Meta-Analysis of Leukocyte Diversity in Atherosclerotic Mouse Aortas

Abstract

The diverse leukocyte infiltrate in atherosclerotic mouse aortas was recently analyzed in 9 single cell RNA-Seq (scRNA-Seq) and 2 mass cytometry (CyTOF) studies. In a comprehensive meta-analysis, we confirm four known macrophage subsets: resident, inflammatory, IFNIC and Trem2 foamy macrophages and identify a new macrophage subset resembling cavity macrophages. We also find that monocytes, neutrophils, dendritic cells, natural killer cells, innate lymphoid cells-2 (ILC2) and CD8 T cells form prominent and separate immune cell populations in atherosclerotic aortas. Many CD4 T cells express interleukin (IL)-17 and the chemokine receptor CXCR6. A small number of Tregs and Th1 cells is also identified. Immature and naïve T cells are present in both healthy and atherosclerotic aortas. Our meta-analysis overcomes limitations of individual studies that, because of their experimental approach, over- or under-represent certain cell populations. CyTOF studies demonstrate that cell surface phenotype provides valuable information beyond the cell transcriptomes. The present analysis helps resolve some long-standing controversies in the field. First, Trem2⁺ foamy macrophages are not pro-inflammatory, but interferon-inducible cell (IFNIC) and inflammatory macrophages are. Second, about half of all foam cells are smooth muscle cell-derived, retaining smooth muscle cell transcripts rather than transdifferentiating to macrophages. Third, Pf4, which had been considered specific for platelets and megakaryocytes, is also prominently expressed in the main population of resident vascular macrophages. Fourth, a new type of resident macrophage shares transcripts with cavity macrophages. Finally, the discovery of a prominent ILC2 cluster links the scRNA-Seq work to recent flow cytometry data suggesting a strong atheroprotective role of ILC2 cells. This resolves apparent discrepancies regarding the role of Th2 cells in atherosclerosis based on studies that pre-dated the discovery of ILC2 cells.

1. Introduction

Atherosclerosis is a disease of large and mid-sized arteries with devastating consequences. The disease process is initiated by low density lipoprotein (LDL) accumulating in the subendothelial space of arteries. Rapidly, resident vascular macrophages and lymphocytes expand in number and differentiate in phenotype. Early on, neutrophils and monocytes are recruited, along with all known cell types of the adaptive immune system. Mature atherosclerotic lesions therefore contain a wide variety of immune cells.

In early studies, the immune cell infiltrate was characterized by immunohistochemistry¹. Later, methods were developed to generate single cell suspensions from mouse aorta² and human atherosclerotic lesions³. Recently, high dimensional methods of analysis became available that generated a plethora of data allowing fine-grained and comprehensive characterization of the immune cell content of atherosclerotic lesions. Mass cytometry^{4, 5} allows assessment of up to 42 cell surface and intracellular markers. Single cell RNA-sequencing (scRNA-Seq) yields expression matrices of thousands of genes per cell, depending on the single cell sequencing method and platform⁶. The technical aspects of these high dimensional methods in vascular biology, including quality controls and validation, have recently been reviewed⁶.

This resource article is focused on leukocyte diversity in atherosclerosis. We assembled a team of specialists in the field to provide comprehensive and, where possible, authoritative information of the immune cell phenotypes as defined by cell surface phenotype, intracellular proteins and gene expression. We will only discuss mouse data, because there are too few studies of leukocytes in human atherosclerosis^{7, 8}. Since genetic models of atherosclerosis (Apoe^−/−, Ldr^−/−) were generated in 1992^9–11, the mouse has become the de facto workhorse of atherosclerosis research. Later, more sophisticated mouse models of atherosclerosis were generated^{12, 13}, but these are not as widely used.

One goal of this review is to assemble lists of surface markers and genes (“gene signature”) that can be used to delineate the various immune cell types. Of note, we are not concerned with the ontogeny of the immune cells in atherosclerosis. It is known that tissue resident vascular macrophages in mice are not monocyte-derived, are largely seeded before birth and can proliferate locally¹⁴. Other macrophages in atherosclerosis are derived from infiltrating monocytes¹⁵. We are also not concerned with cell lineages. In immunology, the relationship between cell types and their precursors, and whether cell types can transdifferentiate into other cell types, is often quite controversial. As a case in point, there is excellent evidence for¹⁶ and against¹⁷ conversion of regulatory T cells to other T cell types. Thus, we remain agnostic with respect to ontogeny and lineage. Rather, we are focusing on describing the cell types and states in atherosclerotic lesions (“what is”). The goal is to create a resource that will promote systematic and mechanistic investigations. Excellent reviews exist on immune cells in the heart^{18, 19}. Neither heart nor veins or other blood vessels will be considered here. Another goal is to promote clarity in nomenclature, which will enable the field to move forward more rapidly. A distant goal is to translate findings in experimental atherosclerosis into prevention and therapy strategies for people with cardiovascular disease.

This review was spawned at a meeting that Andreas Zirlik organized in Graz, Austria in 2019. This meeting was focused on single cell methods in atherosclerosis research. Several of the authors were present at this meeting. Klaus Ley approached the editors of Circulation Research with the idea of putting together a broad-based review. Bioinformatics re-analysis of existing data resulted in this review becoming a resource paper. Computational biologists sifted through the data to identify transcriptomes, gene signatures and surface markers across labs, approaches and platforms. Most authors who published on mouse atherosclerosis using high dimensional single cell methods accepted the invitation to contribute.

This review is focused on CD4 and CD8 αβ T cells, γδ T cells, NK and NKT cells, B cells, neutrophils, macrophages, monocytes and dendritic cells. In addition, eosinophils, mast cells and innate lymphocyte-like cells (ILCs) are found in atherosclerotic lesions. Certainly, other cell types not mentioned here will emerge as important players in atherosclerosis. In fact, single cell technologies have a unique potential to discover new cell types.

2. Introduction of Immune Cell Types

T cells are cells of the adaptive immune system that develop in the thymus, where they find niche conditions conducive to rearrangement, under the control of the recombinase activating genes (RAG)-1 and 2, of their T cell receptor (TCR) γδ or αβ subunits, expression of their unique TCRs, positive and negative selection, and maturation²⁰. CD4 T cells recognize antigenic peptides bound to major histocompatibility complex (MHC)-II and are either regulatory or helper T cells. CD8 T cells recognize peptide antigens bound to MHC-I and become cytotoxic T lymphocytes (CTLs). Some αβ T cells recognize glycolipids bound to the MHC-like molecule CD1d and become NKT cells²¹.

CD4 T cells recognizing self antigens are mostly regulatory T cells. Regulatory CD4⁺ T cells (T_reg) represent immunosuppressive T cells that are defined by the transcription factor Foxp3 and IL-2 receptor α chain (CD25) expression^{22, 23}. Natural T_regs (nTregs) are thymus-derived, whereas peripherally induced T_regs (iTregs) are generated in response to antigen recognition in the presence of transforming growth factor beta (TGF-β)²⁴. Tregs dampen inflammation by limiting the proliferation and function of effector CD4 and CD8 T cells.

Co-stimulation and antigen presentation by antigen-presenting cells induces T cell activation and differentiation into functionally and phenotypically distinct T-helper (T_H) types that contribute to atherosclerosis in a subset-dependent manner²⁵. Follicular helper (TFH) cells express ICOS and the chemokine receptor CXCR5. They stay in lymphoid organs and have important helper functions for B cell maturation.

T-helper1 (T_H1) CD4⁺ T cells express the lineage-defining transcription factor T-bet (gene name Tbx21) and the pro-atherogenic cytokine IFN-γ^{26, 27}. The deletion of either Tbx21 or Ifng markedly reduce atherosclerosis^{28, 29}. T_H2 cells express the lineage-defining transcription factor GATA3 and secrete interleukins (IL)-4, IL-5, IL-10, and IL-13³⁰. Dependent on the context and model analyzed, T_H2 cells have been reported to contribute to atheroprogression or -protection^{31, 32}. T_H17 cells are characterized by the transcription factor RORγt and secrete the cytokines IL-17A, IL-17F, and IL-21. The contribution of T_H17 to atherosclerosis is controversial, as the hallmark cytokine IL-17 can hamper or enhance atherosclerosis. This is likely related to different functions of different types of IL-17-expressing cells: Th17-Tregs have regulatory function (atheroprotective), whereas true Th17 cells and γδ T cells expressing IL-17 are pro-atherogenic. T_H9 T cells express the cytokine IL-9 and the transcription factor PU.1 and accelerate atherosclerosis³³.

T_regs prevent the progression of atherosclerosis by multiple mechanisms, including the secretion of IL-10 and TGF-β^{34, 35}. At later stages of atherosclerosis, T_regs vanish³⁶ and may obtain a T_H1-like phenotype that includes expression of IFN-γ^{37, 38}. While it was proposed that differentiation of T cells to distinct effector T_H cell lineages is final and irreversible, accumulating evidence suggests that T_H cells, and in particular inducible T_regs and T_H17 cells, are flexible in nature and can be reprogrammed to unique mixed phenotypes or can re-differentiate to other subset of T_H cells. T_H1-like T_regs expressing IFN-γ and Foxp3 are found in tissues and the peripheral blood of patients with type 1 diabetes, multiple sclerosis, and arthritis³⁹. IL-17⁺ T_regs are detected in rheumatoid arthritis, atopic asthma, and colorectal cancer, whereas GATA3⁺Foxp3⁺ cells are found in patients with food allergies. These findings suggest that the microenvironment promotes reprogramming of T_H cells under pathological conditions.

CD8 T cells express inflammatory cytokines, cytotoxins and death-inducing proteins including IFNγ, TNF, granzymes A and B, perforin, FasL and TRAIL. Although CD8 T cells are generally considered pro-inflammatory and cytotoxic, CD25-expressing CD8 regulatory T cells, present in humans and mice, may modulate immune responses⁴². Once they are activated by antigen-presenting cells in an MHC-I-dependent manner, CD8 T cells can potently induce target cell death via either cytotoxins, cytokines or death-inducing proteins⁴³. In healthy arteries, CD8 T cells are found in the adventitia. CD8 T cell abundance increases in atherosclerosis, where they populate the plaque shoulder and the area around necrotic cores⁴⁴. In human rupture-prone lesions, cytotoxic CD8 T cells constitute up to 50% of CD45⁺ leukocytes⁴⁵. CD8 T cells progressively accumulate but abruptly decline after plaque rupture^{46, 47}. CD8 T cells promote atherosclerosis by increasing vascular inflammation and apoptotic cell numbers in lesions^{48, 49}. CD8 T cells can also promote atherosclerosis by controlling monopoiesis and circulating monocyte levels^{44, 50}. Some CD8 T cells recognize epitopes in ApoB, but most CD8 T cells in arteries have unknown specificities.

NKT cells are a minor T cell subset. They are considered a bridge between innate and adaptive immunity. NKT cells do not have immunological memory⁵¹. Besides innate receptor-dependent activation, NKT cells are activated by CD1d-bound glycolipid antigens⁵¹. Upon activation, they rapidly release large amounts of cytokines and cytotoxins⁵², which can aggravate atherosclerosis^53–55. Among different subsets, NKT cell research has been focused on invariant NKT (iNKT) cells, also known as type I NKT cells expressing an invariant TCR α chain (Vα14-Jα18 in mouse). Phenotypically, NKT cells can be identified by using combinations of the NK cell marker, killer cell lectin-like receptor subfamily B member 1, also known as KLRB1, NK1.1 or CD161 and the T cell-defining marker CD3. Human and mouse work using lineage tracking or double immunostaining have identified the presence of NKT cells in atherosclerotic lesions^{54, 56, 57}. Recent single-cell immunophenotyping methods have identified NKT cells in human and mouse atherosclerotic lesions⁷.

γδ T cells were first reported in human atherosclerotic lesions in 1993⁵⁸, yet their exact role in atherosclerosis remains unclear. γδ T cells are a minor subset (1–10% of T cells)⁵⁹. They express TCRs composed of γ- and δ-chains that are activated independent of MHC. Upon activation through innate immune receptors γδ T cells can produce large amounts of inflammatory cytokines, chemokines and cytotoxins⁶⁰. CD27⁺ γδ T cells secrete IFN-γ and CD27⁻ γδ T cells IL-17 ^{61, 62}. Another subset of γδ T cells expressing the Vγ9Vδ2 TCR are reported to be cytotoxic⁶³.

NK cells are cytotoxic cells that kill cells that lack expression of MHC-I. NK cells have no direct effect on atherosclerosis or lesion phenotype⁶⁴. However, NK cells may contribute to atherosclerotic lesion formation when activated, such as under conditions of chronic viral infection^{64, 65}.

B cells are adaptive immune cells, but unlike T cells, do not require the thymus for development. Like the TCR, the B cell receptor (BCR) must undergo V(D)J rearrangement under the Rag genes. These “germline-encoded” BCRs have low affinity to antigen. Immature B cells display IgM and IgD with the same sequence as their BCR. In the presence of follicular helper T cells (TFH), B cells can switch their isotype to various IgG isotypes, IgE or IgA. In germinal centers, their BCR undergoes many cycles of affinity maturation. This is under the control of the AID enzyme and results in high affinity antibodies. B cells express CD19 and CD20.

Many IgM-producing B cells are B1 cells, further divided into B1a and B1b based on the expression of CD5. B2 cells are much more abundant than B1 cells in spleen, bone marrow and blood and make IgG, IgE and IgA. Both B1 and B2 cells express MHC-II and can present antigenic peptides to CD4 T cells. One specialized type of B1 cells are Innate Response Activator (IRA) B cells, which are strongly pro-inflammatory by expressing Csf2 (GM-CSF)⁶⁶. IRA B cells aggravate atherosclerosis by stimulating T_H-1 adaptive immunity⁶⁷.

In healthy arteries, B cells are found in the adventitia² and perivascular adipose tissue (PVAT)⁶⁸. B cells in the artery wall and adventitia of normal arteries probably do not secrete antibodies, but can traffic to other sites and become antibody-secreting plasma cells. Antibody production in the artery appears localized to PVAT⁶⁸ and arterial tertiary lymphoid organs (ATLOs) adjacent to advanced plaques⁶⁹.

Neutrophils are short-lived cells of the innate immune system. They are the main defenders against bacterial and fungal infections. A few neutrophils are present in normal arteries. Under conditions of atherosclerosis, some neutrophils are found on the endothelial surface and others in the plaque. Although their numbers are small, neutrophils have important roles in atherosclerosis and thrombosis⁷⁰. Neutrophils are highly granular cells with a distinct lobed nuclear morphology. They are typically produced in the bone marrow in a process called granulopoiesis. Under conditions of metabolic stress such has hypercholesterolemia, however, neutrophil production also occurs in extramedullary tissues including the spleen, thus contributing to heightened circulating neutrophil counts⁷¹. Neutrophils can be identified in mouse blood as well as in single cell suspensions of tissues including aortic tissue as CD45⁺CD11b⁺Ly6G⁺CD115⁷².

Monocytes are innate immune cells that express CD115 (Csf1r) and CD11b. In mice, they come in two main flavors, classical and non-classical. Classical monocytes express Ly-6C and the chemokine receptor CCR2⁷³. Non-classical monocytes highly express the transcription factor Nr4a1⁷⁴ and the chemokine receptor CX3CR1⁷⁵. In healthy arteries, monocytes are rare. However, in atherosclerosis-prone regions (such as the lesser curvature of the ascending aortic arch), classical monocytes can be recruited through VCAM1-dependent interactions on vascular endothelium^{76, 77}. Monocytes rapidly infiltrate atherosclerotic lesions. Some differentiate into macrophages, others to monocyte-derived dendritic cells, and some stay monocytes. Some monocytes express high levels of MHC-II and can present antigenic peptides to CD4 T cells. Non-classical monocytes patrol the endothelial surface and promote endothelial integrity⁷⁸. Patrolling is highly intensified under conditions of atherosclerosis^{79, 80}. Nonclassical monocytes patrol in response to oxidized LDL and ingest greater amounts of oxidized LDL than classical monocytes, processes that are dependent upon the scavenger receptor CD36⁸⁰.

Dendritic cells (DCs) are found in secondary lymphoid organs, where they present antigens to naive T cells and start a specific immune response. DCs can be divided into conventional DCs (cDCs, CD11c⁺) and plasmacytoid DCs (pDCs, CD123⁺). Among the conventional DCs, lymphoid organ resident CD8α⁺ and non-lymphoid organ CD103⁺ type 1 cDC1 expressing Xcr1 and Clec9a (DNGR1) can be discriminated from cDC2⁸¹. Plasmacytoid DCs produce type I interferons (IFNα and β)⁸².

Macrophages are the most common and probably the most important cell type in atherosclerosis. They express F4/80 and CD64, which is an activating Fcγ receptor. In healthy arteries, macrophages are in the adventitia, where they contribute to the physiology and diameter of the vessel wall via cross-talk with smooth muscle cells⁸³. These vascular macrophages are largely derived from embryonal precursor cells and self-renew in situ¹⁴. A small subset of macrophages is located below the endothelium in certain areas of the arterial circulation such as the aortic arch. Atherosclerosis is characterized by an accumulation of a very large number of macrophages, derived from infiltrating monocytes and proliferating vascular macrophages¹⁵. Macrophages function to clear apoptotic cells (efferocytosis), to phagocytose debris, to produce inflammatory and inflammation-resolving cytokines and lipid meditators, and to present antigens^{84, 85}.

3. Meta-Analysis of scRNA-Seq Data

This review and meta-analysis is based on scRNA-Seq and CyTOF data from mouse aortas (figure 1). scRNA-Seq data from 9 data sets (Online Data Set) were analyzed using the latest bioinformatics integration method Harmony⁸⁶. Harmony allows to simultaneously account for multiple experimental and biological batches across data from different labs. After quality control, a total of 15,288 cells were projected into a shared embedding space in which cells were assigned and corrected for dataset-specificity using fuzzy clustering. Then, all cells were visualized using the non-linear high-dimensionality deduction method, UMAP (Uniform Manifold Approximation and Projection)⁸⁷. Louvain-based clustering was then used with the Seurat method⁸⁸ that yielded 17 clusters (figure 2; differentially expressed genes for each of the clusters in Online table I).

Figure 1. Overview.

This analysis is based on published studies where atherosclerosis was induced in mouse models using genetic knockouts Ldlr^{−/− 91, 96}, Apoe^{−/− 5}, diet (chow, western diet, high fat diet, refs) or AAV-PCSK9 induced lipoprotein changes⁹², resulting in atherosclerotic aortas. In some studies, lineage tracking (Cx3cr1-Cre)⁹² or chemical labeling (Bodipy)⁹¹ labeled specific cell types. All studies used enzymatic digestion with the attendant problems of cell death and loss of surface markers⁶. Single cells were phenotyped by RNA-Seq^{91, 92, 96}, CyTOF¹⁰⁶ or both⁵. Dimensionality reduction and clustering identified cell types and gene signatures. Matching CyTOF with scRNA-Seq data is challenging. Based on gene signatures, genetic labeling was used to visualize, image and sort some cell types to gain functional insights and deep transcriptomes¹²⁹.

Figure 2. Integration of scRNA-Seq data from 9 atherosclerosis studies on mouse aortas.

scRNA-Seq data was retrieved from NCBI GEO for Harmony integration and visualized using UMAP⁸⁶. 17 clusters were identified by Louvain clustering. One cluster (lilac) was dominated by non-leukocyte genes including Acta2 and one cluster (dark yellow) was dominated by proliferation genes including Top2a and Tuba1b, leaving 15 bone fide leukocyte clusters: 2 B cell clusters (red and pink), one CD8 T cell cluster (light green), 4 macrophage clusters (inflammatory, blue, IFNIC, light purple, resident, purple and Trem2, light blue), 2 CD4 T cell clusters (Th17, brown and Th2, teal), CD4⁺CD8⁺ cells (turquoise), 2 mixed monocyte/macrophage/DC clusters (orange), Xcr1⁺ DCs (hot pink), neutrophils (light pink), NK cells (dark green). Total n= 15,288 cells

Overall, macrophages are the most abundant cell type in atherosclerotic mouse aortas. The Louvain clustering algorithm implemented in Harmony clearly identified four subsets. We also identified clusters of two monocyte/DC subsets and one cluster of Xcr1⁺ cDC1 cells (Xcr1, Irf8). Xcr1 is a chemokine receptor characteristic of cDC1. One of the monocyte/DC clusters has a gene signature suggesting it may contain classical monocytes (Ccr2), the other a mixture of monocyte-derived DCs (Cd209a), and yet unidentified cells. The B cell compartment shows 2 clusters. The B1-like cluster is enriched in B1 cell genes and the B2-like cluster shares gene expression with germinal center and marginal zone B cells. The three T cell clusters include the expected CD8 T cells, CD4⁺ Th17-like cells and an unexpected subset of CD4⁺CD8⁺ T cells. The fourth cluster, although located right next to the T cells, mainly contains ILC2 cells based on marker genes like Klrg1, Il1rl1 and Areg and the absence of Cd3 expression. It is striking that Th1 cells, shown to be abundant in atherosclerotic lesions in mice⁸⁹ and humans⁹⁰ by immunostaining, do not appear to form an identifiable cluster. However, Tbx21⁺ T cells are found in the T cell and NK cell clusters (Online figure I). Similarly, Tregs are not easily identified in scRNA-Seq experiments, because low-expressed transcription factors like FoxP3 are often missed by scRNA-Seq. Also, FoxP3 expression can be pulsatile. Although they do not form a separate cluster, we find a few FoxP3⁺ CD4 T cells on the edge of the Th17-like cluster (Online figure I). Th1 Tregs³⁷, a type of T cells with features of Tregs and Th1 cells, were not found in this analysis.

The cellular composition of mouse aortic leukocytes varied widely among the 9 data sets (figure 3). This is likely due to differences in the experimental protocols used (Online Data Set) and to biological differences between the different atherosclerosis models. Some datasets show a bias towards or directly focus on macrophages^{91, 92}, while others are more biased towards lymphoid cells, especially Winkels et al.⁵. A very interesting finding is that the foam cells, identified by Bodipy staining in Kim⁹¹, are exclusively Trem2 macrophages⁹³. Thus, based on this meta-analysis, we propose to call this cell type foamy Trem2 macrophages, a classic cell type in atherosclerosis⁸⁴. About half of the foam cells as defined by Bodipy staining are non-leukocytes, probably smooth muscle cells, as expected^{94, 95}.

Figure 3. Different abundance of aortic leukocyte subsets in 9 scRNA-Seq mouse atherosclerosis studies.

The UMAP from figure 2 was separated into the 9 different studies. Studies identified by first author^{5, 91, 92, 96} and abbreviated conditions (see online data set for full conditions). Cell subsets as in figure 2.

CD8 T cells are found in all data sets except in the sorted foam cells⁹¹ and in the regression study that focuses on flow-sorted CX3CR1⁺ CD11b⁺ myeloid cells⁹². Th17-like T cells are probably a mixture of Th17 and γδ T cells. The ILC2 cluster was only found in the Cochain⁹⁶ and Winkels⁵ datasets. Interestingly, these cell types are already present in the healthy aorta, prior to onset of atherosclerosis, whereas the CD8 T cells are rare in healthy aortas⁹⁶. CD4⁺CD8⁺ T cells are present in healthy and atherosclerotic aortas.

Among the B cells, B2-like cells are found in healthy aortas. They become less abundant after 11 or 20 weeks of high fat diet⁹⁶. B1-like cells are only found in the Winkels dataset⁵, which contains data from Apoe^−/− mice. When comparing the different models of atherosclerosis, it seems that the Apoe^−/− model shows preferential accumulation of B2 cells and the appearance of a small cluster of B1 cells.

As expected¹⁴, the healthy aorta contains only resident macrophages. Also expected, macrophage frequencies increase with the duration of high fat diet in Ldlr^−/− mice⁹⁶ and with western diet in Apoe^−/− mice⁵. Interestingly, macrophage diversity increases with atherosclerosis progression. Four clear subsets are seen in all datasets: Trem2⁺ foamy macrophages, resident macrophages, inflammatory macrophages and IFNIC macrophages.

The gene expression driving the various cell types is presented as a dot plot (figure 4). Inflammatory macrophages highly express the inflammatory chemokines Ccl4, Cxcl2, Ccl3 and Ccl2. They also express Cd14 at a higher level than the other macrophage subsets. Many express Il1b and Tnf and a few express Cxcl1. The Trem2 foamy macrophages have a gene expression profile that partially overlaps with the inflammatory macrophages. They highly express Lgals3, the gene encoding galectin-3, a known biomarker for atherosclerosis⁹⁷, the tetraspanin Cd9 and Ctsd, the gene encoding cathepsin D. Some express the fatty acid binding protein Fabp5. Resident macrophages are very different from the other subsets. They express the chemokine Ccl8, the chemokine-like molecule CXCL4 (Pf4), which was previously thought to be platelet-specific⁹⁸, coagulation factor XIII (F13a1) and Wfdc17. Some express Lyve1, a marker associated with tissue resident perivascular macrophages in the aorta^{14, 83, 99}. The IFNIC macrophages have a type I interferon signature, prominently expressing Isg15, Irf7, Ifit 1 and Ifit 3. They express the chemokine Ccl12, which encodes an important ligand for CCR2. A few neutrophils are found in all data sets except the sorted foam cells, even in healthy aorta. This is somewhat surprising and unlikely to be due to blood contamination, as Cochain⁹⁶ used intravascular CD45 staining to exclude blood contamination. Thus, this finding suggests that a few neutrophils may be present in the mouse aorta even under control conditions. Neutrophils share many genes with monocytes and macrophages, but characteristically express Cd7, S100a9 and S100a8.

Figure 4. Top gene signatures for 15 aortic leukocyte subsets.

The top 10 up-regulated genes in each subset compared to all other subsets are shown by Dot plot. Expression level indicated by saturation of blue (dark blue is highest expression, log2 scale where 0 is global average). Dot diameter represents percentage of cells in cluster expressing corresponding genes (largest circle is 100%).

The B cell clusters express typical B cell genes like CD79a and b, Ccr7 and Mzb1. The expression profile between B1-like cells express Tppp3, S100a6 and Cd9, whereas B2-like cells express Fcer2a and Cd23.

Two mixed monocyte/DC subsets are well represented in all datasets with atherosclerosis. As mentioned above, few monocytes are present in healthy aortas, and monocytes are not found among foam cells. The cluster denominated mixed monocytes/ DCs/Ccr2 contains cells expressing genes found in classical monocytes, i.e. Lyz1, and the chemokine receptor Ccr2, but also Fn1, encoding fibronectin, Retnla encoding Relm-α, and the chemokine Ccl6. The other monocyte cluster (mixed monocytes/DC/Cd209a) expresses aspartic peptidase Napsa1 and Cd209a, encoding DC-SIGN, which may suggests the presence of monocyte-derived DCs and cDC2. This cluster also shows expression of interferon genes Ifi30 and Ifitm1. The Xcr1⁺ DCs highly express the transcription factor Irf8, Cst3 encoding cystatin 3, the N-acylethanolamine acid amidase Naaa, the N-acylgalactosaminidase Naga and the phospholipase Plbd1.

NK cells express Nkgh7, Klre1, Klra7, Klrk1 and granzyme B (Gzmb). Interestingly, they also express the chemokine Xcl1, the only known ligand for dendritic cell receptor Xcr1, and the chemokine Ccl5, also known as RANTES.

ILC2 cells are negative for CD3-encoding transcripts (Cd3d, Cd3g) and express genes associated with ILC2 (Rora, Gata3, Areg, Il1rl1 encoding ST2). The Th17-like T cells express CD3-encoding genes, but are negative for Cd4, Cd8a or Cd8b1 transcripts. These cells are enriched for Il17a expression. This cluster likely contains γδT cells also expressing lymphotoxin B (Ltb), Il7r and Icos.

One cluster had an overwhelming proliferation signature. When regressing out cell cycle genes, this cell cluster primarily expressed myeloid markers such as Cd11b, Adgre1, Cd68, Trem2, and Cxcr1. Thus, they are likely proliferating macrophages. Lesional macrophages that have differentiated from monocytes proliferate to locally expand the macrophage population within plaques. Macrophage turnover becomes increasingly dependent on local proliferation as atherosclerosis progresses¹⁵. In a normal mouse aorta, resident adventitial macrophages comprise the vast majority of arterial macrophages. This population expresses Lyve1. Adventitial macrophages arise from CX3CR1⁺ yolk sac progenitors and fetal liver monocytes. Local proliferation maintains stable cell numbers throughout adulthood in the steady state. Upon exposure to a potent systemic inflammatory stimulus, the number of Lyve1⁺ adventitial macrophages falls dramatically and gradual recovery is dependent on a local proliferative response. Thus, a tightly regulated proliferative response is critical for maintaining resident arterial macrophages within the adventitial niche both during homeostasis and post-inflammation¹⁴. Whether resident aortic macrophages proliferate in atherosclerosis and contribute to this proliferating macrophage cluster is unclear.

One cluster contained Acta2⁺ cells that were not leukocytes, but smooth muscle cell-derived foam cells. The remaining 15 clusters distributed into 4 macrophage clusters, 2 monocyte/DC clusters, one DC cluster, one neutrophil cluster, one NK cell cluster, 2 B cell clusters, 3 T cell clusters and one innate lymphoid cell (ILC) cluster.

To improve clustering resolution, we selected myeloid cells (except neutrophils), the T/NK cells/ILCs and the B cells each separately for reclustering (figure 5). This reclustering confirmed the four macrophage subsets, resident, inflammatory, Trem2-foamy and IFNIC, and revealed a new fifth subset of macrophages. Leading differentially expressed genes (DEGs) of cells in this subset included Fn1, Clec4b1, Sept11, and Ear2. A full representation of the representative genes (with less than 30% expressed in the rest) is shown as a dot plot in Online figure II. Comparing the top DEGs of the new macrophage subset with the ImmGen database placed it most closely to peritoneal macrophages. However, these cells could also come from the pleural or pericardial cavities. Since the exact location of origin is unknown, we labeled them “cavity macrophages”. The monocytes and DCs were also better identified, revealing clusters of mature DCs, which express CCR7 and fascin-1, known markers of mature DCs^{100, 101}. These cells were previously identified in ⁹⁶. Reclustering confirmed cDC1 cells and putative pDCs. A rather large cluster contained cells with transcriptomes consistent with monocyte-derived DCs¹⁰² and cDC2s. Leading genes included CD209a, Ifitm1 and Klrd1 (Online figure II).

Figure 5. Reclustering of myeloid cells and T cells.

We selected myeloid cells (Macrophages, Mixed monocyte/macrophage/DC clusters except neutrophils, left panel), and T cells (T cell, IL-17⁺, NK and ILC cell clusters, right panel) from the UMAP from figure 2 for reclustering. The following myeloid clusters were retrieved: 5 macrophage clusters (inflammatory, dark green, IFNIC, light green, resident, dark pink, Trem2, light pink, and cavity, red), monocytes (blue), MoDC/cDC2 (yellow), and 3 DC clusters (Mature DC, turquoise, cDC1, yellow green, and pDC, purple). T cells clustered into naïve T cells (turquoise), CD4⁺Cd8⁺ (light pink), IL17⁺Cxcr6⁺ (light green), Treg (dark pink), CD8 (yellow), NK cells (purple) and ILC2s (dark green).

We formally compared the prevalence of these 10 myeloid cell types among the 9 data sets (Online figure III), projecting each dataset onto the same UMAP of myeloid cells and constructing stacked bar graphs (Online figure IV). The myeloid cell composition of the Cochain et al. 11 weeks HFD and the Kim et al. Ldlr^−/− 12 weeks HFD datasets are almost identical. Thus, we have confidence in the myeloid cell composition of aortas harvested from HFD-fed Ldlr^−/− mice, because the two studies confirm each other. The data from Winkels et al. 12 weeks HFD-fed Apoe^−/− is also similar. However, the Winkels et al. studies are systematically missing foam cells, which could be caused by the harsher digestion procedure used (see Online Data Set). The Lin et al. progression and regression show no difference in myeloid cell composition. The data also show that foam cell enrichment as used in the Kim et al. study (Apoe^−/− intimal foam cells) works well. Effectively, more than 80% of the foam cell-enriched preparation are actually Trem2⁺ foamy macrophages based on this meta-analysis. The other myeloid cell types are removed in about equal proportions. Finally, as expected, myeloid cells in the healthy mouse aorta are dominated by resident macrophages with almost no foam cells in healthy aortas (Cochain et al. Ldlr^−/− mice chow). Although Winkels et al. did not investigate healthy aortas, the myeloid composition of their chow-fed Apoe^−/− mouse aorta is similar, with the exception of a higher proportion of monocytes and monocyte-derived DCs.

Reclustering the T cells, NK cells, and ILCs revealed a new cluster of regulatory T cells, which express GITR (Tnfrsf18), Cd134, also known as Ox40L or Tnfrsf4, Nt5e (CD73), Ctla4 and Nrp1 (figure 5, Online figure V). In some of these cells, the hallmark transcription factor Foxp3 is also detectable (Online figure VI). Regulatory T cells are known to be important in limiting atherosclerosis³⁴. The CD8 T cluster cells express Cd8b, Cd8a and the transcription factor Eomes. Part of the CD4⁺CD8⁺ T cells expressed the transcription factor Tox (Online figure VI). Tox is associated with exhaustion¹⁰³, but is also involved in beta selection¹⁰⁴. All of the Cd4⁺Cd8⁺ cells exclusively expressed Sox4 (Online figure VI), a transcription factor known in stem cells and early progenitors in different tissues¹⁰⁵. Thus, this finding suggests that CD4⁺CD8⁺ lymphocytes in the aorta may be immature. The nature of the ILC2 cells was confirmed by overlaying expression of Il1rl1, which encodes the IL-32 receptor ST-2 (Online figure VI). The IL-17-expressing cells were also confirmed; many of them express the chemokine receptor CXCR6. Reclustering the B cells did not reveal any new subset with statistically significantly DEGs, and confirmed a small subset of B1 cells and a larger subset of B2 cells (data not shown).

Lymphocyte populations are similar in HFD-fed Ldlr^−/− (Cochain et al. 11 weeks HFD and Kim et al. Ldlr^−/− 12 weeks HFD) and Apoe^−/− mice (Winkels et al. 12 weeks HFD) (Online figure VII). An expansion of CD8⁺ T cells is notable in all datasets in atherosclerosis. In the healthy aorta, lymphocytes dominate (figure 3), because there are so few myeloid cells². This meta-analysis suggests that proportions of potentially immature CD4⁺CD8⁺ T cells are increased and Tregs are absent in healthy aortas (Cochain et al. Ldlr^−/− mice chow).

Meta-Analysis of CyTOF Data

Two of the studies of mouse aortic leukocytes^{5, 106} used CyTOF to characterize the immune cell landscape of Apoe^−/− mouse aortas. Each study used 35 markers, 33 of which were conjugated to monoclonal antibodies recognizing cell surface and intracellular molecules. Although the number of markers (33) is much smaller than in scRNA-Seq (~2,000), these markers are very informative, because they were picked based on 30 years of experience with flow cytometry and 15 years of experience with flow cytometry of mouse aortas².

Cole¹⁰⁶ used 19 myeloid markers, 5 T cell markers, 2 B cell markers and 8 others. They identified 4 macrophage subsets (table 1). All macrophages expressed CD11b, CD64 and CD68¹⁰⁷. High levels of CD64 distinguish macrophages from other mononuclear phagocytes including monocytes Markers that differed among subsets were CD11c⁺CD44⁺ for Mac 1, CD11c^low CD44⁺ CCR2+ CD206^med CD169⁻ for Mac 2, CD11c⁻ CD44⁻ CD206⁺ CD169⁺ CD209b⁻ for Mac 3 and CD11c⁻ CD44⁻ CD206⁺ CD169⁺ CD209b⁺ for Mac 4.

Table 1. Identification of immune cells from atherosclerotic mouse aortas by CyTOF.

Data compiled from Winkels et al. and Cole et al.

Author	Winkels et al.	Cole et al.
Macrophages (CD11b+)	• Cl. 14 (CD11cmed F4/80low CD64low CD43high) • Cl. 3 (F4/80med CD64high CD103med FcεRIhigh) • Cl. 9 (F4/80high CD64med CD103low CD160high CD138high)	• Mac 1 (CD68+ CD64+ F4/80− CD11c+ CD44+ CCR2low) • Mac 2 (CD68+ CD64+ F4/80+ CD11clow CD44+ CCR2+ CD206med CD169−) • Mac 3 (CD68+ CD64+ F4/80+ CD11c− CD44low CD206+ CD169+ CD209b−) • Mac 4 (CD68+ CD64+ F4/80+ CD11c− CD44− CD206+ CD169+ CD209b+)
Monocytes (CD11b+ F4/80−)	• Cl. 24 Ly6C+ Monocytes (CD43+ Ly6Chigh MHC-IIlow CD64low CD11c−) • Cl. 6 Ly6C- Monocytes (CD43+ Ly6Clow MHC-IImed CD64low CD11clow)	• Ly6C+ Monocytes (CD68+ Ly6C+ CCR2+ CD44+ CD64med MHC-IIlow CD11c−) • Ly6C- Monocytes (CD68+ Ly6C− CCR2− CD44+ CD64low MHC-II− CD11c+)
Granulocytes (CD11b+)	• Cl.11 Neutrophils (Ly6C+ Ly6G+) • Cl.28 Eosinophils (Ly6C− SiglecFhigh F4/80high)	• Neutrophils (Ly6C+ Ly6G/C+) • Eosinophils (Ly6C− SiglecF+ F4/80+)
DCs (CD11c+)	• Cl. 16 (CD11bmed MHC-IImed CD117low CD4low CD103high) • Cl. 22 (CD11bmed MHC-IIhigh CD117high CD4med CD64low) • Cl. 10 (CD11blow MHC-IImed CD117med CD4neg)	• cDC1 (CD68+ MHC-II+ CD103+ CD11b− CD24+ XCR1high CD172a−) • cDC2 (CD68+ MHC-II+ CD103− CD11b+ CD24low XCR1+ CD172a+) • pDC (CD68+ MHC-II+ SiglecH+ B220med CD4med)
B cells (CD19+)	• Cl. 7 (B220high MHC-IImed CD117med IgMhigh CD43−) • Cl. 4 (B220high MHC-IIhigh CD117high IgMmed CD43med) • Cl. 1 (B220med MHC-IImed CD117− IgMlow CD43−) • Cl. 13 (B220low MHC-IImed CD117low CD11bhigh CD43high)	• B220+ MHC-II+ • B220− MHC-II+
T cells	• Cl. 15 (CD4high CD5med Ly6Chigh FR4high) • Cl. 17 (CD4high CD5high Ly6C− FR4high) • Cl. 25 (CD4low CD5low CD8high FR4−) • Cl. 21 (CD8high CD5med Ly6Chigh FR4high) • Cl. 23 (CD8high CD5med Ly6C− FR4high) • Cl. 20 (TCRβlow IgMlow CD11clow) • Cl. 27 (CD4low CD25high CD43high) • Cl. 19 (TCRγδhigh CD4low CD25high CD43high)	• CD3ε+ TCRβ+ CD4+ Ly6C+ • CD3ε+ TCRβ+ CD4+ Ly6C− • CD3ε+ TCRβ+ CD8+ Ly6C+ • CD3ε+ TCRβ+ CD4+ Ly6C− • CD3ε+ TCRγδ+
ILCs + NK cells	• Cl. 26 (NKp46high NK1.1high CD11bmed CD43high)	• NKp46+ NK1.1+ CD11b+ CD11c+ CD90.2+ • CD3ε− TCRβ− CD90.2+ IL7Rα+

Three of these four subsets were also identified in Winkels et al⁵, who used 7 myeloid, 7 T cell, 7 B cell and 10 other markers. Again, all macrophages expressed CD11b and CD64; CD68 was not in the panel of Winkels et al. CD68 in Cole et al. was expressed by all clusters of monocytes, macrophages and dendritic cells, including pDCs¹⁰⁶. Winkels’ Mac cluster 14 was CD11c^med and CD43⁺ and likely corresponds to Mac 1 in Cole¹⁰⁶. Winkels’ Mac cluster 3 was CD11c⁻ and expressed CD103 and FcεR1. It might correspond to Cole’s Mac 3 or Mac 4, which are CD11c⁻ CD206⁺ CD169⁺ CD209b^±. Winkels’ Mac cluster 9 was also CD11c- but expressed CD160 and CD138. Cole et. al.¹⁰⁶ identified a shift in macrophage populations induced by Western diet, with a reduction in resident-like macrophages (CD11c⁻ CD206⁺ CD169⁺) and an increase in the CD11c⁺ CD44⁺ population.

Both Cole¹⁰⁶ and Winkels⁵ found the two known subsets of mouse monocytes, classical and non-classical. Both subsets express CD11b and CD68. Ly6C⁺ classical monocytes expressed CD68, CCR2, CD44, CD43 but not MHC-II or CD11c (table 1). Ly6C⁻ nonclassical monocytes expressed CD68, CD43, CD44 and some MHC-II and CD11c. It is interesting that the CyTOF studies both identify classical and non-classical monocytes in atherosclerotic mouse aortas, but the scRNA-Seq data (figures 2 and 5) do not yield such clear signatures. This suggests that the monocytes may rapidly change their transcriptome upon entering the arterial wall.

Both Winkels and Cole found cDC1 cells, called cluster 16 in Winkels et al.⁵. They expressed CD68, CD11c, MHC-II, CD103 and XCR1, but little CD11b. cDC2 (cluster 22 in Winkels) also expressed CD68, CD11c and MHC-II, but not CD103. Instead, they expressed CD11b, CD41 and CD172a. Winkels found a third DC subset (cluster 10) that differed from cluster 22 by lower expression of CD11b and absence of CD41, while Cole identified pDCs defined by expression of B220 and SiglecH.

CyTOF identified two granulocyte populations, neutrophils (CD11b⁺Ly6C⁺Ly6G⁺) and eosinophils (CD11b⁺ SiglecF⁺ F4/80⁺). B lymphocytes (CD19⁺) were clustered into 4 clusters in Winkels et al.⁵ based on the expression of B220, CD117, IgM and CD43, and 2 clusters (B220⁺ and B220⁻) in Cole et al.¹⁰⁶.

The two studies showed 5 common T lymphocyte populations including CD4⁺ Ly6C⁻, CD4⁺ Ly6C⁺, CD8⁺ Ly6C⁻, CD8⁺ Ly6C⁺ and TCRβ⁻ TCRγδ⁺ cells^{5, 106}. Additionally, the Winkels panel identified CD4^low CD5^low CD8^high FR4⁻ and TCRβ^low IgM^low CD11c^low T cell clusters⁵. NK cells that express NKp64 and NK1.1 were identified in both studies^{5, 106} and innate lymphoid cells defined as CD3ε⁻ TCRβ⁻ CD90.2⁺ IL7Rα⁺ were detected in Cole et al.’s study¹⁰⁶.

Overall, the CyTOF studies, albeit more limited in scope when compared to the scRNASeq date, offer a solid steer on an extensive panel of antibodies that are useful to disambiguate the leukocyte populations within the atherosclerotic aorta (e.g. macrophages from dendritic cells), leading the way towards more effective and multi-analyte lesion phenotyping. Further integration of CyTOF with scRNASeq-led markers will further refine our capture of other subsets that emerge from the scRNASeq, including Trem2 foamy macrophages.

Macrophage Subsets in Atherosclerosis

As revealed by our meta-analysis, five subsets of macrophages, namely resident-like macrophages, foamy Trem2 macrophages, inflammatory macrophages, IFNIC macrophages and cavity macrophages are distinguishable in atherosclerosis.

Resident-like macrophages expressed genes previously associated with resident aortic macrophages, such as Lyve1 and Mrc1^{14, 83}. These marker genes have been described in yolk sac-derived and embryonic precursor-derived tissue resident macrophages in the adventitia of the aorta and other organs. GSEA analysis of the transcriptome of resident aortic macrophages, obtained by Affimetryx¹⁴, and our integrated scRNA-seq data set indeed showed that vascular macrophages of healthy mouse arteries were most similar to the resident-like macrophages found in healthy and atherosclerotic mouse aortas⁹⁶. Resident-like macrophages are not found in the normal arterial intima, but it is possible that these cells may also be located in the atherosclerotic plaque.

Within the population of resident-like macrophages, Cochain et al. identified a subset of macrophages that showed higher expression of Ccr2 (but lower expression of Lyve1)⁹⁶, suggesting that atherosclerotic aortas may contain monocyte-derived macrophages that adopt a phenotype similar to bone fide resident aortic macrophages, which self-renew independently of CCR2-dependent monocyte recruitment¹⁴.

Among the genes expressed in this cell cluster (figure 4), resident-like macrophages also showed enrichment in Pf4. Pf4 was originally thought to be specific for platelets and megakaryocytes. Pf4 expression was proposed to reflect platelet/macrophage conjugates. However, several studies in reporter mice driven by Pf4-Cre have demonstrated widespread macrophage expression of this gene, notably in peritoneal macrophages¹⁰⁸ and (peri)vascular macrophages¹⁰⁹. Thus, Pf4 can no longer be considered platelet-specific. If Pf4-Cre mice are used to drive a gene modification, the intended gene modification will occur in megakaryocytes, platelets and resident macrophages of the peritoneum and artery wall¹¹⁰.

The foamy Trem2 macrophage population predominantly found in atherosclerotic aortas and almost absent in healthy aortas was identified by Cochain et al⁹⁶. This macrophage population displays high expression of Mmp12, Mmp14, Itgax (CD11c) and markers of lipid loading (Abcg1, Trem2, Fabp4), lysosomal cathepsins (Ctsd, Ctsl) Cd9, and Spp1 (osteopontin). Establishing a flow cytometry-based method of lipid staining of macrophages, Kim et al. detected lipid-laden foam cells in the intima of atherosclerotic aortas⁹¹. Cross-comparison of data from Kim and Cochain et al. indicated that foamy macrophages and Trem2^hi macrophages are identical^{93, 111}. Gene ontology or KEGG pathway analysis assigned foamy Trem2 macrophages putative functions in lipid metabolism, cholesterol efflux and lysosome function^{91, 96}. Foamy Trem2 macrophages had low inflammatory gene expression, in line with prior findings showing that foam cells are characterized by low expression of inflammatory-response genes¹¹². This resolves an important controversy: atherosclerotic arteries show a pro-inflammatory milieu^{113, 114}, yet foam cells are not pro-inflammatory¹¹². With the improved resolution of macrophage subsets provided by scRNA-Seq and CyTOF, it is now clear that the Trem2 macrophages⁹⁶ are the foam cells⁹¹, but do not supply the inflammatory cytokines and chemokines so characteristic of atherosclerotic arteries.

Mononuclear phagocytes sharing transcriptomic features with aortic foamy Trem2 macrophages have been observed across various disease contexts in mice. In particular, this signature is associated with disease-associated microglia in neurodegeneration^{115, 116}, and demyelinating disease¹¹⁷, non-alcoholic hepatosteatosis- and fibrosis-associated macrophages in the liver^{118, 119}, and lipid-associated macrophages in obese adipose tissue¹²⁰. How these cells acquire this specific transcriptomic signature remains to be fully determined. Recent evidence implicates TREM2 in downstream control of cholesterol metabolism in phagocytes^{115–117, 120}, and suggests a TREM2-ApoE pathway¹²¹. Insights from other diseased tissues suggest that macrophage ontogeny may have little impact on acquisition of the foamy Trem2 signature. Both yolk sac-derived microglia¹¹⁵ and monocyte-derived macrophages in adipose tissue¹²⁰ can acquire the foamy Trem2 macrophage state. Importantly, results so far indicate that ApoE deficiency does not affect acquisition of the foamy Trem2 macrophage transcriptional signature in atherosclerosis (Cochain et al.⁹⁶ and the meta-analysis here).

Inflammatory macrophages⁹⁶ primarily evidenced in atherosclerotic aortas by scRNA-seq, show enrichment in inflammatory transcripts, including the chemokines Cxcl1, Cxcl2, Ccl2, Ccl3, Ccl4, and the inflammatory cytokines Il1b, and Tnf. Inflammatory macrophages, also designated as chemokine^high macrophages⁹² or non-foamy macrophages⁹¹, show a strong proinflammatory gene profile. They are not foam cells and express CCR2, suggesting they are likely derived from blood monocytes. How Trem2 macrophages and inflammatory-macrophages are interrelated is currently unclear. An algorithm to predict differentiation pathways¹²² from scRNA-seq data suggested that differentiation pathways into these macrophage subsets may be distinct and independent from each other¹¹⁴

Interferon-inducible macrophages (IFNIC) form a small cluster with a signature characteristic of a type 1 interferon response. IFNIC are enriched for numerous interferon-inducible genes including Ifit3, Irf7 and Isg15. IFNIC were found in the two studies with large numbers of macrophages^{91, 92}, but not by Cochain et al.⁹⁶ or Winkels et al.⁵ Identification of the IFNIC cluster was likely favored by the large number of macrophages sequenced in^{91, 92}. After the meta-analysis (figure 3), small numbers of IFNIC macrophages became visible in the Winkels⁵ and Cochain⁹⁶ studies. IFNIC macrophages are reminiscent of macrophages described in the ischemic heart by King et al.¹²³. IFNIC may originate from remote type I IFN-mediated priming of monocytes and their progenitors in the bone marrow after tissue injury¹²⁴. However, macrophages with a similar gene expression signature have been found at steady state in the heart¹²⁵. Thus, it is not clear whether IFNIC macrophages serve a homeostatic role or whether they are strictly disease-induced. Given the pro-atherogenic role of type I IFN signaling¹²⁶, this cell cluster could be relevant to disease progression.

Cavity macrophages represent a small cell cluster expressing Cd226, Itgax (CD11c), Ccr2, Retnla and MHCII encoding genes, thus showing similarities to previously identified monocyte-derived CD226⁺CD11c+MHCII⁺ small peritoneal macrophages that also populate other serous cavities (pleura)¹²⁷. The origin and function of these macrophages, never observed in the aorta previously, is unclear.

These five aortic macrophage populations can be observed across studies and in three distinct models of atherosclerosis in mice: Ldlr^−/− mice, Apoe^−/− mice^{5, 91, 96}, and in progressing and regressing atherosclerotic lesions of C57Bl/6 mice treated with PCSK9-AAV⁹². Thus, the present meta-analysis resolves the number of macrophage subsets in atherosclerotic arteries.

Relationship between Macrophage Subsets Identified by scRNA-Seq and CyTOF

How the 5 macrophage subsets identified by scRNA-Seq correspond to the 4 subsets identified by CyTOF is not completely clear. Future work with CITE-Seq where the cell surface phenotype and the transcriptome is available for the same cells⁷ may provide a definitive answer. However, at this point, the evidence suggests that the resident-like, foamy Trem2 and inflammatory macrophage populations may correspond to the main macrophage populations identified by mass cytometry. The largest macrophage cluster displayed a CD11b⁺ CD64⁺ CD206⁺ CD169⁺ cell surface phenotype, and could further be discriminated by expression of the C-type lectin receptor CD209b (SIGNR1) into CD209b⁻ and CD209b⁺ cell clusters¹⁰⁶. A large majority of these CD206⁺ macrophages also expressed other resident macrophage markers such as Lyve1 and Tim4 (Monaco et al., unpublished observation). Thus, CD11b⁺ CD64⁺ CD206⁺ CD169⁺ macrophages, encompassing Mac 3 and Mac4 in Cole et al.¹⁰⁶ likely correspond to the resident-like macrophages identified by scRNA-seq.

Another cluster of vascular macrophages expressed CD11c, CD44, CD11b and CD64 on the cell surface but lacked resident markers such as CD206 and CD169¹⁰⁶. MHCII expression appeared to subdivide this subset further into two clusters, however, it is yet to be determined whether these two are functionally or ontogenetically distinct. Given high expression of CD11c in foamy intimal macrophages⁹¹, this cell cluster may correspond to foamy Trem2 macrophages. Mass cytometry also identified a small macrophage subset expressing monocyte chemokine receptor CCR2 and low/intermediate levels of CD206¹⁰⁶, which may map to inflammatory macrophages.

Macrophage and Smooth Muscle Foam Cells

It has been postulated that vascular smooth muscle cells can transdifferentiate to macrophage-like cells in murine and human atherosclerotic lesions^{94, 95, 128}. These cells acquire some macrophage markers like CD68. However, scRNA-Seq of fate-mapped SMCs demonstrated that smooth muscle cells transform into unique fibroblast-like cells, termed ‘fibromyocytes’, rather than into macrophages. Their portfolio of macrophage genes is limited⁸. Such ‘fibromyocytes’ are not recovered by cell isolation strategies focusing on leukocytes or cells derived from Cx3cr1⁺ precursors^{5, 92, 96}. However, in a study focused on foam cells, almost half of the foam cells were of non-leukocytic origin and contained mostly smooth muscle cells (Acta2) and few endothelial cells (Pecam1)⁹¹. In our meta-analysis, intimal foam cells (figure 3, dataset “Apoe−/− intimal foam cells (Kim et al.)”) clustered with foamy Trem2 macrophages, as expected, and cells enriched in SMC genes (Acta2, Tagln). Thus, the present meta-analysis corroborates that smooth muscle cells significantly contribute to foam cells. In all data sets, single cells were recovered that clustered with these SMC-derived foam cells (figure 3), suggesting contaminating SMCs.

Comparing Macrophage Subsets Defined by scRNA-Seq, bulk RNA-Seq and Gene Chips

It is also not clear how the four macrophage subsets identified by scRNA-Seq correspond to the four macrophage subsets GFP⁺, YFP⁺, GFP⁺YFP⁺ and unlabeled in Cx3cr1-GFP⁺/CD11c-YFP Apoe^−/− mice¹²⁹. McArdle et al. re-analyzed the macrophage data from the Winkels⁵ study. Using genomic gating, they found 4 macrophage subsets (rather than the 3 originally reported in Winkels). One of the macrophage subsets identified by scRNA-Seq expressed Cx3cr1, one expressed Itgax, one expressed both and one expressed neither. Thus, it is possible that the GFP⁺ macrophages correspond to the Cx3cr1⁺, the YFP⁺ to the Itgax⁺, the GFP⁺YFP⁺ to the Cx3cr1⁺Itgax⁺ and the unlabeled to the Cx3cr1⁻Itgax⁻. To address this question, we formally mapped the four subsets from the McArdle study to the 4 subsets from the Cochain and Winkels studies^{5, 96}. We compared the genes that were significantly up- or downregulated in one macrophage subset versus the three others in McArdle et al. with the genes that were significantly up- or downregulated in one macrophage subset versus the three others in the present meta-analysis (figure 6). Cx3cr1-GFP macrophages were most similar to resident macrophages as suggested by 36 genes upregulated in common (figure 6) and 10 genes downregulated in common. CD11c-YFP macrophages were most similar to Trem2 foamy macrophages suggested by 21 genes upregulated in common (figure 6) and 24 genes downregulated in common. GFP+YFP+ cells did not map clearly. They showed 11 and 12 genes downregulated in common with inflammatory and Trem2 foamy macrophages, respectively. Thus, based on the intravital 2-photon microscopy analysis in ¹²⁹ the Trem2 foamy macrophages may be motile and the resident macrophages not.

Figure 6. Relationship between Cx3cr1-GFP⁺ and CD11c-YFP⁺ macrophages and the resident, IFNIC, inflammatory and Trem2 foamy macrophages from the scRNA-Seq studies.

Significantly differentially expressed (DE) genes were determined for Cx3cr1⁻GFP⁺, CD11c⁻YFP⁺, GFP⁺YFP⁺ and unlabeled macrophages against the other 3 subsets. The same was done for resident, IFNIC, inflammatory and Trem2 foamy macrophages. The DE gene lists were intersected. Only the genes that were upregulated in common are shown.

Monocytes in Atherosclerosis

Monocyte recruitment intensifies in the setting of vascular inflammation. Parabiosis studies have shown that recruited monocytes persist within the tissue or become tissue macrophages in early lesions^{15, 130, 131}. This is supported by observing the integrated data from Ldlr^−/− and Apoe^−/− mice under normo- and hypercholesterolemic conditions. This meta-analysis revealed a distinct monocyte cell cluster (Ly6c2). Monocytes have been characterized in the bone marrow and blood using RNA sequencing of sorted monocyte subsets. These studies revealed the enrichment of several genes including Lgals3, Mmp8, Ccr2, Ly6c2, and Cebpd in classical Ly6C⁺ monocytes. Nonclassical Ly6C^int/Ly6C⁻ monocytes expressed elevated levels of Nr4a1, Cebpb, Mef2a, Pparg, Cd209a, and Itgal transcripts¹³². The monocyte population in the atherosclerotic aorta does follow a classical Ly6c⁺ (Ly6c2a⁺Ccr2⁺) or non-classical Ly6c⁻ (Ccr2⁻Nr4a1^hiCx3cr1^hi) transcriptomic profile as defined by Thomas and colleagues¹³³ and Mildner and colleagues¹³². Thus, aortic monocytes possess characteristics of both classical and non-classical monocytes. It remains to be determined how soon and how completely they differentiate into macrophages or monocyte-derived DCs. Monocyte differentiation programs may initiate soon upon entry to the atherosclerotic lesion or adventitia. However, in a model of skin inflammation, monocytes have been shown to retain transcriptomes similar to blood monocytes during their migration to draining lymph nodes¹³⁴. Indeed, the meta-analysis conducted on scRNA-Seq data from atherosclerotic aortas shows gene signatures compatible with classical and non-classical monocytes.

The present meta-analysis shows that the vast majority of monocytes were detected in aortas of WD or HFD-fed mice. Furthermore, when comparing cell populations from CD-fed Ldlr^−/− and Apoe^−/− mice, the small number of monocytes contributed from “healthy” mice are from the Apoe^−/− contributed cells. This is likely due to the 2-fold elevated circulating plasma cholesterol in Apoe^−/− mice that occurs under basal conditions, which is known to induce spontaneous lesion formation in the absence of diet-induced hypercholesterolemia¹³⁵. This is in keeping with previous functional studies reporting that the recruitment of classical monocytes to atherosclerotic lesions begins to increase approximately 2 weeks after the initiation of diet-induced hypercholesterolemia¹³⁶ and that monocyte recruitment to the ascending aorta is elevated in 10 week-old Apoe^−/− mice⁷⁷. It remains to be fully understood why there is a delay between the recruitment of monocytes to nascent lesions (immediately) and the detection of macrophage foam cells (detectable within a few days of initiating hypercholesterolemia). Although recruitment of monocyte persists in advanced lesions^{77, 130, 131}, parabiosis experiments showed that recruited monocytes contribute minimally to lesion macrophages¹⁵.

Hypercholesterolemic mice also display monocytosis in the bone marrow¹³¹. Hypercholesterolemia in mice expands monopoiesis, resulting in increased numbers of Ly6C⁺ monocytes in blood¹³⁷. High levels of hypercholesterolemia in mice induce the formation of foamy monocytes¹³⁸. The cytoplasmic lipid droplets result in a high side scatter when analyzed by flow cytometry. Monocyte lipid uptake is associated with upregulated expression of CD11c, chemokine receptors, and activation of α4 integrin (CD49d), which mediates adhesion to vascular cell adhesion molecule-1^{80, 139}. Studies designed to detect monocyte fate-differentiation during atherosclerotic lesion development will require the utilization of peripheral monocyte-inducible systems (driven by Ccr2, Ms4a3¹⁴⁰ or Cxcr4¹⁴¹) with periodic retrieval to map differentiation programs from these cells. These approaches will answer long-standing questions regarding kinetics of fate-specification and heterogeneity of differentiation programs of monocytes entering inflamed lesions.

B Cells in Atherosclerosis

Although some B cells are found, especially in the Apoe^−/− data set⁵ (figure 3) but also in the healthy aorta⁹⁶ (figure 3), single cell analyses of atherosclerotic plaques show that B cells are not a predominant cell type in atherosclerotic aortas^{5, 7, 68}. Resting B cells do not produce antibodies, but can differentiate to antibody-producing plasma cells in supportive niches of the spleen and bone marrow. Hence, it is not surprising that antibody production in the artery wall itself is minimal⁶⁸. However, antibodies can be produced in aortic tertiary lymphoid organs (ATLO)^{142, 143} located in the adventitia adjacent to atherosclerotic plaques and in perivascular adipose tissue (PVAT) of aged mice. In 1 year-old mice, B cells are abundant in ATLOs^{5, 68} and may be more differentiated. Earlier studies^{2, 144} suggesting a greater abundance of B cells in atherosclerotic vessels likely included the adventitial and PVAT compartments. Flow cytometry⁶⁸, CyTOF and RNAseq⁵ confirm the presence of both B1 and B2-like cells in atherosclerotic vessels (figure 2). Although the exact location is not known, many B cells may reside in the adventitia and PVAT. These findings underscore the need to further study adventitia and PVAT as vascular compartments important in immune regulation of atherosclerosis. Moreover, future characterization of the unique BCR sequences associated with different B cell subsets and functions will provide important insights into the specific antigens that are recognized and enhance our understanding of B cell immunity in the context of atherogenesis.

Neutrophils in Atherosclerosis

While neutrophils are consistently found in atherosclerotic lesions and aortic tissue when using antibody-based detection methods, some studies employing scRNAseq of both human and mouse atherosclerotic tissue have failed to present a neutrophil cluster^{5, 7}. This discrepancy is likely explained by the very low mRNA content of neutrophils when compared to other leukocyte subsets. In addition, neutrophils are rich in easily releasable ribonucleases that rapidly and potently degrade endogenous RNA¹⁴⁵. Regardless of such intrinsic obstacles, neutrophils have been detected in healthy and atherosclerotic arteries by scRNAseq⁹⁶, and in this meta-analysis (figure 2). In atherosclerosis, these neutrophils have recently been proposed to segregate into Siglecf^hi and Siglecf^low neutrophil subsets, similar to the neutrophil subpopulations infiltrating the heart after myocardial infarction¹⁴⁶.

T cells in Atherosclerosis

Four studies employed single cell RNA-sequencing of flow-sorted CD45⁺ leukocytes in murine and human atherosclerosis^{5, 7, 96, 147}. In these studies, T cells were identified as cells expressing the mRNA coding for CD3d or CD3e (Cd3d, Cd3e). Coding genes for the T cell receptor (TCR) as frequently used in flow cytometry are less reliably found in scRNAseq data sets. In these studies, T cells were present at all stages of atherosclerosis development. Depending on models, diets and time points, Cd3e⁺ cells accounted for ~46 to 65% of leukocytes enzymatically released from mouse aortas. The presence of CD3d/e⁺ cells in different layers of atherosclerotic arteries has been inferred from a genetic deconvolution strategy of microdissected tissues¹⁴⁸ with highest absolute cell numbers in atherosclerotic lesions and the highest fraction among all leukocytes in the adventitial layer of established atherosclerosis. In aged mice, the adventitia may contain arterial tertiary lymphoid organs (ATLOs)^{5, 148, 149}. Direct scRNAseq of adventitial leukocytes confirmed the presence of T cells in WT and Apoe^−/− mice in the adventitia of aortas from 12-week-old mice on a chow diet¹⁴⁷. Notably, scRNAseq identified T cells even in healthy arteries from Apoe^−/− mice⁵. Fractions of Cd3d/e⁺ cells among all lesional leukocytes seem to be relatively higher in Apoe^−/− and Ldlr^−/− mice fed with a chow diet (61 and 54%, respectively), while the fraction of monocytes and macrophages increases during WD and HFD feeding in Apoe^−/− and Ldlr^−/− mice. This suggests that in the setting of aggravated atherosclerosis, the contribution of myeloid cells increases relatively to T cells. In humans, CD4⁺ and CD8⁺ T cells account for ~65% of all leukocytes in carotid endarteriectomy specimens as measured by CyTOF⁷. These results are at odds with the predominance of macrophages and the low percentage of T cells (5 to 20%) in immunohistochemistry of atherosclerotic plaques^{1, 150}, suggesting that enzymatic isolation, which is required before flow cytometry and scRNAseq, may overestimate relative T cell content due to a loss of fragile myeloid cells including macrophages during tissue digestion. Genetic deconvolution of human carotid plaque gene expression data sets¹⁵¹ suggests that T cells may account for about 1/8 of all leukocytes, while macrophages dominate in atherosclerotic plaques.

The role of T cells in atherosclerosis was recently reviewed¹⁵². As introduced above, phenotypes of T cells are traditionally attributed to lineages that are defined by the expression of specific transcription factors and cytokines¹⁵³. In CD4 T cells, the expression of adhesion and chemokine receptors correlates with T_H-lineage assignment and may be used as surrogates of T_H phenotypes¹⁵⁴. One technical limitation of scRNAseq, in particular of commercial drop-sequencing approaches, is the only incomplete coverage of transcription factor expression at a single cell level. Transcription factors are not highly expressed. Thus, scRNA-seq may reveal a transcription factor-coding mRNA or not, depending whether transcripts were present at the time of sampling¹⁵⁵. In addition, expression of T_H-specific cytokines at the mRNA and protein levels may only be detectable after cell stimulation, e.g. by PMA and ionomycin, for several hours. The latter considerations make scRNAseq a less reliable tool for detecting expression of T_H-defining transcription factors or cytokines that would identify T_H-lineage commitment. Dimensionality reduction tools may overcome this limitation by clustering non-transcription factor and non-cytokine coding T_H-specific genes.

So far, T cell heterogeneity in atherosclerosis has only been characterized in single cell suspensions of aortic leukocytes. This provides sufficient discrimination between major principal leukocyte lineages, e.g. T- vs myeloid cells, but underestimates heterogeneity and differentiating genes within principal lineages. For example, one macrophage cluster in the study of Winkels et al.⁵ was re-clustered and revealed three distinct subsets¹²⁹. Only one study focusing on T_reg plasticity has employed scRNAseq of flow sorted CD4⁺ T cells³⁷. In the available studies, several distinct T cell clusters were identified: Winkels et al. detected 5 T cell populations in Apoe^−/− mice⁵, Cochain et al. 4 in Ldlr^−/− mice⁹⁶, and Gu et al. 3 populations in adventitial cell preparations from Apoe^−/− and Ldlr^−/− mice¹⁴⁷.

In the atherosclerotic plaque, one of the T cell clusters across studies represents CD8⁺ (Cd8b1⁺) cytotoxic T cells^{5, 96}. Differentially expressed genes in the CD8 T cell cluster included Nkg7 (Natural Killer Cell Granule Protein 7), Ms4a4b, the CD20 homologue in T cells, Ccl5, and Gzmk (Granzyme K). Notably, this cluster was also present in healthy arteries of 8-week-old Apoe^−/− mice, indicating the existence of tissue-resident CD8⁺ T cells. Another cluster of potentially immature T cells expresses both CD4 and CD8 (previously termed ‘mixed’^{5, 96}).

The most discrete cluster among non-CD8⁺ T cells identified in two independent studies expresses the chemokine receptor Cxcr6 and the transcription factor Rora⁺ (ROR-α). These cells accounts for ~9 % of cells (Ldlr^−/−) and ~15 % (Apoe^−/−) of non-CD8 T cells, respectively. CXCR6 is a chemokine receptor that is expressed by CD4⁺ T cells, Natural Killer T (NKT) cells and γδ T cells. Global deficiency of CXCR6 reduced atherosclerosis and CD4⁺ T cell accumulation¹⁵⁶, particularly of the pro-inflammatory IL-17A-producing CD4⁺ T cell subset¹⁵⁷, consistent with the concomitant expression of Rora in this subset. In this cluster of Th17-like cells, Cd4 was not detected, arguing that γδ T cells and not Th17 cells dominate.

Our meta-analysis identified another cell cluster displaying an ILC2 gene signature⁵. This cluster was mostly negative for Cd3 and Cd4, but showed enrichment in Areg encoding amphiregulin¹⁵⁸ and Il1rl1 (encoding the IL-33 receptior ST2), consistent with type 2 innate lymphoid cells (ILC2s) that require RORA and GATA3 for their development^{159, 160}. This cell cluster may also contain a few bona fide Th2 cells. Further analysis of larger cell numbers combined with CITE-seq detection of surface markers and 5’-seq of T cell receptor encoding transcripts will likely help resolve the full spectrum of T cell heterogeneity in atherosclerotic vessels.

CyTOF and scRNA-Seq data inherently do not contain spatial information. Thus, it remains unknown in which regions of the atherosclerotic plaque T cells accumulate. Immunohistochemistry has indicated that T cells are mostly found in the fibrous cap regions of the plaque¹. Spatial transcriptomics could precisely address this question by combining single cell gene expression and locations. No such studies are available in mouse or human atherosclerosis.

Association Between Clinical Outcome and Plaque T cell Phenotypes

A recent study describes the T cell landscape of carotid artery endarteriectomy specimens from patients with symptomatic (recent stroke or transient ischemic attack) or asymptomatic disease⁷. Single-cell proteomic and transcriptomic analyses revealed a distinct distribution of leukocytes between blood and atherosclerotic plaques with a population of CD4⁺ and CD8⁺ cells being the most abundant in the plaques. CyTOF analysis identified 13 clusters of aortic CD4⁺ T cells that include central memory, effector memory, terminally differentiated effector memory, and regulatory T cells. One of the most striking phenotypic features of plaque-derived T cells is elevated expression of the activation marker CD69, the chemokine receptor CCR5, and PD-1, a negative regulator of T cell activation and a marker of T cell exhaustion, within CD4⁺ T cells. These data further support the idea that continuous and repeated activation of T cells in the aorta results in dysfunctional T cells that correlate with inflammation. Based on additional CD4⁺ subsets discovered by CyTOF and scRNA-Seq, it will be important to elucidate transcriptional regulation of T_H cell subsets and to identify molecules that drive phenotypes and functions of Th cells in atherosclerosis. In a genetic deconvolution strategy, a gene signature of a scRNAseq-derived T cell cluster from mouse plaques negatively correlated with the clinical outcome of carotid stenosis, corroborating the concept that T cell phenotypes may predict plaque outcomes¹⁶¹.

Limitations of this study

scRNA-Seq preferentially yields highly expressed genes, because genes with low expression or periodic expression have many dropouts and thus are less likely to be significantly different between cell types. This is particularly obvious for transcription factors. One way around this issue is to define gene signatures of abundantly expressed genes that correlate with the low-expressed genes of interest.

Although a total of over 15,000 cells were analyzed in this meta-analysis, this number may still be too low to truly capture all cell subsets and their intermediate states. Rapidly evolving scRNA-Seq technology makes the analysis of larger numbers of cells feasible, approaching the numbers of cells analyzed by FACS or CyTOF.

The cellular stress response during enzymatic processing can induce gene expression artifacts⁶. Expression of immediate early genes induced during enzymatic digestion could induce artifactual clustering of macrophages. This is of particular concern, as many of these genes encode transcripts involved in immune responses, such as heat shock proteins (Hspa1b) that have been used to identify some clusters in atherosclerosis⁹². New protocols e.g. employing cold active protease digestion coupled with inhibition of new RNA synthesis during cell processing may help avoid such biases in future studies⁶.

mRNA levels for cell surface markers are poorly correlated with protein expression on the cell surface. This is because cell surface proteins must undergo glycosylation, cleavage of the signal peptide, vesicular trafficking and sometimes enzymatic modifications to appear at the cell surface. Experience teaches that it is useful to measure the cell surface phenotype in addition to single cell transcriptomes. Although 35 CyTOF markers provide a fine-grained picture of leukocyte heterogeneity in atherosclerosis⁵, it remains to be determined how the CyTOF clusters relate to transcriptional cell states and unique cellular functions. CITE-Seq allows linking cell surface phenotype to transcriptomes by using oligonucleotide-tagged antibodies¹⁶², combining scRNA-Seq with quantitative measurement of 50-200 cell surface markers, which will help to further resolve cluster phenotypes. There is no CITE-Seq dataset available for mouse aortas.

Enzymatic tissue digestion methods may lead to enrichment or loss of specific cell types, which may explain the high variability in aortic immune cell composition observed across the different studies. Macrophages are likely much more sensitive to isolation and sorting procedures than T cells and thus may be lost preferentially. In particular, it is unclear if large, lipid-laden foam cells that are embedded deep within lesions can be efficiently isolated and sorted. The overall low proportion of foamy Trem2 macrophages enumerated in scRNA-seq analysis may thus underestimate the contribution of this cell subset to lesion cellularity. Alternative tissue preparation techniques⁶ may overcome this technical limitation.

Because of differences in cell isolation procedures, different animal models and experimental variations, a direct comparison of cell types in healthy and atherosclerotic arteries is still challenging. Additional studies directly comparing cell subset distributions during atherosclerotic lesion formation, e.g. using cell hashing techniques at multiple time points of disease progression are warranted.

scRNA-Seq and CyTOF retain no information on cell position within the vessel wall or relative to other immune cells. Spatial transcriptomics methods^{163, 164} are on the horizon, but we currently still lack information on spatial gene expression, including the exact localization of immune cells in the plaque versus the adventitia.

scRNA-Seq is a discovery tool. As such, it is great at identifying new cell types. For example, scRNA-Seq provided clarity with respect to the macrophage types in atherosclerotic mouse aortas. Once the phenotype of the cells of interest is known, with the cell surface phenotype defined by CITE-Seq, flow cytometry-based cell sorting can be used to sort homogeneous cell populations, extract RNA and perform bulk RNA-Seq. This yields much better, deeper transcriptomes than scRNA-Seq¹²⁹.

Conclusions

The modern single cell methods scRNA-Seq and CyTOF consistently defined 3-5 different macrophage subsets, two monocyte subsets, 3-5 T cell subsets, 2 B cell subsets, one NK cell subset, neutrophils, eosinophils and dendritic cells in atherosclerotic mouse aortas. This approach, albeit in its infancy, has promise in terms of disambiguation of leukocyte cell populations and states within lesions giving us the tools for multi-dimensional atherosclerotic lesion phenotyping.

By integrating 9 different datasets, we identified the cell subsets found in healthy and atherosclerotic aortas at unprecedented resolution. Macrophages are the predominant cell type in atherosclerotic aortas and can be distributed into 5 different subsets. In addition, neutrophils, monocytes, moDCs, mature DCs and pDC-like cell clusters can be discriminated. Among the T cells, a cluster of CD4⁺CD8⁺ T cells exclusively expressing Sox4 was also found in healthy aortas and may represent T cells in an immature state. In addition, naive T cells were found in both healthy and atherosclerotic aortas, in line with their constitutive migration into the aorta². Some pro-atherogenic Th1 T cells expressing Tbx21 were dispersed among T and NK cells but did not form an identifiable cell cluster. Reclustering T cells revealed a cluster of Tregs. Il17⁺ Cxcr6⁺ T cells were consistently found. Some B1-like and mostly B2-like cells were also present in both healthy and atherosclerotic aortas.

This meta- resolved several apparent controversies. First, foam cells in atherosclerotic aortas are macrophage-derived and smooth muscle cell-derived. Although these smooth muscle foam cells acquire some markers including CD68, they do not become macrophages. Second, macrophage-derived foamy Trem2 cells do not express a pro-inflammatory gene signature. This reconciles a study by Glass et al.¹¹² with many studies showing pro-inflammatory functions of atherosclerotic macrophages. This is resolved by the discovery of two inflammatory macrophage subsets, called IFNIC and inflammatory macrophages. Third, resident-like macrophages are found in the atherosclerotic aorta and express Pf4. Pf4 is not platelet-specific, because Pf4 is consistently expressed in these resident vascular macrophages. Reclustering aortic myeloid cells revealed a fifth macrophage subset resembling cavity macrophages. Their origin and function are unclear. We also identified proliferating macrophages, a process known to dominate macrophage accumulation in atherosclerosis¹⁵.

The discovery of an ILC2 cluster links the scRNA-Seq work to recent work by Mallat’s group¹⁶⁵, showing that ILC2 cells are strongly atheroprotective. This could resolve the controversy about the role of Th2 cells in atherosclerosis: Some of these Th2 cells in earlier studies may have been ILC2 cells instead. Finally, the observation of SiglecF⁺ neutrophils in atherosclerosis¹⁴⁶ is striking. Such neutrophils had previously been observed in cancer¹⁶⁶, and in the infarcted mouse heart^{124, 146}.

This meta-analysis also generated new hypotheses that can be tested in future work. The Treg switch hypothesis has received much support from the recent single cell studies, but it remains a hypothesis that needs to be tested. In particular, the switch mechanism is not known, neither is the switch direction. There is evidence for a switch to Th1^{37, 38}, Th2⁵, Th17¹⁶⁷ and TFH¹⁶⁸. Other hypotheses flowing from this work include the NK cell hypothesis. In mouse models, NK cells do not accelerate or curb atherosclerosis⁶⁴, but are more like bystanders. The reason for this cannot be gleaned from the NK cell transcriptomes consistently found in the scRNA-Seq studies and awaits further experimental testing.

The single cell interrogation techniques have ushered in a new wealth of information that will eventually lead to new depths of understanding. Here, we identify 19 types of leukocytes in mouse aortas: 5 types of macrophages, 5 types of T cells, 2 types of monocytes, 2 types of DCs, B1, B2, NK and ILC2 cells (figure 7). Future work will combine the power of cell surface phenotype assessment by CITE-Seq with single cell transcriptome analysis. Highly detailed analysis of the immune cell infiltrate in atherosclerotic lesions in experimental model systems is needed to translate the insights from the immunology of atherosclerosis into therapies that ultimately can benefit patients.

Figure 7. Summary of cell clusters detectable in the atherosclerotic aorta.

The leukocyte infiltrate in atherosclerotic mouse aortas as analyzed in the comprehensive meta-analysis of 9 single cell RNA-Seq studies identified 19 types of leukocytes: neutrophils, monocytes and DCs, 5 types of macrophages, 5 types of T cells, B1, B2, NK and ILC2 cells. Top expressed and characteristic genes are indicated in italics.

Nonstandard Abbreviations and Acronyms:

AAV

adeno-associated virus

Acta2

actin alpha 2

ATLO

artery tertiary lymphoid organ

Apoe

apolipoprotein E

CC

C-C chemokine

CXC

C-X-C chemokine

CX3C

C-X3-C chemokine

CD

Cluster of differentiation

CITE-seq

cellular indexing of transcriptomes and epitopes by sequencing

CTL

cytotoxic T lymphocytes

CyTOF

mass cytometry

DC

dendritic cell

DEG

differentially expressed gene

GFP

green fluorescent protein

GSEA

gene set enrichment analysis

GWAS

genome-wide association study

HFD

high-fat diet

IFNIC

interferon-inducible cells

ILC

innate lymphocyte-like cells

LDL

low density lipoprotein

Ldlr

low density lipoprotein receptor

MHC

major histocompatibility complex

NK cell

natural killer cell

NKT

natural killer T cell

RAG

recombinase activating gene

scRNA-seq

single cells RNA sequencing

TCR

T cell receptor

TFH cell

T follicular helper cell

T_H cell

T helper cell

Treg

regulatory T cell

TREM2

triggering receptor expressed on myeloid cells 2

TRM

tissue resident memory

UMAP

uniform manifold approximation and projection

pDC

plasmacytoid dendritic cell

PCSK9

protein convertase subtilisin/kexin type 9

PVAT

perivascular adipose tissue

VCAM

vascular cell adhesion molecule

WD

Western diet

YFP

yellow fluorescent protein