Microanatomy of the Human Atherosclerotic Plaque by Single-Cell Transcriptomics

Supplemental Digital Content is available in the text.

pathophysiology of atherogenesis is lacking. Moreover, genome-wide association studies (GWAS) have identified many loci associated with increased risk for CVD, but the translation of these findings into new therapies 2 has been hampered by the lack of information on specific cell communities in atherosclerotic plaques and the cell-specific expression patterns of druggable candidate genes at the site of disease. Recently, the immune cell composition of murine and human aortic atherosclerotic plaques has been described using cytometry by time of flight and single-cell RNA sequencing (scRNAseq). [3][4][5][6][7] Yet, the full cellular composition of human carotid plaques, including nonimmune cells, remains elusive. Therefore, we performed scRNA-seq and single-cell ATAC sequencing (scATAC-seq) on advanced human atherosclerotic plaques obtained during carotid endarterectomy and report a comprehensive overview of the various cell types in plaques and their activation status, which reveals an active, ongoing inflammation and multiple cellular interactions as well as cellular plasticity with respect to endothelial cells (EC) and macrophages. In addition, we identified cell type-specific expression of GWAS risk loci for CVD.

METHODS
In silico data analysis was performed using custom R Scripts (R version 3.5.3) designed especially for this research or based on the recommended pipelines from the preexisting packages listed in the individual segments above. R

Novelty and Significance
What Is Known?
• Atherosclerotic lesions show a complex cellular composition that has mainly been studied using selected marker molecules. • The benefit of using single-cell RNA sequencing as unbiased method has been shown for immune cells in both murine and human atherosclerosis.
What New Information Does This Article Contribute?
• Single-cell RNA sequencing of a broad cohort of human carotid plaques now provides a detailed cellular atlas of the various cell types and their phenotypes, including different clusters of endothelial and smooth muscle cells. • Chromatin accessibility of macrophages and T cells is mapped at a single-cell level and identifies relevant transcription factor binding sites. • Mapping of cardiovascular susceptibility genes identified by genome-wide association studies to cellular subsets identifies potential cell-specific targets.
It is important to determine the exact cell (sub)types and their interactions at play in atherosclerosis to devise novel therapeutic strategies. Here, we describe the total cellular composition of atherosclerotic plaques taken from carotid arteries of a broad cohort of patients. Our data suggest that the main immune cell subset consists of T cells, which can be subdivided by activation status. Macrophages are found in distinct populations with diverse activation patterns, inflammatory status, and foam cell characteristics. We shed light on plaque endothelial and smooth muscle cell gene expression and show cell clusters with gene expression patterns pointing towards characteristics of endothelial to mesenchymal transition. To further investigate the dynamic intraplaque niche, we assessed ligand-receptor interactions driving our cell communities and investigated potential transcription factor activity underlying myeloid and T-cell populations in the plaque by studying chromatin accessibility at the single-cell level. Finally, we identified cell types enriched for cardiovascular susceptibility genes by integrating available genome-wide association studies data. Together, our data provide an in-depth map of the human atherosclerotic plaque and give valuable insights into cell types, pathways, and genes that are relevant for future research aiming at the development of novel therapeutic strategies.
MicroanatomyHumanPlaque_scRNAseq]. Other data is available from the corresponding authors upon reasonable request. Please see the Data Supplement for detailed methods.

Single-Cell RNA Sequencing Identifies 14 Distinct Cell Populations in Human Atherosclerotic Plaques
To examine the transcriptome of human atherosclerotic plaques, carotid endarterectomy tissue from 18 patients (77% male sex) was enzymatically digested, viable nucleated cells were isolated by fluorescenceactivated cell sorting (FACS; Figure 1A, Figure IA, Table I in the Data Supplement), and scRNA-seq libraries were prepared. After filtering cells based on the number of reported genes (see Methods in the Data Supplement), we applied unbiased clustering on 3282 cells, identifying 14 cell populations ( Figure 1B and 1C, Table II in the Data Supplement). Correlation of our scRNA-seq data with bulk RNA-seq data ( Figure IB in the Data Supplement) and examining interpatient variation of cluster distribution ( Figure IC in the Data Supplement) and size ( Figure ID in the Data Supplement) confirmed uniformity of the data except for patient 1.

ECs Exhibited a Gene Expression Profile Indicative of Activation and Potential Transdifferentiation
ECs were represented by cluster 9; expressing COL4A1, (collagen type IV alpha 1 chain) COL4A2, SPARCL1, and PLVAP, and cluster 10; expressing MPZL2, SULF1, VWF, and EDN1 ( Figure 1B and 1C, Table II in the Data Supplement). Isolating and reclustering these clusters revealed 4 distinct subclasses (E.0-E.3, E indicates EC, Figure 2A, Table II in the Data Supplement). We could assign EC phenotypes to the subclasses by assessing marker genes ( Figure 2B). E.0, E.1, and E.2 displayed classical endothelial markers CD34 and PECAM1, and the vascular endothelial marker TIE1. E.0 showed distinct expression of ACKR1, which has been associated with venous ECs and the vasa vasorum in mice 18,19 and PRCP, 20 involved in angiogenesis and regeneration of damaged endothelium ( Figure 2B and 2C). E.1 and E.2 separated on expression of extracellular matrix genes in E.1 and cell mobility markers FGF18 and HEG1 in E.2. Both populations expressed VCAM1 ( Figure 2C), which is expressed by activated endothelium and facilitates adhesion and transmigration of leukocytes, such as monocytes and T cells. 21 Together, this suggests that E.0, E.1, and E.2 represent activated endothelium which actively aggravates inflammation in the advanced lesion by cell adhesion and neovascularization and mediating leukocyte extravasation. 22 Of note, subclass E.3 expressed typical SMC markers, such as ACTA2, NOTCH3, and MYH11, next to the aforementioned endothelial markers ( Figure 2C). This, combined with its clustering among the EC clusters and enrichment of transitory and SMC-related pathways ( Figure 2D), indicated that this subset may be undergoing endothelial to mesenchymal transition or vice-versa.
To validate these findings, we looked into the expression of ACTA2 and CD34 on sequential histological slides. Figure 2E shows cells lining the intraplaque vasculature that shows overlapping expression.

Synthetic Phenotype Dominates in Plaque Smooth Muscle Cells
SMCs were represented by cluster 8, expressing MYH11, PDGFRB, NOTCH3, and MFAP4 23-25 ( Figure 1B and 1C, Table II in the Data Supplement), which separated into 2 subclasses ( Figure

Intraplaque T Cells Are Defined by Activation Status
Lymphocyte clusters consisted of one small, but homogenous cluster of B cells (cluster 11; expressing CD79A, FCER2, CD22, and CD79B) [27][28][29] ( Figure 1B and 1C, Table  II in the Data Supplement), and 4 T-cell clusters. To define the T cells in more detail, we assessed the CD4 + T cells (expression CD4>CD8) and the CD8 + T cells (expression CD8>CD4) from CD3 enriched clusters 0, 1, 3, and 4. Isolating and reclustering the CD4 + T cells revealed 5 subclasses (CD4.0-CD4.4, Figure 3A, Table II in the Data Supplement) of which the primary difference was their activation state rather than the transcription factors and cytokines commonly used to define CD4 + T-helper (T H ) subsets ( Figure 3B and 3C). CD4.0 and CD4.1 exerted a cytotoxic gene expression profile exemplified by expression of GZMA, GZMK, and PRF1. Apart from these cytotoxic transcripts, cells in CD4.0 also showed   very little CD28 expression and some GZMB expression, suggesting that these cells are cytotoxic CD4 + CD28 null cells that have previously been correlated with unstable angina and increased risk of Major Adverse Cardiovascular Events. 30,31 In addition, gene expression in this cluster confirmed an enrichment in proinflammatory pathways associated with adaptive immune responses ( Figure 3D). Using flow cytometry, we confirmed the cytotoxic character of the CD4 + CD28 null cells, which showed that significantly more CD4 + CD28 − cells contained granzyme B as compared to the CD4 + CD28 + cells ( Figure 3E, Figure Table II in the Data Supplement). Interestingly, we also found some coexpression of FOXP3 with transcription factors RORA (RAR Related Orphan Receptor A) and GATA3 (GATA binding protein 3) in this cluster ( Figure IVB in the Data Supplement), which has, respectively, been associated with the enhanced immunosuppressive function of regulatory T cells 33 and with the prevention of polarization towards other T H subsets. 34 Expression of the T H cell subset-specific transcription factors TBX21 (Tbet [T-box transcription factor 21]; Th1), GATA3 (Th2), and RORC (RORγT; Th17) was not linked to a specific cluster ( Figure IVC in the Data Supplement), which seems to be a common phenomenon when dealing with T-cell scRNA-seq data. 35,36 By analyzing the CD4 + T cells in a clustering-independent method by selecting all cells that have the expression of both CD3E and CD4 and subsequently analyzing the expression of single T H -specific transcription factors, we find that a large population of T cells did not express a clear signal of the transcription factors ( Figure IVD in the Data Supplement). Analysis of CD8 + T cells revealed 3 subclasses ( Figure  VA and VB in the Data Supplement), which were similar to CD4 + T cells defined by differences in activation state. CD8.0 was identified as an effector-memory subset, characterized by expression of GZMK, GZMA, and CD69, indicating recent T-cell receptor activity ( Figure VC in the Data Supplement). A clear, terminally differentiated, cytotoxic CD8 + T-cell profile was observed in CD8.1, which showed expression of GZMB, TBX21, NKG7, GNLY, ZNF683, and CX3CR1, and in line, this subclass lacked CD69 expression. Finally, CD8.2 displayed a quiescent, central-memory CD8 + T-cell phenotype with expression of LEF1, SELL, IL7R, and LTB. In contrast to Fernandez et al 7 and previous scRNA-seq data obtained from various cancers, we did not detect a clear exhausted phenotype in the CD8 + T cells. [35][36][37] The CD8 clusters with reduced cytotoxic potential show expression of CD69, suggesting recent T cell receptor (TCR) activation and it will be of future interest to examine how these CD8 + populations were activated and how they affect the pathogenesis of atherosclerosis. This could indicate that not the cytotoxic but the more quiescent CD8 + T-cell subsets are responding to plaque-specific antigens and may be more relevant in the pathogenesis of atherosclerosis. Using experimental mouse models of atherosclerosis it has been shown that the majority of CD8 + T cells in the plaque are antigen-specific, 38 but so far little is known regarding the plaque-antigen(s) they respond to. Whereas CD4 + T cells have been shown to respond to (ox)LDL ([oxidized] low-density lipoprotein) and its related apo B 100 peptide, plaque-antigen(s) for CD8 + remain mostly indefinable. 39 Therefore, we are unable to define which antigens have activated the T cells in the atherosclerotic lesion.

Both Proinflammatory and Anti-inflammatory Macrophage Populations Reside in the Plaque
Atherosclerotic myeloid cells were represented by 5 clusters. A small, distinct mast cell population was defined by expression of HDC, KIT, CMA1, and TPSAB1. 40 The remaining myeloid clusters, cluster 6, 7, 8, and 12, expressed CD14 and CD68 ( Figure 1B  represented macrophages that differentially expressed tumor necrosis factor (TNF) and toll-like receptors (Figure 4B). Interestingly, both My.0 and My.1 expressed KLF4 (kruppel like factor 4), albeit at a low level, which is known to drive macrophages towards an anti-inflammatory phenotype by repressing the NF-κB (nuclear factor κB) gene program. 41 Our data may suggest that an inhibitory feedback loop in the proinflammatory macrophage populations is actively mediated by KLF4 expression.
In  To further characterize the 3 subclasses, we next examined pathways differentially enriched per population ( Figure 4D) as well as the upstream regulators that possibly govern these populations by Ingenuity Pathway Analysis ( Figure 4E). My.0 and My.1 showed enrichment for classical inflammatory and immune pathways clearly suggesting cellular activation, recruitment, and immune cell interactions driving their phenotype. In line, Ingenuity Pathway Analysis predicted that My.0 and My.1 are mainly controlled by proinflammatory factors, such as IL1A, IFN (interferon) A, IFNG, and IL1B.
My.2 was enriched for metabolic pathways and LXR/ RXR (liver X receptor/retinoid X receptor) activation, consistent with a foamy phenotype. Hence, this cluster was uniquely driven by anti-inflammatory pathways such as STAT6 (signal transducer and activator of transcription 6) and had typical foam cell-related factors including APOE and the LXR family (Nr1h [nuclear receptor subfamily 1 group H]: NR1H2,3,4), which interestingly showed some overlap with My.1. The latter may indicate that unlike the more recently recruited My.0, My.1 cells are gaining foamy characteristics.
My.3 is characterized by dendritic cell markers, such as CD1C, CLEC10A, and FCER1A ( Figure VIA in the Data Supplement) and this population most likely represents CD1c + dendritic cells. 13,48,49 In line with their dendritic cell phenotype, this cluster showed the highest expression of multiple class II HLA genes indicative of their enhanced activation status as a consequence of antigen-specific interaction with plaque T cells ( Figure  Finally, we compared our macrophage subclasses with monocyte and macrophage populations from 4 recent articles on scRNA-seq analysis of atherosclerotic plaques in mice. [3][4][5][6] Eight mouse populations showed significant overlap with our human subclasses (Figure 4F). My.0 showed no statistically significant overlap, but most resembled inflammatory mouse macrophages ( Figure 4G). My.1 resembled inflammatory, resident-like mouse macrophages, and My.2 overlapped with foamy, anti-inflammatory, Trem2 + macrophages. Together, this confirms the recently migrated and embedded inflammatory phenotypes we defined, respectively, for My.0 and My.1 and matches the foamy phenotype we saw in My.2. It also showcases a decent concordance between human patients and mouse models in relation to cell type diversity.

Intercellular Communication Drives Inflammation Within the Plaque
We next examined potential ligand-receptor interactions between cell types to predict intercellular communication within the lesion based on CellPhone DB v2.0. 50 Lymphocytes and mast cells showed the lowest absolute numbers of potential interactions while myeloid, endothelial, and SMCs displayed higher numbers of interactions ( Figure 5A). The low interaction between myeloid and T cells may be a consequence of the apparent lack of detection of TCR-related genes (TRA, TRB, TRG) in our scRNA-seq dataset and the fact that CD4-class II and CD8-class I interactions are not included in this database.
Subsequently, we specifically examined the top unique interactions within the myeloid populations, split by myeloid ligands ( Figure 5B) and receptors ( Figure 5C). We found multiple chemotactic interactions, including endothelial ACKR1 51 with myeloid-derived CCL2, CXCL8, CCL8, and CXCL1, of which the last 2 ligands were specifically expressed in My.1. We also observed an interaction between CSF1R on all myeloid subsets and CSF1 on ECs, smooth muscle cells, mast cells, and myeloid cells. CCR1 (C-C motif chemokine receptor 1) and CCR5 interacted with CCL5 from both CD4 + and CD8 + T cells and CXCR4 on B cells interacted with CXCL12 on My.1 cells. In addition, we identified communication patterns that are potentially involved in extravasation of myeloid cells, including CD44 (My)-SELE (EC), SELL (My)-CD34 (EC), SELPLG (My)-SELP, and SELL (both EC). Myeloid cells showed potential capability to attract other leukocytes, for example CCR5 + T cells through expression of CCL3 (My.1). Moreover, myeloid cells were also predicted to interact with T cells leading to mutual activation, through SIRPA (My)-CD47 (T) 52 ICAM1 (My)-ITGAL (CD8), inducing cytotoxicity and multiple interactions involved in antigen presentation. Lastly, interaction of PDGFB on myeloid subsets with PDGFBR on ECs suggest a possible myeloid-driven induction of angiogenesis, which has been associated with plaque destabilization. 53,54

Chromatin Accessibility of Myeloid and T-Cell Populations Reveals Transcription Factors Involved in Gene Regulation
Next, we aimed to further define the genomic landscape that accounts for the obtained cluster-specific patterns of gene expression and potentially uncover disease driving transcription factors. Using scATAC-seq, we examined the open chromatin promoter and enhancer landscape of myeloid and T cells in human plaques. We identified 4 myeloid and 5 T cell clusters by scATACseq. Population label transfer from scRNA-seq to scATAC-seq populations showed good agreement with the native scATAC-seq cluster borders and retrieved the majority of the scRNA-seq populations ( Figure 6A and 6B). Open chromatin at macrophage (CSF1R, IL1B)

Figure 5. Ligand-receptor interaction analyses to assess intracellular communication in the plaque. A, Heatmap showing logarithmic interaction scores between all cell subsets. Top quartile of unique ligand-receptor interactions between all cells and myeloid cells for both (B) ligands expressed by myeloid cells and (C) receptors expressed by myeloid cells. E indicates endothelial cells;
My, myeloid cells; and SMC, smooth muscle cells. and T cell-specific genes (NKG7), as well as enrichment of motifs of cell type TFs for macrophages and T cells (SPI1 55 and ETS1 56 ), confirmed the delineation between cell types ( Figure 6C and 6D). Transferred myeloid populations were reclustered analogous to the scRNA-seq clusters ( Figure 6E).
IRF4 has been shown to be a CD1c + dendritic cellspecific transcriptional regulator 57 and its motif was indeed enriched in My.3 ( Figure 6F). In line, we found specific open chromatin at the promoter region of IL12A, the subunit that is specific for the cytokine IL12, in all myeloid populations and an enhancer specifically in My.3 dendritic cells ( Figure VIIE in the Data Supplement). IL12 is required to induce a proinflammatory, T H 1-like cytotoxic phenotype of T cells and actively induces atherosclerosis. 58,59 Potentially, as a result of the My.3-specific IL12, we observed open chromatin at the IFNG and TNF loci in CD4.0, confirming its activated, cytotoxic phenotype and suggesting that this cluster has T H 1-like properties ( Figure VIIG and VIIH in the Data Supplement). Additionally enriched accessible motifs within the T cells ( Figure  VIIA and VIIB in the Data Supplement) were observed for the RUNX3 motif in CD4.0, normally a CD8 + T-cell linage specific TF that is also known to induce expression of cytotoxic genes in CD4 + T cells, [60][61][62][63] as well as the STAT3 motif, which is downstream of IL6 and IL2 signaling. The BATF_JUN motif ( Figure VIIC in the Data Supplement) that is known to be critical for effector function in T cells was also enriched in this cluster. 64 The effector function could be further confirmed by differential open chromatin of the GZMB and GZMH loci in both CD4.0 and all CD8 clusters ( Figure VIID  In line with the scRNA-seq data, My.1 showed enrichment of proinflammatory TF motifs ( Figure 6F), which matches the proinflammatory gene expression seen in these cells. This cluster was especially enriched in INF signaling induced TFs including IRF1 (interferon regulatory factor 1), IRF9, STAT1, and STAT2. The STAT1-STAT2 complex is known to interact with IRF9 upon IFNγ stimulation and hence induces the upregulation of proinflammatory cytokines as TNF, indicating an IFNγ pathway driven activation, possibly secreted by the T cells. 65 Indeed, the IRF9 motif was accessible and the IRF9 locus was open predominantly in My.1 cells (Figure 6G and 6H). Moreover, these IRF and STAT TFs are also key mediators of type I IFN responses which have previously been shown to associate with atherosclerotic disease as well. 66 My.0 cells were specifically enriched for the NFATC3 (nuclear factor of activated T cells 3) motif ( Figure 6F), a TF that has previously been linked to activated TLR-pathway signaling and has been shown to partially regulate subsequent TNFα and IL-1ß secretion. 67,68 Finally, My.2 cells were enriched for anti-inflammatory, foam cell-associated TFs in the scATAC-seq data similar as in the scRNA-seq data. We observed increased chromatin accessibility at loci harboring the KLF4 motif, which next to repressing proinflammatory programs was shown to implement an anti-inflammatory macrophage activation state and is also known to be involved in the transformation of vascular SMCs to macrophages 41,46 (Figure 6F). This is in contrast with the scRNA-seq data where KLF4 was expressed at a low level, indicating that while the KLF4 locus is poised, its associated gene program is not necessarily executed in all foamy macrophages. Furthermore, My.2 was enriched for the de novo motif MA1149.1, which was annotated to RAR_RXR, a motif with high similarity to the LXR_RXR motif ( Figure 6I). Moreover, LXR_RXR motif accessibility is enriched in My.2 cells and the NR1H3 (LXRα) locus is opened specifically in the My.2 population ( Figure 6J). In line, the scRNA-seq data likewise shows NR1H3 upregulation specifically in My.2 ( Figure 6K).
We could not map the regulatory T-cell cluster CD4.3 to a scATAC-seq cluster. The FOXP3 locus hardly showed open chromatin in any population in the scATAC-seq data set and neither did the Treg-associated cytokine gene IL10 ( Figure VIII and VIIJ in the Data Supplement).

Cell Type-Specific Enrichment of Genes in GWAS Loci.
GWAS have discovered 163 genetic susceptibility loci linked to coronary artery disease (CAD) through literature search and effects on expression. 69 However, the challenge remains in identifying the potential causal genes linked to these loci for functional testing as novel therapeutic targets. In part, this is due to the underlying genetic architecture where multiple causative variants in a gene might be involved and variants in linkage disequilibrium only show marginal significance in a GWAS. Another reason is that many of the risk variants are not causal and ambiguously linked to genes. A gene-centric analysis considers all variants in a gene and solves these issues, yet such analyses fail to identify the cells potentially involved. Here, we aimed to (1) identify genes associated with CAD that are (2) also highly expressed in specific cell types, effectively identifying tangible candidates for functional follow-up. To this end, we mapped genes near GWAS loci associated with CAD and assessed expression of these genes across our scRNA-seq cell populations to investigate their expression in disease-relevant tissue. We prioritized 317 protein-coding genes based on the summary statistics of a recent CAD GWAS 70 (see Methods and Table III in the Data Supplement). Next, we selected the genes that would best represent each individual cell population. To achieve this, we determined differentially expressed genes ( Figure VIII and Methods in the Data Supplement).
Three thousand eight hundred seventy-six genes were differentially expressed and differentially expressed genes were grouped into 15 gene expression patterns that best matched the scRNA-seq populations (Figure 7A). We overlapped the 317 CAD-associated genes with the 3876 differentially expressed genes, resulting in a significant overlap (permutation over random data P=2.67×10 -5 ) of 74 genes. These genes are distributed over multiple individual CAD loci (Table III in (Figure 7A) and contained AMPD2, CTSS, IL6R,  CAPG, GPX1, GNAI2, TRIB1, SH2B3, FES, C19orf38, and VASP ( Figure 7B and 7D). Genes in pattern 14 were predominantly associated with higher expression in both the smooth muscle cell population 8 and the CD34 + EC population 10 ( Figure 7E). This pattern contained ITGB, ARHGEF26, CXCL12, PTPN11, COL4A1, COL4A2, KANK2, and GGT5. Our results suggest that macrophages, smooth muscle cells, and ECs are of particular interest as a starting point for functional testing.
Furthermore, given that for many of the genes previously mapped to the 163 CAD loci the mechanisms and cellular expression are still unknown, 69 we examine whether these genes show cell type-specific expression in carotid plaques. We found that 24 of the 75 genes previously classified as unknown by Erdmann et al. 69 were differentially expressed between cell populations in carotid plaques, and included 3 genes that also showed association with CAD (CHD13, SNRPD2, and ARH-GEF26; Table III in the Data Supplement) in our analysis.

DISCUSSION
In the past 2 years, single-cell technologies have advanced our knowledge of atherosclerosis tremendously. scRNA-seq has been applied to specifically describe the immune cell landscape of murine and human atherosclerotic lesions. [3][4][5][6][7] The recent study by Fernandez et al 7 gave a first overview of the human immune cell landscape during atherosclerosis by showing a data set based on extensive cytometry by time of flight analyses and by comparing RNA expression profiles of T cells and macrophages in plaque and blood of symptomatic and asymptomatic patients. 7 They provide insight into which immune cells reside in the plaque and described their different activation states. Yet, both the mouse and human studies lack coverage of nonimmune cell types in the plaque and so far only a limited number of patients have been included in the scRNA-seq studies. Here, we applied scRNA-seq to all live cells in advanced human atherosclerotic plaques of 18 patients and revealed a highly diverse cellular landscape consisting of 14 main cell populations.
We detected a predominance of T cells in the leukocyte population of the human lesions. In contrast, murine scRNA-seq studies describe a more prominent presence of myeloid cells, which may be caused by the previously described declining myeloid content upon progression of human atherosclerotic plaques, whereas T cells reciprocally increase in human atherosclerosis. 71,72 Both CD4 + and CD8 + T-cell subsets were characterized by their activation state, rather than classical T H or T C subclasses. We could confirm the presence of activated T cells that in the plaque could especially be characterized by the expression of multiple granzymes. 7 In addition, we show that these granzymes are not only expressed by CD8 + T cells but also by a substantial number of CD4 + T cells in the plaque. The CD4 + T cells showed a dominant cytotoxic T-cell pool, characterized by expression of PRF1 and multiple granzymes, with granzyme B production confirmed by flow cytometry. The lack of CD28 expression in these cells indicates that this pool constitutes most likely a subset of cytotoxic CD4 + CD28 null T cells, which has previously been associated with atherosclerosis as they have been detected in peripheral blood of patients with CAD. 30,73 Although the presence of a similar TCR clone as observed in peripheral CD4 + CD28 null cells was found in bulk coronary artery tissue, 30,73 we can now confirm the presence of these cells on a single-cell level suggesting a functional role in patients with CVD. As cytokine expression could not be retrieved from the scRNA-seq data, but we were able to detect open chromatin at various cytokine gene loci within the T-cell populations using scATACseq suggesting active cytokine genes. Among others, IFNG showed open chromatin in the cytotoxic and effector T-cell subclasses. Apart from confirming the cytotoxic, T H 1-like phenotype within the plaque, this also suggests that the proinflammatory macrophage subclasses we observe in our dataset may be primed for classical activation by secretion of IFNγ by the T cells. 74 These T H 1 cells acting on macrophages may, in turn, be driven by activated CD1c + dendritic cells that were characterized by an active IL12 gene (ie, open enhancer), which has previously been found on protein level in plaque lysates 75,76 and the enrichment of HLA-DR (major histocompability complex, class II, DR beta 1) subtypes. 49,77 Each of the macrophage clusters seemed to have been activated differently, one expressing TNF and TLR4, which can be activated by oxLDL and IFNγ, 78 as well as IL1B, and the other more selectively expressing IL1B, which correlated with caspase expression suggesting inflammasome activation. 79 The recent CAN-TOS trial (Canakinumab Anti-Inflammatory Thrombosis Outcome Trial), which targeted IL-1β, 80  The IL12-IFNγ axis, as found in our scRNA-seq, data may form an important feature of T-cell activation in the plaque, and subsequent activation of myeloid cells contributes to the inflammation profile within the plaque. This is in line with several experimental studies that show the proatherogenic role of both IL12 and IFNγ in cardiovascular disease. 59,81,82 The more anti-inflammatory foam cell-like cluster was characterized by expression of ABC cholesterol efflux transporters and lipid-related genes whose expression is most likely driven by intracellular lipid accumulation. 83 The lipid-phenotype was confirmed by the enriched LXR_RXR TF motifs in the scATAC-seq data. LXR is a well-known nuclear receptor, active in foam cells and inducing ABC transporters. 46,84 The notice that foam cell formation per se is not proinflammatory is a recent ongoing paradigm shift in the field. Several studies have previously shown clear proinflammatory characteristics of foam cell formation either through engagement of TLRs by oxLDL, [85][86][87] induction of oxidative responses, 88 or through other pathways. [89][90][91] However, recent data studying foam cells in in vivo model systems 92 or isolating foam cells from murine plaques 5 clearly demonstrate that foam cells do not necessarily show proinflammatory characteristics and even may be considered anti-inflammatory. 92,93 In line, our data show that cells exhibiting the foam cell-driven LXR activation program do not express high levels of IL1B and TNF. This further confirms that lipid accumulation leads to LXR activation and induces an anti-inflammatory phenotype. We also observed TREM2 and CD9 expression within this cluster, resembling the TREM2 + macrophages found in murine atherosclerosis. 4,6 In other tissues, these TREM2 + CD9 + macrophages have been described as either lipid-associated macrophages 45 in obesity, or as scar associated macrophages 44 in liver cirrhosis. Key phenotypes of these cells were shown to involve profibrotic characteristics and this is also of high relevance for human atherosclerosis as it may indicate a plaque stabilizing macrophage population.
Our study provides further supports the notion that trans-differentiation of cells is likely to occur in human atherosclerosis. About a quarter of the My.2 macrophages expressed smooth muscle cell actin, which may indicate derivation from SMCs, or conversely macrophages showing an SMC-like fibrotic phenotype. 17,94 Presence of myeloid lineage-specific TF expression in these cells (eg, SPI and CEBPB) and absence of SMC TFs (eg, MYOCD and MRTFA) suggests that the latter is more likely. This is in line with previous reports applying SMC lineage tracing that showed that unidentified SMC-derived cells in atherosclerotic lesions exhibit phenotypes of other cell lineages, including macrophages and mesenchymal stem cells. 17,95 Also EC cluster E.3 was characterized by expression of smooth muscle cell markers, such as ACTA2, MYH11, and NOTCH3, suggesting that these cells could be in endothelial to mesenchymal transition. Mature ECs can exhibit considerable heterogeneity and can transdifferentiate into mesenchymal-like cells, a biological process called endothelial to mesenchymal transition. 95 There is accumulating evidence that endothelial to mesenchymal transition plays a role in atherosclerotic lesion progression and which has been linked with inflammatory stress and endothelial dysfunction. 96,97 Our study shows that distinct EC clusters are present within atherosclerotic lesions and the gene signatures identify a cluster that shares both SMC and EC characteristics further providing human supportive evidence that endothelial to mesenchymal transition may occur in advanced human atherosclerotic plaques.
Apart from cellular plasticity within the endothelial and macrophage population, our study also provides new insights regarding intercellular communication within the plaque and its role in progression of atherosclerosis. We have shown that the within the plaque this was predicted to be most prevalent between myeloid, endothelial, and smooth muscle cells. In addition to previous studies predicting interactions between macrophages and T cells in human lesions, 7 we were also able to predict interactions between ECs and SMCs, which were mainly involved with chemotaxis and extravasation of myeloid cells. We also show activation and recruitment of other immune cells, in particular T cells. Future development of therapeutics may benefit from detailing these interactions, providing specific pathways to target.
One of the significant post-GWAS challenges is the identification of candidate genes and pathways with clinical potential. 98 Here, we mapped genes based on common variants (minor allele frequency >1%) in susceptibility loci and used single-cell resolution expression in diseaserelevant tissue to identify putative targets for future functional follow-up. Our analysis showed enriched expression of CAD-associated genes in myeloid, endothelial, and smooth muscle cells. Furthermore, some of these genes are involved in cell-cell interactions, such as SORT1 and CXCL12. Interestingly, the candidate genes did not show a significant overlap with T-cell-specific transcriptional signatures. Our approach is pragmatic in that we explicitly focus on (1) common variants in risk loci associated with CAD, (2) map protein-coding genes that are associated with CAD to these risk loci, and (3) select CAD-associated genes that are also differentially expressed between cell populations. This identifies tangible potential targets as starting points for future functional testing in macrophages, endothelial, and smooth muscle cells. However, we note that rare loss-of-function variants and underrepresented genes may have significant effects on these and other cells. Future studies focusing on loss-of-function variants and under-expressed genes could identify potentially other cell-specific targets.
There are several limitations that come with the use of human plaque endarterectomy samples. The vast majority of carotid endarterectomy samples also contain an inevitable small medial smooth muscle cell layer that potentially has contributed to the contractile smooth muscle cell cluster. There is a fine line between increasing digestion time to isolate more cells and generating a pure sample containing a high number of viable cells. We, therefore, did not exclude that the ratio of cell types that we detected in the plaques based on gene expression profiles was affected by the digestion procedure.
In summary, we provide an in-depth characterization of the highly diverse cellular communities in advanced human atherosclerotic plaques. Based on RNA expression and chromatin accessibility profiles of individual cells, we uncover among others the presence of proinflammatory, cytotoxic T-cell populations, multiple activation states of macrophages and their interactions, and functionally distinct EC populations that all can be considered modulators of human disease development. Furthermore, we show that by incorporating GWAS data, scRNA-seq data can be applied to map CVD susceptibility loci to specific cell populations and define potential patient-driven relevant targets for drug intervention of specific cell types. Our approach thus provides a powerful tool to aid research into the mechanisms underlying human disease and discover novel drug targets for intervention.