Single-cell T cell receptor sequencing of paired human atherosclerotic plaques and blood reveals autoimmune-like features of expanded effector T cells

Atherosclerosis is a lipid-driven chronic inflammatory disease; however, whether it can be classified as an autoimmune disease remains unclear. In this study, we applied single-cell T cell receptor seqencing (scTCR-seq) on human carotid artery plaques and matched peripheral blood mononuclear cell samples to assess the extent of TCR clonality and antigen-specific activation within the various T cell subsets. We observed the highest degree of plaque-specific clonal expansion in effector CD4 + T cells, and these clonally expanded T cells expressed genes such as CD69 , FOS and FOSB , indicative of recent TCR engagement, suggesting antigen-specific stimulation. CellChat analysis suggested multiple potential interactions of these effector CD4 + T cells with foam cells. Finally, we integrated a published scTCR-seq dataset of the autoimmune disease psoriatic arthritis, and we report various commonalities between the two diseases. In conclusion, our data suggest that atherosclerosis has an autoimmune compondent driven

Atherosclerosis is a lipid-driven chronic inflammatory disease; however, whether it can be classified as an autoimmune disease remains unclear. In this study, we applied single-cell T cell receptor seqencing (scTCR-seq) on human carotid artery plaques and matched peripheral blood mononuclear cell samples to assess the extent of TCR clonality and antigen-specific activation within the various T cell subsets. We observed the highest degree of plaque-specific clonal expansion in effector CD4 + T cells, and these clonally expanded T cells expressed genes such as CD69, FOS and FOSB, indicative of recent TCR engagement, suggesting antigen-specific stimulation. CellChat analysis suggested multiple potential interactions of these effector CD4 + T cells with foam cells. Finally, we integrated a published scTCR-seq dataset of the autoimmune disease psoriatic arthritis, and we report various commonalities between the two diseases. In conclusion, our data suggest that atherosclerosis has an autoimmune compondent driven by autoreactive CD4 + T cells.
Atherosclerosis is the major underlying pathology of acute cardiovascular events, such as myocardial infarction and stroke. It is characterized by accumulation of lipids and subsequent inflammation of the medium and large arteries. As low-density lipoprotein (LDL) particles are important instigators of atherosclerosis, cardiovascular disease (CVD) has primarily been treated as a lipid-driven disorder, with a treatment focus on lowering LDL cholesterol levels. Nonetheless, inflammation plays a critical role in perpetuating the growth and instability of atherosclerotic lesions, highlighted by the success of recent clinical trials with anti-inflammatory agents 1,2 . Elucidating the dominant inflammatory Article https://doi.org/10.1038/s44161-022-00208-4 CD8 + T cells were specifically enriched in the plaque and/or had recently undergone antigen-specific interactions.
Here we present an approach to identify the T cell subsets that are specifically enriched in atherosclerotic lesions and whether these subsets underwent antigen-specific interaction in the plaque. We combine scRNA-seq and single-cell TCR sequencing (scTCR-seq) of human carotid plaques and matched PBMC samples. With this approach, we observed the highest degree of plaque-specific clonal expansion in both effector CD4 + T cells and, to a smaller extent, in the T reg population. By integrating the data from our patients with atherosclerosis with the scTCR-seq data from patients with psoriatic arthritis (PSA), we show that atherosclerosis has major similarities with another prominent autoimmune disease. Thus, our data suggest that atherosclerosis is characterized by an autoimmune component driven by autoreactive CD4 + T cells.

Signature of antigen-specific T cells in atherosclerosis
Recent scRNA-seq studies in human atherosclerosis have shown a prominent accumulation of T cells in the plaque 3,4 . However, it remains unclear whether these T cells are bystanders or whether they actively contribute to lesion progression through antigen-specific activation. To examine potential recent antigen encounter and activation, CD69 expression was measured on the surface of both PBMCs and plaque T cells through flow cytometry (cohort 1; Fig. 1a and Supplementary Table 1). pathways that drive atherosclerosis may, therefore, allow identification of new druggable targets independent of cholesterol lowering.
Single-cell RNA sequencing (scRNA-seq) and mass cytometry have allowed detailed mapping of the leukocyte contents of atherosclerotic plaques 3,4 . These studies show that T cells are the largest leukocyte population and that the number of effector T cells within the lesion associates with plaque instability. In combination with previous murine work, this suggests that inflammatory processes inside the plaque are driven by T cells, and atherosclerosis could be considered an autoimmune-like disease. In support of that, autoreactive (LDL-specific) CD4 + T cells have previously been reported in human atherosclerotic lesions and have been identified in elevated levels in the circulation of patients with CVD [5][6][7] . Moreover, vaccination approaches aimed at the reduction of self-reactive T cells or induction of regulatory T (T reg ) cells have shown promise in murine models of atherosclerosis 8,9 . However, when self-reactive CD4 + T cells are indeed the culprit T cells that propagate disease, clonal expansion and accumulation of these cells in the lesions is to be expected. Interestingly, recent work examining the T cell receptor (TCR) distribution in human coronary plaques showed primarily clonal expansion of CD8 + T cells inside the plaque and identified some of these TCRs to be specific for common viral antigens, such as influenza, cytomegalovirus (CMV) and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) 10 . However, this work did not include patient-matched peripheral blood mononuclear cell (PBMC) controls, rendering it impossible to assess whether the virus-specific   Fig. 1a,b). Because CD69 is known to rapidly upregulate after TCR/HLA engagement on T cells 11 , these data suggest that T cells actively engage in TCR-specific interactions within the atherosclerotic plaque. However, CD69 expression may also indicate the presence of resident memory T cells or may be upregulated by exposure to type I interferon (IFN) 12,13 . To determine whether the elevated CD69 expression was due to antigen-specific interactions in the plaque, we aimed to assess whether these T cells were clonally expanded as well. We combined scRNA-seq with scTCR-seq on paired PBMCs and carotid artery plaques from three male patients (cohort 2; Supplementary Table 1). The plaques were enzymatically digested, and live CD45 + cells were isolated by fluorescence-activated cell sorting (FACS) (Extended Data Fig. 2a). Both PBMCs and plaque cells were stained for CD3, CD4, CD8 and CD14 on a protein level with feature barcoding to properly distinguish between myeloid and T cell subsets on both RNA and protein level. All cells were subsequently processed with droplet-based single-cell 5′ RNA sequencing (10x Genomics) and sequenced (Fig. 2a). Unsupervised clustering revealed clusters consisted of T cells, natural killer (NK) cells, myeloid cells and B cells, originating from both PBMCs and plaque cells and with limited interpatient variability (Extended Data Fig. 2b-e). We did not further characterize all non-T cells, as we specifically focused on characterizing T cells to assess their clonal expansion in atherosclerosis. Therefore, all T cells were selected based on both RNA and protein expression, and, subsequently, unsupervised clustering was performed independent of the variable TCR genes to prevent clustering based on clonality (Methods). Subclustering of both PBMCs and plaque T cells revealed 13 distinct T cell subsets (Fig.  2b,c and Extended Data Fig. 2f). Within the T cells, we observed one memory (C0) and three naive (C1, C2 and C10) T cell clusters based on different expression levels of TCF7, LEF1, SELL and CCR7 (Fig. 2b,d and  Supplementary Table 2). Furthermore, three effector T cell clusters (C3, C4 and C5) were detected, expressing a multitude of different cytotoxic genes, such as GZMB, GZMK and GZMA (Fig. 2b, Next, we compared expression of CD69 as well as FOS and FOSB genes, which are also upregulated downstream of TCR signalling 17 , between plaque and blood. In line with the increased CD69 + protein expression measured through flow cytometry, all three genes showed an increased mRNA expression in plaque T cells compared to their PBMC counterparts (Fig. 2e). Subsequently, we applied VDJ sequencing to map paired α-chains and β-chains of the TCR and to define the clonal composition of the paired PBMCs and plaque T cells. Clonal expansion levels were calculated to indicate the clonotype abundance as percentage of the total measured TCRs per patient, per tissue ( Fig.  2f and Methods). 'Single' represents a single clonotype occurrence. Expanded T cells were divided into multiple categories characterized by increasing frequencies of clonotype occurrences, labelled as 'Small', 'Medium', 'Large' and 'Hyperexpanded'.
Taken together, a small increase in the percentage of total expanded T cells is observed in the plaque compared to PBMCs (PBMC 23% versus plaque 29%; Fig. 2f Table 3). One clonotype, originating from patient 1, was defined as Hyperexpanded in the PBMC and Large in the plaque.
The TCRα sequence of this clonotype matched with a TCRα sequence previously associated with CMV in the VDJdb database (https://vdjdb. cdr3.net/) 18 . The CD8 + T cell-specific clonotype, however, was only expressed in T cells that had little expression of CD69, FOS and FOSB, suggesting that this was not an active viral infection (Extended Data Fig. 4a-c). In addition, the tissue enrichment of clonotypes was assessed to investigate whether certain clonotypes specifically accumulated within either of the tissues or whether the clonotype abundance was unaffected by the location. T cells with clonotypes more present in the PBMC were identified as PBMC-enriched and vice versa for plaque-enriched T cells. Indeed, within the plaque, an increased percentage of plaque-enriched T cells was observed in all patients, suggesting a potential plaque-restricted antigen-induced clonal expansion (Fig. 2h,i, Extended Data Fig. 3d,e and Supplementary Table 3). To confirm these findings, bulk TCRβ sequencing was performed on matched blood and plaque T cells from ten patients (cohort 3; Supplementary Table 1). Both clonal expansion levels and tissue enrichment were similar between TCRβ bulk sequencing and the scTCR-seq data (Fig. 2j,k and Extended Data Fig. 5a).

Increased percentage of expanded CD8 + T cells in PBMCs
To properly isolate CD4 + and CD8 + T cells for further analysis, a selection was made of CD4 + and CD8 + single-positive T cells based on expression of these proteins as measured by feature barcoding (Extended Data Fig. 6a). Subclustering of CD8 + T cells resulted in 11 distinct subsets. Most CD8 + T cells had an activated phenotype as indicated by expression of multiple genes with a cytotoxic signature. One naive (C6) and one memory (C2) cluster were mainly detected in the PBMC (TCF7, LEF1, SELL and CCR7; Fig. 3a, Extended Data Fig. 6b,c and Supplementary Table 2). Four effector clusters were characterized, of which C0 and C10 mostly reside in the PBMC and C3 and C5 predominantly in the plaque. C0, C3 and C10 expressed a multitude of different cytotoxic genes, including GZMK and GZMA, at different levels. C5 was characterized by expression of CD69, FOS and FOSB (Fig. 3a, Extended Data  Table 2). T EMRA clusters were primarily associated with a gradual increase in expression of, among others, KLRD1, KLRG1 and FCGR3A, indicating various stages of terminal differentiation (Extended Data Fig. 6d). Using Seurat multimodal reference mapping, which maps your dataset to a large PBMC dataset with feature barcoding data, expression of CD45RA and CD45RO could be predicted. Indeed, T EMRA subsets were predicted to express CD45RA, whereas the effector T cells were predicted to be CD45RO + (Extended Data Fig. 6e). Finally, a cluster of γδ-T cells (C8) and a cluster of MAIT cells (C9) were detected within the CD8 + T cell subsets (Fig. 3a, Extended Data Fig. 6a,b and Supplementary Table 2). Subsequently, clonal expansion levels were examined and quantified within the CD8 + T cells in PBMC and plaque. A large percentage of clonally expanded CD8 + T cells was detected in the plaque; however, a higher percentage of expanded CD8 + T cells was detected in the PBMC (Fig. 3b,c, Extended Data Fig. 6f and Supplementary Table 3). Nevertheless, within the plaque, most expanded CD8 + T cells remained plaque enriched (Fig. 3d, Extended Data Fig. 6g and Supplementary Table 3). Expanded CD8 + T cells showed upregulation of multiple genes involved in CD8 cytotoxicity-for example, GZMH, KLRD1, PRF1 and GZMB (Fig. 3e). Interestingly, when comparing PBMCenriched versus plaque-enriched CD8 + T cells, PBMC-enriched cells expressed cytotoxic genes, such as GNLY, PRF1 and members of the killer cell lectin-like subfamily (KLRG1 and KLRD1), whereas plaque-enriched CD8 + T cells seemed to have experienced recent antigen-induced TCR activation (Fig. 3f). To further illustrate the plaque-expanded CD8 + T cell clusters, we selected C1, C3, C5 and C9, which had relatively the most plaque-enriched expansion (Fig. 3g). C1, C3 and C5 all expressed   19 and of TCR activation genes. To identify potential dynamics of different CD8 + populations, we applied lineage tracing analyses using Monocle3 and RNA velocity. RNA velocity shows that, within the CD8 + clusters, cells tend to be less prone to switch into another subset. A small trajectory appeared between the memory CD8 +   Article https://doi.org/10.1038/s44161-022-00208-4 T cells (C2) and the antigen-experienced effector T cells (C5), yet this was not clearly retrieved with pseudotime analysis (Fig. 3i,j).

Increased percentage of expanded CD4 + T cells in plaque
Unsupervised clustering revealed 11 subsets of CD4 + T cells (Fig. 4a). As previously described, CD4 + T cell clusters are mainly defined by a shift in activation status 3,4 . Two naive T cell clusters (C1 and C2) and a memory T cell cluster (C0) were mainly detected within the PBMC (Fig. 4a, Extended Data Fig. 7a,b and Supplementary Table 2). Furthermore, a T-helper (T h ) 17-like cluster (C4) expressing RORC, RORA and CCR6, as well as a T reg cluster (C5; Fig. 4a, Extended Data Fig. 7b and Supplementary Table 2), were identified. Whereas T reg cells were found in both PBMC and plaque, T h17 -like cells were mainly detected in PBMC (Extended Data Fig. 7c). A T cell cluster with genes involved in cell migration (T migr , C6) mainly resided in PBMC (Supplementary Table 2). Two different effector subsets were characterized, of which one was more plaque specific with high expression of CD69, FOS, JUN and GZMA (C3), and one was found in both tissues specifically enriched for GZMK (C8; Fig. 4a, Extended Data Fig. 7a,b and Supplementary Table 2). Moreover, a cytotoxic CD4 + T cell cluster, which resembled the previously described CD4 + CD28 null cells 3,20,21 , was defined by expression of GZMB and PRF1 and lack of CD28 and was found in both PBMC and plaque (Fig. 4a, Extended Data Fig. 6a,b and Supplementary Table 2). Finally, a cluster of T cells was observed in the PBMCs that expressed genes involved in IFN I signalling and a small mast cell cluster in the plaque ( Fig. 4a and Supplementary Table 2). Subsequently, CD4 + T cell clonality was assessed. Clonal expansion levels were projected on the CD4 + T cell uniform manifold approximation and projection (UMAP) and quantified. In line with a recent study by Chowdhury et al. 10 , the percentage of clonal expanded CD8 + T cells in the plaque is larger than those in CD4 + T cells. However, in contrast to CD8 + T cells, a marked increase in the percentage of expanded CD4 + T cells in the plaque was revealed compared to the PBMCs (Fig. 4b,c, Extended Data Fig. 7e and Supplementary Table 3). Furthermore, the expanded clonotypes in the plaque CD4 + T cells were mostly plaque enriched (Fig. 4d, Extended Data Fig. 7f and Supplementary Table 3). When comparing expanded CD4 + T cells to their single counterparts with a unique clonotype, upregulation of genes involved in T cell activation and cytotoxicity, such as GNLY, GZMH, PRF1 and CX3CR1, were particularly observed in the expanded T cells, whereas single T cells expressed genes upregulated in naive and memory T cells (CCR7, LTB, LEF1, SELL and CD27) (Fig. 4e). Interestingly, when comparing clonally expanded PBMC-enriched versus the plaque-enriched expanded CD4 + T cells, plaque-enriched CD4 + T cells showed enhanced expression of genes upregulated shortly after antigen-specific TCR interaction (JUN, CD69, FOS and FOSB) (Fig. 4f), suggesting that there are CD4 + T cells that undergo antigen-specific interactions in the plaque. Next, we quantified the absolute number of plaque-enriched clones per CD4 + T cell cluster (Fig. 4g), which revealed cluster C3 as the major contributor in absolute number of plaquespecific clonally expanded T cells. Furthermore, C7 and C8 consisted of a relatively large number of plaque-enriched clones compared to the other CD4 + T cell clusters. The C7 cluster, characterized by an increase in cytotoxic genes, including GZMB, NKG7 and PRF1, has little to no expression of CD69, FOS and FOSB, indicating that, although these cells have substantial expanded clonotypes, they do not express genes involved in antigen-induced activation (Fig. 4h). The effector populations C3 and C8 displayed increased expression of TCR proximal genes CD69, FOS and FOSB. Interestingly, whereas we did not observe increased accumulation of clonally expanded T reg cells (C5) in plaque, we did observe upregulation of FOS, FOSB and JUN in plaque-derived T reg cells compared to PBMC-derived T reg cells, suggesting that T reg cells are encountering antigen in the plaque. Expression of various functional T reg markers (FOXP3, IL2RA, TIGIT, CTLA4 and TNFRSF4 (OX40) and TNFRSF18 (GITR)) in the plaque compared to the PBMC indicated increased activity of T reg cells (Fig. 4h,i).
To identify the origin of the antigen-specific effector CD4 + T cell subsets in the plaque, we applied lineage tracing analyses to define the dynamics of the different CD4 + T cell populations. Pseudotime analysis using Monocle3 showed a trajectory ranging from naive T cells toward either the T reg cells (branch 1) or the effector T cell population (branch 2) (Fig. 4j). The first pseudotime branch directing toward T reg cells is projected through the T h17 -like CD4 + T cell cluster, potentially suggesting a plasticity between both subtypes. However, if the complementary RNA velocity analysis is assessed (time-resolved analysis based on spliced and unspliced mRNA 22 ), the T reg cluster does not seem to be derived from the T h17 -like cells (branch 1; Fig. 4k). Moreover, T reg cells in tissue also cluster further away from the circulating T h17 -like cells compared to the PBMC T reg cells, indicating that the plaque environment is less likely to induce a phenotype switch from T reg to T h17 . In addition, no overlapping clonotypes were found between both clusters, and FOXP3 and RORC did not co-express (Extended Data Fig. 7b,c), suggesting that, in our dataset, we were not able to detect the previously described T reg /T h17 plasticity 23 . Looking at the other branch in both pseudotime analysis and RNA velocity (branch 2), a clear path ranging from the T migr cluster (C6) toward the CD69 + T eff cluster (C3) was observed. Their migratory phenotype, highlighted by expression of CCR4 and CCR10 previously described to be expressed on infiltrating T cells in the inflamed skin 24 , suggests that this T migr subset could be the precursor population for the antigen-specific CD4 + T cells in the plaque (Extended Data Fig. 7d). Indeed, when comparing overlap in TCR sequence between the different CD4 + subpopulations, 37 clonotypes overlapped between both cluster C6 and cluster C3. Within the top five most expanded clonotypes, four plaque-enriched clonotypes were detected and exhibited marked expansion in C3 compared to C6, further confirming our hypothesis that the clonally expanded T eff cells could originate from the circulating migratory T cell subset ( Fig. 4l and Extended Data Fig. 7c).

TREM2 + macrophages can activate antigen-induced CD4 + T cells
Our data suggest that atherosclerotic plaques harbor one major CD4 + T cell subset that regularly undergoes antigen-specific interactions. To understand whether and how these clonally expanded T cells interact with myeloid subsets in the plaque, we selected five plaque myeloid cell populations from the overall dataset: myeloid-derived dendritic cells (DC-M), plasmacytoid dendritic cells (DC-P), proliferating macrophages (M-Prol), inflammatory macrophages (M-Inf) and foamy TREM2 hi macrophages (M-TREM2) (Extended Data Fig. 8a) 3 . Using CellChat, we examined potential signalling pathways between these myeloid subsets and the CD4 + and CD8 + T cells in the plaque 25 . CellChat can predict incoming (receptor) and outgoing (ligand) activity of cell signalling pathways based on scRNA-seq data, accounting for the multimeric structure of ligand-receptor complexes and the effect of co-factors on the ligand-receptor interactions. Predicted outgoing and incoming pathway signalling was displayed per cluster. Overlap between outgoing and incoming signals of a certain pathway within or between clusters indicates a possible interaction through this pathway. The different CD4 + T cell clusters showed different levels of relative signalling strength in the outgoing signalling patterns (top bar plot heat map, relative to outgoing signals of all pathways in the heat map), whereas CD8 + T cells showed little difference between the clusters (Fig. 5a and Extended Data Fig. 8b). In general, the most upregulated signalling pathway was MHCII as outgoing signal on all myeloid subsets and incoming signals in multiple CD4 + T cell subsets, including cluster 3 (C3). The plaque-enriched CD69 + C3 displayed elevated outgoing signalling patterns. Interestingly, one of the pathways that was enriched in this cluster was the CD40 pathway, involved in antigen-specific T cell activation 26 . Next, we assessed whether the CD40 pathway was also enriched as an incoming signalling pattern (Fig. 5b). Specific enrichment was observed in the M-TREM2 (foam cell) subset. Apart from the CD40 pathway, multiple other enriched  GZMK  NKG7  PRF1  CD28  CCL5  TNF  EOMES  FOXP3  IL2RA  TIGIT  CTLA4  CD69 FOS FOSB    pathways involved in immune synapse formation and co-stimulation could be defined between C3 and M-TREM2, including the CD99, CD6, CD40, macrophage inhibitory factor (mIF) and annexin A1 pathways (Fig. 5b) [27][28][29][30] . Together, this suggests that M-TREM2 could be involved in activation of the clonally expanded CD4 + T cells in atherosclerotic lesions.

Common autoimmune phenotype in expanded plaque T cells
Based on the accumulation of plaque-enriched CD4 + and CD8 + T cell clonotypes, we hypothesized that human atherosclerosis could be characterized as an autoimmune-driven T cell response. To further confirm this hypothesis, we integrated an scTCR-seq dataset of the autoimmune disease PSA, containing data from PBMCs and synovial fluid (SF) 31 . As in this study CD45RA − T cells were isolated, we excluded the naive T cell clusters from our dataset. Moreover, this study did not include feature barcoding. CD4 + and CD8 + T cells were, therefore, selected based on the labels predicted by multimodal reference mapping (Extended Data Fig.  9a-f). Subsequently, CD4 + and CD8 + T cells of both diseases were integrated (Extended Data Fig. 9g,h) and projected on the atherosclerosis CD4 + and CD8 + UMAP as reference. Remarkably, a clear overlap between PBMCs from atherosclerosis and PSA was observed in both CD4 + and CD8 + T cells. In addition, this overlap was also seen between plaque and SF for both T cell subsets (Fig. 6a,b). Next, clonal expansion levels were recalculated for both atherosclerosis and PSA (percentage of all CD4 + or CD8 + TCRs). Indeed, clonally expanded T cells were found in similar CD4 + and CD8 + T cell clusters in both diseases (Fig. 6c,e). Moreover, quantification of this clonal expansion revealed a similar distribution. An increased percentage of expanded CD8 + T cells versus expanded CD4 + T cells was detected in SF. However, as seen in atherosclerosis, the percentage of expanded CD4 + T cells was increased in SF compared to PBMC, whereas expanded CD8 + T cells did not differ between both tissues (Fig. 6d,f). Tissue enrichment scores were also determined and again displayed similarities between atherosclerosis and PSA. Tissueenriched T cells were located in overlapping clusters in both diseases. Quantification resulted in an increase in tissue-enriched T cells in both CD4 + and CD8 + in plaque and SF compared to their matched PBMCs, although this enrichment was more prominent in SF versus plaque T cells (Fig. 6g-j). Finally, we defined the genes supporting the overlap between the atherosclerosis and PSA subsets in C3 and C5 of both CD4 + and CD8 + T cells. CD4 + T cells from C3 were characterized by high expression of CCL5, GZMK and GZMA in both plaque and SF ( Fig. 6k and Extended Data Fig. 10a). Atherosclerosis-specific C3 CD4 + T cells had a b     Fig. 6l and Extended Data Fig. 10b). Nevertheless the T reg subset showed some disparity between SF and plaque-derived cells as plaque T reg also increasingly expressed ICOS and ENTPD1, compared to PSA SF-derived T regs . Interestingly, atherosclerosis T reg cells in both PBMC and plaque had increased expression of TGFB1 compared to the PSA T reg cells. In both PSA and atherosclerosis CD8 + C3 T cells, expression profiles displayed a similar phenotype with high expression of T cell effector genes-for example, CCL5, GZMH, GZMA, GZMK and NKG7 ( Fig.  6m and Extended Data Fig. 10c). Lastly, CD8 + T cells from C5 showed upregulation of genes involved in antigen-induced TCR activation in both affected tissues (FOS and JUN) ( Fig. 6n and Extended Data Fig. 10d). FOSB was upregulated in plaque only, similarly to CD4 + C3, and JUNB expression was increased in PSA compared to atherosclerosis. Furthermore, increased expression of ZNF683 was observed in both diseased tissues. GZMH was particularly upregulated in plaque CD8 + T cells. To summarize, these data support the hypothesis that atherosclerosis has a considerable autoimmune component, as it has phenotypically similar clonally expanded T cells compared to the autoimmune disease PSA.

Discussion
Atherosclerosis has a long history of being treated as metabolic and/ or lifestyle disease, with its inflammatory component being overlooked as a potential target of intervention. Groundbreaking work earlier this century has shown that inflammation is an integral part of the disease pathophysiology, and considerable health benefits can be obtained by intervening in inflammatory cascades. Our work here takes these observations a step further and suggests that atherosclerosis is an autoimmune-like disease, with autoreactive T cells driving the inflammation process inside the plaque (Fig. 7). Classic autoimmune diseases that involve inflammation of distinct tissue, such as type I diabetes, multiple sclerosis and rheumatoid and psoriatic arthritis, are usually associated with specific HLA class II alleles, suggesting that a pathogenic CD4 + T cell response is a major cause of disease. Moreover, accumulation of antigen-specific T cells at the site of inflammation is a hallmark of autoimmune disease. The absence of clear associations of HLA alleles and atherosclerosis argue against the autoimmune theory in CVD 32 , yet the multifactorial nature of the disease and the large population that it affects make such associations difficult to establish. Accumulation of T cells in atherosclerotic plaques, however, is well established. Moreover, earlier studies investigating TCR diversity using TCRβ sequencing in the plaque indicated an increased clonality in the lesions compared to blood samples from patients with CVD 33 . By taking advantage of scTCR-seq here, we can combine data on distribution of TCRs with their activation state and functionality. Using this approach, we show that a selected number of effector CD4 + T cells and CD8 + T cells accumulate in the lesions and probably undergo antigen-specific activation similarly to autoimmune diseases, such as PSA. Recent work by Chowdhury et al. 10 using a similar approach reached the same conclusion 10 ; however, by using matched PBMC controls, we were able to determine that a large fraction of clonally expanded CD8 + T cells did not specifically accumulate in the plaque and were equally represented, or even overrepresented, in the circulation. One CD8 + T cell clone in particular, whose Vα TCR sequence was identified as specific for CMV, was hyperexpanded and accounted for a substantial percentage of clonally expanded T cells in the plaque while also contributing to the clonally expanded CD8 + T cell pool in the PBMCs of this patient. Moreover, this clone did not show a signature of recent antigen encounter. Apart from classical CD4 + and CD8 + T cells, we also identified a pro-inflammatory MAIT cell population. MAIT cells have been described in multiple autoimmune and inflammatory diseases, including PSA, with contradicting or unknown contributions to disease development. How MAIT cells contribute to atherosclerosis development and whether they are activated through their non-polymorphic MHC class I-like protein MR1 or through TCR-independent activation induced by e.g. IL-12 and IL-18 (refs. [34][35][36] ) needs further elucidation. By instead focusing on the clonally enriched T cells specific for the plaque, we observed that one subset of effector CD4 + T cells was considerably enriched in clonally expanded TCRs and expressed genes indicative of recent antigen engagement. Although we found two such populations in the CD8 + T cells, their clonal enrichment was less pronounced. Interestingly, we also observed an antigen activation signature in the plaque-residing T reg cells, suggesting that these T cells undergo antigen-specific interactions in the plaque. However,

Fig. 7 | Expansion of T cells with autoimmune-like features in CVD patients. Schematic presentation of the main conclusions.
Article https://doi.org/10.1038/s44161-022-00208-4 these T reg cells did not show substantial clonal expansion, suggesting that these cells do not expand in the plaque. Instead, RNA velocity analysis suggests that T reg cells are not derived from any other T cell population that we detected in PBMC or plaque. Also, we observed minimal overlapping TCR sequences between T reg cells and other T cells in the plaque, in contrast to the effector CD4 + T cell population, which showed considerable TCR overlap with a migratory CD4 + T cell subset in the circulation. Previous work suggests that T reg cells can lose their suppressive capacity and gain expression of pro-inflammatory markers 37 . A shift of autoreactive (ApoB100-specific) T reg cells toward a T h17 phenotype has been associated with severity of CVD. Although the authors show in mice that this shift happens independent of the TCR clonotypes, our data argue against such a shift and suggest that T reg cells and effector CD4 + T cells do not derive from the same ancestor but, rather, develop independent of one another. Alternatively, the number of TCRs detected here may not have been sufficient to find overlapping sequences between T reg cells and effector CD4 + T cells. Also, it is unknown whether ApoB100-specific T cells undergo antigenspecific interaction in the plaque, and, because the antigen specificity of T cells investigated in this study are unknown, it is possible that we did not examine ApoB100-specific CD4 + and CD8 + T cells here. We attempted to cluster the TCRs in silico using GLIPH2 and GIANA algorithms 38,39 , which are based on CDR3β similarity, as this is proposed to be an attractive way to cluster TCRs for a specific antigen together. However, a convincing clustering of plaque-enriched clonotypes was not observed in our dataset. The current clustering algorithms may have some limitations, which, in our data, was illustrated by co-clustering of CD4 + T cell-derived and CD8 + T cell-derived clonotypes, which was resolved only if the CDR3α sequence was included. Moreover, we observed diffuse clustering of clonotypes previously reported as ApoB100 specific 40 , suggesting that the current algorithms are not specific enough to resolve TCR clustering in atherosclerosis. Therefore, we think that a more stringent approach that includes both CDR3α and CDR3β needs to be developed.
As we observe antigen-specific activation in both the effector and T reg subsets, it is currently unclear what the overall effect of TCR engagement in the lesion is. Previous work in mice has shown mixed results with MHCII −/− apoE −/− mice, suggesting that this interaction is protective, whereas various papers suggest a pathogenic role for CD4 + T cells in atherosclerosis 41,42 . Interestingly, our work identifies several pathways involved in co-stimulation and immunological synapse formation that potentially drive pathogenic interactions of effector CD4 + T cells with the M-TREM2 (foam cell) population. When limited to effector CD4 + T cell populations, these may be specific and druggable targets. For instance, the expression of CD40LG on the clonally enriched effector population suggests active signalling to foam cells through CD40. This co-stimulatory pathway and that of other TNF superfamily member has been extensively studied in mouse models of atherosclerosis and is the subject of a clinical study 43,44 . The observation of antigen-specific T reg interaction also provides a rationale for potential therapeutic possibilities, such as expanding these cells by means of vaccination or development of tolerogenic chimeric antigen receptor (CAR) T cells. Identification of the antigen(s) driving T reg interaction in the plaque will be crucial for this development. Potential antigens, such ApoB100, heat shock proteins and fibronectin, have been suggested as potential selfantigens and have shown therapeutic potential as antigens in mouse models [45][46][47] and may serve as a potential starting point for vaccine development. Thus, here we highlight an autoimmune component to the pathophysiology of atherosclerosis, and we confirm a rationale for immunotherapeutic interventions in CVD.

Patient cohorts
For flow cytometry (cohort 1) and bulk TCRβ sequencing (cohort 3), whole blood and atherosclerotic plaques were obtained from, respectively, 61 and 10 patients who underwent carotid endarterectomy (CEA) surgery at the Haaglanden Medical Center Westeinde (HMC; The Hague, The Netherlands). The study was approved by the Medical Ethics Committee of the HMC (study approval number, cohort 1: 17-046, protocol number NL57482.098.17; study approval number, cohort 3: Z19.075, protocol number NL71516.058. 19). For scTCR-seq, whole blood and atherosclerotic plaques were obtained from three male patients who underwent CEA (cohort 2). Patients were included in the Athero-Express biobank (www. atheroexpress.nl), an ongoing biobank study at the University Medical Centre Utrecht (UMCU) 48 . The study was approved by the Medical Ethics Committee of the UMCU (study approval number: TME/C-01.18, protocol number 03/114). All blood samples were collected by venipuncture before surgery. Atherosclerosis specimens were obtained from primary CEAs, and estenotic plaques were excluded due to their different plaque composition as compared to primary atherosclerotic plaques 49 . Informed consent was obtained from all patients involved in this study.

Whole blood processing
Peripheral venous blood was collected in K2-EDTA blood tubes (BD Vacutainer). For scTCR-seq, blood was processed within 10 minutes after withdrawal (cohort 2). For both cohort 1 and cohort 2, blood was diluted 1:2 in PBS containing 2% FCS. A density gradient was created using SepMate PBMC isolation tubes (STEMCELL Technologies) containing Ficoll-Paque Premium (GE Healthcare). Cells were centrifuged at 1,200g for 10 minutes at room temperature. The intermediate layer containing PBMCs was isolated and washed twice with PBS + 2% FCS (250g, 10 minutes, room temperature). Cells were taken up in PBS + 1% BSA until further processing. For cohort 3, whole blood samples were lysed twice with ACK lysis buffer in PBS (1:10) for 10 minutes at room temperature and washed with PBS (300g, 5 minutes). Cells were taken up in RPMI + 1% FCS and cryostored in CryoStor cell cryopreservation medium (Sigma-Aldrich) until further use.

Human atherosclerotic plaque cell isolation
Human carotid plaques were collected during CEA; the culprit segment (5 mm) was used for histology and embedded in paraffin as described elsewhere 48 . In brief, culprit segments were fixed in 4% formaldehyde and decalcified in 10% EDTA, pH 7.5. Afterwards, culprit segments were embedded in paraffin. Time between surgical removal and plaque processing did not exceed 10 minutes. The inclusion of a small medial layer in the dissected tissue could not be excluded during the surgical procedure. The remainder of the plaque was washed in RPMI and minced into small pieces with a razor blade. The tissue was then digested in RPMI 1640 containing 2.5 mg ml −1 of collagenase IV (Thermo Fisher Scientific), 0.25 mg ml −1 of DNAse I (Sigma-Aldrich) and 2.5 mg ml −1 of Human Albumin Fraction V (MP Biomedicals) at 37 °C for 30 minutes. In cohort 2, 1 µM flavopiridol (Selleck Chemicals) was added to the digestion mixture. Subsequently, the plaque cell suspension was filtered through a 70-µm cell strainer and washed with RPMI 1640. Cells were kept in RPMI 1640 with 1% FCS until subsequent staining for flow cytometry (cohort 1), feature barcoding and FACS (cohort 2) or cryostored in CryoStor cell cryopreservation medium (Sigma-Aldrich) until further use.

Flow cytometry
Single-cell suspensions from blood and plaque from cohort 1 were stained with a mixture of extracellular antibodies for 30 minutes at 37 °C (Supplementary Table 4). All measurements were performed on a CytoFLEX S (Beckman Coulter) and analysed with FlowJo version 10.7 (Tree Star). A Shapiro log-normality test was performed, and a two-tailed Mann-Whitney test was performed using GraphPad analysis software to determine significance.

Antibody staining for feature barcoding and FACS
PBMC. PBMCs of cohort 2 were stained with TotalSeq-C antibodies against CD3, CD4, CD8 and CD14 (Supplementary Table 4). Antibody Article https://doi.org/10.1038/s44161-022-00208-4 pools containing 0.25 µg per antibody were prepared in labeling buffer (PBS + 1% BSA) and spun down at 14,000g for 10 minutes at room temperature, and supernatant was collected for further staining. First, cells were stained with Human TruStain FcX (BioLegend) for 10 minutes at 4 °C. Next, the antibody pool supernatant was added and incubated for 30 minutes at 4 °C. Cells were washed three times with labeling buffer at 400g for 5 minutes at 4 °C. Next, cells were taken up in PBS + 0.4% BSA and further processed with 10x Genomics.
Plaque. Single-cell suspensions of plaques of cohort 2 were stained with TotalSeq-C antibodies against CD3, CD4, CD8 and CD14 (Supplementary Table 4). Antibody pools containing 0.25 µg per antibody and plaque (1 µg per antibody) single-cell suspensions were prepared in labeling buffer (PBS + 1% BSA) and spun down at 14,000g for 10 minutes at room temperature, and supernatant was collected for further staining. First, cells were stained with Human TruStain FcX (BioLegend) for 10 minutes at 4 °C. Next, the antibody pool supernatant was added together with Calcein AM (1:1,000, Thermo Fisher Scientific), Hoechst (1:1,000, Thermo Fisher Scientific) and CD45-PECy7 (1:200, clone HI30, BD Biosciences) and incubated for 30 minutes at 4 °C. Cells were washed three times with labeling buffer at 400g for 5 minutes at 4 °C. Next, cells were taken up in PBS + 2% FBS. Live CD45 + plaque cells were sorted using the BD FACSAria II (BD Biosciences) in PBS + 0.04% BSA and further processed with 10x Genomics.

scTCR-seq by 10x Genomics
scTCR-seq was performed on PBMCs and live CD45 + plaque cell suspensions from cohort 2 using 10x Genomics 5′ Single Cell Immune Profiling technology. Sequencing libraries were prepared using the 5′ version 1.1 chemistry following standard 10x Genomics protocol. Sequencing was performed using the Illumina NovaSeq 6000 (Novogene).

Bulk TCRβ sequencing
Genomic DNA was extracted from plaque single-cell suspensions and matched PBMC samples (cohort 3) using a DNA extraction kit in accordance with the manufacturer's instructions (Qiagen). Sequencing of the VDJ locus was performed using the Adaptive Biotechnologies TCRβ sequencing platform. scTCR-seq data processing, clustering and clonotype quantification scTCR-seq data analyses were executed in R-4.0.1 and R-4.1.3 environments, primarily using Seurat (version 4.0.0-4.1.1) 50,51 . scTCR-seq data were processed as previously described 51,52 . In short, reads were filtered for mitochondrial, ribosomal genes and long non-coding RNA genes. To remove apoptotic cells, low-quality cells and doublets, only cells with a gene expression below 2% for KCNQ1OT1, below 2% for UGDH-AS1, below 2% for GHET1 and expressing between 200 and 5,000 genes were used for further analysis. Quality control (QC)-filtered PBMC and plaque Seurat objects were first merged per patient, after which the patient-merged Seurat objects were normalized using the SCT method, integrated using rpca reduction and clustered according to the Seurat 'scRNA-seq integration' vignette. VDJ sequencing data were imported into Seurat using the combineExpression function of scRepertoire (version 1.4.0) 53 . The complete integrated dataset was mapped to the pbmc_multimodal.h5seurat dataset (https://atlas.fredhutch.org/data/ nygc/multimodal/pbmc_multimodal.h5seurat) to transfer cell type labels to the integrated Seurat object.
For subclustering, T cells were selected from the complete integrated dataset, taking the clusters with protein expression of CD3, CD4 and CD8 and without CD14 expression (ADT assay). Before reclustering the T cells, variable TCR genes were removed from the variable genes list, before principal component analysis (PCA) and clustering, to avoid clustering based on TCR, interfering with clustering on T cell phenotypes. However, TCR genes were not removed from the dataset.
Separate CD4 + T cell and CD8 + T cell objects were then created by subsetting the T cell object based on, respectively, protein expression of CD4 > 0.75 and CD8 > 1.0 in the ADT assay. Custom clonotype counting functions were used to quantify the clonotype content of the individual samples based on the amino acid sequences of the TCRs. Clonotype frequencies relating to the total TCR repertoire per patient, per tissue are depicted in the atherosclerosis figures. Volcano plots were created using EnhancedVolcano (version 1.8.0) 54 . For all volcano plots, the FindMarkers function of Seurat was used to define differential genes between both groups by using a non-parametric Wilcoxon rank-sum test to determine significance. To assess the differentiation trajectories of the CD4 + T cells and CD8 + T cells, Monocle3 and velocyto.R (version 0.6) were used 22,55 . To assess possible interactions of antigen-presenting cells and T cells in the plaque, CellChat (version 1.4.0) was used 25 .

Definition of clonotype expansion levels and tissue enrichment scores
The TCR amino acid sequences were used to define the clonotypes. The clonotype abundance of a clonotype was calculated as the percentage of cells expressing a certain clonotype within a tissue of a patient, divided by the total number of cells in which a TCR was detected in the same tissue of the same patient. Based on the number and percentage of cells expressing the same clonotype, clonotypes were classified as Hyperexpanded, Large, Medium, Small or Single in the tissues of the patients (Supplementary Table 5). Furthermore, the tissue enrichment of clonotypes was determined according to the parameters listed in Supplementary Table 6.
Integration with PSA scTCR-seq data T cells from our scTCR-seq atherosclerosis dataset were compared with TCR-seq data from donor-matched PBMCs and synovial tissue from patients with PSA (ArrayExpress: E-MTAB-9492; European Genomephenome Archive: EGAS00001002104) 31 . The same QC and processing steps were performed for the PSA dataset as described above for our atherosclerosis dataset. Subsequently, the integrated PSA dataset was mapped to the UMAP reduction of our complete T cell object, using our atherosclerosis dataset as reference. Because CD4 + T cells and CD8 + T cells could not be separated cleanly based on the clustering, and the PSA dataset does not contain protein expression data, the atherosclerosis dataset and the PSA dataset were divided based on the predicted cell type (CD4 T cell or CD8 T cell), derived from the pbmc_multimodal. h5seurat dataset. Subsequently, the atherosclerosis and PSA CD4 + T cell and CD8 + T cell datasets were split by patient and reintegrated as previously described for the atherosclerosis object, to form a CD4 + T cell object and a CD8 + T cell object containing atherosclerosis-derived and PSA-derived T cells. Then, the integrated datasets were mapped to our original CD4 + T cell and CD8 + T cell UMAP reductions. Because the PSA dataset is devoid of naive T cells due to the T cell isolation procedure used by Penkava et al. 31 , naive T cell clusters were removed from the CD4 + T cell dataset (clusters 1 and 2) and the CD8 + T cell dataset (cluster 6) before quantification of the clonotype abundance 31 .

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
The raw scTCR-seq data from the Athero-Express cohort are not publicly available due to research participant privacy/consent. These data and the bulk TCRβ sequencing data can be accessed via DataverseNL at this address: https://doi.org/10.34894/DDYKLL. There are restrictions on use by commercial parties and on sharing openly based on (inter)national laws and regulations and written informed consent. Therefore, these data (and additional clinical data) are available only upon discussion and signing a data sharing agreement (see Terms of Access in DataverseNL) Article https://doi.org/10.1038/s44161-022-00208-4 and within a specially designed UMCU-provided environment. Open-source scTCR-seq data from donor-matched PBMCs and synovial tissue from patients with PSA that we used in this study are publicly available (ArrayExpress: E-MTAB-9492; European Genome-phenome Archive: EGAS00001002104) 31 .

Code availability
In silico data analysis was performed using custom-made R scripts designed specifically for this study and/or based on the recommended pipelines from pre-existing packages listed above. R scripts are available via Zenodo (https://doi.org/10.5281/zenodo.7415207).