ERG preserves endothelial identity to limit atherosclerosis

Abstract

Endothelial function is a safeguard against atherosclerosis, yet endothelial‑to‑mesenchymal transition (EndMT) can occur at atheroprone sites. How endothelial identity loss contributes to atherogenesis remains unclear. Here we investigated the role of the ETS transcription factor ERG, a key regulator of endothelial identity, in atheroprogression. Inducible endothelial Erg deletion increased plaque burden in hypercholesterolemic mice. Lineage tracing and single‑cell transcriptomics showed that ERG loss drove endothelial dedifferentiation, mesenchymal fate acquisition, and migration and expansion of EndMT cells within plaques, which interacted with pro‑atherogenic cells. ERG loss also caused early disruption of barrier function, promoting lipid uptake and foam cell accumulation in normally atheroresistant regions. In human atherosclerotic plaques, ERG chromatin accessibility and expression were reduced, with transcriptional changes mirroring those in mice. Restoring ERG in cultured EndMT cells reversed mesenchymal programs and re‑established endothelial identity. These findings identify ERG as an enforcer of endothelial function that restricts EndMT to limit atherosclerosis.

Endothelial cell function protects against atherosclerosis, but how loss of endothelial identity contributes to disease progression remains unclear. Here, the authors show that reduced ERG drives endothelial dysfunction and EndMT, accelerating plaque growth, while ERG restoration reverses these changes.

Introduction

Endothelial cells (ECs) comprise the inner lining of all blood vessels and perform critical gatekeeper functions that limit atherosclerotic disease^1–3. In atheroprone regions of the arterial vasculature, disturbed blood flow disrupts endothelial homeostasis, leading to EC activation, lipid accumulation, and monocyte recruitment in developing plaques^1–3. At the cellular level, atherosclerosis involves dynamic changes in EC state, including endothelial-to-mesenchymal transition (EndMT), during which ECs lose their typical identity and adopt pro-migratory and fibroblast-like features^4,5. While EndMT has been associated with atherosclerotic lesions, where ECs can migrate from the lumen to the plaque interior⁴, how this transition occurs across atheroprogression and its contribution to endothelial dysfunction and plaque formation remain poorly understood. Emerging studies have begun to investigate the role of endothelial identity loss in atherosclerosis⁶ and have called into question whether EndMT-derived cells promote atherogenesis^7,8 or serve a protective role in stabilizing plaques⁹. Recent evidence has also characterized a metastable or partial EndMT state in lesions^10,11, suggesting that reinforcing endothelial identity may offer a therapeutic avenue to prevent or reverse pathological EC transitions.

The transcription factor E26 transformation-specific (ETS)-related gene (ERG) is a key regulator of EC identity; controlling chromatin accessibility^12,13 and orchestrating essential homeostatic EC functions such as angiogenesis^14,15, suppression of vascular inflammation^16,17, and maintenance of the endothelial barrier^18,19. ERG downregulation has been associated with induction of EndMT in chronic inflammatory diseases including liver and pulmonary fibrosis and pulmonary hypertension^20–22. In mice, EC-specific Erg deletion promotes inflammation, EndMT, and organ-specific dysfunction^20,21, though without major effects on arterial homeostasis or survival^13,20. In human atherosclerosis, a preliminary report suggests that ERG expression may be reduced in coronary endothelium overlying plaque¹⁶. Recent genome-wide association studies (GWAS) have also identified ERG as a risk locus for abdominal aortic aneurysms, where atherosclerosis is known to have an etiologic role^23,24. However, the role of ERG in governing chromatin accessibility and transcriptional programs that prevent EndMT and protect against plaque development within the atherosclerotic intima has not been examined.

In this study, we investigated the impact of EC-specific Erg deletion in hypercholesterolemic mouse models of atherosclerosis and identified a marked increase in atherogenesis in both atheroprone (disturbed laminar flow) and atheroresistant (uniform laminar flow) arterial regions. EC lineage tracing in Erg-deficient mice revealed a profound disruption of intimal EC organization, whereby ECs delaminated from the intima and infiltrated plaque. Single-cell RNA sequencing (scRNA-seq) profiling of aortic intimal cells revealed that Erg loss induced widespread erosion of endothelial identity prior to mesenchymal reprogramming, alongside pro-atherosclerotic changes in macrophages and smooth muscle cells (SMCs). This loss of endothelial function was further associated with early reductions in junctional signaling, increased endothelial permeability, and accumulation of foam cells in nascent lesions. To assess the role of ERG dysregulation in human atherosclerosis, we examined human carotid plaques and integrated our mouse data with carotid artery scRNA-seq and coronary artery single-nucleus assay for transposase-accessible chromatin with sequencing (ATAC-seq) datasets, which identified reduced ERG accessibility in plaques, ERG downregulation in EndMT-associated cells, and shared EndMT transcriptional programs across murine and human disease. In cultured human aortic ECs that had undergone EndMT due to loss of ERG, recovery of ERG resulted in a progressive reversal of EndMT chromatin remodeling (through modulation of AP-1 and ETS motifs) and gene regulation that attenuated TGF-β signaling and restored endothelial transcriptional identity. Collectively, our findings establish ERG as a mechanism for safeguarding EC identity and function that protects against atherogenesis by restricting EndMT programming–an event that appears to be a linchpin in atherosclerotic plaque progression.

Results

Deletion of endothelial Erg expands plaque burden and topology

To investigate the role of ERG in atherogenesis, we leveraged Cdh5^CreERT2; Rosa26^{lox-STOP-lox-Sun1-sfGFP}; Erg^fl/fl mice, which have an endothelial-specific Cre-inducible Sun1-sfGFP lineage tag and a conditional Erg knockout allele. Atherosclerosis was initiated in 8-week-old Cre-negative mice (Rosa26^{lox-STOP-lox-Sun1-sfGFP}; Erg^fl/fl, referred to as ERG^Ctrl) and Cre-positive mice (Cdh5^CreERT2; Rosa26^{lox-STOP-lox-Sun1-sfGFP}; Erg^fl/fl, referred to as ERG^EC-KO) by inducing hypercholesterolemia via adeno-associated virus-mediated overexpression of PCSK9, followed by 12 weeks of high-cholesterol diet (HCD) (Fig. 1a). We first validated loss of ERG in the arterial endothelium of normocholesterolemic ERG^EC-KO mice, where ERG-expressing cells were decreased by 85% following Cre induction (5 × 2 mg tamoxifen injections and 1 week recovery) and 78% after 12 weeks of normal chow diet (Fig. 1b, c). Following exposure to HCD, Erg knockout was associated with a 1.7-fold increase in aortic plaque development and extensive plaque accumulation in the brachiocephalic artery (BCA) and in the atheroresistant descending aorta compared to controls (Fig. 1d–f and Supplementary Fig. 1a-c).

Fig. 1. Endothelial Erg knockout exacerbates murine atherosclerosis.

a Schematic of Erg knockout and atherosclerosis induction with PCSK9 overexpression in Cre-negative (ERG^Ctrl) and Cre-positive mice (ERG^EC-KO). IP, intraperitoneal; IV, intravenous. Cartoon schematics were created in BioRender. Botts, S. (https://BioRender.com/b1lghtl). b Immunofluorescence staining of ERG in descending aorta cross sections from normocholesterolemic mice after 12 weeks of normal chow diet and quantification of ERG⁺ endothelial cells (c; ERG^Ctrl week 0/12, n = 3/6; ERG^EC-KO week 0/12, n = 3/4; one-sided Mann-Whitney tests; no correction). d Representative en face oil red O (ORO)-stained aortas from hypercholesterolemic mice after 12 weeks of high-cholesterol diet (HCD) and quantification of plaque in the total aorta (e) and descending aorta (f; ERG^Ctrl, n = 18; ERG^EC-KO, n = 19; one-sided unpaired t-tests). g Aortic arches stained with ORO and prepared en face at 12 weeks of HCD and quantification of total plaque (h; ERG^Ctrl, n = 9; ERG^EC-KO, n = 10; one-sided Mann-Whitney test) and plaque distribution from the midline (LC, lesser curvature) to the greater curvature (GC) (i; ERG^Ctrl, n = 9; ERG^EC-KO, n = 10; two-sided Mann-Whitney tests; Benjamini, Krieger, and Yekutieli FDR correction). Significant p values are indicated. j H&E and Movat stained aortic arch cross sections from atherosclerotic mice at 12 weeks of HCD. Dashed cyan lines denote plaque cap thickness and red arrows denote elastin breaks. k Quantification of aortic arch plaque area, intimal coverage, cellularity, and cap thickness, and medial elastin quantity and breaks (ERG^Ctrl, n = 10; ERG^EC-KO, n = 9; one-sided unpaired t-tests). l Representative immunofluorescence staining of CD68⁺ macrophages in brachiocephalic artery plaques at 12 weeks of HCD. m Quantification of CD68 expression in lesions (ERG^Ctrl, n = 5; ERG^EC-KO, n = 5; one-sided Mann-Whitney test). n Quantification of plasma total and low-density lipoprotein (LDL) cholesterol in ERG^Ctrl and ERG^EC-KO mice over 12 weeks of HCD (ERG^Ctrl, n = 9; ERG^EC-KO, n = 8; two-sided Mann-Whitney tests; Holm-Šídák correction). o Enumeration of peripheral blood (PB) Ly6C^hi monocytes in atherosclerotic mice after 12 weeks of HCD (ERG^Ctrl, n = 9; ERG^EC-KO, n = 9; one-sided unpaired t-test). Bar and line graphs are presented as mean ± standard error of the mean (SEM). For all bar graphs, male data points are depicted with circles and female data points are depicted with triangles. Source data are provided as a Source Data file.

To further examine whether ERG loss alters site-specificity of atherosclerosis, we quantified intimal lipid deposition in the lesser curvature of the aortic arch, a region with proatherogenic disturbed flow, and the greater curvature, a region of uniform laminar flow and limited plaque formation (Fig. 1g–i). While lipid in ERG^Ctrl mice was mainly restricted to the lesser curvature, we observed unrestricted expansive lipid deposition in ERG^EC-KO mice, showing that ERG loss in a hypercholesterolemic setting drives atherogenesis in regions typically considered atheroresistant. Histological examination of aortic arch and BCA lesions supported increased plaque area and intimal coverage with Erg deletion (Fig. 1j, k and Supplementary Fig. 1b, c). Moreover, ERG^EC-KO lesions were associated with features of plaque vulnerability including decreased cap thickness and increased medial elastin degradation. ERG^EC-KO lesions also contained more CD68⁺ macrophages compared to controls (Fig. 1l, m).

Importantly, we observed no differences in total or LDL cholesterol or circulating triglycerides over 12 weeks of HCD in ERG^EC-KO mice compared to controls (Fig. 1n and Supplementary Fig. 1d). We concurrently assessed whether loss of ERG altered peripheral blood leukocytes in atherosclerotic mice and found no elevations in total leukocyte or Ly6C^hi monocyte populations, which indicated that increased macrophage accumulation in ERG^EC-KO mice may result from direct effects of Erg knockout on the arterial endothelium (Fig. 1o and Supplementary Fig. 1e–h). Together, these data highlight that ERG loss unleashes widespread atherogenesis in hypercholesterolemic mice and implicate perturbations of the endothelium as an inciting mechanism.

ERG loss drives endothelial-to-mesenchymal transition

To enable lineage tracing of ECs in both ERG^Ctrl and ERG^EC-KO groups and to control for Cre expression, we introduced the lox-STOP-lox Sun1-sfGFP reporter allele into both unfloxed (Cdh5^CreERT2; Rosa26^{lox-STOP-lox-Sun1-sfGFP}; Erg^+/+) and floxed lines (Cdh5^CreERT2; Rosa26^{lox-STOP-lox-Sun1-sfGFP}; Erg^fl/fl) and assessed plaque development following the induction of hypercholesterolemia (Supplementary Fig. 2a). These mice were used for all subsequent experiments. Loss of ERG increased aortic plaque at 12 weeks of HCD (1.8-fold), with prominent atherosclerosis in the descending aorta of ERG^EC-KO mice compared to control mice (Supplementary Fig. 2b–d). We next characterized the effect of Erg deletion on early atherogenesis by assessing aortic intimal lipid deposition at four weeks of HCD. While histological examination did not reveal pronounced lesions in either ERG^Ctrl or ERG^EC-KO mice (Supplementary Fig. 2e), Erg knockout at this early timepoint did result in a marked 3.9-fold increase in intimal lipid accumulation (Supplementary Fig. 2f–h).

To characterize the fate of endothelial-derived cells in atherosclerotic plaque, we visualized expression of the EC lineage tag in established BCA and aortic arch lesions from ERG^Ctrl and ERG^EC-KO mice following 12 weeks of HCD (Fig. 2a and Supplementary Fig. 2i). Erg knockout resulted in plaque formation in both the atheroprone lesser curvature and atheroresistant greater curvature of the aortic arch (Supplementary Fig. 2i). Loss of ERG was associated with a nearly 2-fold increase in both BCA plaque size and cellularity indicative of a more complex intraplaque environment (Fig. 2b, c). At plaque and adjacent non-plaque regions of the artery, Erg knockout resulted in a 1.7-fold and 3.0-fold reduction in ERG⁺ ECs, respectively (Fig. 2d). We next enumerated lineage-tagged (GFP⁺) cells to assess EC turnover and expansion following loss of ERG (Fig. 2e). We observed a decrease in lineage-tagged cells (GFP⁺ cells) in ERG^EC-KO mice at the vascular lumen in both plaque and non-plaque sites, which suggested that Erg knockout cells are replaced with GFP^-ERG⁺ ECs (Fig. 2e and Supplementary Fig. 2i). Remarkably, loss of ERG resulted in a 5.1-fold expansion in the number of EC-lineage cells (GFP⁺ERG^-) within BCA lesions, which accounted for nearly 60% of intraplaque cells (Fig. 2e–g). The substantial migration of EC-lineage cells into plaque was associated with EndMT in ERG^EC-KO mice, as there was a 31.5-fold increase in GFP⁺ACTA2⁺ cells (Fig. 2h, i).

Fig. 2. Loss of ERG promotes endothelial-to-mesenchymal transition and endothelial cell ingression into plaque.

a Representative immunofluorescence staining of brachiocephalic artery (BCA) cross sections from unfloxed Erg (Cdh5^CreERT2; Rosa26^{lox-STOP-lox-Sun1-sfGFP}; Erg^+/+, referred to as ERG^Ctrl) and floxed Erg mice (Cdh5^CreERT2; Rosa26^{lox-STOP-lox-Sun1-sfGFP}; Erg^fl/fl, referred to as ERG^EC-KO) with AAV8-PCSK9-induced atherosclerosis and 12 weeks of high-cholesterol diet (HCD). Dashed yellow lines denote plaque. Magenta arrows denote replenishment of the endothelium with GFP^-ERG⁺ endothelial cells (ECs) in ERG^EC-KO mice. Cyan arrows denote ingression of GFP⁺ EC-lineage cells in normal atherosclerosis (ERG^Ctrl mice). Quantification of plaque area (b; ERG^Ctrl, n = 10; ERG^EC-KO, n = 16; one-sided Mann-Whitney test) and cellularity (c; ERG^Ctrl, n = 6; ERG^EC-KO, n = 15; one-sided unpaired t-test). Enumeration of ERG⁺ cells (d; two-sided Mann-Whitney tests; Holm-Šídák correction) and GFP⁺ cells by percentage (e; two-way ANOVA; Holm-Šídák correction) in the non-plaque adjacent endothelium (AE), plaque endothelium (PE), and plaque interior (PI) of BCA cross sections (ERG^Ctrl, n = 6; ERG^EC-KO, n = 15). Quantification of ERG⁺ (f) and GFP⁺ plaque cells (g; ERG^Ctrl, n = 6; ERG^EC-KO, n = 15; one-sided unpaired t-tests). h Immunofluorescence staining of the mesenchymal marker ACTA2 in BCA plaques following 12 weeks of HCD. White arrows denote EC-lineage cells with co-expression of GFP and ACTA2. i Quantification of GFP⁺ACTA2⁺ cells in BCA lesions (ERG^Ctrl, n = 3; ERG^EC-KO, n = 5; one-sided Mann-Whitney test). Bar graphs are presented as mean ± standard error of the mean (SEM). Male data points are depicted with circles and female data points are depicted with triangles. Source data are provided as a Source Data file.

To validate our findings in a second murine atherosclerosis model, we bred ERG^Ctrl and ERG^EC-KO mouse lines into an ApoE-deficient background (Apoe^{tm1Unc/tm1Unc}) (Supplementary Fig. 3a). Compared to deleting Apoe alone (ERG^apCtrl), co-deletion of Erg and Apoe (ERG^apEC-KO) resulted in a 2.2-fold increase in aortic plaque development following 12 weeks of HCD (Supplementary Fig. 3b–d) and also increased lung weight (Supplementary Fig. 3e–h). Given the robust effect of ERG loss in complementary models, we assessed whether Erg knockout alone (without hypercholesterolemia) was sufficient to promote plaque development in ERG^EC-KO mice over an extended 6-month timeframe (Supplementary Fig. 4a). Following 25 weeks of normal chow diet without an additional atherogenic insult, aged ERG^Ctrl and ERG^EC-KO mice had comparable body and tissue weight and loss of ERG was not associated with increased morbidity or mortality, in line with previous reports (Supplementary Fig. 4b–e)^13,20. Likewise, sustained Erg knockout did not result in aortic plaque development or migration of lineage-tagged cells into the vessel wall (Supplementary Fig. 4f–h), despite evidence that luminal ECs had acquired expression of the mesenchymal marker FN1 (Supplementary Fig. 4g–i). This suggests that hyperlipidemia is required for ingress of Erg knockout ECs. Together, these findings establish a critical role for ERG in maintaining arterial endothelial identity and function in the setting of atherosclerotic disease conditions.

ERG restricts pro-atherogenic intimal cell states

To comprehensively profile the cellular composition of plaques in unfloxed ERG^Ctrl (Cdh5^CreERT2; Rosa26^{lox-STOP-lox-Sun1-sfGFP}; Erg^+/+) and floxed ERG^EC-KO mice (Cdh5^CreERT2; Rosa26^{lox-STOP-lox-Sun1-sfGFP}; Erg^fl/fl), we leveraged scRNA-seq on intimal cell populations isolated following a brief en face digestion of aortic tissue (Fig. 3a). Hypercholesterolemia was induced with PCSK9 overexpression and intimal cells from Sun1-sfGFP EC lineage-tagged mice were harvested to investigate early changes that occurred prior to EC migration into the intima (week 4 of HCD) and in established plaque (week 12 of HCD). We next examined aortic intimal cell populations shared among ERG^Ctrl and ERG^EC-KO mice during early and late atherogenesis (Fig. 3b, c and Supplementary Fig. 5a, b). Intimal cell enrichment yielded ~9500 ECs (26%) in three clusters defined by endothelial transcripts including Calcrl, Pecam1, and Ptprb, of which EC 1 and EC 2 expressed the GFP lineage tag (Supplementary Fig. 5c). We likewise noted ~3000 macrophage cells (8%) characterized by marker genes such as Adgre1, Cd68, and Mertk, and ~20,000 SMCs (55%) that were broadly classified by marker genes including Acta2, Myh11, and Tagln (Supplementary Fig. 5d, e).

Fig. 3. Erg knockout induces pro-atherosclerotic phenotypes among multiple aortic intimal cell populations.

a Schematic of aortic intimal isolation, single-cell RNA sequencing (scRNA-seq), and endothelial cell (EC) lineage tracing in samples from unfloxed Erg (ERG^Ctrl) and floxed Erg mice (ERG^EC-KO) with AAV8-PCSK9-induced atherosclerosis and 4 or 12 weeks of high-cholesterol diet (HCD) (ERG^Ctrl, n = 6 pooled mice; ERG^EC-KO, n = 6 pooled mice). Cartoon schematics were created in BioRender. Botts, S. (https://BioRender.com/b1lghtl). b, c Uniform manifold approximation and projection (UMAP) plots of aortic intimal cells from ERG^Ctrl and ERG^EC-KO mice at 4 or 12 weeks of HCD and enumeration of cell clusters (d). Prolif. cell, proliferating cell; SMC, smooth muscle cell. Differential ligand-receptor (LR) interaction plots for ECs and macrophages (e) and ECs and SMCs (f) after 12 weeks of HCD (p.adj <0.05; two-sided Wilcoxon rank-sum tests; Bonferroni correction). Select LR interactions enriched in ERG^Ctrl and ERG^EC-KO mice are shown in blue and red text respectively. g–i LR probabilities for select interactions between ECs, macrophages, and SMCs (p.adj <0.05; two-sided Wilcoxon rank-sum tests; Bonferroni correction). Mφ, macrophage.

Stratification of intimal cell populations by genotype and lesion stage revealed a marked increase in the proportion of macrophages in ERG^EC-KO mice compared to ERG^Ctrl in both early (3.8-fold) and late lesions (3.4-fold) with concomitant loss of contractile SMCs (Fig. 3c, d). Loss of ERG in early lesions preferentially drove expansion of Macrophage 1, which enriched for both inflammatory (Ccl2 and Il1b) and foam cell genes (Fabp5 and Trem2) and was associated with a Trem2⁺ macrophage phenotype (lipid transport and metabolism, oxidative phosphorylation, and leukocyte proliferation) (Fig. 3c, d and Supplementary Fig. 5f). A comparable increase by fold-change was observed for Macrophage 2, which enriched for inflammatory pathways including leukocyte activation, cytokine signaling, and inflammatory response (Fig. 3c, d and Supplementary Fig. 5g). In established lesions, Erg knockout also favored an increase in Macrophage 3 that expressed anti-inflammatory and efferocytic markers (Abca1, Mertk, and Pparg) (Fig. 3c, d and Supplementary Fig. 5d). Likewise, a population of foamy SMCs (SMC 2 and SMC 3), which enriched for immune and Trem2⁺ pathways (Supplementary Fig. 5h), expanded with loss of ERG in early plaques (2.4-fold increase) (Fig. 3c, d). Finally, Erg knockout induced a pro-atherosclerotic shift in T cell populations from an immature Cd4⁺Cd8⁺ phenotype (T cell 1) towards a cytotoxic state (T cell 2) in early and late lesions (Fig. 3c, d and Supplementary Fig. 5i). Together, these data highlight that loss of ERG accelerates the development of cell states (foamy and inflammatory macrophages, loss of contractile SMCs with acquisition of foamy SMCs, and cytotoxic T cells) that promote a pro-atherosclerotic and proinflammatory environment in the aortic intima.

To investigate the intercellular communication networks between intimal cell populations, we next inferred differential ligand-receptor (LR) interactions for ERG^Ctrl and ERG^EC-KO mice from scRNA-seq data in early and established atherosclerosis. In early lesions, loss of ERG reduced interaction strength among ECs and contractile SMC 1 and increased strength between ECs and pro-atherosclerotic cells (foamy and inflammatory macrophages, foamy SMCs, and cytotoxic T cells) (Supplementary Fig. 6a). Intimal cells in established lesions demonstrated markedly increased interaction strength, where EC-EC and EC-SMC LR pairs were enriched in control mice and EC-Macrophage LR pairs were enriched with Erg knockout (Supplementary Fig. 6b). In early and established plaques, the top LR interactions for ERG^Ctrl mice occurred between ECs and were associated with regulation of angiogenesis (ANGPT1-TEK and JAG1-NOTCH1) and cell-cell adhesion (CADM1-CADM1 and GJA5-GJA5) (Fig. 3e–g and Supplementary Fig. 6c–e). In contrast, the top LR pairs for ERG^EC-KO mice involved more robust communication between ECs and other cell types (macrophages and SMCs), and in established lesions featured a prominent association between the promigratory proteoglycan receptor syndecan-4 (SDC4) in ECs and multiple ligands across several cell types (COL1A1, COL1A2, FN1, and THBS1) (Fig. 3e, f, h and Supplementary Fig. 6f). Erg knockout in early and established plaque also increased interactions between several EC ligands (COL1A1, COL4A1, FN1, and SPP1) and the cell-surface glycoprotein CD44 on macrophages, which promotes myeloid recruitment in atherosclerosis (Fig. 3e, i and Supplementary Fig. 6c, f)²⁵. These cell-cell networks collectively support a model for reduced signaling amongst ECs upon loss of ERG and markedly increased interactions between ECs (or EC-derived cells) and macrophages/SMCs, including through augmented SDC4-mediated signals received by ECs and CD44-mediated signals initiated by ECs, which have been associated with disruption of the endothelial barrier, enhanced EC migration, and myeloid accumulation in lesions^25–27.

Erg deletion compromises endothelial transcriptional identity

Our examination of the plaque microenvironment in ERG^EC-KO mice revealed that loss of ERG profoundly disrupted EC organization and signaling in the arterial intima. To determine whether these changes were associated with alterations in transcriptional identity, we focused on intimal ECs from our scRNA-seq dataset and explored gene expression programs in early and established atherosclerosis. Reclustering the EC-specific subset yielded a cell population that was stratified by GFP expression, and GFP-enriched ECs (EC 1’ and EC 2’) were prioritized for further transcriptional analysis (Supplementary Fig. 7a–c). The resulting unintegrated subset contained ~4000 ECs that clustered according to both genotype and stage of plaque development (Fig. 4a).

Fig. 4. Loss of ERG promotes transcriptional programs for arterial endothelial cell dedifferentiation, mesenchymal transition, and cell migration in atherosclerosis.

a Uniform manifold approximation and projection (UMAP) plot of aortic intimal endothelial cells (ECs) from unfloxed Erg (ERG^Ctrl) and floxed Erg mice (ERG^EC-KO) with AAV8-PCSK9-induced atherosclerosis and 4 or 12 weeks of high-cholesterol diet (HCD) (ERG^Ctrl, n = 6 pooled mice; ERG^EC-KO, n = 6 pooled mice). Composite score UMAP plots for differentially expressed (DE) EC (b) and endothelial-to-mesenchymal transition transcripts (c; EndMT) with top transcripts by p value highlighted. d Quantification of EC and EndMT composite scores between ERG^Ctrl and ERG^EC-KO ECs (ERG^Ctrl week 4/12, n = 626/1851; ERG^EC-KO week 4/12, n = 575/1273; Kruskal-Wallis tests with Dunn’s multiple comparisons; Bonferroni correction). e DE analysis of EC and EndMT transcripts from ERG^Ctrl and ERG^EC-KO ECs at 4 or 12 weeks of HCD (p.adj <0.05; two-sided Wilcoxon rank-sum tests; Bonferroni correction). f Pseudotime analysis of gene expression in aortic intimal ECs with trajectory progression from ERG^Ctrl to ERG^EC-KO ECs (shown with the blue line). g Select EC and EndMT gene expression plots from pseudotime trajectory modules (p.adj <0.05; Moran’s I statistic; Benjamini-Hochberg correction). h Radar plots depicting enrichment of pathway categories with Erg knockout in aortic intimal ECs (p.adj <0.05; one-sided Fisher’s exact tests; Benjamini-Hochberg correction). i Global pathway enrichment following Erg knockout with top pathways by gene ratio highlighted for categories of interest (p.adj <0.05; one-sided Fisher’s exact tests; Benjamini-Hochberg correction). j Gene-pathway associations depicting the top DE transcripts and the top EC and EndMT transcripts by enrichment score (Enrch. score; p.adj <0.05; one-sided Fisher’s exact tests; Benjamini-Hochberg correction). End., endothelial; Mes., mesenchymal; Adh., adhesion; Mig., migration. k DE analysis of adhesion and migration-associated transcripts from ERG^Ctrl and ERG^EC-KO ECs at 4 or 12 weeks of HCD with genes of interest highlighted (p.adj <0.05; two-sided Wilcoxon rank-sum tests; Bonferroni correction). Adhesion and migration genes are denoted in light blue (week 4) and dark blue (week 12). l UMAP plots for DE adhesion and migration-associated transcripts of interest.

We first assessed shifts in transcriptional programs associated with cell identity that occurred across time (between early and established plaque) and identified a shared EndMT program for ERG^Ctrl and ERG^EC-KO mice, which included increased expression of the mesenchymal transcripts Ccn2, Mgp, Vim, and the stem cell marker Ly6a (Supplementary Fig. 7d, e). This appeared to be particularly pronounced at 12 weeks in the ERG^EC-KO group. To next characterize the contribution of Erg knockout to endothelial identity loss, we leveraged a rank-based signature enrichment analysis²⁸ to assess a composite of 139 differentially expressed (DE) endothelial transcripts and 130 DE transcripts associated with EndMT (Fig. 4b–d). Notably, we observed a robust decrease in endothelial identity genes with loss of ERG in early atherosclerosis, which was not accompanied by increased expression of EndMT genes at this stage (Fig. 4d). In contrast, in more established plaques, loss of ERG resulted in both a decrease in endothelial composite score and a marked increase in EndMT score (Fig. 4d). These data suggest that in the setting of atherosclerosis, the loss of ERG induces a dedifferentiation of ECs away from an established endothelial identity and that this precedes the acquisition of a mesenchymal cell phenotype.

Globally, Erg knockout altered ~6% of the EC transcriptome in early plaque and ~24% in established plaque, which resulted in downregulation of endothelial transcripts including Calcrl, Foxp1, Nos3, and Vwf, and upregulation of mesenchymal transcripts including Col1a2, Fbln5, Ly6a, and S100a4 (Fig. 4e and Supplementary Fig. 7e). We next conducted pseudotime analysis among EC clusters across genotype and time (Fig. 4f). We focused on a pseudotime trajectory in established plaque that progressed from ERG^Ctrl to ERG^EC-KO mice in three transcriptome modules, where endothelial genes were associated with module 1 and mesenchymal genes were associated with modules 2 and 3 (Fig. 4g and Supplementary Fig. 7f). Interestingly, module 1 genes including Epas1 and Nos3 were characterized by a linear decrease in expression in ERG^Ctrl mice over time before plateauing in ERG^EC-KO mice, while module 2 genes including Col8a1 and Fn1 showed a linear increase in ERG^Ctrl mice over time and a plateau in ERG^EC-KO mice. These expression patterns indicate that Erg knockout may augment the gain of EndMT phenotypes that occurs during normal atherosclerosis progression.

Pathway analysis of DE transcripts in ERG^EC-KO versus ERG^Ctrl mice highlighted an enrichment of cell adhesion and migration processes that was prominent in early and established plaque development (Fig. 4h). Notably, loss of ERG resulted in enhanced EndMT-associated Gene Ontology (GO) terms including TGF-β receptor superfamily signaling, epithelial-to-mesenchymal transition, and mesenchymal cell differentiation in addition to altering adhesion (cell-substrate adhesion, cell junction assembly) and migration pathways (epithelial cell migration, tissue migration) (Fig. 4i and Supplementary Fig. 7g). We next investigated the transcripts that drove these pathway predictions and identified strong associations between downregulated endothelial genes (Foxp1, Tek), upregulated EndMT genes (Fn1, Col8a1), and altered adhesion and migration, in accordance with our observation of EC ingression into plaques from ERG^EC-KO mice (Fig. 4j and Supplementary Fig. 8a–d). A large proportion of DE transcripts enriched for dysregulation of cell adhesion and migration during plaque development (2.0% of transcriptome in early plaque and 5.4% in established plaque) (Fig. 4k, l). In line with reduced EC-EC signaling predicted by our LR interaction analysis, we observed a downregulation of genes encoding EC junction proteins including Cadm1, Cgnl1, Gja5, and Ptprb with loss of ERG. Homeostatic regulators of angiogenesis (Angpt1, Tek, Tie1, Vegfc) were also decreased with Erg knockout. In contrast, ERG^EC-KO ECs were highly enriched for extracellular matrix proteins including Col8a1, Ecm1, Efemp1, and Fn1, which are associated with fibrotic deposition in plaque^29,30, and migration-associated genes including Cdh13, Lgmn, Sdc4, and Sox4, which have been implicated in pathological aging, endothelial inflammation, and atherosclerosis^27,31–33.

ERG loss promotes lipid uptake in early lesions

Given that EC identity supports several atheroprotective roles including maintenance of endothelial barrier, and that early changes in EC identity are seen in ERG^EC-KO mice, we sought to determine the functional impact of ERG loss on EC adhesion and barrier integrity. We therefore administered fluorescently labeled dextran to unfloxed ERG^Ctrl and floxed ERG^EC-KO normocholesterolemic mice and measured dextran accumulation in the aortic intima (Fig. 5a). Loss of ERG markedly increased endothelial permeability to dextran in both the atheroprone lesser curvature and atheroresistant greater curvature of the aorta (2.5-fold and 4.5-fold increase in integrated density), which was consistent with the altered intimal lipid deposition observed in ERG^EC-KO mice during atherogenesis (Fig. 5b, c). Increased barrier disruption with Erg knockout was not associated with changes in the area or number of CD45⁺ intimal myeloid cells at this early timepoint (Fig. 5d, e).

Fig. 5. Erg knockout increases arterial endothelial permeability, lipid uptake, and intimal myeloid cell accumulation during atherogenesis.

a Representative immunofluorescence staining of TRITC-dextran and CD45⁺ myeloid cells in aortic arch cross sections from normocholesterolemic unfloxed Erg (ERG^Ctrl) and floxed Erg mice (ERG^EC-KO). GC, greater curvature; LC, lesser curvature. Dashed white lines denote the aortic arch LC. White arrows denote TRITC-dextran accumulation in the aortic arch GC of ERG^EC-KO mice. Quantification of subendothelial TRITC-dextran mean fluorescence intensity (MFI) (b; one-sided Welch’s t-test and two-way ANOVA; Holm-Šídák correction) and integrated density (IntDen) (c; ERG^Ctrl, n = 8; ERG^EC-KO, n = 8; one-sided unpaired t-test and two-way ANOVA; Holm-Šídák correction). Enumeration of CD45⁺ intimal myeloid cells by area (d; one-sided unpaired t-test and two-way ANOVA; Holm-Šídák correction) and by cell number (e; ERG^Ctrl, n = 8; ERG^EC-KO, n = 8; one-sided unpaired t-test and Mann-Whitney tests; Holm-Šídák correction). f, Representative immunofluorescence staining of DiI-labeled low-density lipoprotein (LDL) and CD45⁺ myeloid cells in aortic arch cross sections from unfloxed Erg (ERG^Ctrl) and floxed Erg mice (ERG^EC-KO) with AAV8-PCSK9-induced atherosclerosis and 1 week of high-cholesterol diet. Dashed white lines denote the aortic arch LC. White arrows denote DiI-LDL accumulation in the aortic arch GC of ERG^EC-KO mice. Quantification of myeloid cell DiI-LDL uptake by MFI (g) and IntDen (h; ERG^Ctrl, n = 9; ERG^EC-KO, n = 7; one-sided unpaired t-tests and two-way ANOVAs; Holm-Šídák correction). Enumeration of CD45⁺ intimal myeloid cells by area (i) and by cell number (j; ERG^Ctrl, n = 9; ERG^EC-KO, n = 7; one-sided unpaired t-tests and two-way ANOVAs; Holm-Šídák correction). Bar graphs are presented as mean ± standard error of the mean (SEM). Male data points are depicted with circles and female data points are depicted with triangles. Significant p values are shown in bold text. Source data are provided as a Source Data file.

Next, we induced hypercholesterolemia in unfloxed ERG^Ctrl and floxed ERG^EC-KO mice with PCSK9 overexpression and administered DiI-labeled low-density lipoprotein (LDL) at 1 week of HCD to assess early intimal myeloid lipid uptake (Fig. 5f and Supplementary Fig. 9a). We noted a robust increase in myeloid cell uptake of DiI-LDL in the aortic arch greater curvature (5.6-fold) and BCA (5.8-fold) of ERG^EC-KO mice (Fig. 5g, h and Supplementary Fig. 9b, c, respectively). Consistent with this observation, loss of ERG resulted in a considerable accumulation of myeloid populations in the greater curvature and BCA (2.3-fold and 1.8-fold increase in CD45⁺ intimal cells) (Fig. 5i, j and Supplementary Fig. 9d, e). Increased LDL uptake with Erg knockout did not reach significance in the atheroprone lesser curvature, likely due to the normal propensity for LDL accumulation in this region. These data collectively demonstrate that loss of ERG increases permeability of the arterial endothelium and compromises endothelial barrier in early atherogenesis, which results in intimal myeloid lipid accumulation during hypercholesterolemia.

Dysregulation of ERG in human plaque

To determine the relevance of our genetic loss-of-function mouse model to human atherogenesis, we assessed whether ERG expression was altered in lesions obtained from human carotid endarterectomy, by comparing ERG protein in ECs overlying plaque with endothelial lineage cells (CD31⁺ERG⁺) observed in the plaque interior (Fig. 6a, b). Notably, ERG expression decreased 2.3-fold in intraplaque endothelial lineage cells relative to ECs at the lumen, which suggested that loss of ERG is associated with migration of ECs into human atherosclerotic lesions. We next investigated disease associations among our scRNA-seq data obtained from Erg knockout in murine atherosclerosis. In addition to prominent and progressive enrichments for cardiovascular disease from 4 to 12 weeks of HCD, cancer terms were among the top pathways (tumor progression, neoplasm metastasis, carcinogenesis), which is likely related to the induction of a dedifferentiated, mesenchymal, and migratory phenotype (Fig. 6c). Notably, both cancer and cardiovascular disease enrichment in ERG^EC-KO mice (atherosclerosis, vascular diseases, coronary artery disease) were driven by pro-migratory genes including Cdh13 and Lgmn and extracellular matrix genes including Col8a1, Fbln5, and Fn1 (Supplementary Fig. 10a, b).

Fig. 6. ERG loss is associated with endothelial dysfunction in human atherosclerosis.

a Representative immunofluorescence staining of ERG in human carotid plaque and quantification of ERG expression in EC-lineage cells by mean fluorescence intensity (MFI) and integrated density (IntDen) (b PE, n = 69; PI, n = 124; one-sided Mann-Whitney tests). Magenta arrows indicate endothelial cells (ECs) that form the plaque endothelium (PE). Cyan arrows denote CD31⁺ERG⁺ EC-lineage cells that are observed in the plaque interior (PI). c Disease enrichment of differentially expressed (DE) transcripts between aortic intimal ECs from unfloxed Erg (ERG^Ctrl) and floxed Erg mice (ERG^EC-KO) with AAV8-PCSK9-induced atherosclerosis and 4 or 12 weeks of high-cholesterol diet (ERG^Ctrl, n = 6 pooled mice; ERG^EC-KO, n = 6 pooled mice; p.adj <0.05; one-sided Fisher’s exact tests; Benjamini-Hochberg correction). Top pathways by p value are shown. Genome-wide association study (GWAS) enrichment of DE transcripts from ERG^EC-KO ECs and single nucleotide polymorphisms (SNPs) associated with cancer (d), cardiovascular disease (e), coronary artery disease (f), and atherosclerosis (g) (p.adj <0.05; two-sided Wilcoxon rank-sum tests; Bonferroni correction). Select transcripts of interest are highlighted. h Uniform manifold approximation and projection (UMAP) plot of ECs from human carotid atherosclerotic core (AC) and proximal adjacent regions (PA). i Pearson correlation plot between ERG^EC-KO-enriched EC transcripts (mouse) and AC-enriched EC transcripts (human) with select EC and endothelial-to-mesenchymal transition (EndMT) genes highlighted (p.adj <0.05; two-sided Wilcoxon rank-sum tests; Bonferroni correction). j UMAP plots for DE endothelial and mesenchymal transcripts of interest. k Differential bulk chromatin accessibility of ERG in human coronary lesions versus non-lesions with the top three differentially accessible regions (DARs) by p value highlighted (p.adj <0.05; two-sided Wald tests; Benjamini-Hochberg correction). l Lesion category and cell type-specific chromatin accessibility for the top three DARs with the EC contribution depicted in pink and the ENCODE human coronary artery H3K27ac chromatin immunoprecipitation sequencing track depicted in black. m Motif deviations associated with single-nucleus chromatin accessibility in human coronary artery cells with ETS and ERG motifs highlighted. Source data are provided as a Source Data file.

To further investigate ERG loss in the context of human disease, we integrated our mouse scRNA-seq data with the NHGRI-EBI genome-wide association study (GWAS) Catalog³⁴ and identified associations between transcripts dysregulated by ERG loss and genome-wide significant SNPs mapping to CDH13, COL8A1, and FN1 in GWAS datasets for cancer, cardiovascular disease, and atherosclerosis (Fig. 6d–g). In parallel, we analyzed a publicly available scRNA-seq dataset obtained from human carotid endarterectomy samples³⁵ and compared plaque-enriched ECs (EC 2’ and EC 3’) with ERG^EC-KO ECs in mice (Fig. 6h, i and Supplementary Fig. 10c, d). We observed a notable overlap of transcripts that were enriched in both human carotid plaque ECs and in plaque ECs from our Erg knockout mouse model, including CDH13, COL8A1, FBLN5, and FN1 (Fig. 6i, j). Moreover, key human plaque and ERG^EC-KO endothelial transcripts were similarly downregulated including F8, FZD4, POSTN, and VEGFC (Fig. 6i, j), which have been associated with angiogenesis, cell adhesion, and endothelial barrier^36–38, highlighting the potential relevance of ERG-mediated altered EC identity in atherosclerosis.

Finally, we assessed whether ERG accessibility is altered in human atherosclerosis by querying a dataset of bulk ATAC-seq profiling in human atherosclerotic coronary artery tissue (Fig. 6k). We identified 14 differentially accessible regions (DARs) that mapped to ERG and were collectively downregulated in lesion versus non-lesion coronary artery samples. To elucidate the specific contribution of ECs to ERG accessibility loss in human plaque, we leveraged single-nucleus ATAC-seq data obtained from a similar patient cohort³⁹ and assessed cell type contributions for the top three DARs obtained from our bulk ATAC-seq samples (Fig. 6l). In line with our bulk data, global accessibility for each DAR was reduced in lesion vs non-lesion coronary artery tissue. For two of the three top DARs, accessibility enrichment was greatest among ECs, while accessibility of the remaining DAR was largely attributable to fibroblasts. We next predicted transcription factor binding for each DAR and observed that the top motifs with downregulated accessibility in lesions included MYC, FOXM1, and TRAF7, which have been implicated in angiogenesis and endothelial barrier function (Supplementary Fig. 10e–g)^37,40. Moreover, we assessed motif deviation in single-nucleus ATAC-seq data from coronary artery lesion and non-lesion cells and noted that the top variable motifs including ERG, ETS1, and FLI1 belonged to the ETS transcription factor family (Fig. 6m). Together, these data highlight that loss of the ETS transcription factor ERG is a prominent feature of EC dysfunction in human atherosclerotic plaque and in light of our early and established murine plaque models, suggest this could be a linchpin in atheroprogression.

ERG recovery reverses endothelial-to-mesenchymal transition

To determine whether reintroduction of ERG is sufficient to reverse EndMT, we performed siRNA-mediated ERG silencing in telomerase-immortalized human aortic endothelial cells (teloHAECs) and allowed for natural recovery of ERG expression over 10 days following transfection (Fig. 7a). We longitudinally profiled control (Ctrl) and ERG siRNA teloHAECs with RNA-seq and ATAC-seq from day 0 (72 h post-transfection) to day 7 of ERG recovery to assess changes in EndMT chromatin accessibility and transcriptional programs.

Fig. 7. ERG recovery re-establishes endothelial chromatin accessibility signatures and reverses endothelial-to-mesenchymal transition.

a Schematic of small interfering RNA (siRNA)-mediated ERG silencing and recovery in telomerase-immortalized human aortic endothelial cells (teloHAECs) with longitudinal profiling using RNA sequencing (RNA-seq; n = 4 per group) and assay for transposase-accessible chromatin with sequencing (ATAC-seq; n = 3 per group) from day 0 (72 h post-transfection) to day 7. ERG transcript expression is depicted with normalized count (10³) (**, p = 8.15e-107; *, p = 9.65e-11; ++, p = 4.46e-1; +, p = 1.00e0; two-sided Wald tests; Benjamini-Hochberg correction). Ctrl, control; KD, knockdown; EndMT, endothelial-to-mesenchymal transition. Cartoon schematics were created in BioRender. Botts, S. (https://BioRender.com/b1lghtl). b Differentially expressed mRNA plots depicting downregulated transcripts (top) and upregulated transcripts (bottom) observed in aortic intimal endothelial cells from ERG^EC-KO mice with 4 or 12 weeks of hypercholesterolemia (left) and ERG siRNA-treated teloHAECs at 72 h post-transfection (right). EC and EndMT transcripts are enumerated in the plot insets (p.adj <0.05; two-sided Wald tests; Benjamini-Hochberg correction). KO, knockout. c Quantification of EC and EndMT composite scores between Ctrl and ERG siRNA-treated teloHAECs (Kruskal-Wallis tests with Dunn’s multiple comparisons; Bonferroni correction). d Pearson correlation plot between ERG^EC-KO-enriched EC transcripts (mouse; p.adj <0.05; two-sided Wilcoxon rank-sum tests; Bonferroni correction) and ERG siRNA-enriched EC transcripts (teloHAECs; p.adj <0.05; two-sided Wald tests; Benjamini-Hochberg correction) with select EC and EndMT genes highlighted. e Heatmap depicting transcript expression (RNA-seq) and ETS and AP-1 motif accessibility (ATAC-seq) for select EC, EndMT, and TGF-β genes in Ctrl and ERG siRNA-treated teloHAECs (p.adj <0.05; two-sided Wald tests; Benjamini-Hochberg correction). FC, fold-change. RNA-seq violin plots depicting expression of EC (f), EndMT (g), and TGF-β transcripts of interest (h) (two-sided Wald tests; Benjamini-Hochberg correction). Differential chromatin accessibility of Ctrl and ERG siRNA-treated teloHAECs for select EC (i) and EndMT genes (j). H3K27ac chromatin immunoprecipitation sequencing (ChIP-seq) performed in Ctrl versus ERG siRNA-treated human umbilical vein endothelial cells (HUVECs) (GSE124892) and ATAC-seq performed in human MRC-5 fibroblasts (GSE121053) were respectively used as reference tracks for EC and EndMT accessibility. Ctrl (k) and ERG siRNA-enriched motifs (l) rank-ordered by p value in teloHAECs from day 0 to day 7 of ERG recovery (left plots) (p.adj <0.05; one-sided Fisher’s exact tests; Benjamini-Hochberg correction). Top motifs are depicted in heatmaps (right) with ETS and AP-1 motifs respectively highlighted with blue and red text. Violin plots indicate density of data points at corresponding values. Significant p values are shown in bold text (violin plots) or with black dots (heatmaps).

At 72 h post-transfection, ERG knockdown (KD) altered a considerable proportion of the EC transcriptome, where 2040 transcripts (24.6%) were downregulated and 2321 transcripts (28.0%) were upregulated with ERG siRNA treatment (Fig. 7b). A subset of 224 genes was concordantly downregulated in both ERG KD teloHAECs and ERG^EC-KO mice and 244 genes were concordantly upregulated, which highlighted transcriptional programs shared between cultured EC and murine models. We used the previous rank-based signature enrichment approach²⁸ to assess a composite of 41 concordant endothelial transcripts and 31 concordant EndMT transcripts, which identified a progressive ablation of mesenchymal cell fate and restoration of endothelial identity during ERG recovery (Fig. 7c). Notably, several endothelial genes including PECAM1, PLEKHG1, and TIE1 were commonly downregulated with ERG loss in mice and teloHAECs, while EndMT genes including CD44, COL8A1, and FN1 were commonly upregulated (Fig. 7d–g). Moreover, expression of the EndMT transcriptional regulators SNAI2, TWIST2, and ZEB2 and the TGF-β signaling genes TGFB1, TGFB2, and TGFBI paralleled a progressive reversal of EndMT composite score with ERG restoration (Fig. 7e, g, h).

We next examined our ATAC-seq dataset to investigate ERG-mediated chromatin remodeling that contributed to the reversal of EndMT transcriptional programs. Consistent with our RNA-seq data, ERG siRNA treatment reduced chromatin accessibility (i.e., peak height) of ERG-bound regions near the promoters of endothelial genes including PECAM1 and TIE1, which returned to Ctrl siRNA levels by day 7 of ERG recovery (Fig. 7i). Moreover, ERG loss increased chromatin accessibility of fibroblast-associated regions in EndMT genes including CD44 and ZEB2, which similarly returned to Ctrl siRNA levels during ERG restoration (Fig. 7j). Given our observation of ETS motif variability in human coronary plaque and previous associations between ERG and ETS and AP-1 motifs in experimental models^12,13, we performed a gene-level analysis of motif enrichment using TOBIAS⁴¹ to characterize ETS and AP-1 motif binding in endothelial and EndMT genes (Fig. 7e). In line with transcript expression, we observed decreased ETS binding in endothelial genes and increased AP-1 binding in EndMT and TGF-β genes with ERG siRNA treatment, both of which approached Ctrl siRNA levels during ERG recovery. Finally, we assessed enrichment of individual motifs among concordantly downregulated endothelial genes (Fig. 7k) and upregulated EndMT genes (Fig. 7l) using HOMER⁴². Notably, the top Ctrl siRNA-enriched motifs including ELF3, ETS1, and ETV2 belonged to the ETS family and the top ERG siRNA-enriched motifs including AP-1, FOS, and JUN-AP1 belonged to the AP-1 family. Both ETS and AP-1 motif enrichment rapidly decreased with ERG recovery and was comparable between Ctrl and ERG siRNA-treated ECs by day 3, which suggested that early correction of chromatin remodeling is an initiating event in EndMT reversal. Collectively, these findings implicate ERG restoration as a mechanism to enforce EC identity and mitigate endothelial dysfunction in disease processes that feature EndMT.

Discussion

Arterial ECs protect against atherosclerosis, but exposure to cardiovascular risk factors, such as disturbed flow and intimal lipid accumulation can lead to the loss of endothelial homeostasis and acquisition of altered cellular phenotypes including EndMT^1–5. How this change in cell state occurs across atheroprogression and its functional consequences for plaque development remain incompletely understood. Our investigation of ERG, a key enforcer of endothelial identity that restricts EndMT in ECs^20,21, revealed a marked exacerbation of atherosclerosis with ERG loss that was not constrained to atherosusceptible regions of the vasculature. Endothelial Erg knockout in murine lesions induced a profound loss of EC identity that preceded mesenchymal transition and widespread migration of ECs into the atherosclerotic intima. EndMT programs induced by ERG loss also modeled the natural course of endothelial dysfunction that occurred in both mouse and human plaques, where reduced ERG accessibility and expression were discovered. Moreover, reintroduction of ERG in ECs corrected pathological chromatin remodeling that induced EndMT, driving opening of ETS motifs and closure of AP-1 motifs, and suppression of TGF-β signaling.

Although initial observations of EndMT in atherosclerosis described an “end-state” for mesenchymal cells of EC origin in plaque⁴, more recent studies have leveraged single-cell technologies to characterize an EndMT continuum that spans the disruption of endothelial identity and the acquisition of mesenchymal phenotypes^43–45. Herein, we developed an inducible model of augmented EndMT by ERG loss in murine atherosclerosis and used single-cell transcriptomics of lineage-traced ECs to comprehensively profile endothelial dysfunction from early to established lesions. Our study identified a transitory dedifferentiated state for arterial ECs where global endothelial identity was downregulated prior to the robust induction of mesenchymal transition and the migration of ECs into plaque. Further, we discovered an acute and marked increase in endothelial permeability with Erg knockout in the absence of a pro-atherosclerotic stimulus, which confirmed that core EC functions including endothelial barrier are actively maintained by ERG-dependent transcriptional regulation. Indeed, previous work has shown that combined deletion of ERG and the related ETS factor Friend leukemia integration 1 (FLI1) induces systemic vasculopathy and multiple organ failure due to hyperpermeability and coagulability of the endothelium¹³, although loss of ERG or FLI1 alone elicited only modest dysfunction and specific organ deficits^20,46. Our investigation of Erg knockout in normocholesterolemic mice supported these observations, where we observed that despite alterations in endothelial barrier, ERG^EC-KO mice survived to 6 months post-induction without severe morbidity, mortality, or observable aortic lesions. While FN1 was induced in luminal ECs in this model, these cells did not ingress into the vessel wall, suggesting that hypercholesterolemia is required for the acquisition of a migratory EndMT phenotype. In contrast, loss of ERG in hypercholesterolemic mice promoted unrestricted atherosclerosis and EC migration in both atheroprone and atheroresistant regions of the aorta, indicating that focused disruptions of endothelial identity (e.g., Erg knockout) may establish a vulnerability of the endothelium to more profound dysfunction in states of disease. Notably, the enforcement of endothelial identity through induced expression of ERG and the SRY-box transcription factor SOX17 was sufficient to reprogram mesenchymal cardiac fibroblasts into arterial endothelial-like cells in both in vitro and in vivo models⁴⁷, which highlights the utility of key EC transcription factors as safeguards to restrict EndMT and sustain or restore homeostatic function of the perturbed endothelium.

While complete loss of ERG is embryonically lethal in mice⁴⁸, single nucleotide polymorphisms in humans have been identified that alter the regulatory activity of ERG²⁴ and confer increased genetic risk for vascular diseases including aortic dissection and abdominal aortic aneurysm^23,49, where atherosclerosis has a pathogenic role⁵⁰. ERG downregulation has also emerged as a defining feature of endothelial dysfunction in chronic liver fibrosis^20,51, suggesting that ERG may be a viable therapeutic target to mitigate diseases where EndMT contributes. Recent studies have described partial states of EndMT in atherosclerotic plaque¹⁰ and atheroprone regions of arteries¹¹, which indicate that some endothelial identity programs are retained by EndMT cells during atheroprogression and may present an opportunity for EndMT reversibility. Likewise, our single-cell transcriptomic profiling of aortic intimal cells revealed that ECs from ERG^Ctrl and ERG^EC-KO mice in both early and established lesions clustered separately from mesenchymal cell types including SMCs and fibroblasts, despite profound changes in function and localization within plaque. EC-lineage cells in human plaque also retained observable, albeit reduced, expression of ERG and expressed the EC marker CD31, lending support to observations of partial EndMT in atherosclerosis. As a principal regulator of endothelial identity with multi-faceted contributions to EC homeostasis (e.g., repression of inflammation and support of angiogenesis and barrier)^12,19, ERG is well-positioned to reverse or stabilize EndMT and limit atheroprogression. Our present investigation also predicted key interactions between ECs and other aortic intimal cells that were established with loss of ERG, including SDC4-mediated induction of EC migration and CD44-mediated expansion of macrophages in lesions, which may be considered for treatment strategies that promote healthy cell-cell communication to restrict plaque development. In a proof-of-concept study that delivered exogenous ERG to the arterial endothelium, ERG replenishment was shown to markedly decrease EndMT and vascular remodeling in murine pulmonary arterial hypertension²². Herein, we provide evidence that EndMT is a reversible process in cultured arterial ECs, during which ERG can be leveraged to restore chromatin accessibility and transcriptional programs that support homeostatic EC function. Augmentation of ERG or downstream ERG signaling in ECs may hold similar promise for limiting or reversing atherosclerosis, although timing of intervention must consider the potential difficulties of restoring function to EndMT cells that have delaminated from the endothelium and adopted a distinct physiological role in plaque.

There are some limitations of the work presented herein. Histologically, we were unable to quantify the necrotic cores of plaques as increased cellular density in the vessel wall of ERG^EC-KO mice prevented demarcation of an acellular region within lesions. Instead, plaque cellularity was quantified⁵² in the aortic arch and brachiocephalic artery to provide insight prior to our comprehensive scRNA-seq analysis. More detailed histological examination – the composition of the extracellular matrix, for example – should be considered in future work given that loss of ERG has been linked clinically to aortic aneurysm²⁴. Moreover, in evaluating the cell-cell interactions derived from scRNA-seq, we did not include ERG^EC-KO mice fed a normal chow diet. While this may have limited some comparisons, there were importantly no marked phenotypic changes with Erg knockout in the absence of hypercholesterolemia (i.e., we did not observe differences in plaque formation, EC migration, gross morbidity, or mortality with Erg knockout in normocholesterolemic mice). Finally, we acknowledge that there are limitations to the CreERT2 model, with one recent research letter demonstrating that Cre itself has effects on endothelial phenotypes in perinatal retinal angiogenesis⁵³, with the authors suggesting the use of unfloxed Cre mice as controls. Importantly, we controlled for Cre expression by performing the majority of our experiments in Cdh5^CreERT2 mice containing an unfloxed versus floxed Erg allele, although we did not assess how Cre expression alone might impact plaque development.

In summary, we have identified ERG as a safeguard against endothelial identity loss and dysfunction in atherosclerosis that mitigates disruption of endothelial barrier, intraplaque migration of ECs, and accumulation of myeloid cells in lesions. Our establishment of ERG loss and EndMT as etiologic drivers of atheroprogression provide opportunities for future therapeutic approaches that enforce arterial EC identity to limit plaque development in atherosclerotic vessels.

Methods

Erg knockout mice

The Cdh5^CreERT2; Rosa26^{lox-STOP-lox-Sun1-sfGFP}; Erg^fl/fl mice used in this study were generated at the University of Virginia School of Medicine (VA, USA). This strain was derived from the following mice: Tg(Cdh5-cre/ERT2)^1Rha mice (MGI 3848982) obtained from Dr. Ralf Adams (Max Planck Institute for Molecular Biomedicine, Germany)⁵⁴; Rosa26-Sun1-2xsfGFP-6xMyc mice (B6;129-Gt(ROSA)26Sor^{tm5(CAG-Sun1/sfGFP)Nat}/J; Jackson Laboratory 021039); and floxed Erg mice that were derived from Erg^{tm1a(KOMP)Wtsi} mice (MGI 4455490) as previously described¹⁵. Unfloxed and floxed Erg mice on Apoe^+/+ and Apoe^{tm1Unc/tm1Unc} backgrounds were generated by breeding Cdh5^CreERT2; Rosa26^{lox-STOP-lox-Sun1-sfGFP}; Erg^fl/fl mice with ApoE-deficient mice (B6.129P2-Apoe^tm1Unc/J; Jackson Laboratory 002052) and selecting for desired genotypes. Mice were bred and housed under pathogen-free conditions in cages containing up to 5 animals on vented racks with 12/12 h light/dark cycles at 21-22 °C and 35-40% relative humidity at the University Health Network Animal Resources Center (ARC). Sex- and age-matched mice were randomly allocated for experiments. Investigators were not blinded to group allocation, but the analysis of results was performed blinded when possible (e.g., image quantification). Sex-specific effects were not observed for initial experiments in this study and were not evaluated separately thereafter. Sample sizes were determined empirically based on experience in our research group or previously published studies in similar areas of research. Experiments were conducted in accordance with the current Canadian Council on Animal Care guidelines and with institutional research ethics board approval at the University Health Network (ARC Animal Use Protocol 2291).

Erg knockout was induced in male and female 6–8 week-old mice using five consecutive daily intraperitoneal injections of 2 mg tamoxifen (Sigma-Aldrich T5648) in 200 µL corn oil (Sigma-Aldrich C8267). For atherosclerosis experiments, mice were allowed one week of recovery before induction of hypercholesterolemia. Hypercholesterolemia was induced with a single tail vein injection of AAV encoding mutant mPCSK9 (rAAV8/D377Y-mPCSK9; Vector Biolabs 7600; 1 × 10¹¹ viral genomes in 200 µL PBS-/-) and replacement of normal chow with ad libitum semi-synthetic HCD containing 1.25% cholesterol and 40% fat (Research Diets D12108C). Apoe^{tm1Unc/tm1Unc} mice and respective controls were placed on HCD without injection of AAV8-PCSK9 to induce hypercholesterolemia.

Peripheral blood was collected from mice using saphenous vein puncture. Mice used for scRNA-seq and permeability experiments were euthanized with CO₂ using a flow rate of 40% of chamber volume per minute. All other mice were placed on 3% isoflurane with 2 L/min O₂ to allow for terminal blood collection using cardiac puncture. Mice were perfused with 10 mL PBS-/- containing 10 units/mL heparin (Heparin Leo DIN 00453811) prior to tissue collection. Plasma was isolated from blood samples after centrifugation at 1500 g for 15 min at 4 °C. Aorta and BCA tissue were fixed for 1 h in 4% PFA (Electron Microscopy Sciences 15710), cleaned, embedded in O.C.T compound (Fisher Healthcare SGN4585), and stored at -80 °C.

Human carotid artery tissue collection

Human carotid artery biospecimens were obtained at the Peter Munk Cardiac Center (ON, Canada) from patients undergoing carotid endarterectomy (CEA) that provided written informed consent for research studies as previously described⁵⁵, and according to established clinical guidelines (n = 2 unique samples; median age 81; 50% female)⁵⁶. Carotid plaques were suspended in Hanks Balanced Salt Solution (Thermo Fisher Scientific 24020117) immediately after CEA, fixed in 4% PFA (Electron Microscopy Sciences 15710) for 24 h, embedded in Leica Surgipath Paraplast paraffin (Leica Microsystems 39602004), and stored at 4 °C. All human carotid artery research complies with ethical guidelines for human subjects research under approved REB protocols at the University Health Network (no. 18-5282), for the procurement and use of human tissues and information.

Human coronary artery tissue collection

Ischemic and non-ischemic explanted human coronary artery tissue biospecimens from diseased heart transplant donors and normal coronary artery biospecimens from non-diseased donor hearts rejected for orthotopic heart transplantation were obtained with written informed research consent at Stanford University (CA, USA) as previously described (n = 119 unique samples; median age 54; 35% female)^39,57. All normal arteries originated from hearts with left ventricular ejection fraction greater than 50%. All human coronary artery research herein complies with ethical guidelines for human subjects research under approved IRB protocols at Stanford University (no. 4237 and no. 11925) and the University of Virginia (no. 20008 and no. 302407), for the procurement and use of human tissues and information, respectively.

ERG knockdown and recovery in human aortic endothelial cells

ERG silencing by siRNA was performed in teloHAECs (Thermo Fisher Scientific 50-238-1690) grown at 37 °C and 5% CO₂ in Endothelial Cell Growth Medium MV2 (ECGM-MV2; PromoCell C-22121) on rat tail collagen type 1-coated plates (Sigma-Aldrich C3867). ECGM-MV2 was supplemented with 5% fetal calf serum, 5 ng/ml recombinant human epidermal growth factor, 0.5 ng/ml recombinant human vascular endothelial growth factor 165, 10 ng/ml recombinant human basic fibroblast growth factor, 20 ng/ml long R3 insulin-like growth factor-1, 1 μg/ml ascorbic acid, and 0.2 μg/ml hydrocortisone (from ECGM-MV2 Supplement Pack; PromoCell C-39221). Experiments were performed on cells between passage 11 and 20.

At 60-70% confluence, cells were serum starved for 1 h in Opti-MEM Reduced Serum Medium (Thermo Fisher Scientific 31985-070) and transfected in ECGM-MV2. Lipofectamine RNAiMAX (Invitrogen 56532) and a final concentration of 40 nM Ambion™ Silencer™ Select ERG siRNA (Thermo Fisher Scientific 4392421–s4813) or negative control siRNA (Thermo Fisher Scientific 4390844) was used to transfect cells. Media was changed at 4 h post-transfection and subsequently at every 48 h. Cells were cultured for 10 days post-transfection to assess restoration of ERG with dilution of siRNA from cell proliferation. Collection of RNA and cell chromatin occurred at specified timepoints from day 0 (72 h post-transfection) to day 7 of ERG recovery. Biological replicates were performed on separate days.

Whole aorta imaging

Mouse aortas were harvested, cleaned and pinned, and imaged with a Leica MZFLIII Stereo Microscope (Leica Microsystems, Germany) using an exposure time of 200 ms. Total and descending aorta plaque area (thoracic aorta to renal bifurcation) was quantified in ImageJ (v1.54k)⁵⁸ by thresholding images using minimum pixel intensity. Aortas and aortic arches were stained with oil red O (ORO) to visualize lipid. Briefly, aortas were stained in a 24-well plate using 0.31% ORO in 60% isopropanol for 60 min at room temperature with shaking, destained using 60% isopropanol for three cycles of five min at room temperature, and washed in PBS-/- for 10 min at 4 °C. Residual ORO precipitate was removed from aortas, and tissue was prepared en face for imaging. ORO area and integrated density were quantified by thresholding composite images containing green and blue channels using maximum pixel intensity. To assess plaque distribution in the aortic arch, images of en face tissue were segmented every 250 µm from the midline (lesser curvature) to the greater curvature and ORO area was quantified.

Hematoxylin and eosin and Movat pentachrome staining

O.C.T.-embedded mouse aorta and BCA tissue were sectioned at a thickness of 10 µm using a Leica CM1950 Cryostat (Leica Microsystems, Germany). Tissue was stained with hematoxylin and eosin (H&E) or Movat pentachrome stain according to standard protocols in the Pathology Research Program at the University Health Network (ON, Canada). Cross sections were imaged with a Leica Aperio AT2 slide scanner (Leica Microsystems, Germany) and quantified in ImageJ (v1.54k)⁵⁸. Plaque intimal coverage was calculated as a percentage of the total luminal circumference of cross sections. Plaque cells and medial elastin were quantified by thresholding images to retain cells and elastin content⁵⁹. Plaque cap thickness was determined by obtaining the average of three independent thickness measurements. Medial elastin breaks were counted manually by a blinded observer.

Immunofluorescence staining

Mouse aorta and BCA cross sections were warmed to room temperature for 30 min for immunofluorescence staining. Excess O.C.T. compound was removed and samples were stained using the following protocol at room temperature unless otherwise specified and with three PBS-/- washes in between each step: permeabilization with 0.5% Triton X-100 (BioShop TRX777.500) in 0.1 M glycine (BioShop GLN002.5) in PBS-/- for 15 min; incubation with TrueBlack Lipofuscin Autofluorescence Quencher (Biotium 23007) for 30 seconds; blocking in TNB buffer (Akoya Biosciences FP1012) with 5% donkey serum (Sigma-Aldrich D9663) for 15 min; incubation with primary antibodies in TNB buffer with donkey serum overnight at 4 °C; incubation with secondary antibodies in TNB buffer with donkey serum for 1 h; incubation with Vector TrueVIEW Autofluorescence Quenching reagent (Vector Laboratories SP-8400-15) for 5 min; incubation with 40 µM Hoechst 33342 in PBS-/- (Thermo Fisher Scientific 62249) for 25 min; mounting using VECTASHIELD Vibrance Antifade Mounting Medium (Vector Laboratories H-1700) for imaging. Autofluorescence quenching kits were used according to manufacturer recommended protocols. The following primary antibodies were used with a 1:100 dilution unless otherwise specified: anti-human/mouse ACTA2-Cy3 (Sigma-Aldrich C6198; 1:1000), anti-mouse CD68-AF647 (BioLegend 137003), anti-mouse ERG (Abcam ab133264), anti-human/mouse ERG (Abcam ab110639), anti-human/mouse FN1 (Abcam ab2413), and anti-GFP (Rockland Immunochemicals 600-101-215). The following secondary antibodies were used with a 1:200 dilution: anti-goat AF488 (Jackson ImmunoResearch 705-546-147), anti-rabbit AF647 (Jackson ImmunoResearch 711-606-152), and anti-rabbit Cy3 (Jackson ImmunoResearch 711-166-152). Aorta and BCA tissue was imaged with a Leica TCS SP8 confocal microscope (Leica Microsystems, Germany) using a line average of 1 and images were merged manually in Inkscape (v1.4). Plaque area and cells were quantified in ImageJ (v1.54k)⁵⁸ by thresholding images to retain cells⁵⁹. GFP⁺ACTA2⁺ cells were counted manually by a blinded observer following thresholding for GFP⁺ nuclei.

Paraffin-embedded human carotid tissue was sectioned at a thickness of 7 µm using an Epredia HM 325 Rotary Microtome (Epredia, USA). Samples were deparaffinized with immersion in xylene for three sequential washes, each for 3 min, and rehydrated with three sequential washes of absolute ethanol, with 10 dips each, followed by immersion in 95% ethanol for 10 dips and rinsing under water. Cross sections were incubated with 10% citrate buffer (Sigma-Aldrich C9999) for 15 min at 95 °C for antigen retrieval. Samples were cooled for 30 min at room temperature and stained using the above protocol. The following primary antibodies were used with a 1:100 dilution unless otherwise specified: anti-human/mouse ACTA2-Cy3 (Sigma-Aldrich C6198; 1:1000), anti-human CD31 (Synaptic Systems HS-351 004), and anti-human/mouse ERG (Abcam ab110639). The following secondary antibodies were used with a 1:200 dilution: anti-guinea pig AF488 (Jackson ImmunoResearch 706-545-148) and anti-rabbit AF647 (Jackson ImmunoResearch 711-606-152). Tissue was imaged with a Leica TCS SP8 confocal microscope (Leica Microsystems, Germany) using a line average of 1. Plaque cell mean fluorescence intensity (MFI) and integrated density were quantified in ImageJ (v1.54k)⁵⁸ by thresholding images to retain cells⁵⁹. Quantification of immunofluorescence was based on 69-124 cells obtained from 7 distinct plaque regions from 2 patients with CEA.

Clinical chemistry

Mouse plasma was analyzed in aliquots of 34-45 µL using a Beckman Colter AU480 Chemistry Analyzer (Beckman Colter, USA) at The Center for Phenogenomics in Toronto (ON, Canada) to quantify plasma cholesterol, triglycerides, and glucose.

Flow cytometry

Peripheral blood was aliquoted for leukocyte counting and assessment with flow cytometry. For leukocyte counting, 10 µL of each blood sample was diluted 1:20 with 1x RBC lysis buffer (BioLegend 420301), triturated, incubated for seven min on ice, and enumerated to obtain the average of two independent counts. The remaining blood volume was added to 1 mL 1x RBC lysis buffer, triturated, incubated for 12 min on ice, and neutralized with 10 mL PBS-/-. Samples were centrifuged at 300 g for 10 min at 4 °C and supernatant was discarded. Cells were resuspended in 100 µL ice-cold FACS buffer (1x PBS-/-, 0.5% BSA), vortexed, and 100 µL of cells were added to flow tubes. Samples were incubated with primary antibodies for 30 min on ice and fixed with Cytofix fixation buffer (BD Biosciences 554655) prior to flow cytometry analysis. The following primary antibodies were used with a 1:300 dilution: anti-mouse B220-PE-Cy7 (BD Biosciences 552772), anti-mouse CD115-BV711 (BioLegend 135515), anti-mouse CD11b-BV605 (BD Biosciences 563015), anti-mouse CD3-FITC (BD Biosciences 555274), anti-mouse CD45-BV650 (BD Biosciences 563410), anti-mouse Ly6C-AF700 (BioLegend 128024), anti-mouse Ly6G-APC-Cy7 (BD Biosciences 560600), and anti-mouse SiglecF-PerCP-Cy5.5 (BD Biosciences 565526). Data were acquired with an LSR II Flow Cytometer (BD Biosciences, USA) using BD FACSDiva (v6.1.3) and analyzed with BD FlowJo (v10.8.1). The gating strategy used for identifying leukocyte populations is depicted in Supplementary Fig. 1e.

Permeability assays

Mice received 40 µg/g bodyweight of 70,000 MW TRITC-dextran by tail vein injection (Thermo Fisher Scientific D1818) following induction of Erg knockout and one week of recovery. Tissue was harvested at 10 min post-injection to assess endothelial permeability. Mice were perfused with 10 mL PBS-/- containing 10 units/mL heparin (Heparin Leo DIN 00453811) followed by 4% PFA (Electron Microscopy Sciences 15710). Aortic arch and BCA tissue was harvested, fixed in 4% PFA for one h, and cleaned for immunofluorescence staining. Tissue was stained for immunofluorescence imaging as described above, but with the omission of permeabilization and autofluorescence quenching, prepared en face, and mounted using Dako fluorescence medium (Agilent S302380-2). The following primary antibodies were used with a 1:100 dilution: anti-mouse CD45-AF647 (BioLegend 103124) and anti-GFP (Rockland Immunochemicals 600-101-215). The following secondary antibodies were used with a 1:200 dilution: anti-goat AF488 (Jackson ImmunoResearch 705-546-147).

Aortic arch and BCA tissue were imaged with a Leica TCS SP8 confocal microscope (Leica Microsystems, Germany) using a line average of 1 to obtain tiled 4 µm Z-stacks with a step size of 1.33 µm. Image tiles were merged with Leica Application Suite X (v3.7.4.23463) using statistical overlap blending with linear blending. Image Z-stacks were merged in ImageJ (v1.54k)⁵⁸ by summing slices and TRITC-dextran MFI and integrated density and CD45⁺ area and cells were quantified by thresholding images to retain TRITC-dextran and cells⁵⁹. The lesser curvature of the aortic arch was demarcated by the presence of CD45⁺ cells.

For the LDL permeability assay in early atherosclerosis, mice received 1 µg/g bodyweight of human DiI-labeled LDL by tail vein injection (Kalen Biomedical 770230-9) following induction of Erg knockout, one week of recovery, injection of AAV8-PCSK9, and one week of HCD. Tissue was harvested at 16 h post-injection to assess endothelial permeability. Mice were perfused and aortic arch and BCA tissue was harvested, stained, imaged, and quantified as described above.

Single-cell RNA sequencing

Isolation and sequencing of mouse aortic intimal cells

Aortic intimal cell isolates were obtained as previously described with minor modifications to established protocols^60,61. Briefly, aortas were harvested from the aortic arch to renal bifurcation, cleaned, and pinned en face, and cells were isolated following gentle digestion with 0.15 W units/mL collagenase D (Sigma-Aldrich 11088858001) and 167 units/mL DNAse I (Thermo Fisher Scientific EN0521) in PBS-/- for 10 min at 37 °C. Six male and female mice were pooled per sample. Samples were prepared as described in the 10x Genomics Chromium Next GEM Single Cell 3’ Reagent Kit v3.1 user guide and sequenced paired-end using the 10x Genomics Chromium platform (10x Genomics, USA).

Analysis of mouse and human single-cell RNA sequencing

To allow for EC lineage tracing in aortic intimal samples, a custom reference genome was generated by combining the GRCm38 mouse genome assembly (refdata-gex-mm10-2020-A) and the 5215 base-pair reporter sequence containing Sun1-sfGFP, 6x c-Myc tags, and a bGH poly(A) signal. A reporter exon containing sfGFP and the 6x c-Myc tags was defined from base-pair 2821 to 4560 of the reporter sequence. The full reporter sequence can be found in Supplementary Table 1. Samples were demultiplexed and aligned to the custom reference genome using 10x Genomics Cell Ranger (v7.2.0)⁶² and each sample was subsetted to retain the top 10,000 cells by unique molecular identifier (UMI) count. The resulting alignments contained an average of 25,899 mean reads per cell, 624 median genes per cell, and 24,277 total detected genes.

Gene expression matrices were filtered, normalized, and clustered using R (v4.4.1; http://r-project.org/) and Seurat (v5.1.0)⁶³. UMIs that were detected in < 3 cells and cells with ≤ 500 UMI counts or cells with < 200 or ≥ 3000 UMIs were excluded. Matrices were then filtered to remove cells with ≥ 40% mitochondrial transcripts after confirming scRNA-seq data integrity across a range of mitochondrial cut-offs from 25-55% in line with published methods^61,64. Between 8903 and 9956 cells were retained for each sample (38,362 cells total). Normalization was performed using the LogNormalize method with a scale factor of 100,000. Principal component analysis (PCA) was performed on the top 2000 variable features that were identified using the vst selection method. Cells were clustered using the Louvain algorithm based on the first 17 PC dimensions with a resolution of 0.6. Cell cluster identities were manually assigned based on expression of classical marker genes, and two red blood cell clusters were removed from the dataset. Cells were reclustered using the same parameters, and the dataset was integrated using Harmony (v1.2.3)⁶⁵ with automatic detection of the ridge regression penalty parameter (lambda = NULL). Following integration, cells were reclustered as above with a resolution of 0.8 and cell cluster identities were reassigned. UMAP was used for non-linear reduction and two-dimensional data visualization. Gene expression was visualized with violin plots using ggplot2 (v3.5.2; https://ggplot2.tidyverse.org/). Differential gene expression (DGE) analysis between clusters was performed using two-sided Wilcoxon rank-sum tests with Bonferroni correction in Seurat (adjusted p value < 0.05 denoting significance). Pearson correlation coefficients were generated using ggpubr (v0.6.0; https://rpkgs.datanovia.com/ggpubr/).

To profile ECs with Sun1-sfGFP reporter expression, the integrated dataset was subsetted to retain EC clusters, reclustered as above with a resolution of 0.2, further subsetted to retain high-quality GFP-enriched ECs (EC 1’ and EC 2’ clusters) with a UMI count > 1000, and reclustered with a resolution of 0.35. Unintegrated cluster identities were used for analysis of the resulting subset to avoid overcorrection of transcriptional profiles and preserve biological variability between sample types.

Human carotid endarterectomy scRNA-seq data were obtained from a publicly available repository³⁵ and analyzed using the above methods with minor modifications. UMIs that were detected in < 3 cells and cells with ≤ 500 UMI counts or cells with < 200 or ≥ 4000 UMIs were excluded. Data was then randomly downsampled to the number of cells in the smallest sample, which generated a subset containing 2813 cells per sample (16,878 cells total). Cells were clustered as above with a resolution of 0.3, subsetted to retain EC clusters, and reclustered with a resolution of 0.1. DGE analysis between ECs obtained from the atherosclerotic core and proximal adjacent regions was performed as above.

Bulk RNA sequencing

RNA extraction from cultured human aortic endothelial cells

RNA was extracted from cell lysates in 350 µL Buffer RLT using an RNeasy^® Micro Kit (Qiagen 74004) according to manufacturer’s instructions with the following modifications: i) after adding 350 µL of 70% ethanol and centrifuging samples in RNeasy MinElute Spin Columns for 15 seconds at 8000 g, the flowthrough was reapplied to the spin columns and samples were centrifuged for an additional 15 sec at 8000 g; ii) when adding 350 µL of Buffer RW1, samples were incubated at room temperature for 5 min, centrifuged for 15 sec at 8000 g, and flowthrough was discarded; and iii) after adding 14 µL of RNase-free water to the spin columns, samples were incubated at room temperature for 1 min and centrifuged for 1 min at 8000 g. RNA flowthrough was reapplied to spin columns, samples were incubated at room temperature for an additional 1 min, and centrifuged for 1 min at 8000 g to elute RNA.

Library preparation and sequencing

RNA extracted from Ctrl and ERG siRNA-treated teloHAECs was sent to Novogene (CA, USA) for library preparation and sequencing. Samples were assessed for quantity, integrity, and purity using an Agilent 5400 Fragment Analyzer System (Agilent, USA). During library preparation, mRNA was purified from total RNA using poly-T oligo-attached magnetic beads. After fragmentation, the first strand cDNA was synthesized using random hexamer primers, followed by second strand cDNA synthesis. Libraries were complete after end repair, A-tailing, adapter ligation, size selection, amplification, and purification. Each library was quantified with a Qubit^TM DNA High Sensitivity assay (Thermo Fisher Scientific, USA) and fragment sizes were analyzed on the Agilent Bioanalyzer using the High Sensitivity DNA assay prior to sequencing. Paired-end 150 bp sequencing was performed on an Illumina NovaSeq 6000 System (Illumina, USA).

Read preprocessing and differential expression analysis

Raw RNA-seq reads were processed with Trimmomatic (v0.32; using recommended parameters) to remove adaptor sequences and preserve high-quality reads. The trimmed reads were then aligned to the human genome reference assembly (GRCh37/hg38) using STAR (v2.5.1b; default parameters). Quality control metrics for raw and aligned RNA-seq reads were analyzed using FastQC, MultiQC (v1.3), and Picard. After validating sample quality, feature quantification for each sample was performed using Salmon (v1.09) against a transcriptome derived from hg38.

DE analysis was performed using DESeq2 (v1.44.0) in R (v4.4.1) and examined Ctrl and ERG siRNA-treated teloHAECs at day 0, 3, 5, and 7 (4 groups). The DESeq2 object was filtered to retain transcripts with at least 10 counts per sample in the number of samples equal to the size of the smallest group (4 samples). DE comparisons were conducted using default parameters (adjusted p value < 0.05 denoting significance; two-sided Wald tests; Benjamini-Hochberg correction).

Pathway enrichment analysis

Pathway enrichment analysis was conducted using clusterProfiler (v4.12.6)⁶⁶ to predict GO biological process pathways associated with DE genes (org.Mm.eg.db database; adjusted p value < 0.05; log₂ fold-change < -0.5 or > 0.5; one-sided Fisher’s exact tests; Benjamini-Hochberg correction). Pathways with an adjusted p value < 0.05 were considered significant and assigned to categories based on the inclusion of curated pathway terms. Complete lists of terms for each pathway category are available in Supplementary Table 2. Pathway category radar plots were generated using ggradar (v0.2). Disease enrichment was performed with DE genes using enrichR (v3.4)⁶⁷ and the DisGeNET (v24.3)⁶⁸ knowledge database containing 9828 terms, 17,464 genes, and 63 terms per gene (adjusted p value < 0.05; one-sided Fisher’s exact tests; Benjamini-Hochberg correction). Enrichment scores were generated by multiplying -log₁₀ p values with the sign of log₂ fold-changes.

Cell-cell communication analysis

Cell-cell communication was predicted between high-quality GFP-enriched ECs and other cell clusters using CellChat (v2.1.2)⁶⁹ with default parameters. Differential LR interaction analysis was performed between ERG^Ctrl and ERG^EC-KO datasets in early and established atherosclerosis using a threshold p value < 0.05, a ligand p value < 0.05, and a ligand log₂ fold-change < -0.5 or > 0.5 (two-sided Wilcoxon rank-sum tests; Bonferroni correction). LR transcript expression was visualized with violin plots using plotGeneExpression in CellChat.

Generation of cell type composite scores

Composite scores for EC and EndMT transcripts were produced with a gene signature scoring method based on the Mann-Whitney U statistic using UCell (v2.8.0)²⁸. EC or EndMT gene lists were generated by obtaining total elevated single-cell markers for ECs or fibroblasts from the Human Protein Atlas (v23.0; https://www.proteinatlas.org/)⁷⁰, respectively, and supplementing lists with manually curated genes known to be involved in EC function or EndMT. Complete EC and EndMT gene lists are available in Supplementary Table 3. For mouse scRNA-seq data, all DE genes within each gene list were used to generate composite scores. For teloHAEC bulk RNA-seq data, all concordant DE genes between ERG^EC-KO mice and ERG KD teloHAECs within each gene list were used to generate composite scores (i.e., genes downregulated in both datasets and genes upregulated in both datasets). Composite scores between groups were compared using a Kruskal-Wallis test and Dunn’s multiple comparisons test with Bonferroni correction using rstatix (v0.7.2; https://rpkgs.datanovia.com/rstatix/). An adjusted p value < 0.05 denoted significance.

Pseudotime analysis

Pseudotime analysis was conducted for the EC subset using Monocle 3 (v1.3.7)⁷¹. Data was preprocessed with a dimensionality of 100, clustered with a resolution of 5 × 10^-3, and subsetted to retain cells contained within a pseudotime trajectory in established plaque that progressed from ERG^Ctrl to ERG^EC-KO mice. The data subset was preprocessed and reclustered as above and a root node was assigned to denote the beginning of the trajectory in ERG^Ctrl ECs. Transcriptome modules were identified with graph-autocorrelation analysis using an adjusted p value < 0.05 and clustering at a resolution of 5 × 10^-4 (Moran’s I statistic; Benjamini-Hochberg correction).

Integration of single-cell RNA sequencing with genome-wide association study data

GWAS single nucleotide polymorphisms (SNPs) associated with diseases of interest and with pre-computed p-values were obtained from the NHGRI-EBI GWAS Catalog³⁴ for integration with ERG^EC-KO-enriched transcripts in established murine atherosclerosis (week 12). For cases where multiple SNPs mapped to the same gene, the SNP with the minimum p-value was plotted. Enrichment scores were generated by multiplying -log₁₀ p values with the sign of log₂ fold-changes.

Assay for transposase-accessible chromatin with sequencing in human coronary artery tissue

Isolation of coronary artery cell nuclei and sequencing

Approximately 25 mg of frozen human coronary artery segments were used to isolate nuclei following the Omni-ATAC protocol^39,72, yielding ~50,000 nuclei per sample per transposition reaction for 30 min at 37 °C. Transposed DNA was purified, and bulk ATAC-seq libraries were generated by PCR amplification using Nextera index barcodes. Bulk ATAC-seq libraries were sequenced on an Illumina NovaSeq 6000 (Illumina, USA). After sequencing, samples were processed using the PEPATAC pipeline (v0.8.6)⁷³, including hg38 genome alignment, removal of duplicates and mitochondrial reads, and peak-calling. Processed data were then classified into three categories as previously described³⁹: 1) Non-ischemic/non-lesion (control); 2) Ischemic/non-lesion; and 3) Ischemic/lesion based on the presence of atherosclerotic lesions and a primary diagnosis of ischemic or nonischemic cardiomyopathy.

Differential chromatin accessibility analysis

Using the sample disease classifications, DESeq2 (v1.45.1)⁷⁴ was used to define DARs between control and lesion containing coronary artery tissues (category 1, 2, and 3). Briefly, a counts matrix was derived from PEPATAC-processed coronary artery bulk ATAC-seq data and default parameters were used for DESeq2, setting a log₂ fold-change cutoff of 0.45 and a p value of 1 × 10⁻³ (two-sided Wald tests; Benjamini-Hochberg correction).

Single-nucleus assay for transposase-accessible chromatin with sequencing annotation

Bulk ATAC-seq DARs were queried for cell type specific open chromatin signals using previously generated single-nucleus ATAC-seq data³⁹. Previously annotated pseudobulk signals, stored as BigWig files, were used to query cell type accessibility for candidate bulk coronary artery ATAC-seq DARs. Pseudobulk signal tracks from 11 annotated cell types were merged based on disease classification used previously: non-ischemic/non-lesion (category 1), ischemic/non-lesion (category 2), and ischemic/lesion (category 3). Tracks were visualized on the UCSC genome browser⁷⁵ along with merged cell type specific tracks. Signals were overlapped with the ENCODE human coronary artery H3K27ac chromatin immunoprecipitation sequencing track⁷⁶.

Gene and transcription factor binding site annotation

DARs were annotated with nearest protein coding genes using GenomicRanges (v1.48.0)⁷⁷ and Ensembl (v86)⁷⁸. To identify candidate transcription factor binding sites at DARs, hg38 genomic intervals were queried in the Cistrome Database (Cistrome DB; 2015)⁷⁹ using the Cistrome DB Toolkit (http://dbtoolkit.cistrome.org/) (adjusted p value < 0.05; two-sided Fisher’s exact tests with Benjamini-Hochberg correction). Motif deviations in single-nucleus ATAC-seq data from coronary artery cells were generated using chromVAR (v1.12.0)⁸⁰.

Assay for transposase-accessible chromatin with sequencing in cultured human aortic endothelial cells

Sample preparation and sequencing

Approximately 250,000 cells were harvested from each biological replicate. Cells were cryopreserved in BAMbanker freezing medium (Fujifilm cellular dynamics CS-02-001) using a Mr. Frosty (Thermo Fisher Scientific N51000001). Prior to sequencing, cells were washed with 1 mL of 37 °C ECGM-MV2 (PromoCell C-22221) and pelleted at 500 g. Free DNA from cells lysed during the cryopreservation process was removed with a 30 min incubation in ECGM-MV2 supplemented with 200 units of DNase I (Thermo Fisher Scientific EN0521) at 37 °C. Approximately 50,000 live cells were isolated from the recovered population and lysed in non-ionic detergent lysis buffer (10 mM Tris-HCl, pH 7.4, 3 mM MgCl2, 10 mM NaCl, 0.1% NP-40). Nuclei were pelleted and resuspended in 50 μL transposition reaction solution containing 2.5 μL Tn5 transposase attached to sequencing adapters (Illumina Nextera Tagment DNA enzyme TDE1, Nextera® DNA Sample Preparation Kit FC-121-1030), 25 μL Tagment DNA buffer (20 mM Tris-HCl pH 7.6, 10 Mm MgCl2, 20% Dimethyl Formamide), 16.5 μL PBS, 5 μL NFH2O, 0.1% Tween, and 0.01% Digitonin. Samples were resuspended in Tn5 mixture and incubated at 37 °C for 30 min under gentle mixing conditions (300 rpm) using a thermomixer (Eppendorf). The transposed and adapter-ligated DNA fragments were purified using the Zymo Research DNA Clean and Concentrator purification kit and PCR amplified using barcoded primers (Nextera Index Kit FC-121-1011) using cycle numbers determined by RT-qPCR. A SageSciences PippinPrep was used to size select DNA between 125 and 600 bp using a 2% agarose gel. The final ATAC-seq libraries were sequenced with a 2 × 126 bp paired-end run using the Illumina HiSeq 2500 platform to obtain ~50 million paired-end reads per sample.

Read preprocessing and differential peaks analysis

The quality of raw ATAC-seq data was assessed with FastQC (v0.11.8; http://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Reads were trimmed of adapter sequences using Trimmomatic (v0.32; default parameters). Burrows-Wheeler Aligner (BWA) (v0.7.8; default parameters) was used to align the trimmed reads to the hg38 (GRCh38) genome reference assembly. Further quality control, including assessments of the library complexity, read size distribution, and read density at the TSS were performed using ATACseqQC. Peaks were called with Genrich (https://github.com/jsh58/Genrich) using its multiple replicates, deduplication, and ATAC-seq modes.

In preparation for differential peaks analysis, a counts matrix was constructed in R (v4.4.1) using the featureCounts function from the R package Rsubread. A list of BAM files was provided as input and genomic features were defined by a union peaks file generated using the bedops --merge tool. DESeq2 was used for normalization and to call differential peaks (adjusted p value < 0.05; two-sided Wald tests; Benjamini-Hochberg correction). ChIPseeker⁸¹ was used to annotate differential peaks to genes defined by the UCSC hg38 TxDb. Chromatin accessibility genome tracks were visualized with the UCSC Genome Browser (v494; https://genome.ucsc.edu/). For aggregate analysis of motif enrichment among genes, HOMER⁴² was used to call motif enrichments of differential peaks against hg38 using the size given and –nomotif settings (adjusted p value < 0.05; one-sided Fisher’s exact tests; Benjamini-Hochberg correction).

For gene-level analysis of motif enrichment using TOBIAS (v0.14.0)⁴¹, JASPAR’s non-redundant motif set was downloaded (date of download: 26 November, 2021), subsetted to include only transcription factors that were expressed at an average of >10 transcripts per sample in our RNA-seq, and formatted using TOBIAS formatMotifs. ATAC-seq replicates were merged using Samtools (v1.21), and reads were corrected for Tn5 bias using TOBIAS ATACorrect. Footprinting scores were generated using TOBIAS FootprintScores, and footprinting was compared across conditions with TOBIAS BINDetect using the reduced transcription factor set.

Chromatin immunoprecipitation sequencing analysis

To generate an ERG chromatin immunoprecipitation (ChIP-seq) reference genome track for visualization, H3K27ac and input ChIP-seq FASTQs from human umbilical vein endothelial cells (HUVECs) treated with 24 h of Ctrl or ERG siRNA were downloaded from GSE124892¹². Preprocessing was performed as described in the ATAC-seq section above. Peaks were called using HOMER, TAG files were created for each condition using HOMER makeTagDirectory, and peaks were identified using HOMER findPeaks in histone mode against input DNA. Peaksets from both conditions were merged using HOMER mergePeaks to create a union peakfile. HOMER getDifferentialPeaks was run to identify differential peaks between conditions and Ctrl siRNA-enriched peaks were used for genome track visualization.

Statistical analysis of non-omics data

Statistical analysis for non-omics data was performed using GraphPad Prism version 10.1.1 (270) (Dotmatics, USA). Bar and line graphs were composed of 3-23 independent samples per group and were presented as mean ± standard error of the mean (SEM). For plasma cholesterol, triglyceride, and glucose and peripheral blood leukocyte measurements, mice were measured repeatedly over time. For immunofluorescence staining of human carotid plaques, quantification was performed on individual plaque cells. For all remaining data, measurements were taken from distinct mice or distinct biological replicates of teloHAECs. For mouse experiments, male data points were depicted with circles and female data points were depicted with triangles. For n < 6 per group, comparisons between groups were made using non-parametric Mann-Whitney tests with Holm-Šídák correction. For n ≥ 6 per group, data were assessed with a Shapiro-Wilk test for normality and F test (2 groups) or Spearman’s test for equality of variance ( > 2 groups). If normality and equality of variance were assumed, comparisons between groups were made using an unpaired t-test (2 groups) or a two-way ANOVA with Holm-Šídák correction ( > 2 groups). Comparisons between two groups that were normally distributed but had unequal variances were made using a Welch’s t-test. If data were not normally distributed or if for > 2 groups equal variances could not be assumed, data were log transformed, reassessed for normality and equality of variance, and analyzed using the above tests. If transformed data did not meet the criteria for an unpaired t-test, Welch’s t-test, or two-way ANOVA, untransformed data were then assessed using non-parametric Mann-Whitney tests with Holm-Šídák correction ( < 6 comparisons) or Benjamini, Krieger, and Yekutieli FDR correction ( ≥ 6 comparisons). For comparisons between two groups, a priori hypotheses for the directionality of phenotype were made, and a one-sided unpaired t-test, Welch’s t-test, or Mann-Whitney test was used. Adjusted p values < 0.05 were considered significant.

Ethics declarations

All mouse experiments were conducted in accordance with the current Canadian Council on Animal Care guidelines and with institutional research ethics board approval at the University Health Network (ARC Animal Use Protocol 2291). Human carotid artery biospecimens were obtained at the University Health Network (ON, Canada) from patients that provided written informed consent for research studies as previously described⁵⁵ and according to established clinical guidelines⁵⁶. All human carotid artery research complies with ethical guidelines for human subjects research under approved REB protocols at the University Health Network (no. 18-5282), for the procurement and use of human tissues and information. Human coronary artery biospecimens were obtained with written informed research consent at Stanford University (CA, USA) as previously described^39,57. All human coronary artery research herein complies with ethical guidelines for human subjects research under approved IRB protocols at Stanford University (no. 4237 and no. 11925) and the University of Virginia (no. 20008 and no. 302407), for the procurement and use of human tissues and information, respectively. The study and all protocols conform to the ethical guidelines set forth by the Helsinki Declaration (version 2008).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

SRB, JEF, and KLH designed the research study and analyzed and interpreted the results. SRB conducted all mouse experiments. JEF and KLH provided overall supervision and acquired funding for the study. KY assisted with tamoxifen injections in mice. CAS, LCDB, SR, KP, CKP, SZP, and SJH assisted with mouse tissue collection. KS and CH assisted with staining mouse tissue, and ABK assisted with the peripheral blood flow cytometry experiments. CAS provided expertise for mouse scRNA-seq sample preparation and analysis. CC assisted with the cell-cell communication analysis. LCDB and SR obtained human carotid tissue for immunofluorescence staining, and LCDB stained and imaged the tissue. JDW provided the Cdh5^CreERT2; Rosa26^{lox-STOP-lox-Sun1-sfGFP}; Erg^fl/fl mice and advice regarding mating and tamoxifen injection strategies. SRB generated the unfloxed and floxed Erg mouse lines on Apoe^+/+ and Apoe^{tm1Unc/tm1Unc} backgrounds. CLM conducted the bulk and single-nucleus ATAC-seq analyses of human coronary artery tissue and provided expertise for analyzing sequencing data. SRB and KE performed the RNA-seq and ATAC-seq analyses in cultured teloHAECs. CSR and MIC provided expertise for interpreting the mouse atherosclerosis experiments. SRB, JEF, and KLH wrote the manuscript. All authors provided feedback on the manuscript and approved the submission.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by Medicine by Design, which received funding from the Canada First Research Excellence Fund (CSR, MIC, and JEF), the Canadian Institutes of Health Research (PJT-191915; KLH and JEF and PJT-173489; JEF), the National Institutes of Health (R01HL164577 and R01HL148239; CLM) and a Single-Cell Data Insights award from the Chan Zuckerberg Initiative, LLC, and Silicon Valley Community Foundation (CLM). Studentship funding was received from CIHR and the Ontario Graduate Scholarship program (SRB). Support was also received from a New Investigator Award from the Heart and Stroke Foundation of Canada (KLH), the Wylie Scholar Award from Vascular Cures (KLH), and the Blair Early Career Professorship in Vascular Surgery from the University of Toronto (KLH). Infrastructure funding was obtained from the John R. Evans Leaders Fund from the Canada Foundation for Innovation (JEF).

Data availability

All raw and processed mouse aortic intimal scRNA-seq data are available in the Gene Expression Omnibus database (GEO: GSE300155). The raw and processed human carotid artery scRNA-seq dataset is available through GEO: GSE159677³⁵. The raw and processed human coronary artery bulk ATAC-seq datasets are available through GEO: GSE273054 and GEO: GSE188422. The raw and processed human coronary artery single-nucleus ATAC-seq dataset is available through GEO: GSE175621³⁹. The raw and processed teloHAEC RNA-seq data are part of a larger dataset that is available through GEO: GSE292727. The raw and processed teloHAEC ATAC-seq data are part of a larger dataset that is available through GEO: GSE292724. Published ChIP-seq data from Ctrl and ERG siRNA-treated HUVECs are available through GEO: GSE124892¹². Published ATAC-seq data from human MRC-5 fibroblasts are available through GEO: GSE121053⁸². Sample metadata for human studies are available in the cited publications or GEO datasets. Source data are provided with this paper in the Source Data file. All data supporting the findings described in this manuscript are available in the article, the Supplementary Material, the Source Data file, and the aforementioned GEO datasets. Source data are provided with this paper.

Code availability

The code implemented in this study for RNA-seq and ATAC-seq analyses utilized standard bioinformatic approaches and is available at: (https://github.com/srbotts/Botts_et_al_2026).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at (10.1038/s41467-026-75287-z).