Human samples used in this work were totally supported by patients treated in the Drum Tower Hospital (Nanjing, China), which contained human tissues and blood samples. Informed consent was obtained from all patients included in this study. This research was approved by the Institutional Ethics Committee of Nanjing Drum Tower Hospital (2021-608-01). All samples were fresh-frozen and stored at −80 °C.

Human HCC cell lines (L-02, Hep3B, HepG2, MHCC-97L, MHCC-97H, MHCC-LM3, SMMC-7721, and Huh7) and 293T cells were purchased from the China Academy of Science Cell Bank. Cells for experiments were cultured in DMEM supplemented with 10% fetal bovine serum (FBS, Wisent, Saint Bruno, Canada) and routinely monitored for mycoplasma. SNORD13H was knocked out using the lentiviral-mediated CRISPR–Cas9 system. The sgRNA target sequences and protocol were obtained from the Zhang Lab website (http://www.genome-engineering.org). Lentivirus containing short hairpin RNA (shRNA) plasmids was used to knock down RAS and fibrillarin (FBL), whose targeting sequences were obtained from Sigma-Aldrich (Merck, Darmstadt, Germany). Detailed sequences are shown in Supplementary Table 1. Lentiviral packaging plasmids were selected in reference to a standard method. Lipo3000 and P3000 were purchased from Invitrogen (CA, United States), and polybrene was purchased from Sigma-Aldrich (Merck, Darmstadt, Germany). Viral packaging and infection were performed in accordance with the manufacturer’s instruction and usual protocol.

Cell proliferation was monitored via the CCK-8 assay (Dojindo, Kumamoto, Japan) according to the product instructions. Clonogenic potential of HCC cells was determined through a colony formation assay. For cell cycle experiments, a PI Cell Cycle Kit (Vazyme, Nanjing, China) was used to stain DNA. In addition, Annexin V-FITC/PI from the Apoptosis Detection Kit (Vazyme, Nanjing, China) was used in the cell apoptosis assay. Flow cytometry was applied to analyze both the cell cycle and apoptotic cells. The specific operation was carried out according to the manufacturer’s instruction.

The subcutaneous tumor-bearing mouse model was established using BALB/c nude mice, which were purchased from GemPharmatech Co. Ltd (Nanjing, China). Mice used in this research were kept in SPF environment at all times. There were six BALB/c nude mice (5–6 weeks old) in each group. HCC cells (2–3 × 10⁵) were injected subcutaneously into the inguinal area. Solid tumors at the inoculation sites formed approximately 10 days after injection, and tumor sizes were measured every 5 days until sacrifice. Tumor size and weight were measured and recorded for further analysis. All animal care and animal experiments were performed in accordance with the guidelines of ethical regulations and the Helsinki Declaration, and all treatments adhered to the rules of the Animal Care and Use Committee of the First Affiliated Hospital of Anhui Medical University (2024021904).

Total RNA was extracted using the TRIzol RNA extraction reagent (Invitrogen, CA, United States) following the manufacturer’s protocol. After measuring RNA concentration using the NanoDrop technology, equal amounts of RNA were reverse-transcribed to cDNA using HiScript II Q RT SuperMix for qPCR (Vazyme, Nanjing, China). The cDNA was then used for next quantitative real-time PCR with SYBR Green (Vazyme, Nanjing, China). Total plasma miRNA was isolated using the miRNeasy Serum/Plasma Kit (QIAGEN, Hilden, Germany). Reverse transcription and real-time quantitation were carried out with specific kits, such as the miRcute miRNA First-Strand cDNA Synthesis Kit (QIAGEN, Hilden, Germany) and the miRcute miRNA SYBR Green qPCR Detection Kit (QIAGEN, Hilden, Germany). RNA levels were calculated using U6 or β-actin as control. Primer sequences are detailed in Supplementary Table 2.

Based on the previous publications, 2′-O-methylation levels were detected using reverse transcription at low dNTP concentrations followed by PCR (RTL-P) assay to quantify SNORD13H modification efficiency (Dong et al., 2012; Wu et al., 2020). The presence of 2′-O-methyl modifications impedes reverse transcription at low dNTP concentrations, creating a detectable difference compared to high dNTP conditions (Figure 3A). Based on established rRNA modification maps, 18S and 28S rRNAs were divided into five segments, each containing at least one known modification site, and corresponding primer sets (paired Fu, Fd, and R) were designed. The Fu primer amplifies across potential modification sites, whereas the Fd product serves as an internal control. High dNTP reactions provide additional controls for each sample. As reported in previous studies, reverse transcription was performed with 100 ng of RNA in a 25-μL reaction system. RTase M-MLV (RNase H-) (Takara, Kusatsu, Japan) and RNase Inhibitor (Takara, Kusatsu, Japan) were also added. In the following PCR step, Ex taq DNA polymerase (Takara, Kusatsu, Japan) and dNTP (Takara, Kusatsu, Japan) were mixed in a 10-μL reaction system with primers and products from the last step. The specific PCR process and details of experimental design were provided by Dong et al. (2012) and Wu et al. (2020). The products were subjected to agarose gel electrophoresis with 2% agarose gels and visualized through UV-trans-illumination. Primer sequences are shown in Supplementary Table 3.

To display ribosomal translation efficiency more directly, the surface sensing of translation (SUnSET) technique was performed according to Yi et al. (2021) and Schmidt et al. (2009). As described previously, the experimental principle is labeling newly synthesized proteins with puromycin. Cells were treated with 335 μM CHX (Aladdin, Shanghai, China) for 15 min. Then, 91 μM puromycin was added to incubate cells at 37 °C for 5 min. Results were assessed using the Western blot assay with the anti-puromycin antibody (Sigma-Aldrich, Darmstadt, Germany).

The procession of rRNA was simply measured using the qRT-PCR assay according to Cao et al. (2018). After extracting the total RNAs from HCC cells and reverse transcription, rRNA procession was verified using quantitative real-time PCR as usual. Gene-specific primers were designed (Supplementary Figure 4C), whose sequences are shown in Supplementary Table 4. The Ct values for primers 2/1 and b/a represent mature 18S and 28S rRNAs, respectively. Meanwhile, the primers targeting 5′ ETS-18S (primers 4/3 and 6/5) and ITS2-28S (primers d/c and f/e) represent unprocessed rRNA. The ratios of the Ct value (unprocessed group) to the Ct value (mature group) were averaged, meaning for the fraction of unprocessed rRNA. Primer pairs 4/3 over 2/1 and primer pairs 6/5 over 2/1 were for 18S rRNA, and primer pairs d/c over b/a and primer pairs f/e over b/a were for 28S rRNA.

Immunoblotting was carried out with standard methods and protocol using commercially available antibodies. Antibodies used in this study are listed in Supplementary Table 5.

Tumor samples from the subcutaneous tumor-bearing mouse model were fixed in 4% formaldehyde and sent to Servicebio (Wuhan, China) for paraffin-embedded sections. The hematoxylin and eosin (HE) staining experiment was performed by Servicebio (Wuhan, China). Immunohistochemistry was performed using generic methods and standard protocol with a commercial immunohistochemical staining kit (Maixin Biotech, Ltd., Fuzhou, China). Antibodies are commercially available, including anti-Ki-67 antibody (Abcam, Cambridge, United Kingdom) and anti-RAS antibody (Abcam, Cambridge, United Kingdom). The stained sections were scanned and visualized using PANNORAMIC MIDI II and CaseViewer software (3D HISTECH, Budapest, Hungary).

Total cellular RNA was prepared by TRIzol (Invitrogen, CA, United States) for transcriptome sequencing. The total RNA quantity and purity were analyzed using Bioanalyzer 2100 and RNA 6000 Nano LabChip Kit (Agilent, CA, United States), respectively; high-quality RNA samples with RIN number >7.0 were used to construct the sequencing library. Then, mRNA was purified from the total RNA (5ug) using Dynabeads Oligo (dT) (Thermo Fisher, CA, United States). During library preparation, RNA is fragmented to ∼300 bp using divalent cations at 94 °C, followed by first-strand cDNA synthesis using SuperScript™ II Reverse Transcriptase (Invitrogen, CA, United States) and second-strand synthesis using DNA Polymerase I (NEB, MA, United States). An A-base was then added to the blunt ends of each strand, and dual-index adapters were ligated to the fragments. Size selection was performed using AMPureXP beads. The libraries are amplified through eight cycles of PCR. The average insert size for the final cDNA libraries was 300 ± 50 bp. Finally, the libraries were sequenced using 2 × 150 bp paired-end reads (PE150) on the Illumina Novaseq™ 6000 platform by LC Bio Technology Co., Ltd. (Hangzhou, China). The bioinformatics pipeline includes the following: adapter trimming and quality filtering using Cutadapt, raw data quality control using FastQC (Q20, Q30 and GC-content), alignment to reference genomes with HISAT2, gene expression quantification using StringTie and ballgown, differential expression analysis with DESeq2 (|log2FC| ≥ 1, FDR <0.05), principal component analysis (PCA) using R, and functional enrichment analysis through KEGG/GO pathways. Other bioinformatic analysis was performed using the OmicStudio tools at https://www.omicstudio.cn/tool.

To confirm the 2′-O-methylational modification of SNORD13H on mRNA, we downloaded a gene database of published articles from the Gene Expression Omnibus (GEO) database (GSE77027, https://ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE770 27). It is a CLIP-seq dataset which contains RNA in complex with the 2′-O-methylase FBL (Incarnato et al., 2017).

All data were based on at least three independent experiments. Each independent experiment was in triplicates. Data were analyzed using Student’s t-test and ANOVA analysis, which were implemented using GraphPad Prism 9.0 statistical software (CA, United States). Statistically significant differences were deemed only if p < 0.05.

To identify snoRNAs associated with HCC, we performed RNA sequencing on a cDNA library derived from three pairs of human HCC tissues and matched peritumor tissues. A profile of RNA-sequencing results is displayed (Figure 1A). Using Entrez Gene IDs, we categorized differentially expressed ncRNAs (p < 0.05, Log2 [Fold Change (FC)]>1), with particular focus on lncRNA and snoRNA. We generated a volcano plot highlighting snoRNAs (Figure 1B) and violin plots illustrating overall expression changes in snoRNA and lncRNA (Figure 1C). Notably, snoRNA expression exhibited a significant downward trend in HCC tissues. A heatmap was used for further analysis (Figure 1D). Among these snoRNAs, a C/D box snoRNA, SNORD13H, showed the most pronounced fold changes (Figure 1E). We validated its expression in 20 additional HCC tissue pairs using Q-PCR, confirming significant downregulation in tumors compared to peritumor tissues (Figure 1F). Similarly, SNORD13H levels were reduced in plasma samples from HCC patients relative to non-tumor controls (Figure 1G). Given these findings, we proceeded to investigate the functional roles of SNORD13H in HCC development and progression.

We first evaluated SNORD13H expression across seven HCC cell lines and immortalized human hepatocyte cell line L-02 (Supplementary Figure 1A). Hep3B and MHCCLM3 cells were selected for further studies. To mimic the low SNORD13H conditions observed in HCC, we constructed stable SNORD13H-knockout cell lines using the CRISPR/Cas9 system, with Q-PCR confirming knockout efficiency (Figure 2A). Functional assays revealed that SNORD13H-knockout cells displayed significantly increased viability, proliferation, and colony formation capability compared to controls (Figures 2B–D). Conversely, SNORD13H overexpression (Supplementary Figure 1E) suppressed these malignant phenotypes (Supplementary Figures 1F–H). The results support a tumor-suppressive role for SNORD13H in vitro.

To validate these findings in vivo, we subcutaneously implanted SNORD13H-knockout or control Hep3B/MHCCLM3 cells into BALB/c nude mice, establishing xenograft models. Tumors derived from SNORD13H-knockout cells exhibited larger volumes (Figure 2E), faster growth rates (Figure 2G), and heavier weights (Figure 2F). These results demonstrate that SNORD13H loss enhances development and progression of HCC in vivo.

In the exploration on the mechanistic basis of SNORD13H-regulated tumor progression, SNORD13H-knockout cells showed significant resistance to starvation-induced apoptosis (Figures 2H–J). This result could correlate with reduced cleaved Caspase-3 levels (Supplementary Figure 1C), suggesting that SNORD13H modulates apoptosis via Caspase-3 activation. These findings position SNORD13H as a tumor suppressor in HCC that constrains malignant progression through the dual regulation of proliferation and apoptosis.

SnoRNAs primarily function to guide RNA modifications, with rRNA as their main target. As a C/D box snoRNA, it is anticipated that SNORD13H regulates 2′-O-methylation of rRNA, a conserved RNA modification critical for ribosome function. Using RTL-P (Figure 3A), we assessed 2′-O-methylation levels across 18S and 28S rRNA segments in SNORD13H-knockout and overexpressing HCC cells (Supplementary Figures 2, 3), which is referred to previous studies (Dong et al., 2012). Normalized quantification revealed that SNOD13H depletion specifically reduced 2′-O-methylation at the 18S-4 segment, whereas SNORD13H overexpression increased it (Figure 3B).

C/D box snoRNAs are characterized by conserved C box (UGAUGA) and D box (CUGA) motifs. Between these motifs lies the antisense element (ASE), a 10–21 nucleotide guide sequence that base pairs with target RNA to position the modification site and directs 2′-O-methylation (Bratkovic et al., 2020). Unlike canonical Watson–Crick pairing, RNA–RNA interactions in this process are more flexible, permitting noncanonical base pairs. Through comparative analysis, Um1288 was predicted as the most probable SNORD13H-dependent methylation site on 18S rRNA (Supplementary Figure 4C). Negative controls—including known 2′-O-methylate sites guided by other snoRNAs on 18S or 28S rRNA (Dong et al., 2012; Wu et al., 2020; Zhou et al., 2017)—showed no changes in SNORD13H-knockout cells (Supplementary Figure 4A,B). These complementary findings demonstrate a direct mechanistic link between SNORD13H loss and impaired rRNA modification.

Given the role of 2′-O-methylation in rRNA processing and cleavage (Ni et al., 1997), we evaluated its impact on 18S rRNA mutation (Supplementary Figure 4D). SNORD13H knockout decreased the mature-to-precursor ratio of 18S rRNA, whereas overexpression had the opposite effect (Supplementary Figure 4E). No such changes were observed for 28S rRNA, suggesting that SNORD13H specially regulates 18S rRNA maturation via 2′-O-methylation.

Posttranscriptional rRNA modifications play a direct role in regulating ribosome biogenesis and transcript stability (Liu et al., 2019). As ribosomes are vital organelles for protein synthesis (Ahmed et al., 2019), 2′-O-methylation, one of the most abundant rRNA modifications (Gribling-Burrer et al., 2019), has been shown to modulate the ribosome translation efficiency (Yi et al., 2021). The SUnSET assay (Yi et al., 2021; Schmidt et al., 2009) revealed that SNORD13H knockout enhanced the global translation efficiency, whereas overexpression suppressed it (Figures 3C,D). This aligns with its tumor-suppressive role, as elevated translation promotes the proliferation and growth in HCC (Wang B. et al., 2022). Together, these findings demonstrate that SNORD13H loss drives HCC progression by reducing 18S rRNA methylation and increasing protein synthesis.

Although snoRNAs are best known for rRNA modifications, emerging evidence suggests broader regulatory functions. C/D box snoRNAs can bind mRNAs directly, influencing their splicing (Falaleeva et al., 2016; Huang et al., 2017), stability, and protein expression (Wu et al., 2020; Baldini et al., 2022). These promoted us to investigate whether SNORD13H regulates HCC progression through mechanisms beyond rRNA 2′-O-methylation.

RNA-seq analysis of SNORD13H-knockout MHCCLM3 cells revealed significant transcriptomic alterations (Figure 4A). The principal component analysis (PCA) of SNORD13H-knockdown vs. control HCC cells reveals distinct transcriptional programs, with PC1 separating by genotypes and PC2 reflecting cell line-specific signatures (Supplementary Figure 5A). KEGG pathway enrichment highlighted the MAPK signaling pathway as the most affected one (Figure 4B), with upregulated mRNA levels of enriched genes (Figure 4C). Gene Set Enrichment Analysis (GSEA) further confirmed MAPK pathway activation in knockout cells (Figure 4D).

To further characterize MAPK pathway activation, we also analyzed RNA-seq data for different expressions of MAPK-associated genes (Supplementary Figures 5C,D) and validated its consistent mRNA changes in selected conditions (Supplementary Figure 5E). Based on the GO enrichment analysis (FDR <0.05) highlighting significant terms across biological processes, molecular functions, and cellular components (Supplementary Figure 5B), we prioritized one gene as a candidate regulator. However, despite transcriptional upregulation of the filtered gene, its corresponding protein levels did not display significant alternations (Supplementary Figure 5F). This discrepancy suggests that SNORD13H regulates MAPK signaling primarily through other modalities.

SNORD13H knockout specially increased phosphorylated ERK1/2 (p-ERK1/2) levels (Figure 4E), indicating ERK1/2 pathway activation instead of MAPK/p38, MAPK/ERK5 and MAPK/JNK1/2/3. Subsequent profiling of MAPK cascade components revealed that RAS protein levels were elevated in SNORD13H-knockout cells but reduced in SNORD13H-overexpressing cells (Figure 4F). Consistent with this, xenograft tumors derived from SNORD13H-knockout cells exhibited higher Ki67 and RAS expressions (Supplementary Figures 6, 7), corroborating enhanced proliferation and MAPK signaling in vivo. Our data demonstrate that SNORD13H downregulation in HCC promotes RAS accumulation, leading to constitutive MAPK/ERK pathway activation and driving tumor progression. This expands the functional repertoire of SNORD13H to include direct modulation of oncogenic signaling pathways.

The RAS family of small GTPases (including HRAS, KRAS, and NRAS) plays pivotal roles in tumor progression (Akula et al., 2016). To make sure RAS is a key effector of SNORD13H depletion, we generated SNORD13H/RAS double-knockout HCC cells (Supplementary Figure 8A). Compared to SNORD13H-knockout alone, double-knockout cells exhibited rescued viability, impaired colony formation, and increased apoptosis (Supplementary Figures 8B–F, 9). These results indicate that RAS is essential for the tumor progression promotion induced by SNORD13H loss.

When we explored reasons for altered RAS expression, we found that HRAS and KRAS mRNA levels were paradoxically reduced (Figures 5A,C) despite elevated RAS protein levels in SNORD13H-knockout cells. Conversely, SNORD13H overexpression increased HRAS and KRAS mRNA levels (Figure 5B), with positive correlations observed in HCC tissues (Figures 5D–F). This inverse relationship between RAS mRNA and protein levels suggested post-transcriptional regulation.

Certain ncRNAs directly bind mRNAs to modulate gene expression through mechanisms, including translation inhibition and transcript stabilization (Latonen et al., 2018; He and Hannon, 2004). mRNA 2′-O-methylation guided by snoRNAs shares the same functions (Elliott et al., 2019). These findings prompted us to investigate whether SNORD13H regulates RAS mRNA similarly through direct 2′-O-methylation.

Given that FBL is the specific methyltransferase combining with snoRNAs, analysis of FBL CLIP-seq data (GSE77027) separately found two KRAS and HRAS mRNA coding regions. Primers for RTL-P were designed focusing on these four FBL-binding regions. Assays confirmed reduced 2′-O-methylation of RAS mRNA in SNORD13H-knockout cells (Figure 5G; Supplementary Figure 10A) and enhanced methylation upon SNORD13H overexpression (Figure 5H; Supplementary Figure 10B). Meanwhile, FBL protein levels were not affected by SNORD13H/RAS knockout (Supplementary Figure 10C). These revealed RAS mRNA 2′-O-methylation induced by SNORD13H. We also predicted SNORD13H binding sites within KRAS and HRAS mRNAs (Supplementary Figure 10D). Notably, 2′-O-methylation suppresses translation (Elliott et al., 2019), perhaps explaining why RAS protein levels increased despite mRNA downregulation in SNORD13H-deficient cells. Our findings establish a novel mechanism, whereby SNORD13H inhibits HCC progression by guiding 2′-O-methylation of RAS mRNA. Under this condition, RAS protein expression and MAPK/ERK pathway activation are promoted by SNORD13H loss. This dual regulatory role underscores SNORD13H’s tumor-suppressive function in HCC.

FBL, the catalytic component of snoRNPs, is indispensable for snoRNA-guided 2′-O-methylation. To test whether FBL mediates SNORD13H’s effects, we depleted FBL in SNORD13H-overexpressing cells (Figure 6A). Moreover, overexpression efficiency of SNORD13H remained unchanged (Figure 6B). Whereas SNORD13H overexpression alone suppressed HCC cell growth, stemness, and proliferation, FBL knockout reversed these effects (Figures 6C–F). Consistent with this, FBL depletion restored RAS protein levels (Figure 6G) and global translational efficiency (Figure 6H), proving that SNORD13H’s tumor-suppressive function requires FBL-dependent methylation.

Based on CLIP-seq data (GSE77027) and structural predictions, we divided SNORD13H into four fragments (Figure 7A) and generated three mutants (mutation of segment 2, segment 3, and segment 4, named Δ2, Δ3, and Δ4, respectively) using their antisense sequences (Supplementary Table 1). Notably, segment 3 contained FBL-binding sequences. These mutants were transfected into SNORD13H-knockout cells (Figures 7B,C). In addition, it revealed that Δ3 most weakly rescued cell viability and proliferation (Figures 7D,E). Moreover, Δ3 minimally affected RAS protein levels (Figures 7F,G). These indicate that segment 3 is critical for SNORD13H’s ability to 2′-O-methylate and restrain RAS protein levels. To sum up, we confirmed that SNORD13H’s tumor-suppressive effects depend on FBL-mediated 2′-O-methylation on rRNA and RAS mRNA (Figure 7H). Segment 3 of SNORD13H is essential for targeting RAS mRNA. Reduced 2′-O-methylation of RAS mRNA in SNORD13H-deficient cells drives tumor promotion and progression.