NOGOB receptor deficiency increases cerebrovascular permeability and hemorrhage via impairing histone acetylation–mediated CCM1/2 expression

The loss function of cerebral cavernous malformation (CCM) genes leads to most CCM lesions characterized by enlarged leaking vascular lesions in the brain. Although we previously showed that NOGOB receptor (NGBR) knockout in endothelial cells (ECs) results in cerebrovascular lesions in the mouse embryo, the molecular mechanism by which NGBR regulates CCM1/2 expression has not been elucidated. Here, we show that genetic depletion of Ngbr in ECs at both postnatal and adult stages results in CCM1/2 expression deficiency and cerebrovascular lesions such as enlarged vessels, blood-brain-barrier hyperpermeability, and cerebral hemorrhage. To reveal the molecular mechanism, we used RNA-sequencing analysis to examine changes in the transcriptome. Surprisingly, we found that the acetyltransferase HBO1 and histone acetylation were downregulated in NGBR-deficient ECs. The mechanistic studies elucidated that NGBR is required for maintaining the expression of CCM1/2 in ECs via HBO1-mediated histone acetylation. ChIP-qPCR data further demonstrated that loss of NGBR impairs the binding of HBO1 and acetylated histone H4K5 and H4K12 on the promotor of the CCM1 and CCM2 genes. Our findings on epigenetic regulation of CCM1 and CCM2 that is modulated by NGBR and HBO1-mediated histone H4 acetylation provide a perspective on the pathogenesis of sporadic CCMs.


Introduction
Cerebral cavernous malformations (CCMs) are common vascular malformations in the brain that affect 0.16%-0.9% of the population (1)(2)(3). CCMs arise primarily in the endothelium (4-7) and lesions appear in the brain as thin-walled, dilated blood vessels that promote vascular leakage (8)(9)(10)(11). As a consequence, CCMs have been implicated as playing a causal role in headaches, seizures, and acute hemorrhagic strokes (12)(13)(14)(15). The only treatment currently available for CCMs is neurosurgical lesion resection (16)(17)(18)(19). Unfortunately, surgical excision brings a high risk of brain damage, and not all cavernomas can be removed safely (16)(17)(18). Loss-of-function mutations of the CCM genes CCM1/KRIT1 (20), CCM2 (21), and CCM3/PDCD10 (22) at both postnatal and adult stages (4) have been demonstrated in the pathogenesis of CCMs. Based on Knudson's 2-hit hypothesis in CCM pathogenesis (23)(24)(25), the onset of CCMs requires the loss of function/expression of both CCM gene alleles. The first hit would be induced in all cells by a germline mutation that prevents the expression of 1 of the 2 alleles. The second hit would be induced by a somatic mutation or expression loss in the other allele that teams up with the first hit to completely impair CCM gene expression/function (23). Although somatic mutation of CCM genes is also detected in around half of sporadic CCM cases with multiple lesions (26)(27)(28), the molecular mechanism responsible for CCM loss in the rest of the sporadic CCM cases has not been elucidated to the best of our knowledge. Our previous study demonstrated that CCM1 protein expression could be decreased under long-term treatment with high glucose, and Ccm1 heterozygous mice exhibit cerebral hemorrhage under streptozotocin-induced diabetes (29). However, the underlying mechanisms regulating CCM gene expression remain unclear.
HBO1, a member of the MYST lysine acetyltransferase (KAT) family, is also known as histone acetyltransferase 7 (KAT7) and MYST family histone acetyltransferase 2 (MYST2). HBO1 was originally identified as histone acetyltransferase (HAT) binding-origin recognition complex 1 (ORC1) (30)(31)(32). Besides its implicated role in DNA replication within ORC1, many pieces of evidence support the idea that HBO1 is involved in transcriptional regulation. HBO1 forms a HAT complex with ING4/5, hEaf6, and JADE1/2/3 or Brpf1/2/3, targeting chromatin by binding lysine residues of histone H3 or H4 (33). The acetyltransferase activity of HBO1 is responsible for the acetylation of histone H4 at lysines 5, 8, and 12 (H4K5, H4K8, and H4K12) (34), and is essential for histone H3 acetylation at lysine 14 (H3K14) (35). These acetylated histone marks at both promoter and intragenic regions mediate cell-restricted gene expression. Altered HBO1 expression has been reported in human abdominal aortic aneurysm (36), a vascular disease closely related to endothelial dysfunction (37). Hbo1-knockout mice are embryonic lethal (35). Enlarged blood vessels were noted in the embryonic head, suggesting that The loss function of cerebral cavernous malformation (CCM) genes leads to most CCM lesions characterized by enlarged leaking vascular lesions in the brain. Although we previously showed that NOGOB receptor (NGBR) knockout in endothelial cells (ECs) results in cerebrovascular lesions in the mouse embryo, the molecular mechanism by which NGBR regulates CCM1/2 expression has not been elucidated. Here, we show that genetic depletion of Ngbr in ECs at both postnatal and adult stages results in CCM1/2 expression deficiency and cerebrovascular lesions such as enlarged vessels, blood-brain-barrier hyperpermeability, and cerebral hemorrhage. To reveal the molecular mechanism, we used RNA-sequencing analysis to examine changes in the transcriptome. Surprisingly, we found that the acetyltransferase HBO1 and histone acetylation were downregulated in NGBR-deficient ECs. The mechanistic studies elucidated that NGBR is required for maintaining the expression of CCM1/2 in ECs via HBO1-mediated histone acetylation. ChIP-qPCR data further demonstrated that loss of NGBR impairs the binding of HBO1 and acetylated histone H4K5 and H4K12 on the promotor of the CCM1 and CCM2 genes. Our findings on epigenetic regulation of CCM1 and CCM2 that is modulated by NGBR and HBO1-mediated histone H4 acetylation provide a perspective on the pathogenesis of sporadic CCMs. NOGOB receptor deficiency increases cerebrovascular permeability and hemorrhage via impairing histone acetylation-mediated CCM1/2 expression Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/JCI151382DS1).
As the gold-standard confirmatory measurement of BBB disruption, an Evans blue permeability assay showed dramatically increased Evans blue extravasation in the brain of Ngbr ECKO mice ( Figure 1D). Brain edema also was indicated by the significantly increased water content in the brains of Ngbr ECKO mice compared with Ngbr fl/fl littermate controls ( Figure 1E). FITC-conjugated dextran (FITC-dextran) perfusion showed that FITC-dextran was retained within brain capillary beds of littermate control mice but had infiltrated surrounding tissues around the lesion sites in Ngbr ECKO mouse brains, as marked with blue asterisks in Figure 1F. When stained with IgG, the Ngbr ECKO group displayed obvious positive IgG staining around the leaking capillaries, as marked with black asterisks in Figure 1G. These defects were observed in the brain of both postnatal and adult Ngbr ECKO mice.
Ngbr ECKO mouse brains exhibit enlarged microvessels, increased phosphorylation of MLC, and disrupted adherens junctions and tight junctions. To determine vascular morphology changes in the Ngbr ECKO mouse brain, mouse brain and retina sections were stained with the EC marker isolectin B4 (IB4). As shown in Figure  2A, Supplemental Figure 1, B and C, and Supplemental Videos 1 and 2, brain and retina microvessels were markedly enlarged in the lesion sites of Ngbr ECKO mice compared with Ngbr fl/fl mice. In Ngbr ECKO mouse brains, phosphorylation of MLC (phos-MLC) was induced in the ECs of lesion sites (Figure 2, B and C), suggesting a change in EC contractility. Moreover, immunofluorescent staining results showed a significant decrease in adherens junctions (AJs, represented by green VE-cadherin staining) and tight junctions (TJs, represented by green claudin-5 staining) between ECs in the lesions of Ngbr ECKO mouse brains (Figure 2, D-G). These junctional defects contributed to disrupted vessel integrity and increased BBB permeability observed in the brain of Ngbr ECKO mice, as shown in Figure 1. Electron microscopy images and quantification further revealed changes in the ECs in Ngbr ECKO mouse brains ( Figure 2, H and I). An elongated shape and giant nuclei of ECs and a rough luminal surface of capillaries were noted in the Ngbr ECKO group compared with that of Ngbr fl/fl controls. Intracytoplasmic canaliculi (caveolae), protoplasmic protrusions (filopodia), and gaps between ECs were observed in the Ngbr ECKO group but not in the Ngbr fl/fl control group. These findings indicate that Ngbr deficiency promotes the activation of ECs.
Ngbr deficiency diminishes the expression of CCM1 and CCM2 in brain ECs. The enlarged microvessels and cerebral hemorrhage phenotype in the brain of Ngbr ECKO mice were similar to vascular defects presented in the brain of EC-specific Ccm1-or Ccm2-knockout mice (4,7,40). Our previous work showed that NGBR was decreased in human CCM lesions, and Ccm1/2 was suppressed in Ngbr-deficient yolk sac (39). The pathological changes of brain ECs in Ngbr ECKO mice are identical to that of ECs from human CCM lesions (9)(10)(11), such as increased RhoA/phos-MLC signaling and junctional disruption (42)(43)(44).
To test the hypothesis that NGBR is required for regulating the expression of CCM genes in ECs, we extracted mouse brain microvascular ECs (MBMVECs) from both postnatal and adult Ngbr ECKO mice as well as littermate control mice. As compared with Ngbr fl/fl MBMVECs, the expression of Ccm1 and Ccm2 but not Ccm3 was HBO1-dependent histone H3K14 acetylation is essential for maintaining normal transcriptional activity during embryonic development (35). HBO1 is engaged in vascular endothelial growth factor receptor 2 (VEGFR2) transcription by mediating RNA polymerase II binding and the acetylation of histone H3K14 and histone H4 in the intragenic region of VEGFR2 (38).
As shown in our previous paper (39), the NOGOB receptor (NGBR) was decreased in the endothelial cells (ECs) of human CCM lesions. Like Ccm1/2 and Hbo1 (4,7,40), mice with EC knockout of Ngbr are embryonic lethal and present enlarged blood vessels in the head region of the embryos. Although we appreciated that NGBR loss results in the downregulation of CCM1 and CCM2 expression in ECs, the molecular mechanism by which NGBR regulates the expression of CCM1 and CCM2 genes has not been elucidated. Here, we demonstrated the potential link between NGBR loss and CCM pathogenesis by rescuing CCM1/2 expression in mice with inducible EC-specific knockout of Ngbr and elucidated the molecular mechanism by which NGBR regulates the transcription of CCM1/2 genes through HBO1mediated histone acetylation. We demonstrated that genetic depletion of Ngbr in mouse ECs at postnatal and adult stages results in downregulation of CCM1 and CCM2, and consequently blood-brain-barrier (BBB) hyperpermeability and cerebral hemorrhage. Furthermore, these CCM-related lesions can be rescued by adeno-associated virus-mediated (AAV-mediated) gene delivery of either CCM1/CCM2 or HBO1. Mechanistically, NGBR and HBO1 are required to epigenetically regulate the transcription of CCM1 and CCM2 genes via HBO1-mediated acetylation of histone H4K5 and H4K12. Our findings provide a perspective on epigenetic regulation of CCM gene transcription in the context of sporadic CCMs. Our data suggest that modulating NGBR-and HBO1-mediated histone H4 acetylation may be an epigenetic regulation related to the pathogenesis of certain types of sporadic CCMs promoted by CCM1/2 deficiency.

Results
EC-specific knockout of Ngbr results in hemorrhage and BBB disruption in the brain of postnatal and adult mice. Previously, we and others had shown that EC-specific deletion of Ngbr induced embryonic lethality before E12.5 and led to a cerebral vasculogenic defects at the embryonic stage (39,41). To further address the underlying molecular functions of Ngbr in the vasculopathy at the postnatal and adult stage, we conducted EC-specific Ngbr knockout in Cdh5-Cre ERT2 Ngbr fl/fl mice (hereafter termed Ngbr ECKO ) by intraperitoneal injection of tamoxifen at both the postnatal stage (P1) and the adult stage (8-12 weeks), as illustrated in Figure 1A. Ngbr fl/fl mice injected with tamoxifen served as littermate controls (hereafter referred to as Ngbr fl/fl ). Consequently, mice at both the postnatal and adult stages developed visible cerebral hemorrhage after the genetic deletion of Ngbr in ECs. As shown in Figure 1B, hemorrhage sites were randomly distributed in the brain; however, they were more frequently in the hindbrain. Lesions with red blood cell extravasation were readily detected in hematoxylin and eosinstained (H&E-stained) sections ( Figure 1C). Unlike the postnatal stage, hemorrhage in the brain of adult Ngbr ECKO mice was not very severe and more like cerebral microbleeds. Hemosiderin staining further indicated stale hemorrhage in adult mice (Supplemental Representative images of wholebrain fresh tissue from postnatal and adult mice and H&E staining of brain sections. Blue arrows point to bleeding sites (B) and bleeding spots observed on H&E images (C) in the Ngbr ECKO group. Scale bars: 100 μm (low-magnification images) and 50 μm (high-magnification images). (D and E) Evans blue extravasation and brain water content are increased in the brain of postnatal and adult Ngbr ECKO mice. Data are presented as mean ± SD, n = 6 per group for postnatal mice and n = 5 per group for adult mice. Significance was tested by 2-tailed, unpaired Student's t test. ***P < 0.001. (F) Representative images of immunofluorescent staining on 30-μm FITC-dextran-perfused brain sections showing vessel leakage in the Ngbr ECKO brain. ECs were labeled by CD31 immunostaining. Blue asterisks indicate the leaking sites. Scale bars: 100 μm. (G) Representative IgG staining images show IgG leakage into brain parenchyma in the brain of Ngbr ECKO mice. Black asterisks indicate leakage sites. Scale bars: 100 μm.
of NGBR-deficient HBMVECs and showed synergistic effects in reducing the hyperpermeability to a greater extent when overexpressing both CCM1 and CCM2. Therefore, we used CCM1 and CCM2 overexpression for the following rescue experiments. As a result, overexpression of CCM1 and CCM2 eliminated the defects in AJs (VE-cadherin) and TJs (ZO-1) ( Figure 4B) and diminished the induction of phos-MLC in NGBR-deficient HBMVECs ( Figure  4C). The quantitative changes in RhoA and phos-MLC were further determined by Western blotting analysis (Figure 4, D and E). These results show that CCM1 and CCM2 overexpression suppresses the induction of RhoA and phos-MLC in NGBR-deficient HBMVECs.
The rescue effects of CCM1 and CCM2 overexpression were further confirmed in vivo. As reported by Korbelin et al. and Dogbevia et al., AAV serotype BR1 (AAV-BR1) specifically targets brain ECs and has been used for brain EC-specific gene overexpression (46,47). We generated AAV-BR1 harboring either CCM1 (AAV-BR1-CCM1, AAV-CCM1) or CCM2 (AAV-BR1-CCM2, AAV-CCM2) genes. AAV-BR1 expressing GFP (AAV-BR1-GFP) was used as a control (AAV-ctrl). As shown in Supplemental Figure 5, the efficiency and tissue specificity of AAV-BR1-GFP expression were determined by examining GFP expression in various tissues. At 2 weeks after tail vein injection of AAV-BR1-GFP, positive GFP staining was exclusively present in brain ECs but not in any ECs in the heart, lung, liver, and kidney. A schematic of the protocol for animal treatment is shown in Figure 5A. For in vivo rescue experiments, 1 week after AAV-BR1 injection, tamoxifen administration was used to induce genetic deletion of Ngbr in ECs. The overexpression of CCM1 and CCM2 was detected in MBMVECs extracted from the brains of mice injected with both AAV-BR1-CCM1 and AAV-BR1-CCM2 compared with AAV-ctrl-injected mice ( Figure  5B). As shown in Figure 5C, CCM1 and CCM2 overexpression reduced the hemorrhagic lesions in the brain of Ngbr ECKO mice, as visualized by whole-mount brain images and H&E staining. As shown in Figure 5, D and E, CCM1 and CCM2 overexpression in the brain ECs of Ngbr ECKO mice remarkably attenuated the BBB leakage and brain edema, as determined by Evans blue extravasation assay and water content measurement, respectively. The reduced IgG staining further confirmed the effects of CCM1 and CCM2 overexpression on rescuing the BBB integrity of Ngbr ECKO mice ( Figure 5F). Western blotting results and immunofluorescence staining showed the restoration of AJs (VE-cadherin) and TJs (claudin-5) protein levels, as well as a reduction in phos-MLC in mouse brain ECs (Supplemental Figure 6, A-C) and MBMVECs extracted from the brain of Ngbr ECKO mice overexpressing CCM1 and CCM2 ( Figure 5G). These results indicated that downregulation of CCM1 and CCM2 can be attributed to the Ngbr deficiencypromoted BBB disruption and cerebral hemorrhage.
HBO1-mediated histone acetylation is essential for NGBR deficiency-promoted transcriptional downregulation of CCM1 and CCM2 in HBMVECs. Since NGBR is the receptor for soluble NOGOB (sNOGOB), we detected NOGOB expression in Ngbr ECKO MBMVECs and plasma concentration of sNOGOB in Ngbr ECKO mice. The results showed that EC Ngbr knockout neither influences NOGOB expression in MBMVECs nor the serum concentration of sNOGOB (Supplemental Figure 7, A and B). Consistently, NGBR knockdown did not affect the NOGOB expression in HBMVECs (Supplemental Figure 7C). These results indicate that markedly decreased in Ngbr-deficient MBMVECs isolated from the brain of either postnatal or adult Ngbr ECKO mice ( Figure 3, A and B, and Supplemental Figure 2A). In human brain microvascular ECs (HBMVECs) in vitro, NGBR was knocked down with NGBR siRNA, the efficacy of which has been optimized and validated in our previous publications (45). The mRNA and protein levels of CCM1 and CCM2, but not CCM3, were also significantly decreased in NGBR-deficient HBMVECs ( Figure 3, C-E, and Supplemental Figure 2, B and C).
Consistent with the in vivo study, NGBR knockdown in HBM-VECs significantly increased the monolayer permeability (Figure 3F), increased RhoA/phos-MLC signaling ( Figure 3, G-I), and impaired the formation of AJs (represented by VE-cadherin staining) and TJs (represented by ZO-1 staining) ( Figure 3J). Like NGBR deficiency, knockdown of either CCM1 or CCM2 in HBM-VECs resulted in increased permeability (Supplemental Figure  3A). As shown in Supplemental Figure 3B, immunostaining signals of phos-MLC were increased in CCM1-and CCM2-knockdown HBMVECs. Correspondingly, knockdown of either CCM1 or CCM2 increased the RhoA protein levels in HBMVECs (Supplemental Figure 3, C and D). However, CCM1 or CCM2 knockdown did not affect NGBR expression in HBMVECs (Supplemental Figure 3, C and D), suggesting that NGBR is an upstream regulator of CCM1 and CCM2.
Overexpression of CCM1 and CCM2 ameliorates Ngbr deficiency-promoted vascular defects in vitro and in vivo. To further confirm the contributions of CCM1 and CCM2 downregulation to the pathogenesis of NGBR-deficient ECs, we generated lentiviruses harboring either the CCM1 or CCM2 transgene and confirmed the overexpression of CCM1 and CCM2 proteins in HBMVECs in vitro. As shown in Supplemental Figure 4, A and B, the efficiency of lentivirus-mediated overexpression of CCM1 and CCM2 was sufficient. The overexpression of CCM1 and CCM2 did not affect NGBR expression in HBMVECs. As shown in Figure 4A, overexpression of either CCM1 or CCM2 reduced the hyperpermeability Figure 2. EC-specific Ngbr-knockout mice exhibit enlarged microvessels, increased phosphorylation of MLC, and disrupted AJs and TJs in the brain. (A) Immunofluorescent staining showing enlargement of microvessels in the brain of Ngbr ECKO mice. IB4 (red) was used to label ECs and DAPI (blue) was used to label nuclei. Scale bars: 20 μm. (B-G) Immunofluorescent staining and quantification results showing increased phos-MLC and impaired AJs (VE-cadherin, VE-cad) and TJs (claudin-5, Cldn5) coverage of ECs in the brain lesions of Ngbr ECKO mice. Tissue sections were stained for phos-MLC (green)/CD31 (red)/DAPI (blue), claudin-5 (green)/CD31 (red)/ DAPI (blue), or VE-cadherin (green)/CD31 (red)/DAPI (blue). Scale bars: 20 μm (B, D, and F). Results were quantified using ImageJ software. Data are presented as mean ± SD, n = 6 per group . Significance was tested by 2-tailed, unpaired Student's t test. ***P < 0.001 (C, E, and G). (H and I) Electron microscopy images of microvessels in the brain and quantification of filopodia (Fil) and caveolae (Cav) in both Ngbr fl/fl and Ngbr ECKO groups. Electron microscopy images show enlarged vessel lumen (Lum) and irregular endothelial shape with a rough luminal surface in the brain of Ngbr ECKO mice compared with that of Ngbr fl/fl mice. Caveolae, filopodia, and junction gaps (Gap) between ECs were observed in the Ngbr ECKO group but not in the littermate control group (Ngbr fl/fl ). RBC, red blood cells; Nuc, nucleus. Scale bars: 2 μm (low-magnification images and 0.5 μm (high-magnification images). n = 10 views randomly selected from 3 mice per group. Significance was tested by 2-tailed, unpaired Student's t test. **P < 0.01, ***P < 0.001.
NGBR-mediated CCM1/2 expression regulation is independent of the alteration of NOGOB.
To elucidate the underlying mechanisms by which NGBR regulates the transcription of CCM1 and CCM2, we carried out RNA sequencing (RNA-seq) to examine transcriptome alterations in NGBR-deficient HBMVECs. The molecular function enrichment analysis of differentially expressed genes (DEGs) showed high clustering in microtubule binding, cytoskeleton protein binding, and microtubule motor activity ( Figure 6A), which are closely associated with RhoA/phos-MLC signaling (48)(49)(50)(51)(52). Importantly, gene set enrichment analysis (GSEA) of these RNA-seq data showed a good correlation with the DEGs from human CCM lesions (53), with a normalized enrichment score (NES) of 1.75 and nominal P value and FDR q value both less than 0.01 ( Figure 6B). The results of this comprehensive analysis provide a second layer of evidence to support the correlation between NGBR deficiency in ECs and CCM pathogenesis. However, why is NGBR required for preserving the expression of CCM1 and CCM2 in brain ECs? We further searched the RNA-seq data for transcriptional regulators and surprisingly found the HAT HBO1 remarkably decreased in NGBR-deficient HBMVECs compared with control groups. In contrast, other HATs and histone deacetylases (HDACs) were not significantly altered. The expression of genes encoding HATs (including KAT1, GCN5, PCAF, CREBBP, EP300, KAT5, KAT6A, HBO1, and KAT8) and HDACs (including HDAC1-11 and sirtuins [SIRTs], including SIRT1-7) in NGBR-deficient HBMVECs was further examined by reverse transcription quantitative polymerase chain reaction (RT-qPCR) and Western blotting, respectively. The mRNA and protein expression levels showed that HBO1 is the HAT significantly decreased in NGBR-deficient HBMVECs (Figure 6, The overexpression of CCM1 and CCM2 genes using lentiviruses ameliorated NGBR deficiency-promoted HBMVEC hyperpermeability, which was determined by EC-monolayer permeability assay. Statistical significance was determined by 1-way ANOVA and Dunnett's post hoc test. Data are presented as mean ± SD. ***P < 0.001 versus siCtrl-and lentivirus vector control-treated (lenti-vector) group; ### P < 0.001 versus siNGBR-and lenti-vector-treated group; † † † P < 0.001 versus siNGBR-and CCM1 lentivirus-treated (lenti-CCM1) group; § § P < 0.01 versus siNGBR-and CCM2 lentivirus-treated (lenti-CCM2) group. Based on the reported HAT activity of HBO1 (33-35), we examined the effects of either HBO1 or NGBR knockdown on the acetylation of H3K9/K14/K18/K27 and H4K5/K8/K12 in HBM-VECs. Western blotting analysis showed that either HBO1 or NGBR deficiency results in a similar alteration of histone acetylation, i.e., a significant decrease in acetylated H3K14 (H3K14ac), H4K5ac, and H4K12ac ( Figure 6, J-O). Taken together, these findings indicated that NGBR may regulate the transcription of CCM1 and CCM2 through HBO1-mediated histone acetylation.
The enrichment of HBO1-mediated H4K5 and H4K12 acetylation on the promoter regions of CCM1 and CCM2 genes. It has been demonstrated that particular patterns of histone posttranslational modifications represent a code that is recognized by transcription factors via specific chromatin-binding domains (35,57,58). The enrichment of acetylated histone protein at the promoter region is essential for regulating gene transcription (59)(60)(61). ChIP-qPCR assay using an anti-HBO1 antibody could determine the binding of HBO1 on the promotor region of CCM1 and CCM2. The results showed that HBO1 binds the region -921 to -809 of the CCM1 promotor and the region -1035 to -920 of the CCM2 promotor, and HBO1 binding was significantly decreased in NGBR-deficient HBMVECs (Figure 7, A, B, D, and E). However, the binding of HBO1 on the CCM3 promotor region was not altered under the NGBR-knockdown condition (Supplemental Figure 10A). As shown in Figure 6, J-L, acetylation of H3K14, H4K5, and H4K12 is dependent on the expression of HBO1 in HBMVECs. We further carried out the ChIP-qPCR assays using anti-H3K14ac, -H4K5ac, and -H4K12ac antibodies. ChIP-qPCR results showed the enrichment of H4K5ac and H4K12ac on the promotor region of CCM1 and CCM2, respectively (Figure 7, C and F). Like HBO1, the binding of H4K5ac and H4K12ac was significantly decreased in NGBRdeficient HBMVECs (Figure 7, B, C, E, and F). In contrast, NGBR knockdown in HBMVECs did not affect the binding of H3K14ac on the promotor region of CCM1 and CCM2, nor did the binding of H3K14ac, H4K5ac, H4K8ac, and H4K12ac on the promotor region of CCM3 (Supplemental Figure 10, B-E). These results indicated that NGBR regulates the transcription of CCM1 and CCM2 but not CCM3 via HBO1-mediated acetylation of H4K5 and H4K12.
To further confirm whether NGBR is dependent on HBO1 in regulating the transcription of CCM1 and CCM2, we carried out HBO1 rescue experiments. We first generated lentivirus carrying the HBO1 transgene for restoring HBO1 expression in NGBRdeficient HMBVECs in vitro. As we expected, HBO1 overexpression sufficiently restored the expression of CCM1 and CCM2 in NGBR-knockdown HBMVECs (Figure 7, G-I). As noted herein, HBO1 overexpression did not affect NGBR expression in HBM-VECs. These data demonstrated that NGBR is dependent on HBO1 in regulating the expression of CCM1 and CCM2.
HBO1 overexpression restored CCM1/2 expression and vascular defects in the brain of Ngbr ECKO mice. To determine the rescue effect of HBO1 in vivo, HBO1 was overexpressed in brain ECs using AAV-BR1-HBO1-GFP (AAV-HBO1) 1 week before tamoxifen-induced deletion of Ngbr in ECs. The schematic protocol for animal treatment is shown in Figure 8A. The efficacy and specificity of AAV-BR1-mediated delivery of GFP and HBO1 genes in ECs were C-E, and Supplemental Figure 8, A-C). Moreover, Hbo1 was also significantly decreased in MBMVECs isolated from the brain of either postnatal or adult Ngbr ECKO mice ( Figure 6F). To determine the relationship between HBO1 and CCM1/2 expression, HBO1 was knocked down in HBMVECs with validated HBO1 siRNA. The RT-qPCR and Western blotting results demonstrated that HBO1 knockdown resulted in downregulation of CCM1 and CCM2 expression ( Figure 6, G-I). Of note, HBO1 knockdown did not affect NGBR expression ( Figure 6, H and I). As shown in Supplemental Figure 8, C and D, another HAT-encoding gene, GCN5, showed a slight decrease under NGBR knockdown. However, unlike HBO1 ( Figure 6, G-I), GCN5 knockdown did not affect the expression of CCM1 and CCM2 (Supplemental Figure 8E). These results suggest that HBO1 may be an intermediate in the NGBR-mediated pathway to regulate the transcription of CCM1 and CCM2.
As shown in our previous publication (54), NGBR has a hydrophobic cytoplasmic domain that binds to farnesylated proteins, which are involved in regulating the expression of transcription factors (55,56). SREBP-1c is one of the NGBR-dependent transcription factors, and the HBO1 promotor region has an SREBP-1c binding site (SRE-1), which is not present in the promotor regions of CCM1/2/3. As shown in Supplemental Figure 9, loss of NGBR downregulated SREBP-1c expression (Supplemental Figure 9, A and B), while knockdown of SREBP-1c using siRNA attenuated the expression of HBO1 in HBMVECs (Supplemental Figure 9, C and D). As noted herein, SREBP-1c knockdown did not affect the expression of NGBR (Supplemental Figure 9, C and D). Furthermore, ChIP-qPCR assay results showed that loss of NGBR impairs the enrichment of SREBP-1c on the promotor of the HBO1 gene (Supplemental Figure 9, E and F). These Figure 5. The overexpression of CCM1 and CCM2 in brain ECs diminishes Ngbr deficiency-promoted microvessel leakage and hemorrhage in vivo. CCM1 and CCM2 overexpression in brain ECs was achieved by AAV-CCM1 and AAV-CCM2 administration. AAV-BR1-GFP was used as control (AAVctrl). (A) Tamoxifen was injected 1 week after AAV injection, and mice were euthanized 3 weeks after tamoxifen injection as shown in the diagram protocol. (B) The efficacy of AAV-BR1-mediated overexpression of CCM1 and CCM2 was determined by Western blotting of the lysates of MBMVECs extracted from the brain of mice injected with AAV-CCM1 and -2 or AAVctrl. The results demonstrated sufficient overexpression of CCM1 and CCM2 in MBMVECs in vivo. (C) Representative images of hemorrhage in fresh brain tissues and H&E staining. Hemorrhage sites were observed in the brain of AAV-ctrl-injected Ngbr ECKO mice. In contrast, scarce hemorrhage sites were observed in the brain of the AAV-CCM1/2-injected Ngbr ECKO group. Scale bars: 100 μm. (D and E) CCM1 and CCM2 overexpression significantly diminished the Ngbr deficiency-promoted hyperpermeability, determined by calculation of Evans blue extravasation and brain water content. Data are presented as mean ± SD, n = 5 mice per group. Significance was tested by 1-way ANOVA with Dunnett's post hoc test. ***P < 0.001 versus Ngbr fl/fl mice treated with AAV-ctrl; ### P < 0.001 versus Ngbr ECKO vitro (C, E, G, I, K, L, N, and O) and n = 6 mice per group in vivo (F). Significance was tested by 2-tailed, unpaired Student's t test. *P < 0.05, **P < 0.01, ***P < 0.001. J Clin Invest. 2022;132(9):e151382 https://doi.org/10.1172/JCI151382 CCM1/2 expression in brain ECs, and HBO1 overexpression can rescue the CCM-related lesions caused by Ngbr deficiency in ECs.

Discussion
NGBR is a transmembrane protein involved in stimulating chemotaxis and morphogenesis of ECs (62). It is essential for embryonic and vascular development in mice and zebrafish (39,41,63). Our previous study indicated that NGBR expression decreased in the lesions of the human CCM (39). In this study, we provided critical insights into our previous findings, including comprehensive characterization of vascular lesions in the brain of Ngbr ECKO mice at both postnatal and adult stages, demonstrating CCM1 and CCM2 as critical molecules in Ngbr deficiency-induced endothelial dysfunction and cerebral hemorrhage, and providing what we believe is a new perspective on CCM pathogenesis in the context of HBO1-mediated histone acetylation in preserving the expression of CCM genes. determined by GFP immunofluorescence staining ( Figure 8B). As shown in Figure 8B, GFP was specifically expressed in brain ECs. As visualized by the whole-mount picture and H&E staining (Figure 8C), the brain of Ngbr ECKO mice receiving AAV-HBO1 had much fewer hemorrhagic lesions compared with Ngbr ECKO mice receiving the AAV-BR1-GFP control virus. The rescue effects of HBO1 overexpression on BBB integrity were further determined by Evans blue extravasation assay, water content, and IgG staining. Consistently, AAV-mediated HBO1 overexpression could prevent the onset of BBB damage in the brain of Ngbr ECKO mice, as shown by the decreased permeability ( Figure 8D), reduced water content (Figure 8E), and less IgG staining ( Figure 8F). The RT-qPCR results further confirmed that HBO1 overexpression restored CCM1/2 expression in MBMVECs isolated from the brains of Ngbr ECKO mice treated with AAV-HBO1 ( Figure 8G). These data further demonstrated that HBO1 is essential for NGBR-dependent regulation of Figure 7. HBO1-mediated histone H4 acetylation is required for NGBR-regulated CCM1 and CCM1 expression. (A-F) ChIP-qPCR assays were performed for HBMVECs treated with either control siRNA or NGBR siRNA. Chromatin DNA was pulled down using respective antibodies against IgG, HBO1, H3, H3K14ac, H4K5ac, and H4K12ac. Results showed that NGBR knockdown ameliorated the binding of HBO1, H4K5ac, and H4K12ac on the promotor region of the CCM1 gene (-921 to -809) (A-C) and on the promotor region of CCM2 (-1035 to -920) (D-F). Data are presented as mean ± SD, n = 3 samples per group. Significance was tested by 2-tailed, unpaired Student's t test (B, C, E, and F). **P < 0.01, ***P < 0.001. (G-I) Lentivirus-mediated overexpression of HBO1 (lenti-HBO1) restored CCM1 and CCM2 gene transcription and protein expression in NGBR-deficient HBMVECs. Data are presented as mean ± SD, n = 3 samples per group. Significance was tested by 1-way ANOVA with Dunnett's post hoc test. *P < 0.05, ***P < 0.001 versus siCtrl-and lenti-vectortreated group; ### P < 0.001 versus siNGBR-and lenti-vector-treated group.
tion factors and promotes the transcription of EC genes (66)(67)(68). Histone acetylation is regulated by HATs and HDACs and is a chromatin modification involved in transcriptional activation without a change in the gene sequence (69). Arts et al. firstly reported histone H4 acetylation involved in tissue-type plasminogen activator (t-PA) gene expression in human ECs in 1995 (70). Subsequently, growing evidence has shown that acetylated histone-mediated gene expression participates in multiple endothelial functions, including endothelial barrier function. Acetylated histone H3K9 and H3K18 were demonstrated to be involved in the expression regulation of endothelial junction proteins such as claudin-5 and VE-cadherin (71,72). Several physiopathological stimuli, including hypoxia (73), sheer stress (74,75), reactive oxygen species (76), inflammatory cytokine (77), and high glucose (78,79) were shown to regulate EC gene expression via altering the histone acetylation. Our recent review article provides a comprehensive update on the contributions of histone acetylation to EC functions (80). The enrichment of different histone acetylation marks at the proximal promoter regions, and even intragenic regions, corresponds to the changes in gene expression levels in ECs (81,82).
To reveal the molecular mechanism by which NGBR regulates the transcription of CCM1/2 genes, we firstly made an attempt to investigate all possible transcription factors in our RNA-seq data. GSEA results showed that NGBR deficiency does not result in a significant change in CCM-related transcription factors (83) in HBMVECs (Supplemental Figure 11A). Furthermore, we chose 2 transcription factors, CREB1 and ATF2, which are obviously downregulated in NGBR-deficient HBMVECs, and used the siRNA-mediated gene knockdown approach to determine whether and the extent to which CREB1 and ATF2 would regulate CCM1 and CCM2 expression. The knockdown of either CREB1 or ATF2 had no effect on CCM1 and CCM2 expression (Supplemental Figure 11, B and C). Moreover, our RNA-seq data revealed the significant suppression of HBO1 transcription in NGBR-deficient HBMVECs. HBO1, a member of the MYST KAT family, is responsible for the bulk of acetylation at H4K5, H4K8, H4K12 (34), and H3K14 (35). Therefore, HBO1 is essential for embryonic development and blood vessel formation (35). As reported by Kueh et al., Hbo1 knockout in mice was embryonic lethal, and the embryos developed enlarged blood vessels in the head region (35). Matthew et al. further demonstrated that HBO1-mediated histone acetylation regulates EC gene expression (38). In addition, Han et al. reported altered HBO1 expression in human abdominal aortic aneurysm (36), a vascular disease closely related to endothelial dysfunction (37). These pieces of evidence indicate a crucial role of HBO1 in regulating endothelial function. In this study, the results of HBO1 knockdown and HBO1 overexpression established the link between HBO1 and the expression of CCM1 and CCM2 in the context of NGBR deficiency-promoted CCM-related lesions. The contributions of HBO1-mediated histone acetylation to the transcriptional regulation of CCM1 and CCM2 genes were further supported by ChIP-qPCR results. These findings revealed an epigenetic regulation of CCM genes in an NGBR/ HBO1-dependent manner.
Unlike genetic mutation, epigenetic disorders such as histone acetylation alterations are reversible. The tipping point of histone acetylation is maintained by HAT and HDAC enzyme families.
CCM is a vascular lesion that originates from ECs in the brain. Histologically identical lesions with characteristic enlarged popcorn-like capillaries are associated with all genotypes related to CCM genes (64). The lesions are isolated or clusters of enlarged leaking endothelial lumens at the capillary level, with disrupted and reduced EC junctions (8)(9)(10)(11). Stockton et al. showed that Rho kinase hyperactivity occurs in sporadic and familial human CCM endothelium as judged by phos-MLC levels. Mutations in CCM1 and CCM2 genes resulting in Rho kinase hyperactivation and the Rho kinase inhibitor fasudil were found to have a rescue effect in cell culture and animal models (43). Borikova and colleagues reported a marked increase in total RhoA protein levels after the loss of expression of CCM1, CCM2, or CCM3 and demonstrated that knockdown of RhoA effectively reverses endothelial dysfunction caused by CCM deficiency (42). McDonald et al. also presented evidence that the RhoA/phos-MLC pathway is activated in patients with sporadic CCM (65). These findings indicate that aberrant Rho-kinase activation caused by the functional loss of CCM genes are equivalent in terms of contributing to both familial and sporadic lesions. These pathological signatures of capillary-based vascular defects exhibiting enlarged and leaking endothelial lumens with hyperactivated RhoA/phos-MLC signaling and disrupted intracellular junctions are also present in the brain of Ngbr ECKO mice. GSEA comparing NGBR-deficient EC RNA-seq data with human CCM lesion DEGs strengthened correlations between NGBR loss and CCM pathogenesis. Rescue effects of CCM1 and CCM2 overexpression on Ngbr deficiencyinduced endothelial defects and cerebrovascular lesions further demonstrated that CCM1 and CCM2 downregulation contributes to the pathogenesis of CCM-related vascular lesions in the brain of Ngbr ECKO mice. Our data suggest NGBR is an upstream regulator of the CCM1 and CCM2 genes.
Histone lysine residue acetylation, an epigenetic marker on chromatin, creates binding sites for the recruitment of transcrip- Figure 8. HBO1 overexpression in vivo ameliorates Ngbr deficiencypromoted cerebral hemorrhage and BBB leakage. HBO1 overexpression in brain ECs of Ngbr ECKO mice was achieved by the intravenous administration of AAV-BR1-GFP-HBO1 (AAV-HBO1). AAV-BR1-GFP was used as a control. (A) Tamoxifen was injected 1 week after AAV injection, and mice were euthanized 3 weeks after tamoxifen injection as shown in the protocol schematic. (B) Efficacy and localization of AAV-BR1 were determined by immunofluorescent staining of GFP. The images of GFP staining in brain sections showed HBO1-GFP expression in brain ECs (labeled by CD31, red) after tail vein injection of AAV-HBO1. Scale bars: 25 μm. (C) Representative whole-brain images and H&E staining showing hemorrhage sites in AAV-ctrl-injected Ngbr ECKO mice, while no obvious hemorrhage sites were observed in AAV-HBO1-injected Ngbr ECKO mice. Scale bars: 100 μm. (D and E) HBO1 overexpression significantly diminished the hyperpermeability in the brains of Ngbr ECKO mice. Permeability was determined by the quantification of Evans blue extravasation and brain water content. (F) The results of IgG staining showing that AAV-HBO1 injection reduced the IgG-positive staining in the brains of Ngbr ECKO mice compared with AAV-ctrl-injected Ngbr ECKO mice. Scale bars: 200 μm. (G) Ccm1/2 mRNA expression was determined in MBMVECs extracted from mice. HBO1 overexpression in vivo rescued Ccm1 and Ccm2 transcription in Ngbr ECKO MBMVECs. Data are presented as mean ± SD, n = 3 mice per group. Significance was tested by 1-way ANOVA with Dunnett's post hoc test (D, E, and G). *P < 0.05, **P < 0.01, ***P < 0.001 versus Ngbr fl/fl mice treated with AAV-ctrl; # P < 0.05, ## a potential target for preventing the onset and progression of sporadic CCM promoted by CCM1 and CCM2 deficiency.

Animals
Cdh5-Cre ERT2 Ngbr fl/fl mice were generated as previously reported (39). To knock out Ngbr in ECs in vivo, neonatal mice from mated Cdh5-Cre ERT2 Ngbr fl/fl and Ngbr ECKO adult mice were injected with tamoxifen (T5648, Sigma-Aldrich). Fresh tamoxifen solution (20 mg/mL) was prepared in sterile corn oil. Preliminary results showed that postnatal Ngbr ECKO mice started to show symptoms like seizure and ataxia as well as exhibit vessel dilation histologically starting on P6, and most of them died after P12. Adult mice did not show any behavioral symptoms, but they also encountered survival issues nearly 4 weeks after inducible depletion of Ngbr. We checked adult mouse brains at 2, 3, and 4 weeks after tamoxifen injection. Vessel lesions were successfully induced at 3 weeks after tamoxifen injection and most of the mice survived until 4 weeks after inducible depletion. So, we chose P7-P12 as the timing of analyses for the postnatal model and 3 weeks after tamoxifen injection as the timing of analyses for adult mice. Neonatal mice from mated Cdh5-Cre ERT2 Ngbr fl/fl females and males were injected with tamoxifen (50 μg per mouse) for 3 consecutive days after P1, and genotype was determined when euthanized (P7-P12). Alternatively, Ngbr ECKO adult mice at 8-12 weeks were injected with tamoxifen (2 mg per mouse) for 5 consecutive days and euthanized after 3 weeks.
In some experiments, adult mice were injected via tail vein with AAV-BR1 expressing CCM1 (2.5 × 10 11 gc/mouse), CCM2 (1.8 × 10 11 gc/ mouse), HBO1 (2.5 × 10 11 gc/mouse), or AAV-ctrl 1 week before tamoxifen injection. Postnatal mice were genotyped after euthanasia, and adult mice were ear tagged. We performed blinded analyses for the experiments afterward. We did not exclude animals except for unexpected death due to the intrinsic lethality issues.

Evans blue extravasation and FITC-dextran perfusion
Neonatal and adult Ngbr fl/fl or Ngbr ECKO mice with or without AAV-BR1 injection were weighed and anesthetized with isoflurane. Evans Blue (E2129, Sigma-Aldrich) solution (2% in sterile PBS; 10 μL/g body weight) was injected into the retro-orbital plexus of neonatal mice or the tail vein of adult mice. Two hours (for neonatal mice) or 4 hours (for adult mice) after injection, mice were euthanized and transcardially perfused with PBS. To determine water content, brain tissues were harvested, weighed, and dried at 60°C for 48 hours and weighed again. Evans blue was extracted using 1 mL formamide at 55°C for 16 hours and determined by absorbance at 630 nm with a spectrophotometer. FITC-dextran (50 mg/mL in sterile PBS; FD2000S, The respective function of HATs and HDACs is to add or remove the acetyl group to or from the lysine residue, resulting in chromatin opening or closing for initiating transcription. Inhibition of the enzyme's activity is a way to alter histone acetylation status. A wide spectrum of HDAC inhibitors is available to target different HDACs specifically. Among commonly used HDAC inhibitors, such as trichostatin A (TSA), suberoylanilide hydroxamic acid (SAHA), and nicotinamide (NAM), SAHA has been approved in clinical trials for cancer treatment (84,85). Of note, NAM restored histone H4K12 acetylation and CCM1/2 expression in NGBR-deficient HBMVECs (data not shown).
As reported previously, most CCMs exhibit a 2-hit mechanism (23)(24)(25). However, somatic mutation of CCM genes is detected in only around half of sporadic CCM cases with multiple lesions (26)(27)(28), which means that the molecular mechanism responsible for CCM loss in the other half of sporadic CCM cases with multiple lesions and most of the sporadic CCM cases with a single lesion has not been elucidated. Although in recent years Knudson's 2-hit mechanism has been well studied, besides somatic mutation, it has become clear that epigenetic alterations are additional mechanisms for gene silencing (86)(87)(88). Our previous study demonstrated that CCM1 protein expression decreased under long-term treatment with high glucose, and Ccm1 heterozygous mice exhibit cerebral hemorrhage under streptozotocin-induced diabetes (29), which indicated that environment stimulation like high glucose might also serve as a second hit. The mechanisms of sporadic CCM are complicated and still under investigation. More new genes have been found to promote the pathogenesis of sporadic CCM, and even "3-hit mechanisms" have recently been proposed (89). Because there are many unknown contributors, the strategy to upregulate CCM genes might not work for all cases of sporadic CCM, especially those with CCM gene mutation or those with CCM downstream gene mutation, such as gainof-function MEKK3 (MAP3K3) mutations (90,91). Our findings of histone acetylation-mediated regulation of CCM1 and CCM2 genes provide a perspective on the epigenetic contributions to certain types of sporadic CCM pathogenesis, indicating epigenetic interventions may be an alternative approach for preventing the onset of CCM disease. However, our current studies only focus on the early-stage lesions such as CCM1/2 deficiency-promoting EC junction defects and enlarged leaking cerebrovascular lesions. We may need brain EC-specific Ngbr-knockout mice to reveal the late-stage lesions in our future investigations. Our findings also still need further investigation to support their clinical relevance. Meanwhile, our rescue experiments only provide a proof of principle for the contributions of CCM1/2 and HBO1 deficiency to the pathogenesis of cerebrovascular malformations in the brain of Ngbr ECKO mice. Further investigations are needed to support the therapeutic implications.
In summary, our study demonstrated that NGBR-and HBO1-mediated histone acetylation is required for preserving the expression of CCM1 and CCM2 in ECs. Downregulation of CCM1 and CCM2 contributes to the vascular lesions occurring in the brains of Ngbr ECKO mice. Our data provide an insight into the underlying mechanism by which histone acetylation regulates the transcription of CCM1 and CCM2 genes. Our findings suggest that NGBR-and HBO1-mediated histone H4 acetylation may be MBMVEC extraction MBMVECs were extracted from Ngbr fl/fl and Ngbr ECKO mice with or without AAV-BR1 injection as previously described (92,93). Briefly, mouse brains were collected and minced into small pieces after careful removal of the brainstem, surface vessels, and leptomeninges. Tissues were then digested with 1 mg/mL collagenase II (17101015, Thermo Fisher Scientific) and 10 μg/mL DNase I (10104159001, Sigma-Aldrich) at 37°C for 2 hours. After centrifugation at 500g for 5 minutes at 4°C, the lower layer was resuspended in 17% Percoll (P1644, Sigma-Aldrich) and centrifuged at 1000g for 15 minutes at 4°C. The precipitate was collected, digested with 1 mg/mL collagenase (Dispase, SCR139, Sigma-Aldrich) and 0.01 mg/mL DNase I at 37°C for 1 hour and then centrifuged at 500g for 5 minutes at 4°C. The precipitate was resuspended in a Percoll gradient and centrifuged at 1000g for 15 minutes at 4°C. Cells suspended at the interphase were harvested, centrifuged at 500g for 5 minutes at 4°C, and then lysed for Western blotting or mRNA isolation.

EC monolayer permeability assay
The integrity of the HBMVEC monolayer was assessed by the penetration rate of FITC-dextran (FD40S, Sigma-Aldrich) through monolayer cells following a previous report (94). Briefly, HBMVECs in different groups were seeded confluently on Transwell inserts (24-well format; 0.4 mm pore; 3467, Corning). Fresh culture medium (150 mL) with 100 μg/mL FITC-dextran was added into the upper chamber, and 1 mL fresh culture medium without FITC-dextran was added to the lower chamber. Samples (50 μL) were retrieved from the lower chamber at 0, 30, and 60 minutes and the medium replaced with fresh culture medium. Samples were then analyzed by a fluorescence microplate reader (PerkinElmer) with a wavelength setting of 488/510 nm (ex/em). The permeability coefficient (cm/min) was determined using V/(SA × Cd) × (Cr/T), where V is the medium volume in the receiver chamber, SA is the surface area of the cell monolayer, Cd is the concentration of FITC-dextran in the donor chamber at time 0, and Cr is the concentration of FITC-dextran in the receiver chamber at sampling time T. Permeability changes are presented as percentage of control.

Western blotting analysis
Proteins extracted either from HBMVECs or isolated MBMVECs from Ngbr fl/fl and Ngbr ECKO mice were resolved using 8%-14% SDS-PAGE at 20-30 μg/lane, transferred to nitrocellulose membranes (RPN303D, GE Healthcare), and blocked for 1 hour in Tris-buffered saline (TBS) containing 0.5% Tween 20 and 5% nonfat milk at room temperature. The membranes were then incubated with primary antibodies (Supplemental Table 1) overnight at 4°C. After washing 3 times in TBS/0.5% Tween 20, membranes were incubated with corresponding HRP-conjugated secondary antibodies for 2 hours at room temperature. Immunoreactivity was detected by chemiluminescence. See complete unedited blots in the supplemental material.
RT-qPCR RNA was isolated from either HBMVECs or MBMVECs using TRIzol (15596018, Thermo Fisher Scientific). Extracted total mRNA was then reverse transcribed using iScript Reverse Transcription Supermix (1705541, Bio-Rad). qPCR was then performed using iTaq Universal SYBR Green Supermix (1725121, Bio-Rad). Primers used are shown in Supplemental Table 2. The mRNA expression level of GAPDH was set as housekeeping control.
Sigma-Aldrich) solution was prepared, protected from light, and then injected (10 μL/g body weight) into the retro-orbital plexus of neonatal mice or the tail vein of adult mice. At 5 minutes after injection, mice were euthanized, and brains were harvested and fixed for sectional fluorescent immunostaining.

Histologic examination
Mouse brains were harvested, fixed with 4% paraformaldehyde (PFA) overnight, embedded in OCT or paraffin, and sectioned (8 μm or 30 μm thickness). Cells under different treatments were cultured on chamber slides (Nunc Lab-Tek II Chamber Slide System, 154534, Thermo Fisher Scientific). H&E staining, IgG staining, and immunofluorescence staining were performed on either brain tissues or cells.
H&E staining. Paraffin-embedded mouse brain sections were dewaxed and rehydrated and then stained in 0.1% hematoxylin (MHS16, Sigma-Aldrich) for 20 minutes after being rinsed with tap water for 10 minutes. Slides were washed with distilled water and 95% alcohol, and then stained with 0.5% alcoholic eosin Y (1024390500, Sigma-Aldrich) for 30 seconds. After gradual dehydration, slides were mounted and photographed.
IgG staining. Paraffin-embedded mouse brain sections were dewaxed, rehydrated, and then incubated with biotinylated antimouse IgG antibody (1:200; Vector Laboratories) for 1 hour at room temperature. Slides were then washed for 5 minutes in PBS and incubated with prepared VECTASTAIN Elite ABC Reagent (Vector Laboratories). After another 5-minute wash in PBS, slides were incubated in a peroxidase substrate solution for stain development and rinsed in tap water. Slides were then counterstained with hematoxylin, cleared, mounted, and photographed.
Immunofluorescence staining. OCT-embedded mouse brain sections were washed in PBS twice and then blocked and permeabilized in 5% donkey serum in PBS with 0.3% Triton X-100 for 1 hour at room temperature. After blocking and permeabilization, slides were incubated with primary antibodies, as described in Supplemental Table 1, overnight at 4°C, and then washed with PBS 3 times and incubated with proper secondary antibodies at room temperature for 2 hours. After washing in PBS 3 more times, slides were stained with DAPI (0.5 μg/ mL) for 10 minutes and mounted with an anti-fade mounting medium (P36970, Invitrogen), and then photographed. Primary HBMVECs (ACBRI 376, Cell Systems) were fixed with 4% PFA for 15 minutes, washed with PBS, permeabilized with 0.1% Triton X-100 in PBS for 5 minutes, and then blocked with blocking buffer for 30 minutes. Primary and secondary antibody application, DAPI staining, mounting, and photographing methods were the same as for tissue staining.

Electron microscopy
Brain tissue samples were fixed in 4% glutaraldehyde in 0.1 M sodium cacodylate buffer, pH 7.5, washed in sodium cacodylate buffer, postfixed in buffered 1% osmium tetroxide, stained en bloc with a saturated solution of uranyl acetate in 40% ethanol, dehydrated in a graded series of ethanol, infiltrated in propylene oxide with Epon epoxy resin (LADD LX112, Ladd industries), and embedded. The blocks were sectioned with a Reichert Ultracut microtome at 70 nm. The resulting grids were then poststained with 1% aqueous uranyl acetate followed by 0.5% aqueous lead citrate and analyzed on a Zeiss EM 900 transmission electron microscope retrofitted with an SIA L3C digital camera.

Statistics
Quantification of immunofluorescence images was conducted with ImageJ software (NIH). Statistical analysis was performed using GraphPad Prism and SPSS software. Data are presented as mean ± SD. Statistical testing between 2 groups was performed using a 2-tailed, unpaired Student's t test. Multiple comparisons among groups of more than 2 were performed using 1-way ANOVA and Dunnett's post hoc test. Differences were considered statistically significant when P was less than 0.05.

Study approval
All animal studies were approved by the Institutional Animal Care Use Committees of the Medical College of Wisconsin and New York University Langone Health.

Author contributions
ZF, WH, XS, XW, and JM performed and interpreted the majority of the experiments. UR carried out retina staining, and BM carried out transcription factor GSEA. TP and LR carried out electron microscopy experiments. QRM, ZF, and WH conceived the project, designed the experiments, and wrote the manuscript. QRM supervised the project. All the authors commented on the manuscript.