A clinically derived lipid-endothelial signature links serum multi-omics to immune exclusion and clinical stratification in hepatocellular carcinoma

Abstract

Background

Hepatocellular carcinoma (HCC) is driven by extensive metabolic reprogramming, vascular remodeling, and immune microenvironmental dysfunction. Although numerous stratification signatures have been proposed, few are grounded in clinically derived serum multi-omics and biologically linked to endothelial remodeling, endothelial regulation, and immune exclusion.

Objectives

This study aimed to identify a serum-derived lipid-endothelial program associated with immune exclusion and clinical stratification in HCC.

Design

A translational multi-omics study integrating clinically collected serum samples, public transcriptomic cohorts, single-cell RNA sequencing, and experimental validation.

Methods

Proteomic and metabolomic sequencing was performed on serum samples from patients with HCC and normal controls. Dysregulated pathways were integrated with transcriptomic data from TCGA-LIHC and ICGC-LIRI-JP cohorts to identify genes jointly associated with lipid metabolism and leukocyte transendothelial migration. A risk score was calculated using the expression of PON1, TXNRD1, CLDN4, CLDN6, CYP2C9, and CTSA. Higher expression of TXNRD1, CLDN4, CLDN6, and CTSA contributed to a higher risk score, whereas PON1 and CYP2C9 contributed protective coefficients. Immune contexture, tumor mutation burden, and exploratory therapeutic sensitivity patterns were further evaluated using transcriptome-based drug sensitivity prediction, followed by single-cell RNA sequencing and experimental expression validation.

Results

Serum multi-omics analysis revealed prominent dysregulation of lipid metabolic pathways and leukocyte transendothelial migration-related processes in HCC. Integrative analysis identified a six-gene lipid-endothelial signature (PON1, TXNRD1, CLDN4, CLDN6, CYP2C9, and CTSA) that stratified patients into high- and low-risk groups. In the TCGA-LIHC cohort, high-risk patients had significantly poorer overall survival than low-risk patients (log-rank P < 0.0001), and this survival-stratifying association was externally supported in the ICGC-LIRI-JP cohort (log-rank P = 0.016). The high-risk phenotype was associated with immune-excluded features, distinct somatic mutation patterns, and altered predicted sensitivity to several selected anticancer agents. Single-cell analysis and experimental assays further supported the association between the six-gene program, malignant epithelial states, endothelial-related remodeling, and immune microenvironmental heterogeneity.

Conclusion

This study defines a clinically derived lipid-endothelial program associated with immune exclusion, adverse prognosis, and potential differences in therapeutic vulnerability in HCC. The proposed signature provides a biologically informed framework for prognostic assessment and may support future evaluation of targeted interventions in HCC.

Plain language summary

Hepatocellular carcinoma (HCC), the most common type of liver cancer, is not driven only by tumor growth itself. Changes in metabolism, blood vessel function, and the immune environment also play important roles in how the disease develops and progresses. Many studies have proposed gene signatures to predict outcomes in HCC, but most are based mainly on public databases and are not directly linked to clinical serum samples. In this study, we analyzed proteins and metabolites in serum samples collected from patients with HCC and healthy controls, and combined these data with public transcriptomic datasets, single-cell sequencing, and laboratory validation. We identified a six-gene lipid-endothelial signature that separated patients into high-risk and low-risk groups with different survival outcomes. Patients in the high-risk group showed features of immune exclusion, higher tumor mutation burden, and lower predicted sensitivity to several anticancer drugs. These findings suggest that a serum-derived lipid-endothelial program may help explain why some HCC tumors have a more suppressive immune environment and worse outcomes. This approach may support future patient stratification and help guide further studies on treatment vulnerability in HCC.

Introduction

Hepatocellular carcinoma (HCC) is the sixth most commonly diagnosed cancer worldwide and the third leading cause of cancer-related death.^1,2 Because early-stage HCC is often clinically silent and disease progression can be rapid, many patients are diagnosed at an advanced stage and are no longer eligible for curative treatment. ³ Current management includes surgical resection, liver transplantation, locoregional therapies such as radiofrequency ablation and transarterial chemoembolization, and systemic treatments including targeted agents and immune checkpoint blockade. ⁴ In addition, vascular remodeling and aberrant angiogenesis within the tumor microenvironment (TME) further contribute to tumor progression and therapeutic resistance. Among these features, dysregulated lipid metabolism and impaired antitumor immunity appear to be closely linked and jointly contribute to disease progression and therapeutic resistance. ⁵

Lipid metabolic reprogramming and leukocyte transendothelial migration have each been recognized as important components of HCC progression.^6–9 Tumor cells rewire lipid uptake, synthesis, and storage to meet the demands of rapid proliferation and membrane biosynthesis, while lipid-derived mediators such as prostaglandins and lysophosphatidic acid further shape an immunosuppressive microenvironment.^10,11 Excessive lipid accumulation can also induce lipotoxic stress, impair the function of effector immune cells such as CD8⁺ T cells, and favor the expansion of immunosuppressive populations including regulatory T cells and M2 macrophages. ¹² In parallel, dysfunction of tumor-associated endothelial cells in the hepatic vasculature, a key component of tumor angiogenesis, restricts immune cell trafficking into tumor tissue, thereby weakening immune surveillance. ¹³ Emerging evidence suggests that these processes may not be independent. Endothelial remodeling driven by metabolic alterations may also influence vascular structure and permeability, thereby affecting angiogenesis-related processes in the TME. Altered lipid metabolism can disrupt endothelial homeostasis and affect leukocyte transendothelial migration by modulating adhesion molecules such as ICAM-1 and VCAM-1, ultimately influencing immune cell adhesion and transmigration.^14–16 Our previous sequencing analyses likewise revealed a close association between lipid metabolism-related genes (LMRGs) and leukocyte transendothelial migration-related genes (LTMRGs), supporting the existence of a coordinated regulatory network. Nevertheless, how this metabolic-endothelial-immune axis operates in HCC remains insufficiently defined.

Although numerous stratification signatures for HCC have been reported, many are derived primarily from public transcriptomic datasets and offer limited biological anchoring to clinically collected biospecimens.^17–22 As a result, they do not adequately capture the biological interplay between metabolic reprogramming, endothelial remodeling, vascular dynamics, and immune regulation, which may limit their clinical utility for prognosis assessment and treatment stratification. Integrative analytical strategies provide an opportunity to address this limitation. In the present study, we combined proteomic and metabolomic data from clinical serum samples with transcriptomic data from public HCC cohorts (TCGA-LIHC and ICGC). Differentially expressed molecules identified from serum multi-omics data were integrated with transcriptomic profiles, and a stable gene set was subsequently screened via univariate and LASSO-Cox regression analyses.^23,24 Based on this framework, we established a stratification model and nomogram to evaluate its clinical relevance.

By focusing on the intersection of LMRGs and LTMRGs, this study sought to determine whether clinically derived serum multi-omics could define a biologically grounded lipid-endothelial program associated with immune exclusion and clinically relevant stratification in HCC.

Methods

Clinical serum sample collection

The serum discovery cohort included 10 treatment-naïve patients with primary HCC and 10 normal controls (NC), all of whom were recruited at the Second Hospital of Lanzhou University. The participant selection and sample allocation workflow is shown in Figure 1A. Briefly, eligible patients with primary HCC and NC participants were selected according to predefined inclusion and exclusion criteria. The final cohort comprised 20 participants, with five HCC and five NC serum samples allocated to proteomic sequencing and another five HCC and five NC serum samples allocated to metabolomic sequencing. Baseline demographic characteristics and the clinicopathological and laboratory characteristics of the HCC patients are summarized in Table 1.

Figure 1.

Serum proteomic and metabolomic profiling identifies dysregulated endothelial-related and lipid metabolic pathways in HCC. (A) Participant selection and sample allocation workflow for the serum discovery cohort. (B) Volcano plot of DEPs in serum proteomic profiling. (C) GO enrichment analysis of DEPs. (D) KEGG pathway enrichment analysis of DEPs. (E) Volcano plot of DEMs in serum metabolomic profiling. (F) Representative differentially abundant metabolites identified by serum metabolomic profiling. (G) Hierarchical clustering heatmap of lipid-related metabolites. (H) Pathway enrichment analysis of serum metabolites. (I) KEGG pathway mapping of DEM-associated metabolic pathways.

Table 1.

Serum discovery cohort characteristics.

Characteristics	HCC	Normal
n	10	10
T stage, n (%)
T4	4 (40%)	-
T2+T3	2 (20%)	-
T1	4 (40%)	-
N stage, n (%)
N1	3 (30%)	-
N0	7 (70%)	-
M stage, n (%)
M0	9 (90%)	-
M1	1 (10%)	-
cTNM, n (%)
Stage III	6 (60%)	-
Stage IV	1 (10%)	-
Stage I	3 (30%)	-
CNLC, n (%)
Stage III	4 (40%)	-
Stage IV	2 (20%)	-
Stage II	1 (10%)	-
Stage I	3 (30%)	-
Gender, n (%)
Male	5 (50%)	5 (50%)
Female	5 (50%)	5 (50%)
Age, median (IQR)	53 (49.5, 54.75)	59 (52.5, 65.75)
AFP, ng/mL, mean ± SD	945.29 ± 951.21	-
PIVKA-II, mAU/mL, mean ± SD	4798.1 ± 9170.2	-
AST, U/L, mean ± SD	102.8 ± 136.61	-
ALT, U/L, mean ± SD	29.3 ± 16.6	-
γ-GT, U/L, mean ± SD	163.5 ± 193.71	-
ALP, U/L, mean ± SD	164.2 ± 123.22	-
HBsAg, n (%)		-
Positive	8 (80%)	-
Negative	2 (20%)	-

Blood samples were collected into serum-separator tubes and centrifuged according to standard procedures. Serum aliquots were prepared and stored for subsequent proteomic or metabolomic analyses. Exclusion criteria included prior chemotherapy, transarterial chemoembolization, surgical resection, or the presence of other primary malignancies. This study was approved by the Ethics Committee of the Second Hospital of Lanzhou University (Approval No. 2024A-347), and written informed consent was obtained from all participants. These clinically collected serum samples served as the discovery layer for downstream multi-omics integration. To reduce potential technical bias, HCC and NC serum samples were processed using the same collection, centrifugation, aliquoting, storage, and extraction procedures. Samples from different groups were allocated to the proteomic and metabolomic discovery layers before data analysis. Serum discovery analyses and raw data preprocessing were performed without using clinical outcome information.

Table 1 Data are presented as n (%), mean ± standard deviation (SD), or median (interquartile range, IQR), as appropriate. Clinicopathological variables, including T stage, N stage, M stage, cTNM stage, CNLC stage, AFP, PIVKA-II, AST, ALT, γ-GT, ALP, and HBsAg status, were summarized for patients with HCC where applicable. Variables not applicable to NC are indicated by “-”. AFP, alpha-fetoprotein; PIVKA-II, protein induced by vitamin K absence or antagonist-II; AST, aspartate aminotransferase; ALT, alanine aminotransferase; γ-GT, gamma-glutamyl transferase; ALP, alkaline phosphatase; HBsAg, hepatitis B surface antigen; CNLC, China Liver Cancer staging.

Serum proteomics sample analysis

Serum samples were subjected to acetone precipitation, reductive alkylation, and overnight tryptic digestion. After desalting, 200 ng of peptide mixtures were injected into a nanoElute 2 LC system coupled to a timsTOF Pro 2 mass spectrometer via a CaptiveSpray ion source. Chromatographic separation was performed on a PePSep C18 column. Data-independent acquisition (DIA) was performed in PASEF mode, covering a mass range of 100-1,700 m/z and an ion mobility range of 0.6-1.6 Vs cm^-1 (1/K₀). Raw data were analyzed using Spectronaut v18.2 (Biognosys) against the UniProt Homo sapiens reference proteome. Fixed modification: carbamidomethylation of cysteine; variable modifications: methionine oxidation and N-terminal acetylation. All other parameters were set to default.

Serum metabolites sample analysis

Serum samples were transferred into 1.5 mL polypropylene tubes. Samples were vortex-mixed for 30 seconds and sonicated for 10 minutes at 4°C. The mixtures were then centrifuged at high speed for 15 minutes, and the resulting supernatants were analyzed by LC-MS/MS. Chromatographic separation was performed at 40°C. The mobile phase A consisted of an aqueous solution of ammonium acetate and ammonium hydroxide, and mobile phase B was acetonitrile. The flow rate was 0.3 mL/min, with a 12-minute gradient elution. Samples (2 µL) were stored at 4°C in the autosampler prior to injection. High-resolution MS/MS spectra were acquired using an Orbitrap Exploris 120 mass spectrometer. Full-scan spectra were recorded at a resolution of 60,000 (at m/z 200), and data-dependent MS/MS was performed for the top 3 ions per cycle. Raw data were processed using an in-house R pipeline. Peak picking, retention time alignment, and peak integration were performed with XCMS v3.16, followed by feature grouping using CAMERA.

Integrated proteomics-metabolomics profiling and functional enrichment analysis

Serum proteomic and metabolomic datasets were independently processed and normalized according to platform-specific workflows. Differentially expressed proteins and metabolites were identified, followed by functional enrichment analysis using Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) databases. Because multi-omics datasets were generated from different analytical platforms, integration was performed at the pathway and gene-set levels rather than by direct combination of raw abundance values.

Pearson correlation analysis was performed to evaluate relationships between significantly altered metabolites and proteins involved in HCC-associated pathways. Positive and negative correlation coefficients represented concordant and inverse relationships, respectively. Enrichment significance was assessed using false discovery rate (FDR) correction, with FDR <0.25 and |NES| >1 considered significant for gene-set enrichment analyses. Gene sets containing fewer than 10 or more than 500 genes were excluded to minimize statistical bias.

Construction and external validation of the lipid-endothelial risk signature

To evaluate the predictive value of the lipid-endothelial risk signature, univariable Cox proportional hazards models were applied to the TCGA-LIHC dataset using the survival and survminer packages in R. Genes with a nominal P < 0.05 in univariable Cox analysis were retained for subsequent LASSO-Cox regression. The risk score for each patient was calculated as follows: Risk score = (PON1 × −0.095) + (TXNRD1 × 0.265) + (CLDN4 × 0.092) + (CLDN6 × 0.093) + (CYP2C9 × −0.024) + (CTSA × 0.157). Patients were divided into high-risk and low-risk groups according to the median risk score. Survival differences were assessed by Kaplan-Meier survival analysis and log-rank tests. Risk stratification was evaluated using principal component analysis (PCA) and time-dependent receiver operating characteristic (ROC) curves (via the time ROC package). For external validation, risk coefficients derived from the TCGA-LIHC cohort were applied to the ICGC-LIRI-JP cohort.

Construction of a clinical-molecular nomogram and multivariable prognostic analysis

To determine whether the risk signature maintained prognostic value after adjusting for conventional clinical covariates, we extracted state, and age from the TCGA-LIHC dataset. These variables, together with the continuous risk score, were included in multivariable Cox regression models. A clinical-molecular nomogram was constructed using the rms, survival, and replot packages as an exploratory framework for individualized risk estimation rather than as a clinically validated decision tool.

Gene mutation analysis

To characterize the genomic landscape of the signature, somatic mutation data from TCGA-LIHC were analyzed using the maftools R package.

Tumor immune microenvironment analysis

To characterize the immune landscape of HCC, we analyzed leukocyte composition, the expression of immune inhibitory receptors/ligands, and tumor-intrinsic immunogenicity metrics. The abundance of immune cell subsets was quantified by single-sample gene set enrichment analysis (ssGSEA) and visualized using CloudBioSense. ssGSEA evaluated immune-related signatures, while immune cell abundance was estimated separately. Differences in immune cell infiltration between risk groups were assessed using the Immune Oncology Biological Research R package. The expression of checkpoint molecules was compared to predict sensitivity to immunomodulatory agents. Immunophenoscores from The Cancer Immunome Atlas were used to predict response to anti-CTLA-4 and anti-PD-1 therapies.

Drug sensitivity analysis

To explore potential therapeutic implications of the lipid-endothelial risk signature, we performed an exploratory drug sensitivity analysis using the Genomics of Drug Sensitivity in Cancer (GDSC) database. The half-maximal inhibitory concentration (IC₅₀) values of selected anticancer agents were computationally estimated using the pRRophetic algorithm. Differences in predicted IC₅₀ values were compared between the high-risk and low-risk groups. These analyses were intended to generate hypotheses regarding potential therapeutic vulnerability and were not interpreted as experimentally supported drug responses.

Analysis of single-cell data

The publicly available scRNA-seq dataset GSE149614 was obtained from the Gene Expression Omnibus (GEO) database. Single-cell data processing and integration were performed using the Seurat V5 package. Cells were filtered according to predefined quality control criteria based on detected genes, mitochondrial gene proportion, hemoglobin gene proportion, and unique molecular identifier (UMI) counts. Batch correction was performed using Harmony, followed by dimensionality reduction and clustering using PCA and uniform manifold approximation and projection (UMAP). Cell populations were annotated according to established marker genes. Malignant epithelial cells were identified using inferCNV analysis with monocytes as the reference population.

The established six-gene risk signature was projected onto single-cell populations to characterize its distribution across cell types. T/NK cell subsets were further reclustered and annotated using subset-specific markers. Cell composition differences between risk groups were evaluated. Monocle2 was applied for pseudotime trajectory analysis of epithelial cell subsets using selected ordering genes to investigate cellular state transitions.

Validation using the Human Protein Atlas (HPA) and experimental assays

Protein-level expression data for the five prognostic genes in HCC were retrieved from the HPA database (https://www.proteinatlas.org/) and compared with transcriptomic results for downstream validation.

Quantitative Real-Time Polymerase Chain Reaction (qRT-PCR)

An independent validation set of 12 serum samples from patients with HCC and 12 serum samples from NC participants was collected to measure the expression levels of PON1, TXNRD1, CLDN4, CLDN6, CYP2C9, and CTSA. Total RNA was extracted from serum using a commercial RNA extraction kit and reverse-transcribed into cDNA according to standard protocols. Quantitative real-time PCR was performed on a CFX96 Real-Time PCR Detection System using SYBR Green dye for quantification.

ImmunoFluorescence (IF)

The expression of PON1, TXNRD1, CLDN4, CLDN6, CYP2C9, and CTSA proteins was detected in normal human hepatocytes and HCC cells. Cells were seeded in confocal dishes and cultured in 10% FBS DMEM medium for 24 hours. After confirming satisfactory cell adhesion under a microscope, cells were washed three times with PBS to remove residual medium, then fixed with 4% paraformaldehyde. Cell membranes were permeabilized with 0.1% Triton X-100. Cells were incubated with primary antibodies at 4°C overnight, followed by washing with PBS. Fluorescent secondary antibodies were added and incubated for 1 hour. Cells were washed with PBS, counterstained with 4′,6-diamidino-2-phenylindole (DAPI), and observed under a laser scanning confocal microscope for image acquisition.

Western Blotting (WB)

Cells were harvested and lysed in RIPA buffer supplemented with phenylmethylsulfonyl fluoride (PMSF) for 30 minutes at 4°C with gentle shaking. Lysates were centrifuged, and supernatants containing total protein were collected. Protein concentrations were measured using a BCA kit. Proteins were separated by SDS-PAGE and transferred to polyvinylidene fluoride (PVDF) membranes. Membranes were blocked with 5% non-fat milk in TBST for 1 hour at room temperature. Membranes were incubated with primary antibodies against PON1, TXNRD1, CLDN4, CLDN6, CYP2C9, and CTSA overnight at 4°C. After washing with TBST, membranes were incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies (1:10,000 dilution) for 2 hours at room temperature. Protein bands were visualized using an ECL kit with a MiniChemi 610 Plus Imaging System. GAPDH was used as an internal loading control.

ImmunoHistoChemistry (IHC)

IHC staining was performed to detect protein expression in 10 pairs of HCC tissues and matched normal liver tissues. Tissues were fixed with 4% paraformaldehyde, dehydrated, embedded in paraffin, and sectioned. Sections were rehydrated and subjected to antigen retrieval, then blocked with serum for 10 minutes. Sections were incubated with primary antibodies against the six corresponding proteins at a dilution of 1:200 overnight at 4°C. After secondary antibody incubation, sections were stained with 3,3′-diaminobenzidine (DAB) and counterstained with hematoxylin for observation. Two pathologists from the Second Hospital of Lanzhou University independently scored IHC slides using the H-score (product of staining intensity and positive cell percentage).

Quantification and statistical analysis

Statistical analyses were performed using GraphPad Prism and R software (version 4.3.1). Quantitative variables were presented as mean ± SD, mean ± SEM, or median with IQR, as appropriate. Normality was assessed before statistical testing. Comparisons between two groups were performed using Student’s t-test or Mann–Whitney U test, while comparisons among multiple groups were performed using one-way ANOVA or Kruskal–Wallis test as appropriate. Categorical variables were analyzed using Pearson’s chi-square test or Fisher’s exact test. Two-sided P values <0.05 were considered statistically significant unless otherwise specified.

For high-dimensional omics analyses and enrichment analyses, multiple testing correction was performed using the Benjamini–Hochberg method, and adjusted P values or false discovery rates (FDRs) were reported. Survival analyses were performed using Kaplan–Meier curves, log-rank tests, and Cox proportional hazards regression models. Model discrimination was evaluated using time-dependent ROC curves. The reporting of this study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement ²⁵ and the Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK). ²⁶ Completed checklists are provided as Supplementary Files 1 and 2.

Results

Independent pathway enrichment analyses of serum proteomics and metabolomics identify major altered pathways in HCC

We first established the participant selection and sample allocation workflow for the serum discovery cohort (Figure 1A). Proteomic and metabolomic sequencing was then performed on serum samples from 5 NC and 5 HCC patients in each omics layer, and separate pathway enrichment analyses were conducted to identify dysregulated biological processes in HCC. Proteomic profiling identified 271 differentially expressed proteins (DEPs) between the HCC and NC groups, including 100 upregulated and 171 downregulated proteins, as shown in the volcano plot (Figure 1B). GO analysis indicated that these DEPs were mainly associated with cell motility and migration in the Biological Process category, enriched in the extracellular region in the cellular component category, and involved in cell adhesion, heparin binding, signaling receptor binding, and growth factor activity in the molecular function category (Figure 1C). KEGG pathway analysis showed that the DEPs were primarily enriched in leukocyte transendothelial migration, complement and regulation of the actin cytoskeleton (Figure 1D). Among these, leukocyte transendothelial migration and actin cytoskeleton-related pathways, which are closely linked to endothelial function and vascular dynamics, showed particularly strong enrichment.

Metabolomic profiling identified 107 differentially expressed metabolites (DEMs) between the HCC and NC groups, including 67 upregulated and 40 downregulated metabolites, as illustrated in the volcano plot (Figure 1E). Upregulated DEMs were mainly characterized by lipid-associated metabolites, particularly glycerophospholipids and fatty acid derivatives, whereas downregulated metabolites reflected alterations in amino acid-derived metabolites and lipid homeostasis. (Figure 1F). Hierarchical clustering analysis further demonstrated distinct serum metabolite patterns between HCC and NC. Notably, several clustered metabolites were lipid-associated molecules, including 1,2-distearoyl-sn-glycero-3-phospho-L-serine and phosphatidylinositol (PI; 16:1(9Z)/18:1(9Z)), suggesting altered glycerophospholipid metabolism and lipid remodeling in HCC serum (Figure 1G). Pathway enrichment analysis revealed that altered metabolites were predominantly associated with lipid metabolic processes, particularly fatty acid biosynthesis. Additional affected pathways included the tricarboxylic acid cycle, β-alanine metabolism, glutathione metabolism, and pyrimidine metabolism,,indicating broad metabolic remodeling in HCC serum (Figure 1H). KEGG enrichment analysis revealed that DEMs were mainly involved in lipid-associated pathways, including fatty acid biosynthesis, unsaturated fatty acid biosynthesis, arachidonic acid metabolism, and linoleic acid metabolism, together with alterations in ABC transporter and central metabolic pathways. (Figure 1I).

These exploratory serum proteomic and metabolomic analyses suggested prominent enrichment of leukocyte transendothelial migration-related proteins and lipid metabolic pathways in HCC, providing a clinically derived discovery basis for subsequent integrative stratification analyses.

Identification of lipid metabolism- and leukocyte transendothelial migration-associated genes in HCC

Pearson correlation analysis was performed to screen 313 candidate genes associated with both lipid metabolism and leukocyte transendothelial migration in HCC. The top 20 genes are shown in the correlation heatmap (Figure 2A). A volcano plot identified differentially expressed genes between HCC and NC, revealing 77 significantly dysregulated genes (41 upregulated and 36 downregulated genes) (Figure 2B). These genes were retained as the core candidate set for subsequent analyses.

Figure 2.

Identification and functional annotation of lipid metabolism- and leukocyte transendothelial migration-associated candidate genes in HCC. (A) Heatmap showing the top 20 genes associated with both lipid metabolism and leukocyte transendothelial migration based on Pearson correlation analysis. (B) Volcano plot showing differentially expressed metabolite-associated genes between HCC and normal liver tissues. (C) GO enrichment analysis of the 77 candidate genes. (D) KEGG pathway enrichment analysis of the 77 candidate genes. (E) PPI network of proteins encoded by the 77 candidate genes. (F) Functional module analysis showing the relationship among fatty acid metabolism, leukocyte transendothelial migration and oxidoreductase activity & arachidonic acid metabolism.

GO enrichment analysis showed significant over-representation of lipid-related biological processes, including fatty acid metabolic process, steroid metabolic process, long-chain fatty acid metabolic process, and fatty acid derivative metabolic process (Figure 2C). KEGG pathway analysis linked the 77-gene set to several pathways relevant to lipid metabolism and endothelial/immune-associated signaling. Notably,PPAR signaling, fatty acid metabolism, lipid and atherosclerosis, VEGF signaling, and leukocyte transendothelial migration were among the significantly enriched pathways, providing a functional basis for subsequent identification of lipid–endothelial regulatory signatures (Figure 2D).

A protein-protein interaction (PPI) network was constructed for the proteins encoded by the 77 genes, visualizing both positively and negatively correlated interactions (Figure 2E). A circos plot further grouped these genes into three major functional modules: fatty acid metabolism, leukocyte transendothelial migration and oxidoreductase activity-arachidonic acid metabolism (Figure 2F). Notably, genes in the fatty acid metabolism module displayed extensive connections with the other modules, suggesting close relationships between lipid metabolic processes and pathways involved in immune cell trafficking, endothelial function, and redox regulation in HCC.

Together, these analyses highlighted a set of genes jointly associated with lipid metabolism and leukocyte transendothelial migration in HCC, providing a biologically informed basis for subsequent development of a clinically relevant stratification model.

Development of a six-gene lipid-endothelial stratification model in the TCGA-LIHC cohort

To identify survival-associated genes potentially related to lipid metabolism and leukocyte transendothelial migration, univariate Cox regression analysis was performed in the TCGA-LIHC cohort, yielding 21 prognostically relevant genes (P < 0.05; Supplementary File 3). LASSO-Cox regression was then applied to further refine this gene set, with the optimal λ selected according to the minimum partial likelihood deviance (Figure 3A and B).

Figure 3.

Development of the six-gene lipid–endothelial stratification model in HCC. (A and B) LASSO-Cox regression analysis of 21 univariately significant genes. (C) Heatmap showing the expression patterns of the six signature genes and clinicopathological annotations in the TCGA-LIHC cohort. (D) PCA showing separation between high- and low-risk patients. (E) Risk score distribution curve of the prognostic signature. (F) Correlation of higher risk with elevated mortality and shorter overall survival. (G) Kaplan-Meier survival curves of high- and low-risk groups (log-rank test, P < 0.0001). (H) ROC curve evaluating the predictive performance of the stratification signature.

In the TCGA-LIHC cohort, hierarchical clustering showed distinct expression patterns of the six signature genes between the high- and low-risk groups, together with distribution of clinical characteristics including survival status and age, across different risk groups (Figure 3C). The high-risk group was mainly characterized by higher expression of the positive-coefficient genes TXNRD1, CLDN4, CLDN6, and CTSA, together with lower expression of the protective genes PON1 and CYP2C9. Conversely, the low-risk group tended to show higher expression of PON1 and CYP2C9. These expression patterns explain how the six-gene model contributed to risk stratification. Patients were stratified into high- and low-risk groups using the median risk score as the cutoff. PCA demonstrated separation between the two groups, suggesting distinct transcriptomic patterns associated with risk classification (Figure 3D). Consistently, the distribution of risk scores was markedly different between the two groups, with higher scores observed in the high-risk group (Figure 3E). Scatter plot analysis further showed that increased risk scores were accompanied by more death events and shorter overall survival time (Figure. 3F).

To further clarify the clinicopathological context of the high- and low-risk groups, we compared age, sex, T stage, N stage, M stage, TNM component-derived stage category, survival status, and overall survival time between the two risk groups in the TCGA-LIHC cohort (Table 2). No significant differences were observed between the high- and low-risk groups in age, sex, T stage, or TNM component-derived stage category. In contrast, the high-risk group had a significantly higher proportion of death events and shorter overall survival time than the low-risk group. These findings indicate that the six-gene risk classification was associated with survival outcomes, while its distribution was not simply explained by age or tumor stage imbalance. Importantly, age and TNM-related variables were evaluated as clinical annotations and covariates rather than as components used to define the risk groups.

Table 2.

Association between risk status and clinicopathological characteristics in the TCGA-LIHC cohort.

Characteristic	Total (n=345)	Low-risk (n=173)	High-risk (n=172)	P value
Age, years, median (IQR)	61.0 (50.0, 69.0)	61.5 (50.0, 69.0)	60.0 (49.5, 69.0)	0.372
Age group, n (%)				0.706
≤60	168 (48.7%)	82 (47.4%)	86 (50.0%)	​
>60	175 (50.7%)	90 (52.0%)	85 (49.4%)	​
Missing	2 (0.6%)	1 (0.6%)	1 (0.6%)	​
Sex, n (%)				0.679
Male	226 (65.5%)	111 (64.2%)	115 (66.9%)	​
Female	119 (34.5%)	62 (35.8%)	57 (33.1%)	​
Race, n (%)				0.372
Asian	154 (44.6%)	74 (42.8%)	80 (46.5%)	​
White	167 (48.4%)	90 (52.0%)	77 (44.8%)	​
Black or African American	14 (4.1%)	6 (3.5%)	8 (4.7%)	​
Unknown	10 (2.9%)	3 (1.7%)	7 (4.1%)	​
T stage, n (%)				0.434
T1	182 (52.8%)	95 (54.9%)	87 (50.6%)	​
T2	81 (23.5%)	36 (20.8%)	45 (26.2%)	​
T3	70 (20.3%)	34 (19.7%)	36 (20.9%)	​
T4	12 (3.5%)	8 (4.6%)	4 (2.3%)	​
T stage group, n (%)				0.923
T1–T2	263 (76.2%)	131 (75.7%)	132 (76.7%)	​
T3–T4	82 (23.8%)	42 (24.3%)	40 (23.3%)	​
N stage, n (%)				1.000
N0	238 (69.0%)	119 (68.8%)	119 (69.2%)	​
N1	2 (0.6%)	1 (0.6%)	1 (0.6%)	​
Nx/unknown	105 (30.4%)	53 (30.6%)	52 (30.2%)	​
M stage, n (%)				1.000
M0	254 (73.6%)	124 (71.7%)	130 (75.6%)	​
M1	6 (1.7%)	3 (1.7%)	3 (1.7%)	​
Mx/unknown	85 (24.6%)	46 (26.6%)	39 (22.7%)	​
TNM component-derived category, n (%)				0.998
Early (T1–T2/N0/M0)	167 (48.4%)	85 (49.1%)	82 (47.7%)	​
Advanced (T3–T4/N1/M1)	86 (24.9%)	43 (24.9%)	43 (25.0%)	​
Unknown/indeterminate	92 (26.7%)	45 (26.0%)	47 (27.3%)	​
Survival status, n (%)				<0.001
Alive	216 (62.6%)	125 (72.3%)	91 (52.9%)	​
Dead	129 (37.4%)	48 (27.7%)	81 (47.1%)	​
Overall survival time, months, median (IQR)	20.8 (12.0, 39.8)	22.8 (14.9, 44.4)	17.1 (9.8, 32.4)	<0.001

Table 2 Data are presented as median (IQR) for continuous variables and n (%) for categorical variables. Continuous variables were compared using the Mann-Whitney U test. Categorical variables were compared using Pearson’s chi-square test or Fisher’s exact test, as appropriate. Patients were stratified into high- and low-risk groups according to the median risk score in the TCGA-LIHC cohort.

Kaplan-Meier survival analysis demonstrated that patients in the high-risk group had significantly worse overall survival than those in the low-risk group (log-rank test, P < 0.0001; Figure 3G). Time-dependent ROC analysis was used to assess the prognostic performance of the model, yielding AUC values of 0.784, 0.685, and 0.674 for 1-, 3-, and 5-year survival, respectively (Figure 3H). These results indicate that the six-gene model has good discriminative ability for short-term survival and moderate predictive performance over longer follow-up, supporting its potential utility for risk stratification in HCC.

External validation and clinical relevance of the six-gene stratification model

The TCGA-LIHC cohort was used to train the model, and the ICGC-LIRI-JP cohort was used for external validation. To externally validate the stratification model, patients in the ICGC-LIRI-JP cohort were classified into high- and low-risk groups using the TCGA-derived risk-score formula and cutoff. Kaplan-Meier survival analysis was also performed for each individual gene in the signature (Figure 4A). Among the six genes, CLDN6 and TXNRD1 were not significantly associated with overall survival when assessed individually in this cohort. These findings suggest that the prognostic performance of the model is more robust when considering the combined signature rather than individual genes alone.

Figure 4.

External validation and clinical relevance of the six-gene stratification model in HCC. (A) Kaplan-Meier survival analysis of the six prognostic genes. (B) PCA analysis of high- and low-risk groups. (C) Risk-score distribution plot. (D) Rank-ordered risk scores aligned with survival status. (E) Kaplan–Meier survival analysis of overall survival in the high- and low-risk groups (P = 0.016). (F) Time-dependent ROC curves showing the predictive performance of the six-gene signature at 1, 2, and 3 years in the ICGC-LIRI-JP cohort. (G) Nomogram integrating risk score and clinicopathologic features.

PCA showed a degree of separation between the high- and low-risk groups in the ICGC cohort (Figure 4B). The ICGC-LIRI-JP cohort included 72 patients in the high-risk group and 98 in the low-risk group. The same TCGA-derived risk-score formula was applied to the ICGC-LIRI-JP cohort for external validation, and cohort-specific performance was evaluated using Kaplan-Meier survival analysis and time-dependent ROC curves. Risk score distribution analysis showed higher scores in the high-risk group, consistent with the grouping strategy (Figure 4C). Scatter plot analysis further indicated that patients with higher risk scores tended to have more death events and shorter overall survival (Figure 4D). Kaplan-Meier survival analysis demonstrated significantly poorer overall survival in the high-risk group than in the low-risk group (P = 0.016; Figure 4E).

Time-dependent ROC analysis was performed to further assess model performance in the external cohort. The AUC values were 0.73, 0.67, and 0.66 for 1-, 2-, and 3-year overall survival, respectively (Figure 4F). To further evaluate the utility of the prognostic signature, a nomogram was developed by integrating the risk score with available clinicopathological variables (Figure 4G). Overall, external validation in the ICGC-LIRI-JP cohort supported an association between the six-gene signature and survival stratification in HCC.

Validation of the six signature genes and proteins in clinical samples

In the ICGC-LIRI-JP cohort, boxplot analysis showed that CLDN4, TXNRD1, CLDN6, and CTSA were upregulated in HCC tissues, whereas PON1 and CYP2C9 were downregulated in HCC tissues relative to normal liver tissues (Figure 5A). qRT-PCR analysis in clinical serum samples from patients with HCC and NC showed overall expression trends similar to those observed in the ICGC cohort (Figure 5B). Significant differences were detected for CLDN4, CTSA, TXNRD1, CYP2C9, and PON1 (P < 0.05), whereas CLDN6 did not reach statistical significance, consistent with the relatively weaker signal observed in the validation analysis.

Figure 5.

Validation of the six genes and their corresponding proteins in clinical samples. (A) Expression of the six signature genes in the ICGC-LIRI-JP cohort. (B) qRT-PCR analysis of the expression of the six genes in clinical serum samples. (C) Protein expression patterns of five available signature proteins in the HPA database. (D–I) IHC validation of the six signature proteins in human HCC tissues and normal liver tissues. Scale bars = 20 μm.

Protein-level evidence from HPA was generally consistent with the transcriptomic patterns observed in the ICGC cohort, although no HPA data were available for CLDN6 (Figure 5C). IHC staining of 10 paired HCC and normal liver tissues supported the protein-level expression patterns of most markers (Figure 5D-I). Specifically, CLDN4, TXNRD1, and CTSA showed higher expression in tumor tissues, whereas PON1 and CYP2C9 were downregulated compared with normal liver tissues. CLDN6 showed an upward trend in tumor tissues but did not reach statistical significance. Quantitative analysis of IHC-positive staining areas showed statistically significant differences for CLDN4, TXNRD1, CTSA, PON1, and CYP2C9, whereas CLDN6 remained non-significant.

Validation of the six signature proteins in HCC cell lines

Western blot analysis showed protein expression patterns broadly consistent with those observed in clinical samples (Figure 6A). Specifically, CLDN4, CTSA, TXNRD1, and CLDN6 showed higher protein expression in Huh7 and Hep3B cells than in THLE-2 cells, whereas those of PON1 and CYP2C9 were downregulated. Quantification of band intensities showed statistically significant differences in protein expression (Figure 6B; P < 0.05), further supporting the differential expression patterns observed in the clinical validation cohort.

Figure 6.

Validation of the six signature proteins in HCC cell lines. (A) Western blot analysis of PON1, TXNRD1, CLDN4, CLDN6, CYP2C9, and CTSA protein expression in THLE-2, Huh7, and Hep3B cells. (B) Quantitative analysis of Western blot band intensities. (C) Representative IF images showing the expression and localization of the six signature proteins in THLE-2, Huh7, and Hep3B cells. Scale bars = 50 μm. (D) Quantitative analysis of IF intensity. Data are presented as mean ± SEM.

IF staining was subsequently performed to assess the subcellular localization and relative expression levels of the six proteins (Figure 6C). Quantitative analysis of IF fluorescence intensity showed expression trends generally consistent with those observed by Western blotting (Figure 6D; P < 0.05). Specifically, CLDN4, CTSA, and TXNRD1 were upregulated, whereas PON1 and CYP2C9 were downregulated in HCC cell lines relative to THLE-2 cells. In contrast, although CLDN6 showed an upward trend, it did not reach statistical significance in IF analysis.

Somatic mutation patterns and exploratory therapeutic sensitivity associated with the six-gene signature

Somatic mutation profiling revealed the genomic alteration landscape between the high-risk and low-risk groups, with alterations detected in 91.67% and 89.94% of samples, respectively (Figure 7A-C). Commonly altered genes in both groups included TP53, CTNNB1, MUC16, and TTN, consistent with previously reported mutation patterns in HCC. These findings indicate that the two risk groups shared largely similar genomic alteration patterns, suggesting that the prognostic differences captured by the signature may not be primarily explained by differences in overall somatic mutation burden.

Figure 7.

Somatic mutation patterns and exploratory predicted drug sensitivity in high- and low-risk groups. (A) Somatic genomic alteration landscape of the high-risk group. (B) Somatic genomic alteration landscape of the low-risk group. (C) Comparison of mutation patterns between the two risk groups. (D) Comparison of predicted log-transformed IC50 values between high- and low-risk groups.

We next compared predicted IC₅₀ values for multiple anticancer agents between the high- and low-risk groups (Figure 7D). Exploratory drug sensitivity analysis revealed agent-specific predicted response patterns. The low-risk group showed significantly lower predicted IC₅₀ values for several agents, including Gemcitabine, Paclitaxel, Bortezomib, Camptothecin, Vinorelbine, Erlotinib, Doxorubicin, and Etoposide, suggesting greater predicted sensitivity to these agents. In contrast, the high-risk group showed lower predicted IC₅₀ values for selected agents, including Axitinib, Docetaxel, Sorafenib, and Dasatinib, indicating that predicted therapeutic sensitivity was agent-specific rather than uniformly favorable in either risk group.

scRNA-seq highlights immune microenvironment differences associated with the six-gene risk signature in HCC

Re-clustering of hepatocyte subsets identified 18 distinct hepatocyte subpopulations and their sample distribution (Figure 8A). Violin plot analysis showed that TXNRD1, CLDN4, CLDN6, and CTSA were highly expressed in tumor epithelial cells, whereas PON1 and CYP2C9 showed higher expression in normal epithelial cell subsets (Figure 8B). These cell type-specific expression patterns were largely consistent with the trends observed in clinical samples. Dimensionality reduction analysis of immune cells in the high- and low-risk groups showed separation between the two groups, with fewer B cells, plasma cells, and T-cell populations observed in the high-risk group (Figure 8C). The distribution of the six signature genes across immune cells, tumor cells, and normal cells is shown in Figure 8D, highlighting distinct cell-associated expression patterns.

Figure 8.

Single-cell analysis of hepatocyte subsets and immune microenvironment features associated with the six-gene signature in HCC. (A) Clustering of hepatocyte subsets showing distribution of 18 hepatocyte subpopulations and sample allocation. (B) Violin plot analysis of 6 prognostic genes in tumor vs. normal epithelial cell subsets. (C) Single-cell dimensionality reduction analysis of immune cells in high- and low-risk groups. (D) Expression cluster analysis of 6 prognostic genes in immune, tumor, and normal cells. (E) Pseudotime trajectory density plot of samples in high- and low-risk groups. (F) Re-clustering and annotation of T/NK cell subpopulations. (G) UMAP visualization showing the clustering and annotation of T/NK cell subpopulations based on single-cell RNA sequencing analysis. (H) Differences in the proportions of immune cells subpopulations between high- and low-risk groups. (I) Immune-related functional signatures in the high- and low-risk groups. (J) Infiltration levels of 16 immune cell types. (K) IPS scores of CTLA-4 and PD-1. (L) Expression profiles of immune checkpoint genes in different risk groups.

Pseudotime trajectory analysis was performed to characterize disease-related cellular states at the single-cell level (Figure 8E). High-risk samples were mainly distributed in later regions of the trajectory, whereas low-risk samples were more frequently located in earlier regions. Re-clustering of T/NK-cell subsets further characterized distinct T and NK cell populations based on subtype-specific marker expression (Figure 8F and G). Compared with the low-risk group, the high-risk group showed a higher proportion of regulatory T cells and lower proportions of CD8⁺ effector memory T cells and NK cells (Figure 8H), supporting differences in immune cell composition between the two groups.

ssGSEA identified significant differences in 13 immune-related signatures between the two risk groups. The high-risk group showed higher enrichment scores for APC co-stimulation, APC co-inhibition, chemokine receptor signaling, checkpoint pathways, HLA-related transcription, and MHC class I antigen presentation (Figure 8I). Quantification of 16 immune cell populations showed that activated dendritic cells, immature dendritic cells, macrophages, Th2 cells, and Treg cells were more abundant in the high-risk group (Figure 8J). IPS analysis suggested that differences between the high- and low-risk groups were mainly observed in PD-1-negative subgroups (Figure 8K). Comparison of immune checkpoint-related molecules between risk groups revealed that several immune regulatory genes, including HHLA2, CD80, CD44, and CTLA4, showed differential expression between the high- and low-risk groups (Figure 8L).

Collectively, these analyses support an association between the six-gene signature and immune microenvironment heterogeneity in HCC, particularly with respect to immune cell composition and checkpoint-related immune features. However, these findings should be interpreted as association-level evidence and do not by themselves establish direct causal regulation of endothelial remodeling, immune exclusion, or HCC progression by individual signature genes.

Discussion

A key strength of this study lies in the use of clinically collected serum proteomic and metabolomic data as the discovery layer, rather than relying solely on public transcriptomic datasets. In a field already populated by numerous HCC prognostic signatures, this design provides a more direct biological anchor for linking circulating metabolic alterations to endothelial remodeling and immune exclusion. Such an approach distinguishes the current study from many transcriptome-derived HCC signatures by anchoring stratification in clinically collected biospecimens and cross-scale validation. Among the six genes, PON1 and CYP2C9 are highly expressed in non-tumorous liver tissues but markedly downregulated in HCC. Previous studies are consistent with these observations: genetic variation in PON1 has been associated with liver disease progression in obesity-associated fatty liver disease, ²⁷ whereas CYP2C9 has been implicated in lipid-related drug metabolism and may influence the efficacy of statins such as fluvastatin and rosuvastatin. ²⁸ In the context of HCC, the reduced expression of these genes may reflect disrupted hepatic lipid homeostasis and a shift toward a tumor-promoting metabolic milieu. This clinically anchored framework may be relevant to understanding why biologically distinct HCC states show different patterns of therapeutic vulnerability, particularly in the setting of advanced disease where systemic treatment options remain limited.

By contrast, TXNRD1, CLDN4, CLDN6, and CTSA showed preferential expression in tumor epithelial cells, whereas PON1 and CYP2C9 were mainly expressed in normal epithelial populations. TXNRD1 has been reported to be upregulated in HCC and to promote tumor progression through redox regulation and Akt/mTOR signaling. ²⁹ CLDN4, a tight junction-associated protein, has been linked to malignant progression and aberrant epithelial barrier properties. ³⁰ CTSA is overexpressed in advanced HCC and has been implicated in tumor progression and immune-related regulation. ³¹ Although the role of CLDN6 in HCC remains complex, and its validation in our qRT-PCR and IHC cohorts did not reach statistical significance, existing evidence suggests that it participates in tumor-associated signaling and metabolic regulation. ³² Rather than representing an isolated multigene model, the six-gene panel may reflect a broader lipid-endothelial program associated with vascular remodeling and impaired immune access in HCC. However, these results demonstrate association rather than direct causality, and functional perturbation studies will be required to determine whether individual genes actively regulate endothelial remodeling or immune exclusion.

Somatic mutation analysis showed that the high-risk group exhibited a slightly higher proportion of samples with non-silent mutations compared with the low-risk group. The commonly altered genes included TP53, CTNNB1, and MUC16. These differences may reflect distinct molecular backgrounds between risk groups. TP53 mutations are associated with genomic instability and impaired DNA damage responses, whereas CTNNB1 mutations sustain WNT/β-catenin pathway activation and have also been linked to immune exclusion in liver cancer. ³³ The enrichment of these mutations in the high-risk group may therefore partly explain the adverse prognosis and distinct tumor microenvironmental features observed in these patients.^34,35

Both scRNA-seq and ssGSEA analyses indicated that high-risk tumors were characterized by a more immunosuppressive microenvironment, including increased infiltration of regulatory T cells and reduced proportions of CD8⁺effector memory T cells and NK cells. The high-risk TME is characterized by endothelial-associated remodeling and immune-excluded features, which have been implicated in altered therapeutic responses.^36–38 We also observed enrichment of immune regulatory pathways related to antigen presentation, chemokine signaling, and immune checkpoint activity. Together, these findings support the presence of an immune-excluded and immunoregulatory TME in high-risk HCC. Rather than demonstrating a direct causal mechanism, our data suggest that the six-gene signature may mark a tumor state in which lipid metabolic disturbance, endothelial-related remodeling, and impaired immune cell access coexist. The current evidence supports a biologically plausible association between this serum-derived lipid-endothelial program and immune-excluded features, but it does not prove that the individual genes directly drive endothelial remodeling, immune exclusion, or HCC progression. Functional perturbation studies, including gene knockdown or overexpression, endothelial-tumor co-culture systems, immune cell transmigration assays, and in vivo models, will be required to determine whether these genes have causal roles in shaping the HCC microenvironment.

Several limitations should be acknowledged. First, although the signature was associated with predicted therapeutic sensitivity and immune-excluded features, the current study did not include a dedicated cohort of patients treated with immune checkpoint inhibitors or targeted therapy, therefore, its value for prospective treatment selection remains to be established. Second, the serum discovery cohort was relatively small, with five HCC and five control samples analyzed in each serum omics layer. This limited sample size may have introduced statistical bias, reduced the power to detect weaker molecular signals, and limited the generalizability of the identified pathways. Therefore, the serum proteomic and metabolomic findings should be interpreted as an exploratory discovery layer rather than as independently validated biomarkers. Third, although the six-gene signature was supported by external transcriptomic cohorts, single-cell analysis, and expression-level experimental validation, the individual genes have not yet been functionally tested in dedicated in vitro or in vivo models, and their causal roles in regulating lipid metabolism, endothelial remodeling, immune exclusion and HCC progression remain to be clarified. Fourth, the drug sensitivity analysis in this study was based on computational estimation of IC50 values using the pRRophetic algorithm rather than direct pharmacological measurements. Thus, these findings should be interpreted as exploratory and hypothesis-generating rather than evidence of confirmed therapeutic response. Ultimately, to build upon these preliminary insights, larger multicenter studies and in vivo mechanistic models will be essential to fully establish the clinical utility and biological robustness of this serum-based stratification framework.

Conclusions

This study defines a clinically derived lipid-endothelial program associated with immune exclusion and adverse outcome in HCC. The proposed signature provides a biologically informed framework for prognostic assessment and clinical stratification, thereby supporting future evaluations of therapeutic vulnerability in hepatocellular carcinoma.

ORCID iD

Yumin Li https://orcid.org/0000-0002-9267-1412

Ethical considerations

The study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Lanzhou University Second Hospital (Approval No. 2024A-347).

Consent to participate

Written informed consent was obtained from all participants.

Data Availability Statement

The public datasets analyzed during the current study are available in The Cancer Genome Atlas (TCGA-LIHC), the International Cancer Genome Consortium (ICGC-LIRI-JP), the Gene Expression Omnibus (GEO; accession number GSE149614), and the Human Protein Atlas (HPA) repositories. The clinical serum, tissue, and experimental data generated during the current study are not publicly available due to patient privacy considerations but are available from the corresponding author on reasonable request.*