Exploration of the mechanism of 5-Methylcytosine promoting the progression of hepatocellular carcinoma

Highlights

• •In this study, we conducted MeSeq on hepatocellular carcinoma and adjacent tissues, comprehensively describing the map and significance of m5C modification in hepatocellular carcinoma.
• •Based on multi omics analysis, we constructed a meaningful M5C related HCC subtype and identified its clear functional differences. The differences in the immune and tumor pathways of this subtype may indicate the mechanism of poor prognosis in HCC.
• •We attempted to explore the mechanism of m5C modification promoting the progression of hepatocellular carcinoma using RIP, molecular experiments, and molecular interaction analysts. We identified the regulatory axis of m5C modification in HCC. m5C reader YBX1 binds to TPM3 in an m5C dependent manner and promotes HCC progression.

Abstract

5-Methylcytosine (m5C) is a ubiquitous RNA modification that is closely related to various cellular functions. However, no studies have comprehensively demonstrated the role of m5C in hepatocellular carcinoma (HCC) progression. In this study, six pairs of HCC and adjacent tissue samples were subjected to methylated RNA immunoprecipitation sequencing to identify precise m5C loci. Non-negative matrix factorization (NMF) was used to identify HCC subtypes in TCGA-LIHC cohort. Immune, metabolic, and tumor-related pathways in HCC subtypes with differences in methylation status were analyzed and a prognostic model based on m5C-related genes was constructed. Finally, using RIP and molecular interaction analysis, we demonstrated that YBX1 binds to TPM3 in an m5C dependent manner and regulates HCC progression. Widespread m5C sites were identified and found to be differentially distributed in HCC compared with adjacent tissues. Metabolic processes were inhibited in hypermethylated HCC, whereas immune checkpoint and multiple classical tumor pathways were significantly upregulated. More importantly, we have identified an m5C dependent regulatory axis. The m5C reader YBX1 binds to TPM3 in an M5C dependent manner and promotes the progression of hepatocellular carcinoma. These results provide new evidence for further understanding the comprehensive role of m5C in HCC and the regulatory mechanism of m5C.

Introduction

Hepatocellular carcinoma (HCC) is the fourth most common malignant tumor type worldwide, with an extremely poor prognosis, resulting in greater than 800,000 deaths each year [1,2]. Despite the development of multiple treatment modalities, such as surgery, radiotherapy, chemotherapy, targeted therapy, and immunotherapy, which have achieved breakthrough progress in recent decades, the prognosis of HCC patients remains poor due to a low early detection rate, a high rate of HCC, and complex underlying molecular mechanisms [[3], [4], [5], [6]]. Therefore, improving our understanding of HCC pathogenesis and identifying new therapeutic targets for the effective intervention of HCC has the potential to significantly improve patient outcomes.

In recent years, a large body of literature has demonstrated that abnormalities in epigenetic regulatory mechanisms, such as post-transcriptional modifications, play key roles in tumor progression [7]. In particular, the regulatory role of RNA methylation has attracted the attention of researchers. One of the best-studied RNA methylation, 5-methylcytosine (m5C), has been associated with a tumor-promoting role in various malignancies, including bladder cancer and esophageal cancer [8,9]. m5C is a post-transcriptional modification that has been widely detected in messenger RNA (mRNA), translational RNA (tRNA), and ribosomal RNA (rRNA) in mammals, with dynamic, conserved, and tissue-specific features [[10], [11], [12], [13]]. m5C has been shown to play irreplaceable roles in stable tRNA translation and rRNA processing [[14], [15], [16]]. Recent studies have found that m5C is involved in the nuclear export of mRNA and promotes mRNA stability through its regulators Aly/REF export factor (ALYREF) and Y-box binding protein (YBX1), respectively [8,17]. Sun et al. demonstrated that the m5C regulator, NOP2/Sun domain family, member 2 (NSUN2), promotes HCC progression in an m5C-dependent manner [18]. The widespread distribution of m5C throughout cells suggests that this modification has extensive effects on cellular biological behaviors. We speculate that m5C modifications likely alter the biological behaviors of tumor cells through multiple pathways, while research in this area remains relatively scarce.

Although some bioinformatics analyses have been performed examining m5C regulators, the conclusions of these studies are of limited value due to the functional non-specificity of the examined regulators. Therefore, in this study, we examined m5C using methylated RNA immunoprecipitation sequencing (meRIP-Seq) in paired HCC and adjacent tissue samples. The m5C sites were identified and compared between paired samples. We performed a combined analysis including both the methylome and transcriptome to screen for the most prominent m5C regulatory RNAs. We subsequently performed non-negative matrix factorization (NMF) typing of identified RNAs using public datasets, resulting in the identification of two HCC subtypes with significantly different methylation patterns. These subtypes display the differential activation of metabolic, transcriptional regulatory, immune, and various tumor pathways, supporting the extensive impacts of m5C on cell function. The prognosis model based on m5C target genes has effective prognosis and predictive performance. More importantly, we identified an m5C dependent regulatory axis using methods such as RNA immunoprecipitation and molecular interaction analysis. The m5C reader YBX1 binds to TPM3 in an m5C dependent manner and promotes the progression of hepatocellular carcinoma. We believe that our study is of great value for guiding further explorations of the impacts of m5C in HCC and will contribute to the identification of new m5C-based therapeutic targets for HCC.

Materials and methods

Methylated RNA immunoprecipitation sequencing

We collected six pairs of HCC and adjacent tissue samples from the First Affiliated Hospital of Zhengzhou University. The project has been approved by the institution's ethics committee. Total RNA was extracted with TRIzol reagent (Invitrogen Corporation, CA, USA), and rRNA was removed using the Ribo-Zero rRNA Removal Kit (Illumina, Inc., CA, USA). MeRIP-Seq was performed as previously described [19]. Briefly, the GenSeq™ m5C RNA IP Kit (GenSeq Inc, China) was used to cleave RNA into 100-bp fragments, followed by immunoprecipitation with m5C antibody according to the manufacturer's recommendations. An NEBNext® Ultra II Directional RNA Library Prep Kit (New England Biolabs, Inc, USA) was used for RNA library construction. The Illumina HiSeq instrument was used to sequence cDNA with 150-bp paired-end reads (Fig. 1b).

Fig. 1.

a Flowchart of the study. b Schematic of methylated RNA immunoprecipitation sequencing (meRIP-Seq).

After quality assessment, using a threshold of Q30>80 %, we used MACS software to identify m5C peaks in mRNAs [20]. Fold change (FC) >2 or <0.5 (p ≤ 0.00001) in the m5C peak were considered to indicate methylation up- or downregulation, respectively. Dreme software was used to scan the sequences of methylation peaks, including 50 bp on each side of the peak [21].

Combined methylomics and transcriptomics analyses

Principal component analysis (PCA) was performed using the R package “prcomp.” We identified 172 genes with significant differences in both the methylome and transcriptome (p < 0.00001, FC>2). Gene set enrichment analysis (GSEA) was performed using the R software package “GSEA.” Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes analyses were performed using the R package “clusterProfiler” (p < 0.05). In addition, we performed protein–protein interaction (PPI) network analysis on identified genes using Search Tool for the Retrieval of Interacting Genes/Proteins (STRING), with results modified in Cytoscape software for better visualization. The MCODE plugin was used to screen PPI modules with high connectivity. The Node Score Cutoff was set to 0.2, and K-Core was set to 2.

Database processing, identification, and functional analysis of HCC molecular subtypes based on m5C-related genes

Transcriptome data and clinical follow-up information for The Cancer Genome Atlas HCC data set (TCGA-LIHC) were obtained from TCGA on November 9, 2021. A total of 50 normal tissues and 371 HCC tissues were included in the obtained dataset. The 172 identified m5C-related genes were subjected to univariate Cox regression analyses to identify potential prognosis-related genes. We performed NMF on HCC samples in the TCGA-LIHC cohort using the R package “NMF” based on the identified prognostic m5C-related genes. We chose the standard “brunet” and performed 100 iterations. The number of clusters was set from 2 to 6. According to indicators, such as cophenetic correlations, we determined that the optimal cluster number was 4. In addition, single-sample GSEA (ssGSEA) analysis was performed between groups using the R package “GSVA.” Kaplan–Meier (KM) analysis was performed based on the R package "survival" . We calculated immune infiltration scores and cytological pathway scores between the different subtypes, which were normalized for further analysis.

Construction and validity evaluation of prognostic model based on differentially expressed genes

We performed differential expression analysis between clusters 2 and 3 using the R package “limma.” Subsequently, differentially expressed genes (DEGs) were subjected to univariate Cox regression analysis to identify genes with prognostic significance using the function “coxph.” Due to the large number of genes included in the analysis, we utilized the R package “glmnt,” followed by dimensionality reduction using the least absolute shrinkage and selection operator (LASSO) algorithm to identify significant hub genes. The algorithm can obtain a more refined model by constructing a penalty function, which is a biased estimation for dealing with data with complex collinearity and can better solve the multicollinearity problem in regression analyses.

The hub genes were subjected to multivariate Cox analysis by the R function “coxph,” and we determined the coefficients of hub genes. We calculated the product of each gene and its coefficient, and summed all the products to calculate the risk score for each sample. Based on the median risk score, we divided the TCGA-LIHC samples into high-risk and low-risk groups. Receiver operating curve (ROC) analysis was performed using the R package “timeROC.” Nomograms were analyzed using “rms.” All analysis workflows are shown in Fig. 1A.

Construction of YBX1 stable knockout cell line

The hepatocellular carcinoma cell line HepG2 was purchased from ATCC (HB-8065). Cell culture was performed at 37 °C and 5 % CO2 using Dulbecco's modified Eagle's medium (DMEM) containing 10 % fetal bovine serum (HyClone, USA), 100 units/ml penicillin and 100 mg/ml streptomycin. Lentivirus (Lv-ALKBH5) expressing YBX1 was constructed using pGMLV-PA6 vector (Genomedtech, Shanghai, China). The shYBX1 targeting shRNA was constructed by pGMLV-SC5RNAi lentiviral vector (Genomedtech).

Cells were plated at 5 × 10⁵ cells per well in a 6-well plate and cultured in antibiotic-free medium. Plasmids were transfected using Lipofectamine 8000 (Biyuntian, China) in serum and antibiotic-free DMEM at a ratio of 1:2 (m/v). After 20 min incubation at room temperature, the mixture was added to serum and antibiotic free culture media. After 6 h of transfection, replace the culture medium with complete DMEM. Stable transfected cells are selected using puromycin (Sigma Aldrich, St. Louis, Missouri, USA).

WB, qRT-PCR and IHC

Total RNA was extracted from cells using the TRIzol method according to the manufacturer's instructions. Real-time quantitative RT-PCR reactions were performed using the CFX Connect TM real-time system (Bio Rad) and SYBR Green Master Mix (Vazyme, China). The relative RNA expression level was calculated using the 2-ΔΔ Ct method and standardised to GAPDH mRNA. The specific primers for each target gene are listed in the Supplementary Table 1.

Harvest the cells and lyse them in RIPA lysis buffer. Centrifuge the whole cell lysate and transfer it to the PVDF membrane using an electroblotting device. Incubate the membrane overnight at 4 °C with specific primary antibodies. Observe the bands using ECL detection reagent (Beyo time). Using Image Lab software (Bio Rad, Hercules, CA, USA), GAPDH was used as an internal reference to analyse density quantification. All antibodies used in this study were purchased from Abways. For the IHC assay, paraffin sections were incubated with primary antibodies against YBX1 at 37 °C for 60 min, secondary antibodies at 37 °C for 15 min and horseradish enzyme-labelled streptavidin solution for 10 min and then stained with DAB and haematoxylin.

CCK-8, clone formation

Cell viability was measured using the CCK-8 assay. HepG2 cells were treated at 2 × 10³ cells per well in a 96-well plate inoculated with HepG2 cells and incubated for 96 h. Then 10 µl of CCK-8 reagent (Beyotime, China) was added to each well and incubated at 37 °C for 1 hour. Cell viability was assessed by measuring the optical density at 450 nm after 24, 48, 72 and 96 h of incubation.

Cells were inoculated at a density of 400 cells per well in a 6-well plate and cultured for 14 days. Fix the colony in methanol for 30 min and stain with crystal violet for 15 min. Wash cells with phosphate buffered saline, photograph and count under a light microscope.

RNA immunoprecipitation sequencing

GenSeq ® The RIP Assay Kit was used for RNA immunoprecipitation according to the instructions. The precipitated RNA and input RNA were passed through the GenSeq ® RRNA Removal Kit to remove rRNA. RNA samples after rRNA removal were passed through the GenSeq ® Low Input RNA Library Prep Kit to construct an RNA sequencing library. A BioAnalyzer 2100 instrument (Agilent Technologies) was used for library quality control and quantification. 150 bp double-ended sequencing was performed on an Illumina NovaSeq instrument.

Q30 was used for quality control. Cutadapt (v1.9.3) software was used to remove splices, subsequently low quality reads were removed and high quality reads were obtained. Hisat2 software (v2.0.4) was used to align pure reads to the human reference genome (HG19).MACS software was used to identify genes in each RIP and igG control sample. Genes were subjected to GO and pathway analysis.

Molecular interaction

Biacore is a commonly used method for molecular interaction analysis based on the principle of optical surface plasmon resonance. Use the Biocore X100 molecular interaction analyzer system and CM5 sensing chip to detect the dynamics and affinity data of TMCO3, MEX3A, and TPM3 RNA molecules binding to proteins with/without m5C modification. PBS-P+coupling buffer was used to immobilize YBX1 purified protein (Lianmai Biotech, China) onto the CM5 sensing chip. Using artificially synthesized TMCO3, MEX3A, and TPM3 RNA molecules with/without m5C modification flowing through the chip pipeline, multi cycle kinetics mode was selected for detection, and Affinity mode was used to fit the results to obtain affinity data.

Results

Differential distribution of m5c sites in HCC and adjacent tissues

We identified 47,076 m5C sites in HCC tissues and 30,873 m5C sites in adjacent tissues, including 20,849 sites specific to HCC, 4646 sites specific to adjacent tissues, and 26,227 sites found in both HCC and adjacent tissues (Fig. 2a). These results suggest that m5C sites are widely distributed throughout HCC and adjacent tissues and may be involved in a wide range of cellular biological functions. These results also revealed that m5C was greatly upregulated in HCC, suggesting that m5C plays a role in the evolution of normal tissues to HCC tissues. The PCA results showed that both m5C methylation and transcriptomics data were able to significantly distinguish between HCC and adjacent tissue samples, showing the intergroup stability of methylomes (Fig. 2b-c).

Fig. 2.

m5C sites are differentially distributed in hepatocellular carcinoma (HCC) and adjacent tissues. a Significantly more m5C sites were found in HCC than in adjacent tissues. b Principal component analysis (PCA) of HCC and adjacent tissues by methylomic data. c PCA of HCC and adjacent tissues by transcriptomic data. d Most significant motifs for mRNA in HCC and adjacent tissues. e Most significant motifs for lncRNA in HCC and adjacent tissues. f Most significant motifs for circRNA in HCC and adjacent tissues. g Volcano map of differentially methylated genes in mRNA. h Volcano map of differentially methylated genes in lncRNA. i Volcano map of differentially methylated genes in circRNA.

m5C motifs show concordance in HCC and adjacent tissues

We identified the motif with the most significant ES values for mRNA, long noncoding RNA (lncRNA), and circular RNA (circRNA) (Fig. 2d, f). Combined with our previous research, we found that among mRNAs, the most significant motifs associated with m5C was “CUWCM”, “GWRGA” and “ADUACA”, in HCC tissues, whereas the most significant motif was identified as “CUWCV”, “GGADRA” and “CSUGGH” in adjacent tissues. In lncRNA, the most significant motif for both groups was “UARUCCA”. Similar to lncRNA, in circRNA, the most significant motifs for both groups was “AHUACA” [[22], [23], [24]]. These results indicated that m5C sites were typically associated with cytosine- and adenosine-rich regions. The most significantly associated m5C motifs were similar between HCC and adjacent tissues for all 3 RNA types, including mRNA, lncRNA, and circRNA. We speculate that differences in the m5C distribution between HCC and adjacent tissues are due to differences in the activation or abundance of m5C regulators rather than differences in the types of regulators. Subsequently, differential methylation analysis was performed for the three types of RNAs to compare HCC and adjacent tissues. The results showed that in mRNA, 216 genes showed significantly upregulated methylation, and 282 genes showed significantly downregulated methylation in HCC compared with adjacent tissues (Fig. 2g, Supplementary Table 2, p < 0.05 and |logFC|>1). Among lncRNAs, 337 genes displayed significantly upregulated methylation, and 291 displayed significantly downregulated methylation in HCC compared with adjacent tissues (Fig. 2h, Supplementary Table 3, p < 0.05 and |logFC|>1). Due to the scarcity of circRNAs, only 26 genes with upregulated methylation and 20 genes with downregulated methylation were identified in HCC compared with adjacent tissues (Fig. 2i, Supplementary Table 4, p < 0.05 and |logFC|>1).

Functional enrichment analysis of m5C-related genes

A large number of studies have indicated that m5C mainly functions as a transcriptional regulator to promote RNA export from the nucleus [25]. To more comprehensively explore the functions of methylated RNAs in cells, We conducted GSEA on the methylation genes and differentially methylated genes of HCC separately. The results showed that multiple biological pathways, such as those involved in extracellular matrix organization, the immune system, collagen formation, and Signaling by GPCR, were enriched in differential methylation RNAs in HCC compared with adjacent tissues (Fig. 3a,b). GO analysis revealed key roles for these genes in human phylogeny, consistent with previous reports (Fig. 3c) [26]. These results demonstrate, on the one hand, the intrinsic effects of m5C in cells and, on the other hand, the critical roles played by this modification in the dysregulation of immune, metabolic, and transcriptional regulatory pathways during HCC progression.

Fig. 3.

Functional enrichment analysis of methylated genes. a-b Gene set enrichment analysis (GSEA) were performed on the methylation genes and differentially methylated genes of HCC. c Gene ontology (GO) biological process terms associated with genes with significant differences in methylation and transcriptomics between HCC and adjacent tissues. d The majority of pathways with correlations >0.8 based on methylation and transcriptomic single-sample GSEA (ssGSEA) scores were negatively correlated. e Among genes with differences in both methylation and transcriptomics, most genes with upregulated transcript levels also showed upregulated m5C levels in HCC. f Protein–protein interaction network based on m5C-related genes. Genes in red circles are high-connectivity modules defined by MCODE.

We also performed ssGSEA to evaluate pathway scores using transcriptome data and methylome data from HCC samples and evaluated correlations between calculated pathway scores based on the two omics analyses. A total of 116 pathways were correlated between the two omics analyses, as identified by absolute values greater than 0.5 (p < 0.05) in the correlation analysis, and we present these pathways in Supplementary Figure 1. Screening at higher thresholds (absolute correlation>0.8) revealed 18 pathways primarily involved in transcriptional regulation; immunity, such as the nuclear factor kappa B (NF-κB) signaling pathway; GPCR transcriptional regulation; and tumor-related pathways, such as cell cycle checkpoints, revealing the extensive functional influence of methylation in cells (Fig. 3d). Confusingly, most pathway scores were negatively correlated between the methylome and the transcriptome, while hypermethylated RNAs appeared to be upregulated at the transcriptional level in HCC (Fig. 3e). We speculate that these m5C-modified RNAs may be negative regulators of certain pathways, which will require more precise molecular experiments to explore further.

Protein–protein interaction network analysis

We defined RNAs with significant differences in both the methylome and transcriptome analyses as m5C-related genes, resulting in the identification of 172 m5C-related genes (Supplementary Table 5). To explore the interactions between genes, we performed PPI network analysis on the 172 identified m5C-related genes. A total of 72 genes encode proteins with interaction relationships. Methylation and transcription levels of most genes are simultaneously upregulated in HCC (Fig. 3f). Subsequently, we defined three interacting modules with high connectivity using the MCODE plugin and annotated their functions. As shown in Fig. 3F, these genes play roles in cancer-related behaviors, such as cell invasion, cell metastasis, and transcriptional regulation, and cellular functions, such as the activation of glutamate receptors and calcium-gated channels.

Identification of NMF-based m5C-related molecular subtypes

We extracted expression profiles and prognostic information for the 172 identified m5C-related genes from the TCGA-LIHC cohort and performed univariate Cox analysis on these genes using the “coxph” function in R software, resulting in the identification of 37 genes with prognostic significance (p < 0.05, Supplementary Table 6). Based on the expression profiles of these 37 genes, we performed NMF analysis on the samples from the TGCA-LIHC dataset. Using various indicators, such as cluster heatmap and cophenetic analysis, we characterized HCC subtypes using K = 4 (Fig 4a,b, Supplementary Fig. 2).

Fig. 4.

Identification of m5C-related hepatocellular carcinoma (HCC) subtypes. a Non-negative matrix factorization clustering heatmap when K = 4. b The changing trends in cophenetic, dispersion, and other indicators when K = 2–6. c Kaplan–Meier analysis between Clusters 2 and 3. d Differences between m5C regulators between Clusters 2 and 3. Almost all regulators were significantly upregulated in Cluster 2. We defined Cluster 2 as the hypermethylated group and Cluster 3 as the hypomethylated group. e Volcano plot of differentially expressed genes between Cluster 2 and 3.

Kaplan–Meier (KM) analysis was performed among the four subtypes. The results showed that only clusters 2 and 3 had significant prognostic differences, with cluster 2 having a significantly worse prognosis than cluster 3 (Fig. 4c). To further clarify differences in the patterns of methylation regulation between clusters 2 and 3, we compared the expression of m5C regulators (writers, erasers, and readers) between the two subtypes. Surprisingly, we found that almost all m5C regulators were significantly upregulated in cluster 2, with 2 readers (YBX1 and ALYREF) having the highest expression in cluster 2 among two clusters (Fig. 4d). This finding indicates that the HCC samples in cluster 2 have a more active m5C regulatory status and implies that the significantly worse prognosis observed for cluster 2 may be associated with the highly active m5C regulatory status identified for this subtype. Therefore, we defined cluster 2 as the hypermethylated group and cluster 3 as the hypomethylated group.

Functional enrichment analysis between the hypermethylated and hypomethylated groups

We hypothesized that the significant differences in prognosis and the distinct patterns of methylation regulation observed between hypermethylated and hypomethylated group indicated that the differences between these 2 subtypes were largely representative of the roles played by m5C in HCC progression. Therefore, we performed differential expression analysis between these two subtypes, which revealed 1189 upregulated and 1753 downregulated genes in the hypermethylated group compared with the hypomethylated group (p < 0.05, logFC>1, Fig. 4e). The results of GO enrichment analysis for these DEGs revealed the enrichment of various metabolic biological processes in the hypermethylated group, such as fatty acid, steroid, alcohol, and small-molecule metabolism. Most of these metabolic processes are involved in the normal physiological functions of the liver, implying that abnormally activated m5C states result in the dysregulation of normal metabolic processes (Supplementary Fig. 3a). Various key enzymes associated with the molecular functions of oxidation and reduction were enriched in the hypermethylated group, such as oxidoreductase activity (Supplementary Figure. 3b). Enriched cellular components in the hypermethylated group included microbodies and various lipoproteins, such as very-low-density lipoprotein (Supplementary Fig. 3c). We also performed GSEA on DEGs, as shown in Supplementary Figure 3D, revealing the enrichment of tumor- and metabolism-related pathways in the hypermethylated group, such as the cell cycle, cytochrome P450 signaling, compound functionalization and binding processes, and signal transduction processes. These results suggest that m5C regulates cellular functions and promotes HCC progression through the inhibition of normal metabolic programming and the activation of tumor-associated pathways.

Immunological differences between the hypermethylated and hypomethylated groups

To further clarify the effects of different methylation states on cells and tumors, we analyzed immune differences in the tumor microenvironment (TME) between groups. The ssGSEA results showed the infiltration and upregulation of various T cell types in the hypermethylated group compared with the hypomethylated group, such as activated CD4⁺ T cells, memory CD8⁺ T cells, effector memory CD4⁺ T cells, Type 17 T-helper cells, and Type 2 T-helper cells (Fig. 5a). In addition, natural killer cells and B cells were upregulated to varying degrees in the hypermethylated group compared with the hypomethylated group. Subsequently, we identified dozens of immune checkpoint genes identified in previous studies and examined changes in the expression of these genes under different methylation states. Unexpectedly, these immune checkpoint genes showed a strikingly consistent trend toward upregulation in the hypermethylated group, especially those encoding programmed cell death protein 1 (PDCD1) and cytotoxic T-lymphocyte associated protein 4 (CTLA4) (Fig. 5b). Previous studies indicate that immune checkpoint inhibitors (ICIs) are highly selective in HCC patients, displaying efficacy in only a subset of patients. The infiltration of ICI target cells (T cells) in the TME and the high expression of immune checkpoints are thought to be favorable factors that improve ICI efficacy. Combined with our results, these findings indicate thatHCCs with a highly activated m5C states may be more likely to respond positively to ICIs.

Fig. 5.

Differences in immune and classical tumor pathways between the hypermethylated and hypomethylated groups. a Differences in immune infiltration scores between groups based on single-sample gene set enrichment analysis (ssGSEA). b Differences in immune checkpoint gene expression between groups. c Metabolism, transcriptional regulation, and multiple canonical tumor pathways differ significantly between the hypermethylated and hypomethylated groups based on ssGSEA.

Tumor-related pathways are significantly abnormally activated in the hypermethylated group

We used the ssGSEA algorithm to calculate a pathway score for each sample in both the hypermethylated and hypomethylated groups and screened relevant pathways with different biological functions (Supplementary Table 7). As shown in Fig. 5C, and consistent with previous results, metabolism-related pathways showed a consistent trend toward inhibition in the hypermethylated group. A variety of transcriptional regulation-related pathways showed extensive upregulation in the hypermethylated group, which is consistent with previous studies showing that m5C modification is strongly correlated with RNA transcriptional regulation. In addition, nine widely recognized pathways with significant impacts on tumor progression were identified in a high-impact study [27], including the cell cycle, p53-related pathways, WNT pathways, RTK/RAS pathways, phosphoinositol 3-kinase (PI3K) pathways, Notch pathways, Hippo pathways, and Myc pathways. These pathways were significantly different between groups, and most were significantly upregulated in the hypermethylated group, especially cell cycle-related pathways. These results, on the one hand, confirm the widespread distribution of m5C within cells and the wide-ranging effects that this modification has on function. On the other hand, these findings further illustrate that an active m5C state predisposes cells toward the significant activation of multiple tumor pathways that contribute to HCC progression.

Establishment and validation of an m5C-related prognostic model

To further explore the key differential RNAs that affect prognosis between HCC subtypes and the prognostic, predictive power of m5C status, we performed univariate Cox proportional hazards regression on differentially expressed RNAs identified between the hypermethylated and hypomethylated groups, using p < 0.05 as the threshold for significance (Supplementary Table 8). Prognosis-related RNAs were used to perform LASSO analysis to identify hub genes. The trajectory of each independent variable is shown in Fig. 6A. As lambda gradually increases, the number of independent variable coefficients tending toward 0 gradually increases. We used 5-fold cross-validation for model construction and analyzed the confidence intervals under each lambda, as shown in Fig. 6B. The model was considered optimal when log(lambda) was −2.5. Based on this lambda, we identified 15 signatures, including EZH2, ANXA10, HAVCR1, MYCN, CDCA8, CCNJL, OSGIN1, G6PD, SLC1A7, CBX2, CREG2, FBLL1, ELFN1, MEX3A, and SNHG4. Multivariate Cox analysis of these genes was performed to identify risk coefficients for each gene. Most signatures were identified as prognostic risk factors. The RiskScore for each sample was the sum of the products between the gene expression level and the risk coefficient for each gene. We divided the TCGA-LIHC samples into high-risk and low-risk groups based on the median RiskScore.

Fig. 6.

Construction of m5C-related prognostic model. a Trajectories of lambda changes in LASSO for prognosis-related differentially expressed genes between hypermethylated and hypomethylated groups. b Trajectories of confidence intervals for lambda versus independent variables. We identified 15 signature prognostic models. c Significant prognostic differences between high-risk and low-risk groups. d Receiver operating characteristic curve analysis based on the RiskScore we identified. e Violin plot for RiskScore and Stage_T. f Violin plot for RiskScore and Stage. g Univariate Cox analysis of RiskScore and clinical characteristics. h Multivariate Cox analysis of RiskScore and clinical characteristics. i Nomogram construction of RiskScore and clinical characteristics.

KM analysis between the high-risk group and the low-risk group showed an extremely significant difference in prognosis between the two groups (Fig. 6c, p < 0.0001). The high-risk group had a significantly worse prognosis than the low-risk group. The results of the ROC analysis showed the good prognostic, predictive power of the RiskScore (Fig. 6d). The areas under the ROC curve of this prognostic model were 0.842 (1 year), 0.754 (3 years), and 0.76 (5 years), showing stable predictive power. We further analyzed the correlation between calculated RiskScores and clinical characteristics. Overall, samples with higher RiskScores were characterized with more advanced T Stage and Stage (p < 0.05, Fig. 6e,f); however, some bias may exist due to the small number of Stage IV HCC samples, resulting in no significant correlation between Stage IV and the RiskScore. In addition, M Stage, N Stage, and sex were not significantly correlated with the RiskScore (Supplementary Fig. 4).

Nomogram construction of a prognostic model

To verify the independence of our constructed prognostic model for use in clinical applications, we performed univariate and multivariate Cox analysis on RiskScores and clinical characteristics, respectively. The results of univariate Cox analysis showed that RiskScores had the highest hazard ratio, showing a significant effect for this score on prognosis (Fig. 6g). The values of M Stage, T Stage, and Stage also showed significant negative effects on prognosis. The results of multivariate Cox analysis showed that the RiskScore was an independent prognostic risk factor for HCC patients (Fig. 6h). Based on the results of both univariate and multivariate Cox analyses, we constructed a nomogram for the TCGA-LIHC cohort. As shown in Fig. 6I, our developed RiskScore represents an accurate, stable, and easily quantifiable factor for predicting HCC patient survival compared with other clinical features.

The expression and prognostic value of YBX1 in HCC

YBX1 is a key protein for the function of m5C modification, which mainly stabilises the structure of specific RNA containing m5C modification by binding to it to enhance its downstream proteins. As shown in Fig. 7A, B, the 2 major HCC databases showed that the expression level of YBX1 was significantly higher in hepatocellular carcinoma than in paraneoplastic tissues (P < 0.05). Sequencing data of hepatocellular carcinoma and paracancerous tissues from our centre showed similar trend (Fig. 7c). Subsequently, the difference in protein levels of YBX1 was verified using 10 pairs of total proteins extracted from hepatocellular carcinoma and paracarcinoma tissues, which were significantly higher in HCC than in paracarcinoma tissues (Fig. 7d). We performed IHCstaining for YBX1 on 137 pairs of HCC and adjacent tissues paraffin sections from the First Affiliated Hospital of Zhengzhou University. The representative image showed significant differences in YBX1 expression. YBX1 staining was more pronounced in HCC tissues and was predominantly expressed in the cytoplasm, and the total IHC staining area and the mean IHC staining area in HCC tissues were significantly higher than that in paracancerous tissues (Fig. 7e-g). These evidences showed that the RNA and protein levels of YBX1 were significantly elevated in HCC tissues. Using the median YBX1 expression level as a cut-off value, we divided the HCC samples from the TGCA-LIHC and GSE14520 databases into YBX1 high-expression and low-expression groups, respectively, and analysed the prognostic differences between the groups. In both databases, the prognosis of the YBX1 high-expression group was significantly worse than that of the low-expression group (Fig. 7h-i, P < 0.05). In addition, the group of patients with high-grade Stage stage (Stage III/IV) and T stage (T3/4) contained more YBX1 high-expressing HCC compared with lower Stage stage (Stage I/ II) and T stage (T1/2), which showed that high-expressing YBX1 was correlated with worse clinical grading (Fig. 7j-k).

Fig. 7.

The expression and diagnostic value of YBX1 in HCC. a-c Differential expression of YBX1 between HCC and normal tissues in TCGA-LIHC, GSE14520 and 6 pairs of tissues used for MeRIP Seq. d Differences in YBX1 protein expression between 10 pairs of liver cancer and adjacent tissues in our center; e In situ hybridization immunohistochemistry results of paraffin sections of HCC and adjacent tissues; f-g The difference in IHC total staining area and average staining area between HCC and adjacent tissues; h-i KM analysis of YBX1 in TCGA-LIHC queue and GSE14520 queue; J-K: Differences in YBX1 expression between different stages in the TCGA-LIHC cohort.

Further, we constructed Hep G2 hepatocellular carcinoma cell lines with stable knockdown of YBX1 using shRNA. The efficiency of stable knockdown of YBX1 RNA expression in stably-transformed and control cell lines reached >90 % (Fig. 8a).The expression of YBX1 protein in sh1 and sh2 stably-transformed cell lines was significantly inhibited, demonstrating that YBX1 stably-knocked-down HepG2 cell lines were successfully constructed (Fig. 8b).The results of CCK8 experiments showed that the cell viability of sh1 and sh2 cell lines was significantly inhibited, indicating that YBX1 knockdown suppressed the proliferative capacity of HepG2 cells (Fig. 8c). The results of the cell clone formation assay showed that compared with the control cell lines, the number of cell clone formation became less in sh1-YBX1 and sh2-YBX1 cell lines, and the clone formation ability was significantly inhibited (Fig. 8d).

Fig. 8.

YBX1 binds to TPM3 in an m5C dependent manner. a RNA levels after stable silencing of YBX1 in HepG2 cell lines; b Protein levels after stable silencing of YBX1 in HepG2 cell lines; c CCK8 cell viability experiment between sh1 and sh2 cell lines and control group cell lines; d Clone formation experiment between sh1 and sh2 cell lines and control group cell lines. e The m5C peaks and corresponding m5C consistency motifs of MEX3A, TPM3, and TMCO3 sequences in 6 pairs of liver cancer and adjacent tissues; f-h Expression levels of MEX3A, TPM3, and TMCO3 in control group cells and YBX1 knockout cell lines. i The affinity between TPM3 RNA molecules without m5C modification and YBX1 protein; j Affinity between TPM3 RNA molecules modified with m5C and YBX1 protein.

YBX1 regulates TPM3 in an m5C-dependent manner

The above results demonstrate that YBX1 has a significant effect on the proliferative capacity of hepatocellular carcinoma cell lines. Based on the existing research reports, we speculated that YBX1 might be regulated by binding to RNAs containing m5C modifications. To further clarify the downstream target genes of YBX1, we performed RNA immunoprecipitation sequencing against YBX1 protein. Based on RIP-Seq, we screened RNAs that could directly bind to the YBX1 protein. Functional annotation of these RNAs showed that these RNAs were mainly involved in cellular protein hydrolysis, synthesis, and macromolecular metabolic functions, as well as cell cycle, senescence, autophagy, and RNA transporter functions (Supplementary Figure 5). We further screened potential downstream target RNAs regulated by YBX1 in an m5C modification-dependent manner by the following conditions:

• (1)RNA with direct binding to YBX1 protein;
• (2)RNAs with significantly upregulated methylation levels and transcript levels in HCC tissues (i.e., the aforementioned m5C-related genes);
• (3)RNAs with a differential methylation peak in the region of direct binding to YBX1 and a consistent motif of m5C modification present there.

Based on the above conditions, three potential target RNAs were identified: MEX3A, TPM3 and TMCO3. Significant positive correlations in expression levels were found between YBX1 and all three target RNAs, with correlations of 0.46 (MEX3A), 0.61 (TMCO3) and 0.6 (TPM3), which were all statistically significant. Meanwhile, high expression of all three RNAs predicted significantly poorer prognosis (P < 0.05) as risk factors for hepatocellular carcinoma prognosis (Supplementary Figure 6). In the binding region with YBX1, all three RNAs had significantly up-regulated methylation sites, and the most confident motif for m5C modification was present there: the CUUCC (Fig. 8e). All three RNAs were signatures in the m5C-related prognostic model we constructed above, indirectly confirming the validity of our prognostic model. Subsequently, we verified the expression of the three RNAs in the YBX1-KO Hep G2 cell line, and the expression levels of MEX3A, TPM3 and TMCO3 were significantly decreased in the sh1 and sh2 cell lines compared to the control cell lines (Fig. 8f-h, P < 0.05). Taken together, the above evidence suggests that YBX1 can directly bind to and upregulate the expression of MEX3A, TPM3 and TMCO3, which in turn affects the prognosis of patients with hepatocellular carcinoma, and that this regulatory mechanism is likely to be carried out in an m5C-dependent manner. The results of molecular interaction analysis showed that the affinity of TPM3 RNA molecules without m5C modification to YBX1 protein was 1.577×10^–8, and the affinity of TPM3 RNA molecules containing m5C modification to YBX1 protein was 1.932×10^–9 (Fig. 8i-j). Smaller values of affinity represent higher affinity. Compared with the TPM3 RNA molecule without m5C modification, the affinity of TPM3 containing m5C modification with YBX1 protein increased nearly 10-fold. This result indicates that the binding of YBX1 to TPM3 RNA molecules is dependent on the m5C modification.

Discussion

Post-transcriptional modification, a classic epigenetic pattern, has been shown to play a role in a variety of diseases including cancer [28]. Among them, N6-methyladenosine (m6A) is one of the most prevalent modifications on RNA, which affects cellular phenotype by affecting RNA decay, splicing, trafficking, localization and translation [29,30]. For example, Zuo et al. found that METTL3-mediated m6A modification led to the upregulation of LINC00958 by stabilizing its RNA transcript, which further increased HDGF expression to promote HCC adipogenesis and progression [31]. High-impact studies have also demonstrated that genes such as ATG14, ETS1, HNF3γ, FOXO3, and SOCS2 alter multiple HCC cell phenotypes through similar m6A-dependent regulatory mechanisms [[32], [33], [34], [35], [36]]. N1-methyladenosine (m1A), has also recently been shown to be associated with the progression and self-renewal of HCC [37]. Specifically, TRMT6/TRMT61A enhances PPARδ translation in an m1A-dependent manner, thereby initiating cholesterol synthesis, leading to activation of Hh signaling, which is essential for HCC progression and renewal. As anthor post-transcriptional modification that is widely detected in cellular RNA, m5C has been identified as playing key roles in inflammation, intellectual disability, nervous system development, intergenerational transmission of acquired phenotypes, and mitochondrial dysfunction [[38], [39], [40], [41], [42], [43]]. The aberrant upregulation of m5C in cancers, such as HCC, bladder cancer, and esophageal cancer, further demonstrates the cancer-promoting effects of this modification [8,9,18]. Several excellent studies have explored this correlation in depth; however, given the widespread presence of m5C in cells, we hypothesized that m5C alters tumor phenotypes by affecting multiple cellular functions and pathways rather than acting on any single pathway. A large number of bioinformatics analyses have explored changes in the expression of m5C regulators (writers, erasers, and readers) to explore their functions [44,45]. However, this approach ignores the ubiquity of m5C in cells, does not address differences in m5C sites between cancer and adjacent tissues, and cannot accurately explore how changes in m5C sites regulate tumor phenotypes by affecting specific RNAs. Therefore, a comprehensive analysis examining accurate m5C distribution profiles remains necessary.

In the present study, we performed methylation sequencing in liver cancer tissues and adjacent tissues to identify precise m5C sites. We found a significant increase in methylation sites in liver cancer tissues compared with adjacent tissues, and most of these sites were specific for HCC, suggesting that m5C plays a specific role in HCC progression. In addition, motif analysis showed that mRNA, lncRNA, and circRNA displayed similar m5C-specific motifs between HCC and adjacent tissues, suggesting that differences in methylation distribution were due to differences in the abundance and activation status of m5C regulatory factors rather than the type of regulators. The three types of RNAs were found to have completely different motifs, which is consistent with previous studies indicating that different RNA types are regulated by different m5C writers [46]. A combined analysis of the methylome and transcriptome was performed to identify m5C-related genes. These genes may indicate how m5C affects the evolution of HCC through the regulation of RNA. Consistent with our hypothesis, identified genes were involved in a wide range of cellular functions, such as immune, metabolic, transcriptional regulation, and signal transduction pathways, all of which may play tumor-promoting roles in HCC progression. For example, m5C modification of the promoter region of interferon γ (IFNγ) plays an important role in transcriptional regulation. Megat et al. demonstrated that IFNγ methylation is inversely correlated with transcription, protein levels, and cytotoxicity and that IFNγ promoter hypermethylation may be responsible for the lack of IFNγ responses in a substantial proportion of human tumor-specific cytotoxic T lymphocytes [47]. This result preliminarily illustrates the close relationship between m5C and immune regulation, although more potential links likely remain to be discovered.

Based on these m5C-related genes, we used an NMF algorithm to screen two m5C-related subtypes with significant differences in the expression of almost all m5C regulators, suggesting that the m5C activation states of these two isoforms are distinct and indicating that the differences between these two subtypes may partly explain the role of m5C in HCC. Multiple functional enrichment analyses showed that metabolic, immune, and tumor pathways appeared to be upregulated by m5C in HCC progression. Specifically, the metabolism of small molecules, such as fatty acids and steroids, was significantly enriched. ssGSEA also showed that >10 metabolic pathways were significantly inhibited in m5C hyperactivated HCC. The inhibition of these metabolic processes reflects the disruption of normal hepatocyte function during HCC development. Interestingly, we found that some immune-related pathways were upregulated in the hypermethylated group, and TME immune infiltration analysis revealed that various T cells, such as activated CD4⁺ T cells, memory CD8⁺ T cells, and effector memory CD4⁺ T cells, showed increased infiltration in the hypermethylated group. Unexpectedly, immune checkpoint gene expression was consistently elevated in this group. Studies have shown that various immune checkpoint genes, such as programmed death-ligand 1 (PD-L1), CD8A, lymphocytic activating 3 (LAG3), and signal transducer and activator of transcription 1 (STAT1), are upregulated in the tumors of ICI responders [48]. In addition, the infiltration of ICI target cells is a favorable condition for positive ICI outcomes [49]. The infiltration of ICI target cells in the TME indicates that m5C-active tumors rely on the upregulated immune checkpoint pathway to promote tumor proliferation, indicating that this type of HCC may be more sensitive to ICI treatment.

In terms of the mechanism of how m5C alters the phenotype of HCC cells, Sun et al. have presented one of the mechanisms. They found that NSUN2-mediated m5C modification of H19 lncRNA could enhance the stability of this RNA and further recruit G3BP1 oncoprotein to promote HCC occurrence and development [18]. Similar m5C regulatory axes have been found in bladder and esophageal cancers—although their target RNAs are different. These high-quality studies have contributed to the progress of the m5C field. Compared to other m5C regulatory factors, as a new reader protein discovered in 2019, although YBX1 has been studied as an RNA-binding protein, its role in m5C regulation is still largely unknown. Another question is whether the previously believed binding of YBX1 to RNA is actually achieved through m5C modification. This is an interesting topic, so we chose YBX1 as our research object. In this study, we identified the m5C dependent regulatory axis in HCC through RNA immunoprecipitation and molecular interaction analysis. YBX1 can bind to m5C modification on TPM3 mRNA and further promote the progression of HCC. However, as highlighted above, the widespread presence of m5C modifications in cells means that their effects on signaling pathways are also diverse. Our study showed that almost all canonical cancer pathways were upregulated in m5C-active HCC subtypes compared with the hypomethylated group. These results also indicate that m5C promotes HCC progression through combinations of abnormal activation and inhibition events that affect various tumor-related pathways, including metabolism, immunity, cell cycle, Wnt pathway, and others, rather than affecting only one pathway. We believe that this research is a supplement to m5C field and is expected to provide direction for future research.

Conclusion

In this study, we performed meRIP-seq on HCC and adjacent tissues and conducted combined methylome and transcriptome analyses to identify m5C-related genes. We found that m5C is widespread in cells and involved in multiple cellular functions. HCC subtypes with increased m5C activation have a worse prognosis, which may be due to the suppression of metabolism and immune checkpoint pathways and the marked upregulation of multiple classical cancer pathways. In addition, our prognostic model based on the m5C subtype has good stability and validity. Our findings comprehensively demonstrate the role of m5C as a broad RNA modification involved in HCC progression. More importantly, we have identified an m5C dependent regulatory axis. The m5C reader YBX1 binds to TPM3 in an m5C dependent manner and promotes the progression of hepatocellular carcinoma. These findings lay the groundwork for further understanding the comprehensive role played by m5C in cells and may lead to the therapeutic targeting of m5C.

Ethical approval and consent to participate

This study has been approved by the Ethics Committee of the First Affiliated Hospital of Zhengzhou University with the approval number 2019-KY-21.

Consent for publication

Not applicable.

Availability of supporting data

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Funding

This work was funded by the National Natural Science Foundation of China (819028328), Leading Talents of Zhongyuan Science and Technology Innovation (214200510027), Henan Provincial Medical Science and Technology Research Plan (SBGJ2018002 and SBGJ202102117), the Gandan Xiangzhao Research Fund (GDXZ2022002), and Outstanding Foreign Scientist Studio in Henan Province (GZS2020004).

CRediT authorship contribution statement

Qiyao Zhang: Writing – review & editing, Writing – original draft, Resources, Formal analysis, Data curation. Zhen Cao: Writing – review & editing, Methodology, Data curation. Yuting He: Validation, Methodology, Formal analysis, Conceptualization. Ziwen Liu: Writing – review & editing, Validation, Software, Resources, Project administration, Formal analysis, Data curation, Conceptualization. Wenzhi Guo: Writing – review & editing, Resources, Methodology, Funding acquisition, Data curation, Conceptualization.

Declaration of competing interest

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