Introduction

Vestibular schwannoma (VS), also known as acoustic neuroma, is a type of tumor originating from the Schwann cells of the 8th cranial nerve [1]. Accounting for approximately 85% of cerebellopontine angle tumors, VS is a slow-growing benign neoplasm that nonetheless poses risks to adjacent intracranial structures due to its location [1]. The most common symptom is progressive hearing loss (~60%), with possible trigeminal neuralgia, vertigo, and headache due to brainstem compression [1,2]. There are two types of VS, unilateral sporadic and bilateral hereditary forms, and approximately 95% of these are sporadic and unilateral tumors [3].

VS is among the most prevalent benign intracranial tumors, following meningiomas and pituitary adenomas, with an estimated incidence of approximately 3.0–5.2 cases per 100 000 person‑years and a lifetime prevalence exceeding 1 in 500 individuals, reflecting increased detection and clinical recognition in recent years [4,5]. The primary molecular driver is biallelic NF2 inactivation, encoding the cytoskeletal regulator merlin; yet VS shows low mutational burden beyond NF2, with additional genetic and epigenetic contributors increasingly implicated [6]. Recent multi-omic studies reveal molecular heterogeneity, including immunogenic and proliferative subtypes, yet these studies have been limited by sample numbers and inconsistent replication across cohorts [7]. Single-cell transcriptomic profiling has uncovered diverse Schwann cell states—including injury-like and vascular endothelial growth factor A (VEGFA)-expressing subpopulations—associated with growth patterns and microenvironmental interactions, providing novel insight yet underscoring the complexity and incomplete understanding of VS pathogenesis [8,9].

A variety of therapeutic strategies are currently available for VS management. Wait-and-scan with periodic imaging is often the initial approach, while additional treatment options include pharmacological agents such as bevacizumab, everolimus, and lapatinib, as well as radiotherapy or surgery [1]. Although promising results have been reported on the use of bevacizumab in treating schwannoma, these data are based solely on cases of neurofibromatosis type 2 (NF2) [10]. Despite ongoing advances that have reduced morbidity and mortality in recent years, there remains no safe and effective pharmacological therapy capable of reliably improving tumor size or preserving hearing function [11]. Surgery is a good option for achieving tumor control in VS patients; however, it is among the most technically challenging tumors to remove from the cranial base [12]. On the other hand, the radiotherapy option carries risks such as radiation-induced tumor formation [13].

There are some previous studies that have investigated the transcriptomic profile of VS but these studies have largely been conducted with limited sample sizes based on single cohorts or have focused on specific biological aspects [14–16]. In a recent single-cell multi-omic study, the cellular heterogeneity of the VS microenvironment and the presence of “injury-like” Schwann cell states were characterized in detail [9]. Additionally, Sung and colleagues conducted a comprehensive study integrating bulk transcriptome data with single-cell RNA-seq analyses to elucidate tumor growth dynamics in VS and identified a 23-gene signature associated with growth rates [16]. While these studies offer important mechanistic and cell-type–specific perspectives, they remain focused on specific phenotypic aspects and do not address genome-wide transcriptional alterations across multiple independent cohorts. Also, an earlier integrated analysis of VS transcriptomic data explored both mRNA and microRNA expression profiles from multiple public microarray datasets, identifying differentially expressed mRNAs and miRNAs and constructing protein–protein interaction and miRNA–mRNA regulatory networks [17]. However, the datasets included in that analysis were generated using different microarray platforms, introducing inherent platform-specific variability that may influence gene coverage, probe design, and expression estimates. Given these limitations, a deeper understanding of the molecular mechanisms underlying VS is important for the development of targeted therapies and the identification of reliable biomarkers. Although recent studies suggest that inflammation and angiogenesis may play critical roles in tumor growth, the molecular pathogenesis of VS remains incompletely understood [7,18]. Transcriptomic profiling offers a valuable approach to address this knowledge gap. However, studies analysing the genetic profiles of patients with VS are limited and often restricted by small sample sizes.

Here, we conducted a meta-analysis using microarray datasets from VS patients in the Gene Expression Omnibus (GEO) database, incorporating analytical models that account for potential covariates. Through this approach, we aimed to comprehensively identify consistent mRNA-level alterations, drug–gene interactions, and biological functions associated with differentially expressed genes (DEGs) in VS. Moreover, we provide a comprehensive overview of the molecular pathobiology of VS by systematically examining a wide range of biological processes, from sensory processes to cell adhesion, tumour growth, immune response, and metabolic functions, using GO analyses.

Materials and methods

Results

We performed a meta-analysis with and without covariate adjustment on 21,113 genes across four schwannoma brain datasets from the GEO database. Substantial between-study heterogeneity was observed across the meta-analysed genes, with a median I² of 89.22% across the background gene set. Among genes classified as significant DEGs, 75.01% showed I² values greater than 50%, and 64.70% had significant Cochran’s Q tests (p < 0.05). This high level of heterogeneity supports the use of a random-effects framework, but also indicates that individual pooled metaLFC estimates should be interpreted with caution. To further evaluate the robustness of the findings in this context, we performed a leave-one-out sensitivity analysis excluding the smallest cohort, GSE108237. MetaLFC values were strongly correlated between the main and leave-one-out analyses across common genes (Spearman’s rho = 0.874), and 82.8% of genes retained the same direction of effect in the unadjusted model, with a similar pattern observed in the covariate-adjusted model (83.2%).

Without including covariates, 7,354 genes reached nominal significance (p < 0.05), and 5,421 genes remained significant after adjusting for multiple comparisons (padj <0.05). Applying thresholds of |metaLFC| > 0.5 and padj < 0.05, 3,204 genes showed differential expression, including 2,132 upregulated and 1,072 downregulated genes. Raising the logFC threshold to |metaLFC| > 1 and padj < 0.05, we identified 683 upregulated and 412 downregulated genes. The overall distribution of these meta-analysed expression changes is shown in the volcano plot in Fig 1A, where gene labels indicate the 20 differentially expressed genes with the smallest adjusted p-values. Among the significant DEGs (|metaLFC| > 1 and padj <0.05), the most strongly downregulated genes based on metaLFC were MFAP5 (−6.06), ADH1B (−5.17), SLC22A3 (−4.95), FABP4 (−4.64) and SFRP2 (−4.2), whereas the most strongly upregulated genes based on metaLFC wereGFRA3 (3.48), OLR1 (3.44), GPR83 (3.39), L1CAM (3.34) and ADAM23 (3.25).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Volcano plot of meta-analysis results.

This figure presents volcano plots of meta analysed top 20 genes in VS patients, divided into two panels: A) results without covariate adjustment and B) results with covariate adjustment. Each gene is displayed as an individual point, plotted according to its meta-analysed log2 fold change (metaLFC) on the x-axis and the –log10 of the adjusted p-value (padj) on the y-axis. Genes with |metaLFC| > 1 and padj < 0.05 were classified as differentially expressed, with upregulated genes shown in red and downregulated genes shown in blue. Gene labels indicate the 20 differentially expressed genes with the smallest adjusted p-values in each model.

https://doi.org/10.1371/journal.pone.0353343.g001

Meanwhile, when covariates were included, 7,350 genes were nominally significant (p < 0.05), and 5,532 genes remained significant after multiple comparison correction (padj < 0.05). Applying a |metaLFC| > 0.5 and padj < 0.05 cutoff resulted in 3,250 differentially expressed genes, comprising 2,143 upregulated and 1,107 downregulated genes. A more stringent cutoff (|metaLFC| > 1 and padj < 0.05) reduced this set to 690 upregulated and 413 downregulated genes. The covariate-adjusted results are shown in Fig 1B, with gene labels representing the 20 significant DEGs with the smallest adjusted p-values. The strongest downregulated genes based on metaLFC were unchanged compared with the unadjusted model and showed comparable effect sizes. Among the upregulated genes, those with the highest metaLFC values were OLR1 (3.579), GFRA3 (3.47), L1CAM (3.36), GPR83 (3.32), and MOXD1 (3.21).

Gene Ontology (GO) enrichment analysis

To investigate the functional enrichment of DEGs, we conducted a Gene Ontology (GO) enrichment analysis on significant genes identified from two meta-analysis models, applying a significance threshold of |metaLFC| > 0.5 and padj < 0.05. We extracted the top 20 biological processes (BP) based on the lowest adjusted p-values (false discovery rate, FDR). The GO analysis results are presented in Fig 2. In the model without covariates, top biological processes were predominantly associated with sensory perception and cell adhesion dynamics, including cell junction disassembly, amoeboidal-type migration, and substrate adhesion-dependent cell spreading (Fig 2A). We used bar plots for each graph, with dot colour indicating enriched gene count per pathway. A total of 115 significant GO term enrichments were identified in the model without covariates, and are provided in S2 Table.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Top GO Enrichment Analysis with Gene Counts.

The top 20 enriched biological processes (GO) were identified using an adjusted p-value threshold of < 0.05 (FDR-corrected) and |metaLFC| > 0.5. GO terms were ranked according to significance and visualized using bar plots, where bar length represents -log10(adjusted p-value) and dot color indicates the number of enriched genes associated with each GO term. A) the results without adjusting for covariates and B) the results with adjusting for covariates.

https://doi.org/10.1371/journal.pone.0353343.g002

In the with covariates model, among the top enriched terms, processes associated with the detection of chemical stimulus and sensory perception, including the detection of sensory stimulus involved in smell, showed the highest statistical significance. In addition, several signalling-related pathways, such as Ras protein signal transduction and regulation of small GTPase-mediated signal transduction, were significantly enriched, indicating the involvement of intracellular signalling mechanisms. Also, immune-related processes, including antigen processing and presentation and lymphocyte proliferation were prominently represented, suggesting a potential role of immune regulation within the analysed gene set. The top 20 significantly enriched biological processes are shown in Fig 2B, while the complete list of 170 enriched GO terms is provided in S3 Table.

DrugBank analysis

DrugBank analysis revealed pharmacological associations between the DEGs and several known drugs, and possible pharmacological relationships. The ADH1B gene was downregulated, and an interaction between this gene and fomepizole was identified. Fomepizole was also associated with ADH1C, another downregulated alcohol dehydrogenase gene. Fomepizole is known as a competitive inhibitor of alcohol dehydrogenase enzymes. Also, the gene FCGR3A, which was upregulated, was found to interact with human immunoglobulin G (IgG). The FCGR3A gene may affect the binding capacity of IgG. A total of 154 genes displaying significant drug–gene interactions were identified using the stringent threshold of |metaLFC| > 1 and padj < 0.05. Among these, the top 10 downregulated and top 10 upregulated genes, ranked by metaLFC values, are presented in Figs 3 and 4, respectively. Several genes were found to interact with multiple drugs, resulting in a cumulative total of 1009 drug–gene interactions. The full list of identified interactions is provided in the S4 Table.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Drug-gene Interaction Network for the Top 10 Downregulated Genes.

Networks were constructed using experimentally supported drug-target interactions from DrugBank, focusing on significantly differentially expressed genes (|metaLFC| > 1 and padj < 0.05).

https://doi.org/10.1371/journal.pone.0353343.g003

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Drug-gene Interaction Network for the Top 10 Upregulated Genes.

Networks were constructed using experimentally supported drug-target interactions from DrugBank, focusing on significantly differentially expressed genes (|metaLFC| > 1 and padj < 0.05).

https://doi.org/10.1371/journal.pone.0353343.g004

Discussion

In this study, we conducted the first transcriptome-wide meta-analysis of vestibular schwannoma (VS), integrating multiple independent Affymetrix-based microarray cohorts to derive a statistically robust and platform-homogeneous assessment of genome-wide transcriptional alterations. Using a random-effects meta-analytic framework, we identified more than 3,200 differentially expressed genes (DEGs), based on a cutoff of |metaLFC| > 0.5 and padj < 0.05, with a high degree of concordance between the covariate-adjusted and unadjusted models. Application of a more stringent threshold (|metaLFC| > 1, padj < 0.05) yielded approximately 1,100 DEGs in both analytical conditions, underscoring the stability and reproducibility of the observed transcriptional signals. Functionally, these DEGs were enriched for several broad biological themes, including pronounced downregulation of extracellular matrix and stromal components (e.g., MFAP5, FABP4, DCN), alterations in sensory- and synapse-related genes implicated in auditory function (such as SLC22A3 and LGI1), and increased expression of immune- and inflammation-associated pathways (including TREM2, CCL3, and CCL4) alongside proliferative regulators (notably CCND1 and RAB31). Notably, many of the most strongly dysregulated genes identified in this meta-analysis have not previously been evaluated in VS, highlighting the added value of a genome-wide, discovery-oriented approach. Collectively, these quantitative findings are consistent with a multifaceted transcriptional landscape in VS that may involve contributions from tumor-intrinsic and microenvironment-associated processes, including immune-related signaling and tumor-associated growth pathways; however, further validation using complementary and higher-resolution transcriptomic approaches (e.g., single-cell, spatial, or integrative multi-cohort analyses) will be required to clarify the underlying cellular and biological context. These findings may also help prioritize candidate genes and pathways for future biomarker development and therapeutic investigation. To contextualize our findings, we compared our DEGs with previously published VS studies, including bulk microarray analyses and recent RNA-seq–derived signature studies [14–16]. For upregulated genes, several were consistently identified across datasets, most notably NRG1, along with SPP1 and CDH2, supporting prior VS transcriptomic findings. Consistent with previous studies [14,15], SLC22A3 and PRRX1 were significantly downregulated in our study. Additional genes, including FABP4, MFAP5, and SFRP2 [15], as well as TNNC1, EBF1, and EBF2 [14], were shared between two studies, indicating a combination of conserved and context-dependent transcriptional changes. GFRA3, GPR83, and MOXD1 were among the most strongly dysregulated genes in our dataset and have not been prominently reported in previous VS transcriptomic studies, suggesting that they may represent additional candidate genes requiring further validation.

Partial concordance with the 24-gene communication signature proposed by Sung et al. was observed, primarily driven by SPP1 and members of the CCL chemokine family, including CCL3 and CCL4, indicating shared enrichment of immune and macrophage-associated pathways [16]. In addition, we assessed the overlap between the 24-component Schwannoma signature reported by Sung et al. and the results of our meta-analysis. Since Sung et al. define the schwannoma signature as a 24-gene set in their article and corresponding Supplementary Table 4, we used the 24-component version for this comparison. Nineteen of the 24 signature components (79.2%) were represented by at least one significantly differentially expressed gene in our study. This component-level comparison identified partial recovery of the corresponding gene families or signaling groups, whereas gene-level overlap and directionality are provided separately in S5 Table. Furthermore, genes reported in recent VS single-cell studies, including SPP1 and NRG1, were also identified as upregulated in our dataset [29,30]. In addition, pathways related to macrophage infiltration and cell–cell communication described in these studies are supported by the differential expression of genes such as CCL3, CCL4, and TREM2 in our analysis [31]. Together, these findings suggest that key microenvironmental and intercellular signaling features observed at the single-cell level are partially reflected in our meta-analysis data. The limited gene-level overlap observed across studies is methodologically expected and should be interpreted in the context of differences in sequencing platforms, statistical thresholds, and growth-stratification strategies, all of which can influence DEG detection.

One of the most prominent and consistent transcriptional features identified in this meta-analysis was the marked downregulation of extracellular matrix (ECM)– and stroma-associated genes in VS. The strongest effect size across the entire dataset was observed for MFAP5 (metaLFC = −6.06), an ECM-associated protein implicated in microfibril organization and stromal structure [32]. In addition to MFAP5, several genes involved in extracellular matrix integrity and stromal organization were among the most significantly downregulated transcripts, including FABP4 (metaLFC = −4.64), DCN (metaLFC = −3.67), OMD (metaLFC = −4.17), and SFRP2 (metaLFC = −4.20). The magnitude of suppression across these genes points toward a large-scale alteration of the extracellular architecture. While MFAP5 expression has been reported to be elevated in stromal compartments of certain malignant tumors, particularly within cancer-associated fibroblasts, its strong downregulation in VS suggests a context-dependent regulatory pattern that may reflect tumor-type–specific stromal remodeling [32–35]. Nevertheless, it remains to be elucidated whether these changes primarily occur within the neoplastic Schwann cells or reflect a relative depletion or phenotypic shift of stromal fibroblast populations within the tumor bulk. To our knowledge, MFAP5 and several of the accompanying ECM-related genes identified here have not previously been evaluated in VS at the transcriptome-wide level. Collectively, these findings suggest that disruption of extracellular matrix composition and stromal organization is a prominent transcriptional feature of VS.

Beyond extracellular matrix remodeling, our meta-analysis revealed consistent downregulation of genes implicated in sensory support and synaptic function, which may reflect gene expression alterations relevant to auditory physiology in VS. SLC22A3 was among the most strongly downregulated genes (metaLFC = −4.95). This gene encodes an organic cation transporter expressed in supporting cells across sensory epithelia and is involved in monoaminergic signaling and glial homeostatic functions [36]. Experimental evidence indicates that SLC22A3-mediated transport influences serotonin availability, epigenetic regulation, and inhibitory neurotransmission, processes relevant to sensory signal modulation [37]. In line with its proposed regulatory role, SLC22A3 downregulation has been reported across multiple cancer types, consistent with a hypothesized tumor suppressor function [38,39]. In addition to SLC22A3, LGI1—a gene highly expressed in neuronal tissues and involved in synaptic stability—was significantly downregulated in VS (metaLFC = −2.55). Reduced LGI1 expression has previously been associated with glial tumors and neurological disorders featuring auditory manifestations [40,41]. These findings are consistent with gene expression changes affecting sensory and synaptic pathways. However, the bulk nature of the data does not allow resolution of cell-type–specific contributions or causal relationships. Within this context, the concurrent suppression of SLC22A3 and LGI1 may represent a molecular correlate of auditory dysfunction in VS that is not solely explained by tumor mass effects and warrants further investigation using higher-resolution transcriptomic approaches. A further prominent feature of the VS transcriptome identified in this meta-analysis was the upregulation of genes associated with immune and inflammatory processes, suggesting transcriptional enrichment of these pathways in VS. Among these, L1CAM was significantly upregulated (metaLFC = 3.34). Although classically involved in neural cell adhesion, aberrant L1CAM expression has been widely reported in cancer and linked to inflammatory signaling, including activation of NF-κB–dependent pathways [42]. Consistent with prior transcriptomic studies reporting inflammatory pathway enrichment in VS, the observed L1CAM upregulation may reflect inflammation-associated transcriptional changes within the tumor microenvironment [43]. In addition, TREM2 was significantly upregulated (metaLFC = 2.58). TREM2 expression has been shown to correlate with immune cell infiltration in VS [44]. In line with this association, the chemokines CCL3 and CCL4 were also markedly upregulated in our analysis (metaLFC = 3.02 and 2.98, respectively). These CC chemokines are key mediators of immune cell recruitment, including monocytes, macrophages, dendritic cells, and T cells [45]. Together, the coordinated upregulation of L1CAM, TREM2, CCL3, and CCL4 may suggest immune-associated gene expression in VS. These immune-related gene expression changes may, in part, reflect contributions from non-neoplastic components of the tumor microenvironment.

Genes associated with cell cycle regulation and tumor growth pathways were also consistently upregulated in VS. Among these, CCND1 was significantly increased in the meta-analysis (metaLFC = 2.16). CCND1 encodes cyclin D1, a key regulator of G1–S phase cell cycle progression, and its dysregulation has previously been associated with increased risk and growth of intracranial tumors, including acoustic neuroma [46]. Elevated CCND1 expression has been shown to promote cellular proliferation and tumor expansion across multiple tumor types, suggesting that its upregulation in VS may be consistent with a transcriptional state supporting sustained, non-malignant growth tendencies [47,48]. In addition, RAB31 was among the most strongly upregulated genes (metaLFC = 2.91). RAB31 is a member of the Ras-related small GTPase family and has been associated with pathways related to tumor development, survival signaling, and resistance to apoptosis in several cancers, including glioma [49]. Functional studies in other tumor systems have linked RAB31 overexpression to enhanced proliferative capacity and reduced cell death [49]. To date, RAB31 has not been evaluated in the context of VS, and its upregulation may reflect shared features of growth- and survival related transcriptional patterns. These associations resemble transcriptional patterns linked to proliferation and survival in other tumors, although the current meta-analysis of bulk transcriptomic data cannot determine whether these signals originate from tumor cells, stromal components, or infiltrating immune cells.

Developmental and neural crest–associated transcriptional patterns were evident in VS. MOXD1 was significantly upregulated (metaLFC = 3.21) and is known to be enriched in neural crest–derived Schwann cell precursors [50]. While MOXD1 expression has been associated with aggressive tumor behavior in glioblastoma, its loss in neuroblastoma correlates with advanced disease, highlighting a context-dependent role across neural crest–related tumors [51,52]. In the context of VS, a benign Schwann cell–derived neoplasm, MOXD1 upregulation may reflect developmental gene expression features rather than malignant progression. The concurrent upregulation of MOXD1 alongside proliferative, sensory, and immune-associated gene expression changes may indicate broader developmental influences within the VS transcriptome, potentially affecting interactions within the tumor microenvironment [51].

Our meta-analysis identified a substantial network of 1,009 drug–gene interactions, providing a pharmacological landscape that partially overlaps with the VS transcriptome. DrugBank analysis revealed that several DEGs intersect with previously characterized pharmacological targets, providing a descriptive, hypothesis-generating framework for candidate prioritization. Several factors constrain direct clinical inference: the blood–brain barrier limits central nervous system bioavailability for many compounds, druggability in the VS context remains unestablished for the majority of identified targets, and tumor specificity and systemic toxicity profiles have not been evaluated. Nonetheless, such mappings may support the prioritization of candidate targets for future mechanistic and preclinical studies.

Various clinical and biological covariates may influence gene expression patterns in VS. In a multiplatform molecular analysis, no significant differences in age or sex were detected between VS subgroups [53]. Similarly, a review on VS tumor growth reported that sex is not a determining factor, while age has been suggested as a potential determinant in only a limited number of studies [54]. The same review also indicated that tumors with cystic components may represent a significant risk factor for tumor growth. Given the potential influence of such variables, covariates including age, sex, radiation status, VS type, and disease state were considered in our analyses when available. To address potential biases arising from the inconsistent availability of clinical metadata in public repositories, we adopted a dual-modeling approach with and without covariate adjustment. As shown in S1 Table, covariate availability varied across studies: while datasets such as GSE141801 provided detailed information on radiation history and VS subtypes, others (e.g., GSE39645) lacked clinical metadata. Although missing covariates may theoretically introduce residual confounding, the high consistency between adjusted and unadjusted models suggests that the identified gene expression signatures are only weakly influenced by age, sex, and disease state.

A key strength of this study lies in its comprehensive meta-analytic design, which integrated multiple independent Affymetrix-based microarray datasets to generate robust, cross-cohort transcriptional signatures while minimizing technical heterogeneity. The inclusion of both covariate-adjusted and non-adjusted models demonstrated high concordance of differential expression, reinforcing the reproducibility of the observed alterations. The genome-wide, hypothesis-free approach enabled identification of convergent biological themes—including extracellular matrix remodeling, sensory- and synapse-related pathways, immune activation, and proliferative signaling—rather than isolated gene-specific effects. Notably, drug–gene interaction mapping uncovered a substantial number of pharmacological associations, suggesting that parts of the VS transcriptomic landscape intersect with existing drug targets. Nonetheless, several limitations should be acknowledged. The number of publicly available VS datasets remains limited, reflecting the challenges of obtaining surgical tissue and restricting opportunities for large-scale, standardized transcriptomic studies. In addition, bulk transcriptome profiling precludes direct resolution of cell-type–specific contributions and spatial heterogeneity within the tumor microenvironment. Despite these constraints, the integrated transcriptional framework provided here delineates a multifaceted molecular landscape of VS, highlights previously uncharacterized genes and pathways, and establishes a foundation for future functional studies and potential translational applications, including biomarker discovery and mechanistic investigation of tumor growth and auditory dysfunction.

In conclusion, this meta-analysis defines a multifaceted transcriptional landscape in VS, yet several key questions remain. The cellular sources of immune and stromal alterations, the functional consequences of sensory- and synapse-related gene dysregulation, and the mechanistic role of previously uncharacterized genes such as MFAP5, RAB31, and MOXD1 remain incompletely understood. Future studies integrating single-cell and spatial transcriptomics, longitudinal growth analyses, and functional validation will be essential to clarify the molecular mechanisms driving tumor progression, sensory deficits, and microenvironmental interactions, ultimately informing biomarker development and therapeutic strategies in VS.

Supporting information

Acknowledgments

The author has no acknowledgements to declare.