Identification of DEGs
Following differential gene expression analysis and filtering with the cut-off criteria of P. value < 0.05 & |logFC| > 1, the DEGs between tumor and normal cells were screened. For the GSE43290 dataset, a total of 1499 DEGs were identified, out of which 73 were up-regulated and 1426 were down-regulated. For the GSE77259 dataset, 4830 DEGs, including 2536 up-regulated and 2294 down-regulated genes, were identified. Moreover, a total of 673 overlapping DEGs, such as MYC, TNFAIP3 and SLC2A3 in both GSE43290 dataset and GSE77259 dataset were obtained (Supplementary Fig. 1).
Furthermore, volcano plot and PCA clustering of DEGs in GSE43290 showed that the DEGs between the two groups could be significantly distinguished (Fig. 1A and B). Additionally, using a threshold of |logFC| > 2, 255 DEGs out of the 1499 DEGs in GSE43290 were further screened, and the expression pattern of these DEGs could divid the samples into two groups (Fig. 1C). To further predict the potential functions of these DEGs in meningioma, enrichment analysis was performed. The results of GO analysis showed that these DEGs were functionally classified by Biological Process (BP), Cellular Component (CC), and Molecular Function (MF) (Fig. 1D). Among them, several BPs related to immune response (leukocyte migration, leukocyte chemotaxis) and responses processes (reponse to metal ion, response to zinc ion, positive regulation of response to external stimulus) were significantly enriched; moreover, CCs and MFs were found to be mainly associated with regulation of cardiac electrical activity. Furthermore, KEGG pathway analysis revealed a series of enriched pathways, such as IL-17 signaling pathway and TNF signaling pathway (Fig. 1E).
The immune cell infiltration landscape
Based on the gene expression profiles of GSE43290, the difference in immune cells infiltration between tumor and control samples among 22 immune cells types, were detected using CIBERSORT algorithm. As shown in Fig. 2A, 48 of the 51 samples were valid (P value < 0.05), including 44 tumor samples and 4 normal samples. Among the 22 immune cell types, the percentage of CD8+ T cells (yellow) and M2 macrophages (blue) in each sample was relatively high. The heatmap showing the 22 tumor immune cells was illustrated in Fig. 2B, which also confirmed the high infiltrating abundance of CD8+ T cells and M2 macrophages in all samples. Based on the violin plot (Fig. 2C), the probability of distribution of memory B cells, regulatory T cells (Tregs) and resting mast cells in tumor were obviously higher than those in the control samples (all, P < 0.05), indicating that the immune infiltration proportions could be useful in distinguishing meningioma patients from healthy patients.
In addition, a lot of literature have shown that mast cells (MC) were associated with the development of meningioma [13, 14, 28], and resting mass cells infiltration ratio was relatively high in this study. Hence, the follow-up analysis focused on resting mast cells.
Weighted correlation network analysis (WGCNA) co-expression networks
When correlation coefficient threshold was set at 0.85, the soft-thresholding power was 6 (Supplementary Fig. 2A). Through WGCNA analysis, 10 co-expression modules were constructed based on the 1499 DEGs (Supplementary Fig. 2B), of which the grey module was the gene set that could not gather into other modules, so there were only 9 valid gene modules.
Subsequently, as shown in Fig. 3, the correlation between each module and the degree of invasion of immune cells in tumor tissue was calculated. The results revealed that yellow, the module with the highest correlation, contained 158 DEGs, consisting of 2 up-regulated and 156 down-regulated genes (Supplementary Table 1). Yellow modules are negatively related to degree of resting mast cells infiltration in meningioma tissues, and the correlation between genes in the yellow module and the degree of immune cell infiltration was summarized in the Supplementary Table 2.
Furthermore, GO functional analysis revealed that the genes in yellow module were dominant in biological processes involved in response to leukocyte migration, molecules of bacterial origin, and inflammatory response (Fig. 4A); KEGG pathway analysis revealed that the genes were mainly enriched in the TNF signaling pathway, cytokine-cytokine receptor interaction and IL-17 signaling pathway (Fig. 4B).
Risk model construction of key genes
In order to identify whether each key gene related to resting mast cells correlated with prognosis of meningioma patients, univariate and multivariate analyses were performed. K-M survival analysis revealed that CXCL8 and MYC were associated with prognosis of meningioma patients. Moreover, based on the COX univariate and multivariate regression analyses showed CXCL8 and MYC might be prognostic biomarkers for meningioma patients (Table 1 and Fig. 5A).
Subsequently, based on the 9 genes in univariate regression analysis above, the risk model was constructed. Firstly, the meningioma patients were categorized into low- or high-risk patients, and the cut off value was the median risk score (Fig. 5B). The scatter plot of survival time of risk model samples showed that the survival time of samples was relatively lower in the high-risk group the low-risk group (Fig. 5C). Monolayer clustering analysis of RNA expression in the risk model of each sample showed differences in the key genes between high and low risk groups (Fig. 5D). Moreover, K-M survival curve indicated that the survival time between the high and low risk groups was significantly different (P value = 0.00436; Fig. 5E), suggesting the risk model of 9 key gene related to resting mast cells was successfully constructed.
TF-miRNA- mRNA co-regulation network
Based on the database (miRWalk3.0, TargetScan, MiRDB, and MirTarBase) with the Score > 0.95, 145 miRNAs acting on 3’UTR region of the genes in the risk model associated with resting mast cells were predicted. Then, according to the retrieval of HMDD V3.2 database, 3 miRNA-mRNA pairs (including MYC-miR-145-5p, TNFAIP3-miR-29c-3p and TNFAIP3-miR-335-3p) were obtained, respectively. Subsequently, and 37 TF-mRNA pairs (including 6 mRNAs and 27 TFs) were screened using the online database of TRRUST. Lastly, based on mRNA-miRNA pairs and TF-mRNA pairs, miRNA-TF-mRNA co-regulation network was constructed using Cytoscape (Fig. 6A). The nine genes in prognostic risk model and the TFs that regulated their expression were considered as important genes in meningeoma. Therefore, expression of these genes and TFs were verified using an external dataset GSE54934. Except for HIST1H2BN, the expression of all genes in prognostic risk model showed decreased trend in meningeoma compared to normal, especially in the expression of CXCL2 and SLC2A3 (Supplementary Fig. 3A). In addition, the expression of the 27 TFs showed decreased trend in meningeoma than normal, especially in the expression of CEBPB, DDIT3, ETS2, FOSL2, SMAD1, TCF4. These findings were consistent with our results (Supplementary Fig. 3B). However, not all genes showed significant differences, and this might be explained by the difference in sample size in meningeoma (n = 22) and normal groups (n = 3).
Additionally, in order to further investigate the potential biological functions of the genes involved in miRNA-TF-mRNA co-regulation network, the GO terms and KEGG pathway analyses were conducted. As illustrated in Fig. 6B, the functional processes of the genes were primarily related to biological regulation, while KEGG pathway analysis revealed that the genes were enriched in Tumor Necrosis Factor (TNF) signaling pathway and IL-17 signaling pathway (Fig. 6C).
PPI network and modules analysis
According to STRING database, the PPI network (PPI score = 0.4) was constructed with 27 TFs in miRNA-TF-mRNA co-regulation network and 9 DEGs in the risk model using Cytoscape. As shown in Fig. 7A, there were 36 nodes and 197 interactions in the PPI network. With score > 12, a module with 15 nodes and 88 interactions was further revealed from the PPI network using the MCODE of Cytoscape software (Fig. 7B).
Similarly, based on the 9 DEGs in the risk model, a total of 50 drug-gene interacting pairs, including 3 target genes (CXCL2, CXCL8 and MYC), and 49 drugs were identified using the DGIdb 3.0 database. Furthermore, drug-gene interaction network was constructed by Cytoscape software (Fig. 8).