Identification of PGF+ endothelial cells associated with plaque instability in carotid atherosclerosis by scRNA-seq and RNA-seq analysis

ABSTRACT

Vascular endothelial dysfunction plays a critical role in the development of atherosclerosis; however, the mechanisms by which endothelial cells contribute to plaque instability remain incompletely understood. In this study, we performed an integrated analysis of single-cell RNA sequencing (scRNA-seq) and bulk RNA-seq data to characterize endothelial cell heterogeneity associated with carotid plaque instability. Clustering analysis, gene set variation analysis (GSVA), differential gene expression analysis, and KEGG pathway enrichment were conducted to identify key endothelial cell subsets and their functional characteristics. We identified three endothelial cell subsets (subsets 9, 10, and 11) that were significantly enriched in unstable plaques and exhibited upregulation of multiple pro-inflammatory cytokines. Pathway analysis revealed that these subsets were associated with activation of the PI3K – Akt signaling pathway and other inflammation-related pathways. Furthermore, findings were validated in an apolipoprotein E-deficient (ApoE−/−) mouse model, where increased expression of placental growth factor (PGF) and a higher proportion of PGF-positive endothelial cells were observed in atherosclerotic lesions. In conclusion, this study reveals the heterogeneity of endothelial cells in atherosclerotic plaques and identifies pro-inflammatory endothelial subsets potentially associated with plaque instability, providing new insights into the pathogenesis of atherosclerosis.

Trial registration: This study was approved by the Experimental Animal Ethics Committee of Guizhou University of Chinese Medicine (approval number: 2,024,010).

1. Background

Atherosclerosis (AS) is an important pathological basis for a variety of cardiovascular and cerebrovascular diseases. Its aggravation may cause arterial stenosis, and even malignant cardiovascular and cerebrovascular events including myocardial infarction and stroke as well as disabling peripheral arterial disease. This condition is an important cause of death in patients around the world [1]. As a chronic inflammatory disease, AS involves lipid deposition in the arterial intima, focal intimal fibrosis, and atherosclerotic plaque formation. The instability of atherosclerotic plaques is the direct cause of acute cardiovascular and cerebrovascular events. Secondary changes such as plaque rupture and thrombosis caused by vulnerable atherosclerotic plaques (VAPs), directly determine clinical outcomes in patients [2]. Therefore, it is clinically important to investigate the mechanisms of formation of VAPs in AS.

Vascular endothelial cell damage is the initiating factor of AS. Inflammatory cytokines are secreted by damaged endothelial cells and they attract monocytes to aggregate and migrate into the subendothelium. A large number of white blood cells and monocytes are activated into macrophages, which adhere to the activated endothelium and then differentiate into foam cells. Foam cells can lead to the thinning of the atherosclerotic fibrous cap and exacerbate the rupture of atherosclerotic plaques. Thus, inflammatory cell infiltration and lipid aggregation are important factors that lead to or exacerbate VAPs [3]. Vascular endothelial dysfunction plays a critical role in the development of atherosclerosis. Therefore, regulating endothelial cell function represents a promising strategy for stabilizing vulnerable plaques and preventing disease progression [4,5].

Genome-wide association studies (GWAS) have identified numerous genetic loci associated with atherosclerosis [6]. In addition, many studies have defined the cellular profile of human atherosclerosis based on single or several marker proteins. However, conventional cell flow cytometry cannot provide information on mRNA, and the involvement of multiple cellular components including endothelial cells, smooth muscle cells, macrophages and lymphocytes involved in the process of atherosclerosis requires the use of single-cell sequencing technology to further reveal the heterogeneity of these cells. Single-cell RNA sequencing (scRNAseq) of atherosclerotic core and peripheral plaques revealed that vascular smooth muscle cells (VSMCs) and endothelial cells exhibited the greatest degree of anatomical diversity, and the two types of cells possibly enhanced the role of core plaques in increasing calcification and extracellular matrix remodeling [7]. However, the heterogeneity and molecular mechanism regarding endothelial cells in the mechanism of carotid plaque stability are still unknown, and previous single-cell transcriptomic studies have not deeply analyzed endothelial cells.

Therefore, in this study, we performed an integrated analysis of publicly available single-cell RNA sequencing (scRNA-seq) and bulk RNA-seq datasets derived from human carotid atherosclerotic plaques to characterize endothelial cell heterogeneity associated with plaque instability. Furthermore, we validated key findings using an apolipoprotein E-deficient (ApoE-/-) mouse model of atherosclerosis.^−/−

1.1. Methods and materials

1.1.1. Acquisition and collection of single-cell RNA sequencing data

The single-cell transcriptomic data set used in this study was GSE159677 from the GEO database. The data set GSE159677 collected human carotid plaques using the 10X Genomics platform. ScRNA-seq data in GSE159677 were obtained from the pathological tissues of plaques from three human patients with asymptomatic carotid stenosis (ACS) who underwent carotid endarterectomy. High-throughput single-cell transcriptomic data were prepared for entire atherosclerotic core (AC) plaques and patient-matched proximal adjacent (PA) portions of carotid artery tissue from the patients undergoing carotid endarterectomy. Available patient-level and sample-level information for GSE159677 and GSE120521 was extracted from the corresponding GEO records and source publications and is summarized in Supplementary Tables S1 and S2.

1.2. Analysis of single-cell data

The raw single-cell data were obtained from published studies and analyzed using Seurat, a widely used R package, in the R software. The complete workflow followed established protocols for single-cell RNA sequencing (scRNA-seq) data processing Liu et al. [8], beginning with setting up the Seurat Object and performing standard pre-processing steps including data normalization, identification of highly variable features, and scaling. Linear dimensional reduction was carried out using PCA, followed by determination of the dataset dimensionality and clustering of cells via the Louvain algorithm. Non-linear dimensional reduction techniques such as UMAP were employed for visualization, and differentially expressed features were identified as cluster biomarkers. Cell type identities were assigned to clusters based on canonical markers, with visualizations generated to represent the results. While this workflow shares similarities with methods described by Liu et al. (2024), it was specifically adapted to address the research questions related to atherosclerotic plaque instability in our study.

1.3. Integration of single-cell transcriptomes

All scRNA-seq count matrices used in this study were obtained from the GEO database. The “Seurat” package in R (version 4.1.0) was used for integration, analysis, and visualization of scRNA-seq data, following approaches similar to those described by Liu et al. [8]. Prior to integration, each dataset underwent individual processing including quality control, cell selection, data normalization, and identification of highly variable features. The integration method provided by Seurat was then applied to combine datasets, enabling identification of cells in comparable biological states while correcting for technical differences between datasets. Human data from different patients and tissues were integrated using this approach to account for sample variability. Subsequent analysis followed the standard Seurat processing workflow, including data scaling, linear (PCA) and non-linear (UMAP) dimensional reduction, and cell clustering. These methods, while consistent with established protocols, were specifically implemented to investigate cellular composition and pathological mechanisms in the context of atherosclerosis.

1.4. Cell type identification

Based on reported marker genes for various cell types and cell type biomarkers in the original papers, we identified the following cell types and representative marker genes: endothelial cells (PECAM1, VWF, RAMP2), macrophages (CD14, CD68, LYZ), T cells (CD3D, CD2, TRAC), B cells (CD79A, MS4A1, IGKC), smooth muscle cells (MYL9, TAGLN, CALD1), and NKT cells (GNLY, NKG7, XCL1).

1.5. Visualization of single-cell data

The visualization of data analysis in this study was performed using Seurat, incorporating techniques commonly employed in scRNA-seq studies Liu et al. [8]. UMAP scatter plots were generated to represent the distribution of cells in reduced dimensional space, where spatial proximity reflected similarity in gene expression patterns. Violin plots were used to display expression levels of specific genes or gene sets across different cell types, while dot plots simultaneously represented average expression levels (color intensity) and detection rates (dot size) of marker genes. Volcano plots illustrated differentially expressed genes by statistical significance and fold change, and heatmaps facilitated comparison of average gene expression patterns across cell types. These visualization approaches, while methodologically aligned with standard practices in the field, were specifically tailored to highlight key findings related to endothelial cell heterogeneity in atherosclerotic plaques.

1.6. Microarray data analysis

The microarray data GSE120521 from the GEO database were derived from published studies. RNAseq data in GSE120521 were derived from stable and unstable regions dissected from fresh human carotid plaques obtained at carotid endarterectomy in 4 symptomatic patients. Plaques were classified into stable and unstable regions based on macroscopic appearance. Ex vivo imaging of plaques using 18F-sodium fluoride confirmed that the unstable regions were separated from the stable regions and characterized by visible areas where plaques ruptured. Data matrices were standardized, analyzed and visualized using limma, a widely used R package, in the R software.

1.7. Gene enrichment analysis

Differentially expressed genes identified through Seurat analysis were further examined using the online DAVID tool (https://david.ncifcrf.gov/summary.jsp), following analytical strategies similar to those described by Liu et al. [8]. Gene lists comprising upregulated or downregulated genes from specific clusters were uploaded to DAVID, a comprehensive functional annotation platform that enables biological interpretation of large gene sets. Gene Ontology (GO) enrichment analysis of biological processes was performed to identify pathways significantly associated with the input gene lists. This approach, while consistent with established bioinformatics pipelines, was specifically applied to elucidate molecular pathways relevant to plaque instability in atherosclerosis, providing context for the single-cell findings in our study.

1.8. Animals

^−/−Male ApoE-/- mice on a C57BL/6J background (8 weeks old, n = 3) and age- and sex-matched wild-type C57BL/6J mice (n = 3) were purchased from Beijing Sipeifu Experimental Animal Sales Co., Ltd. (Certificate No.: SCXK (Beijing) 2019–0010). Mice were randomly assigned to experimental groups using a random number sequence. All animals were housed under specific pathogen-free (SPF) conditions with a 12 h light/12 h dark cycle at 22 ± 2°C and 50% ± 10% humidity, with free access to food and water. All animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC) of Guizhou University of Traditional Chinese Medicine (Approval No.: 2,024,010). All procedures were performed in accordance with the NIH Guide for the Care and Use of Laboratory Animals and the ARRIVE 2.0 guidelines.

1.9. Reagents and antibodies

The Modified Oil Red O Staining Kit (C0158S) was purchased from Beyotime Biotech, Inc. (Shanghai, China). Hematoxylin staining solution (G1080), Glycerol Gelatin Mountant (S2150), and Antifade-Fluorescence Mounting Medium (S2100) were purchased from Solarbio Science & Technology Co., Ltd. (Beijing, China). Rabbit anti-PGF antibody (10642–1-AP; Proteintech Group, Inc., Wuhan, China) was used for both immunohistochemistry and immunofluorescence staining. Mouse anti-PECAM-1 antibody (sc-376764; Santa Cruz Biotechnology, Inc., USA) was used for immunofluorescence staining. Polyclonal anti-rabbit IgG-HRP (SV0002) and anti-rabbit IgG-FITC (BA1105) were purchased from Boster Biological Technology, Ltd. (Wuhan, China). Cy3-conjugated sheep anti-mouse IgG (SA00009-1) was purchased from Proteintech Group, Inc. (Wuhan, China).

1.10. Mouse model of arteriosclerosis

After 1 week of acclimatization, the mice were divided into two groups (n = 3 per group, representing independent biological replicates): an atherosclerosis model group consisting of ApoE-/-^−/− mice fed a Western diet (Hallan TD.88137 containing 21% fat and 0.2% cholesterol; Botai Hongda Biotechnology Co., Ltd., Beijing, China) for 8 weeks, and a control group consisting of wild-type C57BL/6J mice fed a standard chow diet for the same duration. Body weight was monitored weekly during the feeding period. At the end of the experiment, mice were euthanized by CO_2_ asphyxiation. Following transcardial perfusion with 20 mL of 5 mM EDTA and 20 mL of PBS, the heart and aorta were harvested for pathological and molecular analyses. Tissue samples were either embedded in optimal cutting temperature (OCT) compound for cryosectioning or stored at −80°C for subsequent analysis.

1.11. Oil Red O staining

Cryosections of the aortic root (8 μm thickness) were air-dried at room temperature for 2 h. Staining was performed using the Modified Oil Red O Staining Kit according to the manufacturer’s instructions with minor adjustments. Briefly, sections were incubated with pre-filtered washing solution for 20 s, followed by immersion in freshly prepared and filtered Oil Red O working solution for 12 min at room temperature. All sections from the same experiment were stained simultaneously using the same batch of staining solution to ensure consistency. Sections were then differentiated in 85% isopropanol for 30 s, rinsed gently under running distilled water for 20 s, and counterstained with hematoxylin for 25 s. After washing in tap water for 5 min to allow color development, sections were mounted with Glycerol Gelatin Mountant. Images were captured using a bright-field microscope. For each mouse (n = 3 biological replicates), three nonconsecutive sections were analyzed, and three random fields per section were imaged at × 200 magnification for quantitative analysis.

1.12. Immunohistochemical staining

Cryosections were air-dried at room temperature for 2 h, fixed in pre-cooled 4% paraformaldehyde (PFA) for 10 min at 4°C, and washed three times with PBS. Endogenous peroxidase activity was blocked by incubation in 3% H₂O₂ in methanol for 10 min at room temperature, followed by three PBS washes. Nonspecific binding was blocked with 5% normal goat serum in PBS for 30 min at room temperature. For negative controls, the primary antibody was replaced with PBS or normal rabbit IgG at the same concentration. Sections were then incubated overnight at 4°C with rabbit anti-PGF antibody (1:200). After warming to room temperature for 30 min and washing three times with PBS, sections were incubated with HRP-conjugated secondary antibody for 30 min at room temperature. Immunoreactivity was visualized using freshly prepared 3,3′-diaminobenzidine (DAB), and development time was monitored microscopically and kept consistent across samples (typically 2–5 min). The reaction was stopped with distilled water. Sections were counterstained with hematoxylin for 20 s, washed in tap water for 10 min, dehydrated through a graded ethanol series, cleared in xylene, and mounted with neutral balsam. Images were captured using a light microscope. For each mouse (n = 3 biological replicates), three nonconsecutive sections were stained, and three random fields per section were analyzed at × 200 magnification.

1.13. Immunofluorescence staining

Frozen sections were air-dried at room temperature for 2 h, fixed with 4% PFA for 10 min, and washed three times with PBS. Sections were then permeabilized and blocked with blocking buffer containing 1% BSA and 0.3% Triton X-100 in PBS for 1 h at room temperature. After blocking, sections were incubated overnight at 4°C with a mixture of primary antibodies, including mouse anti-PECAM-1 (CD31, 1:50) and rabbit anti-PGF (1:50). For negative controls, primary antibodies were replaced with isotype control antibodies or blocking buffer alone. On the following day, sections were warmed to room temperature and washed three times with PBS. Sections were then incubated for 1 h at room temperature in the dark with a mixture of secondary antibodies, including Cy3-conjugated sheep anti-mouse IgG (1:200) and FITC-conjugated anti-rabbit IgG (1:200), together with DAPI (1 μg/mL) for nuclear staining. After three washes with PBS, sections were mounted with Antifade-Fluorescence Mounting Medium. Images were acquired using an Olympus BX53 fluorescence microscope with consistent exposure settings across all samples within the same experiment. For each mouse (n = 3 biological replicates), three nonconsecutive sections were stained, and three random fields per section were captured at × 200 magnification.

1.14. Statistical analysis

All statistical analyses were performed using R software (version 4.1.0) or GraphPad Prism (version 9.0). Data are presented as mean ± standard deviation (SD). The sample size “n” for all quantitative analyses refers to the number of independent biological replicates (mice). For in vitro experiments, “n” refers to the number of independent experimental repeats. This is explicitly stated in each figure legend. The normality of data distribution was assessed using the Shapiro-Wilk test. The homogeneity of variances was assessed using Levene’s test (for two groups) or Bartlett’s test (for more than two groups). For comparisons between two groups of normally distributed data:If variances were equal, an unpaired two-tailed Student’s t-test was used.If variances were unequal, Welch’s corrected t-test was used. For comparisons between two groups of non-normally distributed data, the non-parametric Mann-Whitney U test (Wilcoxon rank-sum test) was used.For comparisons among more than two groups: if data followed a normal distribution and variances were homogeneous, one-way analysis of variance (ANOVA) was performed, followed by Tukey’s honestly significant difference (HSD) post-hoc test for multiple comparisons.If the assumptions for ANOVA were not met, the non-parametric Kruskal-Wallis test was used, followed by Dunn’s post-hoc test with Bonferroni correction. Qualitative data (e.g. categorical pathology scores) were analyzed using Fisher’s exact test. A p-value ≤ 0.05 was considered statistically significant. All statistical tests were two-sided. The specific test used for each experiment, along with the exact n value, is reported in the corresponding figure legend.

2. Results

2.1. Analysis based on scRnaseq data from human carotid atherosclerotic plaques

To learn about the cellular components involved in the development of atherosclerosis, we obtained scRNA-seq data from human carotid atherosclerotic plaques from the dataset GSE159677. Using the Seurat package in Rstudio software, we analyzed the single-cell transcriptomic data by following a standard single-cell sequencing workflow, including data filtering, data normalization, dimensionality reduction and clustering. Subsequently, based on the cellular marker genes reported in original articles and other studies [7], the cells were classified into the following 6 cellular subtypes (Figure 1(A)): macrophages (CD14, CD68, LYZ), endothelial cells (VWF, PECAM-1, RAMP2), smooth muscle cells (MYL9, TAGLN, CALD1), NKT cells (GNLY, NKG7, XCL1), T lymphocytes (CD3D, CD2, TRAC), and B lymphocytes (CD79A, MS4A1, IGKC). The marker genes for the 6 cell types are shown in the dot plot (Figure 1(B)). To analyze the cellular status of each cell type, we calculated the percentages of mitochondrial genes and ribosomal genes in each cell type (Figure 1(C)). The violin plots show that there was no significant difference in the percentages of mitochondrial genes in various cell types, but the percentages of ribosomal genes in T cells and B cells were significantly higher, suggesting active protein synthesis in these cell types. In the single-cell transcriptomic data, a total of 3 samples from patients with carotid atherosclerosis were collected. Each patient’s sample included atherosclerotic core (AC) plaques and patient-matched proximal adjacent (PA) portions. To assess the effect of integration of cells originating from different patients and different tissue sites in the single-cell sequencing workflow, we showed the cluster distributions of cells from different patients and cells from different tissue sites using UMAP plots (Figures 1(D-E)), respectively. The results showed no significant difference in the overall distribution of cells from different patients or cells from different tissue sites, indicating that integration of single-cell transcriptomes eliminated technical noise and sample errors. Finally, to screen out proteins specific to each cell type, the top 10 differentially expressed (DE) genes within the 6 cell types are presented in a heatmap (Figure 1(F)), including ACKR1, IFI27, RAMP2, VWF, IGFBP4, GNG11, TM4SF1, CALCRL, ID1, and CLEC14A.

Figure 1.

Single-cell data. (A) UMAP plot showing clustering of all cells in atherosclerotic core (AC) plaques and patient-matched proximal adjacent (PA) portions in human carotid arteries; (B) dot plot showing marker genes for all cell types; (C) violin plot showing the percentages of mitochondrial genes and ribosomal genes in all cell types; (D) UMAP plot showing the cluster distributions of cells from different patients; E UMAP plot showing the cluster distributions of different pathological tissues of plaques; F heatmap showing differentially expressed genes in all cell types.

2.2. Analysis of endothelial cell subsets

EC damage is not only an initiating factor promoting the formation of early atherosclerotic plaques, but also an important cause of AS [9]. Vascular proliferation and platelet aggregation play essential roles in the occurrence of AS. A large number of studies have shown that some inflammatory functions of human vascular endothelial cells can be induced by inflammatory cytokines to enhance the adhesion and chemotaxis of endothelial cells to white blood cells, stimulate the proliferation of vascular smooth muscle cells and lead to the aggregation of platelets, thereby resulting in AS [10–12]. Endothelial cell dysfunction usually occurs in early vulnerable atherosclerotic plaques [13].

To further characterize endothelial heterogeneity in carotid atherosclerotic plaques, we extracted 6108 endothelial cells and subdivided them into 15 endothelial cell subsets (Figure 2(A)). To improve the interpretability of these numerically labeled clusters, the representative marker features and annotations of endothelial cell subsets 0–14 are summarized in Supplementary Table S3. We then examined the expression of the endothelial markers VWF and PECAM1 across all subsets. The results showed that all subsets expressed VWF and PECAM1, confirming their endothelial identity (Figure 2(B)).

Figure 2.

Analysis of endothelial cell subpopulations. (A) UMAP plot showing clustering of all endothelial cell subpopulations in human carotid plaques; (B) violin plot showing the expression of marker genes for all endothelial cell subpopulations; (C) violin plot showing the percentages of mitochondrial genes and ribosomal genes in all endothelial cell subpopulations; (D) Heatmap showing differentially expressed genes in all endothelial cell subpopulations; (E) UMAP plot showing the cluster distributions of all endothelial cell subpopulations in different pathological tissues of plaques F Bar graph comparing the percentages of all endothelial cell subpopulations in different pathological tissues of plaques.

To assess the cellular status of each subset, we compared the percentages of mitochondrial genes and ribosomal genes across all endothelial cell subsets (Figure 2(C)). The percentage of mitochondrial genes was highest in subset 5, which may indicate increased cellular stress, injury, or apoptotic tendency in this subpopulation. In contrast, the percentages of ribosomal genes were higher in subsets 11 and 13, suggesting relatively active protein synthesis in these subsets.

We next examined differentially expressed genes across endothelial cell subsets using a heatmap (Figure 2(D)). Notably, subsets 9 and 10 showed high expression of inflammatory cytokine-related genes, including CXCL12, CCL5, and CCL4, suggesting that these subsets may represent pro-inflammatory endothelial programs.

To compare the distribution of endothelial subsets between anatomical regions, we analyzed their proportions in atherosclerotic core (AC) plaques and patient-matched proximal adjacent (PA) portions using UMAP plots and bar graphs (Figure 2(E,F)). Overall, subsets 0, 1, 2, and 3 were the most abundant endothelial populations. Among the subsets, subsets 5, 6, 9, 10, 11, and 14 were enriched in AC plaques, whereas subsets 2, 7, and 12 were relatively enriched in PA portions. These findings indicate substantial endothelial heterogeneity across plaque regions and highlight several candidate endothelial subsets potentially associated with lesion progression.

2.3. Gene set variation analysis (GSVA) of RNAseq data from human carotid atherosclerotic plaques

The above analysis of endothelial cells yielded differential subsets in AC plaques and PA portions, namely subsets 5, 6, 9, 10, 11 and 14 in AC plaques and subsets 2, 7 and 12 in PA portions, respectively. To further evaluate whether endothelial cell subsets identified from the scRNA-seq data (AC vs PA regions) are associated with plaque instability, we analyzed RNA-seq data derived from independent samples of stable and unstable plaques. Such data were extracted from the dataset GSE120521. We first analyzed the differentially expressed genes in the differential subsets included in single-cell data, and subsequently estimated by GSVA the infiltration fraction of each patient’s sample included in the RNAseq data. The RNAseq data contained data from 4 stable and 4 unstable plaque samples from patients. Estimated scores after normalization for each subset in 8 patients are shown in heatmaps (Figure 3(A)). The results showed that subsets 14 and 6 were numerous in stable plaques, and subsets 2, 5, 7, 9, 10, 11 and 12 were numerous in unstable plaques.

Figure 3.

Analysis of endothelial cell subpopulations in bulk RNA-seq data from stable and unstable plaques and their relationship to scRNA-seq-defined subsets. (A) GSVA of the difference in infiltration of endothelial cell subpopulations in stable and unstable plaques in human carotid arteries; (B) UMAP plot showing distribution and clustering of endothelial cell subpopulations 5, 9, 10 and 11 in atherosclerotic core (AC) plaques and patient-matched proximal adjacent (PA) portions in human carotid arteries.

Based on the analysis of RNAseq data and sc-RNAseq data, we found that the subsets in stable plaques included in the analysis of RNAseq data were not consistent with those in PA portions included in the analysis of sc-RNAseq data, but both stable plaques and PA portions were relatively normal tissues. Therefore, considering the rigor of the results, we did not conduct an in-depth analysis of the subsets in stable plaques or PA portions. The relation between these subsets and normal tissues remains to be further explored and confirmed. However, the percentages of subsets 5, 9, 10 and 11 (marked with red boxes) were found to be significantly increased in AC plaques included in the analysis of sc-RNAseq data (Figure 3(B)), and at the same time, these subsets were still found to be increased in unstable plaques included in analysis of RNAseq data, indicating the concordance of the results of analysis of the two transcriptomic datasets. It should be noted that AC and PA represent anatomical regions within the same plaque, whereas stable and unstable plaques are defined based on pathological characteristics from an independent dataset. Therefore, these two classifications are not directly equivalent. Given that the percentages of these four subsets were significantly increased in AC regions, and were also consistently elevated in unstable plaques in an independent dataset, these subsets may be closely associated with plaque instability. However, given the distinction between anatomical regions and pathological states, these findings should be interpreted with caution.

2.4. Differential gene expression (DGE) analysis of endothelial cell subsets in unstable plaques

Important causes of atherosclerotic lesions are high expression of inflammatory genes in endothelial cells, endothelium-dependent vasodilator response impaired and endothelial permeability increased when inflammatory response continues to stimulate endothelial cells, as well as endothelial cell dysfunction [14]. Of differentially expressed genes in subset 5 in unstable plaques, NEAT1 was upregulated. It is widely found in mammalian cells and involved in inflammatory responses in a variety of diseases [14]. Other genes including MT-CO3, MT-CO2 and MT-ND4 were all associated with mitochondrial function, possibly indicating a greater degree of apoptosis or damage in subset 5 (Figure 4(A)). In addition, the elevated mitochondrial gene ratio in subset 5 only suggests a potential tendency of cellular injury and apoptosis at the transcriptomic level, and the exact cell state and its functional role in atherosclerotic plaque progression need to be verified by further in vitro and in vivo functional experiments [15]. CXCL12, CLDN5 and SRGN were upregulated in endothelial cell subset 9 (Figure 4(B)). CXCL12, a highly potent inflammatory chemokine, is involved in almost the entire process from the onset of atherosclerotic fatty streaks to plaque formation [16]. Claudin5 (CLDN5) is a tight junction protein expressed in both endothelial and epithelial cells, and it is involved in endothelial cell proliferation [17]. SRGN is predominantly expressed in endothelial cells. Its synthesis and secretion were shown to increase in response to stimulation by other inflammatory cytokines [18]. CCL4, CCL5, CCL4L2, CXCR4, IL32, IL7R, GZMK and GZMA were upregulated in subset 10 (Figure 4(C)). CCL4, CCR5 and CCL4L2 are cardiac chemokines. CCL4 is a ligand for CCL5. It was upregulated in atherosclerotic plaques [19]. The chemokine receptor CXCR4 plays a role in cell proliferation and tissue regeneration. Both IL32 and IL7R are inflammatory cytokines. IL7R is a receptor for IL-7. It was able to contribute to endothelial cell infiltration through inflammatory responses [20]. It was shown that IL-32-mediated endothelial cell dysfunction played a role in vascular wall dysfunction [21]. Granzyme K (GzmK) belongs to a family of trypsin-like serine proteases. Its expression level is correlated with acute inflammation. An immunoassay for assessing cellular senescence suggested that GzmK was an inflammatory marker [22]. Granzyme A (GzmA) can act as a proinflammatory mediator. It played an important role in the pathogenesis of inflammatory diseases including sepsis, rheumatoid arthritis and pneumonia [23,24]. Genes including LY6H, PGF, ANGPT2, ApoD, CTGF and COL4A1 were upregulated in subset 11 (Figure 4(D)). Of them, PGF and ANGPT2 promoted endothelial cell migration and proliferation [25]. It was shown that the expression of PGF was correlated with human atherosclerotic carotid plaque instability [26]. ApoD is a member of the lipocalin superfamily of lipid transport proteins. Its function is mainly related to lipid metabolism. It was shown that the mechanism underlying how ApoD regulated oxidative stress and inflammatory responses was related to its ability to bind to arachidonic acid (ARA) [27]. The secreted protein CTGF, a key target for neuroinflammation, was able to trigger astrocyte-mediated inflammatory responses [28]. Collagen alpha-1(IV) chain (COL4A1) is a major structural component of basement membranes of various tissues. It was shown to be involved in the specific inhibition of endothelial cell proliferation/migration and angiogenesis [29,30].

Figure 4.

Differential gene expression (DGE) analysis of subpopulations in unstable plaques. Volcano plots show upregulated and downregulated genes in endothelial cell subpopulations 5 (A), 9 (B), 10 (C), and 11 (D).

Differential gene expression (DGE) analysis of endothelial cell subsets 5, 9, 10 and 11 in unstable plaques found that many secreted cytokines related to atherosclerosis or involved in inflammatory responses were significantly upregulated in subsets 9, 10 and 11. Upregulated expression and release of these cytokines might contribute to the development of plaque instability. Therefore, we selected subsets 9, 10 and 11 for subsequent in-depth analysis of signaling pathways.

2.5. KEGG signaling pathway in subsets in unstable plaques

DGE analysis above showed that many genes related to inflammatory response and plaque instability were greatly upregulated in endothelial cell subset 9 (CXCL12⁺), subset 10 (CCL4⁺) and subset 11 (PGF⁺), indicating that these 3 subsets are of great value for investigating the role of endothelial cells in the development of atherosclerosis. Therefore, we further conducted an in-depth enrichment analysis of the KEGG signaling pathway in the 3 subsets.

The top 10 significantly enriched signaling pathways in subset 9, subset 10 and subset 11 are shown in dot plots (Figures 5A-C). P13K-Akt pathway activation was significantly increased in all the 3 subsets. The P13K-Akt pathway often acts synergistically with other pathways to regulate cell growth and proliferation [31]. P13K is a key component for signal transduction in the growth factor receptor superfamily. It is able to activate endothelial nitric oxide synthase (eNOS) to produce threonine-protein kinase (Akt) [32]. Activated Akt is able to regulate the biological activity of downstream targets through phosphorylation, thereby regulating cell proliferation, migration, apoptosis and differentiation. It has been shown that inhibition of the P13K/Akt/mTOR pathways is able to regulate apoptosis of endothelial cells in atherosclerotic lesions [33]. endothelial cell inflammatory response is involved throughout arteriosclerosis and plaque progression. The P13K/Akt pathways are widely involved in the inflammatory process in various diseases. It has been shown that downregulation of the PI3K/Akt/mTOR pathways is able to attenuate the inflammatory responses [34,35].

Figure 5.

Enrichment analysis of signaling pathways in subpopulations in unstable plaques enrichment. analysis of the KEGG signaling pathway in endothelial cell subpopulations 9 (A), 10 (B) and 11 (C) is shown in dot plots.

In addition, the activation of pathways including Rap1 and Chemokine pathways was increased in subset 9. It was shown that Rap1 expression was increased in the aorta of ApoE^−/− mice fed a high-fat diet, and Rap1-mediated pathways were involved in the early stage of atherosclerosis [36]. Chemokines are important regulatory proteins in the transport and activation of white blood cells, and comprise 4 subfamilies (C, CC, CXC and CX3C). CXCR2 and CXCR4 and their ligands CXCL8 and CXCL1 in chemokine pathways were able to regulate the recruitment of vascular progenitor cells and neutrophils, thereby contributing to atherosclerotic lesion formation [37]. The activation of pathways including NF-κB and TNF pathways was increased in subset 10. It was shown that inhibition of metabolic inflammation associated with the NF-κB pathway was able to attenuate atherosclerosis [38]. Furthermore, mature dendritic cell-derived exosomes increased endothelial inflammation and atherosclerosis through the membrane TNF-α-mediated NF-κB pathway [39]. In addition, inhibition of ROS/p38 MAPK signaling pathways was able to suppress ox- LDL-induced HUVEC inflammation, endothelial-to-mesenchymal transition (EndMT) and apoptosis [40].

2.6. Validation of PGF-positive endothelial cells in an ApoE-/- mouse model of atherosclerosis

^−/−To further validate the biological relevance of the transcriptomic findings, we established an ApoE-/- mouse model of atherosclerosis and focused on PGF-positive endothelial cells for orthogonal validation. Differential gene expression analysis of endothelial cell subsets 5, 9, 10, and 11 identified several upregulated cytokines associated with inflammation and atherosclerosis. Among these, PGF, which was enriched in subset 11, was selected for in vivo validation because of its potential relevance to endothelial dysfunction and plaque progression.

Wild-type C57BL/6J mice fed a normal diet and ApoE-/- mice fed a^−/− high-fat diet (HFD) for 8 weeks were used to establish the atherosclerosis model [41]. Oil Red O staining confirmed successful model establishment, showing substantial plaque formation in the aortic roots of ApoE-/- mice, whereas no obvious plaques were observed in wild-^−/−type control mice (Figure 6(A)).

Figure 6.

A Oil Red O (ORO) staining of the aortic roots in the blank and model groups; B immunofluorescence co-localization with CD31/PGF in the blank and model groups; C Bar graph showing the percentage of CD31-positive area in the total area in the blank and model groups, the percentage of the PGF-positive area in the total area in the blank and model groups, and the percentage of cells co-stained with antibodies for PGF and CD31 in CD31-positive cells in the blank and model groups; D immunohistochemistry staining showing the expression of PGF in the blank and model groups; E percentage of the PGF-positive area in the total area in the blank and model groups. *p < 0.05, ***p < 0.001, ^ns p > 0.05 (no difference).

To further validate the presence of PGF-positive endothelial cells in atherosclerotic lesions, we performed immunofluorescence co-localization analysis of PGF and the endothelial marker CD31. The results showed that PGF expression was mainly localized to vascular endothelial cells in the luminal region of atherosclerotic lesions (Figure 6B; the negative control image is provided in the Supplementary Material). Quantitative analysis showed that the proportion of PGF/CD31 double-positive cells among CD31-positive cells was significantly increased in the model group compared with the control group (p < 0.05) (Figure 6(C)). In addition, no significant difference was observed in the overall CD31-positive area between the two groups, whereas the PGF-positive area and the proportion of PGF-positive endothelial cells were both significantly increased in atherosclerotic lesions.

To further assess PGF protein expression at the tissue level, we performed immunohistochemical staining. The results showed that the PGF-positive area was significantly increased in the model group compared with the control group (p < 0.05) (Figures 6D–E). Together, these findings provide experimental support for the increased expression of PGF in endothelial cells within atherosclerotic lesions and support the pathological relevance of the PGF+ endothelial program identified by transcriptomic analysis.

2.7. Validation of PI3K, Akt, and NF-κB expression in the atherosclerotic mouse model

To further evaluate the pathway-related findings suggested by KEGG enrichment analysis, we examined the expression of PI3K, Akt, and NF-κB in the atherosclerotic mouse model. Immunohistochemical staining showed increased PI3K expression in the model group compared with the blank group (Figure 7(A–B)). In addition, immunofluorescence staining demonstrated increased expression of Akt and NF-κB in the model group (Figure 7(C–F)). Quantitative analyses confirmed that the positive areas of PI3K, Akt, and NF-κB were all significantly increased in atherosclerotic lesions compared with the blank group. These findings provide additional experimental support for the pathway-enrichment results and suggest the involvement of PI3K-Akt- and NF-κB-related signaling in the inflammatory processes associated with atherosclerotic lesion progression.

Figure 7.

(A) immunohistochemistry staining showing the expression of PI3K in the blank and model groups; (B) percentage of the PI3K-positive area in the total area in the blank and model groups; (C) immunofluorescence staining showing the expression of Akt in the blank and model groups; (D) percentage of the Akt-positive area in the total area in the blank and model groups; (E) immunofluorescence staining showing the expression of NFκB in the blank and model groups; (F) percentage of the NFκB-positive area in the total area in the blank and model groups; (G) immunofluorescence staining showing the expression of CCL4 in the blank and model groups; (H) percentage of the CCL4-positive area in the total area in the blank and model groups; (I) immunofluorescence staining showing the expression of CXCL12 in the blank and model groups; (J) percentage of the CXCL12-positive area in the total area in the blank and model groups *p < .05, ***p < .001, ns p > 0.05 (no difference).

2.8. Validation of CXCL12 and CCL4 expression in the atherosclerotic mouse model

To provide additional experimental support for the transcriptomic findings related to subsets 9 and 10, we further examined the expression of CXCL12 and CCL4 in the atherosclerotic mouse model. Based on the differential gene expression analysis, CXCL12 and CCL4 were selected as representative markers associated with the CXCL12+ and CCL4+ endothelial programs identified in subsets 9 and 10, respectively.

Immunofluorescence staining showed that both CXCL12 and CCL4 expression levels were increased in the model group compared with the control group (Figure 7(G–J)). Quantitative analysis further confirmed that the positive areas of CXCL12 and CCL4 were significantly higher in atherosclerotic lesions than in the control group (p < 0.05). These findings provide additional experimental support for the pathological relevance of the transcriptomic signatures associated with subsets 9 and 10.

However, because these validation experiments were performed at the tissue level rather than at single-cell resolution, they should be interpreted as supportive evidence for the biological relevance of these molecular programs, rather than definitive one-to-one validation of the single-cell-defined endothelial subsets.

3. Discussion

Atherosclerosis is a complex chronic inflammatory disease and an important pathological basis of various acute and chronic cardiovascular and cerebrovascular diseases. Its progression seriously affects the prognosis, clinical outcome, and quality of life of affected patients. In the present study, we identified endothelial cell subsets enriched in atherosclerotic core (AC) regions compared with proximal adjacent (PA) portions using scRNA-seq data. By integrating an independent bulk RNA-seq dataset comparing stable and unstable plaques, we further observed that several of these subsets were also elevated in unstable plaques, suggesting a potential association with plaque instability rather than a direct equivalence between anatomical regions and pathological states. These subsets also showed increased expression of multiple cytokines related to disease progression, including SRGN, CCL5, CCL4L2, IL32, GZMK, GZMA, ANGPT2, and CTGF, most of which are pro-inflammatory mediators. Since inflammatory microenvironments are known to promote plaque vulnerability and rupture [42], these findings support the possibility that specific endothelial programs may contribute to plaque destabilization. In addition, our KEGG pathway analysis indicated enrichment of PI3K/Akt-related signaling across the three prioritized endothelial programs. Because this pathway is involved in endothelial survival, apoptosis, and inflammatory responses [43], these findings may provide a useful framework for further investigation of endothelial cell involvement in plaque progression. However, the correlative nature of the transcriptomic analyses means that future functional studies are still required to establish causality.

The occurrence and progression of atherosclerosis result from the coordinated involvement of multiple cell populations. With the development of single-cell transcriptomic technologies, increasing attention has been directed toward the role of nonimmune cells in atherosclerosis, among which endothelial cells are of particular interest. Kalluri et al. [44] identified three endothelial cell subsets with distinct heterogeneity in the normal mouse aorta, one of which showed increased expression of genes associated with lipid transport and was potentially linked to early atherosclerotic changes induced by shear stress. In addition, Depuydt et al. [45] identified endothelial heterogeneity in human carotid atherosclerotic plaques and described a subset with features of both endothelial cells and smooth muscle cells, further supporting the involvement of endothelial-to-mesenchymal transition in atherosclerosis. Hu et al. [46] also performed single-cell sequencing of human cardiac arteries and highlighted an important contribution of endothelial cells and macrophages to vascular pathology. In this context, our findings further support the concept that endothelial cells participate in atherosclerosis through multiple biological programs. Nevertheless, rather than claiming a direct one-to-one correspondence between scRNA-seq-defined clusters and pathological plaque states, our study should be interpreted as identifying candidate endothelial programs associated with inflammatory and unstable plaque features across datasets.

We identified three pro-inflammatory endothelial programs of particular interest, characterized by CXCL12, CCL4, and PGF, respectively. Among them, the CXCL12-related endothelial program is notable because CXCL12 has been widely implicated in dyslipidemia, angiogenesis, plaque destabilization, thrombosis, neointimal hyperplasia, and vascular inflammation, all of which are closely related to atherosclerotic remodeling [47]. Interestingly, its major receptor CXCR4 was highly expressed in another inflammatory subset characterized by CCL4, suggesting the possibility of coordinated signaling between endothelial subpopulations and other plaque-resident cells. In addition, the CCL4-related endothelial program also showed increased expression of CCL5, CCL4L2, and IL32, further supporting a role in amplifying local inflammatory signaling. In the revised manuscript, we further strengthened the biological relevance of these transcriptomic observations by supplementary experimental validation showing increased expression of CXCL12 and CCL4 in the atherosclerotic mouse model. However, because these validations were performed at the tissue level rather than at single-cell resolution, they should be interpreted as supportive evidence for the pathological relevance of these molecular programs rather than definitive one-to-one validation of the corresponding endothelial subsets. In addition, SRGN, which was upregulated in the CXCL12-related endothelial program, may also deserve further attention, as previous studies have suggested that it participates in inflammatory signaling through extracellular receptor interactions and cytokine-related pathways [48–50].

The PGF-related endothelial program also deserves particular attention. PGF is a member of the VEGF family and has been implicated in vascular injury, endothelial dysfunction, and plaque destabilization in previous studies [51–53]. In our study, the proportion of PGF-positive endothelial cells was increased in atherosclerotic lesions, and immunofluorescence co-localization together with immunohistochemistry provided more direct orthogonal support for the pathological relevance of this endothelial program. Compared with CXCL12 and CCL4, PGF therefore received stronger experimental support in the current work. In addition, this subset also upregulated genes related to vascular remodeling and inflammatory responses, including ANGPT2, ApoD, and CTGF. These findings suggest that the PGF-related endothelial program may represent a promising marker or mediator of endothelial dysfunction during plaque progression. Nevertheless, the mechanistic role of PGF in endothelial injury, inflammatory amplification, and plaque destabilization remains to be clarified through dedicated functional experiments. Similarly, NEAT1, which was enriched in subset 5 together with a high mitochondrial gene ratio, may represent another candidate regulator of endothelial stress, injury, or apoptosis in atherosclerotic lesions [54].

In addition to these candidate endothelial programs, our pathway enrichment results suggested consistent involvement of PI3K/Akt-related signaling, with additional enrichment of chemokine-, TNF-, and NF-κB-related pathways in specific subsets. These findings are biologically plausible, as PI3K/Akt signaling is known to regulate endothelial survival, proliferation, apoptosis, and inflammatory responses, whereas NF-κB signaling plays a central role in vascular inflammation and immune activation. Importantly, in response to the reviewers’ comments, we added supplementary experimental assessment of PI3K, Akt, and NF-κB in the atherosclerotic mouse model. The increased expression of these molecules in atherosclerotic lesions provides orthogonal support for the pathway-enrichment results and strengthens the biological plausibility of PI3K/Akt- and NF-κB-related signaling in plaque-associated endothelial inflammation. However, these data should still be interpreted as supportive evidence for pathway involvement rather than definitive proof of causal signaling activation.

Several limitations should be acknowledged. First, the sample sizes of both the scRNA-seq and bulk RNA-seq datasets were modest, which may limit generalizability and reduce statistical power for detecting subtle cellular differences. Second, the comparison between AC vs. PA regions and stable vs. unstable plaques involves two related but non-equivalent biological frameworks; therefore, the overlap observed in our analyses should be interpreted as supportive rather than definitive evidence linking specific endothelial subsets to plaque instability. Third, scRNA-seq is subject to technical biases, including dissociation-induced transcriptional stress, batch effects, and selective loss of fragile cell populations, all of which may influence the apparent cellular composition. Fourth, although we added experimental validation for PGF, CXCL12, CCL4, PI3K, Akt, and NF-κB, these validations were performed mainly at the tissue level and only partially bridge the gap between transcriptomic signatures and subset-resolved biology. Therefore, the current study should be considered primarily hypothesis-generating, and future work using larger cohorts, spatial transcriptomics, lineage-resolved approaches, and functional perturbation experiments will be required to establish causality and refine the biological roles of these endothelial programs.

In summary, this study provides an integrated view of endothelial heterogeneity in human carotid atherosclerotic plaques and identifies CXCL12-, CCL4-, and PGF-associated endothelial programs as candidate features linked to inflammatory and unstable plaque phenotypes. By combining scRNA-seq, bulk RNA-seq, and orthogonal experimental validation, our findings extend current understanding of endothelial involvement in plaque biology and generate testable hypotheses regarding intercellular communication and inflammatory pathway convergence in atherosclerosis. These endothelial programs may serve as useful entry points for future mechanistic studies aimed at developing endothelial-targeted strategies for plaque stabilization.

Funding Statement

This work was supported by Key Laboratory of Translational Medicine for the Prevention and Treatment of Diseases with Integrated Traditional Chinese and Western Medicine in Guizhou Province [No.0172023], National Natural Science Foundation of China [No. 82460912], University-level Project of Guizhou University of Traditional Chinese Medicine [No. QNYFZK [2025] 07] and Guizhou Provincial Department of Science and Technology Basic Research Program General Project [Grant No. Qian Ke He Ji Chu MS [2026] 704].

Disclosure statement

No potential conflict of interest was reported by the author(s).

Availability of data and materials

The data used to support the findings of this study are available in the text and can be procured from the corresponding author upon request.

Ethics statement

This study was approved by the Experimental Animal Ethics Committee of Guizhou University of Chinese Medicine (approval number: 2,024,010). All animal studies were performed in accordance with the guidelines and regulations for the use and care of animals at the Center for Laboratory Animal Care, Guizhou University of Chinese Medicine.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15384101.2026.2688663