Abstract Background Chemoresistance and recurrence following treatment are the greatest impediments to the prognosis of glioblastoma (GBM). Increasing evidence indicates that cancer-associated fibroblasts (CAFs) play a significant role in the progression of glioblastoma. Nevertheless, the role and source of CAFs in recurrent and chemotherapy-resistant GBMs still remain ambiguous. Methods Spatial transcriptome (ST) sequencing was conducted on the tissue microarray encompassing primary and recurrent glioma samples in order to characterize the cellular composition. Subsequently, the infiltration of CAFs in our formerly established in vivo temozolomide (TMZ)-resistant model was inspected through immunohistochemical staining. Additionally, we carried out RNA-seq and label-free quantitation (LFQ) proteomics on HCMECs co-cultured with TMZ-sensitive (TMZ-S) or TMZ-resistant (TMZ-R) cells to explore the mechanism. Results This investigation revealed that CAFs and astrocytes are enriched in recurrent GBM, and this phenotype is associated with the expression of extracellular matrix (ECM) proteins associated with COL1A1 and FN1 deposition. Further investigations revealed that tenascin-C (TNC) and filamin C (FLNC), which potentially mediate endothelial-to-mesenchymal transition (EndMT), are the predominant factors that induce the deposition of ECM proteins in the resistance-promoting microenvironment. Additionally, the natural product punicalin (PNC) was found to downregulate EndMT-related proteins, multidrug resistance-associated membrane proteins, and collagen-related proteins by targeting TNC and FLNC, thereby increasing the susceptibility of temozolomide (TMZ)-resistant cells to chemotherapeutic agents both in vitro and in vivo. Conclusion These discoveries indicate that TNC and FLNC induced EndMT was a key resource of CAFs and targeting TNC and FLNC to inhibit EndMT and the collagen pathway is a promising tactic for reversing drug resistance in tumours. The development of combined chemotherapeutic strategies based on the features of tumour microenvironment endothelial cells and ECM deposition has high potential clinical value in increasing the efficacy of tumour treatment. Supplementary Information The online version contains supplementary material available at 10.1186/s12967-025-06743-5. Keywords: Glioblastoma, TMZ resistance, EndMT, Tenascin-C, Filamin C Introduction Glioblastoma (GBM), the most prevalent type of brain tumour, is highly malignant, aggressive and prone to recurrence [[46]1]. The standard therapy for GBM typically involves surgical intervention, accompanied by chemotherapy, radiotherapy or targeted therapy [[47]2]; However, treatment outcomes remain limited, as most patients develop drug resistance within a short period, resulting in disease progression and recurrence. The median survival time following recurrence was less than 1 year [[48]3]. Recent studies have emphasized the critical role of interactions between the tumour microenvironment (TME) and tumour cells in promoting therapeutic resistance and recurrence in GBM [[49]4]. In the TME, immune cells and stromal cells, such as tumour-associated endothelial cells (TECs) and cancer-associated fibroblasts (CAFs), collaborate to remodel the extracellular matrix (ECM) by releasing paracrine chemokines and growth factors. This restructuring, along with the activation of various oncogenic signals in cancer cells, promotes tumour cell proliferation, invasion, immune evasion, and resistance to treatment [[50]5-[51]7]. TECs are considered crucial stromal cells within the TME and play a central role in tumour angiogenesis and blood circulation [[52]8]. It has been gradually revealed that TECs possess phenotypic plasticity, and the endothelial‒mesenchymal transition (EndMT) is a key event in tumour progression [[53]9, [54]10]. During EndMT, vascular endothelial cells lose endothelial traits and acquire mesenchymal phenotypes on account of external stimuli, including those that activate the transforming growth factor-β (TGF-β) and Notch pathway [[55]11]. EndMT is recognized as the primary origin of cancer-associated fibroblasts (CAFs) [[56]12-[57]14] and a significant contributor to glioblastoma (GBM) progression; EndMT exerts its effects act by promoting angiogenesis, chemoresistance, and invasiveness. Targeting EndMT represents a potential therapeutic approach to impede tumour progression and reduce the likelihood of recurrence [[58]9, [59]15]. In this research, we identified the characteristics of CAFs through spatial transcriptomics (ST) sequencing of a tissue microarray of primary and recurrent glioblastoma (GBM) tumours. Given that EndMT is an important source of CAFs in GBM [[60]16], we conducted coculture experiments with human cerebral microvascular endothelial cells (HCMECs) and temozolomide-sensitive (TMZ-S) or temozolomide-resistant (TMZ-R) GBM cells. Our results revealed a significant association between EndMT-derived CAFs and glioma cells exhibiting a mesenchymal phenotype. Overall, our study revealed the functional interplay among TECs and other cells within the GBM TME and indicates that these cells may contribute to chemoresistance and disease recurrence. Addressing this knowledge gap presents an opportunity for future research to develop targeted therapeutic strategies aimed at disrupting the fibrotic TME and improving clinical outcomes for GBM patients. Materials and methods Spatial transcriptome sequencing The primary and recurrent glioma samples for the tissue microarray were collected from Wuxi People’s Hospital and the Affiliated Hospital of Jiangnan University. The 6.5 × 6.5 mm capture area detected by the 10× Genomics CytAssist platform contained 12 microarray tissue samples, and we assessed two slides via this method. Spatial transcriptomics (ST) sequencing was performed, and then samples were subjected to hematoxylin and eosin (H&E) staining and tissue permeabilization. The mRNA released by the cells was captured by the primers on the spots and then amplified. After the quality of the amplified products was checked, the cDNA library was obtained by fragment screening. Space Ranger software ([61]https://support.10xgenomics.com/spatial-gene-expression/software/o verview/welcome) from 10× Genomics was used for data filtering, alignment, and quantification of the raw data (Transcriptome GRCh38) to generate gene expression matrices of the spots. Subsequently, Seurat software [[62]17] was employed for variance and visualization analyses. With the Seurat package, the SCTransform function was used for gene expression normalization, and then principal component analysis (PCA) was performed for dimensionality reduction. Differential gene expression analysis among spot populations was conducted via Seurat’s bimodal likelihood ratio statistical test, which revealed upregulated genes in various spot populations. The clusters were initially identified based on spatial transcriptomic profiles using unsupervised clustering and were subsequently annotated by comparing the top marker genes of each cluster with known cell-type–specific gene signatures from published glioma single-cell RNA-seq datasets. To construct the CAF and astrocyte gene sets, we performed differential expression analysis across spatial clusters using Seurat’s FindMarkers function with the Wilcoxon rank-sum test. The top 10 significantly upregulated genes within the fibroblast- and astrocyte-enriched clusters were selected as the marker gene sets for CAFs and astrocytes, respectively. Additionally, GO enrichment analysis was performed on the differentially expressed genes (DEGs). The enrichment results were visualized using a scatter plot generated with the R package ggplot2, where the Rich factor (defined as the ratio of the number of DEGs mapped to a GO term to the total number of genes annotated to that term) was used to represent the degree of enrichment. A higher Rich factor indicates a greater degree of GO term enrichment. GO terms with an adjusted p-value (FDR) < 0.05 were considered statistically significant. All the Spatial Transcriptome Sequencing Data Analysis was performed using the OmicStudio tools created by LC-BIO Co., Ltd (HangZhou, China) at [63]https://www.omicstudio.cn/spatial-infinity. Online clinical data analysis GBM data were obtained from the Chinese Glioma Genome Atlas (CGGA) ([64]http://www.cgga.org.cn/) and The Cancer Genome Atlas (TCGA) ([65]https://tcga-data.nci.nih.gov). databases. The single-sample gene set enrichment analysis (ssGSEA) scores of the KESHELAVA_MULTIPLE_DRUG_RESISTANCE gene set and the astrocytes and CAFs gene sets were evaluated via the gene set variation analysis (GSVA) package as previously reported in a GBM study [[66]15]. These gene set scores were subsequently utilized for Kaplan‒Meier survival analysis and Pearson correlation analysis via GraphPad Prism 8.0 software. The level of CAF infiltration and the correlations between CSF2, TNC, FLNC, TP53I3, CMBL, DDB2, NFKB2, and extracellular matrix (ECM) proteins, including FN1, COL1A1, and COL1A2, were analyzed via the TIMER 2.0 online database ([67]http://timer.cistrome.org/). Immunohistochemistry (IHC) Tissue microarray chips containing primary and recurrent glioma tissues were obtained from the affiliated hospital. The xenograft tumour tissues were collected, embedded and sectioned for subsequent staining. IHC staining was conducted as described previously [[68]18]. Primary antibodies against α-SMA (1:200) and COL1A1 (1:200) were used. Cell culture TMZ-S and TMZ-R cells were derived from a previously established in vivo model of TMZ-resistant glioblastoma multiforme (GBM) xenografts [[69]19] and maintained in DMEM (HyClone) supplemented with 10% foetal bovine serum (Biological Industries) and 1% penicillin‒streptomycin solution (P/S; HyClone). The human brain microvascular endothelial cell line (HCMEC/d3) was procured from iCellbioscience Company (iCell-h070) and cultured in specific medium containing endothelial cell culture additives, 5% FBS, and 1% P/S. Double immunofluorescence staining HCMECs were seeded in 24-well plates coated with coverslips and treated with TMZ-S/R-CM or 5 µg/ml TNC (MCE) or FLNC (Proteintech) recombinant protein for 48 h. The cells were subsequently fixed with 4% paraformaldehyde (PFA) and processed following our established protocol [[70]19]. The primary antibodies utilized are described in Supplementary Table [71]S1, and staining was visualized using a laser scanning confocal microscope (Leica Microsystems GmbH, Mannheim, Germany). RNA sequencing, data processing and analysis HCMECs were cocultured with TMZ-S or TMZ-R cells for 48 h. Subsequently, the tumour cells in the upper chamber were removed, and the medium was replaced with serum-free medium. After an additional 48 h of culture, the HCMECs and supernatants were collected for RNA sequencing and label-free proteomics analysis, respectively (Hangzhou Lc-Bio Technologies in China). Total RNA was isolated and purified using TRIzol (Thermo Fisher) following the manufacturer’s protocol. The quantity and purity of the RNA were assessed via a NanoDrop ND-1000 (NanoDrop, Wilmington, DE, USA), whereas the RNA integrity was evaluated via a Bioanalyzer 2100 (Agilent, CA, USA). PolyA-containing mRNA was then captured, fragmented, and used for cDNA synthesis to construct strand-specific libraries. These libraries were sequenced from both ends following standard procedures. The obtained clean data were aligned to the GRCh38 reference genome (Ensembl release 109) using HISAT2 (version 2.2.1). Based on the alignment results, transcripts were reconstructed and gene expression levels were quantified using StringTie (version 2.1.6). Transcript biotypes were annotated according to Ensembl classification. Differential gene expression analysis was conducted using the DESeq2 package in R, applying a threshold of|log2-fold change (logFC)| > 1 and an adjusted P value (Padj) < 0.05. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed using the Hiplot (ORG) platform ([72]https://hiplot.cn/). Visualization of upregulated and downregulated DEGs in TMZ-R-cocultured HCMECs was conducted using the ggplot2 package to generate volcano plots. Label-free proteomics, data processing and analysis The above FBS-free culture medium was collected for quantitative detection through nano-LC‒MS/MS following protein extraction and digestion. Analysis of the data-dependent acquisition (DDA) label-free MS/MS data was conducted via MaxQuant software [[73]20]. Statistical analysis was carried out via R (version 4.0.0). The raw protein intensity was normalized by method “medium”, and hierarchical clustering was performed via the pheatmap package. PCA was conducted with the metaX package. Statistical analysis of differences was performed via a t test, with a significance threshold of a p value ≤ 0.05 and a fold change ≥ 1.2 utilized to identify significantly differentially secreted proteins (DSPs). Cell viability assay HCMECs were seeded in 96-well plates and exposed to gradient concentrations of drugs for 48 h. Cell viability post-treatment was assessed via Cell Counting Kit-8 (CCK-8; Bimake, Houston, TX, USA) assays following the manufacturer’s protocol. The optical density at 450 nm was measured with a microplate reader (Thermo, Waltham, MA, USA). Each cell line was assessed in triplicate, and the half-maximal inhibitory concentration (IC50) value was determined using the following formula: (OD[treated]/OD[untreated]) × 100. Western blot analysis HCMECs were directly cocultured with TMZ-S/R cells or treated with the culture medium (CM) of TMZ-S/R cells for 48 h. In drug treatment experiments, the indicated drug was used at a concentration of one-tenth of the IC50 to avoid cell death for 48 h. Subsequently, cell lysates were prepared for Western blot analysis using RIPA lysis buffer (Cell Signaling Technology), with the antibody specifications detailed in Supplementary Table [74]S1. Enzyme-linked immunosorbent assay (ELISA) HCMECs were added to a 6-well plate and treated with ETO (2 µM), TMZ-S-CM or TMZ-R-CM for 48 h; then, the medium was changed to FBS-free culture medium, and the cells were cultured for 48 h. The concentrations of COL1A1, FN1, TNC and FLNC in the supernatant were determined via ELISA according to the manufacturer’s instructions (TNC/FLNC/COL1A1; Cloud-Clone Corp., USA; FN1; Proteintech, China). Molecular docking analysis The full sequences of FLNC (UniProt ID: [75]Q14315) and TNC (UniProt ID: [76]P24821) were obtained from the UniProt database and imported into Discovery Studio software for homologous sequence alignment and calculation of homology similarity. AlphaFold3 software was employed for modelling and predicting the structures of both proteins, and Rosetta Relax software was utilized for energy optimization. Subsequently, Sitemap software was employed to predict the active sites, and the proteins were aligned for comparative analysis. Protein structure preparation, which included water removal and hydrogenation, was conducted via PyMOL. The TCMSP Chinese Medicine Active Small Molecule Database ([77]https://www.tcmsp-e.com/#/database) was selected for the small molecule database. Autodock Vina was employed for the docking process, with the docking box and grid parameters configured to cover the active cavity comprehensively to identify the optimal binding site and conformation of small molecules within the active cavity. The docking box parameters were set as follows: box parameters of the FLNC protein: center_x = 1.596, center_y = -16.151, center_z = 44.464, size_x = 30, size_y = 34, size_z = 32, spacing = 0.581; box parameters of the TNC protein: center_x = -3.673, center_y = 36.207, center_z = -20.521, size_x = 30, size_y = 26, size_z = 20, spacing = 0.631. After docking, the predicted small molecules were ranked by Vina based on the empirical force field prediction of binding energy determined via Python, and the relative positions and relative interaction forces between the small molecules and proteins were analysed and plotted. The ligand‒receptor interaction forces were analysed via PLIP software, and the 3D and 2D conformations of the ligand‒receptor interactions were displayed via PyMOL software. Animal studies Six-week-old male BALB/c nude mice were purchased from Changzhou Cavens Experimental Animal Co., Ltd. (Jiangsu, China). A GBM xenograft model was established via the subcutaneous implantation of 5 × 10^6 TMZ-R cells into the right flanks of each mouse. When the tumour volume reached 100 mm^3, the mice were randomly divided into groups and orally administered TMZ (20 mg/kg) or PNC (40 mg/kg) every other day for 10 d. This dosing regimen was approximated based on the IC50 estimates, and no signs of toxicity were observed in our animal models at this dose. Tumours were harvested on the 11th day after the initial treatment, and tumour weights were recorded. All animal procedures were conducted in compliance with the Institutional Animal Care and Use Committee (IACUC) guidelines of Nanjing Medical University and in accordance with the Animal Welfare Act. Statistical analysis All experiments were repeated at least three times to ensure statistical validity. The data were analyzed via GraphPad Prism 8.0 software and are presented as the means ± standard deviations (SDs). The statistical significance of differences between two groups was determined by Student’s t test. A p value of less than 0.05 was considered to indicate statistical significance. Results Characteristics of CAFs and astrocytes in GBM patients identified via spatial transcriptomic sequencing (ST-seq) To characterize the cellular composition and spatial organization of primary and recurrent GBM comprehensively, we utilized ST-seq (10× Genomics CytAssist platform) on a GBM tissue microarray containing primary and recurrent tumours. Raw sequencing data were processed using Space Ranger (version 2.1) and analyzed following standard procedures as previously described [[78]21], each spot presented an average of 1648 unique molecular identifiers (UMIs) and 1468 uniquely expressed genes in the primary group and 2521 UMIs and 2307 uniquely expressed genes in the recurrent group (Fig. [79]1A and Fig. [80]S1A-B). To identify cell types and their respective functions within each cluster, we constructed t-SNE and U-MAP plots (Fig. [81]1B). We identified nine distinct clusters. For each cluster, top differentially expressed genes were identified using FindAllMarkers, and the corresponding marker gene sets were submitted to CellMarker 2.0 for automated annotation: CD1C^−CD141^− dendritic cells (marked with CHI3L1, SOD2), dendritic cells (marked with MBP, CKB, UCHC1, CST3, OLFM1, SNAP25, PEBP1, PTGDS and TUBB2B), astrocytes (marked with GFAP, CLU, CD74, APOE, S100B, SPARC, VIM and TMSB4X), fibroblasts (marked with COL3A1, COL1A1, COL1A2, FN1, COL6A3, DCN, MMP2 and IGFBP7), cancer cells (marked with GSX2, CHIC2, KIT, CACNG4, PDGFRA, PTPRZ1 and C1orf61), plasmacytoid dendritic cells (marked with VSTM2A, SEC61G, VOPP1, EGFR and LANCL2) and three clusters could not be reliably annotated and were labeled as ‘unidentified,’ indicating ambiguous or mixed lineage markers that do not clearly correspond to any well-defined cell type (Fig. [82]1B). Astrocytes, fibroblasts, and plasmacytoid dendritic cells were more prevalent in the recurrent GBM group than in the primary group (Fig. [83]1C). The unique microenvironment of GBM was initially considered fibroblast free, but increasing evidence suggests the presence of mesenchymal stromal cells expressing markers characteristic of CAFs, such as α-smooth muscle actin (α-SMA), PDGFRβ/CD140b, FAP, and FSP1, in GBM [[84]22-[85]25]. Here, we initially examined astrocytes (176) and fibroblasts (18), analysed cluster-specific DEGs and conducted GO pathway enrichment analysis to elucidate the functions of CAFs and astrocyte-enriched clusters (Fig. [86]1D, E and Fig. [87]S1C). In recurrent GBM (rGBM), Gene Ontology enrichment analysis of subcluster 4 (astrocyte-enriched) and subcluster 5 (CAF-enriched) revealed significant enrichment in terms related to the collagen-containing extracellular matrix (ECM), suggesting a potential shared role in ECM remodeling during tumor progression (Fig. [88]1E and Fig. [89]S1C), suggesting a close link between collagen deposition in the ECM and GBM recurrence. Feature plots depicting canonical markers for CAFs (COL3A1, COL1A1, COL1A2, and FN1) and astrocytes (GFAP, CLU, S100B, and CD74) are shown in Fig. [90]1F and Fig. [91]S1D. In contrast, GO enrichment analysis of the corresponding CAF- and astrocyte-enriched clusters in primary (treatment-naïve) GBM samples did not reveal enrichment of the collagen-containing ECM pathway (Fig. [92]S2), further supporting its potential involvement in GBM recurrence. Fig. 1. [93]Fig. 1 [94]Open in a new tab Identification of cancer-associated fibroblasts (CAFs) and astrocytes in primary and recurrent glioblastoma (GBM) via spatial transcriptomics (ST). A. nFeature_spatial plots provide a detailed visualization of the genes expressed in each individual spot within the capture area of combined primary and recurrent GBM, different color gradients or intensity scales were used to represent the relative abundance of specific gene transcripts in each spot. B. t-SNE and UMAP plots showing 9 clusters in each plot. These plots show nine distinct clusters, each represented by a different color, corresponding to different cell types or states within the tissue samples. C. Proportion of different cell types in primary and recurrent GBMs. D. Heatmap showing the expression of selected marker genes across the different cell clusters. E. Scatterplot showing the significant GO terms of the fibroblast cluster in the recurrent GBM (rGBM) group. F. The expression of specific marker genes within the fibroblast cluster was assessed to characterize the molecular profile of fibroblasts in the rGBM samples CAFs and astrocytes in GBM are associated with chemoresistance, recurrence and clinical prognosis Chemoresistance development is a significant contributor to tumour recurrence, metastasis, and cancer-related mortality [[95]26, [96]27]. Through GSEA of a of a combined 693 + 325 CGGA-GBM dataset and TCGA-GBM-HG U133A database, we determined the scores of the Keshelava_Multiple_Drug_Resistance (KMDR) gene sets, CAF gene set (top 10 marker genes of the CAF cluster in rGBM), and astrocyte gene set (top 10 marker genes of the astrocyte cluster in rGBM). The positive correlation observed among the ssGSEA enrichment scores for the KMDR, CAFs, and astrocytes gene sets suggests a potential association between CAFs/astrocytes and chemoresistance in rGBM (Fig. [97]2A-B). We further applied the same scoring method to treatment-naïve (primary) GBM samples. Interestingly, we observed weaker correlations between CAFs/astrocytes and chemoresistance, suggesting a potentially enhanced interaction or co-activation of these cell types in the recurrent setting (Fig. [98]S3). Higher CAF and astrocyte scores in patients, as depicted in Fig. [99]2C and D, were associated with poorer survival outcomes. Although both CAFs and astrocytes were enriched in recurrent GBM, our study primarily focused on CAFs due to their direct involvement in ECM remodeling and the endothelial-to-mesenchymal transition (EndMT) process, both of which contribute to temozolomide (TMZ) resistance in GBM [[100]15, [101]16]. The role of activated astrocytes will be explored in future investigations. Fig. 2. [102]Fig. 2 [103]Open in a new tab CAFs and astrocytes contribute to chemoresistance and recurrence in GBM. A-B. Correlation assay between the ssGSEA score of CAFs/astrocytes (ssGSEA score of the top 10 upregulated marker genes in the fibroblast or astrocyte cell cluster according to spatial transcriptomics in rGBM) and the Keshelava_Multiple_Drug_Resistance (KMDR) score in the indicated dataset. C‒D. Kaplan‒Meier survival (overall survival, OS) analysis of GBM patients from the CGGA and TCGA databases stratified according to the optimal cut-off value of the ssGSEA score of CAFs and astrocytes. E. Representative images of α-SMA and COL1A1 IHC staining of pGBM and rGBM samples (n = 6, **p < 0.01). Bars, 100 μm. F. Representative α-SMA staining of intracranial xenografts derived from TMZ-sensitive (TMZ-S) or TMZ-resistant (TMZ-R) cells (n = 6, ***p < 0.01). Bars, 100 μm α-SMA, a fibroblast marker, was further selected as an indicator of CAF abundance [[104]28]. IHC analysis of α-SMA-positive cells revealed increased CAF infiltration in rGBMs, and high expression of COL1A1 in rGBM was also observed (Fig. [105]2E). In our previously established in vivo TMZ resistance model [[106]15, [107]19], we observed a scarcity of α-SMA^+ cells in normal brain tissues, a rise in TMZ-sensitive tumour tissues (TMZ-S), and an excessive presence in the TMZ-resistant (TMZ-R) microenvironment (Fig. [108]2F). These findings collectively indicate that increased numbers of CAFs are indicative of unfavourable outcomes in GBM patients, suggesting that inhibiting CAF infiltration may be a potential strategy for overcoming chemoresistance. Transcriptomic and proteomic profiling reveals ECM remodeling and CAF enrichment in TMZ-resistant GBM Our previous study of TMZ resistance in vivo revealed that within the chemotherapy-resistant microenvironment of GBM, there was an aberrantly high density of blood vessels, and the number of α-SMA^+ cells adjoining tumour vascular endothelial cells increased, suggesting that the α-SMA^+ cells enriched around blood vessels might be EndMT-derived CAFs that contribute to GBM chemoresistance [[109]15]. To further investigate this mechanism, we performed RNA-seq and label-free quantitation (LFQ) proteomics on HCMECs that had been co-cultured with either TMZ-sensitive (TMZ-S) or TMZ-resistant (TMZ-R) GBM cells, analyzing both total cellular RNA and secreted proteins from the culture supernatant (Fig. [110]3A). In these analyses, we directly compared HCMECs co-cultured with TMZ-R cells to those co-cultured with TMZ-S cells to identify differentially expressed genes (DEGs) and differentially secreted proteins (DSPs). The volcano plot and differential ranking diagrams illustrate the distributions of 1331 DEGs and 567 DSPs derived from this comparison (Fig. [111]3B-C). GO and KEGG pathway analyses revealed that the upregulated 681 DEGs in the TMZ-R coculture group were enriched in cytokine-associated pathways, the NF-kappa B signaling pathway and the collagen-containing extracellular matrix pathway (Fig. [112]3D). The interactome analysis identified 7 factors that were upregulated both within and outside the cell (Fig. [113]3E). Next, we investigated the correlations among the 7 common factors and ECM proteins, including FN1, COL1A1 and COL1A2, as well as the CAFs infiltration level according to the TIMER 2.0 database (Fig. [114]3F). The most notable positive correlations were observed among TNC expression, FLNC expression and CAF infiltration, FN1 expression, COL1A1 expression, as shown in Fig. [115]3G-H. These findings suggest that the expression of TNC and FLNC is closely related to collagen-rich ECM deposition and CAF infiltration in GBMs. Fig. 3. [116]Fig. 3 [117]Open in a new tab Transcriptomic and Proteomic Profiling Reveals ECM Remodeling and CAF Enrichment in TMZ-Resistant GBM. A. The scheme shows the assay protocol of RNA-seq and label-free protein-seq.B. Volcano plot analysis showing differentially expressed genes (DEGs) in HCMECs cocultured with TMZ-R cells compared to TMZ-S cells. C. Rank diagram displaying the top 50 differentially secreted proteins (DSPs) in the culture medium of HCMECs cocultured with TMZ-R cells compared to TMZ-S cells. D. GO/KEGG pathway enrichment analysis of the up-regulated genes. E. Venn diagram identifying upregulated DEGs and DSPs by RNA-seq and label-free protein-seq respectively. F. Pearson correlation coefficients of the indicated proteins associated with CAF infiltration and FN1/COL1A1/COL1A2 expression according to the TIMER 2.0 database. G-H. Analysis of the associations of TNC/FLNC expression with CAF infiltration levels and COL1A1 and FN1 expression in GBM TNC and FLNC potentially promote EndMT and ECM formation in HCMECs To assess whether the EndMT phenotype in HCMECs is induced by treatment with culture medium (CM) from TMZ-R cells or TNC/FLNC, we conducted immunofluorescence (IF) staining for the mesenchymal marker α-SMA and the endothelial marker VE-cadherin. Our results demonstrated that both TMZ-R CM and the recombinant TNC/FLNC protein suppressed VE-cadherin signaling and promoted α-SMA signaling, leading to significant induction of EndMT (Fig. [118]4A-B). We subsequently evaluated the cytotoxic effects of commonly used anticancer agents, etoposide (ETO) and TMZ, on HCMECs. Our data revealed that ETO and TMZ had IC50 values of approximately 5 µM and 758 µM, respectively, in HCMECs (Fig. [119]4C). Furthermore, treatment with ETO or coculture with TMZ-S/TMZ-R cells resulted in increased expression of TNC, FLNC, and the EndMT inducer TGF-β, as confirmed by immunoblot analysis (Fig. [120]4D). Both TMZ-R coculture and TNC/FLNC recombinant protein (TF) treatment led to elevated levels of TGF-β and the ECM proteins FN1/COL1A1 (Fig. [121]4E). further experiments showed that treating HCMECs cultured with R cell-conditioned medium using either the TGF-β inhibitor SB431542 or the TNC/FLNC inhibitor PNC similarly suppressed TNC and FLNC expression (Fig. [122]S4A).Additionally, COL1A1, FN1, TNC, and FLNC were more abundant in the supernatant of HCMECs following stimulation with TMZ-R or ETO (Fig. [123]4F). To further assess whether prolonged TMZ treatment contributes to EndMT induction, we performed western blotting on tumor tissues derived from a TMZ-responsive orthotopic GBM mouse model. Tumors collected after two weeks of continuous TMZ administration, but prior to the development of overt resistance, displayed a mesenchymal shift, including increased α-SMA and decreased VE-cadherin expression (Fig. [124]S4B). These findings indicate that long-term TMZ exposure can initiate EndMT phenotypes in vivo, even in the absence of fully established resistance, reinforcing the role of the chemotherapy microenvironment in reprogramming endothelial cells. Fig. 4. [125]Fig. 4 [126]Open in a new tab TNC and FLNC potentially promote EndMT and ECM formation in HCMECs. A-B. Double immunofluorescence (IF) staining of HCMECs treated with culture medium (CM) from TMZ-R cells/TMZ-S cells or with vehicle/recombinant TNC and FLNC proteins using α-SMA (green)/VE-cadherin (red)antibodies. DAPI (blue) was used to label the nucleus. Bars, 200 μm. C. IC50 values and inhibitory curves of etoposide (ETO) and TMZ tested in HCMECs. D. Immunoblot (IB) was used to detect TNC, FLNC and TGF-β alterations in HCMECs treated with TMZ-S/R CM or vehicle/ETO (n = 3, *p < 0.05; **p < 0.01). E. IB was used to detect COL1A1, FN1 and TGF-β expression in HCMECs treated with TMZ-S/R CM or vehicle/TNC and FLNC (TF) recombinant protein (n = 3, *p < 0.05). F. ELISAs were used to detect COL1A1, FN1, TNC and FLNC in supernatants derived from the indicated cells (Blank/ETO, S/R treated) (n = 3, *p < 0.05; **p < 0.01; ***p < 0.001) Punicalin-targeted delivery of TNC and FLNC inhibits endothelial-to-mesenchymal transition (EndMT) and restores the chemosensitivity of Temozolomide (TMZ)-resistant tumors Considering the potential roles of TNC and FLNC during EndMT and ECM deposition, targeting of these proteins might overcome GBM resistance. To investigate this hypothesis, we performed a comparative analysis of FLNC and TNC using data retrieved from the UniProt database. Homologous sequence alignment conducted in Discovery Studio revealed a moderate degree of similarity, with an identity value of 14.7% (Fig. [127]S5). Structural models for both proteins were generated using AlphaFold3, followed by energy minimization via Rosetta Relax. Active site prediction with Sitemap, combined with structural alignment, revealed that both proteins possess binding cavities shaped by multi-β-sheet architectures (Fig. [128]5A). The active sites of the two proteins presented high homology and similarity, so they were used for subsequent virtual screening. The Traditional Chinese Medicine Systems Pharmacology (TCMSP) with information on active small molecules ([129]https://www.tcmsp-e.com/#/database) was selected for screening. AutoDock Vina was used to perform docking. The docking box and grid parameters were set in accordance with the mode of encompassing the active cavity, and the sites were fully covered to reveal the optimal binding site and the best conformation of small molecules within the active cavity. The docking results for the TOP3 target compounds for FLNC and TNC are presented in Fig. [130]5B, and the potential modes of action of MOL009272 (punicalin, PNC), MOL001479 (chelidimerine, CHE), and MOL007062 (neoprzewaquinone A, NA) with FLNC and TNC were also analysed (Fig. [131]5C and Fig. [132]S6). The IC50 profiles of the three natural small molecules in HCMECs were determined (Fig. [133]5D). We subsequently assessed their inhibitory potency via IB analysis and discovered that the conditioned medium from tumour cells and TMZ-R-treated cells augmented ECM protein expression and EndMT in HCMECs, as indicated by decreases in the levels of the endothelial markers CD31/VE-cadherin and increases in the levels of the mesenchymal marker α-SMA; notably, these effects could be repressed by CHE, PNC and NA. Among them, PNC had the greatest inhibitory effect on TNC, FLNC and EndMT (Fig. [134]5E). Marked suppression of the multidrug resistance-associated protein ABCB1 in the PNC-treated group was also observed, suggesting that PNC plays a role in drug uptake suppression (Fig. [135]5E). The in vivo studies revealed that the growth rates of TMZ-R tumours treated with TMZ or PNC alone were comparable. However, a significant inhibition of tumour growth was observed upon their combined administration (Fig. [136]5F-H). Consistent with these findings, preliminary in vitro experiments demonstrated that PNC treatment in TMZ-sensitive GBM cells did not produce a significant synergistic effect with TMZ (Fig. [137]S7A–B), suggesting that the therapeutic benefit of PNC is specific to the resistant context. Furthermore, inhibition of TNC and FLNC resulted in suppressed angiogenesis and reduced hypoxia within the tumor microenvironment (Fig. [138]S7C–D). These results indicate that targeted inhibition of TNC and FLNC markedly suppresses EndMT and ECM deposition, thereby restoring chemosensitivity both in vitro and in vivo. Fig. 5. [139]Fig. 5 [140]Open in a new tab PNC-targeted TNC and FLNC inhibits EndMT and increases the chemosensitivity of TMZ-resistant tumours. A. The active sites of FLNC and TNC proteins were predicted via Sitemap software. B. Binding analysis of the ligands against TNC and FLNC via AutoDock Tools. C. Three-dimensional (3D) and two-dimensional (2D) diagrams depicting the interaction of punicalin (PNC) with the amino acid residues of TNC and FLNC. D. IC50 values and inhibitory curves of chelidimerine (CHE), PNC and neoprzewaquinone A (NA) in HCMECs. E. IB assay showing the expression of ECM proteins (COL1A1, FN1, TNC and FLNC), EndMT markers (CD31, VE-cadherin and α-SMA) and MDR1 (ABCB1) in HCMECs treated with DMEM or CM from TMZ-S/R cells with or without the indicated drugs. F. Tumour growth assay of xenografts derived from the indicated TMZ-R cells that received the indicated treatment every two days after the tumours reached 150 mm^3 (n = 6, **p < 0.01). G-H. Representative images of xenografts collected on the final day and tumour weight quantification (n = 6, ***p < 0.001) Discussion Tumour-stromal cross-talk, which is mediated by factors secreted by CAFs, promotes chemoresistance in several forms of cancer [[141]29, [142]30]; nevertheless, the origin of CAFs in GBM and their impact on the chemosensitivity of GBM remain unclear. In this study, ST analysis, a coculture model with HCMECs and drug-resistant GBM cells, and transcriptome and proteome sequencing analyses were employed, and we demonstrated that the infiltration of CAFs is augmented in recurrent and drug-resistant tumours. EndMT, which is potentially induced by TNC and FLNC, constitutes an important source of CAFs, and the natural small-molecule compound PNC can clearly inhibit the EndMT phenotype and ECM protein expression by targeting TNC and FLNC and restore the chemosensitivity of GBM. Our work delineated the potential origin and role of CAFs in the TME of drug-resistant and recurrent GBM (Fig. [143]6). These investigations focused on therapeutic strategies that disrupt the fibrotic TME and improve the clinical outcomes of GBM patients. Fig. 6. [144]Fig. 6 [145]Open in a new tab Schematic overview of Punicalin in treating GBM via blockade ECM deposition and targeting EndMT. CAFs and astrocytes are aboundant in recurrent and chemoresistant GBM patients. FLNC and TNC induced EndMT was a potential source of CAFs in microenvironment of drug-resistant GBM. Punicalin can increase the chemosensitivity of TMZ-resistant tumours by targeted inhibition of TNC and FLNC-induced EndMT and ECM deposition CAFs encompass diverse subtypes with distinct functionalities, demonstrating notable heterogeneity. Their origin cells include quiescent fibroblasts, mesenchymal stem cells (MSCs), pericytes, adipocytes, endothelial cells, and epithelial cells [[146]31]. Such diversity in cell origin facilitates the formation of CAFs with variable phenotypes and functions within the TME. In GBM, CAF-like stromal cell cluster are situated within perivascular niches proximate to tumour-initiating glioma stem cells (GSCs) and consist of mesenchymal stromal cells expressing CAF markers [[147]22, [148]32]. For our investigation, we selected α-SMA as a marker for CAFs, and through cross-validation of the ST-seq results with our formerly established in vivo TMZ resistance model, we discerned that α-SMA expression was markedly increased in the TMZ-resistant group, suggesting that CAFs might play a crucial role in GBM recurrence and TMZ resistance. Research has shown that CAFs can synthesize and deposit ECM components to remodel the ECM within the TME, which affects the mechanical attributes of the ECM and thereby influences the interactions between tumour and immune cells [[149]33]. Proteomic analyses suggest that CAFs in GBM may promote tumour migration and invasion by secreting fibronectin-1 (FN1) [[150]34]. With respect to resistance, CAFs can provide a survival niche for cancer stem cells by secreting inflammatory factors, thereby facilitating tumour formation and chemotherapy resistance [[151]30]. Hence, targeted therapeutics aimed at CAFs, especially those that inhibit CAF generation, have the potential to improve tumour-targeted treatment strategies. In terms of the ST-seq results, we likewise discovered that astrocytes and plasmacytoid dendritic cells (pDCs) abound within the recurrent GBM TME. Astrocytes, which are crucial constituents of the GBM microenvironment, can transmute into tumour-associated reactive astrocytes (TARAs), which secrete factors such as TGF-β and IL-6, engendering an immunosuppressive milieu that facilitates GBM growth and invasion [[152]35]. Astrocytes might also foster GBM resistance by activating the Notch signalling pathway [[153]36, [154]37]. Astrocytes can also transfer mitochondria to GBM cells via tumour microtubules, increasing ATP production rates to promote tumour cell growth or restoration, ultimately resulting in chemotherapy resistance [[155]38, [156]39]. pDCs, which are characterized by the generation of copious amounts of type I interferon (IFN-I/α), contribute to the erection of an immunosuppressive TME [[157]40]. Nevertheless, the role of pDCs in GBM drug resistance remains unreported. In this study, we found that both the signature genes of CAFs/astrocytes in recurrent gliomas and the upregulated genes in HCMECs cocultured with resistant cells were associated with collagen-containing ECM formation. We cocultured HCMECs with supernatants from TMZ-sensitive and TMZ-resistant cells and subjected them to RNA-seq to identify the genes most closely associated with CAFs, namely, TNC and FLNC. In vitro assays revealed that TNC and FLNC can prompt HCMECs to undergo EndMT, promoting their transition into CAF-like cells, which in turn modify the attributes of the ECM and promote chemotherapy resistance. TNC has been recognized as a marker for the “angiogenic switch”, a phenomenon whereby angiogenesis converts quiescent blood vessels into new blood vessels [[158]41-[159]43]. Previous investigations have demonstrated that TNC induces EMT in tumour cells via the TGF-β signalling pathway [[160]44]. TNC secreted from stromal cells can facilitate evasion of immune surveillance, suggesting that the TNC-rich stroma may undergo remodelling during tumour progression [[161]45]. Research has shown that TNC is abundantly expressed in the ECM of GBM, particularly in the fibrous stroma and at the tumour margins, whereas the absence of TNC in mesenchymal GBM inhibits the dedifferentiation of transformed astrocytes and impedes the differentiation of glioma stem cells into tumour-derived endothelial cells. Hence, TNC may play a significant role in the autocrine regulation of plasticity in glioma cells and the ECM [[162]46, [163]47]. FLNC plays a crucial role in bridging the cytoskeleton and ECM, acting as an actin-crosslinking cytoskeletal protein that aids in regulating cell morphology [[164]48]. The high expression of FLNC in GBM, head and neck squamous cell carcinoma, and oesophageal squamous cell carcinoma is strongly correlated with increased tumour invasiveness, suggesting that FLNC might play a significant role in ECM remodelling [[165]49-[166]51]. Clinical pathological investigations have revealed that elevated FLNC expression is associated with microvascular invasion and unfavourable tumour prognosis [[167]52]. Our research reveals that the remodelling of ECs by the TME via the TNC and FLNC proteins results in EndMT in tumour endothelial cells, which alters the mechanical attributes of the tumour ECM and augments angiogenesis while influencing drug permeability, thereby promoting the progression of GBM resistance in both aspects. Hence, TNC and FLNC are promising targets for attenuating GBM resistance. Despite the promising results, this study has several limitations that should be acknowledged. First, although spatial transcriptomics and co-culture models provided valuable insights into the cellular composition and mechanisms of CAF induction in recurrent and chemoresistant GBM, they may not fully capture the dynamic heterogeneity and complexity of the tumor microenvironment in human patients. Second, while the efficacy of punicalin in targeting TNC- and FLNC-mediated EndMT and ECM remodeling was demonstrated in both in vitro and in vivo models, its pharmacokinetic properties, potential off-target effects, and safety profile require further evaluation prior to clinical translation. Third, although our results suggest that TNC and FLNC are potential regulators of EndMT and ECM protein deposition, it is important to acknowledge a key limitation. Despite employing an integrated RNA-seq and proteomics approach to enhance confidence in our selected targets, transcriptomic changes do not always correlate directly with protein synthesis and secretion. This discrepancy arises from the complex layers of post-transcriptional, translational, and post-translational regulation—an inherent limitation of such analyses. And additional mechanistic studies using genetic loss-of-function or in vivo lineage tracing models would provide stronger causal evidence. Future research should also explore the interactions between endothelial cells and other stromal or immune components that may influence chemoresistance. Lastly, A key question in studying chemoresistance is whether it mainly results from impaired drug delivery or tumor-intrinsic mechanisms, such as genetic mutations or enhanced drug efflux. Our findings highlight the role of the tumor microenvironment—particularly EndMT-derived CAFs and excessive ECM deposition—in creating physical and biochemical barriers to drug penetration. However, we did not directly assess TMZ permeability or investigate specific resistance-related mutations, such as MGMT promoter methylation or mismatch repair deficiency. Future studies using intratumoral drug quantification, blood-brain barrier (BBB) permeability assays, and whole-exome sequencing will be important to determine whether PNC reverses resistance by enhancing drug delivery or by modifying tumor-intrinsic responses. These analyses will provide a deeper understanding of the interplay between stromal resistance and intrinsic tumor mechanisms in GBM recurrence. Conclusion In conclusion, the present study underscores the role of TNC and FLNC in promoting endothelial-to-mesenchymal transition (EndMT), which contributes to the CAF population in recurrent and chemoresistant glioblastomas. Additionally, punicalin-mediated inhibition of TNC/FLNC-driven collagen-rich extracellular matrix formation emerges as a potential strategy to enhance chemotherapeutic responsiveness. Electronic supplementary material Below is the link to the electronic supplementary material. [168]Supplementary Material 1^ (15.3MB, zip) Author contributions Mei Wang: Writing– original draft, review & editing, Project administration, Funding acquisition, Formal analysis, Data curation, Supervision, Conceptualization. Jian Zou: Writing– review & editing, Funding acquisition. Koukou Li: Writing– review & editing, Conceptualization. Li Ji: Writing– review & editing, Methodology. Die Xia: Methodology, Investigation, Formal analysis. Yu Zhou: Investigation, Formal analysis. Yaling Hu: Methodology, Formal analysis, Data curation. Zhenkun Yang: Investigation, Formal analysis. Ying Yin: Investigation, Formal analysis. Jingjing Wang: Investigation, Formal analysis. Bo Zhang: Investigation, Formal analysis. Lingli Gong: Investigation, Formal analysis. Funding This work was supported by Natural Science Foundation of China (NFSC) grants (no. 82403426, 82372891, 82303448, 82172954, 82003581 and 82203757), The Natural Science Foundation of Jiangsu Province (Grants No BK20230187, BK20220225), “333” Engineering Project Jiangsu Province ((2022) 2–060), Taihu Talent Plan (JZ), Wuxi Medical Innovation Team (CXTD2021006), Top Talent Support Program for young and middle-aged people of Wuxi Health Committee (HB2023018), General Program of Jiangsu Commission of Health (M2020012), NSFC Cultivation Project of Wuxi Medical Center, Nanjing Medical University (WXMCPY202407 and WXKY202309001). We thank Clarity Manuscript Consultants for assistance with language editing ([169]www.aje.cn). Declarations Ethical approval The study was approved by the Institutional Ethics Committee of The Affiliated Wuxi People’s Hospital of Nanjing Medical University. Consent for publication All the authors read the manuscript and agreed to its publication. Competing interests The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Li Ji, Die Xia and Yu Zhou contributed equally to this work. Contributor Information Koukou Li, Email: kkli@njmu.edu.cn. Jian Zou, Email: zoujan@njmu.edu.cn. Mei Wang, Email: wangmei22@njmu.edu.cn. References