Breast Cancer‐Derived Extracellular Vesicles Modulate the Cytoplasmic and Cytoskeletal Dynamics of Blood‐Brain Barrier Endothelial Cells

2.1. Cell Lines and Cell Culture

The human triple‐negative breast cancer (TNBC) cell line MDA‐MB‐231 was purchased from the American Type Culture Collection (ATCC HTB‐26, VA, USA). The brain‐seeking (MDA‐231Br) variant of the TNBC cell line MDA‐MB‐231 was a gift from Dr. T. Yoneda, Indiana University, USA (Yoneda et al. 2001). Primary human brain microvascular endothelial cells (BECs) were purchased from Cell Systems Co. (cat. no. ACBRI 376, Kirkland, WA). TNBC cells were cultured in Dulbecco's modified Eagle's medium (DMEM, cat. no. 11885084, Thermo Fisher Scientific Inc.) supplemented with 1% penicillin−streptomycin (10000 U/mL) (cat. no. 15140148, Thermo Fisher Scientific Inc.) and 10% foetal bovine serum (FBS, cat. no. S11150, Atlanta Biologicals, Atlanta, GA, USA). TNBC cells were seeded in 150‐mm dishes and cultured for 48 h in DMEM supplemented with 10% EV‐depleted FBS to produce conditioned media for isolating EVs. EV‐depleted FBS was prepared by 16 h of ultracentrifugation at 100,000 × g and 4°C. BECs were cultured with a complete classic medium with serum and culture boost (cat. no. 4Z0‐500, Cell System, WA, USA). All cells were used from passages 4–20, maintained in a 37°C humidified incubator with 5% CO₂. All cultures were regularly screened for mycoplasma contamination using the MycoAlert PLUS Mycoplasma Detection Kit (cat. no. LT07‐710, Lonza, Inc.) according to the manufacturer's instructions.

2.2. EV Isolation and Characterization

Conditioned media was collected from MDA‐MB‐231 and MDA‐231Br cells after 48 h of incubation in EV‐free media. Conditioned media was used for EV isolation only if cell viability was > 95%, as determined by Trypan Blue. EVs were isolated using a differential centrifugation protocol. Briefly, conditioned media was collected and centrifuged at 400 × g for 10 min, 2000 × g for 20 min, and 14,000 × g for 45 min at 4°C (Sorvall Lynx 4000, Thermo Fisher Scientific Inc.) to remove dead cells and large debris, apoptotic bodies, and larger microvesicles, respectively. As previously described by us, the resulting supernatant subsequently underwent ultracentrifugation at 100,000 × g for 90 min at 4°C (Optima XE‐90 Ultracentrifuge, Beckman Coulter Life Sciences). The first pellet was washed in PBS and precipitated by repeating the UC step at 100,000 × g for another 90 min. The washed pellet was resuspended in sterile cell‐grade PBS (100–200 µL) for characterization and experiments or stored at −80 C. EV preparations were characterized according to the latest Guidelines of the International Society for Extracellular Vesicles (Figure S1) (Welsh et al. 2024). Total protein concentration of EVs was measured by Bradford assay (Bio‐Rad Protein Assay Dye Reagent Concentrate, cat. no. 5000006, Bio‐Rad, CA, USA) according to the manufacturer's instructions. Bradford assay was performed in a clear 96‐well plate and 590 nm absorbance was measured using FilterMax F3 multi‐mode microplate reader (Molecular Devices, CA, USA). EV size distribution and particle number/mL of solution were measured by nanoparticle tracking analysis (NTA, NanoSight NS300, Malvern Instruments, UK). EV formulations were diluted (1:500–1:1000) in sterile phosphate‐buffered saline (PBS) and analysed (3 mL) with a Malvern Panalytical Nanosight NS300 (60‐s measurement; 3 capture replicates). The presence of EV markers cluster of differentiation (CD)63, CD73, and annexin A2, and the absence of calnexin, a marker of the endoplasmic reticulum (ER), as a negative control, was evaluated by immunoblot (5–20 µg of total proteins/lane). EV morphology was evaluated by transmission electron microscopy (TEM) as previously described by us (Morad et al. 2019). Imaged by TEM and analysed by NTA, P‐EVs, ad Br‐EVs showed characteristic disc‐shaped morphology and heterogeneous size distribution in the nanometre range with a mode of 166 nm for P‐EVs and 157 nm for Br‐EVs (Figure S1a,b). P‐EVs and Br‐EVs had a total protein content of 4–10 µg/10¹⁰ EVs and were enriched in typical EV markers, such as CD63, CD73 and annexin A2. Both EV samples lacked calnexin, an ER marker used to detect the possible presence of nanosized intracytoplasmic contaminants (Figures S1c,d).

2.3. In Vitro EV Treatments

To evaluate the effect of EVs on BEC intracytoplasmic dynamics, BECs grown to confluence in 6‐well plates were treated daily for a total of 5 days with a fixed quantity of EVs (10³ EVs/cell). After, cells were washed 3 times with sterile cell‐grade PBS and lysed using lysis buffer 1X in sterile water (Cell Signaling Technology, Danvers, MA) supplemented with phenylmethylsulphonyl fluoride protease inhibitor incubated for 5 min on ice. Cell lysates were collected with a scraper, quickly tip sonicated, and centrifuged at 14,000 × g for 10 min at 4°C, the supernatant was collected, total proteins were quantified by Bradford analysis, and analysed by Western blot (20–30 µg of total protein/lane).

2.4. Immunoblot Analyses

Immunoblotting was performed following the protocol previously developed by our lab (Morad et al. 2019). Antibodies targeting the following proteins were utilized for immunoblotting: CD63 (1:500, Abcam, cat. no. 59479), CD73 (1:500, Cell Signaling Technologies, cat. no. 13160), annexin A2 (1:500, Novus Biologicals, cat. no. NBP1‐31310), calnexin (1:1000, GeneTex Inc., cat. no. GTX112886), rab 7 (1:1000, Abcam, cat. no. 137029), rab11fip2 (1:500, Abcam, cat. no. ab180504), rab11fip3 (1:500, Proteintech, cat. no. 25843‐1‐AP), rab11fip5 (1:500, Proteintech, cat. no. 14594‐1‐AP), myosin5b (1:500, Lifespan Biosciences, cat. no. LS‐C410093), ehd‐1 (1:500, Proteintech, cat. no. 24657‐1‐AP), rab4 (1:500, Lifespan Biosciences, cat. no. LS‐B8414), ZO‐1 (1:500, Thermo Fisher Scientific, cat. no. 40–2200), GFAP (1:500, Abcam, cat. no. 7260), NKCC1, SLC12A2 (1:500, Proteintech, cat. no. 13884‐1‐AP), and GAPDH (1:3000, Santa Cruz Biotechnology, cat. no. G‐9, sc‐365062).

2.5. CRISPR/Cas9 Cell Transfection

Eighty percent confluent BECs in a 6‐well plate were transfected using Lipofectamine CRISPRMAX Cas9 Transfection reagent (Thermo Fisher Scientific) using 7.5 µg of TrueCut Cas9 Protein v2 (Thermo Fisher Scientific), 1 µL of TrueGuide Synthetic Rab11fip2 (CRISPR989610_SGM) and Rab7 (CRISPR1002837_SGM) gRNA (100 µM), 15 µL of Cas9 Plus Reagent, and 7.5 µL of CRISPRMAX Reagent. After 48 h, cells were used in imaging experiments.

2.6. Cell Electroporation

A total of 0.5 × 10⁶ BECs were harvested, pelleted at 400 × g for 10 min, and electroporated using Cell Line Nucleofector Kit V (Lonza Walkersville, USA). The cell pellet was resuspended in 100 µL of buffer, mixed with 2 µg of Rab11fip3 (human tagged orf clone rc227246, Origene Technologies Inc.) and Rab11fip5 (human tagged orf clone rc206173, Origene Technologies Inc.) plasmids, and electroporated using the S‐05 Program of the Amaxa Biosystem Nucleofector. After, cells were transferred into a 6‐well plate previously filled with 2 mL of media warmed at 37°C and cultured for 24 h before being used in imaging experiments.

2.7. Quantitative Phase Imaging

BECs were seeded in a 6‐well plate and treated with a fixed quantity of EVs (10³ EVs/cell) daily for 5 days. BECs were treated with CRISPRMAX/Cas9 or electroporated 48 and 24 h before the experiments, respectively. Cells were imaged live for 24 h at 37°C using the quantitative phase‐contrast microscope HoloMonitor (PhiOptics). Three different fields per sample were captured at 10‐min intervals.

2.8. Live Fluorescence Imaging

BECs were treated with a fixed quantity of EVs (10³ EVs/cell) daily for 5 days. BECs were treated with CRISPRMAX/Cas9 and electroporated 48 and 24 h before the experiments, respectively. Twenty‐four hours prior to the live imaging, cells were seeded in a 24‐well plate (2 × 10⁴ cells/well). Cells were stained overnight at 37°C with DiI lipophilic fluorescent staining. Fluorescence images were acquired using a Nikon Ti2‐E inverted perfect focus microscope equipped with a microscope cage incubator from OKOLabs, which is capable of controlling 37°C and 5% CO₂ during live cell imaging. The time series images were acquired every 10 min for duplicated positions within each well (24 wells) for 48 h with bright field and Cy3 (cytoplasmic labelling) channels. The raw images were obtained with NIS‐Elements Advanced Research Package for 6D Image Acquisition software (Nikon). The ND2‐formatted raw images were converted to tiff and movie files for ML‐based analysis using NIS‐Elements Advanced Research Package for 6D Image Acquisition software (Nikon) or FiJi (National Institutes of Health, Bethesda, MD).

2.9. Machine Learning Analysis

The proposed ML‐based framework deals with confocal, quantitative phase imaging (qPI) and live cell imaging data to analyse morphodynamical features for specific cellular phenotypes. First, we generated cell segmentation outcomes by using CellPose (Stringer et al. 2021) (https://www.cellpose.org/), a deep learning‐based algorithm for cell image segmentation, for confocal images and an adaptive image thresholding approach (Margiotta et al. 2017) for qPI and live cell images. Then morphology features were obtained from these outcomes, and TrackMate computational tools (Jaqaman et al. 2008) were utilized to extract motility features from live cell imaging data. There are 10 morphology features and 9 motility features being collected to generate feature distribution maps by the UMAP method (McInnes, Healy, and Melville 2018). To clearly represent the differentiation of BECs’ phenotypes, spectral clustering (Andrew, Ng, and Weiss 2001) with Silhouette score (Rousseeuw 1987) was utilized to automatically identify potential subgroups that have similar morphological or motility characteristics.

2.10. In‐Solution Digestion and Solid‐Phase Extraction With Stage Tips

Isolated microvessels from the blood‐brain barrier were suspended in 100 µL of protein solubilization buffer (1% w/v SDC, 10 mM TCEP, 40 mM CAA, and 50 mM TEAB) and subjected to a 10‐min heat treatment at 70°C. This step was performed to ensure complete protein solubilization, denaturation, reduction of cysteine residues, and alkylation of cysteine residues. Subsequently, the samples were centrifuged at 21,000 × g for 10 min and the resulting supernatants were transferred to fresh Eppendorf tubes. Protein concentrations were quantified using the PierceTM 660 nm Protein Assay Reagent (Thermo Fisher Scientific, Waltham, MA, USA), with a standard curve generated using a dilution series of BSA.

For the preparation of extracts for tryptic digestion, 10 µg of protein from each sample was transferred to a new Eppendorf tube and brought to a total volume of 50 µL with protein solubilization buffer. Following this, 1 µg of trypsin in 50 µL of 50 mM TEAB was added to each sample, and the digestions were allowed to incubate overnight at 37°C in an incubator. The following day, an additional 1 µg of trypsin in 1 µL of 50 mM TEAB per sample was added, and the digestions were incubated for an additional 4 h at 37°C. The resulting peptides were purified using SDB‐RPS stage tips, following the protocol developed by Rappsilber, Mann, and Ishihama (2007).

2.11. Measurement of LC‐MS Data in DIA Mode

The LC‐MS/MS data acquisition process was conducted utilizing a Vanquish Neo nanoLC system equipped with an Orbitrap Eclipse mass spectrometer, an FAIMS Pro Interface, and an Easy Spray ESI source, all supplied by Thermo Fisher Scientific (Waltham, MA, USA). For nanoLC separation, an Acclaim PepMap trap column (75 µm × 2 cm) in conjunction with an EasySpray ES802 column (75 µm × 25 mm, 100 Ȧ) from Thermo Fisher Scientific was employed. A 3 µL peptide extract was introduced into the system. The separation of peptides was achieved using a mobile phase consisting of 0.1% (v/v) formic acid in water (solvent A) and 0.1% (v/v) formic acid in 80% (v/v) acetonitrile (solvent B), with a flow rate of 300 nL/min and a column temperature maintained at 40°C. The column was initially conditioned for 2 min with 3% solvent B, followed by a gradual linear gradient up to 40% solvent B over a 60‐min period. Any remaining peptides bound to the C18 resin were subsequently eluted with 95% solvent B for 11 min. In the positive ion mode, the ion source temperature was set to 305°C and 2000 V, and ionized peptides were routed through the FAIMS Pro unit at −50 V. Mass spectra were acquired with a resolution of 120,000 in MS1 mode over the mass range of m/z 350–2000, with standard automatic gain control (AGC) settings and automatic injection time. For MS/MS fragmentation, a data‐independent acquisition (DIA) mode was employed within the mass range of 375–1200 at a resolution of 30,000. The collision energy was set at 30%, and an AGC target of 1000% was applied, utilizing m/z 25 isolation windows with 0.5 m/z overlaps for the collection of MS2 spectra.

2.12. Data Analysis With DIA‐NN and Statistics

The proteome data analysis was carried out with the DIA‐NN 1.8.1 software package (Demichev et al. 2020). The data from the LC‐MS data containers was processed and analysed using the following settings: FASTA‐database: Mus musculus (UP000000589); FASTA digest for library‐free search/library generation; protease: trypsin/P; missed cleavages: 1, N‐term M excision; C carbamidomethylation; peptide length range: 7–30; precursor charge range: 1–4; precursor m/z range: 300–1800; fragment ion m/z range: 200–1800; precursor FDR (%): 1.0; use isotopologues; heuristic protein interference; no shared spectra; protein interference: genes; neural network classifier: single‐pass mode; quantitation strategy: robust LC (high precision); cross‐run normalization: RT‐dependents; library generation: smart profiling; speed and RAM usage: optimal results. The raw protein peak areas were used for statistical analysis following the descriptions by Schulte, Hasturk, and Hardt (2019). Comparative analysis was conducted on proteins showing a log2 fold‐change absolute value greater than 0.5 and an adjusted p value equal or less than 0.05 (Millar et al. 2002; Wu and Voeltz 2021) (Figure S2a–c).

2.13. MDA‐MB‐231 Adhesion Assay

BECs were seeded onto 6‐well plates, cultured into monolayers, and then treated for 5 days with a fixed quantity of Br‐EVs (10³ EVs/cell) or the corresponding volume of sterile cell‐grade PBS. Following the EV treatments, 5 × 10⁴ MDA‐MB‐231 cells stably expressing CD63‐TdTomato fusion protein were incubated for 1 h at 37°C. Cells were washed three times with PBS, stained with DAPI, fixed in 4%PFA, and imaged by fluorescent microscopy. MDA‐MB‐231 red fluorescent cells adhering to the BECs monolayers were counted at 10X magnification (5–6 fields/well, n = 3 wells/treatment).

2.14. In Vivo Experiments

All in vivo experiments were conducted in compliance with the guidelines of the Institutional Animal Care and Use Committee (IACUC) at Boston Children's Hospital, Boston, MA. For all animal experiments, 6−8‐week‐old female athymic nude mice (Nu/Nu) were purchased from Massachusetts General Hospital. The mice were allowed to acclimate for at least four days before the start of the experiments. The minimum number of animals required to obtain data amenable to statistical analysis was used for animal experiments. For the brain pre‐metastatic niche studies, mice were randomly divided into 3 groups to receive retro‐orbital injections of EVs derived from parental and brain‐seeking MDA‐MB‐231 cells (3 µg of EVs in 100 µL of PBS per injection) or 100 µL of PBS. Injections were conducted on alternate days, switching between the right and left eye, for a total of 10 injections. Twenty‐four hours after the last EV injection, mice were sacrificed by CO₂ or cardiac perfusion under anaesthesia with 50 mL of 1% formaldehyde + 0.5% methanol. Perfused brain tissues were collected, fixed for 1 h at RT in 1% formaldehyde, then preserved at +4°C for 24–48 h in a sterile 30% sucrose PBS solution until completely sunk. Brain tissues in 30% sucrose solution were then sent to iHisto (Salem, MA, USA) for cryo‐sectioning. Fresh brain tissues were immediately processed for the ex vivo isolation of cerebral microvessels from the brain cortex, as described here (Lee et al. 2019). BBB microvessels were isolated and stored at −80 C in the form of a dry pellet until further analysis. For histological analysis, each brain was sectioned whole through sequential coronal cuts. Going from the frontal cortex to the cerebellum, 15–25‐µm‐thick slices were prepared for a total of 15 slides for each brain.

2.15. Immunocytochemistry and Immunohistochemistry

For immunocytochemistry, cells and brain microvessels samples were fixed with 4% paraformaldehyde + 4% sucrose in PBS for 14 min, rinsed thrice in PBS, and then treated with 0.2% Triton X‐100 for 5 min for permeabilization. For immunohistochemical staining, frozen sections were fixed with ice‐cold acetone for 7 min and washed three times in PBS. Cells and microvessel blocking was performed using 5% bovine serum albumin for 1 h at RT, whereas brain tissue section blocking was performed using 10% normal goat serum NGS + 0.3% Triton X‐100 in TBS for 1 h at RT. Cells or tissue sections were incubated with the primary antibody (1:100) in 1% BSA for 1 h at room temperature or overnight at 4°C. After the washes, cells or tissue sections were incubated with the relevant secondary antibody (1:1000) in 1% BSA for 45 min. Cells were washed with sterile PBS and stained with DAPI (cat. no. S33025, Thermo Fisher Scientific, USA). Tissue sections were washed with sterile PBS and mounted with Vectashield Plus antifade mounting medium with DAPI (cat. no. H‐2000‐10, Vector Lab Inc., CA, USA). For immunocytochemistry and/or histochemistry we used the following antibodies: Rab11fip2 (1:500, Abcam, cat. no. ab180504), Rab11fip3 (1:500, Proteintech, cat. no. 25843‐1‐AP), Rab11fip5 (1:500, Proteintech, cat. no. 14594‐1‐AP), CD31 (1:100, GeneTex, cat. no. GTX130274), MAP2 (1:100, Abcam, cat. no. ab5392), GFAP (1:100, Novus Biologicals, cat. no. NBP1‐05198), CD11b (1:100, Abcam, cat. no. ab133357), CD45 (1:100, Abcam, cat. no. ab23910), AlexaFluor488 goat anti‐rabbit (1:1000, Invitrogen, cat. no. A11008), goat pAb to chicken IgY Alexa fluor 647 (1:1000, Abcam, cat. no. ab150171), donkey pAb to rat IgG Alexa fluor 647 (1:1000, Abcam, cat. no. ab150155). Images were taken using a Zeiss LSM 880 Confocal Microscope. Images were processed using ImageJ. Fluorescent signals were normalized with the background signal of each image and quantified as described (Shihan et al., 2021). Sections 2 and 14 of each brain were stained with haematoxylin and eosin (H&E) staining; the absence of oedema and inflammation was confirmed in H&E brain sections in a blinded manner.

2.16. Statistical Analyses

Statistical analyses were performed using GraphPad Prism software, Excel, R and QIAGEN Ingenuity Pathway Analysis. Statistical significance was considered at p values lower than 0.033 for GraphPad Prism and lower than 0.05 for Excel, R and Ingenuity Pathway Analysis. Outliers were excluded when performing a comparative analysis of quantitative proteomic results. We used the Kruskal–Wallis test with Dunn's multiple comparisons test to assess differences among groups for experiments with n = 10 or less and when data normality could not be determined (Vrbin 2022). We used the one‐way ANOVA with Tukey's multiple comparison test for experiments with n ≥ 10 and when normality could be determined (Hazra and Gogtay 2016). In vivo experiments were assessed using the Mann−Whitney test (Vrbin 2022; Hazra and Gogtay 2016). The respective figure legends explain the specific statistical analysis methods for each figure.

3.1. Br‐EVs Modulate Multiple Rab11fip Proteins Involved in the Trafficking, Recycling, and Transcytosis of Rab11‐Positive Intracytoplasmic Vesicles

We have previously reported that Br‐EVs breach the intact BEC monolayer via transcytosis involving recycling endosomes expressing Rab11 (whose protein levels are not impacted by Br‐EV treatments) and downregulate Rab7 in BECs to inhibit the intracytoplasmic degradation pathways and facilitate Br‐EV trafficking toward the brain (Morad et al. 2019). To study the effects of Br‐EVs on BEC recycling and trafficking pathways involving Rab11‐positive vesicles, we treated BECs for five days with a fixed quantity of small P‐EVs and Br‐EVs (100–350 nm) isolated and characterized according to the guidelines published by the International Society of Extracellular Vesicles (Welsh et al. 2024) (Figure S1). We then analysed, by immunoblot and immunofluorescence, BEC protein content for Rab11 effector proteins known to interact with Rab11‐expressing endocytic vesicles, motor proteins, and cytoskeletal components (Figure 1a,b) and regulate the intracytoplasmic vesicle trafficking to different compartments of the cell (Ai and Skop 2009; Machesky 2019). Specifically, Rab11fip2 mediates the binding between Rab11 vesicles and myosinVb, thereby facilitating the recycling of Rab11 vesicles back to the plasma membrane (Machesky 2019, Solinger et al. 2020; Solinger et al. 2022; Schafer et al. 2014). Rab11fip3 contributes to increased stability within the endosomal compartment, facilitates the positioning and movement of Rab11‐positive endosomes relative to the microtubule network, and regulates the placement of pericentrosomal recycling endosomes during both cell division and breast cancer progression (Ai and Skop 2009; Fielding et al. 2005; Tong et al. 2017). Rab11fip5 is a component of the Factor for Endosome Recycling And Rab Interactions (FERARI) (Solinger et al. 2020; Solinger et al. 2022), a multi‐subunit tethering complex required for Rab11‐dependent recycling and reported to be involved in transcytosis and endosomal recycling and trafficking (Machesky 2019). Our results show that compared with PBS‐ or P‐EV‐treated BECs, cells treated with Br‐EVs show a significant decrease in Rab11fip2 (Figure 1b–e) and a significant increase in Rab11fip3 (Figure 1b,f–h) and Rab11fip5 (Figure 1b,i–k). Interestingly, although Br‐EVs modulate multiple Rab11fip proteins involved in cytoplasmic vesicle recycling, other proteins such as myosinVb (a member of the myosin family of actin‐based motor proteins known to interact with Rab11fip2) (Schafer et al. 2014; Hales et al. 2002; Gidon et al. 2012; Ji et al. 2019), rab4 (an early endosome protein that regulates vesicle fast recycling route to the plasma membrane) (Mohrmann et al. 2002), and EHD‐1 (an endocytic recycling protein and a component of the FERRARI complex) (Solinger et al. 2020; Solinger et al. 2022) were not affected by the Br‐EV treatment (Figure S3). These results demonstrate that Br‐EVs decrease the expression of Rab11fip2, which is involved in the actin‐based recycling of vesicles to the apical plasma membrane while increasing the expression of Rab11fip3 and Rab11fip5, which are linked to microtubule‐based vesicle recycling towards deeper cytoplasmic compartments of polarized BECs and transcytosis. These findings are the first to suggest that Br‐EVs modulate intracytoplasmic proteins involved in vesicle trafficking, particularly with respect to endocytosis, recycling pathways, and transcytosis.

3.2. Murine Cerebral Microvessels Show Decreased Rab11fip2 and Increased Rab11fip3 and Rab11fip5 Levels in Response to Br‐EV Treatment

To confirm the in vitro observation of Rab11fip protein modulations in vivo, we assessed the effects of Br‐EVs on murine cerebral microvessels, which consist of BECs, pericytes, and astrocyte feet (Lee et al. 2019). Athymic nude mice received a total of 10 intravenous injections of either PBS or 3 µg (∼3−4 × 10⁹ EVs/100 µL) of P‐EVs or Br‐EVs on alternate days (Morad et al. 2019; Morad et al. 2020). This dose has been shown to be proportional to the concentration of circulating EVs in tumour‐bearing mice and leads to an increased incidence of breast‐to‐brain metastasis in vivo (Morad et al. 2019; Morad et al. 2020). Following treatment, mice were euthanized and their brain cortex was isolated and immediately processed to extract cerebral microvessels (Figure S4). Rab11fip2, Rab11fip3, and Rab11fip5 protein content was then analysed by immunoblot. Br‐EVs can significantly upregulate Rab11fip5 compared to both P‐EVs and PBS, while both P‐EVs and Br‐EVs increase Rab11fip3 and decrease Rab11fip2 when compared to PBS treatments. Immunofluorescence analysis and signal quantification (Figure 2b,c) showed a significantly increased expression of Rab11fip3 and Rab11fip5 in cerebral microvessels of mice treated with P‐EVs and Br‐EVs compared to PBS controls. In contrast, the immunofluorescence signal quantification of Rab11fip2 showed no statistically significant difference between Br‐EVs, P‐EVs, and PBS treatments (Figure 2c), as opposed to the immunoblot analysis (Figure 2a). This may be attributable to the fact that murine cerebral microvessel samples, while enriched in BECs, also included pericytes and astrocyte feet (Andrew, Ng, and Weiss 2001) (Figure S4). These additional cell types may have undergone distinct protein changes in response to EV treatments, potentially masking the BEC Rab11fip2 protein downregulation. Furthermore, the absence of precise molecular weight control inherent in immunofluorescence assays represents a known technical limitation and could have contributed to the observed discrepancies, making the semi‐quantitative interpretation of Rab11fips expression results more challenging.

It is worth noting that, as previously observed by us (Morad et al. 2019; Morad et al. 2020; Busatto et al. 2020) and other groups (Hoshino et al. 2015; Liu et al. 2018; Ruan et al. 2024), P‐EVs and Br‐EVs sometimes display similar effects compared to PBS controls. This might be because both EVs originate from cell lines sharing the same background phenotype and retain similarities in certain pathways of cancer progression and pathogenesis. Despite these similarities, Br‐EVs cause distinct changes at the BBB, both in vitro and in vivo, supporting their efficiency in promoting the brain metastatic process. These in vivo findings corroborate our in vitro results and describe previously unknown mechanisms through which Br‐EVs can hijack the endocytic dynamics, vesicular trafficking, and transcytosis pathways of the BECs of the BBB.

3.3. ML Analysis Identifies Morphological and Motility Changes Induced by Br‐EVs That Prime BECs for Metastatic Cancer Cell Transmigration

The BECs of the BBB form an endothelial barrier facing the bloodstream and interact with its circulating components, such as cells, EVs (Busatto et al. 2021; Morad and Moses 2019; Morad et al. 2019; Morad et al. 2020; Busatto et al. 2020), and other blood‐derived nanoparticles. Immune and metastatic cancer cells have the ability to transmigrate the BBB endothelium via the paracellular or transcellular route (Herman et al. 2019; Motallebnejad et al. 2022; Godinho‐Pereira et al. 2021). Paracellular transmigration requires the disruption of the tight junctions between BECs, whereas transcellular transmigration occurs through the cytoplasm of a single BEC (Herman et al. 2019; Motallebnejad et al. 2022; Godinho‐Pereira et al. 2021). In the context of breast‐to‐brain metastasis, invasive TNBC cells are known to preferentially use the transcellular route (Herman et al. 2019; Godinho‐Pereira et al. 2021; Khuon et al. 2010). When TNBC cells engage with the microvascular endothelium of the BBB, there is a significant cytoskeletal and membrane reorganization that leads to BEC phenotypic alterations (Herman et al. 2019). Unlike normal BECs, which are involved in maintaining BBB homeostasis and constitute a selective barrier that does not interact with circulating cells, engaged BECs play a crucial role in metastatic cell extravasation. In particular, they envelop and facilitate the transcellular migration of metastatic cells by extending membrane protrusions (Herman et al. 2019) and reorganizing their cytoskeleton to form a myosin‐contracting cage (Khuon et al. 2010; Wu, Liu, and Zhou 2017), respectively. Surprisingly, it is currently unknown whether Br‐EVs are capable of inducing early phenotypical changes in BECs that prime the BBB for TNBC cell adhesion and transendothelial migration (Nag 2011).

To address this mechanistic void, we analysed BEC morphology and motility changes after treatment with Br‐EVs using an ML approach. Since alterations in cellular shape and movement reflect underlying changes in cytoplasmic and cytoskeletal dynamics (Romano et al. 2022; Mentor, Makhathini, and Fisher 2022), we hypothesized that a thorough analysis of BEC morphological and motility parameters could serve as a readout of Br‐EV‐induced modulation of BBB properties, such as the maintenance of the ability to act as a highly selective and poorly permeable barrier against metastatic cells. Our ML framework was designed to extract, analyse, and visualize morphology and motility features from live and fixed, labelled, and unlabelled cell imaging techniques following treatment with PBS, P‐EVs, or Br‐EVs. Specifically, we utilized confocal (Figure 3a–g) and fluorescent live‐cell imaging (Figure 3h–n) to assess cell morphology. Quantitative phase imaging (QPI) (Figure 4a–g) and fluorescent live‐cell imaging (Figure 4h–n) were used to analyse cell motility. We collected and analysed 10 and 9 of the most used morphology and motility features (see Figures 3 and 4, panels g and n) (Stringer et al. 2021; Jaqaman et al. 2008; Gonzalez 2008). These selected features, which can be extracted from various types of imaging techniques, are widely utilized across different analysis frameworks and provide a comprehensive understanding of changes in cell size, shape, structure, speed, and directionality. For each condition, we generated a distribution map using the UMAP method (McInnes, Healy, and Melville 2018) (Figures 3a–c, h–j, and 4a–c, h–j). To define BEC phenotypes, we used non‐hierarchical spectral clustering (Andrew, Ng, and Weiss 2001) with a maximized Silhouette score (Rousseeuw 1987) to automatically identify clusters, each representing a phenotypic subgroup characterized by specific morphological and motility characteristics (Figures 3d,k, and 4d,k) (Andrew, Ng, and Weiss 2001, Rousseeuw 1987).

This unsupervised approach identified five phenotypic subgroups of BECs when visualized by confocal microscopy (Figure 3d), four phenotypic subgroups by fluorescent live‐cell imaging (Figures 3k and 4k), and three phenotypic subgroups by QPI (Figure 4d).

In the morphology analysis (Figure 3), control BECs treated with PBS exhibited a more uniform distribution across all subgroup clusters in both confocal microscopy and live‐cell imaging compared to BECs treated with Br‐EVs. When fixed BECs were analysed via confocal microscopy, Br‐EV treatment induced an enrichment in Clusters 3 and 4, defined by an increased Max Feret diameter (Desprez et al. 2023), major axis length and perimeter and decreased solidity and extent (Figure 3c–g). Interestingly, these results were recapitulated when live BECs were analysed by fluorescent live‐cell imaging with significant enrichment in Cluster 1, also defined by an increased Feret diameter, major axis length and perimeter, and decreased solidity and extent (Figure 3j–n). These data lead to the conclusion that Br‐EVs significantly increase BECs with larger size, elongated shape, and irregular boundaries compared to BECs treated with PBS and P‐EVs (Figure 3e–g, and l–n). Collectively, these results indicate that Br‐EV‐treated BECs display altered cell‐to‐cell adhesion dynamics and modifications to the cell cytoskeleton, suggesting impaired barrier function(s).

Motility analysis (Figure 4) demonstrated that treatment with Br‐EVs significantly increased the percentage of cells characterized by fast, spread, and convoluted movements with poor directionality, all of which suggest inefficient cell migration. When live BEC were analysed by QPI following Br‐EV treatment, there was a significant enrichment of Cluster 2, defined by increased speed and decreased linearity of forward movement (Figure 4d–g). When BECs were analysed by live‐cell imaging, there was a ∼2‐fold increase in Cluster 4 (Figure 4k–n). Similar to the QPI results, Cluster 4 is also defined by increased speed and decreased linearity of forward movement. Compared to PBS and P‐EVs, Br‐EVs significantly increase BECs characterized by faster but random movements with poor directionality (Figure 4e–g and l–n). The random motility patterns and the displayed enhanced migratory behaviour of BECs treated with Br‐EVs suggest changes in the cytoskeleton dynamics and the endothelial barrier integrity and permeability, respectively. Taken together, ML analysis consistently indicates that Br‐EVs have a significant impact on BEC phenotype populations and promote the selection of larger BECs with irregular boundaries and displaying fast and random movements. These Br‐EV‐induced morphology and motility changes in BECs are consistent with the cytoskeletal reorganizations required for the transendothelial migration of TNBC cells across the BBB.

3.4. BECs With Rab7 Knockdown (KD) and Rab11fip5 Overexpression Exhibit Morphological and Motility Changes Comparable to Those Induced by Br‐EV Treatment

Br‐EVs cause a significant decrease in Rab7 (Morad et al. 2019) and Rab11fip2 and an increase in Rab11fip3 and Rab11fip5 in the BECs of the BBB. These proteins not only play a role in intracytoplasmic vesicle degradation and sorting (Morad et al. 2019, Solinger et al. 2020; Solinger et al. 2022; Gidon et al. 2012; Cullis et al. 2002; Horgan et al. 2007) but also regulate the connection and transport of endocytic vesicles on cytoskeletal filaments and the maintenance of cell polarity and motility (Machesky 2019; Margiotta et al. 2017; Tong et al. 2017; Mateus et al. 2018). Rab7 regulates cell motility by directly interacting with Rac1 (Margiotta et al. 2017). Its depletion leads to the accumulation and expansion of ER membranes and decreased cell directionality (Mateus et al. 2018). Rab11fip2 regulates cell polarity and migration, activating PI3K/Akt signalling and increasing MMP7 expression (Xu et al. 2016). Rab11fip3 has been reported to regulate cell endocytic transport and motility dynamics by regulating the polarization of the actin‐based cytoskeleton and the formation of lamellipodia (Tong et al. 2017). Finally, the KD of Rab11fip5 resulted in a significant decrease in the membrane levels of pro‐migratory integrin, leading to compromised cell motility and adhesion to the ECM (Mateus et al. 2018).

We employed the same imaging analyses, coupled with ML, to investigate whether the modulation of Rab7, Rab11fip2, Rab11fip3, and Rab11fip5 contributes to the morphological and motility changes observed in BECs treated with Br‐EVs (Figure 5). We imaged BECs knocked down (KD) for Rab7 and Rab11fip2 or overexpressing Rab11fip3 and Rab11fip5 using confocal microscopy (Figure 5a,b), live cell imaging (Figure 5c,d, and g,h), and QPI (Figure 5e,f). We extracted and analysed their morphology (Figure 5a–d) and motility features (Figure 5e–h). Notably, unsupervised ML revealed that BECs with Rab7 KD and Rab11fip5 overexpression recapitulate the cell clustering, morphology, and motility features observed in BECs treated with Br‐EVs (Figures 3c,j, and 4c,j). These results suggest that the phenotypic changes induced by Br‐EVs are significantly influenced by the observed protein modulations, particularly the decrease in Rab7 and Rab11fip2 (the latter being evident only in morphological changes detected via confocal microscopy) and the increase in Rab11fip5.

3.5. Br‐EVs Modulate Cytoplasmic and Cytoskeletal Dynamics in Murine Cerebral Microvessels

The normal cerebral vasculature is characterized by high selectivity and low permeability and transcytosis rates, thereby ensuring proper brain physiology (Busatto et al. 2021; Morad and Moses 2019; Morad et al. 2019; Morad et al. 2020; Godinho‐Pereira et al. 2021; Thomsen, Routhe, and Moos 2017). Our group has demonstrated that Br‐EVs breach the intact BBB in vivo and significantly increase the incidence of breast‐to‐brain metastasis, suggesting a dysfunction of the cerebral microvasculature (Busatto et al. 2021; Morad and Moses 2019; Morad et al. 2019; Morad et al. 2020). To elucidate the mechanisms by which Br‐EVs induce cerebrovascular dysfunction and promote breast‐to‐brain metastasis, we utilized a quantitative global proteomics approach. This method allowed us to identify changes in protein expression and to identify and assess biological functions in murine cerebral microvessels modulated by Br‐EVs in a breast‐to‐brain metastasis context (Rappsilber, Mann, and Ishihama 2007; Schulte, Hasturk, and Hardt 2019; Jedinak, Loughlin, and Moses 2018). We employed label‐free mass spectrometry‐based quantitative proteomics to analyse the protein content of cerebral microvessels from the brains of mice treated with PBS, P‐EVs, or Br‐EVs. A total of 3030, 3213, and 2908 proteins were detected in the microvessel samples isolated from PBS, P‐EVs, and Br‐EV‐treated mice, respectively (Figure S2d–f). Seven hundred ninety‐five proteins were shared among all groups (Figure S2g). The comparative heatmap (Figure 6a) and the principal component analysis (PCA) (Figure S2h) illustrate how PBS, P‐EVs, and Br‐EV sample datasets correlate. The negative correlation values in the comparison between Br‐EV and PBS samples (Figure 6a), along with the distinct clustering highlighted by the PCA (Figure S2h), indicate that Br‐EVs cause significant changes in cerebral microvessel protein expression profiles. In contrast, the comparison between P‐EV and PBS samples results in correlation values that, in most cases, are equal to or close to zero, indicating a less pronounced variation in protein expression at the BBB (Figure 6a). We identified 163 proteins that were differentially expressed, both at a significant level and with high confidence, in the cerebral microvessels of mice treated with Br‐EVs compared to PBS. Additionally, 85 proteins were differentially expressed in cerebral microvessels from mice treated with P‐EVs compared to the PBS group (Figure 6b and Figure S6). Compared to the PBS group, Br‐EVs caused a significant increase in 35 proteins representing several functional categories relevant to the context of cancer‐derived EVs breaching the cerebral microvasculature and compromising BBB function and homeostasis (Morad et al. 2019; Morad et al. 2020) (Figure 6b,c). These include, among others, vesicle endocytosis and trafficking, tight junction formation, cell volume, and motility (Table 1). Similarly, Br‐EVs significantly downregulated 129 proteins (Figure 6b,d) representative of other functional categories such as, but not limited to, clathrin‐based endocytosis, vesicle sorting and trafficking, exocytosis, ER maturation, ER‐Golgi transport, and lysosomal function (Table 1). The proteins differentially expressed in Br‐EV‐treated samples compared to PBS controls predict the decreased or increased activation of 23 cellular functions and diseases. Among the top‐downregulated cellular functions are decreased cytoplasm and cytoskeleton organization, microtubule dynamics, and formation of cell protrusions (Figure 6e). These are associated with reduced cell adhesion and persistent, directional migration in polarized cells (Bouchet and Akhmanova 2017; de Forges, Bouissou, and Perez 2012; Kaverina and Straube 2011), therefore validating our in vitro ML results.

Among the proteins exclusively upregulated in murine cerebral microvessels from mice treated with Br‐EVs, was NKCC1. NKCC1 is a cation‐chloride cotransporter expressed in BECs, glia, and neurons (Loscher et al. 2013; Virtanen et al. 2020), co‐localizes with Rab11+ compartments during recycling (Loscher et al. 2013; Singh et al. 2015) and is known to regulate cell mass, volume, cytoskeletal organization (Das et al. 2018; Virtanen et al. 2020), and BBB integrity (Gong et al. 2021; J. Zhang et al. 2017). NKCC1 upregulation in BECs treated with Br‐EVs compared with PBS was further validated in vitro by immunoblot, while P‐EVs did not cause statistically significant results (Figure 6f,g). Taken together, the upregulation of NKCC1 and the other Rab11fip protein modulations are consistent with our conclusion that Br‐EVs affect the cytoplasmic and cytoskeletal dynamics of BECs. Importantly, NKCC1 inhibitors are already in use in the clinic to treat neurological symptoms of Down syndrome and other brain disorders (Lam et al. 2023; Loscher et al. 2013; Virtanen et al. 2020; Singh et al. 2015; Ben‐Ari 2017; Borgogno et al. 2021; Savardi et al. 2021) and may represent a novel approach to restoring the integrity of the BBB, which in turn could prevent the formation of the brain pre‐metastatic niche and the subsequent onset of brain metastasis.

3.6. Brains of Mice Treated With Br‐EVs Show Increased CD31 Expression

Early metastatic lesions in the brain, typically smaller than 0.25 mm in diameter, are viable within 100 µm away from blood vessels and exhibit a characteristic perivascular growth pattern, surrounded by an intact BBB (Kramer et al. 2014). Immunohistochemical and morphometric analyses showed that experimental brain metastases contained fewer dilated blood vessels composed of CD31‐overexpressing BECs compared to adjacent tumour‐free brain tissue (Kramer et al. 2014). CD31 is crucial for maintaining BBB integrity, as changes in its expression or function can influence BBB permeability and facilitate metastatic cell migration into brain tissue (Fidler et al. 2002; Y. Y. Zhang et al. 2018; Z. Zhang et al. 2023; Lertkiatmongkol et al. 2016; Eslami, Cortes‐Hernandez, and Alix‐Panabieres 2020) by playing a direct role in tumour cell adhesion, growth, and survival (Y. Y. Zhang et al. 2018; Z. Zhang et al. 2023). Our findings indicate that Br‐EVs can modulate the endocytic and cytoskeletal dynamics of BECs (Figures 1, 2, 3, 4, 5 ) and promote the adhesion of TNBC cells to BECs compared to controls (Figure 7a). We hypothesized that Br‐EV‐induced changes in the cerebral microvasculature could lead to variations in brain microenvironment structure and composition. To test this hypothesis, whole brain sections of mice treated with Br‐EVs, P‐EVs, and PBS were stained for microtubule‐associated protein (MAP)2, CD31, glial fibrillary acidic protein (GFAP), CD45, and CD11b to detect mature neurons, BECs, astrocytes, and microglia or peripheral infiltrating macrophages, respectively. Notably, compared to mice administered with PBS and P‐EVs, the brain sections of mice treated with Br‐EVs showed areas with increased CD31 staining, potentially suggesting priming of the cerebral vasculature to future cancer cell adhesion and transendothelial migration (Figure 7b and Figures S7–S9). Interestingly, the cerebral microvessel samples isolated from the brains of mice treated with both P‐EVs and Br‐EVs showed a CD31 immunofluorescence signal significantly higher than that of PBS controls (Figure 7c,d). These preliminary results may suggest that P‐EVs also have the potential to increase CD31 protein expression in brain microvessels despite resulting in different macroscopic expression patterns compared to Br‐EV treatments. While beyond the scope of the present study, further investigation into the correlation between TNBC‐EV treatment, BBB CD31 expression patterns, and the incidence and spatial distribution of brain metastasis merits further investigation. No significant changes were detected in the expression patterns of neuronal, astrocyte, and glial biomarkers nor in the whole brain signals for CD31, GFAP, and MAP2 (Figures S10–S16). These data suggest that Br‐EVs may specifically prime areas of the brain microvascular endothelium, which could facilitate the transendothelial migration of circulating tumour cells into the brain parenchyma.