SARS-CoV-2 infection of human brain microvascular endothelial cells leads to inflammatory activation through NF-κB non-canonical pathway and mitochondrial remodeling

Neurological effects of COVID-19 and long-COVID-19 as well as neuroinvasion by SARS-CoV-2 still pose several questions and are of both clinical and scientific relevance. We described the cellular and molecular effects of the human brain microvascular endothelial cells (HBMECs) in vitro infection by SARS-CoV-2 to understand the underlying mechanisms of viral transmigration through the Blood-Brain Barrier. Despite the low to non- productive viral replication, SARS-CoV-2-infected cultures displayed increased apoptotic cell death and tight junction protein expression and immunolocalization. Transcriptomic profiling of infected cultures revealed endothelial activation via NF-κB non-canonical pathway, including RELB overexpression, and mitochondrial dysfunction. Additionally, SARS-CoV-2 led to altered secretion of key angiogenic factors and to significant changes in mitochondrial dynamics, with increased mitofusin-2 expression and increased mitochondrial networks. Endothelial activation and remodeling can further contribute to neuroinflammatory processes and lead to further BBB permeability in COVID-19.

In the present work we describe cellular and molecular effects of HBMEC infection by SARS-CoV-2 in order to gain insight into possible routes by which the virus affects the BBB and invades the brain parenchyma. Transcriptomic analyses revealed activation of noncanonical NF-κB signaling pathway and changes in mitochondrial quality control, which combined could induce endothelial activation and promote increased neuroin ammation in Neuro-COVID-19.

Cell culture
Human brain endothelial cells (HBMECs) were a gift from Prof. Dennis Grab (Department of Pathology, Johns Hopkins School of Medicine). Cells were immortalized using a SV40-LT plasmid (Stins et al., 2001) and were maintained in 199 medium with 10% fetal bovine serum (FBS) and 1% antibiotics (penicillin/streptomycin, ThermoFisher) up to passage 38. Vero E6 cells (African green monkey kidney epithelial cells) were used as gold standard for viral isolation and propagation and were used in a few experiments as a positive control for e cient SARS-CoV-2 infection. Vero E6 cells culture medium consisted of Dulbecco's Modi ed Eagle Medium (DMEM) formulated with D-glucose (4.5 g/l) and L-Glutamine (3.9 mM) supplemented with 100× penicillin-streptomycin solution (to nal 100 U/ml and 100 µg/ml, respectively) and with inactivated FBS (USDA-quali ed region FBS) at 10%. Both cell and viral cultures were incubated at 37°C and 5% CO 2 .

SARS-CoV-2 isolate
All the procedures associated to the viral culture and further infection assays were performed in biosafety level 3 laboratory, in accordance with the WHO guidelines (WHO, 2021). The SARS-CoV-2 isolate used in this study was previously obtained from a respiratory sample collected from a COVID-19 patient diagnosed in March 2020, in Brazil, as part of the Brazilian Ministry of Health surveillance system. Viral isolation and genetic characterization were previously described (Matos et al, 2022). All procedures transcription and SARS-CoV-2 gene ampli cation were performed in one-step reactions with a quantitative real time PCR kit developed by Biomanguinhos Institute (Fiocruz, Brazil), in an ABI 7500 thermocycler (Applied Biosystems, USA). As a quanti cation standard, we used a SARS-CoV-2 plasmid control containing the reference sequence of viral envelope (E) gene, with a known number of copies (IDT, USA). Therefore, a concentration curve was prepared by performing serial dilutions of the plasmid.

RNA libraries and sequencing
For RNAseq analysis, three independent replicates were prepared for each treatment groups: Mock, MOI 0.01-and MOI 0.1-infected cultures, at 6 and 24 hpi. Total RNA was isolated using the miRNeasy micro kit (Qiagen) according to the manufacturer's instructions. The RNA was quanti ed by O.D. measurement before being assessed for quality by chip-based capillary electrophoresis using Agilent 2100 Bioanalyzer RNA 6000 Pico assays (Agilent Technologies; Part # 5067 − 1513).
Libraries were prepared from 150 nanograms (ng) of DNA-free total RNA using the Universal Plus mRNA- 6. RNA-seq data analysis Reads R1 and R2 were trimmed 12 nucleotides (nt) to remove low quality sequences. Bases with a quality score of less than Q20 were trimmed off the right end of each R1 and R2. Illumina adapter sequences were trimmed from the 3'-end of both R1 and R2 reads. Read pairs in which mate in the pair was less than 30nt after trimming were discarded. These quality-ltered reads were then used for alignment.
The top differentially expressed genes have consistent UC vs MOIpt1 changes for the three replicates at 5% FDR, and absolute log2 fold change of 0.6 were considered cut-offs to generate the DEG list. Computed z-scores of signi cant genes are represented in the heatmap. Heatmap was plotted using complexheatmap R package. (https://academic.oup.com/bioinformatics/article/32/18/2847/1743594).

Western blotting
HBMECs were cultivated in 60 mm 2 dishes and at desired times were washed with PBS and lysed in the presence of 1x Laemmli Buffer (0.0625M Tris, 0.07M SDS, 10% glycerol, 5% β-mercapto-ethanol and bromophenol blue). Protein concentration was measured with BCA Protein Assay Kit according to the manufacturer's instructions (Thermo Fisher Scienti c, Carlsbad, CA, USA). Then, 30 µg of protein were loaded onto 4-20% gradient acrylamide gels (Bio-Rad Laboratories, Hercules, CA, USA). Membranes were blocked with bovine serum album (BSA) 5% in TBS-0.05% Tween20 and incubated overnight at 4°C with the primary antibodies at 1:1,000 dilution in TBST (Table S2). Next day, blots were washed with TBS-0.05% Tween20, incubated for 1 hour at room temperature with secondary antibodies (Lincoln, NE, USA) and analyzed using the Licor CLX imaging system and the Image Studio 4.0 software (LI-COR).  Table S2. 11. Quantitative analysis of mitochondrial network morphology Mitochondrial network morphology was analyzed using the Mitochondrial Network Analysis Tool (MiNA) for the Fiji distribution of ImageJ (Valente et al. 2017). Images were cropped into individual cells. To enhance contrast and sharped mitochondrial images, several pre-processing tools were applied to each image prior to MiNA analysis. First, an unsharp mask (sigma = 3) was used to sharpen images by subtracting a blurred version of the image (i.e. unsharp mask) from the image. The unsharp mask is created by Gaussian blurring the original image and multiplying the blurred image by the mask weight (0.8). Second, a median lter (radius = 1) was applied to each image. The median lter functions by replacing each pixel with the neighborhood median, where the neighborhood size is determined by the radius. Following pre-processing, images underwent thresholding using the Otsu thresholding to produce a binary image (Otsu, 1979). The mitochondrial footprint is calculated as the total number of mitochondria-signal positive pixels from the binarized image. A morphological skeleton is then produced from the binarized image using the Skeletonize 2D/3D plug-in (Arganda-Carreras et al. 2010; Lee et al. 1994). This method employs iterative thinning to create a skeleton of mitochondrial structures, one pixel wide. Length measurements of the mitochondrial structures are then measured using the Analyze Skeleton plug-in, resulting in two additional parameters: mean branch length and mean summed branch length. Mitochondrial form branching networks, in which branches intersect at a node. Mean branch length is calculated as the mean length of mitochondrial structure between two nodes. Mean summed branch length is calculated by determining the sum of branch lengths within an independent network structure and dividing by the total number of independent networks within a cell.

Angiogenesis-related proteins secretome
For generation of HBMEC Conditioned Medium (CM), cells were plated on 6-well plates and after infection were maintained in a total volume of 1 ml per well. Conditioned culture media were collected at 24 hours post infection (hpi) and centrifuged for 5 min at 10,000 rpm at 4 o C and stored at -80 o C until use. Secretion of angiogenesis-related protein levels was detected using a Proteome Pro ler™ Human Angiogenesis Antibody Array kit (R&D Systems) according to manufacturer's instructions. Membranes were incubated with pools of two independent experiments (e.g.: Mock #1 + Mock #2, MOI 0.01 #1 + MOI 0.01 #2, etc.) for each experimental condition, in a total of 3 membranes for 6 biological replicas. Spots were detected with chemoluminescence and X-ray lms were exposed for 1, 5, 10 and 15 minutes to detect differentially expressed proteins. Densitometric analysis was performed with UN-SCAN-IT gel analysis software and relative intensity values for each spot of the 1-minute exposed lm was analyzed with GraphPad Prism software version 9.0.1.

Transmission electron microscopy
Cells were grown on 35 mm petri dishes and infected or treated as described above. At desired time, cultures were washed in PBS, xed with 2.5% glutaraldehyde diluted in 0.1 M cacodylate buffer with 3.5% sucrose and CaCl 2 for 1 h at 20 ºC, followed by washes in cacodylate buffer and post-xation with 1% osmium tetroxide with potassium ferricyanide for one hour at 4 ºC in the dark. Cells were dehydrated in crescent acetone gradient and embedded in Epon resin at 60 o C for 72 h. Ultrathin sections were obtained with Leica ultramicrotome and collected in 300-mesh copper grids, stained with uranyl acetate and lead citrate and visualized at Hitachi Transmission Electron Microscope at Centro Nacional de Bioimagem (CENABIO-UFRJ).

Statistical analyses
For RT-qPCR and western blotting, a minimum of 5 independent cell culture preparations were used and analyzed with Two-Way ANOVA with Bonferroni post-test in GraphPad Prism Software v9.3.1. Morphometrical analysis of ZO-1 immunostaining were performed with ImageJ software for uorescence intensity and Tight Junction Organization Rate (TiJOR) using the TiJOR macro for ImageJ, which in an index of localization of tight junction proteins in membrane-membrane contact region of adjacent cells as described by (Terryn et al. 2013).

Characterization of HBMEC infection by SARS-CoV-2
In order to characterize the pro le of host cell infectivity by SARS-CoV-2, HBMEC and Vero E6 cells were infected in the presence or not of serine endoprotease TPCK trypsin (1 µg/ml), which was shown to increase infectivity in Calu-3 cells, a permissive cell line for the e cient replication of SARS-CoV-2 (Jiang et al., 2021). Cultures were infected at different multiplicities of infection (MOIs) of 0.01, 0.1, 1 and 2 and supernatants were collected at 6-, 24-, 48-and 72-hours post-infection (hpi) and analyzed by RT-qPCR for quanti cation of viral E gene (Fig. 1A). We found that HBMECs showed no increase in viral replication or release in the supernatant over time, whereas Vero E6 cells had time-dependent release of SARS-CoV-2 in the supernatant, as expected (Fig. 1A). TPCK trypsin treatment did not affect the cell infectivity rates; however, for consistency, all subsequent assays were performed in the presence of TPCK. In the same context, HBMECs infected with different MOIs did not show any increase in the expression of SARS-CoV-2 Spike1 and E genes at 6 and 24 hpi (Fig. 1B). We showed recently that HBMEC cells express, to some extent, several of SARS-CoV-2 receptors at both RNA and protein levels (Torices et al., 2021). Therefore, we evaluated the possible effect of infection on the expression of ACE2 and TMPRSS2 in HBMEC cells and found that infection with the MOI 0.1 induced a 40% decrease in ACE2 mRNA expression (p < 0.05), which did not result in ACE2 protein level alterations (Fig. 1C). However, TMPRSS2 showed a 1.77-fold increase in protein levels in MOI 0.1-infected cultures at 24 hpi (p = 0.078). Despite the apparent nonproductive infection of HBMEC, SARS-CoV-2 was able to induce apoptosis of HBMECs, as detected by cleaved caspase-3 immunostaining (Fig. 1D) at 24 hpi. MOIs 0.01 and 0.1 resulted in 2.27 and 4.1% of caspase-3-positive cells, respectively, whereas non-infected dishes showed a physiological rate of 0.7% of stained cells. Apoptotic stimulus was also observed in Vero E6 cells 24 hpi with MOI 0.1 (Fig. 1D). Positive control with 0.5 and 2.0 µg Staurosporine for 2 hours led to 1.9 and 9% of caspase-3 positive HBMECs, respectively (not shown).

SARS-CoV-2 affects tight junction genes expression in BBB-forming cells
The barrier property of BMECs is mostly conferred by the expression and function of tight junction proteins, such as ZO-1 and claudin-5 (Takata et al., 2021). HBMEC and Vero E6 cells were infected as described above and analyzed at 6 and 24 hpi. ZO-1 immunoreactivity was drastically altered in infected Vero cultures ( Fig. 2A) and showed discontinuous staining in cell-cell contacts, as compared to uninfected controls. SARS-CoV-2 viral particles were clearly detected in Vero cells as revealed by Spike1 immunoreactivity. Conversely, infected HBMEC cultures did not present as signi cant differences in distribution of ZO-1 along cells membranes at 24 hpi ( Fig. 2A). To better evaluate ZO-1 organization in TJ we performed densitometric (ZO-1 uorescence intensity) and tight junction organization rate (TiJOR) (Terryn et al. 2013) analyses. We observed that ZO-1 presented a signi cant 1.29-fold increase in TiJOR index with MOI 0.1 at 6 hpi, concomitantly with a 1.3-fold increase in uorescence signal, and such effects were lost 24 hpi. In parallel, MOI 0.01 affected ZO-1 uorescence signal at 24 hpi by 1.19-fold (Fig. 2B). ZO-1 and claudin-5 mRNA expression remained unaltered throughout the infection (Fig. 2C), but their protein levels were signi cantly increased by 2.0-and 1.17-fold by the MOI 0.1 at 24 hpi, respectively (Fig. 2D).

SARS-CoV-2 infection promotes endothelial activation and hyper-in ammatory response in vitro
We determined the transcriptional pro le of HBMECs after SARS-CoV-2 infection by RNA-Seq after 6 and 24 hpi. At 6 hpi, biological replicas had high variability across experiments, as determined by the square root of the common dispersion and visualized by principal component analysis (not shown). Infection with both MOIs 0.01 and 0.1 led to minimal effect on HBMEC's transcriptome, with few signi cantly differentially expressed genes (DEGs) and no pathways enrichment found (Supplementary material). However, at 24 hpi we observed signi cant impact on host cell transcriptome: exposure to SARS-CoV-2 MOI 0.1 led to up-regulation of 23 and down-regulation of 4 genes. Volcano plot and heatmap on Fig. 3 (A-B, respectively) depict the transcriptomic pro le of HBMECs infected with the MOI 0.1 at 24 hpi. Data obtained from RNA-Seq was consistent with an endothelial activation, with high expression levels of cytokines (IL-6, IL-8, TNF) and chemokines (CXCL1, -2, -8 and CCL20) encoding genes (Fig. 3). Accordingly, functional enrichment analysis revealed that main genes found related to "Cytokine signaling in immune system", "TNF signaling" and "TNFR1-induced NFkappaB signaling pathway", among other Reactome terms (Fig. 3C). In fact, TNF was the most up-regulated gene, with 104-fold increase, followed by TNF-c (Lymphotoxin beta, LTB), with 32.8-fold change (Fig. 3, Table 1). Interestingly, LTB is a known inducer of noncanonical NFκB in ammatory pathway  and was found to be up-regulated both at 6 and 24 hpi in SARS-CoV-2 infected HBMEC by RT-qPCR (Fig. 3E). Although our RNA-Seq data revealed an increase in NFκB2 (p100/p52) and NFκBIA (IκBα), we performed RT-qPCRs with additional biological samples for NF-κB1 (p105/p50) and NF-κB2 and observed that, due to biological variability, such genes remained unaltered in infected cultures (Fig. 3D). However, RELB, the main activator of noncanonical NF-κB signaling pathway , was shown to be up-regulated by con rmatory RT-qPCR at 24 hpi (Fig. 4). We also further con rmed by RT-qPCR the upregulation of in ammation-related genes, including LTB, TNF, IL-6, CXCL1, CXCL2 and CXCL8 (Fig. 3E). Pentraxin3 (PTX3) is a glycoprotein involved in the innate immune response and has a relevant role in FGF2-dependent angiogenesis (reviewed in . We found that PTX3 to be 19.6-fold increased in SARS-CoV-2 infected HBMECs (Figs. 3A and 4). Apart from the in ammatory transcriptomic response, KEGG pathways related to ribosomal structure/function and mitochondrial biology were found to be altered by SARS-CoV-2-infected HBMECs at 24 hpi (Fig. 3D).

Angiogenic pro ling of SARS-CoV-2-infected HBMEC cells
Dysfunctional angiogenesis is a common phenomenon observed in neuroin ammatory states and can be a result of BBB damage (Estato et al., 2018). We analyzed the pro le of angiogenesis-related secreted proteins by HBMECs during infection with SARS-CoV-2 and observed that out of 55 spotted targets, 15 had most signi cantly detectable signals (Fig. 4A-B, Table 2). Highest signals were observed for uPA, serpin-E1 (PAI-1), IL-8, thrombospondin-1, VEGF, TIMP-1, endothelin-1 (ET-1), PTX3, angiogenin and amphiregulin, with at least 5,000 pixels each. SARS-CoV-2-infected cultures (MOI 0.1) had most pronounced increase in secretion of PTX-3 and TIMP-1 as compared to Mock cultures, with 113 and 112% levels, respectively. Accordingly, PTX-3 was also one of the most up-regulated genes as determined by RNA-Seq (Fig. 3). We further assessed the expression levels of PTX3 by RT-qPCR and found it to be increased in HBMEC cultures after 6 and 24 hpi with the MOI 0.1, whereas VEGF, Serpin-E2 and ET-1 showed no signi cant changes at the transcriptional level (Fig. 4C). Insulin-like growth factor (IGF) binding protein-3 (IGFBP-3), a member of the IGFBP family was shown to be 166 and 125% more abundant in the HBMEC conditioned media in MOI 0.01 and 0.1-infected dishes, respectively. We performed scratch-wound healing migration assays in infected HBMECs, however no effect in cellular migration was noticed in infected cultures as compared to Mock-infected (not shown). Interestingly, hypoxia-inducible factor-1 alpha (HIF-1α) was also increased by SARS-CoV-2 at both MOIs at 24 hpi (Fig. 4C). Analytes in bold had at least 3,000 pixels and were considered as valid signal, as shown in Fig. 4.

Mitochondrial plasticity is affected by SARS-CoV-2 infection
Because mitochondria play a role in cellular homeostasis and pathology, we sought to investigate the effects of SARS-CoV-2 infection on mitochondrial plasticity in HBMECs. Cells were immunostained for mitochondrial import receptor subunit TOMM20 (Fig. 5A) MOIs at 24 hpi. We next measured mean mitochondrial branch length mean, which is the average length of mitochondrial structures that are either independent or connected to networks (Fig. 5C). We observed a slight, yet signi cant, increase in cells infected with MOI 0.01 at 6 hpi and with MOI 0.1 at 24 hpi, with a 7 and 3% increase, respectively. Furthermore, MiNA analysis revealed that SARS-CoV-2 induced an overall increase in mitochondrial networks, with signi cant increase in summed branch length mean values at 6 (34 and 33% increase for MOIs 0.01 and 0.1, respectively) and 24 hpi (38 and 45% increase for MOIs 0.01 and 0.1, respectively). Mitochondrial morphological analyses were further assessed by TEM (Fig. 5B) and we found that infected HBMECs displayed larger, swollen mitochondria, with reduced cristae and to some extent associated to multivesicular bodies. Moreover, MOI 0.1-infected cultures displayed 356 mitochondria/mm 2 , while Mock-infected cultures had 266 mitochondria/mm 2 (p < 0.05), which corresponded to a 33% increase (Fig. 5D).
Since changes in mitochondrial networks could be in uenced by abnormal ssion or fusion events (Giacomello et al., 2020), we evaluated expression of markers of such processes. TOMM20, used to determine mitochondrial networks by confocal microscopy and MiNA analysis ( Fig. 5A and C), had a 4fold increase in its mRNA level (p > 0.05) in MOI 0.01-infected cells. However, no changes were detected in TOMM20 protein levels by western blotting (Fig. 5E). We then assessed the expression of mitochondrial ssion-related genes. Fis1 and Drp1 mRNA were signi cantly increased by 4-fold and 3-fold, respectively at 24 hpi in HBMECs infected with the MOI 0.01, which did not translate to changes in Fis1 and Drp1 protein content (Fig. 5E) (Fig. 5E).

Discussion
Neurological consequences of COVID-19 still pose a relevant puzzle to medical and scienti c community.
Since its rst cases, CNS invasion has been described ( Exposure to SARS-CoV-2 led to augmentation of ACE2 and TMPRSS2 expression, which is consistent with our previous observations, in which HBMECs were exposed to the S1 subunit of Spike protein other angiogenic-related analytes were found to be increased in COVID-19 patients (including PTX3), which correlated with disease severity (Maldonado et al., 2022). Accordingly, VEGF levels were 8% increased in the supernatants of infected HBMECs, even though vegf transcripts remained unaltered. It is well-known that in ammation, especially IL-6-dependent, can stimulate defective angiogenesis (Gopinathan et al., 2015) and our data further contributes to the notion that following SARS-CoV-2 infection there is an intense brain endothelial activation which leads to defective angiogenic signaling and possibly endothelial permeability. Additionally, we found HIF-1α to be greatly increased after 24 hpi.
HIF-1α is a major angiogenesis inductor and is known to be up-regulated by distinct viral infections (reviewed by Reyes et al., 2020). HIF-1α is activated and translocated to the nucleus upon hypoxic conditions (Ke & Costa, 2006) and it has been shown that COVID-19 patients present massive hypoxia due to vasoconstriction and coagulopathy (Afsar et al., 2020). Interestingly, ACE2 expression is decreased in pulmonary smooth muscle cells upon HIF-1α accumulation , whereas hypoxia leads to a biphasic modulation of both ACE2 and TMPRSS2 expression on brain microvascular endothelial cells (hCMEC/D3), with an initial increase at 6 h and a decrease at 48 h of hypoxic stimulus cells (Imperio et al., 2021). These observations are in accordance with our present data, that ACE2 is decreased while HIF-1α is increased at 24 hpi. Although VEGF is one of the most described downstream targets of HIF-1α activation, apoptotic cell death and IFN-stimulated gene expression are additional targets of HIF-1α activation (Reyes et al., 2020), which can also be dependent on NF-κB signaling pathway (Walmsley et al., 2005). Our data indicates that HIF-1α up-regulation can be a part of a SARS-CoV-2-induced endothelial activation, along with cytokine/chemokine stimulation and NF-κB noncanonical activation.
Our nal series of experiments focused on mitochondrial morphology and dynamics in HBMECs following SARS-CoV-2 infection. It is well known that mitochondria are gatekeepers of BBB endothelium physiology and correspond to higher cytoplasmic volume as compared to non-cerebrovascular endothelial cells (Oldendorf et al., 1977; reviewed by Parodi-Rullan et al., 2021). Moreover, mitochondrial function is important for BBB maintenance and integrity (Doll et al., 2015). We rst employed a morphological/morphometrical approach to determine the mitochondrial contents and cellular distribution. Herein we demonstrated that direct exposure to SARS-CoV-2 led to a remodeling of mitochondrial networks. By using the MiNA plugin, we veri ed that infected HBMECs had increased mitochondrial footprint, as an estimation of overall TOMM20 pixel signal. Recent reports have also shown an effect of SARS-CoV-2 and COVID-19 on mitochondrial biology: monocytes isolated from COVID-19 patients display reduced mitochondrial membrane potential and SARS-CoV-2 viral load was positively correlated with generation of ROS (Romão et al., 2022). Importantly, endothelial cells exposed to SARS-CoV-2 Spike1 protein showed decreased tubular and increased fragmented mitochondrial networks in vitro, which was accompanied by a decrease in oxygen consumption rate and increase extracellular acidi cation rate (Lei et al., 2021). Con rming observation was recently described by ) and are correlated with abnormalities in mitochondrial quality control system, which in turn can lead to increased ROS production. Mitochondrial quality control encompasses biogenesis, ssion, fusion and mitophagy processes, which are essential for its biology and function. We analyzed markers of fusion and ssion processes in SARS-CoV-2-infected cultures and determined that while MOI 0.01 led to an increase in ssion-related gene expression (Fis1 and Drp1), this effect was not observed in protein levels or phosphorylation. However, mitofusin2 protein contents were found to be increased in MOI 0.1-infected cultures, which could explain the increased values in branch lengths.
Moreover, we found mitochondria associated to some extent to multivesicular bodies, which has also    Each point in C and D correspond to independent cultures and in B correspond to microscopic eld from four independent cultures. Representative blots in D from three independent experiments.