A novel risk model of cholesterol metabolism-related mRNAs for predicting overall survival and immune signature in glioma based on machine learning and multi-omics data

Introduction

Gliomas represent the most frequent primary malignant neoplasms of the central nervous system (CNS), contributing substantially to disease burden worldwide owing to their high incidence and mortality. They originate from glial or progenitor cells and are classified into distinct histological and molecular categories, such as astrocytoma, oligodendroglioma, and glioblastoma. According to the current World Health Organization (WHO) classification, meningioma and medulloblastoma are recognized as separate tumor entities rather than glioma subtypes (1,2). According to the 2021 5^th edition of the WHO CNS Tumor Classification (CNS5), gliomas are assigned a grade from I to IV, with increasing grade generally reflecting more aggressive biological behavior and shorter survival. Lesions in grades I–II are collectively termed low-grade gliomas (LGGs), whereas those in grades III–IV are grouped as high-grade gliomas (HGGs) (2). Glioblastoma, classified as grade IV under the WHO system, represents the most aggressive glioma subtype, and patients typically survive fewer than 2 years despite standard treatment (3). Although progress in neurosurgery, radiotherapy, and emerging immunotherapies has extended survival to some extent, outcomes remain dismal in glioblastoma, where prognosis continues to be particularly poor (4-6). There is a pressing need to identify reliable biomarkers in glioma that can refine prognostic evaluation and inform immunotherapeutic strategies.

Cholesterol is a major component of cell and organelle membrane lipids and is essential for maintaining cellular homeostasis and structural integrity (7,8). The high metabolic demand for energy in tumor cells leads to metabolic reprogramming in the cell, in which cholesterol metabolism dysregulation is also an influential contributor to tumorigenesis (9,10). Such dysregulation has also been identified as a risk factor for tumor proliferation, invasion, and metastasis (11,12). In gliomas and other malignancies outside the CNS, cholesterol biosynthesis is largely driven by 3-hydroxy-3-methylglutaryl-coenzyme A reductase (HMGCR) and uptake through low-density lipoprotein receptor (LDLR), whereas efflux is controlled by the nuclear receptor family liver X receptor (LXR) (13). The expression of these regulators is further modulated by sterol regulatory element-binding protein 2 (SREBP2), which acts as a key transcriptional activator in lipid metabolism (14,15). Targeting cholesterol biosynthesis and efflux has emerged as a potential therapeutic approach in oncology. Supporting this notion, HMGCR inhibitors such as statins have shown antitumor activity in several studies (16). Within the tumor microenvironment (TME), cholesterol accumulates not only in stromal and immune cells but also influences cytokine and chemokine signaling, highlighting its critical contribution to shaping immune responses (17,18). Moreover, cholesterol modulates T cell consumption, macrophage phagocytosis, and lymphocyte exhaustion in the TME and can lead to tumor cell immune escape (19).

While cholesterol metabolism is known to influence glioma progression, a systematic multi-omics and machine learning-based integration of cholesterol-related messenger RNAs (mRNAs) for prognosis and immunotherapy prediction across glioma grades remains to be conducted. In this work, we applied integrative bioinformatics analyses to cholesterol metabolism-related genes (CMRGs) and developed a prognostic scoring system for glioma (20-22). Its predictive capacity was examined in both an internal test set and an external validation cohort (23,24). To strengthen these findings, expression characteristics and functional implications of the signature genes were further explored using single-cell transcriptomic data and experimental assays in cultured glioma cells (25,26). Collectively, these analyses underscore the prognostic significance of the selected genes and suggest their potential relevance for informing immunotherapy in glioma. We present this article in accordance with the MDAR and TRIPOD reporting checklists (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2819/rc).

Methods

Data collection and processing

Transcriptomic sequencing data from glioma patients, along with clinical information, were collected from The Cancer Genome Atlas (TCGA) (https://portal.gdc.cancer.gov/projects/TCGA-GBM) and Chinese Glioma Genome Atlas (CGGA) (https://www.cgga.org.cn/download.jsp). Clinical data from public databases for glioma patients included their age, gender, overall survival (OS) time, survival status, WHO grade, tumor origin subtype (proneural, neural, classical, or mesenchymal), isocitrate dehydrogenase (IDH) status, 1p/19q co-deletion status, and O⁶-methylguanine-DNA methyltransferase (MGMT) promoter methylation status. CMRGs were extracted from the Molecular Signatures Database (https://www.gsea-msigdb.org/gsea/msigdb). Before further analysis, transcriptomic data acquired from the TCGA were log₂-transformed for standardization. The R (4.2.1) software and the tidyr R packages were used to synthesize for data synthesis and analysis.

CMRG screening and prognostic model construction

To construct the prognostic model, we applied a machine learning-based strategy consisting of least absolute shrinkage and selection operator (LASSO)-penalized Cox regression, which enables efficient feature selection and risk score calculation. We conducted Cox regression based on OS time and survival status. Then, the results were subjected to LASSO regression for further refinement. The final set of model-associated genes was determined through multivariate Cox regression. Based on the expression levels and their correlations with clinical outcomes, we calculated a risk score for glioma patients using the following formula:

Riskscore=∑i=1ncoefficient×expression[1]

Glioma patients were stratified into high- and low-risk groups based on the median risk score. The effectiveness of the prognostic model was further evaluated using Kaplan-Meier (K-M) survival analysis and receiver operating characteristic (ROC) curve analysis.

Influence of risk score on prognosis and nomogram construction

Cox regression models at both univariate and multivariate levels were applied to test whether the constructed risk score retained prognostic independence from established clinical variables. A nomogram integrating the score with patient characteristics was then generated, and its performance was calibrated against observed outcomes. Model discrimination and clinical utility were further quantified through time-dependent ROC analysis and calibration plotting. All evaluations were performed separately in the TCGA and CGGA glioma cohorts.

Enrichment analysis

Pathway analysis was performed using gene set enrichment analysis (GSEA) (v4.3.2) to determine biological processes differentially enriched between the low- and high-risk subgroups of HSPAGs. Functional annotation through Gene Ontology (GO) categories and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways was carried out with the clusterProfiler package in R, applying thresholds of P<0.05 and false discovery rate (FDR) <0.05.

To further characterize group-specific biology, we applied single-sample GSEA (ssGSEA) to compute individual enrichment scores for angiogenesis, epithelial-mesenchymal transition (EMT), tumor stemness, and cytokine signaling. These quantitative indices were then contrasted between the two risk strata.

In addition, expression patterns of the human leukocyte antigen (HLA) gene family were examined to assess potential alterations in antigen presentation capacity associated with risk status.

Mutation analysis

Tumor mutational burden (TMB) scores were computed for glioma patients using the maftoolsR package, based on their somatic mutation data. Survival analyses were then conducted to compare outcomes among patient subgroups stratified by the median TMB score and the median risk score, respectively.

Comprehensive immune landscape

Immune infiltration was estimated for each glioma case using two complementary approaches: the TIMER resource and the CIBERSORT deconvolution algorithm. In parallel, immune, stromal, and composite ESTIMATE scores were derived to evaluate the TME.

To explore potential variation in susceptibility to immune checkpoint blockade (ICB), we compared PD-L1 (CD274) expression between risk groups. In addition, Tumor Immune Dysfunction and Exclusion (TIDE) analysis was applied across the cohort, and intergroup differences were assessed. These evaluations were intended to highlight which patient subsets might exhibit enhanced benefit from checkpoint inhibitor treatment.

Drug sensitivity analysis

Genomics of Drug Sensitivity in Cancer v2 (GDSC2) (https://www.cancerrxgene.org/) and Cancer Therapeutic Response Portal (CTRP) (https://portals.broadinstitute.org/ctrp/) were downloaded through the “pharmigoGx” package. The “calcPhenotype” function predicted half-maximal inhibitory concentration (IC₅₀) values for glioma patients. The IC₅₀ value of the selected drugs between the training set and the test set was then visualized using PharmacoDB 2.0 (https://pharmacodb.ca/).

Single-cell analysis

The single-cell transcriptomes of the samples were downloaded from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) (GSE167960). In total, 26,574 cells were analyzed using the “Seurat” R package. Quality control criteria included a minimum of 200 genes, a maximum of 4,500 genes, a minimum of 1,000 RNA strips, a maximum of 35,000 RNA strips, and a maximum of 10% mitochondrial RNA. Batch effects across samples were eliminated using the Find Integration Anchors and Integrate Data functions. Cells were clustered and identified using marker genes from the Cell Marker Database.

Patients and samples

Tissue samples were obtained from patients with glioma from the Department of Neurosurgery, The First Affiliated Hospital of Soochow University between January 2020 and January 2024. Glioma specimens were histologically confirmed as adult-type diffuse gliomas in accordance with the CNS5. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Medical Ethics Committee of The First Affiliated Hospital of Soochow University [2023 Ethical Review (Declaration) No. 338]. Informed consent was obtained from all individual participants included in the study.

Immunohistochemistry (IHC) staining

Formalin-fixed, paraffin-embedded (FFPE) tumor specimens were sectioned into 4 µm thick slices and processed as previously described (27). The tissue sections were then incubated with either a TM7SF2 Rabbit Polyclonal Antibody (OriGene, Rockville, MD, USA; catalog No. TA35181) or an Annexin V (H-3) Mouse Monoclonal Antibody (Santa Cruz Biotechnology, Inc., Dallas, TX, USA; catalog No. sc-74438) for 2 hours at room temperature (RT). Positive immunostaining was identified as the presence of a yellowish-brown color in the cell cytoplasm or membrane.

Cell culture

Glioma cell lines U87MG and U251, together with a normal human astrocyte line, were purchased from the Shanghai Institute of Biological Sciences and Xavier Biology (Wuhan, China). Both are well-characterized glioblastoma cell lines commonly used in cholesterol metabolism and glioma biology studies. U87MG is derived from a malignant glioma and has been widely used in metabolic and therapeutic research, while U251 is known for its aggressive phenotype and relevance to immune microenvironment studies. Cells were cultured in Dulbecco’s modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin. Cultures were maintained at 37 ℃ in a humidified incubator with 5% CO₂.

RNA extraction and quantitative real-time polymerase chain reaction (qRT-PCR)

RNA extraction was carried out with TRIzol reagent (Invitrogen, Carlsbad, CA, USA). First-strand cDNA was generated using the RevertAid cDNA Synthesis Kit (K1622; Thermo Scientific, Waltham, MA, USA). qRT-PCR was performed on a StepOne™ Real-Time PCR platform (Applied Biosystems, Waltham, MA, USA). Expression of ANXA5 and TM7SF2 was quantified relative to glyceraldehyde 3-phosphate dehydrogenase (GAPDH), and all assays were repeated in triplicate.

Transfection of small interfering RNA (siRNA)

U87MG and U251 cells were transfected with siRNA using Lipofectamine™ 3000 transfection reagent (Invitrogen). The siRNA targeting ANXA5 was purchased from Quanyang Biological (Shanghai, China) and transfected overnight, and the culture medium was changed for follow-up experiments.

Cell counting kit-8 (CCK-8) assay

Cell proliferation was evaluated with the CCK-8 assay (Beyotime, Hangzhou, China). U87MG and U251 cells transfected with the indicated miRNA or siRNA were seeded into 96-well plates at 0.5×10⁴ cells per well and incubated for 0, 12, 24, 36, or 48 hours. Two hours before each time point, 10 µL of CCK-8 solution was added to the wells. Absorbance at 450 nm was recorded using a Thermo Fisher microplate reader to assess cell viability and growth.

Wound healing assay

The wound healing assay was used to evaluate cell migration. Transfected U87MG and U251 cells (1×10⁶) were seeded into six-well plates and grown to full confluence. A scratch was created with a 200 µL pipette tip, and cells were then maintained in serum-free DMEM. Images were captured at 0 and 24 hours with an ECLIPSE Ti2 microscope (Nikon, Tokyo, Japan), and wound closure was quantified using ImageJ.

Transwell assay

A Transwell migration assay was carried out using 24-well chambers with 8 µm pores (#3422; Corning Costar, Tewksbury, MA, USA). Transfected U87MG and U251 cells (2×10⁴) in 200 µL serum-free DMEM were seeded into the upper chamber, and 600 µL DMEM containing 10% FBS was added to the lower chamber. After 24 hours at 37 ℃ in 5% CO₂, cells remaining on the upper surface were removed, while migrated cells were fixed with 4% paraformaldehyde, stained with 0.1% crystal violet, and counted in three random microscopic fields (ECLIPSE Ti2, Nikon).

Statistical analysis

All results, unless noted otherwise in the figure legends, are shown as mean ± standard error of the mean (SEM) based on a minimum of three independent biological replicates, and statistical analyses were conducted using GraphPad Prism. If error bars are not displayed, they fall within the range of the data point markers. Statistical significance was evaluated using two-tailed Student’s t-tests, with P values of <0.05, <0.01, <0.001, and <0.0001 considered significant.

Results

Determination and model validation of cholesterol metabolism-related prognostic signatures (CMRPSGs) in glioma patients

A total of 141 CMRGs were retrieved from the Molecular Signatures Database. Glioma cases from TCGA were filtered using two criteria: (I) removal of samples without reliable survival data; and (II) exclusion of cases lacking WHO grade information. After screening, 629 tumor samples were retained. Univariate Cox regression (R package survival) identified 17 prognostic candidates (Figure S1). Subsequent LASSO analysis (glmnet package) reduced the panel to nine genes (Figure 1A,1B), and multivariate Cox regression further refined this to five hub genes used for model construction (Figure 1C). Detailed survival outcomes for these genes are presented in the following sections.

Figure 1 Construction of the CMRPSG. (A,B) λ value and coefficient curves derived from LASSO regression. (C) Multivariate Cox proportional hazard ratios with 95% CIs and P values. (D) Prognostic risk score model used to risk-stratify the training set of TCGA glioma patients. CI, confidence interval; CMRPSG, cholesterol metabolism-related prognostic signature; LASSO, least absolute shrinkage and selection operator; TCGA, The Cancer Genome Atlas.

Of the five prognostic genes, elevated expression of ANXA5, GPX8, and TNFRSF12A was linked to shortened survival, whereas reduced levels of TM7SF2 and ANXA2 were likewise associated with adverse outcomes. Using the calculated risk scores, patients were divided into high- and low-risk cohorts (Figure 1D). K-M curves demonstrated a significant separation in OS between the two groups (P<0.001) (Figure S2A). The prognostic model showed robust discrimination, with area under the curve (AUC) values of 0.893, 0.804, and 0.749 for 1-, 3-, and 5-year survival, respectively (Figure S2B), supporting its reliability and predictive capacity. To confirm the reliability of CRMGs, we performed an external validation in CGGA. It was found that the survival probability and time of the high-risk group were significantly lower than those of the low-risk group (Figure S2C). Next, ROC curve analysis was then performed to further verify the accuracy of prognostic CRMGs in predicting the prognosis of patients with glioma in CGGA. The AUC of 1-, 3-, and 5-year is 0.706, 0.776, and 0.787, respectively (Figure S2D). In addition, 929 glioma patients in CGGA were divided into a high-risk group and a low-risk group based on the median risk score (Figure S2E).

Functional enrichment analysis of CMRPSGs and mutation analysis

GSEA revealed distinct enrichment patterns of HSPAGs across the two risk categories. In the high-risk subgroup, genes were enriched in KEGG pathways involving carbohydrate metabolism, cell cycle regulation, glutathione metabolism, leukocyte transendothelial migration, and Escherichia coli infection (Figure 2A). In contrast, the low-risk subgroup showed enrichment in KEGG pathways of alanine, aspartate, and glutamate metabolism, as well as long-term synaptic depression/potentiation (Figure 2B). GO analysis further highlighted enrichment in processes such as negative regulation, protein handling, nuclear division and its response, response to gamma irradiation, and cell-substrate junction organization (Figure 2C). GO terms in this group included five distinct biological processes (Figure 2D).

Figure 2 GSEA based on KEGG and GO databases in different risk groups. (A,C) GSEA analysis results according to the KEGG (A) and GO (C) databases in the high-risk group. (B,D) GSEA analysis results according to the KEGG (B) and GO (D) databases in the low-risk group. GO, Gene Ontology; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Construction and validation of a CMRPSG-based nomogram and risk score model for predicting glioma patient survival

Apart from the risk score, other factors demonstrating independent prognostic significance for glioma patient survival included MGMT promoter status, gender, original subtype, grade, age, and IDH mutation status (Figure 3A). Additionally, the calibration curves for TCGA glioma patients showed a high degree of concordance between the observed OS rates and the model-predicted survival rates at 1, 3, and 5 years (Figure 3B).

Figure 3 Nomogram construction and correlation analysis with tumor-related functional scores. (A) Column line plots of predicted 1-, 2-, and 3-year OS probabilities for glioma patients. (B) Calibration curves assessing the agreement between predicted and observed probabilities of the prognostic risk score model evaluated against the OS of glioma patients in the training set. (C-F) Performance status according to the angiogenic activity (C), mesenchymal EMT scores (D), stemness scores (E), and tumorigenic cytokine scores (F). **, P<0.01; ***, P<0.001. EMT, epithelial-mesenchymal transition; IDH, isocitrate dehydrogenase; MGMT, O⁶-methylguanine-DNA methyltransferase; OS, overall survival.

We then assessed angiogenic activity, EMT, stemness, and tumorigenic cytokine scores in the glioma patients. Angiogenic activity scores positively correlated with risk scores (Figure 3C), while mesenchymal EMT scores were negatively correlated (Figure 3D). Stemness scores did not differ significantly (Figure 3E). Furthermore, tumorigenic factor scores were also positively correlated with risk scores (Figure 3F).

Immune checkpoint and immune infiltration

The ssGSEA results showed significantly higher HLA-related molecule expression in the high-risk group, correlating with a poor prognosis (Figure 4A). Moreover, elevated immune checkpoint molecule PD-1 levels in the high-risk group suggested a potentially favorable response to immune checkpoint inhibitor (ICI) therapy (Figure 4B). Subsequently, we used the ESTIMATE algorithm to evaluate distinct TME patterns across different samples, which aligned with our expectations (Figure 4C-4E). Additionally, stratifying patients by TMB score and risk scores revealed poorer prognosis in the high-TMB + high-risk group (Figure 5A). To better predict the response of HSPAGs to immunotherapy, the TIDE analysis was indicated low TIDE scores and exclusion scores in the high-risk group and no significant differences of dysfunction scores were found between groups (Figure 5B-5D).

Figure 4 Evaluation of the tumor immune microenvironment and immune checkpoint expression between risk groups. (A) Analysis of immune checkpoints for two differential groups. (B) Analysis of PD-L1 expression for two groups. (C-E) Differential analysis of ESTIMATE (C), immune (D), and stromal (E) scores. *, P<0.05; **, P<0.01; ***, P<0.001.

Figure 5 Survival analysis combining TMB with risk scores and prediction of immunotherapy response via TIDE. (A) Survival differences based on TMB and risk score. (B-D) TIDE, tumor immune dysfunction or exclusion scores in TCGA dataset. *, P<0.05; ***, P<0.001; ns, not significant (P>0.05). H-TMB, high-TMB; L-TMB, low-TMB; TCGA, The Cancer Genome Atlas; TIDE, Tumor Immune Dysfunction and Exclusion; TMB, tumor mutational burden.

Drug sensitivity

IC₅₀ values for glioma patient data from the training set TCGA and validation set CGGA were predicted using the “oncoPredict” algorithm. The results of IC₅₀ values for 27 Food and Drug Administration (FDA)-approved drugs in two different glioma datasets are shown in Figure 6A, of which 14 are significantly resistant in the high-risk group, while the rest are significantly susceptible, and the pathways of the drug targets are also shown. Nine drugs (AZD7762, PLX-4720, AZD8055, BI-2536, OSI-027, MG-132, NVP-ADW742, BIBR-1532, and ZD6482) were identified by intersecting four databases (Figure 6B). We demonstrate the correlation results of these nine drugs within the glioma dataset and the mix of drug datasets, respectively (Figure 6C-6F).

Figure 6 Prediction of drug sensitivity and identification of candidate therapeutic drugs based on the risk model. (A) Heatmap of predicted IC₅₀ values for 27 kinds of drugs approved by FDA in GDSC2 between different subgroups. (B) Venn plots: cross-validation of different glioma datasets and drug datasets. (C-F) Correlation between the predicted IC₅₀ values of four selected drugs in combinations of GDSC2-TCGA (C), CTRP-TCGA (D), CTRP-CGGA (E), and GDSC2-CGGA (F) and the expression of five modeled genes. CGGA, Chinese Glioma Genome Atlas; CTRP, Cancer Therapeutic Response Portal; FDA, Food and Drug Administration; GDSC2, Genomics of Drug Sensitivity in Cancer v2; IC₅₀, half-maximal inhibitory concentration; TCGA, The Cancer Genome Atlas.

Validation of CMRPSG expression patterns in the risk model through scRNA-seq analysis

Four distinct cell populations were identified by the Uniform Manifold Approximation and Projection (UMAP) analysis, including the following categories: endothelial, glioma, macrophage, microglia, neural stem, oligodendrocyte, and oligodendrocyte precursor cells (Figure 7A). The marker genes for the seven major subclusters are shown in Figure 7B. Figure 7C,7D show that ANXA5 was relatively highly expressed in all cell clusters. The degree of association in the seven cellular subpopulations can be seen in the cell-cell interaction results, and there was a closer association between the glioma, endothelial, oligodendrocyte, and oligodendrocyte precursor cell clusters in the TNF pathway (Figure 7E), and endothelial, microglia, and neural stem cells have a closer relationship in the CXCL pathway (Figure 7F).

Figure 7 Single-cell RNA sequencing analysis reveals signature gene expression patterns and intercellular communication. (A) Clustering and annotation of single-cell data. (B) Marker gene expression signatures. (C,D) Cholesterol signature genes in the four cell types. (E,F) Circle interaction plots showing the counts inferred from intercellular communication network analysis for each of the two cell populations in the TNF (E) and CXCL (F) pathway. UMAP, Uniform Manifold Approximation and Projection.

Expression and functional analysis of TM7SF2 and ANXA5 in glioma cell lines and patient tissues

mRNA levels of TM7SF2 and ANXA5 were assessed by quantitative polymerase chain reaction (qPCR) in U251, U87, and U87MG glioma cells as well as normal astrocytes. The primer sequences of these two genes are shown in Table S1. Results showed that TM7SF2 expression was high in astrocytes but markedly reduced in glioma lines, whereas ANXA5 displayed the opposite trend. Moreover, U87 and U87MG cells exhibited similar expression profiles (Figure 8A,8B).

Figure 8 Experimental validation of the expression and biological functions of TM7SF2 and ANXA5 in glioma. (A,B) TM7SF2 is highly expressed in astrocytes and lowly expressed in glioma cell lines, whereas ANXA5 shows the opposite pattern. (C) Validation of ANXA5 transfection efficiency in U87MG and U251 cell lines using qPCR. (D-H) Assessment of glioma cell proliferation and migration using CCK-8, Transwell (crystal violet staining), and wound healing assays after ANXA5 knockdown via siRNA in U87MG and U251 cell lines. (I-K) IHC analysis of ANXA5 and TM7SF2 expression in glioma patient tissue samples and brain tissue sections from patients with cerebral hemorrhages (IHC staining). *, P<0.05; **, P<0.01; ***, P<0.001. CCK-8, cell counting kit-8; ICH, intracerebral hemorrhage; IHC, immunohistochemistry; qPCR, quantitative polymerase chain reaction; siRNA, small interfering RNA.

To investigate the functional role of ANXA5, siRNA targeting ANXA5 was introduced into U87MG and U251 glioma cells, and transfection efficiency was verified by qPCR (Figure 8C). CCK-8, Transwell, and wound healing assays were subsequently performed to evaluate proliferation and migration. Silencing ANXA5 markedly suppressed both growth and motility in U87MG and U251 cells (Figure 8D-8H).

In addition, IHC was conducted on tissue specimens from glioma and cerebral hemorrhage patients to assess ANXA5 and TM7SF2 expression. The staining results paralleled the qPCR findings, displaying comparable expression trends (Figure 8I-8K).

Discussion

In recent years, studies have identified potential molecular markers for classifying gliomas, assessing prognosis, and guiding treatments (28). Nonetheless, relying solely on these markers falls short of accurately predicting the intricate progression of gliomas and cannot pinpoint optimal therapeutic interventions. Prior investigation has validated the significant role of CMRGs in predicting prognosis for LGG patients (29). Notably, cholesterol metabolism’s relevance extends beyond LGGs and is broadly implicated in various glioma grades. Thus, our study screened metabolism-related genes, identifying five pertinent genes: ANXA5, GPX8, TNFRSF12A, TM7SF2, and ANXA2. The prognostic index model was established by these five genes. Subsequently, we validated this model, and as anticipated, it exhibited strong correlations with adverse clinical and pathological parameters indicative of poor prognosis in gliomas. Simultaneously, our model assessed glioma patient responsiveness to immunotherapy across various risk categories. This holds significant clinical utility for medication decisions. To the best of our knowledge, this is the first study to integrate machine learning with multi-omics data for a systematic, grade-wide analysis of CMRGs in glioma.

The survival time of glioblastoma patients can be prolonged by regulating the level of cholesterol synthesis (30). Moreover, cholesterol plays two key roles in supporting glioblastoma progression: one is to provide the necessary energy for rapid tumor growth, and the other is to promote immune evasion in glioblastoma cells by inducing immunosuppressive TME. Glioma-supporting macrophages (GSMs) are the key cells for cholesterol supply to the CNS and are important immunosuppressive TME controllers (31). They are responsible for both of these mechanisms (32). TM7SF2 is known to play a pivotal role in regulating cholesterol metabolism in the liver, primarily by modulating the activity and expression of HMGCR (33). Emerging evidence also suggests that TM7SF2 may contribute to tumor progression by influencing key cellular signaling pathways, thereby promoting tumor cell proliferation (34). This gene may also affect tumor growth and development by promoting lipid metabolic reprogramming in cancer cells (35). ANXA5’s mechanism in cholesterol metabolism may include the following: ANXA5 may be bound to cholesterol and assist in transporting cholesterol within or between cells, thus affecting the distribution of cholesterol in different organelles or tissues, regulating the activity of cholesterol-associated enzymes, and regulating the cholesterol content of cell membranes, thus altering the fluidity, permeability, and function of cell membranes (36,37).

In previous investigations, the ANXA5 expression trend in gliomas was validated and is consistent with our findings, but information on TM7SF2 in gliomas is still lacking (38). GSEA GO and GSEA KEGG showed the high expression of various signaling pathways possibly affecting the progress and immune responses of gliomas (39). Notably, glutamate and glutathione metabolism were abundant. Previous studies indicated that accumulated endogenous glutamate could provide a carbon skeleton and energy by participating in the tricarboxylic acid (TCA) cycle, whereas TCA cycle intermediates such as acetyl-coenzyme A (acetyl-CoA) are important precursors for cholesterol biosynthesis (40). Both of these pathways could influence ferroptosis occurrence and sensitivity, suggesting that the cholesterol metabolism pathway may be associated with a co-regulatory mechanism in relation to ferroptosis (41). In addition, the high-risk group indicated higher tumorigenic cytokine and angiogenic activity levels than the low-risk group. A higher tumorigenic cytokine score typically indicates cytokines that promote tumor cell proliferation, survival, and invasion, potentially suppressing the immune system’s response to the tumor (42). Moreover, a higher angiogenic activity score often suggests that the tumor promotes new blood vessels in the surrounding tissue, providing oxygen and nutrients to the tumor and facilitating its migration and invasion (43). Consequently, this leads to an unfavorable prognosis for gliomas.

Nevertheless, there is still a lack of comprehensive analysis of CMRPSGs associated with gliomas at different clinical grades. Without a doubt, this study still has some limitations. First, our current findings were mainly derived from the TCGA and CGGA databases, but further validation in a larger glioma cohort is needed because of the obvious differences between TCGA and CGGA in terms of sample size and racial differences. Second, the functions and specific mechanisms of model genes in cholesterol metabolism in gliomas are unknown and need to be further explored. In addition, although U87MG and U251 cells were used for experimental validation, these long-established cell lines have inherent limitations in recapitulating tumor heterogeneity. Future studies using patient-derived models, such as patient-derived xenografts (PDX) or organoids, are warranted to further validate our findings. Our research was based on a machine learning algorithm to comprehensively analyze the CMRG list and construct a brand-new prediction model. We concluded that CMRPSGs robustly regulate glioma prognosis at a genomic regulative level.

Conclusions

Our research has developed some preliminary prognostic risk assessment models for glioma patients based on cholesterol metabolism pathways, and validated the accuracy of our models using various dimensions and methods. We hope that our research can further deepen and provide new insights for glioma patients and glioma researchers.

Acknowledgments

We thank The Cancer Genome Atlas (TCGA), the Chinese Glioma Genome Atlas (CGGA), and the Gene Expression Omnibus (GEO) team for the ability to use their data.

Footnote

Reporting Checklist: The authors have completed the MDAR and TRIPOD reporting checklists. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2819/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2819/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2819/prf

Funding: This research was funded by the National Natural Science Foundation of China (No. 82002643), the Gusu Talent Program (No. GSWS2023086), and the Suzhou Basic Research Pilot Program (No. SSD2024091).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2819/coif). The authors have no 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 the Medical Ethics Committee of The First Affiliated Hospital of Soochow University [2023 Ethical Review (Declaration) No. 338]. Informed consent was obtained from all individual participants included in the study.

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. Xu S, Tang L, Li X, et al. Immunotherapy for glioma: Current management and future application. Cancer Lett 2020;476:1-12. [Crossref] [PubMed]
2. Yang K, Wu Z, Zhang H, et al. Glioma targeted therapy: insight into future of molecular approaches. Mol Cancer 2022;21:39. [Crossref] [PubMed]
3. Weller M, Le Rhun E. How did lomustine become standard of care in recurrent glioblastoma? Cancer Treat Rev 2020;87:102029. [Crossref] [PubMed]
4. Pesce C, Rodella G, Fragassi A, et al. Localized treatment of glioblastoma: a review of clinical strategies and advances in drug delivery systems. Nanomedicine (Lond) 2025;20:2571-99. [Crossref] [PubMed]
5. Cheng M, Zhang ZW, Ji XH, et al. Super-enhancers: A new frontier for glioma treatment. Biochim Biophys Acta Rev Cancer 2020;1873:188353. [Crossref] [PubMed]
6. Wang H, Xu T, Huang Q, et al. Immunotherapy for Malignant Glioma: Current Status and Future Directions. Trends Pharmacol Sci 2020;41:123-38. [Crossref] [PubMed]
7. Zhu J, Deng M, Wang W, et al. Integrative analysis of cholesterol homeostasis reveals prognostic subtypes and immune features in clear cell renal cell carcinoma. Discov Oncol 2025;16:1261. [Crossref] [PubMed]
8. Espinosa G, López-Montero I, Monroy F, et al. Shear rheology of lipid monolayers and insights on membrane fluidity. Proc Natl Acad Sci U S A 2011;108:6008-13. [Crossref] [PubMed]
9. Pavlova NN, Thompson CB. The Emerging Hallmarks of Cancer Metabolism. Cell Metab 2016;23:27-47. [Crossref] [PubMed]
10. Wu Q, Yu X, Li J, et al. Metabolic regulation in the immune response to cancer. Cancer Commun (Lond) 2021;41:661-94. [Crossref] [PubMed]
11. Garcia JH, Jain S, Aghi MK. Metabolic Drivers of Invasion in Glioblastoma. Front Cell Dev Biol 2021;9:683276. [Crossref] [PubMed]
12. Pirmoradi L, Seyfizadeh N, Ghavami S, et al. Targeting cholesterol metabolism in glioblastoma: a new therapeutic approach in cancer therapy. J Investig Med 2019;67:715-9. [Crossref] [PubMed]
13. Pretto S, Yu Q, Bourdely P, et al. A functional single-cell metabolic survey identifies Elovl1 as a target to enhance CD8(+) T cell fitness in solid tumours. Nat Metab 2025;7:508-30. [Crossref] [PubMed]
14. Costa GA, de Souza SB, da Silva Teixeira LR, et al. Tumor cell cholesterol depletion and V-ATPase inhibition as an inhibitory mechanism to prevent cell migration and invasiveness in melanoma. Biochim Biophys Acta Gen Subj 2018;1862:684-91. [Crossref] [PubMed]
15. Liu Z, Liu X, Liu S, et al. Cholesterol promotes the migration and invasion of renal carcinoma cells by regulating the KLF5/miR-27a/FBXW7 pathway. Biochem Biophys Res Commun 2018;502:69-75. [Crossref] [PubMed]
16. Huang B, Song BL, Xu C. Cholesterol metabolism in cancer: mechanisms and therapeutic opportunities. Nat Metab 2020;2:132-41. [Crossref] [PubMed]
17. Reboldi A, Dang E. Cholesterol metabolism in innate and adaptive response. F1000Res 2018;7 F1000 Faculty Rev 1647;
18. Tall AR, Yvan-Charvet L. Cholesterol, inflammation and innate immunity. Nat Rev Immunol 2015;15:104-16. [Crossref] [PubMed]
19. Ma X, Bi E, Lu Y, et al. Cholesterol Induces CD8(+) T Cell Exhaustion in the Tumor Microenvironment. Cell Metab 2019;30:143-156.e5. [Crossref] [PubMed]
20. Guo X, Zhou S, Yang Z, et al. Cholesterol metabolism and its implication in glioblastoma therapy. J Cancer 2022;13:1745-57. [Crossref] [PubMed]
21. Meng Y, Wang Q, Lyu Z. Cholesterol metabolism and tumor. Zhejiang Da Xue Xue Bao Yi Xue Ban 2021;50:23-31. [Crossref] [PubMed]
22. Yoshioka Y, Sasaki J, Yamamoto M, et al. Quantitation by (1)H-NMR of dolichol, cholesterol and choline-containing lipids in extracts of normal and phathological thyroid tissue. NMR Biomed 2000;13:377-83. [Crossref] [PubMed]
23. Guo D, Reinitz F, Youssef M, et al. An LXR agonist promotes glioblastoma cell death through inhibition of an EGFR/AKTREBP-1/LDLR-dependent pathway. Cancer Discov 2011;1:442-56. [Crossref] [PubMed]
24. Qiu Z, Yuan W, Chen T, et al. HMGCR positively regulated the growth and migration of glioblastoma cells. Gene 2016;576:22-7. [Crossref] [PubMed]
25. Chen J, Ying C, Gu Z, et al. Construction of a Prognostic Model Using RNA Processing Factor Genes and the Key Role of NSUN6 in Glioma Outcomes. J Cell Mol Med 2025;29:e70668. [Crossref] [PubMed]
26. Wang J, Ge H, Yu Z, et al. Non-coding RNAs as potential mediators of resistance to lung cancer immunotherapy and chemotherapy. Oncol Res 2025;33:1033-54. [Crossref] [PubMed]
27. Hu C, Zhao L, Liu W, et al. Genomic profiles and their associations with TMB, PD-L1 expression, and immune cell infiltration landscapes in synchronous multiple primary lung cancers. J Immunother Cancer 2021;9:e003773. [Crossref] [PubMed]
28. Li Z, Jin Y, Que T, et al. Identification of Necroptosis-related Molecular Subtypes and Construction of Necroptosis-related Gene Signature for Glioblastoma Multiforme. Curr Med Chem 2024;31:5417-31. [Crossref] [PubMed]
29. Tao R, Huang R, Yang J, et al. Comprehensive analysis of the clinical and biological significances of cholesterol metabolism in lower-grade gliomas. BMC Cancer 2023;23:692. [Crossref] [PubMed]
30. Cui X, Zhao J, Li G, et al. Blockage of EGFR/AKT and mevalonate pathways synergize the antitumor effect of temozolomide by reprogramming energy metabolism in glioblastoma. Cancer Commun (Lond) 2023;43:1326-53. [Crossref] [PubMed]
31. Wu Q, Chen J, Berglund AE, et al. The reprogramming impact of SMAC-mimetic on glioblastoma stem cells and the immune tumor microenvironment evolution. J Exp Clin Cancer Res 2025;44:191. [Crossref] [PubMed]
32. Dong Y, Zhang J, Wang Y, et al. Intracavitary Spraying of Nanoregulator-Encased Hydrogel Modulates Cholesterol Metabolism of Glioma-Supportive Macrophage for Postoperative Glioblastoma Immunotherapy. Adv Mater 2024;36:e2311109. [Crossref] [PubMed]
33. Lin A, Che C, Jiang A, et al. Protein Spatial Structure Meets Artificial Intelligence: Revolutionizing Drug Synergy-Antagonism in Precision Medicine. Adv Sci (Weinh) 2025;12:e07764. [Crossref] [PubMed]
34. Gatticchi L, Petricciuolo M, Scarpelli P, et al. Tm7sf2 gene promotes adipocyte differentiation of mouse embryonic fibroblasts and improves insulin sensitivity. Biochim Biophys Acta Mol Cell Res 2021;1868:118897. [Crossref] [PubMed]
35. Li J, Luo Q, Lu M, et al. METTL16 Promotes Lipid Metabolic Reprogramming and Colorectal Cancer Progression. Int J Biol Sci 2025;21:4782-97. [Crossref] [PubMed]
36. Hakami Zanjani AA, Ebstrup ML, Nylandsted J, et al. Modulation of Annexin-Induced Membrane Curvature by Cholesterol and the Anionic Lipid Headgroup during Plasma Membrane Repair. J Phys Chem B 2024;128:8701-11. [Crossref] [PubMed]
37. Calpe-Berdiel L, Escolà-Gil JC, Julve J, et al. Differential intestinal mucosal protein expression in hypercholesterolemic mice fed a phytosterol-enriched diet. Proteomics 2007;7:2659-66. [Crossref] [PubMed]
38. Liu W, Xiong T, Sun H, et al. Human annexin A5 promotes glioma progression by targeting the MAPK/CD44 pathway. Front Oncol 2025;15:1550961. [Crossref] [PubMed]
39. Li G, Zhao Y, He Y, et al. Machine learning-based construction of Immunogenic cell death-related score for improving prognosis and personalized treatment in glioma. Sci Rep 2025;15:30417. [Crossref] [PubMed]
40. Garcia BM, Melchinger P, Medeiros T, et al. Glutamine sensing licenses cholesterol synthesis. EMBO J 2024;43:5837-56. [Crossref] [PubMed]
41. Wu Y, Liu L, Li Z, et al. A Risk Model Based on Ferroptosis-Related Genes OSMR, G0S2, IGFBP6, IGHG2, and FMOD Predicts Prognosis in Glioblastoma Multiforme. CNS Neurosci Ther 2025;31:e70161. [Crossref] [PubMed]
42. Khezrian A, Shojaeian A, Khaghani Boroujeni A, et al. Therapeutic Opportunities in Breast Cancer by Targeting Macrophage Migration Inhibitory Factor as a Pleiotropic Cytokine. Breast Cancer (Auckl) 2024;18:11782234241276310. [Crossref] [PubMed]
43. Kaur G, Roy B. Decoding Tumor Angiogenesis for Therapeutic Advancements: Mechanistic Insights. Biomedicines 2024;12:827. [Crossref] [PubMed]