Functional Genomics Link REST to Endothelial Plasticity and Atherosclerosis

Abstract

Background:

Genome-wide association studies (GWAS) have identified multiple novel loci that contribute to coronary artery disease (CAD) pathogenesis, but the mechanisms of these associations remain largely unknown.

Methods:

In this study we used a multi-trait colocalization approach to prioritize novel endothelial specific loci for atherosclerosis. We combined computational methods with in vitro assays and mouse models to study one of those new loci targeting gene REST.

Results:

A multi-trait colocalization approach across expression quantitative trait loci (eQTL) in atherosclerosis-relevant cell types followed by in vitro CRISPR interference revealed that a conserved regulatory element in a chromosome 4 genetic locus increases risk of CAD and decreases the expression of REST, a transcriptional repressor, in endothelial cells. Pcsk9-overexpressing mice with an endothelial-specific knockout of Rest exhibited increased atherosclerotic plaque formation in their aortas, with increased macrophage and lipid deposition within the plaque after 16 weeks of high-fat diet exposure compared to littermate controls. RNA-seq in human aortic endothelial cells (HAEC) after REST silencing followed by assessment of protein expression revealed that REST silencing triggers endothelial-to-mesenchymal transition (endMT). Consistently, REST silencing increased endothelial permeability and migration in vitro. Single nucleus RNA sequencing in endothelial lineage traced atherosclerotic mice with Rest knock-out revealed evidence of endothelial TGFb signalling activation and of transition smooth muscle-like cells in atherosclerotic aortas upon genetic knockout of Rest. CUT&Tag sequencing did not identify any known TGFb effector genes as direct REST transcriptional targets. Instead, joint analysis of CUT&Tag with RNA-seq, highlighted L1CAM, a known endMT activator, and its interactors as the most significant gene-set directly affected by REST in the endothelium. Simultaneous silencing of L1CAM and REST in HAEC inhibited the upregulation of mesenchymal genes and the enhanced migration induced by REST silencing and diminished the upregulation of several TGFb effectors overexpressed upon REST silencing.

Conclusion:

In summary, our data reveal the novel role of REST as a repressor in endothelial cells that functions to constitutively inhibit endMT and protect against atherosclerosis.

Introduction

Despite recent progress in cardiovascular disease prevention, atherosclerotic cardiovascular disease (ASCVD) continues to be a major public health burden. While most existing ASCVD treatments are centred around cholesterol storage, atherogenesis is complex and multifactorial. Recent studies indicate that ~50% of patients hospitalized with an ASCVD event have LDL cholesterol below the primary prevention target. Indeed, cellular processes of native cells of the artery wall and the immune system, and their interactions are all central to the pathogenesis of atherosclerotic plaques^1,2. However, the exact mechanisms involved in ASCVD are incompletely understood which hinders development of new targeted therapies to reduce its impact on public health globally.

Endothelial cells play a crucial role in various biological functions relevant to ASCVD. They oversee essential processes such as vascular tone, wound healing, sensitivity to shear stress, and regulation of inflammation³. Normally, the endothelium acts as a protective anti-atherogenic barrier which safeguards the integrity of the arterial intima⁴. However, under certain conditions, the endothelial cells can undergo a state change which leads to recruitment of immune cells, such as circulating monocytes, thereby promoting plaque formation. This is termed endothelial dysfunction and is now considered sine qua non for the initiation of atherosclerotic plaque. Endothelial dysfunction can also arise from a cell fate change in the endothelium termed endothelial-to-mesenchymal transition (endMT). During endMT, endothelial cells lose their defining characteristics and transition to mesenchymal cells with properties traditionally attributed to fibroblasts and smooth muscle cells. Recent studies have indicated that endMT is common in atherosclerotic plaques⁵ and that inhibiting endothelial pathways that promote endMT can decrease vascular inflammation and atherosclerotic plaque formation^6,7. Consequently, an improved understanding of mechanisms governing endothelial cell function and plasticity could facilitate the development of novel therapeutic strategies against ASCVD.

Genome wide association studies (GWAS) provide an opportunity for unbiased discovery of novel genetic mechanisms involved in complex traits⁸. GWAS for atherosclerosis have been very successful with recent large meta-analyses of coronary artery disease (CAD) GWAS increasing the number of identified genetic associations to > 270 genomic loci⁹. GWAS have shown that a substantial portion of the heritability of CAD can be traced back to vascular pathways. Indeed, recent studies show that endothelial cells are the second most enriched cell type in CAD GWAS signals, trailing only vascular smooth muscle cells¹⁰. However, most of the GWAS loci are found in non-coding DNA regions, not always clearly indicating a particular gene or mechanism. Therefore, many of the endothelial genes and pathways identified by GWAS have a presently unknown function.

In this study, we performed a functional genomics screen to prioritize endothelial GWAS loci that affect CAD pathogenesis. We discovered that an active regulatory element in chromosome 4 is regulating a novel transcription factor, REST, in the endothelium. We showed that REST protects against atherosclerosis and constitutively inhibits cellular pathways that induce endMT. This collective evidence nominates REST and its downstream pathways as novel targets in ASCVD prevention.

METHODS

Data Availability

Genotype and gene expression data for HAEC were obtained by dbGaP (phs002057) and the NHLBI GEO database (GSE30169 and GSE139377). Immune cell eQTL summary statistics were obtained from the DICE database (https://dice-database.org). GTEx eQTL summary statistics were obtained from the GTEx portal (https://www.gtexportal.org). VSMC eQTL summary statistics were obtained from the online resources provided in Liu et al.¹⁰ (https://med.stanford.edu/montgomerylab/Resources.html). Telo-HAEC ATAC-seq data included in Figure 1C were obtained from the NHLBI GEO database (GSE126200). Other chromatin data in Figure 1C were obtained directly from the WashU Epigenome Browser. Narrowpeak files for our CUT&Tag analysis are included as Supplemental Data File 1. All sequencing data will be deposited in the National Center for Biotechnology Information Gene Expression Omnibus and will be accessible to the general public after the manuscript is published. For a list of major resources, please see our Major Resources Table in our Supplemental Materials. For more details on Materials and Methods, please refer to our Extended Materials and Methods section in our Supplement.

Figure 1. Functional genomics screen identifies a role for endothelial REST in coronary artery disease.

A. Multi-trait colocalization of candidate endothelial genes between coronary disease GWAS loci and eQTLs in multiple atherosclerosis-relevant cell types using CAFEH. B. Colocalization between the 4q12 CAD GWAS locus and endothelial eQTLs for REST. Each dot represents a variant within the locus and the axes represent the −log10 p-values in the GWAS and endothelial eQTL datasets. The 95% credible set variants identified by CAFEH are encircled. C. Epigenetic fine-mapping of credible set variants in the 4q12 locus identifies two conserved variants in candidate regulatory elements (CREs) as candidate regulators of REST and drivers of the CAD GWAS signal. D. CADD score for each of the credible set variants in the 4q12 locus identifies rs6853156 as the most likely functional variant. E. CRISPR interference of inducible endothelial cells targeting the CREs surrounding the two candidate fine-mapped variants confirms that the region around rs6853156 regulates REST expression in endothelial cells. Relative expression is estimated by qPCR based on the delta-delta Ct method with ACTB as the loading control and a linear mixed model was used to account for nesting structure from three independent experiments with n=3 replicates each. qPCR technical replicates (n=2 per biological replicate) were averaged prior to the statistical analyses and each dot represents the average of the technical replicates. P-values are adjusted for the two comparisons (each variant compared to control) using FDR. HAEC: Human aortic endothelial cells. NC: Non-targeting guide control. VSMC: Vascular smooth muscle cells.

Functional Genomic Analyses

We used CAFEH, a novel Bayesian multi-trait colocalization framework to prioritize endothelial GWAS loci for further study¹¹. CAFEH was run on the van der Harst et al. GWAS summary statistics for CAD¹², along with expression quantitative trait locus (eQTL) data from relevant cell types and tissues (liver, artery, endothelial cells, smooth muscle cells, immune cells) from multiple sources^10,13-15. eQTLs were meta-analyzed when needed using the z-score method as described previously¹⁶. We computed top component colocalization probability based on CAFEH as previously described¹¹ using 1000 genomes LD reference¹⁷. For prioritized loci, we then confirmed presence of open chromatin in 95% credible set variants using orthogonal published endothelial cell ATAC-seq datasets^18-20.

Generation of inducible endothelial cells (iEC) expressing dCas9-KRAB

Human induced pluripotent stem cells (iPSCs) expressing dCas9-KRAB (Corriell #AICS-0090-391) were differentiated into endothelial cells following a previously published protocol²¹. All iPSC experiments were approved by the institutional stem cell research oversight committee under protocol #00000699.

Culture of Primary Cells

Primary HAEC were obtained from Lonza (#CC-25350) and cultured in complete endothelial cell media according to the manufacturer’s instructions. All experiments were performed using HAEC grown for no more than 10 passages.

RNA-seq

We performed paired-end poly-A enriched RNA sequencing in HAEC that were transfected with siRNA for REST compared to scramble control. We followed the DESeq2 pipeline²² for differential expression (DE). Surrogate variable analysis (SVA) was employed to control for unknown confounders and batch effects²³ while preserving the biological differences between groups. Genes were considered significant if their FDR was < 0.05. The gene set enrichment analysis (GSEA) package with the preranked method was used to identify pathways enriched in our DE results²⁴.

Cleavage Under Targets and Tagmentation (CUT&Tag)

CUT&Tag was performed using the CUT Tag-IT Assay Kit (Active Motif #53160) following the manufacturer’s protocol, using an anti-REST antibody (EMD Millipore #17-6410). Two biological replicates were used for CUT&Tag per the ENCODE CHIP-seq best practice guidelines^25,26. CUT&Tag sequences were aligned to the hg38 human genome using Bowtie2 and peaks were called using the MACS2 software²⁷.

Mouse Strains

The previously published²⁸ Rest floxed mouse strain (JAX #024549) was crossed with Cdh5-creER^T2 (Taconic #13073) mice to generate endothelial specific conditional Rest knockout mice for atherosclerosis experiments. These mice were subsequently crossed with mT/mG mice (JAX #007676) for the lineage tracing experiments. All the mouse experiments were approved by the Johns Hopkins animal care and use committee with protocol number #MO22M12. All mice used in our experiments received two doses of tamoxifen (Sigma #T5648-1G) at 0.2 mg/g body weight, administered with a three-day interval via oral gavage at 8-10 weeks of age to activate the Cre recombinase²⁹. A day after the second tamoxifen dose, all mice were given a single tail-vein injection containing 10¹¹ genomic copies of AAV8-m-Pcsk9 (Vector Biolabs, AAV-268246) followed by continuous administration of a high-fat diet (HFD) (Dyets, # 101511), composed of 21% anhydrous milk fat, 19% casein, and 0.15% cholesterol, thereafter³⁰.

Single nucleus RNA-seq data processing

Single nucleus RNA-seq analysis was performed using Cellranger³¹ with a modified version of the mm10 mouse genome that includes EGFP and tdTomato sequences. Diem 2.4.1 was used to identify and remove low-quality droplets³². Feature barcode matrix files from individual samples were normalized using sctransform³³ and integrated with rPCA as implemented in Seurat v5. We utilized Speckle v1.4.0³⁴ to analyze cell proportions across the different experimental groups. Differential expression between endothelial Rest knock-out mice and controls was performed using a pseudobulk approach for individual cell types. Gene set enrichment analysis was performed using GSEA v4.2.1 with the pre-ranked method. The rank list was tested against the mouse Reactome gene sets downloaded from mSigDB³⁵. To estimate cell-cell signaling pathways differentially active between endothelial Rest knock-out mice and controls we performed CellChat³⁶ with default options. In sensitivity analyses, SoupX 1.6.2³⁷ was applied to correct for ambient RNA contamination.

Statistics & Reproducibility

Statistical analyses were carried out using R version 4.3.2 and Prism version 10. Unless otherwise specified, differences between two groups were assessed using an unpaired two-tailed Student’s t-test, while differences among multiple groups were evaluated using one-way analysis of variance (ANOVA) followed by Tukey’s post-hoc test. A Shapiro-Wilks test was performed to assess for deviation from a normal distribution in each compared group and if the results were significant at p < 0.1, non parametric tests were used instead. In our in-vitro assays, data were analyzed with linear mixed models instead (as implemented in the lme4 package in R) to account for the nested structure generated by multiple independent experiments with multiple replicates in each experiment. Two-tailed p-values <0.05 were considered statistically significant, with all error bars representing the standard error of the mean.

Results

Functional genomics screen prioritizes endothelial REST as a novel gene associated with atherosclerosis

In prior work, we developed a new multi-trait colocalization algorithm (named CAFEH) that outperforms existing colocalization methods in the presence of allelic heterogeneity¹¹. In this study, we leveraged CAFEH using publicly available endothelial cell eQTL data that we meta-analyzed (Supplemental Table 1) to prioritize novel endothelial loci for CAD. We first applied CAFEH using endothelial eQTL and GWAS data for all genes within 1Mb of the sentinel variant for all genome-wide significant CAD GWAS loci from van der Harst et al.¹² In total, we screened 1562 genes and identified 32 genes whose endothelial eQTLs colocalize with CAD GWAS loci in any component, 27 of which were colocalizing in the top component for both studies (Supplemental Table 2). Colocalizing genes were enriched in pathways related to cell differentiation, TGFb and TNFa signaling (Supplemental Figure 1), underscoring the importance of endothelial inflammation in atherosclerosis biology and highlighting a role of endothelial cell differentiation and TGFb signaling in plaque development. To further narrow the list of genes for downstream characterization we used CAFEH to perform multi-tissue colocalization with eQTLs of multiple atherosclerosis-relevant cell types jointly (Figure 1A, Supplemental Table 3). This multi-trait analysis enhances the accuracy of colocalization inference and allows us to prioritize a subset of genes likely to be acting in CAD via endothelial specific mechanisms, defined as those genes that demonstrate the strongest evidence of colocalization in endothelial cells among all atherosclerosis relevant cell types. Out of this list we chose to further study the transcipriton factor REST in GWAS locus 4q12 because of the lack of prior literature on its endothelial effects and the multifaceted roles that transcription factors have been shown to play in atherosclerosis biology.

CAFEH identified a single active GWAS component for the 4q12 locus which had a strong colocalization probability with cis-eQTLs of the gene REST in HAEC and GTEx Aorta, weak colocalization probability with VSMCs and no signal for colocalization with liver or immune cells (Figure 1A, B). This result was robust to the ancestry composition of the LD reference panel (Supplemental Table 4). Epigenetic fine-mapping of the variants belonging to the 95% credible set for the locus identified two endothelial candidate regulatory elements (CRE) denoted by endothelial ATAC-seq peaks in different introns of REST, that each contained a credible set variant (Figure 1C). The alternative alleles of both those variants are associated with lower expression of REST in HAEC and increased risk of atherosclerosis, suggesting a protective effect of endothelial REST on atherosclerosis. CADD scores³⁹ estimated for all variants in the 95% credible set prioritized variant rs6853156 in one of the identified endothelial CREs as the most likely to be functionally active (Figure 1D).

To experimentally validate our in silico findings on the 4q12 locus, we employed CRISPR interference (CRISPRi) for the two prioritized CREs (Figure 1E). After differentiating dCas9-KRAB expressing human iPSCs into endothelial cells (Supplemental Figure 2), CRISPRi was conducted by transfecting inducible endothelial cells with CRE specific sgRNA. These experiments indicated that REST expression was significantly suppressed upon CRISPRi of the proximal but not the distal CRE. Further, head-to-head comparison of CRISPRi in inducible endothelial cells and undifferentiated iPSCs showed a significant downregulation of REST expression in endothelial cells but not in iPSCs, supporting the cell type specificity of the proximal CRE (Supplemental Figure 3). Neither of the CREs were found to affect expression of other genes proximal to the tested region, NOA1 or POLR2B (Supplemental Figure 4). Additionally, a luciferase reporter assay in primary HAEC showed modest evidence for an active regulatory signal for the proximal but not the distal CRE region that is diminished by variant rs6853156 (Supplemental Figure 5). These findings together suggest the hypothesis that genetic down-regulation of endothelial REST increases atherosclerosis risk in humans.

Endothelial Rest KO increases atherosclerosis in mice.

To further probe the effects of endothelial REST in atherosclerosis, we utilized a previously published Rest floxed mouse²⁸. We crossed that mouse with an inducible vascular endothelial cadherin Cre (Cdh5-CreER^T2) mouse. In the Cdh5-CreER^T2;Rest^flox/flox offspring (ecRestKO), Rest exon 4 is excised in the endothelium upon tamoxifen administration rendering Rest inactive. Littermate Rest^flox/flox mice that lack the Cre recombinase were used as controls. After vascular development is complete (8 weeks of age), mice were treated with tamoxifen followed by induction of atherosclerosis via a single tail vein injection of AAV8-m-Pcsk9 which overexpresses a gain of function version of Pcsk9 (D377Y) in the mouse livers followed by HFD exposure to induce atherosclerosis (Figure 2A). Although this is an established protocol for atherosclerosis induction in mice, we first confirmed that the protocol works in our hands. To do so, we treated wild type C57B6/J mice with AAV8-m-Pcsk9 or AAV8-GFP followed by 16 weeks of HFD. We confirmed upregulation of Pcsk9 in mouse livers and significant upregulation of lipids starting as early as 2 weeks after AAV8 injection. Significant atherosclerosis was observed in mouse aortas injected with AAV8-m-Pcsk9 but not in controls after 16 weeks of HFD (Supplemental Figure 6). We then proceeded with investigating the effects of endothelial Rest knockout in atherosclerosis. We observed an increase in atheromatous plaque levels in aortas after 16 weeks of HFD exposure in ecRestKO mice compared to controls (Figure 2B and Supplemental Figure 7). Unsurprisingly, total cholesterol, triglycerides, and non-HDL cholesterol were not different between the groups (Supplemental Figure 8). We then sought to investigate changes in plaque composition between the groups. We found that atherosclerotic plaques of ecRestKO mice were composed of higher levels of macrophages (denoted by CD68 staining), lipids (denoted by Oil Red O) and collagen content compared to controls (Figure 2C). Similar effects of endothelial Rest knock-out were seen in female mice (Figure 2B and Supplemental Figure 9). We also observed increased early-stage atherosclerosis in ecRestKO mice compared to controls after a 9 week exposure to high fat diet (Supplemental Figure 7).

Figure 2. Endothelial Rest protects against atherosclerosis in mice.

A. Outline of experimental groups and procedures. B. Oil Red O staining of whole aortas in male and female mice with endothelial Rest knockout or littermate controls. Results are presented as Oil Red O area divided by total aortic area. C. Analysis of plaque composition in atherosclerotic roots of male mice with endothelial Rest knockout or littermate controls. There is a significant increase in macrophage (defined as CD68 positive area divided by atherosclerotic lesion area), lipid (defined as Oil Red O positive area divided by atherosclerotic lesion area) and collagen (quantified by a Masson’s Trichrome stain) deposition in plaques of endothelial Rest knockout mice with no change in acellular area or fibrous cap size. Two-sided t-tests were performed for all comparisons.

Taken together, those results support a direct effect of endothelial Rest in atherogenesis and plaque composition. Higher macrophage and lipid deposition within plaques could be consistent with disturbed endothelial barrier function so we proceeded with an in vivo endothelial permeability assay using Evans Blue dye injection in atherosclerotic ecRestKO mice and controls after 9 weeks of HFD. This experiment showed a non-significant increase in Evans Blue staining in atherosclerotic lesions of ecRestKO mice (Supplemental Figure 10) suggesting that alternative mechanisms may also contribute to the observed change in plaque composition.

REST silencing induces changes in arterial endothelial cells consistent with endMT

To evaluate cellular mechanisms affected by REST in the endothelium we performed silencing of REST in HAEC followed by RNA-seq (Figure 3A and Supplemental Figure 11). Epithelial to mesenchymal transition and myogenesis were the most upregulated pathways upon REST silencing, implicating REST as a potential mediator of endMT and endothelial plasticity (Figure 3A). Indeed, our analyses revealed that REST silencing leads to the upregulation of multiple smooth muscle and extracellular matrix related genes (ACTA2, TAGLN, MMPs, collagen genes), with concomitant downregulation of endothelial markers (NOS3, PECAM1), and upregulation of TGFb signaling pathway genes (SMAD3, TGFB1, TGFBR1, TGFBR3) (Figure 3B). The full results of the RNAseq and pathway analysis are shown in Supplemental Tables 5 and 6 respectively. To further test those findings at the protein level, we performed Western blotting and Immunofluorescence which confirmed significant upregulation of smooth muscle markers TAGLN, and ACTA2, along with a notable down regulation of PECAM1 and NOS3 (Figure 3C, D). Concordantly, we demonstrated a robust increase in vascular permeability and enhanced cellular migration following REST silencing in HAEC, demonstrating phenotypic consequences of our molecular observations (Figure 3E). Given our in vivo findings of increased Oil Red O deposition within plaques we also evaluated whether REST silencing induces cholesterol uptake in HAEC, but we did not see evidence of increased uptake of either LDL or oxidized LDL in HAEC treated with siRNA for REST compared to scramble controls (Supplemental Figure 12).

Figure 3. Endothelial REST silencing triggers endMT in human cells in vitro.

A. Hallmark pathway analysis of RNA-seq data in HAEC treated with siRNA for REST or siRNA scramble. The top 5 upregulated and downregulated pathways are shown. P-values were calculated with a permutation test implemented in GSEA and all pathways shown are significant at an FDR < 0.05. Each data point reflects a measurement obtained from multiple cells in one biological replicate. B. Heatmap of expression of endMT related genes in HAEC treated with siRNA for REST or siRNA scramble. Colors represent the average log2 expression for each biological replicate normalized with the rlog method based on DESeq2 in RNA-seq. For presentation purposes all genes are row normalized by dividing with the average value in control samples. Groups were compared with a Wald test as implemented in DESeq2 and all genes shown are significant at an FDR < 0.005. C. Western blot with corresponding quantification data and D. Immunofluorescence of endothelial (PECAM1, NOS3) and mesenchymal (ASMA, TAGLN) proteins in HAEC treated with siRNA for REST or siRNA scramble. Each data point reflects a measurement obtained from multiple cells in one biological replicate. E. Endothelial monolayer permeability and endothelial cell migration in transwell assays in HAEC treated with siRNA for REST or siRNA scramble. In C-E, linear mixed models were used to account for nesting structure from 2-3 independent experiments with n=2-4 replicates each. Each data point reflects a measurement obtained from multiple cells in one biological replicate.

Endothelial Rest knock-out promotes endMT in atherosclerosis mouse models

We then ventured to assess whether endMT is observed in vivo during atherosclerosis upon endothelial Rest knock-out. To do that we crossed our endothelial Rest knock-out mice and corresponding endothelial Cre mice controls with a reporter mT/mG mouse. By treating these mice with tamoxifen after 8 weeks of age, we are able to simultaneously knock-out endothelial Rest and label endothelial cells with green fluorescence permanently, thereby permitting lineage tracing of those cells during atherosclerosis progression. We injected those mice with AAV8-m-Pcsk9 and exposed them to 16 weeks HFD as above to induce atherosclerosis. We subsequently isolated their aortas and performed single nucleus RNA-sequencing (Figure 4A, B and Supplemental Table 7). The results showed an increased proportion of VSMCs in atherosclerotic aortas of ecRestKO mice compared to controls (Figure 4C and Supplemental Figure 13). Analysis of GFP positive endothelial lineage cells revealed an increased proportion of proinflammatory (denoted by high Vcam1 expression) over lipid handling (denoted by high Cd36) endothelial cells (Figure 4C and Supplemental Figure 14). In addition, we saw evidence of transdifferentiated lineage positive cells which appeared to mostly cluster with VSMCs (Figure 4D) and were predominantly found in ecRestKO animals (Figure 4E and Supplemental Figure 15), which suggest the presence of endMT. Those findings were corroborated by immunofluorescence of the atherosclerotic aortic roots which demonstrated increased green fluorescence per cell in the atherosclerotic lesions of ecRestKO mice compared to controls (Figure 4F and 4G) and revealed an increased proportion of Tagln positive GFP positive cells in ecRestKO mice (Supplemental Figure 16).

Figure 4. Endothelial Rest knockout triggers endMT in vivo.

A. Experimental groups and outline of experimental procedures. B. UMAP plot of snRNAseq data in atherosclerotic aortas of mice with endothelial lineage tracing, with and without endothelial Rest knockout. C. Relative proportions of cell clusters in the snRNAseq data between endothelial Rest knockout mice and controls. D. UMAP plot of snRNAseq split by expression of EGFP. Note that GFP is expressed in endothelial cell clusters as expected but also ectopically in VSMC suggesting endMT. E. Relative proportions of transition VSMC-like cells among all GFP positive cells in mice with endothelial Rest knockout and controls estimated based on the snRNAseq data using Speckle. Moderated t-tests are performed as implemented in Speckle to compare relative proportions of different cell types between groups and p-values are FDR adjusted. F. GFP fluorescence and G. its corresponding quantification within the atherosclerotic lesions in ecRestKO mice and controls in aortic root sections. Lesion area (plaque) is outlined in white. Quantification of the GFP positive signal from endothelial lineage cells migrating in the intima was performed as GFP positive area per cell inside the intimal lesions. One outlier sample (based on p < 0.05 on Grubbs test in the control group) was removed (p-value without outlier removal remains < 0.05). A two-sided t-test was performed to compare between groups. H. Volcano plot and I. Reactome pathway analysis of pseudobulk differential expression analysis of endothelial cell clusters in mouse aorta snRNAseq between endothelial Rest knockout and control mice using DESeq2. Note enrichment in smooth muscle and extracellular matrix pathways, along with TGFb signaling in endothelial cells of mice with endothelial Rest knockout (red boxes). Groups were compared with a Wald test as implemented in DESeq2 for the Volcano Plot. To perform pseudobulk differential expression analysis, the expression read counts of each gene were aggregated across all endothelial cells in each mouse sample using the AggregateExpression function as implemented in Seurat v5. For the pathway analysis, p-values were calculated with a permutation test implemented in GSEA and adjusted for all independent pathways present within Reactome using FDR.

As a sensitivity analysis, because ambient RNA contamination has been noted as common in snRNAseq datasets and could cause misinterpreted or masked cell types⁴⁰, we performed aggressive ambient RNA correction on our dataset. This analysis successfully removed evidence of ambient contamination (Supplemental Figure 17, 18) and continued to demonstrate transition lineage positive endothelial cells outside canonical endothelial clusters predominantly in ecRestKO animals. Interestingly, this analysis revealed that ecRestKO mice are characterized by a novel VSMC-like cell cluster which is mostly absent from control animals (Supplemental Figure 19A) and has increased GFP expression (Supplemental Figure 19B). That cluster is characterized by strong expression of Tagln, but maintains expression of endothelial markers like Cdh5, Vwf and Pecam1 and also expresses synthetic markers like Col1a1 and Klf4 (Supplemental Figure 19C). Although this may indicate a novel intermediate partial endMT cluster, this data should be interpreted with caution given that the cluster is only seen after pre-filtering. Future studies with larger sample sizes and additional sequencing modalities (eg. scATAC-seq) may further help characterize this phenomenon. Regardless of the exact nature of the transitioning cell population, these results support the effects of endothelial Rest in protection against endMT during atherosclerosis in vivo.

Interestingly, we noted that regardless of the type of analysis, ecRestKO mice continued to demonstrate an overall significant increase in the proportion of true VSMCs in the aortas compared to CT (Supplemental Figure 13). This could be partially explained by endMT. However, the proportion of GFP positive VSMCs remains low in ecRestKO animals. This suggests the possibility that cell-cell signalling stemming from endothelial cells may contribute to VSMC proliferation and/or affect VSMC phenotype. To explore some of those signals we performed CellChat analysis to identify ligand-receptor interactions between ECs and VSMCs that are enriched in ecRestKO mice compared to controls. Among other pathways, we identified BMP, FN1 and netrin signaling as potential pathways that may contribute to increased EC to VSMC interactions upon Rest knock-out (Supplemental Figure 20). Those findings are hypothesis-generating and should be further explored in future research.

L1CAM is one of the direct downstream targets of REST that contributes to endMT

We then wanted to discover the mechanisms via which REST protects against endMT. We first used our snRNAseq data to study pathways triggered in endothelial cells of atherosclerotic mice with endothelial Rest knock-out. That process identified Tgfb signaling, along with pathways related to endMT like smooth muscle contraction and extracellular matrix production as some of the primary upregulated pathways (Figure 4H, I). We then wanted to investigate whether TGFb effectors are direct transcriptional targets of REST. Since REST has known functions as a transcriptional repressor in multiple tissues, we reasoned that genes directly repressed by REST should have their promoter or nearby enhancer sites bound by REST in endothelial cells. To discover those bound regulatory elements, we performed CUT&Tag for REST in HAEC. Joint analysis of the CUT&Tag and RNA-seq results revealed several genes that are both significantly upregulated upon REST silencing and have nearby regulatory elements bound by REST (Figure 5A and Supplemental Table 8). Interestingly, none of the known TGFb effector genes belonged in that list (Supplemental Table 9) so we hypothesized that TGFb signaling is indirectly triggered by REST via an intermediate mediator. Pathway enrichment analysis among direct REST target genes compared to all genes expressed in HAEC revealed L1CAM and its interactors as the most upregulated pathway (Figure 5B). Indeed, L1CAM expression was induced > 100-fold upon REST silencing in HAEC (Figure 5C, 5E and 5F, Supplemental Figure 21) and CUT&Tag confirmed direct binding of REST to the L1CAM promoter (Figure 5D). Concordantly, we found increased expression of L1cam in vivo in our snRNAseq data in the GFP positive compared to GFP negative VSMCs in ecRestKO mice and a strong trend towards increased L1cam in GFP positive transition VSMC-like cells overall in ecRestKO mice compared to controls (Supplemental Figure 22). L1CAM has been previously shown to induce endMT in vitro and in vivo^{41, 42} so we reasoned that its overexpression may at least partially account for the endMT effects of REST inhibition. In order to assess if REST silencing leads to endMT via L1CAM overexpression, we specifically examined whether combined silencing of REST and L1CAM would eliminate the observed endMT phenotype in vitro. Indeed, we were able to observe that combined silencing of L1CAM and REST reduced the upregulation of mesenchymal markers and diminished the increased endothelial migration observed by REST silencing (Figure 5G, Supplemental Figure 23). Concordantly, we found that upregulation of several TGFb signaling effector genes upon REST silencing was mediated by L1CAM (Figure 5H). However, L1CAM silencing was not able to inhibit the downregulation of endothelial markers observed upon REST silencing, suggesting that other REST targets also contribute to the observed endothelial phenotype.

Figure 5. L1CAM is a mediator of the endothelial effects of REST.

A. Joint analysis of RNA-seq and CUT&Tag identifies direct targets of REST in HAEC. The top 5 genes upregulated in RNA-seq upon REST silencing that are also direct REST targets are labeled in the plot. B. Reactome pathway analysis of genes upregulated upon REST silencing and whose regulatory elements are bound by REST based on CUT&Tag compared to all genes expressed in HAEC RNA-seq. Note that L1CAM (also known as L1) and its interactors are among the most enriched pathways. All presented pathways are significant at an FDR < 0.05. C. L1CAM immunofluorescence in HAEC shows expression of L1CAM in the surface of endothelial cells upon REST silencing. D. REST binds to the L1CAM promoter based on CUT&Tag. E. Quantification of the immunofluorescence images in C. A Wilcoxon rank sum test was performed for comparison. F. Western blot of L1CAM with its corresponding quantification in HAEC exposed to siRNA for REST or scramble control. A linear mixed model was used to account for nesting due to two independent experiments. G. Combined silencing of REST and L1CAM eliminates the effects of REST silencing on endothelial migration in vitro. Outlier replicates were removed with the ROUT procedure as implemented in Prism. A linear mixed model was used to account for nesting due to four independent experiments with n=2-3 replicates each and p-values were adjusted for all possible comparisons using FDR. H. qPCR shows that combined silencing of REST and L1CAM reverses effects of REST on TGFb signaling in vitro. ACTB is used as loading control. A linear mixed model was used to account for nesting due to 2-3 independent experiments with n=3 replicates each and p-values were adjusted for all possible comparisons using FDR. qPCR technical replicates (n=2 per biological replicate) were averaged prior to the statistical analyses and each dot represents the average of the technical replicates. ns: Not significant. Each data point reflects a measurement obtained from multiple cells in one biological replicate.

Discussion

In this study, we discovered a novel role for transcriptional repressor REST in atherosclerosis. We showed that a novel variant in the CAD GWAS locus 4q12 acts through transcriptional changes in the expression of REST in endothelial cells. We identified endothelial REST expression as protective against atherosclerosis in vivo and demonstrated that REST inhibition was associated with endMT both in vitro and in vivo. Last, we showed that those effects appear to be at least partially mediated via L1CAM and its influence on the TGFb cascade.

Our work has several strengths. First, the endothelial effects of REST have not been previously studied. REST was originally described as a gene that represses neuronal genes in non-neuronal tissues, with critical functions as a transcriptional switch during the process of neurogenesis⁴³. More recent studies however have demonstrated that REST is not silent in adult cells but is rather actively regulated and involved in multiple cellular processes and disease states including neurodegenerative diseases^44,45 and disorders of cardiac structure and function⁴⁶. In our study we show a novel role for REST in regulating endothelial identity and endothelial function with critical effects in atherosclerosis. Those findings expand the evidence for broad effects of REST and its co-repressors outside the nervous system and suggests the possibility that modulating REST expression and function could be effective in improving vascular homeostasis.

Second, our results highlight endMT as an important phenomenon in atheromatous plaque formation. This finding is concordant with recent literature. Evrard et al in 2016 showed that endMT occurs during atherosclerosis using a lineage tracing mouse model⁵. Further in vitro single cell studies revealed evidence of complete endMT in aortic endothelial cells in culture⁴⁷, whereas a growing number of studies have shown that endothelial inhibition of endMT drivers protects against atherosclerosis^6,7. Our study corroborates those studies by supporting detrimental effects of endMT in atherosclerotic aortas in vivo and reveals a new gene whose regulation influences endMT. Nonetheless, we note that whether the transitioning cells themselves or paracrine signaling from altered endothelial cell populations contributes to atherosclerosis pathogenesis is unclear. For example, in our in vivo model, we observed increased lipid deposition and macrophage infiltration in the aortic roots of ecRestKO mice. Although a plausible explanation for those effects could be disruption of endothelial barrier function, our Evans Blue experiment suggests that other mechanisms may also be at play. Our experiments also do not support an increased direct lipid uptake by endothelial cells upon REST inhibition, implying that cell-cell crosstalk may be responsible for some of the observed findings. Indeed, prior literature has shown crosstalk between endMT and macrophages, with endMT cell-conditioned media promoting macrophage lipid uptake⁴⁸ and endMT triggering active monocyte recruitment⁷. Similarly, given the significant increase in VSMCs we observed in ecRestKO atherosclerotic mice compared to controls, our study indicates potential direct effects of endothelial Rest on VSMC growth via paracrine or juxtacrine signaling. EndMT could therefore have multiple effects in promoting endothelial dysfunction and influencing vascular wall composition that leads to atherosclerosis.

Third, we identified L1CAM as a candidate downstream effector of REST that is partially responsible for the endMT phenotype. Notably L1CAM has previously been associated with atherosclerosis in humans. Specifically, a recently published Mendelian Randomization study using protein QTLs for L1CAM demonstrated that L1CAM is associated with coronary atherosclerosis (MR Odds Ratio 1.12, 95% CI (1.01-1.24)) and ischemic heart disease (MR Odds Ratio 1.1, 95% CI (1.01-1.21))⁴⁹. These findings are concordant with two prior papers showing that L1CAM in the endothelium triggers endMT^{41, 42}. In the first, investigators overexpressed L1CAM in mouse lung endothelial cells and demonstrated induction of endMT⁴¹. In the second, investigators demonstrated that in vivo administration of an antibody inhibitor to L1CAM protects mice against radiation-induced endMT⁴². The mechanism via which L1CAM promotes endMT remain unclear. While prior work suggested a role for IL-6/STAT3 inflammatory signaling triggered by L1CAM as an endMT effector⁴¹, STAT3 signaling was downregulated in vitro in our experiments upon REST silencing despite evidence for endMT. Instead, we showed that L1CAM triggers expression of multiple TGFb signaling effectors which may be responsible for at least part of the effect. The TGFb effect on atherosclerosis is controversial. Several studies suggest that canonical TGFb signaling exerts atheroprotective effects via its anti-inflammatory properties and its function in stabilizing plaques^50-52. However, recent studies have demonstrated that excessive and dysregulated TGFb activation may lead to profibrotic and proinflammatory effects, increasing plaque burden⁵³. In addition, TGFb activation in endothelial cells has been shown to promote endMT, increase endothelial activation and accelerate atherosclerosis in vivo^{7, 54}. Our finding of increased endothelial TGFb signaling upon Rest knock-out supports the notion of a detrimental effect of endothelial TGFb activation in atherogenesis. The exact mechanisms of REST effects on endMT, including its L1CAM and TGFb – dependent and independent effects indicated by our findings deserve further investigation.

Our work is not without limitations. First, while our computational and experimental data collectively support an effect of endothelial REST in atherosclerosis, we cannot exclude that the 4q12 locus effects may also be mediated by REST in other cell types which may be revealed by larger eQTL sample sizes or context specific eQTL studies. Additionally, our inducible Rest knockout model may not fully reflect the gene's in vivo role due to additional mechanisms relevant to disease that potentially act during embryogenesis. Second, given the high LD between variants in our CAFEH credible set for REST, we cannot exclude that additional variants within that set may contribute to the signal or that there is a haplotype that is driving part of the effect. However, we have shown with our experiments that rs6853156 is sufficient to alter REST gene expression in the endothelium. Third, given the inherent lack of plaque rupture events in mouse atherosclerosis, the effects of REST in plaque stability remain uncertain, although we note that GWAS data suggest detrimental effects of the same genetic variants in the incidence of myocardial infarctions⁵⁵. Fourth, even though our study suggests that endMT is an important phenomenon triggered by REST inhibition, future work is needed to understand the exact nature and role of mesenchymal cells arising from endMT in atherosclerosis and disentangle other effects of REST on endothelial function. Last, although we performed multiple testing correction for each individual experiment where applicable, our study does not correct for multiple comparisons across all performed experiments.

Conclusions

In conclusion, we have prioritized a novel endothelial specific locus for CAD and performed functional studies to define its underlying mechanisms. Our study revealed a novel role for REST as a critical regulator of endothelial homeostasis and constitutive inhibitor of endMT with atheroprotective effects. We also highlighted L1CAM as a candidate downstream target for REST with an important role in endothelial function. Further research is warranted to delineate the precise pathways affected by REST in the endothelium, which could pave the way for the development of novel targeted interventions for ASCVD.

Supplementary Material

Extended Materials and Methods

Major Resources Table

Supplemental Figures S1-S23

Supplemental Tables S1-S9

Uncut Gel Blots

Novelty and Significance.

What is known

• Genome-wide association studies have discovered several genomic loci for atherosclerotic cardiovascular disease with unknown functions, many of which are active in the endothelium

• Endothelial cells are critical for vascular wall homeostasis, protecting the inner layers of the arteries from circulating noxious agents

• Endothelial cells can lose their endothelial properties and gain fibroblast or smooth muscle-like functions during atherosclerosis, in a process known as endothelial-mesenchymal transition

What new information does this article contribute

• A human coronary artery disease genomic risk locus regulates the expression of the transcriptional repressor REST in the vascular endothelium

• Endothelial Rest knock-out increases atherosclerotic plaque burden and alters plaque composition in mice

• REST safeguards endothelial identity and protects against endothelial to mesenchymal transition in vitro and in a mouse model of atherosclerosis

While much progress has been made in understanding the genetic underpinnings of atherosclerosis, the mechanisms by which these genetic factors contribute to endothelial changes remain poorly understood. Here we performed a functional characterization of human genomic risk loci for coronary artery disease that affect endothelial function. We discovered a new function for the master transcriptional repressor REST in regulating endothelial plasticity and atherosclerosis. Experimental perturbation showed that lowering REST in endothelial cells activates endothelial to mesenchymal transition (endMT) programs, increases permeability and migration in vitro, and worsens atherosclerosis and alters plaque composition in mice. We discovered that those effects may be in part mediated by direct silencing of the cell adhesion molecule L1CAM. Our results demonstrate a novel function for REST as a molecular switch triggering cell fate transitions in the adult endothelium and promoting plaque formation under proatherogenic conditions. We anticipate our data to spark new research into endothelial REST regulation and boost the growing field of endothelial cell fate plasticity. More broadly, our findings provide a framework for translating noncoding atherosclerosis risk variants into targets, guiding future basic studies of endothelial fate control and suggesting therapeutic strategies aimed at limiting endMT to prevent or treat atherosclerotic disease.

Non-standard Abbreviations and Acronyms

AAV

Adeno-associated virus

ANOVA

Analysis of variance

ASCVD

Atherosclerotic Cardiovascular Disease

ATAC

Assay for transposase-accessible chromatin

CAD

Coronary Artery Disease

CRE

candidate regulatory element

CUT&Tag

Cleavage under targets and tagmentation

dbGaP

database of genotypes and phenotypes

ecRestKO

endothelial cell Rest knock-out mouse

endMT

endothelial to mesenchymal transition

eQTL

expression quantitative trait loci

FDR

False discovery rate

GSEA

Gene-set enrichment analysis

GTEx

Genotype Tissue Expression study

GWAS

Genome-wide association study

HAEC

human aortic endothelial cells

H&E

hematoxylin and eosin

HFD

high fat diet

iEC

inducible endothelial cells

iPSC

inducible pluripotent stem cells

LDL

Low density lipoprotein

MR

Mendelian randomization

SVA

Surrogate variable analysis

TSS

transcription start site

VSMC

vascular smooth muscle cells

WT

wild type