RNA-sequencing data from TCGA-663, CGGA-325, CGGA-693, and CPTAC-GBM cohorts with corresponding clinical information were obtained from The Cancer Genome Atlas (TCGA) database (version 28.0, https://portal.gdc.cancer.gov/), the Chinese Glioma Genome Atlas (CGGA) database (http://www.cgga.org.cn/index.jsp), and the Clinical Proteomic Tumor Analysis Consortium (CPTAC) (18). Transcriptional expression was evaluated using transcripts per million (TPM) and was further normalized to log2 (TPM + 1). The establishment of pyroptosis-related signature and internal validation was performed based on the TCGA-663 cohort, while the CGGA-325, CGGA-693, and CPTAC-GBM cohorts were used for external validation. The baseline clinical characteristics of glioma patients are shown in Supplementary Table 1. Mutation and copy number variation data were retrieved from the cBioPortal (http://www.cbioportal.org) for the analysis of driver gene mutations. Mutation data were downloaded from TCGA and visualized using the maftools package in R to identify the somatic mutation landscape in distinct pyroptosis-related subtypes.

To obtain consensus clustering of glioma patients based on genes related to both canonical and non-canonical pyroptosis as previously described (19) (Supplementary Table 2), unsupervised clustering was performed using ConsensusClusterPlus package in R based on 80% sample resampling for 10 repetitions (20). The optimal number of clusters is determined using an empirical cumulative distribution function plot.

To establish the pyroptosis-related risk signature (PRRS), univariate and multivariate Cox regression analyses were performed using survminer in the R package. Briefly, univariate Cox regression analysis was performed based on 26 pyroptosis-related genes for overall survival. Significant factors (P< 0.05) were then selected for multivariate Cox regression. The risk signature was further calculated based on gene expression and the coefficient in multivariate Cox regression analysis for each sample. Receiver operating characteristic (ROC) curve analysis was performed to assess the prognosis-predicted performance, and the area under the curve (AUC) was calculated using timeROC in the R package.

To explore the potential pathways associated with pyroptosis-related signatures, GSEA software (version 3.0) was obtained from the GSEA website (http://software.broadinstitute.org/gsea/index.jsp), and enrichment analysis was then performed based on differentially expressed genes (DEGs) between pyroptosis-related subgroups (Cluster 1 vs. Cluster 2; high-PRRS group vs. low-PRRS group, separated by the median PRRS value), with the minimum number of genes set to 5, the maximum number of genes set to 5000, and 1000 resampling; P values< 0.05, and FDR< 0.25 were considered statistically significant. The results were visualized using ggplot2 in the R package.

To evaluate immune characteristics, the immuneScore and stromalScore were calculated using the estimate package in R (21). The tumor mutation burden (TMB) score was calculated using maftools package in R (22) and the microsatellite instability (MSI) score was obtained from a previous study to evaluate genomic status (23). To assess the degree of oncogenic differentiation, stemness indices (mRNA expression-based stemness index, mRNAsi) for each sample were calculated using a one-class logistic regression machine learning algorithm as previously described (24). To characterize immune infiltration in glioma tissues, the immunedeconv R package, which contains six different algorithms, including TIMER, EPIC, MCP-counter, quanTIseq, CIBERSORT, and xCell, was used (25).

Twelve clinical glioma samples were obtained from the Department of Neurosurgery, Nanfang Hospital, Southern Medical University, Guangzhou, China. RNA was extracted from glioma tissues using TRIzol reagent and reverse-transcribed (Takara Bio Inc., Shiga, Japan). For RT-qPCR, cDNA was amplified using SYBR Green PCR Master Mix (Yeasen Biotechnology, China), with three independent replicates. Relative mRNA expression levels were normalized to that of β-actin. The primers used for RT-qPCR are listed in Supplementary Table 3. Prior consent was obtained from the patients for the use of their clinical materials for research purposes, and approval was obtained from the ethics committees of Nanfang Hospital.

Paraffin sections prepared from clinical samples were used for IHC to detect CD8 + T cell infiltration as previously described (26). Mouse anti-CD8-α (cat. No. sc-7970, 1:100; Santa Cruz Biotechnology, CA, USA) and goat anti-mouse secondary antibodies (Cat. No. PV-9000, Zhongshan Jinqiao, Beijing, China) were used. After incubation with the secondary antibody, the sections were visualized with a DAB kit (Cat. No. ZLI-9018, Zhongshan Jinqiao, Beijing, China), counterstained with hematoxylin, and analyzed using a bright-field microscope equipped with a digital camera (Nikon, Japan).

Single-cell RNA (scRNA) sequencing data for glioblastoma were retrieved from a previous study that identified malignant cells as CD45^- cells with significant copy number alterations (27). To specifically analyze the PRRS gene expression pattern in tumor-associated macrophages (TAMs), another scRNA dataset focused on CD45⁺ immune cells within glioblastoma tissues was used (28). The data were analyzed using the seurat package in R. Briefly, a series of quality filters was applied to the data to remove cells with too few total transcript counts (< 1,000), possible debris with too few genes expressed (< 200), possibly more than one cell with too many genes expressed (> 7,500), too many counts (> 7,500), and possible dead cells or a sign of cellular stress and apoptosis with a high proportion of mitochondrial gene expression over the total transcript counts (> 10%). Clustering was performed on K-nearest neighbor graph using the Louvain algorithm according to the top 30 principal component and cell types, including malignant and immune cells, were assigned according to previous reports (27, 28).

To evaluate the correlation between PRRS and temozolomide sensitivity, the estimated half-maximal inhibitory concentration (IC50) of temozolomide in each sample was estimated by ridge regression using the pRRophetic package in R based on the Genomics of Drug Sensitivity in Cancer (GDSC; https://www.cancerrxgene.org) (29).

To evaluate the correlation between PRRS and immunotherapy response, the potential response to anti-PD1 and anti-CTLA4 immunotherapy was predicted using the Submap tool in GenePattern (https://cloud.genepattern.org/gp) with human immunotherapy transcriptome data from a previous study (30). Two immunotherapy cohorts [recurrent glioblastoma patients treated with nivolumab or pembrolizumab, which are both anti-PD1 mAb (31); metastatic urothelial carcinoma patients treated with atezolizumab, which is an anti-PD-L1 mAb (32)] were used to validate the predictive value of PRRS in response to PD-L1 blockade.

To further unravel the potential small molecule compounds synergizing immunotherapy based on PRRS, weighted gene co-expression network analysis (WGCNA) was performed to identify the gene modules most associated with PRRS and immuneScore using the WGCNA package in R in both TCGA and CGGA cohorts (33). Then, DEGs between the high- and low-PRRS groups were identified using the limma package in R (P< 0.05 and |FC| > 1.5). Overlapping genes between the module genes and DEGs were confirmed and imported into the STRING database for functional enrichment. Connectivity map (CMap) (https://clue.io/) is a well-established tool to predict potential therapeutic drug candidates for specific genomic perturbation by comparing disease-specific gene signatures with drug-specific gene expression profiles (34, 35). Based on a gene expression matrix of DEGs between the high- and low-PRRS groups, potential small-molecule compounds associated with the expression pattern and their corresponding mechanisms of action were predicted using the CMap database and CMap mode-of-action (MOA) analysis, respectively (35).

All statistical analyses were performed using the R software (version 4.0.2). An unpaired t-test was performed to compare two normally distributed variables, whereas the Wilcoxon rank-sum test was performed for non-normally distributed variables. The Kruskal–Wallis test or one-way analysis of variance was performed to compare three or more variables based on the results of the normal distribution criteria test. Spearman’s correlation coefficient was used to determine correlations between variables. The survminer package in R was used to compare the survival status between the two groups, with Kaplan–Meier analysis used to plot survival curves. The log-rank test was performed to determine statistical significance, set at P< 0.05 (two-tailed).

A schematic of the working flow is shown in Figure 1. Using the 26 pyroptosis-related genes (Supplementary Table 2), the TCGA-glioma cohort was divided into two clusters by consensus clustering (Figure 2A; Supplementary Table 4), which showed satisfactory separation based on PCA analysis (Figure 2B). Cluster 2 showed significantly longer overall survival (OS) (HR = 0.18, 95% CI = 0.14–0.25, P< 0.001) and progression-free survival (PFS) (HR = 0.24, 95% CI = 0.19–0.32, P< 0.001) than Cluster 1 (Figures 2C, D). Within the GSDM family, GSDMB, GSDMC, GSDMD, and GSDME have been well-studied and are known downstream activating enzymes involved in pyroptosis (36). Therefore, the expression patterns of these four genes were examined in two clusters. While upregulated expression of GSDMB and GSDMC was found in Cluster 2 compared to that in Cluster 1, Cluster 1 had higher GSDMD and GSDME expression levels than Cluster 2, indicating that different pyroptosis pathways may be activated in different pyroptosis-related clusters (Figure 2E). Gene mutations are major drivers of tumorigenesis and tumor progression (37, 38). Thus, mutations in several driver genes in gliomas were assessed (Figure 2F). While 39% of the patients in Cluster 2 were found to possess IDH1 mutations, only 5% of the patients in Cluster 1 had IDH1 mutations (Figure 2F). Moreover, phosphatase and tensin homolog deleted on chromosome ten (PTEN), a tumor suppressor previously reported to promote pyroptosis by enabling NLRP3/ASC1 assembly via NLRP3 dephosphorylation (39), was found to be more frequently mutated in Cluster 1 (Figure 2F). In addition, a higher missense mutation rate of the Drosophila gene capicua (CIC) was found in Cluster 2 (Figure 2F), which has been shown to correlate with better survival in glioma patients, even with the co-occurrence of favorable markers including IDH mutation and 1p/19q codeletion (40).

We further characterized the clinical and molecular characteristics of Clusters 1 and 2. The pyroptosis-related genes showed distinct expression patterns between Cluster 1 and Cluster 2, with significantly higher expression levels of GSDMD, CASP1, CASP3, CASP4, and CASP8, and lower expression levels of P2RX7 and SAMR1 observed in Cluster 1. Cluster 1 was mainly composed of glioblastomas and grade III gliomas, and Cluster 2 was mainly composed of grade II gliomas (Figure 3A). Consistently, IDH1 mutations and MGMT promoter methylation were more frequent in Cluster 2 (Figure 3A). The mDNAsi index (which reflects epigenetic stemness features) and the mRNAsi index (which reflects transcriptomic stemness features) were calculated, as previously described (24). Although Cluster 1 possessed a lower mRNAsi index, contradictory results were found for the mDNAsi index. This indicates that diverse stemness regulation processes should be focused upon in Clusters 1 and 2 (Figure 3A; Supplementary Figure 1). A higher ImmuneScore and increased PD-L1 and CTLA-4 expression levels were identified in Cluster 1 (Supplementary Figure 1), suggesting a more immunosuppressive tumor ecosystem. Furthermore, higher TMB and lower MSI scores were observed in Cluster 1 (Figure 3A; Supplementary Figure 1; Supplementary Table 5).

Immune infiltration of different clusters was evaluated using six commonly used algorithms. It has been previously reported that tumor-associated macrophages (TAMs) and CD8⁺ T cells were considered the predominant immune cells involved in glioma progression. Accordingly, most algorithms showed higher CD8+ T cell enrichment in Cluster 1. Additionally, higher infiltration of both M1 and M2 macrophages was observed, especially in the latter (Figure 3B). GSEA results showed that in several immune-related pathways (including the IL-2, IL-6, IFN-γ, and TNF-α pathways) and the inflammatory response were highly enriched in Cluster 1 compared to in Cluster 2 (Supplementary Figure 1), indicating the regulatory role of these pathways in modulating cluster-specific immune microenvironments.

Pyroptosis-related clusters established with pyroptosis-related genes focused on the gene co-expression pattern but were also prognostic for glioma patients. To further confirm the prognostic value of the pyroptosis-related signature and establish a scoring system, the TCGA glioma cohort was divided into a training cohort (n=229) and an internal validation cohort (n=430), and a scoring system, called pyroptosis-related risk signature, based on 26 pyroptosis-related genes was then established by univariate and multivariate Cox regression analysis (Supplementary Figure 2). Eight pyroptosis-related genes were identified, and the final formula for calculating PRRS was as follows:

In the training cohort, patients were divided into low- and high-PRRS groups based on the median value (Figure 4A); more alive-status samples were in the low-PRRS group (Figure 4B). The high-PRRS group had higher expression levels of CASP5, GZMA, GZMB, MEFV, NLRC4, and STAT3 than the low-PRRS group, whereas higher GSDMB and SARM1 expression levels were observed in the low-risk group (Figure 4C). PRRS can satisfactorily divide glioma patients into two groups with distinct survival statuses. Specifically, the low-PRRS group had a significantly longer OS than the high-PRRS group (median OS: 9.506 and 4.085 years in the low-PRRS and high-PRRS groups, respectively; Figure 4D), and PRRS also showed satisfactory performance for 1-year, 3-year and 5-year OS predictions (1-year AUC = 0.84, 3-year AUC = 0.76, and 5-year AUC = 0.62; Figure 4E). In the internal validation cohort, PRRS showed a similar prognostic value for glioma patients (median OS: 5.622 and 1.537 years in the low- and high-risk groups, respectively; HR = 3.576, 95% CI = 2.672–4.789, P<0.001; Figure 4F–I). As in the training cohort, the AUC for the 1-year, 3-year and 5-year OS were 0.80, 0.79, and 0.70, respectively (Figure 4J).

The CGGA-325 and CGGA-693 cohorts were further used for external validation, and PRRS showed robust performance for prognosis prediction in these two cohorts (Supplementary Figures 3, 4). In CGGA-325 cohort, the low-PRRS group had a significantly longer OS compared with the high-PRRS group (median OS: 6.1 and 1.2 years in the low-PRRS and high-PRRS groups, respectively; HR = 2.20, 95% CI = 1.67–2.89, P<0.001; Supplementary Figure 3). PRRS also showed satisfactory prediction performance in the CGGA-693 cohort (median OS: 6.8 and 1.7 years in the low-risk and high-risk groups, respectively; HR = 2.20, 95% CI = 1.80–2.69, P<0.001; Supplementary Figure 4). Collectively, PRRS is an ideal tool for predicting the prognosis of patients with glioma.

PRRS was evaluated in different clinical subgroups. High PRRS was observed in Cluster 1 compared to that in Cluster 2, in accordance with the poorer prognosis found in Cluster 1 (Figure 5A). While PRRS increased as the stage advanced (Figure 5B), higher PRRS was also observed in IDH1 wildtype, 1p/19q non-codeleted, and MGMT-promoter unmethylated gliomas compared to their counterparts (Figure 5C-E). Since lower grade, 1p/19q codeletion, MGMT promoter methylation, and IDH1 mutation were the major favorable prognostic markers for glioma, a subgroup analysis based on the status of these parameters was performed in both TCGA and CGGA cohort.

PRRS showed satisfactory stratification for prognosis in the TCGA-LGG cohort (HR = 2.39, 95% CI = 1.56-3.66, adjusted P-value = 0.001). GBM patients with low PRRS also showed longer survival times than those with high PRRS, although this difference was not statistically significant (HR = 1.66, 95% CI = 1.06-2.60, adjusted P-value = 0.111; Figure 5F). Nevertheless, in the CGGA-325 cohort, PRRS was prognostic in the GBM subgroup (for the LGG subgroup, HR = 1.33, 95% CI = 0.75-2.34, adjusted P-value = 1.000; for the GBM subgroup, HR = 2.34, 95% CI = 1.31-4.20, adjusted P-value = 0.012; Figure 5G), whereas in the CGGA-693 cohort, PRRS also showed an association with decreased overall survival time in both LGG and GBM subgroups, although the difference was not statistically significant (for the LGG subgroup, HR = 2.19, 95% CI = 1.55-3.09, adjusted P-value<0.001; for the GBM subgroup, HR = 1.15, 95% CI = 0.73-1.82, adjusted P-value = 1.000; Supplementary Figure 5). To further consolidate this result, we obtained the expression and survival data from another GBM cohort from CPTAC and found similar results (HR = 1.57, 95% CI = 0.94-2.62, P value = 0.088; Supplementary Figure 5; Supplementary Table 6).

In both TCGA and CGGA cohorts, high PRRS was significantly associated with poor prognosis in both MGMT methylated and unmethylated subgroups (Figure 5H; Supplementary Figure 5). Considering the 1p/19q codeletion status, in TCGA and CGGA-325 cohorts, low PRRS predicted longer OS time in glioma patients without 1p/19q codeletion, whereas in the CGGA-693 cohort, PRRS was a significantly prognostic predictor in patients with or without 1p/19q codeletion (Figure 5I; Supplementary Figure 5). Further investigation found that PRRS could not be a significant prognostic predictor in either the IDH1 wildtype or mutant subgroups in the TCGA cohort, although high PRRS was still correlated with unfavorable survival in both subgroups (Figure 5J). Nevertheless, high PRRS is still a valuable marker for poorer prognosis in the IDH1 wild-type subgroup in the CGGA-325 cohort and in the IDH1 mutant subgroup in the CGGA-693 subgroup (Supplementary Figure 5; Supplementary Table 6).

We previously found that different pyroptosis-related clusters were associated with distinct immune infiltration. To determine the potential value of PRRS in immunotherapy, we first evaluated the correlation between PRRS and CD8+ T cells, which are canonical effector cells in immunotherapy (Figure 6A). Among the six commonly used algorithms, PRRS was significantly correlated with CD8+ T-cell enrichment in the TIMER and MCP-counter algorithms (r = 0.530, P<0.001 for the TIMER algorithm; r = 0.400, P<0.001 for the MCP-counter algorithm; Figure 6B; Supplementary Figure 6). To further validate this result, the correlation between PRRS and CD8A and CD8B was analyzed, and a significant positive correlation between PRRS and CD8A and CD8B was confirmed (r = 0.550, P<0.001 for CD8A; r = 0.630, P<0.001 for CD8B; Figure 6C; Supplementary Figure 6). Consistently, PRRS was positively correlated with both stromal and immune scores (r = 0.750, P<0.001 for stromal score; r = 0.620, P<0.001 for immune score; Figure 6D; Supplementary Figure 6). However, a positive correlation between PRRS and immune checkpoints, including PD-L1 and CTLA-4, was also found (r = 0.660, P<0.001 for PD-L1; r = 0.410, P<0.001 for CTLA-4; Figure 6E, F), indicating that although a high PRRS was associated with more CD8+ T cell infiltration, the infiltrated T cells may be exhausted because of the concomitantly high expression of immune checkpoints. While PRRS was positively correlated with TMB score, it was negatively associated with MSI score (r = 0.310, P<0.001 for TMB score; r = -0.370, P<0.001 for MSI score; Supplementary Figure 6). Consistently, in both CGGA-325 and CGGA-693 cohorts, PRRS was positively correlated with CD8+ T cell enrichment and higher immune checkpoint expression levels (Supplementary Figure 7).

We subsequently validated the relationship between PRRS and CD8+ T cell infiltration in glioma samples. Through quantification of signature genes by RT-qPCR and calculation of PRRS according to the formula mentioned above, 12 glioma samples were divided into low- and high-PRRS groups. Higher expression of CASP5, GZMB, MEFV, and NLRC4 was found in the high-PRRS group (Figure 6G; Supplementary Table 7). CD8+ T cell infiltration was examined in the low- and high-PRRS groups using IHC. It was previously reported that T-cell infiltration can be highly variable in gliomas (41), and we observed that LGG or GBM samples with high PRRS harbored more CD8+ T cells than those with low PRRS (Figure 6H). Moreover, a positive correlation was observed between PRRS and CD8+ T-cell enrichment (Figure 6H).

GZMA and GZMB were previously reported to be specifically expressed in CD8+ T cells (42), indicating that PRRS is a signature involving not only tumor cells but also immune cells. To further illustrate the potential cell type-specific expression pattern, the expression of eight PRRS genes in glioblastoma cells, TAMs, T cells, and oligodendrocytes was analyzed using scRNA data obtained in a previous study (Figure 7A) (27). STAT3 and MEFV are ubiquitously expressed in several cell types. Conversely, higher expression levels of GSDMB, GZMA, and GZMB were detected in T cells, whereas NLR4 and CASP5 showed specifically high expression levels in TAMs, and CASP5, GZMA, GZMB, and NLRC4 were barely detected in glioblastoma cells (Figure 7B). Neftel et al. stratified malignant glioblastoma cells into four cellular states: neural-progenitor-like (NPC-like), oligodendrocyte-progenitor-like (OPC-like), astrocyte-like (AC-like), and mesenchymal-like (MES-like) states (27). It has been suggested that glioblastoma cells in the MES-like state may be more efficiently killed by T cells (43). By calculating the PRRS in each cellular state, we found that MES-like and AC-like glioblastoma cells possessed a higher PRRS score than those in the other two states (Figure 7C). Specifically, while MEFV and STAT3 expression was detected in both glioblastoma cells and immune cells, SARM1 expression level was higher in NPC-like and OPC-like glioblastoma cells than in MES-like, AC-like glioblastoma cells, and other tumor cells (Figure 7D–G). Antunes et al. divided myeloid cells in GBM into several subsets (28). PRRS was calculated for each type of TAMs. A previous study suggested that higher interferon γ (IFNγ) response was associated with immunotherapy response in GBM, and IFNγ could modulate the immune cell composition in the tumor center (44). Intriguingly, a significantly higher PRRS was found in interferon-signature microglial-TAM (IFN_MgTAM) than in non-interferon-signature microglial-TAM (MgTAM) (Figure 7H, I).

Based on the prognostic value of the pyroptosis-related group and PRRS, and the distinct tumor-associated immune microenvironment between different groups, further investigation into their roles in therapeutic response was performed. Both Cluster 1 and the high-PRRS group had a lower temozolomide IC50 index than their counterparts, indicating that Cluster 1 and the high-PRRS group were more sensitive to TMZ therapy (Figure 8A, B). Higher expression levels of several immune checkpoints were also observed in Cluster 1 and the high PRRS group (Figure 8C, D, Supplementary Table 8). Subsequently, a submap algorithm was used to evaluate the potential value of PRRS in predicting immunotherapy responses. The results showed that glioma patients with high PRRS might be more responsive to anti-PD1 therapy (nominal P-value = 0.001 and Bonferroni corrected P-value = 0.008; Figure 8E). A recurrent glioblastoma (rGBM) cohort that received anti-PD1 mAb therapy was used for validation (31). Both PFS and OS of GBM patients with high PRRS showed a trend toward improved response compared with the low-PRRS group. However, these differences did not reach statistical significance, partially due to the limited sample size (Figure 8F, G). In the Imvigor210 cohort, a clinical trial involving urothelial carcinoma patients who received anti-PD-L1 therapy, a survival benefit was not observed. However, a higher percentage of inflamed immune phenotype, which is considered a “hot” tumor, was observed in the high-PRRS group, indicating that PRRS could stratify patients into different immune phenotypes (Supplementary Figure 8).

Connectivity map (CMap) is a well-established tool for predicting potential small-molecule compounds for particular genomic perturbations or diseases by comparing specific gene signatures with drug-specific gene expression profiles in the reference database (35). To further deepen the therapeutic value of PRRS, subsequent screening using CMap was performed for potential therapeutic drug candidates associated with PRRS-related expression patterns. First, WGCNA was performed in the TCGA cohort, and nine co-expression gene modules were identified based on a power of 16 (Figure 9A), while the number of mean connectivity was 17 (Figure 9B), including black, blue, brown, green, red, pink, turquoise, yellow, and gray; the gray module was considered to be a group of genes that could not be assigned to any module (Figure 9C, Supplementary Table 9). Among these nine modules, the brown module showed the most positive correlation between both PRRS and ImmuneScore (Figure 9D). Using the same methodology, the blue module, which was highly associated with PRRS and ImmuneScore, was identified in the CGGA-325 cohort (Supplementary Figure 9, Supplementary Table 10). Through further analysis of DEGs, we identified 2142 and 2865 DEGs between the high- and low-PRRS groups in TCGA and CCGA-325 cohorts, respectively (Supplementary Table 11–12; Figure 9E). The intersection of genes from the TCGA brown module, the CGGA blue module, and DEGs created a list of 183 genes (Figure 9F). Subsequent functional enrichment analysis showed significant enrichment in immune response, immune system process, immune effector process, and regulation of immune response for biological process analysis. For the REACTOME pathway analysis, neutrophil degranulation, interferon signaling, antigen processing-cross presentation, and PD-1 signaling were enriched. For the wikiPathway analysis, the microglia pathogen phagocytosis pathway, complement system, macrophage marker, and type II interferon signaling pathways were enriched (Figure 10A). Ultimately, based on 183 upregulated genes and 63 downregulated genes in the high-PRRS group compared to in the low-PRRS group, potential small molecular compounds and their related mechanisms were predicted by the CMap database and MOA analysis, respectively (Figure 10B, Supplementary Table 13). These results indicate that treatment with targeted drugs, such as Bcl-2 and ATPase inhibitors, may facilitate immunotherapy in gliomas, although further evidence is required.