Small and medium/large glioblastoma extracellular vesicles differentially regulate endothelial and mitochondrial function in blood–brain barrier and peripheral vasculature models

Abstract

Purpose

Extracellular vesicles (EVs) released by glioblastoma (GBM) cells circulate systemically and modulate tumour-host interactions, influencing vascular function both within and beyond the tumour microenvironment. The EVs span a size range from tens to hundreds of nanometres; however, the specific effects of distinct GBM-derived EV size subpopulations on endothelial function across different endothelial cell types remain poorly understood.

Methods

In this study, we investigated the effects of enriched small EVs (sEV) and medium/large EVs (m/lEV) fractions derived from human GBM cell line HROG36 on endothelial barrier integrity (assessed by transendothelial electrical resistance (TEER) and tight junction (TJ) protein analysis), migration (wound healing assay), angiogenic capacity (tube formation on extracellular matrix), and mitochondrial function (assessed by measuring the oxygen consumption rate (OCR)) in human brain microvascular endothelial cells (hCMEC/D3), representing blood–brain barrier (BBB) model, and peripheral endothelial cells (HUVECs).

Results

We show that sEV-enriched fractions significantly impaired endothelial barrier integrity in both cell types, as evidenced by reduced TEER and disrupted TJ protein organisation. In contrast, m/lEV-enriched fractions enhanced endothelial cell migration while suppressing angiogenic network formation, resulting in disorganised, shortened capillary-like structures. These pro-migratory and anti-angiogenic effects of m/lEVs were accompanied by a selective suppression of mitochondrial respiration in hCMEC/D3 cells, with no such effect observed in HUVECs, indicating that the functional alterations are linked to altered bioenergetic activity specifically in BBB endothelial cells.

Conclusion

Our data indicate that sEV and m/lEV-enriched subpopulations derived from GBM exert distinct vascular effects, underscoring their potential as functionally distinct EV-based biomarkers and informing the development of personalised therapeutic strategies in GBM.

Graphical Abstract

(created in BioRender. Kulakauskienė, D. (2026) https://BioRender.com/e8jyszx)

Introduction

Cancer cells secrete a complex secretome that readily enters the circulation and induces systemic alterations. Within this secretome, EVs are increasingly recognised as major mediators of long-range intercellular communication, accounting for many of the biological effects observed at sites distant from the tumour. EVs are lipid bilayer-enclosed particles released into the extracellular space and biological fluids [1, 2]. They transport complex molecular cargo, including proteins, lipids, and nucleic acids, and thereby enable cells to influence the behaviour of distant targets [3–5]. In cancer, EVs are now viewed as active signalling entities that reshape tissue environments and coordinate pathological processes across organ systems, not only passive by-products of cellular activity [6].

A major challenge in EV research is biological and technical heterogeneity. Cells generate EVs through multiple intracellular pathways [7], but their biogenetic origin cannot be reliably inferred after isolation. Overlapping size distributions, shared molecular markers, and isolation-related artefacts further limit the utility of strictly biogenesis-based classification. To address these limitations, the International Society for Extracellular Vesicles (ISEV) has proposed an operational framework that prioritises physical characteristics and transparent reporting over inferred origin [2]. Within this framework, sEV- and m/lEV-enriched fractions are operationally defined groups that allow investigation of size-associated functional effects. Distinguishing EV-enriched fractions is therefore biologically meaningful. Molecular profiling studies further indicate that tumour cells selectively package bioactive cargo into distinct EV subpopulations. Cancer-derived larger EV fractions are enriched in oncogenic proteins, whereas sEV display distinct miRNA and mRNA profiles [6]. Collectively, these findings suggest that EV subpopulations function as parallel signalling systems with potentially divergent biological roles, and that bulk EV analysis may obscure subpopulation-specific mechanisms.

The vascular system is a particularly sensitive target of EV-mediated signalling in cancer. Tumour-associated vascular abnormalities occur both locally and systemically, including increased permeability and dysregulated angiogenesis, suggesting broader alterations in endothelial functional state rather than isolated defects. Growing evidence indicates that tumour-derived EVs contribute to these changes by modulating endothelial metabolic programmes [8]. Given their stability in circulation and ability to travel long distances, EVs are well-positioned to exert vascular effects beyond the tumour microenvironment.

GBM represents a particularly informative model for studying EV-driven vascular dysfunction. As one of the most highly vascularised human tumours, GBM is characterised by profound disruption of the BBB, resulting in a heterogeneous blood–tumour barrier that contributes substantially to disease-associated morbidity and mortality [9–11]. Importantly, GBM-derived EVs are detectable both within the tumour microenvironment and in the peripheral circulation [12], where they are associated with systemic vascular abnormalities, including pro-thrombotic states [13]. These features position GBM as a highly relevant model for investigating EV-mediated vascular effects across central and peripheral compartments.

Despite increasing recognition of the role of GBM-derived EVs in vascular pathology, how distinct EV-enriched subpopulations differentially affect endothelial function remains insufficiently characterised. Most studies rely on bulk EV preparations or focus predominantly on sEV-enriched isolates, limiting the ability to resolve subpopulation-specific effects on vascular behaviour. Furthermore, potential links between EV-induced endothelial dysfunction and endothelial metabolic state remain largely unexplored. To address these gaps, we performed a subpopulation-resolved analysis of GBM-derived EVs, distinguishing between sEV- and m/lEV-enriched preparations. We systematically assessed their effects on endothelial barrier properties, migration, angiogenesis, and metabolic status using both BBB and peripheral endothelial models under matched experimental conditions.

Materials and methods

Cell lines and culture conditions

In this study, we used the immortalised human GBM cell line HROG36 (obtained from CLS Cell Lines Service GmbH, Germany), human umbilical vein endothelial cells (HUVECs; obtained from Thermo Fisher Scientific, USA), and the immortalised human cerebral microvascular endothelial cell line hCMEC/D3 (obtained from Sigma-Aldrich). All commercial cells were maintained in plastic cultureware at 37 °C in a humidified 5% CO₂ incubator. The culture medium was replaced twice weekly. For subculturing, cells were rinsed once with phosphate-buffered saline (PBS), detached with a 0.05% trypsin-1 mM ethylenediaminetetraacetic acid (EDTA) solution (Gibco), resuspended in fresh medium, and then transferred to new culture flasks for continued growth.

For expansion before the EV collection, HROG36 cells were grown in Dulbecco’s Modified Eagle Medium/Nutrient Mixture F-12 (DMEM/F-12; Gibco), supplemented with 10% fetal bovine serum (FBS; Gibco) and 1% penicillin–streptomycin (Gibco). The cells used for EV collection were between passages 20 and 35.

All cultureware used for HUVECs was coated with 0.1% gelatin (Sigma-Aldrich, porcine skin-derived), diluted in distilled water, and incubated at 37 °C for at least 30 min. HUVECs were grown in endothelial cell growth medium (Gibco) supplemented with 1x Large Vessel Endothelial Supplement (Gibco) and 1% penicillin–streptomycin (Gibco). The cells were used for experiments between passages 3 and 6.

All cultureware for hCMEC/D3 cells was precoated with rat-tail collagen type I (Gibco) at a 1:20 dilution in PBS and incubated at 37 °C for at least 1 h. hCMEC/D3 were maintained in EndoGRO-MV complete medium (Merk Millipore) containing EndoGRO basal medium, 5% FBS, L-glutamine (10 mM), EndoGRO-LS supplement (0.2%), heparin sulphate (0.75 U/ml), ascorbic acid (50 µg/ml), hydrocortisone hemisuccinate (1 µg/ml), recombinant human epidermal growth factor (5 ng/ml) and were additionally supplemented with freshly added recombinant human fibroblast growth factor (bFGF; 1 ng/ml; Gibco). hCMEC/D3 cells used for the experiments were between passages 12 and 18.

Collection and isolation of EVs

HROG36 cells were seeded at 3000 cells/cm² and cultured for 4 days under standard conditions. Cells were then washed twice with room temperature PBS and subsequently cultured for an additional 3 days in medium containing EV-depleted FBS (Exolitus, Lithuania). Conditioned supernatants were processed at 4 °C by sequential centrifugation. Firstly, supernatants were centrifuged at 500 × g for 15 min to remove cellular debris, followed by 10,000 × g for 45 min. The 10,000 × g pellet was washed once in ice-cold PBS and recentrifuged at 10,000 × g for 45 min to obtain the m/lEV-enriched fractions. The remaining supernatant was passed through a 0.22 μm filter and subjected to ultracentrifugation at 167 000 × g for 2 h using a SORVALL WX Ultra Series ultracentrifuge equipped with a T-865 fixed-angle rotor (8 × 36 mL). Pellets were gently resuspended in ice-cold PBS and washed by a second ultracentrifugation at 167 000 × g for 2 h in the same tubes/rotor to obtain the sEV-enriched fractions. Final m/lEV and sEV pellets were resuspended in ice-cold PBS to 1,000× relative to the starting conditioned volume (e.g., EVs from 100 ml resuspended in 100 µl) and stored at -80 °C.

Experimental design

After purifying both EV subpopulations from HROG36, the samples were characterised and used to treat target cells. To evaluate biological effects, HUVECs were seeded at 100,000 cells/cm² and hCMEC/D3 at 60,000 cells/cm². After seeding, cells were cultured for 3 days to ensure the establishment of a fully confluent monolayer before treatment. During the experiments, the culture medium was refreshed and supplemented with the designated experimental groups: sEV, m/lEV, vehicle control (PBS), and positive control (VEGF/bFGF; only for the tube formation assay). EV treatments were standardised by applying equal volumes of each preparation rather than normalising to particle number or total protein content. This approach was intentionally selected to more closely recapitulate the physiological tumour microenvironment, in which cells are exposed to the total secretome produced by tumours of varying sizes and activity levels. By avoiding artificial normalisation, we preserved biologically relevant differences in EV yield, thereby modelling the naturally generated cumulative signalling output of tumours. For functional assays, 1000× concentrated m/lEV and sEV were applied at a dose of 10 µL/cm² of culture surface area. Three days post-treatment, functional changes in target cells were assessed.

Cryogenic transmission electron microscopy (cryo-TEM)

Cryo-TEM service for visualisation of sEV and m/lEV samples was acquired from the Life Science Centre at Vilnius University, Lithuania. Cu300 mesh R1.2/1.3 holey carbon grids (Quantifoil) were utilised for sample preparation. Before analysis, the grids were glow-discharged for 45 s at 25 mA with the GloQube Plus system (Quorum Technologies). An aliquot of 3 µL of aqueous EV solution was applied to the carbon side of the grid, blotted for 5.0 s, and then plunge-frozen in precooled liquid ethane using the Vitrobot Mark IV (Thermo Fisher Scientific). Embedding the EVs in a thin amorphous ice layer is crucial to preserving their native state and preventing radiation damage. Imaging was performed using a Thermo Fisher Scientific Glacios cryo-electron microscope at 200 kV, equipped with a Falcon 3EC Direct Electron Detector operating in electron-counting mode (Vilnius University, Lithuania). Images were acquired using EPU software (version 2.14.0) at a magnification of 92,000×, yielding a calibrated pixel size of 1.10 Å. The parameters were an exposure rate of 0.81 e/å²/s, a total dose of 29.7 e/å² per image, and a defocus of -2.0 μm.

Nanoparticle tracking analysis (NTA)

The size distribution and quantification of isolated EV samples were analysed using a NanoSight NS300 system (Malvern Panalytical, UK). All samples were diluted in filtered PBS to achieve an optimal particle count per frame (30–100 particles/frame). The camera level and detection threshold were optimised for each sample individually to ensure precise particle detection and tracking. Five 1-min videos were captured with an sCMOS camera and a 488 nm blue laser for each measurement. The videos were then analysed using the NanoSight Software NTA 3.4 Build 3.4.4, with a detection threshold of 5.

Tunable resistive pulse sensing (TRPS)

Additional particle-size and concentration analyses of EV samples were performed using the Exoid TRPS measurement system (Izon Science Ltd.) according to the manufacturer’s instructions. An NP100 (C03533) nanopore was used for sEV, and an NP300 (C02615) nanopore for m/lEV. Samples were diluted 1:100 prior to analysis. The system was calibrated using carboxylated polystyrene calibration particles (CPC200 for sEV and CPC400 for m/lEV; Izon Science Ltd.) to ensure accurate size and concentration measurements. All measurements were performed under consistent stretch and voltage conditions, and size and concentration were determined by comparing EV blockade signals with those obtained from the calibration particles. Two independent samples for sEV and m/lEV were measured at three different pressure settings. For data analysis, the size and concentration distributions from each measurement were exported with a 2 nm bin size, averaged, and the reported values represent the mean of these measurements.

EV marker characterisation

EV marker expression in sEV and m/lEV fractions was measured using a multiplex bead-based immunoassay and an enzyme-linked immunosorbent assay (ELISA) to compare marker distribution between EV subpopulations. Prior to analysis, EV samples were lysed with 0.04% Triton X-100 in PBS for 30 min at 2–4 °C. Multiplex analysis was performed using the Procartaplex Exosome Panel (Thermo Fisher Scientific; cat. no. EPX060-15845-901) according to the manufacturer’s instructions. Data were acquired on a Luminex 200 instrument (Luminex Corp, Austin, TX, USA), and mean fluorescence intensities were converted to concentrations (pg/mL) using a five-parameter logistic (5-PL) curve fit (xPONENT software) with background subtraction. Annexin A1 levels were measured using a human ELISA kit (Invitrogen; cat. no. EH31RB) according to the manufacturer’s instructions, with absorbance recorded at 450 nm on a Tecan Infinite 200 PRO plate reader (Tecan, Männedorf, Switzerland). All samples and standards were analysed in duplicate, and results from both assays were normalised to the EV particle number determined by NTA.

Immunocytochemistry

Endothelial cells were fixed in 4% paraformaldehyde for 20 min at room temperature to analyse TJ proteins, then permeabilised with 0.1% Triton X-100 for 15 min. After each step, samples were washed three times with PBS at room temperature. They were then blocked with 1% bovine serum albumin for 30 min at room temperature and incubated overnight at 4 °C with primary antibodies targeting Zonula occludens-1 (ZO-1; 1:1000; Cell Signaling Technology, cat. no. 5406) and Claudin-5 (1:1000; Cell Signaling Technology, cat. no. 66879). The following day, samples were washed thrice with PBS and incubated with Alexa Fluor-conjugated secondary antibodies (1:1000) for 1 h at room temperature in the dark. After a further three PBS washes, nuclei were stained with Hoechst H33342 (Bisbenzimide) for 5 min at room temperature. Imaging was performed using an automated Olympus ApexView 100 fluorescence microscope at 40× magnification.

EV labelling and internalisation assay

Purified EVs were labelled with SYTO™ Green Fluorescent Nucleic Acid Stains (Invitrogen) that bind to RNA within the EVs. The labelling was performed according to the manufacturer’s instructions, using a 10 µM concentration and a 30-min incubation at 37 °C in 5% CO₂. After incubation, fluorescence intensity was measured using a spectrofluorometer at excitation/emission wavelengths of 490/530 nm, respectively. To remove excess dye, sEV preparations were washed using ExoSpin columns, while m/lEV were centrifuged twice at 10,000 × g for 45 min at 4 °C. The fluorescence was reassessed after washing to confirm dye removal and stability. The labelled EVs were then applied to cells in 35-mm glass-bottom Petri dishes. After 24 h, cells were stained with CellMask™ Plasma Membrane Stain (Invitrogen) at 1×, prepared according to the manufacturer’s protocol, for 45 min in the dark, followed by two washes with warm culture medium. Nuclei were stained with Hoechst H33342 (1 µg/mL in PBS) for 5 min at room temperature. Finally, samples were imaged with a Carl Zeiss Axio Observer Z1 fluorescence microscope equipped with a 100× immersion objective, and the images were processed in ImageJ.

Assessment of cell migration

Cell migration was assessed using a scratch (wound-healing) assay. After 72 h of EV treatment, a sterile 200-µl tip was used to carefully scratch the cell monolayer in each well. Detached cells were rinsed twice with complete growth media. Fresh full media was added, and the cells were incubated under standard conditions. Images of the wound were taken at 0 and 9 h using a Leica DMi1 light microscope (Germany) at 4x magnification. The wound area was manually traced in the captured images using ImageJ software (version 1.53) to measure the cell-free region. The percentage wound closure was calculated as follows:

  

where W₀ represents the wound area at 0 h and W₉ at 9 h.

TEER measurements

To measure TEER, brain endothelial cells hCMEC/D3 were seeded at 6 × 10⁴ cells/cm² and human HUVEC at 1 × 105 cells/cm² on polyester transwell inserts with 0.4 μm pores (BRAND^®) placed in 24-well plates and coated with rat tail collagen I (1:20 in PBS, Gibco) for hCMEC/D3 and coated with gelatin I (0.1% in PBS) for HUVEC. hCMEC/D3 cells were cultured in EndoGRO medium until a confluent monolayer formed. HUVECs were cultured in endothelial cell growth medium supplemented with large vessel endothelial supplement under the same conditions. Once confluency was reached, the medium in the upper (luminal) compartment was replaced with fresh medium containing sEV or m/lEV. TEER was measured using the Millicell^® ERS-2 Electrical Resistance System (Merck Millipore). Each insert was measured three times at different positions. To calculate TEER (Ω·cm²), the mean resistance of cell-free inserts was subtracted from the mean readings of inserts containing endothelial cell monolayers and multiplied by the membrane surface area according to the following formula:

  

where R_cells represents the resistance of inserts with cells, R_blank the resistance of cell-free inserts, and A the effective membrane area. TEER values were expressed as percentages relative to the untreated control cells.

Endothelial cell tube formation assay

To evaluate angiogenesis in vitro, a 96-well plate was coated with 50 µL of Geltrex (Thermo Fisher) and incubated at 37 °C for at least 30 min to allow polymerisation. Following EV treatment, hCMEC/D3 cells were harvested by trypsinisation, resuspended in fresh growth medium, and seeded onto the Geltrex-coated plates at a density of 32,000 cells per well. After 24 h of incubation, tube formation was assessed using a Leica DMi1 light microscope (Germany). Images for quantitative analysis were acquired at 4× magnification. Quantitative analysis of tube formation was performed using ImageJ (version 1.53) with the Angiogenesis Analyser plugin, following the method described by Carpentier et al. [14]. Untreated hCMEC/D3 cells cultured in EndoGRO-MV complete medium served as the negative control, whereas treatment with 35 ng/mL recombinant human basic fibroblast growth factor (bFGF) was used as the positive control.

HUVECs were cultured under conditions identical to those described for hCMEC/D3 cells, except that wells were coated with GelNest™ extracellular matrix (NEST Scientific, Woodbridge, NJ, USA) instead of Geltrex. HUVECs were seeded at a density of 10,000 cells per well, and tube formation was analysed after 4 h of incubation. Images for quantitative analysis were acquired at 10× magnification using a Leica DMi1 light microscope (Leica Microsystems, Germany). Treatment with 25 ng/mL recombinant human vascular endothelial growth factor (VEGF) isoform 165 (PeproTech^®, USA) was used as the positive control.

Mitochondrial respiration

Mitochondrial function was assessed using the Seahorse XFp Cell Mito Stress Test (Agilent Technologies, Santa Clara, CA, USA) according to the manufacturer’s instructions. hCMEC/D3 cells were seeded at 6 × 10⁴ cells/cm² and HUVECs at 1 × 10⁵ cells/cm² in Seahorse XFp microplates coated with rat tail collagen I (1:20 in PBS, Gibco) for hCMEC/D3 or gelatin I (0.1% in PBS) for HUVECs and cultured until a confluent monolayer was formed, followed by treatment with EVs for 72 h. Prior to the assay, cells were incubated in Seahorse XF assay medium supplemented with 10 mM glucose, 2 mM glutamine, and 1 mM pyruvate at 37 °C in a non-CO₂ incubator for 1 h. OCR was measured under basal conditions and following sequential injection of oligomycin (1 µM), trifluoromethoxy carbonylcyanide phenylhydrazone (FCCP; 2 µM for hCMEC/D3 and 3.5 µM for HUVECs), and rotenone/antimycin A (0.5 µM each) to determine basal respiration, ATP production, proton leak, maximal respiration, spare capacity, and non-mitochondrial respiration.

hCMEC/D3 data were normalised to total protein content per well, determined using the Bradford reagent and measured with a multifunctional plate reader (Infinite 200 Pro Nano Plex, Tecan, Männedorf, Switzerland). HUVEC data were normalised to cell number based on nuclear staining. Briefly, cells were stained with Hoechst 33,342 (Sigma-Aldrich, St. Louis, MO, USA) for 10 min, and four independent images per well were acquired using a benchtop fluorescence microscope (Olympus APX 100-HCU, Tokyo, Japan) at 20× magnification. Nuclei were quantified automatically using ImageJ (version 1.53), and the average per well was used for normalisation.

Statistical analysis

Statistical analysis was performed on at least three independent biological experiments. Data are expressed as the mean and standard deviation (SD). The non-parametric Mann-Whitney U test was used to compare two groups. For comparisons involving three or more groups, data were analysed using a non-parametric Kruskal–Wallis test. All data were analysed using GraphPad Prism^® software version 10.4.2 (534) (GraphPad Software, Inc., USA).

Results

Characterisation and cellular internalisation of GBM-derived sEV and m/lEV in hCMEC/D3 and HUVEC cells

A combined centrifugation and ultracentrifugation approach was used to isolate EV-enriched fractions from the HROG36-conditioned medium. Centrifugation at 10,000 × g and 167,000 × g was applied to obtain m/lEV- and sEV-enriched fractions, respectively. These fractions were characterised based on morphology and size distribution using cryo-TEM, NTA, and TRPS. Cryo-TEM imaging confirmed the presence of round, membrane-bound vesicular structures typical of EVs (Fig. 1a). Both sEV and m/lEV subpopulations exhibited a well-defined lipid bilayer membrane, consistent with their biological origin. Representative micrographs reveal structures with characteristic diameters of approximately 75 nm for sEV and 300 nm for m/lEV (Fig. 1a). NTA revealed that the majority of sEV ranged from 50 to 150 nm, with a mode diameter of 107.10 ± 7.31 nm, whereas m/lEV ranged from 70 to 250 nm, with a mode diameter of 132.10 ± 6.78 nm (Fig. 1b–c). TRPS showed a comparable trend, with sEV primarily distributed within the 100–150 nm range, with a mode diameter of 113.00 ± 5.10 nm, whereas m/lEV were predominantly detected between 150 and 400 nm, with a mode diameter of 184.20 ± 14.54 nm (Fig. 1d–e). Although the size distributions partially overlapped, both methods demonstrated a consistent shift towards smaller particles in sEV and larger, more heterogeneous fractions in m/lEV preparations. Differences between NTA and TRPS measurements reflect methodological characteristics rather than biological discrepancies. NTA, which determines particle size based on Brownian motion and light scattering, is particularly sensitive to small, highly diffusive particles and therefore captured a relatively large proportion of EVs below 100 nm, albeit within a more limited upper detection range for larger EV subpopulations. In contrast, TRPS, which measures individual particles passing through a calibrated nanopore, detected a broader and more polydisperse population extending up to ~ 1 μm. This broader detection range was reflected in higher size heterogeneity in m/lEV samples, as indicated by a d90/d10 ratio of ~ 3.0–3.1, compared with ~ 1.6–1.8 for sEV. Methodological differences also influenced the measured particle concentration. Relative to TRPS, NTA reported higher particle concentrations by approximately 2.21-fold for sEV (2.22 × 10¹⁰ vs. 1.00 × 10¹⁰ particles/ml) and 3.97-fold for m/lEV (1.36 × 10¹⁰ vs. 3.43 × 10⁹ particles/ml; data not shown). Based on these concentration values, a 10 µl dose of each EV preparation used for functional assays corresponded to approximately 10⁸ sEV/cm² and 10⁷–10⁸ m/lEV/cm², depending on the analytical method used, with NTA consistently reporting higher particle numbers than TRPS. Importantly, these discrepancies most likely reflect technical differences in detection principles and sensitivity across size ranges rather than true biological variation between the EV preparations. Thus, NTA and TRPS should be considered complementary rather than conflicting approaches, each providing distinct but supportive information on EV size distribution and concentration. Accordingly, functional treatments were intentionally standardised by preparation volume rather than by particle count, thereby ensuring experimental consistency and preventing bias arising from variability in platform-dependent particle quantification. Additionally, by applying sEV- and m/lEV-enriched preparations isolated from the same initial conditioned medium volume, rather than adjusting treatments based on measured particle concentration, we more closely mimic physiological conditions in which EV subpopulations are naturally present in their relative proportions.

Fig. 1.

EV Characterisation and Internalisation. (a) Representative cryo-TEM images of sEV and m/lEV. (b) EV size distribution assessed by NTA, and (c) by TRPS. Representative particle size distribution profiles of sEV and m/lEV fractions determined by (b) NTA and (d) TRPS. Panels (c) and (e) show the mode of particle diameter calculated from NTA and TRPS measurements, respectively. Data represent mean ± SD from three independent preparations. Statistical significance was determined using a nonparametric Kruskal–Wallis test, with significance indicated as *p < 0.05 and **p < 0.01. (f) Expression of EV markers in sEV and m/lEV fractions was assessed by multiplex bead-based immunoassay (CD9, CD63, CD81, cytochrome c, Syntenin-1, VLA-4) and ELISA (Annexin A1), showing one biological replicate with two technical replicates per sample. (g) Representative fluorescence microscopy images showing internalisation of fluorescently labelled HROG36-derived sEV and m/lEV by hCMEC/D3 and (h) by HUVEC cells after treatment. EVs were labelled with SYTO RNASelect (green), whereas the cells were stained with CellMask (red), and nuclei were counterstained with Hoechst 33342 (blue). Merged images show vesicle uptake and cytoplasmic distribution after 24 h of treatment

Purified EV samples were characterised for marker expression using a multiplex bead-based immunoassay and ELISA. Classical exosomal markers CD9, CD63, CD81, Syntenin-1, and VLA-4 were predominantly detected in sEV samples, consistent with the presence of canonical exosomal proteins in this fraction. While Syntenin-1 was largely confined to sEV-enriched preparations, a lower level was also observed in m/lEVs, reflecting partial overlap in protein composition between the two EV populations. In contrast, cytochrome c was enriched in m/lEVs, reflecting the presence of mitochondrial components typically associated with larger EVs. Annexin A1, assessed by ELISA, was also predominantly present in m/lEV samples, supporting their identity as medium- and large-sized EVs that carry cytosolic and membrane-associated proteins. The distribution of cytochrome c and Annexin A1 in m/lEVs highlights the functional and compositional distinctions among EV populations (Fig. 1f).

A fluorescence-based assay was performed to confirm active EV internalisation by endothelial cells and to visualise their subcellular localisation. After 24 h of incubation, both sEV and m/lEV were internalised by hCMEC/D3 and HUVEC cells. As shown in Fig. 1g, hCMEC/D3 internalised fluorescently labelled sEVs, exhibiting a fine reticular cytoplasmic pattern. In contrast, m/lEV in hCMEC/D3 displayed a more diffuse cytoplasmic distribution, suggesting distinct internalisation or trafficking routes. In HUVECs, fluorescence imaging also revealed distinct internalisation patterns for sEV and m/lEV (Fig. 1h). Following sEV treatment, the fluorescent signal in HUVEC cells was diffusely distributed throughout the cytoplasm without distinct compartmental boundaries, indicating widespread cytosolic localisation. By comparison, m/lEV exposure resulted in fluorescence localised to distinct, vesicular structures reminiscent of endosomes, consistent with retention within membrane-bound intracellular compartments.

sEV-induced changes in TEER and tight junctions

To evaluate the functional and biological effects of EVs on endothelial behaviour, measurements were performed after 72 h, allowing sufficient time for internalised EVs to elicit measurable responses. TEER was chosen as the primary assay for this purpose, as it provides a direct, quantitative, and real-time measure of barrier integrity, offering immediate insight into cell-cell junction status and overall endothelial functionality, and guiding subsequent in-depth analyses. TEER analysis revealed that sEV treatment reduced barrier integrity in both endothelial cell models. In hCMEC/D3 cells, sEV decreased TEER by 36.79 ± 22.00% compared with control, while m/lEV caused a smaller, non-significant decrease of 19.30 ± 6.70% (Fig. 2a). Similarly, in HUVECs, sEV reduced TEER by 34.13 ± 9.53%, whereas m/lEV induced a non-significant decrease of 18.95 ± 10.71% (Fig. 2b). These data indicate that sEV-enriched fractions exert a stronger disruptive effect on endothelial barrier function, whereas m/lEV-enriched fractions have little to no impact on barrier integrity.

Fig. 2.

TEER and Localisation of ZO-1 and Claudin-5. TEER measurements of (a) hCMEC/D3 and (b) HUVEC monolayers after 72 h of treatment with sEV and m/lEV. Data are presented as normalised percentages relative to the untreated control. Statistical significance was determined using a nonparametric Kruskal–Wallis test, with p < 0.05 considered significant. (c) Immunofluorescence staining of expression and localisation of TJ proteins ZO-1 and Claudin-5 in hCMEC/D3 and (d) in HUVECs. Cells were stained with ZO-1 or claudin-5 (green) and counterstained with Hoechst 33342 (blue). Red arrows indicate cell borders

Following significant TEER changes, immunocytochemistry was performed to visualize key TJ proteins, including Claudin-5 and ZO-1, linking functional barrier disruption to specific molecular alterations. Fluorescence microscopy revealed distinct effects of EVs on TJ proteins in both endothelial models. In hCMEC/D3 cells, ZO-1 staining showed fragmented and irregular junctional distribution in both sEV- and m/lEV-treated groups, while Claudin-5 localisation was continuous in control cells but disrupted and discontinuous following EV exposure, most prominently after sEV treatment (Fig. 2c). In HUVECs, control cells displayed continuous membrane-associated localisation of ZO-1 and Claudin-5, forming a coherent junctional network. Exposure to sEV resulted in diffuse ZO-1 signal throughout the cytoplasm and intense Claudin-5 localisation to filopodia, reducing membrane-associated signal, whereas m/lEV did not induce notable redistribution of either protein compared to control cells (Fig. 2d).

m/lEV-induced modulation of endothelial migration and angiogenesis

Subsequently, endothelial migration was evaluated to determine whether EVs modulate cell motility independent of barrier disruption. To assess cell migration, a wound healing assay was performed. After 9 h in hCMEC/D3 monolayers, m/lEV treatment increased the migrated area by 17.40 ± 6.32% relative to control, whereas sEV exposure produced no significant change (Fig. 3a-b). Similarly, in HUVECs, m/lEV increased cell motility by 33.03 ± 10.67%, whereas sEV had no significant effect (Fig. 3c-d). These results indicate that m/lEV-enriched fractions selectively promote endothelial cell migration, whereas sEV-enriched fractions do not alter motility in either cell type. Because proliferation was not directly assessed during the 9-h migration window, a minor contribution of cell division to wound closure cannot be fully excluded.

Fig. 3.

Migration and Tube formation. (a) Representative phase-contrast images of hCMEC/D3 monolayers at T0 and T9 for control, sEV-, and m/lEV-treated cells, and (c) HUVEC monolayers under the same conditions. (b) Quantification of hCMEC/D3 and (d) HUVEC cell migration after treatment with sEV and m/lEV using the wound-healing assay. The percentage of the area covered by migrated cells was calculated from images acquired at T0 and T9 (9 h post-incubation). (e) Representative phase-contrast images of capillary-like networks formed by hCMEC/D3 cells after 24 h and (f) HUVECs after 4 h. (g) Quantitative analysis of hCMEC/D3 and (h) HUVEC capillary-like network formation. Quantitative data are presented as mean ± SD from 3–4 independent experiments. Statistical significance was determined using a nonparametric Kruskal–Wallis test, with p < 0.05 considered significant

The increase in endothelial motility induced by m/lEV suggested functional activation; therefore, angiogenic potential was subsequently examined to determine whether enhanced migration translates into coordinated vascular network formation. Endothelial angiogenic potential was assessed by measuring the cell ability to form capillary-like tubular structures on an extracellular matrix. Quantitative analysis demonstrated that m/lEV treatment impaired angiogenic network complexity in hCMEC/D3 cells, reducing the number of nodes (–23.76 ± 3,13%) and meshes (–23.77 ± 5.11%) relative to the bFGF control, whereas sEV had no significant effect (Fig. 3e, g). In HUVECs, a similar effect was observed, with m/lEV-treated cells showing an even more pronounced reduction in the ability to form organised capillary-like networks compared with VEGF-stimulated positive controls. Microscopic image analysis revealed reductions in the number of nodes (− 45.87 ± 6.90%), meshes (− 55.70 ± 21.57%), segments (− 49.09 ± 13.94%), and junctions (− 44.83 ± 8.18%), accompanied by a decrease in total network length (− 32.07 ± 2.36%). In contrast, sEV treatment did not significantly alter any of the angiogenic parameters. (Fig. 3f, h). These findings indicate that m/lEVs-enriched fractions impair the formation of organised capillary-like networks in both hCMEC/D3 and HUVECs, despite enhancing endothelial cell migration.

EV size-dependent effects on mitochondrial respiration

Maintenance of endothelial barrier integrity, migration, and angiogenic network formation are highly energy-dependent processes; therefore, the functional alterations observed across these assays prompted further investigation of cellular metabolism. Mitochondrial respiration was subsequently evaluated to determine whether EV-induced changes in endothelial behavior are associated with alterations in bioenergetic capacity. This was performed using the Seahorse XFp Cell Mito Stress Test, which allows real-time assessment of key mitochondrial parameters. In hCMEC/D3 cells, treatment with m/lEV significantly reduced maximal respiration from 361.97 ± 39.05 to 243.76 ± 23.46 pmol/min/µg of pure protein, and non-mitochondrial respiration from 179.08 ± 17.05 to 127.08 ± 22.83 pmol/min/µg of pure protein, compared to control. In contrast, sEV treatment did not induce significant changes in either parameter (Fig. 4a, c). These findings indicate a pronounced impairment of mitochondrial function specifically associated with m/lEV. The decrease in maximal respiration may indicate increased energetic demand due to enhanced cell migration, while the simultaneous reduction in tube formation suggests that decreased bioenergetic capacity limits the cells’ ability to form vascular-like networks. Notably, mitochondrial function in HUVECs was not significantly affected by either sEV or m/lEV treatment (Fig. 4b, d). These results highlight a cell type- and EV size-dependent bioenergetic response, particularly reflecting the energy-sensitive nature of BBB endothelial cells.

Fig. 4.

Mitochondrial respiration. Original oxygen consumption rate (OCR) traces showing Seahorse MitoStress responses in (a) hCMEC/D3 and (b) HUVEC cells after sequential addition of oligomycin (Oligo), trifluoromethoxy carbonylcyanide phenylhydrazone (FCCP), and rotenone/antimycin A (Rot/AA), comparing sEV and m/lEV treatments. Summarised OCR parameters are shown in (c) hCMEC/D3 and (d) HUVEC cells. Quantitative data are presented as mean ± SD from 3 independent experiments. Statistical significance was determined using a nonparametric Kruskal–Wallis test, with p < 0.05 considered significant

Discussion

Our study establishes a comparative framework for analysing EV-enriched fractions across different size ranges while prioritising physiological relevance over artificial equalisation. Rather than normalising treatments to particle number or total protein content, we applied equal relative sample volumes generated using an identical isolation and concentration protocol. This approach preserves intrinsic differences in EV yield across subpopulations and more closely reflects the cumulative secretory output experienced by recipient cells in vivo. Although normalisation facilitates direct quantitative comparisons, it may obscure biologically meaningful differences in secretion dynamics, cargo loading, and biogenesis, and protein-based adjustments may include non-vesicular components. By avoiding such corrections, our design captures the integrated signalling potential of each EV fraction. The resulting variability in absolute particle counts is therefore considered a biologically relevant parameter contributing to tumour-driven intercellular communication.

Because our experimental design preserved natural differences in EV secretion by avoiding particle normalisation, careful characterisation of particle size and heterogeneity was essential for interpreting functional outcomes. NTA and TRPS, which rely on fundamentally different detection principles, provided complementary insights: NTA is highly sensitive to small, diffusive particles but may underestimate larger or more heterogeneous populations, whereas TRPS captures a broader size range through single-particle, pore-based measurement. In our study, both methods demonstrated partially overlapping size distributions, while consistently showing a shift toward smaller particles in sEV and larger, more heterogeneous populations in m/lEV. However, measured particle concentrations differed between methods, reflecting methodological characteristics rather than biological discrepancies. These findings highlight that EV quantification is method-dependent and support the use of complementary analytical approaches when characterising operationally defined EV-enriched fractions.

A 24-hour incubation period was established to evaluate the internalisation of HROG36-derived EVs, ensuring complete uptake beyond the initial binding and early endocytic stages. Durak-Kozica et al. [15] similarly assessed EV internalisation in endothelial cells, including HUVECs, after 24 h, demonstrating that this timeframe allows reliable detection of intracellular accumulation. We demonstrated that in hCMEC/D3 cells after sEV treatment, the fluorescent signal exhibited a reticular cytoplasmic pattern, suggesting targeted routing through the endomembrane system, such as the endoplasmic reticulum or Golgi apparatus, a pathway often utilised by exosomes to bypass lysosomal degradation [16]. Conversely, in HUVECs, the signal appears diffusely distributed, suggesting a less compartmentalised pattern. The uptake pathway of m/lEV further highlighted the specialised nature of the BBB; while these EVs were sequestered within endosome-like structures in HUVECs – consistent with classical macropinocytosis [17] – they showed diffuse cytoplasmic dispersion in hCMEC/D3 cells. This suggests that the BBB may facilitate more rapid endosomal escape or alternative membrane fusion for larger cargo to maintain intracellular homeostasis.

Numerous in vitro and in vivo studies have shown that GBM-secreted factors compromise BBB integrity; however, the specific contributions of EV-enriched fractions remain unclear. We demonstrate that GBM-derived sEV disrupt endothelial barrier integrity in both BBB and peripheral endothelial models, as reflected by reduced TEER and reorganisation of TJ proteins ZO-1 and Claudin-5. In our study, TEER served as the primary quantitative measure of barrier integrity, while immunofluorescence analysis of TJ proteins provided complementary insight into junctional organisation. We acknowledge that this qualitative approach may be subject to interpretation, and that additional quantitative image analysis could further strengthen these observations. Phenotypically, this aligns with previous reports that GBM EVs increase vascular permeability.

Notably, Treps et al. [18] showed that GBM stem cell-derived EVs enhance BBB permeability via semaphorin3A/Neuropilin-1 signalling. However, despite similar barrier-disruptive outcomes, the experimental design and likely mechanisms differ. Treps et al. analysed bulk EV preparations containing heterogeneous EV populations from a defined GBM stem cells, focusing on short-term exposure. In contrast, we compared sEV- and m/lEV-enriched fractions from non-stem GBM cultures after prolonged (72 h) treatment and evaluated responses in both the BBB and the peripheral endothelium. Given that semaphorin3A is primarily associated with a GBM stem cell phenotype [19], our findings suggest that barrier disruption by non-stem GBM-derived sEV is likely mediated, at least in part, by additional mechanisms beyond semaphorin3A/Neuropilin-1 signalling. Although Treps et al. demonstrated preferential association of semaphorin3A with the exosomal marker CD63, they did not assess the functional effects of this isolated EV subpopulation. Consequently, it remains unclear whether the observed increase in permeability reflects the combined activity of heterogeneous EV populations or the effect of a specific CD63-positive EVs. Our study has a similar limitation: the combined effects of sEV- and m/lEV-enriched fractions were not evaluated. Together, these observations indicate that although sEV consistently impair endothelial barriers, the underlying molecular pathways are context- and cell-phenotype-dependent, warranting further compositional and mechanistic analyses of distinct EV subpopulations.

While sEV primarily affected barrier integrity, m/lEV selectively enhanced endothelial migration in both hCMEC/D3 and HUVEC cells, indicating activation of pro-migratory signalling pathways that are not a consequence of junctional disruption. This observation is in line with a recent study demonstrating that patient-derived GBM larger EVs promote tumour cell migration via intracellular calcium transients mediated by Connexin-43 gap junctions and subsequent PYK2 phosphorylation [20]. Although these mechanisms were described in the context of tumour cells, they suggest that larger EVs can promote calcium-dependent motility independently of classical VEGF-VEGFR-dependent angiogenic signalling. A similar signalling axis may shape the endothelial response observed in our study; however, direct investigation of Connexin-43-dependent calcium dynamics and downstream kinases in endothelial cells would be required to confirm this hypothesis.

The VEGF-VEGFR-independent mechanism of migration may further explain why increased motility did not translate into enhanced angiogenesis, as m/lEV impaired the formation of organised capillary-like networks in vitro. If m/lEV preferentially activate calcium-dependent pathways, this may promote individual cell movement and inhibit the coordinated network assembly necessary for tube formation. Additionally, another study suggests that EVs and growth factors promote different signalling routes within a tumour, and that their relative contributions may account for clinically relevant differences in vessel morphology and permeability [21]. In this context, endothelial migration and tube formation may not be regulated by the same signalling cascade, and m/lEV-induced motility and angiogenesis may therefore occur independently of the canonical VEGF pathway.

Importantly, discrepancies with earlier reports may be attributable to differences in EV preparation. In contrast to previous studies that mainly examined bulk EVs or sEV and reported predominantly pro-angiogenic, dose- and time-dependent endothelial activation, our findings indicate a distinct effect of size-resolved fractions. For example, Giusti et al. [22] demonstrated that bulk EV preparations from U251 GBM cells stimulate BBB endothelial cell proliferation, migration, and tube formation in a dose-dependent manner (2–8 µg/ml), with stronger effects at 72 h than at earlier time points. Similarly, Yang et al. [23] reported that U251-derived sEV enhance HUVEC migration and tube formation, and Monteforte et al. [24] found that sEV from A-172 increase HUVEC angiogenic activity. However, our size-resolved analysis reveals that m/lEV-enriched fractions can strongly promote endothelial motility while simultaneously impairing organised tube morphogenesis, indicating that different EV subpopulations can drive distinct, sometimes opposing, endothelial programmes.

A key novel contribution of this study is the link between EV-induced functional changes and endothelial mitochondrial bioenergetics. Tumour-derived EVs are increasingly recognised as regulators of metabolic adaptation within the vascular microenvironment [25]. We found that m/lEV markedly impaired mitochondrial respiration in only hCMEC/D3, resulting in reduced maximal and non-mitochondrial respiration. Because endothelial barrier maintenance requires continuous ATP-dependent cytoskeletal tension and junctional protein trafficking, energetic insufficiency is expected to preferentially destabilize TJ organisation. Similar links between mitochondrial dysfunction and BBB permeability have been suggested in neurovascular disorders [26] but have rarely been explored in the context of tumour-derived EVs. Collectively, these observations indicate that, although mitochondrial suppression itself compromises endothelial barrier integrity, our findings identify GBM-derived EVs as a potential upstream trigger that induces comparable mitochondrial dysfunction and barrier weakening. In contrast, m/lEV promoted endothelial migration but impaired the formation of organised vascular networks without substantially affecting respiration in HUVECs. This phenotype is consistent with pathological angiogenesis, characterised by increased endothelial motility and aberrant vascular architecture with compromised structural hierarchy and stability, as described in tumour vasculature models [27]. Notably, m/lEV suppressed maximal and non-mitochondrial respiration in hCMEC/D3, indicating that the metabolically specialised brain endothelium remains vulnerable to mitochondrial perturbation even when the peripheral endothelium is not. Overall, these findings underscore the differential bioenergetic effects of EV subpopulations on endothelial cells and call for deeper investigation into the mechanisms linking mitochondrial modulation to barrier and vascular function.

Conclusion

In this study, we demonstrate that GBM-derived EVs exert size-dependent and functionally distinct effects on endothelial biology in both BBB and peripheral vascular models. SEV-enriched preparations primarily compromise endothelial barrier integrity, whereas m/lEV-enriched fractions enhance endothelial cell migration across both endothelial cell models while disrupting the formation of organised angiogenic networks. These pro-migratory yet anti-angiogenic effects of m/lEVs in BBB endothelium are linked to a selective suppression of mitochondrial respiration, whereas no such impairment was observed in peripheral endothelium. Collectively, our findings establish that sEV and m/lEV are biologically distinct signalling entities that differentially shape vascular behaviour in GBM. This underscores the importance of size-resolved EV analysis for mechanistic studies and provides a foundation for developing EV-based biomarkers and targeted strategies to mitigate tumour-associated vascular dysfunction.

Abbreviations

BBB

Blood–brain barrier

bFGF

Basic fibroblast growth factor

Cryo–TEM

Cryogenic transmission electron microscopy

EVs

Extracellular vesicles

FCCP

Trifluoromethoxy carbonylcyanide phenylhydrazone

GBM

Glioblastoma

m/lEV

Medium/large extracellular vesicles

NTA

Nanoparticle tracking analysis

OCR

Oxygen consumption rate

sEV

Small extracellular vesicles

TEER

Transendothelial electrical resistance

TJ

Tight junction

TRPS

Tunable resistive pulse sensing

ZO–1

Zonula Occludens–1

Author contributions

D.K. and K.K. contributed equally; A.J., E.S. conceived the idea; D.K., K.K. designed the experiments; D.K., K.K. isolated the EV fractions; D.K. performed and analysed all HUVEC experiments; K.K. performed and analysed HROG36 and hCMEM/D3 experiments; A.J., Z.B. contributed to EV characterisation. D.K., K.K., A.J. wrote the paper. All authors reviewed the manuscript.

Funding

This project has received funding from the Research Council of Lithuania (LMTLT), agreement No S-PD-24-107 and The Lithuanian University of Health Sciences Research Fund.

Data availability

The datasets generated during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.