoresis analyzed RNA integrity. Sample preparation and microarray hybridization followed Arraystar's methodology. In brief, anti‐m⁶A rabbit polyclonal antibody (#202003, Synaptic Systems, Goettingen, Germany) immunoprecipitated total RNA. The m⁶A modified RNA was extracted from immunoprecipitated beads and labeled “IP”. Unmodified RNA was collected from the supernatant and labeled “Sup”. Each reaction was conducted individually using the Arraystar RNA labeling procedure to label the “IP” and “Sup” RNAs with Cy5 and Cy3 as cRNAs after RNase R (#14606ES72, YEASEN, Shanghai, China) treatment. These cRNAs were together hybridized to Arraystar human circRNA epitranscriptomic microarray (8 × 15K, Arraystar, MD, US). After washing the slides, the arrays were scanned using the two‐color channel Agilent scanner G2505C (Agilent, CA, US).

Acquired array images were analyzed using Agilent Feature Extraction software (version 11.0.1.1, CA, US). Raw intensities of IP (Immunoprecipitation, Cy5‐labeled) and Sup (Supernatant, Cy3‐labeled) were normalized with the mean of peak RNA intensities on a log₂ scale. After peak normalization, probe signals with Present (P) or Marginal (M) QC markers in at least 1 out of 10 samples were retained for further “m⁶A number” analysis. The amount of m⁶A methylation was calculated based on the IP (Cy5‐labeled) normalized intensity. Differences in m⁶A methylated circRNAs between the two groups were determined by fold change and statistical significance (P value) threshold screening. Hierarchical clustering was performed to show distinguishable patterns of m⁶A methylation among samples.

2.6. R loop structural interference

To generate stable cell lines with the inducible overexpression of RNase H1, HepG2 and HCCLM3 cells were infected with lentivirus expressing TetIIP‐RNase H1‐3FLG‐Ubi‐TetR‐IRES‐Puromysin and selected with puromycin (#T19978, TargetMol, MA, US, final concentration 2 µg/mL) for 1‐2 weeks. The induction of Flag‐RNase H1 was determined by western blot using cellular extracts from cells treated with DMSO or Tetracycline (#T120076, TargetMol, final concentration 1 µg/mL) for 24 h.

2.7. m⁶A single base site PCR

Candidate m⁶A RNA methylation modification sites were predicted using the SRAMP website (http://www.cuilab.cn/sramp). The specific PCR primers (Supplementary Table S1) for these sites and standards (known concentration and corresponding specific primers) were designed. Reverse transcription‐polymerase chain reaction (RT‐PCR) assay was performed for the target gene and standards of each sample, and the standard curve was plotted according to the standards. The total complementary DNA (cDNA) concentration of the target gene, the cDNA concentration of m⁶A unmethylated at the target gene locus and cDNA concentration of the standards for each sample were generated directly. RNA amount corrected for m⁶A methylation at the target gene locus = (total cDNA concentration of the target gene‐m⁶A unmethylated cDNA concentration at the target gene locus) × (RNA concentration of the standard/RNA cDNA concentration of the standard). The absolute amount of specific m⁶A modification sites of circRNA and the relative proportion to total RNA were determined according to it.

2.8. F2‐circRNA pulldown assay

Lentivirus constructs were generated by overexpressing circTACC3 with an F2 tag sequence (F2‐circTACC3). F2 tag sequence (5’‐GGCGCTGACAAAGCGCC‐3’) was divided into two parts and were connected to the head and tail of the circTACC3 junction site, respectively. Sequence 5’‐AAAGCGCC‐3’ is connected to the head of circTACC3, and 5’‐GGCGCTGAC‐3’ is connected to the tail of circTACC3. Subsequent transfection in circTACC3^−/‐ HCCLM3 and HepG2 cells was performed to construct HCCLM3 and HepG2 cell lines stably expressing F2‐circTACC3. F2‐circTACC3 was specifically enriched using the F2‐pulldown Kit (#F788702, FITGENE, Guangzhou, Guangdong, China) with probes complementary paired to the F2 tag sequence according to the manufacturer's instruction. In brief, 2 × 10⁷ cells were collected and were lysed with the lysis buffer. The magnetic beads and NT2 buffer were added to the protein extraction solution for incubation. The beads were mixed with 50 µL of elution buffer, to collect the protein and RNA complex bound to circTACC3.

2.9. Lentivirus infection

The lentivirus for overexpression of circTACC3, circTACC3 with mutated m⁶A modification sites, CRISPR‐Cas9 targeting Alu elements of circTACC3 (sgRNA‐1 sequence: 5’‐TACAGGAAATGGTGGATGCC‐3’; sgRNA‐2 sequence: 5’‐ATTGAGGAGCCCAGTCTACC‐3’). CRISPR‐Cas9‐gRNA lentivirus (dual sgRNA‐ALU‐KO‐CBH‐Cas9‐CMV‐EGFP), F2‐circTACC3, and RNase H1‐Tet‐On were designed by Genechem Company (Shanghai, China). The lentivirus for overexpression of NONO/p54^nrb‐mCherry was constructed by authors. The stable overexpression and knockout cell lines were constructed according to the following steps: HCC cell lines were inoculated in 6‐well plates at 1 × 10⁵ cells per dish, and virus infection was performed 24 h later. The corresponding volume of viral solution was added according to the MOI of the cells and the viral titer.

2.10. Small interfering RNA (siRNA) transfection

The siRNAs were purchased from Biotend Co. (Shanghai, China). Transfection were performed with JetPRIME reagents (Poly plus, Illkirch, France) according to the manufacturers’ instructions. Briefly, siRNA was added into JetPRIME buffer and was mixed well. Then, JetPRIME was added into the siRNA duplexes, and was homogenized by vortex immediately. The mixture was incubated for 10 min at room temperature to allow transfection complexes to form between duplexes and JetPRIME. Then, transfection mix was added into the 2 mL cell culture medium to make a final siRNA concentration at 100 nmol/L. Finally, the transfection medium was homogenized by gently swirling the plate.

2.11. Palmitic acid (PA) and oleic acid (OA) treatment

Cells were seeded in culture plates. After 18 h, when cell confluence reached 60%, the cells were washed once with PBS. HCCLM3 and HepG2 cells were treated with 120 µmol/L PA (#T2908, TargetMol) and 240 µmol/L OA (#T2O2668, TargetMol) in 1% fatty acid‐free BSA (#A8806, Sigma, MA, US) culture medium for 48 h for the PA + OA group, or DMSO (v/v 1/1000) for the mock group.

2.12. TUNEL assay

Cell apoptosis was detected using the TUNEL Kit (#40306ES60, YEASEN) according to the manufacturer's instruction. In brief, cells were fixed using 4% PFA at 4°C for 25 min. The fixed cells were treated with 0.2% Triton X‐100 for 5 min. The TdT incubation buffer containing 3‐OH BrDU probe was then added, and the cells were incubated for 1 h at 37°C in dark. The intensity of FITC fluorescence was determined using STELLARIS 5& STELLARIS 8 STED (Leica, Mannheim, Germany) or Lionheart^TM FX (Agilent).

2.13. RNA isolation and PCR analysis

Cells were subjected to RNA isolation using TRIzol reagent (#AG21101, Accurate Biology, Changsha, Hunan, China), according to the manufacturer's guidelines. The cDNA was synthesized using the Evo M‐MLV RT Premix for qPCR Kit (#AG11706, Accurate Biology), following the protocol specified by the manufacturer. The primers utilized in following RT‐PCR or reverse transcription‐quantitative Real‐time polymerase chain reaction (RT‐qPCR; Supplementary Table S2) were designed with the aid of the Primer‐BLAST tool, which is accessible via www.ncbi.nlm.nih.gov. The RT‐qPCR was carried out by a reaction mixture composed of the SYBR Green Premix Pro Taq HS qPCR Kit (#AG11701, Accurate Biology), using the LightCycler^® 96 Real‐Time PCR System (Roche, Basel, Switzerland). The RT‐PCR was carried out using a reaction mixture composed of the 2 × Taq Master Mix (Dye plus) (#P112‐AA, Vazyme, Nanjing, Jiangsu, China), by the Mastercycler^® NEXUS (Eppendorf, Hamburg, Germany). The levels of both mRNA and circRNA were calibrated against the expression of β‐actin. Each gene subjected to RT‐PCR gel electrophoresis or RT‐qPCR assay was conducted on three individual replicates.

2.14. UV cross‐linking and immunoprecipitation (CLIP) assay

The CLIP assay was performed using CLIP‐PCR Kit (#Bes3014‐1, BersinBio, Guangzhou, Guangdong, China) according to the manufacturer's instructions. In brief, 4‐Thiouridine was added to the cell culture media 16 h before cross‐linking to a final concentration of 100 µmol/L. Then, the cultured cells were placed on ice and irradiated with 0.15 J/cm² at 365 nm for 10 min using UV Crosslinker (HL‐2000 HybriLinker^TM Analytik Jena, Jena, Germany). Cell lysis buffer, DTT, and protease inhibitor was added to cell pellet and incubated on ice for 10 min with vortex. Then, 10 µg of NONO/p54^nrb Rabbit mAb (#10162S, Cell Signaling Technology, MA, US), or 10 µg of IgG antibody was added to the cell lysate, and the mixture was incubated overnight at 4°C. The protein A/G beads were added and incubated at 4°C for 3 h with rotation. Then, the proteinase K was added to digest the protein. The supernatant was collected and 0.8 mL Trizol was added to the supernatant to extract RNA. The 3’‐linker ligation mix was prepared as follows: 1 µL Poly A Polymerase (2 U/µL), 1 µL RTase Mix, 4 µL 5 × PAP/RT Buffer, 10 µL RNA template, and 4 µL ddH₂O (RNase and DNase free). The primer pairs for RT‐PCR assay were designed as follows: the forward primers were designed for every ∼100bp of the circTACC3 sequence (Supplementary Table S3) using Primer 3.0 Input (version 0.4.0, CO, US). the reverse primer was the universal tailing primer.

2.15. Western blot

Cells were lysed using RIPA buffer (Beyotime Biotechnology, Shanghai, China), and then were centrifuged at 4°C, 12,000 × g for 20 min to collect the supernatant. Protein concentrations were measured by bicinchoninic acid (BCA) assay (#23227, Pierce^TM BCA Protein Assay Kit, Thermo Scientific, MA, US). A total of 30 µg protein was loaded and subjected to electrophoresis on a polyacrylamide gel containing sodium dodecyl sulfate (SDS‐PAGE; #P2012, 10% ExpressCast PAGE, New Cell & Molecular Biotech, Suzhou, Jiangsu, China). Following electrophoresis, the protein was blotted onto a nitrocellulose (NC) (#10600001, Cytiva, MA, US) or methanol‐treated polyvinylidene difluoride (PVDF) membrane (#10600023, Cytiva). After blocking, the NC/PVDF membranes were incubated with primary antibodies (NONO/p54^nrb Rabbit mAb, 1:1000, #10162S, Cell Signaling Technology; METTL14 Rabbit polyclonal antibody, 1:1000, #26158‐1‐AP, Proteintech, Wuhan, Hubei, China; METTL3 Rabbit mAb, 1:1000, #86132T, Cell Signaling Technology; KIAA1429 Rabbit mAb, 1:1000, #88358T, Cell Signaling Technology; β‐actin Rabbit mAb, 1:1000, #4970T, Cell Signaling Technology) at 4°C overnight. Immunocomplexes were incubated with the fluorescein‐conjugated secondary antibody (#C61012‐05, mouse, Li‐Cor, NE, US; #C80416‐08, rabbit, Li‐Cor) and then detected using the Odyssey fluorescence scanner (Li‐Cor).

2.16. In situ hybridization (ISH) based on POD system and FISH

The specific oligonucleotide probes for target RNAs (Supplementary Table S4) were traced using the POD chromogenic ISH system (BOSTER, Wuhan, Hubei, China) or the FISH system (BOSTER) according to the manufacturer's instruction. In brief, the paraffin tissue sections, in vitro cultured cells and organoids were fixed for 25 min at room temperature in 4% PFA. The samples were then digested using pepsin for 45 sec at room temperature. Subsequently, the prehybridization solution was added to each section, followed by prehybridization at 40°C for 3 h in a humidity chamber with 20 mL 20% glycerol. Then the excess liquid was aspirated. The hybridization probe solution was added to the sections, followed by overnight incubation at 40°C in a humidity chamber. After washing, the sections were incubated with the biotinylated mouse anti‐digoxin at 37°C for 60 min. Subsequently, the sections were incubated with biotinylated peroxidase at room temperature for 30 min.

2.17. Fluorescence lifetime imaging microscopy‐Forster resonance energy transfer (FLIM‐FRET) assay

In the FLIM‐FRET experiments, PA and OA‐induced HepG2 and HCCLM3 cells were fixed with 4% PFA for 15 min at room temperature and permeabilized with 0.1% Triton X‐100. circTACC3 was specifically labeled with Cy3‐conjugated probes (Ex/Em: 550/570 nm, #MK11176, BOSTER), while NONO/p54^nrb was tagged using Alexa Fluor 647‐conjugated antibodies (Ex/Em: 652/668 nm, #ab150079, Abcam). Fluorescence lifetime measurements were performed on a Leica STELLARIS 5&STELLARIS 8 STED with a 100 × oil immersion objective. The spatial proximity between Cy3 and Alexa Fluor 647 was quantitatively analyzed. When their inter‐fluorophore distance fell below the Forster distance (R0, which is 4.9 nm in this study), Cy3 functioned as the energy donor, while Alexa Fluor 647 served as the energy acceptor, energy transfer between the two fluorophores will exceed the detection threshold, indicating a highly reliable direct interaction between the NONO/p54^nrb and circTACC3.

2.18. CircRNA RNase R resistance assay

Total cellular RNA was harvested, and the RNase R reaction system was as follows: 2 µL RNase R reaction buffer (#14606ES72, YEASEN), 1 µg RNA sample, 3U RNase R (20 U/µL) (#14606ES72, YEASEN), and ddH₂O was added to make up the volume to 20 µL. The mixture was digested at 37°C for 10 to 30 min (depending on the amount of total RNA), and then RNase R was inactivated at 70°C for 10 min.

2.19. Nuclear‐cytoplasmic fractionation assay

RNA was isolated individually from the cytoplasm and the nucleus using PARIS™ Kit (#AM1921, ThermoFisher, MA, US) according to the manufacturer's instructions. In brief, 1 × 10⁶ cells were collected, and 300 µL Cell Fractionation Buffer was added to the cells. The mixture was incubated on ice for 5 min, and then was centrifuged at 500 × g for 5 min at 4°C to separate the nuclear and cytoplasmic component. The supernatant (cytoplasmic component) was carefully collected, and the nucleus was precipitated. Cell Disruption Buffer was added to the nucleus precipitate and was vortexed vigorously until the lysate is homogeneous.

2.20. Immunofluorescence assay

The cultured cells or frozen sections were rinsed for 5 min in PBS. Then, the samples were fixed using 4% PFA for 20 min at room temperature, and washed 3 times with PBS. The samples were incubated for 5 min with 0.4% Triton X‐100 for permeabilization. Then, the samples were incubated with 10% goat serum in 1% BSA in PBS for 1 h to block unspecific binding of the antibodies. For immunostaining, the samples were incubated with the diluted antibody (1:100, #10162S, NONO/p54^nrb Rabbit mAb, Cell Signaling Technology; 1:500, #05‐636, γH2AX, #JBW301, Millipore Sigma; 1:100, #ab234957, S9.6 antibody, Abcam, Cambridge, UK; 1:200, #9129S, Ki‐67, Cell Signaling Technology) in 1% BSA in PBS in a humidified chamber overnight at 4°C. and then were labeled using a fluorescent secondary anti‐rabbit or anti‐mouse antibody (Li‐Cor^® IRDye^®, NE, US) at a ratio of 1:200 for 30 min at 37°C. The fluorescence was observed using STELLARIS 5 & STELLARIS 8 STED (Leica). The 3D scanning was performed after FISH or IF assay described above using STELLARIS 5& STELLARIS 8 STED (Leica). The parameters (0.4 µm/step, 2048 × 2048, speed = 200) were used to characterize the intracellular 3D spatial distribution of target RNAs or proteins.

2.21. RNA Immunoprecipitation (RIP) assay

Target proteins were specifically enriched and all RNAs interacting with them were obtained using the Magna RIP RNA binding protein immunoprecipitation Kit (#17‐700, Millipore) according to the manufacturer's instructions. In brief, 2 × 10⁷ cultured cells were lysed by freeze‐thawing once at ‐80°C using RIP Lysis Buffer. The antibody (5 µg, #202003, m⁶A antibody, Synaptic Systems; 1:50, #10162S, NONO/p54^nrb Rabbit mAb, Cell Signaling Technology) was used to encapsulate the magnetic beads. The antibody‐coated magnetic beads were mixed well with cell lysate, and were incubated overnight at 4°C. The binding complex was eluted by adding 150 µL proteinase K buffer to the magnetic beads followed by incubation at 56°C with shaking. Trizol (1 mL) was added to the supernatant to extract RNA for subsequent RT‐PCR gel electrophoresis and RT‐qPCR assay.

2.22. STM2457, SAH, DAA, and Actinomycin D treatment

Cells were seeded in the culture plates and incubated overnight (37°C, 5% CO₂). At ∼60% confluency, the medium was replaced with fresh medium containing STM2457 (#HY‐134836, MedChemExpress, at final concentration of 10 µmol/L), or SAH (#TQ0208, Targetmol, at final concentration of 20 µmol/L), or DAA (#T10111L, Targetmol, at final concentration of 20 µmol/L), or Actinomycin D (#HY‐17559, MedChemExpress, at final concentration of 5 µg/mL). For STM2457, SAH, and DAA treatment, the cells were incubated for 48 h (37°C, 5% CO₂), and for Actinomycin D treatment, the cells were incubated for 8 h (37°C, 5% CO₂) before the following assays.

2.23. DNA‐RNA immunoprecipitation (DRIP) assay

The whole genomic DNA was extracted using SteadyPure Universal Genomic DNA Extraction Kit (#AG21009, Accurate Biology), according to the manufacturer's instructions. RNase A cannot be added in this step; RNase R was selectively added to eliminate linear RNA. Then, gDNA fragmentation was performed using Covaris^®ML230 Focused‐ultrasonicator according to the 8 microTUBE‐50 AFA Fiber StripV2 250 bp protocol (temperature, 12°C; energy, 104.7 watts; number of repetitions, 33; sonication interval, 10s; PIP, 350.0; DF, 25; CPB, 1000; AIP, 87.5) and the peak of fragmentation was identified by 1.5% agarose gel electrophoresis to make sure the peak of fragmentation was at 250 bp.

CircRNA‐formed R loop samples were prepared as follows: 10 µL RNase R Reaction Buffer 15 U RNase R (20 U/µL) (#14606ES72, YEASEN), and 10 µg DNA, to remove R loops formed by linear RNAs in the genome. Negative control samples were prepared as follows: 10 µL RNase H Reaction Buffer, 10 U RNase H (5 U/µL) (#12906ES60, YEASEN), and 10 µg gDNA, to remove all R loop structures from the genome. For immunoprecipitation assay, gDNA:S9.6 antibody (#ab234957‐100 µg, Abcam) was 60 µg:15 µg, DRIP buffer was added up to 1.5 mL, and then was incubated with rotation at 4°C for 16 h. Protein A magnetic beads (#S1425S, New England Biolabs, MA, US) were added to gDNA and S9.6 antibody solution, and were incubated with rotation at 4°C for 4 h. After incubation, 234 µL Elution Buffer, 12 µL 5 mol/L NaCl and 4 µL Proteinase K (#10409ES03, YEASEN) were added to the collected magnetic beads and were incubated with rotation at 65°C for 2 h. Subsequently, all RNAs were isolated from the R loops mentioned above using the NucleoSpin^® RNA Purification Kit (NucleoSpin^®, Düren, German).

2.24. DRIP combined with chromatin isolation by RNA purification (DRIP‐ChIRP)

circTACC3 BSJ targeted or NC probe with 5’‐labeled biotins was designed (Supplementary Table S5). ChIRP was performed as follows: DEPC water of 100‐fold probe nmol/µL was used to dissolve the probes. circRNA (4 µL) or NC probe (group) solution was denatured for 3 min at 85°C, and was then transferred to an ice bath quickly. 20 µL of magnetic beads (#C0090, TargetMol) for each group (ChIRP NC or circRNA group) were prepared. For each group of samples, twice the volume of the DRIP samples in hybridization working solution was added. The DRIP samples were denatured for 10 min at 65°C on a vertical mixer. Then the probes were added, and the mixtures were hybridized at 37°C for 30 min, denatured at 50°C for 5 min, and were hybridized at 37°C for 120 min. Then, the sample‐probe mixtures were added to the magnetic beads, and were incubated on a vertical mixer at room temperature for 30 min. Sample‐bound magnetic beads were washed with wash buffer (#Bes5104‐3, BersinBio). Then, 1.5 mL of 37°C pre‐heated wash buffer was added to the beads, and the mixtures were dispensed into 600 µL and 800 µL for subsequent RNA and DNA purification, respectively. Then the RNA from the samples was purified according to the steps described in the previous section, and then the enrichment of circTACC3 was determined; ssDNA fragments from the samples were enriched using NucleoSpin^® Gel and PCR Clean‐up Kit (#740609.50, NucleoSpin^®).

The DRIP‐ChIRP‐seq library was prepared using the xGen™ ssDNA Low‐Input DNA Lib Prep (#10009817, IDT, CA, US) and xGen™ UDI Primers (#10009794, IDT) following the manufacturer's instructions. The extracted DNA for DRIP‐ChIRP‐seq was ligated to specific adaptors and then was subjected to deep sequencing on the Illumina Novaseq 6000 with 150 bp paired‐end reads. Raw data in fastq format underwent initial processing using in‐house perl scripts for quality control, resulting in clean reads by removing adapter‐containing, ploy‐N‐containing, and low‐quality reads. All subsequent analyses were conducted based on this high‐quality clean data.

2.25. High‐throughput cleavage under target and tagmentation (CUT&Tag)

CUT&Tag assay was performed as described previously with modifications [19]. Briefly, 1 × 10⁵ mock or PA and OA‐treated HepG2 cells were gently washed twice with wash buffer. Concanavalin A coated magnetic beads (10 µL, Bangs Laboratories, IN, US) per sample were added, and were incubated at room temperature for 10 min. The unbound supernatant was removed. Then, the bead‐bound cells were resuspended with Dig wash buffer and were incubated with a 1:50 dilution of primary antibody against γH2AX (#05‐636, #JBW301, Millipore Sigma) or IgG control antibody (#12‐371, normal mouse IgG, Millipore) on the rotating platform overnight at 4°C. The primary antibody was eliminated using a magnetic separator. The cells were incubated with the secondary antibody (1:100, #ab611709, Rabbit Anti‐Mouse IgG H&L, Abcam) in Dig wash buffer at room temperature for 60 min. Then, the cells were rinsed 2‐3 times using Dig wash buffer. A 1:100 dilution of pA‐Tn5 adapter complex was prepared in Dig‐med buffer, and cells were incubated with it at room temperature for 1 h. The cells were washed 2‐3 times for 5 min in 1 mL Dig‐med buffer. Then the cells were resuspended in tag mentation buffer, and were incubated at 37°C for 1 h. DNA was purified using phenol‐chloroform‐isoamyl alcohol extraction and ethanol precipitation.

To amplify the libraries, 21 µL of DNA was combined with 2 µL of a universal i5 and a uniquely barcoded i7 primer. A total of 25 µL NEB Next HiFi 2× PCR Master Mix (#E0492L, NEB, MA, US) was then added and mixed in. The sample underwent following cycling conditions in a Thermocycler with a heated lid: 72°C for 5 min (gap filling); 98°C for 30 s; followed by 14 cycles of 98°C for 10 s and 63°C for 30 s; final extension at 72°C for 1 min before being held at 8°C. Library clean‐up was carried out using XP beads (#A63881, Beckman Coulter, CA, US). The libraries’ size distribution was analyzed using the Agilent 4200 TapeStation, and was combined to ensure equal representation with a final concentration as per the manufacturer's recommendation. Sequencing was carried out on the Illumina Novaseq 6000 with 150 bp paired‐end reads following the manufacturer's protocol.

2.26. High‐through/resolution chromosome conformation capture (Hi‐C) experiment and analysis

Hi‐C libraries were constructed according to previous studies [20]. Briefly, the samples were treated with 1% PFA for 10 min at room temperature and then neutralized with 0.125 mol/L glycine for 5 min. The cells were lysed after cross‐linking, and endogenous nucleases were deactivated with 0.3% SDS. The chromatin DNA was digested using 100 U MboI (#R0147L, NEB), labeled with biotin‐14‐dCTP (#19524016, Thermo Fisher, MA, US), and ligated using 50 U T4 DNA ligase (#M0202L, NEB). Following reversal of the cross‐links, the ligated DNA was extracted using the QIAamp DNA Mini Kit (#36304, Qiagen, Hilden, Germany) as per the manufacturer's instructions. The purified DNA fragments were sheared to sizes of 300‐500 bp, and underwent blunt‐end repair, A‐tailing, adaptor addition, and purification through biotin‐streptavidin‐mediated pull‐down before PCR amplification. Finally, the Hi‐C libraries were quantified and sequenced on the Illumina Nova‐seq platform in San Diego, CA, US.

All subsequent analyses were conducted using the high‐quality clean data. The raw reads were aligned to the hg38 genome using bowtie2 and processed with the Hi‐C‐Pro pipeline (version 3.1.0, TX, US) with default parameters [21]. ICE normalization was applied to generate Hi‐C count matrices at resolutions of 100 kb, 40 kb, and 10 kb for visualization and downstream analyses. GENOVA (version 1.0.0, Oxford, UK) was utilized to identify A/B compartments, employing analysis procedures consistent with previous studies [22]. The first principal component (PC1) was used to assign compartments, A/B compartments were identified at a resolution of 100 kb. TAD boundaries were defined by insulation score analysis [23]. The insulation score was calculated using 40 kb bin size and 800 kb sliding window size (average size of TAD). The TAD calling results were plotted using R script and GENOVA. The significantly enriched contact interactions were detected using Hi‐C computational Unbiased Peak Search (Juicer HiCCUPS) (version 1.11.08, MA, US) with default settings as described in the previous study [20]. The loops were detected at 10 kb resolution.

2.27. First‐generation sequencing

Purified DNA templates (RT‐PCR amplicons) were mixed with a primer, Taq DNA polymerase (#EP0402, Thermo Scientific), deoxynucleotides (dNTPs, #AB0196, Thermo Scientific), and chain‐terminating dideoxynucleotides (ddNTPs, #03732738001, Roche) labeled with fluorescent tags. The reaction underwent thermal cycling (denaturation, annealing, extension) to generate nested DNA fragments terminating at specific bases. Products were resolved by denaturing polyacrylamide gel electrophoresis (PAGE) for 2‐8 h. Bands were visualized via laser scanning, with sequence readout determined by fragment size and terminal ddNTP identity.

2.28. Gene Ontology (GO) analysis

Differentially expressed genes were analyzed for functional enrichment using Gene Ontology databases. Gene identifiers were mapped to GO terms via clusterProfiler (version 4.16.0, Guangzhou, China). Statistical overrepresentation was calculated using a hypergeometric test with Benjamini‐Hochberg false discovery rate (FDR) correction (significance threshold: FDR‐adjusted P < 0.05). Enriched terms were filtered for relevance, clustered to reduce redundancy, and visualized using dot plots, enrichment maps to interpret biological themes.

2.29. Statistical analysis

All statistical analyses were carried out using SPSS software (version 21.0, NY, US). For continuous variables that followed a normal distribution, they were presented as mean ± standard error (SEM). Student's t‐test was employed for comparisons between two groups. When analyzing multiple groups, one‐way or two‐way ANOVA was utilized. For pairwise comparisons among multiple groups, the Tukey Honestly significant difference (Tukey‐HSD) method was applied. Regarding variables that did not follow a normal distribution, the Wilcoxon signed‐rank test was used for comparisons between two groups. For multiple‐group analyses, the Kruskal‐Wallis H test was adopted. If the P value of the Kruskal‐Wallis H test was less than 0.05, indicating that there were differences between at least two groups, further pairwise comparisons among multiple samples were then performed using the Mann‐Whitney U test, along with Bonferroni/Holm correction (to prevent Type I error). The corrected P values were reported accordingly. Using the Pearson correlation coefficient method, correlation analysis was conducted. Unless otherwise indicated, the results are presented as the mean ± SEM. Results of all statistical tests were two‐sided, and statistical significance was defined as a P value of less than 0.05. In each graph, otherwise specifically mentioned, *P < 0.05, **P < 0.01, ***P < 0.001, NS stands for “not significant”.

3. RESULTS

3.1. Landscape of the m⁶A modification and expression of circRNAs in MASH‐related HCC

Five human MASH‐related HCC tumors and paired peritumoral normal tissues were examined utilizing the m⁶A‐circRNA epitranscriptomic microarray to systematically identify m⁶A‐modified circRNAs in MASH‐related HCC (Figure 1A). Among the relative expression, relative m⁶A modification rate, and absolute m⁶A modification quantity of the circRNAs (Supplementary Figure S1A‐C), the absolute m⁶A modification quantity exhibited the most significant difference between MASH‐related HCC tumors and paired peritumoral normal tissues (Supplementary Figure S1C). A total of 1,105 circRNAs exhibited a significant increase in absolute m⁶A modification quantity (fold change > 2, P < 0.05, Figure 1B, Supplementary Figure S1C). The relative m⁶A modification rate of these circRNAs were positively correlated with their relative expression level (Pearson's R = 0.67, P < 0.001; Supplementary Figure S1D). Among these circRNAs, 38 exhibited notable increases in both the m⁶A modification and relative expression level (fold change > 2, P < 0.05; Figure 1C), and circTACC3 (hsa_circ_406433), a novel circRNA that has not been functionally characterized, exhibited the most significant increase in absolute m⁶A modification quantity (Figure 1C).

FIGURE 1.

m⁶A‐modified circTACC3 is up‐regulated in MASH‐related HCC. (A) Schematic representation of m⁶A‐circRNA epitranscriptomic microarray assay. (B) Hierarchical clustering heatmap of differentially m⁶A modified circRNAs from MASH‐related HCC tumor and paired peritumoral normal tissues (m⁶A‐circRNA epitranscriptomic microarray assay; absolute m⁶A modification quantity shown, n = 5). (C) Venn diagram of overlapping circRNAs in MASH‐related HCC tumors with concurrent increases in absolute m⁶A modification quantity, relative m⁶A modification rate, and relative expression levels. (D‐E) The circTACC3 level in MASLD tissues, MASH‐related HCC tumor tissues, and paired peritumoral normal tissues (n = 62) determined by ISH assay (D) and corresponding expression score analysis (E). **P < 0.01; ***P < 0.001; NS, not significant. (F) MeRIP assay shows the enrichment of m⁶A‐modified circTACC3 in MASLD tissues, MASH‐related HCC tumor tissues and paired peritumoral normal tissues (n = 3). (G) 3D‐FISH performed on MASH‐related HCC tumor and paired peritumoral normal tissue derived organoids. Left upper panel shows circTACC3 (red) and DAPI (blue). Left lower panel shows the merge in 3D view. Right upper panel shows representative z‐stack layer capture. Right lower panel shows depth coding. (H) Nuclear‐cytoplasmic fractionation assay determined circTACC3 expression in nuclear and cytoplasmic fractionation, respectively (n = 4). **P < 0.01; ***P < 0.001; NS, not significant. (I) Schematic representation of exon 4 back‐splicing, circTACC3 forming, and the design of indicated primers. (J) Electrophoresis of RT‐PCR product amplified from cDNA or gDNA. (K) The expression of circTACC3 and TACC3 homologous mRNA from RNA with or without RNase R treatment. (L) Levels of circTACC3 and TACC3 mRNA in indicated cells that were treated with or without actinomycin D (n = 4). ***P < 0.001. Abbreviations: m⁶A, N6‐methyladenosine; MASLD, metabolic dysfunction‐associated steatotic liver disease; MASH, metabolic dysfunction‐associated steatohepatitis; HCC, hepatocellular carcinoma; T, tumor tissues; PT, peritumoral normal tissue; ISH, in situ hybridization‐ immunofluorescence; MeRIP, methylated RNA immunoprecipitation; 3D, three dimensions; FISH, fluorescence in situ hybridization; NAS, non‐alcoholic fatty liver disease activity score; RT‐PCR, reverse transcription‐PCR.

ISH assay was performed on 38 human MASLD tissues (Supplementary Table S6, Supplementary Figure S1E), and 62 pairs of human MASH‐related HCC tumor and peritumoral normal tissues (Supplementary Table S7, Supplementary Figure S1E), as well as 63 pairs of human HBV‐related HCC tumor and peritumoral normal tissues (Supplementary Table S8, Supplementary Figure S1F). ISH assay presented the highest expression of circTACC3 in MASH‐related HCC tumor tissues (Figure 1D‐E), while circTACC3 could not be detected in both HBV‐related HCC tumors or their paired adjacent tissues (Supplementary Figures S1F‐G). Consistent with the findings of the m⁶A‐circRNA epitranscriptomic microarray analysis, the MeRIP assay revealed a high proportion of m⁶A‐modified circTACC3 in MASH‐related HCC tumors but not in paired peritumoral normal tissues or MASLD tissues with an NAFLD activity score (NAS) ≥ 6 (Figure 1F, Supplementary Figure S1H). Intriguingly, circTACC3 demonstrated significantly enhanced nuclear localization in MASH‐related HCC tumor tissues compared to paired peritumoral normal tissues, MASLD tissues, and HBV‐related HCC tumor tissues (Figure 1D, Supplementary Figure S1F and S1I). This finding was consistent with the circTACC3 nuclear distribution observed in NAC‐Organ assembled MASH‐related HCC tumor 3D organoids (Figure 1G).

3.2. circRNA characteristics of circTACC3

The adaptation of hepatocytes to lipotoxicity, along with the generation of reactive oxygen species (ROS) and the activation of DDRs induced by lipid overload, is considered a critical contributor to the tumorigenesis of MASH‐related HCC [24, 25]. Lipid overload‐induced lipotoxicity is considered to be induced mainly by free fatty acids and their metabolites [26]. A total of 9 liver cancer cell lines and 2 normal liver cell lines were stimulated with PA and OA to model lipid overload conditions and lipid metabolism patterns in MASH and MASH‐related HCC [27, 28]. The alteration of circTACC3 subcellular distribution induced by lipid overload conditions varied significantly across cell lines (Figure 1H, Supplementary Figure S2A). After PA and OA treatment, HepG2 and HCCLM3 cells presented the highest intranuclear circTACC3 levels. Huh7 and Hep3B cells, as well as the normal liver cell lines L02 and QSG‐7701, exhibited lower levels of intranuclear circTACC3 expression than HepG2 and HCCLM3 cells (Figure 1H). Moreover, other well‐established MASH risk factors, including cholesterol and fructose, as well as PA alone, were found to promote circTACC3 nuclear translocation with less pronounced effects compared to the PA and OA combination. Notably, OA alone did not induce circTACC3 nuclear localization (Supplementary Figure S2B). Additionally, we observed that the enhanced nuclear localization of circTACC3 in PA and OA‐induced HCC cell lines was abolished in Hep‐AD38 cells stably replicating HBV (Supplementary Figure S2C‐D). The variability in PA and OA‐induced circTACC3 nuclear localization among the 4 indicated liver cancer cell lines under lipid overload condition was further validated by FISH assay (Supplementary Figure S2E). The 3D‐FISH assay revealed not only increased nuclear localization but also substantial aggregation of circTACC3 under lipid overload conditions (Supplementary Figure S2F). Interestingly, PA and OA treatment increased the nuclear circTACC3 ratio in all liver cancer cell lines and normal liver cell lines (Supplementary Figure S2A). A comprehensive study of the ratio (Supplementary Figure S2A) and relative expression (Figure 1H) of circTACC3 in the cytoplasm and nucleus suggested that increased biosynthesis, nuclear retention, or possible nuclear translocation may cause its accumulation in the nucleus under lipid overload conditions.

Based on intranuclear circTACC3 expression, HepG2 and HCCLM3 cells, which have a stronger response to lipid overload, and Huh7 and Hep3B cells which have a mild response, were selected for circTACC3 characterization. The sequence of circTACC3 was found to be identical to the sequence generated by head‐to‐tail splicing of exon 4 of its host gene (Figure 1I‐J, Supplementary Figure S2G). Moreover, circTACC3 was notably resistant to RNase R digestion (Figure 1K, Supplementary Figure S2H), and had a longer half‐life than its homologous mRNA (Figure 1L).

3.3. circTACC3 functionally facilitates the growth, lipid accumulation, and adaptive survival of HCC cells under lipid overload condition in a manner depending on its expression and location

Considering that lipid overload‐induced DSBs contribute to MASH‐related HCC tumorigenesis [29], to investigate the tumorigenic function of circTACC3 in MASH‐related HCC, we treated MASH‐related HCC tumors and paired peritumoral normal tissues derived organoids with PA and OA, and an excess lipid accumulation was detected (Supplementary Figure S3A). MASH‐related HCC tumor organoids exhibited larger sizes (Figure 2A), higher proliferative activity, and lower rates of apoptosis compared to organoids derived from paired peritumoral normal tissues (Figure 2B). The MASH‐related HCC organoids with higher nucleus circTACC3 expression (#2) demonstrated superior survival ability and increased lipid accumulation under lipid overload conditions than those with lower circTACC3 expression (#1) (Figure 2A‐B, Supplementary Figure S3A‐B).

FIGURE 2.

circTACC3 m⁶A modification is associated with its intracellular localization. (A) Representative image of H&E staining of MASH‐related HCC tumor and paired peritumoral normal tissue derived organoids with relatively low (#1) and high (#2) circTACC3 expression, treated with or without PA + OA. (B) 3D fluorescence scanning of 3‐OH BrDU probe (green), anti‐Ki‐67 antibody (magenta), and DAPI (blue) indicated tissue derived organoids, treated with or without PA + OA; maximum projection of signals is shown. (C‐D) Assays of fluorescence staining with the 3‐OH BrDU probe were conducted on the cell lines with different nuclear circTACC3 level (C, n = 6) and circTACC3 knockout cell strains or negative controls (D, n = 9). The numbers on the Y‐axis represent the apoptotic rate normalized to baseline of 1. **P < 0.01; ***P < 0.001; NS, not significant. (E) MeRIP assay shows an enrichment of m⁶A‐modified circTACC3 in cytoplasmic and nuclear fraction of HCCLM3 cells. (F) Representation of predicted m⁶A modification motif of circTACC3 (predicted using the SRAMP website). (G) Absolute quantitative RT‐qPCR of m⁶A RNA shows m⁶A modification ratio of intra‐nuclear circTACC3 in the individual motif scale. (H) The m⁶A modification levels of circTACC3 were evaluated by MeRIP in m⁶A inhibitor treated group compared to control (DMSO) group, and in PA + OA treated group compared to control (Mock) group, respectively (n = 3). ***P < 0.001. (I) Nucleo‐plasmic fractionation to evaluate circTACC3 expression after indicated treatment (n = 4). ***P < 0.001. (J) 3D‐FISH conducted on MASH‐related HCC tumor tissue derived organoids demonstrates the 3D distribution of circTACC3 (red) in nuclei (blue). (K) H&E staining of MASH‐related HCC tumor tissue derived organoids treated with PA + OA and/or m⁶A intervention. Abbreviations: PA, palmitic acid; OA, oleic acid; T, tumor tissues; PT, peritumoral normal tissue; H&E, hematoxylin and eosin; MeRIP, methylated RNA immunoprecipitation; m⁶A, N6‐methyladenosine; NAS, Non‐alcoholic fatty liver disease activity score; SAH, S‐adenosylhomocysteine; DAA, 3‐deazaadenosine; FISH, fluorescence i n situ hybridization.

Based on the relative intranuclear circTACC3 level under PA and OA treatment (Figure 1H), we classified the HCC cell lines into circTACC3‐nucleus‐abundant (HepG2 and HCCLM3), nucleus‐deficient (Huh7 and Hep3B), and nucleus‐moderate (MHCC97H, SNU‐398, CSQT‐2, PLC/PRF/5, and MHCC97L) cell lines. Under PA and OA treatment, circTACC3‐nucleus‐abundant HCC cell lines exhibited reduced apoptosis (Figure 2C, Supplementary Figure S3C), greater colocalization between circTACC3 and DSB marker phosphorylation of the Ser‐139 residue of the histone variant H2AX (γH2AX) (Supplementary Figure S3D), and increased lipid accumulation (Supplementary Figure S3E), when compared to nucleus‐deficient ones or normal hepatocyte lines. We then investigated the role of circTACC3 in controlling HCC cell viability under lipid overload using circTACC3 knockout cells (circTACC3^−/−) created via CRISPR‐Cas9, which targets circTACC3 Alu elements, thereby ensuring that it does not interfere with the formation of mature linear transcripts of TACC3 (Supplementary Figure S3F‐G). Compared with wild‐type cells (circTACC3^+/+), circTACC3^−/‐ cells showed increased apoptosis and decreased lipid accumulation after PA and OA treatment (Figure 2D, Supplementary Figure S3H‐I).

3.4. m⁶A modification regulates the intranuclear localization and speckle‐like aggregation of circTACC3 in MASH‐related HCC

The occurrence of RNA hypermethylation in the context of MASLD has been reported in a previous study [30]. F2‐circTACC3 pulldown followed by western blot assays revealed that methyltransferase‐like 3 (METTL3), methyltransferase‐like 14 (METTL14), and vir‐like m⁶A methyltransferase‐associated protein (KIAA1429), the previously reported components of the m⁶A writer complex [31], could interact with circTACC3 (Supplementary Figure S4A). The expression levels of METTL3 and METTL14, which are involved in the development of various cancers [32, 33], appear to be higher in HCCLM3 and HepG2 cells than in Hep3B and Huh7 cells (Supplementary Figure S4B). Correspondingly, the MeRIP assay revealed higher m⁶A modification of circTACC3 in HCCLM3 and HepG2 cells (Supplementary Figure S4C). Furthermore, nuclear circTACC3 exhibited a greater m⁶A modification than did cytoplasmic circTACC3 (Figure 2E, Supplementary Figure S4D), indicating a correlation between circTACC3 m⁶A modification and intracellular localization. The analysis of specific m⁶A modifications of nuclear circTACC3 analysis at the individual motif level revealed 5 possible positions: four (317, 763, 918, and 1,030) had a modification rate > 95%, and position 153 has a modification rate > 70% (Figure 2F‐G).

We conducted dot blot (Supplementary Figure S4E‐F) and MeRIP assays (Figure 2H), thereby confirming that PA and OA‐induced lipid overload increased both the overall and circTACC3‐specific m⁶A modification levels, respectively. We subsequently interfered with m⁶A modification using STM2457, a METTL3 inhibitor; SAH, a METTL3‐METTL14 heterodimer complex inhibitor; and DAA, an SAH hydrolase inhibitor. Each m⁶A inhibitor decreased both the overall (Supplementary Figure S4E‐F) and circTACC3‐specific m⁶A modification levels (Figure 2H, Supplementary Figure S4G) under both normal culture and lipid overload conditions.

As expected, m⁶A interference reduced lipid overload‐induced circTACC3 nuclear localization and speckle‐like aggregation (Figure 2I, Supplementary Figure S4H‐I). The same phenomenon was confirmed in MASH‐related HCC NAC‐Organ assembled 3D organoids (Figure 2J). In addition, m⁶A modification interference under both mock and PA and OA‐supplemented culture conditions resulted in a smaller size and impaired morphology (Figure 2K), along with decreased intracellular lipid accumulation (Supplementary Figure S4J), suppressed proliferation (Supplementary Figure S4K), and increased apoptosis (Supplementary Figure S4L) in MASH‐related HCC NAC‐Organ assembled 3D organoids.

3.5. Lipid overload‐induced circTACC3‐paraspeckle assembly is essential for the nuclear localization of circTACC3 in MASH‐related HCC cells

As subnuclear structures whose formation is affected by the state of cellular stress, paraspeckles (composed of LncNEAT1 and protein components such as NONO/p54^nrb) play an oncogenic role by responding replication stress and DSBs [34]. Some types of RNAs, especially nucleus‐retained RNAs, have been shown to localize to paraspeckles [35]. However, little research has been conducted on this mechanism in relation to circRNAs. As an RNA localized to nucleus under lipid overload conditions, circTACC3 formed highly aggregated foci in the nucleus (Figure 2J, Supplementary Figure S2E‐F, Supplementary Figure S4H), suggesting that its nuclear localization might be related to paraspeckles.

F2‐circTACC3 pulldown followed by liquid chromatography‐tandem mass spectrometry (Supplementary Figure S5A‐B), and F2‐circTACC3 pulldown followed by western blot (Figure 3A) demonstrated the enrichment of NONO/p54^nrb protein by circTACC3. Meanwhile, RIP assays using an anti‐NONO/p54^nrb antibody (Figure 3B, Supplementary Figure S5C) showed the enrichment of circTACC3 in NONO/p54^nrb‐associated RNA.

FIGURE 3.

The m⁶A modification of circTACC3 regulates its interaction with NONO/p54^nrb. (A) Western blot validation of NONO/p54^nrb pulldown by F2‐circTACC3 in HCCLM3 cells. (B) RIP assay shows enrichment of circTACC3 in NONO/p54^nrb‐associated RNA in HCCLM3 cells (n = 4). ***P < 0.001. (C) FLIM‐FRET assay (left panel) and the schematic diagram of FLIM‐FRET (right panel) in PA and OA treated HCCLM3 and HepG2 cells. (D) CLIP assay followed by RT‐PCR gel electrophoresis in HCCLM3 cells. (E) The location of circTACC3 (red), NONO/p54^nrb (green), and LncNEAT1 (magenta) in nuclei (blue) was evaluated in HCCLM3 and HepG2 cells following PA + OA treatment. (F) The distribution of circTACC3 (red) in nuclei (blue) was assessed following NONO/p54^nrb or LncNEAT1 interference in PA and OA treated HCCLM3 and HepG2 cells, respectively. (G) Nucleo‐plasmic fractionation shows the altered intracellular localization of circTACC3 following NONO/p54^nrb and LncNEAT1 interference in PA and OA treated HCCLM3 and HepG2 cells (n = 4). ***P < 0.001. (H) RIP assay shows the enrichment of circTACC3 in NONO/p54^nrb‐associated RNA following indicated treatment (n = 3) in HCCLM3 and HepG2 cells. *P < 0.05; **P < 0.01; ***P < 0.001; NS, not significant. (I‐K) Representative fluorescence images show the distribution of circTACC3 (red) and NONO/p54^nrb (green) in nuclei (blue) in HCCLM3, HepG2 cells, and Hep3B cells expressed ectopic circTACC3 following indicated treatment. (L) RIP assay shows enrichment of wildtype/m⁶A modification site mutant circTACC3 in NONO/p54^nrb‐associated RNA in PA and OA treated HCCLM3 and HepG2 cells. Data normalization by dividing anti‐NONO/p54^nrb RIP and IgG RIP values by their respective Input group data (n = 3). *P < 0.05; **P < 0.01; ***P < 0.001; NS, not significant. Abbreviations: NONO/p54^nrb, non‐POU domain‐containing octamer‐binding protein; CLIP, ultraviolet cross‐linking immunoprecipitation; RT‐PCR, reverse transcription‐PCR; FLIM‐FRET, fluorescence lifetime imaging microscopy‐forster resonance energy transfer; MFD, Minimal Fraction of Donor; PA, palmitic acid; OA, oleic acid; T, tumor tissues; PT, peritumoral normal tissue; NAS, non‐alcoholic fatty liver disease activity score; SAH, S‐adenosylhomocysteine; DAA, 3‐deazaadenosine; Mut, mutation; RIP, RNA immunoprecipitation.

The fluorescence lifetime imaging microscopy‐Forster resonance energy transfer (FLIM‐FRET) assay further verified the high‐confidence direct interaction between circTACC3 and NONO/p54^nrb, especially within their highly clustered region (Figure 3C). The CLIP assay revealed 4 potential specific NONO/p54^nrb interaction fragments within positions 150‐232, 296‐351, 592‐640, and 719‐790 in circTACC3 (Figure 3D). The aggregation of circTACC3 and NONO/p54^nrb was also observed in both MASH‐related HCC tumor tissues (Supplementary Figure S5D) and tumor tissue derived organoids (Supplementary Figure S5E).

Interestingly, our observations revealed that the lipid overload induced by PA and OA markedly enhanced the nuclear colocalization of circTACC3 and NONO/p54^nrb, which is greater than other risk factors associated with MASH‐related HCC did, such as cholesterol, fructose, and the individual use of PA or OA (Supplementary Figure S6A‐C). Notably, the colocalization of circTACC3 with NONO/p54^nrb was not detected in HBV stable replication cell line Hep‐AD38 (Supplementary Figure S6D‐F).

Furthermore, we identified the interaction (Supplementary Figure S6G) between circTACC3 and LncNEAT1 (another paraspeckle biomarker), as well as the aggregation of circTACC3, NONO/p54^nrb, and LncNEAT1 signals under lipid overload conditions (Figure 3E, Supplementary Figure S6H). Correspondingly, NONO/p54^nrb and LncNEAT1 knockdown reduced lipid overload‐induced intranuclear localization and aggregation of circTACC3 (Figure 3F‐G, Supplementary Figure S6I).

3.6. m⁶A modification regulates the assembly of circTACC3‐containing paraspeckle in MASH‐related HCC cells

Among the identified m⁶A modification sites in circTACC3, positions 153, 317, and 763 were located within the circTACC3‐NONO/p54^nrb interaction fragments (Figure 2F‐G, Figure 3D). Moreover, lipid overload‐induced circTACC3 and NONO/p54^nrb aggregation was markedly hindered by m⁶A modification inhibition in MASH‐related HCC tumor derived organoids (Supplementary Figure S7A). Interference with m⁶A modification inhibited the circTACC3‐NONO/p54^nrb interaction (Figure 3H, Supplementary Figure S7B), as well as circTACC3‐NONO/p54^nrb colocalization and coaggregation (Figure 3I, Supplementary Figure S7C‐D). Additionally, in Hep3B cells with lower methyltransferase expression and circTACC3 m⁶A modification (Supplementary Figure S4B‐C), PA and OA failed to induce the colocalization of overexpressed‐circTACC3 with paraspeckles (Figure 3J, Supplementary Figure S7E‐F). However, after concurrent overexpression of METTL3 to increase m⁶A modification, PA and OA effectively increased the colocalization of overexpressed‐circTACC3 and highly clustered NONO/p54^nrb (Figure 3J, Supplementary Figure S7E‐F). Similarly, in Huh7 cells, which have higher m⁶A modification levels than Hep3B cells do, lipid overload induction notably increased the number of ectopic circTACC3 aggregates in paraspeckles (Supplementary Figure S7E‐G). These results suggested that m⁶A modification was essential for circTACC3 coaggregation with NONO/p54^nrb in MASH‐related HCC cell nucleus. Consistent with these findings, FISH‐IF (Figure 3K, Supplementary Figure S8A‐B) and RIP assays (Figure 3L, Supplementary Figure S8C) revealed that mutation of m⁶A modification sites: 153, 317, and 763, significantly inhibited the circTACC3‐NONO/p54^nrb colocalization and coaggregation.

3.7. Paraspeckles mediate the formation of nucleus‐retained circTACC3‐R loop in the genome of MASH‐related HCC

circRNAs form circR loop structures associated with DDRs in leukemia cells [18]. Here, we performed a comprehensive series of experiments to investigate the presence of the R loop structures formed by circTACC3 in MASH‐related HCC tumors genome (Figure 4A). circR loops were also detected in lipid overload induced HCC cells. These R loops exhibited resistance to RNase R digestion but were sensitive to RNase H and DNase I treatment (Figure 4B), which is consistent with previously reported characteristics of circR loops [18]. Through DRIP in human MASH‐related HCC tumor tissues (Figure 4C, Supplementary Figure S9A), and FISH‐IF assays in both human MASH‐related HCC tumor tissues and tumor tissue derived organoids (Figure 4D‐E), we further verified that highly aggregated circTACC3 participated in R loop formation. To further observe the effects of lipid overload and paraspeckles on circTACC3‐R loop formation, we constructed a vector expressing dRNase H1 fused to EGFP (dRNase H1‐EGFP, Supplementary Figure S9B); in this construct, the functional digestion enzymatic site was inactivated but the ability to recognize RNA‐DNA hybrids was retained [36]. After RNase R digestion, the circR loops were highlighted, thus more specific colocalization between dRNase H1 and circTACC3 was observed (Figure 4F‐G).

FIGURE 4.

circTACC3‐R loop structure formation in the MASH‐related HCC genome. (A) Schematic representation of procedure to identify circR loops and circTACC3‐R loops. (B) Dot‐blot assay to validate the R loop structure. S9.6 antibody was used to specifically recognize DNA‐RNA hybrids. (C) DRIP assay to identify circTACC3 enrichment in R loop structures in HCCLM3 cells (n = 4). **P < 0.01; ***P < 0.001; ND, not detection. (D) Representative fluorescence images show the expression and aggregation of circTACC3 and R loops as glow and spectrum signal intensities in MASLD tissues, MASH‐related HCC tumor tissues and paired peritumoral normal tissues, respectively. (E) The 3D distribution of circTACC3 (red) and R loops (green) were showed by 3D‐FISH‐IF in tissue derived organoids. (F‐I) Representative fluorescence images (F, H) and peak graphs of the linear ROI (region of interest) (G, I) show the location of the indicated molecules in HCC cells following PA + OA treatment. The linear ROI is represented by a solid line in the fluorescence graph. (J) The localization of circTACC3 (red) and S9.6‐stained R loops (green) were validated after NONO/p54^nrb or LncNEAT1 interference in PA and OA treated HCCLM3 and HepG2 cells, respectively. Abbreviations: PA, palmitic acid; OA, oleic acid; DRIP, DNA‐RNA immunoprecipitation; IF, immunofluorescence; FISH, fluorescence i n situ hybridization; MASLD, metabolic dysfunction‐associated steatotic liver disease; MASH, metabolic dysfunction‐associated steatohepatitis; HCC, hepatocellular carcinoma; T, tumor tissues; PT, peritumoral normal tissue; NAS, non‐alcoholic fatty liver disease activity score; NONO/p54^nrb, non‐POU domain‐containing octamer‐binding protein; SAH, S‐adenosylhomocysteine; DAA, 3‐deazaadenosine; Mut, mutation; ROI, region of interest.

Furthermore, the coaggregation of paraspeckles, circTACC3 and R loops in HCC cells was detected under lipid overload conditions (Figure 4H‐I). Correspondingly, both NONO/P54^nrb and LncNEAT1 interference inhibited lipid overload‐induced circTACC3‐R loop formation (Figure 4J, Supplementary Figure S9C‐D).

3.8. m⁶A modification triggers a positive feedback regulation loop between circTACC3‐R loop formation and circTACC3‐containing paraspeckle assembly

As a critical factor regulating circTACC3 intranuclear localization, aggregation, and paraspeckle assembly, m⁶A modification was further studied in circTACC3‐R loop formation. Under lipid overload, global m⁶A modification interfering disrupted circTACC3 enrichment in R loops (Figure 5A‐C), as did their colocalization (Supplementary Figure S9E‐G), in organoids and in vitro HCC cell models. Moreover, both in nuclear circTACC3‐deficient Hep3B cells overexpressing both METTL3 and circTACC3 and in Huh7 cells overexpressing only circTACC3, the number of lipid overload‐induced circTACC3‐R loops were increased (Supplementary Figure S9H‐J). Interestingly, when the R loops in lipid overload induced HCC cells were interfered by inducible expression of RNase H1 (Figure 5D), the nuclear localization of circTACC3 was affected (Figure 5E). Moreover, the circTACC3‐NONO/p54^nrb colocalization and coaggregation were also impeded (Figure 5F‐G). These findings indicate that the R loops are critical for maintaining the highly clustered paraspeckles and circTACC3.

FIGURE 5.

circTACC3‐R loop formation is regulated by lipid overload and m⁶A modification. (A) 3D‐distribution of circTACC3 (glow) and S9.6‐stained R loops (spectrum) in nuclei (blue) of MASH‐HCC tissue derived organoids after indicated treatment. (B‐C) DRIP assay shows the enrichment of circTACC3 in R loop structure following m⁶A modification interference in PA and OA treated HCCLM3 and HepG2 cells (n = 3). *P < 0.05; **P < 0.01. (D) Representative fluorescence images show S9.6‐stained R loops (green) in PA + OA induced HCC cells transfected with the RNase H1‐Tet‐On system after treatment with or without Dox. (E) Nucleo‐plasmic fractionation shows the altered intracellular localization of circTACC3 in PA and OA treated HCCLM3 and HepG2 cells (n = 4). ***P < 0.001. (F‐G) Representative fluorescence pictures (F) and peak graphs of the linear ROI (G) demonstrating the colocalization of circTACC3 (red), S9.6‐indicated R loops (green), and NONO/p54^nrb (yellow) in nuclei (blue) with or without Dox‐inducible RNase H1 expression in PA and OA treated HCCLM3 and HepG2 cells. Abbreviations: MASH, metabolic dysfunction‐associated steatohepatitis; HCC, hepatocellular carcinoma; PA, palmitic acid; OA, oleic acid; T, tumor tissue; PT, peritumoral normal tissue; SAH, S‐adenosylhomocysteine; DAA, 3‐deazaadenosine; DRIP, DNA‐RNA immunoprecipitation; Dox, Doxycycline; ROI, region of interest.

3.9. Mapping of circTACC3‐R loop foci by DRIP‐ChIRP sequencing

To further elucidate the pathological characteristics of circTACC3‐R loops in the MASH‐related HCC genome, circR loops were enriched by DRIP assay from lipid overload‐induced HCC cells. On the basis of this premise, we designed an oligonucleotide probe specifically targeting the BSJ region of circTACC3, and used a ChIRP assay to selectively enrich the circR loops formed by endogenous circTACC3, named “DRIP‐ChIRP” method, by which specific enrichment of circTACC3 was detected within global circR loops and circTACC3‐R loops (Supplementary Figure S10A).

Lipid overload increased the number of circTACC3‐R loops in intergenic and intronic regions (Figure 6A, Supplementary S10B‐C). m⁶A modification inhibition reduced circTACC3‐R loop formation under both normal and lipid overload conditions (Figure 6A). Among the 2,154 increased circTACC3‐R loops induced by lipid overload, 972 (45.13%) were decreased upon STM2457 treatment (Supplementary Figure S10D). The circTACC3‐R loops simultaneously regulated by lipid overload and m⁶A modification were located mainly at intronic, intergenic and promoter regions (Figure 6B). GO enrichment analysis (Supplementary Figure S10E) revealed that the bound loci of these circTACC3‐R loops included genes involved in the cellular functions of regulating intermembrane lipid transfer, intracellular lipid transport, and sterol transport, among other functions.

FIGURE 6.

DSB‐circTACC3‐R loops aggregated to promote the inter‐TADs contact. (A) After DRIP‐ChIRP sequencing, the reads distributions across peaks of all independent biological replicates are presented. (B) Genome‐wide distribution of the circTACC3‐R Loop‐located genes positively correlated with PA + OA induction or negatively correlated with m⁶A modification intervention. (C) Representative fluorescence images of the colocalization of indicated molecules in nuclei of HCC cells treated with or without PA and OA following RNase R treatment. (D) Schematic representation shows combination of DRIP‐ChIRP‐seq and γH2AX CUT&Tag‐seq to analyze the distribution of the DSB‐circTACC3‐R Loop structures in the genome. (E) Genome‐wide distribution of the DSB‐circTACC3‐R loop located genes in PA + OA induced HepG2 cells. (F) Top four enriched DSB‐circTACC3‐R loop‐binding motifs based on de novo motif analysis. (G) The dynamic clustering of paraspeckles (indicated by NONO/p54^nrb‐mCherry) were filmed using STELLARIS Dynamic Signal Enhancement 24 h after PA + OA induction at 5‐min intervals for a duration of 1.5 h. Examples (from 50 min to 85 min) of fusions of several NONO/p54^nrb‐mCherry foci are shown (time points indicated in minutes). (H) Heatmap depicting the fold change(log₂) in Hi‐C contact frequencies between PA + OA‐treated and control cells throughout chromosome 7. Interactions that increase in PA + OA group (red) or decrease in Mock group (blue) are evident. Profile of DRIP‐ChIRP‐seq and γH2AX CUT&Tag‐seq are shown on the top. TADs that had higher inter‐TADs contact frequencies (named “contact‐elevated TADs”) in both long‐range (green box) and between adjacent TADs (red box) are marked. DSB‐circTACC3‐R loops are marked with red arrow. (I) Hi‐C maps around the human STX6 locus that formed DSB‐circTACC3‐R loop structure are shown. DSB‐circTACC3‐R loops are marked with red arrow. Abbreviations: DRIP, DNA‐RNA immunoprecipitation; ChIRP, chromatin isolation by RNA purification; γH2AX, Ser‐139 residue of the histone variant H2AX; CUT&Tag, cleavage under targets and tagmentation; IF, immunofluorescence; FISH, fluorescence i n situ hybridization; PA, palmitic acid; OA, oleic acid; Hi‐C, high‐throughput/resolution chromosome conformation capture; DSB, DNA double‐strand breaks; STX6, Syntaxin 6.

3.10. circTACC3 forms circR loops at DSB foci

Previously, we demonstrated that intranuclear clustered circTACC3 was observed at DSB foci after lipid overload induction, and the degree of colocalization of circTACC3 and DSB foci indicated the survival of HCC cells under lipid overload conditions (Figure 2C, Supplementary Figure S3D). Pathological R loop formation may cause DNA damage, and ultimately results in genome instability [37]. Conversely, DNA breakage may activate the formation of pathological R loops [38]. Based on the verified physical colocalization of paraspeckles and circTACC3‐R loops, we further explored the relationships between these structures and DSBs.

We observed that in HCC cells, paraspeckles, circTACC3‐R loops and DSB foci showed highly clustered colocalization after lipid overload induction (Figure 6C, Supplementary Figure S11A). In addition, via CUT&Tag seq, a global increase in DSBs induced by PA and OA was observed, with the most significant changes occurring in promoter (Supplementary Figure S11B‐D). CUT&Tag‐seq data was then analyzed in combination with DRIP‐ChIRP‐seq data (Figure 6D). A total of 38 lipid overload‐induced circTACC3‐R loops located at DSB foci were identified, and named “DSB‐related and circTACC3‐formed circR loops” (DSB‐circTACC3‐R loops). Among these DSB‐circTACC3‐R loops, the most significant changes in colocalization with DSBs were detected in promoter (Figure 6E). Furthermore, the 4 most enriched motifs that bind with DSB‐circTACC3‐R loops were revealed, all of which were highly complementary to the circTACC3 sequence (Figure 6F).

3.11. DSB‐circTACC3‐R loops aggregate to promote inter‐TAD contacts

To further explore the specific function of circTACC3 in the genome organization, we analyzed the 3D genome organization via Hi‐C data generated from lipid overload‐induced HCC cells. Lipid overload did not notably alter genome‐wide chromosome interactions (Supplementary Figure S11E) or trigger major changes in genome compartmentalization (Supplementary Figure S11F). However, lipid overload induction resulted in a greater interaction within the compartments, particularly between inter‐A compartments and between A compartments and B compartments (Supplementary Figure S11G‐H).

The highly clustered colocalizations formed by paraspeckles, circTACC3‐R loops and DSB foci after lipid overload induction could interact with more than one other highly clustered colocalized region (Figure 6C). This phenomenon was then dynamically visualized through live imaging (Figure 6G, Supplemental Videos S1‐S3) in NONO/p54^nrb‐mCherry HepG2 cells (Supplementary Figure S11I). Analysis of the Hi‐C data integrated with γH2AX CUT&Tag‐seq and circTACC3‐R loop DRIP‐ChIRP‐seq datasets further revealed that DSB‐circTACC3‐R loops might positively influence this process. Upon focusing on the TADs scale, we observed that DSB‐circTACC3‐R loops physically colocalized with TADs exhibiting higher inter‐TAD contact frequencies, both within the long‐range and adjacent TADs (Figure 6H); we called these TADs as “contact‐elevated TADs”. Consistently, TAD fusion was also observed at DSB‐circTACC3‐R loop‐localized foci after lipid overload induction, which was defined as a TAD boundary loss resulting from increased inter‐TAD interactions between adjacent TADs (Figure 6I).

3.12. DSB‐circTACC3‐R loop‐localized genes are selectively activated

Primers targeting the corresponding transcripts of 38 DSB‐circTACC3‐R loop‐localized genes were designed for RT‐qPCR analysis, and 35 of the 38 genes presented higher expression in lipid overload‐induced HepG2 cells than in the mock group (Figure 7A‐B). Interestingly, some of the DSB‐circTACC3‐R loop‐localized genes were annotated to be involved in the regulation of proliferation (such as PDIA2 and STX6), lipid metabolism (such as OSMR, OGDH, and STAR), and the DDR (such as ZBTB46, TRAF3IP2, and Y_RNA) (Figure 7A‐B).

FIGURE 7.

DSB‐circTACC3‐R loop‐localized genes are selectively activated. (A) List of DSB‐circTACC3‐R loop‐localized genes. (B) DSB‐circTACC3‐loop‐localized genes expression in HepG2 cells with/without lipid overload induction (n = 4). *P < 0.05; **P < 0.01; ***P < 0.001; NS, not significant. (C‐D) DRIP‐ChIRP‐seq (C) and γH2AX CUT&Tag‐seq (D) RPKM analysis of DSB‐circTACC3‐loop‐localized genes to compare circTACC3‐R Loop enrichment within the “contact‐elevated TADs” (n = 26) or not within the “contact‐elevated TADs” (n = 12). *P < 0.05; **P < 0.01; ***P < 0.001; NS, not significant. (E‐F) Representative fluorescence images (E) and peaks graphs of the linear ROI (F) show the colocalization of indicated molecules with or without Dox‐inducible RNase H1 expression. (G) NONO/p54^nrb‐mCherry HepG2 cells with or without Dox‐inducible RNase H1 expression were filmed 24 h after PA + OA induction at 5‐min intervals. Abbreviations: PA, palmitic acid; OA, oleic acid; TAD, topologically associated domain; γH2AX, Ser‐139 residue of the histone variant H2AX; CUT&Tag, cleavage under targets and tagmentation; RPKM, reads per kilobase per million mapped reads; Dox, Doxycycline; ROI, region of interest.

Then, we initially concentrated on ENSG00000260496 and VCL, which demonstrated the most significant upregulation trend among DSB‐circTACC3‐R loop‐localized genes (Figure 7A‐B). In lipid overload‐induced HCC cells, we intervened in these genes (Supplementary Figure S12A) and observed a significant reduction in cell proliferation and lipid accumulation, and an increase in cell apoptosis (Supplementary Figure S12B‐D). ENSG00000260496 and VCL have not been previously documented as being associated with the initiation and progression of HBV‐related HCC, it is likely that these two genes are specific to the MASH‐related HCC progression. In addition, we discovered that STX6 (Figure 7A‐B, the 6th elevated gene within contact‐elevated TADs), a gene that has been previously associated with the progression of HCC [39], was transcriptionally activated in response to the lipid overload‐induced chromatin remodeling mediated by circTACC3. The interference of STX6 consistently led to the promotion of apoptosis and the inhibition of HCC cell proliferation (Supplementary Figure S12A‐C); however, interference of STX6 had minimal effects on lipid accumulation (Supplementary Figure S12D), indicating that circTACC3 may selectively regulate particular downstream genes, facilitating the malignant progression of MASH‐related HCC. Furthermore, we observed that the number of genes localized within the DSB‐circTACC3‐R loop in “contact‐elevated TADs” was significantly greater than those outside these TADs (Figure 7B), indicating that these genes were selectively activated in response to lipid overload‐induced TADs contact elevating.

3.13. circTACC3‐R loops contribute to inter‐TAD contacts and circTACC3 clustering at DSB foci

Among the 38 DSB‐circTACC3‐R loops, the portion localized within contact‐elevated TADs exhibited a significantly higher frequency of R loop formation compared to the non‐localized portion under lipid overload conditions (Figure 7C). This difference in circTACC3 R loop enrichment between the contact‐elevated and contact‐non‐elevated TADs disappeared after m⁶A intervention and highlighted under lipid overload conditions (Figure 7C). However, DSB signals were not notably different between these two regions (Figure 7D). Inducible eliminating R loop structures inhibited the aggregation of circTACC3 at DSB foci (Figure 7E‐F). Moreover, live imaging revealed the loss of both DSB‐circTACC3‐R loop and NONO/p54^nrb aggregation after the elimination of the R loops (Figure 7G, Supplemental Videos S4‐S5). This pattern suggested that the circTACC3‐R loop structure is the key factor in increasing inter‐TAD contacts under lipid overload conditions.

4. DISCUSSION

In this study, we identified circTACC3 as the most prevalent m⁶A‐modified circRNA in MASH‐related HCC, significantly influencing its malignant progression. Under lipid overload, circTACC3 interacted with NONO/p54^nrb to form intranuclear paraspeckles in an m⁶A modification‐dependent manner. The circTACC3‐containing paraspeckles were recruited to DSB foci to form DSB‐circTACC3‐R loops, thereby promoting TAD contact and fusion, and selectively activate genes related to the malignant phenotype of MASH‐related HCC. Notably, circTACC3‐R loops positively regulate circTACC3‐paraspeckle assembly and TAD clustering.

The increasing prevalence of MASH‐related HCC represents a substantial clinical challenge, primarily due to the absence of reliable diagnostic biomarkers and effective therapeutic strategies for this HCC subtype. Our study identifies circTACC3 as a nuclear‐enriched circRNA that is specifically upregulated and highly m⁶A‐modified in MASH‐related HCC tumors but not in HBV‐HCC, thereby highlighting its etiological specificity. Functionally, MASH‐related HCC tumor derived organoids with high nuclear expression of circTACC3 demonstrated enhanced proliferation, reduced apoptosis, and exacerbated lipid accumulation under lipid overload condition. Notably, both knockout and m⁶A modification intervention of circTACC3 reversed these malignant phenotypes, confirming the critical role of m⁶A‐modified circTACC3 in promoting MASH‐related HCC malignant progression. These findings establish both expression level and m⁶A modification status of circTACC3 as a promising diagnostic marker and therapeutic target for MASH‐related HCC.

Lipid overload induces oxidative stress, DNA damage, and mutations, which serve as premalignant events conferring selective advantages to hepatocytes during the development of MASH‐related HCC [24, 26, 40]. Moreover, paraspeckle assembly has been shown to be regulated by cellular stress [35, 41]. Consistent with these findings, our study demonstrated that PA and OA‐induced lipid overload significantly promoted circTACC3‐paraspeckle assembly in an m⁶A‐modified dependent manner. We observed that circTACC3, NONO/p54^nrb, and LncNEAT1 colocalized in the nuclei of MASH‐related HCC tumor cells, organoids, and lipid overload‐induced HCC cells. Lipid overload directly regulates circTACC3's functional properties, conversely, circTACC3 regulates lipid accumulation in hepatocytes, thereby promoting lipid‐dependent MASH‐related HCC progression. This reciprocal interplay highlights circTACC3 as both a mediator and effector of lipid‐driven pathogenesis.

Previous studies have shown that circRNAs achieve functional diversity through their specific subcellular localization [8, 13]. While the mechanisms underlying circRNA transport from the nucleus to the cytoplasm have been elucidated [13], the precise mechanisms governing the nuclear retention or transport of certain exon‐derived circRNAs remain to be further explored. Some nuclear‐resident ncRNAs, such as lncRNAs, localize to subnuclear paraspeckles to maintain their nuclear residency [35]. Based on the confirmed direct interactions between circTACC3 and the key paraspeckle scaffold protein NONO/p54^nrb, we further proved that circTACC3 could be incorporated into paraspeckles for nuclear localization and aggregation in MASH‐related HCC cells. Furthermore, while intervention in m⁶A modification markedly inhibits circTACC3 aggregation and its colocalization with paraspeckles, it exerted minimal effects on paraspeckle aggregation, suggesting that circTACC3 may function more as a “passenger” rather than a “scaffold” during lipid overload‐induced paraspeckle formation.

circRNAs have been shown to participate in the formation of R loops within the leukemia cell genome [18]. Here, we demonstrated that intranuclear circTACC3 aggregates at DSB foci and forms R loops in MASH‐related HCC. Despite prior research suggesting that m⁶A alteration may substantially affect R loop development [42, 43], its processes remain unclear. In this study, circTACC3‐R loops were found to be lipid‐induced and m⁶A‐dependent. We observed coaggregation of circTACC3‐R loops with NONO/p54^nrb and LncNEAT1 at DSB foci under lipid overload condition. Inhibition of m⁶A modification and interference of NONO/p54^nrb, LncNEAT1 significantly suppressed circTACC3‐induced R loop formation, indicating m⁶A modification and paraspeckles regulate lipid overload‐induced R loops formation. Whether this regulatory role applies to other RNA types requires further study.

DSBs constitute a highly toxic form of genomic damage that can induce chromosomal translocations or structural remodeling [44, 45], thereby facilitating the malignant transformation of tumor cells [46, 47]. In this study, we identified highly clustered DSB‐circTACC3‐R loop foci that form structures similar to the recently reported damaged TAD clusters at R loop‐enriched foci, termed D compartments [15]. Specifically: (1) Both structures exhibit responsiveness to DSBs; (2) The formation of both structures relies on R loop architecture; (3) Both structures promote transcriptional activation at target gene loci through chromatin structure remodeling and regulation of TAD interactions

Nuclear‐enriched ncRNAs are central regulators of chromosome architecture, and can form highly concentrated regions, shape long‐range DNA contacts [48], and promote DSB focus formation and fusion [49]. Furthermore, nuclear‐enriched ncRNAs form R loops as spatial organizers that promote the gene activation [15, 48]. Here, we revealed the increased contact between two long‐range and adjacent DSB‐circTACC3‐R loop foci further promoted inter‐TAD contact and fusion of adjacent TADs, and the transcription levels of DSB‐circTACC3‐R loop‐localized genes increased under lipid overload. Moreover, DSB‐circTACC3‐R loop‐localized genes within contact elevated TADs presented more significant increases at the transcriptional level than those that were not localized. Therefore, building upon the discovery that circTACC3 can form a unique circR loop structure, we further elucidated the remodeling effects of this structure on the MASH‐related HCC tumor genome, as well as its specific biological functions during MASH‐related HCC malignant progression.

Note that this work still had limitations. First, while hepatoma cell lines are a practical model for mechanistic exploration, advanced systems (e.g., genetically engineered murine models, primary MASH‐related HCC cells) are needed to fully recapitulate disease complexity. Second, beyond the roles of PA and OA, the contributions of other risk factors (e.g., fructose, cholesterol) to MASH‐related HCC progression require further investigation. Finally, although ENSG00000260496 and VCL appear specific to MASH‐related HCC, their functional significance and therapeutic potential necessitate additional validation. We plan to address these gaps through expanded models and mechanistic studies to refine therapeutic targeting strategies in the following study.

5. CONCLUSIONS

Our findings demonstrated that m⁶A modification‐dependent circTACC3‐paraspeckle assembly forms R loops at DSB foci to facilitate MASH‐related HCC malignant progression. Elucidating this mechanism is expected to reveal new targets and treatment options.

AUTHOR CONTRIBUTIONS

Hongyang Wang contributed to study concept design and supervised all works. Jingbo Fu, Yanping Wei, and Liang Li contributed to the analysis and interpretation of data and drafting of the paper. Jingbo Fu, Yun Yang, Xinwei Yang and Yanping Wei contributed to the acquisition of data. Yun Yang, Xinwei Yang, Xianming Wang, Shuzhen Chen, Jing Fu, Pinhua Yang, and Liang Li contributed to the acquisition of clinical samples. Tao Ouyang, Zenglin Liu, Yu Su, Miao Yu, Haihua Qian, Hao Song, Shuo Xu, Ru Zhao, Xue Jiang, Zhao Yang, Yunfei Huo, Man Zhang, and Kui Wang contributed to technical and material support. All authors have read and approved the article.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no competing interests.

ETHICS APPROVAL AND CONSENT TO PARTICIPATE

All research complied with the principles of the Declaration of Helsinki. Patient samples were utilized following criterion approved by the Ethics Committee of Eastern Hepatobiliary Surgery Hospital (Approval No. EHBHKY2019‐01‐001) and the Ethics Committee of The First Affiliated Hospital of Shandong First Medical University (Approval No. YXLL‐KY‐2024[095]), respectively.

Supporting information

Supporting information

Supporting information

Supporting information

Supporting information

Supporting information

Supporting information

Fu J, Wei Y, Yang Y, Yang X, Ouyang T, Wang X, et al. Intranuclear paraspeckle‐circular RNA TACC3 assembly forms RNA‐DNA hybrids to facilitate MASH‐related hepatocellular carcinoma growth in an m⁶A‐dependent manner. Cancer Commun. 2025;45:1583–1610. 10.1002/cac2.70061

DATA AVAILABILITY STATEMENT

All m⁶A‐circRNA epitranscriptomics microarray data in this study have been deposited at the National Genomics Data Center under the accession number GSE271997. All DRIP‐ChIRP‐seq data in this study have been deposited at the National Genomics Data Center under the accession number GSE272486. All CUT&Tag‐seq data in this study have been deposited at the National Genomics Data Center under the accession number GSE272542. All Hi‐C‐seq data in this study have been deposited at the National Genomics Data Center under the accession number GSE273061. Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Prof. Hongyang Wang (hywangk@vip.sina.com).