The SARS-CoV-2 main protease Mpro causes microvascular brain pathology by cleaving NEMO in brain endothelial cells

Coronavirus disease 2019 (COVID-19) can damage cerebral small vessels and cause neurological symptoms. Here we describe structural changes in cerebral small vessels of patients with COVID-19 and elucidate potential mechanisms underlying the vascular pathology. In brains of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2)-infected individuals and animal models, we found an increased number of empty basement membrane tubes, so-called string vessels representing remnants of lost capillaries. We obtained evidence that brain endothelial cells are infected and that the main protease of SARS-CoV-2 (Mpro) cleaves NEMO, the essential modulator of nuclear factor-κB. By ablating NEMO, Mpro induces the death of human brain endothelial cells and the occurrence of string vessels in mice. Deletion of receptor-interacting protein kinase (RIPK) 3, a mediator of regulated cell death, blocks the vessel rarefaction and disruption of the blood–brain barrier due to NEMO ablation. Importantly, a pharmacological inhibitor of RIPK signaling prevented the Mpro-induced microvascular pathology. Our data suggest RIPK as a potential therapeutic target to treat the neuropathology of COVID-19.

reaches the brain by a hematogenous route. In line with a vascular infection, endothelial cells in other organs have been identified as targets of SARS-CoV-2 infection 11,12 . In patients with COVID-19, magnetic resonance imaging detected lesions that are compatible with a cerebral small-vessel disease and with a disruption of the blood-brain barrier (BBB) [13][14][15] . Autopsy studies have confirmed this interpretation [15][16][17][18][19] . However, the microvascular pathology and the underlying mechanisms in COVID-19 are still unclear.
In brains of SARS-CoV-2-infected patients, as well as mouse and hamster models, we found an increase in empty vascular basement membrane tubes, so-called string vessels, reflecting microvascular pathology. The SARS-CoV-2 genome encodes two viral proteases that are responsible for processing the viral polyproteins into the individual components of the replication and transcription complexes. We found that one of them, SARS-CoV-2 M pro (also called Nsp5 or 3CL pro ) 20 , cleaves the host protein nuclear factor (NF)-κB essential modulator (NEMO). NEMO is involved in signaling cascades that regulate the transcription of numerous genes, including the antiviral type I interferons and other immune genes 21 . Beyond gene regulation, NEMO modulates cell survival and prevents apoptosis and necroptosis 22 . The ablation of NEMO in brain endothelial cells induced microvascular pathology in mice that was reminiscent of what we observed in brains of patients with COVID-19. Of note, the widespread death of endothelial cells, rarefaction of capillaries, disruption of the BBB and neuroinflammation due to NEMO ablation were prevented by deleting receptor-interacting protein kinase 3 (Ripk3), a protein kinase that is essential for regulated cell death. Importantly, a pharmacological inhibitor of RIPK signaling prevented the microvascular pathology induced by M pro . These data suggest a potential therapeutic option to interfere with the neurological consequences of COVID-19.

Microvascular brain pathology in SARS-CoV-2 infection.
In patients with COVID-19, magnetic resonance imaging and neuropathological studies have reported parenchymal lesions that are compatible with a small-vessel disease 13,[15][16][17][18] . To search for the underlying microvascular pathology, we stained sections of the frontal cortex for the endothelial cell marker CD34 and the basement membrane component collagen IV. We noted an increase in thin collagen IV-positive strings lacking CD34 staining (Fig. 1a,b). These so-called string vessels were interpreted as remnants of capillaries after endothelial cells have died 23 . Super-resolution microscopy showed that string vessels are tube-like structures with a typical diameter of 0.5-1 µm (Extended Data Fig. 1). To test the association of string vessels with COVID-19, we investigated 17 SARS-CoV-2-infected patients and 23 control patients with similar age and sex distribution (Supplementary Table 1 and 2 and Extended Data Fig. 2). Counting revealed a significant increase in string vessels in patients with SARS-CoV-2 (Fig. 1c). The difference persisted when we stratified the two groups according to comorbidities or sex (Extended Data Fig. 2b-d). Because of pneumonia, more SARS-CoV-2-infected patients were ventilated than controls (Supplementary Table 2). However, neither ventilation nor intensive care unit (ICU) treatment as indicators of severe respiratory disease affected string vessels in control patients (Extended Data Fig. 2h,i), arguing against the possibility that string vessels could be caused by systemic hypoxia. In support, we did not find morphological signs of a global hypoxicischemic encephalopathy in any of the patients (Extended Data  Fig. 2j). In line with the notion that string vessels are formed when endothelial cells die, we observed cells that stained for the apoptosis marker active caspase-3 in brain microvessels. These cells were rare in brain sections of the frontal cortex but significantly more frequent in SARS-CoV-2-infected patients than in controls (Fig. 1d).
To exclude unknown confounding factors that may have occurred in the retrospective clinical study, we sought to reproduce the findings in two animal models of SARS-CoV-2 infection. In hamsters infected with SARS-CoV-2, the length of string vessels increased on day 4 after infection ( Fig. 1e,f and Supplementary Fig. 1a). Interestingly, the increase was transient and normalized at later time points. A similar finding was observed in K18-hACE2 mice, which express the human angiotensin-converting enzyme 2 (hACE2) receptor, under the control of the keratin 18 promoter. In this model, SARS-CoV-2 infection induced the formation of string vessels in a time-dependent manner (Fig. 1g,h and Supplementary Fig. 1b). In summary, the data show that SARS-CoV-2 infection induces a microvascular pathology in the brain in the form of string vessels.

Fig. 2 | Brain endothelial cells express SARS-CoV-2 receptors in
mice and humans. a, RNA-seq in single mouse brain cells characterized the cell-type-specific expression of the SARS-CoV-2 receptors Ace2, Bsg and Nrp1. Colors represent mean gene expression, and diameters denote the percentage of positive cells in the 20 cell clusters. Uniform manifold approximation and projection (UMAP) plot and dot plot for marker genes are shown in Extended Data Fig. 3. b, Representative images of mouse brain co-stained for ACE2, BSG, NRP1, the endothelial markers CD31 or caveolin-1 and the pericyte marker PDGFRβ. BSG and NRP1 were co-localized with caveolin-1 or CD31, respectively, but not with PDGFRβ. The staining was reproduced in at least six mice for each marker. Noteworthy, in the scRNA-seq analysis, the number of Ace2 mRNA-positive cells was low and did not fully reflect the number of ACE2-positive cells identified by immunostainings. In immunostainings, almost all pericytes and tanycytes 59 were positive for ACE2, in contrast to the scRNA-seq data. Scale bars, 5 µm. VLMCs, vascular and leptomeningeal cells; OPCs, oligodendrocyte progenitor cells. c, Cell-type-specific expression of ACE2, BSG and NRP1 in a previously published single nuclear RNA-seq profile of human brain 28 . Gene expression of ACE2, BSG and NRP1 is shown as dot plots for all 30 clusters. UMAP plot and dot plot for marker genes is shown in Extended Data Fig. 4. d, Representative images of the human frontal cortex co-stained for ACE2, BSG and NRP1 with the endothelial marker CD34 confirmed the cell-type-specific localization of the receptors in the vascular unit. ACE2, BSG and NRP 1 were co-localized with the endothelial protein CD34. Images were obtained from a dataset of three patients (three sections per patient). Scale bars, 5 µm. e, Human brain endothelial hCMEC/D3 cells were transfected with human ACE2 and were incubated with SARS-CoV-2 (multiplicity of infection (MOI) of 1). Twenty-four hours after exposure to the virus, the spike glycoprotein was detected in several ACE2-positive cells indicating infection. The experiment was performed three times. Scale bars, 5 µm. f, dsRNA was found in caveolin-1-positive endothelial cells in the cortex of two of four patients with COVID-19 but not in three uninfected controls. Scale bar, 20 µm and 2 µm (inset).
To evaluate the expression of NRP1, BSG and ACE2 in human brain endothelial cells, we analyzed a published single-nuclei RNA-seq dataset that allowed the identification of endothelial and pericyte cell clusters 28 (Extended Data Fig. 4a,b). In line with the mouse data, BSG and NRP1 were highly expressed in endothelial cells (Fig. 2c). In this study, ACE2 was below the detection limit in Endothelial cells (1) Endothelial cells (2) Neurons (1) Neurons (2) Neurons (3) Neurons (   the cells are susceptible to a SARS-CoV-2 infection. Because endothelial expression of ACE2 decreases in vitro due to the lack of blood flow 30,31 , we transfected cultured brain endothelial hCMEC/D3 cells with an expression plasmid of ACE2 and added SARS-CoV-2. In ACE2-positive cells, but not ACE2-negative cells, the spike protein was found in a perinuclear localization as reported for pulmonary endothelial cells (Fig. 2e) 32 . In accordance with the concept that SARS-CoV-2 can infect brain endothelial cells 5,8-10 , we detected double-stranded RNA (dsRNA) in endothelial cells (Fig. 2f) and the S gene of SARS-CoV-2 encoding the spike protein in cerebral vessels (Extended Data Fig. 4c) of a patient with COVID-19. Together, these data show that brain endothelial cells express receptors for SARS-CoV-2 and are susceptible to an infection. M pro cleaves NEMO. SARS-CoV-2 manipulates host cells to promote its survival and propagation. We hypothesized that it might be a successful evolutionary strategy for SARS-CoV-2 to cleave NEMO as an essential component of the antiviral immune response 21 . Indeed, NEMO was degraded in infected Vero E6 and hCMEC/ D3 cells before the cells died (Extended Data Fig. 5a,b). Moreover, immunoblots of NEMO in brain lysates of SARS-CoV-2-infected and control patients indicated that NEMO was cleaved in some of the infected patients (Extended Data Fig. 5c).
The SARS-CoV-2 genome encodes two proteases, M pro and the papain-like protease 20 . Purified M pro cleaved recombinant human NEMO as well as human and mouse NEMO in extracts of brain endothelial or HEK293T cells in a dose-dependent and time-dependent manner (Fig. 3a,b and Extended Data Figs. 6 and 7a,b). Importantly, M pro cleaved NEMO also in intact brain endothelial cells ( Fig. 3c and Extended Data Fig. 7c). When coexpressed with NEMO-2A-GFP that self-processes to NEMO-2A and green fluorescent protein (GFP), HA-tagged M pro completely neutralized NEMO-2A in human hCMEC/D3 cells.
In vitro, M pro produced several NEMO fragments ( Fig. 3a and Extended Data Fig. 6). Tryptic digestion of M pro -treated NEMO followed by mass spectrometry of the generated peptides showed that cleavage occurred at Q83, Q205, Q231, Q304 and Q313 (Fig. 3d,e and Supplementary Fig. 2a-j). The cleavage sites that we identified ( Supplementary Fig. 2k) resemble other known recognition sequences of M pro (ref. 20 ). We verified the cleavage at Q231 by using synthetic peptides as substrates corresponding to both the human and mouse NEMO sequence ( Fig. 3f and Supplementary Fig. 3a-c). With the human NEMO sequence at Q231, the apparent catalytic efficiency (about 43 M −1 s −1 ) was in the range that has been reported for the cleavage site between Nsp4 and Nsp5 ( Supplementary  Fig. 3d) 33 . In keeping with the central role of Q in the recognition sequence, the mutation p.Gln231Ala in the human NEMO sequence prevented cleavage by M pro at this site (Fig. 3f).
NEMO is an essential component of the canonical pathway leading to the activation of NF-κB by inflammatory factors such as interleukin (IL)-1β. Supporting the functional relevance of NEMO cleavage, M pro blocked NF-κB activation. When expressed in human brain endothelial hCMEC/D3 cells, M pro prevented the nuclear translocation of the NF-κB subunit p65, reflecting its activation in response to IL-1β (Fig. 4a). M pro also abolished the stimulation of NF-κB-mediated gene transcription by IL-1β, which we investigated in hCMEC/D3 cells and mouse brain endothelial bEnd.3 cells using luciferase reporter gene assays (Fig. 4b, c). Thus, we obtained unequivocal evidence that M pro cleaves and thereby inactivates NEMO.
M pro -induced damage mimics microvascular pathology. As NEMO is required for the integrity of some but not all cell types 22 , the question arose whether the M pro -mediated cleavage of NEMO compromises endothelial survival. To test whether M pro induces endothelial cell death, we transfected hCMEC/D3 cells with a plasmid encoding M pro -HA and treated the cells with tumor necrosis factor (TNF) to model the elevated TNF serum concentrations in patients with COVID-19 (ref. 34 ). M pro -expressing cells were more often positive for the cell death marker TUNEL, especially when exposed to TNF (Fig. 4d,e), demonstrating that M pro promotes endothelial cell death. To test whether protease activity and NEMO cleavage are required for the toxic effects of M pro , we mutated the amino acid Cys145 in the catalytic site to alanine (p.Cys145Ala-M pro ) 20 . This mutation abrogates the protease activity 35 . As expected, p.Cys145Ala-M pro did not cleave NEMO-2A when coexpressed with NEMO-2A-GFP (Fig. 4f).
In these experiments, levels of the HA-tagged p.Cys145Ala-M pro were higher than those of M pro , probably reflecting a better cell viability with the mutated protease. In support of this concept, the number of GFP-positive hCMEC/D3 cells dropped when M pro -HA but not p.Cys145Ala-M pro -HA were co-transfected (Fig. 4g). In contrast to M pro , p.Cys145Ala-M pro did not inhibit NF-κB activation by IL-1β (Fig. 4h). These data show that NEMO cleavage and cell toxicity by M pro depend on its protease activity.
To explore the function of M pro in vivo, we used the adenoassociated viral vector AAV-BR1, which selectively targets brain endothelial cells when injected intravenously 36 (Fig. 4i). After administering AAV-BR1 vectors encoding GFP or HA-tagged proteins, the accumulation of genomic particles in the brain was similar between the vector groups, and about 10-20% of cerebral capillaries expressed GFP or the HA epitope, while different anti-M pro antibodies did not detect the tagged protein ( Fig. 4i and Supplementary  Fig. 4). When mice received AAV-BR1-M pro 2 weeks before, we observed an increased number of string vessels and a decreased vascular density in their brains ( Fig. 4j-l). The inactive p.Cys145Ala-M pro had no effect on string vessels or vascular density. In line with a Cys145Ala-M pro -HA, immunoblots of cell lysates were performed (representative of at least six experiments). g, More hCMEC/D3 cells survived after expressing the inactive variant p.Cys145Ala-M pro -HA than after expressing M pro -HA. All cells were transfected with pCAG-GFP in parallel. The numbers of GFP + or HA + cells are depicted (N = 6 wells per group). h, In contrast to M pro , p.Cys145Ala-M pro did not inhibit the nuclear translocation of the NF-κB subunit p65 when hCMEC/D3 cells were stimulated with IL-1β (0.25 µg ml −1 ) for 30 min. Cells were transfected with control plasmid, pCAG-M pro -HA or pCAG-p. Cys145Ala-M pro -HA (N = 6 wells per group). i, Schematic of AAV-BR1 vectors to transduce brain endothelial cells in vivo. WPRE, woodchuck hepatitis posttranscriptional regulatory element. j, AAV-BR1-M pro but not AAV-BR1-p.Cys145Ala-M pro led to the formation of string vessels (arrowheads) in the brain of mice. Representative images were taken in the cortex. Scale bar, 20 µm. k, Quantification of string vessel length as a percentage of total vessel length (N = 9-10 mice per group). l, Total vessel length was reduced after mice received AAV-BR1-M pro but not AAV-BR1-p.Cys145Ala-M pro (N = 9-10 mice per group). *P < 0.05, **P < 0.01, ***P < 0.001. Means ± s.e.m. are shown. Detailed information on the exact test statistics, sidedness and values is provided in Supplementary Table 5.
detrimental effect of M pro on endothelial cell survival, there were less HA-positive vessels after expression of M pro -HA in comparison to p.Cys145Ala-M pro -HA ( Supplementary Fig. 4c). M pro -induced string vessels were indistinguishable from those that had occurred in SARS-CoV-2-infected patients.
Capillaries are at risk from ablating NEMO. To test whether NEMO ablation could be responsible for the vascular pathology observed with endothelial M pro expression, we used a mouse model of inducible Ikbkg (Nemo) deletion in brain endothelial cells (Nemo beKO ) 37 . Similar to the M pro -mediated cleavage of NEMO, its genetic ablation led to numerous string vessels in the brain ( Fig. 5a,b). STED microscopy showed that string vessels are tube-like structures with a similar morphology in humans, hamsters and Nemo beKO or K18-hACE2 mice ( Fig. 5b and Extended Data Fig. 1a-c). They were deficient of endothelial cells. Our interpretation of string vessels as a sign of endothelial demise is based on the following observations. NEMO deletion induced endothelial cell death as detected by the TUNEL reaction or by staining for active caspase-3 ( Fig. 5c-e). Most of them were observed in higher-order capillaries ( Fig. 5f and Extended Data Fig. 8), suggesting that capillaries are particularly susceptible to NEMO deficiency. Consequently, small-diameter vessels were predominantly lost and mice developed patchy hypoxia in the brain (Fig. 5g,h) but not in peripheral organs ( Supplementary Fig. 5a,b). Finally, Nemo deletion led to a significant vessel rarefaction in the brain (Fig. 6a) and to a BBB disruption (Fig. 6c,f and Supplementary Fig. 5c). The loss of endothelial cells induced by NEMO deficiency also affected other cell types in the neurovascular unit. While the overall coverage of vessels by pericytes was slightly lowered (Fig. 5i), the number of microglia increased and they exhibited an activated morphology (Fig. 5j). The astrogliosis marker glial fibrillary acidic protein (GFAP) was strongly upregulated, indicating an inflammatory response (Fig. 5k).

NEMO ablation induces vascular pathology via RIPK signaling.
Microglia orchestrate the neuroinflammatory response in the brain, including the activation of astrocytes 38 . Therefore, we tested their role in the microvascular pathology induced by NEMO deficiency. However, ablating microglia by administering the CSF-1R antagonist PLX5622 did neither prevent the formation of string vessels nor prevent the activation of astrocytes ( Supplementary Fig. 6).
To develop alternative therapeutic options, we considered previous reports that NEMO blocks apoptosis or necroptosis in epithelial cells 22 . In Nemo beKO mice, we inactivated the Fas-associated death domain protein (FADD), a component of apoptosis signaling and RIPK3, a kinase central for both necroptosis and apoptosis (Extended Data Fig. 9a). FADD deficiency did not ameliorate the consequences of Nemo deletion but enhanced the damage (Extended Data Fig. 9b-f). FADD not only mediates apoptosis but also inhibits necroptosis, explaining the detrimental effect of FADD deletion 39,40 .
In contrast, RIPK3 deficiency, which by itself did not affect the cerebral microvasculature, prevented the formation of string vessels and the rarefaction of cerebral vessels due to Nemo deletion (Fig. 6a). Probably as a response to the vessel rarefaction, NEMO deficiency stimulated endothelial proliferation indicating angiogenesis, which has been described in patients with COVID-19 previously 11 . Notably, Ripk3 deletion abrogated endothelial proliferation (Fig. 6a). Ripk3 deletion also normalized survival, brain weight and body weight of mice with a NEMO deficiency in brain endothelial cells and reduced the extravasation of IgG and albumin into the parenchyma, showing that disruption of the BBB was mitigated (Fig. 6b,c and Extended Data Fig. 9c-f). Consequentially, Nemo deletion did not activate microglia or astrocytes in the absence of RIPK3 (Fig. 6d).
To explore the mechanisms of BBB protection by RIPK3 deficiency, we quantified the levels of the tight junction protein occludin. NEMO deficiency led to interruptions and the disintegration of occludin-positive tight junctions, which was prevented by RIPK3 deficiency (Fig. 6e). In addition to endothelial tight junctions, the BBB is characterized by a low rate of transcytosis in cerebral capillaries. The increased IgG extravasation in Nemo beKO mice was associated with a lower number of IgG-filled vesicles in brain endothelial cells (Fig. 6f). Super-resolution imaging confirmed the decrease of IgG-loaded vesicles, although the density of all vesicles that were detected by transmission electron microscopy was not reduced ( Supplementary Fig. 7b,c). We have observed a similar reduction of IgG-filled vesicles despite increased overall IgG extravasation in Pdgfb Ret/Ret mice, indicating that the detected population of IgG-filled vesicles limit IgG permeation across the BBB 41 . RIPK3 deficiency did not counteract the effect of NEMO ablation on IgG transcytosis (Fig. 6f). Therefore, we conclude that RIPK3 deficiency improves the BBB tightness of Nemo beKO mice mainly by preventing endothelial cell death and rescuing tight junctions.
Importantly, the M pro -induced string vessel formation also depended on the presence of RIPK3 (Fig. 7a,b), suggesting that inhibitors of RIPK signaling may protect against the microvascular pathology induced by M pro . Although RIPK3 inhibitors block necroptosis, they may induce apoptotic cell death, limiting their translational potential 42 . Therefore, we turned to a small-molecule inhibitor of the upstream kinase RIPK1 that activates RIPK3. The RIPK1 inhibitor abrogated M pro -induced string vessel formation (Fig. 7c,d) and normalized body weight gain that was impaired by M pro expression in brain endothelial cells (Extended Data Fig. 10). Overall, these data indicate that inhibitors of RIPK signaling may prevent cerebral microvascular pathology in COVID-19.

Discussion
In many patients with COVID-19, neurological and psychiatric symptoms occur during the acute disease and determine the post-acute syndrome 4 . Here, we show microvascular pathology in the brains of SARS-CoV-2-infected patients that likely explains signs and symptoms, although systemic effects, including respiratory failure and cytokine release, may contribute to CNS symptoms. Furthermore, we propose a mechanism by which SARS-CoV-2 infection compromises brain endothelial function, damages the BBB and reduces CNS perfusion.
In the brains of SARS-CoV-2-infected patients, we found an increase in string vessels. This finding was confirmed in two animal models of SARS-CoV-2 infection. String vessels are empty basement membrane tubes that often contain pericyte processes. In our view, string vessels are similar or, at least in part, identical to tunneling nanotubes that have been implicated in regulating cerebrovascular coupling 43 . Irrespective of such a function, the association of string vessels with endothelial cell death, BBB disruption and brain ischemia is strong 23,44 , especially because blocking cell death prevented string vessel formation and other changes (Figs. 6 and 7). Therefore, from a technical perspective, string vessels provide a straightforward quantitative measure of capillary damage that allowed a careful statistical comparison between SARS-CoV-2-infected patients and controls. We propose that death of brain endothelial cells in COVID-19 is secondary to their SARS-CoV-2 infection. Although microscopy showed that string vessels are thin tubes with a typical diameter of 0.5-1 µm. The image is representative of three Nemo beKO mice. Scale bars, 1 µm. c,d, String vessels (arrowhead) were often adjacent to dying, TUNEL + or active caspase-3 + endothelial cells in Nemo beKO mice. ERG co-staining indicated TUNEL + nuclei in the endothelium. Representative images in c were obtained from the dataset in e. Images in d are representative of two Nemo beKO mice. Scale bar, 20 µm. e, Increased numbers of TUNEL + cells localized in collagen IV-stained microvessels of Nemo beKO mice (N = 5 mice per genotype). f, More string vessels were present in higher branch orders of the vascular tree in the cortices of Nemo beKO mice (N = 3 mice per genotype). Branch orders were defined by α-SMA staining of arterioles (Extended Data Fig. 8a). g, Nemo beKO mice (N = 3 mice per genotype) preferentially lost vessels with small diameters. The diameters of collagen IV + vessels are shown as a histogram. h, Nemo beKO mice (N = 7) demonstrated patchy cerebral hypoxia detected by the hypoxia probe (HP) in contrast to Nemo fl controls (N = 10). i, Pericyte coverage of vessels was reduced in Nemo beKO mice (N = 8) compared to Nemo fl controls (N = 6). Scale bar, 50 µm. j, Nemo beKO mice (N = 3-4) showed an increased number of activated microglia cells as shown by increased Iba1 + soma area in comparison to that of Nemo fl controls (N = 4-5). Scale bar, 100 µm. k, Nemo beKO mice (N = 4) demonstrated astrogliosis as shown by an increased GFAP + area in the cortex compared to that of Nemo fl controls (N = 5). Scale bar, 100 µm. *P < 0.05, **P < 0.01. Means ± s.e.m. are shown. Detailed information on the exact test statistics, sidedness and values is provided in Supplementary Table 5.
several groups have provided converging evidence for endothelial infection 5,8,11,45 , others have questioned this, mainly based on doubts as to whether endothelial cells express ACE2 (refs. 29,31 ). The jury is still open whether ACE2-positive endothelial cells represent a special endothelial subpopulation or a contamination with pericytes. While ACE2 seems to be essential for SARS-CoV-2 infection, soluble ACE2 may substitute for its lack of expression in certain cell types 46 , suggesting a way how SARS-CoV-2 may infect brain endothelial cells even if they do not express ACE2. The presence of NRP1 and possibly BSG in brain endothelial cells facilitates SARS-CoV-2 cell entry and infectivity 8,24,27,47 .
How could SARS-CoV-2 infection induce the death of brain endothelial cells? Our data demonstrate that M pro of SARS-CoV-2 cleaves host cell NEMO with high efficiency. In infected cells, M pro is located in the cytosol and nucleus, where NEMO is also present 48,49 . Cleavage by M pro inactivated NEMO. This may benefit the virus by preventing the induction of antiviral type I interferons that depends on NEMO 21 . Indeed, levels of type I interferons are low or absent Cleaving NEMO is also a strategy used by other viruses [51][52][53] . However, the tropism of SARS-CoV-2 is likely to limit NEMO inactivation to specific cell types. Accordingly, some NF-κB-dependent cytokines, such as TNF and IL-6, are highly upregulated, indicating that the cells of origin have escaped NEMO inactivation 34 . In addition to its central role in immunity, NEMO supports the survival of some but not all cell types 22 . While neurons, glia and endothelial cells of peripheral vessels seem to resist NEMO deficiency or are even protected by it against inflammatory stimuli 54,55 , the survival of other cells, including brain endothelial cells, is supported by NEMO. Our data suggest that, in COVID-19, brain endothelial cells are at disproportionate risk when being infected by SARS-CoV-2 because of their dependence on NEMO activity for survival. Cleavage of NEMO by M pro mimics the genetic disease incontinentia pigmenti that is caused by inactivating mutations in the NEMO (IKBKG) gene. In the latter condition, patients suffer from a mix of neurological symptoms, such as encephalopathy, stroke and seizures that resemble neurological manifestations of COVID-19 (ref. 56 ). The absence of NEMO in mice induced a loss of endothelial cells and microvascular pathology. Subsequently, patchy hypoxia developed in the brain and the BBB became leaky. In parenchymal  57 ). An increased BBB permeability and astrocyte activation may cause epileptic seizures in patients with COVID-19 as in incontinentia pigmenti 56,58 . M pro -mediated damage of brain endothelial cells suggests that inhibitors of M pro may prevent neurological complications of the SARS-CoV-2 infection 26 . Another therapeutic approach may arise from the observation that deletion of Ripk3 or inhibition of RIPK1 profoundly improved the microvascular pathology. RIPK1 inhibitors are available and have already entered clinical testing 40,42 suggesting potential therapeutic options for COVID-19 as well as for incontinentia pigmenti.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41593-021-00926-1.  Plasmid construction and AAV vector production. Plasmids (pCAG-M pro -HA and pCAG-p.Cys145Ala-M pro -HA; Vectorbuilder) contained inverted terminal repeats of AAV2, the 1,733-bp-long CAG promoter 66 , a Kozak sequence with an ATG followed by the native SARS-CoV-2 sequence encoding M pro (gene ID: 43740578) or a mutated M pro (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 before 36,67,68 . 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 × 10 11 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 previously 69 .
For infection of hCMEC/D3 cells, SARS-CoV-2 virus was isolated and propagated in Caco2 cells as previously described 70,71 . 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).
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 origin 73 . 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 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 studies 8,9,19 . 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.

Patients. The clinical details of SARS-CoV-2-infected and control patients are summarized in
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.

Animals. 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 alleles 60,61 were crossed with the BAC-transgenic Slco1c1-CreER T2 strain 62 , which expresses the tamoxifen-inducible CreER T2 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 previously 63 . 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, Nemo fl/fl ; Slco1c1-CreER T2 mice were indicated as Nemo beKO , while control littermates lacking the Cre recombinase but receiving tamoxifen injection were termed Nemo fl . If not mentioned otherwise, mice were perfused 14 d after receiving the first dose of tamoxifen.
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.
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) The intensity of the target protein was expressed relative to the intensity of actin and normalized to the ratio of the control group (Nemo fl ).

Mass spectrometry.
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 H 2 O 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 × 10 6 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 10 5 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 1.6.1.0, 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 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 cerebellum 28 . The gene expression matrices were further processed as described above using the Seurat package (v.3.2.0).
Proteolytic cleavage of NEMO. The M pro protein was generated as described recently 20 . 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 M pro at the indicated concentrations and time points in protease buffer. For mass spectrometry, NEMO was incubated with M pro (5 µM) in protease buffer for 3 h at 37 °C and samples were lyophilized. Tryptic in-solution digestion was performed as described previously 74 . Briefly, after resuspending samples in 50 µl 6 M urea, a reduction was performed using 2.5 mM DTT (in 100 mM NH 4 HCO 3 ) at 56 °C for 20 min and samples were alkylated using 7.4 mM iodoacetamide (in 100 mM NH 4 HCO 3 ) for 30 min at room temperature in the dark. For tryptic digestion, first NH 4 HCO 3 (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 M pro (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, M pro (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.
Immunoblotting. 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). for 2 h, samples (approximately 1 mm 3 ) 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 publications 37,41,80 . 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.
Non-fluorescence histological staining on human samples. Histological hematoxylin and eosin staining and Nissl staining were performed as described before 19 . For active caspase-3 staining, deparaffinized tissue sections were treated for antigen retrieval as described above and subsequently with 3% H 2 O 2 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 .
Super-resolution microscopy. 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 previously 77 . 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 before 41 . 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 before 37 .
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 vessels 78,79 .
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.
Electron microscopy. 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 Extended Data Fig. 1 | String vessels have a tube-like structure. a, b, c, Super-resolution imaging of human (a), hamster (b) and mouse string vessels (c). In confocal mode with the pinhole closed to 0.4 AU, the string vessels appeared as solid collagen tubes. However, STED microscopy demonstrated a tube-like structure of string vessels in humans, hamsters and mice. In all species, the apparent diameter of string vessels was about 500 -1,000 nm with the collagen walls having a similar thickness. Images are representative for at least 3 experiments per species. Scale bar, 1 µm. control patients showed evidence for a hypoxic-ischemic encephalopathy, also known as 'respirator brain'. Characteristic morphological features of eosinophilic neurons, laminar changes in the cerebral cortex or loss of neurons were not detectible. Representative images of a combined HE and Nissl staining of brain sections from a control and COVID-19 patient as well as a brain section from a patient with hypoxic-ischemic encephalopathy are depicted. Scale bar, 50 µm. Means ± s.e.m. are shown if not stated otherwise. * p < 0.05 *** p < 0.001. Detailed information about the exact test statistics, sidedness and values are provided in Supplementary Table 5.  Fig. 8 | String vessels occur in the deeper part of the vascular tree. a, String vessels were tracked in confocal 3D scans of the cortex after staining against CD31, collagen IV and smooth muscle actin (α-SMA) (left). Since α-SMA is expressed in arteries and arterioles (middle), the branching order can be determined (right). String vessels positive for α-SMA are defined as arteriole string vessels, adjacent string vessels as 1 st capillary string vessels, etc. (right). The image was obtained by cropping a section from a Nemo beKO tilescan from the dataset used for quantification in (b) (N = 3 animals). Scale bar, 50 µm. b, Tracking of string vessels in the hippocampus and hypothalamus of Nemo beKO mice revealed that more string vessels were present in higher branch orders of the vascular tree in Nemo beKO mice. N = 3 animals. Means ± s.e.m. are shown, *P < 0.05, **P < 0.01. Detailed information about the exact test statistics, sidedness and values are provided in Supplementary Table 5. Fig. 9 | Modulation of apoptosis and necroptosis pathways in mice with NEMO deficient brain endothelial cells. a, Scheme showing the role of RIPK3 and FADD in cell death caused by NEMO deficiency. TNF leads to the activation of several adaptor proteins and kinases, such as TAK1. Activation of the IκB kinase complex triggers the translocation of NF-κB into the nucleus and the expression of target genes. NEMO is the regulatory subunit of the IκB kinase complex and is essential for NF-κB pathway activation. In the absence of NEMO, TNF leads to the formation of complex IIa, complex IIb, and the necrosome and induces apoptosis or necroptosis. According to current concepts (Kondylis, V., Kumari, S., Vlantis, K. & Pasparakis, M. The interplay of IKK, NF-kappaB and RIPK1 signaling in the regulation of cell death, tissue homeostasis and inflammation. Immunol Rev 277, 113-127 (2017)), apoptosis is mediated by FADD and RIPK3, while necroptosis depends on RIPK3. b, Deletion of Fadd in brain endothelial cells (FADD beKO ) led to the formation of string vessels and reduced vessel lengths. The results obtained in Nemo fl , Nemo beKO , Ripk3 −/− , and Nemo beKO Ripk3 −/− mice are also included in Fig. 6 of the main article. c, Representative immunoblots showing the extravasation of IgG and albumin into the brain tissue. In total, 9 blots were analyzed with one replicate per animal. d, Quantification of the immunoblots normalized for actin levels demonstrated a disruption of the BBB in FADD-deficient mice as shown by the extravasation of IgG and albumin. Brain weights hint at higher water content in FADD deficient mice. e, Nemo beKO FADD beKO mice died within 9 days after tamoxifen injection while RIPK3 deficiency prevented the death of Nemo beKO mice. f, Ripk3 deletion normalized the body weight of mice with brain endothelial NEMO deficiency, while FADD deficiency worsened body weight loss. Only mice that survived until the end of the experiment were analyzed. Means ± s.e.m. are shown. Detailed information of group sizes are indicated in the individual figure. Corresponding author(s): Markus Schwaninger Last updated by author(s): July 29, 2021 Reporting Summary Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Research policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.

n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A list of figures that have associated raw data -A description of any restrictions on data availability RNAseq data is available at GEO: GSE180984 after July 31st 2021. Other data are available upon request.

nature research | reporting summary
April 2020 Field-specific reporting Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design
All studies must disclose on these points even when the disclosure is negative.

Sample size
No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in previous publications. For other analyses, the availability of patient material determined the sample size.
Data exclusions No data were excluded.

Replication
All experiments and analyses, apart from the analyses of patients, were replicated at least three times if not stated otherwise.
Randomization For treatment allocation, mice were randomized.

Blinding
Whenever possible experimenters were blinded for genotype and treatment.
Reporting for specific materials, systems and methods We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response. Details are reported in the section "Animals" of the manuscript.