Identification of immune-related gene signature for predicting prognosis of glioblastoma patients

Introduction

Glioblastoma (GBM) stands as the most frequently occurring primary malignant tumor of the central nervous system, featuring high invasiveness and heterogeneity (1). Despite advancements in multidisciplinary treatment strategies like surgery, radiotherapy, as well as temozolomide chemotherapy, the prognosis remains grim: median survival under 15 months, a 5-year survival rate of less than 5%, and a high likelihood of recurrence (2,3). The pathological features of GBM include aberrant tumor cell proliferation, active angiogenesis, and widespread necrotic regions. Its therapeutic resistance partially arises from the immunosuppressive characteristics of the tumor microenvironment (4,5).

The immune microenvironment of GBM is marked by the infiltration of immunosuppressive cells like tumor-associated macrophages (TAMs), myeloid-derived suppressor cells, as well as regulatory T cells (6,7). These cells suppress antitumor immune responses by secreting immunosuppressive factors like transforming growth factor-beta and interleukin-10, as well as by upregulating immune checkpoint molecules like programmed death-ligand 1 (PD-L1) and cytotoxic T-lymphocyte-associated protein 4 (8,9). Furthermore, the blood-brain barrier hinders the penetration of immune cells and therapeutic agents, resulting in the limited efficacy of conventional immunotherapies (1). In spite of breakthroughs regarding immune checkpoint inhibitors (ICIs) and chimeric antigen receptor T-cell therapies in some solid tumors, their response rates in GBM remain low, with significant interpatient variability, highlighting the urgent need to identify biomarkers for selecting potential responders (10-14).

Current research primarily focuses on molecular subtyping of GBM, such as isocitrate dehydrogenase (IDH) mutations and O⁶-methylguanine-DNA methyltransferase (MGMT) methylation status. However, prognostic models based on immune characteristics remain underdeveloped (15,16). Tumor heterogeneity, dynamic changes in the immune microenvironment, and the lack of cross-dataset validation limit the clinical applicability of existing biomarkers (8,17). Moreover, the synergistic mechanisms of immune-related genes (IRGs) in GBM development and progression are not yet fully understood, and constructing predictive and interpretable models using multi-omics data remains a challenge.

This study aims to integrate transcriptomic data and clinical information from public databases such as The Cancer Genome Atlas (TCGA) and Chinese Glioma Genome Atlas (CGGA) to identify immune genes closely related to GBM prognosis and to construct an immune scoring model. Through molecular subtyping, immune microenvironment analysis, and drug sensitivity assessment, the study explores treatment response differences among patients with different immune subtypes, ultimately providing new strategies for personalized immunotherapy and prognostic evaluation in GBM. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2391/rc).

Methods

Data source and preprocessing

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.RNA sequencing and corresponding clinical data for the GBM population were sourced from TCGA and the CGGA. The CGGA-693 and CGGA-325 datasets were selected. IRGs’ lists were downloaded from the ImmPort database. The preprocessing steps were as follows: (I) samples with missing survival data or critical clinical information were excluded; (II) ENSEMBL gene identifiers were standardized and converted to official gene symbols; (III) for genes with multiple symbol representations, the median expression value was used; (IV) batch effect correction was performed utilizing the remove Batch Effect function from the limma package on the TCGA, CGGA-693, and CGGA-325 datasets, and validated via principal component analysis. The corrected datasets were merged to form a joint dataset for subsequent analysis.

To ensure data reliability, additional quality control steps were performed: (I) expression filtering: genes with zero expression across all samples were removed; (II) normalization: quantile normalization was applied to eliminate technical bias between platforms; (III) clinical information matching: sample IDs were cross-checked to ensure accurate correspondence between gene expression data and clinical information.

Immune cell infiltration analysis and subtype classification

The infiltration of 28 immune cell types in every sample was assessed through single-sample gene set enrichment analysis (ssGSEA), followed by unsupervised clustering. The partitioning around medoids (PAM) algorithm was used with Pearson correlation as the distance metric. Bootstrap sampling (500 iterations) was performed, selecting 80% of the samples in each iteration. The ideal number of clusters (k) was selected based on the cumulative distribution function (CDF) curve and the Delta area, with k set between 2 and 10.

Our samples were split into the hot and cold immune groups based on the immune scores derived using the ESTIMATE package for every subtype. Inter-group differences in overall survival (OS) were examined via Kaplan-Meier (KM) survival curves. The statistical significance was assessed through the log-rank test.

Differential analysis and functional annotation between hot and cold tumor groups

Differentially expressed genes (DEGs) expression across the groups was enabled by the limma package for DEGs identification. Somatic mutation data from TCGA-GBM were annotated and statistically analyzed via the Maftools package. The mutation rates of high-frequency genes in both immune subgroups were assessed. For the identified DEGs, Gene Ontology (GO) enrichment, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis, and gene set enrichment analysis (GSEA) were carried out through the clusterProfiler package.

Weighted gene co-expression network analysis (WGCNA)

The WGCNA package was leveraged for the analysis of scale-free topology characteristics of gene expression data. The optimal soft threshold (Power value: 1–30) was selected based on the scale-free fit index (R²) and the average connectivity, aiming for R²>0.8 and moderate average connectivity. The selected power value (Power =6) was used to construct an adjacency matrix. Gene hierarchical clustering was enabled by the Dynamic Tree Cut algorithm, and highly similar modules were merged and color-coded. The correlation between module eigengenes (ME) and immune subtypes was calculated and assessed through two-sided t-tests. Modules with high correlation were selected for further analysis. Core genes were identified by correlating gene significance and module membership. The intersection of WGCNA module genes and DEGs from the immune subtypes was determined, and these shared genes were further analyzed for functional enrichment and survival analysis.

Prognostic model development

Based on the foregoing intersection between the hot and cold tumor cohorts, univariate Cox proportional hazards regression analysis was carried out to preliminarily screen genes with significant prognostic associations. Candidate genes significantly linked to OS in the GBM population were identified.

Subsequently, based on the prognostic-related genes identified above, 10 machine learning (ML) methods [StepCox, plsRcox, Ridge, least absolute shrinkage and selection operator (LASSO), random survival forest (RSF), gradient boosting machine, Enet, CoxBoost, SuperPC, support vector machine for survival analysis (survival-SVM)] were integrated in various combinations, yielding 101 ML algorithms to construct a multi-model prognostic prediction framework. The TCGA GBM dataset served as the training set. Ten-fold cross-validation was applied to optimize model parameters, thereby preventing overfitting. Model performance was evaluated by the concordance index (C-index) for quantitative comparison. The model demonstrating the greatest C-index was the most accurate prediction model. To validate the model’s generalizability, external validation was performed using three independent external datasets: CGGA-325, CGGA-693, and Gene Expression Omnibus Series (GSE)108474 (comprising 261 GBM samples).

Further, survival disparities across high- and low-risk cohorts were detected via KM curves. Furthermore, the model’s prognostic accuracy for 1-, 2-, and 3-year survival was validated through receiver operating characteristic (ROC) curves, with the area under the curve (AUC) reflecting prediction performance in training and validation sets. Finally, AUC values from the four cohorts were integrated and analyzed to comprehensively examine the clinical prediction value of the model.

Nomogram construction and validation, and subgroup analysis

Using the gene feature scores from the prognostic model and clinical-pathological characteristics, a multivariate Cox proportional hazards regression model was leveraged for independent prognostic factor integration and visual nomogram development. The nomogram quantifies the contribution weight of each variable to the patient’s survival probability, enabling individualized survival predictions. The nomogram’s performance in prediction was assessed via decision curve analysis and time-dependent ROC curves.

Four joint stratification groups were established after combining risk scores with immune microenvironment profiles: high-risk-cold, high-risk-hot, low-risk-cold, and low-risk-hot tumors. KM curves displayed the variations in OS across four groups, and Log-rank tests were conducted to assess statistical significance. Sankey diagrams were employed to dynamically visualize the flow of samples from the initial risk stratification to the combined grouping, offering an intuitive representation of the relationships and proportional distribution among different subgroups.

Gene screening and immune correlation analysis

To accurately identify core genes highly related to survival in GBM, feature selection was conducted using three ML methods: LASSO regression, support vector machine-recursive feature elimination (SVM-RFE), and random forest (RF). The intersection of genes identified via the foregoing methods was used to obtain a core set of genes significantly correlated with OS, ensuring the robustness and biological relevance of the selection results.

ROC curves revealed the ability of core genes to distinguish between hot and cold tumors. Permutation tests were carried out to validate the statistical significance of the AUC values for every gene. Genes with AUC >0.7 and significant differences were selected as biomarkers to differentiate between hot and cold tumors.

Pearson correlation coefficients were calculated to evaluate the relationship between core gene expression and immune cell infiltration levels in the cold and hot tumor groups. Heatmaps were used to visualize the significant gene-immune cell pairs.

Single-cell expression profiling and immunohistochemistry (IHC) validation

Five GBM single-cell RNA sequencing datasets were selected from the Tumor Immune Single-cell Hub (TISCH), including GSE103224, GSE139448, GSE148842, GSE84465, and GSE89567. The specific expression patterns of core genes in distinct cell subpopulations were examined through differential expression analysis.

IHC images for the core genes were from the Human Protein Atlas (HPA), with selection criteria including (I) GBM or normal brain tissue; (II) high-specificity antibodies validated by HPA; and (III) exclusion of blurred or non-specific staining images. The single-cell RNA sequencing data utilized in this study were obtained from TISCH. IHC images were retrieved from HPA. Both data sources are publicly accessible and distributed under open-access licenses. All analytical results and visualization figures presented in this manuscript, including but not limited to single-cell expression profiles, risk model visualizations, and survival curves, were independently generated by the authors.

Drug sensitivity analysis

To assess possible differences in drug response among the GBM population, drug sensitivity predictions were made based on risk scores from the prognostic model and molecular classification. The pRRophetic package facilitated clinical chemotherapy sensitivity and targeted therapy agent forecasting.

Statistical analysis

All analyses were enabled by R 4.3.1, involving data preprocessing, immune typing (ssGSEA/PAM clustering), DEA (limma), WGCNA, and ML modeling (LASSO/RF). Model performance was assessed using KM and log-rank tests, as well as ROC and C-index evaluations. A significance threshold of FDR-corrected P values <0.05 was applied.

Results

Immune subtypes and survival analysis in GBM

This study included 146 TCGA, 249 CGGA-693, and 139 CGGA-325 tumor samples (Figure 1A). After removing batch effects (Figure 1B), immune cell infiltration analysis of the transcriptomic data was performed through the ssGSEA algorithm. Consensus CDF was employed for clustering stability evaluation, with the AUC indicating that when k was 2, the CDF slope flattened (values clustered between 0.4 and 1.0), suggesting a high consistency for selecting two clusters (k=2) (Figure 1C,1D). The consensus matrix further confirmed that the samples were divided into two subtypes (subtypes 1 and 2), supporting the subsequent classification into hot and cold tumor subtypes (Figure 1E). As per ESTIMATE scores (StromaScore, ImmuneScore, and ESTIMATEScore), GBM patients were categorized into immune hot and cold subtypes, with the hot tumor group showing significantly higher immune scores (Figure 1F). Hot tumors exhibited a notably better OS in contrast to cold tumors (P=0.004, Figure 1G).

Figure 1 Immune subtype classification and survival analysis in GBM. (A) Principal component analysis visualization of transcriptomic data from three independent cohorts (TCGA, CGGA-693, CGGA-325) after batch effect removal, showing distinct sample clustering. (B) Batch effect correction validation using the removeBatchEffect function (limma package). (C) Heatmap of immune cell infiltration levels estimated by ssGSEA across all samples. (D) CDF curve for consensus clustering stability evaluation. The plateau at k=2 indicates optimal cluster number selection. (E) Consensus matrix confirming sample separation into two robust immune subtypes (subtype 1 and subtype 2). (F) ESTIMATE score comparison between immune hot and cold subtypes (StromalScore, ImmuneScore, ESTIMATEScore). Subtype 1 represents hot tumors, and subtype 2 represents cold tumors. Hot tumors show significantly higher immune infiltration. (G) KM overall survival analysis demonstrating significantly better prognosis in immune hot tumors compared to cold tumors (P=0.004, log-rank test). ***, P<0.001. CDF, cumulative distribution function; CGGA, Chinese Glioma Genome Atlas; GBM, glioblastoma; KM, Kaplan-Meier; ssGSEA, single-sample gene set enrichment analysis; TCGA, The Cancer Genome Atlas; TIME, tumor immune microenvironment.

DEGs and functional pathway enrichment between hot and cold tumors

DEGs across hot and cold tumors were identified via the limma package. Markedly upregulated genes were predominantly distributed on the right [log₂fold change (FC) >2], indicating enhanced expression in hot tumors, likely linked to immune activation. In contrast, significantly downregulated genes were concentrated on the left (log₂FC <−2), suggesting immune suppression in cold tumors (Figure S1). Functional enrichment analysis based on GO and KEGG pathways revealed distinct functional profiles between DEGs across hot and cold tumors (Figure 2). Upregulated genes in hot tumors were primarily enriched in immune-related pathways like “immune response”, “inflammatory response”, and “cell surface receptor signaling”, whereas downregulated genes in cold tumors were related to “extracellular matrix remodeling”. Cellular component analysis indicated that in hot tumors, genes were significantly enriched in the “plasma membrane” and “cell surface”, consistent with immune cell interactions, while in cold tumors, genes were predominantly localized to the “collagenous extracellular matrix”, possibly reflecting tumor stromal fibrosis. Molecular function analysis revealed marked upregulation of “cytokine activity” in hot tumors, while in cold tumors, “serine-type endopeptidase activity” and “pattern recognition receptor activity” were suppressed, potentially impairing innate immune response capabilities. KEGG pathway analysis further revealed that in hot tumors, “cytokine-receptor interaction” was significantly activated, indicative of an immune-activated microenvironment, whereas cold tumors exhibited immune suppression through extracellular matrix remodeling and various signaling pathways.

Figure 2 Functional enrichment analysis of differentially expressed genes between immune hot and cold GBM subtypes. The chart displays the top enriched GO terms for DEGs identified between immune hot and cold GBM subtypes. DEG, differentially expressed gene; GBM, glioblastoma; GO, Gene Ontology.

Prognostic model development and validation

The module genes found through WGCNA, namely the MEblack and MEblue modules, showed a notable positive link to the Hot phenotype (r=0.63 and 0.61, P=1e−60 and 2e−56, respectively), suggesting the potential involvement of genes promoting immune activation (Figure 3A-3C). The intersection of these two immune-related core modules with DEGs between the Hot and Cold groups resulted in 527 common genes (Figure 3D). Using Cox regression analysis, candidate genes significantly related to immune hot-cold phenotypes were identified (Figure S2).

Figure 3 WGCNA identifies immune-related modules in GBM. (A) Scale-free topology analysis for WGCNA parameter selection. The left panel shows the scale-free fit index (R2) as a function of soft-threshold power [1–30], with R2>0.8 indicating appropriate network construction. The right panel displays mean connectivity versus soft-threshold power. (B) Hierarchical clustering dendrogram of genes with module assignment colors. Gene clusters are color-coded into distinct co-expression modules. (C) Module-trait correlation heatmap showing relationships between WGCNA module eigengenes (rows) and immune subtypes (hot/cold, columns). The MEblack (r=0.63, P=1e−60) and MEblue (r=0.61, P=2e−56) modules show the strongest positive correlations with the Hot phenotype. (D) Venn diagram illustrating the intersection between DEGs from immune subtype comparison and genes from WGCNA immune-related modules. The overlap comprises 527 genes used for subsequent prognostic modeling. DEG, differentially expressed gene; GBM, glioblastoma; ME, module eigengenes; WGCNA, weighted gene co-expression network analysis.

A prognostic model was constructed by integrating 101 ML algorithms, with the StepCox[both] + SuperPC combination model yielding the best performance (C-index =0.61–0.72), significantly outperforming single models (such as LASSO, RSF) and other combinations (such as CoxBoost + Enet, Figure 4A). The “StepCox[both] + SuperPC” model was selected as the final prognostic model. Its efficacy in risk stratification was evaluated across multiple independent cohorts (TCGA, CGGA_325, CGGA_693, GSE108474). In the TCGA cohort, the high-risk cohort (n=72) exhibited notably lower median survival compared to the low-risk group (n=73) [hazard ratio (HR) =1.65, P=0.006], confirming the relevance of risk stratification. The survival rate was significantly lower in the high-risk group across all cohorts (P<0.05), validating the model’s cross-cohort consistency (Figure 4B). The model demonstrated robust predictive ability, with AUCs of 0.71/0.65/0.72 (TCGA), 0.58/0.62/0.62 (GSE108474), 0.58/0.63/0.61 (CGGA-693), and 0.55/0.69/0.69 (CGGA-325) for 1-/2-/3-year survival. This indicates the model’s overall stability, with high discrimination for short-term prognosis and relative stability for long-term prediction (Figure 4C). Furthermore, ROC curve analysis revealed AUC values >0.6 in most cohorts, underscoring the clinical relevance of the model (Figure 4D-4F).

Figure 4 Construction and validation of the prognostic immune-related gene signature for GBM. (A) Performance comparison of 101 machine learning algorithm combinations based on concordance index (C-index) values. The StepCox[both] + SuperPC combination achieved the highest predictive performance (C-index =0.61–0.72) and was selected as the final model. (B) KM survival curves for high-risk and low-risk groups across four independent cohorts (TCGA, CGGA-325, CGGA-693, GSE108474). High-risk patients show significantly worse OS in all cohorts (P<0.05, log-rank test). (C) Time-dependent AUC values for 1-, 2-, and 3-year survival prediction in the training (TCGA) and validation cohorts. The model demonstrates robust discrimination with AUCs of 0.71/0.65/0.72 (TCGA), 0.58/0.62/0.62 (GSE108474), 0.58/0.63/0.61 (CGGA-693), and 0.55/0.69/0.69 (CGGA-325). (D-F) ROC curves illustrating the prognostic accuracy of the model at 1-, 2-, and 3-year time points in validation cohorts. AUC values exceed 0.6 in most cohorts, confirming the model’s clinical relevance in prognostic stratification. AUC, area under the curve; CGGA, Chinese Glioma Genome Atlas; CI, confidence interval; GBM, glioblastoma; KM, Kaplan-Meier; OS, overall survival; ROC, receiver operating characteristic; TCGA, The Cancer Genome Atlas.

Nomogram construction, validation, and stratification analysis

The nomogram built on multivariate Cox regression indicated that the survival rates at 1, 3, and 5 years displayed a negative relation to the accumulation of every variable, offering clinicians an intuitive tool for prognostic assessment and individualized prediction of OS in GBM patients (Figure 5A). The prediction curves for 1-, 3-, and 5-year survival rates closely approached the ideal diagonal line, especially for the 5-year survival rate, where the prediction error was significantly minimized (root mean square error <0.1), demonstrating the high consistency of the nomogram’s predictions with actual observations and its strong clinical applicability (Figure 5B). ROC analysis further validated the reliability of the nomogram in time-dependent predictions (Figure 5C). Patients were stratified into four groups based on the “Hot/Cold” immune phenotype and “High/Low” risk scores: cold-high, hot-high, cold-low, and hot-low. Survival analysis revealed the lowest survival rate in the Cold-High cohort and the highest in the hot-low cohort, with significant survival differences across all subgroups (P=0.001), highlighting the interaction between immune microenvironment and molecular risk (Figure 5D). The distribution of tumor risk groups across the entire cohort showed that 44% of the low-risk hot-cold tumor patients may benefit from immunotherapy, whereas 37% with high-risk cold tumors align with the clinical features of immune-suppressive phenotypes in GBM (Figure 5E). These distribution proportions underscore the urgency of developing new therapies targeting cold tumors and provide data support for stratified treatment strategies.

Figure 5 Nomogram development and combined risk-immune stratification in GBM. (A) Visual nomogram integrating clinical-pathological variables and the prognostic model risk score for individualized survival probability prediction at 1-, 3-, and 5-year time points. (B) Calibration curves comparing observed versus nomogram-predicted OS at 1, 3, and 5 years. The close alignment along the diagonal indicates high predictive accuracy. (C) Time-dependent ROC curves validating the nomogram’s discriminative ability for 1-, 3-, and 5-year survival (AUC: 0.599, 0.662, and 0.663, respectively). (D) KM survival analysis of four combined subgroups based on immune phenotype (hot/cold) and risk score (high/low). The cold-high subgroup exhibits the worst prognosis, while the hot-low subgroup shows the most favorable survival (P=0.001, log-rank test). (E) Sankey diagram visualizing patient flow from initial risk stratification to combined subgroup classification, illustrating distribution proportions across subgroups. The diagram highlights that 44% of low-risk hot/cold tumor patients may benefit from immunotherapy, while 37% with high-risk cold tumors align with clinically immunosuppressive phenotypes. *, P<0.05; **, P<0.01. AUC, area under the curve; GBM, glioblastoma; KM, Kaplan-Meier; OS, overall survival; ROC, receiver operating characteristic.

Gene screening and immune correlation analysis

Gene screening using LASSO regression, SVM-RFE, and RF (Figure 6A-6E) identified seven genes significantly related to OS in patients: Cardiotrophin-like cytokine factor 1 (CLCF1), Integrin-binding sialoprotein (IBSP), Podoplanin (PDPN), Serpin peptidase inhibitor clade A member 5 (SERPINA5), Solute carrier family 11 member 1 (SLC11A1), Transmembrane protein 176A (TMEM176A), and Tumor necrosis factor superfamily member 14 (TNFSF14), ensuring the robustness and biological relevance of the findings (Figure 6F). ROC analysis of these seven genes demonstrated high efficacy in distinguishing between immune cold and hot GBM types. The AUC values for CLCF1, IBSP, PDPN, SERPINA5, SLC11A1, TMEM176A, and TNFSF14 were 0.792, 0.797, 0.752, 0.749, 0.878, 0.755, and 0.817, respectively, indicating their strong discriminatory power in classifying Cold and Hot tumors (Figure 7).

Figure 6 ML-based feature selection and validation of prognostic IRGs. (A) LASSO coefficient profiles showing coefficient variation paths across different penalty parameters (λ). Vertical dashed lines indicate optimal λ selected through 10-fold cross-validation. (B) LASSO coefficient shrinkage plot demonstrating the relationship between log(λ) and cross-validation error. Dotted lines mark the optimal λ value corresponding to the minimal cross-validation error. (C) SVM-RFE recursive feature elimination curve illustrating feature ranking based on cross-validation accuracy. The red dot indicates the optimal feature subset with maximum predictive performance. (D) RF variable importance ranking plot displaying the mean decrease in Gini index for each gene. Top-ranked genes show the highest contribution to classification accuracy. (E) Cumulative feature importance curve from RF analysis, showing the proportion of total importance captured by the top-ranked genes. (F) Venn diagram intersection of genes identified by LASSO, SVM-RFE, and RF algorithms. The overlapping region contains seven core prognostic genes. IRG, immune-related gene; LASSO, least absolute shrinkage and selection operator; ML, machine learning; RF, random forest; RMSE, root mean square error; SVM-RFE, support vector machine-recursive feature elimination.

Figure 7 ROC curves and corresponding AUC values for seven core IRGs in distinguishing between immune hot and cold GBM subtypes. (A) CLCF1; (B) IBSP; (C) PDPN; (D) SERPINA5; (E) SLC11A1; (F) TMEM176A; (G) TNFSF14. All seven genes demonstrate strong discriminative ability. The dashed diagonal line represents random classification (AUC =0.5). AUC, area under the curve; GBM, glioblastoma; IRG, immune-related gene; ROC, receiver operating characteristic.

Single-cell transcriptomic profiling and IHC validation

Five single-cell RNA-seq datasets for GBM from the TISCH database (GSE103224, GSE139448, GSE148842, GSE84465, GSE89567) were integrated to investigate the expression of these genes at the single-cell level. The expression patterns of seven genes (CLCF1, IBSP, PDPN, SERPINA5, SLC11A1, TMEM176A, TNFSF14) were systematically assessed across different cell subpopulations. CLCF1 exhibited high expression in malignant, AC-like malignant, and oligodendrocyte cells; IBSP and SLC11A1 were overexpressed in Mono/Macro cells; PDPN was upregulated in Mono/Macro, malignant, and AC-like malignant cells. SERPINA5 displayed exclusive high expression in Endothelial cells, whereas TMEM176A was prominent in Astrocyte and Mono/Macro populations. TNFSF14 exhibited elevated expression in Mono/Macro, AC-like malignant, and OC-like malignant cells (Figure 8). The high expression of IBSP, SLC11A1, PDPN, and TMEM176A in Mono/Macro cells suggests their involvement in tumor progression through matrix remodeling, immune suppression, and cellular interactions, highlighting the central role of TAMs in GBM progression.

Figure 8 Single-cell expression profiles of core IRGs across GBM microenvironment. Plots showing normalized expression levels of seven core IRGs across different cell populations in five independent single-cell RNA sequencing datasets (GSE103224, GSE139448, GSE148842, GSE84465, GSE89567) from TISCH. (A) CLCF1: highly expressed in malignant cells, AC-like malignant cells, and oligodendrocytes. (B) IBSP: predominantly expressed in Mono/Macro lineages. (C) PDPN: upregulated in monocyte/macrophage (Mono/Macro) populations and malignant cell subtypes. (D) SERPINA5: specifically enriched in endothelial cells. (E) SLC11A1: overexpressed in Mono/Macro populations, consistent with its role in macrophage-mediated immune regulation. (F) TMEM176A: highly expressed in astrocytes and Mono/Macro cells. (G) TNFSF14: elevated in Mono/Macro, AC-like malignant, and OC-like malignant cells. Expression values are log-transformed [log(TPM/10+1)]. Color intensity indicates the average expression level. AC, astrocyte; GBM, glioblastoma; IRG, immune-related gene; OC, oligodendrocyte; TISCH, Tumor Immune Single-cell Hub; TPM, transcript per million.

Additionally, IHC images were obtained from the HPA database for the proteins encoded by CLCF1, PDPN, SERPINA5, SLC11A1, TMEM176A, and TNFSF14 (Figure 9). These images showed the differential expression across cancerous and normal tissues. The staining intensity of CLCF1, PDPN, SERPINA5, SLC11A1, TMEM176A, and TNFSF14 was significantly stronger in tumor tissues compared to normal tissues, suggesting a potential association with immune cell infiltration, particularly macrophages.

Figure 9 Immunohistochemical validation of core gene protein expression in GBM and normal brain tissues. Immunohistochemistry images from the Human Protein Atlas (HPA) database compare protein expression of six core genes between GBM and normal brain tissues under HPA licensing terms. (A,B) CLCF1: moderate (normal, image available from https://www.proteinatlas.org/ENSG00000175505-CLCF1/tissue/brain#img) vs. strong (GBM, image available from https://www.proteinatlas.org/ENSG00000175505-CLCF1/cancer/glioma#img) expression. (C,D) PDPN: weak focal (normal, image available from https://www.proteinatlas.org/ENSG00000162493-PDPN/tissue/brain#img) vs. intense (GBM, image available from https://www.proteinatlas.org/ENSG00000162493-PDPN/cancer/glioma#img) staining. (E,F) SERPINA5: minimal (normal, image available from https://www.proteinatlas.org/ENSG00000188488-SERPINA5/tissue/brain#img) vs. prominent (GBM, image available from https://www.proteinatlas.org/ENSG00000188488-SERPINA5/cancer/glioma#img) expression. (G,H) SLC11A1: basal (normal, image available from https://www.proteinatlas.org/ENSG00000018280-SLC11A1/tissue/brain#img) vs. upregulated (GBM, image available from https://www.proteinatlas.org/ENSG00000018280-SLC11A1/cancer/glioma#img) in tumor cells. (I,J) TMEM176A: weak diffuse (normal, image available from https://www.proteinatlas.org/ENSG00000002933-TMEM176A/tissue/brain#img) vs. strong (GBM, image available from https://www.proteinatlas.org/ENSG00000002933-TMEM176A/cancer/glioma#img) expression. (K,L) TNFSF14: low (normal, image available from https://www.proteinatlas.org/ENSG00000125735-TNFSF14/tissue/brain#img) vs. intense (GBM, image available from https://www.proteinatlas.org/ENSG00000125735-TNFSF14/cancer/glioma#img) staining. The scale bar indicates 200 µm. GBM, glioblastoma; HPA, Human Protein Atlas; IHC, immunohistochemistry.

Drug sensitivity analysis

Using the pRRophetic algorithm to predict drug sensitivity, our study revealed that high-risk GBM showed significantly lower half maximal inhibitory concentration (IC50) values for a range of drugs, including AKT inhibitor VIII, AP-24534, AZ628, bexarotene, bortezomib, bryostatin1, CGP-60474, crizotinib, cyclopamine, JQ12, LY317615, pazopanib, PHA-665752, phenformin, roscovitine, salubrinal, sorafenib, TGX221, WH-4-023, XMD8-85, as well as Z-LLNle-CHO, compared to low-risk GBM (P<2.2e–16) (Figure 10A-10U). Conversely, low-risk GBM exhibited significantly lower IC50 values for BMS-754807 and listinib (P<2.2e−16) (Figure 10V,10W). The drug sensitivity profile of high-risk GBM suggests that its progression may involve activation of the AKT/mTOR pathway, protein homeostasis imbalance, and angiogenesis, while low-risk GBM progression may be linked to metabolic or growth factor dependency.

Figure 10 Differential drug sensitivity profiles between high- and low-risk GBM groups. Box plots show the comparison of predicted IC50 values for 23 therapeutic compounds between high- and low-risk GBM groups, generated using the pRRophetic algorithm. (A-U) High-risk GBM patients show significantly lower IC₅₀ values (P<2.2e−16) for 21 drugs including AKT inhibitor VIII, AP-24534, AZ628, bexarotene, bortezomib, bryostatin 1, CGP-60474, crizotinib, cyclopamine, JQ12, LY317615, pazopanib, PHA-665752, phenformin, roscovitine, salubrinal, sorafenib, TGX221, WH-4-023, XMD8-85, and Z-LLNle-CHO, indicating higher drug sensitivity. (V,W) Low-risk GBM patients exhibit significantly lower IC₅₀ values (P<2.2e−16) for BMS-754807 and listinib, suggesting potential treatment responsiveness. Boxes represent interquartile range, horizontal lines indicate medians. GBM, glioblastoma; IC₅₀, half-maximal inhibitory concentration.

Discussion

Our study developed a prognostic prediction model for GBM based on IRGs through multi-omics data and ML algorithms. The model uncovered molecular characteristics and treatment response differences between cold and hot tumor subtypes. The study not only validated the central role of the immune microenvironment in GBM progression but also identified a set of key genetic biomarkers, providing new insights for GBM precision stratification and personalized treatment.

In comparison with previous studies, the innovation of this research lies in several key aspects: First, through the integration of multi-cohort data (TCGA, CGGA, GSE108474) and rigorous batch correction, the generalizability of the model was significantly enhanced. Second, combining WGCNA module analysis with the selection of 101 ML algorithms overcame the limitations of traditional univariate regression, ensuring the robustness of the genetic biomarkers. The final selection of seven core genes (CLCF1, IBSP, PDPN, SERPINA5, SLC11A1, TMEM176A, TNFSF14) exhibited high AUC values (0.75–0.88) in distinguishing cold and hot tumor subtypes, outperforming existing markers like IDH mutations or MGMT methylation. Furthermore, the model demonstrated stable predictive performance in external validation sets (AUC =0.55–0.69), providing a reliable tool for clinical prognostic evaluation. The study also further validated gene-specific expression patterns through single-cell transcriptomics and IHC, offering new directions for GBM mechanistic research and targeted interventions.

This study identified seven core genes—CLCF1, IBSP, PDPN, SERPINA5, SLC11A1, TMEM176A, and TNFSF14—that potentially exhibit synergistic effects in the immune microenvironment of GBM:

CLCF1, as a neurotrophic factor and a member of the IL-6 cytokine family, primarily participates in B-cell differentiation and plasma cell survival in immune regulation, and modulates signal transduction within the neuroimmune network (18). In glioma, CLCF1 is transcriptionally activated by BRD4 and promotes tumor cell proliferation and invasion through autocrine or paracrine activation of the STAT3 signaling pathway. Concurrently, it remodels the tumor microenvironment and induces infiltration of immunosuppressive cells, thereby facilitating immune evasion (19,20). Previous studies have identified CLCF1 as a marker of poor prognosis in glioma, and targeting CLCF1 may represent a potential therapeutic strategy by reversing the immunosuppressive microenvironment (21,22). In the present study, CLCF1 expression was significantly upregulated in glioma tissues, further confirming its pivotal oncogenic role in glioma malignant progression.

IBSP encodes a bone matrix protein that primarily activates integrin signaling pathways through binding to the integrin αV receptor. It may participate in tumor microenvironment remodeling in the context of immune regulation, but its direct role in immune modulation remains unclear (23,24). Previous studies have demonstrated that high IBSP expression is associated with poor prognosis in osteosarcoma, breast cancer, and prostate cancer (23,25). This association may be attributable to its ability to enhance cell adhesion and migration, promote tumor invasion and metastasis, and interact with other signaling molecules (such as growth factors and extracellular matrix proteins) to activate pro-tumorigenic pathways (26,27). In GBM, elevated IBSP expression promotes tumor cell migration, proliferation, and mesenchymal transition via activation of integrin signaling, and is closely correlated with poor prognosis in patients with IDH-wildtype tumors (24,28). Although the precise regulatory mechanisms of IBSP within the glioma immune microenvironment remain to be clarified, extrapolation from its roles in other malignancies suggests that it may contribute to tumor immune evasion by modulating immune cell infiltration or function. In this study, IBSP expression was significantly upregulated in glioma tissues, further supporting its oncogenic role in glioma progression. Nevertheless, its potential immunoregulatory functions warrant further investigation.

PDPN encodes a transmembrane glycoprotein that is predominantly expressed in tumor-associated fibroblasts, macrophages, and certain tumor cells. In immune regulation, PDPN binds via its extracellular domain to C-type lectin-like receptor 2, inducing platelet aggregation and forming a physical barrier that facilitates tumor immune escape. Additionally, it modulates immune cell function within the tumor microenvironment, such as promoting macrophage polarization toward an immunosuppressive M2 phenotype and inhibiting T-cell proliferation and activity (29-31). In GBM, high PDPN expression is significantly associated with increased tumor invasiveness, radioresistance, angiogenesis, and poor prognosis, particularly in mesenchymal-like tumor cells. PDPN drives tumor invasion through activation of downstream signaling pathways and promotes tumor progression and immune evasion by extensively remodeling the immunosuppressive microenvironment (32-34). Therefore, PDPN not only directly contributes to GBM malignant progression but has also emerged as a potential therapeutic target; nevertheless, its precise mechanisms of action require further validation (34,35). In this study, PDPN expression was significantly upregulated in glioma tissues, further confirming its critical oncogenic role in glioma progression and immune escape.

SERPINA5 encodes a member of the serine protease inhibitor family, which primarily exerts anti-inflammatory effects and maintains vascular integrity by inhibiting proteases involved in coagulation and inflammatory responses, such as activated protein C and thrombin (36,37). Elevated SERPINA5 expression is significantly associated with poor patient prognosis (38). In GBM, SERPINA5 has been identified as an angiogenesis-related prognostic biomarker and an independent predictor of survival outcomes (33,39). It may promote tumor progression through regulation of coagulation, inflammation, and the tumor microenvironment (33,39,40). This study revealed that SERPINA5 expression was significantly upregulated in glioma tissues, further supporting its role as a potential prognostic biomarker in glioma progression. However, its complex biological functions and regulatory mechanisms within the immune microenvironment remain to be comprehensively elucidated.

SLC11A1 is primarily expressed in macrophages and neutrophils. As a divalent metal ion transporter, it regulates intracellular iron and manganese homeostasis and plays a pivotal role in innate immune responses and immunometabolic reprogramming by influencing reactive oxygen species production and inflammasome activation (41). In GBM, SLC11A1 has been identified as a key gene in ferroptosis-related prognostic models and immune microenvironment regulation. Its high expression is significantly associated with poor prognosis, resistance to immunotherapy, and immunosuppressive features (42-44). Moreover, SLC11A1 promotes GBM progression by enhancing tumor cell proliferation and epithelial-mesenchymal transition. Nevertheless, its specific regulatory mechanisms within the glioma immune microenvironment require further experimental validation (45). The present study showed significantly upregulated SLC11A1 expression in glioma tissues, further confirming its critical oncogenic role in glioma malignant progression and the establishment of an immunosuppressive microenvironment.

TMEM176A, a member of the transmembrane protein 176 family and a cation channel protein, is predominantly expressed in dendritic cells and macrophages. It participates in inflammatory responses and immune tolerance by regulating intracellular calcium homeostasis, thereby playing a crucial role in innate immune regulation (46). In GBM, TMEM176A promotes tumor cell proliferation and inhibits apoptosis through activation of the extracellular signal-regulated kinase 1/2 signaling pathway. Its high expression is significantly associated with poor patient prognosis and has been identified as one of the core genes for constructing prognostic models (47,48). In the context of tumor immunity, TMEM176A may indirectly modulate the immune microenvironment by influencing the polarization status of TAMs. Nevertheless, its precise immunoregulatory mechanisms in glioma remain to be fully elucidated (46,49). In the present study, TMEM176A expression was significantly upregulated in glioma tissues, further validating its critical oncogenic role in tumor malignant progression, although its regulatory effects on the immune microenvironment warrant further in-depth investigation.

TNFSF14, a member of the tumor necrosis factor superfamily, exerts pleiotropic functions in immune regulation. It is primarily expressed in activated T cells, natural killer cells, and immature dendritic cells, and bidirectionally modulates immune responses—enhancing T-cell effector functions while also transmitting inhibitory signals to maintain immune homeostasis (50,51). In GBM, TNFSF14 is highly expressed and is significantly associated with IDH-wildtype status, the mesenchymal subtype, and poor patient prognosis, suggesting its potential value as a prognostic biomarker (52-57). Mechanistically, TNFSF14 promotes tumor progression and immune evasion by remodeling the immunosuppressive microenvironment, synergizing with immune checkpoint molecules such as programmed cell death protein 1 (PD-1)/PD-L1 to suppress antitumor immunity, and facilitating aberrant angiogenesis (24,53,58-61). In this study, TNFSF14 expression was significantly upregulated in glioma tissues, further confirming its pivotal oncogenic role in glioma malignant progression and immune escape. Nevertheless, the precise mechanisms underlying its bidirectional immunomodulatory functions require further elucidation to inform the development of precision immunotherapeutic strategies.

While these genes may have potential roles in the GBM immune microenvironment, their exact mechanisms require further experimental validation.

The risk stratification model divides patients into four groups (cold-high, hot-high, cold-low, hot-low), with the cold-high risk group having the worst prognosis and the hot-low risk group exhibiting a significant survival advantage. This stratification not only provides a refined tool for prognostic assessment but also guides treatment strategies.

There are certain limitations in this study. First, retrospective analysis may introduce selection bias, requiring prospective cohort validation of the model’s clinical applicability. Second, the specific regulatory mechanisms of the key genes need to be explored through further in vivo and in vitro experiments. Moreover, drug sensitivity prediction based on computational models must be validated through in vitro drug sensitivity assays and clinical trials. Future research could focus on the following directions: (I) further investigation of the specific regulatory mechanisms of key genes in GBM; (II) integration of spatial transcriptomics to analyze the spatial heterogeneity of cold/hot tumors; (III) design of clinical trials based on risk stratification to evaluate the efficacy of individualized immunotherapy regimens.

Conclusions

This study constructed an immune-related prognostic model for GBM through multi-omics integration and ML, revealing the molecular characteristics and treatment response differences between cold and hot tumor subtypes. The seven identified core genes provide important targets for GBM mechanistic research, prognostic evaluation, and precision treatment. Future research should further investigate the regulatory mechanisms of these key genes in GBM, advancing translational efforts and applying theoretical findings to clinical practice to improve survival outcomes for the GBM population.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2391/rc

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2391/prf

Funding: This research was supported by a horizontal research project from Chengdu Maipu Medical Technology Development Co., Ltd. (No. 2502261). The funder provided financial and technical support but was not involved in data analysis or manuscript preparation.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2391/coif). All authors report that this research was supported by a horizontal research project from Chengdu Maipu Medical Technology Development Co., Ltd. (No. 2502261). The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

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. Lim M, Xia Y, Bettegowda C, et al. Current state of immunotherapy for glioblastoma. Nat Rev Clin Oncol 2018;15:422-42. [Crossref] [PubMed]
2. Lan Z, Li X, Zhang X. Glioblastoma: An Update in Pathology, Molecular Mechanisms and Biomarkers. Int J Mol Sci 2024;25:3040. [Crossref] [PubMed]
3. Królikowska K, Błaszczak K, Ławicki S, et al. Glioblastoma-A Contemporary Overview of Epidemiology, Classification, Pathogenesis, Diagnosis, and Treatment: A Review Article. Int J Mol Sci 2025;26:12162. [Crossref] [PubMed]
4. Erices JI, Bizama C, Niechi I, et al. Glioblastoma Microenvironment and Invasiveness: New Insights and Therapeutic Targets. Int J Mol Sci 2023;24:7047. [Crossref] [PubMed]
5. Li J, Moqbel AQ, Wang Y, et al. Unraveling the glioblastoma (GBM) tumor microenvironment: future perspective on targeted immunotherapy. Biochem Pharmacol 2026;246:117724. [Crossref] [PubMed]
6. Huang B, Li X, Li Y, et al. Current Immunotherapies for Glioblastoma Multiforme. Front Immunol 2020;11:603911. [Crossref] [PubMed]
7. Hansen LJ, Jackson CM. The glioma microenvironment and its impact on antitumor immunity. Neurooncol Adv 2025;7:iv19-31. [Crossref] [PubMed]
8. Agosti E, Zeppieri M, De Maria L, et al. Glioblastoma Immunotherapy: A Systematic Review of the Present Strategies and Prospects for Advancements. Int J Mol Sci 2023;24:15037. [Crossref] [PubMed]
9. Torky RF, Makboul R, Badary DM, et al. Dual biomarker role of PD-L1 and LC3B in glioblastoma: prognostic and therapeutic potential. Neurosurg Rev 2026;49:153. [Crossref] [PubMed]
10. Obrador E, Moreno-Murciano P, Oriol-Caballo M, et al. Glioblastoma Therapy: Past, Present and Future. Int J Mol Sci 2024;25:2529. [Crossref] [PubMed]
11. Arrieta VA, Dmello C, McGrail DJ, et al. Immune checkpoint blockade in glioblastoma: from tumor heterogeneity to personalized treatment. J Clin Invest 2023;133:e163447. [Crossref] [PubMed]
12. Gómez EG, Morales MAM, San-Juan D, et al. Efficacy and safety of immune checkpoint inhibitors and mTOR inhibitors as targeted therapy for glioblastoma: A systematic review and meta-analysis of randomized clinical trials. Neurosurg Rev 2026;49:95. [Crossref] [PubMed]
13. Tang M, Xu P, Zhang X. Advancing glioblastoma immunotherapy: Molecular pathways and innovative therapeutic strategies. Pathol Res Pract 2026;278:156326. [Crossref] [PubMed]
14. Ijaz M, Tan Q, Yan Y, et al. Overcoming barriers in glioblastoma: The potential of CAR T cell immunotherapy. Theranostics 2025;15:7090-126. [Crossref] [PubMed]
15. Tan AC, Ashley DM, López GY, et al. Management of glioblastoma: State of the art and future directions. CA Cancer J Clin 2020;70:299-312. [Crossref] [PubMed]
16. Kakde GS, Dakal TC, Maurya PK. Emerging multi-omics biomarkers in glioblastoma: Integrative insights from genomics to metabolomics. Biochim Biophys Acta Rev Cancer 2026;1881:189540. [Crossref] [PubMed]
17. Rajakaruna P, Rios S, Elnahas H, et al. Molecular Biomarkers of Glioma. Biomedicines 2025;13:1298. [Crossref] [PubMed]
18. Jones SA, Jenkins BJ. Recent insights into targeting the IL-6 cytokine family in inflammatory diseases and cancer. Nat Rev Immunol 2018;18:773-89. [Crossref] [PubMed]
19. Yuan J, Liu H, Wen J, et al. Cardiotrophin-like cytokine factor 1 regulated by Sry-related HMG-box transcription factor 9 promotes malignant behavior and immune evasion of glioblastoma multiforme. J Neuropathol Exp Neurol 2026;nlaf161. [Crossref] [PubMed]
20. Fang X, Wu X, Xu C. Characterization of hypoxia-related molecular clusters and prognostic riskScore for glioma. Front Oncol 2025;15:1605949. [Crossref] [PubMed]
21. Jiang Y, Ji Q, Long X, et al. CLCF1 Is a Novel Potential Immune-Related Target With Predictive Value for Prognosis and Immunotherapy Response in Glioma. Front Immunol 2022;13:810832. [Crossref] [PubMed]
22. Liu X, Liu X. A novel immune-related gene prognostic signature combining immune cell infiltration and immune checkpoint for glioblastoma patients. Transl Cancer Res 2024;13:6136-53. [Crossref] [PubMed]
23. Gordon JA, Sodek J, Hunter GK, et al. Bone sialoprotein stimulates focal adhesion-related signaling pathways: role in migration and survival of breast and prostate cancer cells. J Cell Biochem 2009;107:1118-28. [Crossref] [PubMed]
24. Ghochani Y, Muthukrishnan SD, Sohrabi A, et al. A molecular interactome of the glioblastoma perivascular niche reveals integrin binding sialoprotein as a mediator of tumor cell migration. Cell Rep 2022;41:111511. [Crossref] [PubMed]
25. Ma Y, Chen B, Zhang B, et al. High expression of integrin-binding sialoprotein (IBSP) is associated with poor prognosis of osteosarcoma. Aging (Albany NY) 2023;16:28-42. [Crossref] [PubMed]
26. Hu L, Liu J, Xue H, et al. miRNA-92a-3p regulates osteoblast differentiation in patients with concomitant limb fractures and TBI via IBSP/PI3K-AKT inhibition. Mol Ther Nucleic Acids 2021;23:1345-59. [Crossref] [PubMed]
27. Genchi GG, Sinibaldi E, Ceseracciu L, et al. Ultrasound-activated piezoelectric P(VDF-TrFE)/boron nitride nanotube composite films promote differentiation of human SaOS-2 osteoblast-like cells. Nanomedicine 2018;14:2421-32. [Crossref] [PubMed]
28. Lu CH, Wei ST, Liu JJ, et al. Recognition of a Novel Gene Signature for Human Glioblastoma. Int J Mol Sci 2022;23:4157. [Crossref] [PubMed]
29. Hwang BO, Park SY, Cho ES, et al. Platelet CLEC2-Podoplanin Axis as a Promising Target for Oral Cancer Treatment. Front Immunol 2021;12:807600. [Crossref] [PubMed]
30. Wang Y, Peng D, Huang Y, et al. Podoplanin: Its roles and functions in neurological diseases and brain cancers. Front Pharmacol 2022;13:964973. [Crossref] [PubMed]
31. Suzuki H, Kaneko MK, Kato Y. Roles of Podoplanin in Malignant Progression of Tumor. Cells 2022;11:575. [Crossref] [PubMed]
32. Modrek AS, Eskilsson E, Ezhilarasan R, et al. PDPN marks a subset of aggressive and radiation-resistant glioblastoma cells. Front Oncol 2022;12:941657. [Crossref] [PubMed]
33. Wan Z, Zuo X, Wang S, et al. Identification of angiogenesis-related genes signature for predicting survival and its regulatory network in glioblastoma. Cancer Med 2023;12:17445-67. [Crossref] [PubMed]
34. Sharma B, Agriantonis G, Shafaee Z, et al. Role of Podoplanin (PDPN) in Advancing the Progression and Metastasis of Glioblastoma Multiforme (GBM). Cancers (Basel) 2024;16:4051. [Crossref] [PubMed]
35. Lei J, Liu S, Zuo M, et al. PDPN is associated with prognosis and immune heterogeneity of glioblastomas. Asian J Surg 2024;47:4763-5. [Crossref] [PubMed]
36. Hayashi T, Suzuki K. Gene organization of human protein C inhibitor, a member of SERPIN family proteins encoded in five exons. Int J Hematol 1993;58:213-24. [PubMed]
37. Jing Y, Jia D, Wong CM, et al. SERPINA5 inhibits tumor cell migration by modulating the fibronectin-integrin β1 signaling pathway in hepatocellular carcinoma. Mol Oncol 2014;8:366-77. [Crossref] [PubMed]
38. Fan M, Xiong X, Han L, et al. SERPINA5 promotes tumour cell proliferation by modulating the PI3K/AKT/mTOR signalling pathway in gastric cancer. J Cell Mol Med 2022;26:4837-46. [Crossref] [PubMed]
39. Zhou M, Deng Y, Fu Y, et al. A new prognostic model for glioblastoma multiforme based on coagulation-related genes. Transl Cancer Res 2023;12:2898-910. [Crossref] [PubMed]
40. Cao M, Chen J, Guo RZ. Evaluating the Predictive Value of a Coagulation-Related Gene Model in Glioma. Turk Neurosurg 2024;34:708-15. [PubMed]
41. Banerjee R, Bintee B, Manickasamy MK, et al. The solute carrier family 11 transporters: a bridge between iron homeostasis and tumor biology. Cell Commun Signal 2025;23:332. [Crossref] [PubMed]
42. Tian Y, Liu H, Zhang C, et al. Comprehensive Analyses of Ferroptosis-Related Alterations and Their Prognostic Significance in Glioblastoma. Front Mol Biosci 2022;9:904098. [Crossref] [PubMed]
43. Yan Z, Chu S, Zhu C, et al. Development of a T-cell activation-related module with predictive value for the prognosis and immune checkpoint blockade therapy response in glioblastoma. PeerJ 2021;9:e12547. [Crossref] [PubMed]
44. Wang Y, Ye S, Wu D, et al. Identification, and Experimental and Bioinformatics Validation of an Immune-Related Prognosis Gene Signature for Low-Grade Glioma Based on mRNAsi. Cancers (Basel) 2023;15:3238. [Crossref] [PubMed]
45. Li J, Wei Y, Liu J, et al. Integrative analysis of metabolism subtypes and identification of prognostic metabolism-related genes for glioblastoma. Biosci Rep 2024;44:BSR20231400. [Crossref] [PubMed]
46. Luo X, Luo B, Fei L, et al. MS4A superfamily molecules in tumors, Alzheimer's and autoimmune diseases. Front Immunol 2024;15:1481494. [Crossref] [PubMed]
47. Liu Z, An H, Song P, et al. Potential targets of TMEM176A in the growth of glioblastoma cells. Onco Targets Ther 2018;11:7763-75. [Crossref] [PubMed]
48. Wang Y, Ji L, Ji C, et al. Multi-omics approaches establishing histone modification based prognostic model in glioma patients and further verification of the carcinogenesis mechanism. Funct Integr Genomics 2023;23:307. [Crossref] [PubMed]
49. Zeng Y, Tan P, Ren C, et al. Comprehensive Analysis of Expression and Prognostic Value of MS4As in Glioma. Front Genet 2022;13:795844. [Crossref] [PubMed]
50. Hou Y, Wang Y, Chen J, et al. Dual Roles of Tumor Necrosis Factor Superfamily 14 in Antiviral Immunity. Viral Immunol 2022;35:579-85. [Crossref] [PubMed]
51. Skeate JG, Otsmaa ME, Prins R, et al. TNFSF14: LIGHTing the Way for Effective Cancer Immunotherapy. Front Immunol 2020;11:922. [Crossref] [PubMed]
52. Long S, Li M, Liu J, et al. Identification of immunologic subtype and prognosis of GBM based on TNFSF14 and immune checkpoint gene expression profiling. Aging (Albany NY) 2020;12:7112-28. [Crossref] [PubMed]
53. Yang Y, Lv W, Xu S, et al. Molecular and Clinical Characterization of LIGHT/TNFSF14 Expression at Transcriptional Level via 998 Samples With Brain Glioma. Front Mol Biosci 2021;8:567327. [Crossref] [PubMed]
54. Moreno DA, da Silva LS, Gomes I, et al. Cancer immune profiling unveils biomarkers, immunological pathways, and cell type score associated with glioblastoma patients' survival. Ther Adv Med Oncol 2022;14:17588359221127678. [Crossref] [PubMed]
55. Zhou Y, Qin X, Hu Q, et al. Cross-talk between disulfidptosis and immune check point genes defines the tumor microenvironment for the prediction of prognosis and immunotherapies in glioblastoma. Sci Rep 2024;14:3901. [Crossref] [PubMed]
56. Zottel A, Šamec N, Jovčevska I. TNFSF14 and CD44 are overexpressed in glioblastoma and associated with immunosuppressive microenvironment. Biomol Biomed 2025;25:1829-43. [Crossref] [PubMed]
57. Yu C, Xun M, Yu F, et al. An MHC-Related Gene's Signature Predicts Prognosis and Immune Microenvironment Infiltration in Glioblastoma. Int J Mol Sci 2025;26:4609. [Crossref] [PubMed]
58. Cao K, Jiang X, Wang B, et al. SAA1 Expression as a Potential Prognostic Marker of the Tumor Microenvironment in Glioblastoma. Front Neurol 2022;13:905561. [Crossref] [PubMed]
59. Han M, Sun Y, Zhao W, et al. Comprehensive characterization of TNFSF14/LIGHT with implications in prognosis and immunotherapy of human gliomas. Front Immunol 2022;13:1025286. [Crossref] [PubMed]
60. Ramachandran M, Vaccaro A, van de Walle T, et al. Tailoring vascular phenotype through AAV therapy promotes anti-tumor immunity in glioma. Cancer Cell 2023;41:1134-1151.e10. [Crossref] [PubMed]
61. Huang J, Tong S, Liu J, et al. Identification and validation of cuproptosis-related immune checkpoint expression for glioblastoma. BMC Cancer 2025;25:1723. [Crossref] [PubMed]