Integrated single-cell analysis-based classification of vascular mononuclear phagocytes in mouse and human atherosclerosis

Abstract

Aims

Accumulation of mononuclear phagocytes [monocytes, macrophages, and dendritic cells (DCs)] in the vessel wall is a hallmark of atherosclerosis. Using integrated single-cell analysis of mouse and human atherosclerosis, we here aimed to refine the nomenclature of mononuclear phagocytes in atherosclerotic vessels and to compare their transcriptomic profiles in mouse and human disease.

Methods and results

We integrated 12 single-cell RNA-sequencing (scRNA-seq) datasets of immune cells isolated from healthy or atherosclerotic mouse aortas, and data from 11 patients (n = 4 coronary vessels, n = 7 carotid endarterectomy specimens) from two studies. Integration of mouse data identified subpopulations with discrete transcriptomic signatures within previously described populations of aortic resident (Lyve1), inflammatory (Il1b), as well as foamy (Trem2^hi) macrophages. We identified unique transcriptomic features distinguishing aortic intimal resident macrophages from atherosclerosis-associated Trem2^hi macrophages. Also, populations of Xcr1⁺ Type 1 classical DCs (cDC1), Cd209a⁺ cDC2, and mature DCs (Ccr7, Fscn1) with a ‘mreg-DC’ signature were detected. In humans, we uncovered macrophage and DC populations with gene expression patterns similar to those observed in mice. In particular, core transcripts of the foamy/Trem2^hi signature (TREM2, SPP1, GPNMB, CD9) mapped to a specific population of macrophages in human lesions. Comparison of mouse and human data and direct cross-species data integration suggested transcriptionally similar macrophage and DC populations in mice and humans.

Conclusions

We refined the nomenclature of mononuclear phagocytes in mouse atherosclerotic vessels, and show conserved transcriptomic features of macrophages and DCs in atherosclerosis in mice and humans, emphasizing the relevance of mouse models to study mononuclear phagocytes in atherosclerosis.

Macrophage, Atherosclerosis, Single-cell RNA-seq

Time for primary review: 28 days

1. Introduction

Atherosclerosis is a chronic disease of the arterial wall characterized by chronic lipid accumulation and inflammation in the vascular intima. Its main clinical manifestations, myocardial infarction, and ischaemic stroke, together constitute the most frequent cause of death worldwide.¹ Many adaptive and innate immune cell types have been proposed to contribute to vascular inflammation and atherosclerosis,² and various types of mononuclear phagocytes such as monocytes, macrophages, and dendritic cells are thought to play major roles in the pathogenesis of atherosclerosis.^3,4

Precisely identifying mononuclear phagocyte cell types and the states they assume under homeostatic and inflammatory conditions is a prerequisite for in-depth investigations of their functional role in health and disease. For instance, it has long been assumed that macrophages with distinct functional cell states populate atherosclerotic vessels, but this has been studied mostly using variations of the simplistic M1/M2 macrophage polarization model.⁵ Furthermore, properly discriminating dendritic cells from macrophages, two functionally distinct cell types, has been a long-standing challenge in atherosclerosis⁴ and immunological research,⁶ and misidentification of cells (e.g. identification of CD11c⁺ macrophages as dendritic cells⁷) might cause major bias in data interpretation. In recent years, single-cell technologies have revolutionized high-dimensional profiling of immune cells in experimental models, and fostered major advances in our understanding of mononuclear phagocytes, providing novel insights into macrophage heterogeneity in disease⁸ and expanding dendritic cell nomenclature.⁹

In atherosclerosis research, single-cell studies advanced our understanding of macrophage diversity in murine atherosclerotic vessels,^{7,10,11,12,13,14} and in particular helped to define the transcriptomic profile of Trem2^hi foamy macrophages,¹⁰ and to identify aortic intimal resident macrophages (MAC-AIR).⁷ First available data in humans suggest that macrophages with distinct transcriptional states are also found in atherosclerotic vessels.^{15,16,17,18,19} However, properly identifying and annotating context-specific cell types and states in single-cell transcriptomics data remains a major challenge,²⁰ and a comprehensive nomenclature of mononuclear phagocytes and their marker transcripts in atherosclerosis are still lacking. As mouse models of atherosclerosis are widely employed to decipher the pathogenesis of atherosclerosis, and to perform pre-clinical investigations of new therapeutic targets,²¹ it is furthermore critical to determine whether mononuclear phagocyte cell types and cell states in mouse atherosclerotic vessels are of relevance to human disease.

Here, we performed a computational integration of single-cell RNA-sequencing (scRNA-seq) data of immune cells from independent studies investigating mouse models of atherosclerosis and human atherosclerotic plaque tissue. Our work shows conserved transcriptomic features of macrophages in atherosclerosis across mouse models, defines precise combinations of markers to identify mononuclear phagocyte subsets, and identifies novel putative atherosclerosis-associated macrophage subpopulations and their markers. We furthermore provide evidence that major features of macrophage states in mouse atherosclerosis are also observed in human lesions, emphasizing the relevance of mouse models to study macrophage biology in atherosclerosis.

2. Methods

Using 12 distinct mouse scRNA-seq datasets from six studies^7,11–14,22 (see Supplementary material online, Table 1), and three human scRNA-seq datasets from two studies,^15,16 we performed three integrated analyses (see Supplementary material online, Figure S1A–C): (i) integration of all mouse data, (ii) integration of human mononuclear phagocyte data, and (iii) integration of mouse and human mononuclear phagocyte data. Deposited cell–gene count matrixes were analyzed in Seurat,²³ where all datasets were first pre-processed individually for quality control filtering, selection of relevant cells for further analysis, and assignment of metadata information (e.g. species, protocol, patient) (see Supplementary material online, Table S1, Supplementary material online, Figure S1). To identify the sex of mice used in studies where this information was unavailable,¹⁴ we interrogated the expression of female (Xist) or male specific transcripts (Ddx3y, Uty, Eif2s3y) (see Supplementary material online, Figure S1D). Integration was performed using Seurat v3.²³ The code used for analysis is provided as R notebooks (Supplementary material online, Supplemental Files). Overlapping marker gene lists shared between human macrophage clusters and their putative mouse counterparts were identified using InteractiVenn.²⁴ Gene Ontology (GO) analysis was performed using http://geneontology.org/.^25,26 Transcription factor enrichment analysis was performed using ChEA3 (https://maayanlab.cloud/chea3/).²⁷ Illustration was created using Servier Medical Art (https://smart.servier.com).

3. Results

3.1 Integrated analysis of mouse aortic immune cell scRNA-seq datasets

The integrated analysis of mouse aortic leucocyte datasets from different models of atherosclerosis and experimental conditions (Table 1, Supplementary material online, Figure S1) allowed us to analyze a final number of 22 852 cells (Figure1A and B). We identified macrophages and dendritic cells based on the expression of canonical markers such as Adgre1 (encoding F4/80), Fcgr1 (encoding CD64), and Itgax (encoding CD11c) (Figure 1C). Several populations of T (Cd3d), B (Cd79a), Cytotoxic/NK cells (Nkg7), and neutrophils (S100a8, S100a9),^11–13 could be discerned. A cluster resembling Type 2 Innate Lymphoid Cells (ILC2) (Il1rl1, Gata3) was identified, possibly also containing mast cells and basophils (Cpa3, Calca) (see Supplementary material online, Figure S2A). We had previously identified these cells as mixed/mast cells in Cochain et al.¹¹ A cluster of non-leucocytic cells was also found, originating from the scRNA-seq analysis of intimal BODIPY⁺ foam cells, which included lipid-rich intimal cells of both immune and non-immune origin.¹²

Figure 1

Integrated scRNA-seq analysis of vascular inflammation in mouse atherosclerotic aortas. (A) UMAP representation of integrated scRNA-seq gene expression data in 22 852 cells from mouse atherosclerotic aortas with identification of the major immune cell lineages (DC, dendritic cells; gdT, gammadelta T cells) and (B) projection of single cells in the UMAP space according to dataset and experimental condition of origin. (C) Expression of the indicated transcripts projected onto the UMAP plot.

Table 1

Mouse aortic cell single-cell RNA-seq datasets used in the study

Reference	Model	Diet	Technology	Time points	Sorting strategy	i.v. CD45 exclusion	Sex
Vafadarnejad et al. (2020)	Ldlr−/−	15% milk fat, 1.25% cholesterol;	10x Genomics Single Cell 3′ v2	10 weeks HFD	Live CD45+	Yes	Male
Cochain et al. (2018)	Ldlr−/−	Normal chow	10x Genomics Single Cell 3′ v2	Normal chow	Live CD45+	Yes	Male
Cochain et al. (2018)	Ldlr−/−	15% milk fat, 1.25% cholesterol;	10x Genomics Single Cell 3′ v2	11 weeks HFD	Live CD45+	Yes	Male
Cochain et al. (2018)	Ldlr−/−	15% milk fat, 1.25% cholesterol;	10x Genomics Single Cell 3′ v2	20 weeks	Live CD45+	Yes	Male
Kim et al. (2018)	Apoe−/−	49.9% carbohydrates, 17.4% protein, 20% fat, 0.15% cholesterol	10x Genomics Single Cell 3′ v2	27 weeks HFD	Intimal BODIPYhi cells	No	Male
Kim et al. (2018)	Ldlr−/−	49.9% carbohydrates, 17.4% protein, 20% fat, 0.15% cholesterol	10x Genomics Single Cell 3′ v2	10 weeks HFD	Live CD11b+	No	Male
Lin et al. (2019)	C57BL6a; PCSK9-AAV	Western Diet; (Dyets Inc. #101977)	10x Genomics Single Cell 3′ v2	18+2 weeks HFD (progression)	CD11b+TdTomato+ macrophagesb	No	Male
Lin et al. (2019)	C57BL6a; PCSK9-AAV	Western Diet; (Dyets Inc. #101977)	10x Genomics Single Cell 3′ v2	18 weeks HFD+2 weeks chow/ApoB ASO (regression)	CD11b+TdTomato+ macrophagesb	No	Female
Williams et al. (2020)	Ldlr−/−	Envigo TD.88137	10x Genomics Single Cell 3′ v2	21 days HFD	Live CD45+	No	Male
Williams et al. (2020)	C57BL6/J	Normal chow	10x Genomics Single Cell 3′ v2	Normal chow	Live CD45+	No	Male
Winkels et al. (2018)	Apoe−/−	Normal chow	10x Genomics Single Cell 3′ v2	Normal chow	Live CD45+	No	Female
Winkels et al. (2018)	Apoe−/−	0.2% cholesterol Envigo TD88137	10x Genomics Single Cell 3′ v2	12 weeks HFD	Live CD45+	No	Female

B6.Cx3cr1CreERT2-EYFP/+Rosa26-tdTomato/+; C57BL6 background.

2 weeks post Tamoxifen gavage in B6.Cx3cr1CreERT2-EYFP/+Rosa26-tdTomato/+ mice.

3.2 Macrophage subsets in mouse atherosclerotic aortas

Focusing our analysis on mononuclear phagocytes, we then interrogated expression of characteristic markers of tissue-resident macrophage subsets (Lyve1, Cd209f), inflammatory macrophages (Ccr2, MHCII encoding genes, e.g. Cd74 and H2-Eb1), foamy macrophages (Trem2), and MAC-AIR (Acp5) (Figure 2A), examined the position of MAC-AIR from data in Williams et al.⁷ (Figure 2B), and performed re-clustering of mononuclear phagocyte subsets (i.e. monocytes, macrophages, dendritic cells and proliferating cells comprising mostly Fcgr1 ⁺ Adgre1⁺ macrophages, Figure 1) at a higher resolution to identify potential subpopulations within macrophages and dendritic cells (Figure2C and D). A full list of marker genes for each cluster is available in Supplementary material online, Excel File S1. Resident/Resident-like macrophages (Lyve1, Timd4, Mrc1, Pf4) (Figure 2D) comprised two clusters, one of which showed higher expression of Cd209d, Cd209f, and Cd209g (Figure 2D), corresponding to Cd209⁺ resident macrophages identified by Cole et al.²⁸ Both populations of resident macrophages highly expressed Lyve1, Timd4, and Folr2 (Figure 2D), thus corresponding to subsets of recently described ‘TLF’ (Timd4 ⁺ Lyve1 ⁺ Folr2⁺) resident macrophages conserved across mouse organs.²⁹ We termed these clusters TLF-Cd209^hi and TLF-Cd209^low. Macrophages previously defined as ‘Inflammatory-Mφ’ in a study analyzing much lower cell numbers¹¹ actually comprised two clusters: Inflammatory-Nlrp3 macrophages displayed high Nlrp3 (encoding NLRP3, the sensor component of the NLRP3 inflammasome) and Il1b (Interleukin-1β) expression, while a second subset of CCR2^intMHCII⁺ macrophages expressed intermediate levels of Ccr2 and of MHCII (major histocompatibility complex class II)-encoding transcripts (Cd74, H2-Aa, H2-Eb1) (Figure2C and D). Proliferating cells could be delineated into S-phase and G2M-phase cells (Figure2C and D, Supplementary material online, Figure S2B). We furthermore observed populations consistent with a previous integrative analysis,¹⁰ including cells with a gene expression profile characteristic of small peritoneal macrophages (SPMs/Cavity cluster: Itgax⁺Cd226⁺Ccr2⁺ MHCII⁺),¹⁰ Type I interferon response cells [interferon inducible cell (IFNIC) cluster: Isg15, Oasl2], and monocytes expressing genes characteristic of Ly6C^hi (Ly6c2, Chil3, Ccr2) and Ly6C^low (Ace, Treml4) subsets³⁰ (Figure2C and D). Immediate-early gene (IEG) expression can be induced in mononuclear phagocytes during tissue digestion and processing.³¹ Based on the expression of 18 IEGs, we applied an IEG expression score to macrophages and dendritic cells, and cells with the highest score mapped to the ‘inflammatory-Mφ’ clusters (see Supplementary material online, Figure S2C). However, stress-induced gene expression is unlikely to have caused a major bias in our analysis, as cell clustering was not substantially affected by regressing out variability caused by the expression of immediate-early genes (see Supplementary material online, Figure S2D–F).

Figure 2

Characterization of mononuclear phagocyte subpopulations in the atherosclerotic mouse aorta. (A) Expression of the indicated transcripts projected onto the mononuclear phagocyte UMAP plot; (B) projection of cells corresponding to steady state C57BL6 aortic Mac-AIR cells (Williams et al. 2020) on the UMAP plot and (C) high-resolution re-clustering identifying an independent cluster with a Mac-AIR signature; (D) average expression of selected differentially expressed genes in each mononuclear phagocyte population; (E) proportion of the indicated populations among total aortic mononuclear phagocytes (MPCs) in mouse single-cell RNA-seq datasets (see Table 1). For consistency, not all datasets were used in E: datasets from Apoe^–/− mice (Winkels et al. 2018) were excluded as chow fed Apoe^–/− mice do not represent real non-atherosclerotic controls; the ‘Kim Foam Cell’ dataset was excluded as it is technically enriched for foam cells; the 21 days HFD dataset from Williams et al. 2020 was excluded as it represents a very early time point of lesion formation. Each data point represents an individual single-cell RNA-seq dataset. (F) Heatmap showing the mean top rank of transcription factors predicted by ChEA3 in each cluster (heatmap generated from the top 3 ranked transcription factors for each cluster), (G) expression of the indicated transcripts, encoding selected transcription factors, projected onto the mononuclear phagocyte UMAP plot.

We further analyzed the expression of transcripts associated with core gene expression signatures of three conserved tissue-resident macrophage subsets as recently described by Dick et al,²⁹ confirming that aortic resident/resident-like macrophages correspond to ‘TLF’ macrophages, and showing that the CCR2^intMHCII⁺ and Inflammatory-Nlrp3 clusters shared some similarities with tissue-resident MHCII^hi macrophages (Cd14, Lilra5, H3f3b) and tissue-resident CCR2⁺ macrophages (Ccr2, Cd52), indicating that they might represent a mix of tissue-resident macrophages, and of recently recruited monocyte-derived CCR2⁺ macrophages (see Supplementary material online, Figure S3A).

We identified two clusters enriched for genes characteristic of the foamy/Trem2^hi signature^10,32Trem2, Spp1, Cd9, and Itgax (Figure2C and D). The first population was further enriched for transcripts such as Slamf9, Ch25h, and Cd72 (Trem2^hi-Slamf9, Figure2C and D). The second population (Trem2^hi-Gpnmb) was enriched for Gpnmb, Atp6v0d2, and transcripts characteristic of a foamy signature and TREM2-response genes^33,34 (Lpl, Lipa, Fabp5, Apoc1, Apoe) (Figure2 C and D). Consistent with previous analyses^11,12 and observations in Apoe^−/−Cx3cr1^GFPCd11c^YFP mice,³⁵ foamy/Trem2^hi macrophages were enriched for Itgax (CD11c) (Figure 1C). We also identified a population with a gene expression signature characteristic of recently described aortic intimal resident macrophages (Mac-AIR) with expression of Acp5, Gngt2, and MHCII encoding transcripts⁷ (Figure2C and D). Accordingly, MAC-AIR identified in data from Williams et al.⁷ mapped predominantly to this cluster (Figure 2B–D).

We analyzed the proportion of these macrophage clusters in control vs. atherosclerotic aortas. While these analyses are statistically limited given the low number of replicates and low cell numbers in some datasets, proportions of Trem2^hi-Gpnmb macrophages appeared clearly increased in atherosclerotic aortas, confirming that Trem2^hi-Gpnmb macrophages with a foamy macrophage gene expression signature are induced by atherogenic conditions. Levels of Inflammatory-Nlrp3 macrophages also showed a trend towards increased levels in atherogenic conditions. Notably, all macrophage subsets were found in the different models analyzed, i.e. Ldlr^−/− mice, Apoe^−/− mice, and mice rendered hypercholesterolemic via PCSK9 (Proprotein convertase subtilisin/Kexin Type 9) expressing AAVs (Adeno-associated viruses)¹⁴ (see Supplementary material online, Figure S3B). To predict transcription factors that might drive the gene expression profile of the various aortic macrophage populations, we performed transcription factor enrichment analysis using ChEA3,²⁷ pinpointing enrichment in the activity of Nr1h3, encoding the liver–X-receptor alpha (LXRα), in Trem2^hi-Gpnmb macrophages and MAC-AIR (Figure 3F). This analysis also identified putative regulatory activity of Nr4a1 in CCR2^intMHCII⁺, Csnrp1 in Inflammatory-Nlrp3 macrophages, Irf9 in IFNIC, and Spic in TLF-Cd209^low macrophages (Figure 3F). Expression of some of the identified transcription factors primarily mapped to the macrophage subsets in which their activity was enriched (Figure 3G).

3.3 Analysis of MAC-AIR vs. Trem2^hi macrophages

Monocytes infiltrating the intima under atherogenic conditions have been proposed to acquire transcriptomic features resembling those of Mac-AIR.⁷ To define the relationship of Mac-AIR with both Trem2^hi populations, we performed a focused analysis of these three populations (Figure 3). Compared to other aortic macrophages, MAC-AIR, Trem2^hi-Gpnmb, and the Trem2^hi-Slamf9 macrophages shared enriched markers and expressed, e.g. high Itgax (encoding CD11c) (Figure 1), Cd9, Lgals3, and Lpl (Figure 3A). Trem2^hi-Gpnmb, Trem2^hi-Slamf9, and Mac-AIR were predominant in the intimal foam cell dataset from Kim et al.¹² in the integrated analysis (see Supplementary material online, Figure S3), representing 38, 14, and 9% of mononuclear phagocytes, respectively. Analysis of this dataset alone recovered three macrophage populations with expression signatures reminiscent of Mac-AIR (Acp5, Cd74, Mmp12, Gngt2), Trem2^hi-Gpnmb (Gpnmb, Fabp5, Cstb, Psap), and Trem2^hi-Slamf9 (Slamf9, Cd72, Cd14, Ch25h) macrophages (see Supplementary material online, Figure S4A and B), further indicating that Trem2^hi-Gpnmb, Trem2^hi-Slamf9, and Mac-AIR clusters contribute to intimal foamy macrophages.

Figure 3

Characterization of subpopulations within Trem2^hi macrophages and comparison to Mac-AIR. (A) Expression of shared markers of MAC-AIR, Trem2^hiGpnmb and Trem2^hi-Slamf9 macrophages shown across all mononuclear phagocyte subsets; (B) heatmap showing scaled average expression of marker transcripts for the indicated macrophage subsets; (C) expression of selected markers of MAC-AIR, Trem2^hiGpnmb and Trem2^hi-Slamf9 macrophages shown across these clusters (violin plots) and projected onto the UMAP plot of total mononuclear phagocytes; (D) Gene Ontology analysis for enriched biological processes for MAC-AIR, Trem2^hiGpnmb and Trem2^hi-Slamf9 macrophages. Selected overlapping and unique biological processes for each macrophage population were selected to generate the data visualization, the full analysis can be found in Supplementary material online, Excel Table S2.

We then performed differential expression analysis between these three clusters and attempted to find specific marker transcripts able to distinguish these cells (Figure3B and C). Expectedly, Mac-AIR were enriched for MHCII encoding transcripts, but also for other genes, e.g. Vcam1, Dnase1l3, or Fcgr4 (Figure3B and C). Relative to Mac-AIR and Trem2^hi-Gpnmb macrophages, the Trem2^hi-Slamf9 cluster was enriched for Cd72, Ch25h, and inflammatory markers (Tnf, Il1b) (Figure3B and C), although these were expressed at lower levels than in bona fide pro-inflammatory macrophages (Figure 1F, and not shown). Trem2^hi-Gpnmb had a specific signature including, e.g. Gpnmb, Syngr1, and Fabp5 (Figure 3C). Importantly, Mac-AIR expressed lower levels of Trem2 and Spp1 (Figure 3D), transcripts generally associated with the disease-associated ‘Trem2^hi’ signature in atherosclerosis^10,11 and other diseases.³⁴ Mac-AIR and Trem2^hi-Gpnmb macrophages, but not Trem2^hi-Slamf9 macrophages expressed Acp5 (Figure 2D). A similar signature was obtained when comparing only Mac-AIR from healthy C57 aortas⁷ vs. the Trem2^hi-Gpnmb and Trem2^hi-Slamf9 populations (not shown). Of note, recent evidence indicates that MAC-AIR can be in direct contact with the vessel lumen.³⁶ Datasets where exclusion of circulating leucocytes via intravenous CD45 antibody injection was performed (Table 1) might thus have also excluded part of MAC-AIR cells. Gene ontology analysis for biological processes indicated overlapping functions in the cellular response to lipids and lipoprotein particles in MAC-AIR and Trem2^hi macrophages (Figure 3D and Supplementary material online, Excel File S2). Each population was furthermore enriched for unique biological processes suggesting functional specialization, such as antigen processing and presentation-related processes in MAC-AIR (consistent with the expression of MHCII encoding transcripts), negative regulation of macrophage colony-stimulating factor (M-CSF) signalling pathway in Trem2^hi-Slamf9 macrophages, and osteoclast differentiation and regulation of macrophage fusion in Trem2^hi-Gpnmb macrophages (Figure 3D and Supplementary material online, Excel File S2). Interestingly, Gpnmb (also termed osteoactivin) has been proposed to be involved in osteoclast formation and multinucleation.³⁷

Altogether, this integrated analysis shows that atherosclerosis-associated macrophage transcriptional states are conserved across experimental mouse models of atherosclerosis, and that discrete subpopulations exist within the main aortic macrophage subtypes. We further confirm that Mac-AIR that reside in the normal mouse intima share transcriptional similarities with Foamy/Trem2^hi macrophages but express higher levels of transcripts encoding MHCII, have some specific marker transcripts (Vcam1, Dnase1l3) and express low levels of characteristic genes of the foamy macrophage signature (Trem2, Spp1).

3.4 Integrated analysis of aortic dendritic cells in mouse atherosclerosis

We identified cells corresponding to monocyte-derived dendritic cells (DCs) and/or cDC2 (MoDC/cDC2: Cd209a, Clec10a, Ifitm1, Napsa), cDC1 (Xcr1⁺Clec9a⁺), and Fscn1⁺Ccr7⁺ mature DCs (Figure2C and D). We had previously identified the MoDC/cDC2 cluster as potential monocyte-derived dendritic cells based on the expression of Cd209a³⁸ and intermediate expression of monocytic markers (Ccr2, Csf1r).¹¹ Under inflammatory conditions, however, cDC2 can acquire an ‘inflammatory-cDC2’ phenotype that can be discriminated from monocyte-derived cells by expression of CD26 (Dpp4) and absence of expression of CD88 (C5ar1).³⁹ CD88 also aptly discriminates monocyte/macrophages from dendritic cells in mice and humans.⁴⁰ The MoDC/cDC2 cluster expressed Dpp4 (CD26) but not C5ar1 (CD88) (Figure 4A), suggesting that these cells likely represent bona fide cDC2. Fscn1⁺Ccr7⁺ ‘mature DCs enriched in immunoregulatory molecules’ (mReg-DC) have been described in mouse and human lung cancer.⁴¹ Compared to cDC1 and cDC2 populations, aortic mature Fscn1⁺Ccr7⁺ DCs were clearly enriched for several genes characteristic of the mReg-DC signature, such as Il4i1, Cd274, Tnfrsf4, Ccl22, Cd40, or Cd86 (Figure 4B).

Figure 4

scRNA-seq data-based classification of mouse aortic mononuclear phagocytes in atherosclerosis. (A) Violin plot showing expression of the indicated macrophage/dendritic cell markers; (B-C) heatmaps showing scaled average expression of mononuclear phagocyte defining marker transcripts in (B) dendritic cell subsets and (C) macrophage subsets; (D) proposed updated, scRNA-seq data-based classification of mononuclear phagocyte subpopulations and their markers in the atherosclerotic mouse aorta.

3.5 Updated classification of mononuclear phagocyte subsets in mouse atherosclerotic aorta

Based on our scRNA-seq data and expression of discrete combinations of markers in aortic macrophages (Figure 4C) and dendritic cells (Figure 4B), we propose an updated classification of mononuclear phagocyte subsets in the mouse atherosclerotic aorta (Figure 4D), based on specific marker combinations. In particular, we propose that beyond classically used markers of macrophages (Adgre1 encoding F4/80) and dendritic cells (Itgax encoding CD11c), the expression of Dpp4 (encoding CD26) and C5ar1 (encoding CD88) can be used to properly distinguish mouse dendritic cells from monocyte/macrophages in atherosclerosis single-cell data, in line with recent reports utilizing this marker combination.^39,40 We furthermore propose marker combinations to identify specific macrophage cell states in atherosclerotic mouse vessels (Figure 4D).

3.6 Integrated analysis of human mononuclear phagocytes in atherosclerosis

To gain insight into the transcriptional state of macrophages in human atherosclerosis, we examined scRNA-seq data from Fernandez et al. investigating carotid endarterectomy specimens,¹⁵ and Wirka et al.¹⁶ analyzing coronary artery samples containing atherosclerotic lesions. While atherosclerotic lesions have site-specific particularities, major features of the inflammatory pathogenesis of atherosclerosis (intimal lipid accumulation, monocyte recruitment and macrophage accumulation, and macrophage foam cell formation³) are overall conserved across vascular sites. We, therefore, reasoned that integrating data from both vascular sites would allow us to obtain a global overview of the mononuclear phagocyte subsets and cell states that can be found in human atherosclerosis, to subsequently compare them to the mononuclear phagocytes we identified in experimental atherosclerosis.

Fernandez et al. and others have in detail analyzed the immune composition of carotid plaques using scRNA-seq.^15,17 However, such an analysis has not been performed on all cells from diseased coronary arteries sampled in Wirka et al.¹⁶ We identified immune cells in this data set, and excluded cells corresponding to phenotypically modulated vascular smooth muscle cells identified in Wirka et al.¹⁶ Clustering analysis identified various populations of adaptive and innate immune cells (see Supplementary material online, Figure S5A and B), including a major macrophage cluster enriched for TREM2, GPNMB, and other genes of the foamy/Trem2^hi signature (CTSD, SPP1, APOC1), and a cluster of macrophages expressing LYVE1 and MRC1 (see Supplementary material online, Figure S5A–C). Although a substantial inter-patient variability was noted, macrophages appeared as the predominant immune cell lineage in these coronary artery samples (see Supplementary material online, Figure S5D and E).

We then integrated mononuclear phagocyte data from both studies. A total of 2890 cells were included in the integrated analysis, and 10 clusters were recovered (Figure 5A, full marker gene lists in Supplementary material online, Excel File S3). Macrophages were identified by the expression of CD68, C1QA, and C5AR1 (Figure 5B). Differential gene expression analyses across clusters identified three major human (h) macrophage populations: hInflammatory-Mφ (CD74, HLA-DRB1), hFoamy-Mφ (APOC1, APOE, FABP5, FABP4), and hLYVE1-Mφ (LYVE1, LGMN, MARCO) (Figure5A and C). Additional minor populations were characterized by expression of C3, JUN and CCL4 (hC3-Mφ) and Type I IFN response macrophages (hIFNIC-Mφ cluster; ISG15, IFI6, MX1). Cells corresponding to monocytes (hMonocytes: VCAN, CD52, S100A8, S100A9, LYZ)⁴² were readily observed and highly enriched for pro-inflammatory markers (CXCL2, IL1B, CXCL8, TREM1). Other minor clusters of proliferating cells (hProlif cluster: TUBB, H2AFZ, STMN1) and B cells (hB_cell cluster: MZB1, JCHAIN) were also observed (Figure5A and C). We also recovered cDC1 (hcDC1: CLEC9A, IRF8, IDO1) and cDC2 (hcDC2: CLEC10A, FCER1A, CD1C)⁴³ populations (Figure5A and B).

Figure 5

Integrated scRNA-seq analysis of macrophages in human atherosclerosis. (A) UMAP representation and clustering analysis of integrated scRNA-seq gene expression data in 2890 human mononuclear phagocytes from atherosclerotic lesions; (B) expression of selected genes projected onto the UMAP plot; (C) heatmap of averaged gene expression (top 10 genes ordered by fold change) in the clusters (Inflamm = Inflammatory); (D) DotPlot showing the expression of transcripts enriched in human inflammatory-Mφ, foamy-Mφ and LYVE1-Mφ that are also enriched in their putative mouse counterparts (i.e. mouse inflammatory-Mφ, foamy/Trem2^hi-Mφ, and resident/resident-like-Mφ respectively). (E) Heatmap showing the mean top rank of transcription factors predicted by ChEA3 in each cluster (heatmap generated from the top 3 ranked transcription factors for each cluster).

To exclude the possibility that our analysis was biased by outlier samples, we analyzed cell distribution across individual patients. The number of analyzed cells per patient ranged from n = 2 (patient SYM2 from ref.¹⁵) to 1053 (Patient 3 from Wirka et al.¹⁶) (see Supplementary material online, FigureS6A and B). Proportions of cell clusters varied greatly across patients, which may reflect technical variability as well as the well-known heterogeneity of plaque composition and morphology in patients.^44,45,46 All the clusters were present in both vascular beds (see Supplementary material online, Figure S6A and B).

Gene expression patterns within the three main human macrophage populations (hInflammatory-Mφ, hFoamy-Mφ and hLYVE1-Mφ) suggested similarity to the major mouse aortic macrophage subsets we previously identified.¹¹ We performed differential gene expression analyses specifically within these three main human macrophage clusters and examined the overlap with marker genes of mouse aortic inflammatory, foamy/Trem2^hi and resident/resident-like macrophages. Relative to hFoamy-Mφ and hLYVE1-Mφ, hInflammatory-Mφ were enriched in MHCII encoding genes (CD74), inflammatory cytokines (CXCL2, CCL3, CCL4, IL1B), receptors (CLEC4E)⁴⁷ and transcriptional regulators (IER3, NFKBIA, NR4A2), similarly found in mouse inflammatory-Mφ. hFoamy-Mφ showed enrichment in markers characteristic of mouse foamy/Trem2^hi macrophages (TREM2, CD9, GPNMB, SPP1, CTSL, LIPA, ACP5), and hLYVE1-Mφ expressed genes associated with mouse resident/resident-like macrophages (LYVE1, CD163, SEPP1, FOLR2, F13A1, MRC1, VSIG4) (Figure 5D).

GO enrichment analyses revealed enrichment in similar biological processes, cellular components, or molecular functions (see Supplementary material online, Excel File S4). In particular, hFoamy-Mφ and mouse foamy/Trem2^hi macrophages were enriched for putative functions related to lipid metabolism (e.g. biological process GO terms: lipid catabolic process, lipid storage, cellular response to lipoprotein particle stimulus). Transcription factor enrichment analysis using ChEA3²⁷ identified NR1H3 in hFoamy-Mφ, similar to mouse foamy/Trem2^hi macrophages (Figure 5E).

3.7 Cross-species integration reveals conserved macrophage transcriptional states in mouse and human atherosclerosis

As an alternative strategy to compare macrophages in mouse and human atherosclerosis, we performed cross-species integration of scRNA-seq data. Mouse gene symbols were converted to their human homologues using the BioMart-Ensembl database. Mouse datasets were pre-processed to identify and extract cells corresponding to mononuclear phagocytes (macrophages, monocytes, dendritic cells), and integrated with the human data. As not all mouse genes have human homologues, this led to a decrease of 15.38 ± 1.73% in the number of detected genes per cell in the ‘converted’ mouse data used for integration (Figure 6A). After integration in Seurat v3 and dimensional reduction, clustering analysis generated 10 clusters (Figure 6B) with a clear mouse-to-human overlap (Figure 6C). By identifying and annotating cell clusters based on the characteristic gene expression patterns identified in the integrated human data and based on the mouse homologues in the integrated mouse data, we could readily recover integrated (int) Inflammatory macrophages (int-Inflammatory-Mφ: CD83, CCRL2, IFRD1), int-Res/Res-like-Mφ (LYVE1, FOLR2, F13A1), int-Foamy/TREM2^hi-Mφ (GPNMB, CD9, SPP1, FABP5), int-MoDC/cDC2 (NAPSA, KLRD1), int-Monocytes (PLAC8, MSRB1, THBS1), int-IFNIC-Mφ (IRF7, ISG15), int-FSCN1/CCR7-DCs, and int-XCR1/IRF8-cDC1 (Figure 6D). Expression of characteristic markers of the main vascular macrophage subsets showed consistent distribution across cell population in both species (Figure 6E). All clusters contained both mouse and human cells (Figure 6B). While such analysis might be biased due to overfitting during the data alignment,⁴⁸ this cross-species integration analysis substantiates the notion that macrophages with similar transcriptional states populate human and mouse atherosclerotic lesions.

Figure 6

Cross-species scRNA-seq integrated analysis of macrophages in murine and human atherosclerosis. (A) Median genes detected per single cell in the mouse dataset used for cross-species integration before and after conversion to human gene symbols; (B) UMAP plot and clustering analysis (with annotation) of the mouse/human integrated data; (C) cell species of origin projected onto the UMAP plot; (D) heatmap of enriched genes in the integrated cluster (top 5 ordered by fold change); and (E) expression of characteristic conserved marker transcripts of macrophage subsets projected onto the UMAP plot, split by species of origin.

Altogether, the results obtained from the analysis of human mononuclear phagocytes with a comparison of markers overlapping in mouse and humans (Figure 5), and from the cross-species data integration (Figure 6), suggest that cDC1, cDC2, mature DCs, and classical monocytes observed in mouse atherosclerotic aortas are also found within human lesions, and that macrophages that populate human lesions display transcriptional states resembling the major mouse aortic macrophage subsets (Resident/Resident-like, Inflammatory, Foamy/Trem2^hi, IFNIC).

4. Discussion

We here show that major macrophage and dendritic cell transcriptional states are conserved across widely employed mouse models of atherosclerosis, and that human lesions are populated by mononuclear phagocytes displaying transcriptional states resembling those found in mouse atherosclerosis. We provide two layers of evidence for this conclusion: (i) integrated analysis of human lesion scRNA-seq data from two independent studies recovered macrophage and dendritic cell states similar to those observed in mice and (ii) direct cross-species data integration that showed a strong overlap between mouse and human mononuclear phagocyte states.

Accumulation of macrophage foam cells in the intima is instrumental to lesion development. Macrophages reminiscent of the foamy/Trem2^hi macrophage state were observed across mouse models of atherosclerosis and in human lesions. Mononuclear phagocytes with a transcriptional signature similar to the foamy/Trem2^hi macrophage state found in atherosclerosis have been observed in mouse models of neurodegenerative disease (disease-associated microglia, DAM⁴⁹), demyelinating disease,³³ non-alcoholic steatohepatitis (NAM: NASH-associated macrophages),⁵⁰ metabolic-associated fatty liver disease,⁵¹ liver fibrosis (SAM: scar-associated macrophages),⁵² and diet-induced obesity (LAM: lipid-associated macrophages).⁵³ In the diseased liver and adipose tissue, features of this transcriptional state were similar in mice and humans, with many transcriptomic or cell surface markers such as TREM2, SPP1, or CD9^50,51,52 being conserved across species. The situation seems more complex in neurodegeneration-associated microglia as the characteristic DAM signature was not readily detected in single-nucleus RNA-seq (snRNA-seq) analysis of human neurodegenerative brain samples.⁵⁴ However, this observation might be due to technical issues, as characteristic genes of the DAM signature such as APOE or SPP1 were poorly detected in single-nucleus compared to single-cell RNA-seq.⁵⁵ This technical limitation, if further confirmed, needs to be taken into consideration in future snRNA-seq analyses of atherosclerotic samples.

LAM, DAM, and atherosclerosis-associated foamy/Trem2^hi macrophages share expression of a set of genes with enrichment in lipid metabolism pathways,⁵³ suggesting that similar mechanisms related to lipid loading may drive acquisition of this macrophage state. Nevertheless, further analyses will be required to determine the fine tissue- and species-specific particularities of these macrophages. Evidence from neurodegenerative disease models indicates that acquisition of the DAM state may depend on Apoe,⁵⁶ which raised the possibility that acquisition of the foamy/Trem2^hi macrophage state might differ between the most widely employed mouse models of atherosclerosis, i.e. Ldlr^−/− and Apoe^−/− mice. However, consistent with previous reports,^11,12 our integrated analysis indicates that Apoe expression appears dispensable for the acquisition of the foamy/Trem2^hi macrophage state in mouse arteries. Fully elucidating the impact of the mouse genotype on macrophage states will require more suitable experimental designs including biological replicates and differential gene expression in the absence of overt batch correction, e.g. by employing single-cell multiplexing technologies such as cell hashing⁵⁷ or MULTI-Seq.⁵⁸ Recently, we identified Trem2^hi macrophages in the ischaemic mouse heart sharing gene expression similarities with the LAM/DAM/foamy signature,³⁰ and two reports identified immunosuppressive macrophages enriched in Trem2 in tumour models,^59,60 indicating that part of this transcriptional signature may not only be related to pathological lipid loading but rather more generally induced in contexts of tissue damage. Our observation that macrophages with a Trem2^hi signature populate the aorta in the context of angiotensin-II-mediated inflammation corroborates this notion.⁶¹ Major transcriptional hubs involved in the regulation of lipid homeostasis such as the liver-X-receptor (LXR) pathway are activated also in response to efferocytosis of dead cells,⁶² raising the possibility that macrophages with high efferocytic activity may also acquire a foamy/Trem2^hi gene expression signature. Accordingly, transcription factor enrichment analysis in mouse and human cells pinpointed LXR as a potential regulator of the Trem2^hi signature, also consistent with a recent report indicating LXR as a driver of the Trem2^hi macrophage signature in atherosclerosis.⁶³

By analyzing a large number of mouse macrophages in our integrated approach, we could identify discrete subpopulations within foamy/Trem2^hi macrophages, including recently identified MAC-AIR.^7,36 We furthermore identified two subsets we termed Trem2^hi-Slamf9 and Trem2^hi-Gpnmb macrophages. Analysis of the gene expression patterns supports the notion that MAC-AIR present a gene expression signature specific to the vascular intimal niche that is acquired by infiltrating monocytes, and that lesion-associated foamy macrophages further attain the expression of a disease-specific gene signature.⁷ Importantly, canonical markers of this signature (Trem2, Spp1) appeared to have a low expression in MAC-AIR. While strong recovery of these three clusters in single-cell data from aortic intimal foamy cells¹² and recent fate-mapping and imaging analysis⁷ clearly suggest that these populations are found in the intima, their precise localization within the complex morphology of arteries remains to be further defined. Likewise, while MAC-AIR are self-renewing resident macrophages seeded during the perinatal period,^7,36 the ontogeny of Trem2^hi-Slamf9 and Trem2^hi-Gpnmb macrophages requires further investigation. However, extrapolation of decades of research on foamy macrophage accumulation in atherosclerosis clearly suggests a monocytic origin of these cells, which could be confirmed in future analyses using recently developed monocyte fate-mapping models based on Ms4a3⁶⁴ or Cxcr4.⁶⁵ Finally, whether similar subpopulations of foamy/Trem2^hi macrophages can be detected in human lesions will require sampling of a larger number of human lesional macrophages.

We observed macrophages corresponding to inflammatory macrophages both in mice and humans. Compared to other macrophages, these are characterized by the expression of genes encoding MHCII/HLA genes, and inflammatory cytokines (e.g. IL1B, inflammatory chemokines). Macrophages with a type I interferon signature (IFNIC)⁶⁶ were also observed in mice and humans.

In mice, resident/resident-like macrophages are defined by the expression of a characteristic set of genes (Timd4, Lyve1, Folr2, Sepp1, F13a1, Pf4, Cd163). In human lesional macrophages, we identified a subset of cells with a clearly overlapping transcriptional state. The exact localization of these cells in human diseased vessels remains to be elucidated. In mice, Lyve1⁺ macrophages are typically located in the adventitia.^67,68 We had previously observed LYVE1 protein expression in macrophages in carotid endarterectomy specimens by immunohistochemistry,¹¹ and we here detected LYVE1⁺ macrophages in scRNA-seq in carotid endarterectomy plaques. As the adventitia is not extracted during the carotid endarterectomy procedure, this indicates that cells with the Resident/Resident-like/LYVE1⁺ state may be found within intimal human lesions. However, in a recent report, Alsaigh et al.⁶⁹ performed single-cell analysis of atherosclerotic lesions from three patients, where cells from the atherosclerotic core and the proximal adjacent coronary artery were analyzed. Macrophages with the foam cell signature (APOE, APOC1) were observed in the atherosclerotic core, while LYVE1-enriched macrophages were in the proximal adjacent coronary artery.⁶⁹ The observation that LYVE1⁺ macrophages are enriched in non-diseased segments of endarterectomies when compared to the atherosclerotic plaque cores, clearly supports the notion that LYVE1⁺ macrophages represent a tissue-resident population both in humans and mice. Future investigations employing spatial transcriptomic methods⁷⁰ will help shed light on the precise localization of macrophage populations within diseased vessels.

While our work provides proof-of-concept of conserved transcriptional features of macrophage states across mouse and human atherosclerosis, further acquisition of high-quality data is clearly needed to fully elucidate the phenotypical landscape of human lesional macrophages, and cross-species characteristics of macrophage states. Altogether, our analysis encompassed cells from 11 patients across two vascular beds (carotid and coronary arteries), with the number of analyzed mononuclear phagocytes ranging from n = 2 cells to n = 1058 cells. While mice employed in experimental models of atherosclerosis are rather homogeneous (same genetic background, age, sex, absence of additional comorbidities), patients represent a highly heterogeneous population, and many factors (e.g. age, sex, comorbidities such as diabetes, etc.) are known to influence plaque immune cell composition.^23,71,72,73 Even in patients with a similar clinical profile, atherosclerotic lesions are highly heterogeneous in their morphology and cellular composition.^44,45,46 Hence, we propose that to identify macrophage transcriptional states correlated to clinically relevant events, analysis of a vast number of cells from many patients, and of cells from different vascular beds, will be necessary. Besides increasing statistical power to balance patient and plaque heterogeneity, technical issues remain an additional important hurdle for single-cell analyses of clinical samples.²⁰ The human and mouse studies included here all employed enzymatic digestion of tissues, which leads to cell recovery that may not represent the true composition of in vivo lesions as some cells (in particular macrophage foam cells) may be more difficult to recover compared to other cells, and which causes artificial gene expression patterns such as induction of immediate-early genes.^31,74 Single-nucleus RNA-seq, which bypasses the need for enzymatic tissue digestion, might be particularly suited to the analysis of human atherosclerosis, although poor detection of key mononuclear phagocyte genes may need to be carefully accounted for.⁵⁵ Recent efforts at establishing cell atlases of healthy or diseased tissues, for example in the heart⁷⁵ or in the lung of severe COVID-19 patients⁷⁶ have combined both single-nucleus and single-cell RNA-seq followed by data integration. Thus, the combination of single-cell and single-nucleus analysis to balance the strengths and weaknesses of each specific method appears as the current ‘gold standard’ to generate single-cell atlases of healthy and diseased tissues, and might be of use also in atherosclerosis research. In contrast to recent findings in carotid plaque samples where T cells appeared as the most abundant immune cell subset in human lesions,^15,17 macrophages predominated in coronary samples analyzed in Wirka et al.¹⁶ This could suggest site-dependent differences in plaque immune cell composition, but technical differences in cell extraction procedures (digestion enzymes used, flow cytometry labelling, and gating strategies) in these various studies might also underly these differences. While the two-layered approach we employed here (i.e. ‘side-by-side’ comparison and direct integration) to compare mouse and human data allows identifying similarities and conserved features between mouse and human mononuclear phagocytes subsets and states, it is not appropriate to investigate fine differences across species within defined cell subsets (e.g. to address the differences between human and mouse foamy TREM2⁺ macrophages). To our knowledge, this remains an unresolved computational challenge.

Altogether, our work provides proof-of-concept that macrophage transcriptomic states in atherosclerosis are conserved across mouse models of the disease and different vascular beds in humans. These findings are of importance for experimental investigations of macrophage function in atherosclerosis and their potential clinical translation. However, further research is critically needed to obtain a better understanding of macrophage populations and their states in human atherosclerosis, and of the fine differences between the human and mouse species that may bias pre-clinical investigations. Such research will benefit both from methodological advances as well as the analysis of substantially increased numbers of cells and patients.

Supplementary material

Supplementary material is available at Cardiovascular Research online.

Author contributions

All authors designed the study, analyzed and interpreted data, edited the manuscript. A.Z. and C.C. supervised the study, wrote the manuscript.

Funding

This work was supported by the Interdisciplinary Center for Clinical Research [Interdisziplinäres Zentrum für Klinische Forschung (IZKF)], University Hospital Würzburg (E-352 and A-384 to A.Z., E-353 to C.C.), the Deutsche Forschungsgemeinschaft (DFG; German Research Foundation, Projects 374031971-TRR 240, 324392634-TR221, ZE827/13-1, 14-1, project numbers 396923792 to A.Z., 432915089 to A.Z. and C.C., 458539578 and 471705758 to C.C.), the DFG SFB1525 project number 453989101 (projects A1 and B3 to A.Z., project PS2 to A.E.-S., F.E. and C.C., project A6 to C.C.), and the BMBF within the Comprehensive Heart Failure Centre Würzburg (BMBF 01EO1504 to C.C. and A.-E.S.). A.-E.S. is supported by the EMBO Young Investigator Program. This study was further supported by the SFB 1123 project A07 (to C.S.), as well as the DZHK (German Centre for Cardiovascular Research) and the BMBF (German Ministry of Education and Research) (grants 81Z0600204 to C.S. and 81X2600252 to T.W.). K.L. was supported by NIH P01 HL136275, R35 HL145241, and R01 HL146134.

Data availability

All the single-cell RNA-sequencing datasets used in this study are publicly available. For mouse datasets, see Table 1. For human datasets, please see the original publications.^15,16

Translational perspective

The inflammatory pathogenesis of atherosclerosis is widely studied in mouse models, but whether the immune cell subsets populating mouse lesions are relevant to human disease is unknown. By integrating single-cell RNA-seq datasets from atherosclerotic mouse aortas and human plaques and focusing on innate immune cells, we here identified that human lesions are populated by mononuclear phagocytes displaying transcriptional states resembling those found in mouse atherosclerosis. This provides evidence of conserved transcriptional features of the main macrophage and dendritic cell populations in mice and humans, a finding of importance for future experimental investigations and their relevance for clinical translation.

Author notes

Conflict of interest: None declared.