Abstract Background Meningioma is a common primary intracranial tumor with high recurrence and metastasis rate. Delineating the pathological ecosystem at the brain-tumor interface (BTI) of meningioma is critical for understanding the mechanisms of tumor metastasis and developing effective new therapies. Methods To identify biomarkers of early metastasis and discover potential therapeutic targets, we integrated single-cell transcriptome datasets of meningioma, and identified the cell populations and molecular signatures uniquely present at the BTI. Results A specific BTI-enriched tumor cell population with a pro-EMT (epithelial mesenchymal transition) characteristics was associated with invasion and metastasis, and ANXA2 and COL5A1 were detected as the biomarkers for these BTI-enriched tumor cells. Additionally, we characterized the BTI-specific immunosuppressive microenvironment composed of SPP1^+ tumor-associated macrophages, as well as specific endothelial cells (ACKR1^high) and pericytes (THY1^high) promoting the highly malignant invasive state of angiogenesis. Conclusions Collectively, BTI in meningioma is a metastatic and immunosuppressive zone. We have discovered potential biomarkers that help detect early metastasis and recurrence of meningioma. Supplementary Information The online version contains supplementary material available at 10.1186/s12967-025-06935-z. Keywords: Meningioma, Brain-tumor interface (BTI), Epithelial-mesenchymal transition (EMT), ANXA2, COL5A1 Introduction Meningioma is a common primary intracranial tumor, accounting for approximately 35% of all cases [[44]1]. Currently, the incidence rate of meningioma shows an upward trend year by year and increases with age [[45]2]. Low-grade meningioma (benign) is classified as WHO (World Health Organization) grade l, accounting for approximately 80.1%. High-grade meningiomas (malignant) are classified as WHO grade II and III, accounting for 18.3% and 1.5%, respectively [[46]3]. At present, the main treatment strategy for meningioma is maximum surgical resection combined with radiotherapy. Most patients with benign meningiomas can be cured due to limited tumor growth. However, high-grade meningiomas growing in an infiltrative manner are highly invasive, and patients with brain invasion and extracranial metastasis have much worse prognosis [[47]4, [48]5]. When meningiomas proliferate in a highly infiltrative manner, the demarcation between the tumor tissue and the surrounding normal brain tissue tends to blur, making it more challenging to distinguish the two. Moreover, the presence of residual tumor cells at the surgical margin is a primary cause of recurrence [[49]6, [50]7]. Thus, accurately characterizing and differentiating the brain-tumor interface (BTI) in meningioma holds immense significance. Previous solid tumor-related studies have shown that the tumor core and tumor margin possessed unique microenvironments [[51]8], such as different tumor cell characteristics, stromal cell features and immune microenvironments. For example, in head and neck squamous cell carcinoma, it was reported that the tumor leading edge might mediate tumor invasion and metastasis [[52]9, [53]10]. In colorectal cancer, patients with high microsatellite instability often exhibit a disorganized tumor-stroma interface accompanied by high-level immune cell infiltration. This condition was also associated with the efficacy of immune checkpoint blockade (ICB) therapy [[54]11]. In gastric cancer, it has been found that the tumor-stroma interface (TSI) was of a unique microenvironment, which was rich in pro-angiogenic endothelial cells and pro-metastasis macrophages [[55]12]. In liver cancer, regions between the tumor and adjacent non-tumor tissues were characterized by a hypoxic microenvironment, an exacerbated inflammatory response, and pronounced local immunosuppression [[56]13]. In brain solid tumors, the BTI also has specific gene expression signatures and a complex microenvironment. For instance, in gliomas, a distinct subset of microglial cells, characterized by high-level expression of chemotactic and pro-inflammatory genes, has been identified at the TSI [[57]14]. In the BTI of medulloblastomas, an increase in endothelial VCAM1 (vascular cell adhesion molecule 1) expression and micro-vessel density was found, along with a significant upregulation of tumor proliferation and enrichment of cancer cells with high stemness [[58]15]. However, little work has been done on the BTI of meningioma. Existing study on BTI in meningioma only focused on the MRI (Magnetic Resonance Imaging) features, which showed that extensive peritumoral brain edema (PTBE) could predict brain invasion of meningioma [[59]16]. The scarcity of research on the BTI of meningioma is likely linked to the formidable challenges in precisely obtaining samples that accurately capture the tumor boundaries. In this study, we integrated scRNA-seq (single-cell RNA sequencing) datasets of meningioma-related samples, encompassing the BTI, from multiple sources. By leveraging high-resolution single-cell transcriptomics, we comprehensively characterized the unique pathological ecosystem in the invasive region of the meningioma BTI. Our analysis offers valuable biological insights into elucidating the invasion mechanism, which is of great significance for identifying potential therapeutic targets for meningioma. Materials and methods Human specimen datasets The analyzed data in this study consists of 3 datasets, which were downloaded from the NCBI GEO (Gene Expression Omnibus) database ([60]GSE206647 and [61]GSE183655) and NCBI SRA (Sequence Read Archive) database (PRJNA826269). Ultimately, we integrated scRNA-seq data from 33 samples of 23 patients. These 33 samples included 11 normal meninges (N), 2 brain-tumor interfaces (BTI) and 20 tumor cores (TC), of which 10 were low-grade and 10 were high-grade meningioma. The clinical information is summarized in Supplementary Table 1. Single-cell sequencing data processing scRNA-seq FASTQ files were processed with human reference genome GRCh38-3.0.0 by the 10X Genomics Cell Ranger software (v6.0.0) using default parameter settings. R package Seurat (v4.3.0) was used for downstream processing. According to the specific characteristics of each scRNA-seq dataset, the data quality control criteria were adjusted accordingly, such that > 90% high-quality single-cells were retained for downstream analysis. Expression matrices of the 33 samples were merged into a single data object. We used harmony (v1.0.3; lambda = 1, theta = 2) to merge the datasets and remove batch effects. After performing principal component analysis and clustering (FindClusters), UMAP (Uniform Manifold Approximation and Projection) visualization was performed in Seurat with the first 20 principal components based on the top 2000 highly variable genes. For the combined dataset, the clustering resolution was set to 0.7. Marker genes for each defined cluster were identified using the FindAllMarkers function in Seurat. For each cluster, the genes expressed in more than 25% of cells with at least a 0.25 log2(fold-change) were considered. In order to ensure the repeatability of the cell populations between samples in three datasets, we removed the cell populations with strong sample bias. The initial combined dataset contained 195,452 cells. Cell populations with more than 60% of cells came from only a single sample, or with fewer than 300 cells were removed. Finally, twenty cell clusters were annotated as 9 distinct major cell types according to the top 20 marker genes of each initial cluster. For the myeloid immune cells and stromal cells, the clustering resolution was set to 0.2 to delineate the subpopulations. Cell numbers of each cluster and cell type per sample are summarized in Supplementary Table 2. Differentially expressed genes of the clusters and cell types are summarized in Supplementary Table 3. Copy number variation (CNV) analysis InferCNV (v1.8.1) was used to analyze the scRNA-seq data to identify evidence of large-scale chromosomal copy number changes in the somatic cells, such as the increase or loss of entire chromosome(s) or large segments of chromosome. InferCNV inferred chromosomal variation by exploring the intensity of gene expression at different locations in the tumorous cell genome compared to a set of normal cells as reference. We selected T cells, myeloid cells, endothelial cells, and pericytes as normal references.