Brain endothelial cells express SARS-CoV-2 receptors
To approach the question of how SARS-CoV-2 can infect brain endothelial cells, we determined the expression of membrane receptors and enzymes that are known to facilitate the entry of SARS-CoV-2 in host cells, namely angiotensin-converting enzyme 2 (ACE2), neuropilin-1 (Nrp1), basigin (Bsg), CD209A, Cd209b, Cd209c, Cd209d, and Tmprss29,14,30−32. When analyzing isolated mouse brain cells by single-cell RNA sequencing (sc-RNA-seq), we identified 20 cell clusters, including two endothelial clusters (Extended data Fig. 1a,b). Some cells in the endothelial cell cluster 2 expressed Ace2, albeit at a lower level than pericytes (Fig. 1a). Imaging confirmed that ACE2 was expressed in cerebral microvessels but co-staining of endothelial and pericytic markers and high-resolution microscopy demonstrated that ACE2 was mainly localized in pericytes (Fig. 1a). In contrast, we observed high levels of Nrp1 and Bsg expression in endothelial cells by sc-RNA-seq and immunostaining (Fig. 1b,c), but no Cd209a-d or Tmprss2 (Extended data Fig. 1c-e). In summary, NRP1, BSG, and possibly ACE2 are potential receptors mediating the infection of brain endothelial cells by SARS-CoV-2.
Mpro cleaves NEMO
SARS-CoV-2-infected cells produce specific viral proteins. A strong immune response in COVID-19 patients implies that the viral main protease Mpro is expressed at a high level33. With respect to potential host substrates of Mpro, we speculated that it could be a successful strategy for SARS-CoV-2 to cleave NEMO as an essential component of the anti-viral immune response34. Indeed, purified Mpro cleaved recombinant human NEMO as well as human and mouse NEMO in extracts of brain endothelial or HEK293T cells in a dose- and time-dependent manner (Fig. 2a,b; Extended data Fig. 2, Extended data Fig. 3a,b). Importantly, Mpro cleaved NEMO also in intact brain endothelial cells (Fig. 2c). When co-expressed with NEMO-2A-GFP that self-processes to NEMO-2A and GFP, HA-tagged Mpro completely neutralized NEMO-2A in human hCMEC/D3 cells (Fig. 2c), although the GFP signal persisted (Extended data Fig. 3c), indicating that Mpro degraded NEMO.
In vitro, Mpro produced several NEMO fragments (Fig. 2a, Extended data Fig. 2). Tryptic digestion of Mpro-treated NEMO followed by mass spectrometry of the generated peptides showed that cleavage occurred at Q83, Q205, Q231, Q304, and Q313 (Fig. 2d-e, Extended data Fig. 4). The cleavage sites that we have identified (Extended data Fig. 4k) resemble other known recognition sequences of Mpro (reference 27). We verified the cleavage at Q231 by using synthetic peptides as substrates corresponding to both the human and mouse NEMO sequence (Fig. 2f, Extended data Fig. 5a-c). With the human NEMO sequence at Q231, the apparent catalytic efficiency (about 43 s− 1M− 1) was in the range that has been reported for the cleavage site between Nsp4 and Nsp5 (Extended data Fig. 5d)35. In keeping with the central role of Q in the recognition sequence, the mutation Q231A in the human NEMO sequence prevented cleavage by Mpro (Fig. 2f).
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, Mpro blocked NF-κB activation. When expressed in human brain endothelial hCMEC/D3 cells, Mpro prevented the nuclear translocation of the NF-κB subunit p65, reflecting its activation in response to IL-1β (Fig. 3a). Mpro 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. 3b,c). Thus, we obtained unequivocal evidence that Mpro cleaves and thereby inactivates NEMO.
Mpro-induced damage mimics microvascular pathology in COVID-19 patients
As NEMO is required for the integrity of some but not all cell types29, the question arose whether the Mpro-mediated cleavage of NEMO compromises endothelial survival. To test whether Mpro induces endothelial cell death, we transfected hCMEC/D3 cells with a plasmid encoding for Mpro-HA and treated the cells with tumor necrosis factor (TNF) to model the elevated TNF serum concentrations in COVID-19 patients36. Mpro-expressing cells were more often positive for the cell death marker TUNEL, especially when exposed to TNF (Fig. 3d), demonstrating that Mpro promotes endothelial cell death.
To explore the function of Mpro in vivo, we employed the vector AAV-BR1 that selectively targets brain endothelial cells when injected intravenously37. After administering the control vector AAV-BR1-GFP, about 10% of cerebral capillaries expressed GFP (Fig. 3e,f). When mice had received AAV-BR1-Mpro two weeks before, we observed a decreased vascular density and an increased number of empty basement membrane tubes in their brains (Fig. 3g,h; Extended data Fig. 6). These so-called string vessels have been reported as a sign of microvascular pathology38. They appeared as thin structures with a typical diameter of 0.5–1 µm. Double staining of the vascular basement membrane marker collagen IV and the endothelial marker CD31 revealed that they did not contain endothelial cells.
Importantly, we also found string vessels in the brain of patients that had died with a SARS-CoV-2 infection (Fig. 3i). As in mice, string vessels appeared as thin basement membrane tubes. In the frontal cortex of six patients with SARS-CoV-2 infection, the density of string vessels was higher than in six age- and sex-matched control patients (Fig. 3i, Extended data Table 1).
Capillaries are at risk from ablating NEMO
To confirm that the vascular pathology induced by Mpro is linked to NEMO ablation, we used a mouse model of inducible Nemo deletion in brain endothelial cells (NemobeKO)39. Similar to the Mpro-mediated cleavage of NEMO, its genetic ablation led to numerous string vessels in the brain (Fig. 4a,b; Extended data Fig. 7a). STED microscopy showed that string vessels are tube-like structures with a similar morphology in humans and NemobeKO mice (Fig. 4b, Extended data Fig. 7a). They were deficient of endothelial cells. Our interpretation of string vessels as a sign of ongoing endothelial demise is based on the following observations. NEMO inactivation induced endothelial cell death as detected by staining for active caspase 3 or the TUNEL reaction (Fig. 4c,d). The death of endothelial cells typically occurred in vessel segments adjacent to string vessels (Fig. 4c,d). Most of them were observed in place of higher-order capillaries (Fig. 4e, Extended data Fig. 7b,c), suggesting that capillaries are particularly susceptible to NEMO deficiency. Consequently, small-diameter vessels were predominantly lost and mice developed patchy cerebral hypoxia (Fig. 4f,g). Finally, NEMO ablation led to a significant vessel rarefaction in the brain (Fig. 5a).
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. 4h), the number of microglia increased and they exhibited an activated morphology (Fig. 4i). The astrogliosis marker GFAP was strongly upregulated, indicating an inflammatory response (Fig. 4j).
RIPK3 mediates microvascular pathology induced by NEMO ablation
Microglia orchestrate the neuroinflammatory response in the brain, including the activation of astrocytes40. 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 the activation of astrocytes (Extended data Fig. 8).
To develop alternative therapeutic options, we considered previous reports that NEMO blocks apoptosis or necroptosis in epithelial cells29. In NemobeKO 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). Unexpectedly, FADD deficiency did not ameliorate the consequences of Nemo deletion but significantly enhanced the damage induced by NEMO ablation (Extended data Fig. 9b-f). The deletion of only Fadd in brain endothelial cells led to the formation of string vessels and a disruption of the BBB as shown by the extravasation of IgG and albumin as well as brain edema (Extended data Fig. 9b-d). Consequently, NemobeKOFaddbeKO mice died within 9 days after tamoxifen injection inducing recombination and knockout (Extended data Fig. 9e).
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 ablation (Fig. 5a). Probably as a response to the vessel rarefaction, NEMO deficiency stimulated endothelial proliferation indicating angiogenesis, which has been described in COVID-19 patients before15. Notably, Ripk3 deletion abrogated endothelial proliferation (Fig. 5a). 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. 5b,c; Extended data Fig. 9d,f). Consequentially, NEMO ablation did not activate microglia or astrocytes in the absence of RIPK3 (Fig. 5d).
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+ tight junctions, which was prevented by RIPK3 deficiency (Fig. 5e). In addition to endothelial tight junctions, the BBB is characterized by a low rate of transcytosis in cerebral capillaries. The increased IgG extravasation in NemobeKO mice was associated with a lower number of IgG-filled vesicles in brain endothelial cells (Fig. 5f). While this may seem counterintuitive, we have observed a similar reduction of IgG-filled vesicles despite increased overall IgG extravasation in Pdgfbret/ret mice, demonstrating that the detected population of IgG-filled vesicles limit IgG permeation across the BBB41,42. RIPK3 deficiency did not counteract the effect of NEMO ablation on IgG transcytosis (Fig. 5f). Therefore, we conclude that RIPK3 deficiency improves the BBB tightness of NemobeKO mice mainly by preventing endothelial cell death and rescuing tight junctions. Overall, these data suggest that inhibitors of RIPK3 signaling may protect against the microvascular pathology induced by Mpro and SARS-CoV-2 infection.