Elucidating the role of SIRT2 in hepatocellular carcinoma through multi-omics and deep learning

Abstract

Background

Hepatocellular carcinoma (HCC) is a prevalent global malignancy characterized by a high incidence and poor prognosis. Current diagnostic and therapeutic modalities are insufficient to meet the demands for effective HCC diagnosis and management. Cuproptosis represents a novel mechanism of cellular death, which may offer new avenues for drug research in oncology.

Methods

We employed the Summary-data-based Mendelian Randomization (SMR) method alongside gene intersection analyses related to Cuproptosis, identifying SIRT2 as the target gene. Single-cell RNA sequencing (scRNA-seq) and spatial transcriptome sequencing (stRNA-seq) were conducted to investigate the role of SIRT2 in HCC. Subsequently, we analyzed the differentially expressed genes of SIRT2 + malignant cell (SIRT2 + Mali) using a Deep Learning Survival Neural Network (deepsurv), leading to the construction of a prognosis model. The protein interactions between SIRT2 and the key genes involved in the model were assessed. Finally, RNA sequencing (RNA-seq) analysis was performed, encompassing gene correlation analysis, immune checkpoint analysis, GO, GSEA, and KEGG enrichment analysis, immune cell infiltration, clinical characteristics analysis and drug sensitivity analysis.

Results

SIRT2 was identified as a gene positively correlated with HCC risk through SMR analysis. The scRNA-seq analysis revealed that, compared to SIRT2- malignant cell (SIRT2- Mali), SIRT2 + Mali exhibited stronger interactions with cancer-associated fibroblasts (CAF), tumor-associated macrophages (TAM), and tumor-associated endothelial cells (TEC), participating in more diverse metabolic pathways. The stRNA-seq analysis indicated a robust spatial correlation of SIRT2 + Mali with CAF, TAM, and TEC. The deepsurv prognostic model demonstrated that the survival rate of patients with elevated risk scores was significantly lower than that of those with low risk scores. RNA-seq analysis confirmed a positive correlation between SIRT2 and various immune checkpoint genes. Immune cell infiltration analyses indicated higher abundance scores of monocytic lineage, endothelial cells, and fibroblasts in the SIRT2 high expression group compared to the SIRT2 low expression group. The expression of SIRT2 was associated with gender, T classification, and Stage.

Conclusion

SIRT2 is posited as a pathogenic gene for HCC. It may facilitate the growth and invasion of HCC via multiple mechanisms, including Cuproptosis, tumor microenvironment modulation, metabolic pathway alteration, and immune checkpoint activation. SIRT2 presents as a potential biomarker for HCC and may serve as a novel therapeutic target for its treatment.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-03879-0.

Introduction

Hepatocellular carcinoma (HCC) is one of the most common cancers worldwide, with a high incidence and poor prognosis. HCC is the primary malignant tumor of liver cancer, accounting for 75%–85% of cases, and the relative 5-year survival rate is about 18% [1, 2]. With the approval of new first-line and second-line drugs and the establishment of immune checkpoint inhibitor-based therapies as standard treatment, the treatment prospects of HCC are more diverse than ever before. Despite the recent progress in comprehensive treatment, however, a considerable number of patients are in advanced stage at the time of diagnosis, and there is still an urgent need for biomarkers to guide treatment selection and diagnosis [3]. Recently, due to the rapid development of omics technology, our understanding of the molecular pathogenesis of HCC has been significantly improved [4, 5]. A series of features based on omics data were generated to predict the outcome and recurrence of HCC [6]. Therefore, the analysis method combining multi omics is very valuable for finding new biomarkers of HCC.

Copper plays an important role in cells and is a catalytic cofactor of essential enzymes involved in energy conversion, oxygen transport, and regulation of cellular oxidative metabolism [7]. Copper is necessary for cell life, but excessive copper can also lead to cell death [8]. Cuproptosis is an independent and novel cell death mode, and its mechanism may be related to mitochondrial respiration level and lipoic acid pathway [9]. Cuproptosis can provide a new direction for drug research on tumors [10].

Multi omics integration is an emerging method in the post-genome-wide association studies (GWAS) era, and summary-data-based Mendelian randomization (SMR) has been developed. SMR extends the concept of Mendelian randomization by combining GWAS data with gene expression data from expression quantitative trait loci (eQTL) studies to explore the causal relationship between potentially pathogenic genes and diseases [11]. Spatial transcriptome is a new sequencing method that can observe the distribution of genes in tissue space and the direction of intercellular signaling and can provide new biological insights for diseases [12]. The development of machine learning and deep learning assistant systems has influenced the field of medicine and has been used in the diagnosis and classification of various cancers [13].

In this study, we combined SMR with scRNA-seq, stRNA-seq and other omics to discover a new cuproptosis-related gene SIRT2 in HCC, explored its role in HCC, and constructed a prognostic model through deep learning survival neural network (deepsurv), laying a solid foundation for the development of new targeted therapeutic strategies for SIRT2.

Methods

Summary-data-based Mendelian randomization and cuproptosis

For SMR analysis, the cis-eQTL of liver tissue in exposure data was sourced from the Genotype Tissue Expression (GTEx) project v8 (https://yanglab.westlake.edu.cn/software/smr/#eQTLsummarydata). The GWAS data concerning HCC outcomes were obtained from Hepatocellular carcinoma, excluding all cancers (controls excluding all cancers) in FINNGEN [14]. The version is R10, all populations are European populations. SMR (v1.3.1) was employed to examine these two independent datasets, with a p-value < 0.05 indicative of a causal relationship between exposure and outcome. The 1000 Genomes Project European individuals (1000G EUR) served as reference genotype data, retaining only protein-coding genes. The HEIDI test was conducted to assess the heterogeneity of results, with a p-value < 0.05 signifying the presence of heterogeneity. Genes implicated in Cuproptosis were obtained from FerrDb (http://www.zhounan.org/ferrdb) (Supplementary File 1). SMR results were intersected with the Cuproptosis gene set to isolate SIRT2, a gene causally associated with HCC and linked to Cuproptosis. SMRLocusPlot and SMREffectPlot of SIRT2 and HCC were generated using R (version 4.3.1).

Single-cell RNA sequencing analysis

The scRNA-seq data for HCC were retrieved from the Gene Expression Omnibus (GEO) under accession number GSE125449 from the National Center for Biotechnology Information [15]. scRNA-seq analysis was performed utilizing the “Seurat” package (version 4.3.0). The data were normalized using the NormalizeData function, with the scale factor set to 10,000. The cell subpopulations in GSE125449 have been annotated. Malignant cells were classified into two groups based on SIRT2 expression levels: SIRT2-positive malignant cells (SIRT2 + Mali) and SIRT2-negative malignant cells (SIRT2- Mali), using a threshold of expression >0 for categorization. Cell communication was analyzed using the CellChat package. Subsequently, the scMetabolism package was used to perform metabolic analysis on SIRT2 + Mali and SIRT2- Mali. The sc.metabolism.Seurat function was used, with Method set to AUCell, imputation set to F, ncores set to 2, and metabolism.type set to KEGG [16]. Pseudotime analysis of SIRT2 in malignant cells was executed using the Vector algorithm.

Spatial transcriptome analysis

Spatial transcriptome data processing

The spatial transcriptome data for HCC were sourced from the study conducted by Ma et al. [17]. Raw data were loaded using the Load10X_Spatial function from the Seurat package. Tissue coordinate information was extracted from the raw data utilizing the GetTissueCoordinates function. Cells located on the tissue were filtered out employing the subset function. Quality control procedures were implemented to exclude the expression of mitochondrial genes and ribosome-related genes. Post-preprocessing, the SCTransform function was utilized to adjust for technical variances. The NormalizeData function was employed for data normalization, while the ScaleData function standardized all features. Subsequently, principal component analysis and UMAP dimensionality reduction were performed.

Reverse compositional transcriptomics deconvolution (RCTD) and MistyR

RCTD is a method for inferring cellular composition from spatial transcriptomics data. This technique involves analyzing the comprehensive gene expression profile of the entire tissue or sample to reverse-calculate the spatial distribution and relative abundance of each cell type [18]. SIRT2 + Mali, SIRT2- Mali, cancer-associated fibroblasts (CAF), tumor-associated macrophages (TAM), and tumor-associated endothelial cells (TEC) were visualized. The spatial expression pattern of the SIRT2 gene in HCC tissue was explored using the mistyR package; a network diagram of these cells was constructed using the SPOTlight package.

Cell degree

To further investigate the interactions among different cell types, we utilized cell degree. Initially, the spatial transcriptomics data of HCC and the previously obtained RCTD data were imported. Spatial coordinate data were acquired and converted to image coordinates using the scaling factor. We employed imagerow_scaled and imagecol_scaled to represent scaled row and column coordinates. The DBSCAN algorithm was applied to construct a spatial adjacency network, setting nNeighbours to 6 and maxdist to 200. An adjacency matrix was generated, with the connection information between cells obtained by screening effective spatial connections (minK was set to 4). The threshold for cell interaction was established at 0.1. SIRT2 + Mali was represented in yellow, while the three tumor microenvironment cells (CAF, TAM, and TEC) were depicted in purple, producing a spatial distribution map of cell interactions. Additionally, we calculated the enrichment scores of intercellular relationships; higher enrichment scores indicated stronger cell interactions in the area, thereby quantifying the spatial interactions between SIRT2 + Mali and the three tumor microenvironment cells.

Deep learning survival neural network

Differentially expressed genes from SIRT2 + Mali were extracted, with the logfc.threshold established at 0.2. Genes associated with ribosomes were excluded. The GDC TCGA Liver Cancer dataset was obtained from Xena (https://xenabrowser.net/datapages/) as the training set for HCC [19]. The HCC dataset was sourced from ICGC (https://www.icgc-argo.org/) as the validation set. Survival information and HCC samples were extracted from both datasets for further analysis. The training and validation sets underwent de-batching. Initially, differentially expressed genes were filtered using univariate Cox regression (pFilter = 0.05), followed by lasso regression analysis on the filtered genes. The lasso regression was implemented using the glmnet package, which automatically generated 100 candidate λ values logarithmically spaced to balance model complexity and sparsity. A 10-fold cross-validation procedure was employed to evaluate the generalizability of each λ. The optimal λ value (lambda.min) corresponding to the minimum cross-validated deviance was selected to identify key prognostic genes. Finally, key genes for deep learning were identified. A PPI network diagram of key genes was generated using STRING (https://string-db.org) (version 12.0). DeepSurv was employed in Python to conduct deep learning analysis for both the training and validation sets. DeepSurv is a novel deep learning method integrating Cox proportional hazards for survival analysis, exhibiting performance comparable to or superior to existing survival models [20]. We established 2000 training cycles and utilized the Nesterov momentum optimization algorithm (lasagne.updates.nesterov_momentum) to update model parameters, ensuring training efficiency. The trained model was subsequently employed to predict risk for both the training and validation sets, generating risk scores for each sample. The samples were stratified into high-risk and low-risk groups based on the median risk score. Following this, we utilized the survival and survminer packages in R to perform survival analysis, with survival curves representing high-risk and low-risk groups depicted in blue and red, respectively. We also employed the timeROC package to compute survival ROC curves at different time points (1 year, 2 years, and 3 years) and constructed a nomogram incorporating age, gender, stage, and risk, alongside relevant calibration curves.

Protein interaction analysis

To evaluate potential interaction relationships between SIRT2 and key genes involved in modeling, we conducted protein interaction analysis using AlphaFold 3 (https://alphafoldserver.com/). The AlphaFold 3.0 model can accurately predict biomolecular interactions [21]. The protein sequences of SIRT2 and modeling genes were sourced from the National Center for Biotechnology Information (NCBI). Subsequently, protein interaction analysis was conducted on AlphaFold 3 to derive the predicted template modeling (pTM) score and the interface predicted template modeling (ipTM) score. A pTM score exceeding 0.5 suggests that the overall predicted folding of the complex may resemble the true structure, while an ipTM score greater than 0.6 indicates high prediction accuracy [22, 23].

RNA sequencing analysis

Enrichment analysis

The GDC TCGA Liver Cancer dataset acquired from Xena was utilized for RNA-seq analysis. HCC samples were stratified into two groups: high SIRT2 expression and low SIRT2 expression, based on the median SIRT2 expression. Filtering criteria included logFC absolute value greater than 1 and FDR less than 0.05 to extract differential genes between the two groups. Then GO, KEGG, and GSEA enrichment analyses were performed.

Gene correlation analysis

A histogram depicting SIRT2 expression in normal tissues and HCC was generated. Following this, gene correlation analysis was performed on HCC gene expression data to explore the relationship between the target gene SIRT2 and other genes. Correlation (corFilter = 0.5) and significance thresholds (pFilter = 0.05) were established. A correlation coefficient (cor) greater than 0 suggests a positive regulatory relationship between the gene and SIRT2, while a coefficient less than 0 indicates a negative regulatory relationship. The top 6 genes with the highest correlation coefficients and the bottom 5 genes with the lowest correlation coefficients were selected for visualization using circle diagrams, with positive regulation indicated in red and negative regulation in green. The expression of SIRT2 across different cancers was analyzed using TIMER2.0 (http://timer.cistrome.org/) [24].

Immune cell infiltration analysis and immune checkpoint analysis

The deconvo_mcpcounter function within the IOBR package was employed to assess immune cell infiltration among high and low expression groups of SIRT2 in HCC. Subsequent immune checkpoint analysis set a filtering threshold (pFilter = 0.05), calculating Pearson correlation coefficients between SIRT2 and immune checkpoint genes. Pearson correlation coefficients greater than 0 indicated positive correlations in red, whereas coefficients less than 0 represented negative correlations in green. Among patients with high SIRT2 expression, survival analysis was performed for patients with high and low CD274 expression.

Clinical feature analysis

We explored the relationships between clinical characteristics such as age, gender, TNM classification, grade, and stage in HCC, and SIRT2 expression. Finally, we examined whether these clinical features differed between the high and low SIRT2 expression groups.

Drug sensitivity analysis

The drug sensitivity analysis file was obtained from the Genomics of Drug Sensitivity in Cancer (https://www.cancerrxgene.org/) [25], and the oncoPredict package was employed to perform drug sensitivity analysis on the two groups with high and low SIRT2 expression. The Wilcoxon rank sum test compared drug sensitivity (IC50 value) differences between high and low SIRT2 expression groups. Sensitivity to the drug increases as the IC50 value decreases.

Results

SIRT2 increases the risk with HCC

Genes with a p_SMR value less than 0.05 and that encode proteins were screened, resulting in a total of 70 genes (Supplementary File 2). After intersecting with cuproptosis genes (Fig. 1A), one gene was identified: SIRT2. The p_HEIDI for SIRT2 was 0.61, indicating the absence of heterogeneity. The b_SMR value was 0.51, implying that SIRT2 expression is positively correlated with the risk of HCC. The SMRLocusPlot and SMREffectPlot for SIRT2 and HCC are illustrated in Fig. 1B, C. This analysis suggests that SIRT2 is a gene related to Cuproptosis and is positively correlated with HCC risk.

Fig. 1.

Result of Summary-data-based Mendelian Randomization. A Venn diagram of Mendelian randomization results and cuproptosis; B SMRLocusPlot of SIRT2 and HCC; C SMREffectPlot of SIRT2 and HCC. HCC: hepatocellular carcinoma

SIRT2 modulates tumor progression through tumor microenvironment cells and metabolic reprogramming

The annotated cells are visualized in Fig. 2A. The results of cell communication demonstrated that, compared to SIRT2-Mali, SIRT2 + Mali exhibited stronger interactions with three tumor microenvironment cells (CAF, TAM, and TEC) (Fig. 2B–E). SIRT2 + Mali primarily interacts with TAM through pathways involving SPP1, HLA, PROS1, and others. It interacts with CAF mainly via SPP1, JAG, and others, and with TEC primarily through VEGF, JAG, ANGPTL, and others. Metabolic analysis indicates that SIRT2 + Mali and SIRT2-Mali engage in distinct metabolic pathways, with SIRT2 + Mali participating in a broader range of metabolic pathways (Fig. 2F). This suggests that SIRT2 may influence tumor development through various metabolic pathways. Pseudotime analysis elucidated the trajectory of SIRT2 development and differentiation in HCC cells (Fig. 2G–I).

Fig. 2.

Results of scRNA-seq dataset analysis. A UMAP of scRNA-seq; B–E The results of intercellular communication. SIRT2 + Mali has stronger interactions with CAF, TAM, and TEC than SIRT2- Mali; F The results of metabolic analysis. SIRT2 + Mali participates in a wider range of metabolic pathways; G–I UMAP of pseudotime analysis. SIRT2 + Mali: SIRT2 + Maligant cell; SIRT2- Mali: SIRT2- Maligant cell; CAF: cancer-associated fibroblasts; TAM: tumor-associated macrophages; TEC: tumor-associated endothelial cells

Spatial co-localization of SIRT2 + Mali with tumor microenvironment cells reveals regional niche-specific interactions

The spatial distribution map indicates that the spatial distributions of SIRT2 + Mali and SIRT2-Mali differ substantially (Fig. 3A–E). Spatial correlation colocalization showed that SIRT2-Mail was negatively correlated with CAF, TAM and TEC, while SIRT2 + Mail was positively correlated with CAF, which illustrates the important interaction between SIRT2 and tumor microenvironment cells (Fig. 3F). The spatial distribution diagram of cell interactions and intercellular signals indicates that interactions between SIRT2 + Mali and the three tumor microenvironment cells predominantly occur in regions where SIRT2 + Mali is localized (Fig. 3G–I). Areas with high enrichment scores are also concentrated in the locations occupied by SIRT2 + Mali (Fig. 3J–L). This illustrates that SIRT2 + Mali has a close association with the three tumor microenvironment cells.

Fig. 3.

Results of stRNA-seq analysis. A–E Visualization of spatial distribution of SIRT2 + Mali, SIRT2- Mali, CAF, TAM, and TEC; F Spatial correlation colocalization revealed that SIRT2-Mail was negatively correlated with CAF, TAM and TEC, while SIRT2 + Mail was positively correlated with CAF. G Spatial distribution of SIRT2 + Mali interaction with CAF; H Spatial distribution of SIRT2 + Mali interaction with TEC; I Spatial distribution of SIRT2 + Mali interaction with TAM; J The enrichment scores of SIRT2 + Mali interaction with CAF; K The enrichment scores of SIRT2 + Mali interaction with TEC; L The enrichment scores of SIRT2 + Mali interaction with TAM; SIRT2 + Mali: SIRT2 + Maligant cell; SIRT2- Mali: SIRT2- Maligant cell; CAF: cancer-associated fibroblasts; TAM: tumor-associated macrophages; TEC: tumor-associated endothelial cells

DeepSurv and protein interaction

The differentially expressed genes of SIRT2 + Mali were extracted, yielding a total of 289 differentially expressed genes after the removal of ribosome-related genes. After extracting HCC and survival information, univariate Cox regression analysis of the differentially expressed genes, 100 genes were identified (Supplementary File 3). Subsequently, lasso regression analysis was performed, resulting in 14 key genes for deep learning (Fig. 4A–C). DeepSurv was employed to construct a prognostic model. As the number of training iterations increased, the negative log likelihood loss decreased, and the Concordance Index rose, indicating improved model performance (Fig. 4D, E). The trained model was utilized to predict the risk for the samples, which were classified into high-risk and low-risk groups based on the median risk score. Survival analysis results for the training set and validation set are presented in Fig. 4F–K, showing a significant difference in survival rates between patients with high risk scores and those with low risk scores (p < 0.05). The nomogram was used to predict survival of HCC patients based on risk scores, gender, stage, and age (Fig. 4L). The calibration curve indicates high accuracy of the nomogram in predicting survival for HCC patients (Fig. 4M). In the protein interaction analysis involving SIRT2 and the 14 key genes identified through machine learning, the pTM scores for SIRT2 with PON1 and SIRT2 with RAB13 were greater than 0.5, with ipTM scores exceeding 0.6 (Fig. 4N–Q). This suggests that SIRT2 may interact with PON1 and RAB13.

Fig. 4.

Deep Learning Survival Neural Network and protein interaction. A Path diagram of LASSO coefficients; B LASSO regression cross-validation curve; C PPI network of 14 key genes; D Concordance Index of the DeepSurv. E Negative log likelihood of the DeepSurv; F–G Survival curve and ROC curve of the training set; H Calibration Plot of the model in training set; I, J Survival curve and ROC curve of the test set; K Calibration Plot of the model in test set; L The nomogram of HCC patients based on risk scores, gender, stage, and age; M The calibration curve of the nomogram; N–O Protein docking diagram and expected position error of SIRT2 with PON1; P–Q Protein docking diagram and expected position error of SIRT2 with RAB13. HCC: hepatocellular carcinoma

RNA sequencing analysis and drug sensitivity analysis

KEGG enrichment analysis revealed that differentially expressed genes were predominantly enriched in Neuroactive ligand-receptor interactions, cAMP signaling pathways, and calcium signaling pathways (Fig. 5A). GSEA enrichment analysis indicated that the differential genes were enriched in stimulation and sensory pathways (Fig. 5B). GO enrichment analysis showed that these genes were primarily enriched in terms of cell adhesion, membrane adhesion, neuron synapse, gated channel activity, and ion channel activity (Fig. 5C, D). The expression levels of SIRT2 in HCC were significantly elevated compared to those in normal tissues (Fig. 5E). Gene correlation analyses indicated that SIRT2 was positively correlated with THAP8, EVI5L, USF2, MED29, GSK3A, and PSMD8 in HCC, while negatively correlated with MT-ND3, AKR1C4, TTC36, UGT2B10, and SLC25A47 (Fig. 5F). Pan-cancer analysis demonstrated that SIRT2 expression was upregulated in cholangiocarcinoma, esophageal carcinoma, and kidney chromophobe (Fig. 5G). Immune cell infiltration analysis revealed that the abundance scores of monocytic lineage, endothelial cells, and fibroblasts in the SIRT2 high expression group exceeded those in the low expression group, corroborated by results from cell communication and stRNA-seq analysis (Fig. 5H). Immune checkpoint analysis showed that SIRT2 exhibited positive correlations with the majority of immune checkpoint genes, indicating that SIRT2 may contribute to HCC immune escape (Fig. 5I). However, among patients with high SIRT2 expression, there was no significant difference in survival status between those with high CD274 expression and those with low CD274 expression (Fig. 5J). Clinical feature analysis indicated that female patients exhibited higher SIRT2 expression compared to male patients (Fig. 6A–G). SIRT2 expression was greater in patients classified as T2 compared to those classified as T1, and expression levels in stage IV patients were higher than in stage I patients. Notable differences were observed in stage and gender between the SIRT2 high and low expression groups (Fig. 6H). Drug sensitivity analysis showed the high SIRT2 expression group was more sensitive to Axitinib, Bortezomib, Gallibiscoquinazole, and et, al than the low SIRT2 expression group (Fig. 7).

Fig. 5.

Results of RNA-seq dataset analysis. A Bar plot of KEGG enrichment analysis results. B Results of GSEA enrichment analysis. C-D Bar plot and circle plot of GO enrichment analysis results. E Expression of SIRT2 in normal tissues and HCC tissues on TCGA. F Circle plot of gene correlation analysis. G Pan-cancer analysis shows increased expression of SIRT2 in multiple cancers. H Results of immune cell infiltration analysis. I Results of immune checkpoint analysis. J Among patients with high SIRT2 expression, the survival status of patients with high CD274 expression was not significantly different from that of patients with low CD274 expression

Fig. 6.

Results of clinical feature analysis. A–G Expression of SIRT2 in different age, gender, TNM, grade and stage of HCC. H Heat map of clinical features in HCC in the two groups with high and low expression of SIRT2. HCC: hepatocellular carcinoma

Fig. 7.

Results of drug sensitivity analysis. The SIRT2 high expression group was more sensitive to Bortezomib, Gallibiscoquinazole, and et al. than the SIRT2 low expression group. IC50 (half maximal inhibitory concentration)

Discussion

This study is the first to combine SMR with scRNA-seq, stRNA-seq and other omics technologies to explore the pathogenic role of SIRT2 in HCC. scRNA-seq analysis showed that SIRT2 + Mali had stronger interactions with tumor microenvironment cells (CAF, TAM, and TEC). SIRT2 + Mali and SIRT2-Mali were involved in different metabolic pathways, suggesting that SIRT2 may affect tumor development by regulating the metabolism of compounds. In stRNA-seq analysis, it can be observed that SIRT2 + Mali was closely associated with three tumor microenvironment cells, CAF, TAM, and TEC, compared with SIRT2-Mali. The location of cell interactions between SIRT2 + Mali and tumor microenvironment cells was concentrated in the spatial location of SIRT2 + Mali. Then we extracted the differentially expressed genes of SIRT2 + Mali and finally obtained 14 key genes through single cox analysis and lasso regression and constructed a prognostic model using deepsurv. RNA-seq analysis showed that the SIRT2 high expression group had a higher abundance score of immune cells. This also suggests that SIRT2 is closely associated with the tumor microenvironment. Immune checkpoint analysis showed that SIRT2 was positively correlated with most immune checkpoints. In conclusion, SIRT2 can be used as a biomarker for HCC and may provide a new target for the treatment of HCC.

Sirtuin 2 (SIRT2) is a nicotinamide adenine dinucleotide-dependent deacetylase expressed in multiple organs, including the liver, and plays an important and complex role by interacting with multiple substrates. Previous studies have shown that the expression of SIRT2 in liver fibrosis tissue is significantly higher than that in non-fibrotic tissue [26]. In the hepatitis B virus, SIRT2 can promote chronic liver damage and immune system disorders and ultimately lead to cirrhosis and liver cancer [27]. Increased expression of SIRT2 reduces hepatocyte ferroptosis [28]. SIRT2 can also deacetylate AKT, promote its phosphorylation, and inhibit the activity of GSK-3β, thereby reducing the phosphorylation and degradation of β-catenin [29, 30]. Elevated expression of β-catenin promotes the growth and proliferation of liver cancer cells [29].

Copper plays an important role in cells and is a catalytic cofactor of essential enzymes involved in energy conversion, oxygen transport, and regulation of cellular oxidative metabolism [7]. Intracellular copper accumulation triggers mitochondrial lipoprotein aggregation and protein instability, leading to a unique type of cell death called cuproptosis [31]. The key role of copper in anti-tumor therapy and tumor immunity cannot be underestimated [32, 33]. In addition, with the emergence of nanotechnology and biomaterials, copper has been designed into various structures to exert its therapeutic properties [34]. Cuproptosis is closely related to cancer progression and is expected to become a new therapeutic target for the treatment of cancer [35]. Previous studies have discovered genes related to cuproptosis in hepatocellular carcinoma, such as LIPT1, CDKN2A, ATP7A, SLC25A3, etc [35–37]. Previous studies have rarely combined cuproptosis-related genes with SMR and stRNA-seq. Our study is the first to combine SMR with multi-omics such as stRNA-seq and found that SIRT2 can be used as a therapeutic target for cuproptosis-related therapeutic target for HCC. SIRT2-mediated deacetylation of phosphoglycerate mutase 5 (PGAM5) activates malic enzyme 1 (ME1) activity, leading to ME1 dephosphorylation, subsequent lipid accumulation and proliferation of hepatoma cells [38]. In gastric cancer, SIRT2 inhibits the lactic acidification process of METTL16, thereby reducing the copper concentration in gastric cancer and inhibiting cuproptosis [38]. SIRT2 can also affect copper death by increasing the level of nicotinamide adenine dinucleotide phosphate (NADPH) in cells [39]. Therefore, SIRT2 may affect cuproptosis in HCC through the fatty acylation components of the tricarboxylic acid (TCA) cycle, the level of NAPDH and N6-methyladenosine modification, thereby promoting tumor development. Further studies are needed in the future to prove these mechanisms.

In scRNA-seq metabolic analysis, it can be seen that SIRT2 + Mali and SIRT2-Mali involve different metabolic pathways, and SIRT2 + Mali involves more metabolic pathways, such as vitamin B6 metabolism, retinol metabolism, terpenoid backbone biosynthesis, porphyrin and chlorophyll metabolism, etc. We speculate that SIRT2 is related to the acetylation of some enzymes in vitamin B6 metabolism. A prospective cohort study showed that total vitamin B6 intake was positively correlated with HCC risk [40]. Higher serum retinol concentrations were negatively correlated with the risk of HCC [41]. This association may involve SIRT2-mediated regulation of retinoic acid receptor acetylation status [42]. Caffeine has been shown to inhibit liver fibrosis by blocking adenosine receptors to inhibit hepatic stellate cell activation [43]. Caffeine can also have a beneficial effect on angiogenesis and liver hemodynamics [44]. Caffeine may affect HCC by improving mitochondrial dysfunction and inhibiting oxidative stress via SITR2 [45]. Terpenoid backbone biosynthesis is a basic pathway for the synthesis of various bioactive agents in the human body and is active in many cancers, including HCC [46]. The inhibitory effect of chlorophyll on cancer can be observed in mouse models [47, 48]. These nutritional factors may create a metabolic microenvironment where SIRT2 + cells gain selective advantage through enhanced detoxification (chlorophyll derivatives) and energy metabolism (terpenoid backbones). Therefore, SIRT2 may promote tumor development by rewiring metabolic networks through both direct enzymatic regulation and microenvironmental adaptation.

Enrichment analysis showed that differentially expressed genes were mainly enriched in neuroactive ligand-receptor interaction, signaling pathways, and cell adhesion. Neuroactive ligand-receptor interaction can affect cell-to-cell necroptosis, thereby accelerating the replication of cancer cells and leading to the rapid spread of cancer cells. Previous studies have shown that the cAMP pathway and Wnt signaling are involved in the pathophysiological process of HCC [49–52]. Changes in intracellular Ca2+ constitute an important signaling pathway that regulates lipid and carbohydrate metabolism in normal hepatocytes [53]. Lipids can activate Ca2 + to flow out of the endoplasmic reticulum and transfer to mitochondria, leading to the inhibition of lipolysis, activation of lipogenesis, generation of reactive oxygen species, activation of Ca2+/calmodulin-dependent kinases, and activation of transcription factor Nrf2, thereby promoting the occurrence of HCC [53]. Proteoglycans are mainly present in particles in the extracellular matrix, cell surface, and cytoplasm, thereby playing an important signal transduction function [54]. The content of proteoglycans in HCC tissue is higher than that in normal liver tissue [55]. Cell adhesion is associated with tumor growth, adhesion, and metastasis [56, 57].

Immune checkpoint analysis showed that SIRT2 can activate multiple immune checkpoints, suggesting that SIRT2 can promote HCC immune escape. CD86, HAVCR2, and CD274 are associated with the abundance of NK cells [58]. CD28 is associated with T cell expansion [59]. CD200 is expressed on T, B, NK, and myeloid cells and is also considered a marker of tumor progression [60]. CD276 expression is associated with the infiltration of immune cells, including high levels of tumor-associated macrophages and low levels of CD8 + T cells [61]. CD276 silencing in HCC cells and co-culture with THP-1-derived macrophages have a regulatory effect on macrophage polarization and macrophage-mediated cell proliferation and migration [61]. PDCD1LG2 is associated with the prognosis of HCC and promotes tumor progression through the p53 and STAT3 signaling pathways [62]. The TNF receptor superfamily (TNFRSF) is associated with apoptosis of cancer cells and stimulation of anti-tumor T cells [63]. These immune checkpoints are all related to the pathogenesis of HCC, and new therapeutic drugs can be developed against these immune checkpoints in the future.

Advantages and limitations

This study is the first to discover a cuproptosis-related gene SIRT2 in HCC through comprehensive analysis of SMR, multi-omics, and deep learning and explore its role in HCC. It lays a solid foundation for the development of new targeted therapeutic strategies for SIRT2. However, there are limitations to our analysis that should be noted. Firstly, the GWAS data used in our study mainly involved European populations, which may limit the generalization of our research results to other populations. Secondly, in the analysis of immune checkpoints, combined with clinical prognostic data analysis, we found that among patients with high SIRT2 expression, there was no significant difference in survival status between patients with high CD274 expression and patients with low CD274 expression. The reason why the p-value is greater than 0.05 is that we speculate that it may be due to bias caused by insufficient sample size. In the future, we will increase the sample size to improve this analysis. Lastly, the results of multi-omics and AlphaFold 3 in this study are only bioinformatics prediction results, lacking experimental verification, which will be the research content of our future plans.

Conclusion

In summary, our study concluded that SIRT2 is a pathogenic gene associated with HCC and cuproptosis. The combination of multiple analytical methods such as SMR, multi-omics, and deep learning can make the research results more credible. SIRT2 may promote the growth and invasion of HCC through multiple mechanisms, such as cuproptosis, promoting tumor microenvironment, affecting metabolic pathways, and activating immune checkpoints. SIRT2 can be used as a biomarker for HCC and may provide a new target for the treatment of HCC. Future studies can focus on verifying the mechanism and function of SIRT2 in HCC and opening up new treatments.

Author contributions

Study concept and design: Y.W., C.C. Data acquisition and analysis: Y.W., C.C., T.Y. Writing original draft: C.C. Interpretation of data: C.C., T.Y., C.Y., L.C., B.C., Z.L. Writing review and editing: S.Y., Y.W. All authors reviewed and approved the final manuscript.

Funding

This study was supported by the Affiliated hospital of Guangdong Medical University High-level Talent Start-up Project (GCC2023018).

Data availability

The original data presented in the study are openly available in the FINNGEN (https://r10.finngen.fi/pheno/C3_HEPATOCELLU_CARC_EXALLC) and GEO.

Declarations

Ethics approval and consent to participate

This study did not require any ethics approval. Only secondary analyses were performed using publicly accessible data. All original studies received approval from the relevant ethical review boards, and informed consent was obtained from all participants.

Informed consent

Not applicable.

Competing interests

The authors declare no competing interests.