Primary central nervous system (CNS) tumors account for approximately 1.6% of all cancers (1), with gliomas representing 25.1% of primary brain tumors and 80.8% of malignant CNS neoplasms (2). The 2021 WHO classification (3) categorizes gliomas into astrocytomas, oligodendrogliomas, and glioblastomas, with molecular subtyping further stratifying prognosis: low-grade gliomas exhibit a median survival of 7 years (4), whereas glioblastomas demonstrate a dismal median overall survival of 14 months (5). Despite prolonged survival in low-grade cases, inevitable progression to high-grade disease persists (6). While therapeutic advances have modestly improved outcomes, accurate assessment of natural tumor behavior remains challenging due to universal treatment interventions. Recent shifts toward molecular classification (7) have redefined histologically low-grade but IDH-wildtype tumors as glioblastomas, underscoring the critical role of biomarkers in elucidating glioma biology (1). Key discoveries include glioma stem cells with self-renewal capacity (8) and profound inter−/intra-tumoral heterogeneity (9). Established biomarkers such as MGMT promoter methylation, 1p/19q co-deletion, IDH mutations, and EGFR amplification guide clinical decisions (10), while stemness markers like CD133 and ALDH1 highlight therapeutic resistance mechanism (11). However, limitations in predictive accuracy and standardization necessitate novel biomarkers to refine prognostic frameworks.

CD9, a member of the tetraspanin family, is a cell surface glycoprotein characterized by four transmembrane domains. This unique structural configuration enables CD9 to interact with multiple transmembrane receptors, forming multiprotein complexes that regulate diverse cellular processes such as cell fusion, adhesion, and motility (12). CD9 plays a pivotal role in intercellular communication and signaling networks. Rather than functioning as a cell surface receptor itself, CD9 acts as an organizer of multimolecular complexes, which include integrins, immunoglobulin superfamily members, heparin-binding epidermal growth factor-like growth factor, claudin-1, and other tetraspanin (13). CD9 has been implicated in critical biological processes such as cell adhesion and migration (14), cancer progression and metastasis (14, 15), platelet activation and aggregation (16), sperm-egg fusion during mammalian fertilization (17, 18), humoral immune responses (19), allergic reactions (20), and viral replication of HIV-1 and influenza (21). This study aims to establish CD9 as a prognostic biomarker for glioma through integrative multi-omics analysis, delineate its interaction networks driving tumor progression, and identify therapeutic agents targeting CD9-associated pathways. By bridging molecular insights with clinical translation, we propose a novel prognostic model and repurposed therapeutics to address unmet needs in glioma management.

Clinical and genetic data of glioma patients were collected from the Chinese Glioma Genome Atlas (CGGA) (22) and The Cancer Genome Atlas (TCGA) databases. Following quality validation, 1,328 glioma patients with Radiotherapy, Chemotherapy, WHO grade, IDH mutation status, MGMT promoter methylation, 1p/19q codeletion status, age, gender, and CD9 expression were included. The patient inclusion and exclusion flowchart is provided in Supplementary Figure 1. Genetic data were preprocessed, and the optimal cutoff value for CD9 was determined using the maximum selection rank statistics method. The correlation between the CD9 gene and glioma prognosis was determined based on Kaplan–Meier (KM) survival curves. Univariate Cox regression analysis was performed using the CGGA database, with variables showing p < 0.05 and meeting the PH assumption considered as glioma prognosis-related factors. These factors were subsequently incorporated into a multivariate Cox proportional hazards model to estimate the partial regression coefficients (β) for each variable, and patient survival probabilities were calculated based on the model. Clinical and molecular features were integrated to construct a nomogram. Calibration curves were plotted to assess the agreement between predicted and observed survival events. The discriminative ability of the model was validated using TCGA as an independent dataset.

The genome-wide association study (GWAS) summary statistics for glioma were obtained from the FinnGen database.¹ Genetic associations between CD9 and single nucleotide polymorphisms (SNPs) were derived from the eQTLGen consortium.² All summary-level data were de-identified, and no additional ethical approval was required. Instrumental variables (IVs) were selected by identifying SNPs significantly associated with CD9 and glioma at a genome-wide significance threshold (p < 5 × 10^–8). Candidate SNPs were further subjected to linkage disequilibrium (LD) analysis to exclude weak instruments (r² < 0.001 within a 10,000 kb window). Exposure and outcome datasets were harmonized to align allele effects, and weakly palindromic SNPs were removed.

Mendelian randomization (MR) analyses were performed using a random-effects inverse variance weighted (IVW) method to estimate the causal effect of CD9 on glioma. Sensitivity analyses included weighted median (WM) estimation to validate robustness. Heterogeneity of individual SNP effects was assessed using Cochran’s Q test. Horizontal pleiotropy was evaluated via MR-Egger intercept analysis and the MR pleiotropy residual sum and outlier (MR-PRESSO) test to detect and remove outliers. A leave-one-out analysis was conducted to examine the influence of individual SNPs on overall estimates. The KEGG enrichment analysis of differentially expressed genes (DEGs) between CD9 high- and low-expression groups was performed using the ClusterProfiler R package. The study employed the Seurat (23) R package to investigate the distribution and dynamics of CD9 across different cell types using single-cell data from 24,131 cells derived from 20 adult glioblastoma samples. First, data normalization was performed to identify highly variable genes across cells. The uniform manifold approximation and projection (UMAP) algorithm was then applied for nonlinear dimensionality reduction. Cell annotation and clustering were conducted based on marker genes for each cell cluster, following the methodology described by Nefel et al. (24). Finally, CD9 gene expression levels were identified and localized across distinct glioma cell subpopulations at the single-cell resolution. Comprehensive methodological details are illustrated in the integrative flowchart provided in Figure 1.

Our analysis of 1,033 glioma patients demonstrated a strong correlation between elevated CD9 expression and poor prognosis, with high CD9 expression associated with significantly shorter overall survival compared to low expression. This positions CD9 as a pivotal prognostic biomarker in glioma, offering independent prognostic value beyond established factors such as age, WHO grade, IDH mutation, and 1p19q codeletion status in multivariate Cox analysis. These findings align with previous studies highlighting tetraspanins in cancer biology (10), yet extend the understanding by integrating CD9 into a multifactorial survival model, thereby enriching insights into glioma progression.

Single-cell sequencing revealed enriched CD9 expression in neural progenitor-like (NPC-like) glioma stem cell subpopulations, with pseudotemporal trajectory analysis showing progressive downregulation during differentiation toward oligodendrocyte precursor-like (OPC-like) cells. This aligns with the stemness maintenance theory, where NPC-like cells contribute to therapy resistance and recurrence (31). Notably, CD9 exhibited differential expression in tumor-associated macrophages (TAMs), and CellChat-predicted enhanced TGF-β signaling suggests its role in modulating tumor-immune crosstalk, warranting validation via spatial transcriptomics or multiplex immunofluorescence. Protein–protein interaction (PPI) network analyses and DsigDB drug predictions further identified CD9 as a therapeutic target, with emetine emerging as a promising candidate due to its reported anti-glioblastoma activity.

While existing glioma prognostic models predominantly rely on clinicopathological features (24), our CD9-integrated model synergizes traditional and molecular biomarkers, achieving superior predictive accuracy. This advancement supports personalized treatment strategies, enabling risk stratification for aggressive therapies (e.g., CD9-targeted agents) in high-risk patients and conservative management in low-risk cohorts. Moreover, MR established the causal relationship between CD9 and glioma by integrating SNPs associated with CD9 and glioma GWAS data. To our knowledge, this study represents one of the first to employ MR in identifying potential therapeutic targets for glioma, leveraging large-scale, publicly available glioma GWAS datasets. By implementing MR analyses through the FinnGen dataset, we mitigated confounding biases inherent in observational measurements, thereby confirming the causal link between CD9 and glioma. This approach provides robust causal evidence to inform subsequent mechanistic and translational investigations. Single-cell sequencing further underscores its utility in dissecting cellular heterogeneity and developmental trajectories, as demonstrated by our cell subpopulation annotation guided by marker genes defined by Neftel et al. (32), which clarified CD9’s role in promoting tumor cell states.

Although this study primarily focuses on the prognostic value of CD9 in glioma, the exploration of its underlying mechanisms provides deeper insights into glioma biology. CD9, a transmembrane protein, has been reported to regulate tumor cell proliferation, apoptosis, migration, and invasion in various cancers (24, 33). Notably, CD9 is preferentially expressed in glioblastoma stem cells (GSCs) and is indispensable for maintaining their self-renewal capacity (34). Co-immunoprecipitation and immunofluorescence analyses revealed that CD9 stabilizes the interleukin-6 (IL-6) receptor subunit gp130 by preventing its lysosomal ubiquitination-mediated degradation, thereby activating the BMX-STAT3 signaling pathway to sustain GSC stemness and tumorigenicity (35). This finding offers critical insights into CD9’s role in glioma pathogenesis. In glioblastoma, CD9 may further promote tumor cell migration and invasion by participating in extracellular vesicle-mediated processed (36). Additionally, CD9’s association with GSC properties, including self-renewal and chemoresistance (37), underscores the need for further mechanistic studies. Future research should delineate CD9’s signaling networks and molecular interactions (38), providing a theoretical foundation for developing CD9-targeted therapies.

KEGG enrichment analysis revealed that CD9 high-expression-associated genes were significantly enriched in the calcium signaling pathway (p = 8.81 × 10⁻⁷), synaptic vesicle cycle pathway (p = 3.97 × 10⁻⁸), and GABAergic synapse pathway (p = 1.39 × 10⁻⁵). Dysregulated calcium signaling may drive glioma proliferation and invasion by modulating glutamate release in the tumor microenvironment (25), while synaptic pathway enrichment underscores the critical role of neuron-glioma interactions in tumor progression, consistent with recent evidence that neuronal activity directly stimulates glioma growth (29). Conversely, CD9 low-expression-associated genes were enriched in immune-regulatory pathways, such as cytokine-cytokine receptor interactions (p = 3.79 × 10⁻⁷), suggesting that CD9 may suppress chemokine-mediated immune cell infiltration (e.g., CCL5, CXCL10) to shape an immunosuppressive microenvironment (28). This dual role of CD9 highlights its potential to drive gliomagenesis through both pro-oncogenic signaling and immune evasion mechanisms.

This study’s retrospective design, though mitigated by robust statistical methods, necessitates prospective validation. While CGGA and TCGA datasets provide broad coverage, expanding to multiethnic cohorts and incorporating normal tissue controls would enhance generalizability. Furthermore, the eQTLGen used in this study included individuals of non-European ancestry, whereas the glioma GWAS data were derived from European populations. This population heterogeneity may introduce potential bias in MR effect estimates. The limited single-cell sample size (n = 20) and absence of spatial resolution constrain tumor-specific expression profiling. Future work should integrate multi-omics data and combine CD9 with emerging biomarkers to refine prognostic models. Mechanistic exploration of CD9’s role in GSC-TAM crosstalk and extracellular vesicle biology, alongside preclinical testing of CD9-targeted therapies (39), remains critical to advancing glioma therapeutics.

This study developed and validated a CD9-based prognostic model for glioma, with MR analyses confirming the causal relationship between CD9 and gliomagenesis and progression. Single-cell analysis and PPI network exploration revealed the distribution patterns, dynamic expression trends, and interaction networks of CD9 across distinct glioma cell subpopulations. These findings suggest that CD9 may promote tumor proliferation and invasion by inducing differentiation of glioma stem cells. Drug prediction analysis further identified emetine as a potential CD9-targeting therapeutic agent. Collectively, this study systematically elucidates the central role of CD9 in glioma through prognostic modeling, molecular mechanism investigation, and therapeutic target prediction, providing a multifaceted framework for advancing glioma research and therapy.