Heart attacks and strokes account for the majority of deaths worldwide. Most are caused by arterial thrombosis, a complication of ruptured or eroded atherosclerotic plaque [1]. The normal arterial wall contains endothelial cells, smooth muscle cells (SMCs), adventitial fibroblasts and leukocytes including vascular macrophages [2], T cells and B cells [3^■■,4]. Several arterial leukocyte lineages were discovered by immunostaining and studied by flow cytometry (FACS) of digested aortae of experimental animals. After onset of atherosclerosis, endothelial cells are activated and SMC phenotype changes [5,6], the number of leukocytes increases in the adventitia and a eointima forms that also contains many leukocytes. Vascular macrophages proliferate locally [7^■], and new leukocyte subsets are recruited to the arterial wall from circulation, either across the luminal endothelium or from vasa vasorum. Stem cells have been observed in the adventitia that may give rise to both SMC and macrophages [8].

Immunohistochemistry uncovered the presence of myeloid, T, B and natural killer (NK) cells in atherosclerotic lesions. FACS allows the simultaneous use of 18 fluorochrome-tagged antibodies and forward scatter (cell size) and side scatter (granularity), thus revealing more leukocyte subsets in the atherosclerotic mouse aorta. FACS-defined myeloid cells include three subsets of macrophages [9], neutrophils [10] and monocytes [11]. Furthermore, dendritic cells [3^■■,12] and small numbers of plasmacytoid dendritic cells [13,14] can be found. Among αβ T cells, CD4 and CD8 [15] were found to be present, and a small number of γδ T cells [16]. FACS allows for intracellular staining of transcription factors, which identified Th1 [17], Th17 [18], follicular helper [19] and regulatory (Treg) [20] cells among CD4 T cells. Th1 cells are the predominant CD4 T-cell subset residing in the plaque [21]. Although Th2 cells are scarce in Apolipoprotein E-deficient (Apoe^–/–) mice on the C57BL6/J background, possibly due to the known Th1 bias of this strain [22], single-cell RNA sequencing (scRNAseq) of murine aortae clearly demonstrated the presence of a Th2 subset residing within the aorta [23^■■]. In addition, natural killer T cells [24], innate lymphoid cells [25,26] and NK cells [24] were found in atherosclerotic aortae. Using Cxcr6^GFP mice, a CXCR6⁺ CD4 T-cell subset was identified [27], which was confirmed by scRNAseq [28^■■]. The FACS approach identified only a few cell types per study. The main advantage of FACS is its low cost per cell and high throughput (Table 1). A 14-marker FACS panel resolved 12 different aortic leukocyte populations simultaneously (Table 1 and [23^■■]) (Fig. 1).

Mass cytometry (CyTOF) allows for the simultaneous detection of up to 42 parameters based on metal-conjugated monoclonal antibodies with no spectral overlap [29]. These are clear advantages over FACS. We recently performed CyTOF to more comprehensively and precisely assess the immune landscape in the Apoe^–/– model of atherosclerosis [23^■■,28^■■]. We applied a 35-marker CyTOF panel followed by Pheno-Graph, a dimensionality reduction and unsupervised clustering algorithm [30]. Of the, at least, 32 billion possible marker combinations, the analysis grouped aortic leukocytes into 23 clusters (Table 2), including four B-cell clusters, five T-cell clusters, NK cells, eosinophils, neutrophils, three dendritic cell clusters, two monocyte clusters, three macrophage clusters and three yet unknown populations with unique marker combinations. Although the frequency of macrophages was unsurprisingly increasing in the aorta of Apoe^–/– mice fed Western diet, one B-cell population and a Ly-6C^hi CD8⁺ T-cell population both vanished [23^■■]. Cole et al. [31^■] identified 13 principal aortic immune populations in aortas of Apoe^–/– mice including one population each of neutrophils, eosinophils, B cells, monocytes, macrophages, CD4⁺ T cells, CD8⁺ T cells, γδ T cells, NK cells, ILCs, plasmacytoid dendritic cells and two populations of dendritic cells. Subclustering of only myeloid cells uncovered 20 different populations of which subsets of CD169⁺CD206⁺ macrophages (one expressing CD209b and the other not) and common dendritic cells were significantly reduced in the aortae of cholate-free high-fat-fed Apoe^–/– mice [31^■].

Trogan et al. [32] dissected macrophage-rich areas of atherosclerotic plaques from Apoe^–/– mice by laser-capture microdissection (LCM) and analyzed the expression of several macrophage-related genes by quantitative PCR. Microarray analysis of LCM-dissected CD68⁺ cells from shoulder regions of human plaques showed 72 genes, among them IRF5 and CSF1, to be upregulated compared with in-vitro stimulated THP1 cells [33]. LCM was used to generate the first aortic transcript atlas of wild-type and atherosclerotic Apoe^–/– mice [34^■,35^■,36]. Although not single-cell, this LCM effort yielded transcriptional profiles of the plaque, media, adventitia and adventitia harboring arterial tertiary lymphoid organs (ATLOs). LCM and microarrays also showed differences in the transcriptional landscape of atherosclerotic plaques obtained by carotid endarterectomy from symptomatic and nonsymptomatic patients [37^■].

Next-generation sequencing allows for a deeper insight into the entire transcriptome without having to decide a priori which transcripts should be measured. This methodology was first used in atherosclerosis to determine the transcriptional differences of effector T cells (Teff), Tregs and a newly defined population of CCR5⁺ Teff cells appearing at later stages of atherosclerosis [38]. FACS-sorting the cells requires a-priori knowledge of the phenotype (surface markers) of the cell type to be analyzed. Thus, bulk transcriptomics is less suitable for discovery purposes of yet undefined cell types.

The first application of single-cell transcriptomics in atherosclerosis focused on T cells and described a population called Th1/Tregs [39^■■], which is very similar to the aforementioned CCR5⁺ Teff cells. This group used the Fluidigm C1 single microfluidics chip device (Fluidigm, California, USA) for capturing and preparing libraries of 270 cells of interest sorted by FACS.

Microfluidic devices encapsulating single cells in solvent droplets embedded in silicone oil recently became commercially available (10× Genomics, Pleasanton, California, USA), allowing for massively parallel transcriptome assessment of single cells (Drop-Seq) [40]. This technology, used in two recent publications, is now robust, reproducible and economical [23^■■,28^■■]. At this time, new scRNAseq techniques are being described every year, and the field is not settled. There are significant differences in sensitivity, coverage, precision, reproducibility, cost, time and scalability [41].

Most of the scRNAseq techniques enrich only for the 3′ ends of the transcripts; however, a switching mechanism at the 5′ end of the RNA template, (SMART)-Seq and its derivatives [42,43], produces reads spanning entire transcripts. Although the former can be suitable for studies exploring the heterogeneity of various tissues, only the latter can be used to study splicing isoforms, structural variation and single-nucleotide polymorphisms.

A known shortcoming of Drop-Seq is the random encapsulation of cells into the drops, so abundant cell types are sequenced more often than rare ones. Another inherent shortcoming is the occurrence of doublets, which can be mitigated by reducing the number of cells per microfluidic channel [44], but this increases the cost per cell.

scRNAseq identified 11– 13 distinct immune cell populations (Table 2) based on SEURAT, an algorithm that provides unsupervised identification of cells based on similarity in their transcriptomes [45]. The Drop-Seq approach detected an average of 1004–1873 genes per cell, which is state of the art and accounts for about 10% of the transcriptome [23^■■,28^■■]. Increasing sequencing depth would not detect more genes, as saturation was already reached at 1 million reads per cell. Genomic gating by Seq-Geq (FlowJo LLC, Ashland, California, USA) of scRNAseq data increased the B-cell diversity from two populations to three, which is the same number found by mass cytometry and FACS (Table 2). A B1-like cell subset was enriched for tumor necrosis factor signaling and produced large amounts of CCL5, suggesting a functional role to recruit more immune cells to the aorta [23^■■]. Furthermore, two B2-like subsets, both producing IFNγ and granuolcyte monocyte colony stimulating factor, were identified.

Macrophages are resident in almost all healthy tissues [46] including healthy mouse aortae [2]. The transcriptomes of FACS-sorted macrophages have been reported [47,48], but the two recent publications [23^■■,28^■■] are the first to show evidence of macrophage heterogeneity in atherosclerotic mouse aortae. All macrophages express the adhesion G-protein-coupled receptor, F4/80, and most also express the Fcg receptor-1 (CD64) and the receptor tyrosine kinase, MerTK [48,49]. In healthy mouse aortae, only a single population of macrophages was found by scRNAseq [28^■■], which corresponded to previously defined resident vascular macrophages [2]. In atherosclerotic Ldlr^–/– aortae, two additional subsets of macrophages were uncovered, one of which uniquely expressed TREM2 and another resembled an inflammatory macrophage phenotype that may be specific for atherosclerosis [28^■■]. In atherosclerotic Apoe^–/– mice, CyTOF identified three macrophage subsets that expressed CX3CR1, CD11c and CD103, respectively (Table 2 and [23^■■]). Genomic gating of scRNAseq data for Adgre1 (F4/80) and Cd68 revealed four macrophage subsets. It is not known if and how these correspond to the subsets seen in aortae of Ldlr^–/– mice.

Direct comparison of the principal immune cell populations across the single-cell transcriptomes reveals similar proportions of NK cells and a higher abundance of T cells in aortae of Apoe^–/– mice, whereas macrophages, although increasing upon Western diet feeding in both mice, are more predominant in Ldlr^–/– mice (Fig. 2). Significantly, the frequency of B cells is higher in Apoe^–/– and Ldlr^–/– mice fed chow diet compared with Ldlr^–/– mice fed a cholesterol-rich diet. This suggests a substantial dynamic range of aortic immune cells across the different models. However, at this point, it is unclear whether a difference in digestion protocols (digest for 60 min [23^■■] vs. 40 min [28^■■]) might favor the isolation of lymphocytes or myeloid cells.

Single-cell technologies, like CyTOF and scRNAseq, have provided promising advances in many diseases and disease models. Spitzer et al. [50] observed a multiorgan immune system response during cancer treatment and found that specific cell-type abundance predicted effective cancer therapy. Immune cell signatures before surgery predict recovery [51]. Activation and responsiveness states of B-cell sub-sets in the bone marrow were found to predict relapse in patients with B-cell precursor acute lymphoblastic leukemia [52]. Gene signatures from bulk transcriptomic data have shown promise as prognostic or diagnostic indicators of disease [53–55]. Gene expression deconvolution algorithms, such as CIBERSORT [56], allow for the in-silico dissection of biopsies from patients by probing the transcriptome of individual leukocyte signatures and estimating their respective abundance [57]. We extracted 11 signatures from the single-cell transcriptomes of the murine aortic leukocytes and tested whether these signatures had predictive value in the BiKE cohort [37^■] (Fig. 1). The BiKE study contains bulk transcriptomes from 100 human atherosclerotic plaques from carotid endarterectomies. As in mouse lesions, the predominant signature is derived from macrophages (55%), whereas T cells, B cells and monocytes make up 40% of the lesional immune cells, and NK cells are the smallest population (Fig. 2). From all 11 single-cell transcriptomes identified in atherosclerotic mouse aortae, the memory T-cell population correlated negatively and significantly with ischemic event-free survival of the patient [23^■■].

Although single-cell methods require dissection and digestion, which means cellular spatial information is lost, we were able to gain insight into the spatial distribution of some immune cell populations by genetically deconvolving the individual mouse immune cell signatures with bulk transcriptomes derived by LCM of murine aortae (lesion, media, adventitia and adventitia with ATLOs) (Fig. 2 and [23^■■]). This confirmed that the majority of aortic B cells reside in the adventitia, whereas macrophages, T cells and monocytes populate the plaque (Fig. 2 and [23^■■]). This approach also corrects for the under-sampling of macrophages caused by the digestion and mechanical disruption of the tissue.

Although CyTOF and scRNAseq provide a wealth of information, they both share the need for making single-cell suspensions by enzymatic and mechanical tissue disruption. Although dead cells are excluded by staining in CyTOF, cell death does not affect all cells equally, and this distorts the proportions. Large, branched cells like macrophages are systematically undersampled, as they are more likely to be destroyed by the enzymatic and mechanical dissociation [58]. As mentioned above, this can be corrected by running the scRNAseq-based gene signatures on bulk transcriptomes of LCM material using CIBERSORT. Contamination with circulating leukocytes can be controlled by injecting an anti-CD45 antibody to stain intravascular leukocytes and exclude them from subsequent analysis.

CyTOF is a destructive method, which means the cells analyzed cannot be recovered. CyTOF is currently limited by 42 lanthanide isotopes that are commercially available, but more metals are likely to become available soon. The most severe drawback of CyTOF is that it depends on user-driven selection of markers.

scRNAseq is truly hypothesis-free, but the spatial location of the interrogated cells is usually lost. An attempt to circumvent this limitation is the use of a photoconvertible fluorochrome to label a subset of immune cells in a known location before harvesting [59]. Downstream analysis of scRNAseq is more complex than bulk RNA sequencing due to the increased technical variability among individual cells, which can be attributed to the cell-cycle state [60], the burstiness of gene expression [61], the drop-out probability of low-expressed genes and the different PCR efficiency across the cells and across transcripts within the same cell. The lack of depth in scRNAseq transcriptomes appears to be fundamental and, in part, represents the fact that rare transcripts may not be in the cell at the time of harvest (stochastic dropouts). This can be overcome, partly, by constructing synthetic transcriptomes from the cells found in each t-distributed stochastic neighbor embedding cluster.

The immune system plays a crucial role in the progression and modulation of atherosclerosis. High-modality technologies including CyTOF and scRNAseq uncovered an unexpected diversity among aortic leukocytes. Furthermore, signatures constructed from single-cell transcriptomes can be useful to predict secondary ischemic events [23^■■]. Although both CyTOF and scRNASeq were capable of detecting the principle immune subsets, an integration of single-cell protein and transcriptome information may uncover a potentially even larger heterogeneity among aortic immune subsets. Also, a single immune cell atlas of human plaques is yet missing. The information gained by these data will help to specifically tailor new therapeutic strategies for altering the immune response in atherosclerosis.