Multi-omics integrative analysis of the glioma microenvironment reveals IDH-dependent cellular communication and prognostic signatures

Background Glioma is the most common primary malignant brain tumor. Isocitrate dehydrogenase (IDH) mutation status is a critical prognostic marker that fundamentally reshapes the tumor microenvironment (TME), particularly tumor-associated macrophages (TAMs). However, the IDH-dependent cellular communication landscape and its translational implications remain incompletely characterized, and prior studies have relied on single analytical frameworks without cross-method convergence. Methods We integrated single-cell RNA-seq data from 54,978 cells across 19 glioma patients (GSE247824, GSE159416) encompassing 33,789 IDH-wildtype (IDHwt) and 21,189 IDH-mutant (IDHmut) cells. We employed cell-type annotation, differential expression, CellChat intercellular communication with IDH-stratified bootstrap validation (1,000 replicates), trajectory inference, and three complementary machine learning classifiers (elastic net, random forest, XGBoost) with nested cross-validation and SHAP interpretation. To extend beyond ligand-receptor inference, we applied SigXTalk, a hypergraph neural network framework modeling downstream regulatory pathway activation. We further performed NicheNet ligand activity analysis (v2 model, prior knowledge matrix of 1,226 ligands × 33,354 target genes), including both a CellChat-guided candidate screen and a genome-wide ligand screen of 1,042 expressed ligands, and integrated all three methods into a systematic cross-method comparison framework. Multi-level therapeutic target analysis was performed incorporating druggability, IDH stratification, BBB penetration, and drug-target mapping. Survival analysis was performed on an independent CGGA cohort (n = 693 total, n = 656 with complete covariate data, 394 events). Results We identified 12 distinct cell types, with TAM_Macrophage predominant (39%). IDHwt tumors showed marked TAM_Macrophage enrichment (84.6%), while TAM_Microglia was nearly IDHmut-exclusive (99.1%, p < 0.001). CellChat analysis ranked MIF, COLLAGEN, and SPP1 as the top pathways, with 1,000-replicate bootstrap confirming high ranking stability (Spearman r = 0.96, 95%CI: 0.91–0.99). IDH-stratified analysis revealed fundamentally distinct communication architectures: IDHwt was characterized by extracellular matrix remodeling (COLLAGEN, 8.6-fold enriched vs IDHmut), whereas IDHmut showed immune-modulatory signaling (MIF, APP, CD99). SigXTalk hypergraph analysis identified distinct downstream regulatory targets, with CEBPB (log2FC = 4.93, p < 1e-300) as the top IDHmut-specific fidelity target. Genome-wide NicheNet screen (1,042 ligands) confirmed ECM and matrix-associated proteins as the dominant IDHwt signals (MXRA8 r = 0.158, HLA-DOB r = 0.158, C1QB r = 0.152) and inflammatory cytokines as IDHmut signals (IL5 r = 0.086, IL21 r = 0.074, IFNB1 r = 0.074), with CellChat-identified ligands COL1A2 (rank 61/1042) and CCL4 (rank 10/1042) ranking among the top expressed signals. Three-method cross-method convergence confirmed IL1B, CCL4, and CXCL8 as consistently supported inflammatory signaling axes (3 of 33 features, 9.1%; 13/33 [39.4%] supported by ≥ 2 methods). Therapeutic target analysis prioritized 28 candidates, identifying CD74 (Priority Score 3.1/4.0), CD44 (3.1), and IL1B (3.0) as the highest-confidence targets, with CD99 as an IDHmut-exclusive target. Genome-wide elastic net with nested cross-validation achieved AUC = 0.858 (95%CI: 0.828–0.889), with Random Forest achieving the highest AUC = 0.882 (95%CI: 0.855–0.908). Survival analysis confirmed the independent prognostic value of WHO grade (time-averaged HR = 0.29 for LGG vs GBM, 95%CI: 0.23–0.36, p < 0.001). Conclusions This study provides a comprehensive IDH-dependent atlas of the glioma TME, identifies TAM-derived MIF and COLLAGEN signaling as dominant communication pathways with distinct IDH stratification, and establishes a rigorously validated machine learning framework for glioma classification. The orthogonal application of SigXTalk hypergraph analysis and genome-wide NicheNet ligand activity screening reveals IDH-specific signaling programs, while cross-method convergence identifies IL1B, CCL4, and CXCL8 as the most robustly supported inflammatory signaling axis, providing a multi-level roadmap for TAM-directed therapeutic development in glioma.