Endothelial Cell Heterogeneity in Carotid Plaque Calcification: An Exploratory Single-Cell Transcriptomic Study with Proteomic Support

Abstract

Background

Carotid plaque calcification is an active multicellular process with heterogeneous clinical implications. However, endothelial cell (EC) heterogeneity and plaque-region-specific EC states associated with calcified lesions remain incompletely characterized.

Methods

We performed an exploratory integrative analysis of the public single-cell RNA sequencing dataset GSE159677, comprising paired calcified core (AC) and proximal adjacent (PA) tissues from three patients, together with a single-center proteomic cohort of three additional patients with paired AC and PA samples. Major plaque cell populations and EC subclusters were identified by unsupervised clustering and canonical markers. Calcium signaling activity, pathway enrichment, ligand–receptor communication, and Monocle2 pseudotime trajectories were analyzed. Transcriptomic findings were compared with differentially expressed proteins to identify cross-omics candidate molecules.

Results

A total of 35,890 cells were classified into seven major cell types. AC and PA tissues showed distinct cellular compositions and signaling patterns. Re-clustering of 4,925 ECs identified six subclusters, including a calcium signaling-high EC cluster enriched for extracellular matrix organization, inflammatory signaling, cytoskeletal regulation, and endothelial-to-mesenchymal transition-related programs. CellChat analysis indicated plaque-region-specific communication networks involving ECs, immune cells, fibroblasts, and smooth muscle cells. Pseudotime analysis suggested heterogeneous EC state transitions rather than a definitive longitudinal progression. Cross-omics comparison identified eight candidate molecules, FABP4, FABP5, MYL12A, POSTN, S100A10, SERPINB1, SOD2, and TMSB10, with concordant changes across transcriptomic and preliminary proteomic analyses.

Conclusion

These exploratory findings characterize plaque-region-specific EC heterogeneity associated with carotid plaque calcification and nominate candidate pathways and molecules for further validation in larger cohorts and functional models.

Introduction

Carotid atherosclerosis is a major cause of ischemic stroke, and calcification is an important feature of plaque remodeling. However, its clinical significance varies according to calcification burden, morphology, distribution, and location. Small or spotty microcalcifications, particularly near the fibrous cap, may increase local mechanical stress and be associated with plaque vulnerability, whereas extensive macrocalcification may occur in more advanced and potentially more stable lesions.¹ Therefore, calcification should not be regarded as a uniform indicator of plaque instability.

Vascular calcification is an actively regulated multicellular process. Vascular smooth muscle cells may undergo osteogenic phenotypic transition, while immune cells promote inflammation and mineral deposition through cytokines and matrix vesicles.^2,3 Endothelial cells (ECs) may also contribute through endothelial-to-mesenchymal transition, inflammatory activation, and extracellular matrix remodeling.^4,5 Single-cell technologies have further revealed substantial heterogeneity among vascular and immune cell populations in atherosclerotic plaques.^6–11 Recent cardiovascular studies have also emphasized the interconnected roles of inflammation, immune-cell activity, mitochondrial dysfunction, oxidative stress, and extracellular matrix remodeling,^12–15 together with the increasing application of computational approaches to complex cardiovascular data.¹⁶

Despite these advances, EC heterogeneity in human carotid plaque calcification remains incompletely characterized. In particular, it is unclear whether EC subpopulations with elevated calcium signaling activity exhibit distinct inflammatory and matrix-remodeling programs, how these states differ between calcified core and proximal adjacent regions, and which intercellular signaling networks are associated with these regional differences. Moreover, cross-sectional comparisons cannot determine whether proximal adjacent tissue represents a true precursor of the calcified core.

In this exploratory study, we integrated a public single-cell RNA sequencing dataset containing paired calcified core and proximal adjacent tissues with an independent single-center proteomic cohort. We characterized cellular composition and EC subpopulations, evaluated calcium signaling and pathway activity, inferred intercellular communication, and examined putative EC state relationships using pseudotime analysis. Transcriptomic and proteomic findings were subsequently compared to identify cross-omics candidate molecules associated with plaque-region-specific EC states.

Methods

Data Sources and Sample Information

We used the publicly available single-cell RNA sequencing (scRNA-seq) dataset GSE159677 from the Gene Expression Omnibus (GEO) database.¹⁷ This dataset comprised six samples from three patients, with each patient contributing paired tissues from the calcified core (AC) and proximal adjacent tissue (PA). AC refers to the calcified core of the carotid atherosclerotic plaque, whereas PA refers to the patient-matched tissue region located proximal to the calcified core. These designations refer to the anatomical sampling regions used for paired regional comparison. All patients were clinically and pathologically confirmed to have carotid plaque calcification.

Additionally, an exploratory single-center proteomic cohort was established at the Second Hospital of Hebei Medical University. This cohort comprised three additional biological replicates, corresponding to three patients and six paired AC and PA samples. Because the publicly available scRNA-seq dataset and the available paired clinical specimens were fixed in size, no a priori sample-size calculation was performed. All available samples were included, and the analyses were conducted within an exploratory framework. Tissue samples were homogenized, subjected to protein extraction and trypsin digestion, followed by proteomic analysis using an Orbitrap Exploris 480 mass spectrometer (Thermo Fisher Scientific) coupled with an EASY-nLC 1200 liquid chromatography system. Raw data were processed through MaxQuant (v1.6.6) with the Andromeda search engine against the Swissprot.Human.20210312.fasta database, applying a 1% false discovery rate (FDR) threshold.^18,19 Quantitative proteomic data were used for differential-abundance analysis and comparison with the transcriptomic findings to identify concordant candidate molecules. Protein intensities were log2-transformed and median-normalized across samples. Because of the small paired design, no imputation was performed; only proteins quantified in all six samples were retained, and paired AC–PA comparisons were performed on this complete-case matrix.

Data Preprocessing and Quality Control

Raw scRNA-seq data were processed in R (v4.1.2) mainly using the Seurat (v4.0) package.²⁰ Matrix files in 10X Genomics format were read from GSE159677. Rigorous quality control was applied by filtering cells with fewer than 200 detected transcripts to avoid noise from insufficient library construction, and excluding cells with mitochondrial gene proportion >10% to reduce artifacts caused by apoptosis or stress.²¹

Data were log-normalized using the LogNormalization method, and 2,000 highly variable genes were identified with the FindVariableFeatures function. Principal component analysis (PCA) was applied for linear dimensionality reduction, and the first 20 PCs were used to construct a neighbor graph. For cell clustering, the Louvain algorithm was applied,^22,23 followed by t-distributed stochastic neighbor embedding (t-SNE) for visualization of cell population distributions.²⁴

For the single-center proteomics dataset, raw spectral files were searched against the reference database using MaxQuant (v1.6.6) with Andromeda. FDR was controlled at 1%, and potential contaminants or low-confidence peptides were removed to ensure reliable protein identification.

Cell Type Annotation

To guarantee annotation accuracy, we used canonical marker genes reported in the literature. ECs were marked by PECAM1, CLDN5, IGFBP7,²⁵ SMCs by ACTA2, MYH11,²⁶ fibroblasts by COL1A1, COL3A1,²⁷ while immune populations were identified as follows: monocytes (S100A8, CXCL8),^28,29 macrophages (C1QB, CD163),^30,31 T cells (CXCR4, CCL5),^32,33 B cells (MS4A1, CD79A),^34,35 and mast cells (S100B, CPA3).^36,37 To minimize annotation bias from single markers, we combined multiple visualization approaches: DotPlot to show expression distributions across clusters, heatmaps to compare marker expression patterns, and violin plots to highlight intra-cluster expression differences.

Cellular Composition Analysis

The number and proportion of each cell type were calculated for AC and PA tissues. Differences in cellular composition between groups were tested using Fisher’s exact test. Results were visualized with bar plots annotated with statistical significance markers, revealing compositional shifts associated with calcification.

Functional Scoring and Enrichment Analysis

We focused on calcium signaling and lipid metabolism activity. Functional gene sets were retrieved from GeneCards (relevance score >10) and supplemented by literature review to create a calcification-related core gene set.³⁸ Based on the normalized expression matrix generated in Seurat, single-sample gene set enrichment analysis (ssGSEA) was performed using the GSVA package (v1.42) to calculate a relative calcium signaling enrichment score for each cell. Higher scores indicated greater relative enrichment of the predefined calcium signaling-related gene set.

Scoring results were visualized with boxplots and violin plots, and between-group differences were assessed by the Wilcoxon rank-sum test. Furthermore, functional scores were mapped along pseudotime trajectories to characterize how module activity was dynamically activated or inhibited during disease progression. This integrative approach enabled us to connect functional alterations with temporal-spatial transitions.

Endothelial Cell Subpopulation Analysis

To investigate the heterogeneity of ECs in plaque calcification, ECs were extracted and re-clustered. After data normalization, identification of highly variable genes, and PCA dimensionality reduction, the top 15 principal components were used for neighbor graph construction and clustering. The final subpopulations were visualized using t-SNE. Characteristic genes of each subpopulation were identified with the FindAllMarkers function and displayed by heatmaps and volcano plots. We further compared the proportion differences of subpopulations between AC and PA groups and performed GO and KEGG enrichment analyses with clusterProfiler (v4.0) to reveal functional differences among EC subpopulations.³⁹

Cell–Cell Communication Analysis

To study communication patterns under different tissue conditions, we used CellChat (v1.1.3) to construct ligand–receptor interaction networks of AC and PA.⁴⁰ CellChat objects were generated using annotated cell types, followed by subsetting, identification of differential genes and interactions, and calculation of communication probabilities. We then compared overall communication numbers and strengths between AC and PA, and specifically analyzed communication patterns of ECs as signal sources and receivers. Significant signaling axes were visualized by bubble plots and network diagrams.

Pseudotime Trajectory Analysis

To reveal dynamic changes of ECs during calcification, we used Monocle2 to construct pseudotime trajectories.⁴¹ Expression matrices, cell metadata, and gene annotation information were extracted, and highly variable genes were selected as ordering genes based on differential expression results. Dimensionality reduction was performed with the DDRTree algorithm, and cells were ordered along pseudotime into different developmental states (State 1–3).⁴² The trajectories were colored by subpopulation, pseudotime, and state, to reveal distribution differences of ECs under different tissue conditions. Furthermore, limma was applied for DEG analysis between states representing distinct differentiation directions, and results were displayed by volcano plots and heatmaps.⁴³

Validation Across Multiple Data Sources

To verify the stability of DEGs, we conducted differential analysis in the single-center dataset and compared the results with public data. Venn diagrams were used to integrate DEG sets from different sources and analytical dimensions, finally identifying eight stable candidate genes.⁴⁴ Dynamic expression of these genes was further validated along pseudotime trajectories, showing progressive changes among different states, suggesting their involvement in EC-mediated plaque calcification.

Statistical Analysis

All statistical analyses were performed in R (v4.1.2). DEGs were analyzed using the Wilcoxon rank-sum test or the limma package. Unless otherwise specified, two-sided p < 0.05 was considered statistically significant. In visualization, statistical significance levels were marked with asterisks to improve readability.

Multiple-Testing Correction

For single-cell differential expression, marker and DEG identification was performed with Seurat FindAllMarkers/FindMarkers (Wilcoxon rank-sum test), and genes were retained only if the Bonferroni-adjusted p value (p_val_adj, corrected for the total number of tested genes) was <0.05 together with |log2FC| > 0.5 of cells in either group. For state-wise comparisons along the pseudotime trajectory, limma-derived p values were adjusted using the Benjamini–Hochberg (BH) false discovery rate (FDR), and adj.P.Val < 0.05 was used. GO and KEGG over-representation analyses were performed in clusterProfiler with pAdjustMethod = “BH”, and only terms with adjusted p < 0.05 and q < 0.05 were reported. For GSVA/ssGSEA scores, between-group and between-cluster comparisons were performed with limma followed by BH correction, and FDR < 0.05 was considered significant. For proteomics, identification was controlled at 1% FDR at both the peptide-spectrum-match and protein levels (target–decoy). Differential protein abundance was assessed with a two-sided Student’s t test; given the exploratory design and the limited number of biological replicates (n = 3 pairs), nominal p < 0.05 was used for candidate nomination. Unless otherwise stated, adjusted p values are reported throughout, and asterisks in the figures denote adjusted significance levels.

Results

The process of this study is showed in Figure 1.

Figure 1.

The process of this study.

scRNA-Seq Revealed Cellular Populations and Group Differences in Carotid Plaques

To comprehensively characterize the cellular structure of carotid atherosclerotic plaques, we performed scRNA-seq on 35,890 cells isolated from human plaque tissues. Unsupervised clustering followed by t-SNE dimensionality reduction revealed several transcriptionally distinct clusters (Figure 2A). Based on canonical marker genes, we annotated seven major cell types, including ECs, SMCs, fibroblasts, monocytes/macrophages, T cells, B cells, and mast cells (Figure 2B). Representative lineage-specific gene expression further supported classification, such as PECAM1 and CLDN5 for ECs, ACTA2 and MYH11 for SMCs, COL1A1 and DCN for fibroblasts, CD163 and SPP1 for macrophages, CCL5 for T cells, and CD79A for B cells (Figure 2C).

Figure 2.

Annotation of carotid plaque single-cell transcriptomes and comparison of cell type composition. (A) t-SNE plot showing unsupervised clustering of 35,890 plaque-derived cells. Each color represents a distinct transcriptomic cluster. (B) Cell type annotation based on canonical markers, identifying ECs, SMCs, fibroblasts, macrophages, T cells, B cells, and mast cells. (C) Dot plot showing representative marker gene expression in each annotated cell type. Dot size represents the percentage of cells expressing the gene, and color intensity indicates average expression. (D) t-SNE visualization showing cells colored by sample type: calcified core (AC) versus proximal adjacent tissue (PA). (E) Bar chart comparing proportions of cell types between AC and PA groups.

We then compared cell composition between AC and matched PA tissues. Visualization on the t-SNE map showed that cells from both groups were broadly distributed across clusters, but relative abundance differed markedly (Figure 2D). Quantitative analysis demonstrated significant differences in cell type proportions (Figure 2E). Specifically, ECs and fibroblasts were increased in AC, whereas macrophages, monocytes, T cells, and mast cells were significantly decreased (p < 0.001). These results indicate that carotid plaque calcification is accompanied by remodeling of the cellular ecosystem, characterized by expansion of ECs and fibroblasts and reduction of immune components.

EC Subpopulation Heterogeneity Suggests Early Activation Signatures in Calcification

To further investigate potential cellular sources of plaque calcification, we calculated calcium signaling scores across the full scRNA-seq dataset. The results showed clear heterogeneity among different cell types (Figure 3A). Comparison between AC and PA revealed significant differences in calcium signaling activity across multiple cell types (Figure 3B).

Figure 3.

Distribution of calcium signaling scores and EC subpopulation analysis. (A) t-SNE map showing calcium signaling scores of all cells; warmer colors indicate higher scores. (B) Boxplots comparing calcium signaling scores of major cell types between AC and PA. (C) t-SNE map of 4,925 ECs showing six identified subpopulations. (D) Dot plot of representative marker gene expression profiles across EC subsets. (E) Stacked bar chart showing proportional differences of EC subpopulations between AC and PA. (F) Violin plot comparing calcium signaling scores among EC subpopulations.

Given that ECs are positioned at the initiation site of plaque formation and play critical roles in inflammation, matrix deposition, and calcification initiation, we focused on ECs (n = 4,925) for subpopulation analysis. Unsupervised clustering identified six subpopulations with distinct transcriptional features (Figure 3C). Each subset displayed unique marker gene expression patterns (Figure 3D), indicating intrinsic functional heterogeneity.

Comparison between AC and PA revealed significant compositional differences. Cluster 0 predominated in AC, whereas clusters 1 and 2 were significantly enriched in PA (Figure 3E). Moreover, clusters 3–5 existed almost exclusively in PA and exhibited higher calcium signaling scores (Figure 3F), suggesting that PA may represent an early activation stage of calcification.

High-Calcification EC Subpopulation is Significantly Enriched in ECM Remodeling, Inflammation, and EMT Pathways

To further dissect EC subpopulations potentially involved in plaque calcification, we performed detailed analysis of subpopulations with high calcium signaling scores. Highlighted UMAP visualization showed that high-calcification subsets (such as cluster 5) displayed relatively independent aggregation patterns among ECs (Figure 4A).

Figure 4.

Molecular features and enrichment analysis of high-calcification EC subpopulation. (A) UMAP map highlighting distribution of high-calcification EC subpopulation (cluster 5). (B) Volcano plot showing DEGs between cluster 5 and other EC subsets. (C) Heatmap of top DEGs across EC subpopulations. (D and E) GO and KEGG enrichment analyses of genes upregulated in cluster 5. (F) GSVA analysis comparing activities of calcification, inflammation, and EMT gene sets across EC subpopulations, with cluster 5 showing significant enhancement. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

Differential gene expression analysis between cluster 5 and other EC subsets revealed a large number of DEGs (Figure 4B). Heatmaps of top DEGs further demonstrated distinct transcriptional profiles of cluster 5 (Figure 4C).

GO and KEGG enrichment analyses of cluster 5 upregulated genes showed significant enrichment in ECM organization, vascular development, and inflammatory response processes (Figure 4D–E).

GSVA showed significantly increased activity of inflammatory cytokine signaling and EMT/TGF-β–related gene sets in cluster 5 (Figure 4F), complementing the elevated calcium signaling score of this subcluster shown in Figure 3F. Complete gene-set membership is listed in Supplementary Table 1.

In summary, cluster 5 represents a specific high-calcification EC subpopulation, with molecular features closely related to ECM remodeling, inflammatory activation, and EMT, which may play a key role in plaque calcification.

Cell–Cell Communication Analysis Revealed Network Remodeling and Key EC Pathways Between AC and PA

To further investigate intercellular interactions during plaque calcification, we applied CellChat to analyze AC and PA communication networks. At the global level, both groups showed complex interaction networks, but with evident differences in network structure and edge strength (Figure 5A and B). Quantitative comparison indicated that communication number and mean intensity were significantly higher in PA (Figure 5C), suggesting more active signal exchange in adjacent tissues.

Figure 5.

Analysis of cell–cell communication networks in AC and PA tissues. (A and B) Global communication networks of AC and PA. Nodes represent cell types, and edge width represents communication strength. (C) Bar chart showing numbers and average strengths of communications in AC and PA. (D and E) Communication strength distributions of ECs as signal sources or receivers. (F) Ranking comparison of communication strengths across different cell types. (G) AC-specific upregulated signaling pathway. (H) PA-specific upregulated signaling pathway.

Focusing on ECs, results revealed their central hub role in both groups. As signal senders, ECs in AC showed significantly stronger interactions with fibroblasts and SMCs, whereas in PA they communicated more strongly with ECs and macrophages. As signal receivers, ECs in AC received stronger inputs from fibroblasts and SMCs, while in PA they also received extensive signaling from immune cells (Figure 5D–E). This suggests different microenvironmental functions: in AC, signaling may relate to ECM remodeling and proliferation, while in PA, signaling may be associated with inflammation and immune cell involvement.

Comparison of communication strength rankings (Figure 5F) showed that in AC, COMPLEMENT, CTSG, SELE, LIFR, GRN, CysLTs, SEMA3, IL1, IGF, FLRT axes were more prominent, while in PA, GAP, PTPRM, IL16, CD34, THY1, IGFBP, ApoE, SPP1 axes were enriched. These findings were consistent with a potential shift from immune regulation and matrix repair in PA, involving IGFBP, ApoE, and SPP1 signaling, toward inflammatory amplification, vascular stress, and extracellular matrix remodeling in AC, involving SELE, IL1, and COMPLEMENT signaling. Within this proposed framework, PA exhibited an “immune activation–homeostasis maintenance” profile, whereas AC exhibited an “inflammation amplification–calcification fixation” profile.

At the pathway level, AC specifically showed upregulation of the SELE–CD44 axis associated with adhesion and inflammation (Figure 5G),⁴⁵ while in PA, the IGFBP3–TMEM219 axis was activated, related to survival and stress response (Figure 5H).⁴⁶ These findings suggest that ECs engage in distinct signaling pathways to regulate microenvironments during plaque calcification.

Pseudotime Analysis of ECs Revealed Different Developmental States and Key Transcriptional Features

To Investigate Dynamic Changes of ECs During Calcification, We Performed Pseudotime Analysis with Monocle

Trajectory results showed clear branching patterns, classified into three major states (State 1–3) (Figure 6A). Pseudotime ordering indicated progressive differentiation, with State 1 as the starting point, and State 2 and State 3 representing distinct developmental directions (Figure 6B). Subpopulation annotation showed that high-calcification cluster 5 was mainly distributed in State 3, indicating close association with a calcified phenotype (Figure 6C). Comparison between groups revealed clear differences: PA cells tended to States 2 and 3, while AC cells were concentrated in State 1 (Figure 6D).

Figure 6.

Pseudotime trajectory analysis of ECs. (A) EC trajectories colored by state (State 1–3). (B) EC trajectories colored by pseudotime. (C) EC trajectories colored by subpopulations (cluster). (D) EC trajectories colored by tissue source (AC vs PA). (E) Scatter plot of key genes with opposite trends between states. (F) Heatmap showing expression patterns of opposite-trend genes across states and tissues. (G) Volcano plot showing DEGs between State 2 and State 3.

On the molecular level, we identified key genes with opposite trends between AC and progressive states (Figure 6E), confirmed by heatmap visualization (Figure 6F). Genes such as MGP, CLU, MT-CO1 showed significant differential expression, suggesting roles in EC differentiation. Volcano plots further revealed numerous DEGs between State 2 and State 3 (Figure 6G), with upregulated genes including MT-CO1, MALAT1, and downregulated genes such as CLU, MGP, ITLN1.

Together, pseudotime analysis demonstrated distinct differentiation trajectories of ECs under AC and PA conditions, highlighting transcriptional programs potentially driving plaque calcification and microenvironment remodeling.

Validation of Differential Genes and Trajectory Dynamics Revealed Potential Key Molecules

To further validate molecular features of ECs across different states and tissue conditions, we analyzed single-center sequencing data (Figure 7A). A large number of DEGs were identified (Figure 7B). For example, ITGAM, ITGB2, MMP12, and DCN were significantly upregulated under disease conditions, whereas FAM184B, NEXN, and MYH10 were downregulated. Functionally, the changes in ITGAM and ITGB2 were associated with cell adhesion and immune-cell interactions, those in MMP12 and DCN with extracellular matrix degradation and remodeling, and those in NEXN and MYH10 with cytoskeletal and structural remodeling.⁴⁷

Figure 7.

Validation of EC DEGs and trajectory-based dynamic analysis. (A) Representative images of paired carotid plaque specimens. (B) Volcano plot of single-center sequencing data showing significantly upregulated (red) and downregulated (blue) genes. (C) Venn diagram showing overlaps and unique distributions of DEGs from different datasets. (D) Dynamic expression patterns of eight key genes across states and pseudotime trajectories.

Venn diagram analysis revealed overlaps among datasets (Figure 7C). Eight genes were consistently identified in “common state DEGs,” “State 2 vs State 3 DEGs,” and “single-center DEGs”: FABP4, FABP5, MYL12A, POSTN, S100A10, SERPINB1, SOD2, TMSB10. These genes are closely linked to lipid metabolism, cytoskeletal dynamics, ECM remodeling, inflammatory response, and oxidative stress,^48–52 suggesting potential roles in EC state transitions during calcification. Dynamic analysis of these eight genes along pseudotime trajectories showed progressive changes across different states (Figure 7D).

Discussion

In this study, through integrative analysis of single-cell transcriptomics and proteomics, we systematically characterized the heterogeneity of ECs in carotid atherosclerotic plaques and their key roles in the calcification process. The results showed that EC states vary continuously under different tissue conditions and, along the progression from proximal adjacent tissue (PA) to calcified core (AC), undergo a dynamic transition from inflammation-driven phenotypes toward calcification- and matrix remodeling–dominated phenotypes. This not only deepens our understanding of the course of carotid plaque calcification but also challenges the traditional view of PA as a simple static control tissue, suggesting instead that PA and AC represent continuous stages of lesion progression rather than two opposing states.⁵³

First, we revealed significant differences in cellular composition between AC and PA. Immune cells were more abundant in PA, reflecting an inflammation-driven pathological environment. With disease progression, ECs and fibroblasts increased in AC, indicating a cellular shift toward fibrosis- and calcification-dominated ecosystems. This differs from the traditional view that emphasized only osteogenic transdifferentiation of SMCs,⁵⁴ and instead suggests that ECs are also important participants in calcification development.^55,56 At the single-cell level, we identified multiple EC subsets with different functional states. Among them, the high-calcification subpopulation (eg, cluster 5) displayed molecular features closely linked to ECM remodeling, amplification of inflammation, and epithelial–mesenchymal transition, and was clustered within progressive stages along the pseudotime trajectory. This indicates that ECs are not passively adapting after calcification formation, but rather drive subsequent calcification and mineralization early on through inflammatory signaling, stress responses, and metabolic changes.

Further cell–cell communication analysis suggested that ECs consistently occupied central hub positions throughout plaque evolution, but their interaction patterns changed significantly as the disease progressed.⁵⁷ Complementary evidence from Mendelian randomization studies in other vascular diseases also supports the potential involvement of immune cell traits and inflammatory proteins in pathological vascular remodeling.¹⁵ In the PA stage, signals exchanged between ECs and immune/stromal cells were more complex and active, involving IGFBP,⁵⁸ ApoE,⁵⁹ and SPP1⁶⁰ pathways, forming a dynamic “inflammation activation–homeostasis maintenance” environment. This state may help sustain a relatively reversible microenvironment but also lays the foundation for later pathological progression. In the AC stage, signaling shifted toward SELE–CD44, IL1, and COMPLEMENT pathways related to inflammatory adhesion and ECM remodeling,⁶¹ showing more fixed and amplified inflammatory signatures that drive calcification into a stable and difficult-to-reverse state. This transition from “immune-driven” to “calcification fixation” confirms our understanding of the staged progression of disease.⁶²

At the molecular level, through cross-dataset validation, we identified stable key genes including FABP4, FABP5, SOD2, and POSTN. These genes exhibited gradual changes along pseudotime trajectories and were strongly associated with inflammation, lipid metabolism, oxidative stress, and matrix remodeling. Continuous upregulation of FABP4 and FABP5 indicates roles of lipid metabolic disorder and membrane signaling in maintaining the calcified environment.⁶³ Activation of SOD2 suggests adaptive regulation of oxidative and antioxidative stress in ECs,⁶⁴ More broadly, inflammation and oxidative stress are recognized contributors to pathological cardiovascular remodeling, providing additional context for the SOD2-related changes observed in this study.¹² Meanwhile, enrichment of POSTN highlights its importance in ECM remodeling and fibrosis.⁶⁵ Recent evidence further suggests that impaired mitochondrial quality control may contribute to atherosclerosis by promoting oxidative stress, inflammation, and extracellular matrix remodeling.¹⁴ These molecular alterations not only provide clues for mechanistic research but also hold potential clinical value. If their dynamic changes could be captured using blood-based antibody assays or imaging methods, early identification of individuals at high risk of plaque calcification may become feasible. Future studies could also evaluate whether SGLT2 inhibitors or other pathway-modulating drugs affect the molecular processes identified in this study.

Compared with previous studies, the unique feature of this research lies in emphasizing the transitional role of PA. Previously, PA was often used as a “normal” control, but our results demonstrate that PA is not a static or healthy state but rather a highly dynamic pre-calcification stage. At this stage, ECs have already shown transcriptional activation and stress features, and frequent communications with immune cells highlight their hub role within the inflammatory network. This finding suggests that clinical interventions during the PA period may be more promising for preventing calcification than interventions at the AC stage, offering an earlier therapeutic window for reducing atherosclerosis-related stroke risk. Moreover, differential changes in pathways such as SELE–CD44 and IGFBP3–TMEM219 also provide strategies for developing targeted therapies aimed at EC–immune cell or EC–stromal cell interactions in the future.⁶⁶

Nevertheless, this study has several limitations. First, the sample size was relatively small, consisting mainly of public databases and single-center data, and thus generalizability needs to be confirmed in larger, multicenter cohorts.⁶⁷ Second, the work mainly relied on scRNA-seq for inference. Although reliability was improved by integrating proteomic analysis, the lack of spatial omics profiling prevented precise mapping of EC subsets and their microenvironmental localization.⁶⁸ Furthermore, our inferences remain correlation-based, lacking validation in animal models or in vitro functional experiments to directly confirm the causal roles of the identified key genes in calcification.⁶⁹ In addition, scRNA-seq is limited in its ability to capture low-abundance transcripts, non-coding RNAs, and protein modifications, meaning that the full molecular dynamics of disease progression may not be represented.⁷⁰ Finally, patient heterogeneity, including comorbidities such as diabetes or hypertension and medication use, may have influenced EC states, calcification processes, and the proteomic findings, resulting in potential residual confounding.⁷¹

In summary, this integrative single-cell and proteomic analysis proposes a potential evolutionary framework for carotid plaque calcification and highlights the possible involvement of ECs in the coordinated processes of inflammation, calcification, and extracellular matrix remodeling. Within this framework, PA exhibited cellular and molecular features consistent with a possible transitional state rather than those of a simple control tissue. The identified candidate molecules and signaling pathways provide a basis for further mechanistic investigation and biomarker development. Future research integrating spatial transcriptomics,⁷² experimental models, and longitudinal clinical follow-up will be valuable for confirming and extending these findings and evaluating their potential clinical relevance.

Conclusion

This exploratory study integrates single-cell transcriptomics with preliminary proteomic support to characterize endothelial cell heterogeneity in carotid plaque calcification. The findings suggest a potential continuum of endothelial and microenvironmental changes encompassing inflammation-associated features in PA and calcification and extracellular matrix remodeling in AC, with PA displaying characteristics consistent with a possible intermediate state. A calcification-associated endothelial cell subset and candidate molecules, including FABP4, FABP5, MYL12A, POSTN, S100A10, SERPINB1, SOD2, and TMSB10, were identified. These findings provide new insights into vascular calcification and a basis for further mechanistic and translational investigation.

Funding Statement

There is no funding to report.

AI Declaration

Authors declare no use of generative AI in the manuscript preparation process.

Data Sharing Statement

The original contributions presented in the study are included in the article, further inquiries can be directed to the corresponding author.

Ethics Statement

This study involving human clinical samples and clinical data was reviewed and approved by the Research Ethics Committee of the Second Hospital of Hebei Medical University, Shijiazhuang, China (Approval No. 2022-R265; approval date: March 24, 2022). The approved project title was “Clinical Phenotypes and Mechanisms of Carotid Plaque Calcification” under which the present study was conducted. All procedures involving human participants were conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants before sample collection and use of clinical data.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conceptualization, study design, analysis, and interpretation; took part in drafting and reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.