High-Dimensional Single-Cell Multimodal Landscape of Human Carotid Atherosclerosis

Abstract

Background:

Atherosclerotic plaques are complex tissues composed of a heterogeneous mixture of cells. However, our understanding of the comprehensive transcriptional and phenotypical landscape of the cells within these lesions is limited.

Methods:

To characterize the landscape of human carotid atherosclerosis in greater detail, we combined cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) and single-cell RNA sequencing (scRNA-seq) to classify all cell types within lesions (n=21; 13 symptomatic) to achieve a comprehensive multimodal understanding of the cellular identities of atherosclerosis and their association with clinical pathophysiology.

Results:

We identified 25 cell populations, each with a unique multi-omic signature, including macrophages, T cells, NK cells, mast cells, B cells, plasma cells, neutrophils, dendritic cells, endothelial cells, fibroblasts, and smooth muscle cells (SMCs). Among the macrophages, we identified 2 proinflammatory subsets enriched in IL1B or C1Q expression, 2 TREM2 positive foam cells (one expressing inflammatory genes), and subpopulations with a proliferative gene signature and SMC-specific gene signature with fibrotic pathways upregulated. Further characterization revealed various subsets of SMCs and fibroblasts, including SMC-derived foam cells. These foamy SMCs were localized in the deep intima of coronary atherosclerotic lesions. Utilizing CITE-seq data, we developed a flow cytometry panel, using cell surface proteins CD29, CD142, and CD90, to isolate SMC-derived cells from lesions. Lastly, we observed reduced proportions of efferocytotic macrophages, classically activated endothelial cells, and contractile and modulated SMC-derived cells, while inflammatory SMCs were enriched in plaques of clinically symptomatic vs asymptomatic patients.

Conclusions:

Our multimodal atlas of cell populations within atherosclerosis provides novel insights into the diversity, phenotype, location, isolation, and clinical relevance of the unique cellular composition of human carotid atherosclerosis. These findings facilitate both the mapping of cardiovascular disease susceptibility loci to specific cell types as well as the identification of novel molecular and cellular therapeutic targets for the treatment of the disease.

Graphical Abstract

Introduction

Atherosclerosis, the major underlying cause of cardiovascular diseases (CVD), is a chronic inflammatory disease that is initiated by the infiltration and modification of circulating low-density lipoproteins into the subendothelial space. As the disease initiates and progresses, continuous recruitment of circulating immune cells contributes to a complex inflammatory microenvironment. Additionally, smooth muscle cells (SMCs) migrate from the arterial media into the developing lesion, where they undergo phenotypic modulation and transition to a variety of cell fates including foam cells, macrophage-like cells, synthetic extracellular matrix-producing fibrotic cells, and osteogenic-like cells. Emerging genetic and experimental data suggest that specific SMC-derived cell types are detrimental and contribute to plaque instability (e.g., macrophage and osteogenic types), whereas others are protective and contribute to plaque stability (e.g., fibrotic cell types)¹. Increased proportions of pro-inflammatory cells relative to matrix-producing cells within the lesion appear to increase plaque vulnerability and the likelihood of rupture, leading to clinical manifestations such as myocardial infarction and ischemic stroke².

Histological studies have identified culprit lesions as having a thin fibrous cap, large necrotic core, and an overabundance of immune cells like macrophages and T cells^3,4. Moreover, recent genome-wide association studies (GWAS) have identified many genetic loci associated with CVD⁵, and many of these have been mapped to genes that modulate disease specifically through SMC-derived cell, endothelial cell (EC), or immune cell functions. However, these discoveries have not translated into new therapies since the cellular mechanisms driving this complex process have yet to be elucidated fully, cell and molecular mechanistic and causal studies are lacking for most loci, and mechanistic mouse models may not fully recapitulate the complexity of human atherosclerosis⁶.

Recently, there has been an effort to characterize all cell types in human atherosclerosis by utilizing rapidly evolving single-cell technologies to identify cell type-specific candidate genes and molecular mechanisms and to drive novel therapies for atherosclerosis. These studies have relied primarily on single-cell RNA sequencing (scRNA-seq), which uncovers cellular heterogeneity by identifying subpopulations of cells with distinct transcriptional profiles in atherosclerotic plaques. In one study, a multi-omic approach identified innate and adaptive immune cell alterations in carotid plaques associated with clinical events⁷. In another important study, a vast heterogeneity of all cell types was revealed using combined scRNA-seq and single-cell ATAC sequencing to profile immune and nonimmune cells in human carotid atherosclerosis⁸. Similarly, a more recent study utilizing a larger sample size identified alterations in immune cells associated with cerebrovascular events through scRNA-seq analysis⁹. To date, these studies have been relatively small in scale and have not provided a precise reference of cell types and canonical protein markers to facilitate clinical and therapeutic translation. Additionally, the primary focus of scRNA-seq may not be sufficient for separating molecularly similar, yet functionally distinct cell types. Considering that crucial sources of cellular diversity might not be captured entirely through transcriptomic data alone, it is essential to utilize other modalities to characterize comprehensively all cell types within lesions.

To achieve greater precision, we performed CITE-seq utilizing a large panel of antibodies on carotid atherosclerotic plaques in combination with scRNA-seq. Our aim was to leverage CITE-seq to identify valuable markers for defining cellular identity and state, surpassing what can be achieved solely through scRNA-seq. Here we report the most comprehensive phenotypic and transcriptional landscape of all cell types within advanced carotid atherosclerosis and identify cell-type specific protein markers as well as perturbed activation states associated with cardiovascular events in the largest set of human samples to date.

Methods

The data that support the findings of this study are available from the corresponding author upon reasonable request. The authors declare that all data that support the findings of this study are available within the article and its online supplementary files. Detail descriptions of materials and methods used in this study including single-cell preparation of human and mouse specimens, fluorescence-activated cell sorting and flow cytometry analysis, analysis of scRNA-seq and CITE-seq data, pathway analysis, gene scoring analysis, immunohistochemistry of human coronary and carotid arteries, can be found in the online Data Supplement.

Human studies

All human subjects research in this study, including the use of human tissues, conformed to the principles outlined in the Declaration of Helsinki. All patient information was de-identified. For single-cell studies, tissues from human carotid atherosclerotic plaques were collected from twenty-one patients undergoing carotid endarterectomy surgery. These human subject studies were performed with approval (protocol number AAAJ2765) of the local Institutional Review Board (IRB) of Columbia University Irving Medical Center, and written informed consent was obtained from all participants. Exclusion criteria include current infection, known immune system disorder, and active or recent (within last three months) radiation, chemotherapy, hormone-based, and/or immunotherapy treatment for cancer. Table S1 summarizes the demographic and clinical characteristics of the entire cohort, and subgroups to perform comparative analysis. Symptomatic patients were defined as having a stroke or transient ischemic attack (TIA) within 6 months of surgery. Asymptomatic patients had no history of ischemic events. For arterial immunohistochemistry, human coronary artery tissues were obtained from either donor hearts rejected for transplantation or explanted hearts during cardiac transplant surgery. These human subject studies were performed with approval (protocol number AAAR6796) of the local Institutional Review Board (IRB) of Columbia University. Written informed consent was obtained upon admission before surgery.

Statistical Analysis

Normality (Shapiro-Wilk test) and equal variance (F test) were first assessed to determine which downstream statistical test to use. For data that did pass the normality and equal variance tests, an unpaired student T test was performed. For data that did not pass the normality and equal variance tests, a nonparametric test (Mann Whitney) was performed. GraphPad PRISM 9.4 statistical software was used and P values <0.05 were considered to be statistically significant. Wilcoxon rank-sum test was performed to compare expression between symptomatic and asymptomatic plaques.

Results

High-dimensional multi-omic profiling identifies 25 distinct cell populations in human atherosclerotic plaques

To examine the single-cell transcriptome and cellular immunophenotypes of human atherosclerotic plaques, we performed multi-omic profiling on carotid endarterectomy tissue from 21 patients. Tissue samples were digested enzymatically, and cells from six of these patients were labeled with a panel of 274 CITE-seq antibodies before sequencing. Data were integrated from 88,093 cells across all patients (CITE-seq and scRNA-seq) using our deep learning model sciPENN, which also predicted protein expression in the query scRNA-seq dataset from the CITE-seq reference dataset¹⁷ (Fig 1A). All cells were visualized by uniform manifold approximation and projection (UMAP), revealing 25 distinct cell populations (Fig 1B). Utilizing well-established canonical proteins, we identified 2 EC populations and 18 leukocyte populations (Fig 1C and Suppl Fig 1), including all major immune cell types such as macrophages, T cells, NK cells, B cells, plasma cells, mast cells, dendritic cells, and neutrophils. Within macrophages, we observed 5 subtypes (Macrophage 1–5) that had high, near-ubiquitous expression of major lineage markers such as CD64, CD11c, CD14, and MHCII. Three populations of CD4⁺ T cells (CD4⁺ T Cell 1–3) and 2 populations of CD8⁺ T cells (CD8⁺ T Cell 1 and 2), determined by CD3, CD7, CD2, and CD194 expression, were identified. Cluster 15, known to be phenotypically like T cells, was distinguished by the specific expression of the NK cell markers CD56 and CD16. B Cell 1, B Cell 2, and plasma cells were identified by the expression of CD19 and CD20, and KIT⁺ mast cells were identified by expression of CD117. Two small dendritic cell populations (clusters 19 and 22) that comprised plasmacytoid (pDC) and conventional DCs (cDC) characterized by the expression of CD303 and CD141, respectively, were also identified. We classified cluster 24 as neutrophils, which can be difficult to observe in scRNA-seq data, by the expression of CD15. Similarly, we identified two EC clusters as having specific expression of the proteins CD31, CD34, CD144, and CD49b.

Figure 1. Multiomic analysis of human carotid atherosclerosis identifies 25 distinct cell populations.

(A) Experimental design. (B) UMAP visualization of clustering revealed 25 cell populations from 88,093 cells across 21 individuals. (C) Canonical Proteins to identify macrophages, T cells, NK cells, B Cells, Plasma Cells, Mast Cells, Neutrophils, dendritic cells, and endothelial cells. (D) Canonical Genes to identify SMCs and Fibroblasts. (E) Distribution of each major subclass across all samples.

Because SMCs and fibroblasts have not been immunophenotyped extensively, we used canonical gene markers to identify these cell types (Fig 1D and Suppl Fig 2). To differentiate SMCs and fibroblasts that clustered together by phenotypic similarities, we examined the expression of known SMC genes ACTA2, MYH11, and CNN1, as well as fibroblast-specific genes COL1A2, LUM, and LOXL1. This process revealed that clusters 5 and 12 (SMC 1 and 2) were contractile SMCs with the highest expression of all SMC-related genes, whereas clusters 17 (SMC 3) and 18 (Modulated SMC) had lower expression of SMC genes. Modulated SMC had higher expression of fibroblast genes relative to SMC 1 and 2, indicating that this SMC-derived population has undergone phenotypic modulation. Cluster 2, with high expression of COL1A2, LUM, and LOXL1, encompassed all fibroblasts in our dataset.

Next, we calculated the proportion of each cluster for each of the 21 individual subjects (Suppl Figure 3A and B). On aggregate, macrophages comprised 32% of the cells in our dataset, T cells 25%, SMCs 14%, fibroblasts 13%, and ECs 8% (Fig 1E). In contrast, other scRNA-seq studies have identified T cells as the dominant cell type in human carotid plaques^7,8, perhaps due to differences in digestion and isolation protocols or patient heterogeneity. Leukocytes comprised the largest proportions of cells within our carotid atherosclerosis dataset, consisting mainly of macrophages and T cells.

Multimodal analysis provides a deep understanding of plaque cellular phenotypes

To extend our findings beyond known cell surface protein markers, we used our CITE-seq data to identify the distinct gene and protein signatures for all 25 populations depicted in Fig. 1B. To assess the added value of incorporating both RNA and protein information into our analysis, we calculated the gene and protein correlation for every gene-protein pair in our dataset for each individual cluster. As shown in Fig 2A, overall, there was a high degree of correspondence between protein and gene expression (Table S3), although it is well-established that there can be variability between protein levels and their coding transcripts²³. In several circumstances, the protein expression contributed critically to the identification of cell types when the RNA was not highly expressed (Suppl Fig 4). For example, established neutrophil protein markers CD15 and CD16 were highly expressed and specific although their corresponding genes, FUT4 and FCGR3A, were not detectable.

Figure 2. Multimodal biomarkers of cells within carotid atherosclerotic plaques.

(A) Heatmap of all cell types identified in carotid atherosclerosis. Markers include top 10 mRNA (Left) and top 10 protein (Right) features identified by differential expression. (B) Feature plots displaying unique protein markers for SMCs (Top) and fibroblasts (bottom). (C) Representative immunohistochemistry images from n=5 samples of CD29, CD142, and CD90 in carotid atherosclerosis plaques in early (top row) and late (bottom row) lesions. Scale bars represent 100uM. (D) Feature plots displaying proteins to differentiate endothelial cell 1 and endothelial cell 2. (E) Pathway analysis comparing endothelial cell 1 and endothelial 2 using genes from differential expression analysis.

Our five macrophage clusters segregated into three major phenotypic classes. Macrophages 1 and 5 had high expression of FOLR2, cadherin 11 and CD93, suggesting roles in inflammation, phagocytosis, and cell adhesion. In contrast, Macrophages 3 and 4 had high expression of CD36 and TREM1, suggesting a role in lipid uptake and metabolism. However, Macrophage 3 uniquely expressed CD35, a receptor for complement activation. Macrophages 2 and 3 shared expression of CLEC12A and CD192 (CCR2), indicating that they are more inflammatory and likely recently recruited from circulation (Suppl Fig 5A). Macrophage 4 had the highest expression of genes involved in lipid metabolism such as ABCA1, LPL, FABP5, and APOE, whereas macrophages 2 and 3 had the highest expression of inflammatory genes such as IL1B, NLRP3, and CCR2 (Suppl Fig 5B). Macrophages 1 and 5, notably, had intermediate expression of a substantial proportion of genes involved in both functions. Because there was considerable overlap in the gene expression profile of all macrophage subtypes, we identified which upregulated and downregulated genes were unique to each population (Suppl Fig 5C). Using this gene set, we performed pathway analysis to identify in more detail the predicted functional differences between these populations (Suppl Fig 5D). This suggested both substantial functional heterogeneity as well as degrees of overlap of plaque macrophages.

CITE-seq profiling was particularly helpful for characterizing SMC, fibroblasts, and ECs using novel protein phenotypic markers. The SMC populations (SMC 1–3) had relatively uniform expression of the proteins CD29, CD142, and EGFR. The modulated SMC population also had a relatively uniform expression of CD29 and EGFR but lacked expression of CD142. In contrast, fibroblasts expressed CD90 to a similar degree as the modulated SMC population, but specifically expressed PDPN and CD13 (Fig 2B). To evaluate the spatial distribution of specific marker proteins in carotid atherosclerosis, immunohistochemical analysis was conducted on tissue sections targeting CD29, CD142, and CD90 (Fig 2C). In early-stage lesions, there was prominent expression of CD29, aligning with the CITE-seq analysis suggesting its presence in all SMCs and fibroblasts. In contrast, early-stage lesions exhibited an absence of CD142 and CD90 expression. CD29 continued to be expressed in advanced lesions, while unlike early lesions there was expression of CD142 and CD90. An analysis of the respective expressions of CD90 and CD142 on contiguous histological sections revealed the presence of CD90+CD142- cells in carotid plaques (Fig 2D), corroborating our CITE-seq observation that the modulated SMC population was characterized by robust CD90 expression and lack of CD142 expression.

Our clustering also revealed two distinct EC populations (Fig 1B and C). To differentiate these, we identified proteins that were expressed uniquely on one but not the other, e.g., CD49f, CD112, and CD200 were expressed on endothelial cell 1 while CD62P, CD61, and CD201 were highly expressed on endothelial cell 2 (Fig 2E). To delineate these cell types further, we performed a differential expression analysis followed by pathway analysis using inferred functional differences (Fig 2F). This analysis revealed that the endothelial cell 1 population represented activated and inflamed endothelium, suggesting that this cluster promotes leukocyte extravasation. In contrast, the endothelial cell 2 population had upregulation of fibrotic pathways, suggesting that this cluster may be predisposed to undergo endothelial to mesenchymal transition.

Cerebrovascular events are associated with dysregulation of macrophages as well as specific smooth muscle cell, and endothelial cell subpopulations

UMAP visualization of samples revealed that the majority of clusters overlapped between asymptomatic and symptomatic patient groups (Fig 3A, 3B). However, some striking differences were identified, including reductions in the proportions of Macrophage 1, Endothelial Cell 1, SMC 2, and modulated SMC populations, and a modest increase in the SMC 3 population in symptomatic patients (Fig 3C). The reduction in Macrophage 1, a macrophage population with efferocytotic functions (Suppl Fig 6A), suggests that the capacity to clear dead and dying cells is compromised. The decrease in the activated Endothelial Cell 1 population may suggest increased plaque erosion. The decrease in modulated SMCs and SMC 3 subpopulation increase suggests a shift to pro-inflammatory SMCs (Suppl Fig 6B). Unlike other studies⁷, we did not see alterations in the proportion of T cell subpopulations (Suppl Fig 6C). Contrary to findings in previous publications⁷, differential expression analysis revealed only three significant DE genes associated with clinical status: S100A8, DDT, and PTGS1 (Suppl Fig 6D).A higher expression of S100A8 predominantly in the macrophage 2 and neutrophil clusters was revealed. Similarly, DDT expression was more pronounced in cDCs, and PTGS1 expression was elevated in mast cells (Suppl Fig 6E). Interestingly, these genes did not show differential expression in the clusters in which they were expressed most abundantly.

Figure 3. Cerebrovascular events are associated with alterations in the distribution of certain cell populations.

(A) UMAP colored by clinical status. (B) Vertical bar graph showing proportion of all clusters by clinical status. (C) Box Plots showing distribution of select cell populations including macrophage 1, endothelial cell 1, SMC 2, SMC 3, and modulated SMC. (D) Gene scoring analysis comparing senescence in macrophage clusters in asymptomatic and symptomatic plaques. (E) Gene scoring analysis comparing glycolysis in SMC and fibroblast clusters in asymptomatic and symptomatic plaques.

To assess biological processes that are associated with clinical events in more depth, we performed a gene scoring analysis for macrophages, SMC, and fibroblasts. Generally, macrophages in symptomatic plaques were more senescent (Fig 3D), exhibited a higher level of ER stress through PERK activation, had a higher degree of inflammasome activation, but also were pro-resolving (Suppl Fig 6F). Like macrophages, SMCs and fibroblasts from symptomatic plaques were more senescent and PERK was more activated than were those from asymptomatic plaques (Suppl Fig 6G). Also, SMCs and fibroblasts were more glycolytic suggesting a metabolic shift in plaques associated with clinical events, as has been noted in plaques with features of plaque instability in mice²⁴ (Fig 3E).

CD90 is a marker for modulated SMCs in advanced atherosclerosis

There is increased recognition of the significance of modulated SMCs in the development of atherosclerosis. Nevertheless, there is a scarcity of promising therapeutic targets. Our CITE-seq analysis identified CD90 as a prominent marker of modulated SMCs. To determine its spatial distribution and pathophysiological relevance in atherosclerosis, with immunohistochemistry we demonstrated CD90 expression at the necrotic core within advanced atherosclerotic lesions (Fig 4A). Notably, CD90 expression levels were substantially greater in symptomatic compared to asymptomatic plaques in our CITE-seq data (Fig 4B). Using flow cytometry, we were able to isolate a distinct population of CD45⁻CD31⁻CD90⁺ cells in suspensions from carotid atherosclerotic plaques (Fig 4C).

Figure 4. CD90 expressing cells are associated with advanced atherosclerosis in humans and mice.

(A) Representative immunohistochemistry staining of CD90 in early lesion and necrotic core of same tissue sections. (B) Violin plot comparing CD90 expression in asymptomatic and symptomatic plaques from the CITE-seq analysis. (C) Flow cytometry analysis identifying a subset of cells in human carotid atherosclerotic cell suspensions that express CD90. (D) Experimental design for generation and atherosclerosis induction inof SMC-lineage traced mice. (E) Feature plot displaying the expression of CD90.2 in ZsGreen+ cells in mouse CITE-seq analysis. (F) Proportion of ZsGreen+CD90+ cells during atherosclerosis progression. Line shows mean of n=3. (G) Flow cytometry analysis of aortas from 16-week WD fed LDLr−/−Myh11-CreERT2ROSA26ZsGreen+/− mice that identified a significant proportion of CD90+ cells are SMC-derived. (H) Quantification of percentage of CD90+ cells that are either ZsGreen- or ZsGreen+ in flow analysis. Values are shown as mean±SD, n=3.

The difficulty in defining SMC lineage in human atherosclerosis confounds the identification of SMC origin of cellular clusters in lesions. To probe this with more certainty, we sought to identify CD90+ SMC-derived cells in mouse atherosclerosis using lineage tracing methods. Thus, we used male LDLr^−/−Myh11-CreER^T2ROSA26^ZsGreen+/− mice, which permanently labels SMCs and their progeny with the fluorescent reporter ZsGreen following tamoxifen administration. The mice were then subjected to western diet feeding for various time points (0, 8, 16, and 26 weeks) (Fig 4D). The aorta was harvested at each time point and CITE-seq profiling was performed using a 119 antibody panel. This analysis revealed 22 distinct cell populations in mouse atherosclerosis (Suppl Fig 7A), each having a unique gene (Suppl Fig 7B) and protein (Supple Fig 7C) signature. Upon examining the expression of CD90.2 on ZsGreen⁺ cells (Fig 4E), we observed an enrichment in modulated SMCs and fibroblast-like SMCs, agreeing with our human CITE-seq analysis. More importantly, there was an increase in SMC-derived CD90.2 cells during the development of atherosclerosis (Fig 4F). Furthermore, we performed flow cytometry analysis to substantiate the presence of CD90⁺ SMC in murine atherosclerosis in 16-week western diet fed mice (Fig 4G). Leukocytes were excluded first by gating for CD45⁻ cells, and then CD90⁺ cells were selected. Of these CD90⁺ cells, approximately 20%, were of SMC origin based on the presence of ZsGreen fluorescence (Fig 4H).

Deep sub-clustering reveals diverse macrophages phenotypes in carotid plaques

Initial clustering identified 5 plaque macrophage clusters (Macrophage 1–5) (Figs. 1 and 2). To explore cellular diversity more deeply, we performed a sub-clustering analysis on macrophage and DC populations. It revealed 10 cell clusters, 7 macrophage subpopulations, 2 DC clusters, and residual T cells expressing CD3 (Fig 5A and Suppl Fig 8A and B). From the initial clustering (Suppl Fig 8C), Macrophage 1 mapped to clusters 0 and 4, Macrophage 2 mapped to cluster 1, Macrophage 3 mapped to cluster 3, and Macrophage 4 mapped to clusters 2 and 5. In contrast, Macrophage 5 was distributed throughout the subclusters. Cluster 7 was comprised of cells for every macrophage population in the initial analysis. Clusters 0 and 1 were highly inflammatory, expressing genes C1Q and IL1B, respectively, and were enriched in inflammatory pathways (Fig 5B). Clusters 2 and 4 expressed foam cell marker genes, in which cluster 2 expressed ABCA1, LPL, CD36 and TREM2 most prominently, whereas cluster 4 expressed APOE in addition to inflammatory genes like CCL18 and C1Q genes. Additionally, apoptotic (cluster 3), proliferative (cluster 5), and ACTA2⁺ (cluster 7) subpopulations were identified. Cluster 3 showed upregulation of the granzyme A apoptotic pathway (Fig 5B), along with high expression of mitochondrial genes (Suppl Fig 8A and B). The proliferative subpopulation had specific expression of many proliferation markers including MKI67, TUBB, and STMN1 (Suppl Fig 8B). Finally, the ACTA2⁺ subpopulation (~6% of all macrophages) was, notably, characterized by high expression of several SMC-related genes such as MYOCD, ACTA2, MYH11, and CNN1 (Suppl Fig 8B), along with protein expression of CD64, CD11c, and MHCII proteins (Suppl Fig 8D), suggesting that it could represent phenotypic switching of SMCs to macrophage-like cells. This subpopulation also exhibited upregulation of fibrotic signaling and epithelial-mesenchymal transition pathways (Fig 5B).

Figure 5. Sub-clustering analysis of myeloid cells reveals further macrophage phenotypic and functional heterogeneity.

(A) Sub-clustering analysis of myeloid cells. Macrophage 1–5, pDC, and cDC clusters were selected from the initial clustering analysis, revealing 10 clusters. (B) Heatmap showing top upregulated pathways for each macrophage subpopulation based on top DE genes. (C) Heatmap showing top predicted transcriptional regulators for each macrophage subpopulation. (D) UMAP overlaying expression of common M1 and M2 signatures (left), and violin plots quantifying median expression for each macrophage subpopulation. (E) Gene scoring analysis for inflammasome activation, DNA damage, and efferocytosis across macrophage subpopulations.

To establish regulatory features of each macrophage subpopulation, we conducted an upstream analysis using top DE genes and identified potential transcriptional regulators (Fig 5C). The C1Q^hi (cluster 0) and IL1B^hi (cluster 1) subpopulations shared regulatory features, like STAT1 and RELA, involved in immune modulation^25,26. The proliferative subpopulation (cluster 5) exhibited the most specific regulatory features with high inferred regulation by the oncogene MYC. The ACTA2⁺ subpopulation showed inferred transcriptional regulation by MRTFB, MRTFA, NOTCH3, and SMAD3, known to control SMC differentiation^27–30. To infer M1/M2 canonical functional properties, we tested enrichment of M1- and M2-associated genes²¹ in our dataset (Fig 5D). Only IL1B^hi macrophages showed enrichment in M1 genes. Conversely, the C1Q^hi, foamy 1, foamy 2, and apoptotic subpopulations exhibited strong expression of M2 genes. The proliferative and ACTA2⁺ subpopulations did not show a strong M1 or M2 signature. These data align with literature suggesting that the M1/M2 classification inadequately captures the in vivo phenotypic diversity of macrophages in atherosclerotic lesions³¹.

Although we did not observe any differences in macrophage subpopulations between asymptomatic and symptomatic plaques (Suppl Fig 8E), we evaluated which cell populations had upregulated gene expression for processes known to impact atherosclerosis. To do so, we utilized a list of genes involved in inflammasome activation, DNA damage, and efferocytosis to perform gene score analyses for these processes (Fig 5E). As expected, the IL1B^hi subpopulation was the only one with marked inflammasome activation. The proliferative subpopulation had highest expression of genes involved in DNA damage, while each of the C1Q^hi, foamy 1, and foamy 2 subpopulations had evidence for enhanced efferocytotic function.

Deep sub-clustering reveals multiple SMC and fibroblast subclasses in plaques

In initial clustering we identified 3 SMCs, 1 modulated SMC, and 1 fibroblast population (Figs. 1 and 2). This high-level clustering may not reveal the true heterogeneity within these cell types. A recent study that performed snRNA-seq of human aortic tissue identified 7 SMC and 4 fibroblast populations³². Therefore, we performed a deep sub-clustering analysis on SMCs and fibroblasts that revealed 10 unique clusters with specific gene expression profiles (Fig 6A). Using classical SMC and fibroblast phenotypic markers (Fig 6B), we found that clusters 0 and 2 (major SMC 1 and 2) represented the two major classes of contractile SMCs with the highest expression of MYH11 and ACTA2. Two smaller contractile SMC populations, clusters 4 and 6 (minor SMC 1 and 2), were also identified. Minor SMC 1 had moderate expression of MYH11 and TAGLN, but very low expression of other contractile SMC genes. Minor SMC 2 had moderate expression of MYOCD, MYH11, and SMTN, but low levels of other SMC genes. The most striking population was cluster 5, which had low expression of SMC genes and the specific expression of many genes involved in lipid metabolism, suggesting that this large population may be SMC-derived foam cells (Suppl Fig 9A). This foamy SMC cluster mapped almost exclusively back to the modulated SMC population in the original analysis (Suppl Fig 9B). We identified 3 fibroblast populations, clusters 1, 3, and 8, determined by the expression of BGN, DCN, and COL1A1. Cluster 9, the last cluster identified, represented a fibromyocyte, having the highest expression of VCAM1 and LY6E (Suppl Fig 9C).

Figure 6. Sub-clustering of SMC and Fibroblasts reveals distinct subpopulations.

(A) Sub-clustering analysis of SMC and fibroblasts. SMC 1–3, modulated SMC, and fibroblast clusters were selected from the initial analysis, revealing 10 clusters. (B) Dotplot showing expression of classic SMC genes with corresponding phenotypes. (C) Radar plot showing pathways for SMCs obtained from DE genes. (D) Radar plot showing pathways for fibroblasts obtained from DE genes. (E and F) Immunohistochemistry staining targeting foamy SMC markers in atherosclerotic tissue sections. (E) carotid artery and (F) coronary artery. DAPI (blue), Smoothelin (green), LipidTOX (red), and CD90 (magenta). White arrows indicate Smoothelin⁺LipidTOX⁺CD90⁺ cells in neointima region of lesion. Colocalization of all 3 markers will appear white. Scale bars=100μm. (G) Box plots showing distribution of select cell populations including foamy SMC, minor SMC 2, and fibromyocyte comparing asymptomatic and symptomatic plaques.

To identify functional differences in the SMC subpopulations, we performed pathway analyses of SMC and fibroblast sub-clusters. Consistent with annotation as prototypical contractile SMC populations by DE gene analysis, our pathway analyses (Fig 6C) revealed that major SMC 1 and 2 were enriched for muscle functions, TGF-β signaling, and actin cytoskeleton signaling. Coincident with downregulation of contractile SMC genes, minor SMC 1 had upregulated pathways involved in mRNA degradation and cell cycle regulation, suggesting that it might be an actively dividing and proliferating subpopulation. Fibroblast clusters had upregulation of several fibrotic pathways, as did necroptotic and apoptotic pathways in the fibromyocytes (Fig 6D).

Since our data and those from recent mouse and human studies suggest that many foam cells may be SMC-derived^33,34, we used a combination of markers revealed by CITE-seq and functional inference to design an immunostaining protocol to spatially locate foamy SMCs in human carotid and coronary artery sections. As noted, the foamy SMC cluster mapped primarily onto the modulated SMC cluster in our initial analysis (Suppl Fig 9B), expressed the SMC-specific gene smoothelin (Fig 6B), and highly expressed the protein CD90 (Suppl Fig 9D). Immunohistochemistry of human carotid and coronary artery sections with a combination of smoothelin, CD90, and LipidTOX (to identify foam cells) (Fig 6E and F and Suppl Fig 10) revealed a sub-cluster of neointimal cells that stained for all three markers, validating and localizing modulated SMC-derived foamy cells in human atherosclerotic lesions. Notably, these cells localize to similar areas of lesions in both coronary and carotid arteries.

Given differences in SMC 2, SMC 3, and modulated SMCs by clinical status in our initial clustering (Fig 3C), we examined the cell distributions by clinical events in our deep cell type clustering and found subtle decreases in the proportions of foamy SMCs and fibromyocytes, with a significant increase in minor SMC 2 and increased PERK activation in symptomatic plaques (Fig 6G and Suppl Fig 9E).

Discussion

Through the largest and most in-depth scRNA-seq and high-dimensional CITE-seq phenotypic characterization of human carotid atherosclerosis to date, we provide novel insights into the cellular and molecular pathogenesis of atherosclerotic CVD. Our study reveals a comprehensive landscape of 25 cell types, including macrophages and their regulatory features, immunophenotypic and functional characterization of SMCs and fibroblasts, and cell type-specific perturbations associated with cerebrovascular events. We identified 7 macrophage, 3 fibroblast, and 7 SMC subpopulations, including 3 phenotypically modulated or ‘switched’ SMC subpopulations. Each cell population displayed a distinct gene and protein expression signature, reflecting their unique functions. For the first time, we also offer marker panels for immune-isolation and localization of key modulated SMC populations to facilitate further functional and translational studies.

Our work expands on recent advancements in flow cytometry and CyTOF that have improved our understanding of atherosclerosis by revealing the diversity of cell types in this disease^7,35,36 and also contributes beyond those made to date by single-cell genomic technologies, as exemplified in the work by Fernandez et al.⁷ and Depuydt et al.⁸. These studies have enhanced our knowledge of cellular heterogeneity and regulatory networks in atherosclerosis^{7,8,29,37–46}. Fernandez et al.⁷ used CyTOF, scRNA-seq, and selective CITE-seq to investigate immune cell diversity in human carotid atherosclerosis, but their study lacked nonimmune cells and only single-cell profiled 6 patients. Depuydt et al.⁸ expanded the scRNA-seq analysis to 18 patients, including immune and nonimmune cells, and performed scATAC-seq to identify transcriptional regulators. However, both studies lacked detailed phenotypic characterization of cell types within lesions, which is critical for understanding their roles in the disease.

Despite extensive evidence in mouse and human atherosclerosis that SMC, SMC-derived cells, and fibroblasts comprise the majority of cells within atherosclerotic lesions⁴⁷, single-cell profiling studies in humans have focused primarily on leukocytes. To address this gap, we performed a comprehensive characterization of SMCs and fibroblasts in human lesions. Our analysis identified unique surface proteins, including CD29 and EGFR, that are expressed universally across these cell types, while CD90 was specific to modulated SMCs and fibroblasts, and CD142 was not expressed on the modulated SMC population. These findings enable isolation of these cell types by flow cytometry and determination of the spatial location of each cell type within atherosclerotic tissue, thereby providing critical insight into their functions and potential cell-cell interactions. Sub-clustering analysis revealed previously unknown heterogeneity, including two major contractile SMC and two modulated SMC subpopulations (foamy and fibromyocyte). Mouse models suggest that fibromyocytes contribute to plaque stabilization³⁸, while the role of SMC-derived foam cells remains unclear. Our preliminary evidence suggests that SMC-derived foam cells are localized to the neointima of advanced coronary lesions, not the fibrous cap, shedding light on their possible functions and lesion domain interactions in atherosclerotic lesion subdomains.

Our extensive plaque characterization suggests that macrophages are the predominant leukocyte population, contrary to findings from Fernandez et al. and Depuydt et al., who identified T cells as the most abundant immune cell type in lesions. Previous scRNA-seq studies of mouse and human lesions indicated a higher prevalence of myeloid cells than T cells^38,41,42, highlighting the need for standardization of sample processing, analytical methods, and detailed cross-species comparisons. Using protein and gene expression, we identified distinct phenotypic, functional, and activation states of plaque macrophages. Sub-clustering analysis confirmed two proinflammatory macrophage subsets (IL1B^hi and C1Q^hi) with enriched inflammatory pathways, consistent with other single-cell studies in different disease contexts⁴⁸. The IL1B^hi subset exhibited the most distinct inflammatory gene and protein expression profile, including the genes IL1B, NLRP3, CCR2, and the protein CD192. This suggests that IL1B targeting with canakinumab, which reduced CVD events in the CANTOS trial⁴⁹, may act preferentially on these macrophages. Of further translational relevance, our recent work in mice with collaborators suggests that inhibiting IL1B in individuals with clonal hematopoiesis (CH) could effectively reduce CH-driven CVD risk by targeting inflammatory macrophages⁵⁰.

Our macrophage findings support, in part, a recent transformative study in mice that identified nonfoamy macrophages as having a distinct proinflammatory transcriptome and suggested that lipid accumulation in macrophages may not be a driver of plaque inflammation and vulnerability³⁹. Yet our data suggest complex functions for macrophage foam cell sub-types. We discovered two distinct foamy macrophage types in lesions. Foamy 1 had a classic foam cell phenotype with enriched lipid metabolism genes (ABCA1, LPL, FABP5) and high CD36 protein expression. This population predominantly expressed TREM2, which is proposed to protect against atherosclerosis by facilitating cholesterol metabolism through inducing expression of APOE, LPL, and LIPA, while simultaneously reducing inflammation⁵¹. However, in potential conflict with this model, a recent preprint revealed that macrophage-specific deletion of Trem2 in mice reduced atherosclerosis progression and plaque burden⁵² suggesting that TREM2 may be atherogenic at least under certain circumstances. These studies highlight the critical need for additional research in this area, e.g., to determine if Trem2 agonism or antagonism and under what circumstances is a viable approach for preventing atherosclerosis. In this context, our Foamy 2 macrophage population expressed TREM2 and certain foam cell genes (APOE) as well as C1Q genes, indicating an intermediate state between Foamy 1 and inflammatory C1Q^hi populations. In fact, we identified more than one cluster in our initial analysis (Macrophage 1 and 5) that had intermediate expression of genes involved in both inflammation and lipid metabolism. Further research is needed to understand the functions of these foamy-inflammatory macrophage types and their impact on disease progression and clinical events.

We found a macrophage subpopulation expressing proliferative genes, like the stem-like macrophages identified in murine atherosclerosis through fate-mapping⁵³. Macrophage proliferation is prominent in advanced stage atherosclerosis in mice⁵⁴ and can also occur distinctly during lipid-induced regression and plaque stabilization⁵⁵. Thus, this proliferative population may be responsible for local macrophage proliferation in advanced human atherosclerosis or may act as a transitional cell state giving rise to various phenotypes, including macrophages involved in plaque regression and stabilization. These findings highlight the diversity and complexity of macrophage subtypes within plaques⁵⁶, emphasizing opportunities but significant uncertainties in targeting specific macrophage populations with precision therapeutics.

Our study supports the concept of marked cell type transitions in human atherosclerosis. We identified a small (~6% of all macrophages) but distinct population of macrophages that expressed SMC-specific genes and displayed a fibrotic pathway phenotype regulated by known SMC transcription factors^29,57. Interestingly, these macrophages also expressed classic macrophage proteins, which has been observed previously in advanced human coronary atherosclerosis³³, suggesting they are of SMC origin. This finding aligns with previous mouse studies showing SMC plasticity during atherosclerosis progression, including SMC-derived macrophages, fibroblasts, osteogenic cells as well as foam cells^34,37,38,42. Additionally, we identified a substantial population of SMC and likely SMC-derived cells that had a foam cell gene signature, with specific expression of genes involved in lipid metabolism, agreeing with recent studies suggesting that a large portion of foam cells within atherosclerotic plaques may be of SMC origin^33,34. We also observed two distinct EC populations, with one showing upregulated fibrotic pathway genes suggestive of endothelial-to-mesenchymal transition. Our findings provide human evidence that supports recent studies in mice, highlighting the heterogeneity and transitions of ECs during atherosclerosis³⁷. Importantly, this EC pathophysiological process has been linked to increased disease severity in mouse models of atherosclerosis and in human coronary disease^58,59.

We found that specific cell-type alterations may associate with symptomatic clinical events. Our clinical analyses, however, may have limitations due to small sample sizes, patient selection biases, and time delays between cerebrovascular events and surgeries. Despite limitations, we observed decreases in efferocytotic macrophages, activated ECs, contractile SMCs, and foamy modulated SMCs, along with an increase in inflammatory SMCs in symptomatic plaques. These alterations indicate a loss of efferocytotic macrophages, erosion of the EC monolayer, and a shift towards an inflammatory SMC/SMC-derived state, in plaque vulnerability². Additionally, macrophages in symptomatic plaques showed increased senescence, also reported to associate with plaque vulnerability^60,61. SMCs in symptomatic plaques exhibited an increased glycolytic gene signature, extending prior findings of altered metabolism in plaques overall of high-risk patients⁶². As a corollary, alterations in bioenergetic mechanisms in mouse models of atherosclerosis have been found to control SMC differentiation and to be associated with features of plaque instability^24,63. These clinical associations provide the foundation for pre-clinical and therapeutic studies targeting specific cell types and molecular programs to improve outcomes in atherosclerotic CVD, particularly in high-risk patients unresponsive to lipid-lowering therapy.

Our study suggests that CD90 may play a significant role in SMC phenotypic changes in the progression to and pathophysiology of advanced atherosclerosis. Originally identified as a thymocyte marker⁶⁴, CD90 has since been found in various cell types, including neurons, retinal ganglion cells, fibroblasts, vascular pericytes, endothelial cells, mesangial cells, and both hematopoietic and mesenchymal stem cells⁶⁵. The specific role of CD90 on SMC phenotypic modulation in atherosclerosis, however, remains uninvestigated. Our findings reveal that CD90 expression localizes almost exclusively to the necrotic core of advanced carotid atherosclerotic lesions and its expression is also associated with clinical events. Using lineage tracing, we confirmed that a substantial portion of the CD90⁺ cells in murine atherosclerosis originate from SMCs. Additionally, CD90-expressing foam cells are observed in the neointima of both coronary and carotid lesions. Collectively, these observations implicate CD90 as an important contributor to the transition of SMCs into foam cells. This transition could accelerate the development of atherosclerosis and exacerbate its severity, thereby leading to more adverse outcomes. If so, the potential of CD90 as a target for therapeutic intervention in atherosclerosis is underscored.

Although our dataset provides extensive insights into human carotid atherosclerosis, its sample size is still small and the currently used analytical methods are prone to technical biases and artifacts. Our findings must be corroborated further to determine their relationship to cardiac events. To gain more confidence in identifying causal alterations associated with clinical CVD events, it is crucial to increase the number of patient samples and diversity, and to utilize advanced computational and integrative genomics methods and large-scale human genetics. These integrative studies will help identify specific cell types and molecular programs that play a causal role, enabling mechanism-based translation for clinical and therapeutic applications.

In summary, we present the most advanced in-depth multimodal atlas of advanced human carotid atherosclerosis that reveals a vast cell-type heterogeneity of macrophages, SMCs, and fibroblasts, and identifies cell type-specific alterations associated with clinical events. We characterized SMC and fibroblasts in unprecedented depth, revealing 3 populations of phenotypically modulated SMCs, including a foamy population that resides in the deep intima of the developing lesion. Further, we have uncovered complex macrophage subpopulations with distinct gene, protein, and transcriptional regulatory features. These data are foundational for future studies mapping CVD susceptible loci to specific cell types, with the goal of identifying new CVD targets specific for certain cell populations fundamental to the progression and regression of disease. Such integrative approaches provide the analytical and conceptual framework for the discovery of novel therapeutics that target specific cell populations in CVD.

Highlights.

• We performed an in-depth and extensive cellular and molecular characterization of human carotid atherosclerosis by generating the largest and most comprehensive multimodal dataset to date, revealing novel insights into the pathogenesis of the disease.

• Cell type-specific perturbations were identified that are associated with cerebrovascular events.

• A diverse array of macrophage phenotypes were identified for the first time, including inflammatory, foamy, proliferative, and ACTA2⁺.

• SMCs and fibroblasts were profiled in unprecedented depth, discovering a CD90⁺ foamy SMC subset that is present in the necrotic core in human atherosclerosis and that expands during the progression of atherosclerosis in mice.

Nonstandard Abbreviations

CVD

Cardiovascular Disease

SMC

Smooth Muscle Cell

scRNA-seq

single-cell RNA-seq

EC

Endothelial Cell

CITE-seq

Cellular Indexing of Transcriptomes and Epitopes by Sequencing

GWAS

Genome-Wide Association Studies

TIA

Transient Ischemic Attack

UMAP

Uniform Manifold Approximation and Projection