Pairing of single-cell RNA analysis and T cell antigen receptor profiling indicates breakdown of T cell tolerance checkpoints in atherosclerosis

Abstract

Atherosclerotic plaques form in the inner layer of arteries triggering heart attacks and strokes. Although T cells have been detected in atherosclerosis, tolerance dysfunction as a disease driver remains unexplored. Here we examine tolerance checkpoints in atherosclerotic plaques, artery tertiary lymphoid organs and lymph nodes in mice burdened by advanced atherosclerosis, via single-cell RNA sequencing paired with T cell antigen receptor sequencing. Complex patterns of deteriorating peripheral T cell tolerance were observed being most pronounced in plaques followed by artery tertiary lymphoid organs, lymph nodes and blood. Affected checkpoints included clonal expansion of CD4⁺, CD8⁺ and regulatory T cells; aberrant tolerance-regulating transcripts of clonally expanded T cells; T cell exhaustion; T_reg-TH₁₇ T cell conversion; and dysfunctional antigen presentation. Moreover, single-cell RNA-sequencing profiles of human plaques revealed that the CD8⁺ T cell tolerance dysfunction observed in mouse plaques was shared in human coronary and carotid artery plaques. Thus, our data support the concept of atherosclerosis as a bona fide T cell autoimmune disease targeting the arterial wall.

Extended Data Fig. 1 |. Experimental approach to pairing scRNA-seq analyses with scTCR-seq profiling and blood leukocyte maps of WT and Apoe^−/− mice.

a, Leukocytes from blood, lymph nodes, ATLOs, and plaques from aged Apoe^−/− and WT mice were purified by FACS sorting and barcoded using the Chromium platform (10x Genomics) for scRNA-seq and scTCR-seq as described in methods. b, 7131 blood leukocytes from WT and Apoe^−/− mice were FACS-purified and analyzed by scRNA-seq. 5 major blood immune cell subgroups were defined as T cells, B-cells, monocytes/DCs, NK-cells and granulocytes based on their top 2000 highly expressed genes and graphed as tSNE plots; gray dots represent doublets. c, Blood leukocyte subgroup distribution in tSNE plots of WT vs Apoe^−/− mice. The percentage of cells in each subgroup is indicated. Chi-square test followed by Benjamini-Hochberg correction was used to analyze differences.

Extended Data Fig. 2 |. Characterization of T cell subsets by scRNA-seq and FACS in LNs and the diseased arterial wall.

a, Heatmap shows the top 10 DEGs in each T cell subtype. b, Marker combinations define 10 T cell subtypes in Violin plots. Foxp3 for T_reg T cells, Cd44 for effector/memory T cells, Sell (L-selectin or CD62L) and Ccr7 for naïve T cells, Klrb1c (Killer cell lectin-like receptor) for natural killer T- (NKT) cells, Tcrg-C1 (TCRγ constant region) for γδ T cells, and Gzmk (Granzyme K) for CD8 effector T cells. Subset 1: CD4 effector regulatory (eT_reg) T cells (Cd4⁺Foxp3⁺Cd44⁺Sell⁻Ccr7^low); subset 2: CD4 T_em T cells (Cd4⁺Foxp3⁻Cd44⁺Sell⁻Ccr7^low); subset 3: CD8+ T_em T cells (Cd8a⁺Cd44⁺Sell⁻Ccr7^low); subset 4: CD8 naïve T cells (Cd8a⁺Cd44⁻Sell⁺Ccr7^high); subset 5: CD8+ T_cm T cells (Cd8a⁺Cd44⁺Sell⁺Ccr7^int); subset 6: NKT/CD8 T_cm T cells (Cd8a⁺Cd44⁺Sell⁺Ccr7^intKlrb1c⁺); subset 7: CD4 naïve T cells (Cd4⁺Cd44⁻Sell⁺Ccr7^highLy6c1⁻); subset 8: CD4 central T_reg (cT_reg) T cells (Cd4⁺Foxp3⁺Ccr7^low); subset 9: γδ T cells (Cd4⁻Cd8a⁻Tcrg-C1⁺Trdc⁺); subset 10: CD4 T_cm T cells (Cd4⁺Cd44⁺Sell⁺Ccr7^highLy6c1⁺); each dot represents one cell. c, The percentages of T cell subtypes in LNs, ATLOs and plaques of aged WT and Apoe^−/− mice were determined by FACS. Protein marker combinations were used to define T cell subtypes: CD4 naïve T cells (CD4⁺CD44⁻CD62L⁺), CD8 naïve T cells (CD8⁺CD44⁻CD62L⁺); CD4 T_reg (CD4⁺Foxp3⁺) T cells, CD8 T_cm (CD8⁺CD44⁺CD62L⁺) T cells, CD8 T_em (CD8⁺CD44⁺CD62L⁻) T cells, CD4 T_cm (CD4⁺Foxp3⁻CD44⁺CD62L⁺) T cells and CD4 T_em (CD4⁺Foxp3⁻CD44⁺CD62L⁻) T cells. WT LNs (n = 4), Apoe^−/− LNs (n = 12), ATLOs (n = 6), plaques (n = 6), Data are mean ± s.e.m. One-way ANOVA with Bonferroni post hoc test was used to perform statistical analysis.

Extended Data Fig. 3 |. Characterization of myeloid-cell subsets in RLNs and the diseased arterial wall.

a, Heatmap shows the top 10 DEGs in each myeloid-cell subtype. b, Marker combinations used to define 8 myeloid-cell subtypes are listed in Violin plots: subset 1, CD11b⁺ cDC; subset 2, CD11c⁻ DC; subset 3, CD8a⁺ cDC; subset 4, pDC; subset 5, granulocytes including basophils and neutrophils; subset 6, Lyve1⁺ tissue-resident like macrophages; subset 7, monocyte/macrophages; subset 8, Trem2^hi macrophages. Each dot represents one cell; Ccr7 (chemokine receptor involving in homing of T cells and DCs to lymph nodes), Ccl9 (chemokine or macrophage inflammatory protein-1 gamma attracts DCs); Xcr1 (chemokine receptor mediating leukocyte migration in response to inflammatory mediators), SiglecH (inhibits pDCs inflammatory responses), Trem2 (mainly expressed by macrophages with potential immunosuppressive activities), Ly6c2 (a cell surface glycoprotein mainly expressed by monocyte-macrophages and pDCs), Lyve1 (marker for lymphatic vessels and tissue-resident macrophages with hyaluronan transport activities), S100a8 (marker for granulocytes with calcium- and zinc-binding activities). c, Column graph of numbers of three major myeloid-cell subgroups (Y-axis) in each of four tissues (X-axis). Percentages of basophils/neutrophils, DCs, and macrophages are displayed in each column. d, Percentages of each DC subset per total DCs in WT RLNs, Apoe^−/− RLNs, ATLOs and plaques. The total number of DCs analyzed in each tissue is shown in the inner circle. e, Percentages of each monocyte/macrophage subset per total macrophages in each tissue. The total number of cells analyzed in each tissue is shown in the inner circle.

Extended Data Fig. 4 |. Characterization of the TCR repertoire in RLNs and the diseased arterial wall.

a, TCRβ chain V- and J-gene usages in WT RLNs, Apoe^−/− RLNs, ATLOs, and plaques. V and J segments are represented by arcs, the size of arcs represents their frequencies. V-J pairings are represented by ribbons. b, Distribution of TCRβ V families (TRBV) of total T cells in WT RLNs, Apoe^−/− RLNs, ATLOs, and plaques. Chi-square post hoc test with Benjamini-Hochberg correction was performed to analyze difference among tissues in each TRBV families. c,d, T cell expansion was calculated by measuring the clonal expansion index (that is Gini index) together with the diversification index (i.e Shannon entropy index) of TCRs. Data from each tissue were randomly down-sampled to the same level (95% of plaque T cells with paired TCRαβ chain, that is 161 cells) and this was repeated 1000 times (c,d). The boxplot shows the 25th percentile, median, and 75th percentile values. Each dot represents one time replication. Shapiro-Wilk test was first used to evaluate the data distribution of each group. Data that do not follow normal distribution were analyzed using the Kruskal-Wallis test with Dunn’s non-parametric all-pairs comparison test. The P values were adjusted by Bonferroni correction.

Extended Data Fig. 5 |. Expression of checkpoint-related transcripts in naïve T cells.

a, Violin plots of Vsir, S1pr1, S100a6, Ifngr1, Cd40lg, and Cd69 transcripts in blood naïve CD4 T cells. b, Violin plots of S1pr1, Ccl5, and S100a6 transcripts by blood naïve CD8 T cells. Two-sided Wilcoxon rank sum test with Benjamini-Hochberg correction was used to perform statistical analysis (a,b). c, Violin plots of Vsir, S1pr1, S100a6, Ifngr1, Cd40lg, and Cd69 transcripts in naïve CD4 T cells in four tissues. d, Violin plots of S1pr1, Ccl5, and S100a6 transcripts by tissue naïve CD8 T cells. Each dot represents one single T cell. The difference was calculated by Kruskal-Wallis rank sum test with Dunn’s non-parametric all-pairs comparison test. The P values were adjusted by Benjamini-Hochberg correction (c,d). e, Volcano plot of differentially expressed genes (DEGs) of plaque CD4 naïve T cells vs WT RLNs CD4 naïve T cells. f, Volcano plot of DEGs of plaque CD8 naïve T cells vs WT RLNs of CD8 naïve T cells. Two-sided Wilcoxon-Rank Sum test was used to identify DEGs. Genes with absolute avg_log2FC value >0.25 and adjusted P value <0.05 were considered as significant DEGs. Bonferroni correction was performed to adjust P values (e,f).

Extended Data Fig. 6 |. Representative gene expression of checkpoint-related transcripts in CD4 and CD8 T_eff/mem T cells.

a, Violin plots show representative function-relevant gene expression (S1pr1, Cxcr6, Gzmk, and Pdcd1) in expanded and non-expanded CD8 T_eff/mem T cells of WT RLNs, Apoe^−/− RLNs, ATLOs and plaques. b, Violin plots show representative function-associated gene expressions (S1pr1, Cxcr6, Pdcd1, and Ctla4) in expanded and non-expanded CD4 T_eff/mem T cells of WT RLNs, Apoe^−/− RLNs, ATLOs and plaques. Kruskal-Wallis rank sum test with Dunn’s non-parametric all-pairs comparison test was perform to analyze differences. The P values were adjusted by Benjamini-Hochberg correction (a,b).

Extended Data Fig. 7 |. Average expression of genes with similarities to tissue-resident memory (TRM)-like T cells in different CD8 T_eff/mem T cells among different tissues.

a, 237 plaque CD8 T_eff/mem T cells were grouped into 6 subtypes resulting from tSNE analyses based on their top 2000 highly expressed genes; Dot-plot shows the average expression of genes associated with T cell egression-, activation-, and TRM-like T cell phenotypes; b, 726 ATLO CD8 T_eff/mem T cells were grouped into 5 subtypes; c, 2339 Apoe^−/− RLNs CD8 T_eff/mem T cells were grouped into 6 subtypes; d, 1688 WT RLNs CD8 T_eff/mem T cells were grouped into 6 subtypes; CD8 T_eff/mem T cells were obtained from three independent experiments. Each dot represents one single cell.

Extended Data Fig. 8 |. Dysregulation of costimulatory- and checkpoint inhibitor-related transcripts in APCs in ATLOs and plaques but not in RLNs of WT or Apoe^−/− mice.

a, Myeloid-cell-derived costimulatory genes (Cd80, Cd86, Cd83, Cd40) and checkpoint inhibitor genes (Cd274/PD-L1, Pdcd1lg2/PD-L2, Fas, Icosl, Lgals3, Cd200) in CD45⁺ leukocytes in four tissues. Expression was normalized: Expression = (Average gene expression per cell type)*(Percentage of cell type per CD45⁺ leukocytes). b, Myeloid-cell-derived checkpoint inhibitor genes (Cd274/PD-L1, Pdcd1lg2/PD-L2, Fas, Icosl, Lgals3, Cd200) and costimulatory genes (Cd80, Cd86, Cd83, Cd40) per CD45⁺ leukocytes between WT blood vs Apoe^−/− blood. c, The average expression of costimulatory genes and checkpoint inhibitor genes by different myeloid-cell subsets. d,e, Violin plots show Cd274 and Lgals3 gene expression by different myeloid-cell subsets. Each dot represents one single cell. Two-sided Kruskal-Wallis rank sum test was used to analyze the differences among groups.

Extended Data Fig. 9 |. Predicted cell-cell interactome of CD8 T_em T cells with other immune cell subsets in WT RLNs, Apoe^−/− RLNs, ATLOs, plaques.

a, The x axis represents the CD8 T_em T cell interaction with other immune cell subsets; the y axis depicts paired ligand-receptor interactions. Interactions contributing to T cell dysfunction in plaques when compared to WT RLNs are listed on the right. The size of circles is based on -log10 p values; color scheme is based on log2 value. b, Predicted interactions between CD8 T_em T cells with CD11b⁺ cDCs, pDCs, CD11c⁻ DCs, Trem2^hi macrophages, and Lyve1⁺ res-like macrophages in WT RLNs, Apoe^−/− RLNs, ATLOs, plaques. Color scale indicates CLR-normalized expression, dot size indicates significance (−log10 p value).

Extended Data Fig. 10 |. Comparison of human plaque T-/myeloid-cells and mouse plaque T-/myeloid-cells.

a, tSNE plot shows the expression of T cell canonical marker genes Cd4, Cd8a, Foxp3, and Ccr7 in the human-mouse integrated dataset as colored contour lines. Gene abbreviations on top of each panel; degree of relative expression indicated in colors with red designating high and gray designating low expression; b, Percentages of CD4 T_eff/mem T cells, T_reg T cells, naїve T cells, and γδ T cells in human plaques vs mouse plaques. c, Percentages of DCs and granulocytes in human plaques and mouse plaques. d,e, Dotplots display the tolerance-related genes in CD4 T_eff/mem T cells and DCs in human plaques and mouse plaques. Patient 1–4: human coronary plaques; patient 5–7: human carotid plaques (b-e).

Fig. 1 |. T cell- and myeloid cell-subset-specific maps emerge in the diseased aorta and aorta-draining lymph nodes.

a, 13,800 T cells were grouped into ten subtypes resulting from t-SNE analyses based on their top 2,000 highly expressed genes. b, T cell-subset distribution in the t-SNE plot of WT RLNs, Apoe^−/− RLNs, ATLOs and plaques. c, Percentages of all T cells of the ten subtypes within each tissue. The size of the colored section represents the percentage of each subset per total T cells. d, 2,648 myeloid cells were grouped into eight subtypes resulting from t-SNE analyses, including four DC subtypes, three monocyte/macrophage subtypes and one granulocyte subset consisting of basophils and neutrophils. e, Myeloid cell-subset distribution in t-SNE plots of WT RLNs, Apoe^−/− RLNs, ATLOs and plaques. f, Percentages of eight myeloid cell subsets within each tissue. The size of the colored section represents the percentages for each subset of total myeloid cells. Statistical analysis was performed using Chi-square test with Benjamini–Hochberg correction in c and f.

Fig. 2 |. Pairing of scTCR-seq analyses with scRNA-seq profiling reveals tissue-specific TCRα/β maps in advanced mouse atherosclerosis.

a, Bar plot displays the proportions of clonally expanded (red) versus non-expanded (gray) T cells in different tissues; ≥2 T cells with identical paired TCRα and TCRβ CDR3 sequences are considered as clonally expanded. b, t-SNE analyses separated a total of 13,800 T cells into ten subtypes, and the cells were grouped into five major subgroups according to their biological functions, including CD4⁺ T_eff/mem cells (group I), T_reg cells (group II), CD8⁺ T_eff/mem cells (group III), naїve T cells (group IV) and γδ T cells (group V). c, Pie charts show percentages of clonally expanded versus non-expanded T cells in different groups in different tissues. The sizes of circles represent the relative cell numbers within each tissue. Each circle represents 100%; red indicates the percentages of clonally expanded T cells within each group. a,c, Chi-square post hoc test with Benjamini–Hochberg correction was used to perform statistical analysis. d, Distribution of clonally expanded T cells in t-SNE plot. The clonally expanded T cells are highlighted and bolded in red. e,f, Circos plots show the TCR distribution among four tissues. The number of TCRs of each tissue is represented by arcs displayed in the outer layer of the circle. The size of arcs represents the sample size. Each TCR clone is listed in the inner circle; the size of the inner circle’s segments represents the relative abundance of each clone. Shared TCR clones by two different tissues are represented by ribbons; red represents the shared TCR clones by plaques and other tissues, and purple represents the shared TCR clones by ATLOs, WT RLNs and Apoe^−/− RLNs. The size of ribbons represents the size of individual shared TCR clones.

Fig. 3 |. Tolerance phenotypes at the level of transcripts in non-expanded versus expanded plaque CD8+ and CD4+ effector/memory cells.

a,b, Dot plots of gene expression profiles of selected tolerance-associated genes in CD8⁺ T_eff/mem cells. Genes related to T cell egress/residency (Ccr7, S1pr1, Sell), activation/migration (Cxcr6, Ccr5, Slamf7, Adgre5, S100a4, S100a6, Ctsw), cytotoxicity/cytokines (Cst7, Gzmb, Gzmk, Gzmm, Ifng, Nkg7, Prf1, Efhd2, Ccl5, Ccl4); genes related to exhaustion-associated surface receptors (Pdcd1, Tigit, Lag3, Havcr2), exhaustion-related signaling molecules (Ptpn6, Ptpn11, Ptpn2) and exhaustion-related transcription factors (Irf4, Nr4a1, Gata3, Tox, Batf, Prdm1, Eomes) are listed in b. Non-expanded and expanded CD8⁺ T_eff/mem cells obtained from WT RLNs, Apoe^−/− RLNs, ATLOs and plaques are compared. c,d, Dot plots of gene expression levels of selected function-associated genes in CD4⁺ T_eff/mem cells. Genes related to T cell egress/residency (Ccr7, S1pr1, Sell), activation/migration (Cxcr6, Ccr5, Slamf7, Adgre5, S100a4, S100a6, H2-D1), transcription factors and cytokines (Ifng, Lta, Tbx21, Il5, Il4, Il13, Gata3, Il17a, Rorc) are listed in c; genes related to exhaustion surface receptors (Pdcd1, Tigit, Lag3, Ctla4), exhaustion signaling molecules (Ptpn6, Ptpn11, Ptpn2) and exhaustion transcription factors (Irf4, Nr4a1, Gata3, Tox, Batf, Prdm1, Eomes) are listed in d. Non-expanded and expanded CD4⁺ T_eff/mem cells obtained from WT RLNs, Apoe^−/− RLNs, ATLOs and plaques are compared.

Fig. 4 |. T_reg–TH₁₇ T cell conversion in atherosclerotic plaques.

a, Dot plots of expression levels of selected function-associated genes in T_reg cells. Genes related to IL-2/STAT5 activation (Foxp3, Il2ra, Stat5a), genes highly expressed by effector T_reg cells (Il10, Nrp1, Ctla4, Cd83) and IL-17-signaling pathway-related genes (Rorc, Rora, Il17a). T_reg cells obtained from WT RLNs, Apoe^−/− RLNs, ATLOs and plaques were compared. b, Violin plots show representative gene expression levels associated with T_reg cells. Each dot represents one single T_reg cell. Kruskal–Wallis rank-sum test with Dunn’s non-parametric all-pairs comparison test. The P values were adjusted by Benjamini–Hochberg correction. c, Mean fluorescence intensity (MFI) of Foxp3 protein in T_reg cells of spleen, RLNs, ATLOs and plaques in aged Apoe^−/− mice. CD4⁺Foxp3⁺ protein markers were used to define T_reg cells. Apoe^−/− spleens (n = 9), Apoe^−/− RLNs (n = 8), ATLOs (n = 6), plaques (n = 6). Data are the mean ± s.e.m. Statistical analysis was performed using one-way analysis of variance with Bonferroni post hoc test. d, Anti-CD3e and anti-Foxp3 were used to stain aortic sections of aged Apoe^−/− mice. Data are representative of three experiments. e, Double-positive cells of T_reg–TH₁₇ converting cells. Coexpression of T_reg and TH₁₇ cell markers in single T_reg cells are labeled in red. Each dot represents a single T_reg cell. 1,079 T_reg cells in WT RLNs, 1,358 T_reg cells in Apoe^−/− RLNs, 429 T_reg cells in ATLOs and 73 T_reg cells in plaques were examined. Chi-square test was used to perform statistical analysis. f, Two T cells with identical paired TCR sequences. Cell barcode, the CDR3 aa sequences of paired TCR sequences, gene expression and tissue origin are shown.

Fig. 5 |. Delineation of plaque-inducible T cell transcript profiles.

a, Schematic of experimental design. b, Heat maps of DEGs of pooled T cells carried identical TCRαβ pairs in different tissues. Each row represents one individual T cell. n = 28 T cells in plaques, 27 T cells in ATLOs and 26 T cells in Apoe^−/− RLNs. Two-sided Wilcoxon rank-sum test was used to identify DEGs and significant DEGs were defined by adjusted P value < 0.05. c, Heat maps of plaque-inducible genes in three individual T cell clones. T cells with identical TCR sequences from different tissues were compared. The CDR3 aa sequences of the paired TCRα/β pairs are listed on top of the heat maps. d, Violin plots showed Cxcr6, Lgals1, S100a6 and Reep5 gene expression by different T cell subsets. Each dot represents one cell. Kruskal–Wallis rank-sum test with Benjamini–Hochberg correction was used to analyze the differences among groups.

Fig. 6 |. Comparing human plaque T cell and myeloid cell subsets with mouse plaque T cells.

a, Integration analysis of human plaque T cells and mouse plaque/aorta T cells. t-SNE plot shows T cells obtained from publicly available databanks: GSE131778 and GSE155512: human coronary and carotid plaques; GSE131776: aorta of Apoe^−/− mice fed with a high-fat diet; GSE155513: the aorta of Apoe^−/− and Ldlr^−/− mice fed with high-fat diet; our data, aged Apoe^−/− mice fed with chow diet. b, Comparison of all T cell subsets in aged WT and Apoe^−/− mice and data from other public datasets. Cells from our data were colored; cells from other public databanks were labeled gray. T cell subsets were grouped to five T cell groups according to their functions in the t-SNE plot. Group I: CD4⁺ T_eff/mem cells; group II: T_reg cells; group III: CD8⁺ T_eff/mem cells; group IV: naïve T cells; group V: γδ T cells. c, Percentages of CD8⁺ T_eff/mem cells in human plaques, plaques of aged Apoe^−/− mice, in aortas of high-fat-diet-fed Apoe^−/− or Ldlr^−/− mice at different ages. d, Myeloid cells obtained from the same databanks were integrated and analyzed. e, Comparison of all myeloid cell subsets in aged WT and Apoe^−/− mice and data from other public datasets. Cells from our data were colored. Myeloid cell subsets were assembled into three major groups according to their functions in t-SNE analyses. Group I: macrophages, group II: DCs; group III: granulocytes. f, Percentages of macrophages in human plaques, in plaques of aged Apoe^−/− mice, in aortas of high-fat-diet-fed Apoe^−/− or Ldlr^−/− mice at different ages. g, Heat maps show the similarities between human plaque T cell subsets and mouse plaque T cell subsets. The top 50 subset-specific genes were used to calculate the Spearman correlation coefficient between human and mouse T cell subsets. Mouse T cell subsets from four tissues were compared with human T cell subsets of six individuals. h, Dot plot displays tolerance-related genes in CD8⁺ T_eff/mem cells and macrophages between human and mouse. Participants 1–4: human coronary plaques; participants 5–7: human carotid plaques (c and f–h).

Fig. 7 |. Landscape of T cell and myeloid cell tolerance breakdown in atherosclerosis.

The immune system harbors a comprehensive system of multilayered checkpoints (see 1–5 in blue) to maintain tolerance in WT RLNs under physiological conditions to avoid autoimmune injury of self: (1) Maintenance of quiescence of naïve T cells; (2) support of effector/memory T cell functions; (3) immunosuppression by T_reg cells; (4) antigen presentation to T cells; (5) control of tissue T cell homeostasis. Checkpoints may be compromised at multiple levels (see 1–5 in red) in a tissue-specific manner in mouse advanced atherosclerosis as follows: Apoe^−/− RLNs may be compromised at checkpoint 2; ATLOs may be compromised at checkpoints 1, 2, 3 and 4; plaques may be compromised at all five checkpoints.