The gene expression profiles of glioma patients and clinical trait information were downloaded from TCGA (https://portal.gdc.cancer.gov/) and the Chinese Glioma Genome Atlas (CGGA; http://www.cgga.org.cn/), respectively. The original gene expression profile was normalized and centralized, and the probe names were annotated as gene names. Before analysis, glioma patients without IDH1mt information and survival information were excluded. As a result, 367 patients with IDH1mt and 229 patients with wtIDH1 were obtained from TCGA, while 167 patients with IDH1mt and 145 patients with wtIDH1 were obtained from the CGGA. Immune and stromal scores of IDH1mt glioma tissues in TCGA were measured using the ESTIMATE algorithm.

The median immune and stromal scores of gliomas with IDH1mt in TCGA were used to separate the high and low groups. The EdgeR package was used to perform the differentially expressed genes (DEGs) analysis, while the threshold for considering significance was set as |logFC | ≥ 1 and adjusted P-value < 0.05. Analysis of the changes in all genes in the high- and low-immune/stromal score groups was visualized using volcano plots, while DEGs were visualized in a heatmap.

The primordial protein–protein interaction (PPI) network was constructed using DEG information from the STRING database (https://cn.string-db.org/). Cytoscape software was used to adjust the primordial PPI network, and the isolated genes were removed. Genes that had a relationship with others were set as hub genes and were enrolled in further studies.

The enriched GO terms of hub genes were analyzed using the Database for Annotation, Visualization, and Integrated Discovery (DAVID; https://david.ncifcrf.gov/). GO analysis was conducted using three categories: biological processes (BP), cellular components (CC), and molecular functions (MF). The significance threshold was set at P < 0.05. The top five terms are presented in a bubble diagram.

Before constructing an immune signature, we first analyzed the hub genes associated with the survival of patients with IDH1mt via univariate Cox regression analysis, and the significance threshold was set at P < 0.05. The least absolute shrinkage and selection operator (LASSO) was used to eliminate genes that shared similar genetic information by adding appropriate penalties (lambda). Utilizing the Akaike information criterion, an optimal prognostic risk model was built using a multivariate Cox regression analysis. The upper limit of the risk score was set to 10. The feasibility of the risk model in IDH1mt and wtIDH1 glioma patients in TCGA and CGGA was checked using receiver operator characteristic curve (ROC) analysis and Kaplan–Meier survival analysis. The ROC cut-off was set as the area under the curve (AUC) ≥ 0.75 and P < 0.05, while the threshold in the Kaplan–Meier survival analysis was set as P < 0.05.

Nomograms are constructed based on multifactor regression analysis by integrating multiple predictors, and graduated line segments are then used to draw on the same plane in a certain proportion (14). In this study, a nomogram was created using the “rms” package to simplify the prediction model using independent clinical prognostic factors.

To explore the signaling pathways differentially activated in high- and low-risk group IDH1mt-glioma patients in TCGA and CGGA, we analyzed the change of genes and performed GSEA analysis in R software with an adjusted P-value < 0.05.

Using the “CIBERSORT” R package, we examined 22 immune cells infiltrating IDH1mt-glioma tissues in TCGA and CGGA. Differentially infiltrated cells in the high- and low-risk-score groups were analyzed using the unpaired t-test, with significance set as P < 0.05.

In total, 54 glioma tissues with IDH1mt were collected from the Affiliated Hospital of Guizhou Medical University with the approval of the Human Ethics Committee of Guizhou Medical University. None of the patients received any radio-chemotherapy before the operation, and written informed consent was obtained from all participants. Immunohistochemistry (IHC) staining was performed as described in our previous study (15). Primary antibodies used were as follows: HLA-DQA2 (1:200; Cat No. 42-669, ProSci, Fort Collins, CO, USA), HOXA3 (1:100; Cat No.ab230879, Abcam, Cambridge, UK), SAA2 (1:500; Cat No. CAU25292; Biomatik, Kitchener, Canada), CD8 (1:4000; Cat No. 66868-1-Ig; Proteintech, Wuhan, China), and CD86 (1:250; Cat No. ab220188, Abcam). The expression of target proteins was assessed based on the product of the intensity of staining (0, 1+, 2+, and 3+) and the percentage of positive cells, which was scored as 0 (0%), 1 (1–2 5%), 2 (26–50%), 3 (51–75%), or 4 (76–100%).

To predict the immune checkpoint blockade (ICB) therapy response, the gene expression profile of glioma tissues with IDH1mt was imported into the T-cell dysfunction and exclusion (TIDE; http://tide.dfci.harvard.edu/) online algorithm to obtain exclusion, dysregulation, and TIDE scores. P < 0.05 was set as the threshold for determining the difference between high- and low-risk-score groups using an unpaired t-test.

Using the ESTIMATE algorithm, stromal and immune scores were calculated in TCGA for patients with glioma and IDH1mt. The detailed scores are shown in Supplementary Table 1 . The Kaplan–Meier survival analysis showed that IDH1mt-glioma patients with high stromal ( Figure 1A ) and immune scores ( Figure 1B ) exhibited shorter overall survival rates than those with low stromal and immune scores (HR = 1.679 and HR = 2.367). We then determined the change in gene expression profiles between the high- and low-stromal/immune groups in IDH1mt-glioma tissues. A total of 242 genes were upregulated in the high-stromal group IDH1mt-glioma tissues compared with those in the low-stromal group IDH1mt-glioma tissues, while 20 genes were downregulated ( Figures 2A, B ). Similarly, 285 genes were upregulated in the high-immune group IDH1mt-glioma tissues compared with those in the low-immune group IDH1mt-glioma tissues, while 135 genes were downregulated ( Figures 2C, D ). Through intersection analysis, 209 upregulated ( Figure 2E ) and 13 downregulated overlapping genes ( Figure 2F ) were identified in the high-stromal- and high-immune-score-group IDH1mt-glioma tissues.

We then constructed a PPI network and found 72 genes that interacted with other genes ( Figure 3A ). These genes were set as candidate hub genes associated with stromal and immune scores in IDH1mt glioma. BP enrichment analysis demonstrated that these candidate hub genes were enriched in “immune-response-activating cell surface receptor signaling pathway”, “humoral immune response”, “response to interferon-gamma”, “cellular response to interferon gamma”, and “interferon-gamma-mediated signaling” ( Figure 3B ). MF enrichment analysis demonstrated that the candidate hub genes were enriched in “receptor ligand activity”, “antigen binding”, “peptide antigen binding”, “MHC class II receptor activity”, and “chemokine receptor binding” ( Figure 3C ). Furthermore, the enriched CC terms of candidate hub genes were “endocytic versicle membrane”, “MHC protein complex”, “MHC class II protein complex”, “luminal side of ER”, and “integral component of membrane of ER” ( Figure 3D ).

Next, we determined whether the expression of candidate hub genes was associated with survival in patients with IDH1mt gliomas. A univariate Cox regression analysis demonstrated that among 72 candidate hub genes, 29 genes were linked to survival in IDH1mt-glioma patients ( Table 1 ). LASSO analysis was performed to create a risk model that could predict the survival of glioma patients with IDH1mt. Five more important hub genes, including HLA-DQA2, HLA-DQB2, HOXA2, HOXA3, and SAA2, were identified and used for further analysis ( Figures 4A, B ). A multivariate Cox analysis also indicated that HLA-DQA2, HOXA3, and SAA2 were independent predictors of survival in glioma patients with IDH1mt ( Figure 4C ). Therefore, the gene expression information of HLA-DQA2, HOXA3, and SAA2 and the survival information of glioma patients with IDH1mt in TCGA were imported into R software to construct a risk model. Using computer optimization, a risk model was constructed with a risk score of 0.249 × HLA-DQA2 expression + 0.179 × HOXA3 expression + 0.227 × SAA2 expression.

To verify the applicability of the risk model in glioma patients with IDH1mt, glioma patients with IDH1mt in TCGA ( Figure 5A ) and CGGA ( Figure 5B ) were divided into high- and low-risk groups according to the median risk score obtained from the TCGA cohort. The results indicated that IDH1mt glioma patients in TCGA with high-risk scores had shorter overall survival rates than those with low-risk scores ( Figure 5C ). ROC analysis indicated that the AUC for predicting the one- and three-year survival of IDH1mt-glioma patients in TCGA were 0.845 and 0.821, respectively ( Figures 5D, E ). Similarly, we found that IDH1mt-glioma patients in CGGA with high-risk scores had shorter overall survival rates than those with low-risk scores ( Figure 5F ), and the AUC for predicting one- and three-year survival of IDH1mt-glioma patients in CGGA were 0.794 and 0.764, respectively ( Figures 5G, H ). Furthermore, we found that IDH1mt-glioma patients in TCGA ( Figure 5I ) and CGGA ( Figure 5J ) with high-risk scores exhibited a higher proportion of deaths. This evidence indicates that the risk model has remarkable diagnostic value for glioma patients with IDH1mt.

Glioma patients with wtIDH1 in TCGA ( Figure 6A ) and CGGA ( Figure 6B ) were divided into high- and low-risk groups according to the median risk score. wtIDH1 glioma patients in TCGA with high-risk scores had lower overall survival than those with low-risk scores ( Figure 6C ). However, the AUC for predicting the one- and three-year survival of wtIDH1 glioma patients in TCGA were 0.644 and 0.682, respectively ( Figures 6D, E ). Similarly, high-risk wtIDH1-glioma patients in CGGA had shorter overall survival rates ( Figure 6F ), but the AUCs of the risk model for predicting the one- and three-year survival of wtIDH1 glioma patients in CGGA were 0.570 and 0.652, respectively ( Figures 6G, H ). Furthermore, the percentage of deaths was not significantly different between wtIDH1 glioma patients in the high- and low-risk groups in TCGA ( Figure 6I ) and CGGA ( Figure 6J ). These results indicated that the risk model constructed using HLA-DQA2, HOXA3, and SAA2 was not suitable for predicting the survival of wtIDH1 glioma patients and may be specific for IDH1mt-glioma patients.

We then performed a multivariate Cox regression analysis, and the immune signature constructed using HLA-DQA2, HOXA3, and SAA2 was found to act as an independent prognostic factor for glioma patients with IDH1mt, with an HR of 1.203 ( Table 2 ). In addition, a nomogram was created based on the signature risk score and clinical characteristics ( Figure 7 ).

Gene set enrichment analysis (GSEA) was used to determine if defined pathways were enriched in high- and low-risk groups of glioma patients with IDH1mt. IDH1mt-glioma tissues with high risk in TCGA were positively associated with “M phase“ (NES=1.89, P<0.01) and “signaling by interleukins” (NES=2.24, P<0.01; Figure 8A ), while those with high risk in CGGA were positively associated with “cell cycle mitotic” (NES=2.19, P<0.01) and “neutrophil degranulation” (NES=2.17, P<0.01; Figure 8B ).

According to previous studies (16, 17), infiltrating immune cells play a critical role in the progression of IDH1mt glioma. We determined the difference in infiltration of 22 immune cells between the high- and low-risk groups of glioma tissues with IDH1mt. The CIBERSORT R package was used to convert the gene expression profile of glioma tissues with IDH1mt in TCGA ( Figure 9A ) and CGGA ( Figure 9B ) to a proportion profile of infiltrated immune cells. Compared with IDH1mt glioma tissues with low-risk scores, those with high-risk scores in TCGA exhibited a high proportion of naïve B cells, plasma cells, CD8 T cells, CD4 memory activated T cells, activated NK cells, M0 macrophages, and M1 macrophages, while the proportion of resting NK cells and activated dendritic cells was reduced ( Figure 9C ). In CGGA, IDH1mt-glioma tissues with high-risk scores had higher memory B cells, CD8 T cells, M1 macrophages, M2 macrophages, and resting dendritic cells and lower M0 and CD4 naïve T cells ( Figure 9D ). In conclusion, both TCGA and CGGA indicated a higher proportion of CD8 T cells and M1 macrophages in IDH1mt-glioma tissues with high risk compared with that in those with low risk.

In total, 54 glioma tissues with IDH1mt from our research group were divided into long- (survival ≥ 15 months) and short-term groups (survival < 15 months). IHC was performed to detect the expression of HLA-DQA2, HOXA3, and SAA2 in glioma tissues, and high expression of HLA-DQA2, HOXA3, and SAA2 was observed in glioma tissues in the short-term group compared with that in the long-term group ( Figures 10A, B ). Similarly, we detected the expression of the M1 biomarker CD86 and the CD8 T-cell biomarker CD8 in glioma tissues using IHC. The expression of CD86 and CD8 increased in IDH1mt-glioma tissues in the short-term survival group ( Figure 10C ). These results suggest that M1 and CD8 T cells infiltrate more deeply into IDH1mt-glioma tissues associated with a lower probability of survival ( Figure 10C ). Furthermore, ROC analysis was performed to determine the diagnostic value of HLA-DQA2, HOXA3, and SAA2 in the survival of IDH1mt-glioma patients, and all showed remarkable diagnostic value (AUC = 0.832, 0.896, and 0.857) ( Figures 10D–F ).

The TIDE online algorithm was used to evaluate the responsiveness of IDH1mt-positive glioma patients in the high- and low-risk groups to ICB therapy. Lower exclusion scores were observed in IDH1mt-glioma patients with high-risk scores than in those with low-risk scores ( Figure 11A ), while the dysregulation score was reduced ( Figure 11B ). Overall, the TIDE score was significantly reduced in IDH1mt-glioma tissues with high-risk scores compared with that in those with low-risk scores ( Figure 11C ). Finally, the responder prediction results indicated that glioma patients with IDH1mt in the high-risk group exhibited high responsiveness to ICB therapy ( Figure 11D ). Based on this evidence, this risk model may be able to guide the clinical treatment of glioma patients with IDH1mt.