JAG1-NOTCH4 Mechanosensing Drives Atherosclerosis

Endothelial cell (EC) sensing of fluid shear stress regulates atherosclerosis, a disease of arteries that causes heart attack and stroke. Atherosclerosis preferentially develops at regions of arteries exposed to low oscillatory shear stress (LOSS), whereas high shear regions are protected. We show using inducible EC-specific genetic deletion in hyperlipidaemic mice that the Notch ligands JAG1 and DLL4 have opposing roles in atherosclerosis. While endothelial Jag1 promoted atherosclerosis at sites of LOSS, endothelial Dll4 was atheroprotective. Analysis of porcine and murine arteries and cultured human coronary artery EC exposed to experimental flow revealed that JAG1 and its receptor NOTCH4 are strongly upregulated by LOSS. Functional studies in cultured cells and in mice with EC-specific deletion of Jag1 show that JAG1-NOTCH4 signalling drives vascular dysfunction by repressing endothelial repair. These data demonstrate a fundamental role for JAG1-NOTCH4 in sensing LOSS during disease, and suggest therapeutic targeting of this pathway to treat atherosclerosis.


Introduction
Blood flow generates mechanical shear stress that has profound effects on the function of blood vessels by altering the physiology vascular endothelial cells (EC) 1 .
Low oscillatory shear stress (LOSS) promotes the initiation and progression of atherosclerosis, a disease characterised by the accumulation of cells, lipids and other materials in the arterial wall that can lead to unstable angina, myocardial infarction or stroke 2,3 . Most regions of the arterial tree are exposed to physiologically high shear stress (HSS), which promotes EC quiescence and protection from disease. However, branches and bends of arteries are exposed to complex blood flow patterns, which LOSS that promotes EC dysfunction and the initiation of atherosclerosis. Intriguingly, LOSS drives both angiogenesis and atherosclerosis, suggesting a common mechanism for these divergent vascular processes 1 .
The Notch pathway was discovered in Drosophila as a master regulator of cell fate decisions and the spatial organisation of tissues 4 . Canonical Notch signalling involves an interaction between a Notch transmembrane receptor and a canonical Notch ligand on a contacting cell, which causes proteolytic cleavage of the Notch receptor by gsecretase. This causes release of the Notch intracellular domain (ICD) which subsequently localises to the nucleus to regulate transcription 5 . Mammals possess four Notch receptors and five ligands. Classic studies revealed that the interaction between NOTCH1 and DLL4 on adjacent EC establishes cellular identity (i.e. 'tip' versus 'stalk' cells) during angiogenesis 6 , and NOTCH1 and DLL4 are critical for arterial specification [7][8][9] and for neovascularisation in ischemic tissues 10,11 . Notch activation requires mechanical force induced by endocytosis of ligand 12 and recent studies found that NOTCH1 sensing of shear stress is important in arterial differentiation 13 , vascular homeostasis 14 and protection from atherosclerosis 15 .
However, the potential ability of the Notch system to sense and respond to LOSS during atherosclerosis initiation has not been studied. Moreover, the function of Notch ligands in atherosclerosis has not been previously analysed.
Here we analysed the function of endothelial Jag1 and Dll4 in atherosclerosis using conditional gene deletion approaches and observed that these Notch ligand genes have opposing effects on atherosclerosis. While Jag1 enhances atherosclerosis specifically at regions of LOSS that are susceptible to atherosclerosis, Dll4 exerted an atheroprotective effect. Mechanistic studies revealed that NOTCH4 is the dominant Notch receptor at sites of LOSS, and that JAG1-NOTCH4 signalling enhances atherosclerosis by repressing EC proliferative reserve. These data fundamentally advance our knowledge of the role of Notch signalling in interpreting shear stress signals to control EC function and have important implications for therapeutic targeting of the Notch pathway in atherosclerosis.

Results
Endothelial Jag1 and Dll4 have divergent effects on atherosclerosis.
To delineate the roles of endothelial Jag1 and Dll4 in atherosclerosis, we generated EC-specific inducible knockout mice by crossing floxed strains with a transgenic strain expressing CDH5 CreERT2/+ . Mice aged 6 weeks were treated with Tamoxifen for 2 weeks in order to delete Jag1 or Dll4 from EC or to generate experimental controls, and were then treated with adenoviral-PCSK9 and exposed to high fat diet for 6 weeks to generate hypercholesterolemia and atherosclerotic lesions ( Fig. 1a and Fig. 2a).
Tamoxifen treatment of Jag1 fl/fl Cdh5 CreERT2/+ mice (Jag1 ECKO ) induced deletion of Jag1 which was validated by qRT-PCR ( Fig. 1b; residual Jag1 mRNA likely due to expression in smooth muscle cells) and en face staining of the aorta (Fig. 1c).
Endothelial deletion of Jag1 did not lead to clinical manifestations and did not affect cholesterol levels ( Supplementary Fig. 1a). In hypercholesterolaemic mice, lesion area was significantly reduced in Jag1 ECKO compared to controls in the whole aorta ( Fig. 1d and Fig. 1e). A sub-analysis revealed that Jag1 ECKO caused a reduction in lesion area in the aortic arch but not in the descending aorta (Fig. 1e). Lesion area in the aortic root was also unaltered in Jag1 ECKO compared to controls ( Fig. 1f and Fig. 1g).
Therefore, Jag1 promotes atherosclerosis specifically at the aortic arch which is a region exposed to LOSS.
Deletion of Dll4 was validated in aortae by qRT-PCR analysis in Dll4 fl/fl Cdh5 CreERT2/+ mice (Dll4 ECKO ; Fig. 2b). Deletion of both Dll4 alleles caused lethality under high fat diet and we therefore analysed atherosclerosis using mice with a single allele deletion of Dll4 (Dll4 ECHet ). En face staining using Oil-red-O revealed that lesion area in the aorta was significantly increased in Dll4 ECHet compared to controls ( Fig. 2c and Fig.  2d; Whole aorta). A sub-analysis of the aortic arch and descending aorta revealed that mean lesion area was enhanced in Dll4 ECHet compared to controls at both sites but this difference reached statistical significance only in the aortic arch (Fig. 1d). Similarly, analysis of aortic root cross-sections revealed significantly enhanced lesion area in Dll4 ECHet compared to controls ( Fig. 1e and Fig. 1f). We also observed a slight increase of cholesterol and triglyceride levels in Dll4 ECHet animals ( Supplementary Fig. 1b).
Thus, we conclude that endothelial Dll4 exerts an atheroprotective function in the aorta and aortic root. In summary, endothelial Dll4 and Jag1 have divergent roles in atherosclerosis; Dll4 is protective whereas Jag1 initiates disease specifically at sites of LOSS. JAG1 and NOTCH4 are enriched at atherosusceptible regions exposed to LOSS.
Given the different effects of Jag1 and Dll4 on atherosclerosis, we hypothesised that Notch receptors and ligands may have a different spatial pattern of expression in arteries. This was tested by qRT-PCR analysis of transcripts of Notch receptors and their ligands isolated from regions of the porcine aorta exposed to LOSS (inner curvature) or HSS (outer curvature) using a shear stress map generated previously by our group 16 . The expression of JAG1 and NOTCH4 mRNA was significantly enhanced at sites of LOSS compared to HSS whereas the expression of DLL4, NOTCH1, NOTCH2 and NOTCH3 was similar between these sites (Fig. 3a). Similarly, en face staining demonstrated that JAG1 and NOTCH4 proteins were expressed at higher levels at a LOSS region compared to a HSS region in the aorta of C57BL/6 mice ( Fig.   3b and Fig. 3c). Moreover, endothelial expression of JAG1 was enhanced at atherosclerotic plaques compared to non-diseased arteries in hypercholesterolemic ApoE -/mice ( Supplementary Fig. 2), suggesting a potential role for JAG1 in atheroprogression.

LOSS activates JAG1-NOTCH4 signalling.
The correlation of JAG1 and NOTCH4 with LOSS does not reveal causality since atheroprone sites are exposed to alterations in mass transport, oxygen levels and inflammation as well as shear stress. We therefore used in vitro and in vivo models to assess directly whether Notch receptors and ligands are shear stress sensitive. A commercial syringe-pump, parallel plate apparatus was used to expose cultured human coronary artery EC (HCAEC) to either HSS or LOSS for 72 h. We validated the approach, by demonstrating that inflammatory MCP1 transcripts are enhanced under LOSS whereas protective KLF4 mRNAs are enhanced by HSS (Fig. 4a).
Analysis of Notch pathway components by qRT-PCR revealed that JAG1, DLL4, NOTCH4 and the Notch target gene HES1 were significantly enhanced under LOSS compared to HSS, whereas NOTCH1, NOTCH2 and NOTCH3 were unaltered (Fig.   4b). We next assessed the activity of Notch receptors using antibodies that bind specifically to the active, cleaved intracellular domains (ICD) of NOTCH4 (N4ICD) and NOTCH1 (N1ICD). It was observed by immunostaining ( Fig. 4c) and immunoblotting ( Fig. 4d) that N4ICD was significantly enhanced under LOSS compared to HSS, whereas N1ICD was not altered by shear stress (Fig. 4d). It should also be noted that detection of N1ICD required exposure of immunoblots for much longer periods compared to detection of N4ICD, suggesting that N4ICD is the dominant Notch receptor under LOSS (Fig. 4d). Immunoblotting also revealed that JAG1 was enhanced by exposure of HCAEC to LOSS, whereas DLL4 did not exhibit change at the protein level (Fig. 4d). Similarly, in vivo studies of JAG1 demonstrated that it is enhanced by LOSS in murine carotid arteries modified with a constructive cuff for 14 days (Fig. 4e).
At a mechanistic level, ATACseq analysis of HAEC exposed to flow revealed two

LOSS induces JAG1-dependent activation of NOTCH4.
We investigated whether JAG1 is required for NOTCH4 activation by LOSS. The ability of ligands to elicit Notch signalling in response to LOSS was tested by applying anti-DLL4 23 or anti-JAG1 24 blocking antibodies and measuring N4ICD as an indicator of NOTCH4 activity. Of note both anti-JAG1 and anti-DLL4 blocking antibodies attenuate Notch activity level as revealed by a decreased expression of Notch target genes such as HES1, HEY1 and HEY2 ( Supplementary Fig. 4). It was observed that blocking antibodies directed against JAG1 significantly reduced N4ICD levels whereas blocking of DLL4 activity had modest effects that did not reach significance (Fig. 5a). Thus, LOSS activation of NOTCH4 requires JAG1. Conversely, the induction of JAG1 by LOSS was abolished by treating HCAEC with the DAPT (an inhibitor of Notch receptor activity; Fig. 5b) indicating the Notch activity drives JAG1 expression. Thus, it was concluded that JAG1 and NOTCH4 interact at a functional level in endothelium exposed to LOSS.

JAG1 sensing of LOSS represses endothelial repair at atherosusceptible sites.
The function of JAG1 was investigated by applying anti-JAG1 blocking antibodies to HCAEC exposed to LOSS for 48 h and quantifying changes in the transcriptome by RNAseq. The expression of 1239 genes was significantly altered by inhibition of JAG1, and functional annotation using DAVID found that the most highly enriched gene ontology terms for the genes negatively regulated by Jag1 are centred around cell division processes including Mitotic Nuclear Division, Cell Division, Positive Regulation of Mitotic Transition, cytokinesis ( Fig. 6a and Table 1). The expression of JAG1-regulated genes was visualised using a volcano plot with the position of cell division regulators superimposed (Fig. 6b). This unbiased assessment of JAG1regulated genes led to the hypothesis that JAG1-NOTCH4 signalling is a central negative regulator of EC proliferation. We confirmed this by demonstrating that HCAEC proliferation under LOSS is significantly enhanced by inhibition of Notch signalling (DAPT; Fig. 6c), by inhibition of JAG1 (blocking antibodies; Fig. 6d) or by NOTCH4 knock-down (siRNA; Fig. 6e, Supplementary Figure 5). Notably, inhibition of JAG1 led to significantly increased EC proliferation in wounded monolayers exposed to LOSS (Fig. 6f), demonstrating that JAG1-NOTCH4 signalling reduces the capacity for endothelial repair. These observations were confirmed in vivo by en face staining of Ki67 which demonstrated that EC proliferation was significantly higher in Jag1 ECKO compared to control mice (Fig. 6g). Collectively, these data indicate that JAG1 reduces the capacity of the endothelium for repair under LOSS by limiting EC proliferation.

DLL4 and JAG1 have divergent functions in atherosclerosis.
Blood flow regulates multiple signalling pathways that control both angiogenesis and atherosclerosis 1 . The Notch pathway has a classical role in developmental angiogenesis 6 , arterial specification 7-9 and neovascularisation 10,11 , and more recently has been characterised as a regulator of atherosclerosis 15 Fig. 7). Inducible deletion of Dll4 from EC of hypercholesterolemic mice led to enhanced atherosclerosis. This function phenocopies Notch1 15 , and we therefore conclude that DLL4-NOTCH1 signalling is essential in limiting atherosclerosis. By contrast, we observed that JAG1-NOTCH4 signalling is an essential driver of atherosclerosis specifically at sites of LOSS. Therefore, we propose that the function of Notch signalling in atherosclerosis differs according to anatomical location; DLL4/NOTCH1 is atheroprotective at HSS regions, whereas JAG1/NOTCH4 induces disease at LOSS sites (Fig. 7). This model suggests that therapeutic targeting of JAG1/NOTCH4 signalling in EC may provide a novel treatment strategy to prevent or treat atherosclerosis.
Using an unbiased RNAseq approach we show that the major function of JAG1/NOTCH4 in EC exposed to LOSS is to limit EC proliferation. This observation was validated experimentally by demonstrating that JAG1 and NOTCH4 are negative regulators of proliferation in cultured EC exposed to LOSS, and that inducible deletion of Jag1 rescues EC proliferation at sites of LOSS in the murine aorta. Several studies found that EC proliferation is enhanced at atherosusceptible LOSS regions of arteries 31,32 , however some have associated EC proliferation with disease processes 33 whereas others suggest that it is atheroprotective 34 . We reconcile these different interpretations by proposing that EC proliferation can be both protective or diseaseinitiating depending on context. On one hand, the ability of EC to proliferate in response to injury at atheroprone sites is vital for vascular repair. Consistent with this, it has been demonstrated that irreversible cell cycle arrest (cellular senescence) of EC at atheroprone sites is an inducer of atherosclerosis 35 . On the other hand, uncontrolled proliferation of intact EC monolayers increases the permeability of arteries to lipoproteins and is therefore pro-atherogenic 33 . In summary, we propose that JAG1/NOTCH4 drives atherosclerosis by reducing proliferative reserve at atheroprone sites, thereby reducing the capacity for repair.

JAG1/NOTCH4 mechanosensing.
Sensing of mechanical force is fundamental to Notch signalling. In the canonical pathway, endocytosis of membrane bound ligand transmits force to the Notch receptor on a contacting cell leading to a structural change that promotes activation via gsecretase cleavage 12 40 and is also activated by HSS in adult arteries to protect them from atherosclerosis 15 . The expression of JAG1 is also enhanced in HUVEC exposed to shear stress compared to static conditions 41 . While physiological HSS has been linked to Notch, we provide the first evidence that Notch signalling is also coupled to pathological disturbed blood flow patterns that drive vascular dysfunction and atherosclerosis. Our conclusion is based on the following In summary, JAG1-NOTCH4 sensing of LOSS is a key driver of atherosclerosis by repressing transcriptional programmes that maintain EC proliferation and repair.
These data demonstrate a fundamental role for JAG1-NOTCH4 in sensing diseasepriming LOSS, and suggest that therapeutic targeting of endothelial JAG1-NOTCH4 could be a novel treatment strategy for atherosclerosis.

Methods
Mice. Mice with inducible deletion of Jag1 or Dll4 in EC were generated by crossing Jag1 fl/fl 42 mice or Dll4 fl/fl mice 43 respectively with CDH5 Cre-ERT2 mice 44 . All mice were on a C57BL/6 background. PCR primers used for genotyping are described in Supplementary Table 1. To activate Cre, tamoxifen (Sigma) in corn oil was administered intraperitoneally (IP) for 5 consecutive days (2 mg/mouse/d). Two weeks after the first injection of tamoxifen, hypercholesterolemia was induced by intraperitoneal (I.P.) injection of adeno-associated virus containing a gain-of-function mutated version of proprotein convertase subtilisin/kexin type 9 (rAAV8-D377Y-mPCSK9) gene (Vector Core, North Carolina) followed by a high fat diet (SDS UK, 829100) for 6 weeks as previously described 45 . Constrictive cuffs were applied to the right carotid artery of anesthetized C57BL/6 mice as described previously 46 Table 2). EC were identified by co-staining using anti-CD31 antibodies. Nuclei were identified using To-Pro-3. Stained vessels were mounted prior to visualization of endothelial surfaces en face using confocal microscopy (Olympus SZ1000 confocal inverted microscope). The expression of particular proteins at each site was assessed by quantification of the mean fluorescence intensities with standard error of the mean.  Table 2). Humanized phage antibody YW152F targeting DLL4 6 was provided by Genentech.

) and
AlexaFluor488-or Alexafluor568-conjugated secondary antibodies. Nuclei were identified using DAPI (Sigma). Images were taken with a widefield fluorescence microscope (LeicaDMI4000B) and analysed using Image J software (1.49p) to calculate the frequency of positive cells. Isotype controls or omission of the primary antibody was used to control for non-specific staining.
In vitro repair assay. HCAEC were cultured until confluent in 6 well plates and exposed to LOSS using an orbital shaking platform housed inside a cell culture incubator and rotating at 210 rpm. This system generated low shear stress (approximately 5 dyn/cm 2 ) with rapid variations in direction at the centre. After three days of culture under these conditions, wounds were created on confluent HCAEC monolayers using a pipette tip. Proliferation at the wound edge was assessed 24h later by immunofluorescence. Isolation of EC from porcine aortae. Pig aortas from 4-6 months old animals were obtained immediately after slaughter from a local abattoir. They were cut longitudinally along the outer curvature to expose the lumen. EC exposed to high (= outer curvature) or low (= inner curvature) wall shear stress were harvested using collagenase (1 mg/ml for 10 minutes at room temperature) prior to gentle scraping.
Real time PCR and RNAseq. RNA from mouse aortas was extracted after mechanical homogenization of the aorta in lysis buffer (Qiagen) using Triple-pure zirconium beads and a microtube homogenizer (Benchmark Scientific) at 4°C. RNA was extracted using the RNeasy Mini Kit (74104, Qiagen) and reverse transcribed into cDNA using the iScript cDNA synthesis kit (1708891, Bio-Rad). QRT-PCR was used to assess the levels of transcripts with gene-specific primers (Supplementary Table   3). Reactions were prepared using SsoAdvanced universal SYBR®Green supermix (172-5271, Bio-rad) and following the manufacturer's instructions, and were performed in triplicate. Expression values were normalized against the house-keeping gene (mouse Tbp, human HPRT or porcine B2M). Data were pooled from at least three independent donors and mean values were calculated with SEM. For RNAseq, the purity and integrity of total RNA samples isolated from HCAEC was assessed using a Bioanalyser (Agilent) and high-quality samples from 4 HCAEC donors were used to prepare RNA-seq libraries that were sequenced on an Illumina HiSeq platform yielding 20 million reads per sample. Library preparation and cDNA sequencing were performed by Novogene. Fastq samples were processed using the RNA-seq pipeline implemented in the bcbio-nextgen project https://bcbionextgen.readthedocs.io/en/latest/index.html. After quality control checking using fastQC, RNA-seq reads were aligned to the human reference genome (assembly GRCh37/hg19) using STAR 47 with the default parameters.
FeatureCounts 48 was used to create a matrix of mapped reads per Ensembl annotated gene. Differential gene expression was performed using the DESeq2 R package 49 .
Functional enrichments for protein coding genes with p value < 0.05 and log2 fold change > 0 were calculated using DAVID 50 using the total genes present as a background set.
Western blotting. Total cell lysates were isolated using lysis buffer (containing 2% SDS, 10% Glycerol and 5% β-mercaptoethanol). Primary antibodies used and concentrations are described in Supplementary Table 2. HRP-conjugated secondary antibodies (Dako) and chemiluminescent detection was carried out using ECL Prime ® (GE Healthcare). Membranes were imaged using the Gel Doc XR+ system (Biorad).
Lipid measurement. Blood samples were collected by terminal cardiac puncture and plasma was separated by centrifugation for further analysis using a COBAS analyser (total plasma cholesterol, non-HDL cholesterol and triglycerides). ATAQseq. ATAC-seq was performed on HAECs subjected to 24 h unidirectional flow or disturbed flow as previously described 51 .

Statistical analysis
Data are presented as means values ± SEM. Statistical analysis were performed with GraphPad Prism software. The degree of significance is as following: *p <0.05; **p<0.01; ***p< 0.001. The test performed is indicated in the figure legend.  The expression of JAG1 protein (red) was visualized in the murine aorta by en face staining. EC were identified using anti-CD31 antibodies (green) and nuclei were costained using TOPRO-3 (blue). (d) Representative images of aortas stained with oil Red O and (e) normalized quantification of plaque burden in the whole aorta, arch, and descending aorta in Jag1 ECKO mice (n=8) compared to littermate Controls (n=9). Representative images (f) and quantification of plaque burden (g) in the aortic roots of controls and Jag1 ECKO mice. In all graphs, each data points represents one mouse and mean ± SEM are shown. Differences between means were analyzed using an unpaired t-test. In all graphs, each data points represents one mouse and mean ± SEM are shown. Differences between means were analysed using unpaired t-tests.  16 . Based on this map, endothelial cells were specifically isolated from LOSS area versus HSS area in the porcine aortic arch. The expression of Notch actors was analysed in each population by qRT-PCR (n=3). (b, c) Aortic arches were isolated from C57BL/6 mice and en face immunostainings were performed using anti-JAG1 (b) or anti-NOTCH4 (c) antibodies (red). The endothelium was stained with anti-CD31 antibodies (green) and co-stained with TO-PRO-3 (DNA; blue). The graphs on the right represent the red mean fluorescence intensity (MFI) (n=4). Differences between means were analysed using paired t-tests.