Retrospective cohort reveals clinical benefits of anti-PD-1 treatment in resolving AS plaques

To explore the potential impact of anti-PD-1 treatment on the progression of human AS plaques, we conducted a retrospective cohort study. Our study enrolled the tumor patients who were diagnosed and treated at The Second Affiliated Hospital of Zhejiang University (SAHZU) from 1st Jan 2018 to 1st May 2022 (Fig. 1a) and had at least two eligible ultrasound imaging records of carotid plaques before and/or during anti-tumor therapy (Supplementary information, Fig. S1). A total of 168 patients were enrolled. Among them, 86 patients received chemotherapy combined with anti-PD-1 treatment. And 82 patients received only chemotherapy without any anti-PD-1 treatment.

Fig. 1: Retrospective cohort analyses reveal that anti-PD-1 therapy impedes and reverses AS plaque progression in vivo.

a Flowchart showing the identification of eligible patients in the retrospective cohort study. b, c Representative ultrasound images of carotid plaques (b) and comparisons of AS plaque areas (c) in patients treated with or without anti-PD-1 mAb at two scanning time points (Scan 1 and Scan 2). Scar bars, 10 mm. d Comparison of the compositions of AS plaque progression (decrease: ΔA < –1 mm2; no decrease: ΔA ≥ –1 mm2) in groups treated without (n = 82) or with (n = 86) anti-PD-1 treatment. e Comparison of the changes of AS plaque areas (ΔA) between patients with or without anti-PD-1 treatment. f Univariate and multivariate (modified Poisson) regression analysis of the relative ratio (RR) of anti-PD-1 treatment to AS plaque progression in tumor patients (n = 168). Multivariate analyses were adjusted using age, gender, ΔBMI, ΔHDL, ΔLDL, statin usage, tumor types, tumor stage, and tumor progression. Data are represented as median with interquartile range (IQR) in c and e. Paired Mann–Whitney test was used in c and unpaired Mann–Whitney test was in e, and the χ2 test was in d.

By double-blinded analysis of the ultrasound images, we measured the changes in the area of carotid plaque between consecutive images (Scan 1 and Scan 2) for individual patient (Fig. 1b). We discovered a significant reduction in plaque areas (P = 8.2e–05) in patients who received anti-PD-1 treatment (Fig. 1c), among which 60.5% (52 out of 86) had decreased plaque areas (Fig. 1d). In contrast, patients who did not receive anti-PD-1 treatment experienced an increase in AS plaque areas (P = 0.002) (Fig. 1c). Although 29.3% (24 out of 82) of these patients also had decreased areas of AS plaques, this was still significantly lower than the group who received anti-PD-1 treatment (Fig. 1d). The median change of AS plaque areas (ΔA: the change of AS plaque area) was reduced in patients who received anti-PD-1 treatment (ΔA = −3.0 (−7.0, 1.0) mm2), in contrast to the increased median areas in those who did not receive anti-PD-1 treatment (ΔA = 1.0 (−1.0, 5.0) mm2) (Fig. 1e). This decrease in carotid plaque area was observed in individuals with relatively older ages and slightly reduced high-density lipoprotein (HDL) levels. However, no significant differences were observed for other clinical variables such as Body Mass Index (BMI), gender, LDL, statin usage, tumor progression, tumor stage, and tumor type in the anti-PD-1-treated group (Supplementary information, Fig. S1). We further conducted univariate and multivariate (Modified Poisson)22 regression analyses and found that anti-PD-1 treatment was an independent and significant protective factor for the reduction of carotid plaques when comparing the patients who received or did not receive anti-PD-1 treatment (relative ratio (RR) = 0.56 (0.36–0.85), P = 0.007; RR = 0.57 (0.36–0.88), P = 0.013) (Fig. 1f). Collectively, our clinical evidence suggests that anti-PD-1 treatment potentially reduces AS plaques in tumors of patients.

T-cell atlas of human atherosclerosis

To investigate the mechanism by which anti-PD-1 mAb reduces human AS plaques, we used single-cell RNA sequencing (scRNA-seq) and paired single-cell TCR sequencing (scTCR-seq) to characterize CD45+ cells from 4 human AS plaques and 3 paired peripheral blood mononuclear cells (PBMCs) (Fig. 2a). In total, we obtained 62,522 CD45+ cells with an average of 1641 genes per cell and partitioned them into 26 clusters. Based on lineage-specific genes, we annotated 17 T cell clusters (42,921 cells), 2 B cell clusters (2099 cells), 5 natural killer (NK) cell clusters (15,289 cells), and 2 myeloid cell clusters (2213 cells) (Fig. 2b; Supplementary information, Fig. S2a–e). Among the cells extracted from human AS plaques through single-cell processing, T cells were the most abundant immune cell type (Supplementary information, Fig. S2f), consistent with the previously published findings.16 We further re-clustered T cells into 7 CD4+ and 11 CD8+ T cell clusters. These T cell clusters had distinct tissue distributions (Fig. 2c, d) and were annotated as different functional phenotypes based on their differentially expressed genes (DEGs) (Fig. 2b, c; Supplementary information, Table S1), including naïve/central memory T (Tn/Tcm) cells, effector memory cells re-expressing CD45RA (Temra) T cells, mucosal-associated invariant T (MAIT) cells, and regulatory T (Treg) cells. We also identified T helper (Th)17-like cells that expressed CCR6 (CD4-C3), tissue-resident memory T (Trm) cells (CD8-C2) that expressed ZNF683 and RUNX3,23 and proliferating T cells (Tpro; CD8-C11) that expressed a series of proliferation-related genes (STMN1, MKI67, HMGB2, and TPX2). Furthermore, we identified two activated CD8+ T cell clusters (Tact; CD8-C8, -C9) that highly expressed genes encoding inflammatory cytokines (TNF and IFNG), T-cell activation (CD69, JUN, and FOS), and mitochondrial-related metabolic programming (MT-ND2, MT-CO1, and MT-ND6)24 rather than cytotoxic-related genes (NKG7, GZMB, and GNLY) (Fig. 2b), indicating their pro-inflammatory rather than cytotoxic phenotypes in AS plaques.

Fig. 2: scRNA-seq profiling reveals the T-cell atlas of human AS plaques.

a Experimental design for paired scRNA-seq and αβTCR-seq analyses. b DEGs of T cell clusters, the phenotypical definition of each cluster was labeled on the top, and the clusters are colored by both clusters (left and top) and tissue sources of individual cells (top). Typical genes of each cluster are labeled on the right. c Uniform Manifold Approximation and Projection (UMAP) plots of 40,985 T cells from scRNA-seq data, colored by clusters (left) and tissue sources (right). d Composition of CD4+ (left) and CD8+ (right) T cell clusters, colored by sample sources, and clusters were ranked by mean frequencies in AS plaques. Data are represented as means ± SEM. e, f Scatter plots showing log2(fold change) of overlapped DEGs (left) and enriched pathways (right) between CD4-C4 and CD4-C5 clusters (e) and between CD8-C3 and CD8-C4 clusters (f). g UMAP plots of T cells in plaque-specific T cell clusters (mapping based on cell–regulon expression matrix), colored by T cell clusters (top) and AUCell clusters (bottom). h Heatmap showing pairwise TF–regulon correlations, the left bar is colored with the most expressed AUCell cluster, and the right bar is labeled with the dominant AUCell cluster and typical TF–regulons. i Violin plots showing AUCell scores of regulons on identified plaque-specific T cell clusters. Data are represented as means ± SEM in d. A two-sided Student’s t-test with Benjamini–Hochberg adjustment was used in d, and the one-way ANOVA test was used in i.

After analyzing the tissue distributions of T cells (Fig. 2d), we identified two CD4+ (CD4-C4, -C5) and four CD8+ (CD8-C2, -C3, -C4, and -C8) T cell clusters that were predominantly or even exclusively distributed in AS plaques (Fig. 2c, d). Among them, CD4-C4 and CD8-C3 clusters were the most abundant CD4+ (14.9%) and CD8+ (14.8%) T cell clusters in AS plaques, respectively. Both clusters expressed higher levels of LMNA, MCL1, CXCR3, and activation genes (CD44, FOS, and KLF6) than other T cells (Fig. 2b, c; Supplementary information, Fig. S2g), indicating their effector memory-like phenotype. Therefore, we defined them as LMNA+ effector memory T (Tem) cells. Besides, CD4-C5 and CD8-C4 clusters were two plaque-specific Tem-like clusters that not only shared a part of DEGs with LMNA+ Tem cells but also highly expressed PDCD1 (defined as PDCD1+ Tem cells) (Fig. 2b, c; Supplementary information, Fig. S2g). We further found that in CD4+ and CD8+ T cells, both LMNA+ and PDCD1+ Tem cells highly expressed genes of chemokine receptors (CXCR3 and CXCR4), ZFP36 and TNFAIP3; and IL21R, DUSP2, and DUSP4 were expressed exclusively in CD8+ T cells. These data suggest that all of these gene expressions may contribute to restraining the effector functions of T cells, forming the long-lived Tem cells. More importantly, pathway analysis further confirms that plaque-specific LMNA+ and PDCD1+ Tem cells were both enriched in the signaling pathways of “leukocyte activation”, “leukocyte homeostasis”, and “cytokine response” (Fig. 2e, f). Our results imply the potential inflammatory role of these T cells, which might contribute to sustaining the chronic inflammatory homeostasis of atherosclerotic plaques.25,26,27,28

We next applied the single-cell regulatory network inference and clustering (SCENIC) pipeline29 to dissect the key regulons that included essential transcription factors (TFs) and their target genes in plaque-specific T cell clusters. As a result, we identified three regulon-based AUCell clusters (AUC-C1, -C2, and -C3) and found uneven distributions of T cell clusters (Fig. 2g). Tem cells expressing LMNA+ (CD4-C4 and CD8-C3) and PDCD1+ (CD4-C5 and CD8-C4) were mainly distributed in the AUC-C1 cluster, indicating that these cells shared similar transcriptional regulatory pathways (Supplementary information, Fig. S2h). Correlated regulation analysis revealed that T cells in the AUC-C1 cluster highly expressed regulons like REL (29 g), RELA (632 g), RELB (13 g), NFKB1 (20 g), and NFKB2 (25 g) (bottom block in Fig. 2h), suggesting that the activation of nuclear factor-κB (NF-κB) signaling was involved in programming the inflammatory states of LMNA+ and PDCD1+ Tem cells (Fig. 2h, i; Supplementary information, Fig. S2i). Meanwhile, these cells were also enriched in BACH2 (332 g) and FOXO1 (113 g) regulons30,31 that were related to the differentiation of Tem cells (Fig. 2h, i). In contrast, the regulons for IRFs and JAK-STAT signaling pathways (top block in Fig. 2h, i; Supplementary information, Fig. S2i) were highly expressed in the AUC-C3 cluster (mainly CD8-C8, -C9), indicating their distinct activation pathways. Collectively, we identified two distinct pro-inflammatory signaling pathways that independently remodeled the functional states of T cells in AS plaques, and that the activation of the NF-κB signaling pathway was dominated in plaque-specific LMNA+ and PDCD1+ Tem cells, supporting the transcriptional regulation of these T cells in sustaining the chronic inflammation of AS plaques.32,33

LMNA + and PDCD1 + Tem cells are exclusively enriched in human AS plaques

To distinguish AS-specific T cells from those found in other tissues and diseases, we integrated our T cell scRNA-seq data with those of normal colon tissue,34 immunotherapy-induced colitis tissue,34 immunotherapy-induced inflammatory arthritis synovial fluid,35 and lung tumor tissue.36 We obtained 12 CD4+, 7 CD8+, and 1 γδ T cell clusters (Supplementary information, Fig. S3a–d and Table S2), which were annotated as resting T cells (Meta_CD4_C1, _C2, and Meta_CD8_C1), Treg cells (Meta_CD4_C10, _C11), Th17 cells (Meta_CD4_C5), CXCR5+ T follicular helper (Tfh) cells (Meta_CD4 _C7), Trm cells (Meta_CD8_C6), and Temra cells (Meta_CD8_C4, _C5). To compare the similarities of T cell clusters from the other disease datasets, we calculated the scaled expressions of (AS plaque) T cell cluster gene signature (top 30 DEGs) in individual Meta-T cell clusters (Supplementary information, Fig. S3e). We found that Meta_CD8_C2 and _C3 most resembled the LMNA+ and PDCD1+CD8+ T cell clusters (CD8-C3 and CD8-C4) in AS plaques, respectively. Meta_CD4_C6 was similar to the PDCD1+CD4+ T cell cluster (CD4-C5) in AS plaques, and Meta_CD4_C12 highly expressed the gene signature of both LMNA+ T cell clusters (CD4-C4 and CD8-C3) in AS plaques. These T cell clusters, especially CD4+ T cell clusters, had the highest median cell frequencies in AS plaques (Supplementary information, Fig. S3f). We also identified T cell clusters that were mostly enriched in different tissues, including Th17 cells (Meta_CD4_C5) in normal colon tissues, γδT cells (Meta_γδT) in colitis tissues, CXCR5+ Tfh cells (Meta_CD4_C7) in lung tumor tissues, and interferon-responsive T cells (Meta_CD8_C7) in arthritis synovial fluid (Supplementary information, Fig. S3f). Altogether, these results support that the enrichments of LMNA+ and PDCD1+ T cells with pro-inflammatory phenotypes are human AS-specific.

Cytometry by Time-Of-Flight (CyTOF) analysis reveals that PD-1+ T cells in AS plaques are still in the activated state

To characterize the T-cell atlas and functional phenotypes of PD-1+ T cells in human AS plaques, we performed single-cell CyTOF analysis of CD45+ cells from 64 human samples, including 44 AS peripheral blood (PB) and 20 AS plaques. We designed two independent antibody-staining panels (T- and myeloid cell panel; Supplementary information, Table S3) to profile T cells and myeloid cells in depth. After removing Gadolinium (Gd) contamination in AS plaque samples37 and pre-processing CyTOF raw data (Supplementary information, Fig. S4a, b), we performed single-cell clustering analyses followed by frequency correlation analysis between major immune cell types from the two antibody-staining panels, confirming the high consistency of major immune cell type distributions in our parallel experiments (Supplementary information, Fig. S4c–e). We confirmed that T and myeloid cells were the two predominant immune subtypes in AS plaques (62% and 18%, respectively) (Supplementary information, Fig. S4f).

We further analyzed T-cell compositions and identified 35 T cell clusters (Fig. 3a, b; Supplementary information, Fig. S4g), including 18 CD4+ (T01-T18), 12 CD8+ (T19-T30), 2 γδT (T31 and T32), 2 NKT (natural killer T; T33 and T34), and 1 DNT (double negative T; T35) cell clusters. T cell compositions dramatically altered across tissues, with increased fractions of CD8+ T cells and decreased fractions of CD4+ T, γδT, and NKT cells in AS plaques compared to those in AS PB (Fig. 3c; Supplementary information, Fig. S4h). Moreover, plaque-specific T cells mostly consisted of non-cytotoxic Tem cells (CD45RA–CCR7–) with reduced fractions of Tn (CD45RA+ CCR7+) and effector T (Teff; Granzyme B+CD45RA+T-bet+) cells.

Fig. 3: T-cell atlas of human AS plaques and AS PB revealed by single-cell CyTOF analysis.

a Heatmap displaying the median expression of 35 T cell clusters (T-cell panel), labeled with major or functional subsets (left) and cluster frequency (right). b t-SNE plots of T cells, colored by clusters or sample groups. c Compositions of major (left) and functional (right) T cell subsets in AS PB and AS plaques. d Volcano plots showing different frequencies of CD4+ (left) and CD8+ (right) T cell clusters in AS plaques compared to those in AS PB, colored by dominating tissue types, and arrows indicate PD-1+ T cell clusters. e Multicolor IFC staining confirming PD-1+CD4+ and PD-1+CD8+ T cells in a representative human AS plaque. Scale bars, 20 μm. f Histograms showing selected functional marker expressions on PD-1+CD4+ (top) and PD-1+CD8+ (bottom) T cell clusters. g Histograms showing ICOS, HLA-DR, CD27, and CD28 expressions on plaque-derived PD-1+ and PD-1– T cells. h Dot plots showing expressions of exhaustion-related co-inhibitory and regulatory genes in T cell clusters, colored by scaled mean expression and sized by fraction of cells expressing specified genes. T cell clusters were ranked by their AUCell scores of PDCD1 gene signature as displayed in Supplementary information, Fig. S4l. i Venn plot of shared DEGs between PDCD1+ T cell clusters (CD4-C5 and CD8-C4), with the numbers of intersected or exclusive genes labeled. j Dot plots showing the calculated regulon specificity scores of PDCD1+ T cell clusters (CD4-C5 and CD8-C4), with the top-10 regulons labeled. Data are presented as median with IQR in c. A two-sided Student’s t-test with Benjamini–Hochberg adjustment was used for statistical analyses in c and d. The Kolmogorov–Smirnov test was used in g and the hypergeometric test in i.

We compared the fold changes in cell frequencies of CD4+ and CD8+ T cell clusters (Fig. 3d) and identified that 5 PD-1+ (T11, T13, and T14 for CD4+; T24 and T27 for CD8+) T cell clusters were exclusively enriched in AS plaques, and the existence of CD4+PD-1+ and CD8+PD-1+ T cells was further supported by immunofluorescence staining of human AS plaques (Fig. 3e; Supplementary information, Fig. S4i). Subsequently, we found that co-stimulatory molecules (CD28 and ICOS) were co-expressed on PD-1+ T cells, activating molecules (HLA-DR and CD27) were particularly highly expressed on PD-1+CD8+ T cells (Fig. 3f). Furthermore, these 4 functional markers were significantly higher expressed on PD-1+ T cells than PD-1– T cells in AS plaques (Fig. 3g), and their expression levels were also highly correlated with PD-1 expression (Supplementary information, Fig. S4j). Besides, activating molecule CD3838 was also highly expressed on PD-1+ T cells (T13 and T27) (Fig. 3f). Altogether, these results indicate that PD-1+ T cells are mainly located in human AS plaques and suggest that they do not resemble terminally-differentiated exhausted T (Tex) cells in cancers.39

scRNA-seq confirms the functionally activated state of PDCD1 + T cells in human AS plaques

PD-1 expression is theoretically induced by T-cell activation and contributes to T-cell inhibition, memory, and homeostasis, as well as immune tolerance.40 The sustained co-expression of PD-1 with other co-inhibitory immune checkpoint receptors, such as lymphocyte-activation gene-3 (Lag-3), T cell immunoglobulin and mucin-domain containing-3 (Tim-3), and T cell immunoglobulin and ITIM (immunoreceptor tyrosine-based inhibitory motif) domain (TIGIT), has been validated as the hallmark of T-cell exhaustion in tumor diseases.41,42 We examined the functionalities of PD-1+ T cells in AS plaques by deeply analyzing the single-cell transcriptomes of plaque-specific T cells. We identified the top 30 genes that were highly correlated with PDCD1 expression in T cells, including DUSP2, DUSP4, CXCR3, CXCR4, ICOS, etc., and defined them as the plaque-specific PDCD1 gene signature (Supplementary information, Fig. S4k). We calculated and ranked the AUCell score29 for each T cell cluster and found that CD4-C5 and CD8-C4 clusters, which expressed the highest level of PDCD1, were ranked at the top and defined as PDCD1+ T cell clusters in AS plaques (Supplementary information, Fig. S4l). Consistent with the aforementioned CyTOF results (Fig. 3f, g), ICOS was co-expressed with PDCD1 at the transcriptome level (Supplementary information, Fig. S4k). However, the typical co-inhibitory molecules or regulators related to T-cell exhaustion, such as HAVCR2, LAG3, TIGIT, and TOX,41 were not co-expressed on these cells, thereby not being ranked within the genes correlated with PDCD1 (Fig. 3h; Supplementary information, Fig. S4k). Furthermore, we did not observe significant expressions of dysfunctional T cell-associated genes43 in our PDCD1+ T cell clusters (Supplementary information, Fig. S4m). Collectively, these findings indicate that PDCD1+ T cells in human AS plaques are functionally distinct from those exhausted PDCD1+ tumor-infiltrating lymphocytes (TILs).43,44,45

The PD-1 signaling pathway, accompanied by the activation of TCR signals,46,47 also contributes to the maintenance of T cell memory. By analyzing the DEGs of PDCD1+ T cells (CD4-C5 and CD8-C4), we identified 30 genes that were shared by these two clusters (Fig. 3i), including genes associated with amino acid transport (SLC7A5), T-cell activation (DUSP2, DUSP4, CREM, ARID5A, PDE4B, RNF125, and NR4A2), and chemotaxis (CXCR3 and CXCR4) in response to inflammation.48,49,50,51,52 We further calculated the regulon specificity scores29 of these two PDCD1+ T cell clusters to identify their transcriptomic regulations (Fig. 3j) and found key regulators in the CD4-C5 cluster, such as transcription factors RORA related to colitis and inflammation,53,54 and PRDM1 related to effector functions of T cells.55 Meanwhile, we also found key regulators in the CD8-C4 cluster, such as the transcription factors IRF4, EOMES, and REL, which all play important roles in regulating the differentiation and effector function of T cells.56,57,58 Herein, these analyses reveal that the transcriptomic regulations of PDCD1+ T cells in AS plaques are different from those of terminally-differentiated exhausted TILs, but rather are more activated.

Besides, we compared our transcriptomic dataset of T cells in AS plaques with that of the previous study16 and identified 8 T cell clusters. Among them, F_C0 and F_C2 highly expressed genes, such as NFKBIA, FOS, DUSP1, DUSP2, and LMNA, which shared a similar phenotype with LMNA+ Tem cells (CD4-C4 and CD8-C3) in our dataset (Fig. 2b; Supplementary information, S5a–c and Table S4). We then calculated the pairwise AUCell scores of T cell clusters in their dataset by using the top 30 DEGs of T cell clusters identified in our dataset and found that F_C2 represented the mixture of LMNA+ Tem cells (CD4-C4 and CD8-C3), PDCD1+ Tem cells (CD4-C5 and CD8-C4), and activated T cells (CD8-C8 and CD8-C9) in our dataset, which also had the highest expression level of PDCD1 (Supplementary information, Fig. S5d). Meanwhile, F_C2 did not express the other T cell exhaustion-related genes, such as HAVCR2, LAG3, CTLA4, TIGIT, and TOX (Supplementary information, Fig. S5e), and was not enriched in the dysfunctional gene signature,43 indicating the existence of unexhausted PDCD1+ T cells also in their datasets (Supplementary information, Fig. S5f). Altogether, this independent study also supports our conclusions about the existence of pro-inflammatory and non-exhausted LMNA+ and PDCD1+ Tem cells in human AS plaques.

Epigenetic footprints and regulations of AS plaque-specific PDCD1 + T cells

Single-cell chromatin landscape can reveal both the chromatin accessibility states of cell types and the critical gene regulators that program cellular functions. To investigate the chromatin accessibility states of T cell activation- or exhaustion-related genes in AS plaque-specific T cells, CD3+ T cells were sorted from four AS plaque samples (Supplementary information, Fig. S6a), and a single-nucleus assay for transposase-accessible chromatin using sequencing (snATAC-seq) analyses was performed. After data processing and quality control, 5598 single nuclei were obtained and segregated into 12 T cell clusters (Supplementary information, Fig. S6b and Table S5). Comparing the inferred gene activity scores of each cluster (Supplementary information, Fig. S6c, d), we annotated these clusters as the resting CD4+ (C8) and CD8+ (C1) T cells (Tres), CD8+ Teff cells (C4), Treg cells (C11), Th17-like cells (C10), CD8+ activated T cells (C6 and C7), and γδT cells (C12). Consistent with the gene signature of LMNA+ and PDCD1+ T cells in AS plaques from scRNA-seq data (Fig. 2b), we found CD4+ Tem cells (C9) and CD8+ Tem cells (C3 and C5) with high gene activity scores of CD44, CD69, DUSP4, LMNA, and PDCD1 (Supplementary information, Fig. S6d). C2 cluster was annotated as Tem-like cells because of their lower gene activities of cytotoxic genes (e.g., PRF1, GZMB, and NKG7) but higher gene activity of LMNA compared to CD8+ Teff cells (C4) (Supplementary information, Fig. S6d). We then used chromVAR59 to identify cluster-specific TF regulatory elements and found LEF1 and TCF7 particularly enriched in the resting T cells (C1 and C8), RORA in Th17-like T cells (C10), and YY1 in Treg cells (C11) (Supplementary information, Fig. S6e–g). Consistent with SCENIC analyses (Fig. 2g–i), the plaque-specific Tem and Tem-like cells (C2, C3, C5, and C9) were enriched in the motifs of activator protein-1 (AP-1) TFs (FOS and JUNB), BATF, and BACH2, whereas the activated CD8+ T cells (C6 and C7) were enriched in motifs of IRF3, STAT1, and STAT2 (Supplementary information, Fig. S6e–g).

To investigate the differential chromatin accessibility of AS plaque-specific CD8+PD-1+ Tem cells (C3 and C5), we integrated our snATAC-seq dataset with the exhausted CD8+ T cells from basal cell carcinoma.60 Compared with CD8+ Tres and Teff cells (C1 and C4), both CD8+ Tex (tumor-specific) and Tem (C3, C5; AS plaque-specific) cells exhibited higher accessibility of +5Kb and –5Kb cis-elements of PDCD1 locus,60,61 and also higher inferred gene activity of PDCD1 (Fig. 4a, b). However, the chromatin accessibilities of other T cell exhaustion-related genes (e.g., CTLA4, HAVCR2, and ENTPD1)60 were not enriched in plaque-specific Tem cells but were exclusively enriched in tumor-specific Tex cells (Fig. 4a, b). This indicates that the chromat