Disturbed flow induces reprogramming of endothelial cells to immune-like and foam cells under hypercholesterolaemia during atherogenesis

Abstract

Aims

Atherosclerosis occurs preferentially in the arteries exposed to disturbed flow (d-flow), while the stable flow (s-flow) regions are protected even under hypercholesterolaemic conditions. We recently showed that d-flow alone initiates flow-induced reprogramming of endothelial cells (FIRE), including the novel concept of partial endothelial-to-immune-cell-like transition (partial EndIT), but it was not validated using a genetic lineage-tracing model. In addition, the combined effect of d-flow and hypercholesterolaemia has not been tested. Here, we tested and validated the two-hit hypothesis that d-flow is an initial instigator of partial FIRE but requires hypercholesterolaemia to induce a full-blown FIRE and atherosclerotic plaque development.

Methods and results

Mice were treated with AAV-PCSK9 and a Western diet to induce hypercholesterolaemia and/or partial carotid ligation (PCL) surgery to expose the left common carotid artery (LCA) to d-flow. Single-cell RNA sequencing (scRNA-seq) analysis was performed using single cells obtained from the LCAs and the control right common carotid arteries at 2 and 4 weeks post-PCL. Immunohistochemical staining was performed on EC-specific confetti mice at 4 weeks post-PCL and hypercholesterolaemia to validate endothelial reprogramming. Human aortic endothelial cells (HAECs) exposed to d-flow and hypercholesterolaemic conditions were used to validate FIRE. Atherosclerotic plaques developed by d-flow under hypercholesterolaemia, but not by d-flow or hypercholesterolaemia alone. The scRNA-seq results of 98 553 single cells from 95 mice revealed 25 cell clusters: 5 EC, 3 vascular smooth muscle cell (SMC), 5 macrophage (MΦ), and additional fibroblast, T cell, natural killer cell, dendritic cell, neutrophil, and B-cell clusters. Our scRNA-seq analysis results raised a hypothesis that d-flow under hypercholesterolaemia transitioned healthy ECs to full immune-like (EndIT) and, more surprisingly, foam-like cells (EndFT), in addition to inflammatory and mesenchymal cells (EndMT). Further, ECs with characteristics of foam cells shared remarkably similar transcriptomic profiles with foam cells derived from SMCs and MΦs. Lineage-tracing studies using immunohistochemical staining of canonical protein and lipid markers in the EC-specific confetti mice exposed to d-flow and hypercholesterolaemia demonstrated evidence supporting the novel FIRE hypothesis, including EndIT and EndFT. Moreover, reanalysis of the two publicly available human plaque scRNA-seq datasets and our immunostaining studies suggest that FIRE occurs in human atherosclerotic plaques. Additionally, HAECs exposed to d-flow, high cholesterol, and proinflammatory cytokines (identified in our scRNA-seq data) show the markers of EndIT and EndFT at the mRNA, protein, and functional levels.

Conclusion

The scRNA-seq study raised a two-hit hypothesis for FIRE, including EndIT and EndFT, which was validated by the lineage-tracing and in vitro HAEC studies. D-flow induces partial reprogramming, including inflammation, EndMT, and partial EndIT. Under hypercholesterolaemia, d-flow fully reprogrammes arterial ECs, including the novel EndIT and EndFT, in addition to inflammation and EndMT, during atherogenesis. This single-cell atlas and FIRE programs provide a crucial roadmap for novel mechanistic understanding and therapeutics targeting flow-sensitive genes, proteins, and pathways of atherosclerosis.

Graphical Abstract

1. Introduction

Atherosclerosis, a chronic inflammatory disease resulting in the accumulation of plaque in the arterial wall and lumen narrowing, is the major underlying cause of myocardial infarction, ischaemic stroke, and peripheral arterial disease.^1–3 Despite the widespread and successful use of therapeutics, including statins that effectively lower one of the most critical proatherogenic risk factors, blood cholesterol level, atherosclerotic disease continues to be a leading cause of death worldwide.⁴ It highlights the dire need to develop additional therapeutics to address the residual non-lipid targets. For example, CANTOS trial, using the IL-1β inhibitor canakinumab, demonstrated potential promise of targeting inflammatory factors as anti-atherogenic therapy, although it was not approved by the FDA due to a concern with fatal infections.⁵ Since atherosclerosis preferentially occurs in flow-disturbed regions suh as branching points, we and others have been studying flow-sensitive factors as potential therapeutic targets.^6–12 Here, we report a single-cell RNA-sequencing (scRNA-seq) study to identify flow-sensitive genes and pathways that may be responsible for atherosclerosis development in response to disturbed flow (d-flow) and hypercholesterolaemic condition in the mouse carotid arteries.

Even though hypercholesterolaemia, diabetes, and hypertension are major systemic risk factors, atherosclerotic plaques preferentially occur focally involving curved, branched, and bifurcated regions of arteries exposed to d-flow.^4,7,8,10 In contrast, straight arterial regions exposed to stable flow (s-flow) tend to be protected from atherosclerosis. The mechanisms underlying the different effects of d-flow and s-flow on atherosclerosis, which involve endothelial gene expression and function, are still unclear.^7,9–11 To address the mechanisms, we developed the mouse partial carotid ligation (PCL) model, by ligating 3 of 4 downstream branches of the left common carotid artery (LCA) to induce d-flow while using the contralateral right common carotid artery (RCA) exposed to s-flow as a control within the same mouse.¹³ With the PCL model, we directly demonstrated that d-flow rapidly induces robust atherosclerotic plaque development within 2–3 weeks under hypercholesterolaemic condition. The study also showed that d-flow alone in the absence of hypercholesterolaemia or hypercholesterolaemia under s-flow failed to develop atherosclerosis, demonstrating that atherosclerosis development is promoted by the interaction between systemic hypercholesterolaemia and focal d-flow. This is consistent with the preferential atherosclerosis development in the branching points exposed to d-flow in hypercholesterolaemic patients.^6,12

D-flow and s-flow differently regulate endothelial function through mechanosensors^14–18 and gene expression, leading to initiation and prevention of atherosclerosis.^9,19–26 To explore the mechanisms, we previously carried out a scRNA-seq and single-cell assay for transposase-accessible chromatin sequencing (scATAC-seq) study in response to d-flow alone under normal cholesterol conditions to avoid the complex mechanisms under the combination of d-flow and hypercholesterolaemia.²⁷ In the previous study, we also collected single cells from the intima of the LCAs and RCAs to obtain a sufficient number of ECs for an in-depth examination of the effects of d-flow alone on endothelial cells (ECs). In the study, C57BL/6 mice underwent PCL surgery and were fed a chow diet, followed by single-cell isolation from the LCAs and RCAs 2 days and 2 weeks post-PCL in a time- and flow-dependent manner.²⁷ The study showed that ECs are highly plastic and heterogeneous. More importantly, healthy ECs transition to proatherogenic phenotypes, including inflammation, endothelial-to-mesenchymal transition (EndMT), and the novel finding of endothelial-to-immune cell-like transition (EndIT), which we collectively named as flow-induced reprogramming of ECs (FIRE).^9,27 Interestingly, our pseudotime trajectory analysis showed that many ECs underwent full EndMT. However, only a small number of ECs underwent EndIT and did not fully transition into immune cells in response to d-flow alone, which we refer to as partial EndIT. These prior results showed that d-flow alone initiates partial FIRE, involving inflammation, EndMT, and partial EndIT. However, the partial EndIT suggested by scRNA-seq required validation by methods such as genetic lineage-tracing, which remained to be done. Since d-flow alone could not induce atherosclerotic plaques in the absence of systemic risk factors such as hypercholesterolaemia,^13,28 we hypothesized that a combination of d-flow and hypercholesterolaemia would induce atherosclerotic plaques by inducing robust FIRE, especially EndIT in addition to inflammation and EndMT.

Here, we tested the hypothesis by performing an extensive scRNA-seq study to compare the effects of d-flow alone, hypercholesterolaemia alone, and d-flow under hypercholesterolaemia on ECs as well as all other arterial cell types in the atherosclerotic plaques. Our scRNA-seq results raised a two-hit hypothesis that d-flow under hypercholesterolaemia induces a robust EndIT and a novel endothelial-to-foam cell-like transition (EndFT) in addition to inflammation and EndMT during atherogenesis. We validated the two-hit hypothesis by lineage-tracing study and in vitro studies using human aortic ECs (HAECs) under the d-flow and hypercholesterolaemic conditions. Moreover, the reanalysis of two publicly available human plaque scRNA-seq datasets and immunostaining validation studies using human coronary atherosclerotic plaques provide evidence that FIRE also occurs in human plaques.

2. Methods

2.1. Transparency and openness

The authors will make the data, methods used in the analysis, and materials used to conduct the research available to any researcher for purposes of reproducing the results or replicating the procedure.

2.2. Mouse scRNA-seq study

All mouse studies reported here were conducted as approved by the Emory University institutional animal care and use committee (PROTO202100052) according to the NIH guide for the care and use of laboratory animals. A total of 75 male C57BL/6 mice (Jackson Lab, 9 to 10-week-old) were used in this scRNA-seq study performed in three independent experiments (see Supplementary material online, Table S1). As described previously, hypercholesterolaemia was induced through a combination of AAV8-PCSK9 administration (1 × 10¹¹ viral genome) (Vector Biolabs, Malvern, PA, #AAV8-D377Y-mPCSK9) via tail-vein injection for low-density lipoprotein receptor (LDLR) knockout and feeding with a Western diet a week before the PCL surgery.²⁹ PCL surgery was performed as previously described through ligation of 3 of the 4 caudal branches (left external carotid, left internal carotid, and left occipital arteries) of the LCAs in mice that were anaesthetized, with the RCAs left intact as the contralateral control.¹³ Mice anaesthesia was induced by inhaling 3% isofluorane and maintained by 1.5% isofluorane (Dechra, Northwich, United Kingdom, Isospire, J123025). Buprenorphine (0.05–0.1 mg/kg, SQ) was used post-operatively as an analgesic. Ultrasound imaging was performed to confirm the exposure of the LCA to d-flow, as described previously.¹³ Two- or 4-weeks post-PCL surgery, mice were sacrificed for scRNA-seq experiments by CO₂ asphyxiation.

2.3. Isolation of arterial cells from the carotid arteries and library preparation for scRNA-seq

Single-cell preparation and sequencing were conducted as we previously reported.²⁷ Briefly, cell dissociation buffer was composed of 500 U/mL of collagenase type I (Millipore, Sigma, Burlington, MA, #SCR103), 500 U/mL of collagenase type II (MP Biomedical, Irvine, CA, #0.2100502.5), 150 U/mL of collagenase type XI (Sigma–Aldrich, St. Louis, MO, #C7657), 60 U/mL of hyaluronase type I-S (Sigma–Aldrich, St. Louis, MO, #H3506), and 60 U/mL of DNASE I (Zymo, Research, Irvine, CA, #E1011) in HBSS (Cytiva, Marlborough, MA, #SH30031) and filtered through a 0.45 μm syringe filter (Celltreat, Ayer, MA, #229753).

Upon euthanization, blood was collected from the inferior vena cava with heparin-coated 25G needle (Air-Tite, Virginia Beach, VA, #N2558) to measure the plasma cholesterol level. A lobe of liver was also collected to validate LDLR knockout by western blot. Then, LCAs and RCAs were cleaned and perfused with saline solution before being injected with the dissociation buffer. The ends of the carotid arteries were ligated using sutures and the arteries were dissected out. After incubation in the 35 mm dishes containing PBS at 37°C for 45 min, sutures were removed from the ends. The carotid arteries were luminally flushed using the dissociation buffer into 1.5 mL Eppendorf tubes containing foetal bovine serum (FBS) for neutralization of enzymatic activity and then kept on ice.

Leftover LCAs and RCAs were placed into new tubes containing dissociation buffer and minced using microscissors. The minced tissues were additionally incubated at 37°C for 45 min, after which the entire cell solution was moved to 1.5 mL Eppendorf tubes containing FBS on ice. Both luminal and leftover digestion samples were filtered through a 70 μm cell strainer to remove large pieces of extracellular matrix, followed by a series of washes with 2% bovine serum albumin (BSA) in PBS solution and centrifugation at 1000g for 5 min. RBC lysis was performed via incubation of the cell solutions in the Hybri-Max buffer (Sigma–Aldrich, St. Louis, MO, #R7757) at RT for 5 min, followed by incubation in accutase (Sigma–Aldrich, St. Louis, MO, #A6964) at 37°C for 5 min to prepare single cell solutions. The single cells were finally resuspended in 2% BSA in PBS solution and immediately encapsulated at the Emory Integrated Genomics Core (EIGC) using the 10× Genomics Chromium Next GEM Single Cell 3′ Kit v3.1 and the Chromium X device. The cDNA libraries were constructed, followed by sequencing on the Illumina NovaSeq instrument to a minimum depth of 25 000 reads per cell.

2.4. scRNA-seq data preprocessing and batch effect correction

scRNA-seq data files were preprocessed and aligned to the mouse reference genome (mm10) with 10× Genomics CellRanger software. CellRanger barcode-ranked plots are provided in Supplementary material online, Figure S1. Downstream analyses of the scRNA-seq data were performed using the Seurat v5 R package.³⁰ Quality control measures were implemented to filter out the cells expressing <200 or >7600 genes and those with >10% counts aligned to mitochondrial genes to remove non-singlets or damaged cells, respectively. Normalization, scaling, clustering, and visualization using UMAP were then performed for each of the scRNA-seq datasets from our previous study²⁷ and the newly conducted studies here, composed of 3 independent libraries (see Supplementary material online, Table S1).

Batch effect correction arising from combining multiple scRNA-seq datasets was performed using the Seurat integration function, where the datasets were merged and normalized, followed by integration using reciprocal principal component analysis (RPCA) in Seurat. The four independent libraries were considered as batches (see Supplementary material online, Table S1). To quantify the changes upon batch effect correction, two types of benchmarking metrics were used. Batch mixing metrics quantify the degree of how well-mixed the datasets are in the projected space, while cell type conservation metrics quantify how well-conserved the biological variations of cells are upon mixing.^31,32 Batch mixing metrics used are iLISI: integration local inverse Simpson’s index (LISI), ASW_batch: batch average silhouette width (ASW), and kBET: k-nearest neighbour batch effect test. Cell type conservation metrics used are ARI: adjusted rand index, cLISI: cell type LISI, and ASW_celltype: cell type ASW. ARI value ranges from 0 to 1, with a higher score indicating better cell type conservation. Value of LISI ranges from 1 to the number of batches, which is 4 in this study. An iLISI score closer to 4 indicates enhanced batch mixing, while a cLISI score closer to 1 indicates higher cell type conservation. ASW value ranges from −1 to 1, where a lower ASW_batch score suggests better batch mixing, while a higher ASW_celltype score suggests enhanced cell type conservation. kBET rejection rates range from 0 to 1, with a lower value indicative of improved batch mixing. Batch mixing metrics quantified using all cells showed better batch mixing as indicated by higher iLISI and lower kBET using a sample size of 10% and 20% after integration, while those using only the ECs also showed better batch mixing as shown by higher iLISI and lower ASW_batch and kBET with a sample of size of 10% after integration (see Supplementary material online, Table S2). Similarly, cell type conservation assessed using all cells showed higher cell type purity as indicated by higher ARI after integration. Due to the presence of a single-cell type, cell type conservation metrics were unable to be measured using only the ECs (indicated as N/A). Furthermore, ASW metrics using all cells were unable to be quantified with the publicly available packages. The Seurat RPCA method was selected for batch effect correction of our datasets since it provided better improvements in the overall benchmarking metrics above compared to Seurat canonical correlation analysis (CCA), Harmony, and LIGER integration methods (data not shown).

ARI was measured using the fossil R package, while iLISI and cLISI were quantified using the lisi package. ASW_batch and ASW_celltype were quantified using the cluster package, while kBET was measured on 10–20% sample size using the kBET package.

2.5. scRNA-seq cell clustering analysis

Following integration, scaling, clustering, and visualization were performed on the integrated scRNA-seq object in Seurat. Manual annotation of cell clusters was performed on the list of enriched genes generated from differential gene expression (DGE) analysis of each cluster against the others and sorted by avglog2FC. Canonical cell markers were used for the overall cell type annotation. More detailed clustering analysis of cell types with multiple clusters was performed by subsetting these cells, followed by DGE analysis of every cluster for each cell type independently.

Gene ontology (GO) biological process (BP) terms were generated by uploading the top 100 differentially expressed genes to the Gene Ontology Consortium website (https://geneontology.org/) and sorted by P-value. A heatmap was generated using Seurat to display the top 20 most highly enriched genes for each cluster. Cell–cell communication analyses were performed using the CellChat R package following the developer’s instructions.³³

2.6. scRNA-seq trajectory analysis

Diffusion map analysis, an unsupervised and unbiased form of trajectory analysis, was performed on the subset of ECs, SMCs, and MΦs and projected into both 2D and 3D using the destiny R package following the developer’s instructions.³⁴ Monocle pseudotime trajectory analysis was also carried out on these three cell types using the Monocle 3 R package following the developer’s instructions.³⁵ EC1, SMC1, MΦ1, and MΦ5 were set as roots manually. RNA velocity analysis was performed by generating loom files for spliced and unspliced matrices using the Velocyto Python package³⁶ and merging them to the adata file created from an annotated Seurat scRNA-seq object using the scVelo Python package.³⁷ Stochastic mode was implemented for RNA velocity. Maps showing velocity length and velocity confidence were generated using scVelo.

2.7. Reference map reanalysis of scRNA-seq data from human atherosclerotic plaques

Our mouse atherosclerosis scRNA-seq data genes were first converted to human gene orthologs using the biomaRt R package. We reanalysed the human carotid endarterectomy scRNA-seq dataset composed of 4811 single cells collected from 44 samples^38,39 and scRNA-seq atlas of human atherosclerotic plaques with 257 459 single cells obtained from 12 independent datasets and 80 human samples from the carotid, coronary, and femoral arteries.⁴⁰ Reference map reanalysis was performed using SingleR package,⁴¹ which effectively predicted the cell annotations in both human scRNA-seq datasets using the transcriptomic profiles of cells and their annotations from our mouse scRNA-seq data. The cell number for each predicted cell cluster was quantified based on the predicted cell annotations from reference map analysis.

2.8. Reanalysis of CRISPRi-Perturb-seq data

The publicly available CRISPRi-Perturb-seq data, knocking down 2285 genes associated with 306 coronary artery disease genome-wide association study (CAD-GWAS) signals at a single-cell resolution in human ECs, was reanalysed.⁴² Perturbation of 2285 genes significantly altered the expressions of 17 849 genes (P value < 0.001). The lists of perturbed genes and genes altered in response to Perturb-seq were compared against 1291 flow-sensitive genes that were significantly and differentially expressed (average log₂fold change > 1) in the five EC clusters. Again, the mouse genes were converted to human gene orthologs using the biomarRt package. These reanalyses enabled the identification of flow-sensitive genes that were perturbed and flow-sensitive genes that were altered in response to perturbations. Furthermore, an alluvial plot linking flow-sensitive genes that altered expressions of FIRE markers used to annotate EC clusters was generated using the ggalluvial package.

2.9. Lineage tracing study using EC-confetti mice

The lineage tracing study was approved (PROTO202100052) by the Emory University Institutional Animal Care and Use Committee and conducted in accordance with federal guidelines and regulations. We established an EC-specific confetti mouse model (EC-Confetti) as previously described⁴³ by crossing loxP-flanked (floxed) cytosolic RFP, cytosolic YFP, nuclear GFP, and membrane CFP (Confetti^FL/WT) mice (Jackson Laboratory, Bar Harbor, ME) with tamoxifen-inducible, endothelial-specific Cdh5-CreER^T2 mice provided by Dr. Ralf Adams. EC-Confetti mice were genotyped and screened by PCR and DNA sequencing, and used for a lineage tracing study⁴³ to validate FIRE. A total of 21 (6 Male/15 Female) 4–6-week-old EC-Confetti mice were injected intraperitoneally with tamoxifen dissolved in corn oil (75 mg/kg body weight) over the course of 1 week (series of five injections).⁴³ Following a 2-week resting period, AAV-PCSK9 injection, Western diet feeding, and PCL surgery were carried out and sacrificed at 4 weeks post-PCL, as described for the scRNA-seq study. The carotid arteries and aortic arches were isolated immediately after mouse euthanization. LCAs and RCAs were post-fixed using 4% PFA (Santa Cruz Biotechnology, Dallas, TX, #sc-281692). As we reported previously, a section of the RCA and two segments of the LCA, as well as the aortic arch, were longitudinally cryosectioned for imaging and subsequent analyses.¹⁹ Immunohistochemical staining against multiple canonical markers of FIRE was performed as previously described.¹⁹ FIRE protein markers were endothelial inflammation (Vcam1 and Icam1), EndMT (Snai1, Acta2, and Cnn1), immune cells (Cd68, C1qa, C1qb, and Lyz2), and foam cells (Spp1, Lgals3, and Trem2).

The specificity of each primary antibody used in this study (see Supplementary material online, Table S3) was validated by conducting dilution curve studies (data not shown) and isotype and secondary antibody alone controls (see Supplementary material online, Figure S17). After blocking at room temperature for 1 h, sections were incubated at 4°C overnight with primary antibodies diluted in PBS supplemented with 0.1% Triton X100 and 5% BSA. AlexaFluor 568- (Thermo Fisher Scientific, Waltham, MA, #A20184) or 647-conjugated secondary antibody (Thermo Fisher Scientific, Waltham, MA, #A20186) was used at room temperature for 2 h. For BODIPY staining, the cryosectioned LCAs and RCAs were stained with 2 μM BODIPY493/503 (Thermo Fisher Scientific, Waltham, MA, #D3922) for 1 h at 37°C as previously described.¹⁹ The stained sections were then Dapi-mounted for fluorescence imaging.

Confetti (nuclear GFP, cytosolic YFP, and cytosolic RFP) and FIRE marker expressions were imaged by using an inverted fluorescence microscope (Keyence #BZ-X800, Japan) as previously described.²² Membrane CFP was not detectable as previously reported.⁴³ Following blind deconvolution and dehazing of out-of-focus illumination, image superimposition and automated image stitching were performed by using custom-coded MATLAB algorithms (MathWorks, MA), ImageJ, and Fiji (NIH, MD). To quantify the level of confetti induction, luminal Confetti⁺ ECs within the LCAs and RCAs (N = 295 slides, 21 mice) were manually defined and segmented as described.⁴⁴ To determine the proportion of confetti⁺ ECs undergoing FIRE, multi-level image thresholding was performed via combined use of custom-developed MATLAB algorithms and ImageJ.⁴⁴ Confetti⁺/FIRE⁺ region was defined via binarized masks, while non-overlap or unmasked region was derived computationally to evaluate Confetti⁺/FIRE⁻ region. Intensity histogram of Confetti⁺/FIRE⁺ region in the LCAs and RCAs was normalized for quantification. Clonal expansion of ECs was assessed by colour distribution of each fluorescent reporter protein as previously described.⁴³ Co-immunofluorescence staining was performed using appropriate combinations of primary antibodies with compatible host species (see Supplementary material online, Table S3).

2.10. Human coronary artery immunostaining

Left anterior descending (LAD) coronary arteries were obtained from LifeLink of Georgia and dissected from deidentified human hearts preserved in Wisconsin solution. Following dissection, the entire LAD was fixed in 4% paraformaldehyde (PFA; Santa Cruz Biotechnology, Dallas, TX, #sc-281692) overnight, paraffin-embedded, and sectioned at a thickness of 8 μm. Tissue sections were deparaffinized and subsequently subjected to antigen retrieval, followed by permeabilization and blocking in PBST supplemented with 0.1% Triton X-100 (Sigma-Aldrich #X100, St. Louis, MO) and 5% bovine serum albumin (Sigma–Aldrich, St. Louis, MO, #A7906). Tissue sections were concurrently incubated with Anti-VE-Cadherin (R&D Systems, Minneapolis, MN, AF938) and Anti-FIRE marker (as described in Supplementary material online, Table S3) at 4°C overnight, followed by Alexa Fluor-568 and −647 (Thermo Fisher Scientific, Waltham, MA, 1:500) secondary antibody for 1 h at room temperature. Nuclei were counterstained using Fluoroshield with DAPI (Sigma Aldrich, St. Louis, MO, #F6057). All images were taken at either 10× or 40× magnification using inverted fluorescence microscope (Keyence #BZ-X800, Japan) as previously described. FIRE expression in CDH5+ ECs was visualized and quantified via the combined use of MATLAB algorithms (MathWorks, MA), ImageJ, and Fiji (NIH, MD) as previously described. Plaque severity (grades I to VI) in coronary sections was stratified by 6 independent graders in accordance with the American Heart Association (AHA) histological classification guidelines. Deidentified donor information is displayed in Supplementary material online, Table S4.

2.11. Cell culture

Primary human aortic endothelial cells (HAECs; Cell Applications #304–05a) were cultured in complete medium composed of MCDB 131 (Corning, Corning, NY, #15–100-CV) supplemented with 10% FBS (R&D Systems, Minneapolis, MN, #S11550), 1% L-glutamine (Gibco, Billings, MT, #25030–081), 1% penicillin-streptomycin (Gibco, Billings, MT, #15140–122), 1% endothelial cell growth supplement (ECGS; bovine brain extract), 50 μg/mL L-ascorbic acid (Sigma-Aldrich, St. Louis, MO, #A5960), 1 μg/mL hydrocortisone (Sigma–Aldrich, St. Louis, MO, #H088), 10 ng/mL EGF (STEMCELL Technologies, Vancouver, Canada, #78006), 2 ng/mL FGF (ProSpec, Mount Vernon, NY, #CYT-218-b), 2 ng/mL IGF-1 (R&D Systems, Minneapolis, MN, #291-G1), and 1 ng/mL VEGF (BioLegend, San Diego, CA, #583706). Cells were seeded on 6-well plates precoated with 0.1% gelatin (Sigma-Aldrich, St. Louis, MO, #G1890) and maintained at 37 °C with 5% CO₂ in a humidified incubator. Media were changed every 2 days, and cells between passages 4 and 8 were used for experiments.

2.12. In vitro FIRE induction

HAECs were cultured in 6-well plates and subjected to unidirectional laminar shear stress (ULS, mimicking s-flow) or oscillatory shear stress (OSS, mimicking d-flow) using an orbital shaker (19 mm orbital radius; Benchmark Scientific, Sayreville, NJ, BT3001) at 150 rpm (5–7 dynes/cm²) or 80 rpm (≤3 dynes/cm²), respectively as we published,²⁷ inside the humidified cell culture incubator at 37 °C with 5% CO₂. For ULS conditions, cells within a 17 mm radius of the well centre were removed prior to sample collection to ensure consistent high shear exposure. To induce FIRE, HAECs were exposed to OSS using the orbital shaker with or without 5 μM 27-hydroxycholesterol (Avanti Research, Renton, WA, #700021) either alone or in combination with a cocktail of pro-inflammatory cytokines: 75 ng/mL IFNγ (PeproTech, Cranbury, NJ, #300–02), 75 ng/mL IL-1β (PeproTech, Cranbury, NJ, #200–01B), 50 ng/mL TNFα (PeproTech, Cranbury, NJ, #300–01A), 10 ng/mL TGF-β1 (PeproTech, Cranbury, NJ, # 100–21C), and 10 ng/mL TGF-β2 (PeproTech, Cranbury, NJ, #100–35B). RNA and protein were harvested after 3 or 7 days of treatment for RT-qPCR and western blot analysis.

2.13. qPCR

Total RNA was isolated from cells lysed in TRIzol^™ Reagent (Thermo Fisher Scientific, Waltham, MA, #15596018) and purified using the Direct-zol RNA Miniprep Kit (Zymo Research, Irvine, CA, #R2052), following the manufacturer’s protocol. Reverse transcription was carried out using PrimeScript RT Master Mix (Takara Bio, San Jose, CA, #RR036B) to generate cDNA. qPCR was then performed with PerfeCTa SYBR Green FastMix (QuantaBio, Beverly, MA, #95073–05K) on the StepOnePlus Real-Time PCR System (Applied Biosystems, Waltham, MA, #4376600). GAPDH was used as the internal control for normalization, and relative gene expression levels were calculated using the ΔΔCt method. List of primers used in this study is included in Supplementary material online, Table S5.

2.14. Western blot

Protein lysates were prepared using RIPA buffer (Boston BioProducts, Milford, MA, #BP-115) supplemented with Halt^™ Protease Inhibitor Cocktail (Thermo Fisher Scientific, Waltham, MA, #78429). Total protein concentrations were quantified via the BCA protein assay kit (Thermo Fisher, #23225). Equal protein amounts were loaded onto 10% SDS–PAGE gels for electrophoresis and transferred to PVDF membranes (Bio-Rad #1620177) using Towbin buffer through overnight wet transfer at 25 V under 4°C. Membranes were blocked for 1 h at room temperature using 5% bovine serum albumin (Sigma–Aldrich, St. Louis, MO, #A7906), followed by overnight incubation at 4 °C with primary antibody. The next day, membranes were washed and incubated at room temperature for 1 h with secondary HRP antibodies at 1:2000 dilution. Antibodies used in this study are listed in Supplementary material online, Table S3. Protein bands were detected using Immobilon Western chemiluminescent substrate (Millipore, Sigma, Burlington, MA, #WBKLS0500), and signals were visualized with the iBright FL1000 imaging system (Thermo Fisher Scientific, Waltham, MA).

2.15. Phagocytosis assay

Phagocytic activity was evaluated using the IgG FITC-Labeled Phagocytosis Assay Kit (Cayman Chemical, Ann Arbor, MI, Cat. No. 500290) according to the manufacturer’s protocol with minor modifications. HAECs were first treated with OSS, 27HC, and a cytokine cocktail for 3 days for FIRE induction. For the assay, FITC-IgG–labeled beads were added directly to the culture medium at a 1:200 dilution and incubated with the FIRE-induced cells at 37 °C for 1 h to allow phagocytic uptake. Following incubation, cells were washed twice with the provided assay buffer to remove unbound beads. External fluorescence was quenched by incubating the cells with 0.2% Trypan Blue solution for 2 min, followed by two additional washes with assay buffer. Cells were then fixed with 4% paraformaldehyde for 15 min at room temperature. After fixation, cells were washed twice with PBS, stained with Hoechst solution at room temperature. Fluorescence imaging was immediately performed at 4× magnification using the inverted fluorescence microscope (Keyence #BZ-X800, Japan). Mean fluorescence intensity of FITC-conjugated beads within a fixed field-of-view was quantified by using ImageJ (NIH, MO).

2.16. Lipid uptake assay

Dil-labeled oxidized LDL (Dil-OxLDL) uptake was assessed using the Invitrogen^™ Dil-OxLDL kit (Cat. No. L34358) following the manufacturer’s protocol with minor modifications. For the assay, HAECs were first treated with OSS, 27HC, and cytokine cocktail for 3 days for FIRE induction, followed by incubation with Dil-OxLDL at a final concentration of 15 μg/mL for 4 h at 37 °C. Following incubation, cells were washed twice with PBS containing Ca²⁺ and Mg²⁺ to remove unbound Dil-OxLDL. Cells were then fixed with 4% paraformaldehyde for 15 min at room temperature. After fixation, samples were washed twice with PBS and incubated with Hoechst nuclear stain (1:1000 dilution in PBS) for 10 min at room temperature. Fluorescence imaging was immediately performed at 4× magnification using the inverted fluorescence microscope (Keyence #BZ-X800, Japan). Mean fluorescence intensity of DiI-oxLDL uptake within a fixed field-of-view was quantified by using ImageJ (NIH, MO).

2.17. BODIPY staining on cultured ECs

After FIRE induction, HAECs were washed, fixed with 4% PFA (Santa Cruz Biotechnology, Dallas, TX, #sc-281692), and permeabilized with 0.1% Triton X100 (Sigma–Aldrich, St. Louis, MO, #X100). Permeabilized cells were then incubated with 0.1 mg/mL BODIPY493/503 (Thermo Fisher Scientific, Waltham, MA, #D3922) for 15 min at room temperature, followed by the nucleus staining using Hoechst Nucleic Acid Stains (Thermo Fisher Scientific, Waltham, MA, #H3570). BODIPY fluorescence images were taken at 4× using an inverted fluorescence microscope (Keyence #BZ-X800, Japan). Mean fluorescence intensity of BODIPY staining within a fixed field-of-view was quantified by using ImageJ (NIH, MO).

2.18. Validation of LDL receptor knockdown by AAV-PCSK9 and hypercholesterolaemia

To validate knockout of LDLR, mouse liver tissues were homogenized using BeadBug^™ 6 microtube homogenizer (#31–216) in tubes containing RIPA buffer (Boston BioProducts, Milford, MA, #BP-115) with complete Mini protease inhibitor cocktail (Roche Diagnostics, Indianapolis, IN, #11836153001) to extract proteins. The total protein concentrations were determined using BCA assay (Thermo Fisher Scientific, Waltham, MA, #23225). For SDS–PAGE analysis, 10% gel was used, and the electrophoresis was run at 110 V for 1.5 h. The proteins were then transferred overnight at 25 V and 4°C onto the PVDF membranes (Bio-Rad, Hercules, CA, #1620174). For protein detection, the PVDF membranes were incubated overnight at 4°C with primary antibodies: anti-LDLR rabbit monoclonal antibody (1:1000; Abcam #AB286156) and anti-β-actin mouse monoclonal antibody (1:1000; Sigma–Aldrich, St. Louis, MO, #A5316), which served as a loading control for the liver protein samples. After a series of washes with TBS-T buffer, the membranes were incubated at room temperature (RT) for 1 h with the corresponding secondary antibodies: goat anti-mouse IgG horseradish peroxidase (HRP) (Cayman Chemical, Ann Arbor, MI, #10004302) and goat anti-rabbit IgG HRP (Cayman Chemical, Ann Arbor, MI, #10004301). Protein bands were visualized using the Immobilon Western chemiluminescent HRP substrate (MilliporeSigma, Burlington, MA, #WBKLS0500) and the iBright FL1000 imaging system, which were quantified using ImageJ software and normalized based on the average expression level in control groups (Ctrl_2wk and Ctrl_4wk).

As reported previously, mouse plasma lipid levels were analysed by the Emory Biomarker Core Laboratory using a Beckman CX7 biochemical analyzer.¹⁹

2.19. Statistical analyses

Statistical analyses were conducted using GraphPad Prism version 10.0.0 software. Sample number N is included in each figure caption, and all data are presented as mean ± S.E.M. All datasets were analysed for normality using the Shapiro–Wilk test, and for equal variance using the F test for datasets with two groups, or the Brown–Forsythe test for groups of two or more. Comparisons between two groups were conducted using either a two-tailed unpaired Student’s t-test for normally distributed data, or a two-tailed unpaired Mann–Whitney test for non-normally distributed data. For normally distributed data with unequal variances, the Student’s t-test with Welch’s correction was used. Comparisons between three or more groups were conducted using the one-way ANOVA for normally distributed data or the Kruskal–Wallis test for non-normally distributed data. For normally distributed data with unequal variances, the Brown–Forsythe ANOVA test was used. The Tukey, Dunnett, and Dunn multiple-comparisons tests were used for post hoc pairwise comparisons following ANOVA. The Tukey test was used when all groups in the analysis were compared. The Dunnett test was used when all groups were compared with a single control group. The Dunn test was used to compare all groups against each other. P < 0.05 was considered significant for all statistical tests that were performed.

3. Results

3.1. D-flow induces drastic shifts in arterial cell populations of mouse carotid arteries during atherogenesis

To define the transcriptomic landscape of arterial cells at a single-cell resolution during atherogenesis in response to d-flow under hypercholesterolaemia, we carried out a scRNA-seq study using the common carotid arteries. For this study, we used 75 male C57BL/6 mice treated with (i) control (s-flow and chow diet), (ii) d-flow alone (chow diet), (iii) hypercholesterolaemia alone under s-flow (s-flow), and (iv) d-flow under hypercholesterolaemic conditions at 2 and 4 weeks after PCL surgery (Figure 1A, Supplementary material online, Figure S2 and Table S1). Hypercholesterolaemia was induced by a single AAV8-PCSK9 injection and feeding with a Western diet, as we previously published.²⁹ D-flow was induced in the LCA by PCL, while the contralateral RCA was continued to be exposed to s-flow. LCAs and RCAs of mice without PCL surgery were exposed to s-flow. As expected, LCAs developed atherosclerotic plaques only in the d-flow under the hypercholesterolaemia group at 2 and 4 weeks post-PCL (Figure 1A, Supplementary material online, Figure S2). AAV-PCSK9 injection effectively knocked out the LDL receptors (LDLR) in mouse liver, as validated by western blots (see Supplementary material online, Figure S3A and B), and hypercholesterolaemia (plasma total cholesterol > 400 mg/dL) was confirmed for hypercholesterolaemia alone and d-flow under hypercholesterolaemia groups (see Supplementary material online, Figure S3C–G).

Figure 1.

scRNA-seq data clustering analysis of mouse carotid artery 1 cells exposed to d-flow and/or hypercholesterolaemia during atherogenesis. (A) C57BL/6 mice were treated with or without AAV-PCSK9 injection and Western diet for 2 or 4 weeks, with or without PCL surgery. Representative macroscopic images of mouse carotid arteries and aortic arch are shown for 4 weeks post-PCL time points. Atherosclerotic plaque development occurred only in the LCAs of the d-flow and hypercholesterolaemia group (white arrow). Scale bar: 1 mm. (B) UMAP plot of 98 553 cells from the scRNA-seq data of Ctrl (s-flow, normal cholesterol), D-flow (d-flow, normal cholesterol), HighChol (s-flow, hypercholesterolaemia), and D-flow_HighChol (d-flow, hypercholesterolaemia) groups at 2 and 4 weeks post-PCL mice reveals 25 unique cell clusters. Major cell populations include endothelial cells (ECs), vascular smooth muscle cells (SMCs), fibroblasts (FBs), macrophages (MΦs), dendritic cells (DCs), neutrophils (NTs), B cells (BCs), T cells (TCs), and natural killer cells (NKs). Leukocytes include MΦs, DCs, NTs, BCs, TCs, and NKs. (C) Stacked violin plot shows the expression levels of canonical marker genes used to annotate each cell cluster. (D) UMAP plot of each experimental condition is shown across time (2 days, 2 weeks, and 4 weeks). S-flow (top): Ctrl (left, boxed in green) and HighChol (right, blue) and D-flow (bottom): D-flow (left, red) and D-flow_HighChol (right, purple) are shown. N = 5–20 mice for each condition. Note that Ctrl_2d, D-flow_2d, and D-flow_2wk conditions contain only luminally collected cells from the previous work.

Single cells were first prepared from luminal digestion and flushing of the LCAs and RCAs to obtain the maximum number of ECs (Figure 1A). The remaining arteries were further digested to obtain single cells from the media and adventitia. Single cells from the lumen and leftovers were then encapsulated, barcoded, library prepared, and sequenced separately. The scRNA-seq datasets from both the lumen and the leftover were pooled together during downstream analysis to enrich the EC numbers. Consistent with our previous publication,²⁷ quality control measures were performed on the scRNA-seq data in Seurat³⁰ by selecting the single cells that expressed 200–7600 genes per cell and <10% mitochondrial genes to exclude the non-singlets or damaged cells, respectively, yielding a total of 88 884 single cells. For a more comprehensive analysis, we included 9709 single cells from 20 mice, mostly ECs obtained from luminal digestion (Luminal Only) from our previous scRNA-seq datasets from the LCAs and RCAs exposed to d-flow alone at both 2 days and 2 weeks post-PCL time points.²⁷ Therefore, we analysed a total of 98 553 single cells from 95 mice (75 mice from the present study and 20 mice from our previously published study, since they were generated using the identical mouse model, single-cell preparation, and scRNA-seq analytical pipeline) for downstream analysis.

We used the Seurat integration function to remove batch effects arising from combining scRNA-seq datasets generated from multiple experiments conducted at different times, including our prior study (see Supplementary material online, Figure S4). There were four independent libraries, or batches, that were constructed, including 1 from the previous study (see Supplementary material online, Table S1). Uniform manifold approximation and projection (UMAP) plot of the cells labelled by batch and by cell type revealed both a cohesive mixing of the different batches while maintaining the cell type identities upon integration, compared to simple merging. Testing multiple benchmarking metrics using both all cells in the dataset and only the ECs showed that Seurat integration effectively mixed the batches while conserving biological variation of different cell types (see Supplementary material online, Table S2).^31,32

UMAP clustering analysis of all single cells revealed 25 unique clusters representing 9 unique cell types (Figure 1B). Each cluster was manually annotated using canonical cell markers: 5 EC (Pecam1, Cdh5, Icam2, and Tie1), 3 vascular smooth muscle cell (SMC) (Cnn1, Myh11, and Acta2), 2 fibroblast (FB) (Dpep1, Medag, and Pdgfra), 5 macrophage (MΦ) (Cd68, Adgre1, Fcgr1, and Itgam), 3 dendritic cell (DC) (Ccr7, Fscn1, Xcr1, and Cd209a), 1 neutrophil (NT) (S100a8 and S100a9), 1 B cell (BC) (Cd79a and Cd79b), 4 T cell (TC) (Cd3d and Cd3g), and 1 natural killer cell (NK) (Klre1 and Klra7) clusters (Figure 1C).^{27,38,45–51} Furthermore, the pan-leucocyte marker Ptprc (Cd45) was expressed only in the leucocytes (Leuco: MΦ, DC, NT, BC, TC, and NK) (Figure 1C). Importantly, none of the EC and SMC clusters expressed Cd45, excluding their leucocyte origin. Furthermore, splitting the UMAP plot by source of single-cell collection into luminal flushing (intimal) vs. leftover digestion (medial/adventitial) (see Supplementary material online, Figure S5A) and the % distribution quantifications (see Supplementary material online, Figure S5B) showed that the majority of EC and immune cell clusters are predominantly collected from the luminal flushing samples, as expected. Furthermore, SMC1 and all fibroblast clusters were mostly extracted from the leftover digestion samples. These results validate our luminal flushing and leftover digestion methodologies in collecting the expected cell types from the mouse carotid arteries.

We next split the UMAP plot of the whole cell populations into 10 panels: (1) Ctrl_2 day (Luminal Only), (2) Ctrl_2 wk (Luminal Only + new Ctrl_2wk group), (3) Ctrl_4 wk, (4) HighChol_2 wk, (5) HighChol_4 wk, (6) D-flow_2 day (Luminal Only), (7) D-flow_2 wk (Luminal Only), (8) D-flow_4 wk, (9) D-flow_HighChol_2 wk, and (10) D-flow_HighChol_4 wk (Figure 1D). Note that panels 1, 6, and 7 contained single cells obtained from luminal digestion only from our previous study, while panel 2 contained luminal only cells and the new additional single cells from the LCAs and RCAs, consistent with the higher EC presence in these experimental groups (see Supplementary material online, Figure S6).²⁷ Comparison of the UMAP plot generated with or without the luminal only groups (Figure 1D, Panels 1, 2, 6, and 7) did not reveal any obvious changes to the unique clusters (data not shown), especially the EC clusters. Hypercholesterolaemia alone groups (Figure 1D, Panels 4–5, Supplementary material online, Figure S6) revealed no remarkable difference compared to the control groups (Panels 2–3). In contrast, d-flow alone for 4 weeks increased the appearance of different cell clusters, especially immune cells (Panel 8 vs. 3, Supplementary material online, Figure S6). Moreover, d-flow under hypercholesterolaemia drastically increased the accumulation of a distinct population of ECs (EC5), SMCs (SMC3), and MΦs (MΦ3 and MΦ4), especially at 4 weeks post-PCL (Panels 9–10, Supplementary material online, Figure S6). These results suggest that (1) hypercholesterolaemia alone has minimal effects while d-flow alone induces significant changes in the arterial cell population and (2) the combination of d-flow and hypercholesterolaemia induces a dramatic shift in arterial cell composition as plaques develop.

3.2. D-flow under hypercholesterolaemia may induce the reprogramming of ECs to immune-like and foam-like cells

To better understand the effects of d-flow alone, hypercholesterolaemia alone, and d-flow under hypercholesterolaemia on EC population change during atherogenesis, 5 EC clusters expressing canonical EC markers (Pecam1, Cdh5, Icam2, and Tie1) were further analysed in depth (Figure 2A). Differential gene expression (DGE) analysis revealed 1291 flow-sensitive genes that were significantly and differentially expressed with average log₂Fold Change > 1 in the 5 EC clusters, as shown in the heatmap (see Supplementary material online, Figure S7A). Gene ontology (GO) analysis using the top 100 differentially expressed genes (DEGs) (see Supplementary material online, Figure S7B) and literature-based predicted functions of each cell cluster (Figure 2A) suggested the following annotations: EC1 as healthy and atheroprotective ECs based on Klk10, Klf2, Klf4, Nos3, and Heg1 expression; EC2 as ECs undergoing EndMT with contractile SMC markers (Acta2, Tagln, and Cnn1); EC3 as proinflammatory (Cxcl1, Cxcl2, Vcam1, and Icam1) and angiogenic ECs (Cyr61 and Ctgf); and EC4 as ECs undergoing EndMT with synthetic SMC markers (Mgp, Col8a1, Fn1, and Fbln5). Both EC3 and EC4 clusters highly expressed many proatherogenic EC genes, including Edn1, Thbs1, Serpine1, Sox4, and Txndc5. Interestingly, EC5 included nearly 2800 ECs (more than 60% of all ECs in D-flow_HighChol_4 wk group) abundantly expressing the markers of immune cells (Cd68, Lyz2, C1qa, C1qb, and C1qc), which we refer to as EndIT. In contrast, d-flow alone induced the accumulation of only a few EC5 cells, consistent with our previous report²⁷ (Figure 2A and B). To our further surprise, EC5 highly expressed the markers of foam cells (Trem2, Lgals3, Gpnmb, Spp1, Lpl, and Fabp5) along with the canonical EC markers Cdh5 and Pecam1, albeit at a reduced level, which we refer to as a potential endothelial-to-foam cell-like transition (EndFT) that may occur during atherogenesis⁴⁶ (Figure 2A). In addition, EC5 cells showed a very low expression of markers of lipid droplet hydrolysis, Pnpla2 (Ctgl) and its co-activator Abhd5 (Cgi58), indicating that these ECs may accumulate endothelial lipid droplets as reported recently^52,53 (Figure 2A).

Figure 2.

ECs may transition to immune-like (EndIT) and foam-like cells (EndFT) by d-flow and hypercholesterolaemia during atherogenesis. (A) UMAP plot of 5 EC clusters (18 527 cells in total). Stacked violin plot shows the expression levels of genes used to annotate each EC cluster. EC clusters include atheroprotective EC1, EndMT (contractile SMC) EC2, proinflammatory/angiogenic EC3, EndMT (synthetic SMC) EC4, and EndIT/EndFT EC5. (B) Cell number and % cell population (cell number for each EC cluster normalized by the total number of ECs per group) quantifications for each EC cluster across all 10 experimental groups. (C) UMAP plot of each experimental group is shown for ECs. N = 5–20 mice for each condition. EndIT/EndFT (EC5) dominate the EC population in the D-flow_HighChol at 4 weeks post-PCL (Panel 10).

We next analysed the effects of hypercholesterolaemia alone, d-flow alone, and d-flow under hypercholesterolaemia on the distribution of EC clusters compared to the control. Hypercholesterolaemia alone induced a transient increase in the EC2 (contractile EndMT) cells at 2 weeks (Figure 2C, Panel 4). In contrast, d-flow alone increased EC3 (proinflammatory/angiogenic) at 2 days, EC4 (synthetic EndMT), and just a few cells in EC5 (EndIT and EndFT) at 2 and 4 weeks post-PCL (Panels 6–8). Most importantly, d-flow under hypercholesterolaemia dramatically increased the number of EC5 cells at 4 weeks post-PCL, which corresponds to robust plaque formation (Panel 10). Based on these results, we propose a 2-hit hypothesis that d-flow alone initiates a partial FIRE, including inflammation, EndMT, and partial EndIT, while d-flow under hypercholesterolaemia induces a robust FIRE, including full EndIT and the novel concept of EndFT, in addition to inflammation and EndMT, during atherogenesis.

3.3. D-flow under hypercholesterolaemia induces reprogramming of SMCs to foam cells during atherogenesis

We next examined the effects of d-flow alone, hypercholesterolaemia alone, and d-flow under hypercholesterolaemia on the SMC phenotypes. 3 SMC clusters were annotated using the markers of contractile (SMC1), synthetic (SMC2), and SMC-derived foam cells (SMC3) based on the DEGs, GO terms, and literature-based prediction of each cluster (see Supplementary material online, Figure S8A). SMC1 cells expressed contractile genes (Acta2, Myh11, Tpm2, Myl9, Tagln, Cnn1, and Smtn).^38,47,48 Hypercholesterolaemia alone did not induce any significant SMC phenotypic switches (see Supplementary material online, Figure S8B and C, Panels 4–5). D-flow alone at 2 and 4 weeks post-PCL induced significant accumulation of SMC2 cells expressing markers of synthetic SMCs (Tpm4, Col8a1, Myh10, Mgp, Col1a1, Col3a1, Fn1, Dcn, Lum, Lox, Thbs1, and Fbln5) (Panels 7–8).^38,47,48 In contrast, d-flow under hypercholesterolaemia induced a further increase in the number of synthetic SMC2 cells at 2 weeks post-PCL, while the number of SMC3 cells (SMC-derived foam cells) expressing markers of foam cells (Trem2, Lgals3, Gpnmb, Spp1, Lpl, and Fabp5) along with the canonical SMC marker Acta2, albeit at a reduced level^46,54 exploded at 4 weeks post-PCL (Panels 9–10). These results show that hypercholesterolaemia alone had minimal effects on SMC phenotypic switch, while d-flow alone stimulates phenotypic switching to synthetic SMC2 cells. The combination of d-flow and hypercholesterolaemia induces dramatic switching to SMC-derived foam cells (SMC3), consistent with prior reports.^54,55

3.4. D-flow under hypercholesterolaemia transforms MΦs into foam cells during atherogenesis

We identified five MΦ clusters expressing canonical MΦ markers (Cd68, Adgre1, Fcgr1, and Itgam) and predicted their functions based on the DEGs, GO terms, and the literature (see Supplementary material online, Figure S9A). MΦ5 cells expressing homeostatic, resident MΦ markers (Lyve1, Pf4, F13a1, Mrc1, Cbr2, Folr2, Maf, and C4b) were present in all conditions⁴⁶ (see Supplementary material online, Figure S9). Interestingly, hypercholesterolaemia alone under s-flow condition had minimal effects on the MΦ population (see Supplementary material online, Figure S9B and C, Panels 4–5). It demonstrates the protective role of s-flow on the prevention of monocyte infiltration into the arterial wall. In contrast, d-flow alone significantly increased the accumulation of MΦ1 (monocyte/macrophage) (Panel 6) and MΦ2 (inflammatory) (Panels 7–8) cells that highly expressed the markers of monocytes (Nr4a1, Mef2a, Plac8, Ccr2, Ly6c2, Clec4e, F10, Cx3cr1) at 2 days post-PCL and inflammatory markers (Il1b, Tnf, Cxcl2, Ccl2, Ccl3, Ccl4, and Ccrl2) at 2 and 4 weeks post-PCL, respectively (see Supplementary material online, Figure S9B). In response to d-flow under hypercholesterolaemia, all MΦ clusters significantly accumulated, most notably MΦ3 and MΦ4 that highly expressed the markers of foam cells (Trem2, Lgals3, Gpnmb, Spp1, Lpl, and Fabp5) (see Supplementary material online, Figure S9B and C, Panels 9–10). Interestingly, as plaques increase at 4 weeks post-PCL under hypercholesterolaemic condition, MΦ4 cells expressing lower levels of efferocytosis markers (Abca1, Axl, Cd36, and Ucp2)^56,57 than those of MΦ3 cells accumulated significantly (Panel 9 vs. 10). Similar to the SMCs, hypercholesterolaemia alone under s-flow condition had minimal effects on macrophage accumulation, whereas d-flow alone stimulated the initial accumulation of MΦ1 (monocyte/macrophage) and MΦ2 (inflammatory). In contrast, d-flow under hypercholesterolaemia induced an explosion of all macrophages, especially macrophage-derived foam cells MΦ3 (Abca1^high) and MΦ4 (Abca1^low).

3.5. D-flow induced accumulation of inflammatory and apoptotic fibroblasts and various immune cell types in the arterial wall during atherogenesis

We identified two fibroblast clusters and annotated as fibroblast progenitor/homeostatic fibroblasts (FB1) and inflammatory/apoptotic fibroblasts (FB2) using the same criteria discussed above (see Supplementary material online, Figure S10).^49–51 Our detailed analysis revealed that hypercholesterolaemia alone had minimal effects while d-flow alone and d-flow under hypercholesterolaemia increased FB2 population similarly (see Supplementary material online, Figure S10B and C).

Analysis of other immune cells, including 4 TCs, 1 NKs, 3 DCs, 1 NTs, and 1 BCs, has been annotated using the canonical marker genes⁴⁶ (see Supplementary material online, Figures S11–S13). Hypercholesterolaemia alone had minimal effects on the accumulation of these cells. In contrast, d-flow alone increased the accumulation of most TC clusters, which were further significantly increased by d-flow under hypercholesterolaemia (see Supplementary material online, Figure S11). Analysis of DCs, BCs, and NTs showed similar changes by d-flow alone and d-flow under hypercholesterolaemia, but not by hypercholesterolaemia alone (see Supplementary material online, Figures S12 and S13), demonstrating the dominant role of d-flow in immune cell infiltration during atherogenesis.

3.6. ECs expressing immune cell and foam cell markers share similar transcriptomic profiles with foam cells derived from MΦs and SMCs

The UMAP plot of our scRNA-seq data suggests an interesting possibility that immune-like and foam-like ECs (EC5) share similar transcriptomic profiles with foam cells derived from classical MΦs (MΦ3 and MΦ4) and SMCs (SMC3) during atherogenesis in response to d-flow under hypercholesterolaemia, especially at 4 weeks post-PCL, as suggested from the close proximity of these foam cell clusters (Figure 1E, Panels 9–10). To address this hypothesis, multiple independent trajectory analyses were performed using destiny, Monocle 3, and scVelo packages.^34,35,37

Diffusion map trajectory analysis using destiny was performed unbiasedly without setting any cell cluster as a root or sink using all EC, SMC, and MΦ clusters and projected into diffusion component space (Figure 3A and B). These projections suggested that ECs may transition from the atheroprotective EC1 towards immune cell-like/foam cell-like (EC5) phenotype, SMCs transitioned from the contractile SMC1 towards SMC-derived foam cell (SMC3) phenotype, and MΦs transitioned from the monocyte/macrophage (MΦ1) towards Abca1^high MΦ-derived foam cell (MΦ3) and then to Abca1^low MΦ-derived foam cell (MΦ4) phenotype. Although EC1, SMC1, and MΦ1 (initial cell populations) all originated from distinct nodes, the foam cells derived from ECs (EC5), SMCs (SMC3), and MΦs (MΦ3 and MΦ4) converged at the centre (Figure 3B). Monocle 3 pseudotime trajectory analysis further confirmed the diffusion map analysis (see Supplementary material online, Figure S14A and B). In addition, RNA velocity analysis was also carried out using scVelo to model cell fate transitions using the entire scRNA-seq data, further validating the diffusion map trajectory analysis (see Supplementary material online, Figure S14C–E). We next quantified the % distribution of the foam cell origins by comparing EC5, SMC3, MΦ3, and MΦ4. As expected, MΦ3 (17%) and MΦ4 (52%) were the major sources of foam cells (totaling 69%), while 20% of foam cells were derived from SMCs, with a surprising presence of foam cell-like ECs at 11% (Figure 3C). These results for the first time suggest that ECs may undergo EndIT and EndFT. Moreover, foam cells could be derived not only from MΦs and SMCs but also potentially from ECs during atherogenesis in response to d-flow under hypercholesterolaemia.

Figure 3.

Transcriptomic profiles of ECs expressing foam cell markers are similar to those of foam cells derived from SMCs and MΦs. (A) UMAP plot of all EC, SMC, and MΦ clusters used for further trajectory analysis in (B). (B) Diffusion map trajectory analysis of the three cell types reveals significant overlap of EC5 clusters with SMC3, MΦ3, and MΦ4 clusters from distinct origins. (C) Pie chart showing % distribution of the origins of foam cells reveals that 11% of foam cells may have derived from ECs, 20% are derived from SMCs, and 69% are derived from MΦs.

To explore the potential cell–cell interaction pathways among all arterial cell clusters during atherogenesis, CellChat cell–cell communication analysis was performed on our scRNA-seq data.³³ We identified 113 predicted signalling pathways across all cell clusters, demonstrating similar levels of interactions among them without any dominant cell type or direction (see Supplementary material online, Figure S15A). To identify cell–cell signalling pathways involving ECs potentially undergoing EndIT and EndFT, we compared the ligand–receptor interactions that were sent by EC5 with those of foam cells derived from SMCs (SMC3) and MΦs (MΦ3 and MΦ4) (see Supplementary material online, Figure S15B). Some signals were uniquely sent by EC5, SMC3, and MΦ3/MΦ4, respectively, to all cell clusters, while there were some signals commonly sent by all foam cell clusters. EC5-originated interactions were Bmp4/Bmp6-Bmp receptors, Edn1-Ednra, Hspg2-Dag1, Thbs1-Itga3 + Itgb1, Thbs1-Itgav + Itgb3, and Vwf-Itgav + Itgb3. SMC3-originated interactions were Col4a2-integrins/Cd44/Sdc and Postn-integrins. MΦ3 and MΦ4-originated interactions included Ccl3/4/6/9-Ccr1/2/5, Cd22-Ptprc, Cxcl2/4-Cxcr2/3, Il1a-Il1r1 + Il1rap, Sema4d-Cd72/Plxnb2, Tgfb1-receptors, and Tnf-receptors. Interestingly, 3 signalling pathways, App-Cd74, Mif-receptors, and Spp1-receptors, were commonly sent by all foam cell clusters. In contrast, our attempt to identify ligand–receptor signals received by the foam cells revealed only App-Cd74 interaction (see Supplementary material online, Figure S15C). Interestingly, MΦ3 foam cells were the most common receivers of many ligand–receptor interactions from various sender cells. These ligand–receptor interactions identified among arterial cells may play an important role in endothelial reprogramming and atherogenesis.

3.7. EC lineage tracing model study validates EndIT and EndFT, in addition to inflammation and EndMT, during atherogenesis by d-flow under hypercholesterolaemia

Next, we validated the FIRE hypothesis proposed from our scRNA-seq study, especially the novel EndIT and EndFT, while using endothelial inflammation and EndMT as controls by an independent immunohistochemical staining approach using an EC lineage tracing model. For this study, we used Cdh5-CreER^T2-Confetti^FL/WT mouse model expressing confetti markers, nuclear GFP, cytoplasmic YFP, or cytoplasmic RFP, specifically in ECs in a tamoxifen-inducible and stochastic manner (see Supplementary material online, Figure S16A), as reported previously.⁴³ We hypothesized that the markers of endothelial reprogramming would be co-expressed in confetti⁺ ECs exposed to d-flow, such as the LCA in the PCL model and lesser curvature (LC) of the aortic arch naturally exposed to d-flow under hypercholesterolaemic conditions. Upon tamoxifen induction, EC-Confetti mice (6 males and 15 females) were injected with AAV-PCSK9, fed a high-fat diet, and underwent the PCL surgery, identical to the scRNA-seq study design (see Supplementary material online, Figure S16B). Similar to the scRNA-seq study in C57BL/6 mice, atherosclerotic plaques developed specifically in the LCAs of the EC-Confetti mice but not in the RCAs at 4 weeks post-PCL (see Supplementary material online, Figure S16C and D). PCSK9-mediated knockout of LDLR in the liver and hypercholesterolaemia were also validated (see Supplementary material online, Figure S16E–G).

For immunofluorescence staining of the FIRE markers, the RCA and two halves of the LCA were longitudinally sectioned as we previously reported¹⁹ (Figure 4A and B). We quantified confetti induction in the ECs across different flow conditions (RCA vs. LCA) in males and females by analysing the number of confetti⁺ cells relative to the total number of luminal ECs of the carotid arteries. Approximately 16 and 17% of ECs expressed confetti in male and female RCAs, which were significantly increased by 16 and 9% in male and female LCAs, respectively (see Supplementary material online, Figure S16H). The results indicate that d-flow increased proliferation of confetti⁺ ECs, as expected,^9,10 in a sex-independent manner. Interestingly, we frequently observed patches of confetti⁺ ECs expressing the same fluorescently coloured proteins, indicating a clonal expansion of ECs under d-flow conditions (see Supplementary material online, Figure S16I).

Figure 4.

Lineage tracing study on EC-Confetti mice validates FIRE (endothelial inflammation, EndMT, EndIT, and EndFT) under d-flow and hypercholesterolaemia at 4 weeks post-PCL. EC-Confetti mice treated with d-flow and hypercholesterolaemia at 4 weeks post-PCL (N = 6 male and 15 female) were imaged macroscopically (A) and LCAs/RCAs were longitudinally sectioned, stained, imaged by fluorescence microscopy (B–R), and quantified (S–V). (A) shows a representative gross image of LCA, RCA, and aortic arch. (B–R). LCAs and RCAs were immunostained with markers of endothelial inflammation (Vcam1 and Icam1, B–D); EndMT (Acta2, Snai1, and Cnn1, E–H); EndIT (Cd68, C1qa, C1qb, and Lyz2, I–M); and EndFT (Spp1, Lgals3, Trem2, and BODIPY, N–R). (B), (E), (I), and (N) show merged images of confetti and FIRE markers at low magnification (10×), while the rest show 40× images. Confetti signals show eGFP (green), YFP (green), and RFP (red). All FIRE markers are shown in white except for green BODIPY (R). White arrows indicate confetti⁺ ECs co-expressing the FIRE markers. (S–V) Percent confetti⁺ ECs co-expressing each FIRE marker was quantified by a combined Matlab and ImageJ analysis. Confetti⁺ ECs expressing markers of inflammation (Icam1 and Vcam1, S); EndMT (Acta2, Cnn1, and Snai1, T); EndIT (C1qa, C1qb, Lyz2, and Cd68, U); and EndFT (Lgals3, Trem2, and Spp1, V). Shown are mean ± SEM, each dot (male is black, female is red) represents % of confetti⁺ ECs co-expressing FIRE markers in each longitudinal section used for quantification (N = 10–13 longitudinal sections for RCA; N = 11–24 longitudinal sections for LCA). P values were calculated by two-tailed unpaired Student’s t-test with or without Welch’s correction for normal data and two-tailed unpaired Mann–Whitney U test for non-normal data.

As controls, we first validated the protein markers of endothelial inflammation (Vcam1 and Icam1) and EndMT (Snai1, Acta2, and Cnn1) by immunofluorescence staining using specific antibodies that have been validated in the carotid arteries (LCA vs. RCA) and the aortic arch (LC vs. greater curvature (GC) exposed to s-flow) (see Supplementary material online, Figure S17). To avoid any potential complication of these protein markers arising from the neighbouring non-confetti ECs or other cell types, we focused on examining the confetti⁺ ECs in the luminal layer with minimal plaques. As expected, confetti⁺ ECs expressed Vcam1 and Icam1 (white signals, arrows) (Figure 4B–D and S), respectively, in the LCAs but not in the RCAs. In addition, EndMT markers (Snai1, Acta2, and Cnn1) were clearly and abundantly expressed in the confetti⁺ ECs in the lumen of LCAs (arrows) but not in the RCAs (Figure 4E–H and T, Supplementary material online, Figure S18). Next, the other regions with more plaques were also examined for these markers, further demonstrating their co-expression in the confetti⁺ ECs (see Supplementary material online, Figure S19A–E). These markers of inflammation and EndMT were also abundantly expressed in the confetti⁺ ECs in the LCs of the aortic arches but not in the GCs (see Supplementary material online, Figure S20A–G). These results validate that endothelial inflammation and EndMT occur in the arterial regions of d-flow induced by surgery (LCA) or natural (LC) conditions, as expected.^58–63

Next, we validated the novel concept of EndIT and EndFT during atherogenesis in response to d-flow under hypercholesterolaemia at the protein levels. Canonical protein markers of immune cells (Cd68, C1qa, C1qb, and Lyz2) and foam cells (Spp1, Lgals3, and Trem2) were examined in the carotid arteries and the aortic arches by immunostaining. To our surprise, we found robust expressions of Cd68, C1qa, C1qb, and Lyz2 in the confetti⁺ ECs in the lumen of the LCAs (arrows) but not in the RCAs (Figure 4I–M and U, Supplementary material online, Figure S19F–I). The confetti⁺ ECs in the LCs also expressed the EndIT markers robustly but not in the GCs (see Supplementary material online, Figure S20H–L). These results demonstrate that the confetti⁺ ECs undergo EndIT in response to d-flow under hypercholesterolaemic conditions during atherogenesis. Similarly, the foam cell markers (Spp1, Lgals3, and Trem2) were abundantly expressed in the confetti⁺ ECs in the lumen of the LCAs (arrows) but not in the RCAs (Figure 4N–Q and V, Supplementary material online, Figure S19J–L). We also used BODIPY staining to determine lipid droplet accumulation in confetti⁺ ECs. We found 10 out of 36 LCA sections showed luminal confetti⁺ ECs with BODIPY⁺ lipid droplet accumulation (arrows), supporting foam cell formation derived from ECs (Figure 4R, Supplementary material online, Figure S19M). However, confetti⁺ ECs stained with BODIPY in the LCs were rare (1 out of 25 sections) as shown by the rare example in Supplementary material online, Figure S20Q. Furthermore, Trem2 and Spp1 were expressed at a much lower frequency in the LCs compared to the LCAs (see Supplementary material online, Figure S20M–P). These results show the correlation between the lack of plaque development due to less severe d-flow conditions and the reduced or lack of EndFT markers in aortic arches compared to the LCAs within 4 weeks, as they require more than 2 months to observe plaques in the LCs.^13,29,64

Interestingly, there are many confetti⁺ ECs expressing FIRE markers found in the subendothelial and medial layers with or without plaques (see Supplementary material online, Figure S19A–E, G–I and L). These results suggest that ECs undergoing FIRE, especially EndMT, EndIT, and EndFT, migrate into the subendothelial and medial layers. We also found luminal (see Supplementary material online, Figure S19G and L) and medial (see Supplementary material online, Figure S19C and I) confetti⁺ ECs with the same confetti colours, indicating a potential clonal expansion and intraplaque angiogenesis (see Supplementary material online, Figure S19A, C and I). To test whether the same ECs undergo EndMT, EndIT, and EndFT, we carried out co-immunofluorescence staining studies. Co-staining with Acta2 (green) and Cnn1 (white) showed that 34% of confetti⁺ ECs coexpressed these EndMT markers in the LCAs (see Supplementary material online, Figure S21A and C). Moreover, 18% of confetti⁺ ECs coexpressed Acta2 and Cd68, indicating that EndMT and EndIT occur together within the same ECs (see Supplementary material online, Figure S21B and C).

Taken together, these results demonstrate that d-flow under hypercholesterolaemic conditions induces robust endothelial reprogramming, including full EndIT and EndFT, in addition to endothelial inflammation and EndMT, during atherogenesis.

3.8. Reanalysis of scRNA-seq data from human atherosclerotic plaques shows ECs with markers of EndIT and EndFT

To test the translational potential of the novel EndIT and EndFT in atherosclerosis, we reanalysed the scRNA-seq results from two independent scRNA-seq studies. The publicly available scRNA-seq data of 4811 single cells collected from 44 human carotid endarterectomy samples were reanalysed by SingleR-based reference mapping using our mouse cluster annotations.^38,39 The original UMAP plot of the human data showed 20 unique clusters, including 4 MΦ, 1 DC, 7 TC, 2 EC, 1 SMC, 2 NK, 1 mast cell (MC), and 2 BC clusters (see Supplementary material online, Figure S22A). Our reanalysis of the human scRNA-seq data using our mouse cluster annotations predicts that human carotid endarterectomy samples contain not only MΦ-derived foam cells (MΦ3/4) and SMC-derived foam cells (SMC3) but also the novel EC-derived immune-like and foam-like cells (EndIT/EndFT EC5) (see Supplementary material online, Figure S22B and C).

In addition, we reanalysed another publicly available scRNA-seq atlas of human atherosclerotic plaques with 257 459 single cells obtained from 12 independent datasets and 80 human samples from the carotid, coronary, and femoral arteries by reference mapping using our mouse scRNA-seq data as a reference.⁴⁰ The published UMAP plot by the authors of the human plaque atlas showed 23 unique cell clusters, including 2 FB, 1 SMC, 6 MΦ, 2 TC, 1 MC, 1 NK, 3 EC, 3 DC, 2 BC, 1 NT, and 1 Undefined cluster (see Supplementary material online, Figure S23A). Consistent with our reanalysis of the human carotid endarterectomy scRNA-seq data, our reference map reanalysis of the human scRNA-seq plaque atlas also predicts that human plaques contain EndIT/EndFT ECs (EC5) as well as SMC-derived foam cells (SMC3) and MΦ-derived foam cells (MΦ3/4) (see Supplementary material online, Figure S23B and C). These results obtained from reanalysing two independent human scRNA-seq datasets from atherosclerotic plaques suggest that ECs in the human plaques may undergo EndIT and EndFT, consistent with our mouse data.

3.9. Immunostaining validates that ECs express markers of EndIT and EndFT in human coronary atherosclerotic plaques

To further validate whether EndIT and EndFT occur at the protein level in human atherosclerotic plaques, we performed immunofluorescence staining using the left anterior descending (LAD) coronary artery sections with varying degrees of atherosclerotic plaques. These coronary arteries were isolated from donated human hearts that were unsuitable for transplantation. The coronary artery sections were divided into two groups based on the American Heart Association (AHA) plaque histology grades: AHA I-III for early atherosclerotic plaques (Figure 5A) and AHA IV-VI (Figure 5B) for advanced plaques. The coronary artery sections (N = 73 sections) obtained from 27 hearts were co-stained with antibodies specific to C1QA, C1QB, LYZ, and CD68 for EndIT (Figure 5C–F) or LGALS3, TREM2, and SPP1 for EndFT (Figure 5G–I) with CDH5 for EC identity. In agreement with our reanalysis of human scRNA-seq datasets (see Supplementary material online, Figures S22 and S23), ECs in the coronary arterial lumen (red) exhibited increased expression of each EndIT (>7–20 fold increase, Figure 5J) and EndFT marker (>3–10 fold increase, Figure 5K) in samples with advanced plaques (AHA IV-VI) compared to early plaque samples (AHA I-III). These results provide strong evidence supporting that FIRE occurs in human coronary arteries with advanced atherosclerotic plaques.

Figure 5.

FIRE increases in human coronary arteries with advanced plaque. Human coronary arteries containing varying degrees of atherosclerotic lesions were counterstained with CDH5 and FIRE markers using American Heart Association plaque severity grades. (A and B) shows representative images of C1QB in CDH5⁺ ECs at different stages of atherosclerosis in a wide field-of-view (grade III vs. V, 10× stitched). Scale bar: 500 μm. (C), (D), (E), and (F) show 40× images of EndIT (C1QA, C1QB, LYZ2, and CD68, overlapped white). Scale bar: 200 μm. (G), (H), and (I) show 40× images of EndFT (LGALS3, TREM2, and SPP1, overlapped white). Scale bar: 200 μm. Both EndIT and EndFT immunofluorescence intensity in CDH5⁺ ECs increased in grade IV–VI plaques compared to grade I to III plaques. (J–K) Mean fluorescence intensity of each FIRE marker in CDH5⁺ ECs was quantified by a combined Matlab and ImageJ analysis. CDH5⁺ EC expressing markers of EndIT (C1QA, C1QB, LYZ, and CD68, J) and EndFT (LGALS3, TREM2, and SPP1, K). Data are from 27 donor hearts and 73 sections. Quantifications presented as mean fluorescence intensity ± SEM. P values were calculated by two-tailed unpaired Student’s t-test with or without Welch’s correction for normal data and two-tailed unpaired Mann–Whitney U test for non-normal data.

3.10. Cultured ECs undergo FIRE in response to d-flow, hypercholesterolaemia, and proinflammatory cytokines

Next, we tested whether cultured ECs can undergo FIRE in the absence of other arterial cell types under the proatherogenic conditions of d-flow and hypercholesterolaemia. For this study, human aortic ECs (HAECs) were exposed to s-flow (ULS) or d-flow (OSS) with or without 27-hydroxycholesterol (27HC), the major cholesterol metabolite elevated in hypercholesterolaemic conditions in vivo,^65,66 for 3 or 7 days. As expected, ULS induced a robust expression of anti-atherogenic KLF2 and KLF4, while OSS significantly reduced their expression (Figure 6A and B). OSS induced robust inflammation and EndMT at the mRNA and protein levels (Figure 6C–F). However, EndIT was modestly increased only at the mRNA level but not the protein, while EndFT markers were not induced by OSS alone (Figure 6G–J). The addition of 27HC under OSS further boosted the expression of the inflammatory marker VCAM1 at the mRNA and protein levels, but the expression of EndMT, EndIT, and EndFT markers at the mRNA and protein levels was not affected (Figure 6C–J). These results suggested that OSS alone or OSS + 27HC can induce inflammation and EndMT, but were insufficient to induce robust EndIT and EndFT.

Figure 6.

FIRE induction in HAECs by d-flow, hypercholesterolaemia, and proinflammatory cytokines. HAECs were treated with OSS alone, OSS with 5 μM 27HC (OSS + 27HC), or OSS with 5 μM 27HC and cytokine cocktail (Cyt) composed of 75 ng/mL IFNγ, 75 ng/mL IL-1β, 50 ng/mL TNFα, 10 ng/mL TGF-β1, and 10 ng/mL TGF-β2 (OSS + 27HC + Cyt) for 3 or 7 days. Following treatments, cells were lysed, and qPCR and western blot assays were conducted. (A and B) mRNA and protein expression of atheroprotective transcription factors KLF2/4 significantly decreased in all OSS conditions relative to the ULS group. (C and D) Proinflammatory marker VCAM1 was upregulated in all OSS conditions, both at the RNA and protein levels. (E and F) Significant induction of EndMT markers (ACTA2, CNN1, SNAI1) was achieved in all OSS conditions after 3 or 7 days. (G and H) EndIT was partially observed in OSS alone and OSS + 27HC conditions at the RNA level only, while robust EndIT was induced with OSS + 27HC + Cyt. (I and J) OSS + 27HC + cytokines strongly upregulated foam cell markers (LGALS3, SPP1, TREM2) in HAECs both at the RNA and protein levels. Note that for TREM2 western blot in (J), TREM2 is known to multimerize and form stable trimers.⁶⁷ Data are presented as mean RNA or protein expression ± SEM. N = 6–33 for qPCR analyses; N = 3 for western blot analyses. P values were determined by two-tailed unpaired Student’s t-test with or without Welch’s correction for normal data and two-tailed unpaired Mann–Whitney U test for non-normal data.

We hypothesized that an additional proatherogenic milieu, such as proinflammatory cytokines secreted by infiltrating immune cells and neighbouring stromal cells under d-flow and hypercholesterolaemia during atherogenesis, is required to induce robust EndIT and EndFT. In support of this hypothesis, our scRNA-seq data revealed that several proinflammatory cytokines (TNFα, IL-1β, IFNγ, TGF-β1, and TGF-β2) are significantly expressed in immune cells and stromal cells in the LCA under d-flow and hypercholesterolaemia (see Supplementary material online, Figure S24). Therefore, we tested this hypothesis by treating HAECs with additional cytokine cocktails under the OSS and 27HC conditions. We found that the cytokine cocktail (50 ng/mL TNFa, 75 ng/mL IL-1b, 75 ng/mL IFN-γ, 10 ng/mL TGF-β1, and 10 ng/mL TGF-β2) together with OSS and 27HC induced robust EndIT and EndFT marker expressions both at the mRNA and protein levels (Figure 6G–J). Interestingly, individual cytokines failed to induce consistent EndIT and EndFT (data not shown), suggesting that the combination of these cytokines is required for the FIRE events. Taken together, these results demonstrate that the combination of OSS, 27HC, and the proinflammatory cytokine cocktail simulating the in vivo proatherogenic conditions of d-flow, hypercholesterolaemia, and proinflammatory milieu robustly induces markers of immune cells and foam cells in ECs, even in the absence of other cell types, suggesting that EndIT and EndFT may occur.

Furthermore, we determined whether EndIT and EndFT occur at the functional level. Since there are no established functional markers of EndIT and EndFT, we used the following three surrogate assays to test whether HAECs gain characteristics of immune cells by studying phagocytosis (FITC-labeled bead uptake)^68,69 and foam cells by studying lipid uptake (Dil-labeled oxLDL)^70,71 and lipid droplet accumulation (BODIPY staining)⁵³ in response to OSS, 27HC, and the cytokine cocktail. Surprisingly, OSS alone induced robust phagocytosis (see Supplementary material online, Figure S25A and B), Dil-oxLDL uptake (see Supplementary material online, Figure S25C and D), and lipid droplet accumulation (see Supplementary material online, Figure S25E and F) compared to the ULS condition. Interestingly, however, OSS + 27HC + cytokine cocktail significantly enhanced Dil-oxLDL uptake (from 10-fold by OSS to 15-fold by OSS + 27HC + cytokine cocktail), while phagocytosis and lipid droplet accumulation did not further increase. These results demonstrate that ECs can gain features of immune-like cells and foam-like cells in response to d-flow, hypercholesterolaemic, and proinflammatory conditions.

Together, these transcript, protein, and functional marker studies clearly demonstrate that the proatherogenic conditions that resemble the d-flow + hypercholesterolaemia in vivo, including the proinflammatory milieu, induce robust changes in the markers of inflammation, EndMT, EndIT, and EndFT in HAECs.

3.11. Reanalysis of perturb-seq data reveals 40 flow-sensitive genes that regulate expression of FIRE markers

To investigate the roles of the 1291 flow-sensitive genes that are significantly and differentially expressed by the 5 EC clusters on FIRE, we reanalysed the recently published and publicly available CRISPRi-Perturb-seq data.⁴² Briefly, Schnitzler et al. knocked down 2285 genes related to 306 coronary artery disease (CAD)-GWAS signals at a single-cell resolution in human ECs and identified sets of genes that may regulate CAD and EC-specific functional programs. Interestingly, 257 out of our 1291 flow-sensitive genes (~20%) from our scRNA-seq data were included among the 2285 target genes in the Perturb-seq study (see Supplementary material online, Figure S26A). They reported a total of 17 849 genes that were significantly altered by perturbing the 2285 CAD-GWAS-related genes. Interestingly, 1045 of our 1291 flow-sensitive genes (~81%) were altered by the gene perturbations.⁴² Moreover, 14 of our 1045 flow-sensitive genes are FIRE markers altered by CAD-GWAS perturbations (see Supplementary material online, Figure S26A), suggesting their potential importance in CAD.

Moreover, our reanalysis revealed that these 14 FIRE marker genes were regulated by 329 CAD-GWAS related genes, of which 40 are also flow-sensitive (see Supplementary material online, Figure S26B). Of the 14 FIRE marker genes, 10 were perturbed by the 40 flow-sensitive and CAD-GWAS-related genes. For example, knockdown of TP53 altered markers of EndMT (ACTA2, FN1, TAGLN) and EndFT (LGALS3) (see Supplementary material online, Figure S26C). EndMT markers (FN1, MGP, TAGLN) were regulated by many flow-sensitive genes. FOXC1 knockdown altered an EndIT marker (CD68), while EndFT marker (FABP5) was regulated by DCTPP1 and TXNRD1. These results provide a causal relationship between the flow-sensitive CAD-GWAS-related genes and FIRE events.

4. Discussion

Here, we showed that d-flow under hypercholesterolaemic conditions rapidly and robustly induced the novel EndIT and EndFT in addition to endothelial inflammation and EndMT, which we collectively coined as FIRE (flow-induced reprogramming of ECs), during atherosclerotic plaque development. These provocative, novel concepts of EndIT and EndFT proposed from the scRNA-seq study were validated by the lineage-tracing studies in the EC-specific confetti mice and in vitro HAEC studies at transcript, protein, and functional levels. Moreover, these mouse and HAEC-based concepts of EndIT and EndFT were further supported by the reanalysis of two publicly available human atherosclerotic plaque scRNA-seq datasets^38–40 and the immunostaining studies of human coronary artery plaques. These findings demonstrate that aortic ECs are highly plastic and heterogeneous and can transition to immune-like (EndIT) and foam-like cells (EndFT) in response to d-flow under hypercholesterolaemic conditions during atherogenesis.

We previously showed that d-flow alone induces partial FIRE, including endothelial inflammation, EndMT, and partial EndIT in a few cells, in the mouse PCL model and in human aortic ECs in vitro.²⁷ As discussed in Section 1, atherosclerosis preferentially develops in arterial regions exposed to d-flow conditions, such as branching points, in the presence of hypercholesterolaemia,^7,8,10 demonstrating the interaction between the focal flow condition and systemic hypercholesterolaemia through unclear mechanisms. Here, we tested the two-hit hypothesis that d-flow is the initial instigator of partial FIRE and systemic hypercholesterolaemia is the major fuel, and the combination of the two leads to a full-blown FIRE and atherosclerotic plaque development (Figure 7).

Figure 7.

Summary and two-hit hypothesis of d-flow and hypercholesterolaemia in atherogenesis. D-flow is the initial instigator of partial FIRE, including endothelial inflammation, EndMT, and partial EndIT. D-flow under hypercholesterolaemic conditions triggers a robust FIRE, involving endothelial inflammation, EndMT, full EndIT, and EndFT, leading to atherosclerotic plaque development.

To test the two-hit hypothesis, we for the first time compared the effects of d-flow alone, hypercholesterolaemia alone, or d-flow under hypercholesterolaemia on single-cell transcriptomics using more than 18 000 single ECs in vivo. Our data first confirmed that d-flow alone induces partial EndIT only in a few cells. Surprisingly, hypercholesterolaemia alone had a minimal impact on arterial cell heterogeneity except for transient EndMT (EC2), indicating the dominant atheroprotective effect of s-flow even under the hypercholesterolaemic conditions. In contrast, we found that d-flow under hypercholesterolaemia induced a full and robust expression of EndIT markers in the majority of ECs (~60%) (Figure 2B), with the transcriptomic profiles that are nearly identical to those of foam cells derived from MΦs and SMCs (Figure 3B). Interestingly, those ECs expressing the markers of EndIT/EndFT did not express most canonical MΦ markers, while expressing Cd68 and other markers of immune and foam cells (Figure 2A), indicating that these are not of leucocyte origin. These scRNA-seq results strongly supported the two-hit hypothesis.

We validated the provocative hypothesis proposed from the scRNA-seq data, revealing the novel concepts of EndIT and EndFT by comprehensive immunohistochemical staining studies using an EC-specific lineage-tracing model. EC-specific confetti mice were used to examine whether genetically marked ECs with fluorescent confetti (GFP, YFP, or RFP) co-expressed the canonical markers of EndIT and EndFT, in addition to inflammation and EndMT. To minimize potential contributions from non-confetti ECs and other cell types, we initially focused on examining luminal confetti⁺ ECs adjacent to the internal elastic lamina with minimal plaques in the LCAs. These data were further validated in additional confetti⁺ ECs with plaques in the LCAs as well. We also demonstrated that these FIRE markers are co-expressed in confetti⁺ ECs in the LCs of aortic arches exposed to natural d-flow conditions in comparison to the LCAs exposed to the PCL model. In addition to the protein markers of EndFT, we also demonstrated lipid accumulation in the EndFT cells by using BODIPY staining in the confetti⁺ ECs of LCAs. In support of our EndFT concept, recent studies demonstrated that ECs indeed accumulate lipid droplets under hypercholesterolaemic conditions.^52,53 We also found that EndFT cells lose the markers of lipid droplet hydrolysis, Ctgl and its co-activator Cgi58, which were reported to cause endothelial lipid droplet accumulation (Figure 2A). We speculated that the previously reported endothelial lipid droplets are related to the foam cell development reported here. Interestingly, BODIPY⁺/confetti⁺ ECs were relatively rare in the LCs of aortic arches, which is due to a minimum plaque development in these areas under this acute condition. Interestingly, OSS alone was able to induce lipid droplet accumulation in HAECs (see Supplementary material online, Figure S25C). In addition, we found that the number of confetti⁺ ECs was significantly higher in the LCAs (see Supplementary material online, Figure S16H) compared to the RCAs, indicating that d-flow induced proliferation of ECs, as expected.^9,10 Moreover, as shown in Supplementary material online, Figure S19G and L, we frequently observed patches of luminal ECs with the same confetti colours, suggesting a potential clonal expansion. In addition, some confetti⁺ ECs found in the subintimal and medial layers show the same fluorescent colours, further indicating a potential clonal expansion (see Supplementary material online, Figures S16I, S19C and I). However, these results and their pathophysiological implications need additional validation and studies. It must also be noted that the observed flow-induced endothelial reprogramming is the result of both d-flow-induced activation and s-flow-mediated suppression of pro- and anti-atherogenic events or vice versa. For example, d-flow leads to the suppression of s-flow-mediated expression of atheroprotective transcription factors KLF2 and KLF4, causing ECs to gain proinflammatory and proatherogenic phenotypes.^9,72 In parallel, d-flow also activates the NF-κB and TGFβ pathways that promote endothelial inflammation and EndMT.^9,72 Taken together, these mechanisms synergistically contribute to the induction of FIRE.

We identified four different foam cell clusters derived from SMCs (SMC3), MΦs (MΦ3 and MΦ4), and potentially from ECs (EC5). Most surprisingly, our data for the first time suggests that foam cells are derived not only from the well-known MΦs and SMCs but ECs may also be a source of foam cells. Our quantification indicates that 11% of foam cells may be derived from ECs along with 69% from MΦs and 20% from SMCs (Figure 3C). Previous studies showed that >50% and 60–70% of total foam cells in human plaques and ApoE^−/− mice, respectively, could be derived from SMCs as determined by flow cytometry and lineage tracing methods.^54,55,73 The reasons for quantitative differences among these studies for the % of SMC-derived foam cells are unclear. Potential reasons include different mouse models (ApoE^−/− vs. C57BL/6 treated with AAV-PCSK9) and the quantitative methods (flow cytometry-based method vs. scRNA-seq). Nevertheless, in our model, we confirm the abundant presence of SMC-derived foam cells while reporting the novel EC-derived foam cells as well.

While the overall transcriptomic profiles of these foam cell clusters derived from ECs, SMCs, and MΦs largely overlap (Figure 3B), in-depth DEG analysis also showed their unique characteristics. Foam cell clusters derived from ECs, SMCs, and MΦs show expression of their respective canonical cell markers (see Supplementary material online, Figure S27, EC5: Pecam1, Cdh5; SMC3: Myh11, Acta2; MΦ3: Cd68, Adgre1, Fcgr1, Itgam; MΦ4: Cd68). Similar to the MΦ- (MΦ3/4) and SMC-derived foam cells (SMC3), EC-derived foam cells (EC5) abundantly express the canonical markers of foam cells (Trem2, Lgals3, Gpnmb, Spp1, Lpl, and Fabp5) and their population increases in a time-dependent manner during atherogenesis (Figure 2).⁴⁵ MΦ3 (Abca1^high) that highly expresses the efferocytosis markers (Abca1, Axl, Cd36, and Ucp2)^55,56 is predominant in early atherogenic phase at 2 weeks post-PCL under hypercholesterolaemia. In contrast, EC5, SMC3, and MΦ4 (Abca1^low) foam cells, predominant in advanced plaques at 4 weeks post-PCL under hypercholesterolaemia, show a significantly reduced expression of efferocytosis markers (see Supplementary material online, Figure S27), suggesting a potentially impaired efferocytotic capacity. Efferocytosis is a clearance of apoptotic cells by phagocytosis, which is effective during the early phase of atherosclerosis development but becomes impaired as atherosclerosis progresses, leading to necrotic core development.⁶⁵ Likewise, there is a reduced expression of lipid hydrolysis markers (Pnpla2 and Abhd5) in EC5, SMC3, and MΦ4, suggesting intracellular lipid droplet accumulation in these cells. Our results suggest an intriguing possibility that foam cells derived from ECs, SMCs, and Abca1^low MΦs may have impaired efferocytotic and lipid handling capacities, contributing to advanced plaque development.

HAEC studies provide strong evidence supporting EndIT and EndFT at the transcript, protein, and functional levels in response to the proatherogenic conditions that resemble the d-flow + hypercholesterolaemia in vivo, including the proinflammatory milieu (Figure 6, Supplementary material online, Figure S25). However, we found a discrepancy between the transcript/protein markers and the surrogate functional markers of EndIT and EndFT in response to OSS alone. At the level of transcript and protein expression, OSS alone or OSS + 27HC did not induce robust changes in EndIT and EndFT markers. In contrast, OSS alone was able to induce the functional markers of immune cells and foam cells in HAECs. While additional refinement is likely in the future, our current conservative criteria of EndIT and EndFT require ECs to express the markers of immune cells and foam cells at the transcript, protein, and surrogate functional levels. Using the criteria and the requirement of both d-flow and hypercholesterolaemia to induce plaque development, we interpret that our data support the 2-hit hypothesis: d-flow (OSS) alone induces inflammation and EndMT but not robust EndIT and EndFT, while the combination of d-flow (OSS), hypercholesterolaemia (27HC), and proinflammatory milieu (the cytokine cocktail) induces robust EndIT and EndFT.

Our current study identified and validated flow-sensitive genes involved in FIRE during atherosclerosis, revealing a treasure trove of information. Defining the genes and mechanisms regulating FIRE, especially the novel EndIT and EndFT, is a crucial future direction in understanding the role of flow in atherogenesis and developing anti-atherogenic therapies. Interestingly, a recent Perturb-seq study⁴² provides important direction and clues for the mechanisms. Our reanalysis of the Perturb-seq data identified 40 CAD-GWAS-related genes that are also flow-sensitive, regulating 10 FIRE markers involved in endothelial inflammation, EndMT, EndIT, and EndFT (see Supplementary material online, Figure S26C). These results reveal the potential clinical importance of these flow-sensitive CAD-GWAS genes in FIRE and atherosclerosis for both mechanistic insights and therapeutic targets.

There are several limitations in this study. Due to the low number of cells (mostly ECs and immune cells) collected in the luminal digestion carotid samples, we had to pool single-cell preparations from 5 or 10 mice to obtain a sufficient number of single cells for sequencing. With this approach of pooling the single cells from multiple mice, the mouse-level information for each single cell is lost, making it difficult to conduct statistical analysis across different conditions. However, to overcome this limitation, we conducted extensive immunostaining studies using the EC-Confetti mice for a genetic lineage-tracing study (Figure 4). Our extensive immunostaining studies using 12 different antibodies and a lipid staining method (BODIPY) in N = 21 mice (6 males and 15 females) validate our key conclusions on FIRE revealed by the scRNA-seq study. Furthermore, the underlying mechanisms responsible for EndIT and EndFT under d-flow and hypercholesterolaemic conditions are unclear. The potential list of genes and pathways responsible for these pathways includes more than 1000 flow-sensitive genes revealed in this study. Systematic screening approaches are needed to address these mechanisms. The current scRNA-seq study was conducted using only male mice due to the high cost of scRNA-seq studies. We addressed this limitation by carrying out validation studies using both male and female EC-specific confetti mice. Our quantifications comparing male and female samples revealed no major sex-based differences in FIRE, as both male and female EC-Confetti mice exhibited FIRE specifically in the LCA endothelium in response to d-flow under hypercholesterolaemia (Figure 4S–V). These results strongly validated the novel EndIT and EndFT during atherogenesis in a sex-independent manner.

Furthermore, we induced hypercholesterolaemia in our mice using a combination of Western diet and AAV-PCSK9. AAV-PCSK9 was shown to induce systemic inflammation,⁷⁴ which may have affected our results on EndIT and EndFT. However, hypercholesterolaemia alone caused by AAV-PCSK9 did not induce EndIT and EndFT, although EndMT was transiently increased at 2 weeks but not at 4 weeks post-PCL. The % confetti expression depends on when the tamoxifen injection is performed. Treatment with tamoxifen at an earlier time point, at 4 weeks, induced much higher confetti induction than that induced at 6 weeks, reflecting that their carotid EC development is reaching a young adult level with a relatively low EC proliferation rate. In our initial studies, mice were tamoxifen-treated at 6 weeks, but we switched to an earlier timepoint at 4 weeks. Nevertheless, the low confetti labelling efficiency of ECs remains a limitation.

In summary, we demonstrate that d-flow and hypercholesterolaemia are two major atherogenic hits, inducing a full-blown FIRE, especially the novel EndIT and EndFT, as well as endothelial inflammation and EndMT, and atherosclerotic plaque development (Figure 7). The comprehensive single-cell atlas of mouse atherosclerotic plaques and the flow-sensitive genes, proteins, and pathways under d-flow and hypercholesterolaemic conditions identified in this study are invaluable resources for mechanistic understanding and therapeutic targets of atherosclerosis.

Supplementary Material

Supplementary material

Supplementary material is available at Cardiovascular Research online.

Translational perspective.

Atherosclerosis, a leading cause of death worldwide, preferentially occurs in d-flow-exposed regions under hypercholesterolaemia by unclear mechanisms. Here, we demonstrate d-flow-induced reprogramming of ECs under hypercholesterolaemia (FIRE) in vivo and in vitro. Our scRNA-seq result using the mouse model of d-flow-induced atherosclerosis under hypercholesterolaemia suggested a FIRE hypothesis, including the novel concept of EndIT and EndFT, during atherogenesis. We validated it by the mouse lineage-tracing study, reanalysis of human plaque scRNA-seq data, immunostaining of human coronary plaques, and FIRE induction of HAECs alone by d-flow, hypercholesterolaemia, and cytokines. The genes and pathways of FIRE are potential novel therapeutic targets.

Data availability

The datasets presented in this study can be found in NCBI BioProject repository (accession number: PRJNA1112537).