Liver X receptor alpha ensures blood-brain barrier function by suppressing SNAI2

In Alzheimer’s disease (AD) more than 50% of the patients are affected by capillary cerebral amyloid-angiopathy (capCAA), which is characterized by localized hypoxia, neuro-inflammation and loss of blood-brain barrier (BBB) function. Moreover, AD patients with or without capCAA display increased vessel number, indicating a reactivation of the angiogenic program. The molecular mechanism(s) responsible for BBB dysfunction and angiogenesis in capCAA is still unclear, preventing a full understanding of disease pathophysiology. The Liver X receptor (LXR) family, consisting of LXRα and LXRβ, was reported to inhibit angiogenesis and particularly LXRα was shown to secure BBB stability, suggesting a major role in vascular function. In this study, we unravel the regulatory mechanism exerted by LXRα to preserve BBB integrity in human brain endothelial cells (BECs) and investigate its role during pathological conditions. We report that LXRα ensures BECs identity via constitutive inhibition of the transcription factor SNAI2. Accordingly, deletion of brain endothelial LXRα is associated with impaired DLL4-NOTCH signalling, a critical signalling pathway involved in vessel sprouting. A similar response was observed when BECs were exposed to hypoxia, with concomitant LXRα decrease and SNAI2 increase. In support of our cell-based observations, we report a general increase in vascular SNAI2 in the occipital cortex of AD patients with and without capCAA. Importantly, SNAI2 strongly associated with vascular amyloid-beta deposition and angiopoietin-like 4, a marker for hypoxia. In hypoxic capCAA vessels, the expression of LXRα may decrease leading to an increased expression of SNAI2, and consequently BECs de-differentiation and sprouting. Our findings indicate that LXRα is essential for BECs identity, thereby securing BBB stability and preventing aberrant angiogenesis. These results uncover a novel molecular pathway essential for BBB identity and vascular homeostasis providing new insights on the vascular pathology affecting AD patients.


Introduction
The blood-brain barrier (BBB) is an important cellular interface that is essential for the maintenance of brain homeostasis.The BBB is composed of specialized brain endothelial cells (BEC), which form a physical barrier between the blood and the brain.The highly differentiated BECs form a continuous barrier by means of tight junctions (TJs), consisting of proteins such as claudin-5 (CLDN5) and occludin (OCLDN), which limit the exchange of molecules and cells (1).In addition, BECs express specialized in ux transporters, including the glucose transporter and e ux transporters such as the ATP binding cassette (ABC) transporters.Together they tightly regulate the bidirectional transcellular transport of metabolites, ensuring the brain metabolic demand is met (2).By forming this tight and selective barrier, the BBB protects the central nervous system (CNS) from unwanted neurotoxic compounds or cells.
Under healthy conditions, the specialized and highly differentiated BECs identity is secured by multiple signalling pathways including the Wnt/β-catenin and many others (3,4).However, during disease, BECs display remarkable phenotypic plasticity, highlighted by their ability to undergo endothelial to mesenchymal transition (EndMT) (5,6).During EndMT, transcription factors such as SNAI1 and SNAI2 drive endothelial cells to lose their speci c markers (e.g.CLDN5, OCLDN) and progressively acquire a mesenchymal phenotype (7).This associates with elongated BEC morphology, loss of cell-cell junctions and polarity as well as gaining motility, invasive and contractile properties (8,9).Although EndMT is essential during embryonic development, this process has also been associated with different CNS disorders, including multiple sclerosis and cerebral cavernous malformation, thereby contributing to BBB dysfunction (7,10).
Recently, the family of Liver x receptors (LXRs) has been implicated in the process of epithelial to mesenchymal transition (29).LXRs are members of the nuclear receptor family of ligand-activated transcription factors, and consist of two isoforms, LXRα and LXRβ.They regulate the expression of several genes involved in cholesterol and fatty acid metabolism.Once activated, LXRs binds as a heterodimer with the obligate partner retinoid x receptor (RXR) to LXR-responsive elements and promote gene expression (30,31).Important downstream target genes of the LXR pathway are ABC transporter genes (ABCA1 and ABCG1), Apolipoprotein E, and the E3 ubiquitin ligase IDOL (32).Importantly, the LXR-RXR heterodimers are considered permissive, and thus may be activated by agonists of either heterodimeric receptor (33,34).Apart from cholesterol metabolism, LXRs are important for other biological functions, including in ammation and BBB function (35)(36)(37)(38)(39)(40).
We previously reported that LXRα is important for maintaining BBB integrity and its immune quiescence under normal and in ammatory conditions (41).In the present study, we aim to unravel the regulatory mechanism exerted by LXRα to preserve BBB integrity.Here, we report that LXRα secures BEC identity.Our data suggest that LXRα limits the interaction of LXRβ with RXR, thereby inhibiting the expression of SNAI2.The knockdown of LXRα in BECs results in increased SNAI2 expression, loss of BEC markers, and aberrant angiogenesis via the suppression of the DLL4-NOTCH signalling pathway.Moreover, we show that hypoxia speci cally affects LXRα and recapitulates the effects observed in LXRα knockdown cells.Finally, we show an increased expression of SNAI2 in Aβ-affected vessels in post-mortem tissue from AD patients with and without capCAA.Collectively, our ndings elucidate a novel mechanism via which LXRα secures BBB integrity and how its impairment might contribute to the observed BBB dysfunction in AD patients with capCAA.

RNA isolation and qRT-PCR
Recombinant hCMEC/D3 cells (1x10 6 cells/ml) transduced with either LXRα shRNA, LXRβ shRNA, or nontargeting shRNA were seeded in 24-well plates in culture medium.RNA was isolated using the TRIzol® method (Life Technologies, Bleiswijk, The Netherlands) and cDNA was synthesized using the Reverse Transcription System kit (Promega, Leiden, The Netherlands).Sequences of primers used are listed in Supplementary Table .1. Quantitative Reverse Transcriptase PCR (qRT-PCR) was carried out using SYBR green master mix (Applied Biosystems, Waltham, MA, USA) and a Step One Plus detection system (Applied Biosystems).Quanti cation of gene expression was accomplished using the comparative cycle threshold method.Expression levels were normalized to Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) or Ribosomal Protein Lateral Stalk Subunit P0 (RPLP0) expression.

RNA sequencing-based transcriptional pro ling and analysis
The hCMEC/D3 cells were cultured and treated as described above.Total RNA was extracted using TRIzol reagent (Invitrogen, Carlsbad, CA, USA) and converted into strand-speci c cDNA libraries using the TruSeq Stranded mRNA sample preparation kit (Illumina, San Diego, CA, USA) according to the manufacturer's instructions.Brie y, polyadenylated RNA was enriched using oligo-dT beads and subsequently fragmented, random primed and reverse transcribed using SuperScript II Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA).Second strand synthesis was performed using Polymerase I and RNaseH with replacement of dTTP for dUTP.The generated cDNA fragments were 3' end adenylated, ligated to Illumina paired-end sequencing adapters, and subsequently ampli ed by 12 cycles of PCR.The libraries were analysed on a 2100 Bioanalyzer using a 7500 chip (Agilent, Santa Clara, CA, USA) and subsequently sequenced with 65 base single reads on a HiSeq2500 using V4 chemistry (Illumina, San Diego, CA, USA) Transcripts were aligned to the Human Feb. 2009 (GRCh37/hg19) assembly using TopHat (version 2.1) (46).Gene expression sets were prepared using ICount, which is based on HTSeq-count.Uniquely mapped reads were normalized to 10 million reads followed by log2 transformation.In order to avoid negative normalized values, 1 was added to each gene expression value.Data were analysed using Gene Set Enrichment Analysis software (47, 48) (University of California San Diego, San Diego, CA, USA) and the differentially expressed gene sets (nominal p-value < 0.05) displayed.The results of the gene set enrichment analysis are displayed in Supplementary Table .2. The heat map was created using Heat mapper online tool, plotting the count per million per sample (49).The RNA-seq data were analysed as follow, transcripts with more than 2 counts in 3 or more of the samples were kept, for data normalization TMM (edgeR), weighted trimmed mean of M-values (to the reference) was used (50), the data were annotated using biomaRt, using Ensembl.The count data was transformed to log2-counts per million (logCPM) using voom, estimating the mean-variance relationship and the differential expression was assessed using a moderated t-test using the linear model framework from the limma package and the adjusted p-value calculated by using FDR, Benjamini-Hocheberg correction (Supplementary Table .3).Differentially expressed genes (FDR adj.p-value < 0.05 and Log2 fold change of 0.5) were displayed in a volcano plot.

Western blot and nuclear fractionation
After washing with ice-cold phosphate-buffered saline (PBS), hCMEC/D3 cells were lysed with cell lysis buffer (Cell Signaling Technology, Boston, MA, USA) containing a protease and phosphatase inhibitor cocktail (Roche, Almere, The Netherlands, and Cell Signaling Technology, Boston, MA, USA, respectively) on ice, following the manufacturer's instructions.Nuclear fractions were isolated using the NE-PER extraction kit (Thermo Fisher Scienti c, Rockford, IL, USA), following the manufacturer's guidelines.All samples were diluted in sample buffer (BioRad Hercules, CA, USA) (65.8 mM Tris-HCl, pH 6.8, 2.1% SDS, 26.3% (w/v) glycerol, 0.01% bromophenol blue) and heated to 95°C for 3 min.For whole cell lysates, hCMEC/D3 were removed from the media and lysed in sample buffer.Lysates were separated on SDS-PAGE followed by transfer to nitrocellulose for immune-blot analysis.Blots were blocked for 1 h at room temperature with blocking buffer (Azure Biosystems, Inc, Sierra CT, Dublin, CA, USA).Subsequently, membranes were incubated in blocking buffer containing 0.1% Tween-20 with antibodies against claudin-5 (Santa Cruz, Dallas, TX, USA), SNAI2 (Abcam, Cambridge, United Kingdom) and GAPDH (Proteintech, Manchester, United Kingdom).Primary antibodies were detected and quanti ed by incubation with IRDye secondary antibodies (LI-COR) and use of Azure Sapphire Biomolecular Imager (Azure Biosystems, Inc, Sierra CT, Dublin, CA, USA).

Immunohistochemistry on post-mortem human brain tissue
Brain tissue from 5 patients with clinically diagnosed and neuropathologically con rmed capCAA, 6 AD and 6 non-demented control (NDC) cases without neurological diseases was obtained after autopsy (post-mortem delay < 8hr) and immediately frozen in liquid nitrogen (in collaboration with the Netherlands Brain Bank, Amsterdam).The Netherlands Brain Bank received permission from the ethical committee of the VU University Medical Center Amsterdam, the Netherlands to perform autopsies, for the use of the material and for access to medical records for research purposes.Cortical grey matter samples from the superior occipital gyrus (SOG) were selected and used for staining.All patients and controls, or there next of kin, had given informed consent for autopsy and use of their brain tissue for research purposes.Clinical data are presented in Supplementary Table  For immunohistochemical analysis, 5 µm thick cryosections of frozen brain tissues were xed in ice-cold acetone for 10 min.After washing with PBS, sections were incubated overnight at 4°C with primary antibodies against SNAI2 (Abcam, Cambridge, United Kingdom).Subsequently, sections were washed with PBS and incubated with Envision Dual Link (DAKO, Glostrup, Denmark) for 1h at room temperature, followed by visualization with the peroxidase substrate 3,3′-diaminobenzidine (DAKO, Glostrup, Denmark).Sections were incubated with hematoxylin (Sigma-Aldrich, Saint Louis, MO, USA) for 1 min and thoroughly washed with tap water for 10 min.Ultimately, sections were dehydrated with ethanol followed by xylene (Sigma-Aldrich, Saint Louis, MO, USA) and mounted with Entellan (Merck, Darmstadt, Germany).Immuno uorescent labelling was performed as follows: after xation in ice-cold acetone for 10 minutes, the sections were incubated for 30 min with 10% normal goat serum and 0.1% Triton X-100 (Sigma-Aldrich, Saint Louis, MO, USA) and afterwards incubated overnight at 4°C with antibodies against SNAI2 (Abcam, Cambridge, United Kingdom) or ANGPTL4 (Abcam, Cambridge, UK), and sections were stained with UEA-1 (Vector Lab, Burlingame, CA, USA).

Image acquisition and analysis
Images of the DAB-stained tissue were obtained using a DM6000 (Leica, Mannheim, Germany), 4 random regions of interest (ROIs) were collected per sample and the results presented as average staining intensity per section.Fluorescent images were obtained using an Olympus VS200 (Olympus, Tokyo, Japan) slide scanner or a SP8 confocal microscope (Leica, Mannheim, Germany).Five speci c ROIs with a Z-stacks of 6 µm and a 60x magni cation were recorded and the results are presented as average staining intensity per section.Image deconvolution and analysis were done using Huygens Professional

Statistical analysis
Data were statistically analysed using GraphPad Prism v9 (GraphPad Software, La Jolla, CA, USA).
Ordinary one-way ANOVA (3 groups), or two-tailed paired or unpaired student t-test (2 groups) with original FDR method of Benjamini and Hochberg multiple comparison correction were used for normally distributed data sets.The Kruskal-Wallis (3 groups) or paired Wilcoxon signed-rank test and Mann-Whitney (2 groups) analysis with original FDR method of Benjamini and Hochberg correction was used for non-parametric data sets.*P < 0.05, **P < 0.01, ***P < 0.001 (adjusted p value).

LXRα de cient BECs display a profound loss of brain endothelial markers
To investigate the underlying mechanism of BBB dysfunction caused by depletion of LXRα in BECs, we compared the transcriptional pro le of control (shRNA non-targeting (NTC)) and LXRα knockdown cells (LXRα KD) using RNA-seq.The transcriptional pro le of the LXRα KD cells showed remarkable differences with NTC cells as highlighted in the principal component analysis (Fig. 1A).An in-depth analysis revealed a total of 4223 differentially expressed genes between the two groups, of which 2038 were up-and 2185 down-regulated in LXRα KD cells compared to NTC cells.The Gene Set Enrichment Analysis (GSEA) showed 12 signi cantly upregulated gene set (p < 0.01) and 9 signi cantly downregulated gene set (p < 0.01) (Fig. 1B).Among the downregulated gene set, we recognized the KEGG-ABC transporters and KEGG-cell adhesion molecules (Fig. 1B).These gene sets include important brain endothelial markers such as ABCA1, ABCG1, OCLDN and CLDN5 (Supplementary Fig. 1).
Since none of the signi cantly different gene sets could explain the loss of BEC markers, we explored the gene sets according to their Normalized Enrichment Score.The KEGG-adherens junction, ranked as one of the most different GS between NTC and LXRα KD cells.Among the genes present in this set, we recognized the transcription factors SNAI1 and SNAI2, which were increased in LXRα KD cells (Fig. 1C).
The volcano plot shows a 4-fold increase of SNAI2 (p < 0.05) but no statistical difference in SNAI1 expression (Fig. 1D).Collectively, these results indicate that silencing of LXRα reduces the expression of BEC markers and that this mechanism is potentially driven by SNAI2.

LXRα constitutively inhibits SNAI2 expression in brain endothelial cells
We next set out to validate the increase of SNAI genes observed in the RNA-seq of LXRα KD cells, via targeted quantitative mRNA analysis.Furthermore, we generated speci c LXRβ KD BECs to elucidate whether the regulation of SNAI1 and SNAI2 expression is LXR isoform speci c.The interference with shRNA signi cantly and speci cally decreased the mRNA expression of LXRα (p = 0.008) and LXRβ (p = 0.005) in BECs (Fig. 2A) LXRα silencing increased the expression of SNAI2 3-fold (p = 0.015) and SNAI1 to a lesser extent (p = 0.019) (Fig. 2B).This change in SNAI2 was LXRα speci c as silencing of LXRβ failed to increase SNAI2 expression (Fig. 2B).No signi cant differences were found regarding the levels of other transcription factors involved in EndMT (e.g., ZEB1, ZEB2, TWIST1) (Fig. 2B).We next validated the expression of BEC markers.LXRα KD cells resulted in a signi cant downregulation of the BEC speci c markers CLDN5 (p = 0.044), OCLDN (p = 0.047), ABCA1 (p = 0.002) and ABCA7 (p = 0.049) (Fig. 2C).These markers were unchanged in LXRβ KD cells, with the exception of ABCA1 (p = 0.049) (Fig. 2C).
To decipher whether the regulatory action exerted by LXRα on SNAI2 expression is related to its activity, we treated NTC, LXRα and LXRβ KD cells with a broad LXR antagonist GSK2033 (1µM for 72h).
Treatment of BECs with GSK2033 resulted in complete inhibition in the transcription of the LXR target gene ABCA1 in all three conditions (NTC (p = 0.001), LXRα KD (p = 0.003), LXRB KD (p = 0.001)) (Fig. 2D).Moreover, inhibiting the activity of LXRα KD cells (by treating the LXRβ KD cells) did not result in an increase of SNAI2.Together, these results suggest that LXRα is essential to inhibit SNAI2 but that this likely involves a ligand-independent and indirect mechanism.

LXRα prevents LXRβ-RXR interaction in brain endothelial cells
As permissive receptors, both LXRs and RXRs are activated by oxysterols and retinoic acid (RA) (51), thereby inducing gene expression of both pathways (Fig. 3A).To further dissect the regulatory circuit underlying LXRα and SNAI2 expression, we rst treated BECs with the pan LXR agonist GW3965 (1µM for 48h) and investigated the expression of LXR target genes.GW3965 signi cantly increased the expression of ABCA1 in NTC (p = 0.005) and LXRβ KD cells (p = 0.030) but not in LXRα KD cells (Fig. 3B).In parallel, the relative SNAI2 expression, which was already substantially higher in LXRα KD cells, was signi cantly further increased upon stimulation (p = 0.020) (Fig. 3B).
To investigate whether activation of RXR induces LXR target gene expression, we next treated the cells with retinoic acid, which is an agonist of the RXR pathway (5µM for 72 h).RA was able to activate the LXR pathway in NTC as shown by the signi cant increase of ABCA1 (p < 0.001) and LXRα (p < 0.001).More importantly, the treatment with RA signi cantly increased the expression of SNAI2 in LXRα KD cells (p < 0.001), while the compound had no effect on SNAI2 expression in NTC and LXRβ KD cells (Fig. 3C).
To test whether the interaction of LXRβ and RXR is responsible for the increased SNAI2 expression observed in LXRα KD cells, we treated BECs with a speci c inhibitor for RXRα (PA452, 1µM for 48h) (52).PA452 signi cantly decreased SNAI2 expression (p = 0.020) in LXRα KD cells, while it did not affect the expression in NTC and LXRβ KD cells (Fig. 3D).Furthermore, the RXRα inhibitor had no effect on ABCA1 and LXRα expression in neither of the BECs.Collectively these ndings suggest a model in which the presence of LXRα prevents LXRβ-RXR interaction thereby counteracting SNAI2 expression.

LXRα regulates DLL4-Notch signalling
The transcription factor SNAI2 is directly responsible for the loss of endothelial markers (e.g.CLDN5, CD31) and aberrant angiogenesis via the suppression of DLL4-NOTCH signalling in human umbilical cord endothelial cells (53).In light of this, we next assessed the NOTCH pathway in LXRα KD cells.Our RNA-seq data indicated a signi cant (p < 0.05) downregulation of key players in the Notch pathway (e.g.DLL4, NOTCH1, NOTCH4, HES1, HEY1) as depicted in the heat map (Fig. 4A).We con rmed these ndings in LXRα KD cells, where we determined a signi cant decrease of both DLL4 (p = 0.018) and NOTCH1 transcripts (p = 0.019) in comparison to control cells (Fig. 4B).KDR in contrast, which is the gene encoding for the vascular endothelial grow factor receptor 2 (VEGFR2) and essential during the tip-cell formation process, was increased in LXRα KD cells (p = 0.002).No signi cant differences were found for the other NOTCH ligands measured such as JAG1 and JAG2 (Supplementary Fig. 1).Moreover, DLL4 and NOTCH1 expression did not change upon GW3965 stimulation (Supplementary Fig. 1).We further con rmed the decrease of DLL4 and CD31 protein content in LXRα KD cells, which was accompanied by a corresponding loss of cellular polarization (Fig. 4C).To assess whether the increase of SNAI2 expression in LXRα KD cells modulates Notch signalling, we performed a sprouting assay (Fig. 4D).LXRα KD cells showed a signi cant increase in the number of sprouts (p < 0.001), while these newly formed sprouts were on average shorter than those in NTC cells (p < 0.001) (Fig. 4E).All together, these data highlight the importance of LXRα in the DLL4-NOTCH axis maintenance and associated vessel formation.

Hypoxia selectively suppresses LXRα in BECs
Reduced tissue oxygen pressure can induce tip cell formation, thereby stimulating sprouting angiogenesis (54).In light of our previous results reporting enhanced angiogenesis in LXRα KD cells we questioned the role of hypoxia in the regulation of LXRα expression, BECs identity and the angiogenic process.After 48 hours in 1% O 2 , BECs maintained their con uence yet showed loss in cell polarization as highlighted by the altered cell morphology (Fig. 5A).Importantly, hypoxia markedly reduced the mRNA (p = 0.020) and protein level of LXRα, while expression of the LXRβ isoform was largely refractory to this treatment (Fig. 5A).Despite its fundamental role during hypoxia, HIF-1α is not responsible for LXRα suppression, the knockdown of HIF-1α did not rescue LXRα decrease in hypoxic BECs (Supplementary Fig. 2).The transcripts of BEC markers CLDN5 (p = 0.034), ABCA1 (p = 0.015) and OCLDN (p = 0.060) were also downregulated by hypoxia (Fig. 5B).Immuno uorescent detection of CLDN5 in BEC validated the mRNA results by showing that the corresponding protein is also decreased, as its junctional localization (Fig. 5C).A mirror image was observed with SNAI2, for which the mRNA expression (p = 0.011) and protein levels were increased in hypoxic BECs (Fig. 5D).
To assess the activation of the Notch pathway under hypoxia, we evaluated DLL4 and NOTCH1 expression via qPCR and assessed the angiogenic ability using the sprouting assay.Expression of both genes was reduced in hypoxic BECs, however while DLL4 was signi cantly downregulated (p = 0.033) NOTCH1 expression did not reach statistical signi cance (p = 0.114) (Fig. 5E).The sprouting assay showed major differences in cells grown in hypoxia (Fig. 5F).These cells had an increased number of sprouts (p = 0.001) which were shorter (p = 0.001) when compared to control (Fig. 5G).These ndings reveal the inhibitory role of hypoxia on LXRα-dependent signalling, which triggers phenotypical changes reminiscent of those observed in LXRα KD cells.Furthermore, the data provide insights on the mechanism of hypoxia induced angiogenesis.

SNAI2 is increased in the vasculature of AD patients with capCAA
As hypoxia and vascular dysfunction strongly associate with AD pathology, we next set out to determine whether SNAl2 participates in the endothelial dysfunction present in AD with and without capCAA (16, 55, 56).In our immunohistochemical analysis we made use of ANGPTL4 expression in astrocytes as a recently discovered marker of hypoxia in capCAA (57).Due to the renowned impossibility to stain LXRα in tissue and based on our in vitro data we therefore assessed the localization and expression of SNAI2.We performed immunohistochemistry on the occipital cortex of AD patients with and without capCAA pathology compared to non-demented controls.Immunohistochemical analysis revealed a signi cant upregulation of SNAI2 in AD patients with capCAA (p = 0.038) compared to non-demented controls, while AD patients without capCAA showed only a marginal increase in SNAI2 expression in their vessels (p = 0.052) (Fig. 6A).Further immuno uorescent analysis showed that the increased expression of SNAI2 in AD patients with capCAA was associated with vessels affected by Aβ (Fig. 6B).
We next evaluated the implication of hypoxia in the promotion of SNAI2 in AD patients with capCAA, by assessing the expression of ANGPTL4.We con rmed previous results showing that ANGPTL4 is expressed in astrocytes end-feet, primarily in the vicinity of Aβ aggregates (Supplementary Fig. 2).Aβ positive vessels showed a signi cant increase of ANGPTL4 (p = 0.031) compared to vessels negative for Aβ (Fig. 6C and D).The vessels affected by perivascular accumulation of Aβ that were also positive for ANGPTL4 had a signi cantly higher level of SNAI2 (p = 0.031) (Fig. 6C and D).Finally, we evaluated the expression of SNAI2 in ANGPTL4-positive vessels independent of Aβ, and report a consistent increase of SNAI2 (p = 0.031) (Fig. 6D).Together, these results suggest that hypoxia induces the expression of SNAI2 in AD patients expressing capCAA pathology.

Discussion
CapCAA is frequent in AD and is associated with BBB dysfunction, disturbances of cerebral blood ow, and might contribute to cognitive decline (11)(12)(13)(14)(15)(16).However, the underlying mechanism leading to these microvascular changes remain unknown.In our previous study we showed that LXRα is important in maintaining proper barrier function (41).In the current study, we decipher how LXRα maintains BEC identity.We report a novel indirect inhibitory activity of LXRα on SNAI2, where the loss of LXRα leads to an increase of SNAI2.This results in the de-differentiation and sprouting of BECs via disruption of the DLL4-Notch axis.Finally, we show that hypoxia affects speci cally LXRα expression in vitro and translate our ndings to the observed endothelial dysfunction in AD patients with and without capCAA.
Our results suggest that the presence of LXRα is necessary for the constitutive inhibition of SNAI2, where LXRα might compete with LXRβ for the RXRα monomer, which seems the key mechanism for SNAI2 inhibition.In-silico ligand-binding kinetic measurements, used to explore the a nity of LXRα/β for RXRα, indicate a stronger interaction for the LXRα-RXRα dimer upon stimulation (51).These results support our hypothesis of competitive inhibition.However, LXRα has been reported to partner with RXRβ as well (58), which requires us to underline that the proposed mechanism might not be the only active one.
Nevertheless, only a few studies so far have addressed the LXR-RXR interaction in such great detail, indicating the necessity to further explore these complex dynamics.
The increase of SNAI2 in LXRα KD cells is accompanied by a loss of endothelial markers.These ndings may re ect the initiation of EndMT, which induces dedifferentiation of the BECs (7).LXRα has been shown to inhibit EMT in myo broblast (29).In addition, SNAI2 overexpression was described to induce partial-EndMT in human umbilical vascular endothelial cells (53).Both SNAI1 and SNAI2 are important transcription factors driving epithelial to mesenchymal transition (EMT) (59)(60)(61).However, although SNAI1 also showed a signi cant increase in LXRα KD cells, the limited fold change increase in LXRα KD cells and the discrepancy with the RNA-seq data, led us to focus on the transcription factor SNAI2.
Moreover, we demonstrate the importance of LXRα in the DLL4-NOTCH axis, further highlighting the role of SNAI2.Where transforming growth factor beta is the main inducer of EndMT via SNAI1, EndMT induction by Notch requires SNAI2 (62, 63).In addition, the increased expression of SNAI2 is associated with an impaired Notch signalling, marked by DLL4 reduction, KDR (VEGFR2) increase and aberrant angiogenesis (53).During development the Notch pathway is essential for functional angiogenesis.The leader tip cell expresses DLL4 and signals to the adjacent cells via NOTCH1 to become stalk cells.DLL4-NOTCH1 signalling imposes to the stalk-cells a differential gene expression, ensuring regulated sprouting.
In adult vessels, NOTCH1 is necessary to maintain endothelial quiescence (64) and in BECs DLL4-NOTCH signalling regulates their permeability (55)(56)(57).These ndings highlight the importance of this pathway for normal vascular behaviour and emphasizes the necessity for a functional LXRα activity.
We report that hypoxia, a known driver of EMT and EndMT (65), inhibits LXRα expression, leaving the β isoform unaffected.However, this process is not driven by HIF-1α as shown by our in vitro experiments where we combined HIF-1α KD in BECs and hypoxia.Under hypoxic conditions macrophages drastically reduce cholesterol synthesis leading to the intracellular accumulation of cholesterol esters (66, 67).In BECs, the diminished cholesterol synthesis might result in reduced oxysterols, which are cholesterol derivatives necessary to maintain homeostatic levels of LXRα (68).The downregulation of LXRα, but not LXRβ, was also shown in heart tissue of mice after myocardial ischemia (69), which is in line with our ndings.Importantly, hypoxic BECs closely recapitulate the changes observed in LXRα KD cells, including increased SNAI2 expression, decreased expression of BEC markers, and impaired DLL4-NOTCH signalling.These results bestow novel molecular insights on the role of hypoxia in BBB structural maintenance as well as advances our knowledge on angiogenesis.
In the present study we report a vascular increase of SNAI2, associated with partial-EndMT, in vessels of AD patients suffering from capCAA.The general increase of SNAI2 was strongly correlated with the presence of ANGPTL4 and perivascular accumulation of Aβ.The renowned di culty in visualizing hypoxic genes, including HIF-1α, in post-mortem tissue required an alternative strategy thus we opted for ANGPTL4 as hypoxic marker.ANGPTL4 has been extensively studied in cancer research and reported to increase upon hypoxia and co-localize with HIF-1α (70-72).In the CNS, ANGPTL4 is expressed by reactive astrocytes in post-mortem tissue of patients with capCAA (57).In line with these results, we report an increase of ANGPTL4 in astrocytes located in the vicinity of Aβ affected vessels, indicating a local hypoxic environment.Moreover, our results show that Aβ affected vessels marked by ANGPTL4 have increased SNAI2 expression, possibly leading to partial-EndMT and angiogenesis.Whether Aβ is the driving force inducing localized hypoxia is still unclear and much debated (73).
The formation of new vessels is a delicate process with many facets.During stroke for instance, angiogenesis takes place rapidly after injury (74,75) and associates with higher survival rate (76).However, in diabetic retina aberrant angiogenesis results in haemorrhage and oedema (77), emphasizing the double role exerted by this mechanism.In AD, cerebrovascular defects including reduced cerebral blood ow (16, 55, 56) correlate with higher vessel number (57, 78-80) suggesting reduction in brain oxygen concentration and angiogenesis promotion.This process is accentuated in capCAA vessels, . 4.

Figure 6 SNAI2
Figure 6 The primary antibodies were visualized by incubation with goat anti-mouse Alexa 555 (Molecular Probes, Eugene, OR, USA), donkey anti-rabbit Alexa 647 (Molecular Probes, Eugene, OR, USA) or Streptavidin 488 (Molecular Probes, Eugene, OR, USA) for 1 hour at RT. Next, to visualize Aβ aggregates, sections were incubated for 5 minutes with Thio avin-S (Sigma-Aldrich, Saint Louis, MO, USA) and washed with ethanol afterwards.After washing with PBS, Hoechst (Molecular Probes, Eugene, OR, USA) was used for nuclear staining and slides were mounted in Mowiol (Sigma-Aldrich, Saint Louis, MO, USA).