Twist1 promotes endothelial phenotypic transition and unstable plaque phenotype during atherosclerosis

Abstract

Background:

Comprehensive investigation of endothelial cell (EC) dysfunction during atherosclerosis with single-cell ‘omics’ has resulted in the proposal that ECs undergo multiple alternative cell fate decisions during disease, but lack of lineage tracing and/or spatial localization complicates interpretation of these data. TWIST1, a causal gene for multiple atherosclerotic vascular diseases, is activated with low shear stress in endothelial cells (ECs), and EC-Twist1 knockout results in reduced atherosclerosis. However, it remains unclear how Twist1 affects EC phenotype and plaque biology.

Methods:

We performed EC lineage tracing, in situ analysis and scRNA-Seq in ApoE^−/− mice, both prior to disease and after 16 weeks of high-fat diet. We also performed these studies with two mouse models of EC Twist1 deletion. We overexpressed TWIST1 in human coronary artery endothelial cells (HCAECs) exposed to different flow conditions, followed by bulk RNAseq. Human scRNA-Seq data were used to validate key findings in the mouse model.

Results:

We found that EC phenotypic modulation during atherosclerosis is characterized by both pro-inflammatory and EndMT gene programs, occurring simultaneously along a single cell fate transition. Human scRNA-Seq data validated a similar EndMT transition during disease. We found that the commonly-used Twist1 conditional allele is hypomorphic, leading to reduced Twist1 expression in multiple cell types. Using a mouse model of EC-specific Twist1 deletion, we found reduced EC phenotypic modulation, decreased lesion size, and a more stable lesion phenotype. Integration of TWIST1 overexpression in HCAECs with the mouse scRNAseq data identified specific TWIST1-induced targets including CXCL12 and E-selectin during EC phenotypic modulation.

Conclusions:

Our study revealed important aspects of EC phenotypic modulation during atherosclerosis, unifying disparate observations in the field. We identified key cellular and molecular mechanisms underlying a top risk locus for multiple vascular diseases, highlighting the promotion of inflammatory EndMT by TWIST1 as a key driver of disease risk.

Graphical Abstract

INTRODUCTION

Endothelial cell (EC) dysfunction, characterized by increased EC permeability, recruitment of circulating leukocytes, and increased thrombogenicity, is an important process that drives the progression of atherosclerosis and risk for myocardial infarction. These phenomena have been investigated extensively in vitro, but studies in vivo during atherosclerosis have typically relied upon a small set of marker genes to describe EC phenotype. More recently, single-cell multi-omic studies have allowed more comprehensive phenotyping of multiple cell types, including ECs, during atherosclerosis. For example, a recent study revealed the potential for EC transitions to hematopoietic cell (EndHT), immune-like cell (EndICLT), stem/progenitor-like cell and mesenchymal cell (EndMT) phenotypes under disturbed flow and high-fat diet -induced atherosclerosis ¹. However, these studies lacked either lineage tracing ^1,2, comparison to non-diseased states ³, or spatial localization of cell populations identified in the single-cell data ^2,3, leading to disparate views of this process due to the use of differing models. It is also unclear how potential EC populations identified using single-cell approaches correspond to the previously described aspects of EC dysfunction.

TWIST1, a basic helix-loop-helix transcription factor that regulates myriad processes during development, was found to be upregulated in the atheroprone, low shear stress, lesser aortic curvature in the porcine aorta ⁴. Importantly, EC-specific Twist1 knockout resulted in reduced lesion formation at an early atherosclerosis timepoint in a small number of mice ⁴. A more recent preprint report utilizing 8 weeks of atherosclerosis, followed by Twist1 deletion and an additional 8 weeks of atherosclerosis, found reduced lesion size but a paradoxical increase in markers of plaque instability⁵.

We and others have identified TWIST1 as the causal gene at the 7p21.1 locus associated at genome-wide significance with multiple atherosclerotic vascular diseases including coronary artery disease, ischemic stroke, and peripheral arterial disease ⁶. Given the limited phenotyping of ECs during atherosclerosis in mice and humans, and the above studies suggesting an important role for Twist1 in ECs during atherosclerosis, we sought to investigate the role of Twist1 in ECs during atherosclerosis. The present study used i) multiple mouse models for EC lineage tracing, and EC-specific Twist1 deletion in combination with detailed lesion phenotyping and single-cell transcriptomics in the setting of atherosclerosis; ii) integration of mouse and human datasets; and iii) bulk transcriptomics in human coronary artery endothelial cells (HCAECs) exposed to TWIST1 overexpression and differing flow conditions, to comprehensively define EC phenotypic modulation during atherosclerosis and the cellular and molecular mechanisms by which TWIST1 modifies this process to influence lesion vulnerability and disease risk.

We found that EC phenotypic modulation is largely defined by a single phenotypic transition that encompasses both EndMT-like fibrotic changes and ‘inflammatory’ EC phenotypes, but we did not find evidence of hematopoietic or immune cell transitions. Furthermore, we found that this EC phenotypic transition was similar in human samples. We show that Twist1 deletion in ECs impairs this inflammatory EndMT transition in both the aortic root and arch, but only reduces atherosclerosis in the arch, suggesting additional factors are necessary for Twist1-induced EndMT to influence lesion formation. In the aortic arch, we found that EC-specific Twist1 deletion reduced EndMT and resulted in reduced lesion size and a more stable lesion phenotype, consistent with the genetic evidence that TWIST1 expression is associated with increased vascular disease risk in humans, and in contrast to a recent preprint report of Twist1 deletion in mice⁵. Integration of i) bulk RNAseq of TWIST1 overexpression in HCAECs, and ii) scRNAseq of EC phenotypic modulation during atherosclerosis identified specific molecular mechanisms by which TWIST1 modifies EC phenotype during disease, including upregulation of CXCL12 and E-selectin, as well as modification of EC junctions. The present study unites previous disparate findings regarding EC dysfunction and phenotypic modulation in vivo during atherosclerosis. It also demonstrates that a top genetic locus for multiple vascular diseases acts through TWIST1 in ECs to promote an inflammatory EndMT phenotype resulting in lesion instability.

METHODS

Data Availability:

The data that support the findings of this study are available from the corresponding author upon reasonable request. The single-cell RNA-sequencing data from mouse aorta and the bulk RNA-sequencing data from human coronary artery endothelial cells generated in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE342905. The Major Resources Table, which lists detailed information about the resources used in this study, can be found in the Supplemental Material.

Statistical Methods

Analysis of Lesion Characteristics -

Statistical analyses and data visualization were performed using GraphPad Prism (v9.5.0). For each quantitative comparison, data were first screened for outliers using the ROUT method (Q=1%), and the remaining data were tested for normality using the Shapiro-Wilk test. For two-group comparisons (Twist1-wt vs. Twist1-ECKO, or Twist1-flx vs. Twist1-ECKO), normally distributed data were compared using an unpaired, two-tailed Student’s t-test with Welch’s correction (not assuming equal standard deviations), and non-normally distributed data were compared using an unpaired, two-tailed Mann-Whitney U test. For comparisons across more than two groups (e.g., Twist1-wt, Twist1-ECKO, and Twist1-flx aortic Twist1 expression), a one-way ANOVA with Tukey’s post hoc test or a Kruskal-Wallis with Dunn’s post hoc test for multiple comparisons was used. Distributions of cells across pseudotime were compared using a two-sample Kolmogorov-Smirnov test. Analyses were performed separately for male and female animals; data from the two sexes were not pooled. For all tests, a two-sided P<0.05 was considered statistically significant. Exact P-values are reported in the text and figure legends. Data are presented as mean ± SEM, with individual data points shown for all quantitative comparisons. The number of biological replicates (individual animals) or independent experiments for each comparison is stated in the corresponding figure legend.

Bulk RNA-sequencing differential expression -

Differential expression between TWIST1-overexpression and empty-vector (pWPI) HCAECs under each flow condition was determined using DESeq2 (v1.28.0) on RSEM expected counts, applying an adjusted P-value (Benjamini-Hochberg) cutoff of 0.05 and an absolute log2 fold-change cutoff of 1.

Single-cell RNA-sequencing analysis -

Differential gene expression between clusters and between experimental groups was performed in Seurat (v4.1.1) using the FindMarkers function with the default Wilcoxon rank-sum test. All differential expression testing was performed on the RNA assay (normalized counts). Genes were tested if detected in at least 10% of cells in either of the two populations being compared and if they exhibited a log fold-change of at least 0.25 between groups. P-values were adjusted for multiple comparisons by Bonferroni correction based on the total number of genes in the dataset, and an adjusted P-value < 0.05 was considered statistically significant.

Mouse strains:

A tamoxifen-inducible Cre recombinase driven by the Cdh5 promoter (Tg(Cdh5-cre/ERT2)1Rha) line was bred with a tdTomato fluorescent reporter line (B6.Cg Gt(ROSA)26Sortm14(CAGtdTomato)Hze/J; 007914; JAX) to allow for the labeling of ECs and their progeny. A Twist1^flox allele ⁷ was also bred onto these alleles. The Twist1^flox allele contains the following LoxP sites: 5’ loxP site 472 bp upstream of Exon 1 and 3’ loxP site within the non-coding Exon 2, located 201bp upstream of the end of the exon. The loxP sites thus flank the Twist1 promoter and entire coding sequence contained in Exon 1. All alleles were on the C57Bl/6 ApoE ^−/− background (B6.129P2-Apoe^tm1Unc/J). The final genotypes of the EC lineage-tracing (Twist1-wt) mice were: Tg^Cdh5-CreERT2, Twist1 ^wt/wt, ROSA^tdT/+, ApoE^−/−. The final genotypes of the lineage-traced Twist1 knockout (Twist1-ECKO) mice were: Tg^Cdh5-CreERT2, Twist1^flox/flox, ROSA^tdT/+, ApoE^−/−. Given our discovery that the Twist1^flox allele was hypomorphic in multiple cell types, we also generated an additional cohort of mice to ensure EC-specific deletion of Twist1. This additional cohort was all homozygous for the Twist1^flox allele, but included littermates that were both Cre+ and Cre−, thereby ensuring that any differences were the result of EC-specific Twist1 deletion. Cre+ mice (still denoted Twist1-ECKO) had a final genotype of Tg^Cdh5-CreERT2, Twist1^flox/flox, ROSA^tdT/+, ApoE^−/−, and Cre− controls had a genotype of Twist1^flox/flox, ROSA^tdT/+, ApoE^−/−. Both male and female mice were included in this study. All mice were housed and bred in a pathogen-free facility at the University of North Carolina at Chapel Hill (UNC-CH) and experiments were conducted in accordance with the guidelines of UNC-CH’s Institutional Animal Care and Use Committee (IACUC). The animal study protocol was approved by the Administrative Panel on Laboratory Animal Care at UNC-CH.

Induction of lineage marker and Twist1 deletion by Cre recombinase:

For all scRNA-Seq, RNAscope, and immunohistochemistry experiments, the tamoxifen gavage schedule was as follows: two doses of tamoxifen (T5648-56; Sigma Aldrich), at 0.2 mg g⁻¹ bodyweight, were administered by oral gavage at 6-7 weeks of age, with each dose separated by 48 hours. High fat diet (HFD) (101511, Dyets; 21% anhydrous milk fat, 19% casein and 0.15% cholesterol) was started following the administration of both doses of tamoxifen, and animals remained on HFD for a total of 6 weeks or 16 weeks. A cohort of animals that were administered tamoxifen but did not receive HFD, deemed “baseline”, were included in the study. These animals were sacrificed 72 hours following the last dose of tamoxifen.

Mouse dissection and perfusion:

Mice were anesthetized using the open drop procedure with isoflurane and humanely euthanized by cervical dislocation. After sacrifice, a 22-gauge needle attached to a valve-controlled gravity perfusion system was inserted into the left ventricle, and 1X phosphate-buffered saline (PBS) was used to gently flush blood from the vasculature.

Mouse aortic root dissociation for scRNA-Seq:

For scRNA-Seq experiments, the aortic root and aortic arch (extending to just after the left subclavian artery) were excised. The tissue was washed three times in ice cold 1X PBS then placed into 1 mL of an enzymatic dissociation cocktail (2 U/ml Liberase (5401127001; Sigma–Aldrich) and 2 U/ml elastase (LS002279; Worthingtonin Hank’s Balanced Salt Solution (HBSS)) and minced with dissecting scissors. After incubation at 37°C for 30 minutes, the cells suspended in the enzyme cocktail were agitated by pipetting to ensure complete dissociation. The mixture was incubated at 37°C for an additional 30 minutes, resulting in a total incubation time of 1 hour. The cell suspension was then passed through a 70 μm strainer and centrifuged at 500 g for 5 min at 4°C to pellet the cells. The enzyme solution supernatant was discarded and cells were resuspended in fresh 1X HBSS and placed on ice until ready for sorting.

FACS of mouse aortic root/aortic arch cells:

Immediately prior to sorting, the cell suspension was filtered through a 35 μm strainer. Cells were sorted on a benchtop SH800S cell sorter under UNC-CH’s Advanced Analytics Core (UNCAAC; supported by the CGIBD center grant, P30 DK034987). Forward/side scatter parameters were used to exclude cell debris. Cells were then gated on forward scatter height versus forward scatter area to exclude doublet events. Cells that passed the aforementioned criteria were gated based upon tdTomato fluorescence and sorted into either a 1.5 mL Eppendorf tube collecting lineage-traced ECs or a separate 1.5 mL Eppendorf collecting cells without detected fluorescence. Cells were then pooled to achieve enrichment for ECs (90% tdTomato-positive, 10% tdTomato-negative). ECs constitute a small population (~1-5%) of cells within the aortic tissue. Therefore, aortas from approximately 10 mice were pooled per 10X capture to obtain sufficient ECs for downstream analysis. A total of 5 captures were performed: 1 baseline capture (wild-type mice approximately 7 weeks of age on chow diet), 2 captures from Twist1-wt mice after 16 weeks of HFD, and 2 captures from Twist1-ECKO mice after 16 weeks of HFD. Each pool included animals of both sexes, and sex was subsequently inferred at the single-cell level using Ddx3y and Xist expression. Cells of both sexes were distributed across all identified clusters (Suppl. Fig. 1A), supporting a sex-agnostic analytical approach for cell state identification.

Single-cell capture and library preparation

All single-cell capture and library preparation was performed at UNC-CH’s Advanced Analytics Core using the 10X Chromium platform according to the manufacturer’s instructions. The Chromium Next GEM Single Cell 3’ Reagent Kit (v3.1 Dual Index) was used for library preparation (PN-1000268; 10X Genomics), and the sequencing platform used was the UNC High-Throughput Sequencing Facility (HTSF) NextSeq 2000 instrument (p3, flow cell). Datasets were later combined for all preceding analyses.

Mouse scRNA Data Analysis

FASTQ files from each experiment were aligned to a reference genome including the tdTomato transgene using CellRanger software version 7.1.0. Filtered matrix files were then processed using Seurat v4.1.1 ⁸. Cells expressing fewer than 500 genes and genes expressed in fewer than five cells were removed from each data set. To identify potential doublets and poor-quality cells, the number of genes, unique molecular identifiers, and percentage of mitochondrial genes were examined. Further analysis of scRNA-Seq data was performed using Slingshot v2.2.1 ⁹, Dynamo v 1.3 ¹⁰ and CellChat v1.0.0 ¹¹.

In total, 24,854 high-quality cells were sequenced across Twist1-wt at baseline, Twist1-wt after 16 weeks of HFD, and Twist1-ECKO after 16 weeks of HFD data sets. Among these cells, 13,262 were identified as tdT+/Cdh5+ endothelial cells. Of these, 2,856 aortic endothelial cells underwent further analysis.

Cells expressing fewer than 500 genes and which had fewer than 5000 molecules detected within a cell were excluded from further analysis. Library size normalization was performed on gene expression values to normalize raw gene counts from each cell relative to the total number of read counts present in that cell. The values were multiplied by 10,000 and log-transformed. Further analysis was performed using the 2000 most highly variable genes. Principal component analysis (PCA) was used on the scaled data for dimensionality reduction, followed by clustering in PCA space using a graph-based clustering approach. UMAP was used to visualize the resulting clusters in two-dimensions. To analyze the endothelial and endothelial-derived cells (ECs), only cells expressing the lineage-marker tdTomato (tdT) above 1.5 were included. Considering that not all sequenced tdT+ cells are truly ECs nor EC-derived, due to potential doublet events, the endothelial marker Cdh5 was used to validate the identity of EC and EC-derived cells. To focus analysis on aortic endothelial cells, an Arterial EC module score was used, comprised of the following genes: Gja5, Fbln5, Hey1, Ltbp4, Gja4, and Sox17 ¹², ¹³. Furthermore, we positively selected ECs expressing Gja5, Cdh5 and tdT (above a threshold of 1.5). We excluded cells based upon expression of Alpl (microvascular marker), and also excluded lymphatic endothelial cells (LECs; Lyve1+Prox1+), interstitial microvascular ECs (Car4+), and endocardial ECs and valve cells (Tmem108+Prox1+) ¹⁴.

Mouse and Human scRNAseq Integration

Two human carotid datasets were used: one dataset contained normal carotid samples derived from organ donors ¹⁵, and the other contained atherosclerotic vascular samples from patients undergoing carotid endarterectomy ¹⁶. ECs were isolated from these datasets using the markers CDH5 and GJA5, and microvascular ECs expressing ALPL were excluded. Human and mouse cells were integrated using the CONCORD algorithm (v1.0.13) ¹⁷. Downstream analyses were performed using Seurat v4.1.1 ¹⁸ and scanPy v1.7.2 ¹⁹.

Mouse aortic root and arch preparation for staining and cryoblocks

Immediately following perfusion with 1X PBS, tissues being prepared for staining underwent a second perfusion with 0.4% ice cold paraformaldehyde (PFA) delivered by a separate valve-controlled gravity perfusion system. The aortic root was isolated separately from the remainder of the aorta, from the ascending aorta to just above the iliac arteries. The aortic root was left embedded in the myocardium of the upper half of the heart for on-axis mounting. Distal arch sections were isolated from the remainder of the arch. The tissues were then submerged in 4% PFA overnight at 4°C, followed by immersion in a sucrose gradient up to 30% sucrose at 4°C. The tissue was embedded in OCT compound and frozen for cryoblock preparation.

En face Oil Red O Staining and Quantification

Oil Red O (ORO) staining solution was prepared following the protocol by Chen et al. ²⁰ by dissolving Oil Red O (A12989; Alfa Aesar) in isopropanol, followed by filtration with a 0.45 μm syringe filter (25-246; GenClone). The staining reagents were prepared fresh and used within 2 hours. Aortic arches were submerged in freshly filtered ORO for 1 hour at room temperature, then placed in 60% isopropanol for 20 minutes at room temperature and rinsed 3 times for 5 minutes in double distilled water. A stereomicroscope was then used to carefully trim away any perivascular adipose tissue and pin the aorta for en face imaging of the ORO stain. All ORO-stained areas were annotated using Fiji ImageJ and ORO area was quantified as the %Area of the vessel.

Cryoblock Sectioning

Blocks were sectioned on a cryostat (Leica CM1950) into 7 μm thick sections, mounted on SuperFrost Plus Microscope Slides (1255015; fisherscientific) and kept at −80°C until slide preparation for staining, RNAscope or IHC.

Fixation and slide mounting for tdTomato signal

Slides were allowed to come to room-temperature and air dried before washing with 1X PBS to remove OCT. Slides were then placed in 95% ethanol for 15 seconds, air-dried, and immersed in cold 4% PFA for 30 seconds for post-fixation. Slides were then washed four times with deionized water before drying and mounting with a medium containing a DAPI stain (H-1200; FisherScientific).

Immunohistochemistry

Mouse aortic root slides were thawed at room temperature and the OCT was removed by two washes of deionized water. Slides were submerged in 4% PFA for 2 min, followed by four washes in deionized water. A liquid-blocking pen was used to encircle section of dried slides, followed by application of Peroxidazed (PX968; Biocare Medical) for 5 min. Sections were washed three times in deionized water before incubated with Rodent Block M reagent (RBM961; Biocare Medical) for 30 min. Sections were washed twice in Tris-buffered saline (TBS; pH 7.4), then incubated overnight at 4°C with an anti-Cd68 rabbit polyclonal antibody (1:300 dilution; ab125212; Abcam), an anti-Cnn1 rabbit monoclonal (1:300 dilution; ab46794; Abcam), or a rabbit monoclonal isotype control (1:400 dilution; ab172730; Abcam). Sections were washed twice with TBS before incubated with Rabbit-on-Rodent HRP Polymer (RMR622; Biocare Medical) for 30 min. Sections were washed twice with TBS and then incubated with the Betazoid DAB chromogen reagents (BDB2004; Biocare Medical) for 4 min. Sections were washed twice with deionized water and air dried before mounting with Vectashield antifade mounting medium with DAPI (H-1200; FisherScientific).

RNAscope:

Slides were washed once in PBS for 5 min to remove OCT, then baked at 60°C for 30 min. Slides were post-fixed by immersion in 4% PFA at 4°C for 15 min, followed by a dehydration series: 50% ethanol 5 min, 70% ethanol 5 min, 100% ethanol 5 min, 100% ethanol 5 min. Slides were air dried, then treated with Hydrogen Peroxide (322335; ACD Bio.) for 10 min. Slides were rinsed once in deionized water, then submerged in boiling 1x Target Retrieval reagent (322000; ACD Bio.) for 5 min. Slides were washed twice in deionized water, then immersed in 100% ethanol and air dried. Sections were encircled with a liquid-blocking pen and incubated with Protease Plus (322331; ACD Bio.) reagent for 30 min at 40°C, then washed twice with deionized water. Sections were incubated with probes against mouse Edn1 (435221; ACD Bio.), Alpl (441161; ACD Bio.), Klk10 (584551; ACD Bio.) or a negative control probe (320751; ACD Bio.) for 2 hours at 40°C. Slides were processed according to the manufacturer’s instructions, and all reagents were obtained from ACD Bio (322373). Sections were mounted with EcoMount (EM897L; Biocare Medical).

BODIPY Staining

After fixation, as previously described, each section was covered with a 35μM solution of BODIPY 493/503 in PBS. The sections were incubated at RT for 30 minutes, protected from light, before washing 3 times with PBS and mounting with VectaMount Express Mounting Medium (H-5700-60; Vector Laboratories).

Masson Trichrome Staining

Fixed slides were stained with Mayer’s Hematoxylin Solution (MHS32-1L; Sigma-Aldrich) for 10 minutes, before washing with ddH2O for 1 minute. Slides were then immersed in a Biebrich Scarlet-Acid Fuchsin solution (ICN1548552 and 857408; VWR) for 5 minutes, before washing with ddH2O. Next, slides were differentiated in a phosphomolybdic-phosphotungstic acid solution (221856 and P4006; Sigma-Aldrich) for 10 minutes before immediately being dipped into an aniline blue (AC22966205; Thermo Scientific) solution. Excess aniline blue was removed by washing slides in ddH2O, followed by differentiation in a 1% acetic acid (A6283; Sigma-Aldrich) solution for 2 minutes. Slides were then washed with ddH2O, followed by 95% EtOH, then 100% EtOH, and then cleared with xylenes for 4 minutes. Slides were then mounted with DPX Slide Mounting Medium (06522; Sigma-Aldrich) and allowed to dry overnight before imaging.

Microscopy

Images were captured with the VS-200 SlideScanner at the UNC Hooker Imaging Core. Images for quantification of tdTomato and DAPI signal were captured with the fluorescent camera at 10X magnification. The normal Z-planes setting was selected for image capture and the following channels at their respective exposure times were used: TRITC 50% (160.00 ms), FITC 25% (75.70 ms), DAPI 50% (33.24 ms). Images for quantification of RNAscope-probed slides (Edn1, etc.) and for quantification of IHC-stained slides (CD68, Cnn1) were performed at 20X. Images were processed and quantified using the FIJI ImageJ software ²¹. Fluorescent images were quantified for tdT and DAPI signal and brightfield images were quantified for RNAscope and IHC samples. Annotations of the lumen, internal elastic lamina (IEL), and external elastic lamina (EEL) were outlined using the polygon selections tool. The endothelial monolayer was annotated as a 10-μm-wide segment ²² adjoining the luminal surface. In each case, the software’s enlarge selection function (Edit>Selection>Enlarge…) was used to a duplicate of the lumen annotation with an input value of 10 μm, based on the image’s scale. To ensure areas annotated by the automated enlarged annotation were not double counted in cases where there was overlap with regions within the IEL, annotations of the inverse area of the IEL and its overlap with the lumen annotation were subtracted for final calculations of area and signal in the EC monolayer. These annotations are deemed “regions of interest (ROIs)” by the software. 2-3 serial sections were analyzed for each aortic root sample for each feature quantified. The average of these serial sections was calculated to determine measurements per sample.

Vessel Feature Areas:

The EC monolayer area was calculated by subtracting the lumen area from the area outlined 10μm within the lumen annotation. Lesion area included the endothelial monolayer and was determined by subtracting the lumen area from the IEL area. Medial area was determined by subtracting the IEL area from the EEL area, and the vessel area was defined by the EEL area. Finally, an additional annotation which combined the inverse of the IEL and the EC monolayer was used to measure areas and signal areas overlapping between the EC monolayer and IEL, to prevent counting areas and signal areas twice. Researchers were blinded to the genotype of the animals until completion of the analysis.

Fluorescent Signal Measurement:

All raw fluorescent images were split by channel and converted to an 8-bit image file. Each channel was processed independently using the Li automated thresholding algorithm, which implements Li’s Minimum Cross Entropy thresholding method ²³ to binarize the image. Annotations were overlaid onto each channel and each image was measured to obtain the area of each ROI and the %Area with signal for each ROI, based on the Li threshold.

Brightfield IHC Signal and RNAscope Signal Quantification:

Following the image processing method outlined by Yu in this paper ²⁴, the background was subtracted from brightfield images of CD68 prior to further processing. To ensure only true IHC signal was included in quantification, color deconvolution was performed using the H DAB vector and only the red Color 2 channel was further processed and quantified. Each image Look Up Table (LUT) was inverted so that the white background would not be determined as signal area upon applying the automated Li threshold. Measurements were recorded as described above. Edn1 RNAscope signal was quantified by hand. Individual puncta were hand-counted using the multi-point tool. Puncta were quantified at the endothelial monolayer, within the lesion, and within the media separately. Edn1 puncta are reported as an absolute count.

Calculations:

The area of detected ‘signal’ over a defined threshold (tdT and Cd68 signal) for each ROI was calculated by multiplying the ‘area fraction’ (percent of ROI area with signal) by the area for the respective region of interest. The percentage of endothelial monolayer, lesion, media, and total vessel area with signal was determined by dividing the signal area over total area for each respective region.

To determine the number of cells expressing a gene of interest (example: tdT), the thresholded channels for each signal channel of interest (ex: DAPI channel and tdT channel) were combined using the (Process > Math > AND…) function to generate a composite image that only showed colocalized signal area of the two channels. Using just the DAPI channel, 5 nuclei were selected at random for each of 3 images to be measured and the average nuclei area was calculated. This calculated average was used as a “standardized nucleus area”; colocalized signal area was then divided by this standardized nucleus area to determine the number of cells expressing a gene of interest.

qPCR of Aorta for Twist1 expression

To validate the expression of Twist1 in the aortic tissue collected from Twist1-wt, Twist1-ECKO, and Twist1-flx mice that did not receive tamoxifen, tissue was harvested and dissociated as previously described for scRNA-Seq sample preparation. For this validation, no FACS was performed, and all cell types were included. Following centrifugation, instead of resuspending cells in 1X HBSS, cells from each mouse were resuspended in 250uL of RNA Lysis Buffer from the Zymo Quick-RNA Microprep Kit (R1050, Zymo). RNA isolation was performed following the kit’s standard protocol. cDNA synthesis was performed using the iScript cDNA Synthesis Kit (1708890, BIORAD). qPCR reactions were performed using the TaqMan Universal MasterMix (4304437, ThermoFisher), B2m VIC expression assay (4448489, ThermoFisher), and Twist1 FAM expression assay (4331182, ThermoFisher) and run on a QuantStudio 6 Flex Real-Time PCR System.

HiFi Cloning of TWIST1 into pWPI

The pWPI plasmid was obtained from Addgene (12254) and prepared using the ZymoPURE II Plasmid Midiprep Kit (D4201; Zymo Research). All plasmids were sequence-validated by GENEWIZ from Azenta Life Sciences. To clone TWIST1 into the pWPI vector, pWPI plasmid was digested with Pme1 (R0560S; New England Biolabs) and 1X rCutSmart Buffer (B6004; New England Biolabs) at 37°C for 1 hour, then at 65°C for 20 minutes. A second incubation with QuickCIP (M0525S; New England Biolabs) in rCutSmart Buffer was performed at 37°C for 10 minutes, then 80°C for 2 minutes. The Zymo DNA Clean & Concentrator-5 Kit (D4003; Zymo Research) was used on the suspended open backbone before continuing with assembly. Human TWIST1 was obtained from Origene (NM_000474 Human Tagged ORF Clone, SKU RC202920; Origene) and cloned into the pWPI vector using NEBuilder High Fidelity (HiFi) DNA Assembly. After transformation into competent bacteria, plasmid DNA was purified with the ZymoPURE II Plasmid Midiprep Kit (D4201; Zymo Research). Primer sequences used for amplification of the TWIST1 fragment with pWPI-compatible ends are as follows:

Forward primer: 5’-TATCGATCACGAGACTAGCCTCGAGGTTTGGAACAAAAGTTGATTTCTGAAGAAGATTT -3;

Reverse primer: 5'-CCGCCGCTTCCGCCGGAGCTGCCGCCTCCGGATCCGTGGGACGCGGACATGG-3'. For assembly of TWIST1 into pWPI, a DNA molar ratio of 2:1 was used for TWIST1 (insert) to pWPI (backbone) with 2X NEBuilder HiFi DNA Assembly Master Mix (E2621L; New England Biolabs). The reaction in addition to an independent open-vector only negative control was incubated at 50°C for 60 minutes, as instructed by the manufacturer’s protocol.

Lentivirus production

All lentivirus production was performed in a sterile cell culture hood. 293T cells (CRL-3216, ATCC: The Global Bioresource Center) were seeded into 100 mm Tissue Culture Dishes (fisherbrand; FB012924) with 10 mL of Dulbecco’s Modified Eagle’s Medium (DMEM) (76176-654; VWR) with L-Glutamine (25-509; Genesee Scientific) and fetal bovine serum (FBS) (F4135; Sigma-Aldrich). At 70% confluence, cells were transfected using the invitrogen Lipofectamine 3000 Transfection Kit (L3000015; Invitrogen) protocol for pWPI and TWIST1-overepression vectors. Transfected cells were incubated at 37°C in a cell culture CO₂ incubator for 6 hours, followed by a media change with 10 mL of warmed DMEM. Plates were incubated at 37°C in a cell culture CO₂ incubator for approximately 62 hours before virus collection. GFP-signal was detected after approximately 24 hours of virus production. Lentivirus-containing media was aspirated with a 10 mL syringe and filtered through a 0.45 μm PES membrane (25-246; GenClone) and stored in 1mL aliquots in cryovials (24-203; Genesee Scientific) at −80°C until ready for experimental use.

HCAEC culture, transduction, and flow model

Human coronary artery endothelial cells (HCAECs) were ordered from Cell Applications (300-05a, Lot 2139) and cultured in Lonza’s EBM-2 medium (CC-3156; Lonza) with EGM-2 MV Microvascular Endothelial Cell Growth Medium-2 BulletKit (CC-3202; Lonza), excluding the gentamicin sulfate.

Once confluent in a T75 flask (25-209; GenClone), HCAECs were seeded for experiments at passage 4 into 6-well flat bottom tissue culture plates (FB012927; fisherbrand,). At 80% confluence, each well of the 6-well plate containing HCAECs was incubated with 1 mL of EGM-2 MV medium, 1 mL of lentivirus, and 2 μL of Polybrene to increase transduction efficiency. For each condition (pWPI and TWIST1-overexpression), 4 biological replicates were prepared. After overnight incubation at 37°C in a cell culture CO₂ incubator, the medium was changed with each well receiving 2mL of fresh EGM-2 MV medium. Cells were incubated for 48 hours to allow for expression from the vectors; GFP-signal became apparent after 24 hours. After 48 hours, the HCAECs reached 100% confluence, and another medium change was performed before placing the cells on an orbital shaker (MS-NOR-30; major science), with 1.5 mL of medium per well. Plates were positioned at the corners of the shaking platform to ensure maximum orbital radius. HCAECs were exposed to 180 rpm for 5 days, and each day the 1.5 mL of EGM-2 MV medium was changed. After 5 days of shaking at 180rpm, HCAECs were harvested for RNA extraction using the Zymo Quick-RNA MicroPrep Kit (11-327M; Zymo Research). For each well, a sterile cell scraper (12-565-58; Fisher Scientific) was used to detach LSS-exposed HCAECs from the periphery of the well and the cells suspended in medium were transferred to an Eppendorf tube and centrifuged at 500 xg for 5 minutes at 4°C. The supernatant was aspirated and discarded, and the pellet of HCAECs was resuspended in 300 μL of RNA Lysis Buffer. After collecting LSS-exposed HCAECs from the periphery of each well, 1 mL of 1X PBS was added to each well to gently wash the well of uncollected detached cells. After aspirating the PBS, 300 μL of RNA Lysis Buffer was added to each well and a scraper was used to detach the OSS-exposed HCAECs at the center of each well. The suspended HCAECs were transferred to Eppendorf tubes and RNA was extracted from all LSS-exposed and OSS-exposed HCAECs following the manufacturer’s protocol, including an on-column DNAse clean-up. RNA was suspended in DNAse-free water and submitted to UNC’s Advanced Analytics Core for quality control and library preparation. RNA quantification and normalization was performed using the DeNovix RNA Fluorescence Assay, and RNA quality control was performed using the Agilent Bioanalyzer RNA Pico Kit. All RNA samples passed quality control with RIN scores above 9. The mRNA-Seq library preparation was performed using the NEBNext Ultra II Directional Kit, library quantification and normalization was performed with the DeNovix DNA Fluorescence Assay, and library quality control was performed using the Agilent Bioanalyzer HS DNA Kit. Nf-SeqMatic Sequencing was performed by Admera.

Bulk RNA-Seq data processing and analysis

FastQ files were processed using the nf-core/rnaseq (version 3.17.0) ²⁵ pipeline ²⁶. Software environments from Bioconda ²⁷ and Biocontainers ²⁸ were used. Nextflow (v24.10.0) was used to execute the pipeline ²⁹. Pre-processing to infer strandedness was performed using fq (v0.12.0) and Salmon (v1.10.3). The reference genome GRCh38.primary_assembly reference genome was used. Reads were trimmed using fastp (v0.23.4) and alignment was performed using STAR (v2.7.10a) and the count matrix data were generated using RSEM (v1.3.1) with a minimum mapped reads value of 5. Count matrix data was processed using DESeq2 (v1.28.0) ³⁰ and expected counts data was used to determine differentially expressed genes between conditions. For all analyses, a pval_adj cut-off of 0.05, and a log2FC cut-off of 1 was applied. DE gene analysis was performed comparing TWIST1-overexpression to pWPI empty vector (EV) under LSS conditions and OSS conditions. QIAGEN’s IPA software ³¹ was used to determine the top predicted pathways and top predicted upstream regulators of DE genes.

RESULTS

Aortic EC population expands during atherosclerosis and contributes to the plaque

To characterize EC phenotypic modulation during atherosclerosis, we generated an EC-specific lineage-tracing mouse model (Cdh5^(CreERT2), ROSA26^(tdT), ApoE^(−/−), or Twist1-wt). We administered tamoxifen to adult mice at 6-7 weeks of age prior to placing the animals on high-fat diet (HFD) for a course of 0, 6, or 16 weeks. At each respective time point, we isolated the aortic root for histological analysis. Fluorescent images of transverse sections of the aortic root were captured and annotated to define the different features of the vessel (Supp. Fig. 1B). In situ analysis of the aortic root after 0, 6, and 16 weeks of HFD identified the presence of lineage-traced ECs at the EC monolayer of the aortic root, interfacing with the lumen. We also identified other ECs located outside of the vessel within the adventitia and myocardium, and lining the aortic valve leaflets (Fig. 1A). We performed quantification of tdT and DAPI colocalization within the vessel (defined as the area between the lumen and the external elastic lamina) at each time point. This tdT-positive cell count revealed a significant increase in the absolute number of tdT-positive cells from 0 weeks to 16 weeks of HFD (Kruskal-Wallis with Dunn’s post-hoc, p = 0.004) and from 6 weeks to 16 weeks of HFD (Kruskal-Wallis with Dunn’s post-hoc, p = 0.047) (Fig. 1B), indicating the expansion of ECs in the aortic root. Within the atherosclerotic lesion (defined as the area between the EC monolayer and the internal elastic lamina), the absolute number of tdT-positive cells significantly increased from 6 to 16 weeks of HFD (two-tailed Mann-Whitney U test p <0.001) (Fig. 1C). tdT-positive cells comprised 20% of all cells within the lesion at 6 weeks of HFD, and 15% of all cells in the lesion at 16 weeks of HFD. We found no significant difference in the number of tdT-positive cells within the 10 μm-thick EC monolayer with disease progression (Supp. Fig. 1C).

FIGURE 1: Identification of aortic luminal ECs and their phenotypic transition during disease.

(A) Transverse sections of the aortic root after 0 (n=6), 6 (n=9) and 16 (n=16) weeks of HFD with nuclei represented in blue, lineage-traced tdT cells represented in red, and tissue features represented by autofluorescence (AF) in green. (B) Quantification of tdT+ cells, determined by tdT and DAPI colocalization count, in the vessel of male mice at each time point. Data were analyzed by Kruskal-Wallis with a Dunn’s post-hoc test for multiple comparisons, and presented as mean ± SEM. (C) Quantification of tdT+ cells within the lesion of male mice at each time point. Data were analyzed by two-tailed Mann-Whitney U test. Each point denotes the number of cells in one mouse. (D) Schematic of lineage-traced Cre-lox system in a murine model of high-fat diet-induced atherosclerosis, outlining the experimental workflow of aortic arch isolation, FACS, and pooling of tdT+/tdT− cells for 10X scRNA-Seq. (E) UMAP of all cells sequenced, grouped by broad cell types. (F) A subset of endothelial cells (ECs) co-expressing lineage marker tdT, endothelial marker Cdh5, and arterial marker Gja5, highlighted in red. Cells are further grouped as Aortic ECs or microvascular (MV) ECs. (G) UMAP of cells collected from the aortic arch after 0 weeks of HFD (orange) and after 16 weeks of HFD (purple) for all cells (left) or for tdT+/Cdh5+/Gja5+, “aortic root ECs” (right). (H) UMAP of the aortic root ECs captured after 16 weeks of HFD, grouped by either quiescent or disease-exclusive ModEC phenotypes. (I) Heatmap of DE gene expression for quiescent (green bar) and ModECs (red bar). (J-L) Expression of the quiescent aortic EC marker Klk10 after 16 weeks of HFD, visualized by (J) Feature Plot in the scRNAseq data or by (K-L) RNAscope in situ hybridization within the distal arch. (M-O) Expression of the ModEC aortic EC marker, Edn1, after 16 weeks of HFD, visualized by FeaturePlot (M) in the scRNAseq data or by RNAscope (N-O) within the distal arch. All scale bars represent a length of 100 μm.

At the 0 week and 16-week timepoints, aortic tissue, extending from the aortic root to the distal edge of the left subclavian artery takeoff, was isolated for enzymatic dissociation and fluorescence activated cell sorting (FACS). We enriched for tdT positive cells by pooling at a ratio of 90% tdT-positive cells to 10% tdT-negative cells, followed by scRNA-Seq capture using the 10X platform (Fig. 1D). scRNA-Seq experiments included a total of 6 mice at 0 weeks of HFD (3 male, 3 female) and 14 mice at 16 weeks of HFD (9 male, 5 female).

scRNA-Seq analysis performed on the cells collected from the aortic root and arch revealed a diverse population, including various subpopulations (sub-clusters) of ECs (Fig. 1E) expressing unique markers (Supp. Fig. 1D-E). As the aortic root is partially embedded in the myocardium and also contains valvular tissue, we used cluster-specific markers to localize these EC populations within the aortic root. Non-aortic EC clusters were identified using previously-described markers ¹⁴ (Supp. Fig. 1F-I) and excluded from further analyses. To positively identify arterial ECs lining the aortic lumen that participate in atherosclerosis in our mouse model, we used an arterial endothelial cell module score (Supp. Fig. 1J), as well as co-expression of the well-established arterial EC marker Gja5 ^32,33, the general EC marker Cdh5 ³⁴, and our lineage marker tdT (Fig. 1F. A sub-cluster of these “triple-positive” cells, characterized by expression of the gene Alpl, was located exclusively in smaller vessels within the myocardium (“Microvascular (MV) ECs”, Fig. 1F, Supp. Fig. 1K-L), so these were also excluded from further analyses. Thus, we focused on the population labeled “Aortic ECs” in Fig. 1F. Visualization of the data grouped by timepoint (0 weeks of HFD versus 16 weeks of HFD) showed that this “Aortic EC” population was the only EC population that contained a disease-specific cell state (Fig. 1G), further confirming that this group indeed represented the EC population participating in disease. Other disease-specific populations within the root included cells within the macrophage and smooth muscle cell (SMC) populations, which have been described previously ³⁵.

Upon isolating the tdT+/Cdh5+/Gja5+/Alpl− aortic ECs in UMAP space, we identified cells present at 0 weeks of HFD and deemed these ‘baseline ECs’. Some cells from the 16-week HFD (disease) timepoint were also present in this cluster, but many cells from the 16-week HFD timepoint formed a distinct ‘disease-exclusive’ cluster comprised only of cells from that timepoint (Fig. 1H). Thus, cells from the 16-week HFD disease timepoint appear to contain cells transiting from a ‘quiescent’ to a phenotypically modulated EC (ModEC) phenotype found only in disease.

Among ECs present at the 16-week HFD disease timepoint, we analyzed differentially-expressed (DE) genes between the phenotypically modulated ECs and the quiescent baseline ECs (Fig. 1I). The heatmap featuring some of the most significant DE genes shows that genes involved in tight junctions (such as Cldn5 ³⁶ and Cgnl1 ³⁷ and polarized epithelial cells (such as Klk8, Klk10, and Mal ³⁸) are highly expressed in the quiescent ECs but are lost in the ModEC clusters (Fig. 1I). In contrast, genes such as Plvap, Bmp6, and Edn1 show increased expression in the ModEC clusters, with genes Ngf and Nrg1 being expressed in the most advanced ModEC cluster. Among the DE genes, we identified kallikrein related peptidase 10, Klk10, as a specific marker of the quiescent EC cluster, as it was not expressed in the ModEC cluster (Fig. 1J) nor any other cell clusters in the scRNAseq data (data not shown). RNAscope in situ hybridization (ISH) of transverse sections of the mouse distal arch after 16 weeks of HFD show the expression of Klk10 at the quiescent EC monolayer (Fig. 1K-L, Suppl. Fig. 1M). Endothelin 1 (Edn1) was identified as a specific marker of ModECs as it was not highly expressed in the quiescent aortic EC cluster (Fig. 1M, Suppl. Fig. 1N), nor any other cell clusters in the data. RNAscope for Edn1 in the distal arch sections after 16 weeks of HFD revealed expression of Edn1 at the endothelial monolayer of the lesion and within the lesion below the monolayer (Fig. 1N-O), confirming its ability to mark ECs undergoing phenotypic modulation, and also suggesting that ECs undergo phenotypic modulation at the monolayer before migrating into the lesion. Nrg1, a marker of advanced EC phenotypic modulation (Supp. Fig. 1O), is also present at the EC monolayer and within the lesion (Supp. Fig. 1P), although it is also present to a lesser extent within the valves and endocardium. The presence of Nrg1 at the monolayer additionally suggests that the degree of EC phenotypic modulation does not necessarily correlate with EC migration into the lesion.

Single-cell analysis of EC phenotypic modulation reveals a single cell fate transition defined by EndMT and inflammatory activation

Analysis of aortic ECs using the Slingshot package predicts a single trajectory beginning with the baseline ECs present at 0 weeks of HFD and continuing through the ModECs present at 16 weeks of HFD (Fig. 2A), and RNA velocity analysis shows a similar cellular transition (Suppl. Fig. 2A). Slingshot pseudotime values along the trajectory are visualized in Fig. 2B. As previous studies have identified various ECs phenotypes during atherosclerosis, we investigated known markers of these phenotypes in our data. To investigate whether ECs adopt an EndMT phenotype, we visualized an ‘EndMT’ Module Score (Fig. 2C), comprised of well-established EndMT markers (Acta2, Lum, Dcn, Fn1, Snail, Mmp2, Tagln, Tgfb1, Tgfb2, Tubb3, Cdh2). Selected components are shown in Fig. 2D-F over Slingshot Pseudotime. The EndMT module score increased in the disease-exclusive ModECs (Fig. 2D, Supp. Fig. 2B), suggesting that these cells are undergoing or have undergone EndMT. We then assessed canonical endothelial markers (Pecam1, Cdh5, Cd43, Cldn5, Klf2 and Vwf), which were visualized both as a composite ‘EC’ Module Score (Fig. 2G) and individually (Fig. 2H-J, Supp. Fig. 2C) over Slingshot pseudotime using tradeSeq ³⁹, and observed an appreciable decrease in these markers. However, some canonical EC markers such as Cdh5 were preserved even at advanced stages of EC phenotypic modulation (Supp. Fig. 2D). Due to the persistence of Cdh5 expression, it is possible that this represents a partial EndMT process.

FIGURE 2: Characterization of EC phenotypic modulation and its predicted regulators during atherosclerosis.

(A) Slingshot trajectory on UMAP of aortic luminal ECs illustrates the predicted path through which ECs undergo phenotypic modulation. (B) UMAP of aortic root ECs colored by Slingshot pseudotime. (C) UMAP of the aortic root ECs colored by an EndMT module score. (D-F) Tradeseq smoothened plots for selected genes included in EndMT module score, including Acta2 (D), Dcn (E), and Cdh2 (F) expression over Slingshot Pseudotime.(G) UMAP of the aortic root ECs colored by an EC module score based on canonical EC markers. Tradeseq smoothened plots for selected genes included in the EC module score, including Pecam1 (H), Cldn5 (I), and Klf2 (J) expression over Slingshot Pseudotime. (K) UMAP of the aortic root ECs colored by a Pro-Inflammatory EC score. (L-N) Tradeseq smoothened plots for selected genes included in Pro-Inflammatory Module Score, including Sele (L), Cxcl12 (M), and Ccl2 (N) expression over Slingshot Pseudotime. (O) Ingenuity Pathway Analysis (IPA) top biological pathways enriched in DE gene list between ModECs and quiescent aortic ECs. The length of the bar represents –log10(p-value) assessed by Fisher’s exact test with Benjamini-Hochberg correction.Color of the bar represents z-score computed from the IPA Ingenuity Knowledge base. (P) Overview of CellChat inferred cell to cell communications in Twist1-wt aortic arch cells, from source quiescent (left) or ModEC (right) aortic ECs to target cell types (middle). The ligands listed in orange represent the predicted secreted ligands, while the ligands listed in purple represent the predicted cell contact ligands.

Because many studies have described ECs acquiring a pro-inflammatory phenotype ⁴⁰ we also constructed a ‘Pro-Inflammatory’ Module Score (Fig. 2K), which included Icam1, Selp, Sele, Cxcl12, Mmp2, and Hif1a. Selected examples are shown in Fig. 2L-N over Slingshot Pseudotime. This analysis revealed that the Pro-Inflammatory module score modestly increases during EC phenotypic modulation in the same cells undergoing EndMT (Fig. 2K, Supp. Fig. 2E). This indicates that the inflammatory cell phenotype is not a distinct alternative EC phenotype, but rather one aspect of EC phenotypic modulation occurring simultaneously with partial EndMT.

In a more recent single cell study, ECs have been described to acquire an immune cell-like phenotype (EndICLT) ^{1, 41.} However, upon generating an EndICLT module score based upon the genes supporting this finding (C1qa, C1qb, Tnf, C5ar1, Lyz2), we did not find evidence of this cell state in our aortic ECs (Supp. Fig. 2F-G). Similarly, we did not detect a significant transition signal for the EndHT phenotype, defined by the markers Kit, Sca1, Cd41, Cd45, Sox7, Sox17, Gata2, Runx1, Notch1, Bmp4, Tek, Cd44 ^1,42 (Supp. Fig. 2H-I). Together, our data indicates that EC phenotypic modulation is best defined by EndMT and upregulation of a pro-inflammatory gene program, occurring together simultaneously in the same cells.

Shifting to unbiased characterization of the gene programs unique to the disease-exclusive ModECs, pathway analysis of the DE gene lists using Ingenuity Pathway Analysis (IPA) ³¹ showed upregulation of integrin cell surface interactions, fibrosis signaling, and extracellular matrix (ECM) organization in ModECs (Fig. 2O). Top predicted upstream regulators were also identified (Supp. Fig. 2J).

Interestingly, we observed distinct patterns of gene expression during EC phenotypic modulation. While some genes, such Acta2, Dcn and Cxcl12 (Fig. 2D,E, M) increased monotonically over pseudotime, others increased early in the ModEC transition and then plateaued (Fig. 2L, Supp. Fig. 2K). In contrast, other genes were only expressed at the most advanced stage of EC phenotypic modulation (Fig. 2F, N, Supp. Fig. 2L). This may suggest that, although the process of EC phenotypic modulation occurs across a continuum, distinct cell states may exist at different stages of phenotypic modulation (i.e. early versus more advanced ModECs).

Cell-cell communication analysis (Fig. 2P) identified key differences between quiescent and ModECs with respect to their predicted signaling to other cell types within the plaque. For example, the Delta-like ligand 1 (Dll1) signaling to multiple cell types, present in quiescent ECs, is abolished in ModECs. In contrast, upregulation of E-selectin (Sele) and several secreted signaling molecules such as Cxcl12 and multiple Pdgf factors are upregulated in ModECs and predicted to interact with multiple cell types in the lesion. These results provide multiple mechanisms by which EC modulation might affect phenotypic changes in other cells during disease.

Comparative scRNA-seq analysis of mouse and human atherosclerotic samples reveals conserved endothelial transition

To determine whether the EC phenotypic changes we observed in the mouse model of atherosclerosis are applicable to human atherosclerosis, we analyzed a scRNAseq dataset from coronary artery samples with no macroscopic atherosclerosis (‘non-diseased’ condition ¹⁵, as well as human scRNA-Seq data from carotid endarterectomy samples as a ‘diseased’ condition ¹⁶. CDH5 was used to broadly identify the clusters of endothelial cells within each data set (Fig. 3A, C). Then, both GJA5 (Fig. 3B, D) and a human arterial EC module score (Supp. Fig. 3A-B, see Methods) were used to identify arterial ECs. The CDH5+/GJA5+ ECs from each human data set were integrated with the mouse Twist1-wt aortic root cells using the CONCORD algorithm ¹⁷ (Fig. 3E).

FIGURE 3: Comparative analysis of mouse and human EC phenotypic modulation.

(A) UMAP of all cells sequenced from non-diseased human samples. CDH5+ ECs are highlighted in green. (B) Subsetted UMAP of CDH5+ human ECs from non-diseased sample, with GJA5+ arterial ECs highlighted in green. (C) UMAP of all ‘diseased’ cells sequenced from human carotid endarterectomy samples. CDH5+ ECs are highlighted in red. (D) Subsetted UMAP of CDH5+ human carotid endarterectomy ECs, with GJA5+ arterial ECs highlighted in red. (E) UMAP of CONCORD-integrated human ‘non-diseased’ (green), human ‘diseased’ (red), and mouse aortic ECs (blue). (F) Feature plot of EDN1 expression in all integrated cells. (G) Feature plot of NRG1 expression in all integrated cells. (H) Feature plot of CLDN5 expression in all integrated cells. (I) Subsetted UMAP of all integrated cells, annotated as ‘Quiescent EC’ or ‘Modulated EC’. (J) Bar plot representing the distribution of human ECs from ‘diseased’ and ‘non-diseased’ conditions in each cluster from (I). (K-L) Violin plot of EDN1 (K) or NRG1 (L) expression, grouped by non-diseased and diseased conditions. For (I-L), proportions are descriptive; no statistical tests were performed. (M-N) Native human UMAP of all diseased ECs, with cells assigned to Quiescent ECs in the joint CONCORD object colored in purple (M) or cells assigned to Modulated ECs colored in orange (N). (O-Q) UMAP of all diseased ECs with (O) EndMT module score, (P) Pro-inflammatory module score or (Q) EC module scores plotted.

In the integrated dataset two markers of modulated ECs, EDN1 and NRG1, are shown in Fig. 3F-G, and a marker of quiescent ECs, CLDN5 is shown in Fig. 3H. Interestingly, these markers roughly correspond to unbiased clustering at low resolution (Fig. 3I) which likely represent quiescent and modulated ECs. Focusing exclusively on the human cells, we first tested whether the proportion of cells in the modulated EC cluster differed between the ‘diseased’ and ‘non-diseased’ samples. This analysis revealed that there was a greater proportion of modulated ECs in the ‘diseased’ samples compared to the ‘non-diseased’ samples (33% vs 15.6%, respectively, Fig. 3J). Furthermore, expression of modulated EC-associated EC markers EDN1 and NRG1 were higher in ECs from the ’diseased’ samples compared to the ‘non-diseased’ samples (Fig. 3K-L). In contrast, we observed decreased expression of the quiescent-associated genes CLDN5 and KLF2 in the ‘diseased’ samples (Supp. Fig. 3C-D).

To determine whether the quiescent and disease-associated human ECs are indeed distinct populations when not co-clustered with the mouse data, we visualized the human EC clusters from the carotid endarterectomy dataset in the context of their native UMAP (Fig. 3M-N). This demonstrated that the quiescent and modulated EC populations we identified do indeed have distinct transcriptional profiles in the human dataset. In the carotid endarterectomy ECs, we next visualized the canonical EC module score based on expression of CDH5, CLDN5, KLF2, VWF, PECAM1, and CD44 (Fig. 3O). We also quantified the Pro-Inflammatory EC module score comprised of ICAM1, SELP, SELE, CXCL12, MMP2, and HIF1A (Fig. 3P) and the EndMT module score comprised of ACTA2, LUM, DCN, FN1, SNAI1, MMP2, TAGLN, TUBB3, TGFBR1, TGFBR2, and CDH2 (Fig. 3Q). Additionally, an EndHT module score comprised of KIT, SOX7, SOX17, GATA2, RUNX1, NOTCH1, BMP4, TEK, CD44 (Supp. Fig. 3E) and an EndICLT module score comprised of C1QA, C1QB, TNF, C5AR1, and LYZ2 (Supp. Fig. 3F) were investigated. Altogether, the data revealed that the EndMT phenotype was most clearly recapitulated in the diseased human ECs (Fig. 3O). The pro-inflammatory phenotype was also present, albeit to a lesser extent, in a subset of modulated ECs (Fig. 3P). There was also a downregulation of canonical endothelial markers (Fig. 3Q) in modulated ECs. However, we did not observe an EndHT or EndICLT transition in the human data. To determine the pathways and regulators involved in EC phenotypic modulation in human disease, we performed DE gene analysis comparing the quiescent cluster (Fig. 3M) to the modulated EC cluster (Fig. 3N). We used IPA to identify the top predicted pathways (Supp. Fig. 3G), which were enriched with processes related to extracellular matrix production and consistent primarily with an EndMT phenotype. The top predicted upstream regulators of EC phenotypic modulation in human arterial ECs are shown in Supp. Fig. 3H. Interestingly, some of these factors in the human data were also predicted to be drivers of EC phenotypic modulation in the mouse data (Supp. Fig. 2I), such as the TEAD family of transcription factors, the MRTFA and MRTFB transcription factors. This is consistent again with a mesenchymal transition, and strongly implicates these factors in driving EndMT during atherosclerosis.

Twist1 deletion results in decreased EC phenotypic modulation

To study the effect of Twist1 deletion on EC phenotypic modulation and lesion characteristics during atherosclerosis, we crossed a conditional Twist1 allele ⁷ with the Twist1-wt line to generate an inducible Twist1 knockout line (Cdh5^(CreERT2), Twist1^(flx/flx), ROSA26^(tdT), ApoE^(−/−)) or Twist1-ECKO (Fig. 4A). All experimental animals used in the study were age matched when administered Tamoxifen. The 16-week HFD treatment was also consistent for all cohorts included in the comparison of phenotypic differences between genotypes. We isolated aortic arches from Twist1-wt and Twist1-ECKO male and female animals from above the aortic root until the level of the diaphragm and performed en face Oil Red O staining (Fig. 4B). Quantification of Oil Red O percent area (male n=12 Twist1-wt, n=11 Twist1-ECKO; female n=9 Twist1-wt, n=10 Twist1-ECKO) revealed that Twist1-ECKO mice have significantly reduced lesion area in the aortic arch (Welch’s test male p < 0.001, female 0.0079, Fig. 4C). Although lesion area was reduced in the whole aorta in males (Mann-Whitney test p = 0.037, Fig. 4D), this was driven by the difference in the aortic arch, as we found no significant difference in lesion area in the descending aorta in either sex (Supp. Fig. 4A) (male Welch’s test p = 0.88, female 0.57). Similarly, in transverse sections of the aortic root (male n=16 Twist1-wt, n=19 Twist1-ECKO; female n=11 Twist1-wt, n=13 Twist1-ECKO), there was no significant difference in lesion area between genotypes (Welch’s test male p = 0.32, female 0.41, Supp. Fig. 4B).

FIGURE 4: EC-Twist1 deletion inhibits EC phenotypic modulation and reduces lesion burden.

(A) Schematic of lineage-traced Cre-lox system for endothelial Twist1 deletion. (B) Representative en face Oil Red O-stained aortae of Twist1-wt (left) and Twist1-ECKO (right) male mice after 16 weeks of HFD. The aortic arch area quantified is outlined with the white dashed line. All scale bars represent a length of 2 mm. (C) Quantification of lesion area based upon Oil Red O-staining in the aortic arch and (D) whole aorta, grouped by genotype and separated by sex (male n=12 Twist1-wt, n=11 Twist1-ECKO; female n=9 Twist1-wt, n=10 Twist1-ECKO). Data in (C) were analyzed with a two-tailed Welch’s t-test and data in (D) were analyzed with a two-tailed Mann-Whitney U test. (E) Aortic root ECs with Twist1-wt cells displayed in purple and Twist1-ECKO cells displayed in green. (F) UMAP of Twist1-wt and Twist1-ECKO aortic root ECs plotted over Slingshot Pseudotime. (G) Distribution of Twist1-wt (purple) and Twist1-ECKO (green) ECs over Slingshot Pseudotime. Data were analyzed with a two-sample Kolmogorov-Smirnov test. (H) Feature plot of aortic Edn1 expression. (I) RNAscope ISH for Edn1, in transverse sections of the aortic root (dashed black line). Representative sections of Twist1-wt (top) and Twist1-ECKO (bottom) are shown. All scale bars represent a length of 250 μm. (J-K) Quantification of Edn1 RNAscope staining in the EC monolayer (J) or below the EC monolayer (K) of mice after 16 weeks of HFD, grouped by genotype and separated by sex (male n=17 Twist1-wt, n=16 Twist1-ECKO; female n=9 Twist1-wt, n=7 Twist1-ECKO). Data in (J-K) were analyzed with a two-tailed Mann-Whitney U test. All data are presented as mean ± SEM.

TdT-positive ECs were also quantified in transverse sections of the aortic root (Supp. Fig. 4C, annotated as in Supp. Fig. 1A). There was no significant difference in the number of tdT-positive cells at the EC monolayer (Welch’s test male p = 0.24, female 0.67, Supp. Fig. 4D) (male n=16 Twist1-wt, Twist1-ECKO n=19; female n=11 Twist1-wt, Twist1-ECKO n=13). However, male Twist1-ECKO mice contained significantly fewer tdT-positive cells in the lesion below the EC monolayer (Welch’s test p = 0.003, Supp. Fig. 4E, n=16 Twist1-wt, Twist1-ECKO n=18) and a slight decrease in tdT-positive ECs in the vessel as a whole (Mann-Whitney p = 0.036, Supp. Fig. 4F). This indicates a reduction in EC phenotypic modulation and/or reduced proliferation/migration of phenotypically modulated ECs into the lesion.

To characterize the effect of Twist1 deletion in ECs, we performed scRNA-Seq on a total of 19 Twist1-ECKO animals (8 males, 11 females) after 16 weeks of HFD and the data were analyzed jointly with the Twist1-wt cells. Notably, among the ECs collected after 16 weeks of HFD, Twist1 expression was identified exclusively in the aortic EC and valve EC populations (Supp. Fig. 4G). Endothelial deletion of Twist1 was validated in the scRNA-Seq data, showing an approximately 95% decrease in Twist1 expression in the Twist1-ECKO ECs (Supp. Fig. 4H). To determine how EC Twist1 deletion affects EC phenotype during disease, the distributions of WT and ECKO cells were visualized in UMAP space (Fig. 4E) and across EC phenotypic modulation pseudotime (Fig. 4F-G). This revealed a bi-modal distribution of cells, with quiescent cells at earlier pseudotimes and modulated ECs at later pseudotimes. In contrast to WT cells, which exhibited a greater density at later pseudotimes, the density of ECKO cells was greater at earlier pseudotimes (Fig. 4G). Thus, Twist1 deletion shifted ECs earlier in pseudotime, resulting in significantly less EC phenotypic modulation during disease (p=0.002, two-sample Kolmogorov-Smirnov test). In addition, among modulated ECs Twist1 deletion appeared to disproportionally reduce modulated ECs at earlier pseudotimes, with no differences observed at the most advanced stages of phenotypic modulation Fig. 4G).

To validate the scRNA-Seq data, we performed ISH of transverse aortic root sections to quantify Edn1, a highly-specific ModEC marker (Fig. 4H) in the vessel (Fig. 4I). Quantification of expression (male n=17 Twist1-wt, n=16 Twist1-ECKO; female n=9 Twist1-wt, n=7 Twist1-ECKO) revealed that Edn1 is significantly reduced at both the EC monolayer (Mann-Whitney test male p = 0.001, female p < 0.001, Fig. 4J) and in the lesion below the monolayer (Mann-Whitney test male p < 0.001, female p=0.009, Fig. 4K) of Twist1-ECKO mice relative to Twist1-wt controls. Together, these data indicate that Twist1 deletion in ECs indeed results in reduced EC phenotypic modulation, and that this reduction in EC phenotypic modulation occurs independently of migration into the lesion.

Surprisingly, our scRNA-Seq data revealed that in addition to reduced expression in ECs (reduced by approx. 95%), Twist1-ECKO mice also exhibited reduced Twist1 expression in smooth muscle cells (SMCs; approx. 83%) and adventitial fibroblasts (AdvFibroblasts; approx. 33%) (Supp. Fig 4I). Twist1 was virtually undetectable in macrophages in both genotypes (data not shown). To determine whether this was because Twist1 was being deleted from the genome, we visualized the scRNAseq reads in each cell type in relation to the 3’ loxP site (Supp. Fig. 4J). In ECs from Twist1-wt mice, scRNAseq reads extended past the 3’ loxP site to the end of Exon 2, indicating that the Twist1 gene had not undergone excision from the genome. In ECs from Twist1-ECKO mice, the few existing reads terminated prior to the 3’ loxP site, indicating effective deletion of the Twist1 locus from the genome. In contrast, in SMCs and adventitial fibroblasts, although the Twist1 transcript levels were significantly reduced in Twist1-ECKO mice, the Twist1 transcript reads extended past the 3’ loxP site. This indicates that the Twist1 locus remained intact within the genome, and suggests that Twist1 expression is reduced in these mice by a mechanism other than Cre-mediated Twist1 deletion. To confirm that reduced expression of Twist1 in non-EC cell types was an intrinsic property of the Twist^flx allele and not a Cre-mediated effect, we isolated RNA from aortic tissue extending from the aortic root through the distal arch from adult mice (Twist-wt, Twist1-ECKO, and Twist1-flx genotypes) that did not receive any tamoxifen (Tmx) and assayed Twist1 expression by qPCR. We found that the presence of the floxed Twist1 allele alone significantly reduces Twist1 expression in the absence of Tmx (Twist1-ECKO vs Twist1-wt one-way ANOVA, Tukey’s posthoc test p = 0.006) and in the absence of both Tmx and Cre (Twist1-ECKO vs Twist1-flx, one-way ANOVA, Tukey’s posthoc p = 0.004, Supp. Fig 4K).

Endothelial-specific Twist1 deletion results in reduced lesion size and increased lesion stability in the distal arch

While our original model comparing Twist1-wt and Twist1-ECKO was necessary for lineage tracing of EC-derived cells, this led to reduced Twist1 expression in multiple cell types and confounded investigation of EC-specific effects of Twist1 on lesion phenotype. We therefore generated a cohort of mice that were homozygous for the Twist1^flx allele, with littermates either inheriting the Cdh5-Cre (Twist1-ECKO) or no Cre (Twist1-flx, Fig 5A). Because our Oil Red O data showed a significant reduction in lesion size within the aortic arch, we characterized lesions within the distal arch adjacent to the left subclavian artery.

FIGURE 5: EC-Twist1 deletion reduces lesion size at the distal arch.

(A) Schematic of Twist1-floxed genetic model, without and with Cdh5-Cre. (B) Representative images of Masson’s Trichrome staining (right) and location of sections in the distal arch of the aorta (lower left, black dashed line). Scale bars represent a length of 250 μm. Quantification in male mice of (C) distal arch lesion area (n= 9 Twist1-flx, n= 10 Twist1-ECKO), (D) acellular area as a percentage of lesion area (n= 9 Twist1-flx, n= 11 Twist1-ECKO), and (E) lesion collagen content as a percentage of lesion area (n= 10 Twist1-flx, n= 14 Twist1-ECKO). (F) IHC staining for Cd68 to determine macrophage content. (G) Quantification of Cd68-positive area as a percentage of lesion area, indicating macrophage content (n= 11 Twist1-flx, n= 17 Twist1-ECKO). (H) IHC staining for Cnn1-positive smooth muscle cells. (I) Quantification of Cnn1-positive lesion area as a percentage of lesion area (n= 11 Twist1-flx, n= 17 Twist1-ECKO). (J) RNAscope was performed to probe for Edn1 in the distal arch. (K) Quantification of Edn1-positive cells at the monolayer (n= 7 Twist1-flx, n= 11 Twist1-ECKO) and (L) within the lesion below the monolayer (n= 7 Twist1-flx, n= 12 Twist1-ECKO). Data in (C-K) were analyzed with a two-tailed Welch’s t-test, and data in (L) were analyzed with a two-tailed Mann-Whitney U test. All data are presented as mean ± SEM.

Transverse sections of the distal arch were prepared with Masson’s Trichrome staining (Fig. 5B). We observed a significant reduction in lesion area in Twist1-ECKO mice compared to Twist1-flx controls (males Twist1-flx n=9, n=10 Twist1-ECKO, Welch’s p = 0.048, Fig. 5C). Importantly, lesions from Twist1-ECKO mice had significantly reduced acellular area as a percentage of the lesion size (Welch’s p = 0.005, Fig. 5D), and significantly increased collagen content within the lesion (Welch’s p = 0.047, Fig. 5E) compared to Twist1-flx controls. There was no significant difference in media area (Supp. Fig. 5A) nor vessel area (Supp. Fig. 5B) between genotypes.

IHC staining for Cd68-positive macrophages (Fig. 5F) revealed a significant reduction in Twist1-ECKO mice (males Twist1-flx n=11, Twist1-ECKO n=17, Welch’s p = 0.048, Fig. 5G). IHC staining was also performed for the SMC marker Cnn1 (Fig. 5H), which revealed a significant increase in SMCs content per lesion area in Twist1-ECKO mice (Welch’s p = 0.001, Fig. 5I). RNAscope ISH was then performed to detect Edn1 expression in the distal arch (Fig. 5J). Quantification of Edn1-positive cells revealed a significant decrease at the monolayer (Twist1-flx n =11, Twist1-ECKO n = 17 Welch’s p = 0.020, Fig. 5K) and within the lesion below the monolayer (Mann-Whitney p=0.017, Fig 5L) in Twist1-ECKO animals. Sections were also stained with BODIPY 493/503 to assess neutral lipid content in the lesion (Supp. Fig. 5C), which was found to be no different between genotypes (Supp. Fig 5D).

Taken together, these data strongly support that Twist1 deletion inhibits endothelial phenotypic modulation in the lesion and at the distal arch, and that this results in smaller lesions with a more stable lesion phenotype. However, because the reduction in EC modulation mediated by Twist1 deletion is not atheroprotective in all regions of the aorta, other interacting factors such as flow conditions are likely necessary for Twist1 to affect lesion size and stability.

TWIST1 regulates downstream targets in ECs depending on flow conditions

To gain additional insight into the molecular mechanisms by which TWIST1 alters EC biology, we transduced human coronary artery endothelial cells (HCAECs) with a TWIST1 overexpression vector or an empty vector (EV) control and placed cells on an orbital shaker for 5 days (Fig. 6A). Cells at the center of each well, exposed to oscillatory shear stress (OSS) disturbed-flow conditions, were isolated for RNA extraction and prepared for bulk RNA Sequencing. DE genes induced by TWIST1 overexpression were analyzed with IPA for top predicted pathways, which included Agranulocyte Adhesion & Diapedesis, RHOGDI Signaling ^, and Pathogen Induced Cytokine Storm Signaling Pathway (Fig. 6B). Additionally, TWIST1 was identified as the top predicted upregulated upstream regulator, along with other factors with previously-described roles in endothelial biology and pathobiology ^{43,44,45,46,47,48,49,50,51,52} (Fig. 6C).

FIGURE 6: Identification of Twist1-specific mechanisms during flow and disease states.

(A) Schematic of 6-well plate orbital shaker flow model, with oscillatory shear stress (OSS)-exposed HCAECs at the center of each well. Cells were transduced with lentivirus containing either an empty vector (EV) or a TWIST1-overexpression construct. (B) Ingenuity Pathway Analysis (IPA) top biological pathways enriched in DE gene list between HCAECs with TWIST1 overexpression compared to EV-treated cells, under OSS conditions. Grey bars indicate no z-score value predicted. (C) IPA top predicted upstream regulators of OSS-exposed HCAECs treated with a TWIST1 overexpression vector as compared to the EV control. For (B-C), length of the bar represents –log10(p-value) assessed by Fisher’s exact test with Benjamini-Hochberg correction. Color of the bar represents z-score computed from the IPA Ingenuity Knowledge base. (D) Schematic of analysis performed to identify the subset of ‘co-occurrent’ genes involved in EC phenotypic modulation that are regulated by TWIST1. (E) log2FC values of selected co-occurrent genes SELE, CXCL12, SEMA6A, FGFR3 and KLK10 in the HCAEC bulk RNA-Seq data. (F) Feature plots of the same co-occurrent genes in the luminal aortic ECs of the Twist1-wt mouse scRNA-Seq data after 16 weeks of HFD. (G) Violin plots of the co-occurrent gene expression levels in Twist1-wt and Twist1-ECKO mouse scRNA-Seq data after 16 weeks of HFD. (H) Predicted cell-cell signaling pairs in Twist1-wt mice after 16 weeks of HFD for the selected co-occurrent genes. Klk10 was not present in the cell-cell communication database.

In addition to investigating the effects of TWIST1-overexpression under pro-atherogenic OSS conditions ⁵³, transduced HCAECs were collected from the periphery of each well, as these cells were exposed to athero-protective laminar shear stress (LSS) ⁵⁴ (Supp. Fig. 6A). In contrast to the OSS-exposed transduced HCAECs, the top predicted pathways for LSS-exposed cells did not include upregulated pathways with clear links to atherosclerosis (Supp. Fig. 6B). Instead, upstream regulator analysis predicted suppression of inflammatory factors such as TNF, NFKB1, NONO, and IRF3 and IRF1 with TWIST1 overexpression (Supp. Fig. 6C).

When comparing DE genes with TWIST1-overexpression to EV controls separately according to LSS or OSS conditions, we identified 303 DE genes specific to the LSS condition, 229 DE genes specific to the OSS condition, and 402 DE genes shared between both flow conditions (Supp. Fig. 6D). Analysis of top predicted pathways exclusive to either flow condition revealed downregulated inflammatory pathways in the LSS-only genes (Supp. Fig. 6E), while atherosclerosis-relevant and extracellular matrix pathways were retained in the OSS-only genes (Supp. Fig. 6F). Together, our data strongly suggests that the effect of TWIST1 varies significantly with flow conditions in ECs, and that disturbed flow conditions are important for TWIST1 to regulate atherosclerosis-relevant pathways in ECs.

TWIST1 regulates a subset of genes involved in EC phenotypic modulation

To more clearly identify which genes in vivo in EC modulation are specifically regulated by TWIST1, we queried both the mouse aortic EC scRNA-Seq data from the 16-week HFD timepoint and the TWIST1 overexpression HCAEC Bulk RNA-Seq data (Fig. 6D). Because TWIST1 promotes EC modulation, we sought to identify two groups of genes: i) genes that were both activated by TWIST1 overexpression in HAECs and activated during EC modulation in disease; and ii) genes that were both suppressed by TWIST1 overexpression in HAECs and suppressed during EC modulation in disease. Together, these lists comprised 26 genes that TWIST1 may regulate during EC phenotypic modulation (Supp. Table 1). Several upregulated genes have clear functions related to EC phenotypic modulation, including TGF-β and FGF signaling and immune cell chemotaxis/adherence. Genes related to semaphorin-plexin signaling were also coordinately downregulated. Selected genes are displayed in Fig. 6E-F. When stratified by mouse genotype, genes that were upregulated by TWIST1 overexpression in HCAECs were downregulated in Twist1-ECKO mice, and those downregulated by TWIST1 overexpression in HCAECs were upregulated in Twist1-ECKO mice (Fig. 6G, Supp. Table 1). As these genes are regulated in opposite directions with TWIST1 overexpression versus TWIST1 deletion, they represent a higher-confidence TWIST1 regulatory program in ECs during atherosclerosis.

To determine how these genes might affect the multicellular plaque environment, we identified potential secreted and cell-cell contact communications using CellChat in the larger scRNA-Seq dataset containing all cell types (Fig. 6H). Fgfr3 was predicted to only receive Fgf1 signaling on quiescent aortic ECs, which is consistent with observations from previous studies showing that FGF signaling promotes vascular resisilience and prevents EndMT ⁵⁵. In contrast, C-X-C Motif Chemokine Ligand 12 (Cxcl12) and E-selectin (Sele) are predicted to signal from ModECs to other cell types in the plaque, particularly macrophages. Indeed, the induction of CXCL12 and E-selectin by Twist1 in ModECs is a potential molecular mechanism underlying our observation that fewer macrophages are present in the distal arch lesions of Twist1-ECKO mice.

DISCUSSION

The response of ECs to atherogenic stimuli has been studied extensively in vitro, leading to fundamental insights regarding the potential role these cells during disease. The study of EC-specific processes in vivo has been more challenging due to the multi-cellular composition of the atherosclerotic plaque. This has improved with the use of lineage tracing, but the traditional assessment of only a small number of markers has been a fundamental barrier to the comprehensive assessment of EC phenotype. More recently, with the ability to assay thousands of genes simultaneously in individual cells, single-cell ‘omics’ have the promise both to reveal new aspects of EC biology and to place previous observations in context. However, as most current techniques require dissociation of the tissue, spatial information is lost. As a result, ‘bystander’ ECs residing in the surrounding structures, including the vascular adventitia, valve apparatus, or adjacent tissues, can be misinterpreted as EC heterogeneity at the vessel lumen. This can result in erroneous conclusions regarding ECs transitioning between clusters, when in reality, the clusters represent unrelated ECs in different locations. In the present study, we leveraged i) established markers of arterial EC identity; ii) our previous work localizing EC subtypes within the aortic root ¹⁴; and iii) additional EC subtype localization, to identify the EC population at the luminal surface that directly participates in the atherosclerotic lesion. Importantly, this was the only EC population to undergo a significant phenotypic shift between baseline and 16 weeks of HFD, reinforcing this conclusion.

We found that these cells appear to undergo a single cell fate transition, which encompasses both ‘pro-inflammatory’ and ‘EndMT’ phenotypes. Thus, rather than being two alternative phenotypes, these attributes occur simultaneously in the same cells during EC phenotypic modulation. Notably, we did not find evidence of ECs undergoing transition to immune-like (EndICLT) or hematopoietic-like (EndHT) cells in our model system. This contrasts with a recent report ¹ that used scRNAseq and scATACseq in a carotid model of disturbed flow and atherosclerosis. This discrepancy is likely due to the different models used between the studies. Overall, it is vital that only ECs that are located at the monolayer overlying the lesion or within the lesion are included in the analysis. Inclusion of ‘bystander’ ECs in other anatomical locations (valvular ECs, adventitial microvascular ECs, lymphatic ECs) in cell fate trajectory analyses can interfere with interpretations of such cellular transitions. More broadly, we did not identify multiple discrete EC fate trajectories in mouse or human scRNAseq data, but this possibility is difficult to rule out given current technologies. Our analysis also revealed evidence of distinct pseudo-temporal gene expression patterns (Fig. 2D-F, H-J, K-M, Supp. Fig. 2L-N), which may indicate that a distinct phenotype exists at later pseudotime values within the ModEC population. Interestingly, ECs that demonstrated the most advanced degrees of phenotypic modulation (later pseudotime values, and marked by expression of Nrg1) were found at the monolayer as well as within the lesion, suggesting that EC migration may be uncoupled from de-differentiation and acquisition of a mesenchymal/inflammatory phenotype. In contrast, it is likely that proliferation and migration are tightly coupled during EC phenotypic modulation, as we only observed increased numbers of ECs where cells had migrated into the lesion, and not at the monolayer. Future studies using spatial transcriptomics may be able to identify factors common to ECs that proliferate and migrate into the lesion versus those that remain at the monolayer. Importantly, the human data analyzed in our study mirrors many of the same phenotypic changes we observed in lineage-traced mouse ECs, primarily with respect to the EndMT phenotype. The human ECs showed upregulation of genes in the ‘Pro-inflammatory’ module score to a lesser extent than the mouse ECs. It is possible that an initial inflammatory EC phenotype is reduced over time, with the EndMT phenotype predominating at the time of human tissue collection.

The CAD risk allele at 7p21.1 is associated with increased TWIST1 expression in vascular tissue in the GTEx dataset, indicating that TWIST1 expression promotes lesion formation and risk for myocardial infarction. This is directionally consistent with our current findings that EC-specific Twist1 deletion results in reduced lesion burden and a more stable plaque phenotype. Our findings contrast with a recent study ⁵ reporting a lower lesion burden but a more stable lesion phenotype with Twist1 deletion in ECs. This discrepancy could be due to differences in experimental design. For example, the investigators induced atherosclerosis by giving HFD for 8 weeks prior to deleting Twist1 in ECs, followed by another 8 weeks of HFD, whereas we deleted Twist1 prior to the onset of atherosclerosis. Our study may be more consistent with human genetic data because it more closely mimics the long-term difference in TWIST1 expression mediated by an inherited allele. In contrast, it is possible that reduction in EC Twist1 has different consequences when induced later during disease, which may have implications for potential therapies targeting TWIST1. Another major difference is our discovery that the Twist1^flx allele is hypomorphic, resulting in reduced basal Twist1 expression in multiple cell types even without Cre-mediated excision of the Twist1 gene. Because plaque composition is influenced by multiple cell types, altered Twist1 expression in multiple cell types fundamentally confounds any interpretation of the role of Twist1 specifically in ECs. We therefore generated a second cohort of mice (Twist1^flx/flx, +/− Cre) which allowed us to cleanly determine the effect of EC-specific Twist1 deletion on plaque architecture. Thus, our study also differs in the genetic composition of mice studied.

Although the hypomorphic Twist1^flx allele confounds interpretation of plaque phenotype in our first (lineage tracing) cohort of mice, cell-autonomous effects of Twist1 in ECs, such as the effect of Twist1 on EC phenotypic modulation, should be preserved. Indeed, we observed that deletion of Twist1 in ECs in both cohorts of mice markedly reduced EC phenotypic modulation, as evidenced by reduced Edn1 expression at the EC monolayer and within the plaque. In addition to promoting EC phenotypic modulation, it is also likely that Twist1 affects both EC proliferation and migration based upon the changes in EC numbers we observed in the lesion and monolayer in our first lineage tracing cohort of mice (Twist1-ECKO versus Twist1-wt).

Integration of bulk RNA-seq data of OSS-exposed, TWIST1-overexpressing HCAECs with the scRNA-seq data of mouse aortic ECs provided more insight into the regulatory effects of TWIST1. As examples, our data collectively show that TWIST1 upregulates CXCL12 and SELE. Importantly, CXCL12 is the causal gene at the10q11 CAD risk locus ⁵⁶, and a recent study identified ECs as the relevant source of CXCL12 during atherosclerosis with EC-specific CXCL12 deletion resulting in decreased CXCL12 plasma levels and reduced lesion area in the aorta ⁵⁷. Our study directly links these two important CAD GWAS signals, suggesting that this is a critical nexus of CAD risk mediated through ECs. Another TWIST1 target, E-selectin (SELE), is synthesized during inflammation by activated ECs ⁵⁸, ⁵⁹ and found on the surface of fibrous and lipid-containing plaques ⁶⁰, ⁵⁸ where it likely mediates monocyte adhesion during atherosclerosis. Notably, when comparing OSS versus LSS conditions, SELE was differentially regulated only in the setting of TWIST1 overexpression, and not in cells transduced with empty vector. This suggests that both TWIST1 and disturbed flow conditions are required for SELE upregulation, which may provide a mechanism underlying the finding that Twist1 deletion reduces disease only in specific regions of the aorta.

In contrast, our data collectively show that TWIST1 downregulates KLK10 and FGFR3 expression. With regard to FGFR3, loss of fibroblast growth factor (FGF) signaling has been reported to promote EndMT, promote deposition of fibronectin, and increase neointima formation in a mouse model of atherosclerosis ⁶¹, possibly via upregulation of TGF-β signaling ⁵⁵, ⁶².

Together, our data demonstrates that TWIST1 affects risk for atherosclerosis via ECs by promoting proliferation, migration and phenotypic modulation, ultimately leading to an unstable plaque phenotype. Importantly, we have also identified specific mechanisms in ECs by which TWIST1 likely promotes lesion progression. However, Twist1-induced EC phenotypic modulation does not appear to be sufficient to increase lesion burden, as we did not observe changes in lesion size in the aortic root nor the descending aorta. Rather, as demonstrated by the change in disease burden and lesion stability in the aortic arch with EC Twist1 deletion, and by the significant difference in gene expression programs between TWIST1-overexpressing cells in OSS versus LSS conditions, it is likely that the effect of TWIST1 on disease progression depends strongly on the local flow conditions within the vessel. Finally, although we have provided evidence for a causal role for TWIST1 in ECs, it will be important to determine whether TWIST1 is relevant in other cell types that may modify lesion size and architecture, and the resulting risk for vascular disease.

Supplementary Material

Online Figures S1-S6

Online Table S1

Major Resources Table

CLINICAL IMPLICATIONS.

Cardiovascular disease remains the leading cause of morbidity and mortality worldwide. While current therapies primarily target risk factors, a deeper understanding of the biological processes driving atherosclerotic plaque formation is needed to develop more precise treatments. Endothelial cells line the inner surface of blood vessels, playing a central role in maintaining vascular homeostasis, and are among the first cells to respond to disease-promoting stimuli. Their activation initiates inflammatory signaling and phenotypic changes within the vessel wall that contribute to plaque development and plaque instability.

We find that TWIST1, a major genetic risk factor for several vascular diseases, promotes inflammatory endothelial-to-mesenchymal transition (EndMT), a process in which normally protective endothelial cells adopt a more inflammatory and fibrotic disease-promoting state. We also demonstrate that these mechanisms are conserved between mouse models and human vascular tissue, supporting the translational relevance of our findings. Deletion of Twist1 leads to smaller plaques with fewer inflammatory cells and an overall more stable plaque phenotype. Importantly, we also identify key molecular signatures driven by TWIST1 during disease.

Together, these insights identify potential therapeutic targets aimed at preventing or reversing detrimental endothelial cell transitions, offering new strategies to limit the development and progression of cardiovascular disease.

NON-STANDARD ABBREVIATIONS AND ACRONYMS

DEG

Differentialy expressed gene

EC

Endothelial cell

EEL

External elastic lamina

EndHT

Endothelial-to-hematopoeitic transition

EndICLT

Endothelial-to-immune cell-like transition

EndMT

Endothelial-to-mesenchymal transition

EV

Empty vector control

FACS

Fluorescence activated cell sorting

HCAECs

Human coronary artery endothelial cells

HFD

High fat diet

IEL

Internal elastic lamina

LSS

Laminar shear stress

ORO

Oil Red O staining

OSS

Oscillatory shear stress

TdT

Tandem dimer Tomato

Tmx

Tamoxifen