Prognostic and immunological implications of sialylation-associated gene signatures in hepatocellular carcinoma Authors: Guan-Qing Wang, Kang Du, Yu-Peng Wang Published: 2025 ABSTRACT AbstractObjectiveThe absence of effective biomarkers continues to limit early diagnostic accuracy and prognostic evaluation in patients with hepatocellular carcinoma (HCC). Aberrant sialylation (SI) has been demonstrated to contribute to therapeutic resistance and tumor progression. The aim of this investigation was to identify a sialylation-related gene (SRG) signature, evaluate its prognostic significance, and investigate associated immunological characteristics in HCC.MethodsTranscriptomic profiles and corresponding clinical data for patients with HCC were obtained from UCSC Xena, the International Cancer Genome Consortium (ICGC), and the Molecular Signatures Database (MsigDB). Differential expression analysis, Cox regression analysis modeling, and least absolute shrinkage and selection operator (LASSO) regression analysis were applied to identify independent prognostic markers and develop predictive models. The tumor immune microenvironment and its relationship with the identified SRGs were assessed by evaluating immune infiltration patterns. A gene co-expression network for the prognostic SRGs was constructed using GeneMANIA to identify potentially targetable signaling pathways.ResultsFour SRGs (ST6GALNAC4, B4GALT5, B4GALNT1, and NEU1) were significantly associated with the prognosis of patients with HCC. Prognostic models constructed using these genes demonstrated strong predictive performance. Notable differences were observed in immune cell populations and immune checkpoint expression between the high-risk and low-risk groups. Additionally, the half-maximal inhibitory concentration values for 101 therapeutic compounds varied between these groups. Lipopolysaccharide and sphingolipid metabolism were identified as key biological processes linked to tumor progression and modulation of the immune microenvironment.ConclusionThe four identified SRGs were significantly associated with clinical outcomes and immunological features in HCC. These findings provide a foundation for advancing early diagnostic strategies, refining prognostic assessments, and guiding personalized therapeutic approaches for patients with HCC.Keywords: Biomarker, hepatocellular carcinoma, immune microenvironment, prognosis, sialylation Introduction Hepatocellular carcinoma (HCC) is the third leading cause of cancer-related mortality worldwide and remains a major public health challenge [1]. However, most patients are diagnosed at an advanced stage, which contributes to a poor overall survival rate [2]. Advancements in prevention, early detection, and the development of novel therapeutic modalities remain essential for improving long-term survival in this population [3]. Tumor-associated molecular markers can reflect the biological characteristics of HCC and may serve as predictors for early recurrence and survival [4]. Consequently, the identification and evaluation of prognostic biomarkers in HCC are vital for guiding clinical decision-making. Sialylation (SI) refers to the increased modification of glycoproteins with sialic acid at their terminal ends on the cell membrane, a process mediated by sialyltransferases. This post-translational modification is critical for several biological functions, including cell–cell recognition, immune regulation, and signal transduction [5–9]. Aberrant SI plays a key role in tumorigenesis and metastasis by promoting tumor cell survival, invasiveness, angiogenesis, immune evasion, and resistance to therapy [10,11]. Elevated expression of sialic acid on tumor cell surfaces may contribute to immune escape by masking antigenic epitopes and activating immunosuppressive receptors belonging to the Siglec family [12,13]. Existing evidence has indicated the presence of abnormal sialidase protein modifications and dysregulated sialic acid–sialidase interactions in the pathogenesis of HCC [9,10,14]. ST6GAL1 has been demonstrated to suppress HCC metastasis by modulating the SI index of MCAM on the tumor cell surface [9]. Additionally, sialylated IgG was reported by Wu et al. to inhibit the cGAS-STING signaling pathway via the DC-SIGN/ATF3 axis, thereby suppressing the type I interferon-mediated anti-tumor immune response [10]. In patients with HCC, irregular expression of the sialidase NEU1 has been associated with unfavorable prognosis, potentially due to its involvement in oncogenic pathways and immunosuppressive mechanisms [15]. These findings indicate that genes involved in SI may serve as candidate biomarkers for predicting HCC prognosis and represent promising targets for therapeutic intervention. The present study utilized publicly available datasets to comprehensively assess the relationship between SI-related genes and prognosis in HCC. The findings offer a systematic overview of SI-related gene expression in HCC and highlight their potential relevance to disease progression. Moreover, this analysis provides insight into the prospective clinical utility of immunotherapeutic and targeted treatment approaches aimed at modulating SI pathways. Materials and methods Data collection Transcriptomic and clinical data for HCC were obtained from multiple public databases. The TCGA-LIHC dataset was downloaded from UCSC Xena (https://xena.ucsc.edu/). A total of 365 HCC tumor samples and 50 control samples with comprehensive gene expression and survival data were included. Additionally, the ICGC-LIRI-JP dataset, comprising 231 patients with HCC with complete gene expression and survival information, was obtained from the ICGC database (https://dcc.icgc.org/). A total of 106 sialylation-related genes were retrieved from the MSigDB (https://www.gsea-msigdb.org/gsea/index.jsp) [16]. Analysis of differential gene expression Differential expression analysis was performed using the ‘DEseq2’ package by comparing tumor and normal samples [17]. Significantly differentially expressed genes were identified using an adjusted p-value < 0.05 and |log2 fold change (FC)| > 0.5 as cut-off criteria. A volcano plot was generated using the ‘ggplot2’ package, and a heatmap of all differentially expressed genes was produced using the ‘ComplexHeatmap’ package [18,19]. Identification and functional analyses of candidate genes The intersection between differentially expressed genes and sialylation-related genes was identified using the ‘Evenn’ tool, and the overlapping genes were defined as candidate genes associated with sialylation in HCC. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were conducted using the ‘clusterProfiler’ package [20]. Additionally, the candidate genes were submitted to STRING (https://string-db.org/) to construct a protein–protein interaction (PPI) network. The resulting network was visualized using Cytoscape [21]. Acquisition of prognostic genes Based on the constructed PPI network, five algorithms within Cytoscape were applied to rank genes. The top 30 genes from each algorithm were intersected to identify core genes for further analysis. To explore associations between these core genes and clinical outcomes, gene expression data with corresponding survival duration and status from TCGA-LIHC were analyzed. Univariate Cox regression analysis was conducted using the ‘survival’ package [22]. Genes meeting these criteria were further analyzed using LASSO regression analysis with the ‘glmnet’ package, employing 10-fold cross-validation and Cox proportional hazards modeling [23]. The model with the smallest lambda (λ) value was selected, and the corresponding genes were identified as prognostic genes. Prognostic model construction and validation Risk scores in the TCGA-LIHC dataset were calculated using the following formula: risk score=∑i=1n(coef(genei)×expr(genei)). Samples were divided into high-risk and low-risk groups based on the median risk score. Kaplan–Meier survival curves were plotted using the ‘survminer’ package, and time-dependent receiver operating characteristic (ROC) curves were generated using the ‘survivalROC’ package to assess predictive performance [24,25]. Model validation was performed using the same methodology in the ICGC-LIRI-JP cohort. Independent prognostic analysis and nomogram construction Cox regression analyses were conducted to identify independent prognostic factors by combining risk scores and clinical variables from TCGA-LIHC. A prognostic nomogram for overall survival (OS) was developed using the ‘rms’ package [26]. Calibration curves and time-dependent ROC curves were generated using the ‘rms’ and ‘timeROC’ packages to evaluate model accuracy [26]. Decision curve analysis (DCA) was performed using the ‘ggDCA’ package to assess the clinical utility of the nomogram [27]. Analysis of immune microenvironment To assess immune infiltration between high-risk and low-risk groups, the CIBERSORT algorithm was applied using the ‘IOBR’ package [28,29]. The Wilcoxon test was used to determine statistical significance (p < 0.05) of differences in immune cell infiltration. Correlations between prognostic genes and immune cell subtypes were evaluated using Spearman analysis, with correlation coefficient |cor| > 0.3 and p < 0.05 considered significant. Immunotherapy response analysis TIDE (Tumor Immune Dysfunction and Exclusion) scores, which predict immunotherapy response, were obtained from the TIDE database (http://tide.dfci.harvard.edu/). Comparisons between high-risk and low-risk groups were conducted using the Wilcoxon test (p < 0.05). Correlations between prognostic genes and differentially expressed immune checkpoint genes were analyzed using Spearman correlation (|cor| > 0.3, p < 0.05). Drug sensitivity analysis Chemotherapy response in patients with TCGA-LIHC was estimated using the ‘pRRophetic’ package, which calculates half-maximal inhibitory concentration (IC50) values for each sample [30,31]. Functional similarity and co-expression network To evaluate functional similarity among prognostic genes, the ‘GOSemSim’ package was used based on GO Biological Process (GO-BP), Cellular Component (GO-CC), and Molecular Function (GO-MF) annotations [32]. Genes were ranked based on average similarity scores. Additionally, the GeneMANIA tool (http://genemania.org/) was used to construct a co-expression network to identify functionally related genes. Reverse transcription-quantitative polymerase chain reaction Reverse transcription-quantitative polymerase chain reaction (RT-qPCR) was conducted to validate the expression levels of prognostic genes in HCC. Tissue samples from 24 patients with HCC and 25 controls were collected from Bethune International Peace Hospital. Primer sequences are presented in Supplementary Table 1. Statistical analysis All analyses were performed in R. The Wilcoxon and Kruskal–Wallis tests were used to compare groups. For RT-qPCR data, independent t-tests were applied. Statistical significance was defined as p < 0.05. Results Identification of candidate prognostic genes associated with SI A total of 8525 differentially expressed genes (DEGs) were identified between tumor and normal tissues in the TCGA-LIHC dataset, including 5853 upregulated and 2672 downregulated genes (|log2FC| > 0.5, adj. p < 0.05) (Figures 1A,B). Among these, 50 candidate genes were identified by intersecting the DEGs with 106 sialylation-related genes (SRGs) (Figure 1C). GO and KEGG analyses were conducted to investigate the biological functions and pathways associated with these candidate genes. GO enrichment analysis identified 115 significantly enriched terms, including 72 in biological processes, 18 in cellular components, and 25 in molecular functions. KEGG pathway analysis revealed 8 significantly enriched pathways, including those involving cell adhesion molecules. Networks demonstrating the relationships between the top 5 enriched GO and KEGG terms and candidate genes were constructed (Supplementary Figures S1A,B). Full details of enrichment pathways are presented in Supplementary Tables S2 and S3. Based on the PPI network and 5 ranking algorithms, 24 core genes were identified (Figures 1D,E). Univariate Cox regression analysis indicated that 7 genes among these 24 were associated with poor prognosis in patients with HCC (Supplementary Table S4). LASSO regression analysis further identified 4 prognostic genes—B4GALNT1, B4GALT5, NEU1, and ST6GALNAC4—at a lambda.min of 0.0193 (Figures 1F,G). To experimentally validate the bioinformatically identified SRGs, RT-qPCR was performed on tissue samples obtained from individuals diagnosed with HCC and matched adjacent non-tumor tissues. All four genes (B4GALNT1, B4GALT5, NEU1, and ST6GALNAC4) exhibited significantly elevated expression levels in HCC tissues relative to control tissues (Figure 1H). These results indicated the transcriptomic findings and supported the potential of these SRGs as prognostic biomarkers in HCC. Identification and validation of prognostic sialoglycan-related genes (SRGs). (A,B) Heatmap (A) and volcano plot (B) showing DEGs; (C) Venn diagram depicting the overlap between DEGs and SRGs; (D,E) intersection analysis of five prioritization methods (degree, DMNC, EPC, MCC, and MNC) using flower plot (D) and upset (E); (F,G) feature selection using the LASSO Cox regression model. (H) Validation of identified four SRGs expression by RT-PCR. (HCC group: n = 24; control group: n = 25). Bars represent the mean ± SEM; ****p < 0.0001. Establishment and evaluation of prognostic model based on sialylation-related genes and baseline characteristics Based on the four identified sialylation-related genes (ST6GALNAC4, B4GALT5, B4GALNT1, and NEU1), a prognostic risk score formula was developed as follows: Risk score = (0.2227 × exp ST6GALNAC4) + (0.2218 × exp B4GALT5) + (0.0962 × exp B4GALNT1) + (0.1038 × exp NEU1). Using the median risk score of 1.6170 as the threshold, the 365 HCC samples from TCGA-LIHC were divided into a high-risk group (n = 182) and a low-risk group (n = 183). Similarly, based on a median risk score of 6.7332, the 231 HCC samples from ICGC-LIRI-JP were categorized into high-risk (n = 115) and low-risk (n = 116) groups. Distribution plots revealed that higher risk scores were associated with increased mortality and shorter overall survival in both datasets (Figure 2A and Supplementary Figure S2A). The area under the curve (AUC) values at 1, 2, and 3 years exceeded 0.6 in the TCGA-LIHC cohort and surpassed 0.7 in the ICGC-LIRI-JP cohort, indicating satisfactory prognostic discrimination (Figure 2B and Supplementary Figure S2B). Kaplan–Meier survival analysis further showed that patients in the high-risk group had significantly worse survival outcomes compared to those in the low-risk group across both datasets (p < 0.0001) (Figure 2C and Supplementary Figure S2C). Construction and validation of prognostic model based on risk scores generated from the identified four SRGs in TCGA cohort and other baseline characteristics. (A) Risk scores distribution; (B) ROC curves at 1, 2, and 3 years according to the risk scores; (C) survival analysis of HCC patients’ prognosis in different groups; (D–F) violin plots showing the distribution of risk scores across different genders, tumor grades and tumor stages; (G) individualized prediction nomogram model; (H) the receiver operating characteristic (ROC) curve of the nomogram prediction model. (I) The decision curve analysis (DCA) plot showing the net benefit of the model at different high-risk thresholds. Bars represent the mean ± SEM. Statistical comparisons of baseline characteristics indicated significant differences in risk scores across multiple clinical features (p < 0.05). Notably, female patients had higher risk scores than male patients, and individuals with advanced tumor stages and grades also demonstrated elevated risk scores (Figures 2D–F). Univariate and multivariate Cox regression analyses were conducted to identify independent prognostic factors. Both tumor stage and the risk score derived from the four SI-related genes were found to be independently associated with overall survival (Supplementary Table S5). Based on these two factors, a prognostic nomogram was constructed to predict 1-, 2-, and 3-year mortality risk in patients with HCC (Figure 2G). Higher total points on the nomogram indicated increased mortality risk. Calibration plots for the nomogram showed that predicted survival closely matched observed outcomes at all three time points, confirming the model’s accuracy (Supplementary Figure S3). The nomogram yielded AUC values of 0.730, 0.660, and 0.690 at 1, 2, and 3 years, respectively, reflecting reliable prognostic performance (Figure 2H). Furthermore, DCA demonstrated that the nomogram provided substantial net clinical benefit across a wide range of threshold probabilities (Figure 2I). Collectively, these results supported the clinical utility of the developed prognostic model based on the four SI-related genes, which may serve as a valuable tool for predicting survival outcomes and informing treatment decisions in patients with HCC. Exploration of tumor immune microenvironment (TIME) heterogeneity and drug sensitivity between risk groups Substantial evidence has indicated that the tumor microenvironment (TME), particularly the phenotype of immune cells within the TME, plays a critical role in determining both prognosis and therapeutic efficacy in HCC [33]. To gain further insight into the immune distinctions between high-risk and low-risk groups, the relative abundance of immune cell infiltration was analyzed across individual samples, as presented in Figure 3A. Statistically significant differences were identified in the infiltration levels of 14 immune cell types between the two groups (Figure 3B), indicating that shifts in immune cell composition may critically influence HCC progression. Correlation analysis between the four prognostic SRGs and immune cell types revealed a negative association between NEU1 and M2 macrophages (cor = −0.14, p = 0.0078), and a strong positive association between ST6GALNAC4 and regulatory T cells (Tregs) (cor = 0.34, p = 2.25 × 10−11) (Figure 3C), with full correlation data presented in Supplementary Table S6. The TIDE score, which reflects immune escape potential and immunotherapy responsiveness, was significantly elevated in the high-risk group compared to the low-risk group (Figure 3D). Additionally, expression levels of eight immunosuppressive checkpoint genes were higher in the high-risk group (Figure 3E). Correlation analysis showed that all four prognostic SRGs were significantly associated with the expression of immune checkpoint genes, with ST6GALNAC4 demonstrating consistent associations with all eight, and the strongest correlation observed between ST6GALNAC4 and LGALS9 (cor = 0.64, p = 1.17 × 10−42) (Figure 3F; Supplementary Table S7). Drug sensitivity analysis revealed that 77 compounds exhibited greater efficacy in the high-risk group, while 24 compounds were more effective in the low-risk group (Figure 3G), with the top 10 drugs based on statistical significance displayed in Figure 3H. Among these, GW.441756 demonstrated higher sensitivity in the low-risk group (p < 0.0001), whereas nine compounds, including BI.2536 and FTI.277, demonstrated increased sensitivity in the high-risk group (p < 0.0001). These findings support the utility of the SRG-based prognostic model in guiding individualized therapeutic strategies and highlight these SRGs as promising biomarkers or therapeutic targets in HCC. Immune landscape and drug sensitivity differences between high- and low-risk groups. (A) Relative proportions of infiltrating immune cells in different risk subgroups; (B) differences in immune cell types in different risk subgroups; (C) heatmap showing the correlations between the identified four SRGs and immune cells types; (D) violin plot of the TIDE score in different risk subgroups; (E) differences in immune checkpoint genes in different risk subgroups; (F) heatmap showing the correlations between the identified four SRGs and immune checkpoint markers; (G) waterfall plot illustrating the differential sensitivity in different risk subgroups to various drugs determined by the difference in median IC50 values; (H) boxplots displaying the IC50 values for selected drugs in different risk subgroups. Bars represent the mean ± SEM; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001; nsp > 0.05. In-depth exploration of SRG-related interactions Considering that genes often participate in multiple biological processes, and individual biological functions can be regulated by several genes, further exploration of SRG-related interactions was conducted. Analysis of the 50 candidate genes revealed strong interaction networks (Figure 4A). Functional similarity analysis identified high similarity among ST6GALNAC4, B4GALT5, and B4GALNT1 (Figure 4B). GeneMANIA analysis further demonstrated that the prognostic SRGs interact with other functionally relevant genes through co-expression, co-localization, and physical interactions, and were notably enriched in lipopolysaccharide and membrane lipid metabolic processes (Figure 4C). The above findings not only clarify the potential molecular regulatory roles of SRGs in HCC but also lay a solid foundation for subsequent in-depth research. In-depth exploration of SRG-related interactions. (A) Protein-protein interaction network of SRGs; (B) functional similarity analysis for the identified four SRGs; (C) protein-protein interaction network of the identified four SRGs (GeneMANIA). Discussion Unlike malignancies driven by single genetic mutations or pathway disruptions, complex HCC’s pathogenesis makes the identification of reliable prognostic biomarkers and therapeutic targets particularly challenging [34]. Despite previous studies suggesting a potential role for sialylation in tumor development, definitive evidence linking specific SRGs to HCC progression has remained limited [10,11]. The present study identified four SRGs (ST6GALNAC4, B4GALT5, B4GALNT1, and NEU1) with robust associations with HCC survival and TIME features through data analysis combined with experimental verification, providing new insights into personalized prognosis and treatment strategies. Aberrant sialylation is a hallmark observed in several pathologic conditions, especially in malignancies. Man et al. demonstrated ST6GALNAC4 promotes HCC cell proliferation and migration by inducing abnormal sialylation of cell surface proteins, which directly supports our observation of ST6GALNAC4 upregulation in HCC tissues and its positive correlation with high-risk status [35]. Additionally, Wang et al. showed that B4GALNT1 enhances HCC cell invasiveness via O-glycosylation-dependent activation of the SPP1-TAM axis, which aligns with our data on B4GALNT1’s association with poor prognosis and immunosuppressive infiltration [36]. Meanwhile, Wu et al. reported sialylated IgG inhibits anti-PD-1 immunotherapy efficacy in HCC by suppressing the cGAS-STING pathway, which have strengthened our discussion of TIDE score differences between high/low-risk groups and explain why high-risk patients may have poorer immunotherapy responses, reinforcing the translational relevance of our SRG signature for treatment stratification [37]. Moreover, Bie et al. showed B4GALNT1 mediates O-glycosylation of the RTK MET, preventing its degradation and sustaining activation of the PI3K-AKT-mTOR pathway, which directly supports our earlier finding that B4GALNT1 is an independent prognostic factor and explains why its upregulation correlates with high-risk status [38]. Jiang et al. demonstrated that B4GALT5 enhances EGFR signaling via O-glycosylation of EGFR’s extracellular domain, which clarifies our observation of B4GALT5′s positive correlation with PD-L1 expression, as EGFR activation is known to upregulate PD-L1 in HCC [39]. The present study builds on but distinctively advances prior SRG research in HCC. Specifically, whereas prior studies focused on isolated SRGs, our four-SRG signature demonstrated superior prognostic performance across independent cohorts. Additionally, unlike earlier research that merely established gene-prognosis associations, we have developed a nomogram with robust clinical utility for risk stratification. Notably, this is the first study to systematically connect SRGs to immunosuppressive TIME characteristics and metabolic pathways, addressing critical gaps in mechanistic understanding. Furthermore, this study provides evidence that SRG expression could guide both chemotherapy and immunotherapy, a translational direction that was absent from prior literature. To provide new insights for the development of HCC therapeutic targets, the roles of each SRG in mediating metabolic-immune crosstalk and the potential biological mechanisms underlying the associations between SRGs, TIME remodeling, and HCC progression were further explored. ST6GALNAC4, a sialyltransferase that catalyzes α-2,6-sialylation of glycoproteins/glycolipids, mediates α-2,6-sialylation of LPS-binding glycolipids on HCC cell surfaces, which reduces LPS binding to TLR4 by masking the LPS-recognition domain of TLR4 co-receptor CD14, thereby attenuating acute NF-κB-dependent pro-inflammatory cytokine release [40,41]. However, long-term ST6GALNAC4 overexpression paradoxically enhances the secretion of immunosuppressive cytokines by activating the PI3K-AKT pathway, which recruits regulatory T cells and M2-type macrophages [40,42]. B4GALT5, a galactosyltransferase involved in glycosphingolipid biosynthesis, directly controls the production of lactosylceramide (LacCer)—a central intermediate in sphingolipid metabolism [37,43]. In HCC, B4GALT5 overexpression increases LacCer accumulation, which activates the downstream sphingosine-1-phosphate (S1P) pathway [41]. S1P, a bioactive sphingolipid, binds to S1P receptor 1 (S1PR1) on T cells to impair their migration into tumor tissues, while promoting the recruitment of myeloid-derived suppressor cells [44]. Additionally, LacCer enhances the expression of hypoxia-inducible factor 1α (HIF-1α) via reactive oxygen species generation [45]. HIF-1α further upregulates glycolytic enzymes to increase lactate secretion, which acidifies the TME and inhibits CD8+ T cell function [42]. B4GALNT1, catalyzing β-1,4-N-acetylgalactosamine addition to glycosphingolipids, integrates sphingolipid metabolism with oncogenic signaling to shape immunosuppression. B4GALNT1 overexpression promotes the synthesis of ganglioside GD3, which activates the PI3K-AKT-mTOR pathway [36,38]. This pathway not only drives HCC cell proliferation but also upregulates the expression of PD-L1 by phosphorylating the transcription factor STAT3 [41]. This cross-talk between B4GALNT1, sphingolipids, and LPS signaling reinforces immune evasion in HCC. NEU1, a lysosomal sialidase that cleaves terminal sialic acids from glycoconjugates, regulates both LPS metabolism and sphingolipid turnover [46]. In HCC, NEU1 overexpression increases the desialylation of LPS-binding glycoproteins, enhancing LPS retention in the TME [15]. Persistent LPS exposure activates the TLR4-NF-κB pathway in TAMs, promoting the secretion of TGF-β and arginase 1, molecules that suppress CD8+ T cell cytotoxicity [46]. Additionally, NEU1 mediates the desialylation of ganglioside GM2, leading to its degradation into ceramide [46,47]. While ceramide typically induces apoptosis, in HCC cells with activated PI3K-AKT, ceramide is converted to S1P, which further enhances MDSC recruitment [41]. This ‘ceramide-S1P switch’ explains why NEU1 expression correlates with both high-risk scores and reduced apoptosis of immunosuppressive cells. Based on the above observations, we propose a unifying mechanistic hypothesis: the four SRGs drive LPS-sphingolipid metabolic crosstalk, which remodels the TME into an immunosuppressive state via LPS-TLR4-NF-κB axis, sphingolipid-S1P pathway, and hypoxia-glycolysis axis [44,48]. These findings suggest that SRGs drive HCC progression via metabolic-immune crosstalk, offering a new framework for understanding sialylation’s role in disease pathogenesis. This study has several limitations. First, the precise molecular mechanisms of SRGs regulating TIME and metabolism remain unclear, and in vivo validation is still required to confirm their pro-tumor roles, leading to mechanistic and experimental gaps. Second, CIBERSORT and TIDE analyses are descriptive, and the signature’s ability to predict ICI response has not been validated in prospective immunotherapy-treated HCC cohorts; future studies should collect pre-/post-treatment tumor tissues and clinical data from ICI-treated patients to address this in silico analysis constraint. Third, IC50 differences in this study’s drug sensitivity analysis are all computationally simulated, only reflecting the computational association between SRG expression patterns and drug responses. They cannot exclude physiological factors like transcriptome-proteome discrepancies or TME-mediated drug distribution, nor replace in vitro/in vivo validation, and their value is only to narrow the drug range for subsequent experiments rather than guide clinical medication. Fourthly, this study explored SRG characteristics and their associations with lipid metabolism and immune infiltration using public databases, but lacked a multi-omics perspective. FastUKB overcomes traditional cloud platform limitations by supporting batch extraction of unlimited variables, automatic multi-type data entity matching, and built-in medical data quality control and future studies can use it to integrate multi-dimensional data in the UK Biobank for cross-population validation [49]. Additionally, AI-enhanced ‘genome-transcriptome-proteome-metabolome integration’ is key to resolving complex disease mechanisms [50]. Future studies could integrate multi-omics data to clarify SRGs’ role in glycosylation-mediated lipid metabolism and train AI models for better personalized treatment recommendations [50–52]. Aberrant sialylation is a hallmark observed in several pathologic conditions, including malignancies. Identifying SRGs may facilitate early detection and offer new directions for improving therapeutic efficacy and survival. Despite previous studies suggesting a potential role for sialylation in tumor development, definitive evidence linking specific SRGs to HCC progression has remained limited. In this study, both integrated computational analysis and experimental validation led to the identification of four novel SRGs associated with HCC. ST6GALNAC4 encodes a sialyltransferase involved in glycoprotein biosynthesis and glucose and lipid metabolism [35]. Prior research reported that ST6GALNAC4 promotes tumorigenesis and is associated with immune infiltration by mediating aberrant glycosylation patterns in HCC [36]. B4GALT5, a galactosyltransferase family member, functions in glycosylation and glucose metabolism and has been implicated in cell proliferation [37]. Elevated expression of B4GALT5 has been associated with the progression of various malignancies, including esophageal squamous cell carcinoma and ovarian cancer [38,39]. B4GALNT1 contributes to glucose and glycosphingolipid metabolism through its activity on β-1,4-galactose and N-acetylgalactosamine residues, facilitating HCC progression via the activation of the PI3K-AKT-mTOR pathway and modulation of the HES4-SPP1-TAM/Th2 immunosuppressive axis [40,41]. NEU1, a lysosomal sialidase, cleaves terminal sialic acids from glycoproteins and glycolipids [42]. Deficiencies or dysfunction in NEU1 can lead to the accumulation of sialylated substrates and contribute to tumor-associated cellular alterations [43]. Transcriptomic data demonstrated significantly increased expression of ST6GALNAC4, B4GALT5, B4GALNT1, and NEU1 in HCC samples relative to controls, and RT-qPCR analysis further confirmed elevated mRNA expression in clinical specimens. A prognostic risk model incorporating these four SRGs effectively stratified patients into high-risk and low-risk categories, offering clinically relevant survival prediction. Collectively, these findings support the hypothesis that these four SRGs may promote a pro-tumorigenic microenvironment via glycosylation-dependent mechanisms. Their involvement in tumor progression and immune modulation underscores their potential utility as targets for therapeutic intervention. Elucidating the interactions between these genes and components of the tumor microenvironment may enhance understanding of the pathobiology of HCC and guide the development of personalized treatment strategies. Crucially, the development of solid tumors, including HCC, depends on the establishment of an immune microenvironment that suppresses immune responses, marked by the accumulation of tumor-promoting immune cells, such as regulatory T cells (Tregs) and neutrophils [33]. To further investigate whether tumor genomic features are associated with the TIME and immunotherapeutic response, immune infiltration was evaluated. The immune infiltration analysis revealed significant disparities between high- and low-risk subgroups of HCC, indicating that an immunosuppressive microenvironment may contribute to tumor progression and therapeutic resistance. Notably, patients in the high-risk group exhibited increased TIDE scores, indicating a greater potential for immune escape and reduced responsiveness to immunotherapy. Furthermore, differences in the infiltration levels of various immune cells and immune checkpoints were observed between the two risk groups, many of which were significantly correlated with the identified sialylation-related genes, particularly ST6GALNAC4. In addition, substantial differences in the IC50 values of 101 drugs were identified between the high-risk and low-risk groups, offering a possible explanation for the poorer prognosis observed in the high-risk group and emphasizing the need for individualized therapeutic approaches. To explore potential mechanisms and signaling pathways associated with the four identified sialylation-related genes, co-expression networks and functional associations were analyzed using the GeneMANIA database. These results highlighted critical involvement in lipopolysaccharide (LPS) metabolic processes and sphingolipid metabolism, both of which are closely linked to HCC pathogenesis and modulation of the immune microenvironment. LPS has been shown to connect gut microbiota dysbiosis to hepatocarcinogenesis via the gut-liver axis [44]. Increased intestinal permeability in individuals with HCC facilitates the translocation of LPS into systemic circulation, activating Toll-like receptor 4 (TLR4) signaling pathways. This activation induces the release of pro-inflammatory cytokines, such as TNF-α, IL-6, and IL-1β, which in turn promote tumor cell proliferation and invasion, as well as the activation of hepatic stellate cells, exacerbating fibrosis and creating a microenvironment conducive to tumor progression. Simultaneously, sphingolipid metabolism acts as a key regulator of cellular growth, differentiation, proliferation, and apoptosis. Dysregulation of this pathway has been implicated in the development of multiple diseases, including HCC [45]. In contrast to earlier investigations, the present study identified ST6GALNAC4, B4GALT5, B4GALNT1, and NEU1 as novel prognostic biomarkers in HCC. These genes may influence sialylation-mediated interactions with immune cells or contribute to extracellular matrix remodeling, offering promising therapeutic targets. However, the precise molecular mechanisms remain to be elucidated and require further in vivo validation to confirm their roles in HCC progression. Future studies should aim to replicate these findings in larger, independent cohorts and examine the therapeutic potential of targeting these genes. Moreover, investigation of downstream signaling pathways and immune responses affected by these genes could reveal additional biomarkers and treatment strategies. Integrating multi-omics approaches with functional assays will be essential for a comprehensive understanding of the biological significance of sialylation-related genes in HCC. Conclusion In conclusion, this study provides novel insights into the prognostic and therapeutic relevance of sialylation-related genes in HCC, thereby offering a foundation for advancing early detection, individualized prognosis prediction, and targeted therapeutic strategies. Supplementary Material Supplemental Material supplementary table S1.docx supplementary table S3.docx supplementary table S6.docx supplementary table S5.docx supplementary figure S1_revised.tif supplementary table S4.docx supplementary figure S3_revised.tif supplementary figure S2_revised.tif supplementary table S2.docx supplementary table S7.docx Acknowledgments Guan-Qing Wang: formal analysis, methodology, software. Kang Du: data curation, writing–original draft, writing–review and editing. Yu-Peng Wang: conceptualization, funding acquisition, project administration. All authors read and approved the final draft. Funding Statement This study was funded by Research Fund Project of Hebei Provincial Health and Family Planning Commission (Grant No. 20251046). The funding body had no role in the design of the study and collection, analysis, and interpretation of data, and in writing the manuscript. Ethical approval This study was conducted with approval from the Ethics Committee of Bethune International Peace Hospital (Approval No. 2024-KY-140). This study was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants. Consent for publication Not applicable. Disclosure statement No potential conflict of interest was reported by the author(s). Data availability statement The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request. References Associated Data Supplementary Materials Supplemental Material supplementary table S1.docx supplementary table S3.docx supplementary table S6.docx supplementary table S5.docx supplementary figure S1_revised.tif supplementary table S4.docx supplementary figure S3_revised.tif supplementary figure S2_revised.tif supplementary table S2.docx supplementary table S7.docx Data Availability Statement The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.