Endothelial NOTCH1 is suppressed by circulating lipids and antagonizes inflammation during atherosclerosis

To assess the relative levels of Notch receptors in HAECs, we first compared transcripts for all Notch receptors by quantitative RT-PCR (qRT-PCR). Among the four members of the family, NOTCH1 was the most prevalent (Fig. 1 A). At the protein level, NOTCH1 was sixfold higher in HAECs than in smooth muscle cells isolated from the human aorta (HASMCs; Fig. 1, B and C). Immunodetection of NOTCH1 in human coronary arteries further confirmed its expression in adult endothelium (Fig. 1 D, arrows) and also indicated conspicuous absence from vascular smooth muscle cells (Fig. 1 D). Additional measurements of NOTCH1 expression of 14 human coronary arteries from individuals lacking atheroma (NOTCH1 area/CD31 area) uncovered high variability across the specimens (ranging from 7 to 53%; Fig. 1 E). A broad range of expression of Notch1 in large vessel human endothelium was also confirmed by microarray analysis of HAECs isolated from 147 individuals (Fig. 1 F). Therefore, although NOTCH1 is abundant in the endothelium, levels appear to be variable across donors. We were intrigued by this variation and also by the unclear function of Notch1 in adult endothelium. Potential reasons for variability in Notch expression included age, gender, genetic modifiers, and/or epigenetic modifications imposed by predisposing conditions such as diabetes, hyperlipidemia (diet), etc. Evaluation of gender and age provided no insight. Thus, we turned our attention to other factors.

The initial stages of atherosclerosis are characterized by activation and disruption of the endothelial barrier, events that are induced in mice by hyperlipidemia. We next inquired whether hyperlipidemia was a reason behind the variability observed in NOTCH1 expression in vivo. HFD (Fig. 2 A) in mice induced a rapid increase in plasma cholesterol levels (Fig. 2 B). Transcriptional analysis of aortic ECs from these mice fed for 4–13 d revealed that a proatherogenic diet was able to significantly decrease Notch1 and its canonical target gene: Hey1 (Fig. 2 C). Restoration of normal plasma cholesterol levels (4 d HFD followed by 3 d chow; Fig. 2 B) rescued Notch1 signaling (as per Notch1 and Hey1 levels; Fig. 2 C), supporting a direct impact of circulating lipids on Notch1. In addition, we found that endothelial Notch1 and Hey1 mRNA levels were inversely correlated with circulating cholesterol levels at early time points (Fig. 2 D). These data revealed that hyperlipidemia negatively affects Notch1 signaling in adult endothelium in mice. Next, we inquired whether this was also true for human cells, particularly in relation to oxidized (proatherogenic) lipids and cytokines.

Hyperlipidemia and increased levels of inflammatory cytokines are hallmarks of atherosclerosis; we thus tested whether proatherogenic mediators such as oxidized lipids, TNF, and IL-1β could affect NOTCH1 in human ECs. Using HAECs, we found that levels of NOTCH1, as well as the target genes HES1 and HEYL, were reduced by all three stimuli. Although JAG1 was up-regulated upon cytokine treatment, its expression was repressed by Ox-PAPC (Fig. 3, A and B). This suggested that some components of Notch signaling are differentially modulated by inflammatory cytokines versus Ox-PAPC; both common and divergent mechanisms are likely involved in Notch regulation by the two types of stimuli.

To uncover possible common mechanisms that could explain how two distinct effectors (TNF and Ox-PAPC) mediate NOTCH1 suppression, we evaluated a large group of pharmacological inhibitors. Particularly, we asked whether specific inhibitors could revert the ability of TNF and/or Ox-PAPC to suppress NOTCH1 and HES1. From the inhibitors examined, blockade of STAT3 (Stattic) abolished the transcriptional repression of NOTCH1 and HES1 by TNF or Ox-PAPC (Fig. 3, C and D). Importantly, even in the absence of external stimuli, inhibition of STAT3 increased basal levels of NOTCH1 and HES1 (Fig. 3, C and D). These data highlighted a unifying process for the suppression of NOTCH1 downstream two distinct, athero-promoting effectors (TNF or Ox-PAPC) in HAECs.

To assess the reproducibility and biological impact of Ox-PAPC on NOTCH1, we took advantage of our previously published microarray experiments that include assessment of cultured HAECs from 147 individual donors (Romanoski et al., 2010). 40 µg/ml Ox-PAPC suppressed NOTCH1 expression in 80% of the donors after 4 h of treatment (Fig. 3 E). In addition, time course analysis confirmed that NOTCH1 protein was significantly affected after Ox-PAPC treatment and that the repression was maintained over time (Fig. 3, F and G).

A more comprehensive evaluation indicated that although the response was variable between individuals, NOTCH1 was always decreased between 6 and 8 h after Ox-PAPC exposure (Fig. 3, H–J). Furthermore, the Notch ligand JAG1 and target genes HES1 and HEYL were rapidly and irreversibly decreased upon Ox-PAPC exposure (Fig. 3, J and K).

The variability of NOTCH1 at basal conditions, as well as the level and kinetics of down-regulation by Ox-PAPC treatment suggested that genetic variants contribute to the regulation of NOTCH1 by Ox-PAPC. Information from single nucleotide polymorphisms (SNPs) obtained from the 147 donors was used to seek further clarification. We assessed which variants were involved in the response of NOTCH1 to Ox-PAPC by using a genome-wide association study (GWAS) focusing on the level of NOTCH1 down-regulation in response to Ox-PAPC as a quantitative trait. These experiments identified a suggestive peak on chromosome 11 (Fig. 3 L). The most significantly associated SNP (rs2923084) has also been shown to be associated with high-density lipoprotein (HDL) levels in a human GWAS of ∼100,000 individuals (Teslovich et al., 2010; Willer et al., 2013). The SNP is located in a noncoding RNA (CAD11.1) whose function in ECs has not been studied. The nearest protein coding genes are AMPD3 (adenosine monophosphate deaminase 3) and ADM (Adrenomedullin); however, the expression level of neither of these genes is associated with rs2923084. The G allele of the SNP was determined to be the risk allele linked with lower HDL levels in humans compared with the A allele. We found that the individuals with the G allele displayed greater decrease in NOTCH1 transcript in response to Ox-PAPC (Fig. 3 M), suggesting that the genetic variant regulating HDL levels has an effect on the response of NOTCH1 to Ox-PAPC in ECs. Together with the observation that Notch1 levels were correlated with circulating cholesterol in mice (Fig. 2), these data on primary human cells support the notion that endothelial NOTCH1 levels and activity are modulated by hyperlipidemia.

To investigate the biological relevance of Notch1 suppression in early atherogenic events triggered by inflammatory phospholipids, we compared the transcript profile of HAEC knockdown for NOTCH1 with profiles from HAECs treated with Ox-PAPC (Fig. 4). Gene microarray analysis revealed that ∼14% of the transcripts regulated by Ox-PAPC were also regulated in the same direction when NOTCH1 was reduced (Fig. 4 A). Validation by qRT-PCR performed on the same samples showed that Heme oxygenase (HMOX-1) was increased by Ox-PAPC and siRNA-mediated knockdown of NOTCH1 (Fig. 4, B and C). In addition, we observed that suppression of NOTCH1 led to an increase in IL-8 and CXCL1 at the mRNA and protein levels (Fig. 4, C–E). These are two powerful chemokines that favor recruitment of leukocytes to the endothelium during atherosclerosis. These proinflammatory mediators were also consistently overexpressed in the 147 donors after 4 h of Ox-PAPC treatment (Fig. 4, F and G). Noticeably, whereas overexpression of IL8 was rapid and sustained over time, CXCL1 up-regulation showed an early and transient response to Ox-PAPC (Fig. 4, I and J). Moreover, Ox-PAPC treatment and down-regulation of NOTCH1 led to a significant increase in the endoplasmic reticulum stress factor TDAG51 (Fig. 4, B–D), which has been described to favor atherosclerosis (Hossain et al., 2003). Although TDAG51 levels showed high variability across donors at 4 h after Ox-PAPC treatment (Fig. 4 H), a time course analysis revealed that this increase can occur at early time points and transiently in some individuals (Fig. 4 K). Finally, we also observed some differences. Knockdown of NOTCH1 led to an increase in SELE (encoding CD62E/E-Selectin) and CHST1 (Li et al., 2001), a protein involved in leukocyte rolling; this increase was not observed when HAECs were exposed to Ox-PAPC (Fig. 4, B–D).

The shared transcriptional profile between Ox-PAPC and NOTCH1 suppression suggested that repression of Notch1 signaling may be a prerequisite for some of the effects mediated by Ox-PAPC. To test this possibility, we evaluated the effect of Notch1 retention in the presence of Ox-PAPC.

We first transduced HAECs with lentiviral particles expressing the intracellular domain of rat Notch1 that included the γ-secretase cleavage-dependent region (ZEDN1 [Shawber et al., 1996]). This form enables overexpression, but it retains some level of physiological regulatory control through enzymatic processing (Fig. 5 A). ZEDN1 includes the transmembrane domain of Notch1, and thus it is anchored until the γ-secretase complex releases the intracellular domain of the receptor (NICD; in S3; Fig. 5 A). This is important because we and others found that overexpression of the unbound cytoplasmic fragment NICD (independent of γ-secretase complex activity) is toxic and results in senescence and apoptosis (Liu et al., 2012). Using the membrane-retaining construct, we confirmed that rat Notch1 was present only in the cells infected with lentivirus-expressing ZEDN1 but was absent in HAECs transduced with control lentivirus (Fig. 5 B). ZEDN1 was also detected by protein analysis at the anticipated molecular weight (Fig. 5 C, red arrow) and below the endogenous p120-S1 (Fig. 5 C, black arrow). To reduce potential off-target effects caused by supraphysiological expression, we performed all of the evaluations using the lowest possible dose of ZEDN1 (Fig. 5 C). HAECs transduced with ZEDN1 in the absence of Ox-PAPC showed a slight but significant decrease in TDAG51 transcript, whereas JAG1 and HEYL were clearly increased, confirming that the transgene was active on canonical Notch targets. No changes were observed in IL8, CXCL1, and HES1 (Fig. 5 D) in the absence of Ox-PAPC treatment. Expression of ZEDN1 in the presence of Ox-PAPC rescued JAG1, HES1, and HEYL transcripts, indicating that Notch1 activity was retained, even under circumstances when endogenous human NOTCH1 was suppressed (Fig. 5 E). Maintenance of Notch1 signaling was also associated with a partial but significant (50%) rescue of IL8 levels, whereas the effect of Ox-PAPC on TDAG51 was completely abolished in the presence of ZEDN1. In contrast, regulation of CXCL1 by Notch1 and Ox-PAPC appeared to occur through an alternative pathway, as retention of Notch1 signaling did not affect its up-regulation by Ox-PAPC (Fig. 5 F). Thus, in HAECs, repression of NOTCH1 appears to mediate some of the effects of Ox-PAPC (for IL8 and TDAG51), but it also affects gene expression through independent mechanisms (for CXCL1).

Our data demonstrated that NOTCH1 reduction impacted expression of proinflammatory and proatherogenic molecules (Fig. 4). These events are predicted to facilitate recruitment and binding of leukocytes to the endothelium. Therefore, we reduced the expression of NOTCH1 in confluent HAEC monolayers with siRNA (Fig. 6 A) and performed leukocyte adherence assays with labeled THP-1 monocytes. Decrease of NOTCH1 in the confluent monolayer of HAECs in the absence of external stimuli led to a significant increase in monocyte binding (Fig. 6, B and C). Similarly, CD45 immunodetection in endothelial-specific heterozygous deletion of Notch1 (VeCad-Cre⁺, Notch1^F/+; N1^EC+/−) revealed an enhanced binding and subendothelial infiltration of immune cells compared with wild-type mice (N1^ECWT; Fig. 6 D). In accordance with our in vitro data on HAECs (Fig. 4), partial reduction of Notch1 in ECs in vivo resulted in increased CXCL1 detection in the endothelium of N1^EC+/− animals when compared with wild type (Fig. 6 E).

To further assess the biological consequence of loss of Notch1 in the mature aorta, we induced its deletion from the endothelium of adult mice (Cdh5-Cre^ERT2 Notch1^F/F; N1^ECKO). These animals also expressed Cre-inducible Tomato reporter, allowing for visualization of recombined cells. En face fluorescence imaging revealed single and multiple leukocytes bound to the endothelium of N1^ECKO aortae (Fig. 6, G–J, arrows). These findings further support the notion that Notch1 is constantly required to maintain a quiescent status and prevent EC activation and inflammation in the aorta.

Given the previous findings, we speculated that repression of Notch1 might contribute to the onset of atherosclerosis through regulation of EC activation. To test this hypothesis, we crossed mice with specific heterozygous EC-Notch1 (N1^EC+/−) to a diet-induced atherosclerosis model expressing a liver-specific dominant-active form of IDOL (L-sIDOL; Fig. 7 A; Calkin et al., 2014).

Adult N1^EC+/− mice, maintained on a standard diet, showed significantly higher numbers of CD45^POS cells in the aortic arch when compared with the N1^ECWT/L-sIDOL. Similarly, hemizygous deletion of Notch1 in the L-sIDOL background (N1^EC+/−/L-sIDOL) led to increased accumulation of inflammatory cells (Fig. 7, B and C). In addition, we observed that CXCL1 was enhanced in the endothelium of the mice lacking one allele of endothelial Notch1 when compared with L-sIDOL animals (Fig. 7 D).

To assess the contribution of decreased endothelial Notch1 to atherosclerosis, 6-wk-old mice (N1^EC+/−, N1^ECWT/L-sIDOL, and N1^EC+/−/L-sIDOL) were fed a normal (chow) or HFD for 28 wk. Regardless of their genotypes, HFD resulted in elevated levels in plasma cholesterol (Fig. 8 A). No Sudan IV–positive lesions were observed in mice fed a chow diet (Fig. 8, B and D). When fed an HFD, small lesions were detected in the descending aorta and femoral arteries of N1^EC+/− animals (Fig. 8, C and E). Transgenic mice on HFD developed early atherosclerosis; however, N1^EC+/−/L-sIDOL animals showed a significant increase in en face aorta lesions when compared with the N1^ECWT/L-sIDOL (3.58-fold; Fig. 8, C and E; and Table 1). Aortic root lesion size followed the same trend with an increased mean surface in the N1^EC+/−/L-sIDOL compared with the N1^ECWT/L-sIDOL (2.81-fold; Table 1). We also observed extended lesions that invaded the arch branches in N1^EC+/−/L-sIDOL (Fig. 8 E). Together, these findings indicate that hemizygous loss of Notch1 in the endothelial compartment increases susceptibility to atherosclerosis. Altogether, these data indicate that suppression of NOTCH1 in the endothelium may contribute to the onset of atherosclerosis by de-repressing inflammatory and atherogenic molecules (IL-8, CXCL1, and TDAG51).

In this study we showed that constant basal Notch1 signaling protects against EC activation in large vessels. We observed that atherogenic stimuli rapidly repressed Notch1 in vitro and in vivo. Furthermore, reduction in endothelial Notch1, even in the absence of external stimuli triggers endothelial activation, increased leukocyte binding, and overexpression of proatherogenic molecules. Importantly, data presented here also showed that Notch1 levels are affected by circulating lipids. Finally, this study demonstrates that Notch1 reduction in the endothelium increases diet-induced atherosclerosis. Additional evidence for a role of Notch signaling in atherosclerosis was provided recently by two integrative genomic studies that examined GWAS results for coronary artery disease. These studies evaluated data-driven tissue-specific gene networks (Mäkinen et al., 2014) and Reactome pathways (Ghosh et al., 2015) and identified genetic signals to be enriched in NOTCH pathway genes. In accordance with these findings, our data provide a link between endothelial Notch1 regulation and atherosclerosis and reveal mechanisms responsible for this outcome.

The role of Notch signaling in atherosclerosis has been previously studied in human and mouse models but yielded contradictory conclusions (Quillard et al., 2008, 2010; Aoyama et al., 2009; Nus et al., 2011; Fukuda et al., 2012; Liu et al., 2012; Rizzo et al., 2013; Schober et al., 2014). Administration of γ-secretase inhibitors, and therefore blockade of global Notch signaling, was associated with a decrease in diet-induced plaque burden in ApoE^−/− mice. This effect was attributed to a reduction of Notch activity in MOMA2⁺ macrophages (Aoyama et al., 2009). Although relevant when considering a systemic blockade of the disease, γ-secretase inhibitors are broad spectrum and impact several enzymatic pathways, signaling molecules, and receptors (De Strooper, 2003). Other studies have also found that administration of antibodies to the ligand Dll4 attenuates atherosclerosis. Once again the mechanism was linked to an effect on macrophages (Fukuda et al., 2012). Conversely, other studies have proposed a protective role for Notch signaling in several vascular diseases. Nus et al. (2011) showed that mice with global haploinsufficiency of the Notch pathway effector RBP-jk developed diet-induced calcific aortic valve disease with accumulation of macrophages and collagen deposition. A protective role for Notch signaling has also been supported by a study of Notch2 and Notch4 on ECs (Quillard et al., 2010) of small coronary vessels in rats with heart transplants (Quillard et al., 2008). There, the authors observed that decrease in Notch4 impaired vascular repair after transplantation in vivo and proposed a role for Notch signaling in the maintenance of EC quiescence. They also proposed that Notch was required for optimal survival and repair in response to injury. More recently, it was demonstrated that endothelial Mir-126 was antiinflammatory and essential for the maintenance of a proliferative pool of progenitor ECs in the aorta. Administration of this microRNA to mice limited atherosclerosis via the repression of the Notch antagonist Dlk1, once again implying an inhibitory role for Notch signaling in atherosclerosis (Schober et al., 2014).

In the present work, we provide evidence that NOTCH1 is the most predominant Notch receptor in arterial ECs of human and mouse (Fig. 1; Briot et al., 2014). We also showed that Notch1 is a key regulator of arterial EC homeostasis and that it functions as a sensory link between circulating lipids and atherosclerosis susceptibility. We observed that suppression of Notch1 in the absence of external stimuli leads to overexpression of major proinflammatory and atherogenic mediators (IL8, CXCL1, SELE, and TDAG51) and increased binding of leukocytes in vitro and in vivo. In accordance with previous studies, it is undeniable that partial loss of Notch1 has a broad impact on the EC transcriptome. Theodoris et al. (2015) showed that heterozygous mutation of NOTCH1 in iPSC-derived ECs altered the transcriptional regulation of multiple genes, with predominant signatures in osteogenic and inflammatory genes. Similarly, using siRNA, we found a large cohort of transcription factors altered when NOTCH1 was reduced in adult HAECs. Many of these factors are known to regulate the inflammatory response (such as STAT, NF-κB, CEBP, and EGR1). Interestingly, some of these genes are known to be activated in response to Ox-PAPC treatment (CREB, SREBP, ATF3, and STAT3; Lee et al., 2012). We also noted changes in cofactors involved in NOTCH signaling and chromatin modification (SIN3, HDAC, CoREST, and p300; Table 2). Thus, repression of Notch1 in arterial endothelium results in profound transcriptional and functional changes, leading to proinflammatory states.

Among its proinflammatory targets, NOTCH1 knockdown in HAECs led to a significant increase in CXCL1 expression. Accumulation of CXCL1 on the luminal side of the endothelium has been shown to promote monocyte recruitment in ApoE^−/− mice that was blocked by injection of antibodies targeting the cytokine (Zhou et al., 2011). Thus, the intrinsic increase of CXCL1 observed in HAECs with knockdown of NOTCH1 and in the endothelium of N1^EC+/− mice indicates that this transcription factor could be protective during the early phases of atherosclerosis. This notion was further supported by our observation that hemizygous loss of Notch1 in the endothelium (N1^EC+/−) increased diet-induced atherosclerosis in the L-sIDOL background. Thus, we propose that the intrinsic endothelial activation state of the N1^EC+/−/L-sIDOL aggravates their sensitivity to the development of atherosclerosis.

Our data provide evidence for the role of oxidized lipids and cytokines associated with atherosclerosis in the regulation of endothelial Notch1. Indeed, in vitro TNF as well as atherogenic doses of Ox-PAPC rapidly repress NOTCH1 through a Stat3-dependent mechanism. Importantly, both Ox-PAPC and TNF were previously shown to activate STAT3 (Guo et al., 1998; Lee et al., 2012); therefore, this appears to be a unifying mechanism that results in NOTCH1 suppression by distinct effectors.

It is important to highlight, however, that Ox-PAPC and suppression of Notch1 (by siRNA) are not identical in the regulation of all inflammatory genes. For example, although Notch1 suppression increases E-Selectin and CHST1 Ox-PAPC does not. This might be at first counterintuitive because Ox-PAPC induces Notch1 repression. However, Ox-PAPC is composed of several active oxidized phospholipids such as 1-Palmitoyl-2-(5-oxovaleroyl)-sn-glycero-3-phosphorylcholine (POVPC) and 1-palmitoyl-2-glutaroyl-sn-glycero-3-phosphorylcholine (PGPC; Watson et al., 1997). These subproducts have been shown to selectively activate distinct pathways. In particular, it has been shown that although POVPC represses the induction of E-Selectin by LPS, PGPC induces its expression (Leitinger et al., 1999). Thus, this apparent competing effect of Ox-PAPC active subproducts may account for the differences observed between “simple” reduction of NOTCH1 by siRNA and treatment with Ox-PAPC.

Although the effect of lipid products in vitro and long-term HFD have been shown to impact the endothelium transcriptome (Lee et al., 2012; Erbilgin et al., 2013b), we show that transcriptional repression of Notch1 signaling molecules in vivo occurs as early as 4 d after starting an HFD. This effect was reversible, as switching from HFD to normal diet rescued Notch1 signaling, demonstrating a link between diet and endothelial Notch1 signaling. The unexpected regulation of Notch1 by circulating lipids was also supported by additional evidence from mice and human studies. In fact, we observed a negative correlation between circulating cholesterol levels and transcriptional expression of Notch1 in vivo. Human genetic experiments showed a significant association between a genetic variant on chromosome 11 with NOTCH1 responses to Ox-PAPC in HAECs (Fig. 3, L and M). This variant was previously associated with HDL levels in a human GWAS (Teslovich et al., 2010). In fact, the G allele of this polymorphism was associated with a greater decrease of NOTCH1 under Ox-PAPC stimulus (our data, Fig. 3 M) and a lower level of HDL in the previous study compared with the A allele (Teslovich et al., 2010). These data indicate that genetic variants that modulate NOTCH1 (as per our study with 147 endothelial donors) are playing an important role in the effect of lipids on atherosclerosis.

The protective role of HDL against oxidation of lipids, the effect of oxidized lipids on the endothelium, and subsequent onset of atherosclerosis (Mineo and Shaul, 2013; Riwanto and Landmesser, 2013; Emert et al., 2014; Rye and Barter, 2014) reinforces the concept that low levels of NOTCH1 constitute an important predisposing and permissive factor to endothelial activation and atherosclerosis. In addition, the data presented here offer a link between diet and endothelial changes, leading to the acquisition of a proinflammatory status that precedes development of the disease. In this regard, Notch1 could be considered as a signaling hub that antagonizes the transduction of proatherogenic stimuli, thus preventing endothelial activation.

Tissue sections were deparaffinized, and antigen retrieval was performed with boiling citrate buffer (10 mM in PBS, pH 6) for detection of NOTCH1, CD45, αSMA, and CXCL1. Dako S1700 solution was used for detection of CD31. The source of the primary antibodies used include NOTCH1 and CD31 human (Cell Signaling Technology), αSMA and β-Catenin (Sigma-Aldrich), CXCL1/GROA (Abcam), and CD45 (BD). Imaging was performed with a BX40 light microscope (Olympus) or LSM 710 multiphoton confocal microscope (Carl Zeiss) using Zen software (Carl Zeiss). 3D reconstruction was generated using Imaris software (Bitplane).

We retrospectively reviewed microscopic slides of coronary arteries from 16 patients who received a heart transplant at the University of California, Los Angeles (UCLA), from 2008 to 2013. Tissues that underwent decalcification were excluded from the study (5 patients) because of poor immunohistochemical staining. In some cases, more than one cross section was used from a single coronary. Tissues were fixed with neutral-buffered formalin and embedded in paraffin, and 4-µm serial sections were taken. Staining of the serial sections was performed as described in the previous paragraph using EDTA, pH 8, antigen retrieval. Counter staining with hematoxylin was not performed on the slides stained with CD31 and NOTCH1 to optimize visualization. For image analysis, slides stained for NOTCH1 and CD31 were scanned with a digital slide scanner at 20× (Aperio XT scanner; Aperio Technologies). Randomly chosen relatively linear regions of endothelium (500 µm in length) were manually circled, and the circled areas were examined by Tissue Studio software (Definiens Inc.) to calculate the area positive for NOTCH1 and CD31. The same endothelial regions were carefully circled on NOTCH1 and CD31 slides from each tissue. For the visual analysis, we compared NOTCH1 and CD31 staining. Two observers examined these images. The study was approved by the Institutional Review Board (IRB) at UCLA (IRB#12-000608).

For atherosclerosis measurement, aortae were dissected, fixed (4% PFA, 5% sucrose, and 20 µM EDTA), and stained with Sudan IV. Images were captured with a CCD camera (DXC-97MD 3; Sony). Atherosclerosis in the aortic roots and the descending aortas (en face) were quantified by computer-assisted image analysis as described previously (Tangirala et al., 1995, 1999; Shih et al., 1998). Atherosclerotic lesions at the aortic valve were stained with Oil Red O and analyzed as described previously (Shih et al., 1998). Lesion development is expressed as the percentage of total aortic surface covered by lesions (Bradley et al., 2007). Measurement of the lesions was performed by blinded investigator (mouse genotype and diet unknown).

HAECs were isolated from aortic explants of heart transplant donors in association with the UCLA Transplant Program. Isolation of HAECs involves selective enzymatic digestion and results in high purity of the ECs as previously described (Navab et al., 1988; Romanoski et al., 2011). Confluent HAECs in Medium 199 (Corning) supplemented with 1% heat-inactivated fetal bovine serum and 1% antibiotics were treated with Ox-PAPC as described in a previous publication (Romanoski et al., 2010), 10 ng/ml recombinant human TNF, or 10 ng/ml IL-1β. The dose of Ox-PAPC (40–50 µg/ml) used for these experiments was chosen based on experiments from our laboratory and others showing consistent atherogenic effects at this concentration (Lee et al., 2012). Although Ox-PAPC can have antiinflammatory action on bacterial products at much lower concentrations (Oskolkova et al., 2010), aortic extracts of oxidized phospholipid at the concentration used in this study have been shown to have proinflammatory effects on ECs (Subbanagounder et al., 2000; Oskolkova et al., 2010). For rescue experiments, confluent HAECs were pretreated with 10 µM Stattic (Selleck Chemicals) or vehicle control for 30 min; the cells were then treated with control media or media containing 10 ng/ml TNF or 40 µg/ml Ox-PAPC in the presence or not of the inhibitor for 4 h.

C57BL/6 mice were fed an HFD (#TD88137; Harlan Laboratories, Inc.) for the indicated time and fasted overnight before blood and tissue collection. For rescue experiments, mice were fed an HFD for 4 d and then switched to a standard diet for an additional 3 d.

Endothelial Notch1 heterozygous (N1^EC+/−) mice were crossed with L-sIDOL mice. 6-wk-old males were fed standard chow or an HFD (#D09062501R; Research Diet Inc.) for 28 wk. Mice were fasted overnight before euthanasia. VeCad-Cre/Notch1-floxed mice and L-sIDOL were previously described (Hofmann et al., 2012; Calkin et al., 2014).

For inducible endothelial deletion of Notch1, mice expressing floxed alleles of Notch1 (Radtke et al., 1999) were crossed to Cdh5(PAC)-CreERT2 transgenic animals (Wang et al., 2010) and mice expressing tdTomato reporter (Madisen et al., 2010). 5-wk-old mice received intraperitoneal injections of tamoxifen (1 mg/mouse; MP Biomedicals) for three consecutive days. After 2 wk, the animals were injected with methacholine chloride (6.5 mg/mouse; MP Biomedicals) in PBS and euthanized. Injection of methacholine chloride 2 min before euthanasia induces vascular smooth muscle relaxation and allows for en face imaging of the aorta. Aortae were harvested and fixed in 2% PFA before staining and mounting for en face confocal imaging. Mouse experiments were conducted in accordance with UCLA Department of Laboratory Animal Medicine’s Animal Research Committee guidelines.

Total mouse serum cholesterol and triglyceride were measured using Pointe Scientific Reagent Set (#C7509 and #T7531, respectively; Pointe Scientific). Data are represented as mean ± SEM.

Endothelial RNA-enriched fractions from mouse aortae were prepared as described previously (Briot et al., 2014). Total RNA from HAECs was purified using the RNeasy Mini kit (QIAGEN). Complementary DNA synthesis was performed with Superscript III reverse transcription First-Strand synthesis kit (Invitrogen) using oligo-dT primers.

qRT-PCR was performed using primers designed for human targets, provided in Table S1. Primers for mouse targets have been described previously (Briot et al., 2014). Each reaction was run in duplicate and normalized with HPRT housekeeping gene. P-values were calculated using the appropriate Student’s t test. Data are represented as mean ± SEM. For gene microarray analysis, only RNA samples with RNA integrity numbers (RIN) of 7.0 or higher were used for subsequent processing.

The microarray results used for the analysis were described previously (Romanoski et al., 2010). Confluent cells from 147 donors were treated in duplicate with either control media or 40 µg/ml Ox-PAPC for 4 h. mRNA expression profiles from four samples per donor (two control treated and two Ox-PAPC treated) meeting quality control conditions were determined using the HT HG-U133A microarray (Affymetrix), which contains 18,630 probes. Intensity values were normalized with the robust multiarray average normalization method implemented in the affy package in Bioconductor (Gautier et al., 2004).

For expression profiling after siRNA-mediated NOTCH1 silencing, HAECs transfected with siRNA targeting NOTCH1 or scrambled siRNA in the presence or absence of 50 µg/ml Ox-PAPC (6 h) in triplicate were used. Total RNA was hybridized to human HT-12 v4 Expression BeadChips (Illumina). Genome Studio software (2010.v3) was used for determination of probe fluorescence intensities. HT-12 BeadChip contains 48,804 expression and 786 control probes. The probes were processed using nonparametric background correction followed by quantile normalization with both control and expression probes used by the neqc function in the limma package (R v2.13.0; Ritchie et al., 2007). 20,202 probes with detection p-values <0.01 in any of the samples were applied for any further analysis. Differential gene expression analyses to compare cells transfected with siRNA against NOTCH1 or scrambled siRNA and cells treated with Ox-PAPC or cells transfected with scrambled siRNA was performed using the limma package in R (v2.13.0). Calculated p-values from the moderated Student’s t statistics were adjusted using the Benjamini-Hochberg method for multiple testing. Microarray probes with adjusted values of P < 0.01 (1% FDR) and fold change >25% were considered differentially expressed. The dataset for the gene microarray analysis on HAECs treated with NOTCH1 siRNA or Ox-PAPC was deposited in the National Center for Biotechnology Information Gene Expression Omnibus database under accession no. GSE72633.

Association of the SNPs and endothelial gene expression was described previously (Erbilgin et al., 2013a). In brief, HAEC genomic DNA was isolated using the DNeasy kit (QIAGEN). SNP genotyping was performed using the SNP 6.0 microarray platform (Affymetrix) as described previously (Romanoski et al., 2010). Microarray images were processed using the Genotyping Console 4.1 (Affymetrix) to make the SNP calls. SNPs were filtered out based on the following criteria: 296,063 SNPs with minor allele frequency <10%; 18,030 SNPs with <95% genotyping rate among the donors; 13,213 SNPs were set to missing because they were detected as heterozygous in haploid genotypes; and 3,890 SNPs were excluded because they failed the Hardy-Weinberg equilibrium test (P < 10⁻⁴). A mixed model approach to account for the population structure as implemented in the EMMAX program was used to perform the association analysis (Kang et al., 2010).

We applied the linear mixed model

where µ = mean, x = SNP, β = SNP effect, and u = random effects caused by genetic relatedness, with Var(u) = σ_g²K and Var(e) = σ_e², where K = IBS (identity by state) matrix across all genotypes in the panel. We computed a restricted maximum likelihood estimate for σ_g² and σ_e², and we performed association based on the estimated variance component with an F test to test β does not equal 0. The association analysis of SNPs with the difference in the expression value of the microarray probe set for NOTCH1 between Ox-PAPC treatment and control conditions was performed using the 574,391 informative SNPs that passed the filtering criteria. 108 males and 39 females were present in the donor population; therefore, sex was considered as a covariate in the mixed model for association analysis.

Cells were lysed in modified RIPA buffer containing 1% Triton X-100 and 10% SDS. Protein lysate were resolved and analyzed by Western blot using the following primary antibodies: NOTCH1 and VEGFR2 (Cell Signaling Technology), SM22 (Santa Cruz Biotechnology, Inc.), and Calponin and γ-Tubulin (Abcam). Quantification of bands by densitometry analysis was performed using ImageLab Software (Bio-Rad Laboratories).

HAECs were transfected with stealth RNAi targeting NOTCH1 (NM_017617.3_Stealth_775) or control duplexes (Invitrogen) using siPORT Amine (Ambion) as the transfection agent. For better efficiency, the cells were transfected twice with 24 h of recovery in between; cultures were then used for experiments 24 h after the second transfection.

IL-8 and CXCL1/GROα levels in HAEC culture supernatants were measured with IL-8 and CXCL1 Quantikine ELISA kit (R&D Systems) according to the manufacturer’s recommendation.

Lentivirus-based vectors encoding ZEDN1 (Shawber et al., 1996) and GFP genes were generated by transient cotransfection of 293T cells with a three-plasmid combination, as described previously (Naldini et al., 1996). In brief, 100-mm dishes of nonconfluent 293T cells were cotransfected with 6.5 µg pMDLg/pRRE, 3.5 µg pMDG (encoding the VSV-G envelope), 2.5 µg pRSV–REV, and 10 µg pRRL-CMV-ZEDN1-Ires-GFP by the CaPi-DNA coprecipitation method (Chen and Okayama, 1987; Sakoda et al., 1992). The following day, the medium was adjusted at a final concentration of 10 mM sodium butyrate, and the cells were incubated for 5 h to obtain high-titer virus production as previously described (Sakoda et al., 1999). After 5 h, cells were washed and incubated in fresh medium without sodium butyrate. Conditioned media was harvested the next day and passed through 0.45-mm filters. Viral titer was determined by assessing viral p24 antigen concentration by ELISA (the Alliance HIV-I p24 ELISA kit) and hereafter expressed as micrograms of p24 equivalent units per milliliter.

HAECs at 50% confluency were transduced overnight with an equal amount of lentivirus expressing ZEDN1 (Shawber et al., 1996) or GFP control in the presence of 4 µg/ml protamine sulfate. Cells were allowed to recover and express the construct for an additional 48 h before experimentation. Primers designed to detect rat Notch1 are forward, 5′-CAACTCTCACGCTGATGTCAA-3′; and reverse, 5′-GCAACACTTTGGCAGTCTCA-3′.

Confluent HAEC monolayers were cocultured with human monocytic cell line THP-1 stained with Vybrant CFDA-SE vital dye (Invitrogen) as previously described (Valenzuela et al., 2013). After 10 min at 37°C, 5% CO₂ in static conditions, unbound THP-1 cells were washed and the co-cultures were fixed with 4% PFA. Bound CFDA-SE⁺ THP1 cells were imaged with an inverse fluorescent microscope (Carl Zeiss). Bound cells were counted in 10 fields per condition using ImageJ software (National Institutes of Health).

Appropriate Student’s t test was used to determine significant differences between two groups using Prism software (GraphPad Software). P < 0.05 was considered significant and reported to the graphs. Data are represented as mean ± SEM.

The IRB evaluated the study and determined that the specimens used for isolation of HAECs qualified for exception number 4. The study of microscopic slides of coronary arteries was approved by the IRB at UCLA (IRB#12-000608). Animal experiments were conducted in accordance with UCLA Department of Laboratory Animal Medicine’s Animal Research Committee guidelines (#2005-113-23).

Table S1 includes primer sequences for human cDNA amplification.

The authors wish to thank Michelle Steel, Benjamin Emert, and Longsheng Hong for valuable technical assistance; Dr Geraldine Weinmaster for providing the plasmid-encoding ZEDN1 construct; Dr. Freddy Radtke for the Notch1 floxed mice; the contribution of the Tissue Procurement Core Laboratory Shared Resource at UCLA; and the UCLA Vector Core supported by CURE/P30 DK041301 for the cloning and virus production. The Illumina Beadchip hybridizations were performed at UCLA Neurosciences Genomics Core.

This study was supported by funds from the American Heart Association #14POST20380819 (to A. Briot) and #13BGIA17110079 (to C. Hong); American Diabetes Association grant 1-14-JF-33 (to C. Hong), Ruth L. Kirschstein National Research Service Award T32HL69766 and K99HL121172 (to M. Civelek), HL28481 (to A.J. Lusis) and HL30568 (to M.L. Iruela-Arispe, A.J. Lusis, J.A. Berliner, A.M. Fogelman, P. Tontonoz, and M. Navab), the Piansky Family Endowment (to M.C. Fishbein), and a grant from the National Institutes of Health (R01 HL085618 to M.L. Iruela-Arispe). S.D. Lee was supported by a postdoctoral fellowship from the Canadian Institutes of Health Research.

A.M. Fogelman and M. Navab are principal investigators in Bruin Pharma, and A.M. Fogelman is an officer in Bruin Pharma. All other authors declare no competing financial interests.