This study was approved by the Institutional Review Board of Soonchunhyang University (Approval Number: 202405-SB-049) and conducted in accordance with the Declaration of Helsinki. DNA methylation, RNA-seq, proteomic, and clinical datasets of intracranial meningiomas were obtained from cBioPortal. All analyses were performed in R version 4.4.1. The overall workflow is summarized in Figure 1.

Normalized DNA methylation (21,814 CpG sites), RNA-seq (17,338 genes), and proteomic (6,158 proteins) data were analyzed. Ninety patients with matched multi-omics profiles were included. Patients were grouped by recurrence status and WHO grade (Grade 1 = 37, Grade 2 = 37, Grade 3 = 16). Differentially methylated regions (DMRs), differentially expressed genes (DEGs), and differentially expressed proteins (DEPs) were identified using two-group t-tests, and significant features were visualized with volcano plots and heatmaps. Selection thresholds for each omics layer are summarized in Table 1.

The identified DMRs, DEGs, and DEPs were integrated into a unified multi-omics matrix. Principal component analysis (PCA) and multi-omics clustering were performed to explore overall data structure. For connectome construction, Pearson correlation matrices were constructed, and features with correlation coefficients ≥ 0.7 (p = 1.63 × 10⁻¹⁴) for recurrence analysis or ≥ 0.75 (p = 1.79 × 10⁻¹⁷) for WHO grade analysis were retained for connectome generation. Node importance was evaluated by degree, betweenness, closeness, and eigenvector centralities, and connectome maps were visualized to highlight molecular interaction hubs. Gene Ontology enrichment was performed to explore biological processes related to significant modules.

Independent public datasets were retrieved from the NCBI GEO database using the keyword “meningioma.” Datasets containing methylation, transcriptomic, or proteomic profiles were processed using the same analytic pipeline to validate DMRs, DEGs, and DEPs associated with recurrence and grade. Overlapping genes between discovery and validation sets are listed in Supplementary Table 1.

From the multi-omics datasets, 21,814 CpG sites, 17,338 genes, and 6,158 proteins were analyzed. Based on the selection criteria summarized in Table 1, 29 DMRs, 32 DEGs, and 33 DEPs were identified as significantly associated with recurrence (p < 0.001). These features were visualized using volcano plots and heatmaps (Supplementary Figure 1).

For WHO grade progression, differential analyses between Grades 1–2 and 2–3 revealed 70 DMRs, 61 DEGs, and 56 DEPs meeting the same statistical thresholds (Table 1; Supplementary Figure 2). Distinct hypo- and hypermethylation as well as expression patterns were observed across grades, reflecting progressive molecular changes related to tumor aggressiveness.

To characterize molecular interactions underlying these features, we integrated DMRs, DEGs, and DEPs into multi-layer networks. The recurrence network consisted of 94 features (29 DMRs, 32 DEGs, 33 DEPs), and the grade network included 187 features (70 DMRs, 61 DEGs, 56 DEPs). Pearson correlation matrices were constructed, and features with correlation coefficients ≥ 0.7 (recurrence) or ≥ 0.75 (grade) were retained for connectome generation. Graph-based analysis revealed seven distinct molecular clusters for recurrence and six for WHO grade (Figure 2). Centrality mapping identified key network hubs, with LINC01397 emerging as a recurrent central node across omics layers. Gene ontology enrichment indicated involvement of extracellular matrix organization, cell adhesion, and signal transduction pathways (Table 2).

PCA and iCluster analyses were performed using integrated methylation, transcriptomic, and proteomic data from 90 patients. Three molecular clusters were identified (Figure 3A). Recurrence status showed clear separation along the first principal component (PC1; Figure 3B), while WHO grade separation was less distinct (Figure 3C). Cluster 3 predominantly represented nonrecurrent tumors and was characterized by hypomethylation of LINC01397, CHN2, and FAM181A-AS1 with corresponding upregulation of LINC01397 and UBL4B (Figures 3D, E). Boxplots demonstrated hypermethylation and reduced expression of LINC01397 in recurrent tumors (Figure 4). In contrast, cluster 2 was enriched for recurrent and Grade 2 tumors, showing hypomethylation of MMP23A and SSPOP. Progressive methylation and expression shifts of SPAST, NXPH2, SNORD16, and SNORA51 were associated with increasing WHO grade (Figures 5, 6).

Thirteen public omics datasets were analyzed for validation, including five DNA methylation, seven transcriptomic, and one proteomic dataset. DMRs, DEGs, and DEPs associated with recurrence and grade were reanalyzed using the same thresholds, and overlapping genes with our discovery set were summarized (Supplementary Table 1). Consistent differential patterns were observed for several key genes, including CHN2 (methylation), SYNPO2, GSTM5, TCEAL2, MRAP2, SOX11 (gene expression), and KCNMA1, ALPL, SYNPO2, LEPR, GSTM5 (protein expression). Boxplots demonstrated concordant expression or methylation trends across datasets (Supplementary Figure 5), supporting the reproducibility of our findings.

The patterns observed here point to several directions for future work. First, larger and independently collected multi omics cohorts will be needed to confirm whether the signatures identified in this study are robust and reproducible. Genes such as LINC01397, CHN2, and FAM181A-AS1 are particularly strong candidates for further investigation, given their consistent associations with recurrence and grade.

From a clinical perspective, the integration of genomic and epigenomic profiling into routine assessment may eventually help refine diagnosis and guide individualized management. Methylation based markers are especially appealing because of their biological stability and the practicality of incorporating them into clinical workflows. Nevertheless, translating these findings into clinical practice will require validation in more diverse populations and prospective studies that assess their predictive value in real time.

Overall, the present study demonstrates that examining multi omics data through a network connectome framework can reveal patterns that remain hidden when each dataset is analyzed independently. This approach offers a broader view of tumor biology and may ultimately contribute to more precise risk stratification and improved outcomes for patients with meningioma.

This study has several limitations. Survival analysis and formal predictive performance metrics were not performed, reflecting the exploratory design and reliance on publicly available multi-omics datasets with limited longitudinal clinical information. In addition, recurrence and WHO grade were used as surrogate clinical endpoints, which may not fully capture patient outcomes. Validation in larger, well-annotated cohorts and prospective studies will be necessary to confirm the biological and clinical relevance of the identified molecular features.