A novel glutamine metabolism-based classification system for characterizing the heterogeneity of hepatocellular carcinoma

Introduction

The prevalence of liver cancer is on the rise globally, with approximately 1 million new cases being diagnosed in 2021 (1). Hepatocellular carcinoma (HCC) accounts for 90% of all cases of liver cancer (2). Although patients with early-stage HCC may receive surgery or transplantation, those at advanced stages may miss the appropriate time for surgery (3). A substantial amount of research has been conducted to develop therapeutics for treating advanced HCC, including targeted tyrosine kinase inhibitors (4), immunotherapy (5), or combined chemotherapeutic regimens (6). However, the survival of patients with advanced-stage disease is poor, and the 5-year survival rate in China remains at less than 12.5% (7). Classifying the molecular phenotypes of HCC may help prolong survival and aid in developing curative treatment.

Malignant cells undergo considerable metabolism-related changes to compete with surrounding normal cells for limited nutrient resources and proliferate uncontrollably (8). Clarifying the nature of these alterations is critical to identifying effective biomarkers and therapeutic targets for HCC. Glutamine is the most abundant amino acid in the blood and is a key source of carbon in energy production and anabolic processes (9,10). Similar to their requirement for glucose, malignant cells depend on glutamine as a bioenergy source (11,12). Glutamine can provide a source of nitrogen and carbon for the synthesis of macromolecules that support tumor cell growth (13,14). Although considerable progress has been made in elucidating the metabolism-related alterations of HCC and identifying potential druggable vulnerabilities (15), a comprehensive characterization of glutamine metabolism and its association with functional genomic traits remains lacking (16). To enhance the understanding of glutamine metabolism-related heterogeneity in HCC, we developed a novel glutamine metabolism-based classification and a classifier to facilitate the application of this system to clinical practice. We present this article in accordance with the MDAR reporting checklist (available at https://jgo.amegroups.com/article/view/10.21037/jgo-2026-0706/rc).

Methods

Data acquisition

The transcriptome profiles of HCC samples was acquired from public datasets, including The Cancer Genome Atlas-Liver Hepatocellular Carcinoma (TCGA-LIHC; https://portal.gdc.cancer.gov/) (n=368), the GSE76427 dataset from the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE76427) (n=115), and the Liver Cancer-France (LICA-FR) (n=161) and Liver Cancer-Japan (LIRI-JP) (n=232) datasets from the International Cancer Genome Consortium (ICGC) (https://xena.ucsc.edu/) (17).

Consensus clustering analysis

Glutamine metabolism-related genes (GLS, GLS2, GLUD1, GLUD2, GLUL, GOT2, KYAT1, OAT, PYCR1, PYCR2, PYCR3, RIMKLA, and RIMKLB) were identified from the gene set collection of the Molecular Signature Database (MSigDB) (http://www.broad.mit.edu/gsea/msigdb/) (18). Based on their transcriptomic data, the classification of TCGA-LIHC samples was conducted with the ConsensusClusterPlus package in R (The R Foundation for Statistical Computing, Vienna, Austria) (19). The repeatability of the consensus clustering-based classification was verified in the GSE76427, LICA-FR, and LIRI-JP datasets.

Somatic mutation evaluation

The collected somatic data with Mutation Annotation Format (MAF) from the TCGA-LIHC dataset were assessed via Maftools in R (20).

Gene set enrichment analysis (GSEA)

Through GSEA software (21), the difference in Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment between cluster 1 and cluster 2 was investigated across the TCGA-LIHC samples. The enrichment score of tumorigenic pathways was estimated via single-sample GSEA (ssGSEA) and the GSVA package in R (22).

Estimation of immune infiltration

The infiltration levels of immune compositions were estimated in TCGA-LIHC tissues via the CIBERSORT approach (23). The overall infiltration of immune or stromal cells was evaluated via the ESTIMATE package in R (24).

Estimation of drug sensitivity

Drug sensitivity was determined based on the Genomics of Drug Sensitivity in Cancer (GDSC) resource (25). By integrating drug sensitivity information and transcriptome profiling of TCGA-LIHC samples, we inferred the half-maximal inhibitory concentration (IC₅₀) values using the oncoPredict package in R (26).

Prediction of response to immune checkpoint blockade (ICB)

The response to ICB of patients from TCGA-LIHC was determined in accordance with several predictors. Tumor mutation burden (TMB) was calculated to reflect cancer mutation (27). T cell-inflamed gene expression profile score was calculated according to a weighted linear combination of 18 immune-related signatures (28). The Tumor Immune Dysfunction and Exclusion (TIDE) tool, based on mechanisms of tumor immune escape (including T-cell dysfunction and T-cell exclusion), was used to predict responders and nonresponders to immunotherapy (29). The messenger RNA levels of known immune checkpoint molecules were quantified.

Differential expression analysis

Based upon the criteria of a q value ≤0.01 and |log₂fold change (log₂FC)| >1, genes with differential expression between clusters across TCGA-LIHC cases were selected via the edgeR package in R (30).

Establishment of a classifier

Through the glmnet R package (31), a least absolute shrinkage and selection operator (LASSO) binomial classifier was built for TCGA-LIHC patients based on the identified differentially expressed genes (DEGs) with |log₂FC| >4. The minimum lambda value was selected through 10-fold cross-validation, and feature genes were identified. TCGA-LIHC samples were placed into training and test sets at a 1:1 ratio. Time-dependent receiver operating characteristic (ROC) curve via the pROC R package were used to assess classification efficacy, and the area under the curve (AUC) values were calculated.

Patients and clinical specimens

Tissue samples from 50 patients with HCC were analyzed via immunohistochemistry (IHC). Patients with HCC undergoing radical resection between June 2021 and January 2022 at the Department of Pathology of The Second Affiliated Hospital of Soochow University were randomly selected. None had received chemotherapy or radiation therapy before surgery. HCC diagnoses were confirmed via hematoxylin and eosin (H&E) staining after surgical resection. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by institutional ethics committee of The Second Affiliated Hospital of Soochow University (No. JD-LK2024047-I01) and informed consent was taken from all the patients.

Immunohistochemical staining

All the surgically resected specimens and biopsy samples were fixed with 10% neutral buffered formalin (Thermo Fisher Scientific, Waltham, MA, USA), embedded in paraffin, and serially sectioned at 4 µm. IHC was performed on selected slides via the ChemMate Envision/horseradish peroxidase (HRP) technique (Dako, Carpinteria, CA, USA) (32). Briefly, the sections were deparaffinized and dehydrated; following the blocking of endogenous peroxidase activity via H₂O₂, the sections were incubated with primary antibodies for SLC2A1 (cat. No. 21829-1-AP; Proteintech, Rosemont, IL, USA), SLC2A5 (cat. No. 20350-1-AP; Proteintech), G6PD (cat. No. 25413-1-AP; Proteintech), and MEX3A (cat. No. ARG43260; Arigo Biolaboratories, Hsinchu City, China), as well as secondary antibodies, and visualized with diaminobenzidine (DAB) (Dako, Glostrup, Denmark). Finally, the slides were counterstained with hematoxylin (33).

Cell lines

The human HCC MHCC-97L and Mahlavu cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM; cat. No. SH30243.01; Cytiva, Marlborough, MA, USA) supplemented with 10% fetal bovine serum (FBS; cat. No. A5256701; Gibco, Thermo Fisher Scientific) and 1% penicillin-streptomycin (cat. No. C100C5; NCM Biotech, Suzhou, China). The medium was replaced every 3 to 4 days, and the cells were passaged with 0.25% trypsin (cat. No. C100C1; NCM Biotech). The cell lines were maintained at 37 ℃ in a humidified atmosphere with 5% CO₂. Following treatment, the culture medium was replaced with fresh medium containing varying concentrations of L-glutamine (cat. No. 25030149; Gibco, Thermo Fisher Scientific), which was prepared with glutamine-free culture medium (cat. No. 11960044; Gibco, Thermo Fisher Scientific) supplemented with 10% FBS (cat. No. A5256701; Gibco, Thermo Fisher Scientific) and 1% penicillin-streptomycin (cat. No. C100C5; NCM Biotech).

Cell Counting Kit-8 (CCK-8) cytotoxicity assay

Cell viability was assessed with CCK-8 (cat. No. K1018; APExBIO Technology, Houston, TX, USA) according to the manufacturer’s protocol. Briefly, human HCC lines MHCC-97L and Mahlavu were seeded in 96-well plates at a density of 5,000 cells per well in 100 µL of complete growth medium. After 24 hours of incubation to allow for cell attachment, the medium was aspirated and replaced with 100 µL of fresh medium containing varying concentrations (0, 0.02, 0.2, 0.5, 1, 2, 4, 6, 8, 10, and 20 mM) of L-glutamine (cat. No. 25030149; Gibco, Thermo Fisher Scientific). Cells were then incubated under standard conditions (37 ℃ and 5% CO₂) for 48 hours.

Following the treatment period, 10 µL of the CCK-8 solution was added directly to each well. The plate was returned to the incubator for 2 hours. The absorbance of each well at 450 nm was then measured with a microplate reader (BioTek Instruments, Agilent Technologies, Santa Clara, CA, USA). Wells containing culture medium and CCK-8 reagent without cells served as blanks. The cell viability for each treatment group was calculated as a percentage relative to the untreated control group (4 mM of L-glutamine), as per the following formula:

Cellviability(%)=[(AS−Ab)/(Ac−Ab)×100%][1]

where As is the absorbance of the treated sample, Ac is the absorbance of the untreated control, and Ab is the absorbance of the blank.

Cell proliferation assay

Cell proliferation was dynamically assessed via CCK-8 assay. Briefly, MHCC-97L and Mahlavu cells were seeded in 96-well plates at a density of 3,000 cells per well (in 100 µL of medium) and allowed to adhere for 24 hours. The medium was then replaced with fresh medium containing 0, 0.2, 4, or 10 mM of L-glutamine. For each time point (0, 24, 48, 72, and 96 hours posttreatment), a separate plate was prepared. At the designated time, 10 µL of CCK-8 reagent was added to each well, followed by a 2-hour incubation. Absorbance at 450 nm was measured with a microplate reader, and cell proliferation relative to the 24-hour control was calculated. Each condition was performed with at least six replicates, and experiments were repeated three times independently.

Transwell assay

The cell migration and invasion assays were conducted on the MHCC-97L and Mahlavu cell lines. Cells were seeded at a density of 5×10⁵ cells/mL in each well. For both assays, a total volume of 100 µL of cell suspension was added to the upper chamber of the Transwell insert. The lower chamber was filled with 700 µL of culture medium containing 10% FBS and different concentrations of glutamine (0.2, 4, and 10 mM). For the migration assay, the upper chamber was supplied with medium containing no serum but with the same glutamine concentrations. For the invasion assay, Matrigel was used as a coating for the upper chamber and prepared in medium containing 0.2, 4, or 10 mM glutamine and no serum at a 1:9 ratio. A total of 50 µL of Matrigel solution was added to the upper chamber for each well. Both migration and invasion assays were conducted for 24 hours at 37 ℃ in a 5% CO₂ incubator. After incubation, the cells that had migrated or invaded to the lower side of the membrane were fixed, stained, and quantified via counting under a light microscope.

Measurement of intracellular glutamine content

Intracellular glutamine levels were measured with a commercial glutamine assay kit (cat. No. MAK438; Sigma-Aldrich, St. Louis, MO, USA) according to the manufacturer’s instructions. This assay is based on an enzymatic cycling reaction. The principle of the assay involves two sequential enzymatic steps: glutaminase catalyzes the hydrolysis of glutamine to form glutamate and ammonia, and glutamate dehydrogenase subsequently converts glutamate to α-ketoglutarate, which is accompanied by the reduction of nicotinamide adenine dinucleotide (NAD)⁺ to NAD (with hydrogen) (NADH). The generated NADH then reacts with a colorimetric probe, such as WST-8, producing a formazan dye. The absorbance of the resulting dye was measured at 450 nm, and the intensity of the absorbance was directly proportional to the glutamine concentration in the sample.

Protein extraction and Western blotting

MHCC-97L and Mahlavu cells were seeded in six-well plates and cultured for 48 hours in glutamine-free DMEM supplemented with 10% dialyzed FBS and 0.2, 4, or 10 mM of L-glutamine. For protein extraction, cells were lysed using cell radioimmunoprecipitation assay buffer (cat. No. R0020; Solarbio, Beijing, China) containing phenylmethylsulfonyl fluoride (cat. No. IKM1140; Solarbio) and a mixture of protease and phosphatase inhibitors (cat. No. K1015-A; APExBIO, Houston, TX, USA). Protein samples were separated via sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred onto polyvinylidene difluoride membranes. The membranes were blocked with 5% nonfat milk for 1 hour at room temperature, followed by incubation overnight at 4 ℃ with primary antibodies against SLC2A1 (cat. No. 21829-1-AP; Proteintech), SLC2A5 (cat. No. 20350-1-AP; Proteintech), G6PD (cat. No. 25413-1-AP; Proteintech), MEX3A (cat. No. ARG43260; Arigo Biolaboratories), cleaved notch1 polyclonal antibody (cat. No. 10062-2-AP; Proteintech), NRF2/NFE2L2 polyclonal antibody (cat. No. 16396-1-AP; Proteintech), beta catenin polyclonal antibody (cat. No. 51067-2-AP; Proteintech), HO-1/HMOX1 polyclonal antibody (cat. No. 10701-1-AP; Proteintech), phospho-GSK3B (Ser389) polyclonal antibody (cat. No. 14850-1-AP; Proteintech), glyceraldehyde-3-phosphate dehydrogenase (GAPDH), and β-actin. After a wash, they were treated with HRP-conjugated secondary antibodies for 1 hour at room temperature (≈25 ℃). Protein detection was performed with a chemiluminescent HRP substrate and captured with an electrochemiluminescence imaging system (Tanon, Shanghai, China).

Statistical analysis

The data from the CCK-8 assay, intracellular glutamine content measurement, and Transwell migration and invasion assays were analyzed with one-way analysis of variance (ANOVA) followed by the Tukey post-hoc test. The CCK-8 proliferation assay data were analyzed via two-way ANOVA, which was also followed by the Tukey test. Statistical significance was defined as a P value <0.05. Data are presented as the mean ± standard deviation (SD). Statistical analysis was performed with GraphPad Prism software (Dotmatics, Boston, MA, USA).

Results

Tumor heterogeneity and novel glutamine metabolism-based classification of HCC

Based on the transcriptomic data of glutamine metabolism-related genes (GLS, GLS2, GLUD1, GLUD2, GLUL, GOT2, KYAT1, OAT, PYCR1, PYCR2, PYCR3, RIMKLA, and RIMKLB), we classified TCGA-LIHC patients into two clusters: cluster 1 and cluster 2 (Figure 1A). Across the two clusters, there were differences in glutamine metabolism-related genes (Figure 1B) and transcriptomics between the classes (Figure 1C), with the latter being confirmed via principal component analysis (PCA). Next, we examined the differences in clinical features between the clusters. Cluster 1, compared to cluster 2, had a more advanced T stage and higher histologic grade and pathologic stage as compare to cluster 2, but there were no significant differences in sex, overall survival (OS) status, node stage (N stage), or metastasis stage (M stage) (Figure 1D-1J). Through univariate Cox regression analysis, we found that cluster 2 acted as a protective factor for patients with HCC, being associated with longer survival (Figure 1K). In addition, patients in cluster 2 had a significantly longer OS, disease-free survival (DFS), and progression-free survival (PFS) as compared to those in cluster 1 (Figure 1L-1N). Overall, therefore, patients in cluster 2 had more favorable clinical outcomes.

Figure 1 Tumor heterogeneity and novel glutamine metabolism-based classification in TCGA-LIHC patients. (A) Consensus matrix when k=2 based on the transcriptomic expression levels of glutamine metabolism-related genes. (B) Heatmap of the transcriptome profiling of glutamine metabolism-related genes and clinical features across different glutamine metabolism-based clusters. Red indicates upregulation, while blue indicates downregulation. (C) PCA for the distribution of two clusters on the basis of the transcript levels of glutamine metabolism-related genes. Red indicates cluster 1, while blue indicates cluster 2. (D-J) Bar plots depicting the distribution of sex, OS status, T stage, N stage, M stage, histologic grade, and pathologic stage between the clusters. (K) Univariate Cox regression analyses of the associations between glutamine metabolism-based clusters, clinicopathological characteristics, and HCC survival. (L-N) K-M curves for the OS, DFS, and PFS of the two clusters. CI, confidence interval; DFS, disease-free survival; HR, hazard ratio; K-M, Kaplan-Meier; M, metastasis; N, node; OS, overall survival; PC, principal component; PCA, principal component analysis; PFS, progression-free survival; T, tumor; TCGA-LIHC, The Cancer Genome Atlas-Liver Hepatocellular Carcinoma.

Clinical implication of glutamine metabolism-based clusters in HCC

To ascertain the clinical significance of glutamine metabolism-based clusters in HCC, we conducted stratified survival analysis. Patients in the TCGA-LIHC were classified as female or male, with N0 or M0 stage, with histologic grade 1–2 or 3–4, and with pathologic stage I–II or III–IV. Cluster 2 patients demonstrated better OS (Figure 2A-2H), DFS (Figure 2I-2P), and PFS (Figure 2Q-2X) outcomes in each subgroup, indicating different survival outcomes between the two clusters.

Figure 2 Clinical implication of glutamine metabolism-based clusters in TCGA-LIHC according to stratification analysis. (A-H) K-M curves for OS of the two clusters in each subgroup stratified according to (A,B) female vs. male sex, (C,D) N0 vs. M0 stage, (E,F) histologic grade 1–2 vs. 3–4, and (G,H) pathologic stage I–II vs. III–IV. (I-P) K-M curves for DFS of the two clusters in each subgroup stratified according to (I,J) female vs. male sex, (K,L) N0 vs. M0 stage, (M,N) histologic grade 1–2 vs. 3–4, and (O,P) pathologic stage I–II vs. III–IV. (Q-X) K-M curves for PFS of the two clusters in each subgroup stratified according to (Q,R) female vs. male sex, (S,T) N0 vs. M0, (U,V) histologic grade 1–2 vs. 3–4, and (W,X) pathologic stage I–II vs. III–IV. DFS, disease-free survival; HR, hazard ratio; K-M, Kaplan-Meier; M, metastasis; N, node; OS, overall survival; PFS, progression-free survival; TCGA-LIHC, The Cancer Genome Atlas-Liver Hepatocellular Carcinoma.

External verification of the reproducibility of the glutamine metabolism-based classification in independent datasets

The robustness and reproducibility of the proposed glutamine metabolism-based classification system were verified in three independent datasets. Patients with HCC were robustly classified into two classes according to glutamine metabolism-related genes. Two classes were clearly distinguished in the GSE76427 (Figure 3A-3C), LICA-FR (Figure 3D-3F), and LIRI-JP (Figure 3G-3I) datasets, confirming the robustness and reproducibility of this classification method.

Figure 3 External verification of the reproducibility of the glutamine metabolism-based clusters in independent datasets. (A-I) Consensus matrix k=2 based on the transcriptomic expression levels and PCA of the two clusters in the (A-C) GSE76427, (D-F) LICA-FR, and (G-I) LIRI-JP datasets. LICA-FR, Liver Cancer-France; LIRI-JP, Liver Cancer-Japan; PC, principal component; PCA, principal component analysis.

Mapping of the glutamine metabolism-based clusters onto existing classes

To further characterize the features of the proposed glutamine metabolism-based classification, we mapped our classifications onto the metabolism-associated classifications reported by Yang et al. (34). Figure 4A illustrates the previously identified three metabolism-associated classes across TCGA-LIHC patients, and it was determined that our glutamine metabolism-based classification was different from that of Yang et al. (34) (Figure 4B).

Figure 4 Mapping between the proposed glutamine metabolism-based classification system and previous systems. (A) Consensus matrix when k=3 for three metabolism-associated clusters reported by Yang et al. (34). (B) Heatmap of the distribution of the glutamine metabolism-based classification, previously reported classification, and clinical features across TCGA-LIHC. M, metastasis; N, node; TCGA-LIHC, The Cancer Genome Atlas-Liver Hepatocellular Carcinoma.

Somatic mutations of the two glutamine metabolism-based clusters

We further examined the genetic mutations of the two glutamine metabolism-based clusters. Overall, somatic mutations were more frequent in cluster 1 than in cluster 2 (Figure 5A,5B). Cluster 2 had higher mutation frequencies for F5, MAGEL2, CHSY3, HCN1, HMGCR, MAN2A2, and LRRC7, while cluster 1 had higher mutation frequencies for CTNNB1, SETD2, and BSN (Figure 5C-5E). Figure 5F,5G depict the co-occurrence and mutual exclusion of mutated genes in cluster 1 and cluster 2, respectively.

Figure 5 The heterogeneous somatic mutations in two glutamine metabolism-based clusters. (A,B) Landscape of somatic mutations in cluster 1 and cluster 2. (C-E) Comparison of the mutation frequency between cluster 1 and cluster 2. (F,G) Co-occurrence or mutual exclusion of mutated genes in cluster 1 and cluster 2. *, P value <0.05; **, P value <0.01; ***, P value <0.001. CI, confidence interval; DEL, deletion; INS, insertion; SNP, single nucleotide polymorphism; SNV, single nucleotide variant.

Oncogenic pathways and the tumor microenvironment of the two glutamine metabolism-based classes

DNA replication, mismatch repair, and Fanconi anemia pathways were significantly enriched in cluster 1, while tyrosine metabolism, retinol metabolism, and primary bile acid biosynthesis were significantly enriched in cluster 2 (Figure 6A). Among the classic oncogenic pathways, the Wnt, NRF2, and Notch pathways were more activated in cluster 1 than in cluster 2 (Figure 6B,6C), while cell cycle progression (CCP) was more activated in cluster 1 (Figure 6D,6E). Next, the tumor microenvironment of the two clusters was characterized. There was high infiltration of memory B cells, eosinophils, M0 macrophages, activated mast cells, plasma cells, activated memory CD4 T cells, and regulatory T cells (Tregs) in cluster 1, while cluster 2 exhibited high infiltration of resting memory CD4 T cells and M1 and M2 macrophages (Figure 6F-6H). The ESTIMATE method indicated a lower stromal score in cluster 1 than in cluster 2 (Figure 6I-6L). Overall, these findings indicate the greater activation of oncogenic pathways and immune infiltration in cluster 1 than in cluster 2.

Figure 6 The heterogeneous tumorigenic pathways, and tumor microenvironment in two glutamine metabolism-based clusters across TCGA-LIHC. (A) GSEA of the enriched KEGG pathways in cluster 1 or cluster 2. (B,C) Comparison of the enrichment levels of classic oncogenic pathways in the two clusters. (D,E) Comparison of the enrichment levels of tumor progression-relevant pathways: EMT and CCP. (F-H) Comparison of the infiltration of immune compositions between the two clusters. (I-L) Comparison of ESTIMATE score, immune score, stromal score, and tumor purity between the two clusters. *, P value <0.05; **, P value <0.01; ***, P value <0.001. CCP, cell cycle progression; EMT, epithelial-mesenchymal transition; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; NK, natural killer; TCGA-LIHC, The Cancer Genome Atlas-Liver Hepatocellular Carcinoma.

Response to small-molecule inhibitors and ICB of the glutamine metabolism-based clusters

The sensitivity to small-molecule inhibitors was compared between the glutamine metabolism-based cluster. It was found that cluster 1 patients were more sensitive to sepantronium bromide, ulixertinib, pevonedistat, paclitaxel, ML323, ulixertinib, MG-132, MK-1775, palbociclib, ERK_6604, lapatinib, dasatinib, and telomerase inhibitor IX (Figure 7A,7B). Meanwhile, cluster 2 patients were more sensitive to TAF1_5496, doramapimod, and JAK1_8709. Subsequently, we inferred the response to ICB via TMB score and found that cluster 1 had a higher score compared to cluster 2 (Figure 7C,7D). Additionally, the T cell-inflamed gene expression profile score was higher in cluster 1 than in cluster 2 (Figure 7E). The TIDE algorithm indicated that there were more ICB responders in cluster 1 than in cluster 2 (Figure 7F). Most immune checkpoints (CD276, CD80, VTCN1, PDCD1, CD86, IDO1, and HAVCR2) displayed higher transcript levels in cluster 1 (Figure 7G). Altogether, these findings suggested that patients in cluster 1 are more responsive to ICB.

Figure 7 The differences in response to small-molecule inhibitors and ICB between the glutamine metabolism-based clusters in TCGA-LIHC. (A,B) Comparison of the IC₅₀ values of small-molecule inhibitors between the two clusters. (C,D) The distribution of TMB score in TCGA-LIHC, and the difference in TMB between the two clusters. (E) Comparison of the T cell-inflamed gene expression profile score between the two clusters. (F) The distribution of responders and nonresponders to ICB in the two clusters. (G) Comparison of the transcript values of immune checkpoints in the two clusters. *, P value <0.05; **, P value <0.01; ***, P value <0.001; ns, no significance. CPM, counts per million; IC₅₀, half-maximal inhibitory concentration; ICB, immune checkpoint blockade; TCGA-LIHC, The Cancer Genome Atlas-Liver Hepatocellular Carcinoma; TMB, tumor mutation burden.

Development of a classifier for glutamine metabolism-based clusters

To expand the potential application value of the glutamine metabolism-based classification system, we developed a classifier for distinguishing between the two clusters. We first selected genes with differential expression between the clusters using the following criteria: q value ≤0.01 and |log₂FC| >1 (Figure 8A-8C). Next, the genes with |log₂FC| >4 were selected for LASSO analysis (Table 1). According to 10-fold cross-validation, 27 feature genes were screened for building the classifier: KIRREL2, IGFN1, PLD5, KLK11, WIF1, ODAM, PART1, RTL1, IGDCC3, ALDH3B2, CRTAC1, RP11-54H7.4, CHGA, PGC, SH3GL3, BPIFA1, LGALS14, SYT8, SERPINB3, KLK4, SI, CYP2A13, C16orf89, GCG, PITPNM2-AS1, RBPJL, and GLS2 (Figure 8D,8E). Via a combination of the transcript levels and coefficients of the identified feature genes (Table 2), the classifier was used for predicting the two glutamine metabolism-based clusters. The AUC values in the training, test, and complete sets all exceeded >0.9, indicating excellent efficacy in classification prediction (Figure 8F). Univariate Cox regression analysis indicated that classifier-predicted cluster 2 was a protective factor of HCC survival (Figure 8G). In line with this, patients in classifier-predicted cluster 1 had a shorter OS and DFS time (Figure 8H,8I). Additionally, cluster 1 had higher transcript levels of immune checkpoint molecules (CD276, CD80, VTCN1, PDCD1, CD86, IDO1, and HAVCR2) (Figure 8J). These findings confirmed that the classifier had good accuracy in identifying the glutamine metabolism-based clusters.

Figure 8 Definition of a classifier for glutamine metabolism-based clusters in the TCGA-LIHC dataset. (A,B) Genes with differential expression in the two clusters. (C) The top 20 DEGs in each cluster. (D,E) LASSO regression results and 10-fold cross-validation. (F) ROC curves of the classification efficacy of the classifier in the training, test, and complete sets. (G) Univariate Cox regression results on the classifier and clinical traits with TCGA-LIHC survival. (H,I) K-M curves of OS and DFS in the two predicted clusters. (J) The transcriptome differences in immune checkpoint molecules between the two classifier-predicted clusters. *, P value <0.05; **, P value <0.01; ***, P value <0.001. AUC, area under the curve; CI, confidence interval; CPM, counts per million; DEG, differentially expressed gene; DFS, disease-free survival; HR, hazard ratio; K-M, Kaplan-Meier; LASSO, least absolute shrinkage and selection operator; M, metastasis; N, node; OS, overall survival; ROC, receiver operating characteristic; T, tumor; TCGA-LIHC, The Cancer Genome Atlas-Liver Hepatocellular Carcinoma.

IHC staining of tissue samples from patients with HCC

Tissue samples from 50 patients with HCC were analyzed via IHC. HCC diagnoses were confirmed via H&E staining after surgical resection (Figure 9A). Our analysis of selected genes differentially expressed between clusters revealed that SLC2A1, SLC2A5, G6PD, and MEX3A expression was related to glucose metabolism and bile acid metabolism. The results were validated in HCC tissues (Figure 9B-9E). Increased SLC2A1, SLC2A5, G6PD, and MEX3A expression was significantly correlated with tumor differentiation.

Figure 9 Immunohistochemical staining of the HCC patients’ tissue samples. (A) HCC diagnoses were confirmed via H&E staining after surgical resection. (B) Expression of SLC2A1 in HCC tissues. (C) Expression of SLC2A5 in HCC tissues. (D) Expression of G6PD in HCC tissues. (E) Expression of MEX3A in HCC tissues. Increased SLC2A1, SLC2A5, G6PD, and MEX3A expression was significantly correlated with tumor differentiation. Scale bars: 100 μm for 100× magnification; 200 μm for 40× magnification. H&E, hematoxylin and eosin; HCC, hepatocellular carcinoma.

Glutamine availability differentially modulated metastatic potential and metabolism-related protein expression in HCC cells independent of proliferation

Based on the results from the CCK-8 cytotoxicity assay, glutamine deprivation (0.2 mM) and high-glutamine (10 mM) models were constructed, with 4 mM of glutamine serving as the control condition (Figure 10A). To assess the impact of varying glutamine concentrations on HCC cells, we first measured the glutamine content in MHCC-97L and Mahlavu cells under different conditions. Compared to the control group (4 mM), the glutamine deprivation model (0.2 mM) exhibited a significant reduction in intracellular glutamine levels, while the high-glutamine model (10 mM) had significantly increased glutamine content (Figure 10B). These findings confirmed that the glutamine concentrations used in the experimental conditions effectively modulated intracellular glutamine levels, and we then examined their subsequent effects on HCC cell behavior. The cell proliferation assays (Figure 10C) revealed no significant differences in cell proliferation across the different glutamine concentrations in MHCC-97L and Mahlavu cell lines, suggesting that glutamine concentrations within the tested range did not markedly alter the proliferative capacity of these liver cancer cells. However, Transwell migration and invasion assays (Figure 10D) indicated that, compared to the normal culture medium (4 mM), glutamine deprivation suppressed cell migration and invasion, while high-glutamine conditions significantly promoted these metastatic behaviors. This suggested that glutamine may be involved in enhancing HCC cell aggressiveness.

Figure 10 Effects of glutamine concentration on HCC cell behaviors and protein expression. (A) Experimental setup: glutamine deprivation (0.2 mM), control (4 mM), and high-glutamine (10 mM) conditions. (B) The expression status of glutamine within cells. (C) CCK-8 assay indicated no significant differences in proliferation across glutamine concentrations in the MHCC-97L and Mahlavu cells. (D) Transwell assays indicated that glutamine deprivation and elevation inhibited and promoted migration and invasion, respectively. Scale bars: 100 μm for 100× magnification; 20 μm for 400× magnification. Cells were stained with crystal violet. (E) WB measured the expression levels of GLUT1, GLUT5, G6PD, and MEX3A at different glutamine concentrations (0.2, 4, and 10 mM). Glutamine concentration was found to affect the metabolism-related proteins and metastatic behaviors of HCC cells. (F) WB measured the expression levels of p-GSK-3β, β-catenin, NRF2, HO-1, and the cleaved NICD, and MEX3A at different glutamine concentrations. *, P value <0.05; **, P value <0.01; ***, P value <0.001; ****, P value <0.0001; ns, no significance. CCK-8, Cell Counting Kit-8; HCC, hepatocellular carcinoma; NICD, Notch1 intracellular domain; OD, optical density; WB, Western blotting.

Western blot analysis further supported these findings, revealing that the expression of SLC2A1 (GLUT1), SLC2A5 (GLUT5), and G6PD was significantly reduced in the glutamine deprivation group compared to the control group, whereas the high-glutamine group exhibited increased expression levels of these proteins (Figure 10E). In contrast, the expression of MEX3A was lower in the glutamine deprivation group compared to the control group, while no significant change was observed between the high-glutamine and control groups (Figure 10E). These results suggest that the alterations in glutamine availability can modulate key metabolism-related proteins and signaling pathways involved in HCC progression, with high-glutamine conditions potentially promoting metastasis through the regulation of metabolism-related pathways. Crucially, consistent with the observed phenotypic changes, Western blot analysis further demonstrated that the protein levels of key signaling nodes—including phosphorylated GSK-3β (Ser9), β-catenin, NRF2, HO-1, and the cleaved Notch1 intracellular domain (NICD)—were all markedly upregulated in response to increasing glutamine concentrations (Figure 10F). These results suggest that alterations in glutamine availability can modulate key metabolism-related proteins and signaling pathways involved in HCC progression, with high-glutamine conditions potentially promoting metastasis through the coordinated regulation of metabolic and oncogenic pathways.

Discussion

A dependence on glutamine is a characteristic feature of tumor cell metabolism, and in the context of HCC, further molecular classification of glutamine metabolism and its clinical implications may prove valuable. In line with the transcriptomic expression levels of glutamine metabolism-related genes (GLS, GLS2, GLUD1, GLUD2, GLUL, GOT2, KYAT1, OAT, PYCR1, PYCR2, PYCR3, RIMKLA, and RIMKLB), we classified TCGA-LIHC patients into two clusters, with each class exhibiting distinct survival outcomes and clinical traits. The relevance of these glutamine metabolism-related genes has been established in previous research. For instance, macrophage activation and metabolism-related phenotypes can be fueled by SLC2A1 (35). The alteration in glutamine to glutamic acid at codon 166 can influence the substrate-binding specificity via transforming SLC2A5-independent transport from fructose to glucose (36). Moreover, glutamine mediates activity and functions of G6PD through NRF2 in malignant cells (37,38), and MEX3A serves as an independent predictor of HCC survival (39,40).

Despite the identification of molecular etiology, drivers, and molecular and immune classifications, manipulable mutations are rare in HCC (only 25% of tumors present a druggable target) (41,42). Therefore, neither oncogenic driver factors or molecular classifications have been translated effectively into clinical decision-making (43). In our study, the two glutamine metabolism-based clustered demonstrated markedly distinct somatic mutations, with mutations being more frequent in cluster 1. Dai et al. reported that glutamine synthase attenuates the growth of β-catenin-mutated HCC through sustaining nitrogen homeostasis and inhibiting mTORC1 (44). In our study, CTNNB1, which encodes β-catenin, was more frequently mutated in cluster 1. Cluster 1 also had stronger activation of oncogenic, DNA replication, mismatch repair, Fanconi anemia, Wnt, NRF2, Notch pathways, and CCP pathways, suggesting the role of aberrant glutamine metabolism in HCC progression.

The varied tumor microenvironment exerts constant selection pressure (45), resulting in intratumoral heterogeneity, which is a crucial feature of HCC (42,46). Glutamine is a critical nutrient for the proliferative capacity and functions of immune compositions within the tumor microenvironment (47,48). The rapid consumption of glutamine by malignant cells leads to a tumor microenvironment characterized by depleted glutamine availability for immune cells (49). Hence, glutamine utilization by malignant cells may be a metabolism-related checkpoint that suppresses the immune composition-induced antitumor response (11). In our study, cluster 1 had a higher infiltration of memory B cells, eosinophils, M0 macrophages, activated mast cells, plasma cells, activated memory CD4 T cells, and Tregs, while cluster 2 had a higher infiltration resting memory CD4 T cells and M1 and M2 macrophages. The high glutamine demand in cluster 1 (the high-malignancy subtype) creates a unique metabolic niche. First, excessive glutamine catabolism fuels the tricarboxylic acid (TCA) cycle and supports the biosynthesis of glutathione, leading to elevated antioxidant capacity. This reduces the accumulation of reactive oxygen species (ROS) within the tumor milieu, thereby protecting Tregs from oxidative stress-induced apoptosis and promoting their suppressive function. Second, glutamine-derived metabolites (e.g., α-ketoglutarate) can act as epigenetic modifiers, potentially enhancing the stability of transcription factors like FOXP3 in Tregs, further consolidating the immunosuppressive landscape. The glutamine metabolism traits of HCC considerably hinder immune cell functions and ICB. Glutamine inhibition triggers a variety of metabolism-related programs that aid in overcoming tumor immune escape (50). Given the higher TMB and T cell-inflamed gene expression profile score, along with results related to immune checkpoints and TIDE, cluster 1 patients may respond more favorably to ICB.

Patients with advanced HCC are chemotherapy- and radiotherapy-resistant, thus limiting their treatment options. We identified novel small-molecule inhibitors for each glutamine metabolism-based cluster. Cluster 1 patients had stronger sensitivity to sepantronium bromide, ulixertinib, pevonedistat, paclitaxel, ML323, ulixertinib, MG-132, MK-1775, palbociclib, ERK_6604, lapatinib, dasatinib, and telomerase inhibitor IX, while those in cluster 2 were highly sensitive to TAF1_5496, doramapimod, and JAK1_8709. However, further in-depth experimental verification is needed to confirm the treatment efficacy of each of these therapeutic agents for the respective cluster.

We applied the LASSO classifier (composed of KIRREL2, IGFN1, PLD5, KLK11, WIF1, ODAM, PART1, RTL1, IGDCC3, ALDH3B2, CRTAC1, RP11-54H7.4, CHGA, PGC, SH3GL3, BPIFA1, LGALS14, SYT8, SERPINB3, KLK4, SI, CYP2A13, C16orf89, GCG, PITPNM2-AS1, RBPJL, and GLS2) to predict the two glutamine metabolism-based clusters. Patients in classifier-predicted cluster 1 had a shorter survival time but a more favorable response to ICB. The classifier achieved AUC values >0.9, indicating excellent accuracy in the prediction of clusters.

The results of this study furnish an enhanced understanding of the role of glutamine metabolism in HCC. IHC confirmed the differential expression of key glutamine metabolism-related genes in clinical samples, supporting the findings from the bioinformatics analysis. These genes may play significant roles in the metabolism-related reprogramming observed in HCC. Furthermore, the CCK-8 assays indicated no significant differences in cell proliferation at different glutamine concentrations, suggesting that glutamine availability does not directly affect the proliferative capacity of MHCC-97L and Mahlavu cells within the tested concentration range. This further implies that while glutamine is critical to cell survival, it may not directly influence cell growth in the context of HCC, which aligns with our bioinformatics-based prediction that glutamine contributes to metabolism-related homeostasis and not proliferation.

Meanwhile, the Transwell migration and invasion assays highlighted the functional significance of glutamine in HCC metastasis. Glutamine deprivation (0.2 mM) significantly reduced cell migration and invasion, whereas high-glutamine (10 mM) conditions promoted these aggressive behaviors. These findings are consistent with the bioinformatics analysis, which indicated that altered glutamine metabolism is linked to enhanced migratory and invasive potential in HCC. Moreover, the Western blot analysis revealed that high-glutamine conditions increased the expression of key proteins involved in cellular metabolism, such as SLC2A1, SLC2A5, and G6PD, while glutamine deprivation led to a reduction in their expression. Interestingly, the expression of MEX3A was reduced in the glutamine deprivation group, while no significant changes were observed in the high-glutamine group. This suggests that high-glutamine conditions not only influence metabolism-related pathways but may also impact the regulation of proteins involved in cellular movement and invasion, further supporting the role of glutamine in HCC progression as indicated by our computational analysis.

Certain limitations to this study should be acknowledged. The proposed glutamine metabolism-based classification system remains to be validated in prospective studies. Moreover, the distinct response to small-molecule agents and ICB should be further verified. Additionally, our study primarily focused on in vitro results, and while these findings provide valuable insights, they need to be corroborated by in vivo models. The complexity of the tumor microenvironment and its interaction with metabolism-related reprogramming must be considered to better understand the clinical relevance of our findings. Future research should aim to establish in vivo models that better reflect the heterogeneity of human HCC and clarify how glutamine metabolism interacts with other metabolism-related pathways and therapeutic modalities. Ultimately, a comprehensive understanding of these interactions will be essential for the development of novel and effective treatment strategies for HCC. However, future experimental studies involving isotope tracing and targeted metabolomics are warranted to mechanistically validate these clusters.

Notably, our in vitro mechanistic investigations provided deeper insights into the signaling cascades governing this metabolic reprogramming. Western blot analyses demonstrated that escalating glutamine concentrations concurrently upregulated critical nodal proteins across multiple oncogenic pathways. Specifically, elevated glutamine enhanced the inhibitory phosphorylation of GSK-3β (Ser9) and stabilized β-catenin, confirming the activation of the Wnt/β-catenin axis—a key driver of the proliferative and stemness features observed in cluster 1. Concomitantly, glutamine supplementation increased the protein levels of NRF2 and its downstream effector HO-1, substantiating the link between glutamine availability and the antioxidant program that fosters an immunosuppressive niche. Furthermore, the accumulation of the cleaved NICD under high-glutamine conditions revealed a previously important role for glutamine in modulating Notch signaling, which likely contributes to the aggressive phenotype and altered differentiation status of HCC cells. These findings collectively illustrate that glutamine functions not merely as a metabolic substrate, but as a central signaling integrator that coordinates Wnt, NRF2, and Notch pathways to orchestrate the high-malignant phenotype characteristic of cluster 1.

Conclusions

We developed two glutamine metabolism-based clusters to reflect the heterogeneity of HCC. The classifier we devised may facilitate the translation of data related to glutamine metabolism into precision oncology and offer practical tool for more accurate risk assessment and personalized therapeutic strategies. Beyond simple stratification, this classifier enables the integration of metabolism-related classification into clinical practice. Importantly, experimental validations confirmed that glutamine metabolism contributes significantly to regulating HCC aggressiveness by modulating cell migration and invasion—but not proliferation—highlighting its pivotal role in metastatic behavior. At the molecular level, mechanistic investigations revealed that glutamine functions as a critical signaling integrator, coordinately activating the Wnt/β-catenin, NRF2, and Notch pathways. The upregulation of p-GSK-3β, β-catenin, NRF2, HO-1, and the cleaved NICD under high glutamine availability underscores the capacity of glutamine to orchestrate oncogenic signaling networks that drive the aggressive phenotype of cluster 1. These findings underscore the therapeutic potential of targeting glutamine metabolism to curb HCC progression. However, further prospective studies are needed to validate the clinical applicability of this classification system and its derived classifier.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the MDAR reporting checklist. Available at https://jgo.amegroups.com/article/view/10.21037/jgo-2026-0706/rc

Data Sharing Statement: Available at https://jgo.amegroups.com/article/view/10.21037/jgo-2026-0706/dss

Peer Review File: Available at https://jgo.amegroups.com/article/view/10.21037/jgo-2026-0706/prf

Funding: This work was supported in part by the National Natural Science Foundation of China (Nos. 82372636 and 82102826), the State Key Laboratory of Radiation Medicine and Protection (No. GZK1202242), the Science Projects of Suzhou Technology Bureau (No. SKY2023171), the Suzhou Municipal Project for the Development of Traditional Chinese Medicine Science and Technology (No. ZYMS202601), and the Talent Research Project of Suzhou Gusu Health Talents Program (No. GSWS202037).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jgo.amegroups.com/article/view/10.21037/jgo-2026-0706/coif). All authors report that this work was supported in part by the National Natural Science Foundation of China (Nos. 82372636 and 82102826), the State Key Laboratory of Radiation Medicine and Protection (No. GZK1202242), the Science Projects of Suzhou Technology Bureau (No. SKY2023171), the Suzhou Municipal Project for the Development of Traditional Chinese Medicine Science and Technology (No. ZYMS202601), and the Talent Research Project of Suzhou Gusu Health Talents Program (No. GSWS202037). The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by institutional ethics committee of The Second Affiliated Hospital of Soochow University (No. JD-LK2024047-I01) and informed consent was taken from all the patients.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Llovet JM, Pinyol R, Kelley RK, et al. Molecular pathogenesis and systemic therapies for hepatocellular carcinoma. Nat Cancer 2022;3:386-401. [Crossref] [PubMed]
2. Yoon SM, Ryoo BY, Lee SJ, et al. Efficacy and Safety of Transarterial Chemoembolization Plus External Beam Radiotherapy vs Sorafenib in Hepatocellular Carcinoma With Macroscopic Vascular Invasion: A Randomized Clinical Trial. JAMA Oncol 2018;4:661-9. [Crossref] [PubMed]
3. Di Martino M, Vitale A, Ferraro D, et al. Downstaging Therapies for Patients with Hepatocellular Carcinoma Awaiting Liver Transplantation: A Systematic Review and Meta-Analysis on Intention-to-Treat Outcomes. Cancers (Basel) 2022;14:5102. [Crossref] [PubMed]
4. Koeberle D, Dufour JF, Demeter G, et al. Sorafenib with or without everolimus in patients with advanced hepatocellular carcinoma (HCC): a randomized multicenter, multinational phase II trial (SAKK 77/08 and SASL 29). Ann Oncol 2016;27:856-61. [Crossref] [PubMed]
5. Shitara K, Özgüroğlu M, Bang YJ, et al. Molecular determinants of clinical outcomes with pembrolizumab versus paclitaxel in a randomized, open-label, phase III trial in patients with gastroesophageal adenocarcinoma. Ann Oncol 2021;32:1127-36. [Crossref] [PubMed]
6. Kang YK, Chen LT, Ryu MH, et al. Nivolumab plus chemotherapy versus placebo plus chemotherapy in patients with HER2-negative, untreated, unresectable advanced or recurrent gastric or gastro-oesophageal junction cancer (ATTRACTION-4): a randomised, multicentre, double-blind, placebo-controlled, phase 3 trial. Lancet Oncol 2022;23:234-47. [Crossref] [PubMed]
7. Wei Y, Tang X, Ren Y, et al. An RNA-RNA crosstalk network involving HMGB1 and RICTOR facilitates hepatocellular carcinoma tumorigenesis by promoting glutamine metabolism and impedes immunotherapy by PD-L1+ exosomes activity. Signal Transduct Target Ther 2021;6:421. [Crossref] [PubMed]
8. Wang Z, Yao J, Dong T, et al. Definition of a Novel Cuproptosis-Relevant lncRNA Signature for Uncovering Distinct Survival, Genomic Alterations, and Treatment Implications in Lung Adenocarcinoma. J Immunol Res 2022;2022:2756611. [Crossref] [PubMed]
9. Dai W, Xu L, Yu X, et al. OGDHL silencing promotes hepatocellular carcinoma by reprogramming glutamine metabolism. J Hepatol 2020;72:909-23. [Crossref] [PubMed]
10. Nan D, Yao W, Huang L, et al. Glutamine and cancer: metabolism, immune microenvironment, and therapeutic targets. Cell Commun Signal 2025;23:45. [Crossref] [PubMed]
11. Huang M, Xiong D, Pan J, et al. Targeting Glutamine Metabolism to Enhance Immunoprevention of EGFR-Driven Lung Cancer. Adv Sci (Weinh) 2022;9:e2105885. [Crossref] [PubMed]
12. Yin X, Peng J, Gu L, et al. Targeting glutamine metabolism in hepatic stellate cells alleviates liver fibrosis. Cell Death Dis 2022;13:955. [Crossref] [PubMed]
13. Liu T, Han C, Fang P, et al. Cancer-associated fibroblast-specific lncRNA LINC01614 enhances glutamine uptake in lung adenocarcinoma. J Hematol Oncol 2022;15:141. [Crossref] [PubMed]
14. Zou W, Han Z, Wang Z, et al. Targeting glutamine metabolism as a potential target for cancer treatment. J Exp Clin Cancer Res 2025;44:180. [Crossref] [PubMed]
15. Fang L, Gao D, Jiang Z, et al. Glutamine’s double-edged sword: fueling tumor growth and offering therapeutic hope. Front Immunol 2025;16:1578940. [Crossref] [PubMed]
16. Butrin A, Beaupre BA, Kadamandla N, et al. Structural and Kinetic Analyses Reveal the Dual Inhibition Modes of Ornithine Aminotransferase by (1S,3S)-3-Amino-4-(hexafluoropropan-2-ylidenyl)-cyclopentane-1-carboxylic Acid (BCF(3)). ACS Chem Biol 2021;16:67-75. [Crossref] [PubMed]
17. Grinchuk OV, Yenamandra SP, Iyer R, et al. Tumor-adjacent tissue co-expression profile analysis reveals pro-oncogenic ribosomal gene signature for prognosis of resectable hepatocellular carcinoma. Mol Oncol 2018;12:89-113. [Crossref] [PubMed]
18. Liberzon A, Birger C, Thorvaldsdóttir H, et al. The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst 2015;1:417-25. [Crossref] [PubMed]
19. Wilkerson MD, Hayes DN. ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking. Bioinformatics 2010;26:1572-3. [Crossref] [PubMed]
20. Mayakonda A, Lin DC, Assenov Y, et al. Maftools: efficient and comprehensive analysis of somatic variants in cancer. Genome Res 2018;28:1747-56. [Crossref] [PubMed]
21. Subramanian A, Tamayo P, Mootha VK, et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 2005;102:15545-50. [Crossref] [PubMed]
22. Hänzelmann S, Castelo R, Guinney J. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics 2013;14:7. [Crossref] [PubMed]
23. Chen B, Khodadoust MS, Liu CL, et al. Profiling Tumor Infiltrating Immune Cells with CIBERSORT. Methods Mol Biol 2018;1711:243-59. [Crossref] [PubMed]
24. Yoshihara K, Shahmoradgoli M, Martínez E, et al. Inferring tumour purity and stromal and immune cell admixture from expression data. Nat Commun 2013;4:2612. [Crossref] [PubMed]
25. Yang W, Soares J, Greninger P, et al. Genomics of Drug Sensitivity in Cancer (GDSC): a resource for therapeutic biomarker discovery in cancer cells. Nucleic Acids Res 2013;41:D955-61. [Crossref] [PubMed]
26. Maeser D, Gruener RF, Huang RS. oncoPredict: an R package for predicting in vivo or cancer patient drug response and biomarkers from cell line screening data. Brief Bioinform 2021;22:bbab260. [Crossref] [PubMed]
27. McGrail DJ, Pilié PG, Rashid NU, et al. High tumor mutation burden fails to predict immune checkpoint blockade response across all cancer types. Ann Oncol 2021;32:661-72. [Crossref] [PubMed]
28. Ayers M, Lunceford J, Nebozhyn M, et al. IFN-γ-related mRNA profile predicts clinical response to PD-1 blockade. J Clin Invest 2017;127:2930-40. [Crossref] [PubMed]
29. Jiang P, Gu S, Pan D, et al. Signatures of T cell dysfunction and exclusion predict cancer immunotherapy response. Nat Med 2018;24:1550-8. [Crossref] [PubMed]
30. Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics 2010;26:139-40. [Crossref] [PubMed]
31. Hastie T, Tibshirani R, Tibshirani R. Best subset, forward stepwise or lasso? Analysis and recommendations based on extensive comparisons. Statist Sci 2020;35:579-92.
32. Xie F, Wang Q, Chen Y, et al. Costimulatory molecule OX40/OX40L expression in ductal carcinoma in situ and invasive ductal carcinoma of breast: an immunohistochemistry-based pilot study. Pathol Res Pract 2010;206:735-9. [Crossref] [PubMed]
33. Li Y, Yang X, Wu Y, et al. B7-H3 promotes gastric cancer cell migration and invasion. Oncotarget 2017;8:71725-35. [Crossref] [PubMed]
34. Yang C, Huang X, Liu Z, et al. Metabolism-associated molecular classification of hepatocellular carcinoma. Mol Oncol 2020;14:896-913. [Crossref] [PubMed]
35. Freemerman AJ, Zhao L, Pingili AK, et al. Myeloid Slc2a1-Deficient Murine Model Revealed Macrophage Activation and Metabolic Phenotype Are Fueled by GLUT1. J Immunol 2019;202:1265-86. [Crossref] [PubMed]
36. Ebert K, Ewers M, Bisha I, et al. Identification of essential amino acids for glucose transporter 5 (GLUT5)-mediated fructose transport. J Biol Chem 2018;293:2115-24. [Crossref] [PubMed]
37. Feng K, Zhang X, Li J, et al. Neoantigens combined with in situ cancer vaccination induce personalized immunity and reshape the tumor microenvironment. Nat Commun 2025;16:5074. [Crossref] [PubMed]
38. Polat IH, Tarrado-Castellarnau M, Benito A, et al. Glutamine Modulates Expression and Function of Glucose 6-Phosphate Dehydrogenase via NRF2 in Colon Cancer Cells. Antioxidants (Basel) 2021;10:1349. [Crossref] [PubMed]
39. Ji PX, Zhang P, Zhou HL, et al. MEX3A promotes cell proliferation by regulating the RORA/β-catenin pathway in hepatocellular carcinoma. World J Gastrointest Oncol 2025;17:102084. [Crossref] [PubMed]
40. Yang D, Jiao Y, Li Y, et al. Clinical characteristics and prognostic value of MEX3A mRNA in liver cancer. PeerJ 2020;8:e8252. [Crossref] [PubMed]
41. Raman SS, Cutler NR, Sramek JJ, et al. Advancements in intratumoral therapies for liver tumors. Front Oncol 2026;16:1726128. [Crossref] [PubMed]
42. Xia Y, Tang W, Qian X, et al. Efficacy and safety of camrelizumab plus apatinib during the perioperative period in resectable hepatocellular carcinoma: a single-arm, open label, phase II clinical trial. J Immunother Cancer 2022;10:e004656. [Crossref] [PubMed]
43. Liu X, Niu X, Qiu Z. A Five-Gene Signature Based on Stromal/Immune Scores in the Tumor Microenvironment and Its Clinical Implications for Liver Cancer. DNA Cell Biol 2020;39:1621-38. [Crossref] [PubMed]
44. Dai W, Shen J, Yan J, et al. Glutamine synthetase limits β-catenin-mutated liver cancer growth by maintaining nitrogen homeostasis and suppressing mTORC1. J Clin Invest 2022;132:e161408. [Crossref] [PubMed]
45. Chen L, Niu X, Qiao X, et al. Characterization of Interplay Between Autophagy and Ferroptosis and Their Synergistical Roles on Manipulating Immunological Tumor Microenvironment in Squamous Cell Carcinomas. Front Immunol 2021;12:739039. [Crossref] [PubMed]
46. Khatib S, Pomyen Y, Dang H, et al. Understanding the Cause and Consequence of Tumor Heterogeneity. Trends Cancer 2020;6:267-71. [Crossref] [PubMed]
47. Zhang X, Zhang Y, Huang J, et al. Glutamine antagonist DON attenuates chikungunya virus-induced myositis by suppressing inflammatory activation in a murine model. Emerg Microbes Infect 2026;15:2622213. [Crossref] [PubMed]
48. Liu Y, Wu Z, Li Y, et al. Metabolic reprogramming and interventions in angiogenesis. J Adv Res 2025;70:323-38. [Crossref] [PubMed]
49. Kang HW, Kim JH, Jeong JW, et al. SLC6A14-mediated glutamine promotes SYTL4-CXCL8 axis activation to drive gemcitabine resistance and immune evasion in pancreatic cancer. Exp Mol Med 2025;57:2943-56. [Crossref] [PubMed]
50. Su M, Qin H, Shen J, et al. Beyond the tumor: Enhancing pancreatic cancer therapy through glutamine metabolism and innovative drug delivery. J Cell Commun Signal 2025;19:e70033. [Crossref] [PubMed]

(English Language Editor: J. Gray)