BLVRA promotes glioblastoma progression by regulating fatty acid metabolism

Abstract

Fatty acid metabolism is critically involved in glioblastoma (GBM) pathogenesis; however, its regulatory mechanisms remain incompletely understood. In this study, we identified biliverdin reductase A (BLVRA) as a novel metabolic driver and prognostic biomarker in GBM by integrating bulk and single-cell RNA sequencing with in vitro functional validation. Using ten machine learning algorithms, we developed a fatty acid metabolism–related gene prognostic index (FAMRGPI), which demonstrated strong prognostic value and highlighted the importance of metabolic reprogramming and immune modulation in GBM. Among FAMRGPI components, BLVRA emerged as an independent prognostic factor, with elevated expression associated with poor clinical outcomes. Single-cell transcriptomic analysis revealed that BLVRA expression correlated with tumor heterogeneity and differentiation potential. Experimental validation confirmed that BLVRA was markedly upregulated in GBM tissues and cell lines. Functional assays showed that BLVRA knockdown suppressed GBM cell proliferation and colony formation, disrupted lipid homeostasis (decreased carnitine and triglyceride levels, increased CPT1A and coenzyme A), and induced oxidative stress and DNA damage. Mechanistically, BLVRA depletion increased ROS accumulation, upregulated γ-H2AX, p53, and cleaved caspase-3 expression, and led to downregulation of key cell cycle regulators including CDK2, CDK4, CDK6, Cyclin D1, and c-MYC, thereby promoting cell cycle arrest and apoptosis. This study is the first to establish BLVRA as a critical regulator of fatty acid metabolism in GBM, revealing its dual role in maintaining metabolic homeostasis and supporting tumor growth. These findings uncover a novel metabolic vulnerability in GBM and suggest that BLVRA may serve as a potential therapeutic target for metabolic intervention.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-19026-2.

Introduction

GBM is the most common and aggressive malignant brain tumor in adults, accounting for approximately 56.6% of all gliomas and 49% of all brain tumors¹. Despite advancements in standard treatments such as surgery, radiotherapy, and chemotherapy, GBM is still characterized by high recurrence rates and drug resistance. The median survival of GBM patients is less than two years, and the five-year survival rate is as low as 5.6%². Alarmingly, the incidence of GBM has been rising among younger individuals in recent years^3–5. In our clinical cohort, the youngest patient diagnosed was 35 years old, underscoring the urgent need for more effective therapeutic strategies.

Metabolic reprogramming is a hallmark of GBM pathogenesis. Cancer cells adapt to the tumor microenvironment and sustain growth by enhancing glycolysis (the Warburg effect), modulating glutamine metabolism, and remodeling fatty acid metabolism^6,7. While glucose and glutamine metabolism have been extensively studied, the role of dysregulated fatty acid metabolism in GBM remains comparatively underexplored. Fatty acids not only serve as fundamental components of cell membranes but also function as energy sources through β-oxidation while regulating biological processes such as oxidative stress, signal transduction, and ferroptosis⁸. Previous studies have demonstrated that aberrant fatty acid metabolism promotes GBM proliferation, invasion, and therapy resistance⁹. Elucidating the molecular mechanisms underlying lipid metabolic reprogramming may therefore uncover novel therapeutic vulnerabilities in GBM.

In this study, we integrated bulk RNA sequencing (RNA-seq) and single-cell RNA sequencing (scRNA-seq) datasets to identify core genes associated with fatty acid metabolism in GBM and to characterize their expression patterns and functional roles within the tumor ecosystem. Our analysis identified biliverdin reductase A (BLVRA) as a key metabolic regulator and independent prognostic biomarker in GBM. Beyond its canonical role in the biliverdin–bilirubin redox cycle, BLVRA is known to participate in maintaining redox homeostasis, modulating intracellular signaling, and regulating lipid metabolism^10,11. Although previous studies have implicated dysregulated BLVRA expression in cancer cell proliferation and therapy resistance¹², its specific role in fatty acid metabolism and metabolic adaptation in GBM remains largely undefined. We hypothesized that BLVRA drives GBM progression by orchestrating fatty acid metabolic reprogramming. To test this, we conducted comprehensive in vitro experiments and untargeted metabolomic profiling to assess its role in lipid homeostasis, lipid droplet formation, oxidative stress regulation, and metabolic plasticity. Our findings provide novel mechanistic insights into BLVRA-mediated metabolic remodeling and highlight its potential as a therapeutic target in GBM.

Method

Data integration and differential expression

To comprehensively assess GBM-related gene expression patterns within the tumor microenvironment, both bulk RNA-seq and scRNA-seq data were integrated. Only GBM samples were retained for analysis; non-GBM samples were excluded on the basis of the clinical and pathological annotations provided in the datasets. The detailed sample characteristics and filtering criteria are provided in the supplementary data table. Differential expression analysis was performed via DESeq2 (v1.44.0)¹³, edgeR (v4.2.1)¹⁴, and limma (v3.60.5)¹⁵, with a threshold of a |log2 fold change (logFC)| > 2. To identify overlapping differentially expressed genes (DEGs), Venn diagrams (Venn plots) were constructed. In addition, bulk transcriptomic data from The Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression (GTEx) datasets were analyzed via limma with the criteria |logFC| > 1 and a false discovery rate (FDR) < 0.05. A curated List of 330 fatty acid metabolism-related genes was retrieved from the molecular signature database (MSigDB) and intersected with the DEGs identified from the analyses above for downstream modeling and functional characterization.

Prognostic model construction

To construct a robust and generalizable prognostic model for glioblastoma based on fatty acid metabolism-related genes, we systematically applied ten machine learning algorithms commonly used in high-dimensional survival analysis: least absolute shrinkage and selection operator (LASSO), ridge regression, elastic net (Enet), random survival forest (RSF), stepwise Cox regression (both forward and backward), CoxBoost, generalized boosted regression modeling (GBM), partial least squares regression for Cox (plsRcox), survival support vector machine (survival-SVM), and supervised principal component analysis (SuperPC). These algorithms were selected to encompass a broad spectrum of statistical learning approaches, including penalized regression, ensemble learning, boosting, and kernel-based classification. A total of 87 model combinations were generated by pairing different feature selection methods with survival modeling algorithms. Model training was performed via the TCGA_GBM dataset, and validation was conducted across five independent external cohorts (GSE83300, GSE74187, CGGA301, CGGA325, and CGGA693). Prognostic genes were initially identified through univariate Cox regression (p < 0.05) and served as candidate features for downstream modeling. For each model, the predictive performance was evaluated via Harrell’s concordance index (C-index) and time-dependent receiver operating characteristic (ROC) curves at 1, 3, and 5 years. The area under the curve (AUC) was used as an additional metric to assess prognostic discrimination. Clinical covariates such as age and sex were also incorporated into multivariate Cox analyses to evaluate the independence of the model-derived risk score. Among all combinations, the best-performing strategy was the combination of stepwise Cox regression for feature selection and survival-SVM with a radial basis function (RBF) kernel for model construction, which consistently yielded the highest average C-index across all cohorts. In this approach, stepwise Cox regression was used to identify the most informative prognostic genes on the basis of the Akaike information criterion (AIC), and the selected genes were then input into the survival-SVM algorithm to construct a classification model and calculate individualized prognostic risk scores. The final prognostic index (PI) for each sample was computed via the following formula derived from multivariate Cox regression coefficients:where denotes the Cox coefficient of gene j, and where ​ is the normalized expression of gene j in sample i. Patients were stratified into high- and low-risk groups on the basis of the median PI value. Survival differences between the two groups were evaluated via Kaplan–Meier (KM) analysis and the log-rank test.

Functional analysis of the risk signature

To further investigate the biological functions of the fatty acid metabolism–related gene prognostic index (FAMRGPI), we utilized the biomarker exploration for solid tumors (BEST) platform, a web application for comprehensive biomarker exploration of large-scale data in solid tumors¹⁶. This web-based tool enables comprehensive biomarker analysis across large-scale datasets derived from various solid tumor tissues, including glioblastoma. Using BEST, we evaluated the clinical associations, prognostic value, functional enrichment, and immune landscape related to the FAMRGPI in GBM patients.

Single-cell RNA-seq data processing

Single-cell RNA-seq data were processed via the Seurat package (v5.1.0)¹⁷ following rigorous quality control. The cells were retained if they expressed between 300 and 7,000 genes, had a UMI greater than 1,000, had a mitochondrial gene content less than 10%, or had an erythroid gene content less than 3%. Batch effects were corrected via the Harmony package. Dimensionality reduction was performed via principal component analysis (PCA), and the top 20 principal components were used to construct a shared nearest neighbor graph (FindNeighbors). Clustering was subsequently performed via the Louvain algorithm (FindClusters, resolution = 0.3). Cell type annotation was conducted via the SingleR package (v2.6.0) in combination with reference data from the Celldex package (v1.14.0). The marker genes for each cluster were identified via the FindAllMarkers function (log2FC > 0.25, adjusted p < 0.05). The annotation results were further refined through manual curation, incorporating information from the CellMarker 2.0 database¹⁸, GPT-based model analysis, and relevant literature. To assess cell-cell interaction dynamics, communication networks were constructed under different conditions via the CellChat package (v1.6.1). To investigate glioma-specific heterogeneity, the glioma cell population was extracted for secondary analysis. A new Seurat object was created, low-quality cells (fewer than 500 total transcripts or fewer than 200 detected genes) were removed, and the data were re-normalized. PCA and Louvain-based clustering (resolution = 0.2) were repeated, and the results were visualized using UMAP. Importantly, doublet removal was performed at this stage, after the glioma cell population was isolated. We originally considered global doublet filtering but found it suboptimal in highly heterogeneous datasets involving tumor–normal mixtures. This is consistent with prior studies suggesting that doublet detection algorithms may over-filter biologically relevant intermediate states when applied before subclustering in complex tumors^19–22. Therefore, doublets were selectively removed from the glioma subpopulation using the DoubletFinder package (v2.0.4), enhancing the specificity of downstream malignant cell identification while preserving transcriptional diversity¹⁹. To refine the classification of glioma cell populations beyond canonical marker-based annotation, copy number variation (CNV) analysis was performed using the inferCNV package (v1.20.0). Importantly, this analysis was conducted only on the glioma cell clusters that had been extracted following initial unsupervised clustering and annotation. Cells from peripheral-derived glioma clusters were used as internal reference populations to control for lineage specificity while minimizing batch effects. CNV profiles were computed for each individual cell within the glioma subpopulation to assess chromosomal instability. Cells were then stratified into high-CNV and low-CNV subgroups based on the extent of large-scale chromosomal alterations and their aggregate CNV scores (CNVscore), representing differential levels of genomic instability. This classification was not intended to dichotomize malignant and non-malignant cells in a binary fashion, but rather to reflect intra-tumoral heterogeneity within glioma cells. The resulting subgroups were used for downstream analyses, including the assessment of BLVRA expression patterns and cellular differentiation potential via CytoTRACE.

Gene set variation analysis (GSVA)

The functional states of different cell types were assessed via GSVA²³, which is based on Human KEGG and hallmark gene sets from the MSigDB database. The top 50 significantly enriched pathways were visualized. To assess the activity of specific genes in different GBM tissues, GSVA was performed on individual key genes, and the relationships between target genes and the activity of fatty acid metabolism-related pathways (KEGG/HALLMARK) were analyzed.

Immunohistochemistry (IHC)

Paraffin-embedded tissue samples from 17 glioma patients, as well as adjacent nontumor tissues, were collected for this study. The research protocol was approved by the Ethics Committee of the Fourth Affiliated Hospital of Harbin Medical University (Approval No. 2024-LUNLISHENCHA-110). Patients who had received preoperative treatments were excluded. The tissue sections were deparaffinized, subjected to antigen retrieval, blocked, and then incubated with a 1:200 dilution of anti-BLVRA antibody at 4 °C overnight. Signal detection was performed via streptavidin-peroxidase, followed by DAB staining and hematoxylin counterstaining. Images were captured via electron optical microscopy, and quantitative analysis was performed with ImageJ software.

Transfection

Human glioblastoma cell lines (U87, U251, and LN229) and the human astrocyte cell line (NHA) were purchased from Procell Life Science & Technology Co., Ltd. (Wuhan, China). The cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Gibco), supplemented with 10% fetal bovine serum (FBS, Gibco) and 1% penicillin-streptomycin (HyClone), and maintained in a Humidified incubator at 37 °C with 5% CO₂. When the cells reached 70% confluence, they were transfected with short hairpin RNA (shRNA) sequences targeting BLVRA: shBLVRA1 (GTGCCAAATGTAGGAGTGAAT) and shBLVRA2 (CAGCGTTCCTGAACCTGATTG).

Quantitative real‑time PCR (qRT-PCR)

Total RNA was extracted via TRIzol reagent and first-strand cDNA was synthesized via the TOYOBO FSQ-101 kit according to the standard protocol. qRT-PCR was performed via SYBR qPCR mix and the FQD96A real-time PCR detection system. Each experiment was performed in triplicate, and the mRNA expression of genes was analyzed via the ­2^−ΔΔCt method. The following primers were used: 5’-CTGATTGGCTTCGTGTCGAGA-3’ and 5’-CATGGGGTATTCCACAAGGAC-3’ for BLVRA, and 5’-GACCTCCCTGAACACCGCAAA-3’ and 5’-ATAGCCGTCATCACAACCGG − 3’ for CPT1A. The gene expression data were normalized to that of β-actin, and the results are presented as the means ± SDs from three independent biological replicates.

Western blotting

The cells were lysed with RIPA buffer supplemented with protease and phosphatase inhibitors, and the protein concentrations were determined via a BCA assay kit. The samples were subsequently boiled at 100 °C for 3 min. Equal amounts of protein (20 µg) were subsequently subjected to sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto polyvinylidene difluoride (PVDF) membranes (Sigma-Aldrich, Germany). The membranes were blocked with 3% nonfat dry milk at room temperature for 1 h, followed by incubation with primary antibodies at 4 °C overnight. Each target protein was detected on a separate membrane to ensure specificity and avoid signal interference. After target protein detection and chemiluminescent imaging, membranes were stripped once using a commercial stripping buffer (Epizyme Biotech, China) and re-probed with antibodies against internal controls (β-actin, α-tubulin, or GAPDH). No membrane was subjected to multiple rounds of stripping and re-probing. Protein bands were quantified via ImageJ software and normalized to the internal controls. The quantitative data are presented as the means ± SDs from three independent experiments. All the antibody details are provided in Table S1.

Cellular function experiment

Cell proliferation was assessed via a CCK-8 assay kit (Beyotime, China). U251 and LN229 cells from both the BLVRA knockdown and control groups were seeded into 96-well plates at a density of 3 × 10³ cells per well and cultured for 72 h. At each indicated time point, 10% of CCK-8 solution was added to each well, and the absorbance was measured at 450 nm via a microplate reader. For colony formation assays, cells were plated in 6-well plates at a low density (approximately 200–300 cells per well) and cultured for 10–14 days. Colonies were fixed with 4% paraformaldehyde, stained with 0.1% crystal violet, and visualized under an optical microscope (Olympus IX71). The number of colonies was quantified using ImageJ software.

Cellular phenotype assays

Lipid droplets were detected via a BODIPY 493/503 staining kit. The ROS levels were measured via an ROS assay kit. Nuclear DNA damage was assessed via γ-H2AX immunofluorescence. All reagents were obtained from Beyotime, China. The experiments were performed according to the manufacturer’s instructions, and images were captured via a fluorescence microscope (Olympus IX71).

Nontargeted metabolomics

Metabolites were extracted from U251 cells in the BLVRA-knockdown group and the corresponding parental U251 control group via a methanol-water mixture (2:2:1, v/v/v). After 100 mg of glass beads were added, they were rapidly frozen in liquid nitrogen and subjected to grinding (60 Hz, 2 min) three times. The samples were subsequently centrifuged at 12,000 rpm at 4℃ for 10 min, after which the supernatants were collected and concentrated to dryness. The metabolites were redissolved in 300 µl of 0.1% formic acid-acetonitrile (1:9, v/v), spiked with a 4 ppm 2-chloro-L-phenylalanine internal standard, filtered through a 0.22 μm membrane, and then used for LC-MS analysis. Liquid chromatography separation was performed via a Thermos Vanquish UPLC system (Thermos Fisher Scientific, USA) equipped with an ACQUITY UPLC^® HSS T3 column (2.1 × 100 mm, 1.8 μm, Waters, USA). The column temperature was set to 40 °C, the flow rate was 0.3 mL/min, and the injection volume was 2 µL. Data were collected in both positive and negative ion modes, with the mobile phase consisting of 0.1% formic acid in water and 0.1% formic acid in acetonitrile. Mass spectrometry analysis was performed via a Thermos Orbitrap Exploris 120 mass spectrometer (Thermos Fisher Scientific, USA) equipped with an electrospray ionization (ESI) source. The mass scanning range was m/z 100–1000, with a resolution of 60,000, HCD collision energy of 30%, secondary scan resolution of 15,000, and dynamic exclusion to remove irrelevant MS/MS signals. Data processing was performed via Proteowizard (v3.0.8789) to convert the raw data to mzXML format and XCMS (v3.12.0) for peak detection, alignment, and normalization^24,25. Metabolite identification is based on databases such as HMDB, KEGG, LipidMaps, MassBank and mzCloud^26–30, which are used to match the parent ion m/z, mass deviation (ppm < 30), and secondary fragment spectra. All the experiments were conducted with six independent biological replicates per group to ensure reproducibility and statistical robustness.

Statistical analysis

Differences between two groups were analyzed via two-tailed Student’s t-test, whereas comparisons among multiple groups were performed via one-way analysis of variance (ANOVA), followed by post-hoc tests such as Tukey’s HSD test or Dunnett’s test. Kaplan-Meier survival analysis was conducted via the log-rank test. For metabolomics data analysis, two different multivariate statistical analysis models, unsupervised and supervised, were applied to discriminate the groups (PCA; OPLS-DA) via the R ropls (v1.22.0) package³¹. The statistical significance of the P value was determined by statistical tests between groups. Finally, by combining the P value, VIP (variable importance in projection from OPLS-DA), and FC (fold change between groups), biomarker metabolites were screened. By default, metabolites with a P-value < 0.05 and VIP > 1 were considered to have significant differential expression. Differentially abundant metabolites were subjected to pathway analysis via MetaboAnalyst³², which integrates pathway enrichment analysis with pathway topology analysis. The metabolites identified via metabolomics were then mapped to the KEGG pathway for biological interpretation of higher-level systemic functions. The metabolites and corresponding pathways were visualized via the KEGG Mapper tool. All the statistical analyses were performed via R (v4.4.1) or GraphPad Prism (v9.0), and a P-value of less than 0.05 was considered statistically significant. Statistical significance levels are denoted as follows: * for p < 0.05, ** for p < 0.01, *** for p < 0.001, and **** for p < 0.0001.

Results

Development and validation of the FAMRGPI on the basis of an integrated computational framework

To investigate the expression patterns and clinical significance of fatty acid metabolism-related genes in GBM, we analyzed transcriptome data from 160 GBM patients. Differential expression analysis was performed via edgeR, DESeq2, and Limma, and 19,138 differentially expressed genes (DEGs) were identified via Wayne plot crossover analysis (Fig. 1A, Table S2). Given the Limited number of control samples in the TCGA database, we integrated transcriptome data from GTEx and applied the sva package for batch effect correction. This refinement yielded a secondary differential analysis, identifying 6,701 DEGs (3,111 downregulated and 3,590 upregulated) (Fig. 1B, Table S3). Among these DEGs, 330 fatty acid metabolism-related genes were extracted from the MSigDB and overlapped with the DEGs, resulting in 109 DEGs associated with fatty acid metabolism (Fig. 1C, Table S4-S5). Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses revealed that these genes were significantly involved in metabolic pathways such as fatty acid metabolism, the tricarboxylic acid (TCA) cycle, carbon metabolism, and the PPAR signaling pathway. In addition, these genes participate in key biological processes, including mitochondrial matrix organization, peroxisomal function, and lipoyl-coenzyme A binding (Fig. S1). Via machine learning algorithms, we constructed the FAMRGPI by testing 87 different model combinations. The C-index was calculated for both the training and validation cohorts across all the models (Fig. 1D, Table S6). Among the 87 models, the StepCox + survivalSVM model demonstrated the highest predictive performance, with an average C-index of 0.617. Given its robust predictive ability and translational potential, we selected this model for further analysis. On the basis of this model, we identified nine key genes (ACADVL, ACOT7, ALDH1B1, APEX1, BLVRA, G0S2, IL4I1, MIF, and SLC25A20) (Table S7). Univariate Cox regression analysis revealed that BLVRA, G0S2, IL4I1, and SLC25A20 were significantly associated with patient prognosis (Fig. 1E). Kaplan-Meier survival analysis further confirmed that high expression of these genes was correlated with poor prognosis (Fig. 1F). A risk score was calculated for each patient on the basis of these key genes, and patients were classified into high-risk and low-risk groups according to the median score. Survival analysis revealed that as the risk score increased, patient survival rates progressively decreased (Fig. 1G). The survival curves for the training cohort demonstrated a strong association between high-risk classification and poor prognosis, a finding that was validated in the external cohort (Fig. 1H). The AUC values for the model’s 1-, 3-, and 5-year survival predictions were 0.729, 0.845, and 0.923, respectively. Additionally, the AUC value for the risk score (0.794) was significantly greater than those for age (AUC = 0.593) and sex (AUC = 0.496) (Fig. 1I, J, K), indicating that the risk score is a strong independent prognostic factor capable of effectively predicting patient survival.

Fig. 1.

Construction and validation of the FAMRGPI through a machine learning-based integrative program. (A) Venn diagram showing the intersection of differentially expressed genes (DEGs) identified through edgeR, DESeq2, and Limma. A total of 19,138 DEGs were identified. (B) Volcano plot illustrating secondary differential expression analysis integrating the TCGA and GTEX datasets, revealing 6,701 DEGs (3,111 downregulated, 3,590 upregulated) after batch effect correction. (C) Venn diagram showing the overlap between fatty acid metabolism-related genes from the MSigDB and the 6,701 DEGs, yielding 109 candidate genes. (D) Machine learning-based model selection for the fatty acid metabolism-related gene score (FARMGs) model, evaluated via 87 machine learning algorithms. The StepCox + survival SVM model was identified as the optimal predictive model. (E) Univariate Cox regression analysis identifying BLVRA, G0S2, IL4I1, and SLC25A20 as significant prognostic factors. (F) Kaplan-Meier survival curves demonstrating that higher expression of BLVRA, G0S2, IL4I1, and SLC25A20 correlates with poorer prognosis. (G) Distributions of survival time, survival status, and the six genes included in the model in the high- and low-risk groups. (H) Kaplan-Meier survival analysis comparing overall survival between the high-risk and low-risk groups stratified by the risk score. Patients in the high-risk group presented significantly lower survival rates than those in the low-risk group did. (I) Time-dependent and (J) clinical characteristic-related ROC curves. (K) Association of age, sex, and risk score with event outcomes: Cox proportional hazards analysis.

Exploration of FAMRGPI biomarkers for the treatment of solid GBM

To assess the clinical relevance and biological functions of the FAMRGPI in glioblastoma, we performed a systematic biomarker analysis via the BEST web application, which integrates transcriptomic and clinical data from solid tumor cohorts. The analysis revealed that FAMRGPI expression was significantly elevated in glioblastoma tissues compared with normal brain tissues (Fig. 2A). Stratification by molecular subtype demonstrated marked heterogeneity, with the mesenchymal subtype exhibiting the highest FAMRGPI expression. Furthermore, elevated FAMRGPI were observed in tumors characterized by wild-type IDH, nonmethylated MGMT promoter, and EGFR-amplified phenotypes (Fig. 2B). Kaplan-Meier survival analysis and multivariate Cox regression revealed that high FAMRGPI expression was significantly associated with shorter overall survival (OS) and disease-specific survival (DSS) in multiple independent cohorts (Fig. 2C-D). The DSS was included to specifically assess mortality attributable to GBM, thereby reducing potential confounding from noncancer-related causes of death. Subgroup analysis revealed that the prognostic impact of FAMRGPI was particularly pronounced in male patients over 60 years of age (Fig. S2), suggesting that the index may capture age- and sex-specific risk factors. To elucidate the underlying biological mechanisms, we performed gene ontology (GO), hallmark, and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses for the genes in the FAMRGPI. GO analysis revealed enrichment in immune regulatory processes such as adaptive immune responses, antigen presentation, and immune cell migration, as well as in synaptic signaling pathways (Fig. S3). GSEA and KEGG analyses revealed that the FAMRGPI genes were enriched in multiple hallmark oncogenic and immunosuppressive pathways, including interferon-γ/α responses, IL6-JAK-STAT3 signaling, complement and coagulation cascades, glycolysis, amino acid metabolism, epithelial-mesenchymal transition (EMT), extracellular matrix (ECM) receptor interaction, hypoxia, and glutathione metabolism (Fig. 2E). These findings suggest that the FAMRGPI may promote GBM progression by facilitating immune evasion, metabolic reprogramming, oxidative stress resistance, and EMT-driven invasion. We next assessed the association between FAMRGPI expression and the tumor immune microenvironment (TME). Across multiple independent GBM datasets, FAMRGPI expression was positively correlated with immune infiltration metrics, including ImmuneScore, StromalScore, and ESTIMATEScore (Fig. 2F, Fig. S4). Notably, in the GSE43378 dataset, the FAMRGPI showed the strongest correlation with immune and stromal components (r = 0.71), suggesting its involvement in TME remodeling. Consistent with these findings, high FAMRGPI expression was associated with increased M2 macrophage infiltration and the activation of immunosuppressive pathways, particularly JAK-STAT signaling and the complement cascade (Fig. 2G). To identify the key drivers within the FAMRGPI, we performed multivariate Cox regression analysis and found that BLVRA (HR = 1.411, 95% CI: 1.068–1.863, P = 0.015) and IL4I1 (HR = 1.196, 95% CI: 1.040–1.375, P = 0.012) remained independent prognostic factors after adjusting for clinical covariates (Fig. 2H). We next explored the potential metabolic roles of these genes by correlating their expression with fatty acid metabolic pathway activity. Pearson correlation analysis revealed that BLVRA expression was significantly and positively correlated with fatty acid metabolism activity, whereas IL4I1 showed no such association (Fig. 2I, Fig. S5). This metabolic specificity, combined with its prognostic impact, led us to designate BLVRA as the core candidate gene for subsequent functional investigation. Expression pattern analysis further confirmed that BLVRA was highly expressed in the mesenchymal subtype (Fig. S6), a clinically aggressive phenotype, suggesting a subtype-specific relevance. Together, these results suggest that the FAMRGPI represents a prognostic marker in GBM, that functions through immunosuppressive remodeling and metabolic reprogramming. Among its components, BLVRA has emerged as a key regulator potentially linking fatty acid metabolism with mesenchymal transition and poor prognosis.

Fig. 2.

Exploration of FAMRGPI biomarkers for solid GBM tumors. (A) Differential expression of the FAMRGPI in GBM tissues across multiple datasets. (B) Expression characteristics of the FAMRGPI in different molecular subtypes of GBM. (C) Univariate Cox regression and (D) Kaplan–Meier survival analysis of the FAMRGPI across three survival endpoints in independent GBM cohorts, only nontreatment samples were included in the analysis. (E) Functional enrichment analysis of the FAMRGPI. (F) Correlation analysis of the FAMRGPI with GBM immune microenvironment characteristics and (G) immune cell infiltration. (H) Multivariate Cox regression analysis confirming BLVRA (HR = 1.411, P = 0.015) and IL4I1 (HR = 1.196, P = 0.012) as independent prognostic factors, alongside age. (I)Pearson correlation analysis showing a significant positive correlation between BLVRA expression and fatty acid metabolism pathway activity.

Single-cell atlas reveals the expression and functional role of the BLVRA in GBM

To investigate the expression patterns and molecular functions of BLVRA in glioblastoma (GBM), we analyzed single-cell RNA sequencing (scRNA-seq) data from two datasets: GSE214966, which includes four unsorted GBM tumor samples, and GSE162631, which contains CD31⁺ endothelial cells from four GBM tumors and four matched non-malignant brain tissues. Following quality control, we obtained 52,004 cells from tumor samples and 43,518 cells from peripheral tissues (Fig. 3A). By performing highly variable gene analysis and applying batch effect correction via the Harmony algorithm, we identified 13 major cell clusters (Fig. 3B). Cell annotation (Fig. 3C) classified these clusters into eight primary cell types: macrophages, endothelial cells, glioma cells, mesenchymal (MES)cells, neutrophils, T cells, oligodendrocytes, and monocytes (Fig. 3D). Notably, monocytes and endothelial cells are predominantly derived from peripheral tissues, whereas the remaining cell types are primarily localized within tumor tissue, highlighting the distinct compositional differences between the tumor and its surrounding microenvironment. On the basis of single-cell transcriptomic data, we analyzed the expression patterns and heterogeneity of BLVRA across different cell types. The results demonstrated that BLVRA was highly expressed in glioma cell subpopulations within GBM patients but exhibited heterogeneous expression in other cell types (Fig. 3E). Wilcoxon rank-sum tests were performed to assess the differential expression of BLVRA between the tumor and control groups within each cell type. Significant differences were observed in endothelial cells, mesenchymal (MES)cells, glioma cells, macrophages, and monocytes (p < 0.0001) (Fig. 3F), suggesting a pivotal role for BLVRA in GBM heterogeneity. Furthermore, we conducted GSVA via KEGG and HALLMARK gene sets to evaluate metabolic and immune-related pathways across glioblastoma cell types and identified the top 50 enriched pathways. The results revealed that BLVRA⁺ glioma cells exhibited marked activation of metabolic pathways, including fatty acid metabolism, glycolysis, and the TCA cycle. In contrast, immune-related pathway scores were significantly reduced, indicating that BLVRA⁺ glioma cells may facilitate tumor immune evasion by suppressing antigen presentation, interferon signaling, and NK cell-mediated cytotoxicity (Fig. 3G). These findings suggest that high BLVRA expression is associated with poor prognosis in glioblastoma patients and is characterized by enhanced metabolic reprogramming and pronounced immune evasion. CellChat analysis revealed notable intercellular communication between BLVRA⁺ glioma cells and macrophages. Specifically, enriched ligand–receptor interactions between these populations (Fig. 3H, I) suggest a potential role for BLVRA⁺ glioma cells in modulating macrophage-mediated immune responses within the TME.

Fig. 3.

Single-cell analysis reveals BLVRA expression patterns and their role in glioblastoma metabolism and immune modulation. (A) UMAP plot showing single-cell transcriptomic data from four GBM tumor samples and eight peripheral tissue samples after quality control. Cells are color-coded by sample origin, including 52,004 tumor-derived cells and 43,518 peripheral-derived cells. (B) Unsupervised clustering identified 13 major cell clusters, classified via the Harmony algorithm. (C) Heatmap illustrates cell type annotation on the basis of marker gene expression. (D) Distribution of eight primary cell types, including macrophages, endothelial cells, glioma cells, mesenchymal (MES)cells, neutrophils, T cells, oligodendrocytes, and monocytes. Monocytes and endothelial cells are derived primarily from peripheral tissues, whereas glioma cells are enriched in tumor samples. (E) Bubble plot depicting BLVRA expression levels across different cell types. BLVRA was highly expressed in glioma cells but exhibited heterogeneous expression in other cell populations. (F) Wilcoxon rank-sum test results showing significant differential expression of BLVRA between the tumor and control groups in glioma cells, macrophages, monocytes, endothelial cells, and mesenchymal (MES)cells (p < 0.0001). (G) Gene set variation analysis (GSVA) of the KEGG and HALLMARK gene sets in BLVRA⁺ glioma cells, highlighting enrichment in fatty acid metabolism, glycolysis, and the TCA cycle pathways, whereas immune-related pathways, including antigen presentation and NK cell cytotoxicity, were suppressed. (H) CellChat analysis illustrating strong ligand-receptor interactions between BLVRA⁺ glioma cells and macrophages, suggesting their role in immune modulation. (I) Quantitative assessment of BLVRA-mediated cell-cell communication, highlighting enhanced signaling between glioma cells and macrophages, suggesting a role in tumor immune evasion.

To further investigate the relationship between BLVRA expression, intratumoral heterogeneity, and cellular differentiation potential in GBM, we performed a refined analysis focusing on glioma cell subpopulations. Initial clustering of the integrated single-cell transcriptomic data revealed that glioma cells constituted the predominant malignant population in GBM, while macrophages were the most abundant immune cell subset (Fig. 4A). Reclustering of the glioma compartment identified ten transcriptionally distinct subpopulations, reflecting pronounced heterogeneity within the tumor cell lineage (Fig. 4B). To dissect the malignant potential of these subpopulations, we applied inferCNV analysis to quantify chromosomal instability and stratified glioma cells into high-CNV and low-CNV groups based on their CNV burden (Fig. 4C–D, Fig. S7). UMAP visualization further illustrated the spatial organization of these subgroups (Fig. 4E). Notably, BLVRA expression was significantly elevated in the high-CNV subgroup, suggesting that BLVRA may contribute to enhanced metabolic plasticity in cells experiencing elevated genomic stress, thereby supporting oxidative stress resistance and promoting proliferative capacity (Fig. 4F). Moreover, CytoTRACE analysis revealed that cells in the high-CNV subgroup exhibited greater differentiation potential, indicative of a less differentiated and more plastic state that may be associated with increased malignancy and tumor-initiating potential (Fig. 4G). Together, these findings implicate BLVRA as a potential driver of metabolic adaptation and malignant plasticity in genomically unstable glioma cells.

Fig. 4.

BLVRA expression is correlated with genomic instability and differentiation potential in glioblastoma. (A) Clustering analysis of GBM single-cell transcriptomes identified glioma cells as the predominant population, with macrophages constituting the major immune subset. The x-axis labels represent abbreviated sample identifiers from the GSE162631 and GSE214966 datasets. (B) Reclustering of the glioma cell population revealed ten transcriptionally distinct subclusters, indicating marked intratumoral heterogeneity. (C) CNV heatmap generated via inferCNV, on the basis of comparisons with peripheral-derived glioma-like reference cells, revealed variable genomic instability among subclusters. (D) Distribution of CNV scores across glioma cells, allowing stratification into high-CNV and low-CNV subgroups on the basis of inferred chromosomal aberration burden. (E) UMAP visualization of annotated glioma subclusters after CNV-based stratification, highlighting the spatial distribution of high-CNV and low-CNV populations across the malignant landscape. (F) BLVRA was significantly upregulated in the high-CNV subgroup, suggesting a potential role in enhancing metabolic adaptability under conditions of elevated genomic instability. (G) CytoTRACE analysis demonstrated that high-CNV tumor cells present greater differentiation potential, reflecting a less differentiated and potentially more aggressive phenotype.

Knockdown of BLVRA inhibits GBM cell growth

To explore the role of BLVRA in GBM, we conducted immunohistochemical analysis of clinical GBM tissue samples. The analysis revealed a marked upregulation of BLVRA expression in GBM tissues (Fig. 5A, Fig. S8). qPCR and Western blot analyses further confirmed that BLVRA expression was substantially elevated in GBM cell lines compared with normal astrocytes (Fig. 5B-C, Fig. S9). These findings imply that the BLVRA is linked to the proliferative and invasive characteristics of GBM. To further validate the role of BLVRA in GBM cell proliferation and invasion, we utilized lentivirus-mediated gene knockdown to effectively suppress BLVRA expression in GBM cell lines. The knockdown efficiency was verified via qPCR and Western blotting (p < 0.001) (Fig. 5D, Fig. S10). Cell proliferation assays revealed that BLVRA knockdown led to a significant reduction in cell proliferation (p < 0.001) (Fig. 5E). Moreover, BLVRA knockdown significantly reduced the colony formation rate of U251 and LN229 cells, reinforcing the hypothesis that BLVRA deficiency suppresses GBM cell proliferation (p < 0.01) (Fig. 5F).

Fig. 5.

BLVRA promotes GBM cell proliferation, invasion, and temozolomide resistance. (A) Immunohistochemical analysis of clinical GBM tissue samples revealed marked upregulation of BLVRA protein expression. (B, C) qPCR and Western blot analyses confirmed significantly higher BLVRA expression in GBM cell lines than in normal astrocytes. (D) Efficient BLVRA knockdown in GBM cell lines was validated by Western blot analysis (p < 0.001). (E) Cell proliferation assays indicated that BLVRA knockdown significantly inhibited GBM cell proliferation (p < 0.001). (F) Colony formation assays demonstrated that BLVRA knockdown significantly reduced the proliferative capacity of U251 and LN229 cells (p < 0.01).

Inhibition of BLVRA disrupts fatty acid metabolism and lipid homeostasis in GBM cells

To elucidate the role of BLVRA in lipid homeostasis in GBM cells, we first examined the impact of BLVRA knockdown on lipid droplet (LD) formation. BODIPY 493/503 staining revealed a significant reduction in the number of LDs in BLVRA-deficient GBM cells (Fig. 6A, Fig. S11), suggesting that BLVRA suppression impairs fatty acid storage capacity, thereby limiting the ability of cells to buffer dynamic lipid demands. To further characterize the metabolic alterations induced by BLVRA depletion, we conducted untargeted metabolomics analysis. Principal component analysis (PCA) clearly revealed separation between the knockdown and control groups in both positive and negative ion modes (Fig. 6B), indicating substantial metabolic reprogramming upon BLVRA loss. The R²X (cum) values for the positive and negative ion modes were 0.542 and 0.528, respectively, confirming the model’s strong explanatory power in distinguishing metabolic differences between the two groups. Orthogonal partial least squares discriminant analysis (OPLS-DA) identified significantly altered metabolites (VIP > 1), which were further visualized via S-plots (Fig. 6C). Through primary and secondary mass spectrometry analysis, we identified 437 significantly dysregulated metabolites (p < 0.05, VIP > 1), comprising 233 upregulated and 204 downregulated compounds (Fig. 6D, Table S8). These metabolic changes spanned multiple categories, including lipid, amino acid, and energy metabolism, underscoring BLVRA’s pivotal role in metabolic regulation in GBM. In terms of fatty acid metabolism, BLVRA knockdown resulted in a significant decrease in carnitine levels (p < 0.0001) (Fig. 6E), indicating impaired fatty acid oxidation (FAO) due to reduced transport capacity. Consistent with these findings, the mRNA and protein levels of CPT1A, a key rate-limiting enzyme of FAO, were significantly increased (Fig. 6F, Fig. S12), suggesting that cells attempt to compensate for FAO dysfunction through adaptive responses. However, excessive FAO activation can exacerbate oxidative stress and induce cellular damage in GBM³³. Additionally, coenzyme A levels were significantly elevated (p < 0.05) (Fig. 6G), indicating the accumulation of metabolic intermediates due to FAO blockade, which may contribute to metabolic toxicity. BLVRA depletion also led to significant changes in the levels of key lipid species associated with membrane structure. Specifically, the phosphatidylcholine (PC) level was markedly reduced (p < 0.05) (Fig. 6H), whereas the phosphatidylethanolamine (PE) and sphingomyelin levels were significantly increased (p < 0.05) (Fig. 6I). These findings suggest potential alterations in membrane lipid balance and may reflect compensatory lipid remodeling aimed at maintaining membrane integrity. However, definitive conclusions regarding membrane composition await further membrane-targeted validation. In the context of energy metabolism, KEGG pathway analysis revealed that BLVRA deficiency significantly suppressed the TCA cycle, as evidenced by reduced levels of key metabolites, including oxaloacetate, malate, succinate, and fumarate (Fig. 6J, K).

Fig. 6.

BLVRA knockdown disrupts fatty acid metabolism and lipid homeostasis in GBM cells. (A) BODIPY 493/503 staining revealed a significant reduction in lipid droplet (LD) formation following BLVRA knockdown, indicating impaired fatty acid storage capacity. (B) Principal component analysis (PCA) clearly shows metabolic separation between BLVRA-knockdown and control GBM cells in both positive and negative ion modes. (C) Orthogonal partial least squares discriminant analysis (OPLS-DA) identified significantly altered metabolites (VIP > 1), as visualized in S-plots. (D) Untargeted metabolomics analysis revealed 437 significantly dysregulated metabolites (p < 0.05, VIP > 1), including 233 upregulated and 204 downregulated compounds. (E) BLVRA knockdown led to a significant reduction in carnitine levels (p < 0.0001), suggesting impaired fatty acid oxidation (FAO). (F) The mRNA (a) and protein (b) levels of CPT1A, a key FAO enzyme, were significantly increased in response to BLVRA knockdown. (G) Coenzyme A levels are elevated (p < 0.05), indicating an accumulation of metabolic intermediates due to FAO blockade. (H, I) Lipidomic analysis revealed a significant reduction in phosphatidylcholine (PC) levels (p < 0.05) and increases in phosphatidylethanolamine (PE) and sphingomyelin levels (p < 0.05), suggesting compensatory lipid remodeling in the cell membrane. (J, K) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis demonstrated that BLVRA deficiency suppresses the tricarboxylic acid (TCA) cycle, as evidenced by decreased levels of key metabolites, including oxaloacetate, malate, succinate, and fumarate, potentially impairing GBM cell proliferation and survival. The data are presented as the means ± SDs from six independent biological replicates.

Inhibiting BLVRA leads to severe oxidative stress and DNA damage that kills GBM cells

To investigate the mechanisms potentially contributing to reduced cell viability following BLVRA inhibition, we analyzed metabolic alterations associated with lipid homeostasis and oxidative stress. Metabolomic analysis revealed that BLVRA knockdown led to a significant accumulation of free fatty acids and a marked reduction in triglyceride (TG) levels (Fig. 7A), confirming impaired fatty acid storage. Additionally, increased levels of lipid peroxidation products and a decreased GSH/GSSG ratio indicated elevated oxidative stress and compromised antioxidant defenses (Fig. 7B). Fluorescence imaging confirmed significantly increased ROS levels in BLVRA-deficient cells (Fig. 7C, Fig. S13), while Western blot analysis showed a concomitant downregulation of Nrf2 (Fig. 7D), further suggesting impaired redox homeostasis. These data collectively indicate that BLVRA depletion disrupts antioxidant capacity, thereby promoting ROS accumulation and lipid peroxidation. To assess genomic stability, we performed immunofluorescence analysis of the DNA damage marker γ-H2AX, which showed a notable increase in BLVRA-deficient cells (Fig. 7E, Fig. S14), suggesting elevated DNA damage under oxidative stress conditions. To further determine the impact of BLVRA depletion on lipid-induced cytotoxicity, we treated cells with exogenous palmitic acid (C16:0). While control cells remained viable at concentrations up to 40 µM, BLVRA-knockdown cells exhibited significantly reduced viability in a dose-dependent manner (Fig. 7F, G). These findings suggest that BLVRA deficiency increases susceptibility to fatty acid–induced lipotoxicity by impairing lipid storage and redox buffering capacity. To elucidate the molecular consequences of BLVRA knockdown on cell fate, we next examined the expression of tumor suppressor and apoptosis-related proteins. Western blot analysis revealed upregulation of p53, p21, and cleaved caspase-3 (Fig. 7H), indicating activation of DNA damage response and apoptotic pathways. In parallel, the expression of key regulators of cell cycle progression—including CDK2, CDK4, CDK6, Cyclin D1, and C-MYC—was significantly reduced following BLVRA knockdown (Fig. 7I, Fig. S15), suggesting impaired proliferative capacity due to cell cycle arrest. Together, these results suggest that BLVRA deficiency suppresses tumor growth by inducing oxidative stress, DNA damage, and apoptosis while concurrently impairing cell cycle progression, thereby inhibiting GBM cell proliferation.

Fig. 7.

BLVRA inhibition induces severe oxidative stress and DNA damage and promotes cell cycle arrest and apoptosis in GBM cells. (A) Based on previously acquired untargeted metabolomics data (see Fig. 6), reanalysis revealed a significant increase in free fatty acid levels and a reduction in triglyceride (TG) levels following BLVRA knockdown, indicating impaired fatty acid storage capacity. (B) Lipid peroxidation assays revealed a marked increase in lipid peroxidation products and a reduced GSH/GSSG ratio, reflecting increased oxidative stress and compromised antioxidant defenses. (C) Fluorescence imaging of U251 cells revealed a significant increase in intracellular reactive oxygen species (ROS) levels (green fluorescence) upon BLVRA knockdown. (D) Western blot analysis of U251 cells revealed a decrease in NRF2 protein expression, confirming the suppression of antioxidant defense pathways. (E) Immunofluorescence staining of γ-H2AX revealed an increase in DNA damage following BLVRA knockdown. (F, G) Exogenous palmitic acid (C16:0) treatment led to a dose-dependent decrease in viability in BLVRA-knockdown U251 cells but not in control cells, indicating increased sensitivity to fatty acid-induced toxicity. (H) Western blot analysis revealed that BLVRA knockdown significantly increased the expression of tumor suppressor proteins p53 and p21, and it also demonstrated caspase-3 activation by showing both pro-caspase-3 and cleaved caspase-3 levels, indicating the activation of DNA damage and apoptotic pathways. (I) Expression of key cell cycle regulators—including CDK2, CDK4, CDK6, Cyclin D1, and c-MYC—was markedly downregulated following BLVRA knockdown, suggesting impaired cell cycle progression and proliferative arrest.

Discussion

GBM is a highly malignant tumor characterized by profound metabolic reprogramming, in which dysregulated fatty acid metabolism plays a pivotal role in tumor progression⁹. Our study is the first to establish BLVRA as a key regulator of fatty acid metabolism and oxidative stress adaptation in GBM. Immunohistochemical analysis revealed significant upregulation of BLVRA in GBM tissues compared with adjacent normal tissues and low-grade gliomas. Using a multilayered experimental approach, we systematically investigated how BLVRA modulates lipid homeostasis, oxidative stress, DNA damage, and cell cycle regulation, thereby sustaining GBM growth.

Bioinformatics analysis further revealed that high BLVRA expression is correlated with poor prognosis in GBM patients, highlighting its potential as an independent prognostic factor. By integrating fatty acid metabolism-related gene sets with GBM transcriptomic data, we identified BLVRA as a key metabolic regulator and confirmed its prognostic significance. Notably, we analyzed its expression across the three classical GBM molecular subtypes defined by the Knn classification: classical, mesenchymal, proneural³⁴. Since our key findings focus on the mesenchymal subtype, which is unaffected by this concern, the inclusion of the neural subtype does not compromise our conclusions. The results revealed that BLVRA expression was markedly elevated in the mesenchymal subtype, which is known for its association with increased invasiveness, immune evasion, and resistance to therapy. This enrichment suggests that the BLVRA may contribute to the aggressive and immunosuppressive phenotype characteristic of mesenchymal GBM. These findings underscore the potential of BLVRA as a subtype-specific therapeutic target, particularly in patients with mesenchymal features.

Single-cell analysis revealed that BLVRA expression is closely associated with CNV and the cellular differentiation potential, suggesting its critical role in tumor cell heterogeneity and metabolic adaptability. Functionally, BLVRA knockdown significantly reduced the proliferative and colony-forming abilities of GBM cells, underscoring its essential role in tumor growth. Mass spectrometry analysis demonstrated that BLVRA knockdown profoundly disrupted fatty acid metabolism, leading to decreased carnitine levels, increased coenzyme A, and the accumulation of lipid peroxidation products, collectively indicating impairments in fatty acid transport and oxidation. These findings suggest that BLVRA deficiency disrupts the fatty acid metabolic balance and ultimately inhibits GBM cell proliferation. Furthermore, we observed a marked increase in free fatty acids accompanied by a reduction in triglyceride levels, indicating impaired lipid storage capacity in BLVRA-deficient cells.

Importantly, BLVRA knockdown induced severe oxidative stress, as evidenced by the accumulation of ROS and increased lipid peroxidation, coupled with a reduced GSH/GSSG ratio. These changes were associated with downregulation of NRF2, suggesting compromised antioxidant defenses. Elevated oxidative stress is known to induce DNA damage^35,36. Consistently, BLVRA-deficient cells exhibited higher levels of γ-H2AX, indicating increased genomic instability. Excessive ROS can also induce apoptosis through multiple pathways³⁷. wild type p53 activation in response to ROS stress leads to cell cycle arrest or apoptosis, whereas mutant p53 upregulates Nrf2, enhancing ROS detoxification as a survival mechanism³⁸. To further elucidate the downstream effects of oxidative stress, we analyzed key regulatory proteins and found that BLVRA knockdown significantly upregulated the expression of tumor suppressor proteins (p53, p21) and cleaved caspase 3 (Fig. 7H), while simultaneously downregulating cell cycle-related proteins including CDK6, CDK4, CDK2, Cyclin D1, and C-MYC (Fig. 7I). Interestingly, BLVRA knockdown markedly increased ROS production through dysregulated fatty acid metabolism, leading to enhanced DNA damage (p53 and p21 upregulation). This cascade further promoted caspase-3 activation, as evidenced by the stronger cleavage observed in shBLVRA cells. These results provide molecular evidence that BLVRA depletion triggers oxidative stress–mediated cell cycle arrest and apoptosis.

Interestingly, BLVRA not only exerts direct antioxidant effects by scavenging ROS but also may indirectly regulate oxidative stress by modulating fatty acid metabolism. Notably, BLVRA knockdown significantly decreased carnitine levels, thereby limiting fatty acid transport into mitochondria for β-oxidation. Concomitantly, the expression of CPT1A, a key enzyme in fatty acid oxidation, was significantly increased. Excessive FAO activation can promote lipid peroxidation, further exacerbating oxidative stress. Moreover, dysregulated fatty acid metabolism may impair mitochondrial function, leading to inefficient electron transport chain activity and increased ROS leakage. This vicious cycle of oxidative stress may ultimately compromise GBM cell viability. Nevertheless, it is important to recognize the context-dependent roles of ROS in GBM. While excessive ROS can induce cell death, moderate ROS concentrations act as signaling molecules that promote tumor cell proliferation, migration, and therapeutic resistance by activating pathways such as the PI3K/AKT and MAPK pathways³⁹. Thus, the dual nature of ROS highlights the complexity of redox regulation in GBM and warrants further investigation to delineate how BLVRA modulates ROS dynamics in a context-specific manner.

In addition to its metabolic functions, BLVRA has been implicated as a prognostic marker and therapeutic target in multiple cancers, particularly hepatocellular carcinoma and colorectal cancer, where its prognostic significance has been validated^40–42. Our study extends this understanding by demonstrating that BLVRA plays a similarly crucial role in GBM progression through fatty acid metabolism regulation. BLVRA catalyzes the conversion of biliverdin to bilirubin, which activates PPARα, a master regulator of lipid metabolism, thereby promoting the expression of lipid-burning genes such as CPT1A, which suppresses lipid accumulation^43,44. Our findings suggest that the BLVRA regulates fatty acid transport and oxidation by modulating carnitine metabolism, providing further evidence of its role in metabolic remodeling and tumor growth.

LDs serve as crucial energy reservoirs in cancer cells, maintaining lipid homeostasis and preventing lipotoxicity while sustaining energy demands under metabolic stress⁴⁵. We found that BLVRA knockdown significantly reduced LD accumulation, which coincided with an increase in CPT1A, suggesting a shift toward increased FAO. Metabolomic analysis further revealed a marked downregulation of TCA cycle–associated pathways following BLVRA knockdown. This metabolic suppression coincided with reduced tumor growth, suggesting that impaired TCA activity may contribute to the observed proliferative defects. While this association supports a key role for BLVRA in sustaining energy metabolism via fatty acid oxidation, further mechanistic studies are needed to establish causality directly. Carnitine is essential for mitochondrial FAO, serving as a shuttle for acyl group transport from coenzyme A to the mitochondrial matrix, ensuring efficient fatty acid oxidation. Additionally, carnitine regulates the acetyl-CoA/CoA ratio, a critical determinant of cancer cell metabolic flexibility^46,47. Disruptions in coenzyme A metabolism may alter metabolic homeostasis, as coenzyme A has been identified as a biomarker of p53 activation^48–51. Notably, we observed that BLVRA knockdown significantly reduced carnitine levels (p < 0.0001) while increasing coenzyme A levels (p < 0.0001), highlighting metabolic disruptions associated with BLVRA depletion. These alterations may lead to p53 activation, impairing GBM cell growth. Thus, BLVRA deficiency may weaken FAO, force GBM cells to switch to alternative metabolic pathways, disrupt redox balance, and deplete critical metabolic intermediates, ultimately restricting energy supply and suppressing GBM proliferation.

BLVRA may also contribute to GBM drug resistance and immune evasion by altering the metabolic status of immune cells within the TME. Studies have shown that BLVRA expression is greater in M2 macrophages than in M1 macrophages and that its overexpression is closely associated with chemotherapy resistance^52–54. These findings suggest that the BLVRA not only influences tumor cells but also may modulate GBM drug resistance by reshaping immune responses in the TME. Thus, further investigation into the role of the BLVRA in tumor immune regulation is warranted. The FAMRGPI risk score demonstrated excellent prognostic performance, with an AUC value of 0.923, significantly surpassing traditional clinical variables such as age and sex. The observation that the FAMRGPI has a stronger prognostic impact on male patients over 60 years of age is consistent with the epidemiological trend of GBM, which is more aggressive in older individuals. Age-associated metabolic dysregulation diminishes immune surveillance, and potential sex-specific factors may contribute to the increased predictive value of the FAMRGPI in this subgroup. This finding indicates that the risk score provides a more precise prognostic assessment, particularly for long-term survival prediction. Future clinical studies should further validate the utility of the FAMRGPI as an individualized treatment decision-making tool for GBM. Despite offering valuable insights into the role of BLVRA in GBM metabolic reprogramming, this study has certain limitations. First, the research relied primarily on in vitro models, which, while informative, cannot fully recapitulate the complexity of the TME and its metabolic dynamics in vivo. Therefore, future studies should incorporate animal models and clinical samples to validate the physiological function of BLVRA and its role in tumor progression. Second, although BLVRA deficiency significantly altered tumor cell metabolic pathways, the precise molecular mechanisms involved remain largely unexplored, necessitating further investigation. Future studies should aim to elucidate the regulatory role of BLVRA in these metabolic pathways to better define its contributions to tumor metabolism. Third, although U87 cells were used for initial validation experiments, we acknowledge the limitations of this cell line due to its long-term in vitro passaging and divergence from the primary GBM phenotype. U87 cells, despite their widespread historical use, exhibit notable differences from patient-derived glioma models in terms of genetic and transcriptomic profiles, particularly regarding invasiveness and tumor heterogeneity^55,56. In our study, U87 cells were employed only in preliminary Western blot validation, wherein BLVRA expression was found to be highest among the tested GBM cell lines. For all subsequent functional assays and mechanistic analyses, we deliberately utilized U251 and LN229 cell lines, which exhibited consistent BLVRA upregulation and are commonly used to study GBM proliferation and metabolic regulation. This model selection strategy ensured greater biological relevance and reproducibility.

Conclusion

In this study, we identified BLVRA as a potential regulator of fatty acid metabolism in GBM. Our results suggest that BLVRA may contribute to tumor progression by maintaining lipid homeostasis and modulating oxidative stress and DNA damage responses. Preliminary evidence from qPCR and Western blot analyses indicates that BLVRA depletion disrupts fatty acid metabolism, increases ROS levels, induces cell cycle arrest, and enhances sensitivity to apoptotic signaling, likely through activation of the p53 pathway. These findings offer initial mechanistic insights and provide a rationale for further investigation into the therapeutic potential of targeting BLVRA-mediated metabolic vulnerabilities in GBM.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

GBM

Glioblastoma

BLVRA

Biliverdin reductase A

LC-MS

Liquid chromatograph-mass spectrometer

scRNA-seq

single-cell RNA sequencing

TCGA

The cancer genome atlas

GEO

Gene expression omnibus

CGGA

Chinese glioma genome atlas

GTEX

Genotype-tissue expression

TME

Tumor microenvironment

GSVA

Gene set variation analysis

ROC

Receiver operating characteristics

AUC

Area under the curve

PCA

Principal component analysis

OPLS-DA

Orthogonal partial least squares-discriminant analysis

VIP

OPLS-DA variable projection importance

DEG

Differentially expressed gene

FAMRGPIBEST

Fatty Acid Metabolism-Related Gene Prognostic Index Biomarker Exploration for Solid Tumor

CNV

Copy number variation

DAB

Diaminobenzidine

shRNA

Short hairpin RNA

KEGG

Kyoto encyclopedia of genes and genomes

GO

Gene ontology

CCK8

Cell counting kit 8

LD

Lipid droplet

TCA

Tricarboxylic acid

FAO

Fatty acid oxidation

ROS

Reactive oxygen species

PC

Phosphatidylcholine

PE

Phosphatidylethanolamine

GSH/GSSG

Reduced glutathione/oxidized glutathione

Author contributions

L.L. led the study conceptualization, experimental design, primary data analysis, and manuscript writing. J.-H.Y. contributed to the data analysis and manuscript revision. J.F. and H.-B. S participated in database analysis and data interpretation. Q.Z. and M.-S.Z. were responsible for the collection and analysis of the clinical samples. X.-F.W. played a key role in study initiation, supervision, project management, man-uscript writing, and final revision. All the authors read and approved the final manu-script.

Funding

This study was supported by grants from the Harbin Medical University Postgraduate Research and Innovation Program (Grant No. 0204-22992230029) and the Harbin Medical University Affiliated Hospital No. 4 Special Funded Research Project (Grant No. HYDSYTB202232).

Data availability

The datasets analyzed in the current study are publicly available. Bulk RNA-seq data for glioblastoma multiforme (GBM, TCGA-GBM, CGGA-301, CGGA-325, CGGA-693) were obtained from The Cancer Genome Atlas (TCGA, https://portal.gdc.cancer.gov/) and Chinese Glioma Genome Atlas (CGGA, https://www.cgga.org.cn), and bulk RNA-seq data for normal brain (GTEX-brain) were obtained from the Genotype-Tissue Expression (GTEX, https://gtexportal.org/home/) projects. GBM transcriptome datasets (GSE83300, GSE74187) and single-cell RNA sequencing datasets (GSE162631, GSE214966) were obtained from the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/). All accession numbers and persistent links are provided above and in the manuscript.

Declarations

Ethics approval and consent to participate

The protocol and all procedures involving clinical tissue samples in this study were reviewed and approved by the Medical Ethics Committee of the Fourth Affiliated Hospital of Harbin Medical University (Approval No. 2024-LUNLISHENCHA-110). Informed consent was obtained from all patients prior to the use of paraffin-embedded tissues in this study. All study procedures were conducted in accordance with relevant ethical guidelines and local regulations.

Competing interests

The authors declare no competing interests.