Context-dependent regulation of endothelial inflammation and atherosclerosis by endothelial microRNA-33

Abstract

Background and Aims

Atherosclerosis arises through the metabolic and inflammatory perturbation of numerous cells, including immune cells and endothelial cells (ECs). microRNA-33 (miR-33) regulates lipid metabolism and inflammatory responses of immune cells, but the impact of miR-33 on atherosclerosis progression has mixed effects, pointing to context- and cell-type-specific functions. Notably, the role of EC miR-33 in atherosclerosis remains unexplored, despite the central involvement of metabolic and inflammatory pathways in EC function. We sought to determine the definitive role of EC miR-33 in atherosclerosis progression.

Methods

We generated mice with an inducible EC-specific miR-33 knockout (iECKO), followed by PCSK9-AAV8 injection and western diet feeding. Detailed plaque analyses and scRNAseq were performed. For acute inflammation, we analyzed TNFα-mediated leukocyte recruitment in the air pouch model. In vitro approaches included the culture of human aortic ECs to analyze gene expression under inflammatory and lipid-laden conditions with miR-33 mimicry.

Results

iECKO mice showed accelerated lesion initiation, but this effect did not persist in advanced atherosclerosis, which is likely due to chronic hypercholesterolemia-driven downregulation of miR-33 that masks its deletion at later stages. Transcriptomic analyses revealed that cholesterol loading alters EC responses to inflammation, which can be partially rescued by miR-33 mimicry. Accordingly, iECKO mice exhibited heightened sensitivity to acute, normocholesterolemic inflammation, which is paralleled with regulation of E-selectin levels.

Conclusions

Our work underscores the nuanced effects of miR-33 manipulation and highlights how well-described regulators of atherosclerosis progression may have unique cell type- and disease stage-dependent roles. Additionally, our work further implicates miR-33 as a regulator of EC function and identifies a new potential role in acute inflammation via E-selectin regulation.

Graphical Abstract

INTRODUCTION

One of the earliest events in atherosclerosis is the activation of endothelial cells (EC) by minimally modified low density lipoproteins (LDLs) that accumulate in the subendothelial space during hypercholesterolemia.[1–3] Subsequent EC inflammation and barrier disruption facilitate recruitment of leukocytes, which further perpetuate endothelial activation and inflammation by secreting cytokines such as tumor necrosis factor alpha (TNFα).[2,3] These cytokines are considered major regulators of EC inflammation, however, growing evidence suggests that microRNAs (miRNAs) and metabolic cues also play a role.[4–9] Notably, EC metabolism and inflammatory status differ between atheroprone and atheroprotected aortic regions,[8,10] and pro/anti-atherogenic stimuli are directly linked to changes in EC metabolism.[7,11–13] Given these connections, targeting EC metabolism to prevent EC activation is of therapeutic interest. In fact, microRNAs are known regulators of EC metabolism and inflammation and have the potential to finely tune cell behavior in cardiovascular disease.[14,15]

Intronic miRNA-33 (miR-33) is co-transcribed with host Srebp genes, which are master transcriptional regulators of lipid metabolism.[16,17] Humans express two miR-33 isoforms: miR-33a, (within Srebf2); and miR-33b (within Srebf1), while mice express only miR-33a.[16,18,19] However, the targets of miR-33a and miR-33b are highly conserved, including cholesterol efflux transporters ATP binding cassette transporters (ABCA1 and ABCG1 in mouse, only ABCA1 in humans) and fatty acid oxidation (FAO) enzymes Carnitine Palmitoyltransferase 1 (CPT1a), Carnitine O-octanoyltransferase (CROT), and Hydroxyacyl-CoA Dehydrogenase Subunit Beta (HADHB),[16,20–22] among numerous other transcripts involved in proliferation and signaling.[23] However, most studies on miR-33 have focused on its role in hepatic and inflammatory cells,[24,25] and its role in endothelial cells in atherosclerosis is yet unknown. Determining the cell type-specific functions of miR-33 in atherosclerosis is critical, since previous studies have shown that deletion or therapeutic inhibition of miR-33 have varying effects depending on the model used and stage of disease investigated.[25–30]

Importantly, the targets of miR-33 have been independently implicated in EC pathobiology in atherosclerosis, thus highlighting the potential relevance of gene expression regulation by endothelial SREBP/miR-33 themselves in atherogenesis: for example, Abca1/g1 deletion in mice promotes EC oxysterol accumulation, leading to vascular dysfunction and atherosclerosis, suggesting that miR-33-regulated cholesterol efflux pathways may play a role in EC in this disease.[31–33] In fact, cholesterol efflux has been broadly implicated in overall EC function.[34] Additionally, FAO enzymes regulate EC proliferation,[35] permeability,[36] and atherosclerosis-related processes such as the endothelial-to-mesenchymal transition (EndoMT).[37]

To investigate the role of EC miR-33 in atherosclerosis, we generated mice with an inducible, EC-specific knockout of miR-33 (iECKO). Here, we report that while iECKO mice exhibit an increase in early lesion initiation, this effect is ablated in later stages of atherosclerosis under prolonged hypercholesterolemia. This phenomenon coincides with the downregulation of the miR-33 host gene Srebf2 in control mice, likely obscuring the regulatory role of EC miR-33 in advanced disease. We confirmed that miR-33 is indeed expressed in EC and regulates its bona fide targets in cholesterol efflux and FAO. However, miR-33 gene regulatory functions are significantly altered in EC under inflammatory and hypercholesteremic conditions. Thus, we further tested its role in acute inflammation under, ‘normolipidemic’ conditions using an air pouch model of inflammation and found a significant increase of immune cell recruitment in miR-33 iECKO mice. Thus, our work demonstrates the highly context-dependent nature of miR-33 in inflammation and atherosclerosis and suggests that miR-33 is a key regulator of EC-dependent acute inflammation.

MATERIAL AND METHODS

Additional information and procedures can be found in Supplementary Material.

Experimental Animals

tdTomato reporter mice (Jackson Laboratory #007909) were bred to mice with an inducible endothelial Cdh5-creERT2.[38] miR-33 flox/flox mice[28] were then crossed to generate miR-33 f/f; Cdh5CreERT2; tdTom mice (iECKO). Cdh5CreERT2; tdTom were used as control mice (WT). All mice were in the C57BL/6J background. To induce deletion of miR-33, both WT and iECKO mice were intraperitoneally injected with tamoxifen (2 mg/mouse/day, #T5648) once a day for five days.[39] To induce atherosclerosis, mice received a single injection of the proprotein convertase subtilisin/kexin type 9 (PCSK9) adeno-associated virus serotype 8 (AAV8) (Penn Vector Core) to induce hypercholesterolemia[40] then fed a western diet (WD) containing 40% kcal from fat and 1.25% added cholesterol (Research Diets D12108). In other instances, after tamoxifen, WT and iECKO mice were used in an air pouch model of acute inflammation, described below. At the time of sacrifice, mice were euthanized via an overdose of isoflurane, followed by physical euthanasia by exsanguination and removal of vital organs (heart, liver, aorta). For temporary anesthesia, the drop method was used in which mice were placed in a container containing 20% isoflurane until loss of righting reflex and a breath rate of 60 breaths per minute. All animals were housed in a barrier animal facility with constant temperature and humidity on a 12h light-dark cycle with ad libitum food and water access. All the experiments were approved by the Institutional Animal Care Use Committee of Yale University School of Medicine and are in accordance with the National Institutes of Health (NIH) Guide for the Care and Use of Laboratory Animals.

Cell Culture, Treatment, and Transfections

Human aortic endothelial cells (HAEC, Lonza #CC-2535) were cultured in EGM-2 media (Lonza #CC-3162) at 37°C and 5% CO₂. For conditions of miR-33 inhibition or mimicry, transfection was carried out with Oligofectamine (Invitrogen #12252011) and Optimem media (Gibco #31985070) starting at ~50% confluency.[41] In some cases, cells were treated with TNFα (R&D #210TA020CF) at 10 ng/mL for 24h prior to cell harvesting. To induce cholesterol loading, native low density lipoprotein (LDL), isolated as previously described,[42] were added to a final concentration of 50 mg/dL of cholesterol for 48h prior to harvest. To investigate regulation of E-selectin levels by miR-33, HUVEC and HDMEC were cultured in M199 Media (Cytiva SH30253.FS) + 1:200 ECGS + 20% FBS and EGM-2 MV (Lonza #CC-3202) respectively at 37°C and 5% CO2.

Air Pouch Model

TNFα was used to induce neutrophil recruitment via the air pouch model of acute inflammation.[43] Briefly, the pouch as created on Day 0, re-inflated on Day 3, and 10 ng of TNFα (R&D 410-MT-025) dissolved in 1 mL sterile saline was injected into the pouch on Day 5. After 12h, the mouse was sacrificed and pouch lavaged with sterile saline. The total number of pouch-recruited cells was then quantified using a hemocytometer. The exudate was then centrifuged for 8 min at 350xg at 4°C and samples were fixed in 2% PFA at room temperature for 10 min, washed thrice with PBS, and stored at 4°C until FACS analysis.

Sex as a biological variable

Atherosclerosis analysis was performed in both sexes. For in vitro studies, HAEC were commercially purchased from both male and female donors.

Plasma lipid measurements

Blood was collected from the retro-orbital venous plexus of overnight fasted mice. Plasma was collected via centrifugation for 10’ at 4°C at 10,000xg. Total plasma cholesterol, high density lipoprotein (HDL)-cholesterol and triglycerides were enzymatically measured according to the manufacturer’s instructions (Wako #999–02601, 998–02992, 992–02892).

Histology and Immunofluorescence and immunohistochemistry

At the time of sacrifice, mice were euthanized by an isoflurane overdose and exsanguinated via cardiac puncture through the right ventricle. Mice were then perfused and heart and aorta were harvested followed by overnight fixation in 4% PFA at 4°C. The aorta was then opened longitudinally and incubated in Oil-Red-O stain prior to mounting on slides for imaging. The apex of the post-fixed heart was then discarded, and the remaining tissue was incubated in 30% sucrose in PBS overnight at 4°C. Samples were then mounted in Optimal-Cutting-Temperature compound (OCT) and oriented in transverse with respect to the aortic root. Cryosections were obtained at 6 μm thickness. Histological and immunofluorescent staining were performed as described in Supplemental Methods.

Western blotting

HAEC were lysed and sonicated in RIPA buffer supplemented with protease inhibitors. Total protein concentration was then measured using the BioRad Protein Assay (#5000116). Equal amounts of reduced and boiled protein were loaded for SDS-PAGE electrophoresis. Immunoblots were performed as indicated in Supplementary Table 1 overnight at 4°C.

Mouse Endothelial Cell Isolation

Following sacrifice, lungs were excised and digested followed by isolating viable tdTom+ cells by FACS directly into Trizol LS reagent (Qiagen #10296010).

Bulk RNA sequencing and analysis

EC were lysed in Trizol (#15596–018) and RNA was isolated with the miRNeasy Mini Kit (Qiagen) according to the manufacturer’s instructions. Library prep and sequencing was performed as previously described.[39] Paired-end reads were imported into Partek Flow (Copyright ©; 2018 Partek Inc., St. Louis, MO, USA) for STAR alignment to GRCh38 (hg38) and annotated with Ensembl. Raw FASTQ files and processed data are available at Gene Expression Omnibus (GSE299560). Resultant gene count table imported into R for further analysis as described in Supplemental Methods.

RNA isolation and quantitative real-time PCR

HAEC were lysed in Trizol and RNA was isolated with the Direct-Zol RNA Miniprep Kit (Zymo Research #R2052). cDNA was prepared using Bio-Rad iSCRIPT RT Supermix (#1708841). qPCR was performed using EvaGreen Supermix (Bio-Rad #P1725201). mRNA levels were normalized to 18S rRNA. Primer sequences in Supplementary Table 3. For qPCR of microRNAs, cDNA was prepared with the miRCURY LNA RT Kit (Qiagen #339340) and qPCR performed using the miRCURY LNA SYBR Green PCR Kit (Qiagen #339346). The U6 snRNA assay (Qiagen #YP02119464) used as a normalization control.

FACS Analysis

Fixed exudate cells were blocked for 15 minutes using CD16/CD32 antibody, followed by staining in an immune cell antibody cocktail (Supplementary Table 1). Immune cell subpopulations were analyzed by flow cytometry.

FAO Assay

FA oxidation in HAECs was determined using [¹⁴C] palmitate (Perkin Elmer Revvity # NEC075H050UC).[39] HAECs were incubated with reaction mixture (0.3% BSA/50 μM palmitate/0.5 μCi/mL ¹⁴C-palmitate) for 3h. The reaction mixture was then transferred to a 1.5 mL Eppendorf tube and ¹⁴CO₂ was then released by perchloric acid and trapped in sodium hydroxide-soaked Whatman papers. Radioactivity adsorbed onto the filter disc was quantified by a liquid scintillation counter.

Single Cell RNA Sequencing Analysis

Single cell RNA sequencing data from the aortae of mice fed a western diet for 0 and 16 weeks was downloaded from GSM7816168 and GSM7816158, respectively. An additional condition of 8 weeks on diet was generated in-house as follows (n = 4 mice, GSM9041573): after euthanasia, the thoracic aorta and arch were harvested digested (DMEM with 1.5 mg/mL Collagenase A (Sigma # C756V80), 0.5 mg/mL elastase (Worthington LS002292), 125 U/mL DNAseI, and 10% FBS) with gentle agitation. Samples were stained with eBioscience Fixable Viability Dye eFluor 780 (Invitrogen #65-0865-14) and TotalSeqB Hashtag Oligos (BioLegend) prior to pooling samples from different mice, followed by FACS sorting to enrich tdTom positive cells to ~20%. Single cell suspensions were then prepared for sequencing with the 10X Genomics 3’ version 4 kit and analyzed as described in Supplemental Methods.

Statistics

All data are reported as mean +/− SEM unless otherwise indicated in the figure legend. Significance was determined by a non-parametric Mann-Whitney U-Test with an alpha of 0.05. When comparing more than two conditions, a non-parametric Kruskal-Wallis Test with Dunn’s post-hoc was used. When comparing at least two conditions between at least two groups, a Two-Way ANOVA with Tukey’s post-hoc was used. For RNA-sequencing, limma’s decideTests was used.

RESULTS

EC knockout of miR-33 increases early lesion initiation but does not alter advanced plaque phenotypes.

To investigate the role of EC miR-33 in atherosclerosis progression, we generated inducible EC-specific miR-33 knockout mice with a tdTomato reporter (iECKO), or control mice with an EC-specific Cre and reporter yet without floxed alleles for miR-33 (‘WT’) (Figure S1A). We first confirmed Cre-mediated recombination in iECKO and WT mice by fluorescent visualization of tdTomato expression in aortic root EC (Figure S1B). As expected, tdTomato fluorescence localized to the luminal side of the aortic root, the coronary arteries, and in the surrounding tissue - this staining reflects the expected anatomical distribution of microvasculature at the junction of the aorta and the cardiac tissue (Figure S1B). Specific EC miR-33 deletion was confirmed by qRT-PCR of FACS-sorted tdTomato+ ECs from iECKO and WT mice (Figure S1C). As expected, miR-33 iECKO mice had significantly decreased levels of the 5p strand of miR-33. However, there were no significant differences in the 3p strand, which is likely due to its extremely low levels in relation to the 5p strand and thus is at the limit of detection capability. Thus, deletion of the 5p strand would likely be the primary strand of interest for any observed phenotypes in EC. Additionally, no changes in miR-33 were observed in tdTomato-negative cells (non-ECs) from either group, highlighting the specificity of the deletion (Figure S1C). Lastly, no significant changes in Srebf2, the host gene for miR-33, were observed in tdTomato+ ECs (Figure S1C).

To induce atherosclerosis, iECKO and WT mice received a single injection of the liver-targeted, gain-of-function D377Y PCSK9-AAV8, followed by WD feeding.[40] We first assessed whether EC miR-33 deletion influenced early lesion initiation due to the well-established role of ECs in early atherosclerosis and leukocyte recruitment. To this end, mice were fed a WD for only 5 weeks. At this early stage, iECKO mice exhibited a significant increase in lesion area compared to WT controls (Figure 1A–D). However, this increase was not associated with changes in plaque-associated Oil-Red-O content or leukocyte adhesion molecule expression, as levels of intercellular adhesion molecule 1 (ICAM-1) and vascular cell adhesion molecular 1 (VCAM-1) in plaque-associated EC remained similar between groups (Figure 1C,E–F). Additionally, there were no changes in the fractional CD68-positive area, suggesting that the increase in plaque size may be due to greater overall monocyte recruitment as opposed to preferential recruitment of other immune subtypes (Figure 1G–H).

Figure 1. Endothelial knockout of miR-33 increases lesion initiation.

PCSK9-AAV8-injected male (n = 6 WT, 7 iECKO, triangles) and female (n = 4 WT, 10 iECKO, circles) mice were fed WD for 5 weeks, at which point mice were sacrificed for atherosclerosis analysis. A) H&E and Oil-Red-O images of sectioned aortic roots from the respective representative genotypes. Scale bars: left panels 500 μm, center/right panels 100 μm. B) Quantification of plaque area from (A). C) Quantification of Oil-Red-O staining from (A). D) Oil-Red-O staining of whole aortas and quantification. E-F) Immunofluorescent images and quantification of mean fluorescent intensity for ICAM-1 and VCAM-1 in sectioned aortic roots. Scale bars: 100 μm. G-H) Immunofluorescent images and quantification of CD68 percent coverage in sectioned aortic roots. Scale bars: 500 μm, 100 μm for insets below. The black symbol on the bar graphs indicates the representative image. Data are presented as mean ± SEM. Significance determined by Mann Whitney U-Test with an alpha of 0.05.

Next, to ascertain whether this phenotype persists later in disease, we kept mice on diet for longer timepoints (i.e. 12 weeks of western diet to assess plaque progression and 20 weeks of western diet to assess development of severe plaque burden). As expected, plaques developed in the aortic root, brachiocephalic artery (BCA) and throughout the aorta in both groups. However, iECKO mice showed no significant differences in plaque size, necrotic core area, or lipid accumulation by Oil-Red-O staining in the aorta and aortic root compared to WT controls after 12 weeks of WD feeding (Figure 2A–G). Likewise, WD-induced bodyweight gain, circulating plasma lipids, and circulating leukocytes remained similar between groups (Figure S2). Importantly, we still observed robust tdTomato expression in ECs (Figure 2I, second panel from the left), indicating no issues with knockout efficiency at this time point. Since previous studies suggest that plaque composition can be altered without changes in plaque size,[44] we assessed classical markers of plaque stability including collagen content and fibrous cap thickness, as well as inflammatory and smooth muscle cell infiltration. Quantification of these parameters by immunofluorescence showed no significant differences between groups (Figure 2H–I). Notably, these findings were consistent in both female and male mice (Figure 2). Together, these results indicate that EC miR-33 deletion does not significantly influence atherosclerosis development or plaque stability at the assessed time point.

Figure 2. Endothelial knockout of miR-33 does not affect atherosclerosis progression.

PCSK9-AAV8-injected female mice (n = 8 WT, 7 iECKO, circles) and male mice (n = 5 WT, n = 4 iECKO, triangles) were fed WD for 12 weeks, at which point mice were sacrificed for atherosclerosis analysis. A-B) H&E images of sectioned aortic roots (A) and brachiocephalic arteries (B) from the respective representative genotypes. Scale bars: left panel 500 μm, middle and right panels 100 μm. C) Quantification of plaque area from (A). D) Quantification of plaque area from (B). E) Quantification of necrotic core area from (A). F) Oil-Red-O staining of whole aortas. G) Oil-Red-O staining of aortic roots. Scale bars: 100 μm. H) Aortic roots were stained for Sirius Red and visualized under brightfield (left) and polarized light (middle) (scale bars: 500 μm) or for Masson’s Trichrome (right) (scale bars: 100 μm). Yellow region indicates largest necrotic core. Black bracket indicates measured fibrous cap thickness. Quantification of plaque collagen content and plaque fibrous cap thickness are on the right. I) Immunofluorescent images of sectioned aortic roots with the indicated staining. Scale bars: 100 μm. Quantification of percent αSMA+ and CD68+ plaque area are on the right. The black symbol on the bar graphs indicates the representative image. Data are presented as mean ± SEM. Significance determined by Mann Whitney U-Test with an alpha of 0.05.

Despite not observing any differences in plaque development at 12 weeks of WD, we wondered whether differences might emerge at later stages of atherosclerosis as the plaque composition changes and continues to become more necrotic. Thus, we assessed mice after 20 weeks of WD and found no differences in plaque size, lipid accumulation in the aorta and aortic root, or plaque necrotic core area between groups (Figure 3A–G). Further analysis of plaque characteristics also revealed that the deletion of miR-33 in ECs did not alter plaque fibrosis/collagen content or fibrous cap thickness when compared to WT mice (Figure 3H).

Figure 3. Endothelial knockout of miR-33 does not affect severe atherosclerotic plaque burden.

PCSK9-AAV8-injected male mice (n = 8 WT, 11 iECKO) were fed WD for 20 weeks, at which point mice were sacrificed for atherosclerosis analysis. A) H&E images of sectioned aortic roots from the respective representative genotypes. Scale bars: left panels 500 μm, right panels 100 μm. B) H&E images of sectioned brachiocephalic arteries from the respective representative genotypes. Scale bars: 100 μm. C) Quantification of plaque area from (A). D) Quantification of plaque area from (B). E). Quantification of necrotic core area from (A). F) Oil-Red-O staining of whole aortas. G) Oil-Red-O staining of aortic roots. H) Aortic roots were stained for Sirius Red and visualized under brightfield (left) and polarized light (middle) (scale bars: 500 μm) or for Masson’s Trichrome (right) (scale bars: 100 μm). Yellow region indicates largest necrotic core. Black bracket indicates measured fibrous cap thickness. Quantification of plaque collagen content and plaque fibrous cap thickness are on the right. The black symbol on the bar graphs indicates the representative image. Data are presented as mean ± SEM. Significance determined by Mann Whitney U-Test with an alpha of 0.05.

The Srebf2/miR-33 axis is downregulated in EC in atherosclerosis.

Given these findings, we next sought to understand why the effect of EC miR-33 deletion on early lesion initiation does not persist at later stages of disease. We hypothesized that this could be due to the extreme hyperlipidemia present during atherosclerosis progression, particularly since our lab has previously shown that miR-33 expression is sensitive to hypercholesterolemia.[16] We therefore postulated that downregulation of miR-33 in advanced atherosclerosis in WT mice may contribute to the lack of phenotypic differences between groups at these time points. To test this, we integrated previously published single cell data of murine aortas after 0 and 16 weeks of WD with an 8-week diet timepoint generated in our laboratory (Figure S3A). This integration yielded comparable UMAP embedding across data sets (Figure S3B) and enable the identification of diverse cell types present in the aorta throughout atherosclerosis progression (Figure S3C–D).

We then analyzed expression of Srebf2, the host gene of miR-33, in various cell types throughout disease progression, as a proxy for miR-33 levels, since this microRNA is co-transcribed with its host gene.[16,45] In fact, while Srebf2 is detectable in EC at all disease stages, its expression decreases throughout the disease (Figure 4A). This downregulation is especially pronounced when compared to the macrophage subpopulation (Figure 4A, pink boxes). In macrophages, the percent of cells expressing Srebf2 declined from 72% at baseline to 44% (39% decrease) after 16 weeks of WD, and the average relative expression dropped from 1.9 to 0.8. However, in EC, the percentage of Srebf2- expressing cells decreased by 50%, and the average relative expression was also lower overall for the same period of exposure to diet. This diet-mediated hypercholesterolemia downregulation that occurs in WT EC likely contributes to the absence of a relative phenotype upon miR-33 deletion observed in late-stage atherosclerosis. In other words, the miR-33 expression levels likely approach equivalency between WT and iECKO EC in advanced disease, thus masking any effect of EC-specific miR-33 deletion. We then validated these findings in vitro by exposing HAEC to LDL for various timepoints ranging from 0–72h to mimic acute and chronic hyperlipidemia, followed by qPCR quantification of SREBF2 and miR-33 levels. As expected, sustained exposure to hyperlipidemia further downregulates SREBF2 and miR-33a-5p, consistent with our phenotypes observed in vivo. We also analyzed the changes in expression of the 3p passenger strand of miR-33, and while there were no changes in expression during prolonged hypercholesterolemia, the levels of this isoform of miR-33 were near the detection threshold limit and much lower than that of the main 5p strand, suggesting that the 5p strand is the most relevant miR-33 isoform in this cell type (Figure 4B).

Figure 4. Bioinformatic analysis of miR-33 expression and regulation in hyperlipidemia.

A) Publicly available datasets were integrated with our generated single cell sequencing data to analyze the expression of Srebf2 in the indicated cell types at the indicated time on diet (0, 8, 16 weeks, each sample was a pool of 3, 4, or 3 animals, respectively). B) HAEC were treated with LDL to a final cholesterol concentration of 50 μg/mL, followed by qPCR analysis of SREBF2 and miR-33 levels. C) HAEC were treated with TNF with or without cholesterol loading and miR-33 mimicry, followed by RNA-sequencing to assess the regulation of gene expression by miR-33 under various conditions. Experiment was performed in technical triplicate of a pool from three distinct biological donors. KEGG pathway analysis of genes whose expression is perturbed under hypercholesteremic inflammation but rescued by miR-33 mimicry is shown. D) Heatmap visualization of genes rescued by miR-33 that fall under the KEGG pathways of cytokine-cytokine receptor interaction and fluid shear stress and atherosclerosis. Significance determined by limma’s DecideTests function with BH multiple testing correction.

The gene expression regulation by miR-33 is altered by cholesterol loading.

To address this further, we turned to in vitro culture of HAEC. We first sought to validate that miR-33 is indeed expressed and regulates its bona fide targets such as FAOrelated enzymes CPT1A and CROT as well as the cholesterol efflux transporter ABCA1. By transfecting HAEC with a miR-33 mimic, we validated that miR-33 overexpression decreases the RNA and protein levels of CPT1A (Figure S4A,E–F), CROT (Figure S4B,E–G), and ABCA1 (Figure S4C,E–H) (see Supplementary File 1 for uncropped membranes). However, while transfection with a miR-33 inhibitor increased the RNA and protein levels of CPT1A (Figure S4A,E–F) and CROT (Figure S4B,E–G), it did not result in de-repression of ABCA1 (Figure S4C,E–H), suggesting that the physiological levels of miR-33 in EC are not sufficient to potently regulate ABCA1 abundance. Due to the effect of miR-33 inhibition on CPT1A/CROT, we validated the functional implications of this targeting by confirming that miR-33 inhibition increases fatty acid β-oxidation (Figure S4I).

Next, we sought to investigate whether overall gene expression regulation by miR33 is dependent on the lipidemic status of the environment. We cultured HAEC in regular media, or media supplemented with cholesterol-rich LDL to mimic hypercholesterolemia, followed by TNFα stimulation with and without miR-33 mimicry. While EC are exposed to both native and oxidized LDL throughout atherosclerosis progression, the vast majority of LDL that ECs are luminally exposed to are not oxidized. However, a small amount of oxidized LDL present in the subendothelial space can impact EC activation and inflammation due to the presence of oxysterols, oxidized phospholipids, and aldehydes.[46,47] To understand true role of cholesterol overloading and induce a controlled and reproducible inflammation, we utilized native LDL combined with TNFα to mimic the inflammatory and lipid-laden environment of atherosclerosis. After validating efficacy of the various treatments (Figure S5A), we sought to determine which genes have altered TNF-mediated expression patterns when the media is also supplemented with LDL. To determine whether these alterations are due to cholesterol (LDL)-mediated repression of the SREBP2/miR-33 axis, we transfected HAEC with a miR-33 mimic in the same conditions to rescue the cholesterol-mediated downregulation of miR-33. We found that the expression of ~80 genes could be rescued to levels seen in normolipidemic inflammation. Interestingly, KEGG pathway analysis of these genes revealed that several are linked to inflammation-associated processes and atherosclerosis (Figure 4C), with example heatmaps of genes involved in the “Cytokine-cytokine receptor interaction” and “Fluid shear stress and atherosclerosis” pathways highlighted in Figure 4D.

EC miR-33 regulates acute inflammation.

To address whether miR-33 has any role in endothelial inflammatory responses, we performed experiments in normolipidemic conditions to avoid any confounding effects of hypercholesterolemia on miR-33 expression. We utilized the murine air pouch model, which is a robust model that facilitates easily quantifiable analysis of both immune cell recruitment and endothelial inflammation after inflammatory insult.[43,48] Briefly, after induction and reinflation of a sterile air pouch in WT and iECKO mice, 10 ng of TNFα was injected into the pouch to induce leukocyte recruitment (Figure 5A). Twelve hours postinjection, lavage of the pouch and quantification of recruited cells revealed a significant increase in total leukocyte recruitment in iECKO mice compared to WT controls (Figure 5B). This enhanced recruitment was localized to the site of inflammation and did not reflect systemic changes, as complete blood counts revealed no differences in circulating leukocyte numbers between groups either before or after TNFα injection (Figure S6A). Flow cytometry analysis of the immune cell populations in the exudate (Figure 5C, Figure S6B) showed no significant differences in the proportion of neutrophils, Ly6C^hi monocytes, or Ly6C^lo monocytes between groups (Figure 5D). However, consistent with the overall increase in cell recruitment, the absolute number of each of these populations was significantly elevated in iECKO mice (Figure 5E). To determine whether this effect reflected an intrinsic pre-conditioning of the endothelium in the miR-33 KO mice, we analyzed cell recruitment to the pouch only after 4 hours of TNFα injection and found no differences in total or subset-specific cell recruitment in WT versus iECKO mice (Figure S6C). These findings suggest that the enhanced recruitment seen at 12 hours results from active miR-33–dependent regulation of gene expression in response to inflammatory stimulation in WT mice, rather than baseline alterations in endothelial function caused by the knockout of miR-33.

Figure 5. EC miR-33 deletion promotes acute inflammation.

WT and miR-33 iECKO mice (n = 14 WT, 21 iECKO mice) were injected with sterile air to form an air pouch, followed by TNFα injection and analysis of leukocyte recruitment. A) Scheme of experimental design. B) Batch-corrected quantification of total number of cells recruited to each pouch. C) Representative flow cytometric analysis of CD45+ CD11b+ cells, stained for Ly6G and Ly6C. Black box indicates Ly6G+ cells. Pink box indicates Ly6C^hi cells, and green Ly6C^lo. D) Proportion of Ly6G+, Ly6C^hi, and Ly6C^lo cells per animal, quantified from (C). E) Absolute quantification of indicated cell populations, obtained by back-calculating based on percentages from (D) and total cell content from (B). F) The air pouches from WT and miR-33 iECKO mice (n = 8 WT, 14 iECKO mice) were processed for immunofluorescent staining to visualize E-selectin levels, quantified to the right. Scale bars: 25 μm. G) Visualization of predicted miR-33 binding sites in the 3’ UTR of human and murine E-Selectin transcripts from TargetScan. H) Various EC types were transfected with miR-33 mimic or respective control and treated with TNF for 8h, followed by western blot analysis of E-selectin expression, two representative samples per condition from the quantification on the right are shown. Data are presented as mean ± SEM. Significance determined by Mann Whitney U-Test with an alpha of 0.05 (B, D-F) or Two-Way ANOVA with Tukey’s post-hoc (H).

In order to examine potential mechanisms responsible for the endothelial miR-33-dependent alterations in leukocyte recruitment, we analyzed E-selectin levels in the murine air pouch from mice 12h after TNF injection, since E-selectin is a prominent adhesion molecule induced by the endothelium during acute inflammation.[1] Consistent with the increase in leukocyte recruitment after TNF stimulation, vasculature from the air pouch of miR-33 iECKO mice had increased levels of E-selectin as compared to control mice (Figure 5F). In order to ascertain whether this could be due to targeting of E-selectin by miR-33, and subsequent de-repression upon miR-33 deletion in the endothelium, we searched for potential miR-33 binding sites within the E-selectin 3’UTR using TargetScan, and found a match in both the human and murine E-selectin 3’ UTR (Figure 5G). We next validated regulation of E-selectin expression by miR-33 following TNF stimulation in HUVEC, a widely generally accepted model for studying endothelial inflammation particularly in relation to E-selectin levels, and observed that miR-33 mimicry does indeed impair TNF-mediated induction of E-selectin (Figure 5H). To ascertain whether this regulation may be relevant across different endothelial subtypes, we also validated this regulation in HAEC and human dermal microvascular EC (HDMEC) (Figure 5H). Lastly, to determine whether this regulation was indeed due to direct inhibition by miR-33 versus indirect inhibition, we performed a luciferase reporter assay and indeed confirmed that miR-33 does regulate E-selectin levels to the same levels as previously reported for known regulators of E-selectin such as miR-31 (Figure S6D). Thus, the increase in leukocyte recruitment seen in miR-33 iECKO mice is likely due to the direct de-repression of E-selectin expression in these animals.

DISCUSSION

In this study, we assessed the contribution of EC miR-33 in atherosclerosis and acute inflammation. miR-33 has extensively been implicated in lipid metabolism and atherosclerosis, but most studies to date have focused on whole-body or tissue-specific deletion in either hematopoietic cells or hepatocytes.[24,25,28] However, the role of miR33 in other cell types remains poorly defined. While a handful of studies have explored the role of miR-33 in EC, these have been limited to in vitro analyses of cellular behaviors, such as apoptosis and senescence, along with analysis of the levels of some miR-33 target genes.[49,50]

ECs are known to play key roles throughout atherosclerosis progression, contributing to both early inflammatory responses[1] – by mediating leukocyte recruitment and regulating vascular permeability – as well as to plaque instability in advanced disease by undergoing EndoMT.[51,52] These functions have been independently reported to be controlled by pathways of which miR-33 is an upstream regulator, namely cholesterol efflux and FAO,[33,36,37] suggesting that miR-33 could influence pro-atherosclerotic EC behaviors in a stage-dependent manner. However, our data show that EC miR-33 deletion accelerates only lesion initiation while intermediate and late stages were unaffected. This indicates that the increase in early lesion formation is insufficient to drive long-term differences in atherosclerosis progression, highlighting an interesting emergence of stage-dependent actions of this well-described miRNA in this cell type. One possible explanation for this lies in the dynamic regulation of miR-33 that is related to its hypercholesterolemia sensitivity.[16] As disease progresses, its expression declines, potentially masking the impact of its deletion. Indeed, analysis of publicly available single-cell RNA-seq data[53] combined with our own across different time points of WD feeding revealed a progressive reduction of Srebf2 expression, the host gene of miR-33 which can be used as a proxy for its expression. Notably, this decline is particularly pronounced in ECs at advanced disease states. This potentially explains the lack of phenotype at later stages as the downregulation of miR-33 masks any effect of differences that may arise in the genetic knockout.

Our phenotypic and single cell analyses are in accordance with the existing body of literature reporting on the role of miR-33 in atherosclerosis. Briefly, studies of either global miR-33^−/− mice or mice treated systemically with miR-33 inhibitors show varying effects depending on the type of diet and genetic background. Specifically, previous studies have shown that global miR-33 deletion only has an effect in lower-added-cholesterol diets (0.3% compared to 1.25% added cholesterol).[27,30] In addition to global deletion, reconstitution of Ldlr^−/− mice with miR-33^−/−;Ldlr^−/− bone marrow decreases atherosclerosis, suggesting that miR-33 in the hematopoietic compartment is pro-atherogenic.[25] Based on experiments with miR-33^−/− bone marrow-derived macrophages and other in vitro approaches, these findings have been largely ascribed to macrophages, although definitive validation with macrophage- or myeloid-specific knockout models has yet to be performed.[22,25,54,55] Notably, despite WD feeding, other cell types have higher expression levels of the miR-33 host gene, Srebf2, as compared to EC. This is likely why extensive literature has been able to report a role for miR-33 in advanced atherosclerosis. This is particularly interesting, considering the fact that EC miR-33 deletion actually increases atherosclerosis, suggesting an anti-atherogenic role in this cell type, thus highlighting further the context-dependent nature of gene- and phenotype-regulation by this miRNA. Future studies should include further dissection of the mechanisms behind this phenotype, especially regarding the extent it is due to direct versus indirect transcript level regulation, or indirect effects due to metabolic changes induced by miR-33 in EC.

Hypercholesterolemia appears to alter the regulatory influence of miR-33 in ECs, and our in vitro studies using cultured HAEC in inflammatory and cholesterol-laden conditions support this notion. Under these conditions, cholesterol-rich LDL loading significantly altered the gene expression response to TNFα. Notably, restoring miR-33 levels via mimic-based rescue partially normalized this response, reversing the perturbed expression of ~80 out of ~250 TNFα-responsive genes. These rescued genes were enriched for inflammatory pathways, indicating that lipid-induced downregulation of miR-33 disrupts its regulatory network. This context-dependent regulation likely contributes to the lack of sustained phenotypic differences at later atherosclerosis stages, where miR-33 cression of single genes, but more broadly orchestrate expression of related genes to fine-tune cell behaviors at the pathway level,[56] as exemplified by literature on other miRNAs such as miR-21[57] as well as previous literature on miR-33.[23]

These transcriptomic differences led us to consider whether EC miR-33 might play a more prominent role in inflammatory responses in normolipidemic conditions. This is particularly relevant given our observation of accelerated lesion initiation during early atherosclerosis, a stage characterized by comparatively lower systemic cholesterol and early activation of the endothelium.[3] Indeed, when we used a classical model of acute sterile inflammation to assess TNFα-mediated leukocyte recruitment into a murine air pouch,[43] we found increased leukocyte recruitment in the absence of miR-33, emphasizing an anti-inflammatory role for this microRNA in the endothelium. Interestingly, this is the opposite effect as reported in macrophages, where miR-33 is described to have a more pro-inflammatory role. While the overall increase in immune cell infiltration was not skewed toward a specific cell type, the findings support a model in which miR-33 broadly modulates endothelial responsiveness to inflammatory cues.

Regarding potential mechanisms, while it has been reported that miRNAs can regulate EC leukocyte adhesion molecules (ELAMs)[6] and TNFα responses with implications for atherosclerosis and chronic inflammation,[58] EC miR-33 does not regulate ELAMs such as ICAM-1 and VCAM-1 in vivo during lesion initiation, where we did see atherosclerosis phenotypes consistent with an increase in EC inflammation (Figure 3E). However, we did indeed identify miR-33 binding sites within the E-selectin 3’ UTR, which occurs with a greater induction of E-selectin in the air pouch of miR-33 iECKO mice and is also consistent with miR-33-mediated repression of E-selectin in HUVEC during shorter timepoints of TNF stimulation. It is noteworthy that while several adhesion molecules are upregulated during acute inflammation, our analysis focused on E-selectin as a direct miR-33 target. Unlike Icam1 and Vcam1, which lack miR-33 binding sites in the mouse 3’ UTR, Sele contains a highly conserved target site that we confirmed to be responsive to miR-33 downregulation of E-selectin.

One limitation of this study is that we have used a pan-endothelial inducible cre to delete miR-33 in EC.[38] While this cre driver is widely used in the field to perform inducible endothelial gene deletions, it is possible that the phenotypes observed in early atherosclerosis are due to indirect effects of miR-33 gene deletion in EC in other tissues that control atherosclerosis progression and systemic lipid homeostasis such as the liver, despite that in the latter we did not observe differences in systemic lipid levels after deletion of mir-33. Regardless, our study is the first to definitively address the role of miR-33 in EC during atherosclerosis using a novel inducible EC-specific KO mouse model with detailed analysis across multiple disease stages. We show that EC miR-33 diminishes lesion initiation, although this effect is blunted at later stages, likely due to hypercholesterolemia-driven downregulation of miR-33 itself, a phenomenon to which EC appear particularly sensitive. Our present results provide additional insight into the potential therapeutic potential of miR-33 manipulation. This is especially important regarding conflicting data on the effect of miR-33 deletion either globally or in the hematopoietic compartment,[25] since we observed an increase in lesion initiation upon miR-33 EC deletion, while others have reported either no differences or a decrease in atherosclerosis with systemic miR-33 deletion.[25,27,29,30] Additionally, we also demonstrate that miR-33 shapes local inflammatory responses, as deletion of miR-33 in ECs enhances TNFα-induced leukocyte recruitment in an air pouch model. We also show that E-selectin is a novel target of miR-33 in the endothelium during inflammation. Given the central role of ECs in sensing inflammatory cues and coordinating leukocyte trafficking,[1] this finding highlights a broader regulatory function for miR-33 in vascular inflammation.

Altogether, these data raise the possibility that miR-33 influences early atherogenesis and acute local inflammation through overlapping or context-dependent mechanisms. Together, our work highlights the importance of dissecting cell-specific roles when considering miR-33-targeted therapies and underscores the nuanced and environment-specific functions of miR-33 in endothelial biology.

HIGHLIGHTS.

• Knockout of miR-33 in endothelial cells increases atherosclerotic lesion initiation without affecting advanced plaque progression

• Knockout of miR-33 in endothelial cells does not affect advanced plaque progression due to chronic hypercholesterolemia-driven downregulation of miR-33-SREBF2 in control mice, masking its deletion at later stages

• Knockout of miR-33 in endothelial cells exacerbate acute inflammation in normocholesterolemic conditions via absence of E-Selectin targeting.