The CXCL Family Contributes to Immunosuppressive Microenvironment in Gliomas and Assists in Gliomas Chemotherapy

Abstract

Gliomas are a type of malignant central nervous system tumor with poor prognosis. Molecular biomarkers of gliomas can predict glioma patient's clinical outcome, but their limitations are also emerging. C-X-C motif chemokine ligand family plays a critical role in shaping tumor immune landscape and modulating tumor progression, but its role in gliomas is elusive. In this work, samples of TCGA were treated as the training cohort, and as for validation cohort, two CGGA datasets, four datasets from GEO database, and our own clinical samples were enrolled. Consensus clustering analysis was first introduced to classify samples based on CXCL expression profile, and the support vector machine was applied to construct the cluster model in validation cohort based on training cohort. Next, the elastic net analysis was applied to calculate the risk score of each sample based on CXCL expression. High-risk samples associated with more malignant clinical features, worse survival outcome, and more complicated immune landscape than low-risk samples. Besides, higher immune checkpoint gene expression was also noticed in high-risk samples, suggesting CXCL may participate in tumor evasion from immune surveillance. Notably, high-risk samples also manifested higher chemotherapy resistance than low-risk samples. Therefore, we predicted potential compounds that target high-risk samples. Two novel drugs, LCL-161 and ADZ5582, were firstly identified as gliomas' potential compounds, and five compounds from PubChem database were filtered out. Taken together, we constructed a prognostic model based on CXCL expression, and predicted that CXCL may affect tumor progression by modulating tumor immune landscape and tumor immune escape. Novel potential compounds were also proposed, which may improve malignant glioma prognosis.

Keywords: CXCL; chemotherapy; gliomas; immune checkpoint genes; immunosuppressive.

Figure 1

Relationships between CXCL expression patterns and clinical characters of gliomas. The expression profile of the CXCL family in the LGGGBM group (A) from the TCGA database and the validation cohort (B, C). The heatmap of the CXCL family expression in LGG group (D–F) and GBM group (G–I) from the training and validation cohorts. *p < 0.05, **p < 0.01, ***p < 0.001.

Figure 2

The prognostic value of the cluster model in gliomas. Kaplan–Meier survival analysis were used to show overall survival outcome difference between the two clusters in LGG, GBM, and LGGGBM samples from TCGA (A–C), CGGA1 (D–F), and CGGA2 (G–I).

Figure 3

Constructing the risk model. (A) The co-expression network of the CXCL family. (B, C) The construction of the risk model based on the expression profile of the CXCL family by performing elastic net regression algorithm. (D) Heatmap displayed the alternation of the CXCL family members’ expression according to the risk model, and corresponding clinical features were also mapped. The distribution of the risk in gliomas’ pathological grade (E), IDH status (F), cancer type (G), MGMG status (H), 1p19q status (I), and subtype (J). NS, not significant, ***p < 0.001.

Figure 4

The overall survival analysis and biofunction prediction based on the risk model. Kaplan–Meier survival analysis based on the risk model from the training cohort (A) and the validation cohort, including Xiangya cohort (B), CGGA1 (C), CGGA2 (D), CGGA688 (E), GSE108474 (F), GSE43378 (G), GSE16011 (H), GSE68838 (I). (J) GO/KEGG enrichment analysis based on the GSVA analysis in the training cohort.

Figure 5

The expression of ICGs and immunocyte infiltration ratio. (A) ICG expression was mapped based on the risk model. The ImmuneScore (B), StromalScore (C), ESTIMATEscore (D), and purity (E) difference between high- and low-risk group in TCGA database. (F) Immunocyte infiltration ratio enrichment score based on the risk model. (G) The correlation between the enrichment score of immunocytes and the risk model. NS, not significant, ***p < 0.001.

Figure 6

Chemotherapy suggestion based on the risk model. Chemotherapy efficacy difference between high- and low-risk group in the CGGA1 (A) and CGGA2 (B) datasets. (C) Correlation between risk and the AUC value of 17 candidate compounds. (D–F) Distribution of the AUC value of candidate compounds in the risk model. (G) Correlation between compounds’ IC50 and risk from Cellminer dataset. (H) The difference between IC50 of compounds and the risk model ***p < 0.001.

Figure 7

Comparing the prognostic ability of the cluster model, the risk model, and glioma pathological grade. (A) The Sankey diagram revealed the potential connection between glioma pathological grade, risk, cluster, IDH status, 1p19q status, and MGMT status. ROC curve generated based on the risk model by taking the IDH status (B), OS (C), and 1p19q status (D) as outcome variable in the training cohort. The validation of ROC curve in the CGGA1 (E–G) and CGGA2 (H–J) database.

Figure 8

Prognostic nomogram based on the risk model. (A–D) The Schoenfeld test of the factors involved in the construction of the nomogram. (E) The nomogram based on the risk model. (F) The calibration curve of 3-year and 5-year OS based on the nomogram. ROC curves and AUC values from the nomogram of 3-year and 5-year OS, in TCGA datasets (G), CGGA1 datasets (H), and CGGA2 datasets (I). (J–L) Survival analysis based on the nomogram in TCGA datasets (J), CGGA1 datasets (K), and CGGA2 datasets (L).