The clinical details of SARS-CoV-2-infected and control patients are summarized in Supplementary Table 1. SARS-CoV-2 infection was diagnosed by RT–PCR from pharyngeal swabs. The results of a systematic neurological examination were not available. Patients were autopsied at the University Medical Center Hamburg-Eppendorf or at the University Medical Center Göttingen. SARS-CoV-2-infected patients had partially been included in previous studies8,9,19. The study was approved by the local ethics committees in Hamburg and Göttingen (Hamburg approval no. PV7311; Göttingen approval no. 42/8/20). Control participants were matched to SARS-CoV-2-infected patients according to age and sex. Comorbidities did not differ between the groups, but more SARS-CoV-2-infected patients were ventilated than controls (Supplementary Table 2). Brains were fixed in buffered 4% formaldehyde, examined macroscopically and underwent routine neuropathological workup that did not show morphological signs of a global hypoxic–ischemic encephalopathy in any case (Extended Data Fig. 2j). We analyzed 3–12-µm-thick paraffin-embedded sections of the frontal lobe. The section thickness did not differ between groups (Supplementary Table 2). String vessel measurements were normalized to image volume.
For immunoblotting, frozen tissue samples of medulla oblongata from patients deceased with/from COVID-19 or controls were homogenized in cold RIPA buffer (50 mM Tris-HCl, pH 8, 150 mM NaCl, 1% NP-40, 0.5% Na-deoxycholate and 0.1% SDS) supplemented with protease and phosphatase inhibitor (cOmplete mini EDTA-free, PhosSTOP, Roche). Samples were incubated on ice for 30 min and centrifuged at 12,000g at 4 °C for 10 min. The resulting supernatant was mixed with Laemmli sample buffer and denatured for 10 min at 96 °C.
Transgenic mouse models
Mice were housed in individually ventilated Green Line cages (Tecniplast) under a 12-h light–dark cycle and fed an autoclaved pelleted mouse diet ad libitum. We performed all studies in accordance with the German Animal Welfare Act and the corresponding regulations. Experimental procedures were approved by the local animal ethics committee (Ministerium für Landwirtschaft, Umwelt und ländliche Räume, Kiel, Germany). All mouse lines were established on a C57BL/6 background. In all experiments, adult littermate mice at the age of 6 to 24 weeks were used that were matched by age and sex between experimental groups. Unless stated otherwise, male and female mice were used.
For brain endothelial knockout of Ikbkg (Nemo) and Fadd, mice with the respective loxP-flanked alleles60,61 were crossed with the BAC-transgenic Slco1c1-CreERT2 strain62, which expresses the tamoxifen-inducible CreERT2 recombinase under the control of the mouse Slco1c1 regulatory sequences in brain endothelial cells and epithelial cells of the choroid plexus. Ripk3−/− mice have been reported previously63. Tamoxifen (dissolved in 90% Miglyol 812 with 10% ethanol, 50 mg per kg body weight, intraperitoneally (i.p.), twice per day for five consecutive days; Sigma-Aldrich) was injected to induce recombination. After receiving tamoxifen, Nemofl/fl; Slco1c1-CreERT2 mice were indicated as NemobeKO, while control littermates lacking the Cre recombinase but receiving tamoxifen injection were termed Nemofl. If not mentioned otherwise, mice were perfused 14 d after receiving the first dose of tamoxifen.
For microglia depletion, mice were fed the CSF-1R inhibitor PLX5622-containing (1,200 ppm) AIN-76A rodent diet (D11100404i) 14 d before the start of tamoxifen injection. Controls received a control diet (OpenSource Diets, D10001i). Mice received pimonidazole HCl (100 mg ml−1 in 0.9% NaCl; Hypoxyprobe) 1 h before perfusion (i.p., 60 μg per gram of body weight).
For RIPK inhibition, a specific RIPK1 inhibitor (GSK′547, RA15777187)64,65 was suspended in 0.6% methyl cellulose and mice were treated with 60 mg per kg body weight every 12 h by oral gavage. For control treatment we used methyl cellulose. Treatment started 1 d before injection of viral vectors and was stopped at the day of perfusion 2 weeks after viral injection.
Male 8- to 10-week-old golden Syrian hamsters and K18-hACE2-expressing C57BL/6 mice (B6.Cg-Tg(K18-hACE2)2Prlmn/J) were purchased from the Janvier Laboratory (Le Genest-St-Isle, France) and the Jackson Laboratory, respectively. The BetaCoV/France/IDF0372/2020 strain of SARS-CoV-2 was supplied by the French National Reference Center for Respiratory Viruses hosted by the Institut Pasteur (Paris, France). Hamsters were anesthetized by i.p. injection of ketamine (100 mg per kg body weight), atropine (0.75 mg per kg body weight) and diazepam (2.5 mg per kg body weight) and intranasally infected with 100 µl of DMEM containing (or not, in mock samples) 2 × 104 TCID50 (50% tissue culture infectious dose) of SARS-CoV-2. Male 8- to 10-week-old mice were anesthetized by i.p. injection of ketamine (100 mg per kg body weight) and xylazine (10 mg per kg body weight) and intranasally infected with 50 µl of DMEM containing 5 × 103 TCID50 of SARS-CoV-2. For brain preparation, animals were euthanized by i.p. injection of pentobarbital (140 mg per kg body weight) on day 2, 4, 7 or 24 after infection.
All experiments were performed within the biosafety level 3 suite on the Institut Pasteur de Lille campus and complied with current national and institutional regulations and ethical guidelines (Institut Pasteur de Lille/B59-350009). The protocols were approved by the institutional ethical committee (Comité d’Ethique en Experimentation Animale 75, Nord Pas-de-Calais, France) and authorized by the ‘Education, Research and Innovation Ministry’ (APAFIS no. 25041-2020040917227851; APAFIS no. 25517-2020052608325772 v3).
Plasmid construction and AAV vector production
Plasmids (pCAG-Mpro-HA and pCAG-p.Cys145Ala-Mpro-HA; Vectorbuilder) contained inverted terminal repeats of AAV2, the 1,733-bp-long CAG promoter66, a Kozak sequence with an ATG followed by the native SARS-CoV-2 sequence encoding Mpro (gene ID: 43740578) or a mutated Mpro (codon of Cys145 (TGT) changed to Ala145 (GCT)). The 3′-end was labeled with a HA tag and a TAA codon followed by WPRE and the bovine growth hormone polyadenylation signal. We produced AAV vectors by triple transfection of HEK293T cells or Sf9 insect cells as described before36,67,68.
Genomic titers were determined by quantitative PCR against CAG (forward primer: 5′- AACGCCAATAGGGACTTTC-3′; reverse primer: 5′-GTAGGAAAGTCCCATAAGGTCA-3′). Vectors were injected into the tail veins of mice (1.8 or 3.3 × 1011 genomic particles per mouse, 100 µl). Except in the experiment using Ripk3−/− mice (Fig. 7a,b), for which we used male and female mice, only male C57BL/6 mice were used for vector injection. Mice were perfused under deep anesthesia with PBS and paraformaldehyde (PFA, 2% or 4%) 2 weeks after administering the vector. Total DNA of a sagittal brain section (50-µm thick) was extracted using the DNeasy tissue kit (Qiagen) according to the manufacturer’s instructions. We quantified DNA with a spectral photometer (Nanodrop ND-2000C, Peqlab) as described previously69.
Cell culture and transfection
hCMEC/D3 and bEnd.3 cells
The human brain endothelial cell line hCMEC/D3 (Merck SCC066, RRID: CVCL_U985) and the mouse brain endothelial cell line bEnd.3 (American Type Culture Collection (ATCC), CRL-2299, RRID: CVCL_0170) were cultivated as described previously36,37. We used 24-well plates for luciferase assays, 48-well or 96-well plates or chamber slides for immunocytochemistry and 6-well or 12-well plates for immunoblotting.
After withdrawing heparin (hCMEC/D3) or penicillin–streptomycin (bEnd.3) from the medium, we transfected the cells using Lipofectamine 3000 (Thermo Fisher Scientific) and the following plasmids: pNF-κB-Luciferase (200 ng per well; Stratagene), pCAG-hACE2-TMPRSS2 (100 ng per well on 8-well chamber slides; 2,500 ng per well on 6-well plates, Invivogen), pCAG-GFP, pCAG-p.Cys145Ala-Mpro-HA or pCAG-Mpro-HA (400 ng per well on 24-well and 48-well plates; 1,000 ng per well on 12-well plates), pCAG-NEMO-2A-eGFP (1,000 ng per well on 12-well plates)36 and pRL-SV40 (40 ng per well). The DNA was filled up with pBluescript to equal amounts per well. One day after lipofection, we treated the cells with IL-1β (0.25 µg ml−1; PeproTech) and measured luciferase activity using the Dual Luciferase Reporter Assay (Promega) after 6 h. Immunocytochemistry of p65 or the TUNEL reaction was performed after stimulating cells for 30 min with IL-1β (0.25 µg ml−1) or for 4.5 h with TNF (100 ng ml−1), 2–3 d after transfection. For immunoblotting, we lysed cells 2–3 d after transfection.
For infection of hCMEC/D3 cells, SARS-CoV-2 virus was isolated and propagated in Caco2 cells as previously described70,71. To infect endothelial cells, the viral stock was diluted to the desired MOI in culture medium supplemented with 1% FCS and incubated for 2 h with the cells 24 h after transfection. Twenty-four hours after infection, endothelial cells were fixed in 4% PFA for 10 min or lysed for immunoblotting in lysis buffer (20 mM Tris-HCl, pH 7.5, 20 mM NaF, 150 mM NaCl, 10 mM NaPPi and 1% Triton X-100).
Vero E6 cells
Vero E6 cells (ATCC CRL-1008) cultivated in DMEM containing 3% FCS, 1% penicillin–streptomycin, 2 mM l-glutamine, 1% sodium pyruvate and 1% non-essential amino acids (all Gibco/Thermo Fisher) were seeded in 6-well plates and infected with SARS-CoV-2 isolate HH-1 at a MOI of 1 (ref. 72). At the indicated time points after infection, cells were centrifuged (2,000 r.p.m., 5 min) and lysed in SDS inactivation buffer (6% SDS, 150 mM Tris, pH 6.8, 30% glycerol, 100 mM dithiothreitol (DTT) and bromophenol blue).
Single-cell RNA sequencing
Two 10-week-old male C57BL/6 mice were killed by decapitation. The hypothalamic region was microdissected and digested with the Papain Dissociation System (Worthington, LK003150) at 37 °C. After triturating, the cell suspension was centrifuged at 700g for 5 min at room temperature, and the supernatant removed. We resuspended the cell pellet (500 µl EBSS, 56 µl reconstituted albumin–ovomucoid inhibitor solution and 28 µl DNase, Papain Dissociation System) and passed it through a 40-µm cell strainer. After another centrifugation step (700g for 5 min at room temperature), we resuspended the cell pellet in HBSS containing 5% glucose and stored it on ice.
Single-cell capture was achieved by random distribution of the single-cell suspension across >200,000 microwells through a limited dilution approach with the BD Rhapsody system. Cells were sorted as described by the manufacturer (BD Rhapsody cartridge reagent kit, 633731). In total, 20,448 viable cells were captured. Upon cDNA synthesis (BD Rhapsody cDNA kit; 633773), each cDNA molecule was tagged on the 5′ end with a molecular index and cell label indicating its cell of origin73. Whole-transcriptome libraries were prepared with half of the beads using the BD Resolve single-cell whole-transcriptome amplification workflow (BD targeted and AbSeq Amplification Kit, 633774) with a randomer primer pool (tcagacgtgtgctcttccgatctNNNNNNNNN). In brief, second-strand cDNA was synthesized, followed by the ligation of the adaptor for universal amplification. Eighteen cycles of PCR were used to amplify the adaptor-ligated cDNA products. Libraries were quantified using a High Sensitivity DNA chip (Agilent) on a Bioanalyzer 2100 and the Qubit High Sensitivity DNA assay (Thermo Fisher Scientific). Libraries were sequenced using High Output sequencing kits (75 × 2 bp; Illumina) by a commercial provider (Novogene).
Bioinformatic analysis of single-cell RNA-sequencing data
Raw gene expression matrices were generated for each sample by a custom pipeline combining kallisto (v.0.46.1) and bustools (v.0.46.1) coupled with mouse reference version GRCm38. The output filtered gene expression matrices were analyzed by R software (v.4.2.0) with the DropletUtils (v.1.8.0) and Seurat (v.3.2.0) packages. In brief, for each sample, cells were detected by ranking cell barcodes according to their number of unique molecular identifiers (UMIs) captured using the barcodeRanks function. Low-ranked cells from this process were labeled as false positives and were discarded.
Only genes expressed in >0.5% of the dataset and cells with >200 genes assigned were selected for further analyses. Low-quality cells were removed if they included >20% UMIs derived from the mitochondrial genome. Gene expression matrices were normalized by the NormalizeData function and 2,000 features with high cell-to-cell variation were calculated using the FindVariableFeatures function. For both samples, we identified ‘anchors’ between individual datasets with the FindIntegrationAnchors function and fed these anchors into the IntegrateData function to create a batch-corrected expression matrix of all cells, which allowed cells from different datasets to be integrated and analyzed together. The dimensionality of the data was reduced by principal-component analysis followed by visualization with UMAP clustering using the Louvain algorithm. Finally, we clustered cells by using the FindClusters and FindNeighbors functions. Cluster-specific markers were identified by the FindAllMarkers function and assigned to cell types. Clusters were then classified and annotated based on expressions of canonical markers of particular cell types. All details regarding the Seurat analyses performed in this work can be found in the website tutorial (https://satijalab.org/seurat/v3.2/pbmc3k_tutorial.html).
For analysis of single-nuclei RNA-seq data from human brain, preprocessed expression matrices were obtained from the Gene Expression Omnibus (GEO; GSE97942) consisting of >60,000 single nuclei from the human adult visual cortex, frontal cortex and cerebellum28. The gene expression matrices were further processed as described above using the Seurat package (v.3.2.0).
Proteolytic cleavage of NEMO
The Mpro protein was generated as described recently20. The purified protein was stored at −80 °C in protease buffer (20 mM Tris, 150 mM NaCl, 1 mM EDTA, 1 mM DTT, pH 7.8) until usage.
For immunoblotting, recombinant human NEMO with an N-terminal GST tag (4.1 µg, ab206008, Abcam) was incubated with Mpro at the indicated concentrations and time points in protease buffer. For mass spectrometry, NEMO was incubated with Mpro (5 µM) in protease buffer for 3 h at 37 °C and samples were lyophilized. Tryptic in-solution digestion was performed as described previously74. Briefly, after resuspending samples in 50 µl 6 M urea, a reduction was performed using 2.5 mM DTT (in 100 mM NH4HCO3) at 56 °C for 20 min and samples were alkylated using 7.4 mM iodoacetamide (in 100 mM NH4HCO3) for 30 min at room temperature in the dark. For tryptic digestion, first NH4HCO3 (425 µl, 100 mM) and then trypsin solution (sequencing grade, 1.5 µl, 0.05 µg µl−1 in 50 mM acetic acid, Promega) were added. After incubation for 18 h at 37 °C, samples were desalted using C18 SPE cartridges (Sep-Pak, Waters) and resuspended in 30 µl 0.1% formic acid for liquid chromatography (LC)–MS/MS.
To validate the cleavage of human and mouse NEMO at Q231, synthetic peptide substrates as shown in the respective figures and reference h-NEMO_222-231 (EEKRKLAQLQ), consisting of the native or mutated human or mouse NEMO sequence, were commercially obtained (Peptide Specialty Laboratories). Peptide substrates (10 µM) were incubated with Mpro (2.5 µM) for 1 h at 37 °C in water. We precipitated proteins using ice-cold acetonitrile and then kept samples at −20 °C for 10 min followed by centrifugation at 4 °C and 20,817g for 10 min. The supernatant was lyophilized and samples were dissolved in 30 µl 0.1% formic acid and further diluted at a 1:1 ratio with 0.1% formic acid for LC–MS/MS. To determine the apparent catalytic efficiency, Mpro (2.5 µM) was incubated with different concentrations of h-NEMO_222-241 for 30 min at 37 °C in protease buffer and samples were processed as described above.
Samples were supplemented 1:4 with SDS buffer (0.75 M Tris-HCl, 0.08 g ml−1 SDS, 40% glycerol, 0.4 mg ml−1 bromophenol blue and 62 mg ml−1 DTT) and incubated at 95 °C for 10 min. After loading on SDS–PAGE gels, we transferred proteins to nitrocellulose membranes, which were incubated with primary antibodies (Supplementary Table 3) overnight at 4 °C. Subsequently, we incubated membranes with HRP‐conjugated secondary antibodies (Supplementary Table 4) for 2 h at room temperature. For detection, we used enhanced chemiluminescence (SuperSignal West Femto Substrate, Thermo Scientific) and a digital detection system (Fusion Solo S, Vilber). Immunoblots of IgG and albumin were analyzed by ImageJ (National Institutes of Health, RRID: SCR_002285). The intensity of the target protein was expressed relative to the intensity of actin and normalized to the ratio of the control group (Nemofl).
Dextran (4 kDa, labeled with FITC; BD) was suspended in PBS (12 mg ml−1) and intravenously injected (100 µl per mouse) 30 min before perfusion. Brains were homogenized as described previously37. In supernatants, fluorescence was detected using a microplate reader (CLARIOstar, BMG LABTECH).
Analyses were performed on a UHPLC system (Dionex Ultimate 3000, Thermo Scientific) coupled to a quadrupole orbitrap mass spectrometer (Orbitrap Q-Exactive, Thermo Scientific). We separated the samples on a RP separation column (ACQUITY UPLC BEH C18, 130 Å, 1.7 µm, 2.1 × 100 mm, Waters) using H2O and acetonitrile (LC–MS grade, Merck KGaA), both containing formic acid (0.1%), as eluents at a flow rate of 200 µl min−1. We used stepwise and linear gradients: For tryptic NEMO peptides, 3–45% B in 43 min, 45–70% B in 7 min; for synthetic NEMO peptides, 3–70% B in 10 min. All spectra were acquired in positive-ion mode and capillary voltage was set to 3,500 V, capillary temperature to 320 °C, sheath gas flow to 30 and auxiliary gas flow to 10. Full scan spectra were acquired with a scan range of 150–2,000 m/z (tryptic NEMO peptides) or 400–2,000 m/z (synthetic NEMO peptides) and the resolution was set to 70,000, AGC target to 3 × 106 and maximum injection time to 100 ms. For precursor selection, ions of charge states of 1+ and >6+ were excluded from fragmentation. We fragmented precursor ions using higher-energy C-trap dissociation with a stepped normalized collision energy of 30. MS/MS spectra were acquired with a resolution of 17,500, AGC target was set to 105 and the maximum injection time to 50 ms. Data analysis was performed using the Xcalibur software (version 3.0.63, Thermo Scientific). To identify NEMO-derived peptides, a database search using the MaxQuant software (version 184.108.40.206, Max Planck Society)75 with the Andromeda search engine was performed. We searched raw data against a human NEMO database (https://www.uniprot.org/; July 2020). Digestion mode was set to unspecific and minimum peptide length to 3.
RNAscope (in situ hybridization) for detecting SARS-CoV-2 in human samples
Human brains were immersion fixed in 10% formalin for 1 week at room temperature followed by 4% PFA and PBS 0.1 M (pH 7.4) for an additional 48 h at 4 °C, cryoprotected in 30% sucrose for one additional week at 4 °C, Tissue-tek embedded and frozen in liquid nitrogen at the crystallization temperature of isopentane. The SARS-CoV-2 S gene encoding the spike protein was detected on 20-µm-thick sections using RNAscope Multiplex Fluorescent Reagent kit v2 Assay and the V-nCov2019-S probe (848561, both Advanced Cell Diagnostics) according to the manufacturer’s instructions.
Immunofluorescence staining and confocal microscopy
Human brain sections were deparaffinized in xylene and ethanol, rehydrated in water and rinsed in 0.1% Triton X-100 in PBS for 10 min and 0.1% Tween-20 in PBS for 5 min. To retrieve antigens, we incubated the sections in sodium citrate buffer (10 mM, pH 6, 95 °C, 10 min). The sections were blocked in PBS containing 5% BSA and 0.1% Triton X-100 for 30 min. Primary antibodies (Supplementary Table 3) diluted in blocking solution were incubated at 4 °C overnight. Secondary antibodies (Supplementary Table 4) diluted in blocking solution were incubated at room temperature for 1 h in the dark.
For the staining of cryosections of mouse brains, we perfused mice under deep anesthesia with PBS containing heparin (10 IU ml−1). Brains were frozen on dry ice and stored at −80 °C. Sections (20-µm thick) were postfixed in methanol for 10 min at −20 °C or in 4% PFA in PBS for 15–20 min at room temperature, if not indicated otherwise. Specimens were blocked with either 1–3% BSA in PBS (methanol post-fixation) or 1–3% BSA and 0.1–0.3% Triton X-100 in PBS (PFA post-fixation) for 1 h and stained as described for human sections.
For vibratome sections, mice were either not perfused or perfused with freshly prepared PFA (2% in PBS, 4% for GFP staining). Brains were postfixed in 2% PFA for 7 h (4% PFA for 2 h for GFP staining) at 4 °C before sectioning using a vibratome (Leica, VT1200S). Before the staining of brain sections from SARS-CoV-2-infected hamsters or mice, pepsin antigen retrieval was performed for 10 min at 37 °C (0.1 mg ml−1 pepsin in PBS, 0.2 N HCl). Sections (50- or 100-µm thick) were blocked with 3% BSA in PBS containing 0.1–0.3% Triton X-100 for 6 h at room temperature, and incubation with primary antibodies (Supplementary Table 3) was performed at 4 °C for 48–72 h, while incubation with secondary antibodies (Supplementary Table 4) was performed in blocking solution at 4 °C overnight.
For the TUNEL assay, mouse brains were postfixed in 4% PFA in PBS at 4 °C overnight and transferred to a 30% sucrose solution the next day. On the following day, brains were frozen and stored at −80 °C. Cryosections (40-µm thick) were prepared and a heat-induced epitope retrieval was performed using 10 mM sodium citrate buffer at 95 °C for 20 min. TUNEL assay was applied after immunohistochemistry staining according to the manufacturer’s instructions (In Situ Cell Death Detection Kit, Fluorescein; Roche, 11684795910).
Images were taken by confocal laser scanning microscopes (Leica, SP5 or SP8) or a fluorescence microscope (Leica, DMI 6000B). Images for determining the number of string vessels in human samples as well as super-resolution images were taken using a STED microscope, custom made by Abberior Instruments. All images were produced using the same setting.
For all analyses, we imaged four fields from two sections per individual unless stated otherwise.
Non-fluorescence histological staining on human samples
Histological hematoxylin and eosin staining and Nissl staining were performed as described before19. For active caspase-3 staining, deparaffinized tissue sections were treated for antigen retrieval as described above and subsequently with 3% H2O2 before blocking with PBS containing 10% FCS. Primary antibodies were applied overnight and visualized using the EnVision+ System for rabbit and mouse (Dako). We briefly counterstained sections with hemalaun. To evaluate the number of active caspase-3-positive cells, sections were scanned (magnification of ×200) using the Virtual Slide Microscope VS120. Image visualization and manual analysis were performed using Omero Server software (5.6.3)76.
We used stimulated emission depletion (STED) imaging and expansion microscopy. For STED, 640- and 561-nm diode excitation lasers, a 775-nm STED laser, all pulsed at 40 MHz, and a ×100 1.4-NA Olympus UPlanSApo were utilized. A spatial light modulator (Hamamatsu) was used to produce either a doughnut-shaped (two-dimensional (2D) STED) or a top-hat (three-dimensional (3D) STED) phase mask, shaping different depletion beams without changing the optical setup. To reduce photobleaching at an optimal signal-to-noise ratio, we used DyMIN adaptive illumination.
For expansion microscopy, after immunofluorescence staining, gelation, digestion and expansion were performed as described previously77. Notably, we extended incubation time in monomer solution to 45 min and gelation time to 2.5 h, and digestion was performed overnight. Images were taken with an HC PL APO CS2 ×40/1.10 water objective. Expansion microscopy was used for qualitative representation of occludin and ZO-1 morphology.
Quantitative analysis of immunostainings
To analyze IgG transcytosis, z-stacks were taken with an HCX PL APO CS ×63/1.4 oil objective for confocal microscopy or a ×100 1.4-NA Olympus UPlanSApo for STED microscopy. Deconvolution was performed for confocal images by Huygens Software (Scientific Volume Imaging). IgG vesicle quantification was performed using Imaris 9.3.0 (Bitplane) as described before41. The 3D vasculature mask was smoothed under surface details (1.0 µm confocal/0.4 µm for STED), and spots were identified with an estimated diameter of 0.5 µm. IgG+ vesicles and IgG extravasation were quantified on 10–15 images on one section for each animal (confocal) or three images for each animal (STED).
String vessels and their localization in the vascular tree, vessel length, vessel diameter, occludin interruptions and GFAP+ astrocytes were analyzed using ImageJ. Mouse and hamster string vessels were analyzed as before37.
For vascular tree analysis, string vessels were counted manually and tracked in tile scans of the cortex, hippocampus and hypothalamus (one section per animal). As the starting point of the vascular tree, we chose α-SMA+ vessels, indicating arterioles. The number of string vessels was normalized to the total area of the image.
Because CD34 staining was unreliable in human brain sections, human string vessels were identified as collagen IV-positive tubes of <4 µm in apparent outer diameter in images produced by confocal laser scanning microscopy. Because the theoretical minimal diameter for a functional capillary is 2.7 µm and more than 90% of brain capillaries in aged humans have an inner diameter of >3.5 µm, we expect this threshold to be selective for string vessels78,79.
Vessel diameter was measured by using the DiameterJ plugin for ImageJ. For the TUNEL analysis, collagen IV+ and TUNEL+ vessels were counted and normalized to the total image area.
GFAP-positive astrocytes were quantified as the percentage of the GFAP+ area relative to the total area of the image. Quantifications were obtained from 4–6 images per animal (one section per animal). Images were taken from the cortex if not stated otherwise.
Pericyte coverage was analyzed using MotionTracking software (MPI-CBG, v8). In general, the CD13+ area inside the collagen IV+ vessels was normalized to the total collagen IV+ area. For measuring occludin interruptions, the lengths of occludin and ZO-1 tight junctions were traced manually by using the Simple Neurite Tracer plugin (ImageJ). Then, occludin length was normalized to ZO-1 length for 15 images per animal (one section per animal).
For the soma size measurement of Iba1+ cells, confocal imaging was obtained with 25 steps and a step size of 0.99 µm. Maximal-intensity z-stacks were generated using ImageJ. A threshold was applied (Li autothreshold), converted to a mask and speckles and outliers removed. Then, all the processes were removed from the soma and the soma area was measured for 15 microglia per sample.
For electron microscopy, mice were perfused with heparinized Ringer’s solution and with freshly prepared PFA and glutaraldehyde (2.5% glutaraldehyde and 2% PFA in PBS). Until further processing, brains were postfixed in Monti-Graziadei solution (2% glutaraldehyde, 0.6% PFA, 0.03% calcium chloride in 0.06 M sodium cacodylate buffer, pH 7.35) for at least 48 h at 4 °C. After further fixation in 1% osmium tetroxide in 0.1 M cacodylate buffer for 2 h, samples (approximately 1 mm3) were dehydrated in an ascending series of ethanol and incubated in propylene oxide followed by a 1:1 mixture of propylene oxide and araldite (Sigma-Aldrich) and subsequently embedded in araldite. Ultrathin sections were cut at approximately 80 nm and were transferred to copper grids. Sections were contrasted in a contrasting system for ultrathin sections using uranyl acetate ready-to-use solution, followed by lead citrate ready-to-use solution (all Leica Microsystems). Images of vessels smaller than 10 µm were taken by an electron microscope (Jeol JEM 1011). After putting images of a vessel into a collage by using Inkscape 1.0.1 (RRID: SCR_014479), vesicles in the range of 30–200 nm were manually counted from three capillaries for each animal, and luminal membrane length was measured using ImageJ.
Statistics and reproducibility
No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in previous publications37,41,80. Data were analyzed using Prism 8 (GraphPad) and SPSS 25 (IBM). Significance was considered when P < 0.05. Depending on the dataset and experimental design, different statistical methods were used as indicated in Supplementary Table 5. Parametric statistics (for example, t-test and ANOVA) were only applied if assumptions were met, that is, datasets were examined for Gaussian distribution using the D’Agostino–Pearson test, aided by visual inspection of the data and homogeneity of variances by Brown–Forsythe, Levene’s or F-test (depending on the statistical method used). If assumptions for parametric procedures were not met or could not be reliably assumed due to small sample size, non-parametric methods were used as indicated. Two-tailed tests were applied if not indicated otherwise. Greenhouse–Geisser correction was used in ANOVA statistics if the sphericity assumption was violated (Mauchly test). No data points were excluded. Cell culture studies were performed at least three times in independent experiments with at least three replicates per condition and per experiment unless stated otherwise. Animal experiments were repeated as stated by the N number. Animals were randomly allocated to diet or treatment groups as long as age-matched, sex-matched and littermate conditions were fulfilled. All analyses were performed blinded without the knowledge of the genotype, treatment or infection status if not needed for subsequent processing.
Further information on research design is available in the Nature Research Reporting Summary linked to this article.