Cross-species single-cell RNA sequencing reveals divergent phenotypes and activation states of adaptive immunity in human carotid and experimental murine atherosclerosis

Abstract

Graphical Abstract

Open in new tabDownload slide

Time of primary review: 50 days

1. Introduction

Atherosclerosis is a chronic inflammatory disease of middle-to-large-sized arteries, leading to the build-up of vessel-occluding atherosclerotic plaques.¹ It is now acknowledged that atherosclerosis, as well as its clinical complications, myocardial infarction (MI) and stroke, is regulated by pro- and anti-inflammatory leucocytes and an ongoing autoimmune response involving autoreactive T cells and autoantibodies recognizing apolipoprotein B (ApoB).^2,3 Recent advancements in high-parametric immunophenotyping tools with single-cell RNA-sequencing (scRNA-seq) and mass cytometry (CyTOF) have enabled the detailed characterization of local immune cell populations in cardiovascular research.⁴ In human peripheral and coronary arterial blood, distinct leucocyte populations associate with the features of complicated atherosclerosis, including plaque erosion, and rupture.⁵ An scRNA-seq of atherosclerotic aortas in mice has recently uncovered an unexpected diversity of lesional leucocytes, almost reaching the level of cellular complexity observed in lymphoid organs.^6,7 A few pioneering studies indicate that the composition of carotid atherosclerotic plaques in humans may predict cerebral ischaemia.^7–9 It remains largely unclear whether the immune cell landscape of atherosclerotic lesions in mice resembles human disease, and whether the abundance of leucocyte phenotypes in the plaque predicts clinical complications of atherosclerosis beyond cerebral ischaemia. To address these questions, we performed an scRNA-seq of human carotid plaques and an integrative analysis with mouse data sets.

2. Methods

An expanded method section is available in the online-only supplement.

2.1 Clinical studies

We performed an scRNA-seq of isolated leucocytes from eight surgical specimens after carotid thrombendarteriectomy (TEA) at the Department for Vascular Surgery, Medical University of Graz, Austria (discovery cohort). Forty-three patients undergoing surgical carotid TEA were included in a validation cohort at the University Heart Centre Freiburg, Germany.

2.2 Leucocyte isolation

Atherosclerotic plaques were enzymatically digested, and leucocytes were analysed in flow cytometry or by scRNA-seq. Blood samples were collected in trisodium citrate.

2.3 Flow cytometry

Plaque and blood cell suspensions were incubated with fluorochrome-coupled antibodies against cell surface markers. After incubation, cells were washed, fixed with formaldehyde, and analysed on a flow cytometer (BD Fortessa, BD Biosciences, San Jose, CA, USA)). For cell sorting and scRNA-seq, cell suspensions were stained and collected in serum-free media containing at a concentration of 1 × 10⁶ cells/mL.

2.4 scRNA-seq

Single-cell and gel bead emulsion were generated from flow-sorted viable CD45⁺ cells on a Chromium Controller (10× Genomics, Pleasanton, CA, USA). Samples were sequenced on an Illumina NovaSeq sequencer using paired-end sequencing (BioCenter Core Facility, Vienna, Austria). Sequencing data were processed using the Cell Ranger Single Cell software suite. Raw sequencing reads were analysed with the Bioconductor analysis package SEURAT.

2.5 scRNA-seq integration

Data sets were merged with the SEURAT v3 variance-stabilizing transformation (VST)-integration workflow. During pre-processing, only leucocytes in clusters with a positive signal for PTPRC (CD45) were included. Uniform Manifold Approximation and Projection (UMAP) embedding and Louvain-based cluster detection were employed for subsequent analyses.

2.6 Statistics

Paired groups were compared with a paired Wilcoxon-matched pairs signed rank test or a paired Student’s t-test, and unpaired samples were compared with a Mann–Whitney U test. For multiple groups, a one-way analysis of variance (ANOVA) or the Kruskal–Wallis test with Dunnett’s or Tukey’s multiple correction was used as indicated. A P-value <0.05 was considered indicative of statistical significance. Data are presented as mean ± standard deviation (SD).

2.7 Ethical statement

Patients undergoing carotid TEA were prospectively enrolled in this study. All patients gave informed written consent on the day before surgery. Studies were performed according to the Declaration of Helsinki and approved by the local institutional review boards (EK31-367 ex 18/19, EK249/14, and EK22-1050-retro).

3. Results

3.1 The leucocyte repertoire in human atherosclerotic plaques is dominated by T cells

The composition of human atherosclerotic plaques has been established by immunohistochemistry and flow cytometry, but these strategies involve an inherently strong bias of marker pre-selection.⁷ To characterize lesional leucocytes in an unbiased approach, we performed an scRNA-seq of eight human carotid plaques after surgical carotid endarterectomy (Supplementary material online, Table S1). Our workflow included enzymatic tissue digestion, flow sorting, and drop sequencing (Figure 1A). We excluded CD45^neg non-leucocytes and dead cells tested positive for the viability dye (L/D) and high autofluorescence (Figure 1B). We retrieved a total of 34 904 leucocyte transcriptomes with an average of 1624 ± 285 unique genes expressed per cell (Supplementary material online, Table S2). Transcriptomes were visualized by UMAP. Louvain-based cluster detection revealed 22 distinct leucocyte and one non-leucocyte populations (Figure 1C) that were detectable across all patients (Supplementary material online, Figure S1). To interrogate cluster identities (IDs), we combined the manual curation of established cell-type-specific genes with semi-automated cluster annotation based on published cell atlases and one human scRNA-seq data set⁹ (Figure 1D–F, Supplementary material online, Data File S1, Figure S1D, and Table S3). Applying this strategy, we noticed six clusters containing CD4⁺ T-helper and two clusters of CD8A⁺ cytotoxic T cells that differed mostly in activation and memory status (Supplementary material online, Data Files S2 and S3) and one cluster with a CD3E⁺CD8A⁺NKG7⁺ natural killer (NK) T-cell signature. In the remaining 14 clusters, we identified C1QC⁺ and TREM2⁺ macrophages, classical and non-classical monocytes, conventional dendritic cells (cDC1/2), plasmacytoid DCs, MKI67⁺ proliferating cells, CD16⁺ and CD16^neg  NKG7⁺ NK cells, B cells, and innate lymphoid cells (ILCs). T cells (CD4⁺, CD8⁺, and NK T cells) dominated the repertoire of plaque leucocytes (65.5 ± 14.1%), while the myeloid compartment with macrophages, monocytes, and cDCs represented only 15.6 ± 16.2% of all leucocytes (Figure 1G and H, Supplementary material online, Table S4). We validated the high fraction of T cells by immunohistochemistry for CD3 (T cells) and CD68 (macrophages, monocytes, DCs) in sections of four additional human carotid atherosclerotic plaques (Supplementary material online, Figure S2). These data demonstrate that T cells represent the most frequent cell type among viable leucocytes in human atherosclerotic plaques.

Figure 1

The leucocyte repertoire in human atherosclerotic plaques is dominated by T cells. (A) Eight human carotid plaques after surgical endarterectomy were enzymatically digested. Single, viable CD45⁺ leucocytes were isolated by flow sorting, subjected to droplet scRNA-seq, and analysed by dimensionality reduction algorithms. (B) Gating strategy for cell isolation. (C) The resulting 34 904 transcriptomes, pooled from 8 patients, were visualized by UMAP. Louvain-based cluster detection revealed 23 distinct clusters. (D) Genes encoding for T cells (CD3E, CD8A, CD4), B cells (CD19), and myeloid cells (LYZ, CD68) are displayed on the UMAP plot. Log-normalized gene expression per cell (ranging from 0 to 6.25) was binned (≤1, 1–3, ≥3) and colour coded. (E) Log-normalized RNA expression of selected marker genes in each of the 23 clusters displayed as violin plots. (F) Differentially expressed genes in each cluster in comparison with all other clusters were calculated using a Wilcoxon rank sum test with a multiplicity-adjusted significance cut-off of P < 0.05. Top 3 up-regulated genes with the highest log₂(fold change) per cluster. (G, H) Cell-type percentages (% of all leucocyte transcriptomes per sample) displayed as a mean of eight patients and per patient. P1–P8 indicate the eight patients (n = 8). (H) The Y-axis indicates 100% of all leucocyte transcriptomes per sample. T_EM, effector/memory T cells; T_CM, central-memory T cells.

Open in new tabDownload slide

3.2 Integration of leucocyte transcriptomes from mouse and human atherosclerotic lesions

To compare immune cells residing in human carotid atherosclerotic plaques and atherosclerotic mouse aortas, human single-cell transcriptomes were integrated with seven mouse data sets generated by us and others, including scRNA-seq from atherosclerotic aortas from Apoe^−/−7 and Ldlr^−/−10 and adventitial preparations from Apoe^−/− and wild-type (WT) mice¹¹ (Supplementary material online, Table S5). The data sets were processed by VST, anchor detection, and integration (Figure 2A). The resulting 43 291 transcriptomes (Figure 2B) from 2 species, 3 genetic backgrounds, and different locations uniformly distributed in UMAP without batch effects (Figure 2C, Supplementary material online, Figure S3A). The expression of marker genes indicated principal haematopoietic lineages (Figure 2D). Louvain-based cluster detection set to a high resolution to detect even rare cell types retrieved 51 leucocyte ID of cell types and differentiation states (Supplementary material online, Figure S3B and Table S6) with distinct signatures (Supplementary material online, Figure S3C and Data File S4). While most IDs could be assigned to single haematopoietic lineages, we detected two clusters that expressed proliferation-associated genes (ID31, 40) and contained mixed cell types (Figure 2E and F). The addition of mouse leucocytes enabled the distinction of rare human leucocyte subsets, including a T-cell subset with a strong interferon (IFN) signature (ID27), g/d T cells (ID32), ILC-2 (ID44), ILC-3 (ID51), and mast cells (ID47). Grouping all IDs into principal meta-clusters (MCs) revealed a total of 14 leucocyte MCs with distinct gene signatures (Figure 2G and H, Supplementary material online, Data File S5): CD4⁺ T cells (CD3E⁺IL7R⁺), CD8⁺ T cells (CD3E⁺CD8A⁺NKG7^dim), NK cells (NKG7^high), NK T cells (CD3E⁺CD8A⁺NGK7^high), ILCs (AREG⁺GATA3⁺), B cells (CD79A⁺), neutrophils (CD14^highCD9⁺), monocyte-derived DCs (FLT3⁺ITGAX⁺), plasmacytoid DCs (FLT3⁺CD19^dim), monocytes (CD68^dimCD14⁺C1QC^neg), macrophages (CD14⁺CD68^highC1QC⁺), mast cells (TPSB2⁺, GATA2⁺), proliferating cells (MKI67⁺), and a cluster containing g/d, IFN⁺, and pre-apoptotic T cells (‘other T cells’). We defined MC-specific genes that were best conserved between the two species (Supplementary material online, Figure S4) and may serve as transcriptional markers for future studies.

Figure 2

Integration of leucocyte transcriptomes from mouse and human atherosclerotic lesions. (A) Leucocyte single-cell transcriptomes from 15 individual mouse (n = 7) and human (n = 8) data sets generated by scRNA-seq were integrated with a VST and anchor-detection and anchor-integration algorithm (SEURAT V3). The resulting 43 291 single-cell transcriptomes were projected by UMAP. Louvain-based cluster detection revealed 51 distinct leucocyte clusters. (B) MCs were constructed by grouping these 51 leucocyte phenotypes in clusters of one haematopoietic lineage. (C) Distribution of the transcriptomes across species, genetic mouse models, and locations. Only samples meeting the criteria indicated in each plot are shown. (D) Expression of the indicated lineage marker genes were displayed on the UMAP plot. Log-normalized gene expression per cell (ranging from 0 to 5.5) was binned (≤1, 1–3, ≥3) and colour coded. (E) The assignment of proliferation states (S-, G1-, and G2/M-cell cycle phases) based on a proliferation score/cell quantified by gene module enrichment. (F) Absolute number of unique transcripts expressed per cell (‘genes’), the percentage of mitochondrial genes (of all expressed transcripts, ‘% mito’), S and G2M scores indicative of cell-cycle phase were displayed as violin plots per MC. (G) Differentially expressed genes in each cluster in comparison with all other clusters were calculated using a Wilcoxon rank sum test with a multiplicity-adjusted significance cut-off of P < 0.05. (H) Expression of established lineage marker genes across all MCs as log-normalized RNA counts on a violin plot. % mito, % mitochondrial genes of all genes.

Open in new tabDownload slide

3.3 Existence of species-conserved leucocyte IDs with distinct transcriptional programmes

We detected a considerable divergence of the immune cell composition between murine and human tissues (Figure 3A, Supplementary material online, Figure S5 and Table S7). Because human carotid TEA samples contain only plaques and no adventitia, we selectively compared human plaques to plaque-containing tissue from Apoe^−/− and Ldlr^−/− mice on a chow and western diet, and high-fat diet (HFD), respectively. Human carotid plaques contained on average 67.8 ± 14.3% T cells (CD4⁺, CD8⁺, other, NK T cells), 15.4 ± 16.0% myeloid cells (macrophages, monocytes, cDCs, neutrophils), and 3.6 ± 2.0% B cells. Contrastingly, atherosclerotic mouse aortas contained 36.6 ± 9.4% T cells, 32.7 ± 20% myeloid cells, and 15.2 ± 9.3% B cells. Aortas from HFD-fed Ldlr^−/− mice showed the greatest expansion of myeloid cells (∼56%) and aortas from chow diet-fed Apoe^−/− mice the highest fraction of T cells (∼45%). Besides myeloid and B cells, ILCs (0.3 vs. 2.6%) and proliferating cells (2.2 vs. 6.7%) were over-represented, while CD4⁺ T cells (35.5 vs. 19.1%) were under-represented in murine atherosclerotic aortas (Supplementary material online, Table S7 and Figure S5). The adventitia, which may dilute plaque-derived cells in whole aortic preparations in murine tissue, showed a distinct immune cell composition with high fractions of B cells (∼35% in WT mice) and neutrophils (∼26% in Apoe^−/− mice), suggesting an overall numerically low contribution of adventitial cells in whole plaque preparations. A hierarchical clustering of MC abundance (Figure 3B) indicated that none of the genetic models, diets, and locations resembled human plaque composition. We also detected a substantial difference between species at the level of the 51 IDs. Eleven of the 21 most regulated IDs in human vs. mouse plaques (P < 0.05, fold change <0.5 or >2) could be assigned to either CD4⁺ or CD8⁺ T cells (Supplementary material online, Table S6 and Figure S6). We also observed a considerable high divergence among non-T cells from both species (Supplementary material online, Tables S6 and S8): for instance, mouse plaques mostly contained ILC-2 (ID44), while we detected a small population of ILC-3 (ID51) exclusively in human plaques. Human plaques contained fully differentiated pDC (CLEC4C⁺TCF4⁺TLR7⁺IL3RA⁺, ID25), while we found only a pre-pDC subset with a stronger residual B-cell signature in mice (ID30). Naïve B cells (ID15), resident/inflammatory macrophages (ID24), and a subset of classical monocytes (ID18) were more abundant in mouse plaques.

Figure 3

Existence of species-conserved leucocyte IDs with distinct transcriptional programmes. Leucocyte single-cell transcriptomes from mice and humans generated by scRNA-seq were integrated bioinformatically. (A) Fractions of MCs across samples. (B) Hierarchical clustering of colour-coded MC abundance (% of sample/row) on a linear scale. (C) The fraction of genes differentially expressed (DE) between mouse and human in each MC was calculated using a Wilcoxon Rank Sum test. DE genes with a multiplicity-adjusted P-value >0.05 were classified as not DE. Genes with P < 0.05 and a positive log₂(fold change) were considered significantly increased in human, and genes with P < 0.05 and a negative log₂(fold change) were considered significantly increased in mouse. (D) Fraction of genes regulated only in one (colour coded, legend as in A) or in more MCs (gray). Values are displayed as % of all genes DE between mouse and human. (E) DE genes between human plaques and mouse atherosclerotic aortas. Sharing of DE genes between samples shown as Circos plot, where each purple line represents a shared gene. The outer circle represents the individual data sets. The inner circle represents the fraction of shared (dark orange) and unique DE genes (light orange) per data set. (F) Venn diagram with absolute numbers of unique and shared genes between atherosclerotic mouse data sets. Pathway analysis (Metascape) of DE genes in human plaques and mouse atherosclerotic aortas. Up-regulated genes in each data set [P < 0.05 and log₂(fold change) > 1] served as input for pathway analysis in (G) CD4⁺ T cells, (H) macrophages, and (I) B cells. Significance of regulated pathways is displayed as colour-coded Z score. WD, western diet; chow, standard diet; HFD, high-fat diet; wt, wild type.

Open in new tabDownload slide

Next, we determined differentially expressed genes between both species. To account for non-orthologous genes in mice and humans, we removed genes coding for MHC-II, immunoglobulins, T-cell receptors (TCRs), mitochondrial genes, and genes that were not expressed in either of the two species. At the level of MCs across all data sets, we observed a low transcriptional variation between species in NK T cells (75.2% not differentially expressed genes) and ILCs (83%), while B-cell transcriptome (21.1%) and CD4⁺ T-cell transcriptome (22.5%) were less well conserved (Figure 3C). About 82.5% of all regulated transcripts were shared by more than one MC in a fine-grained analysis (Figure 3D), indicating the existence of general transcriptional programmes in each species, not confined to single MCs. Among all transcripts that were not conserved between both species across all MCs, we observed an enrichment of pathways linked to adaptive immune cell function, haemostasis, glucose metabolism, and hypoxia in humans (Supplementary material online, Data File S6). Next, we compared the non-conserved transcripts only in plaque-containing tissue: None of the genes up-regulated in human data sets was shared by one of the mouse data sets (Figure 3E), while we detected a separation of transcriptomes from Apoe^−/− and Ldlr^−/− mice (Figure 3F). Because transcriptional programmes may reflect compositional differences in the individual data sets, we next compared pathways at the level of MCs: Human CD4⁺ T cells expressed more transcripts linked to IFN, IL-10, and IL-2-signalling as well as to cell activation. On the contrary, CD4⁺ T cells from Ldlr^−/− mice were more strongly enriched for IL-18, IL-17, and pro-inflammatory signalling (Figure 3G, Supplementary material online, Data File S7). Likewise, we observed distinct transcriptional programmes selectively up-regulated in other human MCs, such as enhanced chemokine and IL-1b signalling in macrophages (Figure 3H), anti-viral and IFN-a responses in B cells (Figure 3I), and IFN signalling in CD8⁺ T cells (Supplementary material online, Figure S7). Compared with other genotype/diet combinations, Ldlr^−/− on HFD mice showed the strongest immune signature in mice. At the level of all MCs (Figure 3I) and CD4⁺ T cells (Supplementary material online, Table S9), human T cells more strongly expressed target genes of STAT3, a known transcription factor for T_H17/T_FH/T_reg development, HIF-1a, a hypoxia-inducible transcription factor, and of TWIST-1/2, which are known to inhibit T-cell function. Collectively, these data indicate that human atherosclerotic plaques are characterized by distinct immune cell populations and specific immunological transcriptional programmes.

3.4 Human T cells are enriched in states of dysfunctionality, activation, and exhaustion

Because T cells represented the largest cell population in human plaques, we revisited their heterogeneity in-depth. We re-clustered T cells from the CD4⁺-, CD8⁺-, and ‘other T-cell’-containing MCs (Figure 2) and down-sampled data sets to comparable cell numbers (∼2000 cells/data set). Louvain-based cluster detection revealed 15 T-cell clusters (TCLs) with distinct differentiation and activation states (Figure 4A). The patterns of CD8A/CD4 expression grouped TCLs in classical CD4⁺ (TCL1, 2, 8, 9, 10, 11), CD8⁺ (TCL3, 4, 5, 6, 13), and one cluster containing mixed CD4⁺/CD8⁺ T cells with a strong IFN signature (TCL12, IFN⁺; seeSupplementary material online, Data Files S8 and S9). In addition, we detected that TCL7, 14, and 15 expressed unusual combinations of CD4/CD8 at the single-cell level with CD4⁺CD8⁺ double positive (DPT, TCL7), double negative (DNT, TCL15), as well as T cells in transition from DNT to conventional T cells (TCL14; Figure 4B–D, Supplementary material online, Table S10). These highly expressed DNTT, TCF7, RAG1, and CCR9, genes indicating a premature developmental stage reminiscent of immature thymic CD4⁺CD8⁺ T cells (Supplementary material online, Data File S9). Consistent with earlier observations,^12,13 immature T cells were found at a high abundance in murine aortas (23.3 ± 5.8%) but were sparse in human plaques (3.3 ± 1.5%; Figure 4E, Supplementary material online, Figure S8A and B). Human plaques contained more CD4⁺ and IFN⁺ T cells than all mouse locations (Figure 4E). With the exception of FOXP3⁺IL2RA⁺ T_reg in TCL11, TCLs containing CD4⁺ and CD8⁺ cells clustered according to their memory but not to their differentiation status: expression of CCR7 (CCR7), CD62L (SELL), and CD44 (CD44) indicated naïve (T_naïve, CD44⁻SELL⁺CCR7⁺), effector memory (T_EM, CD44⁺SELL⁻CCR7⁻), and central-memory (T_CM, CD44⁺SELL⁺) T cells (Figure 4F, Supplementary material online, Figure S8C). In human plaques, we detected 2.6- to 4.1-fold more CD4⁺ T_CM/T_EM and CD8⁺ T_EM than in mouse plaque- and non-plaque-containing vascular tissue (Figure 4G). We observed a higher expression of markers of immediate activation and inflammation (CD40LG, CD69, TNF, NFKBIA, NR4A1) and survival factors (TNFAIP3, FOS, TUBA1B) in one CD4⁺ T_EM cluster (TCL8) in human samples suggesting highly activated and proliferation-prone T cells. Likewise, we observed a CD8⁺ T_EM-cell cluster that expressed the activation markers CD69 and KLRB1 (TCL13), reminiscent of recently discovered lesional mucosal-associated invariant T (MAIT) cells,^9,14 which was 17.9-fold more abundant in human plaques (Figure 4H, Supplementary material online, Data File S9). These data indicate that T cells in human plaques develop a considerable stronger T-cell memory than in mice.

Figure 4

Human lesional T cells are characterized by an activated memory phenotype. (A) T cells from mouse (n = 7) and human (n = 8) data sets were used as input for integration, UMAP, and Louvain-based cluster detection, which revealed 15 distinct TCLs. Marker gene expression overlaid on the UMAP plot. Log-normalized gene expression per cell (ranging from 0 to 6.25) was binned (≤1, 1–3, ≥3) and colour coded. (B) TCLs were grouped into four MCs displayed on the lower right UMAP plot. (C) DE genes in each MC in comparison with all other MCs were calculated using a Wilcoxon rank sum test with a multiplicity-adjusted significance cut-off of P < 0.05. (D) Percentages of cells expressing combinations of CD4 and CD8 in selected TCLs. Positive expression for either CD4 or CD8 was defined as a raw gene count ≥1. (E) Percentages of the four MCs separated by species (expressed as % of all cells). (F) Expression of T-cell memory marker genes coding for CD62L (SELL), CD44, CCR7, IL7-receptor (IL7R), KLRB1, and the T_reg transcription factor FoxP3 (FOXP3) displayed based on kernel density estimation by Nebulosa. The abundance of the indicated (G) memory TCLs and the (H) activated (act.) CD4⁺ and CD8⁺ TCLs expressed as % of all CD45⁺ in the individual input samples. (I) Enrichment Scores for distinct T_H transcriptomes: positive values indicate an enrichment, while negative values indicate an under-representation of the module. Quantification of the T_reg and T_H17 module in each of the six CD4⁺ TCLs displayed as bar charts. (J) Expression of T_H-marker genes for cytokines, transcription factors, and costimulatory molecules visualized in the six CD4⁺ TCLs as dot plot; fractions of CD4⁺ T cells (% all cells) with positive enrichment scores for T_reg, T_H1, and/or T_H17 transcriptomes expressed in (K) Venn diagram or (L) bar graph. (G, H, K, L) Samples split in plaque-containing (P) and non-plaque-containing (NP) mouse samples. (G, H, K, L) Statistical significance by multiplicity-adjusted one-way ANOVA with Dunnett’s post hoc test. P-values are indicated in the figure.

Open in new tabDownload slide

A recent scRNA-seq of mouse plaques indicated mixed T_H1, T_reg, and T_H17 T-helper cells, contradicting the classic immunological concept of distinct and exclusive T_H types at the single-cell level.³ Only TCL11 had a predominant signature of T_regs already at the level of TCLs and was enriched for transcriptional programmes of TCR signalling, T-cell activation and differentiation in humans, while murine cells in TCL11 had an additional T_H17 signature (Supplementary material online, Figure S9). To interrogate T_H differentiation in other TCLs, we applied reported T_H-gene modules to the six CD4⁺ clusters (Figure 4I, Supplementary material online, Figure S10A) and found mixed signals for T_H1 and T_H17 pathways (Supplementary material online, Figure S10B). When comparing the enrichment of T_H modules and T_H marker genes (Supplementary material online, Data File S10) between both species, we detected a stronger signal for T_H1 in T_EM (TCL1) from mouse plaques (Figure 4J, Supplementary material online, Figure S10C). In addition, T_FH were enriched in the adventitia of mice and less frequent in plaque-containing murine tissue and human carotid plaques (Supplementary material online, Figure S10C). We recently proposed that multi-lineage committed plaque T cells evolve from FoxP3⁺ T_regs that can transit from T_regs into other phenotypes.^1,12 To quantify multi-lineage commitment, we assigned all CD4⁺ T cells to groups of cells expressing only modules of one or several T_H phenotypes (Figure 4K and L, Supplementary material online, Figure S10D). The proportion of T cells enriched in only one gene module was higher in humans (∼60%) than in mice (∼34%). Human plaques contained more T_H17 and T_reg and less triple positive T_reg/T_H17/T_H1 cells (Figure 4L) than plaque-containing mouse tissues, indicating a lower degree of T_reg plasticity or impaired proliferation of ex-T_regs in humans. We observed a relative enrichment of a gene module of CD4⁺ T-cell dysfunction/exhaustion, which typically occurs in chronic infection and counteracts cellular proliferation,¹⁵ in T_EM and multi-lineage committed human CD4⁺ T cells (Supplementary material online, Figure S11). Likewise, CD8⁺ TCLs from naïve TCLs showed a lower enrichment of cytotoxicity and secretory gene modules than activated, T_CM or T_EM CD8⁺ T cells. The majority of CD8⁺ T_EM from human plaques was enriched in a CD8⁺ dysfunctionality module (Supplementary material online, Figure S12). Because dysfunctionality/exhaustion typically occurs after frequent TCR engagement, we interrogated gene expression of T-cell activation markers across all CD4⁺ and CD8⁺ TCLs (Supplementary material online, Figure S13). Indeed, we observed that expression of CD69, an immediate marker of antigen recognition and of genes indicating TCR-signalling events (NFATC, ZAP70, FYN) was higher in human than in mouse T cells. To explore a potential relationship of T-cell phenotypes with ageing, we mapped a recently published data set of aortic leucocytes from 22-month-old Ldlr^−/− mice on a chow diet¹⁶ on our integrated data sets. We found an increased fraction of B cells and neutrophils (Supplementary material online, Figure S14A and B), indicative of a relevant adventitial proportion in these aged mice consistent with earlier observations.¹⁷ While CD8⁺, but not CD4⁺ T cells, showed a higher expression of exhaustion-associated transcripts than observed in data sets from younger mice (Supplementary material online, Figure S14C), we observed higher expression of TCR-signalling transcripts in cells from both, CD4⁺ and CD8⁺ T cells, in aged mice—albeit not reaching the level observed in human plaques (Supplementary material online, Figure S14D). The numerically smaller compartment of CD4⁺ T cells (Supplementary material online, Figure S14E and F) had a T_H-lineage profile similar to younger mice with the exception of a higher T_FH commitment (Supplementary material online, Figure S14G). Contrasting to stronger TCR signalling, we did not detect a higher abundance of memory cells than in younger atherosclerotic mice (Supplementary material online, Figure S14H) suggesting that ageing is a factor that fails to explain the full extent of T-cell activation in human plaques. Overall, our analysis proposes a higher degree of activation and dysfunction in human compared with T cells in murine atherosclerotic aortas.

3.5 Protein surface marker–defined immune cell landscape of human carotid plaques

To validate the immune cell composition in human plaques in an independent cohort, we characterized 43 additional carotid plaques from surgical endarterectomy (Table 1) with a flow cytometric multi-colour panel capable of identifying principal haematopoietic lineages (Supplementary material online, Table S11). We defined leucocyte clusters by t-distributed stochastic neighbour embedding (t-SNE) and gated these by surface marker expression profiles (Figure 5A and B). This strategy revealed 21 leucocyte clusters (CLs; Figure 5C). Overall, we detected nine TCLs including CD4⁺ (CL7, 20) and CD8⁺ T cells (CL3, 19), one immature CD4 ⁺ CD8⁺ (CL18), one CD4^negCD8^neg (CL1), and two NK TCLs (CL4, 13). In addition, we noticed one CD19⁺ B-cell cluster (CL2), three NK-cell clusters (CL5, 10, 14) and nine myeloid clusters. On average, T cells accounted for 67.5 ± 17% of all lesional cells, while myeloid cells represented only 10 ± 9% of all leucocytes (Figure 5D, Supplementary material online, Table S12 and Figure S15A). These data confirmed the higher abundance of T cells in human plaques, while a proportion of cells (10.3%) remained not assignable to one of the one leucocyte lineages (‘undefined’). We next asked whether the high abundance of T cells in plaques may reflect the composition of blood leucocytes or a relative over-representation in the plaque. Therefore, we applied the same strategy to matched peripheral arterial blood samples in a subgroup of patients. Indeed, we detected an enrichment of CD4⁺ and CD8⁺ TCLs in the plaque (1.6- to 4.9-fold), but not of other haematopoietic lineages (Figure 5E, Supplementary material online, Table S13 and Figure S15B). Expression of the activation markers CCR7, CD45RO, CD45RA, and CD69 (Supplementary material online, Figure S16A, Tables S14 and S15) indicated that the majority of T cells in the plaque had a T_EM phenotype (Supplementary material online, Figure S16B and C). T_EM cells and activated CD69⁺ T_EM cells were more abundant in the plaque than in matched blood samples (Supplementary material online, Figure S16D and Figure 5F). In addition, plaque-derived T_H cells expressed higher levels of the activation markers CD40L, CD25, and the exhaustion marker OX40 than their circulating counterparts (Supplementary material online, Figure S17). This is consistent with our observation that intermediate CD4- and CD8-expressing cells (CL19/20), which likely represent activated T cells,¹⁸ were highly enriched in plaques. These findings demonstrate that most lesional T cells have a memory phenotype with hallmarks of cellular activation.

Figure 5

Protein surface marker–defined immune cell landscape of human carotid plaques. Leucocytes were isolated from human carotid plaques after surgical dissection by enzymatic digestion and analysed by flow cytometry with a (A–E) pan-leucocyte (n = 43) or a (F) T-cell activation (n = 33) marker panel. Unsupervised cluster detection by tSNE of plaque leucocytes (n = 43) and blood (n = 19). (B) Gating strategy for the identification of tSNE-retrieved 21 distinct leucocyte clusters. Hierarchical column and row clustering of mean fluorescence intensity expressed as Z score. (C) Z scores for each value were calculated by subtracting the column mean and dividing by the column SD. The abundance of the 21 leucocyte clusters in carotid plaques and blood shown as the mean of each percentage (% of all leucocytes). (D) Cell-type annotation based on indicated marker expression. Volcano plot of the log₂(fold change) of cluster percentages in blood vs. plaque (n = 21). (E) Significances were calculated by an unadjusted Wilcoxon-matched pairs signed rank test. Paired percentages of T-cell activation states in blood vs. plaque from the same patient (n = 21). (F) Statistical significance was calculated by a Wilcoxon-matched pairs signed rank test. Numeric, unadjusted P-values are indicated in the figure. MMA, mature-macrophage antigen; CL, cluster according to the nomenclature in D.

Open in new tabDownload slide

3.6 The abundance of leucocyte phenotypes in the carotid plaque associates with complicated and generalized atherosclerotic disease

To assess whether immune cell populations in carotid plaques associate with plaque vulnerability, we evaluated clinical complications of carotid atherosclerosis (stroke or a transient-ischaemic attack) within 6 months before surgery in our validation cohort. While we could not confirm a significant association with lesional CD4⁺ or CD8⁺ T cells as previously suggested,⁸ we found that NK T cells (CL4,13) and B cells (CL2) were more abundant in patients with an ischaemic event compared with asymptomatic patients. Among T cells, only the population of activated CD25⁺CD8⁺ T cells was slightly increased in patients with ischaemic events (Figure 6A and B). In addition, we observed that the abundance of NK cells (CL10, 14) was higher in the blood of patients with an ischaemic event (Supplementary material online, Figure S18). Next, we tested the association of pre-existing coronary artery disease (CAD) with the abundance of leucocyte phenotypes in blood and carotid plaques. In patients with CAD, more CD8⁺, CD8⁺ T_EM, and CD4⁺ T cells expressed the activation marker CD69, which was associated with dysregulated T cells in the integrative scRNA-seq analysis. Conversely, the fraction of non-activated CD8⁺ T cells (CD69^neg) and atheroprotective CD25 ⁺ CD4⁺ T_regs was larger in carotid plaques from patients without CAD. These results indicate that the activation status of T cells in carotid plaques associates with an increased likelihood of generalized atherosclerosis.

Figure 6

The abundance of carotid plaque leucocytes associates with clinical complications and generalized atherosclerotic disease. Leucocytes from human carotid plaques isolated by enzymatic digestion and paired blood samples were analysed by flow cytometry with a pan-leucocyte or a T-cell activation marker panel. (B, D) Individual regulation or (A, C) volcano plots of the log₂(fold change) of cluster abundance in patients with (A, B) an ischaemic cerebral event 6 months prior to surgery (I.E.) and no ischaemic cerebral event (no I.E.) or in patients with (C, D) simultaneous CAD. The number of plaque samples included for I.E. vs. no I.E. analysed with pan-leucocyte panel: I.E. (n = 18), no I.E. (n = 25); and T-cell activation panel: I.E. (n = 20), no I.E. (n = 13). The number of plaque samples included for CAD vs. no CAD analysed with pan-leucocyte panel: CAD (n = 27), no CAD (n = 16); and with T-cell activation panel: CAD (n = 11), no CAD (n = 22). Statistical significance was assessed by Mann–Whitney U test. (B, D) Numeric, unadjusted P-values are indicated in the figure. Bar graphs show the mean. CL, cluster according to the nomenclature in Figure 5D.

Open in new tabDownload slide

4. Discussion

While Ldlr- and Apoe-deficient C57BL/6 mice are frequently used in pre-clinical atherosclerosis research, their translational value remains disputed:¹² both models exhibit dramatically higher LDL-C levels than in humans. Apoe^−/− mice develop organ-wide autoimmunity.¹ In contrast, human atherosclerosis is promoted by hypertension, diabetes, and other metabolic disorders not existent in standard mouse models. Humans, but not mice, develop spontaneous rupture of atherosclerotic plaques and MI. In addition, human atherosclerosis has mostly been studied in carotid plaques with only two available reports of scRNA-seq in coronary arteries.^19,20 These discrepancies render the translation from mouse models to human disease challenging. Novel immunophenotyping by scRNA-seq and mass cytometry has pointed out that human atherosclerotic plaques are mainly populated by T cells that associate with plaque vulnerability.^8,21 How these findings relate to mouse atherosclerosis has remained unclear. Here, we provide a simultaneous analysis of murine and human atherosclerotic tissues.

Our strategy affords several advantages. First, it allows identifying species-related differences in the composition of atherosclerotic plaques. We validate earlier findings⁸ that the leucocyte repertoire in human plaques is dominated by T cells and contradict the classic perception of atherosclerosis as a macrophage-dominated disease. Our analysis reveals a T-cell fraction (CD4⁺, CD8⁺, other T cells, NK T cells) of 65.–67.8% in humans (Supplementary material online, Figure S19), while atherosclerotic mouse aortas contained on average 36.6 ± 9.4% T cells. These findings are in line with previous report from Fernandez et al. (68.6%)⁸ and Dib et al. (52%).⁹ The low macrophage counts in our and other data sets (10–15.6%, Supplementary material online, Table S16) might be explained by a low isolation efficiency during digestion and cell sorting with a stringent autofluorescence and size gate.⁷ Yet, histologic evidence from >30 years ago already suggested a higher abundance of T cells than of macrophages in human plaques.²² We validate these findings by immunohistochemistry employing CD3 and the myeloid/phagocyte marker CD68. Our findings are consistent with protein- and transcript-based cell-type identification in other scRNA-seq studies of human plaques^8,9,21 that consistently report a T-cell content in the range of ∼52–69%. Notably, we present the so far largest scRNA-seq data set of human atherosclerotic plaque leucocytes (with an average of 4363 leucocytes/plaque) that exceeds previous studies from Fernandez et al. (1194/plaque),⁸ Depuydt et al. (147/plaque),²¹ and Dib et al. (3490/plaque)⁹ (Supplementary material online, Table S17).

Second, our data indicate that human atherosclerosis—in contrast to mice—is shaped by transcriptional programmes comprising the biological response to glucose, hypoxia, coagulation, and infection/autoimmunity. These programmes match the pathophysiological characteristics of human atherosclerosis: enhanced hypoxia within larger atherosclerotic plaque and necrotic cores in humans, increased platelet reactivity, and a strong dependency on metabolic risk factors such as diabetes. Mouse plaques were highly enriched for pathways of myeloid-driven inflammation, general traits of cell activation, and chemokine signalling. Along with these findings, our data indicate a substantial transcriptional difference between Ldlr^−/− and Apoe^−/− mouse models, while diet regimens accounted for only slight differences in the transcriptional landscape. These transcriptional differences were not consistent across all cellular phenotypes and lineages: while cells of the adaptive immune system showed an unexpected transcriptional divergence, others, including ILCs, NK T cells, and macrophages were transcriptionally better conserved. Consistently, one recent report showed an effective alignment of myeloid cell types in human and murine atherosclerotic plaques.²³

Third, the combination of mouse and human data sets provided a substantial technical advantage over testing either of the species alone: because mapping, integration, and cluster detection by deep learning algorithms require sufficiently high numbers of cells with the same IDs, the scRNA-seq of human lesional leucocytes alone failed to detect rare IDs in human data sets. This is evidenced by the selective analysis of our human data sets (Figure 1) and by recent studies,^8,21 where a fine-grained cell-type resolution was only observed within the large T-cell population but not among rare cell types. For instance, mast cells, ILC-3, pre-pDCs, naïve B cells, and immature-like T cells remained undetected in single-species analyses. Our approach with an intentionally high resolution for improved detection of rare cell types detected 51 IDs, which in large parts could be tracked back to known leucocyte subtypes. Our flow cytometry–based approach and a recent study using antibody-epitope mapping by CyTOF underrated the actual heterogeneity⁸ of some cell types, including CD4⁺, NK T cells, and NK cells. Because our strategy has a high sensitivity, at which the positive matching of two populations strongly suggests a transcriptional relation, but a low specificity, an exact analysis of non-integrated populations will require future investigations.

Fourth, our study addresses whether cell types in the plaque associate with clinical outcomes, as recently demonstrated for peripheral leucocyte numbers and the risk of cerebrovascular events.²⁴ We used a protein-based flow cytometry to validate the principal composition of human plaques. We demonstrate that lesional CD4⁺ T_H cells and cytotoxic CD8⁺ T cells are enriched in the human plaque compared with peripheral blood of the same patient. Our data indicate that the higher fraction of circulating memory T cells in humans (compared with laboratory mice)²⁵ alone does not suffice to explain the dominance of T cells in plaques. In addition, we demonstrate that plaques containing higher numbers of some leucocyte subsets are more prone to cause clinical complications. While Fernandez et al.⁸ suggested that numbers of CD4⁺ cells in the plaque associate with a recent cerebrovascular event, we could not confirm such an association. Instead, we found that NK T cells and B cells were enriched in plaques from patients with a recent ischaemic event. Although B cells are less abundant in human plaques, clinical association studies have identified distinct circulating B-cell subsets that correlated positively or negatively with the incidence of stroke.²⁶ These observations are consistent with the notion that protective and pathogenic B-cell subsets exist.¹ More strikingly, we observed an association of CD69⁺CD4⁺ or CD8⁺ T_EM cells in the carotid plaque with coronary atherosclerotic disease, suggesting that plaque inflammation and activation of T cells represent a systemic, rather than a local event. CD69 is up-regulated in activated T cells and may serve as an indirect marker of antigen exposure.²⁷ Additionally, we confirmed previous reports demonstrating an enrichment of exhaustion and dysfunctionality pathways in human T cells.⁸ These findings together with recent parallel TCR and scRNA-seq of human plaque T cells¹⁴ indicate a higher degree of sustained T-cell engagement in humans than in mice and propose that fraction of human plaque T cells may be reactive against foreign or self-antigens.^3,19 The weaker engagement of T cells in murine samples was consistent across adventitial and plaque-containing data sets and, therefore, could not be explained by an anatomical bias. In contrast to previous assumptions,³ multi-lineage committed T cells that appear to be transitioning from antigen-specific T_regs to composite T_H1/T_H17 cells were less abundant in human plaques. This is consistent with our observation that transcriptomes in human T_regs were enriched for pathways linked to T-cell activation, TCR signalling, and differentiation, while murine T_regs showed a higher engagement of pathways involved in T_H17 differentiation and TNF-a signalling. Likewise, we observed stronger expression of T_reg-defining genes, such as FOXP3, IL10, CTLA4, and TIGIT, in humans than in mice. These findings may indicate that antigen recognition by T_regs is a more likely feature of human than murine atherosclerosis, albeit antigen-specific T_regs that recognize ApoB-100 have been described in both species.^3,28 T_reg instability on the other hand may represent a specific feature of highly inflamed murine plaques rather than of antigen-specific immunity alone.¹ Overall, our findings suggest that autoimmunity may be more important in human atherosclerosis and likely serves as predictive marker of clinical disease severity.^3,28

This study, along with other scRNA-seq studies, has technical limitations: first, transcriptional signatures do not necessarily predict protein-defined phenotypes. To overcome this problem, Fernandez et al.⁸ presented a combination of protein phenotyping and scRNA-seq, which revealed a consistent overlap of the protein- and transcriptome-based approaches. We confirm that the abundance of flow-cytometry- and scRNA-seq-defined phenotypes correlates well among our analyses and the study of Fernandez et al.,⁸ particularly for CD4⁺, CD8⁺, and B cells (Supplementary material online, Figure S19). Notably, we have employed a tSNE-based clustering of flow cytometry data to reduce cell types not annotated by traditional gating strategies. This is exemplified by antibody-based immunophenotyping that underestimates myeloid cell abundance and retains 8–10% of cells that cannot be assigned to known haematopoietic lineages (Supplementary material online, Table S16). This is also exemplified by Fernandez et al.,⁸ which did not detect ILCs, mast cells, and proliferating cells in CyTOF of human atherosclerotic plaques. Second, scRNA-seq-defined relative abundance must be considered carefully. Until now, scRNA-seq is challenging and most experimental data rely on pooled cell samples with low technical and biological replication. Sample preparation, particularly from mouse aortas, can introduce a major technical bias and, therefore, it remains unclear whether observed differences are affected by different cell isolation strategies. For instance, gating strategies in flow-based cell sorting must be designed carefully to exclude dead cells and to retain cell types, such as macrophages, with an inherently higher signal for live/dead dyes and autofluorescence. Adventitial tissue is not present in human carotid endarterectomy samples, rendering the direct comparison of mouse aortic tissue per se difficult. To overcome this limitation, we present separate data sets on adventitial tissue in mice and selectively compare plaque- and non-plaque-containing mouse tissue. While we cannot enumerate the proportion of adventitial cells in whole aortic preparations by our approach, the high abundance of B cells, neutrophils, and T_FH signatures in adventitial, but not in other murine data sets, suggests that adventitial contamination is overall low. Third, some of the differences between mice and humans may reflect stage-dependent effects. A precise age match of humans (in our study, the median age was 69 years) would require an age of 72–104 weeks in mice.²⁹ At the age of 22 months (∼88 weeks), we observed a stronger expression of TCR-signalling transcripts in Ldlr^−/− mice, which confirms that ageing predisposes for auto-reactivity.³⁰ Therefore, our results partially render the translatability of the usually accepted age of atherosclerotic mice (24–28 weeks) questionable. Fourth, it remains unknown to which extent different vascular beds shape the cellular composition of atherosclerotic plaque. For instance, only two studies have investigated atherosclerotic coronary arteries and obtained mixed results: an analysis of four human right coronary arteries suggested a higher macrophage content than observed in carotid plaques.²⁰ Another study with 35 donors found clonal expansion and a high fraction of memory T cells, which is consistent with the findings in our study.¹⁹ Future studies will have to clarify these location-specific effects.

Taken together, we present a cross-species atlas of mouse and human leucocytes in atherosclerosis that defines cell types and transcriptional programmes. We identify several cellular IDs that are unique to human disease and associated with plaque vulnerability and systemic atherosclerosis in humans. While previous studies had already detected activated T cells in human atherosclerotic plaques,⁸ we show that T-cell activation and exhaustion need to be considered as a more likely characteristic of human, not murine, atherosclerosis, likely as a result of antigen-specific immunity³ and clonal expansion of autoreactive T cells in the human plaque.^14,19 Adaptive immune cells may therefore represent future cellular targets for cardiovascular immunotherapy or atheroprotective vaccination.³¹ The definitive function of the findings made in this explorative and hypothesis-generating study will have to be corroborated by future studies.

Supplementary material

Supplementary material is available at Cardiovascular Research online.

Authors’ contributions

D.Wo. and A.Zi. designed the study. Cell isolation and single-cell sequencing were conducted by H.H. and N.A.M. X.S. analysed flow cytometry data. K.P. helped in the preparation of leucocytes for single-cell sequencing. Plaque samples for single-cell sequencing were provided by T.C. and W.O. H.H., A.L., S.H., L.S.M., T.B.-N.M., T.A., X.L., T.M., and M.C.G. performed flow cytometry and tissue digestion. M.C.F., J.M., P.K., and H.H. performed, analysed, and interpreted immunohistochemical staining. The analysis of scRNA-seq data was conducted by D.Wo. and H.H. The patients were enrolled by A.L., I.H., T.R., and N.V. scRNA-seq data sets were provided and discussed with A.C.F. and C.v.z.M. H.W., H.B., T.H., K.B., P.S., S.P., T.G., C.v.z.M., A.Z., P.M.S., D.W., A.Zi., and D.Wo. analysed the data and wrote and discussed the manuscript.

Acknowledgements

The authors thank Qingbo Xu (Queen Mary University of London, Great Britain) for providing the mouse scRNA-seq data sets. The authors thank the SCI-MED imaging facility at the Institute for Experimental Cardiovascular Medicine in Freiburg for granting access to the slide scanner.

Funding

This study was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation), SFB1425, project #422681845. A.Zi. received funding from DFG (ZI 743/8-1). This project received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement no. 853425). M.C.G. was funded by the IMM-PACT programme for Clinician Scientists, Department of Medicine II, Medical Center—University of Freiburg and Faculty of Medicine, University of Freiburg, funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)—413517907.

Data availability

All sequencing data are available via Gene Expression Omnibus (GSE245373). Custom R scripts used for the in silico scRNA-seq analysis are available in a GitHub Repository (https://github.com/haukeh90/cross_species_scRNAseq_in_atherosclerosis).

References

Author notes