Ferroptosis Activation Scoring Model Assists in Chemotherapeutic Agents' Selection and Mediates Cross-Talk With Immunocytes in Malignant Glioblastoma

Abstract

Gliomas are aggressive tumors in the central nervous system and glioblastoma is the most malignant type. Ferroptosis is a programmed cell death that can modulate tumor resistance to therapy and the components of tumor microenvironment. However, the relationship between ferroptosis, tumor immune landscape, and glioblastoma progression is still elusive. In this work, data from bulk RNA-seq analysis, single cell RNA-seq analysis, and our own data (the Xiangya cohort) are integrated to reveal their relationships. A scoring system is constructed according to ferroptosis related gene expression, and high scoring samples resistant to ferroptosis and show worse survival outcome than low scoring samples. Notably, most of the high scoring samples are aggressive glioblastoma subtype, mesenchymal, and classical, by calculating RNA velocity. Cross-talk between high scoring glioblastoma cells and immunocytes are explored by R package 'celltalker'. Ligand-receptor pairs like the TRAIL or TWEAK signaling pathway are identified as novel bridges implying how ferroptosis modulate immunocytes' function and shape tumor microenvironment. Critically, potential drugs target to high scoring samples are predicted, namely, SNX2112, AZ628, and bortezomib and five compounds from the CellMiner database. Taken together, ferroptosis associates with glioblastoma aggressiveness, cross-talk with immunocytes and offer novel chemotherapy strategy.

Keywords: cell–cell communication; ferroptosis; glioblastoma; immunocytes; mesenchymal.

Figure 1

The construction of the FeAS model. (A) Flow chart shows the construction of the FeAS model. (B) Ferroptosis related gene expression and corresponding clinical feature based on the clustering model was illustrated with heatmap. (C) Ferroptosis related gene expression and corresponding clinical feature based on the FeAS model was illustrated with heatmap. Survival analysis based on the FeAS model in the GBM cohort in TCGA database (D, P value = 0.0051), GSE108474 database (E, P value = 0.021), CGGA1 database (F, P value = 0.016) and the Xiangya cohort (G, P value = 0.013). (H) Prognostic efficiency ability comparison between the FeAS model and other three ferroptosis models by introducing ROC curve.

Figure 2

Association between ferroptosis and GBM aggressiveness. (A) Distribution of FeAS in GBM subtypes according to bulk RNA-seq analysis in the TCGA, CGGA1, and GSE108474 database. (B) Distribution of FeAS in single cell RNA-seq analysis. (C) The subtype of GBM cells in the FeAS model. (D) RNA velocity illustrated by pseudo-time analysis indicating GBM cells aggressiveness difference. (E) Integration of RNA velocity and the subtype of GBM cells. (F) Integration of RNA velocity and FeAS of GBM cells. CL, classical; MES, mesenchymal; PN, proneural. NS, no significant; **P < 0.01; ***P < 0.001.

Figure 3

Transcription factor activation difference between high and low FeAS GBM cells. (A) GBM cells can be grouped in five modules according to the cooperation of different transcription factors. (B–F) Potential relationship between the scoring system and those modules based on RNA velocity. (G) Top 10 differential activated transcription factor in high and low FeAS samples respectively.

Figure 4

Biofunction analysis based on bulk RNA-seq analysis and single cell RNA-seq analysis in GBM. (A) GO enrichment analysis based on the GSVA algorithm in bulk RNA-seq analysis. (B) KEGG enrichment analysis based on the GSVA algorithm in bulk RNA-seq analysis. (C) GO enrichment analysis based on differential expression genes between high and low FeAS samples in single cell RNA-seq analysis. (D) KEGG enrichment analysis based on differential expression genes between high and low FeAS samples in single cell RNA-seq analysis. GSEA enrichment analysis based on bulk RNA-seq analysis (E) and single cell RNA-seq analysis (F).

Figure 5

Tumor immune landscape based on bulk RNA-seq analysis in TCGA database. (A) Correlation of ESTIMATE score, stromal score, immune score and tumor purity with FeAS. Immunocytes infiltration ratio in high and low FeAS samples according to CIBERSORT algorithm (B) and xCell algorithm (C). NS, no significant. *P < 0.05; **P < 0.01; ***P < 0.001.

Figure 6

Novel ligand–receptor pairs difference between high and low FeAS samples. (A) High FeAS cells communicate with macrophage, naïve T cell, and dendritic cells through PRSS3-F2R. (B) High FeAS cells communicate with naïve T cell and plasmacytoid dendritic cell through TNFSF12-TNFSF12A. (C) High FeAS cells communicate with plasmacytoid dendritic cell through WNTSA-FZD3. (D) High FeAS cells communicate with plasmacytoid dendritic cell through RETN-CAP1. (E) High FeAS cells communicate with macrophage, naïve T cell and T cells through NAMPT-(ITGA5+ITGB1). (F) High FeAS cells communicate with macrophage, microglial cell, naïve T cell, T cells and plasmacytoid dendritic cell through TNFSF10-TNFRSF10B.

Figure 7

Potential targeted drugs according to the FeAS model. (A) Venn chart shows the number of drugs in PRISM dataset and two CTRP database. (B) Flow chart illustrates the potential compounds based on the FeAS model. (C) Correlation between the AUC value of potential drugs and the FeAS of each sample. (D) Distribution of the AUC value of potential drugs in the FeAS model. (E) Correlation between GI50 and FeAS. (F) The distribution of GI50 of each compound based on the FeAS model. *P < 0.05; **P < 0.01; ***P < 0.001.