Animals
All experiments were conducted under the UK Home Office Animals (Scientific Procedures) Act 1986, in agreement with local ethical and veterinary approval (Biomedical Research Resources, University of Edinburgh), and authorized by Project Licenses PP8310627 and PP2636156. Most experiments were performed using C57Bl/6J mice (Charles River Laboratories, UK) with male and female mice in equal proportion. For brain endothelial cell RNA-seq experiments Cdh5-Cre/ERT2 (Tg(Cdh5-cre/ERT2)1Rha)64 mice were bred with B6N.129-Rpl22tm1.1Psam/J (RRID: IMSR_JAX: 011029)63. Mice were housed in techniplast GM 500 cages with a rodent roll, dome home, chew stick and handling tube, in a 12-h light/dark cycle with food and water ad libitum, at 20–24 degrees Celsius and with humidity between 45-65%. Mice used were between 12 and 22 weeks of age.
Drug administration
Lipopolysaccharide (LPS) (Sigma Cat #L5024) from Escherichia coli O127:B8 was dissolved in sterile PBS at a concentration of 1 mg/ml and stored in aliquots at −80 °C. LPS-treated mice received one single intraperitoneal (i.p.) injection of 5 mg LPS/kg body weight. Control mice received an equivalent volume of sterile phosphate buffer saline (PBS). All mice used for proteomics in this manuscript received either LPS or PBS. Mice were weighed before and 24 h after the injection. For inducible gene expression in CRE-dependent mouse lines, tamoxifen (Sigma Cat #T5648) was dissolved in sunflower oil at a concentration of 20 mg/ml and was administrated at 100 mg tamoxifen/kg body weight by oral gavage for 5 consecutive days.
Generation of adeno-associated viruses
The pUCmini-iCAP-PHP.eb plasmid used for the adeno-associated AAV capsids was kindly shared by Dr. Viviana Gradinaru (Addgene Cat# 103005). The pZac2.1-GfaABC1D-tdTomato plasmid used as the control AAV was kindly shared by Dr. Baljit S. Khakh (Addgene Cat# 44332). pZac2.1-GfaABC1D-GFP was generated by excising GFP with BamHI from pZac2.1-GfaABC1D-AQP4-GFP, a gift from Dr. Baljit S. Khakh, and amplifying with the following primers before inserting into the BamHI site of the pZac2.1-GfaABC1D vector using the In-fusion® HD Cloning kit (Takara Cat# 639650): Forward: 5’ CCTCGAGCTCGGATCCGCCACCATGGTGAGCAAGGGCGAGGAG 3’; Reverse: 5’ TAAGCGAATTGGATCCTTACTTGTACAGCTCGTCCAT 3’. CMV-V5-TurboID-NES-pCDNA3 was kindly shared by Dr. Alice Ting (Addgene Cat# 107169) and we cloned the cDNA of the TurboID enzyme gene into a pZac2.1 plasmid containing the astrocyte specific promoter GfaABC1D. To do so, the sequence containing the V5 Tag-TurboID-HA Tag-nuclear export signal (NES) was amplified by PCR using the following primers: Forward: 5’ CCTCGAGCTCGGATCCATGGGCAAGCCCATCCCCAA 3’; Reverse: 5’ TAAGCGAATTGGATCCTTAGTCCAGGGTCAGGCGCTCCAGGGG 3’. The vector GfaABC1D-BioID2-13xLinker-BioID2, kindly shared by Dr. Baljit S. Khakh, was cut with the restriction enzyme BamHI to remove the BioID2 gene. The V5-TurboID-NES-HA fragment was then inserted into the BamHI pZac2.1-GfaABC1D vector using the In-fusion® HD Cloning kit (Takara Cat# 639650). Plasmids were confirmed by sequencing (Source Bioscience).
In preparation for AAV packaging, HEK 293T cells ((ATCC)293T/17 CRL-11268) were cultured at 37 °C, 5% CO2 in Dulbecco’s Modified Eagle Medium (DMEM, Gibco Cat# 11965092) with 10% heat-inactivated fetal bovine serum (FBS; Gibco Cat# A5256801). Cells were cultured in 145 mm2 plates (Greiner Bio Cat# 639160) at 80% confluence on the day of transfection. Plasmids pZac2.1-GfaABC1D-TurboID or pZac2.1-GfaABC1D-tdTomato were co-transfected into HEK 293T cells with the pALD-X80, AAV helper Plasmid (Aldevron, Cat# 5017-10), and pUCmini-iCAP-PHP.eb packaging plasmid at the molecular ratio of 1:3:1. Viral particles were harvested at 72 h post-transfection and purified by POROS™ CaptureSelect™ AAVX Affinity Resin (Thermo Fisher, Cat# A36741). The purified rAAVs were titered by ddPCR using the ITR primers (BioRad Cat# QX200, Forward primer: 5’ GGAACCCCTAGTGATGGAGTT 3’; Reverse primer: CGGCCTCAGTGAGCGA). The viral vectors were aliquoted and stored at −80 °C until use.
AAV administration and in vivo TurboID protein biotinylation
Mice received a single intravenous injection in the tail vein of either PHP.eb-GfaABC1D-V5-TurboID-HA-WPRE or PHP.eb-GfaABC1D-tdTomato (1.5 ×1011 genome copies) and given at least three weeks for the virus to express. Biotin (Sigma Cat# B4501) was dissolved in PBS at 5 mM, and mice were given a series of three biotin subcutaneous injections (24 mg biotin/kg body weight each) at 24, 18, and 3 h prior to dissection. The first biotin injection (24 h before dissection) occurred immediately following PBS/LPS treatment.
For assessment of FZD7 signal within endfeet, mice received PHP.eb-GfaABC1D-GFP AAVs (1.5 ×1011 genome copies) to tag astrocytes with the fluorescent reporter GFP and given at least three weeks for the virus to express.
Immunofluorescence
For mouse brain immunofluorescence studies, mice were given a terminal dose of anesthesia (Dolethal, 200 mg/mL, 0.1 mL/animal) and transcardially perfused with PBS containing 10 units/mL of heparin followed by 15 mL 10% formalin. Brains were dissected, stored in 10% formalin at 4 °C overnight, and moved to 30% sucrose until sectioning. Coronal or sagittal sections (40 µm each) were cut using a cryostat (Leica) and kept in antifreezing solution (0.05 M PBS, 250 mM sucrose, 7 mM MgCl2 and 50% glycerol) at −20 °C.
Mouse brain sections were washed twice in PBS for 10 min and incubated at room temperature for 2 h in blocking solution (10% normal goat (Stratech, Cat# 005-000-121-JIR) or donkey (Stratech, Cat# 017-000-121-JIR) serum in PBS with 0.2% Triton X-100 (Sigma, Cat# 102533092) with agitation. Sections were subsequently incubated overnight in primary antibodies diluted in the blocking solution at 4 °C. The following primary antibodies were used: rabbit anti-S100β (1:1000, Abcam Cat# ab41548), mouse anti-S100β (1:1000, Sigma Cat# S2532), mouse anti-NEUN (1:500, Sigma Cat# MAB377), guinea pig anti-S100β (1:500, Synaptic Systems Cat# 287004), rabbit anti-NEUN (1:2000, Cell Signaling Cat# 12943S), rabbit anti-RFP (1:1000, Rockland Cat# 600-401-379), goat anti-IBA1 (1:1000, Abcam Cat# ab5076), rabbit anti-IBA1 (1:1000, Wako Cat# 019-19741), rabbit anti-aquaporin4 (1:1000, Millipore Cat# ab3594), mouse anti-β-dystroglycan (1:1000, DSHB Cat# 7D11), rabbit anti-HA (1:500, Cell Signaling, Cat# C29F4), rat anti-PECAM1 (1:50, BD Biosciences Cat# 550274), goat anti-CD13 (1:100, R&D Cat# AF2335), mouse anti-CC1 (1:300, Sigma Cat# OP80), rabbit anti-FZD7 (1:500, Abcam Cat# ab64636), rabbit anti-WNT10B (1:200, Abcam Cat# ab70816), biotinylated lectin (1:500, Vector Cat# B-1175-1), and rabbit anti-PYGB (1:500, Atlas antibodies Cat# HPA031067). After overnight incubation at 4 °C with primary antibodies, sections were washed three times in PBS for 10 min each and incubated with the secondary antibody diluted in blocking solution. The following secondary antibodies were used at a concentration of 1:1000 unless otherwise indicated: goat anti-mouse IgG Alexa 488 (Invitrogen Cat# A11001), goat anti-rabbit IgG Alexa 488 (Invitrogen Cat# A11008), donkey anti-mouse IgG Alexa 488 (Invitrogen Cat# A21202), donkey anti-rabbit IgG Alexa 488 (Invitrogen Cat# A21206), goat anti-rabbit IgG Alexa 546 (Invitrogen Cat# A11010), donkey anti-rabbit IgG Alexa 546 (Invitrogen, Cat# A10040), donkey anti-rat IgG Cy3 (Jackson Immuno Research Cat# 712-165-150), goat anti-mouse IgG Alexa 647 (Invitrogen Cat# A21235), goat anti-rabbit IgG Alexa 647 (Invitrogen Cat# A21244), goat anti-rat IgG Alexa 647 (Invitrogen Cat# A21247), donkey anti-goat IgG Alexa 647 (Invitrogen Cat# A21447), and conjugated streptavidin Alexa 405 (Invitrogen, Cat# S32351). For rabbit anti-WNT10B staining, antigen retrieval was performed by incubating tissue in methanol at −20 °C for 30 min. To visualize biotinylated proteins, brain sections were incubated with streptavidin Alexa 488 (1:200, Invitrogen Cat# S32354). Nuclei were identified with either Hoechst 33342 (ThermoFisher, Cat# H3570) or DAPI (Sigma, Cat# D9542). The sections were mounted on microscope slides in ProLong Gold antifade reagent (Invitrogen, Cat# P36930) and coverslipped (170 ± 5 µm, Marienfeld, Cat# 0107222).
For immunofluorescence of expanded tissue, CUBIC-X expansion microscopy and tissue clearing68 was performed on 200 µm thick brain slices from wild-type mice that had received PHP.eb-GfaABC1D-gGFP AAVs to tag astrocytes with the fluorescent reporter eGFP. CUBIC-X reagents were provided by TCI Chemicals and the ‘CUBIC – Animal Tissue-Clearing Reagents – Technical Guidebook’ available on the TCI Chemicals website was used to develop our protocol. Free-floating sections were initially washed with 1X PBS (3 × 10 min each) to remove anti-freezing media. CUBIC-L (TCI Chemicals, Cat# T3740) was diluted 1:1 with distilled water to prepare a 50% working solution. The slices were immersed in 50% CUBIC-L solution and incubated at 37 °C for 12 h with gentle agitation. After 12 h, the CUBIC-L solution was removed and the slices were washed with 0.5% Triton-X diluted in 1X PBS at room temperature with gentle agitation 3 × 10 minutes each, followed by immersion in CUBIC-X1 solution (TCI Chemicals, Cat# T3866) and incubated at 4 °C for 24 h with gentle agitation. Our normal immunofluorescence protocol was used for protein detection, and the following primary and secondary antibodies were used: primaries chicken anti-GFP (1:1000, Abcam, Cat# ab13970), rabbit anti-FZD7 (1:500, Abcam, Cat# ab64636) and biotinylated-lectin (1:200, Vector, Cat# B-1175-1); secondaries goat anti-Rabbit Alexa 647 (1:1000, Invitrogen, Cat# A21244), goat anti-chicken Alexa 488 (1:1000, Invitrogen, Cat# A11039) and Streptavidin Alexa 405 (1:200, Invitrogen, Cat#S32351). After immunostaining, the slices were immersed in CUBIC-X2 solution (TCI Chemicals, Cat# T3867) and incubated for 24 h at room temperature with gentle agitation. The slices were then mounted onto SuperFrost Plus glass slides with parafilm chambers to confine the tissue without compression (mounting medium TCI Chemicals, Cat# M3292).
For immunofluorescence of isolated microvessels, the area containing the vessels was outlined with a PAP pen (Abcam, ab2601) to create a hydrophobic barrier, and all washes and incubation steps were performed directly on the slide. Mouse and human isolated vessels were visualized by incubating with fluorescein-labeled lectin (1:200 for mouse vessels, 1:1000 for human vessels, Vector Cat# 1171), and/or biotinylated lectin (1:200, Vector Cat#B-1175), the endfeet with rabbit anti-aquaporin4 (1:1000, Millipore Cat# ab3594) and smooth muscle cells with mouse anti-α-Smooth Muscle Actin (ACTA2) (1: 200, Sigma-Aldrich Cat#A5228). For FZD7 staining, antigen retrieval was performed by incubating the vessels in 0.01 M sodium citrate, pH = 6, at 97.5 °C for 20 min. Mouse isolated vessels were visualized by incubating with fluorescein-labeled lectin 1:200 (Vector Cat# 1171) or biotinylated lectin (Vector Cat# B-1175-1), the endfeet with mouse anti-aquaporin4 (1:100, Abcam, Cat# ab9512), and rabbit anti-Frizzled7 (1:250, Proteintech, 16974-1-AP). The following secondary antibodies were used at a concentration of 1:1000: goat anti-mouse 546 (Invitrogen, Cat# A11003), goat anti-rabbit IgG Alexa 647 (Invitrogen, Cat# A21244), and Streptavidin Alexa 405 conjugate (1:200, Invitrogen, #S32351) to visualize biotinylated lectin. The sections were mounted on microscope slides in ProLong Gold antifade reagent (Invitrogen, Cat# P36930) and coverslipped (170 ± 5 µm, Marienfeld, Cat# 0107222).
Human tissue for vessel purification and staining was sourced from Edinburgh Brain and Tissue Bank (Table 1). Informed consent was either obtained from participants in life and/or from the nearest relative for deceased participants, in keeping with legal requirements. Sex information is based on clinical data.
Table 1 Tissue used for immunostaining
For immunofluorescence studies in human brain slices, formalin-fixed paraffin-embedded (FFPE) frontal cortex (BA46) human brain sections (8 µm) were obtained from the Edinburgh Brain Bank, a Medical Research Council funded facility with research ethics committee (REC) under the ethical approval (21/ES/0087). For immunofluorescence labeling, sections were first deparaffinized with two 5 min xylene washes and rehydrated in consecutive 5 min washes in 99.8% and 70% ethanol. Sections were rinsed in distilled water for 5 min prior to performing antigen retrieval in 0.01 M sodium citrate (pH 6.0) at 97.5 °C in a water bath for 30 min. Sections were cooled in PBS for 10 min. A hydrophobic border was applied using a PAP pen (Merck, Cat # Z672548) and sections were incubated for 1 h in blocking solution (5% normal donkey serum (Stratech, Cat# 017-000-121-JIR) in PBS with 0.3% Triton X-100 (Sigma, Cat# 102533092)). Subsequently, sections were incubated with primary antibodies diluted in antibody diluent solution (1% normal donkey serum in PBS with 0.3% TritonX-100). The following primary antibodies were used: rabbit anti-FZD7 (1:200, Abcam Cat# ab64636), rabbit anti-WNT10B (1:200, Abcam Cat# ab70816), rabbit anti-PYGB (1:200, Atlas antibodies Cat# HPA031067), mouse anti-ALDH1L1 (1:200, NeuroMab, Cat# 75-140), and goat anti-VE-cadherin (1:200, R&D Systems, Cat# AF938). Sections were washed three times in PBS for 10 min each and incubated with fluorophore-conjugated secondary antibodies in antibody diluent solution for 1 h at room temperature in the dark. The following secondary antibodies were used: donkey anti-rabbit IgG Alexa 488 (1:250, Invitrogen Cat# A21206), donkey anti-mouse IgG Alexa 546 (1:250, Invitrogen, Cat# A10036), donkey anti-goat IgG Alexa 647 (1:250, Invitrogen Cat# A21447). Following three washes in PBS, tissue autofluorescence was quenched by incubating sections in TrueBlack Lipofuscin Autofluorescence Quencher (1:20 in 70% ethanol, Biotium, Cat# 23007) for 30 s followed by five washes in PBS of 5 min each with gentle shaking. Nuclei were stained using DAPI (Sigma, Cat# D9542). Sections were coverslipped (170 ± 5 µm, Marienfeld) using ProLong Gold antifade reagent.
RNAscope in situ hybridization
RNAscope (Advanced Cell Diagnostics; ACD) in situ hybridization was performed on 40 µM cortical mouse brain tissue sections. Free-floating sections were initially washed three times in 1X PBS for 10 min each to remove anti-freezing media and mounted onto SuperFrost Plus glass slides. Slides were left to air dry for at least 3 h at room temperature, then baked at 60 °C for 30 min, followed by incubation with pre-chilled 10% formalin diluted in 1X PBS for 25 min at 4 °C. RNAscope Multiplex Fluorescent V2 Assay (ACD Cat# 323100) was used to label multiplex probes. The assays were performed according to the manufacturer’s instructions and user manual for fixed-frozen tissue, including slightly adjusting and following steps of tissue dehydration, pre-treatment with hydrogen peroxide and 1X Target Retrieval. This pre-treatment method was optimized for recognition of probes and combined immunofluorescence antibodies for both RNA and protein detection. For the 1X Target Retrieval steps, the assay was optimized by initially pre-heating the 1X Target Retrieval solution to 97.5 °C for 15 min before use. After hydrogen peroxide incubation (following the user manual), the slides were then immersed into the pre-heated 1X Target Retrieval solution for 15 min only and immediately washed with distilled water, followed by immersion into 100% ethanol for 5 min. The slides were then dried at 60 °C for 10 min and left overnight at room temperature prior to protease treatment. RNAscope Protease III solution was used for protease treatment, following the user manual, and adjusting the incubation step to 15 min at 40 °C to limit over-digestion. RNAscope probe Mm-Wnt10b (ACD Cat# 401071) paired with TSA Vivid 650 (1:1500) fluorophores were used in this study. After developing with HRP and prior to the addition of blocking buffer following our immunofluorescence protocol, the slides were washed with PBS for 10 min. It is important to note that the slides were not left dry at any point during the immunofluorescence steps. Immunofluorescence was performed as described above, and the following primary and secondary antibodies were used: primary antibody rabbit anti-ERG (1:500, abcam, Cat# ab92513) and biotinylated-lectin (1:500, Vector, Cat# B-1175-1); secondary antibody goat anti-rabbit 488 (1:1000, Invitrogen, Cat# A11008) and streptavidin Alexa 405 (1:1000, Invitrogen, Cat# S32351).
Microscopy and image analysis
Fluorescent images were taken on a Zeiss LSM900 confocal laser-scanning microscope.
To examine the distribution of expression of astrocyte-TurboID and astrocyte-tdTomato in the mouse brain (Supplementary Fig. 1a, f), a human-guided automatic epifluorescence tile scan was carried out. For the rest of images, confocal microscopy was used.
For AAV cell-specificity validation (Supplementary Fig. 1d, e, g, h), 3 images (1024 ×1024 pixels) were taken in the prefrontal cortex (PFC) area using a 25 × 0.8 NA objective (Plan-Apochromat, Zeiss Cat# 420852-9871-000). Image analysis was conducted using FIJI (ImageJ). Three images were taken in the PFC area for each cell marker (S100β, NeuN, Iba1, CC1). Z-stacks of 30 µm (30 slices, 1 µm step size) were taken. For each image, the total number of cells expressing the cell-specific marker, and the number of cells co-stained with streptavidin, was counted manually. The percent co-localization for each image was calculated by dividing the total number of double-labeled cells (i.e., labeled with streptavidin and the cell-specific marker) by the total number of cells expressing cell specific markers.
For streptavidin intensity measurements (Supplementary Fig. 1b, c), the laser power and gain were kept consistent for all images. To calculate the streptavidin signal intensity as a percentage of total area, a minimum threshold for streptavidin signal was calculated based on the background streptavidin signal present in brains expressing the tdTomato control AAV. Then, the signal intensity was captured through the “measurements” tool within FIJI. The average minimum threshold calculated was applied to all images from brains expressing TurboID AAV, and the area of streptavidin signal intensity above the threshold was captured and recorded.
To measure endfoot-vasculature distance (Fig. 1e), a 20 µm line was drawn from the center of the vessel in 63× (oil objective, 1.2 NA, Plan-Apochromat, Zeiss Cat# 420882-9870-799) confocal single planes and the intensity profile was plotted. Intensity values for each channel were exported from FIJI and analyzed separately with Clampfit (Molecular Devices) to determine the position of the highest intensity peak for each marker. This data was then used to calculate the distance between markers.
The proportion of different sizes of vessels after vessel purification (Fig. 1j) was quantified using FIJI. A ROI for all vessels in the image was generated using the lectin channel and the percent area was recorded. Next, vessels with a diameter less than 10 µm were selected and deleted from the image and the percent area was recorded again. This process was subsequently repeated for vessels less than 30 µm and greater than 30 µm. The vessel composition was calculated by subtracting the percent areas for each vessel size from the full lectin ROI in the order they were acquired and normalizing each area by the total percent area of the identified vessels.
To examine the aquaporin-4 coverage in PBS and LPS isolated vessel samples (Fig. 1j–n; Supplementary Fig. 7d–j), 1-7 images of each vessel type (i.e., artery, arteriole, capillary, vein and venule) were taken in each sample using a 25× objective. α-SMA signal was observed in all vessels >8–10 μm in diameter53. Thus, vessel type was determined by the combination of vessel diameter and α-SMA morphology (i.e., arteries and arterioles: 30–90 μm and 8–30 μm diameter respectively, with cross-vessel striated α-SMA signal; veins and venules: 30–90 μm and 8–30 μm diameter respectively, with irregular α-SMA signal along the vessel). Using FIJI, Z-projections of the lectin and aquaporin-4 channels were generated and a ROIs representing the vessel were created thresholding the lectin signal. AQP4 percentage area of colocalization was then measured within the vessel ROIs. In bigger vessels containing more parts of the vascular tree, a ROI of the contour of the vessel was created prior to measuring AQP4 coverage.
The proportion of astrocyte endfeet obtained from isolated different sized vessels (Fig. 1n) was calculated by first averaging the percent AQP4 coverage for each vessel size category (i.e., for 10–30 µm, the percent AQP4 coverage for arterioles and venules and for >30 µm, the percent AQP4 coverage for arteries and veins were averaged together). The average percent AQP4 area was then multiplied by the normalized proportion of each vessel size (in Fig. 1j) and divided by the total to obtain the proportion of endfeet coming from each vessel category in our sample.
For PYGB analyses (Supplementary Fig. 5), a β-dystroglycan and lectin signal intensity threshold were applied to generate two ROIs representing endfeet and vessels respectively in mouse, and ALDH1L1 or VE-cadherin to create ROIs for astrocyte endfeet or vessels in human samples. Then the PYGB percent area of colocalization was measured within the endfoot or vessel ROIs.
For FZD7 analysis in expanded tissue (Fig. 5a, b), CUBIC-X sections were imaged with a 63× oil objective using Airyscan mode and processing (Zeiss) on a Zeiss LSM900 confocal laser-scanning microscope. For image analysis with FIJI, a GFP or lectin signal intensity threshold was applied to generate a region of interest representing either the endfoot or vessel, respectively. FZD7 percentage area of colocalization was then measured within the endfoot or vessel ROIs.
To measure FZD7, AQP4, and lectin intensity profile and distance in isolated vessels (Fig. 5c–i) a 15 µm line was drawn across the vessel in 63× (oil objective, 1.2 NA, Plan-Apochromat, Zeiss Cat# 420882-9870-799) confocal single planes and the intensity profile was plotted as shown by representative examples in Fig. 5d, e. Intensity values and their respective distance across the drawn line for each channel were exported from FIJI and analyzed separately with Clampfit (Molecular Devices). The position in µm of the highest intensity peak for each marker was used to calculate the distance of FZD7 signal from AQP4 and lectin as well as the distance of AQP4 from lectin as shown in Fig. 5f and plotted in Fig. 5g. In addition, for each marker the fluorescence intensity ratio was determined as the difference between the highest intensity peak and the baseline intensity in the middle of the vessel, as shown in Fig. 5h and plotted in Fig. 5i. Both, the distance between markers and the intensity ratio values that are shown in the figures are the result of the average value between the peaks of each side of a single the vessel.
For RNAscope image analysis (Supplementary Fig. 8), sections were imaged with a 63× objective using Airyscan mode on a Zeiss LSM900 confocal laser-scanning microscope and analyzed using FIJI. A lectin signal intensity threshold was applied to generate a region of interest representing whole astrocytes or vessels, respectively. Probe signal of Wnt10b within these respective ROIs was analyzed by recording the percent area of Wnt10b signal within the lectin signal, respectively. An additional threshold was applied to the Wnt10b signal to allow for the manual counting of each RNA dot found within the lectin ROIs.
For WNT10B and FZD7 immunohistochemistry quantification analyses in mice (Fig. 6), vessels <10 µm in diameter were imaged with a 63× objective using the Airyscan mode and processing (Zeiss) and analyzed using FIJI. For FZD7, a β-dystroglycan signal intensity threshold was applied to generate a region of interest (ROI) representing endfeet. FZD7 signal within this ROI was analyzed by recording the percent area and mean-intensity of FZD7 signal within the β-dystroglycan signal. In addition, we generated a histogram representing how many FZD7 positive pixels were found at each signal intensity value within the endfoot (defined by the β-dystroglycan signal) and recorded the intensity at which most of the FZD7 positive pixels are found (peak intensity). A similar process was followed for WNT10B, but due to WNT10B being a ligand that could be released by the BECs, the WNT10B signal was quantified within a ROI that contained both the astrocyte endfoot and the vessel by using β-dystroglycan staining to define the endfoot boundary.
For FZD7 and WNT10B analyses in postmortem human brain (Supplementary Fig. 9), fluorescent images were acquired with a 63× objective and analyzed using FIJI. To determine the percent area labeled by FZD7, an ALDH1L1 signal intensity threshold was applied to generate a ROI for astrocyte endfeet. To measure WNT10B, a vasculature ROI that encompassed both the vessel and the surrounding endfeet was first generated. The ROI was defined by VE-cadherin staining, which was expanded outward by thirty pixels to approximate the endfoot ALDH1L1 immunopositive boundary. Within this vessel ROI, the percent area labeled by WNT10B was calculated.
For analysis of human isolated vessels (Fig. 7), the overlap between aquaporin-4 (AQP4) and human isolated vessels of <10 µm in diameter was imaged with a 63× objective and analyzed using FIJI. A region of interest (ROI) representing the blood vessel was generated by applying a threshold to lectin signal intensity, and the overlap of AQP4 signal with the lectin ROI was recorded as percent area.
Mouse brain microvasculature isolation
Isolation of brain microvessels was conducted by adapting a published protocol53. Briefly, mice were culled by cervical dislocation, and the brain was quickly removed and dissected in cold PBS. Isolated cortices were snap-frozen in liquid nitrogen and stored at −80 °C until microvessel isolation. For each mouse, the cortex was divided into two hemispheres: one hemisphere was used for microvessel isolation and endfoot proteomics, and the other hemisphere was used for whole astrocyte proteomics.
For isolation of brain microvessels, one cortex hemisphere was homogenized in 2 mL of MCDB-131 media (Gibco Cat# 10372019) with Halt protease inhibitor (ThermoFisher Cat# 78430) using a dounce tissue grinder (Kimble Cat# 885303-002). The homogenate was centrifuged (2000 × g/5 min/4 °C), the pellet was resuspended in 15% dextran in DPBS (Gibco Cat# 14190144) and centrifuged again (10,000 × g/15 min/4 °C) to yield a pellet containing blood vessels. The pellet was resuspended in DPBS, transferred to a 40 μm cell strainer, and washed with 10 mL of DPBS. From this step, the microvessels were either processed for immunofluorescence or proteomics.
To deplete isolated vessels of endfeet, we adapted a published protocol101. The entire previously frozen cortex was homogenized in 3 mL of Buffer 1 (B1; 10 mM HEPES in HBSS) after being cut into 2 mm pieces with a scalpel and centrifuged at 600 × g/5 min/4 °C. The resulting pellet was resuspended in 5 ml of B1 containing 18.75 µg/ml of Liberase DL (Roche, Cat # 05401054001) together with 40 U/mL of DNaseI (Invitrogen, Cat# 18047019) and incubate for 15 min at 37 °C while gently mixed every 2 min. The digested samples were moved on ice and gently homogenated using a 10 ml stripette. The homogenate was then centrifuged at 2000 × g/5 min/4 °C and the supernatant resuspended in 2 mL of MCDB-131 media (Gibco Cat# 10372019) and processed for microvessel isolation as described above.
For immunofluorescence: The microvessels were fixed by placing the cell strainer in 10% formalin for 15 min with gentle shaking. Fixed microvessels were collected by reversing the cell strainer and applying 1% BSA in DPBS (Merckmillipore Cat# 126609) to the surface. The solution was centrifuged (2000 × g/10 min/4 °C), the pellet resuspended in DPBS, added dropwise to microscope slides and allowed to dry. Slides were stored at −80 °C until further use.
For proteomics: To collect microvessels, the cell strainer was reversed and 0.5% BSA in MCDB-131 was applied to the surface. Samples were centrifuged (4100 × g/15 min/4 °C) and the vessel-containing pellet resuspended in lysis buffer A (50 mM HEPES, 150 mM NaCl, 1 mM EDTA, 1X Halt protease inhibitor (ThermoFisher Cat# 78430)).
Purification of biotinylated proteins
Tissue (hemicortices for whole astrocyte protein purification or isolated microvessels for endfoot protein purification – see above) was homogenized using a dounce tissue grinder (Kimble Cat# 885303-002) in lysis buffer A (50 mM HEPES, 150 mM NaCl, 1 mM EDTA, 1X Halt protease inhibitor (ThermoFisher Cat# 78430)). An equivalent volume of lysis buffer B (50 mM HEPES, 150 mM NaCl, 1 mM EDTA, 0.4% SDS, 2% Triton X-100, 2% Na-deoxycholate) was added after 5 min. The samples were then sonicated (30 s on, 30 s off; for whole astrocyte – 60 cycles, for endfoot – 10 cycles at 4 °C) in a Bioruptor® Pico sonication device (Diagenode Cat# B01060010). Sonicated homogenates were centrifuged (15,000 × g/15 min/4 °C) and the supernatant ultracentrifuged (100,000 × g/45 min/4 °C; Beckman-Coulter Optima Max-XP). Cleared supernatant was supplemented with SDS to a final concentration of 1%. Samples were heated at 95 °C for 5 min. Streptavidin-coated magnetic beads (ThermoFisher Cat# 88817) were combined with the samples and incubated overnight at 4 °C with agitation. Bound beads were subsequentially washed with 2% SDS wash buffer (1% Triton X-100, 1% Na-deoxycholate, 25 mM LiCl), 1 M NaCl and 50 mM triethylammonium bicarbonate (Sigma Cat# T7408). All steps were performed on ice or at 4 °C. After the last wash, beads were resuspended in 50 mM triethylammonium bicarbonate. The samples were snap-frozen and stored at −80 °C until further use.
Samples from both Astrocyte-TurboID or Astrocyte-tdTomato injected animals underwent this process. We used Astrocyte-tdTomato as a negative control because it accounts for three potential issues – (1) binding of endogenously biotinylated proteins to the streptavidin beads, (2) potential additional endogenous protein biotinylation due to the biotin injected to the mice, (3) non-specific protein binding to the streptavidin beads.
On-bead protein digestion
Streptavidin beads were washed three times with ice-cold wash buffer (20 mM HEPES pH 7.5, 150 mM NaCl) and resuspended in 20 mM HEPES. Bead-bound proteins were reduced by adding 10 mM TCEP (Tris(2-carboxyethyl)phosphine hydrochloride; Sigma Cat# C4706) and incubated on a shaking thermomixer (Eppendorf; 1000 RPM/30 min/56 °C). To alkylate the proteins, 2 M urea (Thermo Scientific Cat# 140750010) and 20 mM iodoacetamide (Sigma Cat# I1149) were added and incubated on a thermomixer (1000 RPM/30 min/RT).
To digest the proteins, beads were incubated with 200 ng trypsin+Lys-C (Thermo Scientific A40007) on a thermomixer (1250 RPM/1 h/37 °C). The supernatant was transferred to a new tube and beads were resuspended in an equal volume of 50 mM triethylammonium bicarbonate. To both the supernatant and beads, an additional 200 ng trypsin+Lys-C was added and incubated on a thermomixer (1250 RPM/overnight/37 °C). Resulting supernatant and bead digests were combined before proceeding. The beads were centrifuged (2500 × g/2 min / RT), supernatant collected and trypsin digestion quenched by adding 1% trifluoroacetic acid (Fisher Scientific, A116) and centrifuging at max speed for 10 min.
Peptides were isolated using homemade stage tips. A single 16-gauge SDB-RPS disk (3 M Empore #2241; SDB-RPS, polystyrenedivinylbenzene-reversed phase sulfonate, 12 μm particle size, 47 mm, CDS analyticals) was isolated using a flat needle (Sigma, Cat# Z261378) and blown into a 250 µL pipette tip with compressed air. The stage tips were activated by adding wash buffer 1 (1% trifluoroacetic acid in 100% mass spectrometry-grade isopropanol) and centrifuging (1500 × g/5 min/RT). The peptides were twice added to the activated stage tip and centrifuged (1500 × g/5 min/RT) to ensure binding. The bound peptides were washed with wash buffer 1 (1% trifluoroacetic acid in 100% mass spectrometry-grade isopropanol) and wash buffer 2 (0.2% trifluoroacetic acid in 3% acetonitrile). The peptides were eluted by subsequent additions of elution buffer 1 (1.25% ammonium hydroxide in 50% acetonitrile) and elution buffer 2 (1.25% ammonium hydroxide in 80% acetonitrile). The eluted peptides were stored at −20 °C until further use. Prior to LC-MS/MS analysis, samples were evaporated to dryness (Savant SPD140DDA concentrator with a Savant RVT5105 refrigerated vapor trap). All used solutions were LC-MS/MS grade.
Human vessel isolation and protein digestion
As described above for mouse but with the following changes:
Unfixed, snap-frozen cortical human brain tissue (Table 2 and Supplementary Data 6) was sourced from Edinburgh Brain and Tissue Bank, a Medical Research Council funded facility with research ethics committee (REC) under the ethical approval (21/ES/0087). Informed consent was either obtained from participants in life and/or from the nearest relative for deceased participants, in keeping with legal requirements. Sex information is based on clinical data. 250–800 mg of tissue was used for vessel isolation. Three pieces from each sample (i.e. bulk tissue samples) were punched out with a 1.5 mm biopsy punch (Kai, Cat# BPP-15F) and placed directly into lysis buffer (50 mM HEPES pH 7.5, 150 mM NaCl, 1 mM EDTA, 0.2% SDS, 1% Triton X-100, 1% sodium deoxycholate, 1X Halt protease inhibitor (ThermoFisher, Cat# 78430)). Remaining tissue was used for vessel isolation. All samples were homogenized with stainless steel beads (Biospec, Cat# 11079132ss) in a bead homogenizer (Bertin, Cat# P000669-PR240-A). To obtain the microvessel pellet, vessel homogenate was centrifuged in MCDB-131 media (2000 × g/5 min/4 °C) and then in 15% dextran (4200 × g/45 min/4 °C).
Table 2 Tissue used for proteomics analysis
Human microvessels captured on a 40 μm cell strainer were lysed with the direct addition of lysis buffer. The resulting vessel lysate and bulk tissue homogenate were sonicated (30 s on, 30 s off for 10 cycles at 4 °C) in a Bioruptor® Pico sonication device (Diagenode, Cat# B01060010) and ultracentrifuged (100,000 × g/45 min/4 °C; Beckman-Coulter Optima Max-XP). Proteins were precipitated with the addition of 20% (v/v) trichloroacetic acid (Merck, Cat# 91228-100 G), washed three times with ice-cold acetone, dried and resuspended in DLT buffer (50 mM TEAB, 0.5% sodium deoxycholate, 12 mM sodium N-lauroylsarcosine). Samples were snap-frozen and stored at −80 °C until further use. Isolated protein content was quantified with BCA assay (ThermoScientific, Cat# 23225). All steps were performed on ice unless otherwise indicated.
Isolated proteins were reduced with the addition of 10 mM TCEP (Sigma Cat# C4706) and incubated on a thermomixer (1250 RPM/10 min/60 °C), then alkylated by adding 40 mM iodoacetamide (Sigma Cat# I1149) and incubated on a thermomixer (1250 RPM/40 min/RT). Finally, 5% (v/v) SDS, 1.2% (v/v) phosphoric acid and 6 volumes of S-TRAP buffer (90% (v/v) methanol in 100 mM triethylammonium bicarbonate) were added.
Proteins were bound to S-TRAP micro columns (Protifi, Cat# C02-micro), washed four times with S-TRAP buffer, topped with 0.5–1 μg trypsin/Lys-C (Thermo Scientific, A40007) and incubated overnight at 37 °C. Digested peptides were eluted with one wash with 50 mM TEAB, one wash with elution buffer I (0.2% (v/v) formic acid), and three washes with elution buffer II (0.2% (v/v) formic acid, 50% (v/v) acetonitrile). Digested peptides were snap-frozen and stored at −80 °C until further use. Prior to LC-MS/MS analysis, the solution was evaporated to dryness (Savant SPD140DDA concentrator with a Savant RVT5105 refrigerated vapor trap). All used solutions were LC-MS/MS grade.
LC-MS/MS
For mouse samples, peptides were resuspended in 0.1% formic acid prior to EvoTip (EvoSep, Odense, Denmark) preparation. EvoTips were prepared according to the manufacturer’s instructions. Briefly, EvoTips were washed with 0.1% formic acid in acetonitrile, conditioned with 1-propanol, and equilibrated with 0.1% formic acid. Samples were loaded onto the tip and washed 4 times with 0.1% formic acid. A preservation reservoir of 100 µL 0.1% formic acid was left in the tip to avoid drying out. 10% of the astrocyte and 50% of the endfoot peptide sample volume were injected into a Bruker timsTOF SCP mass spectrometer using an EvoSep One LC system, using the “30 samples per day” method with a 44-min gradient. The analytical column used was an “endurance” column (C18, 1.9 µm, 15 cm × 150 µm; EV1106). Data was acquired in the standard DIA PASEF template mode.
For human samples, peptides were analyzed on a Orbitrap Fusion Lumos Tribrid mass spectrometer interfaced with a 3000 RSLC nano liquid chromatography system. 1 µg of each sample was loaded on to a 2 cm trapping column (PepMap C18 100 A – 300 mm, Thermo Fisher Scientific Cat# 160454) at a 5 mL/min flow rate using a loading pump. Samples were analyzed on a 50 cm analytical column (EASY-Spray column, 50 cm 75 mm ID, Cat# ES803) at a 300 nL/min flow rate that is interfaced to the mass spectrometer using Easy nLC source and electrosprayed directly into the mass spectrometer. A linear gradient between solvent A (0.1% formic acid in water) and solvent B (0.1% formic acid in acetonitrile) was established as such: 3% to 30% of solvent B at a 300 nL/min flow rate for 105 min; 30% to 40% solvent B for 15 min; 35–99% solvent B for 5 min, which was maintained at 90% B for 10 min. The column was washed with 3% solvent B for another 10 min comprising a total 140 min run with a 120 min gradient in a data independent acquisition (DIA) mode. 45 isolated 350–1500 m/z windows were used. Collision induced dissociation (CID) was used for fragmentation of peptides.
Proteomics data analyses
Raw mass spectra files were analyzed using DIA-NN v.1.8.1102 in library free mode. For mouse samples, peptides were searched against a Uniprot mouse database (downloaded March 2022) supplemented with TurboID and tdTomato sequences. For human samples, peptides were searched against a Uniprot reference human proteome (downloaded March 2023). Both mouse and human peptides were also searched against a custom database of common contaminants based on the cRAP database (https://www.thegpm.org/crap/). The default parameters for double pass neural network classifier were used with the following exceptions: the mass accuracy and MS1 accuracy was set to 10.0, and “Heuristic protein inference” was unselected. Precursor false discovery rate was set to 0.1.
All proteomics data presented here originates from the “report.pg_matrix.tsv” output file from DIA-NN. All analysis was conducted using R (R Core Team, 2023), RStudio (RStudio Team, 2023), and the tidyverse package103 (version 2.0.0).
Ambiguous, species-mismatched, contaminant proteins, and proteins not mapped to genes were removed from analysis. For comparisons between mice expressing TurboID and tdTomato AAV, proteins found in at least 33% of the samples from either astrocyte or endfoot protein isolations regardless of AAV were kept for further analysis due to the inherent difference in isolated proteins between TurboID and tdTomato samples. For comparisons between LPS and PBS samples, which only received TurboID AAV and are therefore expected to yield a similar number of isolated proteins between conditions, proteins were kept for further analysis if they were found in at least 50% of both PBS and LPS samples. Human proteins were kept if they appeared in at least 2 samples from bulk or at least 2 samples from vessel protein isolations. Quantile normalization was performed using the preprocessCore package104 (version 1.66.0). Mouse astrocyte and endfoot datasets were normalized separately, human bulk and vessel datasets were normalized together. For mouse samples, differential expression analysis was conducted with the package limma105 (version 3.56.2). If a protein was not observed in a sample, its abundance was imputed with a value of 1. Differential expression was considered significant when p-value < 0.05. For human samples, differential expression analysis was conducted with the package DEP106 (version 1.26.0), no imputation was used, and differential expression was considered significant when p-value < 0.05. Gene ontology and pathway analyses were generated using Qiagen IPA58.
For Supplementary Fig. 2b, c, all peptides mapped to dystrophin (DMD) were extracted by filtering the DIANN output “report.tsv” for peptides found in the PBS endfoot samples. A total of 30 peptides were identified and the number of times each was recorded. The peptide sequences were manually mapped to the FASTA sequences for DMD and isoforms Dp71 and Dp40, which were obtained from NCBI. The FASTA sequence for mouse Dp140 is not annotated but is ~1200 amino acids (AA) long in humans and conserved through the C-terminus, so the same was assumed for this analysis. The AA positions of each peptide sequence was mapped onto the full-length DMD sequence and counted to generate the graphs.
For Supplementary Fig. 3, we compared our endfoot proteome to previously published datasets36,37,38,39. For each published dataset, we used the provided data with some additional filters. For Stokum et al., we filtered the data for endfoot enriched proteins (log(endfoot/cell body) > 0). For Kameyama et al., we used the proteins unique to the purified perivascular astrocyte endfeet (PV-AEF) and the proteins enriched in purified PV-AEF identified by the following filters: a) enriched in the crude PV-AEF fraction (CA/B0 > 0), b) enriched in the pure PV-AEF fraction (PA/B0 > 0), c) ANOVA q-value < 0.05, d) removed rows with multiple gene names separated by a semicolon. For Alonso-Gardon et al., no additional filtering was applied; we used Hepacam (HECAM), Mlc1 (MLC1), and all identified proteins. For Soto et al., we used the proteins found to be enriched and unique to AQP4-BioID2 versus AQP4-GFP (1) or astrocyte BioID2 (2) and removed proteins with no gene name. Plots were generated using the UpSetR package107. To directly compare the endfoot and astrocyte proteomes (PBS and TurboID AAV), a relative normalized abundance for each protein shared between the endfoot and astrocyte proteomes (1151 proteins) was generated using the “report.stats” output from DIANN. For each sample, the TotalQuantity value (reflecting the overall protein abundance per sample) was extracted, and divided by the highest value. This sample-specific normalization factor was used to scale the abundance of each shared protein (1151) protein in each sample. The resulting normalized values were used for differential expression analysis with Limma, as described above.
For Supplementary Fig. 4, transporters were identified by comparing the endfoot proteome to a database of SLCs (https://slc.bioparadigms.org/).
For Fig. 2i, receptors in endfeet were identified by comparing receptors from CellTalkDB61 (http://tcm.zju.edu.cn/celltalkdb/) with the endfoot proteome and manually curating the list to select only the ones with known downstream molecular cascades.
For Supplementary Figs. 2a, 3b, 6c, and 11d, cell type markers were included due to their widespread acceptance as cell type markers in immunohistochemical and/or western blot applications. Additional markers were identified in single-cell RNA-seq datasets of bulk brain tissue (UCSC Cell Browser v1.2.12108) and isolated vasculature73, and review manuscripts109,110.
Quantitative real time-polymerase chain reaction (qRT-PCR)
To check cell specificity of the Cdh5-CRE/RiboTag mouse line, purified mRNA from Input and IP samples was transcribed to cDNA using Superscript IV reverse transcriptase (ThermoFisher Cat# 18090010). Cell markers abundance was assessed by qRT-PCR using Fast SYBR Green (ThermoFisher Cat# 4385612) and previously published primers111. Gene expression was calculated relative to the housekeeping Rplp0 gene based on their Ct values using the formula: 2-ΔCt (ΔCt = Ct (gene of interest) – Ct (housekeeping gene).
Mouse whole tissue and BEC-specific RNA-seq and differential expression analysis
Mice were culled by cervical dislocation and the brain was quickly removed and cut into 1.5 mm coronal slices using a rodent brain matrix. To obtain enough RNA, each biological sample contained PFCs from two mice. The detailed protocol has been published111. In brief, brain slices were placed in cold RNAse-free PBS and PFC was further dissected and collected for RNA extraction. The dissected tissue was homogenized using a dounce tissue grinder (Kimble Cat# 885303-002) in 1 mL of homogenization buffer (47 mM Tris, 94 mM KCl, 11.3 mM MgCl2, 1%NP-40, 1 mM DTT, protease inhibitor (Sigma Cat# P8340), 200 U/mL of RNasin (Promega Cat# N2115), 100 μg/mL cycloheximide, 1 mg/mL heparin). The homogenate was centrifuged (10,000 × g/10’/4 °C) to obtain a clear lysate, 10% (100 μL) of which was used to extract whole tissue RNA (Input). The remaining lysate (IP) was incubated with mouse anti HA-antibody (1:200, Biolegend Cat# 901514) for 4 h at 4 °C in rotation, followed by addition of 200 μL of Pierce protein A/G magnetic beads (ThermoFisher Cat#88803) and incubation overnight at 4 °C in rotation. The next day, beads were washed three times using high salt solution (50 mM Tris, 300 mM KCl, 12 mM MgCl2, 1% NP-40, 1 mM DTT and 100 μg/mL). RNA was purified using RNeasy Plus Micro kit (Qiagen Cat# 74004). The concentration and quality of the RNA was assessed with an Agilent 2100 Bioanalyzer and only RNA samples with a concentration higher than 4 μg/μL and an RNA integrity number (RIN) greater than 7 were further sequenced. Sequencing libraries were prepared using TruSeq RNA stranded mRNA kit (Illumina Cat# 20020594) and libraries were sequenced on a NextSeq 500 system (Illumina).
Reads were mapped to the primary assembly of the mouse reference genome contained in Ensembl release 104, using the STAR RNA-seq aligner, version 2.7.9a112. Tables of per-gene read counts were then generated from the mapped reads with featureCounts, version 2.0.2113. Differential gene expression was performed in R using DESeq2, version 1.30.1114.
Identification of BEC-endfoot ligand-receptor pairs
The full list of mouse ligand-receptor pairs was downloaded from CellTalkDB61 (http://tcm.zju.edu.cn/celltalkdb/). Ligands were identified by comparing the genes identified in BEC RNA-seq data to CellTalkDB. For ligands at baseline (PBS), CellTalkDB ligands were compared to genes identified with FPKM > 1 for BEC IP in PBS samples. For ligands altered by LPS, CellTalkDB ligands were compared to upregulated (LPSvsPBS, FPKM > 1, logFoldChange >1, FDR < 0.05) or downregulated genes (LPSvsPBS, FPKM > 1, logFoldChange < −1, FDR < 0.05). Receptors in endfeet were identified as described above. Ligand-receptor pairs were identified by comparing ligands found in BEC RNA-seq with receptors found in the endfoot proteome. For PBS ligand-receptor pairs, only ligands and receptors identified in the PBS samples were considered. For ligand-receptor pairs that change with LPS, up- or downregulated ligands after LPS were matched with receptors found in the endfoot proteome samples receiving either PBS or LPS.
Comparison of ligand-receptor pairs involved in mouse BEC-endfoot communication against human data
To evaluate the overlap of signaling pathways identified between mouse BECs and endfeet with human data, we generated our own young adult (34–40-years-old) healthy human vessel dataset (Fig. 7, Supplementary Fig. 11 and Supplementary Data 6), and sourced three previously published datasets on neurodegeneration-associated molecular changes in the human vasculature: single-nucleus RNA sequencing (snRNA-seq) of multiple sclerosis (MS)71, snRNA-seq of Alzheimer’s disease (AD) patients73 and proteomics of human vasculature isolated from AD patients and controls72 (Supplementary Data 7). To analyze single-cell data from Yang et al. and Macnair et al., per-cell, per-gene count matrices were loaded using Seurat115 (R package version 4.3.0). Pseudo-bulk differential expression analysis was then performed by summarizing single cell gene expression profiles for different cell types at the sample level using the aggregateBioVar (https://github.com/jasonratcliff/aggregateBioVar; R package version 1.6.0), then differentially expressed genes between conditions were calculated using DESeq2114 (R package version 1.36.0). To analyse proteomics data from Wojtas et al., we used publicly available data. In downstream analysis, we focused on differential expression between cortical gray matter demyelinated lesions and controls in the Macnair et al. study, and between cortex of Alzheimer’s disease patients and controls in the Yang et al. and Wojtas et al. studies. A ligand-receptor pair was defined as “identified” in a human dataset if both the ligand and the receptor were detected. For snRNA-seq datasets, ligands were searched in the capillary cluster and receptors in the astrocyte cluster. For proteomics datasets, ligands and receptors were searched in the entire vascular proteome. A ligand-receptor pair was defined as “differentially expressed” in a human dataset if either the ligand or the receptor was differentially expressed (FDR < 0.1 for snRNA-seq studies, Bonferroni-adjusted p < 0.1 for proteomic studies) in the dataset, and its corresponding ligand or receptor was identified in the dataset. See Supplementary Data 7 for further details on the downstream analysis.
Statistical analysis
Data were analyzed with GraphPad Prism or R. Unpaired Student’s two-tailed t-test was used for Fig. 1i (parametric data). Datasets containing more than one datapoint per animal were analyzed using linear mixed-effects modeling (LMM), which included random factors to control for pseudo-replications. After assembling an initial LMM, the normality of the residuals was assessed using Shapiro-Wilk test. To meet model assumptions, data with non-normally distributed residuals were transformed using Tukey Ladder of Powers. To identify significance main effects, Type 3 ANOVA with Satterthwaite approximation was run on the LMM modifying accordingly the code available at https://github.com/Diaz-Castro-Lab/Regional-blood-brain-barrier-ageing-LMM-script. When relevant, the emmeans R package (version 1.10.7) was used for multiple comparisons by Tukey post hoc tests. For distribution analysis, a chi-squared test was used to determine significant differences between groups. No sex-based comparisons were made as this study was not powered for it. Investigators were blinded during analyses.
Representative images are based on at least three independent experiments.
Graphs were generated using both GraphPad Prism and R. Most graphs generated in R were made with the ggplot2 package (version 3.5.1). Sankey plots in Figs. 4, 7, and Supplementary Fig. 12 were made using the networkD3 (version 0.4) and htmlwidgets (version 1.6.4) packages. UpSet plots in Figs. 1, 8, and Supplementary Fig. 3 were made using the UpSetR package (version 1.4.0). Circle dendrogram in Supplementary Fig. 4 was generated the ggraph (version 2.2.1) and igraph (version 2.1.2) packages. Data are represented as mean ± standard error of the mean (SEM) or median depending on the distribution of the data and as indicated in the figure legends. Significance level was considered at p-value < 0.05, FDR < 0.05, FDR < 0.1 or pBonf < 0.1, depending on the dataset as indicated in the figure legends.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.